summaryrefslogtreecommitdiff
path: root/dviware
diff options
context:
space:
mode:
authorNorbert Preining <norbert@preining.info>2023-11-16 03:01:15 +0000
committerNorbert Preining <norbert@preining.info>2023-11-16 03:01:15 +0000
commite0db3f0709a737e7258fd40a7cbfe9a461ea0756 (patch)
treee81cfc35b6602ffceefa73b8971ce3b94312820a /dviware
parentf5460c92605f1b3b17448b27ce53375157c3e4b1 (diff)
CTAN sync 202311160301
Diffstat (limited to 'dviware')
-rw-r--r--dviware/dvisvgm/Makefile.in2
-rw-r--r--dviware/dvisvgm/NEWS16
-rw-r--r--dviware/dvisvgm/aminclude_static.am2
-rwxr-xr-xdviware/dvisvgm/configure22
-rw-r--r--dviware/dvisvgm/configure.ac4
-rw-r--r--dviware/dvisvgm/doc/conf-dblatex-man.xsl26
-rw-r--r--dviware/dvisvgm/doc/conf-dblatex-pdf.xsl112
-rw-r--r--dviware/dvisvgm/doc/dvisvgm.152
-rw-r--r--dviware/dvisvgm/doc/dvisvgm.txt.in26
-rw-r--r--dviware/dvisvgm/doc/tweak-db-refentry.xsl84
-rw-r--r--dviware/dvisvgm/doc/tweak-dblatex-pdf.xsl42
-rw-r--r--dviware/dvisvgm/libs/brotli/Makefile.am5
-rw-r--r--dviware/dvisvgm/libs/brotli/Makefile.in41
-rw-r--r--dviware/dvisvgm/libs/brotli/common/constants.c2
-rw-r--r--dviware/dvisvgm/libs/brotli/common/constants.h3
-rw-r--r--dviware/dvisvgm/libs/brotli/common/context.c2
-rw-r--r--dviware/dvisvgm/libs/brotli/common/dictionary.c8
-rw-r--r--dviware/dvisvgm/libs/brotli/common/platform.c3
-rw-r--r--dviware/dvisvgm/libs/brotli/common/platform.h173
-rw-r--r--dviware/dvisvgm/libs/brotli/common/shared_dictionary_internal.h75
-rw-r--r--dviware/dvisvgm/libs/brotli/common/transform.c2
-rw-r--r--dviware/dvisvgm/libs/brotli/common/version.h37
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/backward_references.c100
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/backward_references.h9
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/backward_references_hq.c200
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/backward_references_hq.h11
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/backward_references_inc.h36
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/bit_cost.c15
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/bit_cost.h9
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/block_splitter.c51
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/block_splitter.h9
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/block_splitter_inc.h131
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.c242
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.h17
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/cluster.c19
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/cluster.h13
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/cluster_inc.h45
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/command.c2
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/command.h9
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/compound_dictionary.c207
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/compound_dictionary.h74
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/compress_fragment.c176
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/compress_fragment.h39
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.c186
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.h24
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/dictionary_hash.c4
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/encode.c875
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/encoder_dict.c615
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/encoder_dict.h120
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/entropy_encode.c5
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/entropy_encode.h3
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/entropy_encode_static.h7
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/fast_log.c2
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/fast_log.h3
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/find_match_length.h37
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash.h330
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_composite_inc.h37
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_forgetful_chain_inc.h38
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_longest_match64_inc.h73
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_longest_match_inc.h16
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_longest_match_quickly_inc.h8
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_rolling_inc.h8
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/hash_to_binary_tree_inc.h11
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/histogram.c8
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/histogram.h13
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/literal_cost.c35
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/literal_cost.h6
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/memory.c34
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/memory.h21
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/metablock.c130
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/metablock.h17
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/metablock_inc.h14
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/params.h7
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/prefix.h5
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/quality.h43
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/ringbuffer.h7
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/state.h104
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/static_dict.c64
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/static_dict.h5
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/static_dict_lut.h2
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/utf8_util.c2
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/utf8_util.h3
-rw-r--r--dviware/dvisvgm/libs/brotli/enc/write_bits.h3
-rw-r--r--dviware/dvisvgm/libs/brotli/include/brotli/decode.h69
-rw-r--r--dviware/dvisvgm/libs/brotli/include/brotli/encode.h57
-rw-r--r--dviware/dvisvgm/libs/brotli/include/brotli/port.h39
-rw-r--r--dviware/dvisvgm/libs/brotli/include/brotli/shared_dictionary.h100
-rw-r--r--dviware/dvisvgm/libs/xxHash/xxh3.h2
-rw-r--r--dviware/dvisvgm/libs/xxHash/xxhash.c2
-rw-r--r--dviware/dvisvgm/libs/xxHash/xxhash.h2960
-rw-r--r--dviware/dvisvgm/src/Font.cpp15
-rw-r--r--dviware/dvisvgm/src/Makefile.am3
-rw-r--r--dviware/dvisvgm/src/Makefile.in13
-rw-r--r--dviware/dvisvgm/src/PSFilter.hpp41
-rw-r--r--dviware/dvisvgm/src/PSInterpreter.cpp23
-rw-r--r--dviware/dvisvgm/src/PSInterpreter.hpp2
-rw-r--r--dviware/dvisvgm/src/PSPreviewHandler.cpp (renamed from dviware/dvisvgm/src/PSPreviewFilter.cpp)85
-rw-r--r--dviware/dvisvgm/src/PSPreviewHandler.hpp (renamed from dviware/dvisvgm/src/PSPreviewFilter.hpp)25
-rw-r--r--dviware/dvisvgm/src/PsSpecialHandler.cpp22
-rw-r--r--dviware/dvisvgm/src/PsSpecialHandler.hpp6
-rw-r--r--dviware/dvisvgm/src/SVGCharPathHandler.cpp4
-rw-r--r--dviware/dvisvgm/src/optimizer/ClipPathReassigner.cpp1
-rw-r--r--dviware/dvisvgm/src/optimizer/SVGOptimizer.cpp56
-rw-r--r--dviware/dvisvgm/src/options.xml386
-rw-r--r--dviware/dvisvgm/src/psdefs.cpp253
-rw-r--r--dviware/dvisvgm/src/utility.cpp7
-rw-r--r--dviware/dvisvgm/src/utility.hpp2
-rw-r--r--dviware/dvisvgm/tests/UtilityTest.cpp3
-rw-r--r--dviware/dvisvgm/tests/normalize.xsl220
109 files changed, 6447 insertions, 3027 deletions
diff --git a/dviware/dvisvgm/Makefile.in b/dviware/dvisvgm/Makefile.in
index 4096c579b8..215bcf3963 100644
--- a/dviware/dvisvgm/Makefile.in
+++ b/dviware/dvisvgm/Makefile.in
@@ -15,7 +15,7 @@
@SET_MAKE@
# aminclude_static.am generated automatically by Autoconf
-# from AX_AM_MACROS_STATIC on Wed Jul 5 12:26:42 CEST 2023
+# from AX_AM_MACROS_STATIC on Thu Oct 12 14:50:13 CEST 2023
VPATH = @srcdir@
am__is_gnu_make = { \
if test -z '$(MAKELEVEL)'; then \
diff --git a/dviware/dvisvgm/NEWS b/dviware/dvisvgm/NEWS
index cf7a347f47..0e996e2a74 100644
--- a/dviware/dvisvgm/NEWS
+++ b/dviware/dvisvgm/NEWS
@@ -1,3 +1,19 @@
+dvisvgm-3.1.2 (2023-10-12)
+- improved evaluation of graphic extents created by the 'preview' package
+ (GH issue #246)
+- fixed computation of character boxes when using --exact-bbox (GH issue #248)
+- fixed memory error in optimizer module (GH issue #250)
+- updated bundled brotli library to version 1.1.0
+
+dvisvgm-3.1.1 (2023-08-29)
+- added evaluation of optional dash prefixes to command-line option --optimize
+ to exclude selected optimizer modules
+- fixed invalid removal of clipping paths referenced in 'defs' section
+- fixed automatic calls of optimizer module 'remove-clippaths'
+- suppress Ghostscript warning about deprecated option -dNEWPDF (GH issue #245)
+- updated bundled xxHash library to version 0.8.2
+- few fixes in the manual page
+
dvisvgm-3.1 (2023-07-06)
- added command-line option --currentcolor to replace a given color with
CSS variable 'currentColor' (GH issue #214)
diff --git a/dviware/dvisvgm/aminclude_static.am b/dviware/dvisvgm/aminclude_static.am
index 51bd6c838b..b13d91489c 100644
--- a/dviware/dvisvgm/aminclude_static.am
+++ b/dviware/dvisvgm/aminclude_static.am
@@ -1,6 +1,6 @@
# aminclude_static.am generated automatically by Autoconf
-# from AX_AM_MACROS_STATIC on Wed Jul 5 12:26:42 CEST 2023
+# from AX_AM_MACROS_STATIC on Thu Oct 12 14:50:13 CEST 2023
# Code coverage
diff --git a/dviware/dvisvgm/configure b/dviware/dvisvgm/configure
index 470bf23060..c878cdc515 100755
--- a/dviware/dvisvgm/configure
+++ b/dviware/dvisvgm/configure
@@ -1,6 +1,6 @@
#! /bin/sh
# Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.71 for dvisvgm 3.1.
+# Generated by GNU Autoconf 2.71 for dvisvgm 3.1.2.
#
# Report bugs to <martin.gieseking@uos.de>.
#
@@ -621,8 +621,8 @@ MAKEFLAGS=
# Identity of this package.
PACKAGE_NAME='dvisvgm'
PACKAGE_TARNAME='dvisvgm'
-PACKAGE_VERSION='3.1'
-PACKAGE_STRING='dvisvgm 3.1'
+PACKAGE_VERSION='3.1.2'
+PACKAGE_STRING='dvisvgm 3.1.2'
PACKAGE_BUGREPORT='martin.gieseking@uos.de'
PACKAGE_URL=''
@@ -1437,7 +1437,7 @@ if test "$ac_init_help" = "long"; then
# Omit some internal or obsolete options to make the list less imposing.
# This message is too long to be a string in the A/UX 3.1 sh.
cat <<_ACEOF
-\`configure' configures dvisvgm 3.1 to adapt to many kinds of systems.
+\`configure' configures dvisvgm 3.1.2 to adapt to many kinds of systems.
Usage: $0 [OPTION]... [VAR=VALUE]...
@@ -1508,7 +1508,7 @@ fi
if test -n "$ac_init_help"; then
case $ac_init_help in
- short | recursive ) echo "Configuration of dvisvgm 3.1:";;
+ short | recursive ) echo "Configuration of dvisvgm 3.1.2:";;
esac
cat <<\_ACEOF
@@ -1653,7 +1653,7 @@ fi
test -n "$ac_init_help" && exit $ac_status
if $ac_init_version; then
cat <<\_ACEOF
-dvisvgm configure 3.1
+dvisvgm configure 3.1.2
generated by GNU Autoconf 2.71
Copyright (C) 2021 Free Software Foundation, Inc.
@@ -2133,7 +2133,7 @@ cat >config.log <<_ACEOF
This file contains any messages produced by compilers while
running configure, to aid debugging if configure makes a mistake.
-It was created by dvisvgm $as_me 3.1, which was
+It was created by dvisvgm $as_me 3.1.2, which was
generated by GNU Autoconf 2.71. Invocation command line was
$ $0$ac_configure_args_raw
@@ -3110,7 +3110,7 @@ ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $
ac_compiler_gnu=$ac_cv_c_compiler_gnu
-DATE="June 2023"
+DATE="October 2023"
ac_config_headers="$ac_config_headers config.h"
@@ -3704,7 +3704,7 @@ fi
# Define the identity of the package.
PACKAGE='dvisvgm'
- VERSION='3.1'
+ VERSION='3.1.2'
printf "%s\n" "#define PACKAGE \"$PACKAGE\"" >>confdefs.h
@@ -21279,7 +21279,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
# report actual input values of CONFIG_FILES etc. instead of their
# values after options handling.
ac_log="
-This file was extended by dvisvgm $as_me 3.1, which was
+This file was extended by dvisvgm $as_me 3.1.2, which was
generated by GNU Autoconf 2.71. Invocation command line was
CONFIG_FILES = $CONFIG_FILES
@@ -21347,7 +21347,7 @@ ac_cs_config_escaped=`printf "%s\n" "$ac_cs_config" | sed "s/^ //; s/'/'\\\\\\\\
cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
ac_cs_config='$ac_cs_config_escaped'
ac_cs_version="\\
-dvisvgm config.status 3.1
+dvisvgm config.status 3.1.2
configured by $0, generated by GNU Autoconf 2.71,
with options \\"\$ac_cs_config\\"
diff --git a/dviware/dvisvgm/configure.ac b/dviware/dvisvgm/configure.ac
index 4d4432a3ce..9338a0c87b 100644
--- a/dviware/dvisvgm/configure.ac
+++ b/dviware/dvisvgm/configure.ac
@@ -4,8 +4,8 @@
# Process this file with autoconf to produce a configure script.
AC_PREREQ([2.69])
-AC_INIT([dvisvgm],[3.1],[martin.gieseking@uos.de])
-DATE="June 2023"
+AC_INIT([dvisvgm],[3.1.2],[martin.gieseking@uos.de])
+DATE="October 2023"
AC_CONFIG_SRCDIR(src)
AC_CONFIG_HEADERS([config.h])
AC_CONFIG_MACRO_DIR([m4])
diff --git a/dviware/dvisvgm/doc/conf-dblatex-man.xsl b/dviware/dvisvgm/doc/conf-dblatex-man.xsl
index 23fb9ff2c4..a1e336376c 100644
--- a/dviware/dvisvgm/doc/conf-dblatex-man.xsl
+++ b/dviware/dvisvgm/doc/conf-dblatex-man.xsl
@@ -2,18 +2,18 @@
<!-- This file is part of dvisvgm -->
<!-- Copyright (C) 2015-2023 Martin Gieseking <martin.gieseking@uos.de> -->
<xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform" version="1.0">
- <!-- output monospaced text in bold -->
- <xsl:template match="literal">
- <xsl:text>\fB</xsl:text>
- <xsl:apply-templates/>
- <xsl:text>\fR</xsl:text>
- </xsl:template>
+ <!-- output monospaced text in bold -->
+ <xsl:template match="literal">
+ <xsl:text>\fB</xsl:text>
+ <xsl:apply-templates/>
+ <xsl:text>\fR</xsl:text>
+ </xsl:template>
- <!-- append URL to hyperlinked text pointing to external targets -->
- <xsl:template match="ulink[not(contains(., '://')) and not(contains(@url, 'mailto:'))]">
- <xsl:apply-templates/>
- <xsl:text> (\m[blue]</xsl:text>
- <xsl:value-of select="@url"/>
- <xsl:text>\m[])</xsl:text>
- </xsl:template>
+ <!-- append URL to hyperlinked text pointing to external targets -->
+ <xsl:template match="ulink[not(contains(., '://')) and not(contains(@url, 'mailto:'))]">
+ <xsl:apply-templates/>
+ <xsl:text> (\m[blue]</xsl:text>
+ <xsl:value-of select="@url"/>
+ <xsl:text>\m[])</xsl:text>
+ </xsl:template>
</xsl:stylesheet>
diff --git a/dviware/dvisvgm/doc/conf-dblatex-pdf.xsl b/dviware/dvisvgm/doc/conf-dblatex-pdf.xsl
index d39e03c3a3..98a630d849 100644
--- a/dviware/dvisvgm/doc/conf-dblatex-pdf.xsl
+++ b/dviware/dvisvgm/doc/conf-dblatex-pdf.xsl
@@ -2,67 +2,67 @@
<!-- This file is part of dvisvgm -->
<!-- Copyright (C) 2015-2023 Martin Gieseking <martin.gieseking@uos.de> -->
<xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform">
- <xsl:param name="xetex.font">
- <xsl:text>\setmainfont{Source Serif Pro}&#10;</xsl:text>
- <xsl:text>\setsansfont{Source Sans Pro}&#10;</xsl:text>
- <xsl:text>\setmonofont[Scale=0.9]{Source Code Pro}&#10;</xsl:text>
- </xsl:param>
+ <xsl:param name="xetex.font">
+ <xsl:text>\setmainfont{Source Serif 4}&#10;</xsl:text>
+ <xsl:text>\setsansfont{Source Sans 3}&#10;</xsl:text>
+ <xsl:text>\setmonofont[Scale=0.9]{Source Code Pro}&#10;</xsl:text>
+ </xsl:param>
- <xsl:param name="doc.toc.show">0</xsl:param>
- <xsl:param name="refentry.numbered">0</xsl:param>
- <xsl:param name="figure.caution">caution</xsl:param>
- <xsl:param name="figure.note">note</xsl:param>
- <xsl:param name="literal.layout.options"/>
- <xsl:param name="table.default.tabstyle">tabular</xsl:param>
- <xsl:param name="table.in.float">0</xsl:param>
- <xsl:param name="term.breakline">1</xsl:param>
+ <xsl:param name="doc.toc.show">0</xsl:param>
+ <xsl:param name="refentry.numbered">0</xsl:param>
+ <xsl:param name="figure.caution">caution</xsl:param>
+ <xsl:param name="figure.note">note</xsl:param>
+ <xsl:param name="literal.layout.options"/>
+ <xsl:param name="table.default.tabstyle">tabular</xsl:param>
+ <xsl:param name="table.in.float">0</xsl:param>
+ <xsl:param name="term.breakline">1</xsl:param>
- <xsl:param name="literal.layout.options">
- <xsl:text>basicstyle=\ttfamily\small,backgroundcolor=\color[gray]{0.9},columns=fullflexible,frame=single</xsl:text>
- </xsl:param>
+ <xsl:param name="literal.layout.options">
+ <xsl:text>basicstyle=\ttfamily\small,backgroundcolor=\color[gray]{0.9},columns=fullflexible,frame=single</xsl:text>
+ </xsl:param>
- <!-- ensure processing of refmiscinfo elements -->
- <xsl:template match="refentry">
- <xsl:apply-templates select="refmeta/refmiscinfo"/>
- <xsl:apply-imports/>
- </xsl:template>
+ <!-- ensure processing of refmiscinfo elements -->
+ <xsl:template match="refentry">
+ <xsl:apply-templates select="refmeta/refmiscinfo"/>
+ <xsl:apply-imports/>
+ </xsl:template>
- <!-- put refmiscinfo data into the TeX file -->
- <xsl:template match="refmiscinfo[@class]">
- <xsl:value-of select="concat('\def\refmiscinfo', @class, '{', ., '}&#10;')"/>
- </xsl:template>
+ <!-- put refmiscinfo data into the TeX file -->
+ <xsl:template match="refmiscinfo[@class]">
+ <xsl:value-of select="concat('\def\refmiscinfo', @class, '{', ., '}&#10;')"/>
+ </xsl:template>
- <!-- output TOC before synopsis section -->
- <xsl:template match="refsynopsisdiv">
- <xsl:text>\tableofcontents&#10;</xsl:text>
- <xsl:call-template name="makeheading">
- <xsl:with-param name="level">2</xsl:with-param>
- <xsl:with-param name="num">1</xsl:with-param>
- </xsl:call-template>
- <xsl:apply-templates/>
- </xsl:template>
+ <!-- output TOC before synopsis section -->
+ <xsl:template match="refsynopsisdiv">
+ <xsl:text>\tableofcontents&#10;</xsl:text>
+ <xsl:call-template name="makeheading">
+ <xsl:with-param name="level">2</xsl:with-param>
+ <xsl:with-param name="num">1</xsl:with-param>
+ </xsl:call-template>
+ <xsl:apply-templates/>
+ </xsl:template>
- <!-- force non-star \subsection commands -->
- <xsl:template match="refsect1">
- <xsl:call-template name="makeheading">
- <xsl:with-param name="level">2</xsl:with-param>
- <xsl:with-param name="num">1</xsl:with-param>
- </xsl:call-template>
- <xsl:apply-templates/>
- </xsl:template>
+ <!-- force non-star \subsection commands -->
+ <xsl:template match="refsect1">
+ <xsl:call-template name="makeheading">
+ <xsl:with-param name="level">2</xsl:with-param>
+ <xsl:with-param name="num">1</xsl:with-param>
+ </xsl:call-template>
+ <xsl:apply-templates/>
+ </xsl:template>
- <!-- add named list items (e.g. command-line options, specials) to PDF bookmarks -->
- <xsl:template match="refsect1/variablelist/varlistentry/term[emphasis]">
- <xsl:text>\phantomsection\pdfbookmark[3]{</xsl:text>
- <xsl:choose>
- <xsl:when test="contains(emphasis, '--')">
- <xsl:value-of select="substring-after(emphasis, '--')"/>
- </xsl:when>
- <xsl:otherwise>
- <xsl:value-of select="emphasis"/>
- </xsl:otherwise>
- </xsl:choose>
- <xsl:value-of select="concat('}{', generate-id(emphasis), '}&#10;')"/>
- <xsl:apply-imports/>
- </xsl:template>
+ <!-- add named list items (e.g. command-line options, specials) to PDF bookmarks -->
+ <xsl:template match="refsect1/variablelist/varlistentry/term[emphasis]">
+ <xsl:text>\phantomsection\pdfbookmark[3]{</xsl:text>
+ <xsl:choose>
+ <xsl:when test="contains(emphasis, '--')">
+ <xsl:value-of select="substring-after(emphasis, '--')"/>
+ </xsl:when>
+ <xsl:otherwise>
+ <xsl:value-of select="emphasis"/>
+ </xsl:otherwise>
+ </xsl:choose>
+ <xsl:value-of select="concat('}{', generate-id(emphasis), '}&#10;')"/>
+ <xsl:apply-imports/>
+ </xsl:template>
</xsl:stylesheet>
diff --git a/dviware/dvisvgm/doc/dvisvgm.1 b/dviware/dvisvgm/doc/dvisvgm.1
index b4bb1cb3c8..430a1de890 100644
--- a/dviware/dvisvgm/doc/dvisvgm.1
+++ b/dviware/dvisvgm/doc/dvisvgm.1
@@ -1,13 +1,13 @@
'\" t
-.\" Title: dvisvgm Manual
+.\" Title: dvisvgm
.\" Author: Martin Gieseking <martin.gieseking@uos.de>
.\" Generator: DocBook XSL Stylesheets vsnapshot <http://docbook.sf.net/>
-.\" Date: 2023-07-05
+.\" Date: 2023-09-12
.\" Manual: dvisvgm Manual
-.\" Source: dvisvgm 3.1
+.\" Source: dvisvgm 3.1.2
.\" Language: English
.\"
-.TH "DVISVGM MANUAL" "1" "2023\-07\-05" "dvisvgm 3\&.1" "dvisvgm Manual"
+.TH "DVISVGM" "1" "2023\-09\-12" "dvisvgm 3\&.1\&.2" "dvisvgm Manual"
.\" -----------------------------------------------------------------
.\" * Define some portability stuff
.\" -----------------------------------------------------------------
@@ -309,7 +309,7 @@ which holds the current foreground color set by CSS property
The optional argument of
\fB\-\-currentcolor\fR
specifies the RGB color value to be replaced and must be given as either
-three or six digit hex color value (\m[blue]https://www.w3.org/TR/css-color/#typedef-hex-color\m[])
+six or three digit hex color value (\m[blue]https://www.w3.org/TR/css-color/#typedef-hex-color\m[])
preceded by character
\fB#\fR, e\&.g\&.
\fB#0000ff\fR
@@ -619,7 +619,15 @@ By default, dvisvgm creates CSS styles and class attributes to reference fonts\&
\fB\-O, \-\-optimize\fR[=\fImodules\fR]
.RS 4
Applies several optimizations on the generated SVG tree to reduce the file size\&. The optimizations are performed by running separate optimizer modules specified by optional argument
-\fImodules\fR\&. It may consist of a single module name or a comma\-separated list of several module names\&. The corresponding modules are executed one by one in the given order and thus transform the XML tree gradually\&.
+\fImodules\fR\&. It may consist of a single module name or a comma\-separated list of several module names\&. The corresponding modules are executed one by one in the given order and thus transform the XML tree gradually\&. A leading hyphen (\fB\-\fR) can be used to exclude a module from processing\&. For example
+\fB\-Oall,\-collapse\-groups\fR
+performs all optimizations except
+\fBcollapse\-groups\fR\&. If the first module name in the list starts with a hyphen, dvisvgm automatically inserts
+\fBall\fR
+at the beginning, i\&.e\&.
+\fB\-O\-collapse\-groups\fR
+is identical to
+\fB\-Oall,\-collapse\-groups\fR\&.
.sp
The following list describes the currently available optimizer modules\&.
.PP
@@ -639,9 +647,9 @@ can\(cqt be combined with other module names\&.
.RS 4
Performs all optimizations listed below\&. This is also the default if option
\fB\-\-optimize\fR
-is used without argument\&. The modules are executed in a predefined order that usually leads to the best results\&.
+is used without any arguments\&. The modules are executed in a predefined order that usually leads to the best results\&.
\fBall\fR
-can\(cqt be combined with other module names\&.
+must be the only or the first entry in the list of module names\&.
.RE
.PP
\fBcollapse\-groups\fR
@@ -879,7 +887,9 @@ special command which is forwarded to dvisvgm\(cqs PostScript special handler\&.
for further information\&.
.sp
As of Ghostscript 10\&.01\&.0, this will no longer work due to the introduction of a new PDF interpreter\&. Therefore, an alternative conversion module based on
-\fImutool\fR, a utility which is part of the [MuPDF](\m[blue]\fBhttps://mupdf\&.com\fR\m[]) package, has been introduced\&. It\(cqs automatically invoked if Ghostscript can\(cqt be used and if a working
+\fImutool\fR, a utility which is part of the
+MuPDF (\m[blue]https://mupdf.com\m[])
+package, has been introduced\&. It\(cqs automatically invoked if Ghostscript can\(cqt be used and if a working
\fBmutool\fR
executable is present in a directory which is part of the system\(cqs search path\&.
.sp
@@ -1251,7 +1261,7 @@ attributes are omitted\&.
.sp
dvisvgm supports several \fIspecial commands\fR that enrich the DVI command set with additional instructions for features, like color, graphics, and hyperlinks\&. The term \fIspecial command\fR, or just \fIspecial\fR, originates from the TeX command \fB\especial{\&...}\fR which does almost nothing\&. TeX only forwards the whole command to the DVI file and delegates its execution to the DVI driver\&. The DVI driver can then decide to either process or ignore it, depending on the supported statements\&. The parameter of TeX\(cqs \fB\especial\fR command is given in curly braces and may consist of an arbitrary character sequence representing the actual instruction, for example \fBcolor Red\fR\&.
.sp
-+ dvisvgm delegates the evaluation of special commands to dedicated handlers\&. Each handler is responsible for all special statements of the same command set, i\&.e\&. commands usually \(en but not necessarily \(en beginning with a common prefix\&. For example, all PDF specials start with the prefix \fBpdf:\fR, while the TPIC special set consists of 13 different commands without common identifier\&. The PDF specials are processed by dvisvgm\(cqs \fIpdf\fR handler, the TPIC ones by the \fItpic\fR handler\&. To get a list of the actually provided special handlers, use option \fB\-\-list\-specials\fR (see above)\&. The following list gives an overview of the special sets currently supported\&. The names of the handlers can also be used to disable the processing of individual sets of specials\&. For further information on this, see option \fB\-\-no\-specials\fR\&.
+dvisvgm delegates the evaluation of special commands to dedicated handlers\&. Each handler is responsible for all special statements of the same command set, i\&.e\&. commands usually \(en but not necessarily \(en beginning with a common prefix\&. For example, all PDF specials start with the prefix \fBpdf:\fR, while the TPIC special set consists of 13 different commands without common identifier\&. The PDF specials are processed by dvisvgm\(cqs \fIpdf\fR handler, the TPIC ones by the \fItpic\fR handler\&. To get a list of the actually provided special handlers, use option \fB\-\-list\-specials\fR (see above)\&. The following list gives an overview of the special sets currently supported\&. The names of the handlers can also be used to disable the processing of individual sets of specials\&. For further information on this, see option \fB\-\-no\-specials\fR\&.
.PP
\fBbgcolor\fR
.RS 4
@@ -1423,16 +1433,6 @@ allows for applying the current transformation matrix to the bounding box\&.
This variant of the bbox special assigns an absolute (final) bounding box to the resulting SVG\&. After executing this command, dvisvgm doesn\(cqt further alter the bounding box coordinates, except this special is called again later\&. The points (\fIx1\fR,\fIy1\fR) and (\fIx2\fR,\fIy2\fR) denote the absolute coordinates of two diagonal corners of the rectangle\&. As with the relative special variant described above, the optional modifier
\fBtransform\fR
allows for applying the current transformation matrix to the bounding box\&.
-.RE
-.PP
-\fBdvisvgm:message\fR \fImsg\fR
-.RS 4
-Prints the given text
-\fImsg\fR
-to the console\&. It may also contain the macros
-\fI{?\&...}\fR
-mentioned above (see
-\fBdvisvgm:raw\fR)\&.
.sp
The following TeX snippet adds two raw SVG elements to the output and updates the bounding box accordingly:
.sp
@@ -1464,6 +1464,16 @@ but doesn\(cqt require an explicit color argument\&. Instead, it takes the curre
functionality \(en can be omitted\&. When called with argument
\fBoff\fR, the functionality is deactivated again\&.
.RE
+.PP
+\fBdvisvgm:message\fR \fImsg\fR
+.RS 4
+Prints the given text
+\fImsg\fR
+to the console\&. It may also contain the macros
+\fI{?\&...}\fR
+mentioned above (see
+\fBdvisvgm:raw\fR)\&.
+.RE
.RE
.PP
\fBem\fR
@@ -1502,7 +1512,7 @@ pdfTeX and dvipdfmx introduced several special commands related to the generatio
\fIpdf:pagesize\fR
is similar to the
\fIpapersize\fR
-special (see above) which specifies the size of the current and all folowing pages\&. In order to actually apply the extents to the generated SVG files, option
+special (see above) which specifies the size of the current and all following pages\&. In order to actually apply the extents to the generated SVG files, option
\fB\-\-bbox=papersize\fR
must be given\&.
.sp
diff --git a/dviware/dvisvgm/doc/dvisvgm.txt.in b/dviware/dvisvgm/doc/dvisvgm.txt.in
index 5b31c8d092..8060b53aa9 100644
--- a/dviware/dvisvgm/doc/dvisvgm.txt.in
+++ b/dviware/dvisvgm/doc/dvisvgm.txt.in
@@ -22,7 +22,7 @@ Martin Gieseking <@PACKAGE_BUGREPORT@>
:man source: dvisvgm
:man version: @VERSION@
:man manual: dvisvgm Manual
-:revdate: 2023-07-05 10:37 +0200
+:revdate: 2023-09-12 16:36 +0200
Name
----
@@ -212,7 +212,7 @@ holds the current foreground color set by CSS property `color`. This way an appl
can change the color via CSS without the need to modify the SVG code.
+
The optional argument of `--currentcolor` specifies the RGB color value to be replaced and must
-be given as either https://www.w3.org/TR/css-color/#typedef-hex-color[three or six digit hex color value]
+be given as either https://www.w3.org/TR/css-color/#typedef-hex-color[six or three digit hex color value]
preceded by character `#`, e.g. `#0000ff` or `#00f` for blue. If the optional argument is omitted, black
is assumed.
@@ -430,6 +430,10 @@ are performed by running separate optimizer modules specified by optional argume
It may consist of a single module name or a comma-separated list of several module names.
The corresponding modules are executed one by one in the given order and thus transform the
XML tree gradually.
+A leading hyphen (+-+) can be used to exclude a module from processing. For example
++-Oall,-collapse-groups+ performs all optimizations except +collapse-groups+. If the first
+module name in the list starts with a hyphen, dvisvgm automatically inserts +all+ at the
+beginning, i.e. +-O-collapse-groups+ is identical to +-Oall,-collapse-groups+.
+
The following list describes the currently available optimizer modules.
@@ -442,8 +446,8 @@ The following list describes the currently available optimizer modules.
*all*;;
Performs all optimizations listed below. This is also the default if option *--optimize* is
- used without argument. The modules are executed in a predefined order that usually leads to the
- best results. *all* can't be combined with other module names.
+ used without any arguments. The modules are executed in a predefined order that usually leads
+ to the best results. *all* must be the only or the first entry in the list of module names.
*collapse-groups*;;
Combines nested group elements (+<g>+...+</g>+) that contain only a single group each. If possible,
@@ -590,7 +594,7 @@ See option *--libgs* for further information.
+
As of Ghostscript 10.01.0, this will no longer work due to the introduction of a new PDF interpreter.
Therefore, an alternative conversion module based on _mutool_, a utility which is part of the
-[MuPDF](https://mupdf.com) package, has been introduced. It's automatically invoked if Ghostscript
+https://mupdf.com[MuPDF] package, has been introduced. It's automatically invoked if Ghostscript
can't be used and if a working +mutool+ executable is present in a directory which is part of the
system's search path.
+
@@ -773,7 +777,7 @@ DVI file and delegates its execution to the DVI driver. The DVI driver can then
ignore it, depending on the supported statements. The parameter of TeX's +\special+ command is given in curly
braces and may consist of an arbitrary character sequence representing the actual instruction, for example
+color Red+.
-+
+
dvisvgm delegates the evaluation of special commands to dedicated handlers. Each handler is responsible for
all special statements of the same command set, i.e. commands usually &#x2013; but not necessarily &#x2013;
beginning with a common prefix. For example, all PDF specials start with the prefix +pdf:+, while the TPIC
@@ -880,10 +884,6 @@ dvisvgm offers its own small set of specials. The following list gives a brief o
The points ('x1','y1') and ('x2','y2') denote the absolute coordinates of two diagonal corners of the rectangle.
As with the relative special variant described above, the optional modifier +transform+ allows for applying the
current transformation matrix to the bounding box.
-
- *dvisvgm:message* 'msg';;
- Prints the given text 'msg' to the console. It may also contain the macros '{?...}' mentioned above
- (see *dvisvgm:raw*).
+
The following TeX snippet adds two raw SVG elements to the output and updates the bounding box accordingly:
+
@@ -903,6 +903,10 @@ The following TeX snippet adds two raw SVG elements to the output and updates th
placement of the special and the preceding color changes. Argument `on` &#x2013; which activates the 'currentColor'
functionality &#x2013; can be omitted. When called with argument `off`, the functionality is deactivated again.
+ *dvisvgm:message* 'msg';;
+ Prints the given text 'msg' to the console. It may also contain the macros '{?...}' mentioned above
+ (see *dvisvgm:raw*).
+
*em*::
These specials were introduced with the 'emTeX' distribution by Eberhard Mattes. They provide line drawing
statements, instructions for embedding MSP, PCX, and BMP image files, as well as two PCL commands.
@@ -931,7 +935,7 @@ only 'pdf:mapfile', 'pdf:mapline', 'pdf:pagesize', and PDF hyperlink specials ar
The latter are the PDF pendants to the HTML HyperTeX specials generated by the hyperref package in PDF mode.
+
'pdf:pagesize' is similar to the 'papersize' special (see above) which specifies the size of the current
-and all folowing pages. In order to actually apply the extents to the generated SVG files,
+and all following pages. In order to actually apply the extents to the generated SVG files,
option *--bbox=papersize* must be given.
+
'pdf:mapfile' and 'pdf:mapline' allow for modifying the font map tree while processing the DVI file.
diff --git a/dviware/dvisvgm/doc/tweak-db-refentry.xsl b/dviware/dvisvgm/doc/tweak-db-refentry.xsl
index f47e18d93d..24e4343f14 100644
--- a/dviware/dvisvgm/doc/tweak-db-refentry.xsl
+++ b/dviware/dvisvgm/doc/tweak-db-refentry.xsl
@@ -2,51 +2,51 @@
<!-- This file is part of dvisvgm -->
<!-- Copyright (C) 2015-2023 Martin Gieseking <martin.gieseking@uos.de> -->
<xsl:stylesheet version="1.0"
- xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
- xmlns:date="http://exslt.org/dates-and-times"
- xmlns:func="http://exslt.org/functions"
- xmlns:my="my-namespace"
- exclude-result-prefixes="my"
- extension-element-prefixes="date func">
+ xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
+ xmlns:date="http://exslt.org/dates-and-times"
+ xmlns:func="http://exslt.org/functions"
+ xmlns:my="my-namespace"
+ exclude-result-prefixes="my"
+ extension-element-prefixes="date func">
- <xsl:output doctype-public="-//OASIS//DTD DocBook XML V4.5//EN" doctype-system="http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd"/>
+ <xsl:output doctype-public="-//OASIS//DTD DocBook XML V4.5//EN" doctype-system="http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd"/>
- <!-- adapt ISO date created by keyword expansion -->
- <xsl:template match="refentryinfo/date | revision[1]/date">
- <date>
- <xsl:value-of select="my:extract-date(.)"/>
- </date>
- </xsl:template>
+ <!-- adapt ISO date created by keyword expansion -->
+ <xsl:template match="refentryinfo/date | revision[1]/date">
+ <date>
+ <xsl:value-of select="my:extract-date(.)"/>
+ </date>
+ </xsl:template>
- <!-- add date to refmeta element -->
- <xsl:template match="refmeta">
- <xsl:copy>
- <xsl:copy-of select="*"/>
- <refmiscinfo class="date">
- <xsl:value-of select="my:extract-date(../refentryinfo/date)"/>
- </refmiscinfo>
- </xsl:copy>
- </xsl:template>
+ <!-- add date to refmeta element -->
+ <xsl:template match="refmeta">
+ <xsl:copy>
+ <xsl:copy-of select="*"/>
+ <refmiscinfo class="date">
+ <xsl:value-of select="my:extract-date(../refentryinfo/date)"/>
+ </refmiscinfo>
+ </xsl:copy>
+ </xsl:template>
- <func:function name="my:extract-date">
- <xsl:param name="isodate"/>
- <xsl:variable name="shortdate" select="substring($isodate, 1, 10)"/>
- <xsl:choose>
- <xsl:when test="translate($shortdate, '0123456789', '##########') = '####-##-##'">
- <!-- use date part of expanded keyword -->
- <func:result select="$shortdate"/>
- </xsl:when>
- <xsl:otherwise>
- <!-- use current date as fallback -->
- <func:result select="substring(date:date(), 1, 10)"/>
- </xsl:otherwise>
- </xsl:choose>
- </func:function>
+ <func:function name="my:extract-date">
+ <xsl:param name="isodate"/>
+ <xsl:variable name="shortdate" select="substring($isodate, 1, 10)"/>
+ <xsl:choose>
+ <xsl:when test="translate($shortdate, '0123456789', '##########') = '####-##-##'">
+ <!-- use date part of expanded keyword -->
+ <func:result select="$shortdate"/>
+ </xsl:when>
+ <xsl:otherwise>
+ <!-- use current date as fallback -->
+ <func:result select="substring(date:date(), 1, 10)"/>
+ </xsl:otherwise>
+ </xsl:choose>
+ </func:function>
- <!-- copy everything else unchanged -->
- <xsl:template match="@*|node()">
- <xsl:copy>
- <xsl:apply-templates select="@*|node()"/>
- </xsl:copy>
- </xsl:template>
+ <!-- copy everything else unchanged -->
+ <xsl:template match="@*|node()">
+ <xsl:copy>
+ <xsl:apply-templates select="@*|node()"/>
+ </xsl:copy>
+ </xsl:template>
</xsl:stylesheet>
diff --git a/dviware/dvisvgm/doc/tweak-dblatex-pdf.xsl b/dviware/dvisvgm/doc/tweak-dblatex-pdf.xsl
index 15523e3a20..406d771ca2 100644
--- a/dviware/dvisvgm/doc/tweak-dblatex-pdf.xsl
+++ b/dviware/dvisvgm/doc/tweak-dblatex-pdf.xsl
@@ -2,28 +2,28 @@
<!-- This file is part of dvisvgm -->
<!-- Copyright (C) 2015-2023 Martin Gieseking <martin.gieseking@uos.de> -->
<xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform" version="1.0">
- <xsl:output doctype-public="-//OASIS//DTD DocBook XML V4.5//EN" doctype-system="http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd"/>
+ <xsl:output doctype-public="-//OASIS//DTD DocBook XML V4.5//EN" doctype-system="http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd"/>
- <!-- replace default document title -->
- <xsl:template match="refentrytitle[. != ../refmiscinfo[@class='manual']]">
- <xsl:copy>
- <xsl:value-of select="../refmiscinfo[@class='manual']"/>
- </xsl:copy>
- </xsl:template>
+ <!-- replace default document title -->
+ <xsl:template match="refentrytitle[. != ../refmiscinfo[@class='manual']]">
+ <xsl:copy>
+ <xsl:value-of select="../refmiscinfo[@class='manual']"/>
+ </xsl:copy>
+ </xsl:template>
- <!-- add static title to the synopsis section to prevent creating a dynamic one -->
- <xsl:template match="refsynopsisdiv[not(title)]">
- <xsl:copy>
- <xsl:copy-of select="@*"/>
- <title>Synopsis</title>
- <xsl:apply-templates/>
- </xsl:copy>
- </xsl:template>
+ <!-- add static title to the synopsis section to prevent creating a dynamic one -->
+ <xsl:template match="refsynopsisdiv[not(title)]">
+ <xsl:copy>
+ <xsl:copy-of select="@*"/>
+ <title>Synopsis</title>
+ <xsl:apply-templates/>
+ </xsl:copy>
+ </xsl:template>
- <!-- copy everything else unchanged -->
- <xsl:template match="@*|node()">
- <xsl:copy>
- <xsl:apply-templates select="@*|node()"/>
- </xsl:copy>
- </xsl:template>
+ <!-- copy everything else unchanged -->
+ <xsl:template match="@*|node()">
+ <xsl:copy>
+ <xsl:apply-templates select="@*|node()"/>
+ </xsl:copy>
+ </xsl:template>
</xsl:stylesheet>
diff --git a/dviware/dvisvgm/libs/brotli/Makefile.am b/dviware/dvisvgm/libs/brotli/Makefile.am
index 64ad12e6aa..7d7bf1db32 100644
--- a/dviware/dvisvgm/libs/brotli/Makefile.am
+++ b/dviware/dvisvgm/libs/brotli/Makefile.am
@@ -9,6 +9,7 @@ libbrotli_a_SOURCES = \
common/dictionary.h \
common/platform.c \
common/platform.h \
+ common/shared_dictionary_internal.h \
common/transform.c \
common/transform.h \
common/version.h \
@@ -31,6 +32,8 @@ libbrotli_a_SOURCES = \
enc/cluster_inc.h \
enc/command.c \
enc/command.h \
+ enc/compound_dictionary.c \
+ enc/compound_dictionary.h \
enc/compress_fragment.c \
enc/compress_fragment.h \
enc/compress_fragment_two_pass.c \
@@ -68,6 +71,7 @@ libbrotli_a_SOURCES = \
enc/prefix.h \
enc/quality.h \
enc/ringbuffer.h \
+ enc/state.h \
enc/static_dict.c \
enc/static_dict.h \
enc/static_dict_lut.h \
@@ -77,6 +81,7 @@ libbrotli_a_SOURCES = \
include/brotli/decode.h \
include/brotli/encode.h \
include/brotli/port.h \
+ include/brotli/shared_dictionary.h \
include/brotli/types.h
AM_CFLAGS = -Wall -I$(srcdir)/include
diff --git a/dviware/dvisvgm/libs/brotli/Makefile.in b/dviware/dvisvgm/libs/brotli/Makefile.in
index 6518869128..075a208efb 100644
--- a/dviware/dvisvgm/libs/brotli/Makefile.in
+++ b/dviware/dvisvgm/libs/brotli/Makefile.in
@@ -120,15 +120,16 @@ libbrotli_a_LIBADD =
am__libbrotli_a_SOURCES_DIST = common/constants.c common/constants.h \
common/context.c common/context.h common/dictionary.c \
common/dictionary.h common/platform.c common/platform.h \
- common/transform.c common/transform.h common/version.h \
- enc/backward_references.c enc/backward_references.h \
- enc/backward_references_hq.c enc/backward_references_hq.h \
- enc/backward_references_inc.h enc/bit_cost.c enc/bit_cost.h \
- enc/bit_cost_inc.h enc/block_encoder_inc.h \
- enc/block_splitter.c enc/block_splitter.h \
- enc/block_splitter_inc.h enc/brotli_bit_stream.c \
- enc/brotli_bit_stream.h enc/cluster.c enc/cluster.h \
- enc/cluster_inc.h enc/command.c enc/command.h \
+ common/shared_dictionary_internal.h common/transform.c \
+ common/transform.h common/version.h enc/backward_references.c \
+ enc/backward_references.h enc/backward_references_hq.c \
+ enc/backward_references_hq.h enc/backward_references_inc.h \
+ enc/bit_cost.c enc/bit_cost.h enc/bit_cost_inc.h \
+ enc/block_encoder_inc.h enc/block_splitter.c \
+ enc/block_splitter.h enc/block_splitter_inc.h \
+ enc/brotli_bit_stream.c enc/brotli_bit_stream.h enc/cluster.c \
+ enc/cluster.h enc/cluster_inc.h enc/command.c enc/command.h \
+ enc/compound_dictionary.c enc/compound_dictionary.h \
enc/compress_fragment.c enc/compress_fragment.h \
enc/compress_fragment_two_pass.c \
enc/compress_fragment_two_pass.h enc/dictionary_hash.c \
@@ -143,11 +144,11 @@ am__libbrotli_a_SOURCES_DIST = common/constants.c common/constants.h \
enc/histogram_inc.h enc/literal_cost.c enc/literal_cost.h \
enc/memory.c enc/memory.h enc/metablock.c enc/metablock.h \
enc/metablock_inc.h enc/params.h enc/prefix.h enc/quality.h \
- enc/ringbuffer.h enc/static_dict.c enc/static_dict.h \
- enc/static_dict_lut.h enc/utf8_util.c enc/utf8_util.h \
- enc/write_bits.h include/brotli/decode.h \
+ enc/ringbuffer.h enc/state.h enc/static_dict.c \
+ enc/static_dict.h enc/static_dict_lut.h enc/utf8_util.c \
+ enc/utf8_util.h enc/write_bits.h include/brotli/decode.h \
include/brotli/encode.h include/brotli/port.h \
- include/brotli/types.h
+ include/brotli/shared_dictionary.h include/brotli/types.h
am__dirstamp = $(am__leading_dot)dirstamp
@HAVE_BROTLI_FALSE@am_libbrotli_a_OBJECTS = \
@HAVE_BROTLI_FALSE@ common/constants.$(OBJEXT) \
@@ -161,6 +162,7 @@ am__dirstamp = $(am__leading_dot)dirstamp
@HAVE_BROTLI_FALSE@ enc/block_splitter.$(OBJEXT) \
@HAVE_BROTLI_FALSE@ enc/brotli_bit_stream.$(OBJEXT) \
@HAVE_BROTLI_FALSE@ enc/cluster.$(OBJEXT) enc/command.$(OBJEXT) \
+@HAVE_BROTLI_FALSE@ enc/compound_dictionary.$(OBJEXT) \
@HAVE_BROTLI_FALSE@ enc/compress_fragment.$(OBJEXT) \
@HAVE_BROTLI_FALSE@ enc/compress_fragment_two_pass.$(OBJEXT) \
@HAVE_BROTLI_FALSE@ enc/dictionary_hash.$(OBJEXT) \
@@ -197,7 +199,8 @@ am__depfiles_remade = common/$(DEPDIR)/constants.Po \
enc/$(DEPDIR)/backward_references_hq.Po \
enc/$(DEPDIR)/bit_cost.Po enc/$(DEPDIR)/block_splitter.Po \
enc/$(DEPDIR)/brotli_bit_stream.Po enc/$(DEPDIR)/cluster.Po \
- enc/$(DEPDIR)/command.Po enc/$(DEPDIR)/compress_fragment.Po \
+ enc/$(DEPDIR)/command.Po enc/$(DEPDIR)/compound_dictionary.Po \
+ enc/$(DEPDIR)/compress_fragment.Po \
enc/$(DEPDIR)/compress_fragment_two_pass.Po \
enc/$(DEPDIR)/dictionary_hash.Po enc/$(DEPDIR)/encode.Po \
enc/$(DEPDIR)/encoder_dict.Po enc/$(DEPDIR)/entropy_encode.Po \
@@ -422,6 +425,7 @@ top_srcdir = @top_srcdir@
@HAVE_BROTLI_FALSE@ common/dictionary.h \
@HAVE_BROTLI_FALSE@ common/platform.c \
@HAVE_BROTLI_FALSE@ common/platform.h \
+@HAVE_BROTLI_FALSE@ common/shared_dictionary_internal.h \
@HAVE_BROTLI_FALSE@ common/transform.c \
@HAVE_BROTLI_FALSE@ common/transform.h \
@HAVE_BROTLI_FALSE@ common/version.h \
@@ -444,6 +448,8 @@ top_srcdir = @top_srcdir@
@HAVE_BROTLI_FALSE@ enc/cluster_inc.h \
@HAVE_BROTLI_FALSE@ enc/command.c \
@HAVE_BROTLI_FALSE@ enc/command.h \
+@HAVE_BROTLI_FALSE@ enc/compound_dictionary.c \
+@HAVE_BROTLI_FALSE@ enc/compound_dictionary.h \
@HAVE_BROTLI_FALSE@ enc/compress_fragment.c \
@HAVE_BROTLI_FALSE@ enc/compress_fragment.h \
@HAVE_BROTLI_FALSE@ enc/compress_fragment_two_pass.c \
@@ -481,6 +487,7 @@ top_srcdir = @top_srcdir@
@HAVE_BROTLI_FALSE@ enc/prefix.h \
@HAVE_BROTLI_FALSE@ enc/quality.h \
@HAVE_BROTLI_FALSE@ enc/ringbuffer.h \
+@HAVE_BROTLI_FALSE@ enc/state.h \
@HAVE_BROTLI_FALSE@ enc/static_dict.c \
@HAVE_BROTLI_FALSE@ enc/static_dict.h \
@HAVE_BROTLI_FALSE@ enc/static_dict_lut.h \
@@ -490,6 +497,7 @@ top_srcdir = @top_srcdir@
@HAVE_BROTLI_FALSE@ include/brotli/decode.h \
@HAVE_BROTLI_FALSE@ include/brotli/encode.h \
@HAVE_BROTLI_FALSE@ include/brotli/port.h \
+@HAVE_BROTLI_FALSE@ include/brotli/shared_dictionary.h \
@HAVE_BROTLI_FALSE@ include/brotli/types.h
@HAVE_BROTLI_FALSE@AM_CFLAGS = -Wall -I$(srcdir)/include
@@ -567,6 +575,8 @@ enc/cluster.$(OBJEXT): enc/$(am__dirstamp) \
enc/$(DEPDIR)/$(am__dirstamp)
enc/command.$(OBJEXT): enc/$(am__dirstamp) \
enc/$(DEPDIR)/$(am__dirstamp)
+enc/compound_dictionary.$(OBJEXT): enc/$(am__dirstamp) \
+ enc/$(DEPDIR)/$(am__dirstamp)
enc/compress_fragment.$(OBJEXT): enc/$(am__dirstamp) \
enc/$(DEPDIR)/$(am__dirstamp)
enc/compress_fragment_two_pass.$(OBJEXT): enc/$(am__dirstamp) \
@@ -619,6 +629,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/brotli_bit_stream.Po@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/cluster.Po@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/command.Po@am__quote@ # am--include-marker
+@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/compound_dictionary.Po@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/compress_fragment.Po@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/compress_fragment_two_pass.Po@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@enc/$(DEPDIR)/dictionary_hash.Po@am__quote@ # am--include-marker
@@ -810,6 +821,7 @@ distclean: distclean-am
-rm -f enc/$(DEPDIR)/brotli_bit_stream.Po
-rm -f enc/$(DEPDIR)/cluster.Po
-rm -f enc/$(DEPDIR)/command.Po
+ -rm -f enc/$(DEPDIR)/compound_dictionary.Po
-rm -f enc/$(DEPDIR)/compress_fragment.Po
-rm -f enc/$(DEPDIR)/compress_fragment_two_pass.Po
-rm -f enc/$(DEPDIR)/dictionary_hash.Po
@@ -880,6 +892,7 @@ maintainer-clean: maintainer-clean-am
-rm -f enc/$(DEPDIR)/brotli_bit_stream.Po
-rm -f enc/$(DEPDIR)/cluster.Po
-rm -f enc/$(DEPDIR)/command.Po
+ -rm -f enc/$(DEPDIR)/compound_dictionary.Po
-rm -f enc/$(DEPDIR)/compress_fragment.Po
-rm -f enc/$(DEPDIR)/compress_fragment_two_pass.Po
-rm -f enc/$(DEPDIR)/dictionary_hash.Po
diff --git a/dviware/dvisvgm/libs/brotli/common/constants.c b/dviware/dvisvgm/libs/brotli/common/constants.c
index 6bad9f613c..89866b1505 100644
--- a/dviware/dvisvgm/libs/brotli/common/constants.c
+++ b/dviware/dvisvgm/libs/brotli/common/constants.c
@@ -4,7 +4,7 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./constants.h"
+#include "constants.h"
const BrotliPrefixCodeRange
_kBrotliPrefixCodeRanges[BROTLI_NUM_BLOCK_LEN_SYMBOLS] = {
diff --git a/dviware/dvisvgm/libs/brotli/common/constants.h b/dviware/dvisvgm/libs/brotli/common/constants.h
index e848195a0d..31e5bd376e 100644
--- a/dviware/dvisvgm/libs/brotli/common/constants.h
+++ b/dviware/dvisvgm/libs/brotli/common/constants.h
@@ -12,10 +12,11 @@
#ifndef BROTLI_COMMON_CONSTANTS_H_
#define BROTLI_COMMON_CONSTANTS_H_
-#include "./platform.h"
#include <brotli/port.h>
#include <brotli/types.h>
+#include "platform.h"
+
/* Specification: 7.3. Encoding of the context map */
#define BROTLI_CONTEXT_MAP_MAX_RLE 16
diff --git a/dviware/dvisvgm/libs/brotli/common/context.c b/dviware/dvisvgm/libs/brotli/common/context.c
index 2c2dceba9b..7f9c958699 100644
--- a/dviware/dvisvgm/libs/brotli/common/context.c
+++ b/dviware/dvisvgm/libs/brotli/common/context.c
@@ -1,4 +1,4 @@
-#include "./context.h"
+#include "context.h"
#include <brotli/types.h>
diff --git a/dviware/dvisvgm/libs/brotli/common/dictionary.c b/dviware/dvisvgm/libs/brotli/common/dictionary.c
index f9e30417c0..7c015ab0ba 100644
--- a/dviware/dvisvgm/libs/brotli/common/dictionary.c
+++ b/dviware/dvisvgm/libs/brotli/common/dictionary.c
@@ -4,8 +4,8 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./dictionary.h"
-#include "./platform.h"
+#include "dictionary.h"
+#include "platform.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -13,6 +13,7 @@ extern "C" {
#if !defined(BROTLI_EXTERNAL_DICTIONARY_DATA)
static const uint8_t kBrotliDictionaryData[] =
+/* GENERATED CODE START */
{
116,105,109,101,100,111,119,110,108,105,102,101,108,101,102,116,98,97,99,107,99,
111,100,101,100,97,116,97,115,104,111,119,111,110,108,121,115,105,116,101,99,105
@@ -5860,6 +5861,7 @@ static const uint8_t kBrotliDictionaryData[] =
,164,181,224,164,190,224,164,136,224,164,184,224,164,149,224,165,141,224,164,176
,224,164,191,224,164,175,224,164,164,224,164,190
}
+/* GENERATED CODE END */
;
#endif /* !BROTLI_EXTERNAL_DICTIONARY_DATA */
@@ -5895,7 +5897,7 @@ static BrotliDictionary kBrotliDictionary = {
#endif
};
-const BrotliDictionary* BrotliGetDictionary() {
+const BrotliDictionary* BrotliGetDictionary(void) {
return &kBrotliDictionary;
}
diff --git a/dviware/dvisvgm/libs/brotli/common/platform.c b/dviware/dvisvgm/libs/brotli/common/platform.c
index aef39e93c4..25d84a9467 100644
--- a/dviware/dvisvgm/libs/brotli/common/platform.c
+++ b/dviware/dvisvgm/libs/brotli/common/platform.c
@@ -6,9 +6,10 @@
#include <stdlib.h>
-#include "./platform.h"
#include <brotli/types.h>
+#include "platform.h"
+
/* Default brotli_alloc_func */
void* BrotliDefaultAllocFunc(void* opaque, size_t size) {
BROTLI_UNUSED(opaque);
diff --git a/dviware/dvisvgm/libs/brotli/common/platform.h b/dviware/dvisvgm/libs/brotli/common/platform.h
index f5ca4435bc..7406f3fe69 100644
--- a/dviware/dvisvgm/libs/brotli/common/platform.h
+++ b/dviware/dvisvgm/libs/brotli/common/platform.h
@@ -12,12 +12,13 @@
* BROTLI_BUILD_BIG_ENDIAN forces to use big-endian optimizations
* BROTLI_BUILD_ENDIAN_NEUTRAL disables endian-aware optimizations
* BROTLI_BUILD_LITTLE_ENDIAN forces to use little-endian optimizations
- * BROTLI_BUILD_PORTABLE disables dangerous optimizations, like unaligned
- read and overlapping memcpy; this reduces decompression speed by 5%
* BROTLI_BUILD_NO_RBIT disables "rbit" optimization for ARM CPUs
+ * BROTLI_BUILD_NO_UNALIGNED_READ_FAST forces off the fast-unaligned-read
+ optimizations (mainly for testing purposes)
* BROTLI_DEBUG dumps file name and line number when decoder detects stream
or memory error
* BROTLI_ENABLE_LOG enables asserts and dumps various state information
+ * BROTLI_ENABLE_DUMP overrides default "dump" behaviour
*/
#ifndef BROTLI_COMMON_PLATFORM_H_
@@ -40,7 +41,7 @@
#define BROTLI_X_BIG_ENDIAN BIG_ENDIAN
#endif
-#if BROTLI_MSVC_VERSION_CHECK(12, 0, 0)
+#if BROTLI_MSVC_VERSION_CHECK(18, 0, 0)
#include <intrin.h>
#endif
@@ -156,24 +157,6 @@ OR:
#define BROTLI_NOINLINE
#endif
-/* BROTLI_INTERNAL could be defined to override visibility, e.g. for tests. */
-#if !defined(BROTLI_INTERNAL)
-#if defined(_WIN32) || defined(__CYGWIN__)
-#define BROTLI_INTERNAL
-#elif BROTLI_GNUC_VERSION_CHECK(3, 3, 0) || \
- BROTLI_TI_VERSION_CHECK(8, 0, 0) || \
- BROTLI_INTEL_VERSION_CHECK(16, 0, 0) || \
- BROTLI_ARM_VERSION_CHECK(4, 1, 0) || \
- BROTLI_IBM_VERSION_CHECK(13, 1, 0) || \
- BROTLI_SUNPRO_VERSION_CHECK(5, 11, 0) || \
- (BROTLI_TI_VERSION_CHECK(7, 3, 0) && \
- defined(__TI_GNU_ATTRIBUTE_SUPPORT__) && defined(__TI_EABI__))
-#define BROTLI_INTERNAL __attribute__ ((visibility ("hidden")))
-#else
-#define BROTLI_INTERNAL
-#endif
-#endif
-
/* <<< <<< <<< end of hedley macros. */
#if BROTLI_GNUC_HAS_ATTRIBUTE(unused, 2, 7, 0) || \
@@ -226,15 +209,24 @@ OR:
#define BROTLI_TARGET_RISCV64
#endif
+#if defined(__loongarch_lp64)
+#define BROTLI_TARGET_LOONGARCH64
+#endif
+
+#if defined(BROTLI_TARGET_X64) || defined(BROTLI_TARGET_ARMV8_64) || \
+ defined(BROTLI_TARGET_POWERPC64) || defined(BROTLI_TARGET_RISCV64) || \
+ defined(BROTLI_TARGET_LOONGARCH64)
+#define BROTLI_TARGET_64_BITS 1
+#else
+#define BROTLI_TARGET_64_BITS 0
+#endif
+
#if defined(BROTLI_BUILD_64_BIT)
#define BROTLI_64_BITS 1
#elif defined(BROTLI_BUILD_32_BIT)
#define BROTLI_64_BITS 0
-#elif defined(BROTLI_TARGET_X64) || defined(BROTLI_TARGET_ARMV8_64) || \
- defined(BROTLI_TARGET_POWERPC64) || defined(BROTLI_TARGET_RISCV64)
-#define BROTLI_64_BITS 1
#else
-#define BROTLI_64_BITS 0
+#define BROTLI_64_BITS BROTLI_TARGET_64_BITS
#endif
#if (BROTLI_64_BITS)
@@ -278,18 +270,19 @@ OR:
#undef BROTLI_X_BIG_ENDIAN
#endif
-#if defined(BROTLI_BUILD_PORTABLE)
-#define BROTLI_ALIGNED_READ (!!1)
-#elif defined(BROTLI_TARGET_X86) || defined(BROTLI_TARGET_X64) || \
+#if defined(BROTLI_BUILD_NO_UNALIGNED_READ_FAST)
+#define BROTLI_UNALIGNED_READ_FAST (!!0)
+#elif defined(BROTLI_TARGET_X86) || defined(BROTLI_TARGET_X64) || \
defined(BROTLI_TARGET_ARMV7) || defined(BROTLI_TARGET_ARMV8_ANY) || \
- defined(BROTLI_TARGET_RISCV64)
-/* Allow unaligned read only for white-listed CPUs. */
-#define BROTLI_ALIGNED_READ (!!0)
+ defined(BROTLI_TARGET_RISCV64) || defined(BROTLI_TARGET_LOONGARCH64)
+/* These targets are known to generate efficient code for unaligned reads
+ * (e.g. a single instruction, not multiple 1-byte loads, shifted and or'd
+ * together). */
+#define BROTLI_UNALIGNED_READ_FAST (!!1)
#else
-#define BROTLI_ALIGNED_READ (!!1)
+#define BROTLI_UNALIGNED_READ_FAST (!!0)
#endif
-#if BROTLI_ALIGNED_READ
/* Portable unaligned memory access: read / write values via memcpy. */
static BROTLI_INLINE uint16_t BrotliUnalignedRead16(const void* p) {
uint16_t t;
@@ -309,75 +302,6 @@ static BROTLI_INLINE uint64_t BrotliUnalignedRead64(const void* p) {
static BROTLI_INLINE void BrotliUnalignedWrite64(void* p, uint64_t v) {
memcpy(p, &v, sizeof v);
}
-#else /* BROTLI_ALIGNED_READ */
-/* Unaligned memory access is allowed: just cast pointer to requested type. */
-#if BROTLI_SANITIZED
-/* Consider we have an unaligned load/store of 4 bytes from address 0x...05.
- AddressSanitizer will treat it as a 3-byte access to the range 05:07 and
- will miss a bug if 08 is the first unaddressable byte.
- ThreadSanitizer will also treat this as a 3-byte access to 05:07 and will
- miss a race between this access and some other accesses to 08.
- MemorySanitizer will correctly propagate the shadow on unaligned stores
- and correctly report bugs on unaligned loads, but it may not properly
- update and report the origin of the uninitialized memory.
- For all three tools, replacing an unaligned access with a tool-specific
- callback solves the problem. */
-#if defined(__cplusplus)
-extern "C" {
-#endif /* __cplusplus */
- uint16_t __sanitizer_unaligned_load16(const void* p);
- uint32_t __sanitizer_unaligned_load32(const void* p);
- uint64_t __sanitizer_unaligned_load64(const void* p);
- void __sanitizer_unaligned_store64(void* p, uint64_t v);
-#if defined(__cplusplus)
-} /* extern "C" */
-#endif /* __cplusplus */
-#define BrotliUnalignedRead16 __sanitizer_unaligned_load16
-#define BrotliUnalignedRead32 __sanitizer_unaligned_load32
-#define BrotliUnalignedRead64 __sanitizer_unaligned_load64
-#define BrotliUnalignedWrite64 __sanitizer_unaligned_store64
-#else /* BROTLI_SANITIZED */
-static BROTLI_INLINE uint16_t BrotliUnalignedRead16(const void* p) {
- return *(const uint16_t*)p;
-}
-static BROTLI_INLINE uint32_t BrotliUnalignedRead32(const void* p) {
- return *(const uint32_t*)p;
-}
-#if (BROTLI_64_BITS)
-static BROTLI_INLINE uint64_t BrotliUnalignedRead64(const void* p) {
- return *(const uint64_t*)p;
-}
-static BROTLI_INLINE void BrotliUnalignedWrite64(void* p, uint64_t v) {
- *(uint64_t*)p = v;
-}
-#else /* BROTLI_64_BITS */
-/* Avoid emitting LDRD / STRD, which require properly aligned address. */
-/* If __attribute__(aligned) is available, use that. Otherwise, memcpy. */
-
-#if BROTLI_GNUC_HAS_ATTRIBUTE(aligned, 2, 7, 0)
-typedef BROTLI_ALIGNED(1) uint64_t brotli_unaligned_uint64_t;
-
-static BROTLI_INLINE uint64_t BrotliUnalignedRead64(const void* p) {
- return (uint64_t) ((const brotli_unaligned_uint64_t*) p)[0];
-}
-static BROTLI_INLINE void BrotliUnalignedWrite64(void* p, uint64_t v) {
- brotli_unaligned_uint64_t* dwords = (brotli_unaligned_uint64_t*) p;
- dwords[0] = (brotli_unaligned_uint64_t) v;
-}
-#else /* BROTLI_GNUC_HAS_ATTRIBUTE(aligned, 2, 7, 0) */
-static BROTLI_INLINE uint64_t BrotliUnalignedRead64(const void* p) {
- uint64_t v;
- memcpy(&v, p, sizeof(uint64_t));
- return v;
-}
-
-static BROTLI_INLINE void BrotliUnalignedWrite64(void* p, uint64_t v) {
- memcpy(p, &v, sizeof(uint64_t));
-}
-#endif /* BROTLI_GNUC_HAS_ATTRIBUTE(aligned, 2, 7, 0) */
-#endif /* BROTLI_64_BITS */
-#endif /* BROTLI_SANITIZED */
-#endif /* BROTLI_ALIGNED_READ */
#if BROTLI_LITTLE_ENDIAN
/* Straight endianness. Just read / write values. */
@@ -453,6 +377,16 @@ static BROTLI_INLINE void BROTLI_UNALIGNED_STORE64LE(void* p, uint64_t v) {
}
#endif /* BROTLI_LITTLE_ENDIAN */
+static BROTLI_INLINE void* BROTLI_UNALIGNED_LOAD_PTR(const void* p) {
+ void* v;
+ memcpy(&v, p, sizeof(void*));
+ return v;
+}
+
+static BROTLI_INLINE void BROTLI_UNALIGNED_STORE_PTR(void* p, const void* v) {
+ memcpy(p, &v, sizeof(void*));
+}
+
/* BROTLI_IS_CONSTANT macros returns true for compile-time constants. */
#if BROTLI_GNUC_HAS_BUILTIN(__builtin_constant_p, 3, 0, 1) || \
BROTLI_INTEL_VERSION_CHECK(16, 0, 0)
@@ -474,22 +408,34 @@ static BROTLI_INLINE void BROTLI_UNALIGNED_STORE64LE(void* p, uint64_t v) {
#endif
#if defined(BROTLI_DEBUG) || defined(BROTLI_ENABLE_LOG)
+#define BROTLI_ENABLE_DUMP_DEFAULT 1
#define BROTLI_DCHECK(x) assert(x)
+#else
+#define BROTLI_ENABLE_DUMP_DEFAULT 0
+#define BROTLI_DCHECK(x)
+#endif
+
+#if !defined(BROTLI_ENABLE_DUMP)
+#define BROTLI_ENABLE_DUMP BROTLI_ENABLE_DUMP_DEFAULT
+#endif
+
+#if BROTLI_ENABLE_DUMP
static BROTLI_INLINE void BrotliDump(const char* f, int l, const char* fn) {
fprintf(stderr, "%s:%d (%s)\n", f, l, fn);
fflush(stderr);
}
#define BROTLI_DUMP() BrotliDump(__FILE__, __LINE__, __FUNCTION__)
#else
-#define BROTLI_DCHECK(x)
#define BROTLI_DUMP() (void)(0)
#endif
-/* TODO: add appropriate icc/sunpro/arm/ibm/ti checks. */
+/* BrotliRBit assumes brotli_reg_t fits native CPU register type. */
+#if (BROTLI_64_BITS == BROTLI_TARGET_64_BITS)
+/* TODO(eustas): add appropriate icc/sunpro/arm/ibm/ti checks. */
#if (BROTLI_GNUC_VERSION_CHECK(3, 0, 0) || defined(__llvm__)) && \
!defined(BROTLI_BUILD_NO_RBIT)
#if defined(BROTLI_TARGET_ARMV7) || defined(BROTLI_TARGET_ARMV8_ANY)
-/* TODO: detect ARMv6T2 and enable this code for it. */
+/* TODO(eustas): detect ARMv6T2 and enable this code for it. */
static BROTLI_INLINE brotli_reg_t BrotliRBit(brotli_reg_t input) {
brotli_reg_t output;
__asm__("rbit %0, %1\n" : "=r"(output) : "r"(input));
@@ -498,15 +444,14 @@ static BROTLI_INLINE brotli_reg_t BrotliRBit(brotli_reg_t input) {
#define BROTLI_RBIT(x) BrotliRBit(x)
#endif /* armv7 / armv8 */
#endif /* gcc || clang */
+#endif /* brotli_reg_t is native */
#if !defined(BROTLI_RBIT)
static BROTLI_INLINE void BrotliRBit(void) { /* Should break build if used. */ }
#endif /* BROTLI_RBIT */
-#define BROTLI_REPEAT(N, X) { \
- if ((N & 1) != 0) {X;} \
- if ((N & 2) != 0) {X; X;} \
- if ((N & 4) != 0) {X; X; X; X;} \
-}
+#define BROTLI_REPEAT_4(X) {X; X; X; X;}
+#define BROTLI_REPEAT_5(X) {X; X; X; X; X;}
+#define BROTLI_REPEAT_6(X) {X; X; X; X; X; X;}
#define BROTLI_UNUSED(X) (void)(X)
@@ -529,7 +474,7 @@ BROTLI_MIN_MAX(size_t) BROTLI_MIN_MAX(uint32_t) BROTLI_MIN_MAX(uint8_t)
#if BROTLI_GNUC_HAS_BUILTIN(__builtin_ctzll, 3, 4, 0) || \
BROTLI_INTEL_VERSION_CHECK(16, 0, 0)
#define BROTLI_TZCNT64 __builtin_ctzll
-#elif BROTLI_MSVC_VERSION_CHECK(12, 0, 0)
+#elif BROTLI_MSVC_VERSION_CHECK(18, 0, 0)
#if defined(BROTLI_TARGET_X64)
#define BROTLI_TZCNT64 _tzcnt_u64
#else /* BROTLI_TARGET_X64 */
@@ -546,7 +491,7 @@ static BROTLI_INLINE uint32_t BrotliBsf64Msvc(uint64_t x) {
#if BROTLI_GNUC_HAS_BUILTIN(__builtin_clz, 3, 4, 0) || \
BROTLI_INTEL_VERSION_CHECK(16, 0, 0)
#define BROTLI_BSR32(x) (31u ^ (uint32_t)__builtin_clz(x))
-#elif BROTLI_MSVC_VERSION_CHECK(12, 0, 0)
+#elif BROTLI_MSVC_VERSION_CHECK(18, 0, 0)
static BROTLI_INLINE uint32_t BrotliBsr32Msvc(uint32_t x) {
unsigned long msb;
_BitScanReverse(&msb, x);
@@ -571,6 +516,8 @@ BROTLI_UNUSED_FUNCTION void BrotliSuppressUnusedFunctions(void) {
BROTLI_UNUSED(&BROTLI_UNALIGNED_LOAD32LE);
BROTLI_UNUSED(&BROTLI_UNALIGNED_LOAD64LE);
BROTLI_UNUSED(&BROTLI_UNALIGNED_STORE64LE);
+ BROTLI_UNUSED(&BROTLI_UNALIGNED_LOAD_PTR);
+ BROTLI_UNUSED(&BROTLI_UNALIGNED_STORE_PTR);
BROTLI_UNUSED(&BrotliRBit);
BROTLI_UNUSED(&brotli_min_double);
BROTLI_UNUSED(&brotli_max_double);
@@ -586,7 +533,7 @@ BROTLI_UNUSED_FUNCTION void BrotliSuppressUnusedFunctions(void) {
BROTLI_UNUSED(&brotli_max_uint8_t);
BROTLI_UNUSED(&BrotliDefaultAllocFunc);
BROTLI_UNUSED(&BrotliDefaultFreeFunc);
-#if defined(BROTLI_DEBUG) || defined(BROTLI_ENABLE_LOG)
+#if BROTLI_ENABLE_DUMP
BROTLI_UNUSED(&BrotliDump);
#endif
}
diff --git a/dviware/dvisvgm/libs/brotli/common/shared_dictionary_internal.h b/dviware/dvisvgm/libs/brotli/common/shared_dictionary_internal.h
new file mode 100644
index 0000000000..963762e432
--- /dev/null
+++ b/dviware/dvisvgm/libs/brotli/common/shared_dictionary_internal.h
@@ -0,0 +1,75 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+ Distributed under MIT license.
+ See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+/* (Transparent) Shared Dictionary definition. */
+
+#ifndef BROTLI_COMMON_SHARED_DICTIONARY_INTERNAL_H_
+#define BROTLI_COMMON_SHARED_DICTIONARY_INTERNAL_H_
+
+#include <brotli/shared_dictionary.h>
+#include <brotli/types.h>
+
+#include "dictionary.h"
+#include "transform.h"
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+struct BrotliSharedDictionaryStruct {
+ /* LZ77 prefixes (compound dictionary). */
+ uint32_t num_prefix; /* max SHARED_BROTLI_MAX_COMPOUND_DICTS */
+ size_t prefix_size[SHARED_BROTLI_MAX_COMPOUND_DICTS];
+ const uint8_t* prefix[SHARED_BROTLI_MAX_COMPOUND_DICTS];
+
+ /* If set, the context map is used to select word and transform list from 64
+ contexts, if not set, the context map is not used and only words[0] and
+ transforms[0] are to be used. */
+ BROTLI_BOOL context_based;
+
+ uint8_t context_map[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+ /* Amount of word_list+transform_list combinations. */
+ uint8_t num_dictionaries;
+
+ /* Must use num_dictionaries values. */
+ const BrotliDictionary* words[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+ /* Must use num_dictionaries values. */
+ const BrotliTransforms* transforms[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+ /* Amount of custom word lists. May be 0 if only Brotli's built-in is used */
+ uint8_t num_word_lists;
+
+ /* Contents of the custom words lists. Must be NULL if num_word_lists is 0. */
+ BrotliDictionary* words_instances;
+
+ /* Amount of custom transform lists. May be 0 if only Brotli's built-in is
+ used */
+ uint8_t num_transform_lists;
+
+ /* Contents of the custom transform lists. Must be NULL if num_transform_lists
+ is 0. */
+ BrotliTransforms* transforms_instances;
+
+ /* Concatenated prefix_suffix_maps of the custom transform lists. Must be NULL
+ if num_transform_lists is 0. */
+ uint16_t* prefix_suffix_maps;
+
+ /* Memory management */
+ brotli_alloc_func alloc_func;
+ brotli_free_func free_func;
+ void* memory_manager_opaque;
+};
+
+typedef struct BrotliSharedDictionaryStruct BrotliSharedDictionaryInternal;
+#define BrotliSharedDictionary BrotliSharedDictionaryInternal
+
+#if defined(__cplusplus) || defined(c_plusplus)
+} /* extern "C" */
+#endif
+
+#endif /* BROTLI_COMMON_SHARED_DICTIONARY_INTERNAL_H_ */
diff --git a/dviware/dvisvgm/libs/brotli/common/transform.c b/dviware/dvisvgm/libs/brotli/common/transform.c
index f8fa4335e1..49455fc496 100644
--- a/dviware/dvisvgm/libs/brotli/common/transform.c
+++ b/dviware/dvisvgm/libs/brotli/common/transform.c
@@ -4,7 +4,7 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./transform.h"
+#include "transform.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/common/version.h b/dviware/dvisvgm/libs/brotli/common/version.h
index 01b2998e25..8098040f64 100644
--- a/dviware/dvisvgm/libs/brotli/common/version.h
+++ b/dviware/dvisvgm/libs/brotli/common/version.h
@@ -9,18 +9,43 @@
#ifndef BROTLI_COMMON_VERSION_H_
#define BROTLI_COMMON_VERSION_H_
-/* This macro should only be used when library is compiled together with client.
- If library is dynamically linked, use BrotliDecoderVersion and
+/* Compose 3 components into a single number. In a hexadecimal representation
+ B and C components occupy exactly 3 digits. */
+#define BROTLI_MAKE_HEX_VERSION(A, B, C) ((A << 24) | (B << 12) | C)
+
+/* Those macros should only be used when library is compiled together with
+ the client. If library is dynamically linked, use BrotliDecoderVersion and
BrotliEncoderVersion methods. */
-/* Semantic version, calculated as (MAJOR << 24) | (MINOR << 12) | PATCH */
-#define BROTLI_VERSION 0x1000009
+#define BROTLI_VERSION_MAJOR 1
+#define BROTLI_VERSION_MINOR 1
+#define BROTLI_VERSION_PATCH 0
+
+#define BROTLI_VERSION BROTLI_MAKE_HEX_VERSION( \
+ BROTLI_VERSION_MAJOR, BROTLI_VERSION_MINOR, BROTLI_VERSION_PATCH)
/* This macro is used by build system to produce Libtool-friendly soname. See
https://www.gnu.org/software/libtool/manual/html_node/Libtool-versioning.html
+ Version evolution rules:
+ - interfaces added (or change is compatible) -> current+1:0:age+1
+ - interfaces removed (or changed is incompatible) -> current+1:0:0
+ - interfaces not changed -> current:revision+1:age
*/
-/* ABI version, calculated as (CURRENT << 24) | (REVISION << 12) | AGE */
-#define BROTLI_ABI_VERSION 0x1009000
+#define BROTLI_ABI_CURRENT 2
+#define BROTLI_ABI_REVISION 0
+#define BROTLI_ABI_AGE 1
+
+#if BROTLI_VERSION_MAJOR != (BROTLI_ABI_CURRENT - BROTLI_ABI_AGE)
+#error ABI/API version inconsistency
+#endif
+
+#if BROTLI_VERSION_MINOR != BROTLI_ABI_AGE
+#error ABI/API version inconsistency
+#endif
+
+#if BROTLI_VERSION_PATCH != BROTLI_ABI_REVISION
+#error ABI/API version inconsistency
+#endif
#endif /* BROTLI_COMMON_VERSION_H_ */
diff --git a/dviware/dvisvgm/libs/brotli/enc/backward_references.c b/dviware/dvisvgm/libs/brotli/enc/backward_references.c
index a07a617a09..f600e6452f 100644
--- a/dviware/dvisvgm/libs/brotli/enc/backward_references.c
+++ b/dviware/dvisvgm/libs/brotli/enc/backward_references.c
@@ -6,17 +6,19 @@
/* Function to find backward reference copies. */
-#include "./backward_references.h"
+#include "backward_references.h"
+
+#include <brotli/types.h>
#include "../common/constants.h"
-#include "../common/context.h"
#include "../common/dictionary.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./command.h"
-#include "./dictionary_hash.h"
-#include "./memory.h"
-#include "./quality.h"
+#include "command.h"
+#include "compound_dictionary.h"
+#include "dictionary_hash.h"
+#include "encoder_dict.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -52,67 +54,103 @@ static BROTLI_INLINE size_t ComputeDistanceCode(size_t distance,
#define EXPORT_FN(X) EXPAND_CAT(X, EXPAND_CAT(PREFIX(), HASHER()))
#define PREFIX() N
+#define ENABLE_COMPOUND_DICTIONARY 0
#define HASHER() H2
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H3
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H4
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H5
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H6
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H40
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H41
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H42
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H54
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H35
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H55
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
#define HASHER() H65
/* NOLINTNEXTLINE(build/include) */
-#include "./backward_references_inc.h"
+#include "backward_references_inc.h"
#undef HASHER
+#undef ENABLE_COMPOUND_DICTIONARY
+#undef PREFIX
+#define PREFIX() D
+#define ENABLE_COMPOUND_DICTIONARY 1
+
+#define HASHER() H5
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+#define HASHER() H6
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+#define HASHER() H40
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+#define HASHER() H41
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+#define HASHER() H42
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+#define HASHER() H55
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+#define HASHER() H65
+/* NOLINTNEXTLINE(build/include) */
+#include "backward_references_inc.h"
+#undef HASHER
+
+#undef ENABLE_COMPOUND_DICTIONARY
#undef PREFIX
#undef EXPORT_FN
@@ -125,6 +163,29 @@ void BrotliCreateBackwardReferences(size_t num_bytes,
ContextLut literal_context_lut, const BrotliEncoderParams* params,
Hasher* hasher, int* dist_cache, size_t* last_insert_len,
Command* commands, size_t* num_commands, size_t* num_literals) {
+ if (params->dictionary.compound.num_chunks != 0) {
+ switch (params->hasher.type) {
+#define CASE_(N) \
+ case N: \
+ CreateBackwardReferencesDH ## N(num_bytes, \
+ position, ringbuffer, ringbuffer_mask, \
+ literal_context_lut, params, hasher, dist_cache, \
+ last_insert_len, commands, num_commands, num_literals); \
+ return;
+ CASE_(5)
+ CASE_(6)
+ CASE_(40)
+ CASE_(41)
+ CASE_(42)
+ CASE_(55)
+ CASE_(65)
+#undef CASE_
+ default:
+ BROTLI_DCHECK(false);
+ break;
+ }
+ }
+
switch (params->hasher.type) {
#define CASE_(N) \
case N: \
@@ -136,6 +197,7 @@ void BrotliCreateBackwardReferences(size_t num_bytes,
FOR_GENERIC_HASHERS(CASE_)
#undef CASE_
default:
+ BROTLI_DCHECK(false);
break;
}
}
diff --git a/dviware/dvisvgm/libs/brotli/enc/backward_references.h b/dviware/dvisvgm/libs/brotli/enc/backward_references.h
index 9589cc1541..20fb98a4d8 100644
--- a/dviware/dvisvgm/libs/brotli/enc/backward_references.h
+++ b/dviware/dvisvgm/libs/brotli/enc/backward_references.h
@@ -9,14 +9,15 @@
#ifndef BROTLI_ENC_BACKWARD_REFERENCES_H_
#define BROTLI_ENC_BACKWARD_REFERENCES_H_
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/context.h"
#include "../common/dictionary.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./command.h"
-#include "./hash.h"
-#include "./quality.h"
+#include "command.h"
+#include "hash.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.c b/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.c
index 5651caeb7a..6325032e1c 100644
--- a/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.c
+++ b/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.c
@@ -6,22 +6,24 @@
/* Function to find backward reference copies. */
-#include "./backward_references_hq.h"
+#include "backward_references_hq.h"
#include <string.h> /* memcpy, memset */
+#include <brotli/types.h>
+
#include "../common/constants.h"
-#include "../common/context.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./command.h"
-#include "./fast_log.h"
-#include "./find_match_length.h"
-#include "./literal_cost.h"
-#include "./memory.h"
-#include "./params.h"
-#include "./prefix.h"
-#include "./quality.h"
+#include "command.h"
+#include "compound_dictionary.h"
+#include "encoder_dict.h"
+#include "fast_log.h"
+#include "find_match_length.h"
+#include "literal_cost.h"
+#include "memory.h"
+#include "params.h"
+#include "prefix.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -73,6 +75,14 @@ static BROTLI_INLINE uint32_t ZopfliNodeCommandLength(const ZopfliNode* self) {
return ZopfliNodeCopyLength(self) + (self->dcode_insert_length & 0x7FFFFFF);
}
+/* Temporary data for ZopfliCostModelSetFromCommands. */
+typedef struct ZopfliCostModelArena {
+ uint32_t histogram_literal[BROTLI_NUM_LITERAL_SYMBOLS];
+ uint32_t histogram_cmd[BROTLI_NUM_COMMAND_SYMBOLS];
+ uint32_t histogram_dist[BROTLI_MAX_EFFECTIVE_DISTANCE_ALPHABET_SIZE];
+ float cost_literal[BROTLI_NUM_LITERAL_SYMBOLS];
+} ZopfliCostModelArena;
+
/* Histogram based cost model for zopflification. */
typedef struct ZopfliCostModel {
/* The insert and copy length symbols. */
@@ -83,6 +93,12 @@ typedef struct ZopfliCostModel {
float* literal_costs_;
float min_cost_cmd_;
size_t num_bytes_;
+
+ /* Temporary data. */
+ union {
+ size_t literal_histograms[3 * 256];
+ ZopfliCostModelArena arena;
+ };
} ZopfliCostModel;
static void InitZopfliCostModel(
@@ -139,18 +155,15 @@ static void ZopfliCostModelSetFromCommands(ZopfliCostModel* self,
const Command* commands,
size_t num_commands,
size_t last_insert_len) {
- uint32_t histogram_literal[BROTLI_NUM_LITERAL_SYMBOLS];
- uint32_t histogram_cmd[BROTLI_NUM_COMMAND_SYMBOLS];
- uint32_t histogram_dist[BROTLI_MAX_EFFECTIVE_DISTANCE_ALPHABET_SIZE];
- float cost_literal[BROTLI_NUM_LITERAL_SYMBOLS];
+ ZopfliCostModelArena* arena = &self->arena;
size_t pos = position - last_insert_len;
float min_cost_cmd = kInfinity;
size_t i;
float* cost_cmd = self->cost_cmd_;
- memset(histogram_literal, 0, sizeof(histogram_literal));
- memset(histogram_cmd, 0, sizeof(histogram_cmd));
- memset(histogram_dist, 0, sizeof(histogram_dist));
+ memset(arena->histogram_literal, 0, sizeof(arena->histogram_literal));
+ memset(arena->histogram_cmd, 0, sizeof(arena->histogram_cmd));
+ memset(arena->histogram_dist, 0, sizeof(arena->histogram_dist));
for (i = 0; i < num_commands; i++) {
size_t inslength = commands[i].insert_len_;
@@ -159,21 +172,21 @@ static void ZopfliCostModelSetFromCommands(ZopfliCostModel* self,
size_t cmdcode = commands[i].cmd_prefix_;
size_t j;
- histogram_cmd[cmdcode]++;
- if (cmdcode >= 128) histogram_dist[distcode]++;
+ arena->histogram_cmd[cmdcode]++;
+ if (cmdcode >= 128) arena->histogram_dist[distcode]++;
for (j = 0; j < inslength; j++) {
- histogram_literal[ringbuffer[(pos + j) & ringbuffer_mask]]++;
+ arena->histogram_literal[ringbuffer[(pos + j) & ringbuffer_mask]]++;
}
pos += inslength + copylength;
}
- SetCost(histogram_literal, BROTLI_NUM_LITERAL_SYMBOLS, BROTLI_TRUE,
- cost_literal);
- SetCost(histogram_cmd, BROTLI_NUM_COMMAND_SYMBOLS, BROTLI_FALSE,
+ SetCost(arena->histogram_literal, BROTLI_NUM_LITERAL_SYMBOLS, BROTLI_TRUE,
+ arena->cost_literal);
+ SetCost(arena->histogram_cmd, BROTLI_NUM_COMMAND_SYMBOLS, BROTLI_FALSE,
cost_cmd);
- SetCost(histogram_dist, self->distance_histogram_size, BROTLI_FALSE,
+ SetCost(arena->histogram_dist, self->distance_histogram_size, BROTLI_FALSE,
self->cost_dist_);
for (i = 0; i < BROTLI_NUM_COMMAND_SYMBOLS; ++i) {
@@ -188,7 +201,7 @@ static void ZopfliCostModelSetFromCommands(ZopfliCostModel* self,
literal_costs[0] = 0.0;
for (i = 0; i < num_bytes; ++i) {
literal_carry +=
- cost_literal[ringbuffer[(position + i) & ringbuffer_mask]];
+ arena->cost_literal[ringbuffer[(position + i) & ringbuffer_mask]];
literal_costs[i + 1] = literal_costs[i] + literal_carry;
literal_carry -= literal_costs[i + 1] - literal_costs[i];
}
@@ -206,7 +219,8 @@ static void ZopfliCostModelSetFromLiteralCosts(ZopfliCostModel* self,
size_t num_bytes = self->num_bytes_;
size_t i;
BrotliEstimateBitCostsForLiterals(position, num_bytes, ringbuffer_mask,
- ringbuffer, &literal_costs[1]);
+ ringbuffer, self->literal_histograms,
+ &literal_costs[1]);
literal_costs[0] = 0.0;
for (i = 0; i < num_bytes; ++i) {
literal_carry += literal_costs[i + 1];
@@ -418,7 +432,8 @@ static size_t UpdateNodes(
size_t min_len;
size_t result = 0;
size_t k;
- size_t gap = 0;
+ const CompoundDictionary* addon = &params->dictionary.compound;
+ size_t gap = addon->total_size;
EvaluateNode(block_start + stream_offset, pos, max_backward_limit, gap,
starting_dist_cache, model, queue, nodes);
@@ -472,6 +487,24 @@ static size_t UpdateNodes(
len = FindMatchLengthWithLimit(&ringbuffer[prev_ix],
&ringbuffer[cur_ix_masked],
max_len);
+ } else if (backward > dictionary_start) {
+ size_t d = 0;
+ size_t offset;
+ size_t limit;
+ const uint8_t* source;
+ offset = dictionary_start + 1 + addon->total_size - 1;
+ while (offset >= backward + addon->chunk_offsets[d + 1]) d++;
+ source = addon->chunk_source[d];
+ offset = offset - addon->chunk_offsets[d] - backward;
+ limit = addon->chunk_offsets[d + 1] - addon->chunk_offsets[d] - offset;
+ limit = limit > max_len ? max_len : limit;
+ if (best_len >= limit ||
+ continuation != source[offset + best_len]) {
+ continue;
+ }
+ len = FindMatchLengthWithLimit(&source[offset],
+ &ringbuffer[cur_ix_masked],
+ limit);
} else {
/* "Gray" area. It is addressable by decoder, but this encoder
instance does not have that data -> should not touch it. */
@@ -577,7 +610,7 @@ void BrotliZopfliCreateCommands(const size_t num_bytes,
size_t pos = 0;
uint32_t offset = nodes[0].u.next;
size_t i;
- size_t gap = 0;
+ size_t gap = params->dictionary.compound.total_size;
for (i = 0; offset != BROTLI_UINT32_MAX; i++) {
const ZopfliNode* next = &nodes[pos + offset];
size_t copy_length = ZopfliNodeCopyLength(next);
@@ -653,6 +686,23 @@ static size_t ZopfliIterate(size_t num_bytes, size_t position,
return ComputeShortestPathFromNodes(num_bytes, nodes);
}
+static void MergeMatches(BackwardMatch* dst,
+ BackwardMatch* src1, size_t len1, BackwardMatch* src2, size_t len2) {
+ while (len1 > 0 && len2 > 0) {
+ size_t l1 = BackwardMatchLength(src1);
+ size_t l2 = BackwardMatchLength(src2);
+ if (l1 < l2 || ((l1 == l2) && (src1->distance < src2->distance))) {
+ *dst++ = *src1++;
+ len1--;
+ } else {
+ *dst++ = *src2++;
+ len2--;
+ }
+ }
+ while (len1-- > 0) *dst++ = *src1++;
+ while (len2-- > 0) *dst++ = *src2++;
+}
+
/* REQUIRES: nodes != NULL and len(nodes) >= num_bytes + 1 */
size_t BrotliZopfliComputeShortestPath(MemoryManager* m, size_t num_bytes,
size_t position, const uint8_t* ringbuffer, size_t ringbuffer_mask,
@@ -661,21 +711,26 @@ size_t BrotliZopfliComputeShortestPath(MemoryManager* m, size_t num_bytes,
const size_t stream_offset = params->stream_offset;
const size_t max_backward_limit = BROTLI_MAX_BACKWARD_LIMIT(params->lgwin);
const size_t max_zopfli_len = MaxZopfliLen(params);
- ZopfliCostModel model;
StartPosQueue queue;
- BackwardMatch matches[2 * (MAX_NUM_MATCHES_H10 + 64)];
+ BackwardMatch* BROTLI_RESTRICT matches =
+ BROTLI_ALLOC(m, BackwardMatch, 2 * (MAX_NUM_MATCHES_H10 + 64));
const size_t store_end = num_bytes >= StoreLookaheadH10() ?
position + num_bytes - StoreLookaheadH10() + 1 : position;
size_t i;
- size_t gap = 0;
- size_t lz_matches_offset = 0;
- BROTLI_UNUSED(literal_context_lut);
+ const CompoundDictionary* addon = &params->dictionary.compound;
+ size_t gap = addon->total_size;
+ size_t lz_matches_offset =
+ (addon->num_chunks != 0) ? (MAX_NUM_MATCHES_H10 + 128) : 0;
+ ZopfliCostModel* model = BROTLI_ALLOC(m, ZopfliCostModel, 1);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(model) || BROTLI_IS_NULL(matches)) {
+ return 0;
+ }
nodes[0].length = 0;
nodes[0].u.cost = 0;
- InitZopfliCostModel(m, &model, &params->dist, num_bytes);
+ InitZopfliCostModel(m, model, &params->dist, num_bytes);
if (BROTLI_IS_OOM(m)) return 0;
ZopfliCostModelSetFromLiteralCosts(
- &model, position, ringbuffer, ringbuffer_mask);
+ model, position, ringbuffer, ringbuffer_mask);
InitStartPosQueue(&queue);
for (i = 0; i + HashTypeLengthH10() - 1 < num_bytes; i++) {
const size_t pos = position + i;
@@ -684,17 +739,35 @@ size_t BrotliZopfliComputeShortestPath(MemoryManager* m, size_t num_bytes,
pos + stream_offset, max_backward_limit);
size_t skip;
size_t num_matches;
+ int dict_id = 0;
+ if (params->dictionary.contextual.context_based) {
+ uint8_t p1 = pos >= 1 ?
+ ringbuffer[(size_t)(pos - 1) & ringbuffer_mask] : 0;
+ uint8_t p2 = pos >= 2 ?
+ ringbuffer[(size_t)(pos - 2) & ringbuffer_mask] : 0;
+ dict_id = params->dictionary.contextual.context_map[
+ BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+ }
num_matches = FindAllMatchesH10(&hasher->privat._H10,
- &params->dictionary,
+ params->dictionary.contextual.dict[dict_id],
ringbuffer, ringbuffer_mask, pos, num_bytes - i, max_distance,
dictionary_start + gap, params, &matches[lz_matches_offset]);
+ if (addon->num_chunks != 0) {
+ size_t cd_matches = LookupAllCompoundDictionaryMatches(addon,
+ ringbuffer, ringbuffer_mask, pos, 3, num_bytes - i,
+ dictionary_start, params->dist.max_distance,
+ &matches[lz_matches_offset - 64], 64);
+ MergeMatches(matches, &matches[lz_matches_offset - 64], cd_matches,
+ &matches[lz_matches_offset], num_matches);
+ num_matches += cd_matches;
+ }
if (num_matches > 0 &&
BackwardMatchLength(&matches[num_matches - 1]) > max_zopfli_len) {
matches[0] = matches[num_matches - 1];
num_matches = 1;
}
skip = UpdateNodes(num_bytes, position, i, ringbuffer, ringbuffer_mask,
- params, max_backward_limit, dist_cache, num_matches, matches, &model,
+ params, max_backward_limit, dist_cache, num_matches, matches, model,
&queue, nodes);
if (skip < BROTLI_LONG_COPY_QUICK_STEP) skip = 0;
if (num_matches == 1 && BackwardMatchLength(&matches[0]) > max_zopfli_len) {
@@ -710,12 +783,14 @@ size_t BrotliZopfliComputeShortestPath(MemoryManager* m, size_t num_bytes,
i++;
if (i + HashTypeLengthH10() - 1 >= num_bytes) break;
EvaluateNode(position + stream_offset, i, max_backward_limit, gap,
- dist_cache, &model, &queue, nodes);
+ dist_cache, model, &queue, nodes);
skip--;
}
}
}
- CleanupZopfliCostModel(m, &model);
+ CleanupZopfliCostModel(m, model);
+ BROTLI_FREE(m, model);
+ BROTLI_FREE(m, matches);
return ComputeShortestPathFromNodes(num_bytes, nodes);
}
@@ -753,14 +828,15 @@ void BrotliCreateHqZopfliBackwardReferences(MemoryManager* m, size_t num_bytes,
size_t orig_last_insert_len;
int orig_dist_cache[4];
size_t orig_num_commands;
- ZopfliCostModel model;
+ ZopfliCostModel* model = BROTLI_ALLOC(m, ZopfliCostModel, 1);
ZopfliNode* nodes;
BackwardMatch* matches = BROTLI_ALLOC(m, BackwardMatch, matches_size);
- size_t gap = 0;
- size_t shadow_matches = 0;
- BROTLI_UNUSED(literal_context_lut);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(num_matches) ||
- BROTLI_IS_NULL(matches)) {
+ const CompoundDictionary* addon = &params->dictionary.compound;
+ size_t gap = addon->total_size;
+ size_t shadow_matches =
+ (addon->num_chunks != 0) ? (MAX_NUM_MATCHES_H10 + 128) : 0;
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(model) ||
+ BROTLI_IS_NULL(num_matches) || BROTLI_IS_NULL(matches)) {
return;
}
for (i = 0; i + HashTypeLengthH10() - 1 < num_bytes; ++i) {
@@ -772,15 +848,34 @@ void BrotliCreateHqZopfliBackwardReferences(MemoryManager* m, size_t num_bytes,
size_t num_found_matches;
size_t cur_match_end;
size_t j;
+ int dict_id = 0;
+ if (params->dictionary.contextual.context_based) {
+ uint8_t p1 = pos >= 1 ?
+ ringbuffer[(size_t)(pos - 1) & ringbuffer_mask] : 0;
+ uint8_t p2 = pos >= 2 ?
+ ringbuffer[(size_t)(pos - 2) & ringbuffer_mask] : 0;
+ dict_id = params->dictionary.contextual.context_map[
+ BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+ }
/* Ensure that we have enough free slots. */
BROTLI_ENSURE_CAPACITY(m, BackwardMatch, matches, matches_size,
cur_match_pos + MAX_NUM_MATCHES_H10 + shadow_matches);
if (BROTLI_IS_OOM(m)) return;
num_found_matches = FindAllMatchesH10(&hasher->privat._H10,
- &params->dictionary,
+ params->dictionary.contextual.dict[dict_id],
ringbuffer, ringbuffer_mask, pos, max_length,
max_distance, dictionary_start + gap, params,
&matches[cur_match_pos + shadow_matches]);
+ if (addon->num_chunks != 0) {
+ size_t cd_matches = LookupAllCompoundDictionaryMatches(addon,
+ ringbuffer, ringbuffer_mask, pos, 3, max_length,
+ dictionary_start, params->dist.max_distance,
+ &matches[cur_match_pos + shadow_matches - 64], 64);
+ MergeMatches(&matches[cur_match_pos],
+ &matches[cur_match_pos + shadow_matches - 64], cd_matches,
+ &matches[cur_match_pos + shadow_matches], num_found_matches);
+ num_found_matches += cd_matches;
+ }
cur_match_end = cur_match_pos + num_found_matches;
for (j = cur_match_pos; j + 1 < cur_match_end; ++j) {
BROTLI_DCHECK(BackwardMatchLength(&matches[j]) <=
@@ -810,15 +905,15 @@ void BrotliCreateHqZopfliBackwardReferences(MemoryManager* m, size_t num_bytes,
orig_num_commands = *num_commands;
nodes = BROTLI_ALLOC(m, ZopfliNode, num_bytes + 1);
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(nodes)) return;
- InitZopfliCostModel(m, &model, &params->dist, num_bytes);
+ InitZopfliCostModel(m, model, &params->dist, num_bytes);
if (BROTLI_IS_OOM(m)) return;
for (i = 0; i < 2; i++) {
BrotliInitZopfliNodes(nodes, num_bytes + 1);
if (i == 0) {
ZopfliCostModelSetFromLiteralCosts(
- &model, position, ringbuffer, ringbuffer_mask);
+ model, position, ringbuffer, ringbuffer_mask);
} else {
- ZopfliCostModelSetFromCommands(&model, position, ringbuffer,
+ ZopfliCostModelSetFromCommands(model, position, ringbuffer,
ringbuffer_mask, commands, *num_commands - orig_num_commands,
orig_last_insert_len);
}
@@ -827,12 +922,13 @@ void BrotliCreateHqZopfliBackwardReferences(MemoryManager* m, size_t num_bytes,
*last_insert_len = orig_last_insert_len;
memcpy(dist_cache, orig_dist_cache, 4 * sizeof(dist_cache[0]));
*num_commands += ZopfliIterate(num_bytes, position, ringbuffer,
- ringbuffer_mask, params, gap, dist_cache, &model, num_matches, matches,
+ ringbuffer_mask, params, gap, dist_cache, model, num_matches, matches,
nodes);
BrotliZopfliCreateCommands(num_bytes, position, nodes, dist_cache,
last_insert_len, params, commands, num_literals);
}
- CleanupZopfliCostModel(m, &model);
+ CleanupZopfliCostModel(m, model);
+ BROTLI_FREE(m, model);
BROTLI_FREE(m, nodes);
BROTLI_FREE(m, matches);
BROTLI_FREE(m, num_matches);
diff --git a/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.h b/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.h
index 36b75f250d..8acf975ab9 100644
--- a/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.h
+++ b/dviware/dvisvgm/libs/brotli/enc/backward_references_hq.h
@@ -9,15 +9,16 @@
#ifndef BROTLI_ENC_BACKWARD_REFERENCES_HQ_H_
#define BROTLI_ENC_BACKWARD_REFERENCES_HQ_H_
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/context.h"
#include "../common/dictionary.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./command.h"
-#include "./hash.h"
-#include "./memory.h"
-#include "./quality.h"
+#include "command.h"
+#include "hash.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/backward_references_inc.h b/dviware/dvisvgm/libs/brotli/enc/backward_references_inc.h
index 766bf91ffd..752c12e9fd 100644
--- a/dviware/dvisvgm/libs/brotli/enc/backward_references_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/backward_references_inc.h
@@ -28,13 +28,11 @@ static BROTLI_NOINLINE void EXPORT_FN(CreateBackwardReferences)(
const size_t random_heuristics_window_size =
LiteralSpreeLengthForSparseSearch(params);
size_t apply_random_heuristics = position + random_heuristics_window_size;
- const size_t gap = 0;
+ const size_t gap = params->dictionary.compound.total_size;
/* Minimum score to accept a backward reference. */
const score_t kMinScore = BROTLI_SCORE_BASE + 100;
- BROTLI_UNUSED(literal_context_lut);
-
FN(PrepareDistanceCache)(privat, dist_cache);
while (position + FN(HashTypeLength)() < pos_end) {
@@ -43,13 +41,29 @@ static BROTLI_NOINLINE void EXPORT_FN(CreateBackwardReferences)(
size_t dictionary_start = BROTLI_MIN(size_t,
position + position_offset, max_backward_limit);
HasherSearchResult sr;
+ int dict_id = 0;
+ uint8_t p1 = 0;
+ uint8_t p2 = 0;
+ if (params->dictionary.contextual.context_based) {
+ p1 = position >= 1 ?
+ ringbuffer[(size_t)(position - 1) & ringbuffer_mask] : 0;
+ p2 = position >= 2 ?
+ ringbuffer[(size_t)(position - 2) & ringbuffer_mask] : 0;
+ dict_id = params->dictionary.contextual.context_map[
+ BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+ }
sr.len = 0;
sr.len_code_delta = 0;
sr.distance = 0;
sr.score = kMinScore;
- FN(FindLongestMatch)(privat, &params->dictionary,
+ FN(FindLongestMatch)(privat, params->dictionary.contextual.dict[dict_id],
ringbuffer, ringbuffer_mask, dist_cache, position, max_length,
max_distance, dictionary_start + gap, params->dist.max_distance, &sr);
+ if (ENABLE_COMPOUND_DICTIONARY) {
+ LookupCompoundDictionaryMatch(&params->dictionary.compound, ringbuffer,
+ ringbuffer_mask, dist_cache, position, max_length,
+ dictionary_start, params->dist.max_distance, &sr);
+ }
if (sr.score > kMinScore) {
/* Found a match. Let's look for something even better ahead. */
int delayed_backward_references_in_row = 0;
@@ -65,11 +79,23 @@ static BROTLI_NOINLINE void EXPORT_FN(CreateBackwardReferences)(
max_distance = BROTLI_MIN(size_t, position + 1, max_backward_limit);
dictionary_start = BROTLI_MIN(size_t,
position + 1 + position_offset, max_backward_limit);
+ if (params->dictionary.contextual.context_based) {
+ p2 = p1;
+ p1 = ringbuffer[position & ringbuffer_mask];
+ dict_id = params->dictionary.contextual.context_map[
+ BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+ }
FN(FindLongestMatch)(privat,
- &params->dictionary,
+ params->dictionary.contextual.dict[dict_id],
ringbuffer, ringbuffer_mask, dist_cache, position + 1, max_length,
max_distance, dictionary_start + gap, params->dist.max_distance,
&sr2);
+ if (ENABLE_COMPOUND_DICTIONARY) {
+ LookupCompoundDictionaryMatch(
+ &params->dictionary.compound, ringbuffer,
+ ringbuffer_mask, dist_cache, position + 1, max_length,
+ dictionary_start, params->dist.max_distance, &sr2);
+ }
if (sr2.score >= sr.score + cost_diff_lazy) {
/* Ok, let's just write one byte for now and start a match from the
next byte. */
diff --git a/dviware/dvisvgm/libs/brotli/enc/bit_cost.c b/dviware/dvisvgm/libs/brotli/enc/bit_cost.c
index 1f3f7ad5c9..6b7c904ced 100644
--- a/dviware/dvisvgm/libs/brotli/enc/bit_cost.c
+++ b/dviware/dvisvgm/libs/brotli/enc/bit_cost.c
@@ -6,28 +6,29 @@
/* Functions to estimate the bit cost of Huffman trees. */
-#include "./bit_cost.h"
+#include "bit_cost.h"
+
+#include <brotli/types.h>
#include "../common/constants.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./fast_log.h"
-#include "./histogram.h"
+#include "fast_log.h"
+#include "histogram.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
#define FN(X) X ## Literal
-#include "./bit_cost_inc.h" /* NOLINT(build/include) */
+#include "bit_cost_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Command
-#include "./bit_cost_inc.h" /* NOLINT(build/include) */
+#include "bit_cost_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Distance
-#include "./bit_cost_inc.h" /* NOLINT(build/include) */
+#include "bit_cost_inc.h" /* NOLINT(build/include) */
#undef FN
#if defined(__cplusplus) || defined(c_plusplus)
diff --git a/dviware/dvisvgm/libs/brotli/enc/bit_cost.h b/dviware/dvisvgm/libs/brotli/enc/bit_cost.h
index 6586469e62..f6f2773994 100644
--- a/dviware/dvisvgm/libs/brotli/enc/bit_cost.h
+++ b/dviware/dvisvgm/libs/brotli/enc/bit_cost.h
@@ -9,10 +9,11 @@
#ifndef BROTLI_ENC_BIT_COST_H_
#define BROTLI_ENC_BIT_COST_H_
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./fast_log.h"
-#include "./histogram.h"
+
+#include "../common/platform.h"
+#include "fast_log.h"
+#include "histogram.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -45,7 +46,7 @@ static BROTLI_INLINE double BitsEntropy(
const uint32_t* population, size_t size) {
size_t sum;
double retval = ShannonEntropy(population, size, &sum);
- if (retval < sum) {
+ if (retval < (double)sum) {
/* At least one bit per literal is needed. */
retval = (double)sum;
}
diff --git a/dviware/dvisvgm/libs/brotli/enc/block_splitter.c b/dviware/dvisvgm/libs/brotli/enc/block_splitter.c
index deb7c2e151..eba1b691e5 100644
--- a/dviware/dvisvgm/libs/brotli/enc/block_splitter.c
+++ b/dviware/dvisvgm/libs/brotli/enc/block_splitter.c
@@ -6,18 +6,18 @@
/* Block split point selection utilities. */
-#include "./block_splitter.h"
+#include "block_splitter.h"
#include <string.h> /* memcpy, memset */
#include "../common/platform.h"
-#include "./bit_cost.h"
-#include "./cluster.h"
-#include "./command.h"
-#include "./fast_log.h"
-#include "./histogram.h"
-#include "./memory.h"
-#include "./quality.h"
+#include "bit_cost.h"
+#include "cluster.h"
+#include "command.h"
+#include "fast_log.h"
+#include "histogram.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -30,6 +30,7 @@ static const double kCommandBlockSwitchCost = 13.5;
static const double kDistanceBlockSwitchCost = 14.6;
static const size_t kLiteralStrideLength = 70;
static const size_t kCommandStrideLength = 40;
+static const size_t kDistanceStrideLength = 40;
static const size_t kSymbolsPerLiteralHistogram = 544;
static const size_t kSymbolsPerCommandHistogram = 530;
static const size_t kSymbolsPerDistanceHistogram = 544;
@@ -89,19 +90,19 @@ static BROTLI_INLINE double BitCost(size_t count) {
#define FN(X) X ## Literal
#define DataType uint8_t
/* NOLINTNEXTLINE(build/include) */
-#include "./block_splitter_inc.h"
+#include "block_splitter_inc.h"
#undef DataType
#undef FN
#define FN(X) X ## Command
#define DataType uint16_t
/* NOLINTNEXTLINE(build/include) */
-#include "./block_splitter_inc.h"
+#include "block_splitter_inc.h"
#undef FN
#define FN(X) X ## Distance
/* NOLINTNEXTLINE(build/include) */
-#include "./block_splitter_inc.h"
+#include "block_splitter_inc.h"
#undef DataType
#undef FN
@@ -119,6 +120,8 @@ void BrotliDestroyBlockSplit(MemoryManager* m, BlockSplit* self) {
BROTLI_FREE(m, self->lengths);
}
+/* Extracts literals, command distance and prefix codes, then applies
+ * SplitByteVector to create partitioning. */
void BrotliSplitBlock(MemoryManager* m,
const Command* cmds,
const size_t num_commands,
@@ -136,7 +139,9 @@ void BrotliSplitBlock(MemoryManager* m,
/* Create a continuous array of literals. */
CopyLiteralsToByteArray(cmds, num_commands, data, pos, mask, literals);
/* Create the block split on the array of literals.
- Literal histograms have alphabet size 256. */
+ * Literal histograms can have alphabet size up to 256.
+ * Though, to accomodate context modeling, less than half of maximum size
+ * is allowed. */
SplitByteVectorLiteral(
m, literals, literals_count,
kSymbolsPerLiteralHistogram, kMaxLiteralHistograms,
@@ -144,6 +149,10 @@ void BrotliSplitBlock(MemoryManager* m,
literal_split);
if (BROTLI_IS_OOM(m)) return;
BROTLI_FREE(m, literals);
+ /* NB: this might be a good place for injecting extra splitting without
+ * increasing encoder complexity; however, output parition would be less
+ * optimal than one produced with forced splitting inside
+ * SplitByteVector (FindBlocks / ClusterBlocks). */
}
{
@@ -161,7 +170,7 @@ void BrotliSplitBlock(MemoryManager* m,
kCommandStrideLength, kCommandBlockSwitchCost, params,
insert_and_copy_split);
if (BROTLI_IS_OOM(m)) return;
- /* TODO: reuse for distances? */
+ /* TODO(eustas): reuse for distances? */
BROTLI_FREE(m, insert_and_copy_codes);
}
@@ -181,13 +190,27 @@ void BrotliSplitBlock(MemoryManager* m,
SplitByteVectorDistance(
m, distance_prefixes, j,
kSymbolsPerDistanceHistogram, kMaxCommandHistograms,
- kCommandStrideLength, kDistanceBlockSwitchCost, params,
+ kDistanceStrideLength, kDistanceBlockSwitchCost, params,
dist_split);
if (BROTLI_IS_OOM(m)) return;
BROTLI_FREE(m, distance_prefixes);
}
}
+#if defined(BROTLI_TEST)
+size_t CountLiteralsForTest(const Command*, const size_t);
+size_t CountLiteralsForTest(const Command* cmds, const size_t num_commands) {
+ return CountLiterals(cmds, num_commands);
+}
+
+void CopyLiteralsToByteArrayForTest(const Command*,
+ const size_t, const uint8_t*, const size_t, const size_t, uint8_t*);
+void CopyLiteralsToByteArrayForTest(const Command* cmds,
+ const size_t num_commands, const uint8_t* data, const size_t offset,
+ const size_t mask, uint8_t* literals) {
+ CopyLiteralsToByteArray(cmds, num_commands, data, offset, mask, literals);
+}
+#endif
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/block_splitter.h b/dviware/dvisvgm/libs/brotli/enc/block_splitter.h
index a5e006c4b3..6046b90a5d 100644
--- a/dviware/dvisvgm/libs/brotli/enc/block_splitter.h
+++ b/dviware/dvisvgm/libs/brotli/enc/block_splitter.h
@@ -9,11 +9,12 @@
#ifndef BROTLI_ENC_BLOCK_SPLITTER_H_
#define BROTLI_ENC_BLOCK_SPLITTER_H_
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./command.h"
-#include "./memory.h"
-#include "./quality.h"
+
+#include "../common/platform.h"
+#include "command.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/block_splitter_inc.h b/dviware/dvisvgm/libs/brotli/enc/block_splitter_inc.h
index e612d6a370..aa40bfd329 100644
--- a/dviware/dvisvgm/libs/brotli/enc/block_splitter_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/block_splitter_inc.h
@@ -46,17 +46,17 @@ static void FN(RandomSample)(uint32_t* seed,
static void FN(RefineEntropyCodes)(const DataType* data, size_t length,
size_t stride,
size_t num_histograms,
- HistogramType* histograms) {
+ HistogramType* histograms,
+ HistogramType* tmp) {
size_t iters =
kIterMulForRefining * length / stride + kMinItersForRefining;
uint32_t seed = 7;
size_t iter;
iters = ((iters + num_histograms - 1) / num_histograms) * num_histograms;
for (iter = 0; iter < iters; ++iter) {
- HistogramType sample;
- FN(HistogramClear)(&sample);
- FN(RandomSample)(&seed, data, length, stride, &sample);
- FN(HistogramAddHistogram)(&histograms[iter % num_histograms], &sample);
+ FN(HistogramClear)(tmp);
+ FN(RandomSample)(&seed, data, length, stride, tmp);
+ FN(HistogramAddHistogram)(&histograms[iter % num_histograms], tmp);
}
}
@@ -71,46 +71,56 @@ static size_t FN(FindBlocks)(const DataType* data, const size_t length,
double* cost,
uint8_t* switch_signal,
uint8_t* block_id) {
- const size_t data_size = FN(HistogramDataSize)();
- const size_t bitmaplen = (num_histograms + 7) >> 3;
+ const size_t alphabet_size = FN(HistogramDataSize)();
+ const size_t bitmap_len = (num_histograms + 7) >> 3;
size_t num_blocks = 1;
+ size_t byte_ix;
size_t i;
size_t j;
BROTLI_DCHECK(num_histograms <= 256);
+
+ /* Trivial case: single historgram -> single block type. */
if (num_histograms <= 1) {
for (i = 0; i < length; ++i) {
block_id[i] = 0;
}
return 1;
}
- memset(insert_cost, 0, sizeof(insert_cost[0]) * data_size * num_histograms);
+
+ /* Fill bitcost for each symbol of all histograms.
+ * Non-existing symbol cost: 2 + log2(total_count).
+ * Regular symbol cost: -log2(symbol_count / total_count). */
+ memset(insert_cost, 0,
+ sizeof(insert_cost[0]) * alphabet_size * num_histograms);
for (i = 0; i < num_histograms; ++i) {
insert_cost[i] = FastLog2((uint32_t)histograms[i].total_count_);
}
- for (i = data_size; i != 0;) {
+ for (i = alphabet_size; i != 0;) {
+ /* Reverse order to use the 0-th row as a temporary storage. */
--i;
for (j = 0; j < num_histograms; ++j) {
insert_cost[i * num_histograms + j] =
insert_cost[j] - BitCost(histograms[j].data_[i]);
}
}
- memset(cost, 0, sizeof(cost[0]) * num_histograms);
- memset(switch_signal, 0, sizeof(switch_signal[0]) * length * bitmaplen);
+
/* After each iteration of this loop, cost[k] will contain the difference
between the minimum cost of arriving at the current byte position using
entropy code k, and the minimum cost of arriving at the current byte
position. This difference is capped at the block switch cost, and if it
reaches block switch cost, it means that when we trace back from the last
position, we need to switch here. */
- for (i = 0; i < length; ++i) {
- const size_t byte_ix = i;
- size_t ix = byte_ix * bitmaplen;
- size_t insert_cost_ix = data[byte_ix] * num_histograms;
+ memset(cost, 0, sizeof(cost[0]) * num_histograms);
+ memset(switch_signal, 0, sizeof(switch_signal[0]) * length * bitmap_len);
+ for (byte_ix = 0; byte_ix < length; ++byte_ix) {
+ size_t ix = byte_ix * bitmap_len;
+ size_t symbol = data[byte_ix];
+ size_t insert_cost_ix = symbol * num_histograms;
double min_cost = 1e99;
double block_switch_cost = block_switch_bitcost;
size_t k;
for (k = 0; k < num_histograms; ++k) {
- /* We are coding the symbol in data[byte_ix] with entropy code k. */
+ /* We are coding the symbol with entropy code k. */
cost[k] += insert_cost[insert_cost_ix + k];
if (cost[k] < min_cost) {
min_cost = cost[k];
@@ -126,20 +136,21 @@ static size_t FN(FindBlocks)(const DataType* data, const size_t length,
if (cost[k] >= block_switch_cost) {
const uint8_t mask = (uint8_t)(1u << (k & 7));
cost[k] = block_switch_cost;
- BROTLI_DCHECK((k >> 3) < bitmaplen);
+ BROTLI_DCHECK((k >> 3) < bitmap_len);
switch_signal[ix + (k >> 3)] |= mask;
}
}
}
+
+ byte_ix = length - 1;
{ /* Trace back from the last position and switch at the marked places. */
- size_t byte_ix = length - 1;
- size_t ix = byte_ix * bitmaplen;
+ size_t ix = byte_ix * bitmap_len;
uint8_t cur_id = block_id[byte_ix];
while (byte_ix > 0) {
const uint8_t mask = (uint8_t)(1u << (cur_id & 7));
- BROTLI_DCHECK(((size_t)cur_id >> 3) < bitmaplen);
+ BROTLI_DCHECK(((size_t)cur_id >> 3) < bitmap_len);
--byte_ix;
- ix -= bitmaplen;
+ ix -= bitmap_len;
if (switch_signal[ix + (cur_id >> 3)] & mask) {
if (cur_id != block_id[byte_ix]) {
cur_id = block_id[byte_ix];
@@ -185,13 +196,16 @@ static void FN(BuildBlockHistograms)(const DataType* data, const size_t length,
}
}
+/* Given the initial partitioning build partitioning with limited number
+ * of histograms (and block types). */
static void FN(ClusterBlocks)(MemoryManager* m,
const DataType* data, const size_t length,
const size_t num_blocks,
uint8_t* block_ids,
BlockSplit* split) {
uint32_t* histogram_symbols = BROTLI_ALLOC(m, uint32_t, num_blocks);
- uint32_t* block_lengths = BROTLI_ALLOC(m, uint32_t, num_blocks);
+ uint32_t* u32 =
+ BROTLI_ALLOC(m, uint32_t, num_blocks + 4 * HISTOGRAMS_PER_BATCH);
const size_t expected_num_clusters = CLUSTERS_PER_BATCH *
(num_blocks + HISTOGRAMS_PER_BATCH - 1) / HISTOGRAMS_PER_BATCH;
size_t all_histograms_size = 0;
@@ -214,20 +228,25 @@ static void FN(ClusterBlocks)(MemoryManager* m,
static const uint32_t kInvalidIndex = BROTLI_UINT32_MAX;
uint32_t* new_index;
size_t i;
- uint32_t sizes[HISTOGRAMS_PER_BATCH] = { 0 };
- uint32_t new_clusters[HISTOGRAMS_PER_BATCH] = { 0 };
- uint32_t symbols[HISTOGRAMS_PER_BATCH] = { 0 };
- uint32_t remap[HISTOGRAMS_PER_BATCH] = { 0 };
+ uint32_t* BROTLI_RESTRICT const sizes = u32 + 0 * HISTOGRAMS_PER_BATCH;
+ uint32_t* BROTLI_RESTRICT const new_clusters = u32 + 1 * HISTOGRAMS_PER_BATCH;
+ uint32_t* BROTLI_RESTRICT const symbols = u32 + 2 * HISTOGRAMS_PER_BATCH;
+ uint32_t* BROTLI_RESTRICT const remap = u32 + 3 * HISTOGRAMS_PER_BATCH;
+ uint32_t* BROTLI_RESTRICT const block_lengths =
+ u32 + 4 * HISTOGRAMS_PER_BATCH;
+ /* TODO(eustas): move to arena? */
+ HistogramType* tmp = BROTLI_ALLOC(m, HistogramType, 2);
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(histogram_symbols) ||
- BROTLI_IS_NULL(block_lengths) || BROTLI_IS_NULL(all_histograms) ||
+ BROTLI_IS_NULL(u32) || BROTLI_IS_NULL(all_histograms) ||
BROTLI_IS_NULL(cluster_size) || BROTLI_IS_NULL(histograms) ||
- BROTLI_IS_NULL(pairs)) {
+ BROTLI_IS_NULL(pairs) || BROTLI_IS_NULL(tmp)) {
return;
}
- memset(block_lengths, 0, num_blocks * sizeof(uint32_t));
+ memset(u32, 0, (num_blocks + 4 * HISTOGRAMS_PER_BATCH) * sizeof(uint32_t));
+ /* Calculate block lengths (convert repeating values -> series length). */
{
size_t block_idx = 0;
for (i = 0; i < length; ++i) {
@@ -240,6 +259,7 @@ static void FN(ClusterBlocks)(MemoryManager* m,
BROTLI_DCHECK(block_idx == num_blocks);
}
+ /* Pre-cluster blocks (cluster batches). */
for (i = 0; i < num_blocks; i += HISTOGRAMS_PER_BATCH) {
const size_t num_to_combine =
BROTLI_MIN(size_t, num_blocks - i, HISTOGRAMS_PER_BATCH);
@@ -247,8 +267,9 @@ static void FN(ClusterBlocks)(MemoryManager* m,
size_t j;
for (j = 0; j < num_to_combine; ++j) {
size_t k;
+ size_t block_length = block_lengths[i + j];
FN(HistogramClear)(&histograms[j]);
- for (k = 0; k < block_lengths[i + j]; ++k) {
+ for (k = 0; k < block_length; ++k) {
FN(HistogramAdd)(&histograms[j], data[pos++]);
}
histograms[j].bit_cost_ = FN(BrotliPopulationCost)(&histograms[j]);
@@ -257,7 +278,7 @@ static void FN(ClusterBlocks)(MemoryManager* m,
sizes[j] = 1;
}
num_new_clusters = FN(BrotliHistogramCombine)(
- histograms, sizes, symbols, new_clusters, pairs, num_to_combine,
+ histograms, tmp, sizes, symbols, new_clusters, pairs, num_to_combine,
num_to_combine, HISTOGRAMS_PER_BATCH, max_num_pairs);
BROTLI_ENSURE_CAPACITY(m, HistogramType, all_histograms,
all_histograms_capacity, all_histograms_size + num_new_clusters);
@@ -278,6 +299,7 @@ static void FN(ClusterBlocks)(MemoryManager* m,
}
BROTLI_FREE(m, histograms);
+ /* Final clustering. */
max_num_pairs =
BROTLI_MIN(size_t, 64 * num_clusters, (num_clusters / 2) * num_clusters);
if (pairs_capacity < max_num_pairs + 1) {
@@ -285,19 +307,19 @@ static void FN(ClusterBlocks)(MemoryManager* m,
pairs = BROTLI_ALLOC(m, HistogramPair, max_num_pairs + 1);
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(pairs)) return;
}
-
clusters = BROTLI_ALLOC(m, uint32_t, num_clusters);
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(clusters)) return;
for (i = 0; i < num_clusters; ++i) {
clusters[i] = (uint32_t)i;
}
num_final_clusters = FN(BrotliHistogramCombine)(
- all_histograms, cluster_size, histogram_symbols, clusters, pairs,
+ all_histograms, tmp, cluster_size, histogram_symbols, clusters, pairs,
num_clusters, num_blocks, BROTLI_MAX_NUMBER_OF_BLOCK_TYPES,
max_num_pairs);
BROTLI_FREE(m, pairs);
BROTLI_FREE(m, cluster_size);
+ /* Assign blocks to final histograms. */
new_index = BROTLI_ALLOC(m, uint32_t, num_clusters);
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(new_index)) return;
for (i = 0; i < num_clusters; ++i) new_index[i] = kInvalidIndex;
@@ -305,20 +327,21 @@ static void FN(ClusterBlocks)(MemoryManager* m,
{
uint32_t next_index = 0;
for (i = 0; i < num_blocks; ++i) {
- HistogramType histo;
size_t j;
uint32_t best_out;
double best_bits;
- FN(HistogramClear)(&histo);
+ FN(HistogramClear)(tmp);
for (j = 0; j < block_lengths[i]; ++j) {
- FN(HistogramAdd)(&histo, data[pos++]);
+ FN(HistogramAdd)(tmp, data[pos++]);
}
+ /* Among equally good histograms prefer last used. */
+ /* TODO(eustas): should we give a block-switch discount here? */
best_out = (i == 0) ? histogram_symbols[0] : histogram_symbols[i - 1];
- best_bits =
- FN(BrotliHistogramBitCostDistance)(&histo, &all_histograms[best_out]);
+ best_bits = FN(BrotliHistogramBitCostDistance)(
+ tmp, &all_histograms[best_out], tmp + 1);
for (j = 0; j < num_final_clusters; ++j) {
const double cur_bits = FN(BrotliHistogramBitCostDistance)(
- &histo, &all_histograms[clusters[j]]);
+ tmp, &all_histograms[clusters[j]], tmp + 1);
if (cur_bits < best_bits) {
best_bits = cur_bits;
best_out = clusters[j];
@@ -330,6 +353,7 @@ static void FN(ClusterBlocks)(MemoryManager* m,
}
}
}
+ BROTLI_FREE(m, tmp);
BROTLI_FREE(m, clusters);
BROTLI_FREE(m, all_histograms);
BROTLI_ENSURE_CAPACITY(
@@ -337,6 +361,9 @@ static void FN(ClusterBlocks)(MemoryManager* m,
BROTLI_ENSURE_CAPACITY(
m, uint32_t, split->lengths, split->lengths_alloc_size, num_blocks);
if (BROTLI_IS_OOM(m)) return;
+
+ /* Rewrite final assignment to block-split. There might be less blocks
+ * than |num_blocks| due to clustering. */
{
uint32_t cur_length = 0;
size_t block_idx = 0;
@@ -357,28 +384,41 @@ static void FN(ClusterBlocks)(MemoryManager* m,
split->num_types = (size_t)max_type + 1;
}
BROTLI_FREE(m, new_index);
- BROTLI_FREE(m, block_lengths);
+ BROTLI_FREE(m, u32);
BROTLI_FREE(m, histogram_symbols);
}
+/* Create BlockSplit (partitioning) given the limits, estimates and "effort"
+ * parameters.
+ *
+ * NB: max_histograms is often less than number of histograms allowed by format;
+ * this is done intentionally, to save some "space" for context-aware
+ * clustering (here entropy is estimated for context-free symbols). */
static void FN(SplitByteVector)(MemoryManager* m,
const DataType* data, const size_t length,
- const size_t literals_per_histogram,
+ const size_t symbols_per_histogram,
const size_t max_histograms,
const size_t sampling_stride_length,
const double block_switch_cost,
const BrotliEncoderParams* params,
BlockSplit* split) {
const size_t data_size = FN(HistogramDataSize)();
- size_t num_histograms = length / literals_per_histogram + 1;
HistogramType* histograms;
+ HistogramType* tmp;
+ /* Calculate number of histograms; initial estimate is one histogram per
+ * specified amount of symbols; however, this value is capped. */
+ size_t num_histograms = length / symbols_per_histogram + 1;
if (num_histograms > max_histograms) {
num_histograms = max_histograms;
}
+
+ /* Corner case: no input. */
if (length == 0) {
split->num_types = 1;
return;
- } else if (length < kMinLengthForBlockSplitting) {
+ }
+
+ if (length < kMinLengthForBlockSplitting) {
BROTLI_ENSURE_CAPACITY(m, uint8_t,
split->types, split->types_alloc_size, split->num_blocks + 1);
BROTLI_ENSURE_CAPACITY(m, uint32_t,
@@ -390,7 +430,8 @@ static void FN(SplitByteVector)(MemoryManager* m,
split->num_blocks++;
return;
}
- histograms = BROTLI_ALLOC(m, HistogramType, num_histograms);
+ histograms = BROTLI_ALLOC(m, HistogramType, num_histograms + 1);
+ tmp = histograms + num_histograms;
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(histograms)) return;
/* Find good entropy codes. */
FN(InitialEntropyCodes)(data, length,
@@ -398,7 +439,7 @@ static void FN(SplitByteVector)(MemoryManager* m,
num_histograms, histograms);
FN(RefineEntropyCodes)(data, length,
sampling_stride_length,
- num_histograms, histograms);
+ num_histograms, histograms, tmp);
{
/* Find a good path through literals with the good entropy codes. */
uint8_t* block_ids = BROTLI_ALLOC(m, uint8_t, length);
diff --git a/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.c b/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.c
index 9348a97e1b..5fa0c69aa8 100644
--- a/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.c
+++ b/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.c
@@ -8,20 +8,21 @@
compression algorithms here, just the right ordering of bits to match the
specs. */
-#include "./brotli_bit_stream.h"
+#include "brotli_bit_stream.h"
#include <string.h> /* memcpy, memset */
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/context.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./entropy_encode.h"
-#include "./entropy_encode_static.h"
-#include "./fast_log.h"
-#include "./histogram.h"
-#include "./memory.h"
-#include "./write_bits.h"
+#include "entropy_encode.h"
+#include "entropy_encode_static.h"
+#include "fast_log.h"
+#include "histogram.h"
+#include "memory.h"
+#include "write_bits.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -286,6 +287,7 @@ void BrotliStoreHuffmanTree(const uint8_t* depths, size_t num,
/* Write the Huffman tree into the brotli-representation.
The command alphabet is the largest, so this allocation will fit all
alphabets. */
+ /* TODO(eustas): fix me */
uint8_t huffman_tree[BROTLI_NUM_COMMAND_SYMBOLS];
uint8_t huffman_tree_extra_bits[BROTLI_NUM_COMMAND_SYMBOLS];
size_t huffman_tree_size = 0;
@@ -400,7 +402,7 @@ static BROTLI_INLINE BROTLI_BOOL SortHuffmanTree(
return TO_BROTLI_BOOL(v0->total_count_ < v1->total_count_);
}
-void BrotliBuildAndStoreHuffmanTreeFast(MemoryManager* m,
+void BrotliBuildAndStoreHuffmanTreeFast(HuffmanTree* tree,
const uint32_t* histogram,
const size_t histogram_total,
const size_t max_bits,
@@ -432,10 +434,7 @@ void BrotliBuildAndStoreHuffmanTreeFast(MemoryManager* m,
memset(depth, 0, length * sizeof(depth[0]));
{
- const size_t max_tree_size = 2 * length + 1;
- HuffmanTree* tree = BROTLI_ALLOC(m, HuffmanTree, max_tree_size);
uint32_t count_limit;
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tree)) return;
for (count_limit = 1; ; count_limit *= 2) {
HuffmanTree* node = tree;
size_t l;
@@ -500,7 +499,6 @@ void BrotliBuildAndStoreHuffmanTreeFast(MemoryManager* m,
}
}
}
- BROTLI_FREE(m, tree);
}
BrotliConvertBitDepthsToSymbols(depth, length, bits);
if (count <= 4) {
@@ -677,7 +675,14 @@ static void RunLengthCodeZeros(const size_t in_size,
#define SYMBOL_BITS 9
+typedef struct EncodeContextMapArena {
+ uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
+ uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
+ uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
+} EncodeContextMapArena;
+
static void EncodeContextMap(MemoryManager* m,
+ EncodeContextMapArena* arena,
const uint32_t* context_map,
size_t context_map_size,
size_t num_clusters,
@@ -687,10 +692,10 @@ static void EncodeContextMap(MemoryManager* m,
uint32_t* rle_symbols;
uint32_t max_run_length_prefix = 6;
size_t num_rle_symbols = 0;
- uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
+ uint32_t* BROTLI_RESTRICT const histogram = arena->histogram;
static const uint32_t kSymbolMask = (1u << SYMBOL_BITS) - 1u;
- uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
- uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
+ uint8_t* BROTLI_RESTRICT const depths = arena->depths;
+ uint16_t* BROTLI_RESTRICT const bits = arena->bits;
StoreVarLenUint8(num_clusters - 1, storage_ix, storage);
@@ -703,7 +708,7 @@ static void EncodeContextMap(MemoryManager* m,
MoveToFrontTransform(context_map, context_map_size, rle_symbols);
RunLengthCodeZeros(context_map_size, rle_symbols,
&num_rle_symbols, &max_run_length_prefix);
- memset(histogram, 0, sizeof(histogram));
+ memset(histogram, 0, sizeof(arena->histogram));
for (i = 0; i < num_rle_symbols; ++i) {
++histogram[rle_symbols[i] & kSymbolMask];
}
@@ -774,7 +779,7 @@ static void BuildAndStoreBlockSplitCode(const uint8_t* types,
++length_histo[BlockLengthPrefixCode(lengths[i])];
}
StoreVarLenUint8(num_types - 1, storage_ix, storage);
- if (num_types > 1) { /* TODO: else? could StoreBlockSwitch occur? */
+ if (num_types > 1) { /* TODO(eustas): else? could StoreBlockSwitch occur? */
BuildAndStoreHuffmanTree(&type_histo[0], num_types + 2, num_types + 2, tree,
&code->type_depths[0], &code->type_bits[0],
storage_ix, storage);
@@ -787,7 +792,8 @@ static void BuildAndStoreBlockSplitCode(const uint8_t* types,
}
/* Stores a context map where the histogram type is always the block type. */
-static void StoreTrivialContextMap(size_t num_types,
+static void StoreTrivialContextMap(EncodeContextMapArena* arena,
+ size_t num_types,
size_t context_bits,
HuffmanTree* tree,
size_t* storage_ix,
@@ -797,9 +803,9 @@ static void StoreTrivialContextMap(size_t num_types,
size_t repeat_code = context_bits - 1u;
size_t repeat_bits = (1u << repeat_code) - 1u;
size_t alphabet_size = num_types + repeat_code;
- uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
- uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
- uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS];
+ uint32_t* BROTLI_RESTRICT const histogram = arena->histogram;
+ uint8_t* BROTLI_RESTRICT const depths = arena->depths;
+ uint16_t* BROTLI_RESTRICT const bits = arena->bits;
size_t i;
memset(histogram, 0, alphabet_size * sizeof(histogram[0]));
/* Write RLEMAX. */
@@ -914,17 +920,17 @@ static void StoreSymbolWithContext(BlockEncoder* self, size_t symbol,
#define FN(X) X ## Literal
/* NOLINTNEXTLINE(build/include) */
-#include "./block_encoder_inc.h"
+#include "block_encoder_inc.h"
#undef FN
#define FN(X) X ## Command
/* NOLINTNEXTLINE(build/include) */
-#include "./block_encoder_inc.h"
+#include "block_encoder_inc.h"
#undef FN
#define FN(X) X ## Distance
/* NOLINTNEXTLINE(build/include) */
-#include "./block_encoder_inc.h"
+#include "block_encoder_inc.h"
#undef FN
static void JumpToByteBoundary(size_t* storage_ix, uint8_t* storage) {
@@ -932,6 +938,13 @@ static void JumpToByteBoundary(size_t* storage_ix, uint8_t* storage) {
storage[*storage_ix >> 3] = 0;
}
+typedef struct StoreMetablockArena {
+ BlockEncoder literal_enc;
+ BlockEncoder command_enc;
+ BlockEncoder distance_enc;
+ EncodeContextMapArena context_map_arena;
+} StoreMetablockArena;
+
void BrotliStoreMetaBlock(MemoryManager* m,
const uint8_t* input, size_t start_pos, size_t length, size_t mask,
uint8_t prev_byte, uint8_t prev_byte2, BROTLI_BOOL is_last,
@@ -945,9 +958,10 @@ void BrotliStoreMetaBlock(MemoryManager* m,
uint32_t num_effective_distance_symbols = params->dist.alphabet_size_limit;
HuffmanTree* tree;
ContextLut literal_context_lut = BROTLI_CONTEXT_LUT(literal_context_mode);
- BlockEncoder literal_enc;
- BlockEncoder command_enc;
- BlockEncoder distance_enc;
+ StoreMetablockArena* arena = NULL;
+ BlockEncoder* literal_enc = NULL;
+ BlockEncoder* command_enc = NULL;
+ BlockEncoder* distance_enc = NULL;
const BrotliDistanceParams* dist = &params->dist;
BROTLI_DCHECK(
num_effective_distance_symbols <= BROTLI_NUM_HISTOGRAM_DISTANCE_SYMBOLS);
@@ -955,21 +969,24 @@ void BrotliStoreMetaBlock(MemoryManager* m,
StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage);
tree = BROTLI_ALLOC(m, HuffmanTree, MAX_HUFFMAN_TREE_SIZE);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tree)) return;
- InitBlockEncoder(&literal_enc, BROTLI_NUM_LITERAL_SYMBOLS,
+ arena = BROTLI_ALLOC(m, StoreMetablockArena, 1);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tree) || BROTLI_IS_NULL(arena)) return;
+ literal_enc = &arena->literal_enc;
+ command_enc = &arena->command_enc;
+ distance_enc = &arena->distance_enc;
+ InitBlockEncoder(literal_enc, BROTLI_NUM_LITERAL_SYMBOLS,
mb->literal_split.num_types, mb->literal_split.types,
mb->literal_split.lengths, mb->literal_split.num_blocks);
- InitBlockEncoder(&command_enc, BROTLI_NUM_COMMAND_SYMBOLS,
+ InitBlockEncoder(command_enc, BROTLI_NUM_COMMAND_SYMBOLS,
mb->command_split.num_types, mb->command_split.types,
mb->command_split.lengths, mb->command_split.num_blocks);
- InitBlockEncoder(&distance_enc, num_effective_distance_symbols,
+ InitBlockEncoder(distance_enc, num_effective_distance_symbols,
mb->distance_split.num_types, mb->distance_split.types,
mb->distance_split.lengths, mb->distance_split.num_blocks);
- BuildAndStoreBlockSwitchEntropyCodes(&literal_enc, tree, storage_ix, storage);
- BuildAndStoreBlockSwitchEntropyCodes(&command_enc, tree, storage_ix, storage);
- BuildAndStoreBlockSwitchEntropyCodes(
- &distance_enc, tree, storage_ix, storage);
+ BuildAndStoreBlockSwitchEntropyCodes(literal_enc, tree, storage_ix, storage);
+ BuildAndStoreBlockSwitchEntropyCodes(command_enc, tree, storage_ix, storage);
+ BuildAndStoreBlockSwitchEntropyCodes(distance_enc, tree, storage_ix, storage);
BrotliWriteBits(2, dist->distance_postfix_bits, storage_ix, storage);
BrotliWriteBits(
@@ -980,34 +997,36 @@ void BrotliStoreMetaBlock(MemoryManager* m,
}
if (mb->literal_context_map_size == 0) {
- StoreTrivialContextMap(mb->literal_histograms_size,
+ StoreTrivialContextMap(
+ &arena->context_map_arena, mb->literal_histograms_size,
BROTLI_LITERAL_CONTEXT_BITS, tree, storage_ix, storage);
} else {
- EncodeContextMap(m,
+ EncodeContextMap(m, &arena->context_map_arena,
mb->literal_context_map, mb->literal_context_map_size,
mb->literal_histograms_size, tree, storage_ix, storage);
if (BROTLI_IS_OOM(m)) return;
}
if (mb->distance_context_map_size == 0) {
- StoreTrivialContextMap(mb->distance_histograms_size,
+ StoreTrivialContextMap(
+ &arena->context_map_arena, mb->distance_histograms_size,
BROTLI_DISTANCE_CONTEXT_BITS, tree, storage_ix, storage);
} else {
- EncodeContextMap(m,
+ EncodeContextMap(m, &arena->context_map_arena,
mb->distance_context_map, mb->distance_context_map_size,
mb->distance_histograms_size, tree, storage_ix, storage);
if (BROTLI_IS_OOM(m)) return;
}
- BuildAndStoreEntropyCodesLiteral(m, &literal_enc, mb->literal_histograms,
+ BuildAndStoreEntropyCodesLiteral(m, literal_enc, mb->literal_histograms,
mb->literal_histograms_size, BROTLI_NUM_LITERAL_SYMBOLS, tree,
storage_ix, storage);
if (BROTLI_IS_OOM(m)) return;
- BuildAndStoreEntropyCodesCommand(m, &command_enc, mb->command_histograms,
+ BuildAndStoreEntropyCodesCommand(m, command_enc, mb->command_histograms,
mb->command_histograms_size, BROTLI_NUM_COMMAND_SYMBOLS, tree,
storage_ix, storage);
if (BROTLI_IS_OOM(m)) return;
- BuildAndStoreEntropyCodesDistance(m, &distance_enc, mb->distance_histograms,
+ BuildAndStoreEntropyCodesDistance(m, distance_enc, mb->distance_histograms,
mb->distance_histograms_size, num_distance_symbols, tree,
storage_ix, storage);
if (BROTLI_IS_OOM(m)) return;
@@ -1016,12 +1035,12 @@ void BrotliStoreMetaBlock(MemoryManager* m,
for (i = 0; i < n_commands; ++i) {
const Command cmd = commands[i];
size_t cmd_code = cmd.cmd_prefix_;
- StoreSymbol(&command_enc, cmd_code, storage_ix, storage);
+ StoreSymbol(command_enc, cmd_code, storage_ix, storage);
StoreCommandExtra(&cmd, storage_ix, storage);
if (mb->literal_context_map_size == 0) {
size_t j;
for (j = cmd.insert_len_; j != 0; --j) {
- StoreSymbol(&literal_enc, input[pos & mask], storage_ix, storage);
+ StoreSymbol(literal_enc, input[pos & mask], storage_ix, storage);
++pos;
}
} else {
@@ -1030,7 +1049,7 @@ void BrotliStoreMetaBlock(MemoryManager* m,
size_t context =
BROTLI_CONTEXT(prev_byte, prev_byte2, literal_context_lut);
uint8_t literal = input[pos & mask];
- StoreSymbolWithContext(&literal_enc, literal, context,
+ StoreSymbolWithContext(literal_enc, literal, context,
mb->literal_context_map, storage_ix, storage,
BROTLI_LITERAL_CONTEXT_BITS);
prev_byte2 = prev_byte;
@@ -1047,10 +1066,10 @@ void BrotliStoreMetaBlock(MemoryManager* m,
uint32_t distnumextra = cmd.dist_prefix_ >> 10;
uint64_t distextra = cmd.dist_extra_;
if (mb->distance_context_map_size == 0) {
- StoreSymbol(&distance_enc, dist_code, storage_ix, storage);
+ StoreSymbol(distance_enc, dist_code, storage_ix, storage);
} else {
size_t context = CommandDistanceContext(&cmd);
- StoreSymbolWithContext(&distance_enc, dist_code, context,
+ StoreSymbolWithContext(distance_enc, dist_code, context,
mb->distance_context_map, storage_ix, storage,
BROTLI_DISTANCE_CONTEXT_BITS);
}
@@ -1058,9 +1077,10 @@ void BrotliStoreMetaBlock(MemoryManager* m,
}
}
}
- CleanupBlockEncoder(m, &distance_enc);
- CleanupBlockEncoder(m, &command_enc);
- CleanupBlockEncoder(m, &literal_enc);
+ CleanupBlockEncoder(m, distance_enc);
+ CleanupBlockEncoder(m, command_enc);
+ CleanupBlockEncoder(m, literal_enc);
+ BROTLI_FREE(m, arena);
if (is_last) {
JumpToByteBoundary(storage_ix, storage);
}
@@ -1131,54 +1151,60 @@ static void StoreDataWithHuffmanCodes(const uint8_t* input,
}
}
-void BrotliStoreMetaBlockTrivial(MemoryManager* m,
- const uint8_t* input, size_t start_pos, size_t length, size_t mask,
- BROTLI_BOOL is_last, const BrotliEncoderParams* params,
- const Command* commands, size_t n_commands,
- size_t* storage_ix, uint8_t* storage) {
+/* TODO(eustas): pull alloc/dealloc to caller? */
+typedef struct MetablockArena {
HistogramLiteral lit_histo;
HistogramCommand cmd_histo;
HistogramDistance dist_histo;
+ /* TODO(eustas): merge bits and depth? */
uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS];
uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS];
uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS];
uint16_t cmd_bits[BROTLI_NUM_COMMAND_SYMBOLS];
uint8_t dist_depth[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE];
uint16_t dist_bits[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE];
- HuffmanTree* tree;
+ HuffmanTree tree[MAX_HUFFMAN_TREE_SIZE];
+} MetablockArena;
+
+void BrotliStoreMetaBlockTrivial(MemoryManager* m,
+ const uint8_t* input, size_t start_pos, size_t length, size_t mask,
+ BROTLI_BOOL is_last, const BrotliEncoderParams* params,
+ const Command* commands, size_t n_commands,
+ size_t* storage_ix, uint8_t* storage) {
+ MetablockArena* arena = BROTLI_ALLOC(m, MetablockArena, 1);
uint32_t num_distance_symbols = params->dist.alphabet_size_max;
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(arena)) return;
StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage);
- HistogramClearLiteral(&lit_histo);
- HistogramClearCommand(&cmd_histo);
- HistogramClearDistance(&dist_histo);
+ HistogramClearLiteral(&arena->lit_histo);
+ HistogramClearCommand(&arena->cmd_histo);
+ HistogramClearDistance(&arena->dist_histo);
BuildHistograms(input, start_pos, mask, commands, n_commands,
- &lit_histo, &cmd_histo, &dist_histo);
+ &arena->lit_histo, &arena->cmd_histo, &arena->dist_histo);
BrotliWriteBits(13, 0, storage_ix, storage);
- tree = BROTLI_ALLOC(m, HuffmanTree, MAX_HUFFMAN_TREE_SIZE);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tree)) return;
- BuildAndStoreHuffmanTree(lit_histo.data_, BROTLI_NUM_LITERAL_SYMBOLS,
- BROTLI_NUM_LITERAL_SYMBOLS, tree,
- lit_depth, lit_bits,
+ BuildAndStoreHuffmanTree(arena->lit_histo.data_, BROTLI_NUM_LITERAL_SYMBOLS,
+ BROTLI_NUM_LITERAL_SYMBOLS, arena->tree,
+ arena->lit_depth, arena->lit_bits,
storage_ix, storage);
- BuildAndStoreHuffmanTree(cmd_histo.data_, BROTLI_NUM_COMMAND_SYMBOLS,
- BROTLI_NUM_COMMAND_SYMBOLS, tree,
- cmd_depth, cmd_bits,
+ BuildAndStoreHuffmanTree(arena->cmd_histo.data_, BROTLI_NUM_COMMAND_SYMBOLS,
+ BROTLI_NUM_COMMAND_SYMBOLS, arena->tree,
+ arena->cmd_depth, arena->cmd_bits,
storage_ix, storage);
- BuildAndStoreHuffmanTree(dist_histo.data_, MAX_SIMPLE_DISTANCE_ALPHABET_SIZE,
- num_distance_symbols, tree,
- dist_depth, dist_bits,
+ BuildAndStoreHuffmanTree(arena->dist_histo.data_,
+ MAX_SIMPLE_DISTANCE_ALPHABET_SIZE,
+ num_distance_symbols, arena->tree,
+ arena->dist_depth, arena->dist_bits,
storage_ix, storage);
- BROTLI_FREE(m, tree);
StoreDataWithHuffmanCodes(input, start_pos, mask, commands,
- n_commands, lit_depth, lit_bits,
- cmd_depth, cmd_bits,
- dist_depth, dist_bits,
+ n_commands, arena->lit_depth, arena->lit_bits,
+ arena->cmd_depth, arena->cmd_bits,
+ arena->dist_depth, arena->dist_bits,
storage_ix, storage);
+ BROTLI_FREE(m, arena);
if (is_last) {
JumpToByteBoundary(storage_ix, storage);
}
@@ -1189,9 +1215,11 @@ void BrotliStoreMetaBlockFast(MemoryManager* m,
BROTLI_BOOL is_last, const BrotliEncoderParams* params,
const Command* commands, size_t n_commands,
size_t* storage_ix, uint8_t* storage) {
+ MetablockArena* arena = BROTLI_ALLOC(m, MetablockArena, 1);
uint32_t num_distance_symbols = params->dist.alphabet_size_max;
uint32_t distance_alphabet_bits =
Log2FloorNonZero(num_distance_symbols - 1) + 1;
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(arena)) return;
StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage);
@@ -1202,8 +1230,6 @@ void BrotliStoreMetaBlockFast(MemoryManager* m,
size_t pos = start_pos;
size_t num_literals = 0;
size_t i;
- uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS];
- uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS];
for (i = 0; i < n_commands; ++i) {
const Command cmd = commands[i];
size_t j;
@@ -1214,61 +1240,50 @@ void BrotliStoreMetaBlockFast(MemoryManager* m,
num_literals += cmd.insert_len_;
pos += CommandCopyLen(&cmd);
}
- BrotliBuildAndStoreHuffmanTreeFast(m, histogram, num_literals,
+ BrotliBuildAndStoreHuffmanTreeFast(arena->tree, histogram, num_literals,
/* max_bits = */ 8,
- lit_depth, lit_bits,
+ arena->lit_depth, arena->lit_bits,
storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
StoreStaticCommandHuffmanTree(storage_ix, storage);
StoreStaticDistanceHuffmanTree(storage_ix, storage);
StoreDataWithHuffmanCodes(input, start_pos, mask, commands,
- n_commands, lit_depth, lit_bits,
+ n_commands, arena->lit_depth, arena->lit_bits,
kStaticCommandCodeDepth,
kStaticCommandCodeBits,
kStaticDistanceCodeDepth,
kStaticDistanceCodeBits,
storage_ix, storage);
} else {
- HistogramLiteral lit_histo;
- HistogramCommand cmd_histo;
- HistogramDistance dist_histo;
- uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS];
- uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS];
- uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS];
- uint16_t cmd_bits[BROTLI_NUM_COMMAND_SYMBOLS];
- uint8_t dist_depth[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE];
- uint16_t dist_bits[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE];
- HistogramClearLiteral(&lit_histo);
- HistogramClearCommand(&cmd_histo);
- HistogramClearDistance(&dist_histo);
+ HistogramClearLiteral(&arena->lit_histo);
+ HistogramClearCommand(&arena->cmd_histo);
+ HistogramClearDistance(&arena->dist_histo);
BuildHistograms(input, start_pos, mask, commands, n_commands,
- &lit_histo, &cmd_histo, &dist_histo);
- BrotliBuildAndStoreHuffmanTreeFast(m, lit_histo.data_,
- lit_histo.total_count_,
+ &arena->lit_histo, &arena->cmd_histo, &arena->dist_histo);
+ BrotliBuildAndStoreHuffmanTreeFast(arena->tree, arena->lit_histo.data_,
+ arena->lit_histo.total_count_,
/* max_bits = */ 8,
- lit_depth, lit_bits,
+ arena->lit_depth, arena->lit_bits,
storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
- BrotliBuildAndStoreHuffmanTreeFast(m, cmd_histo.data_,
- cmd_histo.total_count_,
+ BrotliBuildAndStoreHuffmanTreeFast(arena->tree, arena->cmd_histo.data_,
+ arena->cmd_histo.total_count_,
/* max_bits = */ 10,
- cmd_depth, cmd_bits,
+ arena->cmd_depth, arena->cmd_bits,
storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
- BrotliBuildAndStoreHuffmanTreeFast(m, dist_histo.data_,
- dist_histo.total_count_,
+ BrotliBuildAndStoreHuffmanTreeFast(arena->tree, arena->dist_histo.data_,
+ arena->dist_histo.total_count_,
/* max_bits = */
distance_alphabet_bits,
- dist_depth, dist_bits,
+ arena->dist_depth, arena->dist_bits,
storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
StoreDataWithHuffmanCodes(input, start_pos, mask, commands,
- n_commands, lit_depth, lit_bits,
- cmd_depth, cmd_bits,
- dist_depth, dist_bits,
+ n_commands, arena->lit_depth, arena->lit_bits,
+ arena->cmd_depth, arena->cmd_bits,
+ arena->dist_depth, arena->dist_bits,
storage_ix, storage);
}
+ BROTLI_FREE(m, arena);
+
if (is_last) {
JumpToByteBoundary(storage_ix, storage);
}
@@ -1309,6 +1324,13 @@ void BrotliStoreUncompressedMetaBlock(BROTLI_BOOL is_final_block,
}
}
+#if defined(BROTLI_TEST)
+void GetBlockLengthPrefixCodeForTest(uint32_t len, size_t* code,
+ uint32_t* n_extra, uint32_t* extra) {
+ GetBlockLengthPrefixCode(len, code, n_extra, extra);
+}
+#endif
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.h b/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.h
index 2ed703bf79..a289509af3 100644
--- a/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.h
+++ b/dviware/dvisvgm/libs/brotli/enc/brotli_bit_stream.h
@@ -16,13 +16,14 @@
#ifndef BROTLI_ENC_BROTLI_BIT_STREAM_H_
#define BROTLI_ENC_BROTLI_BIT_STREAM_H_
+#include <brotli/types.h>
+
#include "../common/context.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./command.h"
-#include "./entropy_encode.h"
-#include "./memory.h"
-#include "./metablock.h"
+#include "command.h"
+#include "entropy_encode.h"
+#include "memory.h"
+#include "metablock.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -35,7 +36,7 @@ BROTLI_INTERNAL void BrotliStoreHuffmanTree(const uint8_t* depths, size_t num,
HuffmanTree* tree, size_t* storage_ix, uint8_t* storage);
BROTLI_INTERNAL void BrotliBuildAndStoreHuffmanTreeFast(
- MemoryManager* m, const uint32_t* histogram, const size_t histogram_total,
+ HuffmanTree* tree, const uint32_t* histogram, const size_t histogram_total,
const size_t max_bits, uint8_t* depth, uint16_t* bits, size_t* storage_ix,
uint8_t* storage);
@@ -77,6 +78,10 @@ BROTLI_INTERNAL void BrotliStoreUncompressedMetaBlock(
size_t position, size_t mask, size_t len,
size_t* BROTLI_RESTRICT storage_ix, uint8_t* BROTLI_RESTRICT storage);
+#if defined(BROTLI_TEST)
+void GetBlockLengthPrefixCodeForTest(uint32_t, size_t*, uint32_t*, uint32_t*);
+#endif
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/cluster.c b/dviware/dvisvgm/libs/brotli/enc/cluster.c
index a20dfd385f..b0faf8114c 100644
--- a/dviware/dvisvgm/libs/brotli/enc/cluster.c
+++ b/dviware/dvisvgm/libs/brotli/enc/cluster.c
@@ -6,14 +6,15 @@
/* Functions for clustering similar histograms together. */
-#include "./cluster.h"
+#include "cluster.h"
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./bit_cost.h" /* BrotliPopulationCost */
-#include "./fast_log.h"
-#include "./histogram.h"
-#include "./memory.h"
+
+#include "../common/platform.h"
+#include "bit_cost.h" /* BrotliPopulationCost */
+#include "fast_log.h"
+#include "histogram.h"
+#include "memory.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -38,15 +39,15 @@ static BROTLI_INLINE double ClusterCostDiff(size_t size_a, size_t size_b) {
#define CODE(X) X
#define FN(X) X ## Literal
-#include "./cluster_inc.h" /* NOLINT(build/include) */
+#include "cluster_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Command
-#include "./cluster_inc.h" /* NOLINT(build/include) */
+#include "cluster_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Distance
-#include "./cluster_inc.h" /* NOLINT(build/include) */
+#include "cluster_inc.h" /* NOLINT(build/include) */
#undef FN
#undef CODE
diff --git a/dviware/dvisvgm/libs/brotli/enc/cluster.h b/dviware/dvisvgm/libs/brotli/enc/cluster.h
index bb26124d24..013629c6d9 100644
--- a/dviware/dvisvgm/libs/brotli/enc/cluster.h
+++ b/dviware/dvisvgm/libs/brotli/enc/cluster.h
@@ -9,10 +9,11 @@
#ifndef BROTLI_ENC_CLUSTER_H_
#define BROTLI_ENC_CLUSTER_H_
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./histogram.h"
-#include "./memory.h"
+
+#include "../common/platform.h"
+#include "histogram.h"
+#include "memory.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -28,15 +29,15 @@ typedef struct HistogramPair {
#define CODE(X) /* Declaration */;
#define FN(X) X ## Literal
-#include "./cluster_inc.h" /* NOLINT(build/include) */
+#include "cluster_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Command
-#include "./cluster_inc.h" /* NOLINT(build/include) */
+#include "cluster_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Distance
-#include "./cluster_inc.h" /* NOLINT(build/include) */
+#include "cluster_inc.h" /* NOLINT(build/include) */
#undef FN
#undef CODE
diff --git a/dviware/dvisvgm/libs/brotli/enc/cluster_inc.h b/dviware/dvisvgm/libs/brotli/enc/cluster_inc.h
index 3d4f40e601..d6215ef06e 100644
--- a/dviware/dvisvgm/libs/brotli/enc/cluster_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/cluster_inc.h
@@ -12,8 +12,8 @@
/* Computes the bit cost reduction by combining out[idx1] and out[idx2] and if
it is below a threshold, stores the pair (idx1, idx2) in the *pairs queue. */
BROTLI_INTERNAL void FN(BrotliCompareAndPushToQueue)(
- const HistogramType* out, const uint32_t* cluster_size, uint32_t idx1,
- uint32_t idx2, size_t max_num_pairs, HistogramPair* pairs,
+ const HistogramType* out, HistogramType* tmp, const uint32_t* cluster_size,
+ uint32_t idx1, uint32_t idx2, size_t max_num_pairs, HistogramPair* pairs,
size_t* num_pairs) CODE({
BROTLI_BOOL is_good_pair = BROTLI_FALSE;
HistogramPair p;
@@ -42,10 +42,10 @@ BROTLI_INTERNAL void FN(BrotliCompareAndPushToQueue)(
} else {
double threshold = *num_pairs == 0 ? 1e99 :
BROTLI_MAX(double, 0.0, pairs[0].cost_diff);
- HistogramType combo = out[idx1];
double cost_combo;
- FN(HistogramAddHistogram)(&combo, &out[idx2]);
- cost_combo = FN(BrotliPopulationCost)(&combo);
+ *tmp = out[idx1];
+ FN(HistogramAddHistogram)(tmp, &out[idx2]);
+ cost_combo = FN(BrotliPopulationCost)(tmp);
if (cost_combo < threshold - p.cost_diff) {
p.cost_combo = cost_combo;
is_good_pair = BROTLI_TRUE;
@@ -68,6 +68,7 @@ BROTLI_INTERNAL void FN(BrotliCompareAndPushToQueue)(
})
BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
+ HistogramType* tmp,
uint32_t* cluster_size,
uint32_t* symbols,
uint32_t* clusters,
@@ -87,7 +88,7 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
for (idx1 = 0; idx1 < num_clusters; ++idx1) {
size_t idx2;
for (idx2 = idx1 + 1; idx2 < num_clusters; ++idx2) {
- FN(BrotliCompareAndPushToQueue)(out, cluster_size, clusters[idx1],
+ FN(BrotliCompareAndPushToQueue)(out, tmp, cluster_size, clusters[idx1],
clusters[idx2], max_num_pairs, &pairs[0], &num_pairs);
}
}
@@ -146,8 +147,8 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
/* Push new pairs formed with the combined histogram to the heap. */
for (i = 0; i < num_clusters; ++i) {
- FN(BrotliCompareAndPushToQueue)(out, cluster_size, best_idx1, clusters[i],
- max_num_pairs, &pairs[0], &num_pairs);
+ FN(BrotliCompareAndPushToQueue)(out, tmp, cluster_size, best_idx1,
+ clusters[i], max_num_pairs, &pairs[0], &num_pairs);
}
}
return num_clusters;
@@ -155,13 +156,14 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
/* What is the bit cost of moving histogram from cur_symbol to candidate. */
BROTLI_INTERNAL double FN(BrotliHistogramBitCostDistance)(
- const HistogramType* histogram, const HistogramType* candidate) CODE({
+ const HistogramType* histogram, const HistogramType* candidate,
+ HistogramType* tmp) CODE({
if (histogram->total_count_ == 0) {
return 0.0;
} else {
- HistogramType tmp = *histogram;
- FN(HistogramAddHistogram)(&tmp, candidate);
- return FN(BrotliPopulationCost)(&tmp) - candidate->bit_cost_;
+ *tmp = *histogram;
+ FN(HistogramAddHistogram)(tmp, candidate);
+ return FN(BrotliPopulationCost)(tmp) - candidate->bit_cost_;
}
})
@@ -171,16 +173,16 @@ BROTLI_INTERNAL double FN(BrotliHistogramBitCostDistance)(
Note: we assume that out[]->bit_cost_ is already up-to-date. */
BROTLI_INTERNAL void FN(BrotliHistogramRemap)(const HistogramType* in,
size_t in_size, const uint32_t* clusters, size_t num_clusters,
- HistogramType* out, uint32_t* symbols) CODE({
+ HistogramType* out, HistogramType* tmp, uint32_t* symbols) CODE({
size_t i;
for (i = 0; i < in_size; ++i) {
uint32_t best_out = i == 0 ? symbols[0] : symbols[i - 1];
double best_bits =
- FN(BrotliHistogramBitCostDistance)(&in[i], &out[best_out]);
+ FN(BrotliHistogramBitCostDistance)(&in[i], &out[best_out], tmp);
size_t j;
for (j = 0; j < num_clusters; ++j) {
const double cur_bits =
- FN(BrotliHistogramBitCostDistance)(&in[i], &out[clusters[j]]);
+ FN(BrotliHistogramBitCostDistance)(&in[i], &out[clusters[j]], tmp);
if (cur_bits < best_bits) {
best_bits = cur_bits;
best_out = clusters[j];
@@ -226,7 +228,7 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramReindex)(MemoryManager* m,
++next_index;
}
}
- /* TODO: by using idea of "cycle-sort" we can avoid allocation of
+ /* TODO(eustas): by using idea of "cycle-sort" we can avoid allocation of
tmp and reduce the number of copying by the factor of 2. */
tmp = BROTLI_ALLOC(m, HistogramType, next_index);
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tmp)) return 0;
@@ -257,10 +259,12 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
size_t pairs_capacity = max_input_histograms * max_input_histograms / 2;
/* For the first pass of clustering, we allow all pairs. */
HistogramPair* pairs = BROTLI_ALLOC(m, HistogramPair, pairs_capacity + 1);
+ /* TODO(eustas): move to "persistent" arena? */
+ HistogramType* tmp = BROTLI_ALLOC(m, HistogramType, 1);
size_t i;
if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(cluster_size) ||
- BROTLI_IS_NULL(clusters) || BROTLI_IS_NULL(pairs)) {
+ BROTLI_IS_NULL(clusters) || BROTLI_IS_NULL(pairs)|| BROTLI_IS_NULL(tmp)) {
return;
}
@@ -283,7 +287,7 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
clusters[num_clusters + j] = (uint32_t)(i + j);
}
num_new_clusters =
- FN(BrotliHistogramCombine)(out, cluster_size,
+ FN(BrotliHistogramCombine)(out, tmp, cluster_size,
&histogram_symbols[i],
&clusters[num_clusters], pairs,
num_to_combine, num_to_combine,
@@ -301,7 +305,7 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
if (BROTLI_IS_OOM(m)) return;
/* Collapse similar histograms. */
- num_clusters = FN(BrotliHistogramCombine)(out, cluster_size,
+ num_clusters = FN(BrotliHistogramCombine)(out, tmp, cluster_size,
histogram_symbols, clusters,
pairs, num_clusters, in_size,
max_histograms, max_num_pairs);
@@ -310,7 +314,8 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
BROTLI_FREE(m, cluster_size);
/* Find the optimal map from original histograms to the final ones. */
FN(BrotliHistogramRemap)(in, in_size, clusters, num_clusters,
- out, histogram_symbols);
+ out, tmp, histogram_symbols);
+ BROTLI_FREE(m, tmp);
BROTLI_FREE(m, clusters);
/* Convert the context map to a canonical form. */
*out_size = FN(BrotliHistogramReindex)(m, out, histogram_symbols, in_size);
diff --git a/dviware/dvisvgm/libs/brotli/enc/command.c b/dviware/dvisvgm/libs/brotli/enc/command.c
index 5e6c249196..bf80561bca 100644
--- a/dviware/dvisvgm/libs/brotli/enc/command.c
+++ b/dviware/dvisvgm/libs/brotli/enc/command.c
@@ -4,7 +4,7 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./command.h"
+#include "command.h"
#include <brotli/types.h>
diff --git a/dviware/dvisvgm/libs/brotli/enc/command.h b/dviware/dvisvgm/libs/brotli/enc/command.h
index d84e373d00..ba4de7eab3 100644
--- a/dviware/dvisvgm/libs/brotli/enc/command.h
+++ b/dviware/dvisvgm/libs/brotli/enc/command.h
@@ -9,12 +9,13 @@
#ifndef BROTLI_ENC_COMMAND_H_
#define BROTLI_ENC_COMMAND_H_
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./fast_log.h"
-#include "./params.h"
-#include "./prefix.h"
+#include "fast_log.h"
+#include "params.h"
+#include "prefix.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/compound_dictionary.c b/dviware/dvisvgm/libs/brotli/enc/compound_dictionary.c
new file mode 100644
index 0000000000..a3b5e6933d
--- /dev/null
+++ b/dviware/dvisvgm/libs/brotli/enc/compound_dictionary.c
@@ -0,0 +1,207 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+ Distributed under MIT license.
+ See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+#include "compound_dictionary.h"
+
+#include <brotli/types.h>
+
+#include "../common/platform.h"
+#include "memory.h"
+#include "quality.h"
+
+static PreparedDictionary* CreatePreparedDictionaryWithParams(MemoryManager* m,
+ const uint8_t* source, size_t source_size, uint32_t bucket_bits,
+ uint32_t slot_bits, uint32_t hash_bits, uint16_t bucket_limit) {
+ /* Step 1: create "bloated" hasher. */
+ uint32_t num_slots = 1u << slot_bits;
+ uint32_t num_buckets = 1u << bucket_bits;
+ uint32_t hash_shift = 64u - bucket_bits;
+ uint64_t hash_mask = (~((uint64_t)0U)) >> (64 - hash_bits);
+ uint32_t slot_mask = num_slots - 1;
+ size_t alloc_size = (sizeof(uint32_t) << slot_bits) +
+ (sizeof(uint32_t) << slot_bits) +
+ (sizeof(uint16_t) << bucket_bits) +
+ (sizeof(uint32_t) << bucket_bits) +
+ (sizeof(uint32_t) * source_size);
+ uint8_t* flat = NULL;
+ PreparedDictionary* result = NULL;
+ uint16_t* num = NULL;
+ uint32_t* bucket_heads = NULL;
+ uint32_t* next_bucket = NULL;
+ uint32_t* slot_offsets = NULL;
+ uint16_t* heads = NULL;
+ uint32_t* items = NULL;
+ uint8_t** source_ref = NULL;
+ uint32_t i;
+ uint32_t* slot_size = NULL;
+ uint32_t* slot_limit = NULL;
+ uint32_t total_items = 0;
+ if (slot_bits > 16) return NULL;
+ if (slot_bits > bucket_bits) return NULL;
+ if (bucket_bits - slot_bits >= 16) return NULL;
+
+ flat = BROTLI_ALLOC(m, uint8_t, alloc_size);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(flat)) return NULL;
+
+ slot_size = (uint32_t*)flat;
+ slot_limit = (uint32_t*)(&slot_size[num_slots]);
+ num = (uint16_t*)(&slot_limit[num_slots]);
+ bucket_heads = (uint32_t*)(&num[num_buckets]);
+ next_bucket = (uint32_t*)(&bucket_heads[num_buckets]);
+ memset(num, 0, num_buckets * sizeof(num[0]));
+
+ /* TODO(eustas): apply custom "store" order. */
+ for (i = 0; i + 7 < source_size; ++i) {
+ const uint64_t h = (BROTLI_UNALIGNED_LOAD64LE(&source[i]) & hash_mask) *
+ kPreparedDictionaryHashMul64Long;
+ const uint32_t key = (uint32_t)(h >> hash_shift);
+ uint16_t count = num[key];
+ next_bucket[i] = (count == 0) ? ((uint32_t)(-1)) : bucket_heads[key];
+ bucket_heads[key] = i;
+ count++;
+ if (count > bucket_limit) count = bucket_limit;
+ num[key] = count;
+ }
+
+ /* Step 2: find slot limits. */
+ for (i = 0; i < num_slots; ++i) {
+ BROTLI_BOOL overflow = BROTLI_FALSE;
+ slot_limit[i] = bucket_limit;
+ while (BROTLI_TRUE) {
+ uint32_t limit = slot_limit[i];
+ size_t j;
+ uint32_t count = 0;
+ overflow = BROTLI_FALSE;
+ for (j = i; j < num_buckets; j += num_slots) {
+ uint32_t size = num[j];
+ /* Last chain may span behind 64K limit; overflow happens only if
+ we are about to use 0xFFFF+ as item offset. */
+ if (count >= 0xFFFF) {
+ overflow = BROTLI_TRUE;
+ break;
+ }
+ if (size > limit) size = limit;
+ count += size;
+ }
+ if (!overflow) {
+ slot_size[i] = count;
+ total_items += count;
+ break;
+ }
+ slot_limit[i]--;
+ }
+ }
+
+ /* Step 3: transfer data to "slim" hasher. */
+ alloc_size = sizeof(PreparedDictionary) + (sizeof(uint32_t) << slot_bits) +
+ (sizeof(uint16_t) << bucket_bits) + (sizeof(uint32_t) * total_items) +
+ sizeof(uint8_t*);
+
+ result = (PreparedDictionary*)BROTLI_ALLOC(m, uint8_t, alloc_size);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(result)) {
+ BROTLI_FREE(m, flat);
+ return NULL;
+ }
+ slot_offsets = (uint32_t*)(&result[1]);
+ heads = (uint16_t*)(&slot_offsets[num_slots]);
+ items = (uint32_t*)(&heads[num_buckets]);
+ source_ref = (uint8_t**)(&items[total_items]);
+
+ result->magic = kLeanPreparedDictionaryMagic;
+ result->num_items = total_items;
+ result->source_size = (uint32_t)source_size;
+ result->hash_bits = hash_bits;
+ result->bucket_bits = bucket_bits;
+ result->slot_bits = slot_bits;
+ BROTLI_UNALIGNED_STORE_PTR(source_ref, source);
+
+ total_items = 0;
+ for (i = 0; i < num_slots; ++i) {
+ slot_offsets[i] = total_items;
+ total_items += slot_size[i];
+ slot_size[i] = 0;
+ }
+ for (i = 0; i < num_buckets; ++i) {
+ uint32_t slot = i & slot_mask;
+ uint32_t count = num[i];
+ uint32_t pos;
+ size_t j;
+ size_t cursor = slot_size[slot];
+ if (count > slot_limit[slot]) count = slot_limit[slot];
+ if (count == 0) {
+ heads[i] = 0xFFFF;
+ continue;
+ }
+ heads[i] = (uint16_t)cursor;
+ cursor += slot_offsets[slot];
+ slot_size[slot] += count;
+ pos = bucket_heads[i];
+ for (j = 0; j < count; j++) {
+ items[cursor++] = pos;
+ pos = next_bucket[pos];
+ }
+ items[cursor - 1] |= 0x80000000;
+ }
+
+ BROTLI_FREE(m, flat);
+ return result;
+}
+
+PreparedDictionary* CreatePreparedDictionary(MemoryManager* m,
+ const uint8_t* source, size_t source_size) {
+ uint32_t bucket_bits = 17;
+ uint32_t slot_bits = 7;
+ uint32_t hash_bits = 40;
+ uint16_t bucket_limit = 32;
+ size_t volume = 16u << bucket_bits;
+ /* Tune parameters to fit dictionary size. */
+ while (volume < source_size && bucket_bits < 22) {
+ bucket_bits++;
+ slot_bits++;
+ volume <<= 1;
+ }
+ return CreatePreparedDictionaryWithParams(m,
+ source, source_size, bucket_bits, slot_bits, hash_bits, bucket_limit);
+}
+
+void DestroyPreparedDictionary(MemoryManager* m,
+ PreparedDictionary* dictionary) {
+ if (!dictionary) return;
+ BROTLI_FREE(m, dictionary);
+}
+
+BROTLI_BOOL AttachPreparedDictionary(
+ CompoundDictionary* compound, const PreparedDictionary* dictionary) {
+ size_t length = 0;
+ size_t index = 0;
+
+ if (compound->num_chunks == SHARED_BROTLI_MAX_COMPOUND_DICTS) {
+ return BROTLI_FALSE;
+ }
+
+ if (!dictionary) return BROTLI_FALSE;
+
+ length = dictionary->source_size;
+ index = compound->num_chunks;
+ compound->total_size += length;
+ compound->chunks[index] = dictionary;
+ compound->chunk_offsets[index + 1] = compound->total_size;
+ {
+ uint32_t* slot_offsets = (uint32_t*)(&dictionary[1]);
+ uint16_t* heads = (uint16_t*)(&slot_offsets[1u << dictionary->slot_bits]);
+ uint32_t* items = (uint32_t*)(&heads[1u << dictionary->bucket_bits]);
+ const void* tail = (void*)&items[dictionary->num_items];
+ if (dictionary->magic == kPreparedDictionaryMagic) {
+ compound->chunk_source[index] = (const uint8_t*)tail;
+ } else {
+ /* dictionary->magic == kLeanPreparedDictionaryMagic */
+ compound->chunk_source[index] =
+ (const uint8_t*)BROTLI_UNALIGNED_LOAD_PTR((const uint8_t**)tail);
+ }
+ }
+ compound->num_chunks++;
+ return BROTLI_TRUE;
+}
diff --git a/dviware/dvisvgm/libs/brotli/enc/compound_dictionary.h b/dviware/dvisvgm/libs/brotli/enc/compound_dictionary.h
new file mode 100644
index 0000000000..9c531d5b19
--- /dev/null
+++ b/dviware/dvisvgm/libs/brotli/enc/compound_dictionary.h
@@ -0,0 +1,74 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+ Distributed under MIT license.
+ See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+#ifndef BROTLI_ENC_PREPARED_DICTIONARY_H_
+#define BROTLI_ENC_PREPARED_DICTIONARY_H_
+
+#include <brotli/shared_dictionary.h>
+#include <brotli/types.h>
+
+#include "../common/platform.h"
+#include "../common/constants.h"
+#include "memory.h"
+
+/* "Fat" prepared dictionary, could be cooked outside of C implementation,
+ * e.g. on Java side. LZ77 data is copied inside PreparedDictionary struct. */
+static const uint32_t kPreparedDictionaryMagic = 0xDEBCEDE0;
+
+static const uint32_t kSharedDictionaryMagic = 0xDEBCEDE1;
+
+static const uint32_t kManagedDictionaryMagic = 0xDEBCEDE2;
+
+/* "Lean" prepared dictionary. LZ77 data is referenced. It is the responsibility
+ * of caller of "prepare dictionary" to keep the LZ77 data while prepared
+ * dictionary is in use. */
+static const uint32_t kLeanPreparedDictionaryMagic = 0xDEBCEDE3;
+
+static const uint64_t kPreparedDictionaryHashMul64Long =
+ BROTLI_MAKE_UINT64_T(0x1FE35A7Bu, 0xD3579BD3u);
+
+typedef struct PreparedDictionary {
+ uint32_t magic;
+ uint32_t num_items;
+ uint32_t source_size;
+ uint32_t hash_bits;
+ uint32_t bucket_bits;
+ uint32_t slot_bits;
+
+ /* --- Dynamic size members --- */
+
+ /* uint32_t slot_offsets[1 << slot_bits]; */
+ /* uint16_t heads[1 << bucket_bits]; */
+ /* uint32_t items[variable]; */
+
+ /* [maybe] uint8_t* source_ref, depending on magic. */
+ /* [maybe] uint8_t source[source_size], depending on magic. */
+} PreparedDictionary;
+
+BROTLI_INTERNAL PreparedDictionary* CreatePreparedDictionary(MemoryManager* m,
+ const uint8_t* source, size_t source_size);
+
+BROTLI_INTERNAL void DestroyPreparedDictionary(MemoryManager* m,
+ PreparedDictionary* dictionary);
+
+typedef struct CompoundDictionary {
+ /* LZ77 prefix, compound dictionary */
+ size_t num_chunks;
+ size_t total_size;
+ /* Client instances. */
+ const PreparedDictionary* chunks[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+ const uint8_t* chunk_source[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+ size_t chunk_offsets[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+
+ size_t num_prepared_instances_;
+ /* Owned instances. */
+ PreparedDictionary* prepared_instances_[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+} CompoundDictionary;
+
+BROTLI_INTERNAL BROTLI_BOOL AttachPreparedDictionary(
+ CompoundDictionary* compound, const PreparedDictionary* dictionary);
+
+#endif /* BROTLI_ENC_PREPARED_DICTIONARY */
diff --git a/dviware/dvisvgm/libs/brotli/enc/compress_fragment.c b/dviware/dvisvgm/libs/brotli/enc/compress_fragment.c
index 9e50b2098a..13890eabf6 100644
--- a/dviware/dvisvgm/libs/brotli/enc/compress_fragment.c
+++ b/dviware/dvisvgm/libs/brotli/enc/compress_fragment.c
@@ -12,19 +12,18 @@
Adapted from the CompressFragment() function in
https://github.com/google/snappy/blob/master/snappy.cc */
-#include "./compress_fragment.h"
+#include "compress_fragment.h"
#include <string.h> /* memcmp, memcpy, memset */
-#include "../common/constants.h"
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./brotli_bit_stream.h"
-#include "./entropy_encode.h"
-#include "./fast_log.h"
-#include "./find_match_length.h"
-#include "./memory.h"
-#include "./write_bits.h"
+
+#include "../common/platform.h"
+#include "brotli_bit_stream.h"
+#include "entropy_encode.h"
+#include "fast_log.h"
+#include "find_match_length.h"
+#include "write_bits.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -69,16 +68,18 @@ static BROTLI_INLINE BROTLI_BOOL IsMatch(const uint8_t* p1, const uint8_t* p2) {
and thus have to assign a non-zero depth for each literal.
Returns estimated compression ratio millibytes/char for encoding given input
with generated code. */
-static size_t BuildAndStoreLiteralPrefixCode(MemoryManager* m,
+static size_t BuildAndStoreLiteralPrefixCode(BrotliOnePassArena* s,
const uint8_t* input,
const size_t input_size,
uint8_t depths[256],
uint16_t bits[256],
size_t* storage_ix,
uint8_t* storage) {
- uint32_t histogram[256] = { 0 };
+ uint32_t* BROTLI_RESTRICT const histogram = s->histogram;
size_t histogram_total;
size_t i;
+ memset(histogram, 0, sizeof(s->histogram));
+
if (input_size < (1 << 15)) {
for (i = 0; i < input_size; ++i) {
++histogram[input[i]];
@@ -108,10 +109,9 @@ static size_t BuildAndStoreLiteralPrefixCode(MemoryManager* m,
histogram_total += adjust;
}
}
- BrotliBuildAndStoreHuffmanTreeFast(m, histogram, histogram_total,
+ BrotliBuildAndStoreHuffmanTreeFast(s->tree, histogram, histogram_total,
/* max_bits = */ 8,
depths, bits, storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return 0;
{
size_t literal_ratio = 0;
for (i = 0; i < 256; ++i) {
@@ -124,53 +124,56 @@ static size_t BuildAndStoreLiteralPrefixCode(MemoryManager* m,
/* Builds a command and distance prefix code (each 64 symbols) into "depth" and
"bits" based on "histogram" and stores it into the bit stream. */
-static void BuildAndStoreCommandPrefixCode(const uint32_t histogram[128],
- uint8_t depth[128], uint16_t bits[128], size_t* storage_ix,
- uint8_t* storage) {
- /* Tree size for building a tree over 64 symbols is 2 * 64 + 1. */
- HuffmanTree tree[129];
- uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS] = { 0 };
- uint16_t cmd_bits[64];
-
- BrotliCreateHuffmanTree(histogram, 64, 15, tree, depth);
- BrotliCreateHuffmanTree(&histogram[64], 64, 14, tree, &depth[64]);
+static void BuildAndStoreCommandPrefixCode(BrotliOnePassArena* s,
+ size_t* storage_ix, uint8_t* storage) {
+ const uint32_t* const histogram = s->cmd_histo;
+ uint8_t* const depth = s->cmd_depth;
+ uint16_t* const bits = s->cmd_bits;
+ uint8_t* BROTLI_RESTRICT const tmp_depth = s->tmp_depth;
+ uint16_t* BROTLI_RESTRICT const tmp_bits = s->tmp_bits;
+ /* TODO(eustas): do only once on initialization. */
+ memset(tmp_depth, 0, BROTLI_NUM_COMMAND_SYMBOLS);
+
+ BrotliCreateHuffmanTree(histogram, 64, 15, s->tree, depth);
+ BrotliCreateHuffmanTree(&histogram[64], 64, 14, s->tree, &depth[64]);
/* We have to jump through a few hoops here in order to compute
the command bits because the symbols are in a different order than in
the full alphabet. This looks complicated, but having the symbols
in this order in the command bits saves a few branches in the Emit*
functions. */
- memcpy(cmd_depth, depth, 24);
- memcpy(cmd_depth + 24, depth + 40, 8);
- memcpy(cmd_depth + 32, depth + 24, 8);
- memcpy(cmd_depth + 40, depth + 48, 8);
- memcpy(cmd_depth + 48, depth + 32, 8);
- memcpy(cmd_depth + 56, depth + 56, 8);
- BrotliConvertBitDepthsToSymbols(cmd_depth, 64, cmd_bits);
- memcpy(bits, cmd_bits, 48);
- memcpy(bits + 24, cmd_bits + 32, 16);
- memcpy(bits + 32, cmd_bits + 48, 16);
- memcpy(bits + 40, cmd_bits + 24, 16);
- memcpy(bits + 48, cmd_bits + 40, 16);
- memcpy(bits + 56, cmd_bits + 56, 16);
+ memcpy(tmp_depth, depth, 24);
+ memcpy(tmp_depth + 24, depth + 40, 8);
+ memcpy(tmp_depth + 32, depth + 24, 8);
+ memcpy(tmp_depth + 40, depth + 48, 8);
+ memcpy(tmp_depth + 48, depth + 32, 8);
+ memcpy(tmp_depth + 56, depth + 56, 8);
+ BrotliConvertBitDepthsToSymbols(tmp_depth, 64, tmp_bits);
+ memcpy(bits, tmp_bits, 48);
+ memcpy(bits + 24, tmp_bits + 32, 16);
+ memcpy(bits + 32, tmp_bits + 48, 16);
+ memcpy(bits + 40, tmp_bits + 24, 16);
+ memcpy(bits + 48, tmp_bits + 40, 16);
+ memcpy(bits + 56, tmp_bits + 56, 16);
BrotliConvertBitDepthsToSymbols(&depth[64], 64, &bits[64]);
{
/* Create the bit length array for the full command alphabet. */
size_t i;
- memset(cmd_depth, 0, 64); /* only 64 first values were used */
- memcpy(cmd_depth, depth, 8);
- memcpy(cmd_depth + 64, depth + 8, 8);
- memcpy(cmd_depth + 128, depth + 16, 8);
- memcpy(cmd_depth + 192, depth + 24, 8);
- memcpy(cmd_depth + 384, depth + 32, 8);
+ memset(tmp_depth, 0, 64); /* only 64 first values were used */
+ memcpy(tmp_depth, depth, 8);
+ memcpy(tmp_depth + 64, depth + 8, 8);
+ memcpy(tmp_depth + 128, depth + 16, 8);
+ memcpy(tmp_depth + 192, depth + 24, 8);
+ memcpy(tmp_depth + 384, depth + 32, 8);
for (i = 0; i < 8; ++i) {
- cmd_depth[128 + 8 * i] = depth[40 + i];
- cmd_depth[256 + 8 * i] = depth[48 + i];
- cmd_depth[448 + 8 * i] = depth[56 + i];
+ tmp_depth[128 + 8 * i] = depth[40 + i];
+ tmp_depth[256 + 8 * i] = depth[48 + i];
+ tmp_depth[448 + 8 * i] = depth[56 + i];
}
+ /* TODO(eustas): could/should full-length machinery be avoided? */
BrotliStoreHuffmanTree(
- cmd_depth, BROTLI_NUM_COMMAND_SYMBOLS, tree, storage_ix, storage);
+ tmp_depth, BROTLI_NUM_COMMAND_SYMBOLS, s->tree, storage_ix, storage);
}
- BrotliStoreHuffmanTree(&depth[64], 64, tree, storage_ix, storage);
+ BrotliStoreHuffmanTree(&depth[64], 64, s->tree, storage_ix, storage);
}
/* REQUIRES: insertlen < 6210 */
@@ -369,11 +372,12 @@ static void RewindBitPosition(const size_t new_storage_ix,
*storage_ix = new_storage_ix;
}
-static BROTLI_BOOL ShouldMergeBlock(
+static BROTLI_BOOL ShouldMergeBlock(BrotliOnePassArena* s,
const uint8_t* data, size_t len, const uint8_t* depths) {
- size_t histo[256] = { 0 };
+ uint32_t* BROTLI_RESTRICT const histo = s->histogram;
static const size_t kSampleRate = 43;
size_t i;
+ memset(histo, 0, sizeof(s->histogram));
for (i = 0; i < len; i += kSampleRate) {
++histo[data[i]];
}
@@ -423,11 +427,14 @@ static uint32_t kCmdHistoSeed[128] = {
};
static BROTLI_INLINE void BrotliCompressFragmentFastImpl(
- MemoryManager* m, const uint8_t* input, size_t input_size,
- BROTLI_BOOL is_last, int* table, size_t table_bits, uint8_t cmd_depth[128],
- uint16_t cmd_bits[128], size_t* cmd_code_numbits, uint8_t* cmd_code,
+ BrotliOnePassArena* s, const uint8_t* input, size_t input_size,
+ BROTLI_BOOL is_last, int* table, size_t table_bits,
size_t* storage_ix, uint8_t* storage) {
- uint32_t cmd_histo[128];
+ uint8_t* BROTLI_RESTRICT const cmd_depth = s->cmd_depth;
+ uint16_t* BROTLI_RESTRICT const cmd_bits = s->cmd_bits;
+ uint32_t* BROTLI_RESTRICT const cmd_histo = s->cmd_histo;
+ uint8_t* BROTLI_RESTRICT const lit_depth = s->lit_depth;
+ uint16_t* BROTLI_RESTRICT const lit_bits = s->lit_bits;
const uint8_t* ip_end;
/* "next_emit" is a pointer to the first byte that is not covered by a
@@ -451,9 +458,6 @@ static BROTLI_INLINE void BrotliCompressFragmentFastImpl(
we can update it later if we decide to extend this meta-block. */
size_t mlen_storage_ix = *storage_ix + 3;
- uint8_t lit_depth[256];
- uint16_t lit_bits[256];
-
size_t literal_ratio;
const uint8_t* ip;
@@ -466,25 +470,24 @@ static BROTLI_INLINE void BrotliCompressFragmentFastImpl(
BrotliWriteBits(13, 0, storage_ix, storage);
literal_ratio = BuildAndStoreLiteralPrefixCode(
- m, input, block_size, lit_depth, lit_bits, storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
+ s, input, block_size, s->lit_depth, s->lit_bits, storage_ix, storage);
{
/* Store the pre-compressed command and distance prefix codes. */
size_t i;
- for (i = 0; i + 7 < *cmd_code_numbits; i += 8) {
- BrotliWriteBits(8, cmd_code[i >> 3], storage_ix, storage);
+ for (i = 0; i + 7 < s->cmd_code_numbits; i += 8) {
+ BrotliWriteBits(8, s->cmd_code[i >> 3], storage_ix, storage);
}
}
- BrotliWriteBits(*cmd_code_numbits & 7, cmd_code[*cmd_code_numbits >> 3],
- storage_ix, storage);
+ BrotliWriteBits(s->cmd_code_numbits & 7,
+ s->cmd_code[s->cmd_code_numbits >> 3], storage_ix, storage);
emit_commands:
/* Initialize the command and distance histograms. We will gather
statistics of command and distance codes during the processing
of this block and use it to update the command and distance
prefix codes for the next block. */
- memcpy(cmd_histo, kCmdHistoSeed, sizeof(kCmdHistoSeed));
+ memcpy(s->cmd_histo, kCmdHistoSeed, sizeof(kCmdHistoSeed));
/* "ip" is the input pointer. */
ip = input;
@@ -565,6 +568,8 @@ trawl:
int distance = (int)(base - candidate); /* > 0 */
size_t insert = (size_t)(base - next_emit);
ip += matched;
+ BROTLI_LOG(("[CompressFragment] pos = %d insert = %lu copy = %d\n",
+ (int)(next_emit - base_ip), (unsigned long)insert, 2));
BROTLI_DCHECK(0 == memcmp(base, candidate, matched));
if (BROTLI_PREDICT_TRUE(insert < 6210)) {
EmitInsertLen(insert, cmd_depth, cmd_bits, cmd_histo,
@@ -593,6 +598,12 @@ trawl:
}
EmitCopyLenLastDistance(matched, cmd_depth, cmd_bits, cmd_histo,
storage_ix, storage);
+ BROTLI_LOG(("[CompressFragment] pos = %d distance = %d\n"
+ "[CompressFragment] pos = %d insert = %d copy = %d\n"
+ "[CompressFragment] pos = %d distance = %d\n",
+ (int)(base - base_ip), (int)distance,
+ (int)(base - base_ip) + 2, 0, (int)matched - 2,
+ (int)(base - base_ip) + 2, (int)distance));
next_emit = ip;
if (BROTLI_PREDICT_FALSE(ip >= ip_limit)) {
@@ -630,6 +641,10 @@ trawl:
storage_ix, storage);
EmitDistance((size_t)last_distance, cmd_depth, cmd_bits,
cmd_histo, storage_ix, storage);
+ BROTLI_LOG(("[CompressFragment] pos = %d insert = %d copy = %d\n"
+ "[CompressFragment] pos = %d distance = %d\n",
+ (int)(base - base_ip), 0, (int)matched,
+ (int)(base - base_ip), (int)last_distance));
next_emit = ip;
if (BROTLI_PREDICT_FALSE(ip >= ip_limit)) {
@@ -667,7 +682,7 @@ trawl:
last insert-only command. */
if (input_size > 0 &&
total_block_size + block_size <= (1 << 20) &&
- ShouldMergeBlock(input, block_size, lit_depth)) {
+ ShouldMergeBlock(s, input, block_size, lit_depth)) {
BROTLI_DCHECK(total_block_size > (1 << 16));
/* Update the size of the current meta-block and continue emitting commands.
We can do this because the current size and the new size both have 5
@@ -680,6 +695,8 @@ trawl:
/* Emit the remaining bytes as literals. */
if (next_emit < ip_end) {
const size_t insert = (size_t)(ip_end - next_emit);
+ BROTLI_LOG(("[CompressFragment] pos = %d insert = %lu copy = %d\n",
+ (int)(next_emit - base_ip), (unsigned long)insert, 2));
if (BROTLI_PREDICT_TRUE(insert < 6210)) {
EmitInsertLen(insert, cmd_depth, cmd_bits, cmd_histo,
storage_ix, storage);
@@ -711,20 +728,17 @@ next_block:
/* No block splits, no contexts. */
BrotliWriteBits(13, 0, storage_ix, storage);
literal_ratio = BuildAndStoreLiteralPrefixCode(
- m, input, block_size, lit_depth, lit_bits, storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
- BuildAndStoreCommandPrefixCode(cmd_histo, cmd_depth, cmd_bits,
- storage_ix, storage);
+ s, input, block_size, lit_depth, lit_bits, storage_ix, storage);
+ BuildAndStoreCommandPrefixCode(s, storage_ix, storage);
goto emit_commands;
}
if (!is_last) {
/* If this is not the last block, update the command and distance prefix
codes for the next block and store the compressed forms. */
- cmd_code[0] = 0;
- *cmd_code_numbits = 0;
- BuildAndStoreCommandPrefixCode(cmd_histo, cmd_depth, cmd_bits,
- cmd_code_numbits, cmd_code);
+ s->cmd_code[0] = 0;
+ s->cmd_code_numbits = 0;
+ BuildAndStoreCommandPrefixCode(s, &s->cmd_code_numbits, s->cmd_code);
}
}
@@ -732,20 +746,17 @@ next_block:
#define BAKE_METHOD_PARAM_(B) \
static BROTLI_NOINLINE void BrotliCompressFragmentFastImpl ## B( \
- MemoryManager* m, const uint8_t* input, size_t input_size, \
- BROTLI_BOOL is_last, int* table, uint8_t cmd_depth[128], \
- uint16_t cmd_bits[128], size_t* cmd_code_numbits, uint8_t* cmd_code, \
- size_t* storage_ix, uint8_t* storage) { \
- BrotliCompressFragmentFastImpl(m, input, input_size, is_last, table, B, \
- cmd_depth, cmd_bits, cmd_code_numbits, cmd_code, storage_ix, storage); \
+ BrotliOnePassArena* s, const uint8_t* input, size_t input_size, \
+ BROTLI_BOOL is_last, int* table, size_t* storage_ix, uint8_t* storage) { \
+ BrotliCompressFragmentFastImpl(s, input, input_size, is_last, table, B, \
+ storage_ix, storage); \
}
FOR_TABLE_BITS_(BAKE_METHOD_PARAM_)
#undef BAKE_METHOD_PARAM_
void BrotliCompressFragmentFast(
- MemoryManager* m, const uint8_t* input, size_t input_size,
- BROTLI_BOOL is_last, int* table, size_t table_size, uint8_t cmd_depth[128],
- uint16_t cmd_bits[128], size_t* cmd_code_numbits, uint8_t* cmd_code,
+ BrotliOnePassArena* s, const uint8_t* input, size_t input_size,
+ BROTLI_BOOL is_last, int* table, size_t table_size,
size_t* storage_ix, uint8_t* storage) {
const size_t initial_storage_ix = *storage_ix;
const size_t table_bits = Log2FloorNonZero(table_size);
@@ -762,8 +773,7 @@ void BrotliCompressFragmentFast(
#define CASE_(B) \
case B: \
BrotliCompressFragmentFastImpl ## B( \
- m, input, input_size, is_last, table, cmd_depth, cmd_bits, \
- cmd_code_numbits, cmd_code, storage_ix, storage); \
+ s, input, input_size, is_last, table, storage_ix, storage);\
break;
FOR_TABLE_BITS_(CASE_)
#undef CASE_
diff --git a/dviware/dvisvgm/libs/brotli/enc/compress_fragment.h b/dviware/dvisvgm/libs/brotli/enc/compress_fragment.h
index 80007f5dca..9c0780f8c9 100644
--- a/dviware/dvisvgm/libs/brotli/enc/compress_fragment.h
+++ b/dviware/dvisvgm/libs/brotli/enc/compress_fragment.h
@@ -12,14 +12,43 @@
#ifndef BROTLI_ENC_COMPRESS_FRAGMENT_H_
#define BROTLI_ENC_COMPRESS_FRAGMENT_H_
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./memory.h"
+
+#include "../common/constants.h"
+#include "../common/platform.h"
+#include "entropy_encode.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
+typedef struct BrotliOnePassArena {
+ uint8_t lit_depth[256];
+ uint16_t lit_bits[256];
+
+ /* Command and distance prefix codes (each 64 symbols, stored back-to-back)
+ used for the next block. The command prefix code is over a smaller alphabet
+ with the following 64 symbols:
+ 0 - 15: insert length code 0, copy length code 0 - 15, same distance
+ 16 - 39: insert length code 0, copy length code 0 - 23
+ 40 - 63: insert length code 0 - 23, copy length code 0
+ Note that symbols 16 and 40 represent the same code in the full alphabet,
+ but we do not use either of them. */
+ uint8_t cmd_depth[128];
+ uint16_t cmd_bits[128];
+ uint32_t cmd_histo[128];
+
+ /* The compressed form of the command and distance prefix codes for the next
+ block. */
+ uint8_t cmd_code[512];
+ size_t cmd_code_numbits;
+
+ HuffmanTree tree[2 * BROTLI_NUM_LITERAL_SYMBOLS + 1];
+ uint32_t histogram[256];
+ uint8_t tmp_depth[BROTLI_NUM_COMMAND_SYMBOLS];
+ uint16_t tmp_bits[64];
+} BrotliOnePassArena;
+
/* Compresses "input" string to the "*storage" buffer as one or more complete
meta-blocks, and updates the "*storage_ix" bit position.
@@ -42,15 +71,11 @@ extern "C" {
REQUIRES: "table_size" is an odd (9, 11, 13, 15) power of two
OUTPUT: maximal copy distance <= |input_size|
OUTPUT: maximal copy distance <= BROTLI_MAX_BACKWARD_LIMIT(18) */
-BROTLI_INTERNAL void BrotliCompressFragmentFast(MemoryManager* m,
+BROTLI_INTERNAL void BrotliCompressFragmentFast(BrotliOnePassArena* s,
const uint8_t* input,
size_t input_size,
BROTLI_BOOL is_last,
int* table, size_t table_size,
- uint8_t cmd_depth[128],
- uint16_t cmd_bits[128],
- size_t* cmd_code_numbits,
- uint8_t* cmd_code,
size_t* storage_ix,
uint8_t* storage);
diff --git a/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.c b/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.c
index ca68b38139..a762679c1f 100644
--- a/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.c
+++ b/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.c
@@ -10,20 +10,20 @@
second pass we emit them into the bit stream using prefix codes built based
on the actual command and literal byte histograms. */
-#include "./compress_fragment_two_pass.h"
+#include "compress_fragment_two_pass.h"
#include <string.h> /* memcmp, memcpy, memset */
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./bit_cost.h"
-#include "./brotli_bit_stream.h"
-#include "./entropy_encode.h"
-#include "./fast_log.h"
-#include "./find_match_length.h"
-#include "./memory.h"
-#include "./write_bits.h"
+#include "bit_cost.h"
+#include "brotli_bit_stream.h"
+#include "entropy_encode.h"
+#include "fast_log.h"
+#include "find_match_length.h"
+#include "write_bits.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -66,53 +66,53 @@ static BROTLI_INLINE BROTLI_BOOL IsMatch(const uint8_t* p1, const uint8_t* p2,
/* Builds a command and distance prefix code (each 64 symbols) into "depth" and
"bits" based on "histogram" and stores it into the bit stream. */
-static void BuildAndStoreCommandPrefixCode(
- const uint32_t histogram[128],
- uint8_t depth[128], uint16_t bits[128],
- size_t* storage_ix, uint8_t* storage) {
+static void BuildAndStoreCommandPrefixCode(BrotliTwoPassArena* s,
+ size_t* storage_ix,
+ uint8_t* storage) {
/* Tree size for building a tree over 64 symbols is 2 * 64 + 1. */
- HuffmanTree tree[129];
- uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS] = { 0 };
- uint16_t cmd_bits[64];
- BrotliCreateHuffmanTree(histogram, 64, 15, tree, depth);
- BrotliCreateHuffmanTree(&histogram[64], 64, 14, tree, &depth[64]);
+ /* TODO(eustas): initialize once. */
+ memset(s->tmp_depth, 0, sizeof(s->tmp_depth));
+ BrotliCreateHuffmanTree(s->cmd_histo, 64, 15, s->tmp_tree, s->cmd_depth);
+ BrotliCreateHuffmanTree(&s->cmd_histo[64], 64, 14, s->tmp_tree,
+ &s->cmd_depth[64]);
/* We have to jump through a few hoops here in order to compute
the command bits because the symbols are in a different order than in
the full alphabet. This looks complicated, but having the symbols
in this order in the command bits saves a few branches in the Emit*
functions. */
- memcpy(cmd_depth, depth + 24, 24);
- memcpy(cmd_depth + 24, depth, 8);
- memcpy(cmd_depth + 32, depth + 48, 8);
- memcpy(cmd_depth + 40, depth + 8, 8);
- memcpy(cmd_depth + 48, depth + 56, 8);
- memcpy(cmd_depth + 56, depth + 16, 8);
- BrotliConvertBitDepthsToSymbols(cmd_depth, 64, cmd_bits);
- memcpy(bits, cmd_bits + 24, 16);
- memcpy(bits + 8, cmd_bits + 40, 16);
- memcpy(bits + 16, cmd_bits + 56, 16);
- memcpy(bits + 24, cmd_bits, 48);
- memcpy(bits + 48, cmd_bits + 32, 16);
- memcpy(bits + 56, cmd_bits + 48, 16);
- BrotliConvertBitDepthsToSymbols(&depth[64], 64, &bits[64]);
+ memcpy(s->tmp_depth, s->cmd_depth + 24, 24);
+ memcpy(s->tmp_depth + 24, s->cmd_depth, 8);
+ memcpy(s->tmp_depth + 32, s->cmd_depth + 48, 8);
+ memcpy(s->tmp_depth + 40, s->cmd_depth + 8, 8);
+ memcpy(s->tmp_depth + 48, s->cmd_depth + 56, 8);
+ memcpy(s->tmp_depth + 56, s->cmd_depth + 16, 8);
+ BrotliConvertBitDepthsToSymbols(s->tmp_depth, 64, s->tmp_bits);
+ memcpy(s->cmd_bits, s->tmp_bits + 24, 16);
+ memcpy(s->cmd_bits + 8, s->tmp_bits + 40, 16);
+ memcpy(s->cmd_bits + 16, s->tmp_bits + 56, 16);
+ memcpy(s->cmd_bits + 24, s->tmp_bits, 48);
+ memcpy(s->cmd_bits + 48, s->tmp_bits + 32, 16);
+ memcpy(s->cmd_bits + 56, s->tmp_bits + 48, 16);
+ BrotliConvertBitDepthsToSymbols(&s->cmd_depth[64], 64, &s->cmd_bits[64]);
{
/* Create the bit length array for the full command alphabet. */
size_t i;
- memset(cmd_depth, 0, 64); /* only 64 first values were used */
- memcpy(cmd_depth, depth + 24, 8);
- memcpy(cmd_depth + 64, depth + 32, 8);
- memcpy(cmd_depth + 128, depth + 40, 8);
- memcpy(cmd_depth + 192, depth + 48, 8);
- memcpy(cmd_depth + 384, depth + 56, 8);
+ memset(s->tmp_depth, 0, 64); /* only 64 first values were used */
+ memcpy(s->tmp_depth, s->cmd_depth + 24, 8);
+ memcpy(s->tmp_depth + 64, s->cmd_depth + 32, 8);
+ memcpy(s->tmp_depth + 128, s->cmd_depth + 40, 8);
+ memcpy(s->tmp_depth + 192, s->cmd_depth + 48, 8);
+ memcpy(s->tmp_depth + 384, s->cmd_depth + 56, 8);
for (i = 0; i < 8; ++i) {
- cmd_depth[128 + 8 * i] = depth[i];
- cmd_depth[256 + 8 * i] = depth[8 + i];
- cmd_depth[448 + 8 * i] = depth[16 + i];
+ s->tmp_depth[128 + 8 * i] = s->cmd_depth[i];
+ s->tmp_depth[256 + 8 * i] = s->cmd_depth[8 + i];
+ s->tmp_depth[448 + 8 * i] = s->cmd_depth[16 + i];
}
- BrotliStoreHuffmanTree(
- cmd_depth, BROTLI_NUM_COMMAND_SYMBOLS, tree, storage_ix, storage);
+ BrotliStoreHuffmanTree(s->tmp_depth, BROTLI_NUM_COMMAND_SYMBOLS,
+ s->tmp_tree, storage_ix, storage);
}
- BrotliStoreHuffmanTree(&depth[64], 64, tree, storage_ix, storage);
+ BrotliStoreHuffmanTree(&s->cmd_depth[64], 64, s->tmp_tree, storage_ix,
+ storage);
}
static BROTLI_INLINE void EmitInsertLen(
@@ -330,6 +330,8 @@ trawl:
ip += matched;
BROTLI_DCHECK(0 == memcmp(base, candidate, matched));
EmitInsertLen((uint32_t)insert, commands);
+ BROTLI_LOG(("[CompressFragment] pos = %d insert = %d copy = %d\n",
+ (int)(next_emit - base_ip), insert, 2));
memcpy(*literals, next_emit, (size_t)insert);
*literals += insert;
if (distance == last_distance) {
@@ -340,6 +342,12 @@ trawl:
last_distance = distance;
}
EmitCopyLenLastDistance(matched, commands);
+ BROTLI_LOG(("[CompressFragment] pos = %d distance = %d\n"
+ "[CompressFragment] pos = %d insert = %d copy = %d\n"
+ "[CompressFragment] pos = %d distance = %d\n",
+ (int)(base - base_ip), (int)distance,
+ (int)(base - base_ip) + 2, 0, (int)matched - 2,
+ (int)(base - base_ip) + 2, (int)distance));
next_emit = ip;
if (BROTLI_PREDICT_FALSE(ip >= ip_limit)) {
@@ -395,6 +403,10 @@ trawl:
BROTLI_DCHECK(0 == memcmp(base, candidate, matched));
EmitCopyLen(matched, commands);
EmitDistance((uint32_t)last_distance, commands);
+ BROTLI_LOG(("[CompressFragment] pos = %d insert = %d copy = %d\n"
+ "[CompressFragment] pos = %d distance = %d\n",
+ (int)(base - base_ip), 0, (int)matched,
+ (int)(base - base_ip), (int)last_distance));
next_emit = ip;
if (BROTLI_PREDICT_FALSE(ip >= ip_limit)) {
@@ -447,70 +459,71 @@ emit_remainder:
if (next_emit < ip_end) {
const uint32_t insert = (uint32_t)(ip_end - next_emit);
EmitInsertLen(insert, commands);
+ BROTLI_LOG(("[CompressFragment] pos = %d insert = %d copy = %d\n",
+ (int)(next_emit - base_ip), insert, 2));
memcpy(*literals, next_emit, insert);
*literals += insert;
}
}
-static void StoreCommands(MemoryManager* m,
+static void StoreCommands(BrotliTwoPassArena* s,
const uint8_t* literals, const size_t num_literals,
const uint32_t* commands, const size_t num_commands,
size_t* storage_ix, uint8_t* storage) {
static const uint32_t kNumExtraBits[128] = {
- 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 7, 8, 9, 10, 12, 14, 24,
- 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4,
- 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 7, 8, 9, 10, 24,
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
- 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8,
- 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, 14, 15, 15, 16, 16,
- 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 23, 23, 24, 24,
+ 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5,
+ 6, 7, 8, 9, 10, 12, 14, 24, 0, 0, 0, 0, 0, 0, 0, 0,
+ 1, 1, 2, 2, 3, 3, 4, 4, 0, 0, 0, 0, 0, 0, 0, 0,
+ 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 7, 8, 9, 10, 24,
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+ 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8,
+ 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, 14, 15, 15, 16, 16,
+ 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 23, 23, 24, 24,
};
static const uint32_t kInsertOffset[24] = {
- 0, 1, 2, 3, 4, 5, 6, 8, 10, 14, 18, 26, 34, 50, 66, 98, 130, 194, 322, 578,
- 1090, 2114, 6210, 22594,
+ 0, 1, 2, 3, 4, 5, 6, 8, 10, 14, 18, 26,
+ 34, 50, 66, 98, 130, 194, 322, 578, 1090, 2114, 6210, 22594,
};
- uint8_t lit_depths[256];
- uint16_t lit_bits[256];
- uint32_t lit_histo[256] = { 0 };
- uint8_t cmd_depths[128] = { 0 };
- uint16_t cmd_bits[128] = { 0 };
- uint32_t cmd_histo[128] = { 0 };
size_t i;
+ memset(s->lit_histo, 0, sizeof(s->lit_histo));
+ /* TODO(eustas): is that necessary? */
+ memset(s->cmd_depth, 0, sizeof(s->cmd_depth));
+ /* TODO(eustas): is that necessary? */
+ memset(s->cmd_bits, 0, sizeof(s->cmd_bits));
+ memset(s->cmd_histo, 0, sizeof(s->cmd_histo));
for (i = 0; i < num_literals; ++i) {
- ++lit_histo[literals[i]];
+ ++s->lit_histo[literals[i]];
}
- BrotliBuildAndStoreHuffmanTreeFast(m, lit_histo, num_literals,
- /* max_bits = */ 8,
- lit_depths, lit_bits,
- storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
+ BrotliBuildAndStoreHuffmanTreeFast(s->tmp_tree, s->lit_histo, num_literals,
+ /* max_bits = */ 8, s->lit_depth,
+ s->lit_bits, storage_ix, storage);
for (i = 0; i < num_commands; ++i) {
const uint32_t code = commands[i] & 0xFF;
BROTLI_DCHECK(code < 128);
- ++cmd_histo[code];
+ ++s->cmd_histo[code];
}
- cmd_histo[1] += 1;
- cmd_histo[2] += 1;
- cmd_histo[64] += 1;
- cmd_histo[84] += 1;
- BuildAndStoreCommandPrefixCode(cmd_histo, cmd_depths, cmd_bits,
- storage_ix, storage);
+ s->cmd_histo[1] += 1;
+ s->cmd_histo[2] += 1;
+ s->cmd_histo[64] += 1;
+ s->cmd_histo[84] += 1;
+ BuildAndStoreCommandPrefixCode(s, storage_ix, storage);
for (i = 0; i < num_commands; ++i) {
const uint32_t cmd = commands[i];
const uint32_t code = cmd & 0xFF;
const uint32_t extra = cmd >> 8;
BROTLI_DCHECK(code < 128);
- BrotliWriteBits(cmd_depths[code], cmd_bits[code], storage_ix, storage);
+ BrotliWriteBits(s->cmd_depth[code], s->cmd_bits[code], storage_ix, storage);
BrotliWriteBits(kNumExtraBits[code], extra, storage_ix, storage);
if (code < 24) {
const uint32_t insert = kInsertOffset[code] + extra;
uint32_t j;
for (j = 0; j < insert; ++j) {
const uint8_t lit = *literals;
- BrotliWriteBits(lit_depths[lit], lit_bits[lit], storage_ix, storage);
+ BrotliWriteBits(s->lit_depth[lit], s->lit_bits[lit], storage_ix,
+ storage);
++literals;
}
}
@@ -521,19 +534,19 @@ static void StoreCommands(MemoryManager* m,
#define MIN_RATIO 0.98
#define SAMPLE_RATE 43
-static BROTLI_BOOL ShouldCompress(
+static BROTLI_BOOL ShouldCompress(BrotliTwoPassArena* s,
const uint8_t* input, size_t input_size, size_t num_literals) {
double corpus_size = (double)input_size;
if ((double)num_literals < MIN_RATIO * corpus_size) {
return BROTLI_TRUE;
} else {
- uint32_t literal_histo[256] = { 0 };
const double max_total_bit_cost = corpus_size * 8 * MIN_RATIO / SAMPLE_RATE;
size_t i;
+ memset(s->lit_histo, 0, sizeof(s->lit_histo));
for (i = 0; i < input_size; i += SAMPLE_RATE) {
- ++literal_histo[input[i]];
+ ++s->lit_histo[input[i]];
}
- return TO_BROTLI_BOOL(BitsEntropy(literal_histo, 256) < max_total_bit_cost);
+ return TO_BROTLI_BOOL(BitsEntropy(s->lit_histo, 256) < max_total_bit_cost);
}
}
@@ -555,7 +568,7 @@ static void EmitUncompressedMetaBlock(const uint8_t* input, size_t input_size,
}
static BROTLI_INLINE void BrotliCompressFragmentTwoPassImpl(
- MemoryManager* m, const uint8_t* input, size_t input_size,
+ BrotliTwoPassArena* s, const uint8_t* input, size_t input_size,
BROTLI_BOOL is_last, uint32_t* command_buf, uint8_t* literal_buf,
int* table, size_t table_bits, size_t min_match,
size_t* storage_ix, uint8_t* storage) {
@@ -573,14 +586,13 @@ static BROTLI_INLINE void BrotliCompressFragmentTwoPassImpl(
CreateCommands(input, block_size, input_size, base_ip, table,
table_bits, min_match, &literals, &commands);
num_literals = (size_t)(literals - literal_buf);
- if (ShouldCompress(input, block_size, num_literals)) {
+ if (ShouldCompress(s, input, block_size, num_literals)) {
const size_t num_commands = (size_t)(commands - command_buf);
BrotliStoreMetaBlockHeader(block_size, 0, storage_ix, storage);
/* No block splits, no contexts. */
BrotliWriteBits(13, 0, storage_ix, storage);
- StoreCommands(m, literal_buf, num_literals, command_buf, num_commands,
+ StoreCommands(s, literal_buf, num_literals, command_buf, num_commands,
storage_ix, storage);
- if (BROTLI_IS_OOM(m)) return;
} else {
/* Since we did not find many backward references and the entropy of
the data is close to 8 bits, we can simply emit an uncompressed block.
@@ -597,18 +609,18 @@ static BROTLI_INLINE void BrotliCompressFragmentTwoPassImpl(
#define BAKE_METHOD_PARAM_(B) \
static BROTLI_NOINLINE void BrotliCompressFragmentTwoPassImpl ## B( \
- MemoryManager* m, const uint8_t* input, size_t input_size, \
+ BrotliTwoPassArena* s, const uint8_t* input, size_t input_size, \
BROTLI_BOOL is_last, uint32_t* command_buf, uint8_t* literal_buf, \
int* table, size_t* storage_ix, uint8_t* storage) { \
size_t min_match = (B <= 15) ? 4 : 6; \
- BrotliCompressFragmentTwoPassImpl(m, input, input_size, is_last, command_buf,\
+ BrotliCompressFragmentTwoPassImpl(s, input, input_size, is_last, command_buf,\
literal_buf, table, B, min_match, storage_ix, storage); \
}
FOR_TABLE_BITS_(BAKE_METHOD_PARAM_)
#undef BAKE_METHOD_PARAM_
void BrotliCompressFragmentTwoPass(
- MemoryManager* m, const uint8_t* input, size_t input_size,
+ BrotliTwoPassArena* s, const uint8_t* input, size_t input_size,
BROTLI_BOOL is_last, uint32_t* command_buf, uint8_t* literal_buf,
int* table, size_t table_size, size_t* storage_ix, uint8_t* storage) {
const size_t initial_storage_ix = *storage_ix;
@@ -617,7 +629,7 @@ void BrotliCompressFragmentTwoPass(
#define CASE_(B) \
case B: \
BrotliCompressFragmentTwoPassImpl ## B( \
- m, input, input_size, is_last, command_buf, \
+ s, input, input_size, is_last, command_buf, \
literal_buf, table, storage_ix, storage); \
break;
FOR_TABLE_BITS_(CASE_)
diff --git a/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.h b/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.h
index 928677df42..6d28d9bb78 100644
--- a/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.h
+++ b/dviware/dvisvgm/libs/brotli/enc/compress_fragment_two_pass.h
@@ -13,16 +13,34 @@
#ifndef BROTLI_ENC_COMPRESS_FRAGMENT_TWO_PASS_H_
#define BROTLI_ENC_COMPRESS_FRAGMENT_TWO_PASS_H_
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./memory.h"
+
+#include "../common/constants.h"
+#include "../common/platform.h"
+#include "entropy_encode.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
+/* TODO(eustas): turn to macro. */
static const size_t kCompressFragmentTwoPassBlockSize = 1 << 17;
+typedef struct BrotliTwoPassArena {
+ uint32_t lit_histo[256];
+ uint8_t lit_depth[256];
+ uint16_t lit_bits[256];
+
+ uint32_t cmd_histo[128];
+ uint8_t cmd_depth[128];
+ uint16_t cmd_bits[128];
+
+ /* BuildAndStoreCommandPrefixCode */
+ HuffmanTree tmp_tree[2 * BROTLI_NUM_LITERAL_SYMBOLS + 1];
+ uint8_t tmp_depth[BROTLI_NUM_COMMAND_SYMBOLS];
+ uint16_t tmp_bits[64];
+} BrotliTwoPassArena;
+
/* Compresses "input" string to the "*storage" buffer as one or more complete
meta-blocks, and updates the "*storage_ix" bit position.
@@ -36,7 +54,7 @@ static const size_t kCompressFragmentTwoPassBlockSize = 1 << 17;
REQUIRES: "table_size" is a power of two
OUTPUT: maximal copy distance <= |input_size|
OUTPUT: maximal copy distance <= BROTLI_MAX_BACKWARD_LIMIT(18) */
-BROTLI_INTERNAL void BrotliCompressFragmentTwoPass(MemoryManager* m,
+BROTLI_INTERNAL void BrotliCompressFragmentTwoPass(BrotliTwoPassArena* s,
const uint8_t* input,
size_t input_size,
BROTLI_BOOL is_last,
diff --git a/dviware/dvisvgm/libs/brotli/enc/dictionary_hash.c b/dviware/dvisvgm/libs/brotli/enc/dictionary_hash.c
index 16d853fe5a..1a60eb3cda 100644
--- a/dviware/dvisvgm/libs/brotli/enc/dictionary_hash.c
+++ b/dviware/dvisvgm/libs/brotli/enc/dictionary_hash.c
@@ -7,12 +7,13 @@
/* Hash table on the 4-byte prefixes of static dictionary words. */
#include "../common/platform.h"
-#include "./dictionary_hash.h"
+#include "dictionary_hash.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
+/* GENERATED CODE START */
BROTLI_INTERNAL const uint16_t kStaticDictionaryHashWords[32768] = {
1002,0,0,0,0,0,0,0,0,683,0,0,0,0,0,0,0,1265,0,0,0,0,0,1431,0,0,0,0,0,0,40,0,0,0,
0,155,8,741,0,624,0,0,0,0,0,0,0,0,0,0,0,0,66,503,0,0,0,451,0,0,0,0,0,0,0,835,70,
@@ -1840,6 +1841,7 @@ BROTLI_INTERNAL const uint8_t kStaticDictionaryHashLengths[32768] = {
0,0,0,0,6,0,0,0,0,0,0,0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0,0,
10,7,0,0,0,0,0,0,0,0,9,0,0,0,0,4,0,0,0,0,0,0,0,0,0,5,11,0,0,0,0,0,0,0,8,6,0,0,9,
7,0,0,12,4,0,0,0,0,0,0,12,6,0,6,0,7,0,0,8,5,0,0,0,0};
+/* GENERATED CODE END */
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/encode.c b/dviware/dvisvgm/libs/brotli/enc/encode.c
index e77009e39a..95ff2eaf50 100644
--- a/dviware/dvisvgm/libs/brotli/enc/encode.c
+++ b/dviware/dvisvgm/libs/brotli/enc/encode.c
@@ -7,6 +7,8 @@
/* Implementation of Brotli compressor. */
#include <brotli/encode.h>
+#include <brotli/shared_dictionary.h>
+#include <brotli/types.h>
#include <stdlib.h> /* free, malloc */
#include <string.h> /* memcpy, memset */
@@ -15,24 +17,26 @@
#include "../common/context.h"
#include "../common/platform.h"
#include "../common/version.h"
-#include "./backward_references.h"
-#include "./backward_references_hq.h"
-#include "./bit_cost.h"
-#include "./brotli_bit_stream.h"
-#include "./compress_fragment.h"
-#include "./compress_fragment_two_pass.h"
-#include "./encoder_dict.h"
-#include "./entropy_encode.h"
-#include "./fast_log.h"
-#include "./hash.h"
-#include "./histogram.h"
-#include "./memory.h"
-#include "./metablock.h"
-#include "./prefix.h"
-#include "./quality.h"
-#include "./ringbuffer.h"
-#include "./utf8_util.h"
-#include "./write_bits.h"
+#include "backward_references.h"
+#include "backward_references_hq.h"
+#include "bit_cost.h"
+#include "brotli_bit_stream.h"
+#include "compress_fragment.h"
+#include "compress_fragment_two_pass.h"
+#include "dictionary_hash.h"
+#include "encoder_dict.h"
+#include "entropy_encode.h"
+#include "fast_log.h"
+#include "hash.h"
+#include "histogram.h"
+#include "memory.h"
+#include "metablock.h"
+#include "prefix.h"
+#include "state.h"
+#include "quality.h"
+#include "ringbuffer.h"
+#include "utf8_util.h"
+#include "write_bits.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -40,94 +44,6 @@ extern "C" {
#define COPY_ARRAY(dst, src) memcpy(dst, src, sizeof(src));
-typedef enum BrotliEncoderStreamState {
- /* Default state. */
- BROTLI_STREAM_PROCESSING = 0,
- /* Intermediate state; after next block is emitted, byte-padding should be
- performed before getting back to default state. */
- BROTLI_STREAM_FLUSH_REQUESTED = 1,
- /* Last metablock was produced; no more input is acceptable. */
- BROTLI_STREAM_FINISHED = 2,
- /* Flushing compressed block and writing meta-data block header. */
- BROTLI_STREAM_METADATA_HEAD = 3,
- /* Writing metadata block body. */
- BROTLI_STREAM_METADATA_BODY = 4
-} BrotliEncoderStreamState;
-
-typedef enum BrotliEncoderFlintState {
- BROTLI_FLINT_NEEDS_2_BYTES = 2,
- BROTLI_FLINT_NEEDS_1_BYTE = 1,
- BROTLI_FLINT_WAITING_FOR_PROCESSING = 0,
- BROTLI_FLINT_WAITING_FOR_FLUSHING = -1,
- BROTLI_FLINT_DONE = -2
-} BrotliEncoderFlintState;
-
-typedef struct BrotliEncoderStateStruct {
- BrotliEncoderParams params;
-
- MemoryManager memory_manager_;
-
- uint64_t input_pos_;
- RingBuffer ringbuffer_;
- size_t cmd_alloc_size_;
- Command* commands_;
- size_t num_commands_;
- size_t num_literals_;
- size_t last_insert_len_;
- uint64_t last_flush_pos_;
- uint64_t last_processed_pos_;
- int dist_cache_[BROTLI_NUM_DISTANCE_SHORT_CODES];
- int saved_dist_cache_[4];
- uint16_t last_bytes_;
- uint8_t last_bytes_bits_;
- /* "Flint" is a tiny uncompressed block emitted before the continuation
- block to unwire literal context from previous data. Despite being int8_t,
- field is actually BrotliEncoderFlintState enum. */
- int8_t flint_;
- uint8_t prev_byte_;
- uint8_t prev_byte2_;
- size_t storage_size_;
- uint8_t* storage_;
-
- Hasher hasher_;
-
- /* Hash table for FAST_ONE_PASS_COMPRESSION_QUALITY mode. */
- int small_table_[1 << 10]; /* 4KiB */
- int* large_table_; /* Allocated only when needed */
- size_t large_table_size_;
- /* Command and distance prefix codes (each 64 symbols, stored back-to-back)
- used for the next block in FAST_ONE_PASS_COMPRESSION_QUALITY. The command
- prefix code is over a smaller alphabet with the following 64 symbols:
- 0 - 15: insert length code 0, copy length code 0 - 15, same distance
- 16 - 39: insert length code 0, copy length code 0 - 23
- 40 - 63: insert length code 0 - 23, copy length code 0
- Note that symbols 16 and 40 represent the same code in the full alphabet,
- but we do not use either of them in FAST_ONE_PASS_COMPRESSION_QUALITY. */
- uint8_t cmd_depths_[128];
- uint16_t cmd_bits_[128];
- /* The compressed form of the command and distance prefix codes for the next
- block in FAST_ONE_PASS_COMPRESSION_QUALITY. */
- uint8_t cmd_code_[512];
- size_t cmd_code_numbits_;
- /* Command and literal buffers for FAST_TWO_PASS_COMPRESSION_QUALITY. */
- uint32_t* command_buf_;
- uint8_t* literal_buf_;
-
- uint8_t* next_out_;
- size_t available_out_;
- size_t total_out_;
- /* Temporary buffer for padding flush bits or metadata block header / body. */
- union {
- uint64_t u64[2];
- uint8_t u8[16];
- } tiny_buf_;
- uint32_t remaining_metadata_bytes_;
- BrotliEncoderStreamState stream_state_;
-
- BROTLI_BOOL is_last_block_emitted_;
- BROTLI_BOOL is_initialized_;
-} BrotliEncoderStateStruct;
-
static size_t InputBlockSize(BrotliEncoderState* s) {
return (size_t)1 << s->params.lgblock;
}
@@ -147,7 +63,7 @@ BROTLI_BOOL BrotliEncoderSetParameter(
BrotliEncoderState* state, BrotliEncoderParameter p, uint32_t value) {
/* Changing parameters on the fly is not implemented yet. */
if (state->is_initialized_) return BROTLI_FALSE;
- /* TODO: Validate/clamp parameters here. */
+ /* TODO(eustas): Validate/clamp parameters here. */
switch (p) {
case BROTLI_PARAM_MODE:
state->params.mode = (BrotliEncoderMode)value;
@@ -283,11 +199,9 @@ static void EncodeWindowBits(int lgwin, BROTLI_BOOL large_window,
}
}
+/* TODO(eustas): move to compress_fragment.c? */
/* Initializes the command and distance prefix codes for the first block. */
-static void InitCommandPrefixCodes(uint8_t cmd_depths[128],
- uint16_t cmd_bits[128],
- uint8_t cmd_code[512],
- size_t* cmd_code_numbits) {
+static void InitCommandPrefixCodes(BrotliOnePassArena* s) {
static const uint8_t kDefaultCommandDepths[128] = {
0, 4, 4, 5, 6, 6, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8,
0, 0, 0, 4, 4, 4, 4, 4, 5, 5, 6, 6, 6, 6, 7, 7,
@@ -320,13 +234,13 @@ static void InitCommandPrefixCodes(uint8_t cmd_depths[128],
0x46, 0xe1, 0xb0, 0xd0, 0x4e, 0xb2, 0xf7, 0x04, 0x00,
};
static const size_t kDefaultCommandCodeNumBits = 448;
- COPY_ARRAY(cmd_depths, kDefaultCommandDepths);
- COPY_ARRAY(cmd_bits, kDefaultCommandBits);
+ COPY_ARRAY(s->cmd_depth, kDefaultCommandDepths);
+ COPY_ARRAY(s->cmd_bits, kDefaultCommandBits);
/* Initialize the pre-compressed form of the command and distance prefix
codes. */
- COPY_ARRAY(cmd_code, kDefaultCommandCode);
- *cmd_code_numbits = kDefaultCommandCodeNumBits;
+ COPY_ARRAY(s->cmd_code, kDefaultCommandCode);
+ s->cmd_code_numbits = kDefaultCommandCodeNumBits;
}
/* Decide about the context map based on the ability of the prediction
@@ -401,7 +315,8 @@ static void ChooseContextMap(int quality,
first 5 bits of literals. */
static BROTLI_BOOL ShouldUseComplexStaticContextMap(const uint8_t* input,
size_t start_pos, size_t length, size_t mask, int quality, size_t size_hint,
- size_t* num_literal_contexts, const uint32_t** literal_context_map) {
+ size_t* num_literal_contexts, const uint32_t** literal_context_map,
+ uint32_t* arena) {
static const uint32_t kStaticContextMapComplexUTF8[64] = {
11, 11, 12, 12, /* 0 special */
0, 0, 0, 0, /* 4 lf */
@@ -426,16 +341,17 @@ static BROTLI_BOOL ShouldUseComplexStaticContextMap(const uint8_t* input,
return BROTLI_FALSE;
} else {
const size_t end_pos = start_pos + length;
- /* To make entropy calculations faster and to fit on the stack, we collect
- histograms over the 5 most significant bits of literals. One histogram
+ /* To make entropy calculations faster, we collect histograms
+ over the 5 most significant bits of literals. One histogram
without context and 13 additional histograms for each context value. */
- uint32_t combined_histo[32] = { 0 };
- uint32_t context_histo[13][32] = { { 0 } };
+ uint32_t* BROTLI_RESTRICT const combined_histo = arena;
+ uint32_t* BROTLI_RESTRICT const context_histo = arena + 32;
uint32_t total = 0;
double entropy[3];
size_t dummy;
size_t i;
ContextLut utf8_lut = BROTLI_CONTEXT_LUT(CONTEXT_UTF8);
+ memset(arena, 0, sizeof(arena[0]) * 32 * 14);
for (; start_pos + 64 <= end_pos; start_pos += 4096) {
const size_t stride_end_pos = start_pos + 64;
uint8_t prev2 = input[start_pos & mask];
@@ -449,7 +365,7 @@ static BROTLI_BOOL ShouldUseComplexStaticContextMap(const uint8_t* input,
BROTLI_CONTEXT(prev1, prev2, utf8_lut)];
++total;
++combined_histo[literal >> 3];
- ++context_histo[context][literal >> 3];
+ ++context_histo[(context << 5) + (literal >> 3)];
prev2 = prev1;
prev1 = literal;
}
@@ -457,7 +373,7 @@ static BROTLI_BOOL ShouldUseComplexStaticContextMap(const uint8_t* input,
entropy[1] = ShannonEntropy(combined_histo, 32, &dummy);
entropy[2] = 0;
for (i = 0; i < 13; ++i) {
- entropy[2] += ShannonEntropy(&context_histo[i][0], 32, &dummy);
+ entropy[2] += ShannonEntropy(context_histo + (i << 5), 32, &dummy);
}
entropy[0] = 1.0 / (double)total;
entropy[1] *= entropy[0];
@@ -481,19 +397,21 @@ static BROTLI_BOOL ShouldUseComplexStaticContextMap(const uint8_t* input,
static void DecideOverLiteralContextModeling(const uint8_t* input,
size_t start_pos, size_t length, size_t mask, int quality, size_t size_hint,
- size_t* num_literal_contexts, const uint32_t** literal_context_map) {
+ size_t* num_literal_contexts, const uint32_t** literal_context_map,
+ uint32_t* arena) {
if (quality < MIN_QUALITY_FOR_CONTEXT_MODELING || length < 64) {
return;
} else if (ShouldUseComplexStaticContextMap(
input, start_pos, length, mask, quality, size_hint,
- num_literal_contexts, literal_context_map)) {
+ num_literal_contexts, literal_context_map, arena)) {
/* Context map was already set, nothing else to do. */
} else {
/* Gather bi-gram data of the UTF8 byte prefixes. To make the analysis of
UTF8 data faster we only examine 64 byte long strides at every 4kB
intervals. */
const size_t end_pos = start_pos + length;
- uint32_t bigram_prefix_histo[9] = { 0 };
+ uint32_t* BROTLI_RESTRICT const bigram_prefix_histo = arena;
+ memset(bigram_prefix_histo, 0, sizeof(arena[0]) * 9);
for (; start_pos + 64 <= end_pos; start_pos += 4096) {
static const int lut[4] = { 0, 0, 1, 2 };
const size_t stride_end_pos = start_pos + 64;
@@ -513,7 +431,7 @@ static void DecideOverLiteralContextModeling(const uint8_t* input,
static BROTLI_BOOL ShouldCompress(
const uint8_t* data, const size_t mask, const uint64_t last_flush_pos,
const size_t bytes, const size_t num_literals, const size_t num_commands) {
- /* TODO: find more precise minimal block overhead. */
+ /* TODO(eustas): find more precise minimal block overhead. */
if (bytes <= 2) return BROTLI_FALSE;
if (num_commands < (bytes >> 8) + 2) {
if ((double)num_literals > 0.99 * (double)bytes) {
@@ -613,10 +531,14 @@ static void WriteMetaBlockInternal(MemoryManager* m,
size_t num_literal_contexts = 1;
const uint32_t* literal_context_map = NULL;
if (!params->disable_literal_context_modeling) {
+ /* TODO(eustas): pull to higher level and reuse. */
+ uint32_t* arena = BROTLI_ALLOC(m, uint32_t, 14 * 32);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(arena)) return;
DecideOverLiteralContextModeling(
data, wrapped_last_flush_pos, bytes, mask, params->quality,
params->size_hint, &num_literal_contexts,
- &literal_context_map);
+ &literal_context_map, arena);
+ BROTLI_FREE(m, arena);
}
BrotliBuildMetaBlockGreedy(m, data, wrapped_last_flush_pos, mask,
prev_byte, prev_byte2, literal_context_lut, num_literal_contexts,
@@ -681,12 +603,13 @@ static void ChooseDistanceParams(BrotliEncoderParams* params) {
}
}
- BrotliInitDistanceParams(
- params, distance_postfix_bits, num_direct_distance_codes);
+ BrotliInitDistanceParams(&params->dist, distance_postfix_bits,
+ num_direct_distance_codes, params->large_window);
}
static BROTLI_BOOL EnsureInitialized(BrotliEncoderState* s) {
- if (BROTLI_IS_OOM(&s->memory_manager_)) return BROTLI_FALSE;
+ MemoryManager* m = &s->memory_manager_;
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
if (s->is_initialized_) return BROTLI_TRUE;
s->last_bytes_bits_ = 0;
@@ -728,8 +651,12 @@ static BROTLI_BOOL EnsureInitialized(BrotliEncoderState* s) {
}
if (s->params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY) {
- InitCommandPrefixCodes(s->cmd_depths_, s->cmd_bits_,
- s->cmd_code_, &s->cmd_code_numbits_);
+ s->one_pass_arena_ = BROTLI_ALLOC(m, BrotliOnePassArena, 1);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ InitCommandPrefixCodes(s->one_pass_arena_);
+ } else if (s->params.quality == FAST_TWO_PASS_COMPRESSION_QUALITY) {
+ s->two_pass_arena_ = BROTLI_ALLOC(m, BrotliTwoPassArena, 1);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
}
s->is_initialized_ = BROTLI_TRUE;
@@ -745,7 +672,7 @@ static void BrotliEncoderInitParams(BrotliEncoderParams* params) {
params->stream_offset = 0;
params->size_hint = 0;
params->disable_literal_context_modeling = BROTLI_FALSE;
- BrotliInitEncoderDictionary(&params->dictionary);
+ BrotliInitSharedEncoderDictionary(&params->dictionary);
params->dist.distance_postfix_bits = 0;
params->dist.num_direct_distance_codes = 0;
params->dist.alphabet_size_max =
@@ -754,6 +681,11 @@ static void BrotliEncoderInitParams(BrotliEncoderParams* params) {
params->dist.max_distance = BROTLI_MAX_DISTANCE;
}
+static void BrotliEncoderCleanupParams(MemoryManager* m,
+ BrotliEncoderParams* params) {
+ BrotliCleanupSharedEncoderDictionary(m, &params->dictionary);
+}
+
static void BrotliEncoderInitState(BrotliEncoderState* s) {
BrotliEncoderInitParams(&s->params);
s->input_pos_ = 0;
@@ -769,9 +701,11 @@ static void BrotliEncoderInitState(BrotliEncoderState* s) {
HasherInit(&s->hasher_);
s->large_table_ = NULL;
s->large_table_size_ = 0;
- s->cmd_code_numbits_ = 0;
+ s->one_pass_arena_ = NULL;
+ s->two_pass_arena_ = NULL;
s->command_buf_ = NULL;
s->literal_buf_ = NULL;
+ s->total_in_ = 0;
s->next_out_ = NULL;
s->available_out_ = 0;
s->total_out_ = 0;
@@ -796,13 +730,9 @@ static void BrotliEncoderInitState(BrotliEncoderState* s) {
BrotliEncoderState* BrotliEncoderCreateInstance(
brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
- BrotliEncoderState* state = 0;
- if (!alloc_func && !free_func) {
- state = (BrotliEncoderState*)malloc(sizeof(BrotliEncoderState));
- } else if (alloc_func && free_func) {
- state = (BrotliEncoderState*)alloc_func(opaque, sizeof(BrotliEncoderState));
- }
- if (state == 0) {
+ BrotliEncoderState* state = (BrotliEncoderState*)BrotliBootstrapAlloc(
+ sizeof(BrotliEncoderState), alloc_func, free_func, opaque);
+ if (state == NULL) {
/* BROTLI_DUMP(); */
return 0;
}
@@ -812,19 +742,36 @@ BrotliEncoderState* BrotliEncoderCreateInstance(
return state;
}
+#ifdef BROTLI_REPORTING
+/* When BROTLI_REPORTING is defined extra reporting module have to be linked. */
+void BrotliEncoderOnFinish(const BrotliEncoderState* s);
+#define BROTLI_ENCODER_ON_FINISH(s) BrotliEncoderOnFinish(s);
+#else
+#if !defined(BROTLI_ENCODER_ON_FINISH)
+#define BROTLI_ENCODER_ON_FINISH(s) (void)(s);
+#endif
+#endif
+
static void BrotliEncoderCleanupState(BrotliEncoderState* s) {
MemoryManager* m = &s->memory_manager_;
+
+ BROTLI_ENCODER_ON_FINISH(s);
+
if (BROTLI_IS_OOM(m)) {
BrotliWipeOutMemoryManager(m);
return;
}
+
BROTLI_FREE(m, s->storage_);
BROTLI_FREE(m, s->commands_);
RingBufferFree(m, &s->ringbuffer_);
DestroyHasher(m, &s->hasher_);
BROTLI_FREE(m, s->large_table_);
+ BROTLI_FREE(m, s->one_pass_arena_);
+ BROTLI_FREE(m, s->two_pass_arena_);
BROTLI_FREE(m, s->command_buf_);
BROTLI_FREE(m, s->literal_buf_);
+ BrotliEncoderCleanupParams(m, &s->params);
}
/* Deinitializes and frees BrotliEncoderState instance. */
@@ -832,11 +779,8 @@ void BrotliEncoderDestroyInstance(BrotliEncoderState* state) {
if (!state) {
return;
} else {
- MemoryManager* m = &state->memory_manager_;
- brotli_free_func free_func = m->free_func;
- void* opaque = m->opaque;
BrotliEncoderCleanupState(state);
- free_func(opaque, state);
+ BrotliBootstrapFree(state, &state->memory_manager_);
}
}
@@ -925,6 +869,8 @@ static void ExtendLastCommand(BrotliEncoderState* s, uint32_t* bytes,
uint64_t cmd_dist = (uint64_t)s->dist_cache_[0];
uint32_t distance_code = CommandRestoreDistanceCode(last_command,
&s->params.dist);
+ const CompoundDictionary* dict = &s->params.dictionary.compound;
+ size_t compound_dictionary_size = dict->total_size;
if (distance_code < BROTLI_NUM_DISTANCE_SHORT_CODES ||
distance_code - (BROTLI_NUM_DISTANCE_SHORT_CODES - 1) == cmd_dist) {
if (cmd_dist <= max_distance) {
@@ -935,6 +881,38 @@ static void ExtendLastCommand(BrotliEncoderState* s, uint32_t* bytes,
(*wrapped_last_processed_pos)++;
}
} else {
+ if ((cmd_dist - max_distance - 1) < compound_dictionary_size &&
+ last_copy_len < cmd_dist - max_distance) {
+ size_t address =
+ compound_dictionary_size - (size_t)(cmd_dist - max_distance) +
+ (size_t)last_copy_len;
+ size_t br_index = 0;
+ size_t br_offset;
+ const uint8_t* chunk;
+ size_t chunk_length;
+ while (address >= dict->chunk_offsets[br_index + 1]) br_index++;
+ br_offset = address - dict->chunk_offsets[br_index];
+ chunk = dict->chunk_source[br_index];
+ chunk_length =
+ dict->chunk_offsets[br_index + 1] - dict->chunk_offsets[br_index];
+ while (*bytes != 0 && data[*wrapped_last_processed_pos & mask] ==
+ chunk[br_offset]) {
+ last_command->copy_len_++;
+ (*bytes)--;
+ (*wrapped_last_processed_pos)++;
+ if (++br_offset == chunk_length) {
+ br_index++;
+ br_offset = 0;
+ if (br_index != dict->num_chunks) {
+ chunk = dict->chunk_source[br_index];
+ chunk_length = dict->chunk_offsets[br_index + 1] -
+ dict->chunk_offsets[br_index];
+ } else {
+ break;
+ }
+ }
+ }
+ }
}
/* The copy length is at most the metablock size, and thus expressible. */
GetLengthCode(last_command->insert_len_,
@@ -952,8 +930,8 @@ static void ExtendLastCommand(BrotliEncoderState* s, uint32_t* bytes,
If |*out_size| is positive, |*output| points to the start of the output
data. If |is_last| or |force_flush| is BROTLI_TRUE, an output meta-block is
always created. However, until |is_last| is BROTLI_TRUE encoder may retain up
- to 7 bits of the last byte of output. To force encoder to dump the remaining
- bits use WriteMetadata() to append an empty meta-data block.
+ to 7 bits of the last byte of output. Byte-alignment could be enforced by
+ emitting an empty meta-data block.
Returns BROTLI_FALSE if the size of the input data is larger than
input_block_size().
*/
@@ -968,10 +946,39 @@ static BROTLI_BOOL EncodeData(
MemoryManager* m = &s->memory_manager_;
ContextType literal_context_mode;
ContextLut literal_context_lut;
+ BROTLI_BOOL fast_compress =
+ s->params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY ||
+ s->params.quality == FAST_TWO_PASS_COMPRESSION_QUALITY;
data = s->ringbuffer_.buffer_;
mask = s->ringbuffer_.mask_;
+ if (delta == 0) { /* No new input; still might want to flush or finish. */
+ if (!data) { /* No input has been processed so far. */
+ if (is_last) { /* Emit complete finalized stream. */
+ BROTLI_DCHECK(s->last_bytes_bits_ <= 14);
+ s->last_bytes_ |= (uint16_t)(3u << s->last_bytes_bits_);
+ s->last_bytes_bits_ = (uint8_t)(s->last_bytes_bits_ + 2u);
+ s->tiny_buf_.u8[0] = (uint8_t)s->last_bytes_;
+ s->tiny_buf_.u8[1] = (uint8_t)(s->last_bytes_ >> 8);
+ *output = s->tiny_buf_.u8;
+ *out_size = (s->last_bytes_bits_ + 7u) >> 3u;
+ return BROTLI_TRUE;
+ } else { /* No data, not last -> no-op. */
+ *out_size = 0;
+ return BROTLI_TRUE;
+ }
+ } else {
+ /* Fast compress performs flush every block -> flush is no-op. */
+ if (!is_last && (!force_flush || fast_compress)) { /* Another no-op. */
+ *out_size = 0;
+ return BROTLI_TRUE;
+ }
+ }
+ }
+ BROTLI_DCHECK(data);
+
+ if (s->params.quality > s->params.dictionary.max_quality) return BROTLI_FALSE;
/* Adding more blocks after "last" block is forbidden. */
if (s->is_last_block_emitted_) return BROTLI_FALSE;
if (is_last) s->is_last_block_emitted_ = BROTLI_TRUE;
@@ -991,19 +998,12 @@ static BROTLI_BOOL EncodeData(
}
}
- if (s->params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY ||
- s->params.quality == FAST_TWO_PASS_COMPRESSION_QUALITY) {
+ if (fast_compress) {
uint8_t* storage;
size_t storage_ix = s->last_bytes_bits_;
size_t table_size;
int* table;
- if (delta == 0 && !is_last) {
- /* We have no new input data and we don't have to finish the stream, so
- nothing to do. */
- *out_size = 0;
- return BROTLI_TRUE;
- }
storage = GetBrotliStorage(s, 2 * bytes + 503);
if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
storage[0] = (uint8_t)s->last_bytes_;
@@ -1012,16 +1012,14 @@ static BROTLI_BOOL EncodeData(
if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
if (s->params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY) {
BrotliCompressFragmentFast(
- m, &data[wrapped_last_processed_pos & mask],
+ s->one_pass_arena_, &data[wrapped_last_processed_pos & mask],
bytes, is_last,
table, table_size,
- s->cmd_depths_, s->cmd_bits_,
- &s->cmd_code_numbits_, s->cmd_code_,
&storage_ix, storage);
if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
} else {
BrotliCompressFragmentTwoPass(
- m, &data[wrapped_last_processed_pos & mask],
+ s->two_pass_arena_, &data[wrapped_last_processed_pos & mask],
bytes, is_last,
s->command_buf_, s->literal_buf_,
table, table_size,
@@ -1099,7 +1097,7 @@ static BROTLI_BOOL EncodeData(
const size_t max_commands = max_length / 8;
const size_t processed_bytes = (size_t)(s->input_pos_ - s->last_flush_pos_);
/* If maximal possible additional block doesn't fit metablock, flush now. */
- /* TODO: Postpone decision until next block arrives? */
+ /* TODO(eustas): Postpone decision until next block arrives? */
const BROTLI_BOOL next_input_fits_metablock = TO_BROTLI_BOOL(
processed_bytes + InputBlockSize(s) <= max_length);
/* If block splitting is not used, then flush as soon as there is some
@@ -1192,7 +1190,7 @@ static size_t WriteMetadataHeader(
if (block_size == 0) {
BrotliWriteBits(2, 0, &storage_ix, header);
} else {
- uint32_t nbits = (block_size == 1) ? 0 :
+ uint32_t nbits = (block_size == 1) ? 1 :
(Log2FloorNonZero((uint32_t)block_size - 1) + 1);
uint32_t nbytes = (nbits + 7) / 8;
BrotliWriteBits(2, nbytes, &storage_ix, header);
@@ -1201,228 +1199,6 @@ static size_t WriteMetadataHeader(
return (storage_ix + 7u) >> 3;
}
-static BROTLI_BOOL BrotliCompressBufferQuality10(
- int lgwin, size_t input_size, const uint8_t* input_buffer,
- size_t* encoded_size, uint8_t* encoded_buffer) {
- MemoryManager memory_manager;
- MemoryManager* m = &memory_manager;
-
- const size_t mask = BROTLI_SIZE_MAX >> 1;
- int dist_cache[4] = { 4, 11, 15, 16 };
- int saved_dist_cache[4] = { 4, 11, 15, 16 };
- BROTLI_BOOL ok = BROTLI_TRUE;
- const size_t max_out_size = *encoded_size;
- size_t total_out_size = 0;
- uint16_t last_bytes;
- uint8_t last_bytes_bits;
-
- const size_t hasher_eff_size = BROTLI_MIN(size_t,
- input_size, BROTLI_MAX_BACKWARD_LIMIT(lgwin) + BROTLI_WINDOW_GAP);
-
- BrotliEncoderParams params;
-
- const int lgmetablock = BROTLI_MIN(int, 24, lgwin + 1);
- size_t max_block_size;
- const size_t max_metablock_size = (size_t)1 << lgmetablock;
- const size_t max_literals_per_metablock = max_metablock_size / 8;
- const size_t max_commands_per_metablock = max_metablock_size / 8;
- size_t metablock_start = 0;
- uint8_t prev_byte = 0;
- uint8_t prev_byte2 = 0;
-
- Hasher hasher;
- HasherInit(&hasher);
-
- BrotliEncoderInitParams(&params);
- params.quality = 10;
- params.lgwin = lgwin;
- if (lgwin > BROTLI_MAX_WINDOW_BITS) {
- params.large_window = BROTLI_TRUE;
- }
- SanitizeParams(&params);
- params.lgblock = ComputeLgBlock(&params);
- ChooseDistanceParams(&params);
- max_block_size = (size_t)1 << params.lgblock;
-
- BrotliInitMemoryManager(m, 0, 0, 0);
-
- BROTLI_DCHECK(input_size <= mask + 1);
- EncodeWindowBits(lgwin, params.large_window, &last_bytes, &last_bytes_bits);
- InitOrStitchToPreviousBlock(m, &hasher, input_buffer, mask, &params,
- 0, hasher_eff_size, BROTLI_TRUE);
- if (BROTLI_IS_OOM(m)) goto oom;
-
- while (ok && metablock_start < input_size) {
- const size_t metablock_end =
- BROTLI_MIN(size_t, input_size, metablock_start + max_metablock_size);
- const size_t expected_num_commands =
- (metablock_end - metablock_start) / 12 + 16;
- Command* commands = 0;
- size_t num_commands = 0;
- size_t last_insert_len = 0;
- size_t num_literals = 0;
- size_t metablock_size = 0;
- size_t cmd_alloc_size = 0;
- BROTLI_BOOL is_last;
- uint8_t* storage;
- size_t storage_ix;
-
- ContextType literal_context_mode = ChooseContextMode(&params,
- input_buffer, metablock_start, mask, metablock_end - metablock_start);
- ContextLut literal_context_lut = BROTLI_CONTEXT_LUT(literal_context_mode);
-
- size_t block_start;
- for (block_start = metablock_start; block_start < metablock_end; ) {
- size_t block_size =
- BROTLI_MIN(size_t, metablock_end - block_start, max_block_size);
- ZopfliNode* nodes = BROTLI_ALLOC(m, ZopfliNode, block_size + 1);
- size_t path_size;
- size_t new_cmd_alloc_size;
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(nodes)) goto oom;
- BrotliInitZopfliNodes(nodes, block_size + 1);
- StitchToPreviousBlockH10(&hasher.privat._H10, block_size, block_start,
- input_buffer, mask);
- path_size = BrotliZopfliComputeShortestPath(m, block_size, block_start,
- input_buffer, mask, literal_context_lut, &params, dist_cache, &hasher,
- nodes);
- if (BROTLI_IS_OOM(m)) goto oom;
- /* We allocate a command buffer in the first iteration of this loop that
- will be likely big enough for the whole metablock, so that for most
- inputs we will not have to reallocate in later iterations. We do the
- allocation here and not before the loop, because if the input is small,
- this will be allocated after the Zopfli cost model is freed, so this
- will not increase peak memory usage.
- TODO: If the first allocation is too small, increase command
- buffer size exponentially. */
- new_cmd_alloc_size = BROTLI_MAX(size_t, expected_num_commands,
- num_commands + path_size + 1);
- if (cmd_alloc_size != new_cmd_alloc_size) {
- Command* new_commands = BROTLI_ALLOC(m, Command, new_cmd_alloc_size);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(new_commands)) goto oom;
- cmd_alloc_size = new_cmd_alloc_size;
- if (commands) {
- memcpy(new_commands, commands, sizeof(Command) * num_commands);
- BROTLI_FREE(m, commands);
- }
- commands = new_commands;
- }
- BrotliZopfliCreateCommands(block_size, block_start, &nodes[0], dist_cache,
- &last_insert_len, &params, &commands[num_commands], &num_literals);
- num_commands += path_size;
- block_start += block_size;
- metablock_size += block_size;
- BROTLI_FREE(m, nodes);
- if (num_literals > max_literals_per_metablock ||
- num_commands > max_commands_per_metablock) {
- break;
- }
- }
-
- if (last_insert_len > 0) {
- InitInsertCommand(&commands[num_commands++], last_insert_len);
- num_literals += last_insert_len;
- }
-
- is_last = TO_BROTLI_BOOL(metablock_start + metablock_size == input_size);
- storage = NULL;
- storage_ix = last_bytes_bits;
-
- if (metablock_size == 0) {
- /* Write the ISLAST and ISEMPTY bits. */
- storage = BROTLI_ALLOC(m, uint8_t, 16);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(storage)) goto oom;
- storage[0] = (uint8_t)last_bytes;
- storage[1] = (uint8_t)(last_bytes >> 8);
- BrotliWriteBits(2, 3, &storage_ix, storage);
- storage_ix = (storage_ix + 7u) & ~7u;
- } else if (!ShouldCompress(input_buffer, mask, metablock_start,
- metablock_size, num_literals, num_commands)) {
- /* Restore the distance cache, as its last update by
- CreateBackwardReferences is now unused. */
- memcpy(dist_cache, saved_dist_cache, 4 * sizeof(dist_cache[0]));
- storage = BROTLI_ALLOC(m, uint8_t, metablock_size + 16);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(storage)) goto oom;
- storage[0] = (uint8_t)last_bytes;
- storage[1] = (uint8_t)(last_bytes >> 8);
- BrotliStoreUncompressedMetaBlock(is_last, input_buffer,
- metablock_start, mask, metablock_size,
- &storage_ix, storage);
- } else {
- MetaBlockSplit mb;
- BrotliEncoderParams block_params = params;
- InitMetaBlockSplit(&mb);
- BrotliBuildMetaBlock(m, input_buffer, metablock_start, mask,
- &block_params,
- prev_byte, prev_byte2,
- commands, num_commands,
- literal_context_mode,
- &mb);
- if (BROTLI_IS_OOM(m)) goto oom;
- {
- /* The number of distance symbols effectively used for distance
- histograms. It might be less than distance alphabet size
- for "Large Window Brotli" (32-bit). */
- BrotliOptimizeHistograms(block_params.dist.alphabet_size_limit, &mb);
- }
- storage = BROTLI_ALLOC(m, uint8_t, 2 * metablock_size + 503);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(storage)) goto oom;
- storage[0] = (uint8_t)last_bytes;
- storage[1] = (uint8_t)(last_bytes >> 8);
- BrotliStoreMetaBlock(m, input_buffer, metablock_start, metablock_size,
- mask, prev_byte, prev_byte2,
- is_last,
- &block_params,
- literal_context_mode,
- commands, num_commands,
- &mb,
- &storage_ix, storage);
- if (BROTLI_IS_OOM(m)) goto oom;
- if (metablock_size + 4 < (storage_ix >> 3)) {
- /* Restore the distance cache and last byte. */
- memcpy(dist_cache, saved_dist_cache, 4 * sizeof(dist_cache[0]));
- storage[0] = (uint8_t)last_bytes;
- storage[1] = (uint8_t)(last_bytes >> 8);
- storage_ix = last_bytes_bits;
- BrotliStoreUncompressedMetaBlock(is_last, input_buffer,
- metablock_start, mask,
- metablock_size, &storage_ix, storage);
- }
- DestroyMetaBlockSplit(m, &mb);
- }
- last_bytes = (uint16_t)(storage[storage_ix >> 3]);
- last_bytes_bits = storage_ix & 7u;
- metablock_start += metablock_size;
- if (metablock_start < input_size) {
- prev_byte = input_buffer[metablock_start - 1];
- prev_byte2 = input_buffer[metablock_start - 2];
- }
- /* Save the state of the distance cache in case we need to restore it for
- emitting an uncompressed block. */
- memcpy(saved_dist_cache, dist_cache, 4 * sizeof(dist_cache[0]));
-
- {
- const size_t out_size = storage_ix >> 3;
- total_out_size += out_size;
- if (total_out_size <= max_out_size) {
- memcpy(encoded_buffer, storage, out_size);
- encoded_buffer += out_size;
- } else {
- ok = BROTLI_FALSE;
- }
- }
- BROTLI_FREE(m, storage);
- BROTLI_FREE(m, commands);
- }
-
- *encoded_size = total_out_size;
- DestroyHasher(m, &hasher);
- return ok;
-
-oom:
- BrotliWipeOutMemoryManager(m);
- return BROTLI_FALSE;
-}
-
size_t BrotliEncoderMaxCompressedSize(size_t input_size) {
/* [window bits / empty metadata] + N * [uncompressed] + [last empty] */
size_t num_large_blocks = input_size >> 14;
@@ -1470,7 +1246,8 @@ static size_t MakeUncompressedStream(
BROTLI_BOOL BrotliEncoderCompress(
int quality, int lgwin, BrotliEncoderMode mode, size_t input_size,
- const uint8_t input_buffer[BROTLI_ARRAY_PARAM(input_size)], size_t* encoded_size,
+ const uint8_t input_buffer[BROTLI_ARRAY_PARAM(input_size)],
+ size_t* encoded_size,
uint8_t encoded_buffer[BROTLI_ARRAY_PARAM(*encoded_size)]) {
BrotliEncoderState* s;
size_t out_size = *encoded_size;
@@ -1487,17 +1264,6 @@ BROTLI_BOOL BrotliEncoderCompress(
*encoded_buffer = 6;
return BROTLI_TRUE;
}
- if (quality == 10) {
- /* TODO: Implement this direct path for all quality levels. */
- const int lg_win = BROTLI_MIN(int, BROTLI_LARGE_MAX_WINDOW_BITS,
- BROTLI_MAX(int, 16, lgwin));
- int ok = BrotliCompressBufferQuality10(lg_win, input_size, input_buffer,
- encoded_size, encoded_buffer);
- if (!ok || (max_out_size && *encoded_size > max_out_size)) {
- goto fallback;
- }
- return BROTLI_TRUE;
- }
s = BrotliEncoderCreateInstance(0, 0, 0);
if (!s) {
@@ -1509,6 +1275,7 @@ BROTLI_BOOL BrotliEncoderCompress(
uint8_t* next_out = encoded_buffer;
size_t total_out = 0;
BROTLI_BOOL result = BROTLI_FALSE;
+ /* TODO(eustas): check that parameters are sane. */
BrotliEncoderSetParameter(s, BROTLI_PARAM_QUALITY, (uint32_t)quality);
BrotliEncoderSetParameter(s, BROTLI_PARAM_LGWIN, (uint32_t)lgwin);
BrotliEncoderSetParameter(s, BROTLI_PARAM_MODE, (uint32_t)mode);
@@ -1560,6 +1327,18 @@ static void InjectBytePaddingBlock(BrotliEncoderState* s) {
s->available_out_ += (seal_bits + 7) >> 3;
}
+/* Fills the |total_out|, if it is not NULL. */
+static void SetTotalOut(BrotliEncoderState* s, size_t* total_out) {
+ if (total_out) {
+ /* Saturating conversion uint64_t -> size_t */
+ size_t result = (size_t)-1;
+ if (s->total_out_ < result) {
+ result = (size_t)s->total_out_;
+ }
+ *total_out = result;
+ }
+}
+
/* Injects padding bits or pushes compressed data to output.
Returns false if nothing is done. */
static BROTLI_BOOL InjectFlushOrPushOutput(BrotliEncoderState* s,
@@ -1579,7 +1358,7 @@ static BROTLI_BOOL InjectFlushOrPushOutput(BrotliEncoderState* s,
s->next_out_ += copy_output_size;
s->available_out_ -= copy_output_size;
s->total_out_ += copy_output_size;
- if (total_out) *total_out = s->total_out_;
+ SetTotalOut(s, total_out);
return BROTLI_TRUE;
}
@@ -1676,19 +1455,19 @@ static BROTLI_BOOL BrotliEncoderCompressStreamFast(
if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
if (s->params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY) {
- BrotliCompressFragmentFast(m, *next_in, block_size, is_last, table,
- table_size, s->cmd_depths_, s->cmd_bits_, &s->cmd_code_numbits_,
- s->cmd_code_, &storage_ix, storage);
+ BrotliCompressFragmentFast(s->one_pass_arena_, *next_in, block_size,
+ is_last, table, table_size, &storage_ix, storage);
if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
} else {
- BrotliCompressFragmentTwoPass(m, *next_in, block_size, is_last,
- command_buf, literal_buf, table, table_size,
+ BrotliCompressFragmentTwoPass(s->two_pass_arena_, *next_in, block_size,
+ is_last, command_buf, literal_buf, table, table_size,
&storage_ix, storage);
if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
}
if (block_size != 0) {
*next_in += block_size;
*available_in -= block_size;
+ s->total_in_ += block_size;
}
if (inplace) {
size_t out_bytes = storage_ix >> 3;
@@ -1697,7 +1476,7 @@ static BROTLI_BOOL BrotliEncoderCompressStreamFast(
*next_out += out_bytes;
*available_out -= out_bytes;
s->total_out_ += out_bytes;
- if (total_out) *total_out = s->total_out_;
+ SetTotalOut(s, total_out);
} else {
size_t out_bytes = storage_ix >> 3;
s->next_out_ = storage;
@@ -1766,6 +1545,7 @@ static BROTLI_BOOL ProcessMetadata(
memcpy(*next_out, *next_in, copy);
*next_in += copy;
*available_in -= copy;
+ s->total_in_ += copy; /* not actually data input, though */
s->remaining_metadata_bytes_ -= copy;
*next_out += copy;
*available_out -= copy;
@@ -1776,6 +1556,7 @@ static BROTLI_BOOL ProcessMetadata(
memcpy(s->next_out_, *next_in, copy);
*next_in += copy;
*available_in -= copy;
+ s->total_in_ += copy; /* not actually data input, though */
s->remaining_metadata_bytes_ -= copy;
s->available_out_ = copy;
}
@@ -1803,7 +1584,7 @@ static void UpdateSizeHint(BrotliEncoderState* s, size_t available_in) {
BROTLI_BOOL BrotliEncoderCompressStream(
BrotliEncoderState* s, BrotliEncoderOperation op, size_t* available_in,
- const uint8_t** next_in, size_t* available_out,uint8_t** next_out,
+ const uint8_t** next_in, size_t* available_out, uint8_t** next_out,
size_t* total_out) {
if (!EnsureInitialized(s)) return BROTLI_FALSE;
@@ -1845,6 +1626,7 @@ BROTLI_BOOL BrotliEncoderCompressStream(
CopyInputToRingBuffer(s, copy_input_size, *next_in);
*next_in += copy_input_size;
*available_in -= copy_input_size;
+ s->total_in_ += copy_input_size;
if (s->flint_ > 0) s->flint_ = (int8_t)(s->flint_ - (int)copy_input_size);
continue;
}
@@ -1922,6 +1704,293 @@ uint32_t BrotliEncoderVersion(void) {
return BROTLI_VERSION;
}
+BrotliEncoderPreparedDictionary* BrotliEncoderPrepareDictionary(
+ BrotliSharedDictionaryType type, size_t size,
+ const uint8_t data[BROTLI_ARRAY_PARAM(size)], int quality,
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
+ ManagedDictionary* managed_dictionary = NULL;
+ BROTLI_BOOL type_is_known = BROTLI_FALSE;
+ type_is_known |= (type == BROTLI_SHARED_DICTIONARY_RAW);
+#if defined(BROTLI_EXPERIMENTAL)
+ type_is_known |= (type == BROTLI_SHARED_DICTIONARY_SERIALIZED);
+#endif /* BROTLI_EXPERIMENTAL */
+ if (!type_is_known) {
+ return NULL;
+ }
+ managed_dictionary =
+ BrotliCreateManagedDictionary(alloc_func, free_func, opaque);
+ if (managed_dictionary == NULL) {
+ return NULL;
+ }
+ if (type == BROTLI_SHARED_DICTIONARY_RAW) {
+ managed_dictionary->dictionary = (uint32_t*)CreatePreparedDictionary(
+ &managed_dictionary->memory_manager_, data, size);
+ }
+#if defined(BROTLI_EXPERIMENTAL)
+ if (type == BROTLI_SHARED_DICTIONARY_SERIALIZED) {
+ SharedEncoderDictionary* dict = (SharedEncoderDictionary*)BrotliAllocate(
+ &managed_dictionary->memory_manager_, sizeof(SharedEncoderDictionary));
+ managed_dictionary->dictionary = (uint32_t*)dict;
+ if (dict != NULL) {
+ BROTLI_BOOL ok = BrotliInitCustomSharedEncoderDictionary(
+ &managed_dictionary->memory_manager_, data, size, quality, dict);
+ if (!ok) {
+ BrotliFree(&managed_dictionary->memory_manager_, dict);
+ managed_dictionary->dictionary = NULL;
+ }
+ }
+ }
+#else /* BROTLI_EXPERIMENTAL */
+ (void)quality;
+#endif /* BROTLI_EXPERIMENTAL */
+ if (managed_dictionary->dictionary == NULL) {
+ BrotliDestroyManagedDictionary(managed_dictionary);
+ return NULL;
+ }
+ return (BrotliEncoderPreparedDictionary*)managed_dictionary;
+}
+
+void BrotliEncoderDestroyPreparedDictionary(
+ BrotliEncoderPreparedDictionary* dictionary) {
+ ManagedDictionary* dict = (ManagedDictionary*)dictionary;
+ if (!dictionary) return;
+ /* First field of dictionary structs. */
+ /* Only managed dictionaries are eligible for destruction by this method. */
+ if (dict->magic != kManagedDictionaryMagic) {
+ return;
+ }
+ if (dict->dictionary == NULL) {
+ /* This should never ever happen. */
+ } else if (*dict->dictionary == kLeanPreparedDictionaryMagic) {
+ DestroyPreparedDictionary(
+ &dict->memory_manager_, (PreparedDictionary*)dict->dictionary);
+ } else if (*dict->dictionary == kSharedDictionaryMagic) {
+ BrotliCleanupSharedEncoderDictionary(&dict->memory_manager_,
+ (SharedEncoderDictionary*)dict->dictionary);
+ BrotliFree(&dict->memory_manager_, dict->dictionary);
+ } else {
+ /* There is also kPreparedDictionaryMagic, but such instances should be
+ * constructed and destroyed by different means. */
+ }
+ dict->dictionary = NULL;
+ BrotliDestroyManagedDictionary(dict);
+}
+
+BROTLI_BOOL BrotliEncoderAttachPreparedDictionary(BrotliEncoderState* state,
+ const BrotliEncoderPreparedDictionary* dictionary) {
+ /* First field of dictionary structs */
+ const BrotliEncoderPreparedDictionary* dict = dictionary;
+ uint32_t magic = *((const uint32_t*)dict);
+ SharedEncoderDictionary* current = NULL;
+ if (magic == kManagedDictionaryMagic) {
+ /* Unwrap managed dictionary. */
+ ManagedDictionary* managed_dictionary = (ManagedDictionary*)dict;
+ magic = *managed_dictionary->dictionary;
+ dict = (BrotliEncoderPreparedDictionary*)managed_dictionary->dictionary;
+ }
+ current = &state->params.dictionary;
+ if (magic == kPreparedDictionaryMagic ||
+ magic == kLeanPreparedDictionaryMagic) {
+ const PreparedDictionary* prepared = (const PreparedDictionary*)dict;
+ if (!AttachPreparedDictionary(&current->compound, prepared)) {
+ return BROTLI_FALSE;
+ }
+ } else if (magic == kSharedDictionaryMagic) {
+ const SharedEncoderDictionary* attached =
+ (const SharedEncoderDictionary*)dict;
+ BROTLI_BOOL was_default = !current->contextual.context_based &&
+ current->contextual.num_dictionaries == 1 &&
+ current->contextual.dict[0]->hash_table_words ==
+ kStaticDictionaryHashWords &&
+ current->contextual.dict[0]->hash_table_lengths ==
+ kStaticDictionaryHashLengths;
+ BROTLI_BOOL new_default = !attached->contextual.context_based &&
+ attached->contextual.num_dictionaries == 1 &&
+ attached->contextual.dict[0]->hash_table_words ==
+ kStaticDictionaryHashWords &&
+ attached->contextual.dict[0]->hash_table_lengths ==
+ kStaticDictionaryHashLengths;
+ size_t i;
+ if (state->is_initialized_) return BROTLI_FALSE;
+ current->max_quality =
+ BROTLI_MIN(int, current->max_quality, attached->max_quality);
+ for (i = 0; i < attached->compound.num_chunks; i++) {
+ if (!AttachPreparedDictionary(&current->compound,
+ attached->compound.chunks[i])) {
+ return BROTLI_FALSE;
+ }
+ }
+ if (!new_default) {
+ if (!was_default) return BROTLI_FALSE;
+ /* Copy by value, but then set num_instances_ to 0 because their memory
+ is managed by attached, not by current */
+ current->contextual = attached->contextual;
+ current->contextual.num_instances_ = 0;
+ }
+ } else {
+ return BROTLI_FALSE;
+ }
+ return BROTLI_TRUE;
+}
+
+size_t BrotliEncoderEstimatePeakMemoryUsage(int quality, int lgwin,
+ size_t input_size) {
+ BrotliEncoderParams params;
+ size_t memory_manager_slots = BROTLI_ENCODER_MEMORY_MANAGER_SLOTS;
+ size_t memory_manager_size = memory_manager_slots * sizeof(void*);
+ BrotliEncoderInitParams(&params);
+ params.quality = quality;
+ params.lgwin = lgwin;
+ params.size_hint = input_size;
+ params.large_window = lgwin > BROTLI_MAX_WINDOW_BITS;
+ SanitizeParams(&params);
+ params.lgblock = ComputeLgBlock(&params);
+ ChooseHasher(&params, &params.hasher);
+ if (params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY ||
+ params.quality == FAST_TWO_PASS_COMPRESSION_QUALITY) {
+ size_t state_size = sizeof(BrotliEncoderState);
+ size_t block_size = BROTLI_MIN(size_t, input_size, (1ul << params.lgwin));
+ size_t hash_table_size =
+ HashTableSize(MaxHashTableSize(params.quality), block_size);
+ size_t hash_size =
+ (hash_table_size < (1u << 10)) ? 0 : sizeof(int) * hash_table_size;
+ size_t cmdbuf_size = params.quality == FAST_TWO_PASS_COMPRESSION_QUALITY ?
+ 5 * BROTLI_MIN(size_t, block_size, 1ul << 17) : 0;
+ if (params.quality == FAST_ONE_PASS_COMPRESSION_QUALITY) {
+ state_size += sizeof(BrotliOnePassArena);
+ } else {
+ state_size += sizeof(BrotliTwoPassArena);
+ }
+ return hash_size + cmdbuf_size + state_size;
+ } else {
+ size_t short_ringbuffer_size = (size_t)1 << params.lgblock;
+ int ringbuffer_bits = ComputeRbBits(&params);
+ size_t ringbuffer_size = input_size < short_ringbuffer_size ?
+ input_size : (1u << ringbuffer_bits) + short_ringbuffer_size;
+ size_t hash_size[4] = {0};
+ size_t metablock_size =
+ BROTLI_MIN(size_t, input_size, MaxMetablockSize(&params));
+ size_t inputblock_size =
+ BROTLI_MIN(size_t, input_size, (size_t)1 << params.lgblock);
+ size_t cmdbuf_size = metablock_size * 2 + inputblock_size * 6;
+ size_t outbuf_size = metablock_size * 2 + 503;
+ size_t histogram_size = 0;
+ HasherSize(&params, BROTLI_TRUE, input_size, hash_size);
+ if (params.quality < MIN_QUALITY_FOR_BLOCK_SPLIT) {
+ cmdbuf_size = BROTLI_MIN(size_t, cmdbuf_size,
+ MAX_NUM_DELAYED_SYMBOLS * sizeof(Command) + inputblock_size * 12);
+ }
+ if (params.quality >= MIN_QUALITY_FOR_HQ_BLOCK_SPLITTING) {
+ /* Only a very rough estimation, based on enwik8. */
+ histogram_size = 200 << 20;
+ } else if (params.quality >= MIN_QUALITY_FOR_BLOCK_SPLIT) {
+ size_t literal_histograms =
+ BROTLI_MIN(size_t, metablock_size / 6144, 256);
+ size_t command_histograms =
+ BROTLI_MIN(size_t, metablock_size / 6144, 256);
+ size_t distance_histograms =
+ BROTLI_MIN(size_t, metablock_size / 6144, 256);
+ histogram_size = literal_histograms * sizeof(HistogramLiteral) +
+ command_histograms * sizeof(HistogramCommand) +
+ distance_histograms * sizeof(HistogramDistance);
+ }
+ return (memory_manager_size + ringbuffer_size +
+ hash_size[0] + hash_size[1] + hash_size[2] + hash_size[3] +
+ cmdbuf_size +
+ outbuf_size +
+ histogram_size);
+ }
+}
+size_t BrotliEncoderGetPreparedDictionarySize(
+ const BrotliEncoderPreparedDictionary* prepared_dictionary) {
+ /* First field of dictionary structs */
+ const BrotliEncoderPreparedDictionary* prepared = prepared_dictionary;
+ uint32_t magic = *((const uint32_t*)prepared);
+ size_t overhead = 0;
+ if (magic == kManagedDictionaryMagic) {
+ const ManagedDictionary* managed = (const ManagedDictionary*)prepared;
+ overhead = sizeof(ManagedDictionary);
+ magic = *managed->dictionary;
+ prepared = (const BrotliEncoderPreparedDictionary*)managed->dictionary;
+ }
+
+ if (magic == kPreparedDictionaryMagic) {
+ const PreparedDictionary* dictionary =
+ (const PreparedDictionary*)prepared;
+ /* Keep in sync with step 3 of CreatePreparedDictionary */
+ return sizeof(PreparedDictionary) + dictionary->source_size +
+ (sizeof(uint32_t) << dictionary->slot_bits) +
+ (sizeof(uint16_t) << dictionary->bucket_bits) +
+ (sizeof(uint32_t) * dictionary->num_items) + overhead;
+ } else if (magic == kLeanPreparedDictionaryMagic) {
+ const PreparedDictionary* dictionary =
+ (const PreparedDictionary*)prepared;
+ /* Keep in sync with step 3 of CreatePreparedDictionary */
+ return sizeof(PreparedDictionary) + sizeof(uint8_t*) +
+ (sizeof(uint32_t) << dictionary->slot_bits) +
+ (sizeof(uint16_t) << dictionary->bucket_bits) +
+ (sizeof(uint32_t) * dictionary->num_items) + overhead;
+ } else if (magic == kSharedDictionaryMagic) {
+ const SharedEncoderDictionary* dictionary =
+ (const SharedEncoderDictionary*)prepared;
+ const CompoundDictionary* compound = &dictionary->compound;
+ const ContextualEncoderDictionary* contextual = &dictionary->contextual;
+ size_t result = sizeof(*dictionary);
+ size_t i;
+ size_t num_instances;
+ const BrotliEncoderDictionary* instances;
+ for (i = 0; i < compound->num_prepared_instances_; i++) {
+ size_t size = BrotliEncoderGetPreparedDictionarySize(
+ (const BrotliEncoderPreparedDictionary*)
+ compound->prepared_instances_[i]);
+ if (!size) return 0; /* error */
+ result += size;
+ }
+ if (contextual->context_based) {
+ num_instances = contextual->num_instances_;
+ instances = contextual->instances_;
+ result += sizeof(*instances) * num_instances;
+ } else {
+ num_instances = 1;
+ instances = &contextual->instance_;
+ }
+ for (i = 0; i < num_instances; i++) {
+ const BrotliEncoderDictionary* dict = &instances[i];
+ result += dict->trie.pool_capacity * sizeof(BrotliTrieNode);
+ if (dict->hash_table_data_words_) {
+ result += sizeof(kStaticDictionaryHashWords);
+ }
+ if (dict->hash_table_data_lengths_) {
+ result += sizeof(kStaticDictionaryHashLengths);
+ }
+ if (dict->buckets_data_) {
+ result += sizeof(*dict->buckets_data_) * dict->buckets_alloc_size_;
+ }
+ if (dict->dict_words_data_) {
+ result += sizeof(*dict->dict_words) * dict->dict_words_alloc_size_;
+ }
+ if (dict->words_instance_) {
+ result += sizeof(*dict->words_instance_);
+ /* data_size not added here: it is never allocated by the
+ SharedEncoderDictionary, instead it always points to the file
+ already loaded in memory. So if the caller wants to include
+ this memory as well, add the size of the loaded dictionary
+ file to this. */
+ }
+ }
+ return result + overhead;
+ }
+ return 0; /* error */
+}
+
+#if defined(BROTLI_TEST)
+size_t MakeUncompressedStreamForTest(const uint8_t*, size_t, uint8_t*);
+size_t MakeUncompressedStreamForTest(
+ const uint8_t* input, size_t input_size, uint8_t* output) {
+ return MakeUncompressedStream(input, input_size, output);
+}
+#endif
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/encoder_dict.c b/dviware/dvisvgm/libs/brotli/enc/encoder_dict.c
index c9e963b89d..6602c55a24 100644
--- a/dviware/dvisvgm/libs/brotli/enc/encoder_dict.c
+++ b/dviware/dvisvgm/libs/brotli/enc/encoder_dict.c
@@ -4,18 +4,45 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./encoder_dict.h"
+#include "encoder_dict.h"
+
+#include <stdlib.h> /* malloc, free */
#include "../common/dictionary.h"
+#include "../common/platform.h"
+#include "../common/shared_dictionary_internal.h"
#include "../common/transform.h"
-#include "./dictionary_hash.h"
-#include "./hash.h"
+#include "compound_dictionary.h"
+#include "dictionary_hash.h"
+#include "memory.h"
+#include "quality.h"
+#include "hash.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
-void BrotliInitEncoderDictionary(BrotliEncoderDictionary* dict) {
+#define NUM_HASH_BITS 15u
+#define NUM_HASH_BUCKETS (1u << NUM_HASH_BITS)
+
+static void BrotliTrieInit(BrotliTrie* trie) {
+ trie->pool_capacity = 0;
+ trie->pool_size = 0;
+ trie->pool = 0;
+
+ /* Set up the root node */
+ trie->root.single = 0;
+ trie->root.len_ = 0;
+ trie->root.idx_ = 0;
+ trie->root.sub = 0;
+}
+
+static void BrotliTrieFree(MemoryManager* m, BrotliTrie* trie) {
+ BrotliFree(m, trie->pool);
+}
+
+/* Initializes to RFC 7932 static dictionary / transforms. */
+static void InitEncoderDictionary(BrotliEncoderDictionary* dict) {
dict->words = BrotliGetDictionary();
dict->num_transforms = (uint32_t)BrotliGetTransforms()->num_transforms;
@@ -26,8 +53,588 @@ void BrotliInitEncoderDictionary(BrotliEncoderDictionary* dict) {
dict->cutoffTransformsCount = kCutoffTransformsCount;
dict->cutoffTransforms = kCutoffTransforms;
+
+ dict->parent = 0;
+
+ dict->hash_table_data_words_ = 0;
+ dict->hash_table_data_lengths_ = 0;
+ dict->buckets_alloc_size_ = 0;
+ dict->buckets_data_ = 0;
+ dict->dict_words_alloc_size_ = 0;
+ dict->dict_words_data_ = 0;
+ dict->words_instance_ = 0;
+ dict->has_words_heavy = BROTLI_FALSE;
+ BrotliTrieInit(&dict->trie);
+}
+
+static void BrotliDestroyEncoderDictionary(MemoryManager* m,
+ BrotliEncoderDictionary* dict) {
+ BrotliFree(m, dict->hash_table_data_words_);
+ BrotliFree(m, dict->hash_table_data_lengths_);
+ BrotliFree(m, dict->buckets_data_);
+ BrotliFree(m, dict->dict_words_data_);
+ BrotliFree(m, dict->words_instance_);
+ BrotliTrieFree(m, &dict->trie);
+}
+
+#if defined(BROTLI_EXPERIMENTAL)
+/* Word length must be at least 4 bytes */
+static uint32_t Hash(const uint8_t* data, int bits) {
+ uint32_t h = BROTLI_UNALIGNED_LOAD32LE(data) * kHashMul32;
+ /* The higher bits contain more mixture from the multiplication,
+ so we take our results from there. */
+ return h >> (32 - bits);
+}
+
+/* Theoretical max possible word size after transform */
+#define kTransformedBufferSize \
+ (256 + 256 + SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH)
+
+/* To be safe buffer must have at least kTransformedBufferSize */
+static void TransformedDictionaryWord(uint32_t word_idx, int len, int transform,
+ const BrotliTransforms* transforms,
+ const BrotliEncoderDictionary* dict,
+ uint8_t* buffer, size_t* size) {
+ const uint8_t* dict_word = &dict->words->data[
+ dict->words->offsets_by_length[len] + (uint32_t)len * word_idx];
+ *size = (size_t)BrotliTransformDictionaryWord(buffer, dict_word, len,
+ transforms, transform);
+}
+
+static DictWord MakeDictWord(uint8_t len, uint8_t transform, uint16_t idx) {
+ DictWord result;
+ result.len = len;
+ result.transform = transform;
+ result.idx = idx;
+ return result;
+}
+
+static uint32_t BrotliTrieAlloc(MemoryManager* m, size_t num, BrotliTrie* trie,
+ BrotliTrieNode** keep) {
+ uint32_t result;
+ uint32_t keep_index = 0;
+ if (keep && *keep != &trie->root) {
+ /* Optional node to keep, since address may change after re-allocating */
+ keep_index = (uint32_t)(*keep - trie->pool);
+ }
+ if (trie->pool_size == 0) {
+ /* Have a dummy node in the front. We do not want the result to be 0, it
+ must be at least 1, 0 represents "null pointer" */
+ trie->pool_size = 1;
+ }
+ BROTLI_ENSURE_CAPACITY(m, BrotliTrieNode, trie->pool, trie->pool_capacity,
+ trie->pool_size + num);
+ if (BROTLI_IS_OOM(m)) return 0;
+ /* Init the new nodes to empty */
+ memset(trie->pool + trie->pool_size, 0, sizeof(*trie->pool) * num);
+ result = (uint32_t)trie->pool_size;
+ trie->pool_size += num;
+ if (keep && *keep != &trie->root) {
+ *keep = trie->pool + keep_index;
+ }
+ return result;
+}
+
+/**
+ * len and idx: payload for last node
+ * word, size: the string
+ * index: position in the string
+ */
+static BROTLI_BOOL BrotliTrieNodeAdd(MemoryManager* m, uint8_t len,
+ uint32_t idx, const uint8_t* word, size_t size, int index,
+ BrotliTrieNode* node, BrotliTrie* trie) {
+ BrotliTrieNode* child = 0;
+ uint8_t c;
+ if ((size_t)index == size) {
+ if (!node->len_ || idx < node->idx_) {
+ node->len_ = len;
+ node->idx_ = idx;
+ }
+ return BROTLI_TRUE;
+ }
+ c = word[index];
+ if (node->single && c != node->c) {
+ BrotliTrieNode old = trie->pool[node->sub];
+ uint32_t new_nodes = BrotliTrieAlloc(m, 32, trie, &node);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ node->single = 0;
+ node->sub = new_nodes;
+ trie->pool[node->sub + (node->c >> 4)].sub = new_nodes + 16;
+ trie->pool[trie->pool[node->sub + (node->c >> 4)].sub + (node->c & 15)] =
+ old;
+ }
+ if (!node->sub) {
+ uint32_t new_node = BrotliTrieAlloc(m, 1, trie, &node);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ node->single = 1;
+ node->c = c;
+ node->sub = new_node;
+ }
+ if (node->single) {
+ child = &trie->pool[node->sub];
+ } else {
+ if (!trie->pool[node->sub + (c >> 4)].sub) {
+ uint32_t new_nodes = BrotliTrieAlloc(m, 16, trie, &node);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ trie->pool[node->sub + (c >> 4)].sub = new_nodes;
+ }
+ child = &trie->pool[trie->pool[node->sub + (c >> 4)].sub + (c & 15)];
+ }
+ return BrotliTrieNodeAdd(m, len, idx, word, size, index + 1, child, trie);
+}
+
+static BROTLI_BOOL BrotliTrieAdd(MemoryManager* m, uint8_t len, uint32_t idx,
+ const uint8_t* word, size_t size, BrotliTrie* trie) {
+ return BrotliTrieNodeAdd(m, len, idx, word, size, 0, &trie->root, trie);
+}
+
+const BrotliTrieNode* BrotliTrieSub(const BrotliTrie* trie,
+ const BrotliTrieNode* node, uint8_t c) {
+ BrotliTrieNode* temp_node;
+ if (node->single) {
+ if (node->c == c) return &trie->pool[node->sub];
+ return 0;
+ }
+ if (!node->sub) return 0;
+ temp_node = &trie->pool[node->sub + (c >> 4)];
+ if (!temp_node->sub) return 0;
+ return &trie->pool[temp_node->sub + (c & 15)];
+}
+
+static const BrotliTrieNode* BrotliTrieFind(const BrotliTrie* trie,
+ const uint8_t* word, size_t size) {
+ const BrotliTrieNode* node = &trie->root;
+ size_t i;
+ for (i = 0; i < size; i++) {
+ node = BrotliTrieSub(trie, node, word[i]);
+ if (!node) return 0;
+ }
+ return node;
+}
+
+static BROTLI_BOOL BuildDictionaryLut(MemoryManager* m,
+ const BrotliTransforms* transforms,
+ BrotliEncoderDictionary* dict) {
+ uint32_t i;
+ DictWord* dict_words;
+ uint16_t* buckets;
+ DictWord** words_by_hash;
+ size_t* words_by_hash_size;
+ size_t* words_by_hash_capacity;
+ BrotliTrie dedup;
+ uint8_t word[kTransformedBufferSize];
+ size_t word_size;
+ size_t total = 0;
+ uint8_t l;
+ uint16_t idx;
+
+ BrotliTrieInit(&dedup);
+
+ words_by_hash = (DictWord**)BrotliAllocate(m,
+ sizeof(*words_by_hash) * NUM_HASH_BUCKETS);
+ words_by_hash_size = (size_t*)BrotliAllocate(m,
+ sizeof(*words_by_hash_size) * NUM_HASH_BUCKETS);
+ words_by_hash_capacity = (size_t*)BrotliAllocate(m,
+ sizeof(*words_by_hash_capacity) * NUM_HASH_BUCKETS);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ memset(words_by_hash, 0, sizeof(*words_by_hash) * NUM_HASH_BUCKETS);
+ memset(words_by_hash_size, 0, sizeof(*words_by_hash_size) * NUM_HASH_BUCKETS);
+ memset(words_by_hash_capacity, 0,
+ sizeof(*words_by_hash_capacity) * NUM_HASH_BUCKETS);
+
+ if (transforms->num_transforms > 0) {
+ for (l = SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH;
+ l <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH; ++l) {
+ uint16_t n = dict->words->size_bits_by_length[l] ?
+ (uint16_t)(1 << dict->words->size_bits_by_length[l]) : 0u;
+ for (idx = 0; idx < n; ++idx) {
+ uint32_t key;
+ /* First transform (usually identity) */
+ TransformedDictionaryWord(idx, l, 0, transforms, dict, word,
+ &word_size);
+ /* Cannot hash words smaller than 4 bytes */
+ if (word_size < 4) {
+ /* Break instead of continue, all next words of this length will have
+ same length after transform */
+ break;
+ }
+ if (!BrotliTrieAdd(m, 0, idx, word, word_size, &dedup)) {
+ return BROTLI_FALSE;
+ }
+ key = Hash(word, NUM_HASH_BITS);
+ BROTLI_ENSURE_CAPACITY_APPEND(m, DictWord, words_by_hash[key],
+ words_by_hash_capacity[key], words_by_hash_size[key],
+ MakeDictWord(l, 0, idx));
+ ++total;
+ }
+ }
+ }
+
+ /* These LUT transforms only supported if no custom transforms. This is
+ ok, we will use the heavy trie instead. */
+ if (transforms == BrotliGetTransforms()) {
+ for (l = SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH;
+ l <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH; ++l) {
+ uint16_t n = dict->words->size_bits_by_length[l] ?
+ (uint16_t)(1 << dict->words->size_bits_by_length[l]) : 0u;
+ for (idx = 0; idx < n; ++idx) {
+ int k;
+ BROTLI_BOOL is_ascii = BROTLI_TRUE;
+ size_t offset = dict->words->offsets_by_length[l] + (size_t)l * idx;
+ const uint8_t* data = &dict->words->data[offset];
+ for (k = 0; k < l; ++k) {
+ if (data[k] >= 128) is_ascii = BROTLI_FALSE;
+ }
+ if (data[0] < 128) {
+ int transform = 9; /* {empty, uppercase first, empty} */
+ uint32_t ix = idx + (uint32_t)transform * n;
+ const BrotliTrieNode* it;
+ TransformedDictionaryWord(idx, l, transform, transforms,
+ dict, word, &word_size);
+ it = BrotliTrieFind(&dedup, word, word_size);
+ if (!it || it->idx_ > ix) {
+ uint32_t key = Hash(word, NUM_HASH_BITS);
+ if (!BrotliTrieAdd(m, 0, ix, word, word_size, &dedup)) {
+ return BROTLI_FALSE;
+ }
+ BROTLI_ENSURE_CAPACITY_APPEND(m, DictWord, words_by_hash[key],
+ words_by_hash_capacity[key], words_by_hash_size[key],
+ MakeDictWord(l, BROTLI_TRANSFORM_UPPERCASE_FIRST, idx));
+ ++total;
+ }
+ }
+ if (is_ascii) {
+ int transform = 44; /* {empty, uppercase all, empty} */
+ uint32_t ix = idx + (uint32_t)transform * n;
+ const BrotliTrieNode* it;
+ TransformedDictionaryWord(idx, l, transform, transforms,
+ dict, word, &word_size);
+ it = BrotliTrieFind(&dedup, word, word_size);
+ if (!it || it->idx_ > ix) {
+ uint32_t key = Hash(word, NUM_HASH_BITS);
+ if (!BrotliTrieAdd(m, 0, ix, word, word_size, &dedup)) {
+ return BROTLI_FALSE;
+ }
+ BROTLI_ENSURE_CAPACITY_APPEND(m, DictWord, words_by_hash[key],
+ words_by_hash_capacity[key], words_by_hash_size[key],
+ MakeDictWord(l, BROTLI_TRANSFORM_UPPERCASE_ALL, idx));
+ ++total;
+ }
+ }
+ }
+ }
+ }
+
+ dict_words = (DictWord*)BrotliAllocate(m,
+ sizeof(*dict->dict_words) * (total + 1));
+ buckets = (uint16_t*)BrotliAllocate(m,
+ sizeof(*dict->buckets) * NUM_HASH_BUCKETS);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ dict->dict_words_alloc_size_ = total + 1;
+ dict->dict_words = dict->dict_words_data_ = dict_words;
+ dict->buckets_alloc_size_ = NUM_HASH_BUCKETS;
+ dict->buckets = dict->buckets_data_ = buckets;
+
+ /* Unused; makes offsets start from 1. */
+ dict_words[0] = MakeDictWord(0, 0, 0);
+ total = 1;
+ for (i = 0; i < NUM_HASH_BUCKETS; ++i) {
+ size_t num_words = words_by_hash_size[i];
+ if (num_words > 0) {
+ buckets[i] = (uint16_t)(total);
+ memcpy(&dict_words[total], &words_by_hash[i][0],
+ sizeof(dict_words[0]) * num_words);
+ total += num_words;
+ dict_words[total - 1].len |= 0x80;
+ } else {
+ buckets[i] = 0;
+ }
+ }
+
+ for (i = 0; i < NUM_HASH_BUCKETS; ++i) {
+ BrotliFree(m, words_by_hash[i]);
+ }
+ BrotliFree(m, words_by_hash);
+ BrotliFree(m, words_by_hash_size);
+ BrotliFree(m, words_by_hash_capacity);
+ BrotliTrieFree(m, &dedup);
+
+ return BROTLI_TRUE;
+}
+
+static void BuildDictionaryHashTable(uint16_t* hash_table_words,
+ uint8_t* hash_table_lengths, const BrotliDictionary* dict) {
+ int j, len;
+ /* The order of the loops is such that in case of collision, words with
+ shorter length are preferred, and in case of same length, words with
+ smaller index. There is only a single word per bucket. */
+ /* TODO(lode): consider adding optional user-supplied frequency_map to use
+ for preferred words instead, this can make the encoder better for
+ quality 9 and below without affecting the decoder */
+ memset(hash_table_words, 0, sizeof(kStaticDictionaryHashWords));
+ memset(hash_table_lengths, 0, sizeof(kStaticDictionaryHashLengths));
+ for (len = SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH;
+ len >= SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH; --len) {
+ const size_t num_words = dict->size_bits_by_length[len] ?
+ (1u << dict->size_bits_by_length[len]) : 0;
+ for (j = (int)num_words - 1; j >= 0; --j) {
+ size_t offset = dict->offsets_by_length[len] +
+ (size_t)len * (size_t)j;
+ const uint8_t* word = &dict->data[offset];
+ const uint32_t key = Hash(word, 14);
+ int idx = (int)(key << 1) + (len < 8 ? 1 : 0);
+ BROTLI_DCHECK(idx < (int)NUM_HASH_BUCKETS);
+ hash_table_words[idx] = (uint16_t)j;
+ hash_table_lengths[idx] = (uint8_t)len;
+ }
+ }
}
+static BROTLI_BOOL GenerateWordsHeavy(MemoryManager* m,
+ const BrotliTransforms* transforms,
+ BrotliEncoderDictionary* dict) {
+ int i, j, l;
+ for (j = (int)transforms->num_transforms - 1; j >= 0 ; --j) {
+ for (l = 0; l < 32; l++) {
+ int num = (int)((1u << dict->words->size_bits_by_length[l]) & ~1u);
+ for (i = 0; i < num; i++) {
+ uint8_t transformed[kTransformedBufferSize];
+ size_t size;
+ TransformedDictionaryWord(
+ (uint32_t)i, l, j, transforms, dict, transformed, &size);
+ if (size < 4) continue;
+ if (!BrotliTrieAdd(m, (uint8_t)l, (uint32_t)(i + num * j),
+ transformed, size, &dict->trie)) {
+ return BROTLI_FALSE;
+ }
+ }
+ }
+ }
+ return BROTLI_TRUE;
+}
+
+/* Computes cutoffTransformsCount (in count) and cutoffTransforms (in data) for
+ the custom transforms, where possible within the limits of the
+ cutoffTransforms encoding. The fast encoder uses this to do fast lookup for
+ transforms that remove the N last characters (OmitLast). */
+static void ComputeCutoffTransforms(
+ const BrotliTransforms* transforms,
+ uint32_t* count, uint64_t* data) {
+ int i;
+ /* The encoding in a 64-bit integer of transform N in the data is: (N << 2) +
+ ((cutoffTransforms >> (N * 6)) & 0x3F), so for example the identity
+ transform code must be 0-63, for N=1 the transform code must be 4-67, ...,
+ for N=9 it must be 36-99.
+ TODO(lode): consider a simple flexible uint8_t[10] instead of the uint64_t
+ for the cutoff transforms, so that shared dictionaries can have the
+ OmitLast transforms anywhere without loss. */
+ *count = 0;
+ *data = 0;
+ for (i = 0; i < BROTLI_TRANSFORMS_MAX_CUT_OFF + 1; i++) {
+ int idx = transforms->cutOffTransforms[i];
+ if (idx == -1) break; /* Not found */
+ if (idx < (i << 2)) break; /* Too small for the encoding */
+ if (idx >= (i << 2) + 64) break; /* Too large for the encoding */
+ (*count)++;
+ *data |= (uint64_t)(((uint64_t)idx -
+ ((uint64_t)i << 2u)) << ((uint64_t)i * 6u));
+ }
+}
+
+static BROTLI_BOOL ComputeDictionary(MemoryManager* m, int quality,
+ const BrotliTransforms* transforms,
+ BrotliEncoderDictionary* current) {
+ int default_words = current->words == BrotliGetDictionary();
+ int default_transforms = transforms == BrotliGetTransforms();
+
+ if (default_words && default_transforms) {
+ /* hashes are already set to Brotli defaults */
+ return BROTLI_TRUE;
+ }
+
+ current->hash_table_data_words_ = (uint16_t*)BrotliAllocate(
+ m, sizeof(kStaticDictionaryHashWords));
+ current->hash_table_data_lengths_ = (uint8_t*)BrotliAllocate(
+ m, sizeof(kStaticDictionaryHashLengths));
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ current->hash_table_words = current->hash_table_data_words_;
+ current->hash_table_lengths = current->hash_table_data_lengths_;
+
+ BuildDictionaryHashTable(current->hash_table_data_words_,
+ current->hash_table_data_lengths_, current->words);
+
+ ComputeCutoffTransforms(transforms,
+ &current->cutoffTransformsCount, &current->cutoffTransforms);
+
+ /* Only compute the data for slow encoder if the requested quality is high
+ enough to need it */
+ if (quality >= ZOPFLIFICATION_QUALITY) {
+ if (!BuildDictionaryLut(m, transforms, current)) return BROTLI_FALSE;
+
+ /* For the built-in Brotli transforms, there is a hard-coded function to
+ handle all transforms, but for custom transforms, we use the following
+ large hammer instead */
+ current->has_words_heavy = !default_transforms;
+ if (current->has_words_heavy) {
+ if (!GenerateWordsHeavy(m, transforms, current)) return BROTLI_FALSE;
+ }
+ }
+
+ return BROTLI_TRUE;
+}
+#endif /* BROTLI_EXPERIMENTAL */
+
+void BrotliInitSharedEncoderDictionary(SharedEncoderDictionary* dict) {
+ dict->magic = kSharedDictionaryMagic;
+
+ dict->compound.num_chunks = 0;
+ dict->compound.total_size = 0;
+ dict->compound.chunk_offsets[0] = 0;
+ dict->compound.num_prepared_instances_ = 0;
+
+ dict->contextual.context_based = 0;
+ dict->contextual.num_dictionaries = 1;
+ dict->contextual.instances_ = 0;
+ dict->contextual.num_instances_ = 1; /* The instance_ field */
+ dict->contextual.dict[0] = &dict->contextual.instance_;
+ InitEncoderDictionary(&dict->contextual.instance_);
+ dict->contextual.instance_.parent = &dict->contextual;
+
+ dict->max_quality = BROTLI_MAX_QUALITY;
+}
+
+#if defined(BROTLI_EXPERIMENTAL)
+/* TODO(eustas): make sure that tooling will warn user if not all the cutoff
+ transforms are available (for low-quality encoder). */
+static BROTLI_BOOL InitCustomSharedEncoderDictionary(
+ MemoryManager* m, const BrotliSharedDictionary* decoded_dict,
+ int quality, SharedEncoderDictionary* dict) {
+ ContextualEncoderDictionary* contextual;
+ CompoundDictionary* compound;
+ BrotliEncoderDictionary* instances;
+ int i;
+ BrotliInitSharedEncoderDictionary(dict);
+
+ contextual = &dict->contextual;
+ compound = &dict->compound;
+
+ for (i = 0; i < (int)decoded_dict->num_prefix; i++) {
+ PreparedDictionary* prepared = CreatePreparedDictionary(m,
+ decoded_dict->prefix[i], decoded_dict->prefix_size[i]);
+ AttachPreparedDictionary(compound, prepared);
+ /* remember for cleanup */
+ compound->prepared_instances_[
+ compound->num_prepared_instances_++] = prepared;
+ }
+
+ dict->max_quality = quality;
+ contextual->context_based = decoded_dict->context_based;
+ if (decoded_dict->context_based) {
+ memcpy(contextual->context_map, decoded_dict->context_map,
+ SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS);
+ }
+
+ contextual->num_dictionaries = decoded_dict->num_dictionaries;
+ contextual->num_instances_ = decoded_dict->num_dictionaries;
+ if (contextual->num_instances_ == 1) {
+ instances = &contextual->instance_;
+ } else {
+ contextual->instances_ = (BrotliEncoderDictionary*)
+ BrotliAllocate(m, sizeof(*contextual->instances_) *
+ contextual->num_instances_);
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ instances = contextual->instances_;
+ }
+ for (i = 0; i < (int)contextual->num_instances_; i++) {
+ BrotliEncoderDictionary* current = &instances[i];
+ InitEncoderDictionary(current);
+ current->parent = &dict->contextual;
+ if (decoded_dict->words[i] == BrotliGetDictionary()) {
+ current->words = BrotliGetDictionary();
+ } else {
+ current->words_instance_ = (BrotliDictionary*)BrotliAllocate(
+ m, sizeof(BrotliDictionary));
+ if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+ *current->words_instance_ = *decoded_dict->words[i];
+ current->words = current->words_instance_;
+ }
+ current->num_transforms =
+ (uint32_t)decoded_dict->transforms[i]->num_transforms;
+ if (!ComputeDictionary(
+ m, quality, decoded_dict->transforms[i], current)) {
+ return BROTLI_FALSE;
+ }
+
+ contextual->dict[i] = current;
+ }
+
+ return BROTLI_TRUE; /* success */
+}
+
+BROTLI_BOOL BrotliInitCustomSharedEncoderDictionary(
+ MemoryManager* m, const uint8_t* encoded_dict, size_t size,
+ int quality, SharedEncoderDictionary* dict) {
+ BROTLI_BOOL success = BROTLI_FALSE;
+ BrotliSharedDictionary* decoded_dict = BrotliSharedDictionaryCreateInstance(
+ m->alloc_func, m->free_func, m->opaque);
+ if (!decoded_dict) { /* OOM */
+ return BROTLI_FALSE;
+ }
+ success = BrotliSharedDictionaryAttach(
+ decoded_dict, BROTLI_SHARED_DICTIONARY_SERIALIZED, size, encoded_dict);
+ if (success) {
+ success = InitCustomSharedEncoderDictionary(m,
+ decoded_dict, quality, dict);
+ }
+ BrotliSharedDictionaryDestroyInstance(decoded_dict);
+ return success;
+}
+#endif /* BROTLI_EXPERIMENTAL */
+
+void BrotliCleanupSharedEncoderDictionary(MemoryManager* m,
+ SharedEncoderDictionary* dict) {
+ size_t i;
+ for (i = 0; i < dict->compound.num_prepared_instances_; i++) {
+ DestroyPreparedDictionary(m,
+ (PreparedDictionary*)dict->compound.prepared_instances_[i]);
+ }
+ if (dict->contextual.num_instances_ == 1) {
+ BrotliDestroyEncoderDictionary(m, &dict->contextual.instance_);
+ } else if (dict->contextual.num_instances_ > 1) {
+ for (i = 0; i < dict->contextual.num_instances_; i++) {
+ BrotliDestroyEncoderDictionary(m, &dict->contextual.instances_[i]);
+ }
+ BrotliFree(m, dict->contextual.instances_);
+ }
+}
+
+ManagedDictionary* BrotliCreateManagedDictionary(
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
+ ManagedDictionary* result = (ManagedDictionary*)BrotliBootstrapAlloc(
+ sizeof(ManagedDictionary), alloc_func, free_func, opaque);
+ if (result == NULL) return NULL;
+
+ result->magic = kManagedDictionaryMagic;
+ BrotliInitMemoryManager(
+ &result->memory_manager_, alloc_func, free_func, opaque);
+ result->dictionary = NULL;
+
+ return result;
+}
+
+void BrotliDestroyManagedDictionary(ManagedDictionary* dictionary) {
+ if (!dictionary) return;
+ BrotliBootstrapFree(dictionary, &dictionary->memory_manager_);
+}
+
+/* Escalate internal functions visibility; for testing purposes only. */
+#if defined(BROTLI_TEST)
+void InitEncoderDictionaryForTest(BrotliEncoderDictionary*);
+void InitEncoderDictionaryForTest(BrotliEncoderDictionary* d) {
+ InitEncoderDictionary(d);
+}
+#endif
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/encoder_dict.h b/dviware/dvisvgm/libs/brotli/enc/encoder_dict.h
index a1c329fbf4..27dcbcd2f3 100644
--- a/dviware/dvisvgm/libs/brotli/enc/encoder_dict.h
+++ b/dviware/dvisvgm/libs/brotli/enc/encoder_dict.h
@@ -7,15 +7,56 @@
#ifndef BROTLI_ENC_ENCODER_DICT_H_
#define BROTLI_ENC_ENCODER_DICT_H_
+#include <brotli/shared_dictionary.h>
+#include <brotli/types.h>
+
#include "../common/dictionary.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./static_dict_lut.h"
+#include "compound_dictionary.h"
+#include "memory.h"
+#include "static_dict_lut.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
+/*
+Dictionary hierarchy for Encoder:
+-SharedEncoderDictionary
+--CompoundDictionary
+---PreparedDictionary [up to 15x]
+ = prefix dictionary with precomputed hashes
+--ContextualEncoderDictionary
+---BrotliEncoderDictionary [up to 64x]
+ = for each context, precomputed static dictionary with words + transforms
+
+Dictionary hiearchy from common: similar, but without precomputed hashes
+-BrotliSharedDictionary
+--BrotliDictionary [up to 64x]
+--BrotliTransforms [up to 64x]
+--const uint8_t* prefix [up to 15x]: compound dictionaries
+*/
+
+typedef struct BrotliTrieNode {
+ uint8_t single; /* if 1, sub is a single node for c instead of 256 */
+ uint8_t c;
+ uint8_t len_; /* untransformed length */
+ uint32_t idx_; /* word index + num words * transform index */
+ uint32_t sub; /* index of sub node(s) in the pool */
+} BrotliTrieNode;
+
+typedef struct BrotliTrie {
+ BrotliTrieNode* pool;
+ size_t pool_capacity;
+ size_t pool_size;
+ BrotliTrieNode root;
+} BrotliTrie;
+
+#if defined(BROTLI_EXPERIMENTAL)
+BROTLI_INTERNAL const BrotliTrieNode* BrotliTrieSub(const BrotliTrie* trie,
+ const BrotliTrieNode* node, uint8_t c);
+#endif /* BROTLI_EXPERIMENTAL */
+
/* Dictionary data (words and transforms) for 1 possible context */
typedef struct BrotliEncoderDictionary {
const BrotliDictionary* words;
@@ -32,9 +73,82 @@ typedef struct BrotliEncoderDictionary {
/* from static_dict_lut.h, for slow encoder */
const uint16_t* buckets;
const DictWord* dict_words;
+ /* Heavy version, for use by slow encoder when there are custom transforms.
+ Contains every possible transformed dictionary word in a trie. It encodes
+ about as fast as the non-heavy encoder but consumes a lot of memory and
+ takes time to build. */
+ BrotliTrie trie;
+ BROTLI_BOOL has_words_heavy;
+
+ /* Reference to other dictionaries. */
+ const struct ContextualEncoderDictionary* parent;
+
+ /* Allocated memory, used only when not using the Brotli defaults */
+ uint16_t* hash_table_data_words_;
+ uint8_t* hash_table_data_lengths_;
+ size_t buckets_alloc_size_;
+ uint16_t* buckets_data_;
+ size_t dict_words_alloc_size_;
+ DictWord* dict_words_data_;
+ BrotliDictionary* words_instance_;
} BrotliEncoderDictionary;
-BROTLI_INTERNAL void BrotliInitEncoderDictionary(BrotliEncoderDictionary* dict);
+/* Dictionary data for all 64 contexts */
+typedef struct ContextualEncoderDictionary {
+ BROTLI_BOOL context_based;
+ uint8_t num_dictionaries;
+ uint8_t context_map[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+ const BrotliEncoderDictionary* dict[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+ /* If num_instances_ is 1, instance_ is used, else dynamic allocation with
+ instances_ is used. */
+ size_t num_instances_;
+ BrotliEncoderDictionary instance_;
+ BrotliEncoderDictionary* instances_;
+} ContextualEncoderDictionary;
+
+typedef struct SharedEncoderDictionary {
+ /* Magic value to distinguish this struct from PreparedDictionary for
+ certain external usages. */
+ uint32_t magic;
+
+ /* LZ77 prefix, compound dictionary */
+ CompoundDictionary compound;
+
+ /* Custom static dictionary (optionally context-based) */
+ ContextualEncoderDictionary contextual;
+
+ /* The maximum quality the dictionary was computed for */
+ int max_quality;
+} SharedEncoderDictionary;
+
+typedef struct ManagedDictionary {
+ uint32_t magic;
+ MemoryManager memory_manager_;
+ uint32_t* dictionary;
+} ManagedDictionary;
+
+/* Initializes to the brotli built-in dictionary */
+BROTLI_INTERNAL void BrotliInitSharedEncoderDictionary(
+ SharedEncoderDictionary* dict);
+
+#if defined(BROTLI_EXPERIMENTAL)
+/* Initializes to shared dictionary that will be parsed from
+ encoded_dict. Requires that you keep the encoded_dict buffer
+ around, parts of data will point to it. */
+BROTLI_INTERNAL BROTLI_BOOL BrotliInitCustomSharedEncoderDictionary(
+ MemoryManager* m, const uint8_t* encoded_dict, size_t size,
+ int quality, SharedEncoderDictionary* dict);
+#endif /* BROTLI_EXPERIMENTAL */
+
+BROTLI_INTERNAL void BrotliCleanupSharedEncoderDictionary(
+ MemoryManager* m, SharedEncoderDictionary* dict);
+
+BROTLI_INTERNAL ManagedDictionary* BrotliCreateManagedDictionary(
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+
+BROTLI_INTERNAL void BrotliDestroyManagedDictionary(
+ ManagedDictionary* dictionary);
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/entropy_encode.c b/dviware/dvisvgm/libs/brotli/enc/entropy_encode.c
index b50ccb5d1f..9aed43b6eb 100644
--- a/dviware/dvisvgm/libs/brotli/enc/entropy_encode.c
+++ b/dviware/dvisvgm/libs/brotli/enc/entropy_encode.c
@@ -6,13 +6,14 @@
/* Entropy encoding (Huffman) utilities. */
-#include "./entropy_encode.h"
+#include "entropy_encode.h"
#include <string.h> /* memset */
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/platform.h"
-#include <brotli/types.h>
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/entropy_encode.h b/dviware/dvisvgm/libs/brotli/enc/entropy_encode.h
index 9618e1d359..e1c779cc6f 100644
--- a/dviware/dvisvgm/libs/brotli/enc/entropy_encode.h
+++ b/dviware/dvisvgm/libs/brotli/enc/entropy_encode.h
@@ -9,9 +9,10 @@
#ifndef BROTLI_ENC_ENTROPY_ENCODE_H_
#define BROTLI_ENC_ENTROPY_ENCODE_H_
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/entropy_encode_static.h b/dviware/dvisvgm/libs/brotli/enc/entropy_encode_static.h
index 62b99a954c..ecff1fe9ee 100644
--- a/dviware/dvisvgm/libs/brotli/enc/entropy_encode_static.h
+++ b/dviware/dvisvgm/libs/brotli/enc/entropy_encode_static.h
@@ -9,10 +9,11 @@
#ifndef BROTLI_ENC_ENTROPY_ENCODE_STATIC_H_
#define BROTLI_ENC_ENTROPY_ENCODE_STATIC_H_
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./write_bits.h"
+#include "write_bits.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -76,6 +77,7 @@ static const uint8_t kStaticDistanceCodeDepth[64] = {
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
};
+/* GENERATED CODE START */
static const uint32_t kCodeLengthBits[18] = {
0, 8, 4, 12, 2, 10, 6, 14, 1, 9, 5, 13, 3, 15, 31, 0, 11, 7,
};
@@ -531,6 +533,7 @@ static BROTLI_INLINE void StoreStaticDistanceHuffmanTree(
size_t* storage_ix, uint8_t* storage) {
BrotliWriteBits(28, 0x0369DC03u, storage_ix, storage);
}
+/* GENERATED CODE END */
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/fast_log.c b/dviware/dvisvgm/libs/brotli/enc/fast_log.c
index 2319baeb74..2fa0efcf86 100644
--- a/dviware/dvisvgm/libs/brotli/enc/fast_log.c
+++ b/dviware/dvisvgm/libs/brotli/enc/fast_log.c
@@ -4,7 +4,7 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./fast_log.h"
+#include "fast_log.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/fast_log.h b/dviware/dvisvgm/libs/brotli/enc/fast_log.h
index 2094f13e55..f82f4cffc8 100644
--- a/dviware/dvisvgm/libs/brotli/enc/fast_log.h
+++ b/dviware/dvisvgm/libs/brotli/enc/fast_log.h
@@ -11,9 +11,10 @@
#include <math.h>
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/find_match_length.h b/dviware/dvisvgm/libs/brotli/enc/find_match_length.h
index f8853a70fb..f3de0bdb6b 100644
--- a/dviware/dvisvgm/libs/brotli/enc/find_match_length.h
+++ b/dviware/dvisvgm/libs/brotli/enc/find_match_length.h
@@ -9,9 +9,10 @@
#ifndef BROTLI_ENC_FIND_MATCH_LENGTH_H_
#define BROTLI_ENC_FIND_MATCH_LENGTH_H_
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
@@ -21,31 +22,23 @@ extern "C" {
static BROTLI_INLINE size_t FindMatchLengthWithLimit(const uint8_t* s1,
const uint8_t* s2,
size_t limit) {
- size_t matched = 0;
- size_t limit2 = (limit >> 3) + 1; /* + 1 is for pre-decrement in while */
- while (BROTLI_PREDICT_TRUE(--limit2)) {
- if (BROTLI_PREDICT_FALSE(BROTLI_UNALIGNED_LOAD64LE(s2) ==
- BROTLI_UNALIGNED_LOAD64LE(s1 + matched))) {
- s2 += 8;
- matched += 8;
- } else {
- uint64_t x = BROTLI_UNALIGNED_LOAD64LE(s2) ^
- BROTLI_UNALIGNED_LOAD64LE(s1 + matched);
+ const uint8_t *s1_orig = s1;
+ for (; limit >= 8; limit -= 8) {
+ uint64_t x = BROTLI_UNALIGNED_LOAD64LE(s2) ^
+ BROTLI_UNALIGNED_LOAD64LE(s1);
+ s2 += 8;
+ if (x != 0) {
size_t matching_bits = (size_t)BROTLI_TZCNT64(x);
- matched += matching_bits >> 3;
- return matched;
+ return (size_t)(s1 - s1_orig) + (matching_bits >> 3);
}
+ s1 += 8;
}
- limit = (limit & 7) + 1; /* + 1 is for pre-decrement in while */
- while (--limit) {
- if (BROTLI_PREDICT_TRUE(s1[matched] == *s2)) {
- ++s2;
- ++matched;
- } else {
- return matched;
- }
+ while (limit && *s1 == *s2) {
+ limit--;
+ ++s2;
+ ++s1;
}
- return matched;
+ return (size_t)(s1 - s1_orig);
}
#else
static BROTLI_INLINE size_t FindMatchLengthWithLimit(const uint8_t* s1,
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash.h b/dviware/dvisvgm/libs/brotli/enc/hash.h
index 6362f69b9f..5677d82920 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash.h
@@ -10,33 +10,49 @@
#ifndef BROTLI_ENC_HASH_H_
#define BROTLI_ENC_HASH_H_
+#include <stdlib.h> /* exit */
#include <string.h> /* memcmp, memset */
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/dictionary.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./encoder_dict.h"
-#include "./fast_log.h"
-#include "./find_match_length.h"
-#include "./memory.h"
-#include "./quality.h"
-#include "./static_dict.h"
+#include "compound_dictionary.h"
+#include "encoder_dict.h"
+#include "fast_log.h"
+#include "find_match_length.h"
+#include "memory.h"
+#include "quality.h"
+#include "static_dict.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
typedef struct {
- /* Dynamically allocated area; first member for quickest access. */
- void* extra;
+ /**
+ * Dynamically allocated areas; regular hasher uses one or two allocations;
+ * "composite" hasher uses up to 4 allocations.
+ */
+ void* extra[4];
+
+ /**
+ * False before the fisrt invocation of HasherSetup (where "extra" memory)
+ * is allocated.
+ */
+ BROTLI_BOOL is_setup_;
size_t dict_num_lookups;
size_t dict_num_matches;
BrotliHasherParams params;
- /* False if hasher needs to be "prepared" before use. */
+ /**
+ * False if hasher needs to be "prepared" before use (before the first
+ * invocation of HasherSetup or after HasherReset). "preparation" is hasher
+ * data initialization (using input ringbuffer).
+ */
BROTLI_BOOL is_prepared_;
} HasherCommon;
@@ -62,8 +78,7 @@ typedef struct HasherSearchResult {
for this use.
* The number has been tuned heuristically against compression benchmarks. */
static const uint32_t kHashMul32 = 0x1E35A7BD;
-static const uint64_t kHashMul64 = BROTLI_MAKE_UINT64_T(0x1E35A7BD, 0x1E35A7BD);
-static const uint64_t kHashMul64Long =
+static const uint64_t kHashMul64 =
BROTLI_MAKE_UINT64_T(0x1FE35A7Bu, 0xD3579BD3u);
static BROTLI_INLINE uint32_t Hash14(const uint8_t* data) {
@@ -232,7 +247,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define BUCKET_BITS 17
#define MAX_TREE_SEARCH_DEPTH 64
#define MAX_TREE_COMP_LENGTH 128
-#include "./hash_to_binary_tree_inc.h" /* NOLINT(build/include) */
+#include "hash_to_binary_tree_inc.h" /* NOLINT(build/include) */
#undef MAX_TREE_SEARCH_DEPTH
#undef MAX_TREE_COMP_LENGTH
#undef BUCKET_BITS
@@ -249,7 +264,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define BUCKET_SWEEP_BITS 0
#define HASH_LEN 5
#define USE_DICTIONARY 1
-#include "./hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
+#include "hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
#undef BUCKET_SWEEP_BITS
#undef USE_DICTIONARY
#undef HASHER
@@ -257,7 +272,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define HASHER() H3
#define BUCKET_SWEEP_BITS 1
#define USE_DICTIONARY 0
-#include "./hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
+#include "hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
#undef USE_DICTIONARY
#undef BUCKET_SWEEP_BITS
#undef BUCKET_BITS
@@ -267,7 +282,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define BUCKET_BITS 17
#define BUCKET_SWEEP_BITS 2
#define USE_DICTIONARY 1
-#include "./hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
+#include "hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
#undef USE_DICTIONARY
#undef HASH_LEN
#undef BUCKET_SWEEP_BITS
@@ -275,11 +290,11 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#undef HASHER
#define HASHER() H5
-#include "./hash_longest_match_inc.h" /* NOLINT(build/include) */
+#include "hash_longest_match_inc.h" /* NOLINT(build/include) */
#undef HASHER
#define HASHER() H6
-#include "./hash_longest_match64_inc.h" /* NOLINT(build/include) */
+#include "hash_longest_match64_inc.h" /* NOLINT(build/include) */
#undef HASHER
#define BUCKET_BITS 15
@@ -288,13 +303,13 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define NUM_BANKS 1
#define BANK_BITS 16
#define HASHER() H40
-#include "./hash_forgetful_chain_inc.h" /* NOLINT(build/include) */
+#include "hash_forgetful_chain_inc.h" /* NOLINT(build/include) */
#undef HASHER
#undef NUM_LAST_DISTANCES_TO_CHECK
#define NUM_LAST_DISTANCES_TO_CHECK 10
#define HASHER() H41
-#include "./hash_forgetful_chain_inc.h" /* NOLINT(build/include) */
+#include "hash_forgetful_chain_inc.h" /* NOLINT(build/include) */
#undef HASHER
#undef NUM_LAST_DISTANCES_TO_CHECK
#undef NUM_BANKS
@@ -304,7 +319,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define NUM_BANKS 512
#define BANK_BITS 9
#define HASHER() H42
-#include "./hash_forgetful_chain_inc.h" /* NOLINT(build/include) */
+#include "hash_forgetful_chain_inc.h" /* NOLINT(build/include) */
#undef HASHER
#undef NUM_LAST_DISTANCES_TO_CHECK
#undef NUM_BANKS
@@ -317,7 +332,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define BUCKET_SWEEP_BITS 2
#define HASH_LEN 7
#define USE_DICTIONARY 0
-#include "./hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
+#include "hash_longest_match_quickly_inc.h" /* NOLINT(build/include) */
#undef USE_DICTIONARY
#undef HASH_LEN
#undef BUCKET_SWEEP_BITS
@@ -331,14 +346,14 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define JUMP 4
#define NUMBUCKETS 16777216
#define MASK ((NUMBUCKETS * 64) - 1)
-#include "./hash_rolling_inc.h" /* NOLINT(build/include) */
+#include "hash_rolling_inc.h" /* NOLINT(build/include) */
#undef JUMP
#undef HASHER
#define HASHER() HROLLING
#define JUMP 1
-#include "./hash_rolling_inc.h" /* NOLINT(build/include) */
+#include "hash_rolling_inc.h" /* NOLINT(build/include) */
#undef MASK
#undef NUMBUCKETS
#undef JUMP
@@ -348,7 +363,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define HASHER() H35
#define HASHER_A H3
#define HASHER_B HROLLING_FAST
-#include "./hash_composite_inc.h" /* NOLINT(build/include) */
+#include "hash_composite_inc.h" /* NOLINT(build/include) */
#undef HASHER_A
#undef HASHER_B
#undef HASHER
@@ -356,7 +371,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define HASHER() H55
#define HASHER_A H54
#define HASHER_B HROLLING_FAST
-#include "./hash_composite_inc.h" /* NOLINT(build/include) */
+#include "hash_composite_inc.h" /* NOLINT(build/include) */
#undef HASHER_A
#undef HASHER_B
#undef HASHER
@@ -364,7 +379,7 @@ static BROTLI_INLINE size_t BackwardMatchLengthCode(const BackwardMatch* self) {
#define HASHER() H65
#define HASHER_A H6
#define HASHER_B HROLLING
-#include "./hash_composite_inc.h" /* NOLINT(build/include) */
+#include "hash_composite_inc.h" /* NOLINT(build/include) */
#undef HASHER_A
#undef HASHER_B
#undef HASHER
@@ -391,43 +406,55 @@ typedef struct {
/* MUST be invoked before any other method. */
static BROTLI_INLINE void HasherInit(Hasher* hasher) {
- hasher->common.extra = NULL;
+ hasher->common.is_setup_ = BROTLI_FALSE;
+ hasher->common.extra[0] = NULL;
+ hasher->common.extra[1] = NULL;
+ hasher->common.extra[2] = NULL;
+ hasher->common.extra[3] = NULL;
}
static BROTLI_INLINE void DestroyHasher(MemoryManager* m, Hasher* hasher) {
- if (hasher->common.extra == NULL) return;
- BROTLI_FREE(m, hasher->common.extra);
+ if (hasher->common.extra[0] != NULL) BROTLI_FREE(m, hasher->common.extra[0]);
+ if (hasher->common.extra[1] != NULL) BROTLI_FREE(m, hasher->common.extra[1]);
+ if (hasher->common.extra[2] != NULL) BROTLI_FREE(m, hasher->common.extra[2]);
+ if (hasher->common.extra[3] != NULL) BROTLI_FREE(m, hasher->common.extra[3]);
}
static BROTLI_INLINE void HasherReset(Hasher* hasher) {
hasher->common.is_prepared_ = BROTLI_FALSE;
}
-static BROTLI_INLINE size_t HasherSize(const BrotliEncoderParams* params,
- BROTLI_BOOL one_shot, const size_t input_size) {
+static BROTLI_INLINE void HasherSize(const BrotliEncoderParams* params,
+ BROTLI_BOOL one_shot, const size_t input_size, size_t* alloc_size) {
switch (params->hasher.type) {
-#define SIZE_(N) \
- case N: \
- return HashMemAllocInBytesH ## N(params, one_shot, input_size);
+#define SIZE_(N) \
+ case N: \
+ HashMemAllocInBytesH ## N(params, one_shot, input_size, alloc_size); \
+ break;
FOR_ALL_HASHERS(SIZE_)
#undef SIZE_
default:
break;
}
- return 0; /* Default case. */
}
static BROTLI_INLINE void HasherSetup(MemoryManager* m, Hasher* hasher,
BrotliEncoderParams* params, const uint8_t* data, size_t position,
size_t input_size, BROTLI_BOOL is_last) {
BROTLI_BOOL one_shot = (position == 0 && is_last);
- if (hasher->common.extra == NULL) {
- size_t alloc_size;
+ if (!hasher->common.is_setup_) {
+ size_t alloc_size[4] = {0};
+ size_t i;
ChooseHasher(params, &params->hasher);
- alloc_size = HasherSize(params, one_shot, input_size);
- hasher->common.extra = BROTLI_ALLOC(m, uint8_t, alloc_size);
- if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(hasher->common.extra)) return;
hasher->common.params = params->hasher;
+ hasher->common.dict_num_lookups = 0;
+ hasher->common.dict_num_matches = 0;
+ HasherSize(params, one_shot, input_size, alloc_size);
+ for (i = 0; i < 4; ++i) {
+ if (alloc_size[i] == 0) continue;
+ hasher->common.extra[i] = BROTLI_ALLOC(m, uint8_t, alloc_size[i]);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(hasher->common.extra[i])) return;
+ }
switch (hasher->common.params.type) {
#define INITIALIZE_(N) \
case N: \
@@ -440,6 +467,7 @@ static BROTLI_INLINE void HasherSetup(MemoryManager* m, Hasher* hasher,
break;
}
HasherReset(hasher);
+ hasher->common.is_setup_ = BROTLI_TRUE;
}
if (!hasher->common.is_prepared_) {
@@ -454,10 +482,6 @@ static BROTLI_INLINE void HasherSetup(MemoryManager* m, Hasher* hasher,
#undef PREPARE_
default: break;
}
- if (position == 0) {
- hasher->common.dict_num_lookups = 0;
- hasher->common.dict_num_matches = 0;
- }
hasher->common.is_prepared_ = BROTLI_TRUE;
}
}
@@ -481,6 +505,222 @@ static BROTLI_INLINE void InitOrStitchToPreviousBlock(
}
}
+/* NB: when seamless dictionary-ring-buffer copies are implemented, don't forget
+ to add proper guards for non-zero-BROTLI_PARAM_STREAM_OFFSET. */
+static BROTLI_INLINE void FindCompoundDictionaryMatch(
+ const PreparedDictionary* self, const uint8_t* BROTLI_RESTRICT data,
+ const size_t ring_buffer_mask, const int* BROTLI_RESTRICT distance_cache,
+ const size_t cur_ix, const size_t max_length, const size_t distance_offset,
+ const size_t max_distance, HasherSearchResult* BROTLI_RESTRICT out) {
+ const uint32_t source_size = self->source_size;
+ const size_t boundary = distance_offset - source_size;
+ const uint32_t hash_bits = self->hash_bits;
+ const uint32_t bucket_bits = self->bucket_bits;
+ const uint32_t slot_bits = self->slot_bits;
+
+ const uint32_t hash_shift = 64u - bucket_bits;
+ const uint32_t slot_mask = (~((uint32_t)0U)) >> (32 - slot_bits);
+ const uint64_t hash_mask = (~((uint64_t)0U)) >> (64 - hash_bits);
+
+ const uint32_t* slot_offsets = (uint32_t*)(&self[1]);
+ const uint16_t* heads = (uint16_t*)(&slot_offsets[1u << slot_bits]);
+ const uint32_t* items = (uint32_t*)(&heads[1u << bucket_bits]);
+ const uint8_t* source = NULL;
+
+ const size_t cur_ix_masked = cur_ix & ring_buffer_mask;
+ score_t best_score = out->score;
+ size_t best_len = out->len;
+ size_t i;
+ const uint64_t h =
+ (BROTLI_UNALIGNED_LOAD64LE(&data[cur_ix_masked]) & hash_mask) *
+ kPreparedDictionaryHashMul64Long;
+ const uint32_t key = (uint32_t)(h >> hash_shift);
+ const uint32_t slot = key & slot_mask;
+ const uint32_t head = heads[key];
+ const uint32_t* BROTLI_RESTRICT chain = &items[slot_offsets[slot] + head];
+ uint32_t item = (head == 0xFFFF) ? 1 : 0;
+
+ const void* tail = (void*)&items[self->num_items];
+ if (self->magic == kPreparedDictionaryMagic) {
+ source = (const uint8_t*)tail;
+ } else {
+ /* kLeanPreparedDictionaryMagic */
+ source = (const uint8_t*)BROTLI_UNALIGNED_LOAD_PTR((const uint8_t**)tail);
+ }
+
+ for (i = 0; i < 4; ++i) {
+ const size_t distance = (size_t)distance_cache[i];
+ size_t offset;
+ size_t limit;
+ size_t len;
+ if (distance <= boundary || distance > distance_offset) continue;
+ offset = distance_offset - distance;
+ limit = source_size - offset;
+ limit = limit > max_length ? max_length : limit;
+ len = FindMatchLengthWithLimit(&source[offset], &data[cur_ix_masked],
+ limit);
+ if (len >= 2) {
+ score_t score = BackwardReferenceScoreUsingLastDistance(len);
+ if (best_score < score) {
+ if (i != 0) score -= BackwardReferencePenaltyUsingLastDistance(i);
+ if (best_score < score) {
+ best_score = score;
+ if (len > best_len) best_len = len;
+ out->len = len;
+ out->len_code_delta = 0;
+ out->distance = distance;
+ out->score = best_score;
+ }
+ }
+ }
+ }
+ while (item == 0) {
+ size_t offset;
+ size_t distance;
+ size_t limit;
+ item = *chain;
+ chain++;
+ offset = item & 0x7FFFFFFF;
+ item &= 0x80000000;
+ distance = distance_offset - offset;
+ limit = source_size - offset;
+ limit = (limit > max_length) ? max_length : limit;
+ if (distance > max_distance) continue;
+ if (cur_ix_masked + best_len > ring_buffer_mask ||
+ best_len >= limit ||
+ data[cur_ix_masked + best_len] != source[offset + best_len]) {
+ continue;
+ }
+ {
+ const size_t len = FindMatchLengthWithLimit(&source[offset],
+ &data[cur_ix_masked],
+ limit);
+ if (len >= 4) {
+ score_t score = BackwardReferenceScore(len, distance);
+ if (best_score < score) {
+ best_score = score;
+ best_len = len;
+ out->len = best_len;
+ out->len_code_delta = 0;
+ out->distance = distance;
+ out->score = best_score;
+ }
+ }
+ }
+ }
+}
+
+/* NB: when seamless dictionary-ring-buffer copies are implemented, don't forget
+ to add proper guards for non-zero-BROTLI_PARAM_STREAM_OFFSET. */
+static BROTLI_INLINE size_t FindAllCompoundDictionaryMatches(
+ const PreparedDictionary* self, const uint8_t* BROTLI_RESTRICT data,
+ const size_t ring_buffer_mask, const size_t cur_ix, const size_t min_length,
+ const size_t max_length, const size_t distance_offset,
+ const size_t max_distance, BackwardMatch* matches, size_t match_limit) {
+ const uint32_t source_size = self->source_size;
+ const uint32_t hash_bits = self->hash_bits;
+ const uint32_t bucket_bits = self->bucket_bits;
+ const uint32_t slot_bits = self->slot_bits;
+
+ const uint32_t hash_shift = 64u - bucket_bits;
+ const uint32_t slot_mask = (~((uint32_t)0U)) >> (32 - slot_bits);
+ const uint64_t hash_mask = (~((uint64_t)0U)) >> (64 - hash_bits);
+
+ const uint32_t* slot_offsets = (uint32_t*)(&self[1]);
+ const uint16_t* heads = (uint16_t*)(&slot_offsets[1u << slot_bits]);
+ const uint32_t* items = (uint32_t*)(&heads[1u << bucket_bits]);
+ const uint8_t* source = NULL;
+
+ const size_t cur_ix_masked = cur_ix & ring_buffer_mask;
+ size_t best_len = min_length;
+ const uint64_t h =
+ (BROTLI_UNALIGNED_LOAD64LE(&data[cur_ix_masked]) & hash_mask) *
+ kPreparedDictionaryHashMul64Long;
+ const uint32_t key = (uint32_t)(h >> hash_shift);
+ const uint32_t slot = key & slot_mask;
+ const uint32_t head = heads[key];
+ const uint32_t* BROTLI_RESTRICT chain = &items[slot_offsets[slot] + head];
+ uint32_t item = (head == 0xFFFF) ? 1 : 0;
+ size_t found = 0;
+
+ const void* tail = (void*)&items[self->num_items];
+ if (self->magic == kPreparedDictionaryMagic) {
+ source = (const uint8_t*)tail;
+ } else {
+ /* kLeanPreparedDictionaryMagic */
+ source = (const uint8_t*)BROTLI_UNALIGNED_LOAD_PTR((const uint8_t**)tail);
+ }
+
+ while (item == 0) {
+ size_t offset;
+ size_t distance;
+ size_t limit;
+ size_t len;
+ item = *chain;
+ chain++;
+ offset = item & 0x7FFFFFFF;
+ item &= 0x80000000;
+ distance = distance_offset - offset;
+ limit = source_size - offset;
+ limit = (limit > max_length) ? max_length : limit;
+ if (distance > max_distance) continue;
+ if (cur_ix_masked + best_len > ring_buffer_mask ||
+ best_len >= limit ||
+ data[cur_ix_masked + best_len] != source[offset + best_len]) {
+ continue;
+ }
+ len = FindMatchLengthWithLimit(
+ &source[offset], &data[cur_ix_masked], limit);
+ if (len > best_len) {
+ best_len = len;
+ InitBackwardMatch(matches++, distance, len);
+ found++;
+ if (found == match_limit) break;
+ }
+ }
+ return found;
+}
+
+static BROTLI_INLINE void LookupCompoundDictionaryMatch(
+ const CompoundDictionary* addon, const uint8_t* BROTLI_RESTRICT data,
+ const size_t ring_buffer_mask, const int* BROTLI_RESTRICT distance_cache,
+ const size_t cur_ix, const size_t max_length,
+ const size_t max_ring_buffer_distance, const size_t max_distance,
+ HasherSearchResult* sr) {
+ size_t base_offset = max_ring_buffer_distance + 1 + addon->total_size - 1;
+ size_t d;
+ for (d = 0; d < addon->num_chunks; ++d) {
+ /* Only one prepared dictionary type is currently supported. */
+ FindCompoundDictionaryMatch(
+ (const PreparedDictionary*)addon->chunks[d], data, ring_buffer_mask,
+ distance_cache, cur_ix, max_length,
+ base_offset - addon->chunk_offsets[d], max_distance, sr);
+ }
+}
+
+static BROTLI_INLINE size_t LookupAllCompoundDictionaryMatches(
+ const CompoundDictionary* addon, const uint8_t* BROTLI_RESTRICT data,
+ const size_t ring_buffer_mask, const size_t cur_ix, size_t min_length,
+ const size_t max_length, const size_t max_ring_buffer_distance,
+ const size_t max_distance, BackwardMatch* matches,
+ size_t match_limit) {
+ size_t base_offset = max_ring_buffer_distance + 1 + addon->total_size - 1;
+ size_t d;
+ size_t total_found = 0;
+ for (d = 0; d < addon->num_chunks; ++d) {
+ /* Only one prepared dictionary type is currently supported. */
+ total_found += FindAllCompoundDictionaryMatches(
+ (const PreparedDictionary*)addon->chunks[d], data, ring_buffer_mask,
+ cur_ix, min_length, max_length, base_offset - addon->chunk_offsets[d],
+ max_distance, matches + total_found, match_limit - total_found);
+ if (total_found == match_limit) break;
+ if (total_found > 0) {
+ min_length = BackwardMatchLength(&matches[total_found - 1]);
+ }
+ }
+ return total_found;
+}
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_composite_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_composite_inc.h
index cba156c0e2..3923bc72c8 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_composite_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_composite_inc.h
@@ -30,10 +30,10 @@ static BROTLI_INLINE size_t FN(StoreLookahead)(void) {
typedef struct HashComposite {
HASHER_A ha;
HASHER_B hb;
+ HasherCommon ha_common;
HasherCommon hb_common;
/* Shortcuts. */
- void* extra;
HasherCommon* common;
BROTLI_BOOL fresh;
@@ -43,12 +43,12 @@ typedef struct HashComposite {
static void FN(Initialize)(HasherCommon* common,
HashComposite* BROTLI_RESTRICT self, const BrotliEncoderParams* params) {
self->common = common;
- self->extra = common->extra;
+ self->ha_common = *self->common;
self->hb_common = *self->common;
self->fresh = BROTLI_TRUE;
self->params = params;
- /* TODO: Initialize of the hashers is defered to Prepare (and params
+ /* TODO(lode): Initialize of the hashers is deferred to Prepare (and params
remembered here) because we don't get the one_shot and input_size params
here that are needed to know the memory size of them. Instead provide
those params to all hashers FN(Initialize) */
@@ -59,21 +59,36 @@ static void FN(Prepare)(
size_t input_size, const uint8_t* BROTLI_RESTRICT data) {
if (self->fresh) {
self->fresh = BROTLI_FALSE;
- self->hb_common.extra = (uint8_t*)self->extra +
- FN_A(HashMemAllocInBytes)(self->params, one_shot, input_size);
-
- FN_A(Initialize)(self->common, &self->ha, self->params);
+ self->ha_common.extra[0] = self->common->extra[0];
+ self->ha_common.extra[1] = self->common->extra[1];
+ self->ha_common.extra[2] = NULL;
+ self->ha_common.extra[3] = NULL;
+ self->hb_common.extra[0] = self->common->extra[2];
+ self->hb_common.extra[1] = self->common->extra[3];
+ self->hb_common.extra[2] = NULL;
+ self->hb_common.extra[3] = NULL;
+
+ FN_A(Initialize)(&self->ha_common, &self->ha, self->params);
FN_B(Initialize)(&self->hb_common, &self->hb, self->params);
}
FN_A(Prepare)(&self->ha, one_shot, input_size, data);
FN_B(Prepare)(&self->hb, one_shot, input_size, data);
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
- return FN_A(HashMemAllocInBytes)(params, one_shot, input_size) +
- FN_B(HashMemAllocInBytes)(params, one_shot, input_size);
+ size_t input_size, size_t* alloc_size) {
+ size_t alloc_size_a[4] = {0};
+ size_t alloc_size_b[4] = {0};
+ FN_A(HashMemAllocInBytes)(params, one_shot, input_size, alloc_size_a);
+ FN_B(HashMemAllocInBytes)(params, one_shot, input_size, alloc_size_b);
+ /* Should never happen. */
+ if (alloc_size_a[2] != 0 || alloc_size_a[3] != 0) exit(EXIT_FAILURE);
+ if (alloc_size_b[2] != 0 || alloc_size_b[3] != 0) exit(EXIT_FAILURE);
+ alloc_size[0] = alloc_size_a[0];
+ alloc_size[1] = alloc_size_a[1];
+ alloc_size[2] = alloc_size_b[0];
+ alloc_size[3] = alloc_size_b[1];
}
static BROTLI_INLINE void FN(Store)(HashComposite* BROTLI_RESTRICT self,
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_forgetful_chain_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_forgetful_chain_inc.h
index bfae6ba6a2..48e1cdcdf2 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_forgetful_chain_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_forgetful_chain_inc.h
@@ -49,7 +49,7 @@ typedef struct HashForgetfulChain {
size_t max_hops;
/* Shortcuts. */
- void* extra;
+ void* extra[2];
HasherCommon* common;
/* --- Dynamic size members --- */
@@ -77,14 +77,15 @@ static uint8_t* FN(TinyHash)(void* extra) {
}
static FN(Bank)* FN(Banks)(void* extra) {
- return (FN(Bank)*)(&FN(TinyHash)(extra)[65536]);
+ return (FN(Bank)*)(extra);
}
static void FN(Initialize)(
HasherCommon* common, HashForgetfulChain* BROTLI_RESTRICT self,
const BrotliEncoderParams* params) {
self->common = common;
- self->extra = common->extra;
+ self->extra[0] = common->extra[0];
+ self->extra[1] = common->extra[1];
self->max_hops = (params->quality > 6 ? 7u : 8u) << (params->quality - 4);
}
@@ -92,9 +93,9 @@ static void FN(Initialize)(
static void FN(Prepare)(
HashForgetfulChain* BROTLI_RESTRICT self, BROTLI_BOOL one_shot,
size_t input_size, const uint8_t* BROTLI_RESTRICT data) {
- uint32_t* BROTLI_RESTRICT addr = FN(Addr)(self->extra);
- uint16_t* BROTLI_RESTRICT head = FN(Head)(self->extra);
- uint8_t* BROTLI_RESTRICT tiny_hash = FN(TinyHash)(self->extra);
+ uint32_t* BROTLI_RESTRICT addr = FN(Addr)(self->extra[0]);
+ uint16_t* BROTLI_RESTRICT head = FN(Head)(self->extra[0]);
+ uint8_t* BROTLI_RESTRICT tiny_hash = FN(TinyHash)(self->extra[0]);
/* Partial preparation is 100 times slower (per socket). */
size_t partial_prepare_threshold = BUCKET_SIZE >> 6;
if (one_shot && input_size <= partial_prepare_threshold) {
@@ -116,24 +117,25 @@ static void FN(Prepare)(
memset(self->free_slot_idx, 0, sizeof(self->free_slot_idx));
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
+ size_t input_size, size_t* alloc_size) {
BROTLI_UNUSED(params);
BROTLI_UNUSED(one_shot);
BROTLI_UNUSED(input_size);
- return sizeof(uint32_t) * BUCKET_SIZE + sizeof(uint16_t) * BUCKET_SIZE +
- sizeof(uint8_t) * 65536 + sizeof(FN(Bank)) * NUM_BANKS;
+ alloc_size[0] = sizeof(uint32_t) * BUCKET_SIZE +
+ sizeof(uint16_t) * BUCKET_SIZE + sizeof(uint8_t) * 65536;
+ alloc_size[1] = sizeof(FN(Bank)) * NUM_BANKS;
}
/* Look at 4 bytes at &data[ix & mask]. Compute a hash from these, and prepend
node to corresponding chain; also update tiny_hash for current position. */
static BROTLI_INLINE void FN(Store)(HashForgetfulChain* BROTLI_RESTRICT self,
const uint8_t* BROTLI_RESTRICT data, const size_t mask, const size_t ix) {
- uint32_t* BROTLI_RESTRICT addr = FN(Addr)(self->extra);
- uint16_t* BROTLI_RESTRICT head = FN(Head)(self->extra);
- uint8_t* BROTLI_RESTRICT tiny_hash = FN(TinyHash)(self->extra);
- FN(Bank)* BROTLI_RESTRICT banks = FN(Banks)(self->extra);
+ uint32_t* BROTLI_RESTRICT addr = FN(Addr)(self->extra[0]);
+ uint16_t* BROTLI_RESTRICT head = FN(Head)(self->extra[0]);
+ uint8_t* BROTLI_RESTRICT tiny_hash = FN(TinyHash)(self->extra[0]);
+ FN(Bank)* BROTLI_RESTRICT banks = FN(Banks)(self->extra[1]);
const size_t key = FN(HashBytes)(&data[ix & mask]);
const size_t bank = key & (NUM_BANKS - 1);
const size_t idx = self->free_slot_idx[bank]++ & (BANK_SIZE - 1);
@@ -196,10 +198,10 @@ static BROTLI_INLINE void FN(FindLongestMatch)(
const size_t cur_ix, const size_t max_length, const size_t max_backward,
const size_t dictionary_distance, const size_t max_distance,
HasherSearchResult* BROTLI_RESTRICT out) {
- uint32_t* BROTLI_RESTRICT addr = FN(Addr)(self->extra);
- uint16_t* BROTLI_RESTRICT head = FN(Head)(self->extra);
- uint8_t* BROTLI_RESTRICT tiny_hashes = FN(TinyHash)(self->extra);
- FN(Bank)* BROTLI_RESTRICT banks = FN(Banks)(self->extra);
+ uint32_t* BROTLI_RESTRICT addr = FN(Addr)(self->extra[0]);
+ uint16_t* BROTLI_RESTRICT head = FN(Head)(self->extra[0]);
+ uint8_t* BROTLI_RESTRICT tiny_hashes = FN(TinyHash)(self->extra[0]);
+ FN(Bank)* BROTLI_RESTRICT banks = FN(Banks)(self->extra[1]);
const size_t cur_ix_masked = cur_ix & ring_buffer_mask;
/* Don't accept a short copy from far away. */
score_t min_score = out->score;
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_longest_match64_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_longest_match64_inc.h
index 956fb304b3..e48fc61554 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_longest_match64_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_longest_match64_inc.h
@@ -20,13 +20,12 @@ static BROTLI_INLINE size_t FN(HashTypeLength)(void) { return 8; }
static BROTLI_INLINE size_t FN(StoreLookahead)(void) { return 8; }
/* HashBytes is the function that chooses the bucket to place the address in. */
-static BROTLI_INLINE uint32_t FN(HashBytes)(const uint8_t* BROTLI_RESTRICT data,
- const uint64_t mask,
- const int shift) {
- const uint64_t h = (BROTLI_UNALIGNED_LOAD64LE(data) & mask) * kHashMul64Long;
+static BROTLI_INLINE size_t FN(HashBytes)(const uint8_t* BROTLI_RESTRICT data,
+ uint64_t hash_mul) {
+ const uint64_t h = BROTLI_UNALIGNED_LOAD64LE(data) * hash_mul;
/* The higher bits contain more mixture from the multiplication,
so we take our results from there. */
- return (uint32_t)(h >> shift);
+ return (size_t)(h >> (64 - 15));
}
typedef struct HashLongestMatch {
@@ -35,10 +34,8 @@ typedef struct HashLongestMatch {
/* Only block_size_ newest backward references are kept,
and the older are forgotten. */
size_t block_size_;
- /* Left-shift for computing hash bucket index from hash value. */
- int hash_shift_;
- /* Mask for selecting the next 4-8 bytes of input */
- uint64_t hash_mask_;
+ /* Hash multiplier tuned to match length. */
+ uint64_t hash_mul_;
/* Mask for accessing entries in a block (in a ring-buffer manner). */
uint32_t block_mask_;
@@ -63,16 +60,16 @@ static void FN(Initialize)(
self->common_ = common;
BROTLI_UNUSED(params);
- self->hash_shift_ = 64 - common->params.bucket_bits;
- self->hash_mask_ = (~((uint64_t)0U)) >> (64 - 8 * common->params.hash_len);
+ self->hash_mul_ = kHashMul64 << (64 - 5 * 8);
+ BROTLI_DCHECK(common->params.bucket_bits == 15);
self->bucket_size_ = (size_t)1 << common->params.bucket_bits;
self->block_bits_ = common->params.block_bits;
self->block_size_ = (size_t)1 << common->params.block_bits;
self->block_mask_ = (uint32_t)(self->block_size_ - 1);
self->num_last_distances_to_check_ =
common->params.num_last_distances_to_check;
- self->num_ = (uint16_t*)common->extra;
- self->buckets_ = (uint32_t*)&self->num_[self->bucket_size_];
+ self->num_ = (uint16_t*)common->extra[0];
+ self->buckets_ = (uint32_t*)common->extra[1];
}
static void FN(Prepare)(
@@ -84,8 +81,7 @@ static void FN(Prepare)(
if (one_shot && input_size <= partial_prepare_threshold) {
size_t i;
for (i = 0; i < input_size; ++i) {
- const uint32_t key = FN(HashBytes)(&data[i], self->hash_mask_,
- self->hash_shift_);
+ const size_t key = FN(HashBytes)(&data[i], self->hash_mul_);
num[key] = 0;
}
} else {
@@ -93,15 +89,15 @@ static void FN(Prepare)(
}
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
+ size_t input_size, size_t* alloc_size) {
size_t bucket_size = (size_t)1 << params->hasher.bucket_bits;
size_t block_size = (size_t)1 << params->hasher.block_bits;
BROTLI_UNUSED(one_shot);
BROTLI_UNUSED(input_size);
- return sizeof(uint16_t) * bucket_size +
- sizeof(uint32_t) * bucket_size * block_size;
+ alloc_size[0] = sizeof(uint16_t) * bucket_size;
+ alloc_size[1] = sizeof(uint32_t) * bucket_size * block_size;
}
/* Look at 4 bytes at &data[ix & mask].
@@ -111,8 +107,7 @@ static BROTLI_INLINE void FN(Store)(
const size_t mask, const size_t ix) {
uint16_t* BROTLI_RESTRICT num = self->num_;
uint32_t* BROTLI_RESTRICT buckets = self->buckets_;
- const uint32_t key = FN(HashBytes)(&data[ix & mask], self->hash_mask_,
- self->hash_shift_);
+ const size_t key = FN(HashBytes)(&data[ix & mask], self->hash_mul_);
const size_t minor_ix = num[key] & self->block_mask_;
const size_t offset = minor_ix + (key << self->block_bits_);
++num[key];
@@ -217,14 +212,17 @@ static BROTLI_INLINE void FN(FindLongestMatch)(
}
}
{
- const uint32_t key = FN(HashBytes)(
- &data[cur_ix_masked], self->hash_mask_, self->hash_shift_);
+ const size_t key = FN(HashBytes)(&data[cur_ix_masked], self->hash_mul_);
uint32_t* BROTLI_RESTRICT bucket = &buckets[key << self->block_bits_];
const size_t down =
(num[key] > self->block_size_) ?
(num[key] - self->block_size_) : 0u;
- for (i = num[key]; i > down;) {
+ const uint32_t first4 = BrotliUnalignedRead32(data + cur_ix_masked);
+ const size_t max_length_m4 = max_length - 4;
+ i = num[key];
+ for (; i > down;) {
size_t prev_ix = bucket[--i & self->block_mask_];
+ uint32_t current4;
const size_t backward = cur_ix - prev_ix;
if (BROTLI_PREDICT_FALSE(backward > max_backward)) {
break;
@@ -235,22 +233,19 @@ static BROTLI_INLINE void FN(FindLongestMatch)(
data[cur_ix_masked + best_len] != data[prev_ix + best_len]) {
continue;
}
+ current4 = BrotliUnalignedRead32(data + prev_ix);
+ if (first4 != current4) continue;
{
- const size_t len = FindMatchLengthWithLimit(&data[prev_ix],
- &data[cur_ix_masked],
- max_length);
- if (len >= 4) {
- /* Comparing for >= 3 does not change the semantics, but just saves
- for a few unnecessary binary logarithms in backward reference
- score, since we are not interested in such short matches. */
- score_t score = BackwardReferenceScore(len, backward);
- if (best_score < score) {
- best_score = score;
- best_len = len;
- out->len = best_len;
- out->distance = backward;
- out->score = best_score;
- }
+ const size_t len = FindMatchLengthWithLimit(&data[prev_ix + 4],
+ &data[cur_ix_masked + 4],
+ max_length_m4) + 4;
+ const score_t score = BackwardReferenceScore(len, backward);
+ if (best_score < score) {
+ best_score = score;
+ best_len = len;
+ out->len = best_len;
+ out->distance = backward;
+ out->score = best_score;
}
}
}
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_inc.h
index 27f4463d7f..788e9ef993 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_inc.h
@@ -54,10 +54,6 @@ typedef struct HashLongestMatch {
uint32_t* buckets_; /* uint32_t[bucket_size * block_size]; */
} HashLongestMatch;
-static BROTLI_INLINE uint16_t* FN(Num)(void* extra) {
- return (uint16_t*)extra;
-}
-
static void FN(Initialize)(
HasherCommon* common, HashLongestMatch* BROTLI_RESTRICT self,
const BrotliEncoderParams* params) {
@@ -68,8 +64,8 @@ static void FN(Initialize)(
self->bucket_size_ = (size_t)1 << common->params.bucket_bits;
self->block_size_ = (size_t)1 << common->params.block_bits;
self->block_mask_ = (uint32_t)(self->block_size_ - 1);
- self->num_ = (uint16_t*)common->extra;
- self->buckets_ = (uint32_t*)(&self->num_[self->bucket_size_]);
+ self->num_ = (uint16_t*)common->extra[0];
+ self->buckets_ = (uint32_t*)common->extra[1];
self->block_bits_ = common->params.block_bits;
self->num_last_distances_to_check_ =
common->params.num_last_distances_to_check;
@@ -92,15 +88,15 @@ static void FN(Prepare)(
}
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
+ size_t input_size, size_t* alloc_size) {
size_t bucket_size = (size_t)1 << params->hasher.bucket_bits;
size_t block_size = (size_t)1 << params->hasher.block_bits;
BROTLI_UNUSED(one_shot);
BROTLI_UNUSED(input_size);
- return sizeof(uint16_t) * bucket_size +
- sizeof(uint32_t) * bucket_size * block_size;
+ alloc_size[0] = sizeof(uint16_t) * bucket_size;
+ alloc_size[1] = sizeof(uint32_t) * bucket_size * block_size;
}
/* Look at 4 bytes at &data[ix & mask].
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_quickly_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_quickly_inc.h
index e5ba840ab9..54397ef891 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_quickly_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_longest_match_quickly_inc.h
@@ -49,7 +49,7 @@ static void FN(Initialize)(
self->common = common;
BROTLI_UNUSED(params);
- self->buckets_ = (uint32_t*)common->extra;
+ self->buckets_ = (uint32_t*)common->extra[0];
}
static void FN(Prepare)(
@@ -80,13 +80,13 @@ static void FN(Prepare)(
}
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
+ size_t input_size, size_t* alloc_size) {
BROTLI_UNUSED(params);
BROTLI_UNUSED(one_shot);
BROTLI_UNUSED(input_size);
- return sizeof(uint32_t) * BUCKET_SIZE;
+ alloc_size[0] = sizeof(uint32_t) * BUCKET_SIZE;
}
/* Look at 5 bytes at &data[ix & mask].
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_rolling_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_rolling_inc.h
index 586ae73859..4c7a6b199a 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_rolling_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_rolling_inc.h
@@ -67,7 +67,7 @@ static void FN(Initialize)(
self->factor_remove *= self->factor;
}
- self->table = (uint32_t*)common->extra;
+ self->table = (uint32_t*)common->extra[0];
for (i = 0; i < NUMBUCKETS; i++) {
self->table[i] = FN(kInvalidPos);
}
@@ -88,13 +88,13 @@ static void FN(Prepare)(HashRolling* BROTLI_RESTRICT self, BROTLI_BOOL one_shot,
BROTLI_UNUSED(one_shot);
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
- return NUMBUCKETS * sizeof(uint32_t);
+ size_t input_size, size_t* alloc_size) {
BROTLI_UNUSED(params);
BROTLI_UNUSED(one_shot);
BROTLI_UNUSED(input_size);
+ alloc_size[0] = NUMBUCKETS * sizeof(uint32_t);
}
static BROTLI_INLINE void FN(Store)(HashRolling* BROTLI_RESTRICT self,
diff --git a/dviware/dvisvgm/libs/brotli/enc/hash_to_binary_tree_inc.h b/dviware/dvisvgm/libs/brotli/enc/hash_to_binary_tree_inc.h
index 9880e0aef6..a639d2d4b0 100644
--- a/dviware/dvisvgm/libs/brotli/enc/hash_to_binary_tree_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/hash_to_binary_tree_inc.h
@@ -57,8 +57,8 @@ typedef struct HashToBinaryTree {
static void FN(Initialize)(
HasherCommon* common, HashToBinaryTree* BROTLI_RESTRICT self,
const BrotliEncoderParams* params) {
- self->buckets_ = (uint32_t*)common->extra;
- self->forest_ = &self->buckets_[BUCKET_SIZE];
+ self->buckets_ = (uint32_t*)common->extra[0];
+ self->forest_ = (uint32_t*)common->extra[1];
self->window_mask_ = (1u << params->lgwin) - 1u;
self->invalid_pos_ = (uint32_t)(0 - self->window_mask_);
@@ -78,14 +78,15 @@ static void FN(Prepare)
}
}
-static BROTLI_INLINE size_t FN(HashMemAllocInBytes)(
+static BROTLI_INLINE void FN(HashMemAllocInBytes)(
const BrotliEncoderParams* params, BROTLI_BOOL one_shot,
- size_t input_size) {
+ size_t input_size, size_t* alloc_size) {
size_t num_nodes = (size_t)1 << params->lgwin;
if (one_shot && input_size < num_nodes) {
num_nodes = input_size;
}
- return sizeof(uint32_t) * BUCKET_SIZE + 2 * sizeof(uint32_t) * num_nodes;
+ alloc_size[0] = sizeof(uint32_t) * BUCKET_SIZE;
+ alloc_size[1] = 2 * sizeof(uint32_t) * num_nodes;
}
static BROTLI_INLINE size_t FN(LeftChildIndex)(
diff --git a/dviware/dvisvgm/libs/brotli/enc/histogram.c b/dviware/dvisvgm/libs/brotli/enc/histogram.c
index 6da2ff6bb4..4dbb87f907 100644
--- a/dviware/dvisvgm/libs/brotli/enc/histogram.c
+++ b/dviware/dvisvgm/libs/brotli/enc/histogram.c
@@ -6,11 +6,11 @@
/* Build per-context histograms of literals, commands and distance codes. */
-#include "./histogram.h"
+#include "histogram.h"
#include "../common/context.h"
-#include "./block_splitter.h"
-#include "./command.h"
+#include "block_splitter.h"
+#include "command.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -63,7 +63,7 @@ void BrotliBuildHistogramsWithContext(
BlockSplitIteratorNext(&insert_and_copy_it);
HistogramAddCommand(&insert_and_copy_histograms[insert_and_copy_it.type_],
cmd->cmd_prefix_);
- /* TODO: unwrap iterator blocks. */
+ /* TODO(eustas): unwrap iterator blocks. */
for (j = cmd->insert_len_; j != 0; --j) {
size_t context;
BlockSplitIteratorNext(&literal_it);
diff --git a/dviware/dvisvgm/libs/brotli/enc/histogram.h b/dviware/dvisvgm/libs/brotli/enc/histogram.h
index 42af3c3f9d..d1abd973c1 100644
--- a/dviware/dvisvgm/libs/brotli/enc/histogram.h
+++ b/dviware/dvisvgm/libs/brotli/enc/histogram.h
@@ -11,12 +11,13 @@
#include <string.h> /* memset */
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/context.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./block_splitter.h"
-#include "./command.h"
+#include "block_splitter.h"
+#include "command.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -28,7 +29,7 @@ extern "C" {
#define FN(X) X ## Literal
#define DATA_SIZE BROTLI_NUM_LITERAL_SYMBOLS
#define DataType uint8_t
-#include "./histogram_inc.h" /* NOLINT(build/include) */
+#include "histogram_inc.h" /* NOLINT(build/include) */
#undef DataType
#undef DATA_SIZE
#undef FN
@@ -36,13 +37,13 @@ extern "C" {
#define FN(X) X ## Command
#define DataType uint16_t
#define DATA_SIZE BROTLI_NUM_COMMAND_SYMBOLS
-#include "./histogram_inc.h" /* NOLINT(build/include) */
+#include "histogram_inc.h" /* NOLINT(build/include) */
#undef DATA_SIZE
#undef FN
#define FN(X) X ## Distance
#define DATA_SIZE BROTLI_NUM_HISTOGRAM_DISTANCE_SYMBOLS
-#include "./histogram_inc.h" /* NOLINT(build/include) */
+#include "histogram_inc.h" /* NOLINT(build/include) */
#undef DataType
#undef DATA_SIZE
#undef FN
diff --git a/dviware/dvisvgm/libs/brotli/enc/literal_cost.c b/dviware/dvisvgm/libs/brotli/enc/literal_cost.c
index c231100e34..2ac847f3c9 100644
--- a/dviware/dvisvgm/libs/brotli/enc/literal_cost.c
+++ b/dviware/dvisvgm/libs/brotli/enc/literal_cost.c
@@ -7,12 +7,15 @@
/* Literal cost model to allow backward reference replacement to be efficient.
*/
-#include "./literal_cost.h"
+#include "literal_cost.h"
+
+#include <string.h> /* memset */
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./fast_log.h"
-#include "./utf8_util.h"
+
+#include "../common/platform.h"
+#include "fast_log.h"
+#include "utf8_util.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -54,22 +57,23 @@ static size_t DecideMultiByteStatsLevel(size_t pos, size_t len, size_t mask,
}
static void EstimateBitCostsForLiteralsUTF8(size_t pos, size_t len, size_t mask,
- const uint8_t* data, float* cost) {
+ const uint8_t* data,
+ size_t* histogram, float* cost) {
/* max_utf8 is 0 (normal ASCII single byte modeling),
1 (for 2-byte UTF-8 modeling), or 2 (for 3-byte UTF-8 modeling). */
const size_t max_utf8 = DecideMultiByteStatsLevel(pos, len, mask, data);
- size_t histogram[3][256] = { { 0 } };
size_t window_half = 495;
size_t in_window = BROTLI_MIN(size_t, window_half, len);
size_t in_window_utf8[3] = { 0 };
-
size_t i;
+ memset(histogram, 0, 3 * 256 * sizeof(histogram[0]));
+
{ /* Bootstrap histograms. */
size_t last_c = 0;
size_t utf8_pos = 0;
for (i = 0; i < in_window; ++i) {
size_t c = data[(pos + i) & mask];
- ++histogram[utf8_pos][c];
+ ++histogram[256 * utf8_pos + c];
++in_window_utf8[utf8_pos];
utf8_pos = UTF8Position(last_c, c, max_utf8);
last_c = c;
@@ -85,7 +89,7 @@ static void EstimateBitCostsForLiteralsUTF8(size_t pos, size_t len, size_t mask,
size_t last_c =
i < window_half + 2 ? 0 : data[(pos + i - window_half - 2) & mask];
size_t utf8_pos2 = UTF8Position(last_c, c, max_utf8);
- --histogram[utf8_pos2][data[(pos + i - window_half) & mask]];
+ --histogram[256 * utf8_pos2 + data[(pos + i - window_half) & mask]];
--in_window_utf8[utf8_pos2];
}
if (i + window_half < len) {
@@ -93,7 +97,7 @@ static void EstimateBitCostsForLiteralsUTF8(size_t pos, size_t len, size_t mask,
size_t c = data[(pos + i + window_half - 1) & mask];
size_t last_c = data[(pos + i + window_half - 2) & mask];
size_t utf8_pos2 = UTF8Position(last_c, c, max_utf8);
- ++histogram[utf8_pos2][data[(pos + i + window_half) & mask]];
+ ++histogram[256 * utf8_pos2 + data[(pos + i + window_half) & mask]];
++in_window_utf8[utf8_pos2];
}
{
@@ -101,7 +105,7 @@ static void EstimateBitCostsForLiteralsUTF8(size_t pos, size_t len, size_t mask,
size_t last_c = i < 2 ? 0 : data[(pos + i - 2) & mask];
size_t utf8_pos = UTF8Position(last_c, c, max_utf8);
size_t masked_pos = (pos + i) & mask;
- size_t histo = histogram[utf8_pos][data[masked_pos]];
+ size_t histo = histogram[256 * utf8_pos + data[masked_pos]];
double lit_cost;
if (histo == 0) {
histo = 1;
@@ -125,17 +129,18 @@ static void EstimateBitCostsForLiteralsUTF8(size_t pos, size_t len, size_t mask,
}
void BrotliEstimateBitCostsForLiterals(size_t pos, size_t len, size_t mask,
- const uint8_t* data, float* cost) {
+ const uint8_t* data,
+ size_t* histogram, float* cost) {
if (BrotliIsMostlyUTF8(data, pos, mask, len, kMinUTF8Ratio)) {
- EstimateBitCostsForLiteralsUTF8(pos, len, mask, data, cost);
+ EstimateBitCostsForLiteralsUTF8(pos, len, mask, data, histogram, cost);
return;
} else {
- size_t histogram[256] = { 0 };
size_t window_half = 2000;
size_t in_window = BROTLI_MIN(size_t, window_half, len);
+ size_t i;
+ memset(histogram, 0, 256 * sizeof(histogram[0]));
/* Bootstrap histogram. */
- size_t i;
for (i = 0; i < in_window; ++i) {
++histogram[data[(pos + i) & mask]];
}
diff --git a/dviware/dvisvgm/libs/brotli/enc/literal_cost.h b/dviware/dvisvgm/libs/brotli/enc/literal_cost.h
index 8f53f39d3f..284a8e5af7 100644
--- a/dviware/dvisvgm/libs/brotli/enc/literal_cost.h
+++ b/dviware/dvisvgm/libs/brotli/enc/literal_cost.h
@@ -10,9 +10,10 @@
#ifndef BROTLI_ENC_LITERAL_COST_H_
#define BROTLI_ENC_LITERAL_COST_H_
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
@@ -21,7 +22,8 @@ extern "C" {
ring-buffer (data, mask) will take entropy coded and writes these estimates
to the cost[0..len) array. */
BROTLI_INTERNAL void BrotliEstimateBitCostsForLiterals(
- size_t pos, size_t len, size_t mask, const uint8_t* data, float* cost);
+ size_t pos, size_t len, size_t mask, const uint8_t* data, size_t* histogram,
+ float* cost);
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/memory.c b/dviware/dvisvgm/libs/brotli/enc/memory.c
index f6ed7e3cb7..bb5e364198 100644
--- a/dviware/dvisvgm/libs/brotli/enc/memory.c
+++ b/dviware/dvisvgm/libs/brotli/enc/memory.c
@@ -7,21 +7,22 @@
/* Algorithms for distributing the literals and commands of a metablock between
block types and contexts. */
-#include "./memory.h"
+#include "memory.h"
#include <stdlib.h> /* exit, free, malloc */
#include <string.h> /* memcpy */
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
-#define MAX_PERM_ALLOCATED 128
-#define MAX_NEW_ALLOCATED 64
-#define MAX_NEW_FREED 64
+#define MAX_NEW_ALLOCATED (BROTLI_ENCODER_MEMORY_MANAGER_SLOTS >> 2)
+#define MAX_NEW_FREED (BROTLI_ENCODER_MEMORY_MANAGER_SLOTS >> 2)
+#define MAX_PERM_ALLOCATED (BROTLI_ENCODER_MEMORY_MANAGER_SLOTS >> 1)
#define PERM_ALLOCATED_OFFSET 0
#define NEW_ALLOCATED_OFFSET MAX_PERM_ALLOCATED
@@ -67,6 +68,7 @@ void BrotliWipeOutMemoryManager(MemoryManager* m) {
static void SortPointers(void** items, const size_t n) {
/* Shell sort. */
+ /* TODO(eustas): fine-tune for "many slots" case */
static const size_t gaps[] = {23, 10, 4, 1};
int g = 0;
for (; g < 4; ++g) {
@@ -165,6 +167,28 @@ void BrotliWipeOutMemoryManager(MemoryManager* m) {
#endif /* BROTLI_ENCODER_EXIT_ON_OOM */
+void* BrotliBootstrapAlloc(size_t size,
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
+ if (!alloc_func && !free_func) {
+ return malloc(size);
+ } else if (alloc_func && free_func) {
+ return alloc_func(opaque, size);
+ }
+ return NULL;
+}
+
+void BrotliBootstrapFree(void* address, MemoryManager* m) {
+ if (!address) {
+ /* Should not happen! */
+ return;
+ } else {
+ /* Copy values, as those would be freed. */
+ brotli_free_func free_func = m->free_func;
+ void* opaque = m->opaque;
+ free_func(opaque, address);
+ }
+}
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/memory.h b/dviware/dvisvgm/libs/brotli/enc/memory.h
index 832e7b2b6e..a4417df4cb 100644
--- a/dviware/dvisvgm/libs/brotli/enc/memory.h
+++ b/dviware/dvisvgm/libs/brotli/enc/memory.h
@@ -11,9 +11,10 @@
#include <string.h> /* memcpy */
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
@@ -23,6 +24,16 @@ extern "C" {
#define BROTLI_ENCODER_EXIT_ON_OOM
#endif
+#if !defined(BROTLI_ENCODER_EXIT_ON_OOM)
+#if defined(BROTLI_EXPERIMENTAL)
+#define BROTLI_ENCODER_MEMORY_MANAGER_SLOTS (48*1024)
+#else /* BROTLI_EXPERIMENTAL */
+#define BROTLI_ENCODER_MEMORY_MANAGER_SLOTS 256
+#endif /* BROTLI_EXPERIMENTAL */
+#else /* BROTLI_ENCODER_EXIT_ON_OOM */
+#define BROTLI_ENCODER_MEMORY_MANAGER_SLOTS 0
+#endif /* BROTLI_ENCODER_EXIT_ON_OOM */
+
typedef struct MemoryManager {
brotli_alloc_func alloc_func;
brotli_free_func free_func;
@@ -32,7 +43,7 @@ typedef struct MemoryManager {
size_t perm_allocated;
size_t new_allocated;
size_t new_freed;
- void* pointers[256];
+ void* pointers[BROTLI_ENCODER_MEMORY_MANAGER_SLOTS];
#endif /* BROTLI_ENCODER_EXIT_ON_OOM */
} MemoryManager;
@@ -107,6 +118,12 @@ V: value to append
A[(S) - 1] = (V); \
}
+/* "Bootstrap" allocations are not tracked by memory manager; should be used
+ only to allocate MemoryManager itself (or structure containing it). */
+BROTLI_INTERNAL void* BrotliBootstrapAlloc(size_t size,
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+BROTLI_INTERNAL void BrotliBootstrapFree(void* address, MemoryManager* m);
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/metablock.c b/dviware/dvisvgm/libs/brotli/enc/metablock.c
index 5aa4d4f17c..0c5c078d05 100644
--- a/dviware/dvisvgm/libs/brotli/enc/metablock.c
+++ b/dviware/dvisvgm/libs/brotli/enc/metablock.c
@@ -7,27 +7,27 @@
/* Algorithms for distributing the literals and commands of a metablock between
block types and contexts. */
-#include "./metablock.h"
+#include "metablock.h"
+
+#include <brotli/types.h>
#include "../common/constants.h"
#include "../common/context.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./bit_cost.h"
-#include "./block_splitter.h"
-#include "./cluster.h"
-#include "./entropy_encode.h"
-#include "./histogram.h"
-#include "./memory.h"
-#include "./quality.h"
+#include "bit_cost.h"
+#include "block_splitter.h"
+#include "cluster.h"
+#include "entropy_encode.h"
+#include "histogram.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
-void BrotliInitDistanceParams(BrotliEncoderParams* params,
- uint32_t npostfix, uint32_t ndirect) {
- BrotliDistanceParams* dist_params = &params->dist;
+void BrotliInitDistanceParams(BrotliDistanceParams* dist_params,
+ uint32_t npostfix, uint32_t ndirect, BROTLI_BOOL large_window) {
uint32_t alphabet_size_max;
uint32_t alphabet_size_limit;
uint32_t max_distance;
@@ -41,7 +41,7 @@ void BrotliInitDistanceParams(BrotliEncoderParams* params,
max_distance = ndirect + (1U << (BROTLI_MAX_DISTANCE_BITS + npostfix + 2)) -
(1U << (npostfix + 2));
- if (params->large_window) {
+ if (large_window) {
BrotliDistanceCodeLimit limit = BrotliCalculateDistanceCodeLimit(
BROTLI_MAX_ALLOWED_DISTANCE, npostfix, ndirect);
alphabet_size_max = BROTLI_DISTANCE_ALPHABET_SIZE(
@@ -83,14 +83,14 @@ static BROTLI_BOOL ComputeDistanceCost(const Command* cmds,
size_t num_commands,
const BrotliDistanceParams* orig_params,
const BrotliDistanceParams* new_params,
- double* cost) {
+ double* cost,
+ HistogramDistance* tmp) {
size_t i;
BROTLI_BOOL equal_params = BROTLI_FALSE;
uint16_t dist_prefix;
uint32_t dist_extra;
double extra_bits = 0.0;
- HistogramDistance histo;
- HistogramClearDistance(&histo);
+ HistogramClearDistance(tmp);
if (orig_params->distance_postfix_bits == new_params->distance_postfix_bits &&
orig_params->num_direct_distance_codes ==
@@ -114,12 +114,12 @@ static BROTLI_BOOL ComputeDistanceCost(const Command* cmds,
&dist_prefix,
&dist_extra);
}
- HistogramAddDistance(&histo, dist_prefix & 0x3FF);
+ HistogramAddDistance(tmp, dist_prefix & 0x3FF);
extra_bits += dist_prefix >> 10;
}
}
- *cost = BrotliPopulationCostDistance(&histo) + extra_bits;
+ *cost = BrotliPopulationCostDistance(tmp) + extra_bits;
return BROTLI_TRUE;
}
@@ -147,43 +147,46 @@ void BrotliBuildMetaBlock(MemoryManager* m,
uint32_t ndirect_msb = 0;
BROTLI_BOOL check_orig = BROTLI_TRUE;
double best_dist_cost = 1e99;
- BrotliEncoderParams orig_params = *params;
- BrotliEncoderParams new_params = *params;
+ BrotliDistanceParams orig_params = params->dist;
+ BrotliDistanceParams new_params = params->dist;
+ HistogramDistance* tmp = BROTLI_ALLOC(m, HistogramDistance, 1);
+
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tmp)) return;
for (npostfix = 0; npostfix <= BROTLI_MAX_NPOSTFIX; npostfix++) {
for (; ndirect_msb < 16; ndirect_msb++) {
uint32_t ndirect = ndirect_msb << npostfix;
BROTLI_BOOL skip;
double dist_cost;
- BrotliInitDistanceParams(&new_params, npostfix, ndirect);
- if (npostfix == orig_params.dist.distance_postfix_bits &&
- ndirect == orig_params.dist.num_direct_distance_codes) {
+ BrotliInitDistanceParams(&new_params, npostfix, ndirect,
+ params->large_window);
+ if (npostfix == orig_params.distance_postfix_bits &&
+ ndirect == orig_params.num_direct_distance_codes) {
check_orig = BROTLI_FALSE;
}
skip = !ComputeDistanceCost(
- cmds, num_commands,
- &orig_params.dist, &new_params.dist, &dist_cost);
+ cmds, num_commands, &orig_params, &new_params, &dist_cost, tmp);
if (skip || (dist_cost > best_dist_cost)) {
break;
}
best_dist_cost = dist_cost;
- params->dist = new_params.dist;
+ params->dist = new_params;
}
if (ndirect_msb > 0) ndirect_msb--;
ndirect_msb /= 2;
}
if (check_orig) {
double dist_cost;
- ComputeDistanceCost(cmds, num_commands,
- &orig_params.dist, &orig_params.dist, &dist_cost);
+ ComputeDistanceCost(cmds, num_commands, &orig_params, &orig_params,
+ &dist_cost, tmp);
if (dist_cost < best_dist_cost) {
/* NB: currently unused; uncomment when more param tuning is added. */
/* best_dist_cost = dist_cost; */
- params->dist = orig_params.dist;
+ params->dist = orig_params;
}
}
- RecomputeDistancePrefixes(cmds, num_commands,
- &orig_params.dist, &params->dist);
+ BROTLI_FREE(m, tmp);
+ RecomputeDistancePrefixes(cmds, num_commands, &orig_params, &params->dist);
BrotliSplitBlock(m, cmds, num_commands,
ringbuffer, pos, mask, params,
@@ -284,15 +287,15 @@ void BrotliBuildMetaBlock(MemoryManager* m,
}
#define FN(X) X ## Literal
-#include "./metablock_inc.h" /* NOLINT(build/include) */
+#include "metablock_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Command
-#include "./metablock_inc.h" /* NOLINT(build/include) */
+#include "metablock_inc.h" /* NOLINT(build/include) */
#undef FN
#define FN(X) X ## Distance
-#include "./metablock_inc.h" /* NOLINT(build/include) */
+#include "metablock_inc.h" /* NOLINT(build/include) */
#undef FN
#define BROTLI_MAX_STATIC_CONTEXTS 13
@@ -535,17 +538,21 @@ static void MapStaticContexts(MemoryManager* m,
}
}
-static BROTLI_INLINE void BrotliBuildMetaBlockGreedyInternal(
- MemoryManager* m, const uint8_t* ringbuffer, size_t pos, size_t mask,
- uint8_t prev_byte, uint8_t prev_byte2, ContextLut literal_context_lut,
- const size_t num_contexts, const uint32_t* static_context_map,
- const Command* commands, size_t n_commands, MetaBlockSplit* mb) {
+typedef struct GreedyMetablockArena {
union {
BlockSplitterLiteral plain;
ContextBlockSplitter ctx;
} lit_blocks;
BlockSplitterCommand cmd_blocks;
BlockSplitterDistance dist_blocks;
+} GreedyMetablockArena;
+
+static BROTLI_INLINE void BrotliBuildMetaBlockGreedyInternal(
+ MemoryManager* m, GreedyMetablockArena* arena, const uint8_t* ringbuffer,
+ size_t pos, size_t mask, uint8_t prev_byte, uint8_t prev_byte2,
+ ContextLut literal_context_lut, const size_t num_contexts,
+ const uint32_t* static_context_map, const Command* commands,
+ size_t n_commands, MetaBlockSplit* mb) {
size_t num_literals = 0;
size_t i;
for (i = 0; i < n_commands; ++i) {
@@ -553,20 +560,20 @@ static BROTLI_INLINE void BrotliBuildMetaBlockGreedyInternal(
}
if (num_contexts == 1) {
- InitBlockSplitterLiteral(m, &lit_blocks.plain, 256, 512, 400.0,
+ InitBlockSplitterLiteral(m, &arena->lit_blocks.plain, 256, 512, 400.0,
num_literals, &mb->literal_split, &mb->literal_histograms,
&mb->literal_histograms_size);
} else {
- InitContextBlockSplitter(m, &lit_blocks.ctx, 256, num_contexts, 512, 400.0,
- num_literals, &mb->literal_split, &mb->literal_histograms,
+ InitContextBlockSplitter(m, &arena->lit_blocks.ctx, 256, num_contexts, 512,
+ 400.0, num_literals, &mb->literal_split, &mb->literal_histograms,
&mb->literal_histograms_size);
}
if (BROTLI_IS_OOM(m)) return;
- InitBlockSplitterCommand(m, &cmd_blocks, BROTLI_NUM_COMMAND_SYMBOLS, 1024,
- 500.0, n_commands, &mb->command_split, &mb->command_histograms,
+ InitBlockSplitterCommand(m, &arena->cmd_blocks, BROTLI_NUM_COMMAND_SYMBOLS,
+ 1024, 500.0, n_commands, &mb->command_split, &mb->command_histograms,
&mb->command_histograms_size);
if (BROTLI_IS_OOM(m)) return;
- InitBlockSplitterDistance(m, &dist_blocks, 64, 512, 100.0, n_commands,
+ InitBlockSplitterDistance(m, &arena->dist_blocks, 64, 512, 100.0, n_commands,
&mb->distance_split, &mb->distance_histograms,
&mb->distance_histograms_size);
if (BROTLI_IS_OOM(m)) return;
@@ -574,15 +581,15 @@ static BROTLI_INLINE void BrotliBuildMetaBlockGreedyInternal(
for (i = 0; i < n_commands; ++i) {
const Command cmd = commands[i];
size_t j;
- BlockSplitterAddSymbolCommand(&cmd_blocks, cmd.cmd_prefix_);
+ BlockSplitterAddSymbolCommand(&arena->cmd_blocks, cmd.cmd_prefix_);
for (j = cmd.insert_len_; j != 0; --j) {
uint8_t literal = ringbuffer[pos & mask];
if (num_contexts == 1) {
- BlockSplitterAddSymbolLiteral(&lit_blocks.plain, literal);
+ BlockSplitterAddSymbolLiteral(&arena->lit_blocks.plain, literal);
} else {
size_t context =
BROTLI_CONTEXT(prev_byte, prev_byte2, literal_context_lut);
- ContextBlockSplitterAddSymbol(&lit_blocks.ctx, m, literal,
+ ContextBlockSplitterAddSymbol(&arena->lit_blocks.ctx, m, literal,
static_context_map[context]);
if (BROTLI_IS_OOM(m)) return;
}
@@ -595,21 +602,24 @@ static BROTLI_INLINE void BrotliBuildMetaBlockGreedyInternal(
prev_byte2 = ringbuffer[(pos - 2) & mask];
prev_byte = ringbuffer[(pos - 1) & mask];
if (cmd.cmd_prefix_ >= 128) {
- BlockSplitterAddSymbolDistance(&dist_blocks, cmd.dist_prefix_ & 0x3FF);
+ BlockSplitterAddSymbolDistance(
+ &arena->dist_blocks, cmd.dist_prefix_ & 0x3FF);
}
}
}
if (num_contexts == 1) {
BlockSplitterFinishBlockLiteral(
- &lit_blocks.plain, /* is_final = */ BROTLI_TRUE);
+ &arena->lit_blocks.plain, /* is_final = */ BROTLI_TRUE);
} else {
ContextBlockSplitterFinishBlock(
- &lit_blocks.ctx, m, /* is_final = */ BROTLI_TRUE);
+ &arena->lit_blocks.ctx, m, /* is_final = */ BROTLI_TRUE);
if (BROTLI_IS_OOM(m)) return;
}
- BlockSplitterFinishBlockCommand(&cmd_blocks, /* is_final = */ BROTLI_TRUE);
- BlockSplitterFinishBlockDistance(&dist_blocks, /* is_final = */ BROTLI_TRUE);
+ BlockSplitterFinishBlockCommand(
+ &arena->cmd_blocks, /* is_final = */ BROTLI_TRUE);
+ BlockSplitterFinishBlockDistance(
+ &arena->dist_blocks, /* is_final = */ BROTLI_TRUE);
if (num_contexts > 1) {
MapStaticContexts(m, num_contexts, static_context_map, mb);
@@ -628,14 +638,18 @@ void BrotliBuildMetaBlockGreedy(MemoryManager* m,
const Command* commands,
size_t n_commands,
MetaBlockSplit* mb) {
+ GreedyMetablockArena* arena = BROTLI_ALLOC(m, GreedyMetablockArena, 1);
+ if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(arena)) return;
if (num_contexts == 1) {
- BrotliBuildMetaBlockGreedyInternal(m, ringbuffer, pos, mask, prev_byte,
- prev_byte2, literal_context_lut, 1, NULL, commands, n_commands, mb);
+ BrotliBuildMetaBlockGreedyInternal(m, arena, ringbuffer, pos, mask,
+ prev_byte, prev_byte2, literal_context_lut, 1, NULL, commands,
+ n_commands, mb);
} else {
- BrotliBuildMetaBlockGreedyInternal(m, ringbuffer, pos, mask, prev_byte,
- prev_byte2, literal_context_lut, num_contexts, static_context_map,
- commands, n_commands, mb);
+ BrotliBuildMetaBlockGreedyInternal(m, arena, ringbuffer, pos, mask,
+ prev_byte, prev_byte2, literal_context_lut, num_contexts,
+ static_context_map, commands, n_commands, mb);
}
+ BROTLI_FREE(m, arena);
}
void BrotliOptimizeHistograms(uint32_t num_distance_codes,
diff --git a/dviware/dvisvgm/libs/brotli/enc/metablock.h b/dviware/dvisvgm/libs/brotli/enc/metablock.h
index 334a79a443..db38f8fd21 100644
--- a/dviware/dvisvgm/libs/brotli/enc/metablock.h
+++ b/dviware/dvisvgm/libs/brotli/enc/metablock.h
@@ -10,14 +10,15 @@
#ifndef BROTLI_ENC_METABLOCK_H_
#define BROTLI_ENC_METABLOCK_H_
+#include <brotli/types.h>
+
#include "../common/context.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./block_splitter.h"
-#include "./command.h"
-#include "./histogram.h"
-#include "./memory.h"
-#include "./quality.h"
+#include "block_splitter.h"
+#include "command.h"
+#include "histogram.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -95,8 +96,8 @@ BROTLI_INTERNAL void BrotliBuildMetaBlockGreedy(
BROTLI_INTERNAL void BrotliOptimizeHistograms(uint32_t num_distance_codes,
MetaBlockSplit* mb);
-BROTLI_INTERNAL void BrotliInitDistanceParams(BrotliEncoderParams* params,
- uint32_t npostfix, uint32_t ndirect);
+BROTLI_INTERNAL void BrotliInitDistanceParams(BrotliDistanceParams* params,
+ uint32_t npostfix, uint32_t ndirect, BROTLI_BOOL large_window);
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/metablock_inc.h b/dviware/dvisvgm/libs/brotli/enc/metablock_inc.h
index ed507ef5ef..f9393869ab 100644
--- a/dviware/dvisvgm/libs/brotli/enc/metablock_inc.h
+++ b/dviware/dvisvgm/libs/brotli/enc/metablock_inc.h
@@ -27,6 +27,9 @@ typedef struct FN(BlockSplitter) {
HistogramType* histograms_; /* not owned */
size_t* histograms_size_; /* not owned */
+ /* Temporary storage for BlockSplitterFinishBlock. */
+ HistogramType combined_histo[2];
+
/* The number of symbols that we want to collect before deciding on whether
or not to merge the block with a previous one or emit a new block. */
size_t target_block_size_;
@@ -104,17 +107,16 @@ static void FN(BlockSplitterFinishBlock)(
} else if (self->block_size_ > 0) {
double entropy = BitsEntropy(histograms[self->curr_histogram_ix_].data_,
self->alphabet_size_);
- HistogramType combined_histo[2];
double combined_entropy[2];
double diff[2];
size_t j;
for (j = 0; j < 2; ++j) {
size_t last_histogram_ix = self->last_histogram_ix_[j];
- combined_histo[j] = histograms[self->curr_histogram_ix_];
- FN(HistogramAddHistogram)(&combined_histo[j],
+ self->combined_histo[j] = histograms[self->curr_histogram_ix_];
+ FN(HistogramAddHistogram)(&self->combined_histo[j],
&histograms[last_histogram_ix]);
combined_entropy[j] = BitsEntropy(
- &combined_histo[j].data_[0], self->alphabet_size_);
+ &self->combined_histo[j].data_[0], self->alphabet_size_);
diff[j] = combined_entropy[j] - entropy - last_entropy[j];
}
@@ -141,7 +143,7 @@ static void FN(BlockSplitterFinishBlock)(
split->lengths[self->num_blocks_] = (uint32_t)self->block_size_;
split->types[self->num_blocks_] = split->types[self->num_blocks_ - 2];
BROTLI_SWAP(size_t, self->last_histogram_ix_, 0, 1);
- histograms[self->last_histogram_ix_[0]] = combined_histo[1];
+ histograms[self->last_histogram_ix_[0]] = self->combined_histo[1];
last_entropy[1] = last_entropy[0];
last_entropy[0] = combined_entropy[1];
++self->num_blocks_;
@@ -152,7 +154,7 @@ static void FN(BlockSplitterFinishBlock)(
} else {
/* Combine this block with last block. */
split->lengths[self->num_blocks_ - 1] += (uint32_t)self->block_size_;
- histograms[self->last_histogram_ix_[0]] = combined_histo[0];
+ histograms[self->last_histogram_ix_[0]] = self->combined_histo[0];
last_entropy[0] = combined_entropy[0];
if (split->num_types == 1) {
last_entropy[1] = last_entropy[0];
diff --git a/dviware/dvisvgm/libs/brotli/enc/params.h b/dviware/dvisvgm/libs/brotli/enc/params.h
index 54a7f00736..78b2ab6082 100644
--- a/dviware/dvisvgm/libs/brotli/enc/params.h
+++ b/dviware/dvisvgm/libs/brotli/enc/params.h
@@ -10,13 +10,13 @@
#define BROTLI_ENC_PARAMS_H_
#include <brotli/encode.h>
-#include "./encoder_dict.h"
+
+#include "encoder_dict.h"
typedef struct BrotliHasherParams {
int type;
int bucket_bits;
int block_bits;
- int hash_len;
int num_last_distances_to_check;
} BrotliHasherParams;
@@ -40,7 +40,8 @@ typedef struct BrotliEncoderParams {
BROTLI_BOOL large_window;
BrotliHasherParams hasher;
BrotliDistanceParams dist;
- BrotliEncoderDictionary dictionary;
+ /* TODO(eustas): rename to BrotliShared... */
+ SharedEncoderDictionary dictionary;
} BrotliEncoderParams;
#endif /* BROTLI_ENC_PARAMS_H_ */
diff --git a/dviware/dvisvgm/libs/brotli/enc/prefix.h b/dviware/dvisvgm/libs/brotli/enc/prefix.h
index fd359a478d..0f006f1614 100644
--- a/dviware/dvisvgm/libs/brotli/enc/prefix.h
+++ b/dviware/dvisvgm/libs/brotli/enc/prefix.h
@@ -10,10 +10,11 @@
#ifndef BROTLI_ENC_PREFIX_H_
#define BROTLI_ENC_PREFIX_H_
+#include <brotli/types.h>
+
#include "../common/constants.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./fast_log.h"
+#include "fast_log.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/quality.h b/dviware/dvisvgm/libs/brotli/enc/quality.h
index 5f4d034503..ffdfd72fb4 100644
--- a/dviware/dvisvgm/libs/brotli/enc/quality.h
+++ b/dviware/dvisvgm/libs/brotli/enc/quality.h
@@ -10,9 +10,10 @@
#ifndef BROTLI_ENC_QUALITY_H_
#define BROTLI_ENC_QUALITY_H_
-#include "../common/platform.h"
#include <brotli/encode.h>
-#include "./params.h"
+
+#include "../common/platform.h"
+#include "params.h"
#define FAST_ONE_PASS_COMPRESSION_QUALITY 0
#define FAST_TWO_PASS_COMPRESSION_QUALITY 1
@@ -118,6 +119,41 @@ static BROTLI_INLINE size_t LiteralSpreeLengthForSparseSearch(
return params->quality < 9 ? 64 : 512;
}
+/* Quality to hasher mapping:
+
+ - q02: h02 (longest_match_quickly), b16, l5
+
+ - q03: h03 (longest_match_quickly), b17, l5
+
+ - q04: h04 (longest_match_quickly), b17, l5
+ - q04: h54 (longest_match_quickly), b20, l7 | for large files
+
+ - q05: h05 (longest_match ), b14, l4
+ - q05: h06 (longest_match64 ), b15, l5 | for large files
+ - q05: h40 (forgetful_chain ), b15, l4 | for small window
+
+ - q06: h05 (longest_match ), b14, l4
+ - q06: h06 (longest_match64 ), b15, l5 | for large files
+ - q06: h40 (forgetful_chain ), b15, l4 | for small window
+
+ - q07: h05 (longest_match ), b15, l4
+ - q07: h06 (longest_match64 ), b15, l5 | for large files
+ - q07: h41 (forgetful_chain ), b15, l4 | for small window
+
+ - q08: h05 (longest_match ), b15, l4
+ - q08: h06 (longest_match64 ), b15, l5 | for large files
+ - q08: h41 (forgetful_chain ), b15, l4 | for small window
+
+ - q09: h05 (longest_match ), b15, l4
+ - q09: h06 (longest_match64 ), b15, l5 | for large files
+ - q09: h42 (forgetful_chain ), b15, l4 | for small window
+
+ - q10: t10 (to_binary_tree ), b17, l128
+
+ - q11: t10 (to_binary_tree ), b17, l128
+
+ Where "q" is quality, "h" is hasher type, "b" is bucket bits,
+ "l" is source len. */
static BROTLI_INLINE void ChooseHasher(const BrotliEncoderParams* params,
BrotliHasherParams* hparams) {
if (params->quality > 9) {
@@ -132,10 +168,11 @@ static BROTLI_INLINE void ChooseHasher(const BrotliEncoderParams* params,
hparams->type = 6;
hparams->block_bits = params->quality - 1;
hparams->bucket_bits = 15;
- hparams->hash_len = 5;
hparams->num_last_distances_to_check =
params->quality < 7 ? 4 : params->quality < 9 ? 10 : 16;
} else {
+ /* TODO(eustas): often previous setting (H6) is faster and denser; consider
+ adding an option to use it. */
hparams->type = 5;
hparams->block_bits = params->quality - 1;
hparams->bucket_bits = params->quality < 7 ? 14 : 15;
diff --git a/dviware/dvisvgm/libs/brotli/enc/ringbuffer.h b/dviware/dvisvgm/libs/brotli/enc/ringbuffer.h
index 8dce148039..27245b7f39 100644
--- a/dviware/dvisvgm/libs/brotli/enc/ringbuffer.h
+++ b/dviware/dvisvgm/libs/brotli/enc/ringbuffer.h
@@ -11,10 +11,11 @@
#include <string.h> /* memcpy */
-#include "../common/platform.h"
#include <brotli/types.h>
-#include "./memory.h"
-#include "./quality.h"
+
+#include "../common/platform.h"
+#include "memory.h"
+#include "quality.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/state.h b/dviware/dvisvgm/libs/brotli/enc/state.h
new file mode 100644
index 0000000000..cb82987701
--- /dev/null
+++ b/dviware/dvisvgm/libs/brotli/enc/state.h
@@ -0,0 +1,104 @@
+/* Copyright 2022 Google Inc. All Rights Reserved.
+
+ Distributed under MIT license.
+ See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+/* Encoder state. */
+
+#ifndef BROTLI_ENC_STATE_H_
+#define BROTLI_ENC_STATE_H_
+
+#include <brotli/types.h>
+
+#include "command.h"
+#include "compress_fragment.h"
+#include "compress_fragment_two_pass.h"
+#include "hash.h"
+#include "memory.h"
+#include "params.h"
+#include "ringbuffer.h"
+
+typedef enum BrotliEncoderStreamState {
+ /* Default state. */
+ BROTLI_STREAM_PROCESSING = 0,
+ /* Intermediate state; after next block is emitted, byte-padding should be
+ performed before getting back to default state. */
+ BROTLI_STREAM_FLUSH_REQUESTED = 1,
+ /* Last metablock was produced; no more input is acceptable. */
+ BROTLI_STREAM_FINISHED = 2,
+ /* Flushing compressed block and writing meta-data block header. */
+ BROTLI_STREAM_METADATA_HEAD = 3,
+ /* Writing metadata block body. */
+ BROTLI_STREAM_METADATA_BODY = 4
+} BrotliEncoderStreamState;
+
+typedef enum BrotliEncoderFlintState {
+ BROTLI_FLINT_NEEDS_2_BYTES = 2,
+ BROTLI_FLINT_NEEDS_1_BYTE = 1,
+ BROTLI_FLINT_WAITING_FOR_PROCESSING = 0,
+ BROTLI_FLINT_WAITING_FOR_FLUSHING = -1,
+ BROTLI_FLINT_DONE = -2
+} BrotliEncoderFlintState;
+
+typedef struct BrotliEncoderStateStruct {
+ BrotliEncoderParams params;
+
+ MemoryManager memory_manager_;
+
+ uint64_t input_pos_;
+ RingBuffer ringbuffer_;
+ size_t cmd_alloc_size_;
+ Command* commands_;
+ size_t num_commands_;
+ size_t num_literals_;
+ size_t last_insert_len_;
+ uint64_t last_flush_pos_;
+ uint64_t last_processed_pos_;
+ int dist_cache_[BROTLI_NUM_DISTANCE_SHORT_CODES];
+ int saved_dist_cache_[4];
+ uint16_t last_bytes_;
+ uint8_t last_bytes_bits_;
+ /* "Flint" is a tiny uncompressed block emitted before the continuation
+ block to unwire literal context from previous data. Despite being int8_t,
+ field is actually BrotliEncoderFlintState enum. */
+ int8_t flint_;
+ uint8_t prev_byte_;
+ uint8_t prev_byte2_;
+ size_t storage_size_;
+ uint8_t* storage_;
+
+ Hasher hasher_;
+
+ /* Hash table for FAST_ONE_PASS_COMPRESSION_QUALITY mode. */
+ int small_table_[1 << 10]; /* 4KiB */
+ int* large_table_; /* Allocated only when needed */
+ size_t large_table_size_;
+
+ BrotliOnePassArena* one_pass_arena_;
+ BrotliTwoPassArena* two_pass_arena_;
+
+ /* Command and literal buffers for FAST_TWO_PASS_COMPRESSION_QUALITY. */
+ uint32_t* command_buf_;
+ uint8_t* literal_buf_;
+
+ uint64_t total_in_;
+ uint8_t* next_out_;
+ size_t available_out_;
+ uint64_t total_out_;
+ /* Temporary buffer for padding flush bits or metadata block header / body. */
+ union {
+ uint64_t u64[2];
+ uint8_t u8[16];
+ } tiny_buf_;
+ uint32_t remaining_metadata_bytes_;
+ BrotliEncoderStreamState stream_state_;
+
+ BROTLI_BOOL is_last_block_emitted_;
+ BROTLI_BOOL is_initialized_;
+} BrotliEncoderStateStruct;
+
+typedef struct BrotliEncoderStateStruct BrotliEncoderStateInternal;
+#define BrotliEncoderState BrotliEncoderStateInternal
+
+#endif // BROTLI_ENC_STATE_H_
diff --git a/dviware/dvisvgm/libs/brotli/enc/static_dict.c b/dviware/dvisvgm/libs/brotli/enc/static_dict.c
index 7299ab7203..291d283354 100644
--- a/dviware/dvisvgm/libs/brotli/enc/static_dict.c
+++ b/dviware/dvisvgm/libs/brotli/enc/static_dict.c
@@ -4,13 +4,13 @@
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-#include "./static_dict.h"
+#include "static_dict.h"
#include "../common/dictionary.h"
#include "../common/platform.h"
#include "../common/transform.h"
-#include "./encoder_dict.h"
-#include "./find_match_length.h"
+#include "encoder_dict.h"
+#include "find_match_length.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
@@ -74,10 +74,27 @@ static BROTLI_INLINE BROTLI_BOOL IsMatch(const BrotliDictionary* dictionary,
}
}
-BROTLI_BOOL BrotliFindAllStaticDictionaryMatches(
+/* Finds matches for a single static dictionary */
+static BROTLI_BOOL BrotliFindAllStaticDictionaryMatchesFor(
const BrotliEncoderDictionary* dictionary, const uint8_t* data,
size_t min_length, size_t max_length, uint32_t* matches) {
BROTLI_BOOL has_found_match = BROTLI_FALSE;
+#if defined(BROTLI_EXPERIMENTAL)
+ if (dictionary->has_words_heavy) {
+ const BrotliTrieNode* node = &dictionary->trie.root;
+ size_t l = 0;
+ while (node && l < max_length) {
+ uint8_t c;
+ if (l >= min_length && node->len_) {
+ AddMatch(node->idx_, l, node->len_, matches);
+ has_found_match = BROTLI_TRUE;
+ }
+ c = data[l++];
+ node = BrotliTrieSub(&dictionary->trie, node, c);
+ }
+ return has_found_match;
+ }
+#endif /* BROTLI_EXPERIMENTAL */
{
size_t offset = dictionary->buckets[Hash(data)];
BROTLI_BOOL end = !offset;
@@ -481,6 +498,45 @@ BROTLI_BOOL BrotliFindAllStaticDictionaryMatches(
return has_found_match;
}
+/* Finds matches for one or more dictionaries, if multiple are present
+ in the contextual dictionary */
+BROTLI_BOOL BrotliFindAllStaticDictionaryMatches(
+ const BrotliEncoderDictionary* dictionary, const uint8_t* data,
+ size_t min_length, size_t max_length, uint32_t* matches) {
+ BROTLI_BOOL has_found_match =
+ BrotliFindAllStaticDictionaryMatchesFor(
+ dictionary, data, min_length, max_length, matches);
+
+ if (!!dictionary->parent && dictionary->parent->num_dictionaries > 1) {
+ uint32_t matches2[BROTLI_MAX_STATIC_DICTIONARY_MATCH_LEN + 1];
+ int l;
+ const BrotliEncoderDictionary* dictionary2 = dictionary->parent->dict[0];
+ if (dictionary2 == dictionary) {
+ dictionary2 = dictionary->parent->dict[1];
+ }
+
+ for (l = 0; l < BROTLI_MAX_STATIC_DICTIONARY_MATCH_LEN + 1; l++) {
+ matches2[l] = kInvalidMatch;
+ }
+
+ has_found_match |= BrotliFindAllStaticDictionaryMatchesFor(
+ dictionary2, data, min_length, max_length, matches2);
+
+ for (l = 0; l < BROTLI_MAX_STATIC_DICTIONARY_MATCH_LEN + 1; l++) {
+ if (matches2[l] != kInvalidMatch) {
+ uint32_t dist = (uint32_t)(matches2[l] >> 5);
+ uint32_t len_code = matches2[l] & 31;
+ uint32_t skipdist = (uint32_t)((uint32_t)(1 << dictionary->words->
+ size_bits_by_length[len_code]) & ~1u) *
+ (uint32_t)dictionary->num_transforms;
+ /* TODO(lode): check for dist overflow */
+ dist += skipdist;
+ AddMatch(dist, (size_t)l, len_code, matches);
+ }
+ }
+ }
+ return has_found_match;
+}
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/static_dict.h b/dviware/dvisvgm/libs/brotli/enc/static_dict.h
index 6b5d4eb0c9..ab832207d1 100644
--- a/dviware/dvisvgm/libs/brotli/enc/static_dict.h
+++ b/dviware/dvisvgm/libs/brotli/enc/static_dict.h
@@ -9,10 +9,11 @@
#ifndef BROTLI_ENC_STATIC_DICT_H_
#define BROTLI_ENC_STATIC_DICT_H_
+#include <brotli/types.h>
+
#include "../common/dictionary.h"
#include "../common/platform.h"
-#include <brotli/types.h>
-#include "./encoder_dict.h"
+#include "encoder_dict.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
diff --git a/dviware/dvisvgm/libs/brotli/enc/static_dict_lut.h b/dviware/dvisvgm/libs/brotli/enc/static_dict_lut.h
index e299cda6d8..a465ffde74 100644
--- a/dviware/dvisvgm/libs/brotli/enc/static_dict_lut.h
+++ b/dviware/dvisvgm/libs/brotli/enc/static_dict_lut.h
@@ -22,6 +22,7 @@ typedef struct DictWord {
uint16_t idx;
} DictWord;
+/* GENERATED CODE START */
static const int kDictNumBits = 15;
static const uint32_t kDictHashMul32 = 0x1E35A7BD;
@@ -5856,6 +5857,7 @@ static const DictWord kStaticDictionaryWords[31705] = {
,0,1735},{5,0,598},{7,0,791},{8,0,108},{9,0,123},{7,10,1570},{140,10,542},{142,
11,410},{9,11,660},{138,11,347}
};
+/* GENERATED CODE END */
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
diff --git a/dviware/dvisvgm/libs/brotli/enc/utf8_util.c b/dviware/dvisvgm/libs/brotli/enc/utf8_util.c
index e802b6a751..65ec3f5c8d 100644
--- a/dviware/dvisvgm/libs/brotli/enc/utf8_util.c
+++ b/dviware/dvisvgm/libs/brotli/enc/utf8_util.c
@@ -6,7 +6,7 @@
/* Heuristics for deciding about the UTF8-ness of strings. */
-#include "./utf8_util.h"
+#include "utf8_util.h"
#include <brotli/types.h>
diff --git a/dviware/dvisvgm/libs/brotli/enc/utf8_util.h b/dviware/dvisvgm/libs/brotli/enc/utf8_util.h
index 8fda80c220..a38a95383c 100644
--- a/dviware/dvisvgm/libs/brotli/enc/utf8_util.h
+++ b/dviware/dvisvgm/libs/brotli/enc/utf8_util.h
@@ -9,9 +9,10 @@
#ifndef BROTLI_ENC_UTF8_UTIL_H_
#define BROTLI_ENC_UTF8_UTIL_H_
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
diff --git a/dviware/dvisvgm/libs/brotli/enc/write_bits.h b/dviware/dvisvgm/libs/brotli/enc/write_bits.h
index f6f88b45be..242754b0ee 100644
--- a/dviware/dvisvgm/libs/brotli/enc/write_bits.h
+++ b/dviware/dvisvgm/libs/brotli/enc/write_bits.h
@@ -9,9 +9,10 @@
#ifndef BROTLI_ENC_WRITE_BITS_H_
#define BROTLI_ENC_WRITE_BITS_H_
-#include "../common/platform.h"
#include <brotli/types.h>
+#include "../common/platform.h"
+
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
diff --git a/dviware/dvisvgm/libs/brotli/include/brotli/decode.h b/dviware/dvisvgm/libs/brotli/include/brotli/decode.h
index 0f5c8f9d11..af1aa23f60 100644
--- a/dviware/dvisvgm/libs/brotli/include/brotli/decode.h
+++ b/dviware/dvisvgm/libs/brotli/include/brotli/decode.h
@@ -13,6 +13,7 @@
#define BROTLI_DEC_DECODE_H_
#include <brotli/port.h>
+#include <brotli/shared_dictionary.h>
#include <brotli/types.h>
#if defined(__cplusplus) || defined(c_plusplus)
@@ -85,8 +86,9 @@ typedef enum {
BROTLI_ERROR_CODE(_ERROR_FORMAT_, PADDING_2, -15) SEPARATOR \
BROTLI_ERROR_CODE(_ERROR_FORMAT_, DISTANCE, -16) SEPARATOR \
\
- /* -17..-18 codes are reserved */ \
+ /* -17 code is reserved */ \
\
+ BROTLI_ERROR_CODE(_ERROR_, COMPOUND_DICTIONARY, -18) SEPARATOR \
BROTLI_ERROR_CODE(_ERROR_, DICTIONARY_NOT_SET, -19) SEPARATOR \
BROTLI_ERROR_CODE(_ERROR_, INVALID_ARGUMENTS, -20) SEPARATOR \
\
@@ -155,6 +157,28 @@ BROTLI_DEC_API BROTLI_BOOL BrotliDecoderSetParameter(
BrotliDecoderState* state, BrotliDecoderParameter param, uint32_t value);
/**
+ * Adds LZ77 prefix dictionary, adds or replaces built-in static dictionary and
+ * transforms.
+ *
+ * Attached dictionary ownership is not transferred.
+ * Data provided to this method should be kept accessible until
+ * decoding is finished and decoder instance is destroyed.
+ *
+ * @note Dictionaries can NOT be attached after actual decoding is started.
+ *
+ * @param state decoder instance
+ * @param type dictionary data format
+ * @param data_size length of memory region pointed by @p data
+ * @param data dictionary data in format corresponding to @p type
+ * @returns ::BROTLI_FALSE if dictionary is corrupted,
+ * or dictionary count limit is reached
+ * @returns ::BROTLI_TRUE if dictionary is accepted / attached
+ */
+BROTLI_DEC_API BROTLI_BOOL BrotliDecoderAttachDictionary(
+ BrotliDecoderState* state, BrotliSharedDictionaryType type,
+ size_t data_size, const uint8_t data[BROTLI_ARRAY_PARAM(data_size)]);
+
+/**
* Creates an instance of ::BrotliDecoderState and initializes it.
*
* The instance can be used once for decoding and should then be destroyed with
@@ -333,10 +357,51 @@ BROTLI_DEC_API const char* BrotliDecoderErrorString(BrotliDecoderErrorCode c);
/**
* Gets a decoder library version.
*
- * Look at BROTLI_VERSION for more information.
+ * Look at BROTLI_MAKE_HEX_VERSION for more information.
*/
BROTLI_DEC_API uint32_t BrotliDecoderVersion(void);
+/**
+ * Callback to fire on metadata block start.
+ *
+ * After this callback is fired, if @p size is not @c 0, it is followed by
+ * ::brotli_decoder_metadata_chunk_func as more metadata block contents become
+ * accessible.
+ *
+ * @param opaque callback handle
+ * @param size size of metadata block
+ */
+typedef void (*brotli_decoder_metadata_start_func)(void* opaque, size_t size);
+
+/**
+ * Callback to fire on metadata block chunk becomes available.
+ *
+ * This function can be invoked multiple times per metadata block; block should
+ * be considered finished when sum of @p size matches the announced metadata
+ * block size. Chunks contents pointed by @p data are transient and shouln not
+ * be accessed after leaving the callback.
+ *
+ * @param opaque callback handle
+ * @param data pointer to metadata contents
+ * @param size size of metadata block chunk, at least @c 1
+ */
+typedef void (*brotli_decoder_metadata_chunk_func)(void* opaque,
+ const uint8_t* data,
+ size_t size);
+
+/**
+ * Sets callback for receiving metadata blocks.
+ *
+ * @param state decoder instance
+ * @param start_func callback on metadata block start
+ * @param chunk_func callback on metadata block chunk
+ * @param opaque callback handle
+ */
+BROTLI_DEC_API void BrotliDecoderSetMetadataCallbacks(
+ BrotliDecoderState* state,
+ brotli_decoder_metadata_start_func start_func,
+ brotli_decoder_metadata_chunk_func chunk_func, void* opaque);
+
#if defined(__cplusplus) || defined(c_plusplus)
} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/libs/brotli/include/brotli/encode.h b/dviware/dvisvgm/libs/brotli/include/brotli/encode.h
index b2774cb631..dea9931ebb 100644
--- a/dviware/dvisvgm/libs/brotli/include/brotli/encode.h
+++ b/dviware/dvisvgm/libs/brotli/include/brotli/encode.h
@@ -13,6 +13,7 @@
#define BROTLI_ENC_ENCODE_H_
#include <brotli/port.h>
+#include <brotli/shared_dictionary.h>
#include <brotli/types.h>
#if defined(__cplusplus) || defined(c_plusplus)
@@ -269,6 +270,51 @@ BROTLI_ENC_API BrotliEncoderState* BrotliEncoderCreateInstance(
*/
BROTLI_ENC_API void BrotliEncoderDestroyInstance(BrotliEncoderState* state);
+/* Opaque type for pointer to different possible internal structures containing
+ dictionary prepared for the encoder */
+typedef struct BrotliEncoderPreparedDictionaryStruct
+ BrotliEncoderPreparedDictionary;
+
+/**
+ * Prepares a shared dictionary from the given file format for the encoder.
+ *
+ * @p alloc_func and @p free_func @b MUST be both zero or both non-zero. In the
+ * case they are both zero, default memory allocators are used. @p opaque is
+ * passed to @p alloc_func and @p free_func when they are called. @p free_func
+ * has to return without doing anything when asked to free a NULL pointer.
+ *
+ * @param type type of dictionary stored in data
+ * @param data_size size of @p data buffer
+ * @param data pointer to the dictionary data
+ * @param quality the maximum Brotli quality to prepare the dictionary for,
+ * use BROTLI_MAX_QUALITY by default
+ * @param alloc_func custom memory allocation function
+ * @param free_func custom memory free function
+ * @param opaque custom memory manager handle
+ */
+BROTLI_ENC_API BrotliEncoderPreparedDictionary*
+BrotliEncoderPrepareDictionary(BrotliSharedDictionaryType type,
+ size_t data_size, const uint8_t data[BROTLI_ARRAY_PARAM(data_size)],
+ int quality,
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+
+BROTLI_ENC_API void BrotliEncoderDestroyPreparedDictionary(
+ BrotliEncoderPreparedDictionary* dictionary);
+
+/**
+ * Attaches a prepared dictionary of any type to the encoder. Can be used
+ * multiple times to attach multiple dictionaries. The dictionary type was
+ * determined by BrotliEncoderPrepareDictionary. Multiple raw prefix
+ * dictionaries and/or max 1 serialized dictionary with custom words can be
+ * attached.
+ *
+ * @returns ::BROTLI_FALSE in case of error
+ * @returns ::BROTLI_TRUE otherwise
+ */
+BROTLI_ENC_API BROTLI_BOOL BrotliEncoderAttachPreparedDictionary(
+ BrotliEncoderState* state,
+ const BrotliEncoderPreparedDictionary* dictionary);
+
/**
* Calculates the output size bound for the given @p input_size.
*
@@ -407,7 +453,7 @@ BROTLI_ENC_API BROTLI_BOOL BrotliEncoderHasMoreOutput(
*
* This method is used to make language bindings easier and more efficient:
* -# push data to ::BrotliEncoderCompressStream,
- * until ::BrotliEncoderHasMoreOutput returns BROTL_TRUE
+ * until ::BrotliEncoderHasMoreOutput returns BROTLI_TRUE
* -# use ::BrotliEncoderTakeOutput to peek bytes and copy to language-specific
* entity
*
@@ -433,11 +479,18 @@ BROTLI_ENC_API BROTLI_BOOL BrotliEncoderHasMoreOutput(
BROTLI_ENC_API const uint8_t* BrotliEncoderTakeOutput(
BrotliEncoderState* state, size_t* size);
+/* Returns the estimated peak memory usage (in bytes) of the BrotliCompress()
+ function, not counting the memory needed for the input and output. */
+BROTLI_ENC_EXTRA_API size_t BrotliEncoderEstimatePeakMemoryUsage(
+ int quality, int lgwin, size_t input_size);
+/* Returns 0 if dictionary is not valid; otherwise returns allocation size. */
+BROTLI_ENC_EXTRA_API size_t BrotliEncoderGetPreparedDictionarySize(
+ const BrotliEncoderPreparedDictionary* dictionary);
/**
* Gets an encoder library version.
*
- * Look at BROTLI_VERSION for more information.
+ * Look at BROTLI_MAKE_HEX_VERSION for more information.
*/
BROTLI_ENC_API uint32_t BrotliEncoderVersion(void);
diff --git a/dviware/dvisvgm/libs/brotli/include/brotli/port.h b/dviware/dvisvgm/libs/brotli/include/brotli/port.h
index 825237a335..0d50019042 100644
--- a/dviware/dvisvgm/libs/brotli/include/brotli/port.h
+++ b/dviware/dvisvgm/libs/brotli/include/brotli/port.h
@@ -224,14 +224,6 @@
#define BROTLI_HAS_FEATURE(feature) (0)
#endif
-#if defined(ADDRESS_SANITIZER) || BROTLI_HAS_FEATURE(address_sanitizer) || \
- defined(THREAD_SANITIZER) || BROTLI_HAS_FEATURE(thread_sanitizer) || \
- defined(MEMORY_SANITIZER) || BROTLI_HAS_FEATURE(memory_sanitizer)
-#define BROTLI_SANITIZED 1
-#else
-#define BROTLI_SANITIZED 0
-#endif
-
#if defined(_WIN32) || defined(__CYGWIN__)
#define BROTLI_PUBLIC
#elif BROTLI_GNUC_VERSION_CHECK(3, 3, 0) || \
@@ -247,9 +239,28 @@
#define BROTLI_PUBLIC
#endif
-#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) && \
- !defined(__STDC_NO_VLA__) && !defined(__cplusplus) && \
- !defined(__PGI) && !defined(__PGIC__) && !defined(__TINYC__)
+/* BROTLI_INTERNAL could be defined to override visibility, e.g. for tests. */
+#if !defined(BROTLI_INTERNAL)
+#if defined(_WIN32) || defined(__CYGWIN__)
+#define BROTLI_INTERNAL
+#elif BROTLI_GNUC_VERSION_CHECK(3, 3, 0) || \
+ BROTLI_TI_VERSION_CHECK(8, 0, 0) || \
+ BROTLI_INTEL_VERSION_CHECK(16, 0, 0) || \
+ BROTLI_ARM_VERSION_CHECK(4, 1, 0) || \
+ BROTLI_IBM_VERSION_CHECK(13, 1, 0) || \
+ BROTLI_SUNPRO_VERSION_CHECK(5, 11, 0) || \
+ (BROTLI_TI_VERSION_CHECK(7, 3, 0) && \
+ defined(__TI_GNU_ATTRIBUTE_SUPPORT__) && defined(__TI_EABI__))
+#define BROTLI_INTERNAL __attribute__ ((visibility ("hidden")))
+#else
+#define BROTLI_INTERNAL
+#endif
+#endif
+
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) && \
+ !defined(__STDC_NO_VLA__) && !defined(__cplusplus) && \
+ !defined(__PGI) && !defined(__PGIC__) && !defined(__TINYC__) && \
+ !defined(__clang__)
#define BROTLI_ARRAY_PARAM(name) (name)
#else
#define BROTLI_ARRAY_PARAM(name)
@@ -285,4 +296,10 @@
#define BROTLI_ENC_API
#endif
+#if defined(BROTLI_BUILD_ENC_EXTRA_API)
+#define BROTLI_ENC_EXTRA_API BROTLI_ENC_API
+#else
+#define BROTLI_ENC_EXTRA_API BROTLI_INTERNAL
+#endif
+
#endif /* BROTLI_COMMON_PORT_H_ */
diff --git a/dviware/dvisvgm/libs/brotli/include/brotli/shared_dictionary.h b/dviware/dvisvgm/libs/brotli/include/brotli/shared_dictionary.h
new file mode 100644
index 0000000000..2970c2dcc9
--- /dev/null
+++ b/dviware/dvisvgm/libs/brotli/include/brotli/shared_dictionary.h
@@ -0,0 +1,100 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+ Distributed under MIT license.
+ See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+/* (Opaque) Shared Dictionary definition and utilities. */
+
+#ifndef BROTLI_COMMON_SHARED_DICTIONARY_H_
+#define BROTLI_COMMON_SHARED_DICTIONARY_H_
+
+#include <brotli/port.h>
+#include <brotli/types.h>
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+#define SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH 4
+#define SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH 31
+#define SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS 64
+#define SHARED_BROTLI_MAX_COMPOUND_DICTS 15
+
+/**
+ * Opaque structure that holds shared dictionary data.
+ *
+ * Allocated and initialized with ::BrotliSharedDictionaryCreateInstance.
+ * Cleaned up and deallocated with ::BrotliSharedDictionaryDestroyInstance.
+ */
+typedef struct BrotliSharedDictionaryStruct BrotliSharedDictionary;
+
+/**
+ * Input data type for ::BrotliSharedDictionaryAttach.
+ */
+typedef enum BrotliSharedDictionaryType {
+ /** Raw LZ77 prefix dictionary. */
+ BROTLI_SHARED_DICTIONARY_RAW = 0,
+ /** Serialized shared dictionary.
+ *
+ * DO NOT USE: methods accepting this value will fail.
+ */
+ BROTLI_SHARED_DICTIONARY_SERIALIZED = 1
+} BrotliSharedDictionaryType;
+
+/**
+ * Creates an instance of ::BrotliSharedDictionary.
+ *
+ * Fresh instance has default word dictionary and transforms
+ * and no LZ77 prefix dictionary.
+ *
+ * @p alloc_func and @p free_func @b MUST be both zero or both non-zero. In the
+ * case they are both zero, default memory allocators are used. @p opaque is
+ * passed to @p alloc_func and @p free_func when they are called. @p free_func
+ * has to return without doing anything when asked to free a NULL pointer.
+ *
+ * @param alloc_func custom memory allocation function
+ * @param free_func custom memory free function
+ * @param opaque custom memory manager handle
+ * @returns @c 0 if instance can not be allocated or initialized
+ * @returns pointer to initialized ::BrotliSharedDictionary otherwise
+ */
+BROTLI_COMMON_API BrotliSharedDictionary* BrotliSharedDictionaryCreateInstance(
+ brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+
+/**
+ * Deinitializes and frees ::BrotliSharedDictionary instance.
+ *
+ * @param dict shared dictionary instance to be cleaned up and deallocated
+ */
+BROTLI_COMMON_API void BrotliSharedDictionaryDestroyInstance(
+ BrotliSharedDictionary* dict);
+
+/**
+ * Attaches dictionary to a given instance of ::BrotliSharedDictionary.
+ *
+ * Dictionary to be attached is represented in a serialized format as a region
+ * of memory.
+ *
+ * Provided data it partially referenced by a resulting (compound) dictionary,
+ * and should be kept untouched, while at least one compound dictionary uses it.
+ * This way memory overhead is kept minimal by the cost of additional resource
+ * management.
+ *
+ * @param dict dictionary to extend
+ * @param type type of dictionary to attach
+ * @param data_size size of @p data
+ * @param data serialized dictionary of type @p type, with at least @p data_size
+ * addressable bytes
+ * @returns ::BROTLI_TRUE if provided dictionary is successfully attached
+ * @returns ::BROTLI_FALSE otherwise
+ */
+BROTLI_COMMON_API BROTLI_BOOL BrotliSharedDictionaryAttach(
+ BrotliSharedDictionary* dict, BrotliSharedDictionaryType type,
+ size_t data_size, const uint8_t data[BROTLI_ARRAY_PARAM(data_size)]);
+
+#if defined(__cplusplus) || defined(c_plusplus)
+} /* extern "C" */
+#endif
+
+#endif /* BROTLI_COMMON_SHARED_DICTIONARY_H_ */
diff --git a/dviware/dvisvgm/libs/xxHash/xxh3.h b/dviware/dvisvgm/libs/xxHash/xxh3.h
index f7dc1959b5..7e3ce68e0c 100644
--- a/dviware/dvisvgm/libs/xxHash/xxh3.h
+++ b/dviware/dvisvgm/libs/xxHash/xxh3.h
@@ -1,7 +1,7 @@
/*
* xxHash - Extremely Fast Hash algorithm
* Development source file for `xxh3`
- * Copyright (C) 2019-2020 Yann Collet
+ * Copyright (C) 2019-2021 Yann Collet
*
* BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
*
diff --git a/dviware/dvisvgm/libs/xxHash/xxhash.c b/dviware/dvisvgm/libs/xxHash/xxhash.c
index 0fae88c5d6..083b039d70 100644
--- a/dviware/dvisvgm/libs/xxHash/xxhash.c
+++ b/dviware/dvisvgm/libs/xxHash/xxhash.c
@@ -1,6 +1,6 @@
/*
* xxHash - Extremely Fast Hash algorithm
- * Copyright (C) 2012-2020 Yann Collet
+ * Copyright (C) 2012-2021 Yann Collet
*
* BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
*
diff --git a/dviware/dvisvgm/libs/xxHash/xxhash.h b/dviware/dvisvgm/libs/xxHash/xxhash.h
index 511c4d12bf..a18e8c762d 100644
--- a/dviware/dvisvgm/libs/xxHash/xxhash.h
+++ b/dviware/dvisvgm/libs/xxHash/xxhash.h
@@ -1,7 +1,7 @@
/*
* xxHash - Extremely Fast Hash algorithm
* Header File
- * Copyright (C) 2012-2020 Yann Collet
+ * Copyright (C) 2012-2021 Yann Collet
*
* BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
*
@@ -32,49 +32,142 @@
* - xxHash homepage: https://www.xxhash.com
* - xxHash source repository: https://github.com/Cyan4973/xxHash
*/
+
/*!
* @mainpage xxHash
*
+ * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM speed
+ * limits.
+ *
+ * It is proposed in four flavors, in three families:
+ * 1. @ref XXH32_family
+ * - Classic 32-bit hash function. Simple, compact, and runs on almost all
+ * 32-bit and 64-bit systems.
+ * 2. @ref XXH64_family
+ * - Classic 64-bit adaptation of XXH32. Just as simple, and runs well on most
+ * 64-bit systems (but _not_ 32-bit systems).
+ * 3. @ref XXH3_family
+ * - Modern 64-bit and 128-bit hash function family which features improved
+ * strength and performance across the board, especially on smaller data.
+ * It benefits greatly from SIMD and 64-bit without requiring it.
+ *
+ * Benchmarks
+ * ---
+ * The reference system uses an Intel i7-9700K CPU, and runs Ubuntu x64 20.04.
+ * The open source benchmark program is compiled with clang v10.0 using -O3 flag.
+ *
+ * | Hash Name | ISA ext | Width | Large Data Speed | Small Data Velocity |
+ * | -------------------- | ------- | ----: | ---------------: | ------------------: |
+ * | XXH3_64bits() | @b AVX2 | 64 | 59.4 GB/s | 133.1 |
+ * | MeowHash | AES-NI | 128 | 58.2 GB/s | 52.5 |
+ * | XXH3_128bits() | @b AVX2 | 128 | 57.9 GB/s | 118.1 |
+ * | CLHash | PCLMUL | 64 | 37.1 GB/s | 58.1 |
+ * | XXH3_64bits() | @b SSE2 | 64 | 31.5 GB/s | 133.1 |
+ * | XXH3_128bits() | @b SSE2 | 128 | 29.6 GB/s | 118.1 |
+ * | RAM sequential read | | N/A | 28.0 GB/s | N/A |
+ * | ahash | AES-NI | 64 | 22.5 GB/s | 107.2 |
+ * | City64 | | 64 | 22.0 GB/s | 76.6 |
+ * | T1ha2 | | 64 | 22.0 GB/s | 99.0 |
+ * | City128 | | 128 | 21.7 GB/s | 57.7 |
+ * | FarmHash | AES-NI | 64 | 21.3 GB/s | 71.9 |
+ * | XXH64() | | 64 | 19.4 GB/s | 71.0 |
+ * | SpookyHash | | 64 | 19.3 GB/s | 53.2 |
+ * | Mum | | 64 | 18.0 GB/s | 67.0 |
+ * | CRC32C | SSE4.2 | 32 | 13.0 GB/s | 57.9 |
+ * | XXH32() | | 32 | 9.7 GB/s | 71.9 |
+ * | City32 | | 32 | 9.1 GB/s | 66.0 |
+ * | Blake3* | @b AVX2 | 256 | 4.4 GB/s | 8.1 |
+ * | Murmur3 | | 32 | 3.9 GB/s | 56.1 |
+ * | SipHash* | | 64 | 3.0 GB/s | 43.2 |
+ * | Blake3* | @b SSE2 | 256 | 2.4 GB/s | 8.1 |
+ * | HighwayHash | | 64 | 1.4 GB/s | 6.0 |
+ * | FNV64 | | 64 | 1.2 GB/s | 62.7 |
+ * | Blake2* | | 256 | 1.1 GB/s | 5.1 |
+ * | SHA1* | | 160 | 0.8 GB/s | 5.6 |
+ * | MD5* | | 128 | 0.6 GB/s | 7.8 |
+ * @note
+ * - Hashes which require a specific ISA extension are noted. SSE2 is also noted,
+ * even though it is mandatory on x64.
+ * - Hashes with an asterisk are cryptographic. Note that MD5 is non-cryptographic
+ * by modern standards.
+ * - Small data velocity is a rough average of algorithm's efficiency for small
+ * data. For more accurate information, see the wiki.
+ * - More benchmarks and strength tests are found on the wiki:
+ * https://github.com/Cyan4973/xxHash/wiki
+ *
+ * Usage
+ * ------
+ * All xxHash variants use a similar API. Changing the algorithm is a trivial
+ * substitution.
+ *
+ * @pre
+ * For functions which take an input and length parameter, the following
+ * requirements are assumed:
+ * - The range from [`input`, `input + length`) is valid, readable memory.
+ * - The only exception is if the `length` is `0`, `input` may be `NULL`.
+ * - For C++, the objects must have the *TriviallyCopyable* property, as the
+ * functions access bytes directly as if it was an array of `unsigned char`.
+ *
+ * @anchor single_shot_example
+ * **Single Shot**
+ *
+ * These functions are stateless functions which hash a contiguous block of memory,
+ * immediately returning the result. They are the easiest and usually the fastest
+ * option.
+ *
+ * XXH32(), XXH64(), XXH3_64bits(), XXH3_128bits()
+ *
+ * @code{.c}
+ * #include <string.h>
+ * #include "xxhash.h"
+ *
+ * // Example for a function which hashes a null terminated string with XXH32().
+ * XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed)
+ * {
+ * // NULL pointers are only valid if the length is zero
+ * size_t length = (string == NULL) ? 0 : strlen(string);
+ * return XXH32(string, length, seed);
+ * }
+ * @endcode
+ *
+ * @anchor streaming_example
+ * **Streaming**
+ *
+ * These groups of functions allow incremental hashing of unknown size, even
+ * more than what would fit in a size_t.
+ *
+ * XXH32_reset(), XXH64_reset(), XXH3_64bits_reset(), XXH3_128bits_reset()
+ *
+ * @code{.c}
+ * #include <stdio.h>
+ * #include <assert.h>
+ * #include "xxhash.h"
+ * // Example for a function which hashes a FILE incrementally with XXH3_64bits().
+ * XXH64_hash_t hashFile(FILE* f)
+ * {
+ * // Allocate a state struct. Do not just use malloc() or new.
+ * XXH3_state_t* state = XXH3_createState();
+ * assert(state != NULL && "Out of memory!");
+ * // Reset the state to start a new hashing session.
+ * XXH3_64bits_reset(state);
+ * char buffer[4096];
+ * size_t count;
+ * // Read the file in chunks
+ * while ((count = fread(buffer, 1, sizeof(buffer), f)) != 0) {
+ * // Run update() as many times as necessary to process the data
+ * XXH3_64bits_update(state, buffer, count);
+ * }
+ * // Retrieve the finalized hash. This will not change the state.
+ * XXH64_hash_t result = XXH3_64bits_digest(state);
+ * // Free the state. Do not use free().
+ * XXH3_freeState(state);
+ * return result;
+ * }
+ * @endcode
+ *
* @file xxhash.h
* xxHash prototypes and implementation
*/
-/* TODO: update */
-/* Notice extracted from xxHash homepage:
-
-xxHash is an extremely fast hash algorithm, running at RAM speed limits.
-It also successfully passes all tests from the SMHasher suite.
-
-Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)
-
-Name Speed Q.Score Author
-xxHash 5.4 GB/s 10
-CrapWow 3.2 GB/s 2 Andrew
-MurmurHash 3a 2.7 GB/s 10 Austin Appleby
-SpookyHash 2.0 GB/s 10 Bob Jenkins
-SBox 1.4 GB/s 9 Bret Mulvey
-Lookup3 1.2 GB/s 9 Bob Jenkins
-SuperFastHash 1.2 GB/s 1 Paul Hsieh
-CityHash64 1.05 GB/s 10 Pike & Alakuijala
-FNV 0.55 GB/s 5 Fowler, Noll, Vo
-CRC32 0.43 GB/s 9
-MD5-32 0.33 GB/s 10 Ronald L. Rivest
-SHA1-32 0.28 GB/s 10
-
-Q.Score is a measure of quality of the hash function.
-It depends on successfully passing SMHasher test set.
-10 is a perfect score.
-
-Note: SMHasher's CRC32 implementation is not the fastest one.
-Other speed-oriented implementations can be faster,
-especially in combination with PCLMUL instruction:
-https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735
-
-A 64-bit version, named XXH64, is available since r35.
-It offers much better speed, but for 64-bit applications only.
-Name Speed on 64 bits Speed on 32 bits
-XXH64 13.8 GB/s 1.9 GB/s
-XXH32 6.8 GB/s 6.0 GB/s
-*/
#if defined (__cplusplus)
extern "C" {
@@ -84,21 +177,80 @@ extern "C" {
* INLINE mode
******************************/
/*!
- * XXH_INLINE_ALL (and XXH_PRIVATE_API)
+ * @defgroup public Public API
+ * Contains details on the public xxHash functions.
+ * @{
+ */
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Gives access to internal state declaration, required for static allocation.
+ *
+ * Incompatible with dynamic linking, due to risks of ABI changes.
+ *
+ * Usage:
+ * @code{.c}
+ * #define XXH_STATIC_LINKING_ONLY
+ * #include "xxhash.h"
+ * @endcode
+ */
+# define XXH_STATIC_LINKING_ONLY
+/* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */
+
+/*!
+ * @brief Gives access to internal definitions.
+ *
+ * Usage:
+ * @code{.c}
+ * #define XXH_STATIC_LINKING_ONLY
+ * #define XXH_IMPLEMENTATION
+ * #include "xxhash.h"
+ * @endcode
+ */
+# define XXH_IMPLEMENTATION
+/* Do not undef XXH_IMPLEMENTATION for Doxygen */
+
+/*!
+ * @brief Exposes the implementation and marks all functions as `inline`.
+ *
* Use these build macros to inline xxhash into the target unit.
* Inlining improves performance on small inputs, especially when the length is
* expressed as a compile-time constant:
*
- * https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
+ * https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
*
* It also keeps xxHash symbols private to the unit, so they are not exported.
*
* Usage:
+ * @code{.c}
* #define XXH_INLINE_ALL
* #include "xxhash.h"
- *
+ * @endcode
* Do not compile and link xxhash.o as a separate object, as it is not useful.
*/
+# define XXH_INLINE_ALL
+# undef XXH_INLINE_ALL
+/*!
+ * @brief Exposes the implementation without marking functions as inline.
+ */
+# define XXH_PRIVATE_API
+# undef XXH_PRIVATE_API
+/*!
+ * @brief Emulate a namespace by transparently prefixing all symbols.
+ *
+ * If you want to include _and expose_ xxHash functions from within your own
+ * library, but also want to avoid symbol collisions with other libraries which
+ * may also include xxHash, you can use @ref XXH_NAMESPACE to automatically prefix
+ * any public symbol from xxhash library with the value of @ref XXH_NAMESPACE
+ * (therefore, avoid empty or numeric values).
+ *
+ * Note that no change is required within the calling program as long as it
+ * includes `xxhash.h`: Regular symbol names will be automatically translated
+ * by this header.
+ */
+# define XXH_NAMESPACE /* YOUR NAME HERE */
+# undef XXH_NAMESPACE
+#endif
+
#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
&& !defined(XXH_INLINE_ALL_31684351384)
/* this section should be traversed only once */
@@ -213,21 +365,13 @@ extern "C" {
# undef XXHASH_H_STATIC_13879238742
#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
-
-
/* ****************************************************************
* Stable API
*****************************************************************/
#ifndef XXHASH_H_5627135585666179
#define XXHASH_H_5627135585666179 1
-
-/*!
- * @defgroup public Public API
- * Contains details on the public xxHash functions.
- * @{
- */
-/* specific declaration modes for Windows */
+/*! @brief Marks a global symbol. */
#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
# if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
# ifdef XXH_EXPORT
@@ -240,24 +384,6 @@ extern "C" {
# endif
#endif
-#ifdef XXH_DOXYGEN
-/*!
- * @brief Emulate a namespace by transparently prefixing all symbols.
- *
- * If you want to include _and expose_ xxHash functions from within your own
- * library, but also want to avoid symbol collisions with other libraries which
- * may also include xxHash, you can use XXH_NAMESPACE to automatically prefix
- * any public symbol from xxhash library with the value of XXH_NAMESPACE
- * (therefore, avoid empty or numeric values).
- *
- * Note that no change is required within the calling program as long as it
- * includes `xxhash.h`: Regular symbol names will be automatically translated
- * by this header.
- */
-# define XXH_NAMESPACE /* YOUR NAME HERE */
-# undef XXH_NAMESPACE
-#endif
-
#ifdef XXH_NAMESPACE
# define XXH_CAT(A,B) A##B
# define XXH_NAME2(A,B) XXH_CAT(A,B)
@@ -318,11 +444,39 @@ extern "C" {
/* *************************************
+* Compiler specifics
+***************************************/
+
+/* specific declaration modes for Windows */
+#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+# if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+# ifdef XXH_EXPORT
+# define XXH_PUBLIC_API __declspec(dllexport)
+# elif XXH_IMPORT
+# define XXH_PUBLIC_API __declspec(dllimport)
+# endif
+# else
+# define XXH_PUBLIC_API /* do nothing */
+# endif
+#endif
+
+#if defined (__GNUC__)
+# define XXH_CONSTF __attribute__((const))
+# define XXH_PUREF __attribute__((pure))
+# define XXH_MALLOCF __attribute__((malloc))
+#else
+# define XXH_CONSTF /* disable */
+# define XXH_PUREF
+# define XXH_MALLOCF
+#endif
+
+/* *************************************
* Version
***************************************/
#define XXH_VERSION_MAJOR 0
#define XXH_VERSION_MINOR 8
-#define XXH_VERSION_RELEASE 1
+#define XXH_VERSION_RELEASE 2
+/*! @brief Version number, encoded as two digits each */
#define XXH_VERSION_NUMBER (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
/*!
@@ -331,16 +485,22 @@ extern "C" {
* This is mostly useful when xxHash is compiled as a shared library,
* since the returned value comes from the library, as opposed to header file.
*
- * @return `XXH_VERSION_NUMBER` of the invoked library.
+ * @return @ref XXH_VERSION_NUMBER of the invoked library.
*/
-XXH_PUBLIC_API unsigned XXH_versionNumber (void);
+XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber (void);
/* ****************************
* Common basic types
******************************/
#include <stddef.h> /* size_t */
-typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
+/*!
+ * @brief Exit code for the streaming API.
+ */
+typedef enum {
+ XXH_OK = 0, /*!< OK */
+ XXH_ERROR /*!< Error */
+} XXH_errorcode;
/*-**********************************************************************
@@ -364,29 +524,27 @@ typedef uint32_t XXH32_hash_t;
# include <limits.h>
# if UINT_MAX == 0xFFFFFFFFUL
typedef unsigned int XXH32_hash_t;
+# elif ULONG_MAX == 0xFFFFFFFFUL
+ typedef unsigned long XXH32_hash_t;
# else
-# if ULONG_MAX == 0xFFFFFFFFUL
- typedef unsigned long XXH32_hash_t;
-# else
-# error "unsupported platform: need a 32-bit type"
-# endif
+# error "unsupported platform: need a 32-bit type"
# endif
#endif
/*!
* @}
*
- * @defgroup xxh32_family XXH32 family
+ * @defgroup XXH32_family XXH32 family
* @ingroup public
* Contains functions used in the classic 32-bit xxHash algorithm.
*
* @note
* XXH32 is useful for older platforms, with no or poor 64-bit performance.
- * Note that @ref xxh3_family provides competitive speed
- * for both 32-bit and 64-bit systems, and offers true 64/128 bit hash results.
+ * Note that the @ref XXH3_family provides competitive speed for both 32-bit
+ * and 64-bit systems, and offers true 64/128 bit hash results.
*
- * @see @ref xxh64_family, @ref xxh3_family : Other xxHash families
- * @see @ref xxh32_impl for implementation details
+ * @see @ref XXH64_family, @ref XXH3_family : Other xxHash families
+ * @see @ref XXH32_impl for implementation details
* @{
*/
@@ -395,6 +553,8 @@ typedef uint32_t XXH32_hash_t;
*
* Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
*
+ * See @ref single_shot_example "Single Shot Example" for an example.
+ *
* @param input The block of data to be hashed, at least @p length bytes in size.
* @param length The length of @p input, in bytes.
* @param seed The 32-bit seed to alter the hash's output predictably.
@@ -412,8 +572,9 @@ typedef uint32_t XXH32_hash_t;
* @see
* XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
*/
-XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
+#ifndef XXH_NO_STREAM
/*!
* Streaming functions generate the xxHash value from an incremental input.
* This method is slower than single-call functions, due to state management.
@@ -436,32 +597,7 @@ XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_
*
* When done, release the state using `XXH*_freeState()`.
*
- * Example code for incrementally hashing a file:
- * @code{.c}
- * #include <stdio.h>
- * #include <xxhash.h>
- * #define BUFFER_SIZE 256
- *
- * // Note: XXH64 and XXH3 use the same interface.
- * XXH32_hash_t
- * hashFile(FILE* stream)
- * {
- * XXH32_state_t* state;
- * unsigned char buf[BUFFER_SIZE];
- * size_t amt;
- * XXH32_hash_t hash;
- *
- * state = XXH32_createState(); // Create a state
- * assert(state != NULL); // Error check here
- * XXH32_reset(state, 0xbaad5eed); // Reset state with our seed
- * while ((amt = fread(buf, 1, sizeof(buf), stream)) != 0) {
- * XXH32_update(state, buf, amt); // Hash the file in chunks
- * }
- * hash = XXH32_digest(state); // Finalize the hash
- * XXH32_freeState(state); // Clean up
- * return hash;
- * }
- * @endcode
+ * @see streaming_example at the top of @ref xxhash.h for an example.
*/
/*!
@@ -478,7 +614,7 @@ typedef struct XXH32_state_s XXH32_state_t;
* Must be freed with XXH32_freeState().
* @return An allocated XXH32_state_t on success, `NULL` on failure.
*/
-XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
+XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void);
/*!
* @brief Frees an @ref XXH32_state_t.
*
@@ -546,7 +682,8 @@ XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void*
*
* @return The calculated xxHash32 value from that state.
*/
-XXH_PUBLIC_API XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr);
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
/******* Canonical representation *******/
@@ -597,43 +734,72 @@ XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t
*
* @return The converted hash.
*/
-XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+/*! @cond Doxygen ignores this part */
#ifdef __has_attribute
# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
#else
# define XXH_HAS_ATTRIBUTE(x) 0
#endif
+/*! @endcond */
+/*! @cond Doxygen ignores this part */
+/*
+ * C23 __STDC_VERSION__ number hasn't been specified yet. For now
+ * leave as `201711L` (C17 + 1).
+ * TODO: Update to correct value when its been specified.
+ */
+#define XXH_C23_VN 201711L
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
/* C-language Attributes are added in C23. */
-#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && defined(__has_c_attribute)
# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
#else
# define XXH_HAS_C_ATTRIBUTE(x) 0
#endif
+/*! @endcond */
+/*! @cond Doxygen ignores this part */
#if defined(__cplusplus) && defined(__has_cpp_attribute)
# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
#else
# define XXH_HAS_CPP_ATTRIBUTE(x) 0
#endif
+/*! @endcond */
+/*! @cond Doxygen ignores this part */
/*
-Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
-introduced in CPP17 and C23.
-CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
-C23 : https://en.cppreference.com/w/c/language/attributes/fallthrough
-*/
-#if XXH_HAS_C_ATTRIBUTE(x)
-# define XXH_FALLTHROUGH [[fallthrough]]
-#elif XXH_HAS_CPP_ATTRIBUTE(x)
+ * Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
+ * introduced in CPP17 and C23.
+ * CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+ * C23 : https://en.cppreference.com/w/c/language/attributes/fallthrough
+ */
+#if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough)
# define XXH_FALLTHROUGH [[fallthrough]]
#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
-# define XXH_FALLTHROUGH __attribute__ ((fallthrough))
+# define XXH_FALLTHROUGH __attribute__ ((__fallthrough__))
#else
-# define XXH_FALLTHROUGH
+# define XXH_FALLTHROUGH /* fallthrough */
#endif
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
+/*
+ * Define XXH_NOESCAPE for annotated pointers in public API.
+ * https://clang.llvm.org/docs/AttributeReference.html#noescape
+ * As of writing this, only supported by clang.
+ */
+#if XXH_HAS_ATTRIBUTE(noescape)
+# define XXH_NOESCAPE __attribute__((noescape))
+#else
+# define XXH_NOESCAPE
+#endif
+/*! @endcond */
+
/*!
* @}
@@ -671,7 +837,7 @@ typedef uint64_t XXH64_hash_t;
/*!
* @}
*
- * @defgroup xxh64_family XXH64 family
+ * @defgroup XXH64_family XXH64 family
* @ingroup public
* @{
* Contains functions used in the classic 64-bit xxHash algorithm.
@@ -682,7 +848,6 @@ typedef uint64_t XXH64_hash_t;
* It provides better speed for systems with vector processing capabilities.
*/
-
/*!
* @brief Calculates the 64-bit hash of @p input using xxHash64.
*
@@ -706,32 +871,131 @@ typedef uint64_t XXH64_hash_t;
* @see
* XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
*/
-XXH_PUBLIC_API XXH64_hash_t XXH64(const void* input, size_t length, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);
/******* Streaming *******/
+#ifndef XXH_NO_STREAM
/*!
* @brief The opaque state struct for the XXH64 streaming API.
*
* @see XXH64_state_s for details.
*/
typedef struct XXH64_state_s XXH64_state_t; /* incomplete type */
-XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
+
+/*!
+ * @brief Allocates an @ref XXH64_state_t.
+ *
+ * Must be freed with XXH64_freeState().
+ * @return An allocated XXH64_state_t on success, `NULL` on failure.
+ */
+XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void);
+
+/*!
+ * @brief Frees an @ref XXH64_state_t.
+ *
+ * Must be allocated with XXH64_createState().
+ * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref XXH64_createState().
+ * @return XXH_OK.
+ */
XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr);
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dst_state, const XXH64_state_t* src_state);
-XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH64_state_t* statePtr, XXH64_hash_t seed);
-XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
-XXH_PUBLIC_API XXH64_hash_t XXH64_digest (const XXH64_state_t* statePtr);
+/*!
+ * @brief Copies one @ref XXH64_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ * @p dst_state and @p src_state must not be `NULL` and must not overlap.
+ */
+XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const XXH64_state_t* src_state);
+/*!
+ * @brief Resets an @ref XXH64_state_t to begin a new hash.
+ *
+ * This function resets and seeds a state. Call it before @ref XXH64_update().
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 64-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH64_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ * @pre
+ * The memory between @p input and @p input + @p length must be valid,
+ * readable, contiguous memory. However, if @p length is `0`, @p input may be
+ * `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated hash value from an @ref XXH64_state_t.
+ *
+ * @note
+ * Calling XXH64_digest() will not affect @p statePtr, so you can update,
+ * digest, and update again.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return The calculated xxHash64 value from that state.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_digest (XXH_NOESCAPE const XXH64_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
/******* Canonical representation *******/
+
+/*!
+ * @brief Canonical (big endian) representation of @ref XXH64_hash_t.
+ */
typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
-XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
+
+/*!
+ * @brief Converts an @ref XXH64_hash_t to a big endian @ref XXH64_canonical_t.
+ *
+ * @param dst The @ref XXH64_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH64_hash_t to be converted.
+ *
+ * @pre
+ * @p dst must not be `NULL`.
+ */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash);
+
+/*!
+ * @brief Converts an @ref XXH64_canonical_t to a native @ref XXH64_hash_t.
+ *
+ * @param src The @ref XXH64_canonical_t to convert.
+ *
+ * @pre
+ * @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src);
+
+#ifndef XXH_NO_XXH3
/*!
* @}
* ************************************************************************
- * @defgroup xxh3_family XXH3 family
+ * @defgroup XXH3_family XXH3 family
* @ingroup public
* @{
*
@@ -751,16 +1015,26 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src
*
* XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
* but does not require it.
- * Any 32-bit and 64-bit targets that can run XXH32 smoothly
- * can run XXH3 at competitive speeds, even without vector support.
- * Further details are explained in the implementation.
- *
- * Optimized implementations are provided for AVX512, AVX2, SSE2, NEON, POWER8,
- * ZVector and scalar targets. This can be controlled via the XXH_VECTOR macro.
+ * Most 32-bit and 64-bit targets that can run XXH32 smoothly can run XXH3
+ * at competitive speeds, even without vector support. Further details are
+ * explained in the implementation.
+ *
+ * XXH3 has a fast scalar implementation, but it also includes accelerated SIMD
+ * implementations for many common platforms:
+ * - AVX512
+ * - AVX2
+ * - SSE2
+ * - ARM NEON
+ * - WebAssembly SIMD128
+ * - POWER8 VSX
+ * - s390x ZVector
+ * This can be controlled via the @ref XXH_VECTOR macro, but it automatically
+ * selects the best version according to predefined macros. For the x86 family, an
+ * automatic runtime dispatcher is included separately in @ref xxh_x86dispatch.c.
*
* XXH3 implementation is portable:
* it has a generic C90 formulation that can be compiled on any platform,
- * all implementations generage exactly the same hash value on all platforms.
+ * all implementations generate exactly the same hash value on all platforms.
* Starting from v0.8.0, it's also labelled "stable", meaning that
* any future version will also generate the same hash value.
*
@@ -772,24 +1046,42 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src
*
* The API supports one-shot hashing, streaming mode, and custom secrets.
*/
-
/*-**********************************************************************
* XXH3 64-bit variant
************************************************************************/
-/* XXH3_64bits():
- * default 64-bit variant, using default secret and default seed of 0.
- * It's the fastest variant. */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* data, size_t len);
+/*!
+ * @brief 64-bit unseeded variant of XXH3.
+ *
+ * This is equivalent to @ref XXH3_64bits_withSeed() with a seed of 0, however
+ * it may have slightly better performance due to constant propagation of the
+ * defaults.
+ *
+ * @see
+ * XXH32(), XXH64(), XXH3_128bits(): equivalent for the other xxHash algorithms
+ * @see
+ * XXH3_64bits_withSeed(), XXH3_64bits_withSecret(): other seeding variants
+ * @see
+ * XXH3_64bits_reset(), XXH3_64bits_update(), XXH3_64bits_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length);
-/*
- * XXH3_64bits_withSeed():
- * This variant generates a custom secret on the fly
- * based on default secret altered using the `seed` value.
+/*!
+ * @brief 64-bit seeded variant of XXH3
+ *
+ * This variant generates a custom secret on the fly based on default secret
+ * altered using the `seed` value.
+ *
* While this operation is decently fast, note that it's not completely free.
- * Note: seed==0 produces the same results as XXH3_64bits().
+ *
+ * @note
+ * seed == 0 produces the same results as @ref XXH3_64bits().
+ *
+ * @param input The data to hash
+ * @param length The length
+ * @param seed The 64-bit seed to alter the state.
*/
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);
/*!
* The bare minimum size for a custom secret.
@@ -800,8 +1092,9 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, X
*/
#define XXH3_SECRET_SIZE_MIN 136
-/*
- * XXH3_64bits_withSecret():
+/*!
+ * @brief 64-bit variant of XXH3 with a custom "secret".
+ *
* It's possible to provide any blob of bytes as a "secret" to generate the hash.
* This makes it more difficult for an external actor to prepare an intentional collision.
* The main condition is that secretSize *must* be large enough (>= XXH3_SECRET_SIZE_MIN).
@@ -817,10 +1110,11 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, X
* This is not necessarily the case when using the blob of bytes directly
* because, when hashing _small_ inputs, only a portion of the secret is employed.
*/
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);
/******* Streaming *******/
+#ifndef XXH_NO_STREAM
/*
* Streaming requires state maintenance.
* This operation costs memory and CPU.
@@ -834,23 +1128,53 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len,
* @see XXH3_state_s for details.
*/
typedef struct XXH3_state_s XXH3_state_t;
-XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void);
+XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t* XXH3_createState(void);
XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
-XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state);
-/*
- * XXH3_64bits_reset():
- * Initialize with default parameters.
- * digest will be equivalent to `XXH3_64bits()`.
+/*!
+ * @brief Copies one @ref XXH3_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ * @p dst_state and @p src_state must not be `NULL` and must not overlap.
*/
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t* statePtr);
-/*
- * XXH3_64bits_reset_withSeed():
- * Generate a custom secret from `seed`, and store it into `statePtr`.
- * digest will be equivalent to `XXH3_64bits_withSeed()`.
+XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state);
+
+/*!
+ * @brief Resets an @ref XXH3_state_t to begin a new hash.
+ *
+ * This function resets `statePtr` and generate a secret with default parameters. Call it before @ref XXH3_64bits_update().
+ * Digest will be equivalent to `XXH3_64bits()`.
+ *
+ * @param statePtr The state struct to reset.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ *
*/
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
-/*
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);
+
+/*!
+ * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
+ *
+ * This function resets `statePtr` and generate a secret from `seed`. Call it before @ref XXH3_64bits_update().
+ * Digest will be equivalent to `XXH3_64bits_withSeed()`.
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 64-bit seed to alter the state.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ *
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
+
+/*!
* XXH3_64bits_reset_withSecret():
* `secret` is referenced, it _must outlive_ the hash streaming session.
* Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
@@ -859,10 +1183,44 @@ XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr,
* When in doubt about the randomness of a candidate `secret`,
* consider employing `XXH3_generateSecret()` instead (see below).
*/
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* statePtr);
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH3_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ * @pre
+ * The memory between @p input and @p input + @p length must be valid,
+ * readable, contiguous memory. However, if @p length is `0`, @p input may be
+ * `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated XXH3 64-bit hash value from an @ref XXH3_state_t.
+ *
+ * @note
+ * Calling XXH3_64bits_digest() will not affect @p statePtr, so you can update,
+ * digest, and update again.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return The calculated XXH3 64-bit hash value from that state.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
/* note : canonical representation of XXH3 is the same as XXH64
* since they both produce XXH64_hash_t values */
@@ -883,11 +1241,31 @@ typedef struct {
XXH64_hash_t high64; /*!< `value >> 64` */
} XXH128_hash_t;
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* data, size_t len);
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+/*!
+ * @brief Unseeded 128-bit variant of XXH3
+ *
+ * The 128-bit variant of XXH3 has more strength, but it has a bit of overhead
+ * for shorter inputs.
+ *
+ * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of 0, however
+ * it may have slightly better performance due to constant propagation of the
+ * defaults.
+ *
+ * @see
+ * XXH32(), XXH64(), XXH3_64bits(): equivalent for the other xxHash algorithms
+ * @see
+ * XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding variants
+ * @see
+ * XXH3_128bits_reset(), XXH3_128bits_update(), XXH3_128bits_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* data, size_t len);
+/*! @brief Seeded 128-bit variant of XXH3. @see XXH3_64bits_withSeed(). */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
+/*! @brief Custom secret 128-bit variant of XXH3. @see XXH3_64bits_withSecret(). */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);
/******* Streaming *******/
+#ifndef XXH_NO_STREAM
/*
* Streaming requires state maintenance.
* This operation costs memory and CPU.
@@ -900,12 +1278,77 @@ XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t le
* All reset and streaming functions have same meaning as their 64-bit counterpart.
*/
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t* statePtr);
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+/*!
+ * @brief Resets an @ref XXH3_state_t to begin a new hash.
+ *
+ * This function resets `statePtr` and generate a secret with default parameters. Call it before @ref XXH3_128bits_update().
+ * Digest will be equivalent to `XXH3_128bits()`.
+ *
+ * @param statePtr The state struct to reset.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ *
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr);
+/*!
+ * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
+ *
+ * This function resets `statePtr` and generate a secret from `seed`. Call it before @ref XXH3_128bits_update().
+ * Digest will be equivalent to `XXH3_128bits_withSeed()`.
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 64-bit seed to alter the state.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ *
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
+/*! @brief Custom secret 128-bit variant of XXH3. @see XXH_64bits_reset_withSecret(). */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH3_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ * @pre
+ * The memory between @p input and @p input + @p length must be valid,
+ * readable, contiguous memory. However, if @p length is `0`, @p input may be
+ * `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated XXH3 128-bit hash value from an @ref XXH3_state_t.
+ *
+ * @note
+ * Calling XXH3_128bits_digest() will not affect @p statePtr, so you can update,
+ * digest, and update again.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ * @p statePtr must not be `NULL`.
+ *
+ * @return The calculated XXH3 128-bit hash value from that state.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
/* Following helper functions make it possible to compare XXH128_hast_t values.
* Since XXH128_hash_t is a structure, this capability is not offered by the language.
@@ -915,26 +1358,48 @@ XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr);
* XXH128_isEqual():
* Return: 1 if `h1` and `h2` are equal, 0 if they are not.
*/
-XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
+XXH_PUBLIC_API XXH_PUREF int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
/*!
- * XXH128_cmp():
- *
+ * @brief Compares two @ref XXH128_hash_t
* This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
*
- * return: >0 if *h128_1 > *h128_2
- * =0 if *h128_1 == *h128_2
- * <0 if *h128_1 < *h128_2
+ * @return: >0 if *h128_1 > *h128_2
+ * =0 if *h128_1 == *h128_2
+ * <0 if *h128_1 < *h128_2
*/
-XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2);
+XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2);
/******* Canonical representation *******/
typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
-XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash);
-XXH_PUBLIC_API XXH128_hash_t XXH128_hashFromCanonical(const XXH128_canonical_t* src);
+/*!
+ * @brief Converts an @ref XXH128_hash_t to a big endian @ref XXH128_canonical_t.
+ *
+ * @param dst The @ref XXH128_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH128_hash_t to be converted.
+ *
+ * @pre
+ * @p dst must not be `NULL`.
+ */
+XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash);
+
+/*!
+ * @brief Converts an @ref XXH128_canonical_t to a native @ref XXH128_hash_t.
+ *
+ * @param src The @ref XXH128_canonical_t to convert.
+ *
+ * @pre
+ * @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src);
+
+
+#endif /* !XXH_NO_XXH3 */
#endif /* XXH_NO_LONG_LONG */
/*!
@@ -978,7 +1443,7 @@ struct XXH32_state_s {
XXH32_hash_t v[4]; /*!< Accumulator lanes */
XXH32_hash_t mem32[4]; /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
XXH32_hash_t memsize; /*!< Amount of data in @ref mem32 */
- XXH32_hash_t reserved; /*!< Reserved field. Do not read or write to it, it may be removed. */
+ XXH32_hash_t reserved; /*!< Reserved field. Do not read nor write to it. */
}; /* typedef'd to XXH32_state_t */
@@ -1002,9 +1467,11 @@ struct XXH64_state_s {
XXH64_hash_t mem64[4]; /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
XXH32_hash_t memsize; /*!< Amount of data in @ref mem64 */
XXH32_hash_t reserved32; /*!< Reserved field, needed for padding anyways*/
- XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it, it may be removed. */
+ XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it. */
}; /* typedef'd to XXH64_state_t */
+#ifndef XXH_NO_XXH3
+
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
# include <stdalign.h>
# define XXH_ALIGN(n) alignas(n)
@@ -1038,6 +1505,7 @@ struct XXH64_state_s {
#define XXH3_INTERNALBUFFER_SIZE 256
/*!
+ * @internal
* @brief Default size of the secret buffer (and @ref XXH3_kSecret).
*
* This is the size used in @ref XXH3_kSecret and the seeded functions.
@@ -1070,7 +1538,7 @@ struct XXH64_state_s {
*/
struct XXH3_state_s {
XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
- /*!< The 8 accumulators. Similar to `vN` in @ref XXH32_state_s::v1 and @ref XXH64_state_s */
+ /*!< The 8 accumulators. See @ref XXH32_state_s::v and @ref XXH64_state_s::v */
XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
/*!< Used to store a custom secret generated from a seed. */
XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
@@ -1110,69 +1578,119 @@ struct XXH3_state_s {
* Note that this doesn't prepare the state for a streaming operation,
* it's still necessary to use XXH3_NNbits_reset*() afterwards.
*/
-#define XXH3_INITSTATE(XXH3_state_ptr) { (XXH3_state_ptr)->seed = 0; }
+#define XXH3_INITSTATE(XXH3_state_ptr) \
+ do { \
+ XXH3_state_t* tmp_xxh3_state_ptr = (XXH3_state_ptr); \
+ tmp_xxh3_state_ptr->seed = 0; \
+ tmp_xxh3_state_ptr->extSecret = NULL; \
+ } while(0)
-/* XXH128() :
+/*!
* simple alias to pre-selected XXH3_128bits variant
*/
-XXH_PUBLIC_API XXH128_hash_t XXH128(const void* data, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
/* === Experimental API === */
/* Symbols defined below must be considered tied to a specific library version. */
-/*
+/*!
* XXH3_generateSecret():
*
* Derive a high-entropy secret from any user-defined content, named customSeed.
* The generated secret can be used in combination with `*_withSecret()` functions.
- * The `_withSecret()` variants are useful to provide a higher level of protection than 64-bit seed,
- * as it becomes much more difficult for an external actor to guess how to impact the calculation logic.
+ * The `_withSecret()` variants are useful to provide a higher level of protection
+ * than 64-bit seed, as it becomes much more difficult for an external actor to
+ * guess how to impact the calculation logic.
*
* The function accepts as input a custom seed of any length and any content,
- * and derives from it a high-entropy secret of length @secretSize
- * into an already allocated buffer @secretBuffer.
- * @secretSize must be >= XXH3_SECRET_SIZE_MIN
+ * and derives from it a high-entropy secret of length @p secretSize into an
+ * already allocated buffer @p secretBuffer.
*
* The generated secret can then be used with any `*_withSecret()` variant.
- * Functions `XXH3_128bits_withSecret()`, `XXH3_64bits_withSecret()`,
- * `XXH3_128bits_reset_withSecret()` and `XXH3_64bits_reset_withSecret()`
+ * The functions @ref XXH3_128bits_withSecret(), @ref XXH3_64bits_withSecret(),
+ * @ref XXH3_128bits_reset_withSecret() and @ref XXH3_64bits_reset_withSecret()
* are part of this list. They all accept a `secret` parameter
- * which must be large enough for implementation reasons (>= XXH3_SECRET_SIZE_MIN)
+ * which must be large enough for implementation reasons (>= @ref XXH3_SECRET_SIZE_MIN)
* _and_ feature very high entropy (consist of random-looking bytes).
- * These conditions can be a high bar to meet, so
- * XXH3_generateSecret() can be employed to ensure proper quality.
+ * These conditions can be a high bar to meet, so @ref XXH3_generateSecret() can
+ * be employed to ensure proper quality.
+ *
+ * @p customSeed can be anything. It can have any size, even small ones,
+ * and its content can be anything, even "poor entropy" sources such as a bunch
+ * of zeroes. The resulting `secret` will nonetheless provide all required qualities.
*
- * customSeed can be anything. It can have any size, even small ones,
- * and its content can be anything, even "poor entropy" sources such as a bunch of zeroes.
- * The resulting `secret` will nonetheless provide all required qualities.
+ * @pre
+ * - @p secretSize must be >= @ref XXH3_SECRET_SIZE_MIN
+ * - When @p customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
*
- * When customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
+ * Example code:
+ * @code{.c}
+ * #include <stdio.h>
+ * #include <stdlib.h>
+ * #include <string.h>
+ * #define XXH_STATIC_LINKING_ONLY // expose unstable API
+ * #include "xxhash.h"
+ * // Hashes argv[2] using the entropy from argv[1].
+ * int main(int argc, char* argv[])
+ * {
+ * char secret[XXH3_SECRET_SIZE_MIN];
+ * if (argv != 3) { return 1; }
+ * XXH3_generateSecret(secret, sizeof(secret), argv[1], strlen(argv[1]));
+ * XXH64_hash_t h = XXH3_64bits_withSecret(
+ * argv[2], strlen(argv[2]),
+ * secret, sizeof(secret)
+ * );
+ * printf("%016llx\n", (unsigned long long) h);
+ * }
+ * @endcode
*/
-XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize);
-
+XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize);
-/*
- * XXH3_generateSecret_fromSeed():
- *
- * Generate the same secret as the _withSeed() variants.
- *
- * The resulting secret has a length of XXH3_SECRET_DEFAULT_SIZE (necessarily).
- * @secretBuffer must be already allocated, of size at least XXH3_SECRET_DEFAULT_SIZE bytes.
+/*!
+ * @brief Generate the same secret as the _withSeed() variants.
*
* The generated secret can be used in combination with
*`*_withSecret()` and `_withSecretandSeed()` variants.
- * This generator is notably useful in combination with `_withSecretandSeed()`,
- * as a way to emulate a faster `_withSeed()` variant.
+ *
+ * Example C++ `std::string` hash class:
+ * @code{.cpp}
+ * #include <string>
+ * #define XXH_STATIC_LINKING_ONLY // expose unstable API
+ * #include "xxhash.h"
+ * // Slow, seeds each time
+ * class HashSlow {
+ * XXH64_hash_t seed;
+ * public:
+ * HashSlow(XXH64_hash_t s) : seed{s} {}
+ * size_t operator()(const std::string& x) const {
+ * return size_t{XXH3_64bits_withSeed(x.c_str(), x.length(), seed)};
+ * }
+ * };
+ * // Fast, caches the seeded secret for future uses.
+ * class HashFast {
+ * unsigned char secret[XXH3_SECRET_SIZE_MIN];
+ * public:
+ * HashFast(XXH64_hash_t s) {
+ * XXH3_generateSecret_fromSeed(secret, seed);
+ * }
+ * size_t operator()(const std::string& x) const {
+ * return size_t{
+ * XXH3_64bits_withSecret(x.c_str(), x.length(), secret, sizeof(secret))
+ * };
+ * }
+ * };
+ * @endcode
+ * @param secretBuffer A writable buffer of @ref XXH3_SECRET_SIZE_MIN bytes
+ * @param seed The seed to seed the state.
*/
-XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed);
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed);
-/*
- * *_withSecretandSeed() :
+/*!
* These variants generate hash values using either
- * @seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
- * or @secret for "large" keys (>= XXH3_MIDSIZE_MAX).
+ * @p seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
+ * or @p secret for "large" keys (>= XXH3_MIDSIZE_MAX).
*
* This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
* `_withSeed()` has to generate the secret on the fly for "large" keys.
@@ -1181,7 +1699,7 @@ XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_
* which requires more instructions than _withSeed() variants.
* Therefore, _withSecretandSeed variant combines the best of both worlds.
*
- * When @secret has been generated by XXH3_generateSecret_fromSeed(),
+ * When @p secret has been generated by XXH3_generateSecret_fromSeed(),
* this variant produces *exactly* the same results as `_withSeed()` variant,
* hence offering only a pure speed benefit on "large" input,
* by skipping the need to regenerate the secret for every large input.
@@ -1190,32 +1708,34 @@ XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_
* for example with XXH3_64bits(), which then becomes the seed,
* and then employ both the seed and the secret in _withSecretandSeed().
* On top of speed, an added benefit is that each bit in the secret
- * has a 50% chance to swap each bit in the output,
- * via its impact to the seed.
+ * has a 50% chance to swap each bit in the output, via its impact to the seed.
+ *
* This is not guaranteed when using the secret directly in "small data" scenarios,
* because only portions of the secret are employed for small data.
*/
-XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSecretandSeed(const void* data, size_t len,
- const void* secret, size_t secretSize,
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len,
+ XXH_NOESCAPE const void* secret, size_t secretSize,
XXH64_hash_t seed);
-
-XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSecretandSeed(const void* data, size_t len,
- const void* secret, size_t secretSize,
+/*! @copydoc XXH3_64bits_withSecretandSeed() */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
+XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length,
+ XXH_NOESCAPE const void* secret, size_t secretSize,
XXH64_hash_t seed64);
-
+#ifndef XXH_NO_STREAM
+/*! @copydoc XXH3_64bits_withSecretandSeed() */
XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
- const void* secret, size_t secretSize,
+XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
+ XXH_NOESCAPE const void* secret, size_t secretSize,
XXH64_hash_t seed64);
-
+/*! @copydoc XXH3_64bits_withSecretandSeed() */
XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
- const void* secret, size_t secretSize,
+XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
+ XXH_NOESCAPE const void* secret, size_t secretSize,
XXH64_hash_t seed64);
+#endif /* !XXH_NO_STREAM */
-
+#endif /* !XXH_NO_XXH3 */
#endif /* XXH_NO_LONG_LONG */
#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
# define XXH_IMPLEMENTATION
@@ -1269,7 +1789,7 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
/*!
* @brief Define this to disable 64-bit code.
*
- * Useful if only using the @ref xxh32_family and you have a strict C90 compiler.
+ * Useful if only using the @ref XXH32_family and you have a strict C90 compiler.
*/
# define XXH_NO_LONG_LONG
# undef XXH_NO_LONG_LONG /* don't actually */
@@ -1292,7 +1812,7 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
* Use `memcpy()`. Safe and portable. Note that most modern compilers will
* eliminate the function call and treat it as an unaligned access.
*
- * - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((packed))`
+ * - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))`
* @par
* Depends on compiler extensions and is therefore not portable.
* This method is safe _if_ your compiler supports it,
@@ -1312,20 +1832,48 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
* inline small `memcpy()` calls, and it might also be faster on big-endian
* systems which lack a native byteswap instruction. However, some compilers
* will emit literal byteshifts even if the target supports unaligned access.
- * .
+ *
*
* @warning
* Methods 1 and 2 rely on implementation-defined behavior. Use these with
* care, as what works on one compiler/platform/optimization level may cause
* another to read garbage data or even crash.
*
- * See http://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
+ * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
*
* Prefer these methods in priority order (0 > 3 > 1 > 2)
*/
# define XXH_FORCE_MEMORY_ACCESS 0
/*!
+ * @def XXH_SIZE_OPT
+ * @brief Controls how much xxHash optimizes for size.
+ *
+ * xxHash, when compiled, tends to result in a rather large binary size. This
+ * is mostly due to heavy usage to forced inlining and constant folding of the
+ * @ref XXH3_family to increase performance.
+ *
+ * However, some developers prefer size over speed. This option can
+ * significantly reduce the size of the generated code. When using the `-Os`
+ * or `-Oz` options on GCC or Clang, this is defined to 1 by default,
+ * otherwise it is defined to 0.
+ *
+ * Most of these size optimizations can be controlled manually.
+ *
+ * This is a number from 0-2.
+ * - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations. Speed
+ * comes first.
+ * - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more
+ * conservative and disables hacks that increase code size. It implies the
+ * options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK == 0,
+ * and @ref XXH3_NEON_LANES == 8 if they are not already defined.
+ * - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible.
+ * Performance may cry. For example, the single shot functions just use the
+ * streaming API.
+ */
+# define XXH_SIZE_OPT 0
+
+/*!
* @def XXH_FORCE_ALIGN_CHECK
* @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
* and XXH64() only).
@@ -1346,9 +1894,11 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
*
* In these cases, the alignment check can be removed by setting this macro to 0.
* Then the code will always use unaligned memory access.
- * Align check is automatically disabled on x86, x64 & arm64,
+ * Align check is automatically disabled on x86, x64, ARM64, and some ARM chips
* which are platforms known to offer good unaligned memory accesses performance.
*
+ * It is also disabled by default when @ref XXH_SIZE_OPT >= 1.
+ *
* This option does not affect XXH3 (only XXH32 and XXH64).
*/
# define XXH_FORCE_ALIGN_CHECK 0
@@ -1370,12 +1920,29 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
* XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
* compiler full control on whether to inline or not.
*
- * When not optimizing (-O0), optimizing for size (-Os, -Oz), or using
- * -fno-inline with GCC or Clang, this will automatically be defined.
+ * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if
+ * @ref XXH_SIZE_OPT >= 1, this will automatically be defined.
*/
# define XXH_NO_INLINE_HINTS 0
/*!
+ * @def XXH3_INLINE_SECRET
+ * @brief Determines whether to inline the XXH3 withSecret code.
+ *
+ * When the secret size is known, the compiler can improve the performance
+ * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret().
+ *
+ * However, if the secret size is not known, it doesn't have any benefit. This
+ * happens when xxHash is compiled into a global symbol. Therefore, if
+ * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0.
+ *
+ * Additionally, this defaults to 0 on GCC 12+, which has an issue with function pointers
+ * that are *sometimes* force inline on -Og, and it is impossible to automatically
+ * detect this optimization level.
+ */
+# define XXH3_INLINE_SECRET 0
+
+/*!
* @def XXH32_ENDJMP
* @brief Whether to use a jump for `XXH32_finalize`.
*
@@ -1396,34 +1963,45 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
*/
# define XXH_OLD_NAMES
# undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
+
+/*!
+ * @def XXH_NO_STREAM
+ * @brief Disables the streaming API.
+ *
+ * When xxHash is not inlined and the streaming functions are not used, disabling
+ * the streaming functions can improve code size significantly, especially with
+ * the @ref XXH3_family which tends to make constant folded copies of itself.
+ */
+# define XXH_NO_STREAM
+# undef XXH_NO_STREAM /* don't actually */
#endif /* XXH_DOXYGEN */
/*!
* @}
*/
#ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
- /* prefer __packed__ structures (method 1) for gcc on armv7+ and mips */
-# if !defined(__clang__) && \
-( \
- (defined(__INTEL_COMPILER) && !defined(_WIN32)) || \
- ( \
- defined(__GNUC__) && ( \
- (defined(__ARM_ARCH) && __ARM_ARCH >= 7) || \
- ( \
- defined(__mips__) && \
- (__mips <= 5 || __mips_isa_rev < 6) && \
- (!defined(__mips16) || defined(__mips_mips16e2)) \
- ) \
- ) \
- ) \
-)
+ /* prefer __packed__ structures (method 1) for GCC
+ * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte shifting, so we use memcpy
+ * which for some reason does unaligned loads. */
+# if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && defined(__ARM_FEATURE_UNALIGNED))
# define XXH_FORCE_MEMORY_ACCESS 1
# endif
#endif
+#ifndef XXH_SIZE_OPT
+ /* default to 1 for -Os or -Oz */
+# if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__)
+# define XXH_SIZE_OPT 1
+# else
+# define XXH_SIZE_OPT 0
+# endif
+#endif
+
#ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */
-# if defined(__i386) || defined(__x86_64__) || defined(__aarch64__) \
- || defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) /* visual */
+ /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is available */
+# if XXH_SIZE_OPT >= 1 || \
+ defined(__i386) || defined(__x86_64__) || defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) \
+ || defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) || defined(_M_ARM) /* visual */
# define XXH_FORCE_ALIGN_CHECK 0
# else
# define XXH_FORCE_ALIGN_CHECK 1
@@ -1431,14 +2009,22 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
#endif
#ifndef XXH_NO_INLINE_HINTS
-# if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ \
- || defined(__NO_INLINE__) /* -O0, -fno-inline */
+# if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__) /* -O0, -fno-inline */
# define XXH_NO_INLINE_HINTS 1
# else
# define XXH_NO_INLINE_HINTS 0
# endif
#endif
+#ifndef XXH3_INLINE_SECRET
+# if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) \
+ || !defined(XXH_INLINE_ALL)
+# define XXH3_INLINE_SECRET 0
+# else
+# define XXH3_INLINE_SECRET 1
+# endif
+#endif
+
#ifndef XXH32_ENDJMP
/* generally preferable for performance */
# define XXH32_ENDJMP 0
@@ -1453,6 +2039,24 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
/* *************************************
* Includes & Memory related functions
***************************************/
+#if defined(XXH_NO_STREAM)
+/* nothing */
+#elif defined(XXH_NO_STDLIB)
+
+/* When requesting to disable any mention of stdlib,
+ * the library loses the ability to invoked malloc / free.
+ * In practice, it means that functions like `XXH*_createState()`
+ * will always fail, and return NULL.
+ * This flag is useful in situations where
+ * xxhash.h is integrated into some kernel, embedded or limited environment
+ * without access to dynamic allocation.
+ */
+
+static XXH_CONSTF void* XXH_malloc(size_t s) { (void)s; return NULL; }
+static void XXH_free(void* p) { (void)p; }
+
+#else
+
/*
* Modify the local functions below should you wish to use
* different memory routines for malloc() and free()
@@ -1463,7 +2067,7 @@ XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
* @internal
* @brief Modify this function to use a different routine than malloc().
*/
-static void* XXH_malloc(size_t s) { return malloc(s); }
+static XXH_MALLOCF void* XXH_malloc(size_t s) { return malloc(s); }
/*!
* @internal
@@ -1471,6 +2075,8 @@ static void* XXH_malloc(size_t s) { return malloc(s); }
*/
static void XXH_free(void* p) { free(p); }
+#endif /* XXH_NO_STDLIB */
+
#include <string.h>
/*!
@@ -1515,6 +2121,11 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size)
# define XXH_NO_INLINE static
#endif
+#if XXH3_INLINE_SECRET
+# define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE
+#else
+# define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE
+#endif
/* *************************************
@@ -1540,7 +2151,11 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size)
# include <assert.h> /* note: can still be disabled with NDEBUG */
# define XXH_ASSERT(c) assert(c)
#else
-# define XXH_ASSERT(c) ((void)0)
+# if defined(__INTEL_COMPILER)
+# define XXH_ASSERT(c) XXH_ASSUME((unsigned char) (c))
+# else
+# define XXH_ASSERT(c) XXH_ASSUME(c)
+# endif
#endif
/* note: use after variable declarations */
@@ -1572,11 +2187,19 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size)
* XXH3_initCustomSecret_scalar().
*/
#if defined(__GNUC__) || defined(__clang__)
-# define XXH_COMPILER_GUARD(var) __asm__ __volatile__("" : "+r" (var))
+# define XXH_COMPILER_GUARD(var) __asm__("" : "+r" (var))
#else
# define XXH_COMPILER_GUARD(var) ((void)0)
#endif
+/* Specifically for NEON vectors which use the "w" constraint, on
+ * Clang. */
+#if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__)
+# define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w" (var))
+#else
+# define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0)
+#endif
+
/* *************************************
* Basic Types
***************************************/
@@ -1591,6 +2214,7 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size)
typedef XXH32_hash_t xxh_u32;
#ifdef XXH_OLD_NAMES
+# warning "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly"
# define BYTE xxh_u8
# define U8 xxh_u8
# define U32 xxh_u32
@@ -1664,25 +2288,26 @@ static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr;
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
/*
- * __pack instructions are safer but compiler specific, hence potentially
- * problematic for some compilers.
- *
- * Currently only defined for GCC and ICC.
+ * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
+ * documentation claimed that it only increased the alignment, but actually it
+ * can decrease it on gcc, clang, and icc:
+ * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
+ * https://gcc.godbolt.org/z/xYez1j67Y.
*/
#ifdef XXH_OLD_NAMES
typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
#endif
static xxh_u32 XXH_read32(const void* ptr)
{
- typedef union { xxh_u32 u32; } __attribute__((packed)) xxh_unalign;
- return ((const xxh_unalign*)ptr)->u32;
+ typedef __attribute__((aligned(1))) xxh_u32 xxh_unalign32;
+ return *((const xxh_unalign32*)ptr);
}
#else
/*
* Portable and safe solution. Generally efficient.
- * see: http://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
*/
static xxh_u32 XXH_read32(const void* memPtr)
{
@@ -1758,6 +2383,51 @@ static int XXH_isLittleEndian(void)
# define XXH_HAS_BUILTIN(x) 0
#endif
+
+
+/*
+ * C23 and future versions have standard "unreachable()".
+ * Once it has been implemented reliably we can add it as an
+ * additional case:
+ *
+ * ```
+ * #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN)
+ * # include <stddef.h>
+ * # ifdef unreachable
+ * # define XXH_UNREACHABLE() unreachable()
+ * # endif
+ * #endif
+ * ```
+ *
+ * Note C++23 also has std::unreachable() which can be detected
+ * as follows:
+ * ```
+ * #if defined(__cpp_lib_unreachable) && (__cpp_lib_unreachable >= 202202L)
+ * # include <utility>
+ * # define XXH_UNREACHABLE() std::unreachable()
+ * #endif
+ * ```
+ * NB: `__cpp_lib_unreachable` is defined in the `<version>` header.
+ * We don't use that as including `<utility>` in `extern "C"` blocks
+ * doesn't work on GCC12
+ */
+
+#if XXH_HAS_BUILTIN(__builtin_unreachable)
+# define XXH_UNREACHABLE() __builtin_unreachable()
+
+#elif defined(_MSC_VER)
+# define XXH_UNREACHABLE() __assume(0)
+
+#else
+# define XXH_UNREACHABLE()
+#endif
+
+#if XXH_HAS_BUILTIN(__builtin_assume)
+# define XXH_ASSUME(c) __builtin_assume(c)
+#else
+# define XXH_ASSUME(c) if (!(c)) { XXH_UNREACHABLE(); }
+#endif
+
/*!
* @internal
* @def XXH_rotl32(x,r)
@@ -1880,8 +2550,10 @@ XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
*********************************************************************/
/*!
* @}
- * @defgroup xxh32_impl XXH32 implementation
+ * @defgroup XXH32_impl XXH32 implementation
* @ingroup impl
+ *
+ * Details on the XXH32 implementation.
* @{
*/
/* #define instead of static const, to be used as initializers */
@@ -1915,7 +2587,7 @@ static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
acc += input * XXH_PRIME32_2;
acc = XXH_rotl32(acc, 13);
acc *= XXH_PRIME32_1;
-#if (defined(__SSE4_1__) || defined(__aarch64__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
+#if (defined(__SSE4_1__) || defined(__aarch64__) || defined(__wasm_simd128__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
/*
* UGLY HACK:
* A compiler fence is the only thing that prevents GCC and Clang from
@@ -1945,9 +2617,12 @@ static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
* can load data, while v3 can multiply. SSE forces them to operate
* together.
*
- * This is also enabled on AArch64, as Clang autovectorizes it incorrectly
- * and it is pointless writing a NEON implementation that is basically the
- * same speed as scalar for XXH32.
+ * This is also enabled on AArch64, as Clang is *very aggressive* in vectorizing
+ * the loop. NEON is only faster on the A53, and with the newer cores, it is less
+ * than half the speed.
+ *
+ * Additionally, this is used on WASM SIMD128 because it JITs to the same
+ * SIMD instructions and has the same issue.
*/
XXH_COMPILER_GUARD(acc);
#endif
@@ -1961,17 +2636,17 @@ static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
* The final mix ensures that all input bits have a chance to impact any bit in
* the output digest, resulting in an unbiased distribution.
*
- * @param h32 The hash to avalanche.
+ * @param hash The hash to avalanche.
* @return The avalanched hash.
*/
-static xxh_u32 XXH32_avalanche(xxh_u32 h32)
+static xxh_u32 XXH32_avalanche(xxh_u32 hash)
{
- h32 ^= h32 >> 15;
- h32 *= XXH_PRIME32_2;
- h32 ^= h32 >> 13;
- h32 *= XXH_PRIME32_3;
- h32 ^= h32 >> 16;
- return(h32);
+ hash ^= hash >> 15;
+ hash *= XXH_PRIME32_2;
+ hash ^= hash >> 13;
+ hash *= XXH_PRIME32_3;
+ hash ^= hash >> 16;
+ return hash;
}
#define XXH_get32bits(p) XXH_readLE32_align(p, align)
@@ -1984,24 +2659,25 @@ static xxh_u32 XXH32_avalanche(xxh_u32 h32)
* This final stage will digest them to ensure that all input bytes are present
* in the final mix.
*
- * @param h32 The hash to finalize.
+ * @param hash The hash to finalize.
* @param ptr The pointer to the remaining input.
* @param len The remaining length, modulo 16.
* @param align Whether @p ptr is aligned.
* @return The finalized hash.
+ * @see XXH64_finalize().
*/
-static xxh_u32
-XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
+static XXH_PUREF xxh_u32
+XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
{
-#define XXH_PROCESS1 do { \
- h32 += (*ptr++) * XXH_PRIME32_5; \
- h32 = XXH_rotl32(h32, 11) * XXH_PRIME32_1; \
+#define XXH_PROCESS1 do { \
+ hash += (*ptr++) * XXH_PRIME32_5; \
+ hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1; \
} while (0)
-#define XXH_PROCESS4 do { \
- h32 += XXH_get32bits(ptr) * XXH_PRIME32_3; \
- ptr += 4; \
- h32 = XXH_rotl32(h32, 17) * XXH_PRIME32_4; \
+#define XXH_PROCESS4 do { \
+ hash += XXH_get32bits(ptr) * XXH_PRIME32_3; \
+ ptr += 4; \
+ hash = XXH_rotl32(hash, 17) * XXH_PRIME32_4; \
} while (0)
if (ptr==NULL) XXH_ASSERT(len == 0);
@@ -2017,49 +2693,49 @@ XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
XXH_PROCESS1;
--len;
}
- return XXH32_avalanche(h32);
+ return XXH32_avalanche(hash);
} else {
switch(len&15) /* or switch(bEnd - p) */ {
case 12: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 8: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 4: XXH_PROCESS4;
- return XXH32_avalanche(h32);
+ return XXH32_avalanche(hash);
case 13: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 9: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 5: XXH_PROCESS4;
XXH_PROCESS1;
- return XXH32_avalanche(h32);
+ return XXH32_avalanche(hash);
case 14: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 10: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 6: XXH_PROCESS4;
XXH_PROCESS1;
XXH_PROCESS1;
- return XXH32_avalanche(h32);
+ return XXH32_avalanche(hash);
case 15: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 11: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 7: XXH_PROCESS4;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 3: XXH_PROCESS1;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 2: XXH_PROCESS1;
- XXH_FALLTHROUGH;
+ XXH_FALLTHROUGH; /* fallthrough */
case 1: XXH_PROCESS1;
- XXH_FALLTHROUGH;
- case 0: return XXH32_avalanche(h32);
+ XXH_FALLTHROUGH; /* fallthrough */
+ case 0: return XXH32_avalanche(hash);
}
XXH_ASSERT(0);
- return h32; /* reaching this point is deemed impossible */
+ return hash; /* reaching this point is deemed impossible */
}
}
@@ -2079,7 +2755,7 @@ XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
* @param align Whether @p input is aligned.
* @return The calculated hash.
*/
-XXH_FORCE_INLINE xxh_u32
+XXH_FORCE_INLINE XXH_PUREF xxh_u32
XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
{
xxh_u32 h32;
@@ -2112,10 +2788,10 @@ XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment
return XXH32_finalize(h32, input, len&15, align);
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
{
-#if 0
+#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
/* Simple version, good for code maintenance, but unfortunately slow for small inputs */
XXH32_state_t state;
XXH32_reset(&state, seed);
@@ -2134,42 +2810,39 @@ XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t s
/******* Hash streaming *******/
-/*!
- * @ingroup xxh32_family
- */
+#ifndef XXH_NO_STREAM
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
{
return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
{
XXH_free(statePtr);
return XXH_OK;
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
{
XXH_memcpy(dstState, srcState, sizeof(*dstState));
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
{
- XXH32_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
- memset(&state, 0, sizeof(state));
- state.v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
- state.v[1] = seed + XXH_PRIME32_2;
- state.v[2] = seed + 0;
- state.v[3] = seed - XXH_PRIME32_1;
- /* do not write into reserved, planned to be removed in a future version */
- XXH_memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved));
+ XXH_ASSERT(statePtr != NULL);
+ memset(statePtr, 0, sizeof(*statePtr));
+ statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+ statePtr->v[1] = seed + XXH_PRIME32_2;
+ statePtr->v[2] = seed + 0;
+ statePtr->v[3] = seed - XXH_PRIME32_1;
return XXH_OK;
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH_errorcode
XXH32_update(XXH32_state_t* state, const void* input, size_t len)
{
@@ -2224,7 +2897,7 @@ XXH32_update(XXH32_state_t* state, const void* input, size_t len)
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
{
xxh_u32 h32;
@@ -2242,12 +2915,12 @@ XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
}
-
+#endif /* !XXH_NO_STREAM */
/******* Canonical representation *******/
/*!
- * @ingroup xxh32_family
+ * @ingroup XXH32_family
* The default return values from XXH functions are unsigned 32 and 64 bit
* integers.
*
@@ -2266,7 +2939,7 @@ XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t
if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
XXH_memcpy(dst, &hash, sizeof(*dst));
}
-/*! @ingroup xxh32_family */
+/*! @ingroup XXH32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
{
return XXH_readBE32(src);
@@ -2307,25 +2980,26 @@ static xxh_u64 XXH_read64(const void* memPtr)
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
/*
- * __pack instructions are safer, but compiler specific, hence potentially
- * problematic for some compilers.
- *
- * Currently only defined for GCC and ICC.
+ * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
+ * documentation claimed that it only increased the alignment, but actually it
+ * can decrease it on gcc, clang, and icc:
+ * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
+ * https://gcc.godbolt.org/z/xYez1j67Y.
*/
#ifdef XXH_OLD_NAMES
typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
#endif
static xxh_u64 XXH_read64(const void* ptr)
{
- typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) xxh_unalign64;
- return ((const xxh_unalign64*)ptr)->u64;
+ typedef __attribute__((aligned(1))) xxh_u64 xxh_unalign64;
+ return *((const xxh_unalign64*)ptr);
}
#else
/*
* Portable and safe solution. Generally efficient.
- * see: http://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
*/
static xxh_u64 XXH_read64(const void* memPtr)
{
@@ -2409,8 +3083,10 @@ XXH_readLE64_align(const void* ptr, XXH_alignment align)
/******* xxh64 *******/
/*!
* @}
- * @defgroup xxh64_impl XXH64 implementation
+ * @defgroup XXH64_impl XXH64 implementation
* @ingroup impl
+ *
+ * Details on the XXH64 implementation.
* @{
*/
/* #define rather that static const, to be used as initializers */
@@ -2428,6 +3104,7 @@ XXH_readLE64_align(const void* ptr, XXH_alignment align)
# define PRIME64_5 XXH_PRIME64_5
#endif
+/*! @copydoc XXH32_round */
static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
{
acc += input * XXH_PRIME64_2;
@@ -2444,43 +3121,59 @@ static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
return acc;
}
-static xxh_u64 XXH64_avalanche(xxh_u64 h64)
+/*! @copydoc XXH32_avalanche */
+static xxh_u64 XXH64_avalanche(xxh_u64 hash)
{
- h64 ^= h64 >> 33;
- h64 *= XXH_PRIME64_2;
- h64 ^= h64 >> 29;
- h64 *= XXH_PRIME64_3;
- h64 ^= h64 >> 32;
- return h64;
+ hash ^= hash >> 33;
+ hash *= XXH_PRIME64_2;
+ hash ^= hash >> 29;
+ hash *= XXH_PRIME64_3;
+ hash ^= hash >> 32;
+ return hash;
}
#define XXH_get64bits(p) XXH_readLE64_align(p, align)
-static xxh_u64
-XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align)
+/*!
+ * @internal
+ * @brief Processes the last 0-31 bytes of @p ptr.
+ *
+ * There may be up to 31 bytes remaining to consume from the input.
+ * This final stage will digest them to ensure that all input bytes are present
+ * in the final mix.
+ *
+ * @param hash The hash to finalize.
+ * @param ptr The pointer to the remaining input.
+ * @param len The remaining length, modulo 32.
+ * @param align Whether @p ptr is aligned.
+ * @return The finalized hash
+ * @see XXH32_finalize().
+ */
+static XXH_PUREF xxh_u64
+XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
{
if (ptr==NULL) XXH_ASSERT(len == 0);
len &= 31;
while (len >= 8) {
xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
ptr += 8;
- h64 ^= k1;
- h64 = XXH_rotl64(h64,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
+ hash ^= k1;
+ hash = XXH_rotl64(hash,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
len -= 8;
}
if (len >= 4) {
- h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
+ hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
ptr += 4;
- h64 = XXH_rotl64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
+ hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
len -= 4;
}
while (len > 0) {
- h64 ^= (*ptr++) * XXH_PRIME64_5;
- h64 = XXH_rotl64(h64, 11) * XXH_PRIME64_1;
+ hash ^= (*ptr++) * XXH_PRIME64_5;
+ hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1;
--len;
}
- return XXH64_avalanche(h64);
+ return XXH64_avalanche(hash);
}
#ifdef XXH_OLD_NAMES
@@ -2493,7 +3186,15 @@ XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align)
# undef XXH_PROCESS8_64
#endif
-XXH_FORCE_INLINE xxh_u64
+/*!
+ * @internal
+ * @brief The implementation for @ref XXH64().
+ *
+ * @param input , len , seed Directly passed from @ref XXH64().
+ * @param align Whether @p input is aligned.
+ * @return The calculated hash.
+ */
+XXH_FORCE_INLINE XXH_PUREF xxh_u64
XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
{
xxh_u64 h64;
@@ -2530,10 +3231,10 @@ XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment
}
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t len, XXH64_hash_t seed)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64 (XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
{
-#if 0
+#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
/* Simple version, good for code maintenance, but unfortunately slow for small inputs */
XXH64_state_t state;
XXH64_reset(&state, seed);
@@ -2551,42 +3252,40 @@ XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t len, XXH64_hash_t s
}
/******* Hash Streaming *******/
-
-/*! @ingroup xxh64_family*/
+#ifndef XXH_NO_STREAM
+/*! @ingroup XXH64_family*/
XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
{
return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
}
-/*! @ingroup xxh64_family */
+/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
{
XXH_free(statePtr);
return XXH_OK;
}
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dstState, const XXH64_state_t* srcState)
{
XXH_memcpy(dstState, srcState, sizeof(*dstState));
}
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, XXH64_hash_t seed)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed)
{
- XXH64_state_t state; /* use a local state to memcpy() in order to avoid strict-aliasing warnings */
- memset(&state, 0, sizeof(state));
- state.v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
- state.v[1] = seed + XXH_PRIME64_2;
- state.v[2] = seed + 0;
- state.v[3] = seed - XXH_PRIME64_1;
- /* do not write into reserved64, might be removed in a future version */
- XXH_memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved64));
+ XXH_ASSERT(statePtr != NULL);
+ memset(statePtr, 0, sizeof(*statePtr));
+ statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+ statePtr->v[1] = seed + XXH_PRIME64_2;
+ statePtr->v[2] = seed + 0;
+ statePtr->v[3] = seed - XXH_PRIME64_1;
return XXH_OK;
}
-/*! @ingroup xxh64_family */
+/*! @ingroup XXH64_family */
XXH_PUBLIC_API XXH_errorcode
-XXH64_update (XXH64_state_t* state, const void* input, size_t len)
+XXH64_update (XXH_NOESCAPE XXH64_state_t* state, XXH_NOESCAPE const void* input, size_t len)
{
if (input==NULL) {
XXH_ASSERT(len == 0);
@@ -2636,8 +3335,8 @@ XXH64_update (XXH64_state_t* state, const void* input, size_t len)
}
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t* state)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_digest(XXH_NOESCAPE const XXH64_state_t* state)
{
xxh_u64 h64;
@@ -2655,20 +3354,20 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t* state)
return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
}
-
+#endif /* !XXH_NO_STREAM */
/******* Canonical representation *******/
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash)
{
XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
XXH_memcpy(dst, &hash, sizeof(*dst));
}
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src)
{
return XXH_readBE64(src);
}
@@ -2681,7 +3380,7 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src
************************************************************************ */
/*!
* @}
- * @defgroup xxh3_impl XXH3 implementation
+ * @defgroup XXH3_impl XXH3 implementation
* @ingroup impl
* @{
*/
@@ -2689,11 +3388,19 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src
/* === Compiler specifics === */
#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
-# define XXH_RESTRICT /* disable */
+# define XXH_RESTRICT /* disable */
#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* >= C99 */
# define XXH_RESTRICT restrict
+#elif (defined (__GNUC__) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) \
+ || (defined (__clang__)) \
+ || (defined (_MSC_VER) && (_MSC_VER >= 1400)) \
+ || (defined (__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300))
+/*
+ * There are a LOT more compilers that recognize __restrict but this
+ * covers the major ones.
+ */
+# define XXH_RESTRICT __restrict
#else
-/* Note: it might be useful to define __restrict or __restrict__ for some C++ compilers */
# define XXH_RESTRICT /* disable */
#endif
@@ -2707,17 +3414,33 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src
# define XXH_unlikely(x) (x)
#endif
-#if defined(__GNUC__)
-# if defined(__AVX2__)
-# include <immintrin.h>
-# elif defined(__SSE2__)
-# include <emmintrin.h>
-# elif defined(__ARM_NEON__) || defined(__ARM_NEON)
+#ifndef XXH_HAS_INCLUDE
+# ifdef __has_include
+# define XXH_HAS_INCLUDE(x) __has_include(x)
+# else
+# define XXH_HAS_INCLUDE(x) 0
+# endif
+#endif
+
+#if defined(__GNUC__) || defined(__clang__)
+# if defined(__ARM_FEATURE_SVE)
+# include <arm_sve.h>
+# endif
+# if defined(__ARM_NEON__) || defined(__ARM_NEON) \
+ || (defined(_M_ARM) && _M_ARM >= 7) \
+ || defined(_M_ARM64) || defined(_M_ARM64EC) \
+ || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* WASM SIMD128 via SIMDe */
# define inline __inline__ /* circumvent a clang bug */
# include <arm_neon.h>
# undef inline
+# elif defined(__AVX2__)
+# include <immintrin.h>
+# elif defined(__SSE2__)
+# include <emmintrin.h>
# endif
-#elif defined(_MSC_VER)
+#endif
+
+#if defined(_MSC_VER)
# include <intrin.h>
#endif
@@ -2817,7 +3540,7 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src
* Note that these are actually implemented as macros.
*
* If this is not defined, it is detected automatically.
- * @ref XXH_X86DISPATCH overrides this.
+ * internal macro XXH_X86DISPATCH overrides this.
*/
enum XXH_VECTOR_TYPE /* fake enum */ {
XXH_SCALAR = 0, /*!< Portable scalar version */
@@ -2829,14 +3552,19 @@ enum XXH_VECTOR_TYPE /* fake enum */ {
*/
XXH_AVX2 = 2, /*!< AVX2 for Haswell and Bulldozer */
XXH_AVX512 = 3, /*!< AVX512 for Skylake and Icelake */
- XXH_NEON = 4, /*!< NEON for most ARMv7-A and all AArch64 */
+ XXH_NEON = 4, /*!<
+ * NEON for most ARMv7-A, all AArch64, and WASM SIMD128
+ * via the SIMDeverywhere polyfill provided with the
+ * Emscripten SDK.
+ */
XXH_VSX = 5, /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+ XXH_SVE = 6, /*!< SVE for some ARMv8-A and ARMv9-A */
};
/*!
* @ingroup tuning
* @brief Selects the minimum alignment for XXH3's accumulators.
*
- * When using SIMD, this should match the alignment reqired for said vector
+ * When using SIMD, this should match the alignment required for said vector
* type, so, for example, 32 for AVX2.
*
* Default: Auto detected.
@@ -2852,23 +3580,27 @@ enum XXH_VECTOR_TYPE /* fake enum */ {
# define XXH_AVX512 3
# define XXH_NEON 4
# define XXH_VSX 5
+# define XXH_SVE 6
#endif
#ifndef XXH_VECTOR /* can be defined on command line */
-# if defined(__AVX512F__)
-# define XXH_VECTOR XXH_AVX512
-# elif defined(__AVX2__)
-# define XXH_VECTOR XXH_AVX2
-# elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
-# define XXH_VECTOR XXH_SSE2
+# if defined(__ARM_FEATURE_SVE)
+# define XXH_VECTOR XXH_SVE
# elif ( \
defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
- || defined(_M_ARM64) || defined(_M_ARM_ARMV7VE) /* msvc */ \
+ || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
+ || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* wasm simd128 via SIMDe */ \
) && ( \
defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
|| (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
)
# define XXH_VECTOR XXH_NEON
+# elif defined(__AVX512F__)
+# define XXH_VECTOR XXH_AVX512
+# elif defined(__AVX2__)
+# define XXH_VECTOR XXH_AVX2
+# elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
+# define XXH_VECTOR XXH_SSE2
# elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
|| (defined(__s390x__) && defined(__VEC__)) \
&& defined(__GNUC__) /* TODO: IBM XL */
@@ -2878,6 +3610,17 @@ enum XXH_VECTOR_TYPE /* fake enum */ {
# endif
#endif
+/* __ARM_FEATURE_SVE is only supported by GCC & Clang. */
+#if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE)
+# ifdef _MSC_VER
+# pragma warning(once : 4606)
+# else
+# warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead."
+# endif
+# undef XXH_VECTOR
+# define XXH_VECTOR XXH_SCALAR
+#endif
+
/*
* Controls the alignment of the accumulator,
* for compatibility with aligned vector loads, which are usually faster.
@@ -2897,16 +3640,26 @@ enum XXH_VECTOR_TYPE /* fake enum */ {
# define XXH_ACC_ALIGN 16
# elif XXH_VECTOR == XXH_AVX512 /* avx512 */
# define XXH_ACC_ALIGN 64
+# elif XXH_VECTOR == XXH_SVE /* sve */
+# define XXH_ACC_ALIGN 64
# endif
#endif
#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
|| XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
# define XXH_SEC_ALIGN XXH_ACC_ALIGN
+#elif XXH_VECTOR == XXH_SVE
+# define XXH_SEC_ALIGN XXH_ACC_ALIGN
#else
# define XXH_SEC_ALIGN 8
#endif
+#if defined(__GNUC__) || defined(__clang__)
+# define XXH_ALIASING __attribute__((may_alias))
+#else
+# define XXH_ALIASING /* nothing */
+#endif
+
/*
* UGLY HACK:
* GCC usually generates the best code with -O3 for xxHash.
@@ -2930,111 +3683,130 @@ enum XXH_VECTOR_TYPE /* fake enum */ {
*/
#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
&& defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
- && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+ && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
# pragma GCC push_options
# pragma GCC optimize("-O2")
#endif
-
#if XXH_VECTOR == XXH_NEON
+
/*
- * NEON's setup for vmlal_u32 is a little more complicated than it is on
- * SSE2, AVX2, and VSX.
- *
- * While PMULUDQ and VMULEUW both perform a mask, VMLAL.U32 performs an upcast.
+ * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC -O3
+ * optimizes out the entire hashLong loop because of the aliasing violation.
*
- * To do the same operation, the 128-bit 'Q' register needs to be split into
- * two 64-bit 'D' registers, performing this operation::
- *
- * [ a | b ]
- * | '---------. .--------' |
- * | x |
- * | .---------' '--------. |
- * [ a & 0xFFFFFFFF | b & 0xFFFFFFFF ],[ a >> 32 | b >> 32 ]
- *
- * Due to significant changes in aarch64, the fastest method for aarch64 is
- * completely different than the fastest method for ARMv7-A.
- *
- * ARMv7-A treats D registers as unions overlaying Q registers, so modifying
- * D11 will modify the high half of Q5. This is similar to how modifying AH
- * will only affect bits 8-15 of AX on x86.
- *
- * VZIP takes two registers, and puts even lanes in one register and odd lanes
- * in the other.
+ * However, GCC is also inefficient at load-store optimization with vld1q/vst1q,
+ * so the only option is to mark it as aliasing.
+ */
+typedef uint64x2_t xxh_aliasing_uint64x2_t XXH_ALIASING;
+
+/*!
+ * @internal
+ * @brief `vld1q_u64` but faster and alignment-safe.
*
- * On ARMv7-A, this strangely modifies both parameters in place instead of
- * taking the usual 3-operand form.
+ * On AArch64, unaligned access is always safe, but on ARMv7-a, it is only
+ * *conditionally* safe (`vld1` has an alignment bit like `movdq[ua]` in x86).
*
- * Therefore, if we want to do this, we can simply use a D-form VZIP.32 on the
- * lower and upper halves of the Q register to end up with the high and low
- * halves where we want - all in one instruction.
+ * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so it
+ * prohibits load-store optimizations. Therefore, a direct dereference is used.
*
- * vzip.32 d10, d11 @ d10 = { d10[0], d11[0] }; d11 = { d10[1], d11[1] }
+ * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a safe
+ * unaligned load.
+ */
+#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__)
+XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) /* silence -Wcast-align */
+{
+ return *(xxh_aliasing_uint64x2_t const *)ptr;
+}
+#else
+XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr)
+{
+ return vreinterpretq_u64_u8(vld1q_u8((uint8_t const*)ptr));
+}
+#endif
+
+/*!
+ * @internal
+ * @brief `vmlal_u32` on low and high halves of a vector.
*
- * Unfortunately we need inline assembly for this: Instructions modifying two
- * registers at once is not possible in GCC or Clang's IR, and they have to
- * create a copy.
+ * This is a workaround for AArch64 GCC < 11 which implemented arm_neon.h with
+ * inline assembly and were therefore incapable of merging the `vget_{low, high}_u32`
+ * with `vmlal_u32`.
+ */
+#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 11
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+ /* Inline assembly is the only way */
+ __asm__("umlal %0.2d, %1.2s, %2.2s" : "+w" (acc) : "w" (lhs), "w" (rhs));
+ return acc;
+}
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+ /* This intrinsic works as expected */
+ return vmlal_high_u32(acc, lhs, rhs);
+}
+#else
+/* Portable intrinsic versions */
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+ return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs));
+}
+/*! @copydoc XXH_vmlal_low_u32
+ * Assume the compiler converts this to vmlal_high_u32 on aarch64 */
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+ return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs));
+}
+#endif
+
+/*!
+ * @ingroup tuning
+ * @brief Controls the NEON to scalar ratio for XXH3
*
- * aarch64 requires a different approach.
+ * This can be set to 2, 4, 6, or 8.
*
- * In order to make it easier to write a decent compiler for aarch64, many
- * quirks were removed, such as conditional execution.
+ * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used.
*
- * NEON was also affected by this.
+ * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but only 2 of those
+ * can be NEON. If you are only using NEON instructions, you are only using 2/3 of the CPU
+ * bandwidth.
*
- * aarch64 cannot access the high bits of a Q-form register, and writes to a
- * D-form register zero the high bits, similar to how writes to W-form scalar
- * registers (or DWORD registers on x86_64) work.
+ * This is even more noticeable on the more advanced cores like the Cortex-A76 which
+ * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
*
- * The formerly free vget_high intrinsics now require a vext (with a few
- * exceptions)
+ * Therefore, to make the most out of the pipeline, it is beneficial to run 6 NEON lanes
+ * and 2 scalar lanes, which is chosen by default.
*
- * Additionally, VZIP was replaced by ZIP1 and ZIP2, which are the equivalent
- * of PUNPCKL* and PUNPCKH* in SSE, respectively, in order to only modify one
- * operand.
+ * This does not apply to Apple processors or 32-bit processors, which run better with
+ * full NEON. These will default to 8. Additionally, size-optimized builds run 8 lanes.
*
- * The equivalent of the VZIP.32 on the lower and upper halves would be this
- * mess:
+ * This change benefits CPUs with large micro-op buffers without negatively affecting
+ * most other CPUs:
*
- * ext v2.4s, v0.4s, v0.4s, #2 // v2 = { v0[2], v0[3], v0[0], v0[1] }
- * zip1 v1.2s, v0.2s, v2.2s // v1 = { v0[0], v2[0] }
- * zip2 v0.2s, v0.2s, v1.2s // v0 = { v0[1], v2[1] }
+ * | Chipset | Dispatch type | NEON only | 6:2 hybrid | Diff. |
+ * |:----------------------|:--------------------|----------:|-----------:|------:|
+ * | Snapdragon 730 (A76) | 2 NEON/8 micro-ops | 8.8 GB/s | 10.1 GB/s | ~16% |
+ * | Snapdragon 835 (A73) | 2 NEON/3 micro-ops | 5.1 GB/s | 5.3 GB/s | ~5% |
+ * | Marvell PXA1928 (A53) | In-order dual-issue | 1.9 GB/s | 1.9 GB/s | 0% |
+ * | Apple M1 | 4 NEON/8 micro-ops | 37.3 GB/s | 36.1 GB/s | ~-3% |
*
- * Instead, we use a literal downcast, vmovn_u64 (XTN), and vshrn_n_u64 (SHRN):
+ * It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
*
- * shrn v1.2s, v0.2d, #32 // v1 = (uint32x2_t)(v0 >> 32);
- * xtn v0.2s, v0.2d // v0 = (uint32x2_t)(v0 & 0xFFFFFFFF);
+ * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of the lanes meaning
+ * it effectively becomes worse 4.
*
- * This is available on ARMv7-A, but is less efficient than a single VZIP.32.
+ * @see XXH3_accumulate_512_neon()
*/
-
-/*!
- * Function-like macro:
- * void XXH_SPLIT_IN_PLACE(uint64x2_t &in, uint32x2_t &outLo, uint32x2_t &outHi)
- * {
- * outLo = (uint32x2_t)(in & 0xFFFFFFFF);
- * outHi = (uint32x2_t)(in >> 32);
- * in = UNDEFINED;
- * }
- */
-# if !defined(XXH_NO_VZIP_HACK) /* define to disable */ \
- && defined(__GNUC__) \
- && !defined(__aarch64__) && !defined(__arm64__) && !defined(_M_ARM64)
-# define XXH_SPLIT_IN_PLACE(in, outLo, outHi) \
- do { \
- /* Undocumented GCC/Clang operand modifier: %e0 = lower D half, %f0 = upper D half */ \
- /* https://github.com/gcc-mirror/gcc/blob/38cf91e5/gcc/config/arm/arm.c#L22486 */ \
- /* https://github.com/llvm-mirror/llvm/blob/2c4ca683/lib/Target/ARM/ARMAsmPrinter.cpp#L399 */ \
- __asm__("vzip.32 %e0, %f0" : "+w" (in)); \
- (outLo) = vget_low_u32 (vreinterpretq_u32_u64(in)); \
- (outHi) = vget_high_u32(vreinterpretq_u32_u64(in)); \
- } while (0)
-# else
-# define XXH_SPLIT_IN_PLACE(in, outLo, outHi) \
- do { \
- (outLo) = vmovn_u64 (in); \
- (outHi) = vshrn_n_u64 ((in), 32); \
- } while (0)
+# ifndef XXH3_NEON_LANES
+# if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
+ && !defined(__APPLE__) && XXH_SIZE_OPT <= 0
+# define XXH3_NEON_LANES 6
+# else
+# define XXH3_NEON_LANES XXH_ACC_NB
+# endif
# endif
#endif /* XXH_VECTOR == XXH_NEON */
@@ -3047,27 +3819,42 @@ enum XXH_VECTOR_TYPE /* fake enum */ {
* inconsistent intrinsics, spotty coverage, and multiple endiannesses.
*/
#if XXH_VECTOR == XXH_VSX
+/* Annoyingly, these headers _may_ define three macros: `bool`, `vector`,
+ * and `pixel`. This is a problem for obvious reasons.
+ *
+ * These keywords are unnecessary; the spec literally says they are
+ * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd
+ * after including the header.
+ *
+ * We use pragma push_macro/pop_macro to keep the namespace clean. */
+# pragma push_macro("bool")
+# pragma push_macro("vector")
+# pragma push_macro("pixel")
+/* silence potential macro redefined warnings */
+# undef bool
+# undef vector
+# undef pixel
+
# if defined(__s390x__)
# include <s390intrin.h>
# else
-/* gcc's altivec.h can have the unwanted consequence to unconditionally
- * #define bool, vector, and pixel keywords,
- * with bad consequences for programs already using these keywords for other purposes.
- * The paragraph defining these macros is skipped when __APPLE_ALTIVEC__ is defined.
- * __APPLE_ALTIVEC__ is _generally_ defined automatically by the compiler,
- * but it seems that, in some cases, it isn't.
- * Force the build macro to be defined, so that keywords are not altered.
- */
-# if defined(__GNUC__) && !defined(__APPLE_ALTIVEC__)
-# define __APPLE_ALTIVEC__
-# endif
# include <altivec.h>
# endif
+/* Restore the original macro values, if applicable. */
+# pragma pop_macro("pixel")
+# pragma pop_macro("vector")
+# pragma pop_macro("bool")
+
typedef __vector unsigned long long xxh_u64x2;
typedef __vector unsigned char xxh_u8x16;
typedef __vector unsigned xxh_u32x4;
+/*
+ * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing issue.
+ */
+typedef xxh_u64x2 xxh_aliasing_u64x2 XXH_ALIASING;
+
# ifndef XXH_VSX_BE
# if defined(__BIG_ENDIAN__) \
|| (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
@@ -3119,8 +3906,9 @@ XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
/* s390x is always big endian, no issue on this platform */
# define XXH_vec_mulo vec_mulo
# define XXH_vec_mule vec_mule
-# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
+# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__)
/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
+ /* The IBM XL Compiler (which defined __clang__) only implements the vec_* operations */
# define XXH_vec_mulo __builtin_altivec_vmulouw
# define XXH_vec_mule __builtin_altivec_vmuleuw
# else
@@ -3141,13 +3929,28 @@ XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
# endif /* XXH_vec_mulo, XXH_vec_mule */
#endif /* XXH_VECTOR == XXH_VSX */
+#if XXH_VECTOR == XXH_SVE
+#define ACCRND(acc, offset) \
+do { \
+ svuint64_t input_vec = svld1_u64(mask, xinput + offset); \
+ svuint64_t secret_vec = svld1_u64(mask, xsecret + offset); \
+ svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec); \
+ svuint64_t swapped = svtbl_u64(input_vec, kSwap); \
+ svuint64_t mixed_lo = svextw_u64_x(mask, mixed); \
+ svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32); \
+ svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \
+ acc = svadd_u64_x(mask, acc, mul); \
+} while (0)
+#endif /* XXH_VECTOR == XXH_SVE */
/* prefetch
* can be disabled, by declaring XXH_NO_PREFETCH build macro */
#if defined(XXH_NO_PREFETCH)
# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */
#else
-# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) /* _mm_prefetch() not defined outside of x86/x64 */
+# if XXH_SIZE_OPT >= 1
+# define XXH_PREFETCH(ptr) (void)(ptr)
+# elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) /* _mm_prefetch() not defined outside of x86/x64 */
# include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
# define XXH_PREFETCH(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
@@ -3184,6 +3987,8 @@ XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
};
+static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL; /*!< 0b0001011001010110011001111001000110011110001101110111100111111001 */
+static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL; /*!< 0b1001111110110010000111000110010100011110100110001101111100100101 */
#ifdef XXH_OLD_NAMES
# define kSecret XXH3_kSecret
@@ -3212,7 +4017,6 @@ XXH_mult32to64(xxh_u64 x, xxh_u64 y)
return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
}
#elif defined(_MSC_VER) && defined(_M_IX86)
-# include <intrin.h>
# define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
#else
/*
@@ -3252,7 +4056,7 @@ XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
* In that case it is best to use the portable one.
* https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
*/
-#if defined(__GNUC__) && !defined(__wasm__) \
+#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
&& defined(__SIZEOF_INT128__) \
|| (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
@@ -3269,7 +4073,7 @@ XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
*
* This compiles to single operand MUL on x64.
*/
-#elif defined(_M_X64) || defined(_M_IA64)
+#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
#ifndef _MSC_VER
# pragma intrinsic(_umul128)
@@ -3286,7 +4090,7 @@ XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
*
* This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
*/
-#elif defined(_M_ARM64)
+#elif defined(_M_ARM64) || defined(_M_ARM64EC)
#ifndef _MSC_VER
# pragma intrinsic(__umulh)
@@ -3376,7 +4180,7 @@ XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
}
/*! Seems to produce slightly better code on GCC for some reason. */
-XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
+XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
{
XXH_ASSERT(0 <= shift && shift < 64);
return v64 ^ (v64 >> shift);
@@ -3389,7 +4193,7 @@ XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
{
h64 = XXH_xorshift64(h64, 37);
- h64 *= 0x165667919E3779F9ULL;
+ h64 *= PRIME_MX1;
h64 = XXH_xorshift64(h64, 32);
return h64;
}
@@ -3403,9 +4207,9 @@ static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
{
/* this mix is inspired by Pelle Evensen's rrmxmx */
h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
- h64 *= 0x9FB21C651E98DF25ULL;
+ h64 *= PRIME_MX2;
h64 ^= (h64 >> 35) + len ;
- h64 *= 0x9FB21C651E98DF25ULL;
+ h64 *= PRIME_MX2;
return XXH_xorshift64(h64, 28);
}
@@ -3443,7 +4247,7 @@ static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
*
* This adds an extra layer of strength for custom secrets.
*/
-XXH_FORCE_INLINE XXH64_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
@@ -3465,7 +4269,7 @@ XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_h
}
}
-XXH_FORCE_INLINE XXH64_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
@@ -3481,7 +4285,7 @@ XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_h
}
}
-XXH_FORCE_INLINE XXH64_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
@@ -3498,7 +4302,7 @@ XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_
}
}
-XXH_FORCE_INLINE XXH64_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(len <= 16);
@@ -3568,7 +4372,7 @@ XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
}
/* For mid range keys, XXH3 uses a Mum-hash variant. */
-XXH_FORCE_INLINE XXH64_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
@@ -3577,6 +4381,14 @@ XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
XXH_ASSERT(16 < len && len <= 128);
{ xxh_u64 acc = len * XXH_PRIME64_1;
+#if XXH_SIZE_OPT >= 1
+ /* Smaller and cleaner, but slightly slower. */
+ unsigned int i = (unsigned int)(len - 1) / 32;
+ do {
+ acc += XXH3_mix16B(input+16 * i, secret+32*i, seed);
+ acc += XXH3_mix16B(input+len-16*(i+1), secret+32*i+16, seed);
+ } while (i-- != 0);
+#else
if (len > 32) {
if (len > 64) {
if (len > 96) {
@@ -3591,14 +4403,14 @@ XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
}
acc += XXH3_mix16B(input+0, secret+0, seed);
acc += XXH3_mix16B(input+len-16, secret+16, seed);
-
+#endif
return XXH3_avalanche(acc);
}
}
#define XXH3_MIDSIZE_MAX 240
-XXH_NO_INLINE XXH64_hash_t
+XXH_NO_INLINE XXH_PUREF XXH64_hash_t
XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
@@ -3610,13 +4422,17 @@ XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
#define XXH3_MIDSIZE_LASTOFFSET 17
{ xxh_u64 acc = len * XXH_PRIME64_1;
- int const nbRounds = (int)len / 16;
- int i;
+ xxh_u64 acc_end;
+ unsigned int const nbRounds = (unsigned int)len / 16;
+ unsigned int i;
+ XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
for (i=0; i<8; i++) {
acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
}
- acc = XXH3_avalanche(acc);
+ /* last bytes */
+ acc_end = XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
XXH_ASSERT(nbRounds >= 8);
+ acc = XXH3_avalanche(acc);
#if defined(__clang__) /* Clang */ \
&& (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
&& !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */
@@ -3643,11 +4459,13 @@ XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
#pragma clang loop vectorize(disable)
#endif
for (i=8 ; i < nbRounds; i++) {
- acc += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
+ /*
+ * Prevents clang for unrolling the acc loop and interleaving with this one.
+ */
+ XXH_COMPILER_GUARD(acc);
+ acc_end += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
}
- /* last bytes */
- acc += XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
- return XXH3_avalanche(acc);
+ return XXH3_avalanche(acc + acc_end);
}
}
@@ -3663,6 +4481,47 @@ XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
# define ACC_NB XXH_ACC_NB
#endif
+#ifndef XXH_PREFETCH_DIST
+# ifdef __clang__
+# define XXH_PREFETCH_DIST 320
+# else
+# if (XXH_VECTOR == XXH_AVX512)
+# define XXH_PREFETCH_DIST 512
+# else
+# define XXH_PREFETCH_DIST 384
+# endif
+# endif /* __clang__ */
+#endif /* XXH_PREFETCH_DIST */
+
+/*
+ * These macros are to generate an XXH3_accumulate() function.
+ * The two arguments select the name suffix and target attribute.
+ *
+ * The name of this symbol is XXH3_accumulate_<name>() and it calls
+ * XXH3_accumulate_512_<name>().
+ *
+ * It may be useful to hand implement this function if the compiler fails to
+ * optimize the inline function.
+ */
+#define XXH3_ACCUMULATE_TEMPLATE(name) \
+void \
+XXH3_accumulate_##name(xxh_u64* XXH_RESTRICT acc, \
+ const xxh_u8* XXH_RESTRICT input, \
+ const xxh_u8* XXH_RESTRICT secret, \
+ size_t nbStripes) \
+{ \
+ size_t n; \
+ for (n = 0; n < nbStripes; n++ ) { \
+ const xxh_u8* const in = input + n*XXH_STRIPE_LEN; \
+ XXH_PREFETCH(in + XXH_PREFETCH_DIST); \
+ XXH3_accumulate_512_##name( \
+ acc, \
+ in, \
+ secret + n*XXH_SECRET_CONSUME_RATE); \
+ } \
+}
+
+
XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
{
if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
@@ -3683,6 +4542,7 @@ XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
typedef long long xxh_i64;
#endif
+
/*
* XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
*
@@ -3730,7 +4590,7 @@ XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
/* data_key = data_vec ^ key_vec; */
__m512i const data_key = _mm512_xor_si512 (data_vec, key_vec);
/* data_key_lo = data_key >> 32; */
- __m512i const data_key_lo = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+ __m512i const data_key_lo = _mm512_srli_epi64 (data_key, 32);
/* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
__m512i const product = _mm512_mul_epu32 (data_key, data_key_lo);
/* xacc[0] += swap(data_vec); */
@@ -3740,6 +4600,7 @@ XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
*xacc = _mm512_add_epi64(product, sum);
}
}
+XXH_FORCE_INLINE XXH_TARGET_AVX512 XXH3_ACCUMULATE_TEMPLATE(avx512)
/*
* XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
@@ -3773,13 +4634,12 @@ XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
/* xacc[0] ^= (xacc[0] >> 47) */
__m512i const acc_vec = *xacc;
__m512i const shifted = _mm512_srli_epi64 (acc_vec, 47);
- __m512i const data_vec = _mm512_xor_si512 (acc_vec, shifted);
/* xacc[0] ^= secret; */
__m512i const key_vec = _mm512_loadu_si512 (secret);
- __m512i const data_key = _mm512_xor_si512 (data_vec, key_vec);
+ __m512i const data_key = _mm512_ternarylogic_epi32(key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */);
/* xacc[0] *= XXH_PRIME32_1; */
- __m512i const data_key_hi = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+ __m512i const data_key_hi = _mm512_srli_epi64 (data_key, 32);
__m512i const prod_lo = _mm512_mul_epu32 (data_key, prime32);
__m512i const prod_hi = _mm512_mul_epu32 (data_key_hi, prime32);
*xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
@@ -3794,7 +4654,8 @@ XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
XXH_ASSERT(((size_t)customSecret & 63) == 0);
(void)(&XXH_writeLE64);
{ int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
- __m512i const seed = _mm512_mask_set1_epi64(_mm512_set1_epi64((xxh_i64)seed64), 0xAA, (xxh_i64)(0U - seed64));
+ __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64);
+ __m512i const seed = _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos);
const __m512i* const src = (const __m512i*) ((const void*) XXH3_kSecret);
__m512i* const dest = ( __m512i*) customSecret;
@@ -3802,14 +4663,7 @@ XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
XXH_ASSERT(((size_t)dest & 63) == 0);
for (i=0; i < nbRounds; ++i) {
- /* GCC has a bug, _mm512_stream_load_si512 accepts 'void*', not 'void const*',
- * this will warn "discards 'const' qualifier". */
- union {
- const __m512i* cp;
- void* p;
- } remote_const_void;
- remote_const_void.cp = src + i;
- dest[i] = _mm512_add_epi64(_mm512_stream_load_si512(remote_const_void.p), seed);
+ dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed);
} }
}
@@ -3845,7 +4699,7 @@ XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
/* data_key = data_vec ^ key_vec; */
__m256i const data_key = _mm256_xor_si256 (data_vec, key_vec);
/* data_key_lo = data_key >> 32; */
- __m256i const data_key_lo = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+ __m256i const data_key_lo = _mm256_srli_epi64 (data_key, 32);
/* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
__m256i const product = _mm256_mul_epu32 (data_key, data_key_lo);
/* xacc[i] += swap(data_vec); */
@@ -3855,6 +4709,7 @@ XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
xacc[i] = _mm256_add_epi64(product, sum);
} }
}
+XXH_FORCE_INLINE XXH_TARGET_AVX2 XXH3_ACCUMULATE_TEMPLATE(avx2)
XXH_FORCE_INLINE XXH_TARGET_AVX2 void
XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
@@ -3877,7 +4732,7 @@ XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
__m256i const data_key = _mm256_xor_si256 (data_vec, key_vec);
/* xacc[i] *= XXH_PRIME32_1; */
- __m256i const data_key_hi = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+ __m256i const data_key_hi = _mm256_srli_epi64 (data_key, 32);
__m256i const prod_lo = _mm256_mul_epu32 (data_key, prime32);
__m256i const prod_hi = _mm256_mul_epu32 (data_key_hi, prime32);
xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
@@ -3909,12 +4764,12 @@ XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTR
XXH_ASSERT(((size_t)dest & 31) == 0);
/* GCC -O2 need unroll loop manually */
- dest[0] = _mm256_add_epi64(_mm256_stream_load_si256(src+0), seed);
- dest[1] = _mm256_add_epi64(_mm256_stream_load_si256(src+1), seed);
- dest[2] = _mm256_add_epi64(_mm256_stream_load_si256(src+2), seed);
- dest[3] = _mm256_add_epi64(_mm256_stream_load_si256(src+3), seed);
- dest[4] = _mm256_add_epi64(_mm256_stream_load_si256(src+4), seed);
- dest[5] = _mm256_add_epi64(_mm256_stream_load_si256(src+5), seed);
+ dest[0] = _mm256_add_epi64(_mm256_load_si256(src+0), seed);
+ dest[1] = _mm256_add_epi64(_mm256_load_si256(src+1), seed);
+ dest[2] = _mm256_add_epi64(_mm256_load_si256(src+2), seed);
+ dest[3] = _mm256_add_epi64(_mm256_load_si256(src+3), seed);
+ dest[4] = _mm256_add_epi64(_mm256_load_si256(src+4), seed);
+ dest[5] = _mm256_add_epi64(_mm256_load_si256(src+5), seed);
}
}
@@ -3961,6 +4816,7 @@ XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
xacc[i] = _mm_add_epi64(product, sum);
} }
}
+XXH_FORCE_INLINE XXH_TARGET_SSE2 XXH3_ACCUMULATE_TEMPLATE(sse2)
XXH_FORCE_INLINE XXH_TARGET_SSE2 void
XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
@@ -4028,96 +4884,222 @@ XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTR
#if (XXH_VECTOR == XXH_NEON)
+/* forward declarations for the scalar routines */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
+ void const* XXH_RESTRICT secret, size_t lane);
+
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+ void const* XXH_RESTRICT secret, size_t lane);
+
+/*!
+ * @internal
+ * @brief The bulk processing loop for NEON and WASM SIMD128.
+ *
+ * The NEON code path is actually partially scalar when running on AArch64. This
+ * is to optimize the pipelining and can have up to 15% speedup depending on the
+ * CPU, and it also mitigates some GCC codegen issues.
+ *
+ * @see XXH3_NEON_LANES for configuring this and details about this optimization.
+ *
+ * NEON's 32-bit to 64-bit long multiply takes a half vector of 32-bit
+ * integers instead of the other platforms which mask full 64-bit vectors,
+ * so the setup is more complicated than just shifting right.
+ *
+ * Additionally, there is an optimization for 4 lanes at once noted below.
+ *
+ * Since, as stated, the most optimal amount of lanes for Cortexes is 6,
+ * there needs to be *three* versions of the accumulate operation used
+ * for the remaining 2 lanes.
+ *
+ * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics overlap
+ * nearly perfectly.
+ */
+
XXH_FORCE_INLINE void
XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
- {
- uint64x2_t* const xacc = (uint64x2_t *) acc;
+ XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
+ { /* GCC for darwin arm64 does not like aliasing here */
+ xxh_aliasing_uint64x2_t* const xacc = (xxh_aliasing_uint64x2_t*) acc;
/* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
- uint8_t const* const xinput = (const uint8_t *) input;
- uint8_t const* const xsecret = (const uint8_t *) secret;
+ uint8_t const* xinput = (const uint8_t *) input;
+ uint8_t const* xsecret = (const uint8_t *) secret;
size_t i;
- for (i=0; i < XXH_STRIPE_LEN / sizeof(uint64x2_t); i++) {
+#ifdef __wasm_simd128__
+ /*
+ * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret
+ * is constant propagated, which results in it converting it to this
+ * inside the loop:
+ *
+ * a = v128.load(XXH3_kSecret + 0 + $secret_offset, offset = 0)
+ * b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0)
+ * ...
+ *
+ * This requires a full 32-bit address immediate (and therefore a 6 byte
+ * instruction) as well as an add for each offset.
+ *
+ * Putting an asm guard prevents it from folding (at the cost of losing
+ * the alignment hint), and uses the free offset in `v128.load` instead
+ * of adding secret_offset each time which overall reduces code size by
+ * about a kilobyte and improves performance.
+ */
+ XXH_COMPILER_GUARD(xsecret);
+#endif
+ /* Scalar lanes use the normal scalarRound routine */
+ for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+ XXH3_scalarRound(acc, input, secret, i);
+ }
+ i = 0;
+ /* 4 NEON lanes at a time. */
+ for (; i+1 < XXH3_NEON_LANES / 2; i+=2) {
/* data_vec = xinput[i]; */
- uint8x16_t data_vec = vld1q_u8(xinput + (i * 16));
+ uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput + (i * 16));
+ uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput + ((i+1) * 16));
/* key_vec = xsecret[i]; */
- uint8x16_t key_vec = vld1q_u8(xsecret + (i * 16));
- uint64x2_t data_key;
- uint32x2_t data_key_lo, data_key_hi;
- /* xacc[i] += swap(data_vec); */
- uint64x2_t const data64 = vreinterpretq_u64_u8(data_vec);
- uint64x2_t const swapped = vextq_u64(data64, data64, 1);
- xacc[i] = vaddq_u64 (xacc[i], swapped);
+ uint64x2_t key_vec_1 = XXH_vld1q_u64(xsecret + (i * 16));
+ uint64x2_t key_vec_2 = XXH_vld1q_u64(xsecret + ((i+1) * 16));
+ /* data_swap = swap(data_vec) */
+ uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1);
+ uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1);
/* data_key = data_vec ^ key_vec; */
- data_key = vreinterpretq_u64_u8(veorq_u8(data_vec, key_vec));
- /* data_key_lo = (uint32x2_t) (data_key & 0xFFFFFFFF);
- * data_key_hi = (uint32x2_t) (data_key >> 32);
- * data_key = UNDEFINED; */
- XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
- /* xacc[i] += (uint64x2_t) data_key_lo * (uint64x2_t) data_key_hi; */
- xacc[i] = vmlal_u32 (xacc[i], data_key_lo, data_key_hi);
+ uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1);
+ uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2);
+ /*
+ * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a
+ * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to
+ * get one vector with the low 32 bits of each lane, and one vector
+ * with the high 32 bits of each lane.
+ *
+ * The intrinsic returns a double vector because the original ARMv7-a
+ * instruction modified both arguments in place. AArch64 and SIMD128 emit
+ * two instructions from this intrinsic.
+ *
+ * [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ]
+ * [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ]
+ */
+ uint32x4x2_t unzipped = vuzpq_u32(
+ vreinterpretq_u32_u64(data_key_1),
+ vreinterpretq_u32_u64(data_key_2)
+ );
+ /* data_key_lo = data_key & 0xFFFFFFFF */
+ uint32x4_t data_key_lo = unzipped.val[0];
+ /* data_key_hi = data_key >> 32 */
+ uint32x4_t data_key_hi = unzipped.val[1];
+ /*
+ * Then, we can split the vectors horizontally and multiply which, as for most
+ * widening intrinsics, have a variant that works on both high half vectors
+ * for free on AArch64. A similar instruction is available on SIMD128.
+ *
+ * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi
+ */
+ uint64x2_t sum_1 = XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi);
+ uint64x2_t sum_2 = XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi);
+ /*
+ * Clang reorders
+ * a += b * c; // umlal swap.2d, dkl.2s, dkh.2s
+ * c += a; // add acc.2d, acc.2d, swap.2d
+ * to
+ * c += a; // add acc.2d, acc.2d, swap.2d
+ * c += b * c; // umlal acc.2d, dkl.2s, dkh.2s
+ *
+ * While it would make sense in theory since the addition is faster,
+ * for reasons likely related to umlal being limited to certain NEON
+ * pipelines, this is worse. A compiler guard fixes this.
+ */
+ XXH_COMPILER_GUARD_CLANG_NEON(sum_1);
+ XXH_COMPILER_GUARD_CLANG_NEON(sum_2);
+ /* xacc[i] = acc_vec + sum; */
+ xacc[i] = vaddq_u64(xacc[i], sum_1);
+ xacc[i+1] = vaddq_u64(xacc[i+1], sum_2);
+ }
+ /* Operate on the remaining NEON lanes 2 at a time. */
+ for (; i < XXH3_NEON_LANES / 2; i++) {
+ /* data_vec = xinput[i]; */
+ uint64x2_t data_vec = XXH_vld1q_u64(xinput + (i * 16));
+ /* key_vec = xsecret[i]; */
+ uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16));
+ /* acc_vec_2 = swap(data_vec) */
+ uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1);
+ /* data_key = data_vec ^ key_vec; */
+ uint64x2_t data_key = veorq_u64(data_vec, key_vec);
+ /* For two lanes, just use VMOVN and VSHRN. */
+ /* data_key_lo = data_key & 0xFFFFFFFF; */
+ uint32x2_t data_key_lo = vmovn_u64(data_key);
+ /* data_key_hi = data_key >> 32; */
+ uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32);
+ /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */
+ uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi);
+ /* Same Clang workaround as before */
+ XXH_COMPILER_GUARD_CLANG_NEON(sum);
+ /* xacc[i] = acc_vec + sum; */
+ xacc[i] = vaddq_u64 (xacc[i], sum);
}
}
}
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(neon)
XXH_FORCE_INLINE void
XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
- { uint64x2_t* xacc = (uint64x2_t*) acc;
+ { xxh_aliasing_uint64x2_t* xacc = (xxh_aliasing_uint64x2_t*) acc;
uint8_t const* xsecret = (uint8_t const*) secret;
- uint32x2_t prime = vdup_n_u32 (XXH_PRIME32_1);
size_t i;
- for (i=0; i < XXH_STRIPE_LEN/sizeof(uint64x2_t); i++) {
+ /* WASM uses operator overloads and doesn't need these. */
+#ifndef __wasm_simd128__
+ /* { prime32_1, prime32_1 } */
+ uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1);
+ /* { 0, prime32_1, 0, prime32_1 } */
+ uint32x4_t const kPrimeHi = vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32));
+#endif
+
+ /* AArch64 uses both scalar and neon at the same time */
+ for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+ XXH3_scalarScrambleRound(acc, secret, i);
+ }
+ for (i=0; i < XXH3_NEON_LANES / 2; i++) {
/* xacc[i] ^= (xacc[i] >> 47); */
uint64x2_t acc_vec = xacc[i];
- uint64x2_t shifted = vshrq_n_u64 (acc_vec, 47);
- uint64x2_t data_vec = veorq_u64 (acc_vec, shifted);
+ uint64x2_t shifted = vshrq_n_u64(acc_vec, 47);
+ uint64x2_t data_vec = veorq_u64(acc_vec, shifted);
/* xacc[i] ^= xsecret[i]; */
- uint8x16_t key_vec = vld1q_u8 (xsecret + (i * 16));
- uint64x2_t data_key = veorq_u64 (data_vec, vreinterpretq_u64_u8(key_vec));
-
+ uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16));
+ uint64x2_t data_key = veorq_u64(data_vec, key_vec);
/* xacc[i] *= XXH_PRIME32_1 */
- uint32x2_t data_key_lo, data_key_hi;
- /* data_key_lo = (uint32x2_t) (xacc[i] & 0xFFFFFFFF);
- * data_key_hi = (uint32x2_t) (xacc[i] >> 32);
- * xacc[i] = UNDEFINED; */
- XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
- { /*
- * prod_hi = (data_key >> 32) * XXH_PRIME32_1;
- *
- * Avoid vmul_u32 + vshll_n_u32 since Clang 6 and 7 will
- * incorrectly "optimize" this:
- * tmp = vmul_u32(vmovn_u64(a), vmovn_u64(b));
- * shifted = vshll_n_u32(tmp, 32);
- * to this:
- * tmp = "vmulq_u64"(a, b); // no such thing!
- * shifted = vshlq_n_u64(tmp, 32);
- *
- * However, unlike SSE, Clang lacks a 64-bit multiply routine
- * for NEON, and it scalarizes two 64-bit multiplies instead.
- *
- * vmull_u32 has the same timing as vmul_u32, and it avoids
- * this bug completely.
- * See https://bugs.llvm.org/show_bug.cgi?id=39967
- */
- uint64x2_t prod_hi = vmull_u32 (data_key_hi, prime);
- /* xacc[i] = prod_hi << 32; */
- xacc[i] = vshlq_n_u64(prod_hi, 32);
- /* xacc[i] += (prod_hi & 0xFFFFFFFF) * XXH_PRIME32_1; */
- xacc[i] = vmlal_u32(xacc[i], data_key_lo, prime);
- }
- } }
+#ifdef __wasm_simd128__
+ /* SIMD128 has multiply by u64x2, use it instead of expanding and scalarizing */
+ xacc[i] = data_key * XXH_PRIME32_1;
+#else
+ /*
+ * Expanded version with portable NEON intrinsics
+ *
+ * lo(x) * lo(y) + (hi(x) * lo(y) << 32)
+ *
+ * prod_hi = hi(data_key) * lo(prime) << 32
+ *
+ * Since we only need 32 bits of this multiply a trick can be used, reinterpreting the vector
+ * as a uint32x4_t and multiplying by { 0, prime, 0, prime } to cancel out the unwanted bits
+ * and avoid the shift.
+ */
+ uint32x4_t prod_hi = vmulq_u32 (vreinterpretq_u32_u64(data_key), kPrimeHi);
+ /* Extract low bits for vmlal_u32 */
+ uint32x2_t data_key_lo = vmovn_u64(data_key);
+ /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */
+ xacc[i] = vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo);
+#endif
+ }
+ }
}
-
#endif
#if (XXH_VECTOR == XXH_VSX)
@@ -4128,23 +5110,23 @@ XXH3_accumulate_512_vsx( void* XXH_RESTRICT acc,
const void* XXH_RESTRICT secret)
{
/* presumed aligned */
- unsigned long long* const xacc = (unsigned long long*) acc;
- xxh_u64x2 const* const xinput = (xxh_u64x2 const*) input; /* no alignment restriction */
- xxh_u64x2 const* const xsecret = (xxh_u64x2 const*) secret; /* no alignment restriction */
+ xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
+ xxh_u8 const* const xinput = (xxh_u8 const*) input; /* no alignment restriction */
+ xxh_u8 const* const xsecret = (xxh_u8 const*) secret; /* no alignment restriction */
xxh_u64x2 const v32 = { 32, 32 };
size_t i;
for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
/* data_vec = xinput[i]; */
- xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + i);
+ xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16*i);
/* key_vec = xsecret[i]; */
- xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + i);
+ xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16*i);
xxh_u64x2 const data_key = data_vec ^ key_vec;
/* shuffled = (data_key << 32) | (data_key >> 32); */
xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
/* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
xxh_u64x2 const product = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
/* acc_vec = xacc[i]; */
- xxh_u64x2 acc_vec = vec_xl(0, xacc + 2 * i);
+ xxh_u64x2 acc_vec = xacc[i];
acc_vec += product;
/* swap high and low halves */
@@ -4153,18 +5135,18 @@ XXH3_accumulate_512_vsx( void* XXH_RESTRICT acc,
#else
acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
#endif
- /* xacc[i] = acc_vec; */
- vec_xst(acc_vec, 0, xacc + 2 * i);
+ xacc[i] = acc_vec;
}
}
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx)
XXH_FORCE_INLINE void
XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
- { xxh_u64x2* const xacc = (xxh_u64x2*) acc;
- const xxh_u64x2* const xsecret = (const xxh_u64x2*) secret;
+ { xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
+ const xxh_u8* const xsecret = (const xxh_u8*) secret;
/* constants */
xxh_u64x2 const v32 = { 32, 32 };
xxh_u64x2 const v47 = { 47, 47 };
@@ -4176,7 +5158,7 @@ XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
/* xacc[i] ^= xsecret[i]; */
- xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + i);
+ xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16*i);
xxh_u64x2 const data_key = data_vec ^ key_vec;
/* xacc[i] *= XXH_PRIME32_1 */
@@ -4190,40 +5172,233 @@ XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
#endif
+#if (XXH_VECTOR == XXH_SVE)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_sve( void* XXH_RESTRICT acc,
+ const void* XXH_RESTRICT input,
+ const void* XXH_RESTRICT secret)
+{
+ uint64_t *xacc = (uint64_t *)acc;
+ const uint64_t *xinput = (const uint64_t *)(const void *)input;
+ const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
+ svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
+ uint64_t element_count = svcntd();
+ if (element_count >= 8) {
+ svbool_t mask = svptrue_pat_b64(SV_VL8);
+ svuint64_t vacc = svld1_u64(mask, xacc);
+ ACCRND(vacc, 0);
+ svst1_u64(mask, xacc, vacc);
+ } else if (element_count == 2) { /* sve128 */
+ svbool_t mask = svptrue_pat_b64(SV_VL2);
+ svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+ svuint64_t acc1 = svld1_u64(mask, xacc + 2);
+ svuint64_t acc2 = svld1_u64(mask, xacc + 4);
+ svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+ ACCRND(acc0, 0);
+ ACCRND(acc1, 2);
+ ACCRND(acc2, 4);
+ ACCRND(acc3, 6);
+ svst1_u64(mask, xacc + 0, acc0);
+ svst1_u64(mask, xacc + 2, acc1);
+ svst1_u64(mask, xacc + 4, acc2);
+ svst1_u64(mask, xacc + 6, acc3);
+ } else {
+ svbool_t mask = svptrue_pat_b64(SV_VL4);
+ svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+ svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+ ACCRND(acc0, 0);
+ ACCRND(acc1, 4);
+ svst1_u64(mask, xacc + 0, acc0);
+ svst1_u64(mask, xacc + 4, acc1);
+ }
+}
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_sve(xxh_u64* XXH_RESTRICT acc,
+ const xxh_u8* XXH_RESTRICT input,
+ const xxh_u8* XXH_RESTRICT secret,
+ size_t nbStripes)
+{
+ if (nbStripes != 0) {
+ uint64_t *xacc = (uint64_t *)acc;
+ const uint64_t *xinput = (const uint64_t *)(const void *)input;
+ const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
+ svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
+ uint64_t element_count = svcntd();
+ if (element_count >= 8) {
+ svbool_t mask = svptrue_pat_b64(SV_VL8);
+ svuint64_t vacc = svld1_u64(mask, xacc + 0);
+ do {
+ /* svprfd(svbool_t, void *, enum svfprop); */
+ svprfd(mask, xinput + 128, SV_PLDL1STRM);
+ ACCRND(vacc, 0);
+ xinput += 8;
+ xsecret += 1;
+ nbStripes--;
+ } while (nbStripes != 0);
+
+ svst1_u64(mask, xacc + 0, vacc);
+ } else if (element_count == 2) { /* sve128 */
+ svbool_t mask = svptrue_pat_b64(SV_VL2);
+ svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+ svuint64_t acc1 = svld1_u64(mask, xacc + 2);
+ svuint64_t acc2 = svld1_u64(mask, xacc + 4);
+ svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+ do {
+ svprfd(mask, xinput + 128, SV_PLDL1STRM);
+ ACCRND(acc0, 0);
+ ACCRND(acc1, 2);
+ ACCRND(acc2, 4);
+ ACCRND(acc3, 6);
+ xinput += 8;
+ xsecret += 1;
+ nbStripes--;
+ } while (nbStripes != 0);
+
+ svst1_u64(mask, xacc + 0, acc0);
+ svst1_u64(mask, xacc + 2, acc1);
+ svst1_u64(mask, xacc + 4, acc2);
+ svst1_u64(mask, xacc + 6, acc3);
+ } else {
+ svbool_t mask = svptrue_pat_b64(SV_VL4);
+ svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+ svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+ do {
+ svprfd(mask, xinput + 128, SV_PLDL1STRM);
+ ACCRND(acc0, 0);
+ ACCRND(acc1, 4);
+ xinput += 8;
+ xsecret += 1;
+ nbStripes--;
+ } while (nbStripes != 0);
+
+ svst1_u64(mask, xacc + 0, acc0);
+ svst1_u64(mask, xacc + 4, acc1);
+ }
+ }
+}
+
+#endif
+
/* scalar variants - universal */
+#if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__))
+/*
+ * In XXH3_scalarRound(), GCC and Clang have a similar codegen issue, where they
+ * emit an excess mask and a full 64-bit multiply-add (MADD X-form).
+ *
+ * While this might not seem like much, as AArch64 is a 64-bit architecture, only
+ * big Cortex designs have a full 64-bit multiplier.
+ *
+ * On the little cores, the smaller 32-bit multiplier is used, and full 64-bit
+ * multiplies expand to 2-3 multiplies in microcode. This has a major penalty
+ * of up to 4 latency cycles and 2 stall cycles in the multiply pipeline.
+ *
+ * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL) which does
+ * not have this penalty and does the mask automatically.
+ */
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
+{
+ xxh_u64 ret;
+ /* note: %x = 64-bit register, %w = 32-bit register */
+ __asm__("umaddl %x0, %w1, %w2, %x3" : "=r" (ret) : "r" (lhs), "r" (rhs), "r" (acc));
+ return ret;
+}
+#else
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
+{
+ return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc;
+}
+#endif
+
+/*!
+ * @internal
+ * @brief Scalar round for @ref XXH3_accumulate_512_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc,
+ void const* XXH_RESTRICT input,
+ void const* XXH_RESTRICT secret,
+ size_t lane)
+{
+ xxh_u64* xacc = (xxh_u64*) acc;
+ xxh_u8 const* xinput = (xxh_u8 const*) input;
+ xxh_u8 const* xsecret = (xxh_u8 const*) secret;
+ XXH_ASSERT(lane < XXH_ACC_NB);
+ XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
+ {
+ xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
+ xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
+ xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
+ xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */, data_key >> 32, xacc[lane]);
+ }
+}
+
+/*!
+ * @internal
+ * @brief Processes a 64 byte block of data using the scalar path.
+ */
XXH_FORCE_INLINE void
XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
- xxh_u64* const xacc = (xxh_u64*) acc; /* presumed aligned */
- const xxh_u8* const xinput = (const xxh_u8*) input; /* no alignment restriction */
- const xxh_u8* const xsecret = (const xxh_u8*) secret; /* no alignment restriction */
size_t i;
- XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
+ /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on ARMv6. */
+#if defined(__GNUC__) && !defined(__clang__) \
+ && (defined(__arm__) || defined(__thumb2__)) \
+ && defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes bytes */ \
+ && XXH_SIZE_OPT <= 0
+# pragma GCC unroll 8
+#endif
for (i=0; i < XXH_ACC_NB; i++) {
- xxh_u64 const data_val = XXH_readLE64(xinput + 8*i);
- xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + i*8);
- xacc[i ^ 1] += data_val; /* swap adjacent lanes */
- xacc[i] += XXH_mult32to64(data_key & 0xFFFFFFFF, data_key >> 32);
+ XXH3_scalarRound(acc, input, secret, i);
}
}
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(scalar)
+/*!
+ * @internal
+ * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
XXH_FORCE_INLINE void
-XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+ void const* XXH_RESTRICT secret,
+ size_t lane)
{
xxh_u64* const xacc = (xxh_u64*) acc; /* presumed aligned */
const xxh_u8* const xsecret = (const xxh_u8*) secret; /* no alignment restriction */
- size_t i;
XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
- for (i=0; i < XXH_ACC_NB; i++) {
- xxh_u64 const key64 = XXH_readLE64(xsecret + 8*i);
- xxh_u64 acc64 = xacc[i];
+ XXH_ASSERT(lane < XXH_ACC_NB);
+ {
+ xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
+ xxh_u64 acc64 = xacc[lane];
acc64 = XXH_xorshift64(acc64, 47);
acc64 ^= key64;
acc64 *= XXH_PRIME32_1;
- xacc[i] = acc64;
+ xacc[lane] = acc64;
+ }
+}
+
+/*!
+ * @internal
+ * @brief Scrambles the accumulators after a large chunk has been read
+ */
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+ size_t i;
+ for (i=0; i < XXH_ACC_NB; i++) {
+ XXH3_scalarScrambleRound(acc, secret, i);
}
}
@@ -4238,15 +5413,16 @@ XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
const xxh_u8* kSecretPtr = XXH3_kSecret;
XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
-#if defined(__clang__) && defined(__aarch64__)
+#if defined(__GNUC__) && defined(__aarch64__)
/*
* UGLY HACK:
- * Clang generates a bunch of MOV/MOVK pairs for aarch64, and they are
+ * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are
* placed sequentially, in order, at the top of the unrolled loop.
*
* While MOVK is great for generating constants (2 cycles for a 64-bit
- * constant compared to 4 cycles for LDR), long MOVK chains stall the
- * integer pipelines:
+ * constant compared to 4 cycles for LDR), it fights for bandwidth with
+ * the arithmetic instructions.
+ *
* I L S
* MOVK
* MOVK
@@ -4255,7 +5431,7 @@ XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
* ADD
* SUB STR
* STR
- * By forcing loads from memory (as the asm line causes Clang to assume
+ * By forcing loads from memory (as the asm line causes the compiler to assume
* that XXH3_kSecretPtr has been changed), the pipelines are used more
* efficiently:
* I L S
@@ -4263,23 +5439,20 @@ XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
* ADD LDR
* SUB STR
* STR
+ *
+ * See XXH3_NEON_LANES for details on the pipsline.
+ *
* XXH3_64bits_withSeed, len == 256, Snapdragon 835
* without hack: 2654.4 MB/s
* with hack: 3202.9 MB/s
*/
XXH_COMPILER_GUARD(kSecretPtr);
#endif
- /*
- * Note: in debug mode, this overrides the asm optimization
- * and Clang will emit MOVK chains again.
- */
- XXH_ASSERT(kSecretPtr == XXH3_kSecret);
-
{ int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
int i;
for (i=0; i < nbRounds; i++) {
/*
- * The asm hack causes Clang to assume that kSecretPtr aliases with
+ * The asm hack causes the compiler to assume that kSecretPtr aliases with
* customSecret, and on aarch64, this prevented LDP from merging two
* loads together for free. Putting the loads together before the stores
* properly generates LDP.
@@ -4292,7 +5465,7 @@ XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
}
-typedef void (*XXH3_f_accumulate_512)(void* XXH_RESTRICT, const void*, const void*);
+typedef void (*XXH3_f_accumulate)(xxh_u64* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, size_t);
typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
@@ -4300,82 +5473,63 @@ typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
#if (XXH_VECTOR == XXH_AVX512)
#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
+#define XXH3_accumulate XXH3_accumulate_avx512
#define XXH3_scrambleAcc XXH3_scrambleAcc_avx512
#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
#elif (XXH_VECTOR == XXH_AVX2)
#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
+#define XXH3_accumulate XXH3_accumulate_avx2
#define XXH3_scrambleAcc XXH3_scrambleAcc_avx2
#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
#elif (XXH_VECTOR == XXH_SSE2)
#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
+#define XXH3_accumulate XXH3_accumulate_sse2
#define XXH3_scrambleAcc XXH3_scrambleAcc_sse2
#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
#elif (XXH_VECTOR == XXH_NEON)
#define XXH3_accumulate_512 XXH3_accumulate_512_neon
+#define XXH3_accumulate XXH3_accumulate_neon
#define XXH3_scrambleAcc XXH3_scrambleAcc_neon
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
#elif (XXH_VECTOR == XXH_VSX)
#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
+#define XXH3_accumulate XXH3_accumulate_vsx
#define XXH3_scrambleAcc XXH3_scrambleAcc_vsx
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+#elif (XXH_VECTOR == XXH_SVE)
+#define XXH3_accumulate_512 XXH3_accumulate_512_sve
+#define XXH3_accumulate XXH3_accumulate_sve
+#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
#else /* scalar */
#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
+#define XXH3_accumulate XXH3_accumulate_scalar
#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
#endif
-
-
-#ifndef XXH_PREFETCH_DIST
-# ifdef __clang__
-# define XXH_PREFETCH_DIST 320
-# else
-# if (XXH_VECTOR == XXH_AVX512)
-# define XXH_PREFETCH_DIST 512
-# else
-# define XXH_PREFETCH_DIST 384
-# endif
-# endif /* __clang__ */
-#endif /* XXH_PREFETCH_DIST */
-
-/*
- * XXH3_accumulate()
- * Loops over XXH3_accumulate_512().
- * Assumption: nbStripes will not overflow the secret size
- */
-XXH_FORCE_INLINE void
-XXH3_accumulate( xxh_u64* XXH_RESTRICT acc,
- const xxh_u8* XXH_RESTRICT input,
- const xxh_u8* XXH_RESTRICT secret,
- size_t nbStripes,
- XXH3_f_accumulate_512 f_acc512)
-{
- size_t n;
- for (n = 0; n < nbStripes; n++ ) {
- const xxh_u8* const in = input + n*XXH_STRIPE_LEN;
- XXH_PREFETCH(in + XXH_PREFETCH_DIST);
- f_acc512(acc,
- in,
- secret + n*XXH_SECRET_CONSUME_RATE);
- }
-}
+#if XXH_SIZE_OPT >= 1 /* don't do SIMD for initialization */
+# undef XXH3_initCustomSecret
+# define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+#endif
XXH_FORCE_INLINE void
XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble)
{
size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
@@ -4387,7 +5541,7 @@ XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
for (n = 0; n < nb_blocks; n++) {
- XXH3_accumulate(acc, input + n*block_len, secret, nbStripesPerBlock, f_acc512);
+ f_acc(acc, input + n*block_len, secret, nbStripesPerBlock);
f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
}
@@ -4395,12 +5549,12 @@ XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
XXH_ASSERT(len > XXH_STRIPE_LEN);
{ size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
- XXH3_accumulate(acc, input + nb_blocks*block_len, secret, nbStripes, f_acc512);
+ f_acc(acc, input + nb_blocks*block_len, secret, nbStripes);
/* last stripe */
{ const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
#define XXH_SECRET_LASTACC_START 7 /* not aligned on 8, last secret is different from acc & scrambler */
- f_acc512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
+ XXH3_accumulate_512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
} }
}
@@ -4445,12 +5599,12 @@ XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secre
XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
const void* XXH_RESTRICT secret, size_t secretSize,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble)
{
XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
- XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc512, f_scramble);
+ XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc, f_scramble);
/* converge into final hash */
XXH_STATIC_ASSERT(sizeof(acc) == 64);
@@ -4464,13 +5618,15 @@ XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
* It's important for performance to transmit secret's size (when it's static)
* so that the compiler can properly optimize the vectorized loop.
* This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
+ * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
+ * breaks -Og, this is XXH_NO_INLINE.
*/
-XXH_FORCE_INLINE XXH64_hash_t
+XXH3_WITH_SECRET_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64;
- return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate_512, XXH3_scrambleAcc);
+ return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate, XXH3_scrambleAcc);
}
/*
@@ -4479,12 +5635,12 @@ XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
* Note that inside this no_inline function, we do inline the internal loop,
* and provide a statically defined secret size to allow optimization of vector loop.
*/
-XXH_NO_INLINE XXH64_hash_t
+XXH_NO_INLINE XXH_PUREF XXH64_hash_t
XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64; (void)secret; (void)secretLen;
- return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate_512, XXH3_scrambleAcc);
+ return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate, XXH3_scrambleAcc);
}
/*
@@ -4501,18 +5657,20 @@ XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
XXH64_hash_t seed,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble,
XXH3_f_initCustomSecret f_initSec)
{
+#if XXH_SIZE_OPT <= 0
if (seed == 0)
return XXH3_hashLong_64b_internal(input, len,
XXH3_kSecret, sizeof(XXH3_kSecret),
- f_acc512, f_scramble);
+ f_acc, f_scramble);
+#endif
{ XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
f_initSec(secret, seed);
return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
- f_acc512, f_scramble);
+ f_acc, f_scramble);
}
}
@@ -4520,12 +5678,12 @@ XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
* It's important for performance that XXH3_hashLong is not inlined.
*/
XXH_NO_INLINE XXH64_hash_t
-XXH3_hashLong_64b_withSeed(const void* input, size_t len,
- XXH64_hash_t seed, const xxh_u8* secret, size_t secretLen)
+XXH3_hashLong_64b_withSeed(const void* XXH_RESTRICT input, size_t len,
+ XXH64_hash_t seed, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
(void)secret; (void)secretLen;
return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
- XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+ XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
}
@@ -4557,37 +5715,37 @@ XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
/* === Public entry point === */
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* input, size_t len)
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length)
{
- return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
+ return XXH3_64bits_internal(input, length, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+XXH3_64bits_withSecret(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize)
{
- return XXH3_64bits_internal(input, len, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
+ return XXH3_64bits_internal(input, length, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed)
{
- return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
+ return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
}
XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
{
- if (len <= XXH3_MIDSIZE_MAX)
- return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
- return XXH3_hashLong_64b_withSecret(input, len, seed, (const xxh_u8*)secret, secretSize);
+ if (length <= XXH3_MIDSIZE_MAX)
+ return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+ return XXH3_hashLong_64b_withSecret(input, length, seed, (const xxh_u8*)secret, secretSize);
}
/* === XXH3 streaming === */
-
+#ifndef XXH_NO_STREAM
/*
* Malloc's a pointer that is always aligned to align.
*
@@ -4611,7 +5769,7 @@ XXH3_64bits_withSecretandSeed(const void* input, size_t len, const void* secret,
*
* Align must be a power of 2 and 8 <= align <= 128.
*/
-static void* XXH_alignedMalloc(size_t s, size_t align)
+static XXH_MALLOCF void* XXH_alignedMalloc(size_t s, size_t align)
{
XXH_ASSERT(align <= 128 && align >= 8); /* range check */
XXH_ASSERT((align & (align-1)) == 0); /* power of 2 */
@@ -4653,7 +5811,13 @@ static void XXH_alignedFree(void* p)
XXH_free(base);
}
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
+/*!
+ * @brief Allocate an @ref XXH3_state_t.
+ *
+ * Must be freed with XXH3_freeState().
+ * @return An allocated XXH3_state_t on success, `NULL` on failure.
+ */
XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
{
XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
@@ -4662,16 +5826,23 @@ XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
return state;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
+/*!
+ * @brief Frees an @ref XXH3_state_t.
+ *
+ * Must be allocated with XXH3_createState().
+ * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ * @return XXH_OK.
+ */
XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
{
XXH_alignedFree(statePtr);
return XXH_OK;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API void
-XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state)
+XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state)
{
XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
}
@@ -4703,18 +5874,18 @@ XXH3_reset_internal(XXH3_state_t* statePtr,
statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset(XXH3_state_t* statePtr)
+XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
{
if (statePtr == NULL) return XXH_ERROR;
XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
return XXH_OK;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
{
if (statePtr == NULL) return XXH_ERROR;
XXH3_reset_internal(statePtr, 0, secret, secretSize);
@@ -4723,9 +5894,9 @@ XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t
return XXH_OK;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
{
if (statePtr == NULL) return XXH_ERROR;
if (seed==0) return XXH3_64bits_reset(statePtr);
@@ -4735,9 +5906,9 @@ XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
return XXH_OK;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed64)
+XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed64)
{
if (statePtr == NULL) return XXH_ERROR;
if (secret == NULL) return XXH_ERROR;
@@ -4747,35 +5918,61 @@ XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret,
return XXH_OK;
}
-/* Note : when XXH3_consumeStripes() is invoked,
- * there must be a guarantee that at least one more byte must be consumed from input
- * so that the function can blindly consume all stripes using the "normal" secret segment */
-XXH_FORCE_INLINE void
+/*!
+ * @internal
+ * @brief Processes a large input for XXH3_update() and XXH3_digest_long().
+ *
+ * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a block.
+ *
+ * @param acc Pointer to the 8 accumulator lanes
+ * @param nbStripesSoFarPtr In/out pointer to the number of leftover stripes in the block*
+ * @param nbStripesPerBlock Number of stripes in a block
+ * @param input Input pointer
+ * @param nbStripes Number of stripes to process
+ * @param secret Secret pointer
+ * @param secretLimit Offset of the last block in @p secret
+ * @param f_acc Pointer to an XXH3_accumulate implementation
+ * @param f_scramble Pointer to an XXH3_scrambleAcc implementation
+ * @return Pointer past the end of @p input after processing
+ */
+XXH_FORCE_INLINE const xxh_u8 *
XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble)
{
- XXH_ASSERT(nbStripes <= nbStripesPerBlock); /* can handle max 1 scramble per invocation */
- XXH_ASSERT(*nbStripesSoFarPtr < nbStripesPerBlock);
- if (nbStripesPerBlock - *nbStripesSoFarPtr <= nbStripes) {
- /* need a scrambling operation */
- size_t const nbStripesToEndofBlock = nbStripesPerBlock - *nbStripesSoFarPtr;
- size_t const nbStripesAfterBlock = nbStripes - nbStripesToEndofBlock;
- XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripesToEndofBlock, f_acc512);
- f_scramble(acc, secret + secretLimit);
- XXH3_accumulate(acc, input + nbStripesToEndofBlock * XXH_STRIPE_LEN, secret, nbStripesAfterBlock, f_acc512);
- *nbStripesSoFarPtr = nbStripesAfterBlock;
- } else {
- XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripes, f_acc512);
+ const xxh_u8* initialSecret = secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE;
+ /* Process full blocks */
+ if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) {
+ /* Process the initial partial block... */
+ size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr;
+
+ do {
+ /* Accumulate and scramble */
+ f_acc(acc, input, initialSecret, nbStripesThisIter);
+ f_scramble(acc, secret + secretLimit);
+ input += nbStripesThisIter * XXH_STRIPE_LEN;
+ nbStripes -= nbStripesThisIter;
+ /* Then continue the loop with the full block size */
+ nbStripesThisIter = nbStripesPerBlock;
+ initialSecret = secret;
+ } while (nbStripes >= nbStripesPerBlock);
+ *nbStripesSoFarPtr = 0;
+ }
+ /* Process a partial block */
+ if (nbStripes > 0) {
+ f_acc(acc, input, initialSecret, nbStripes);
+ input += nbStripes * XXH_STRIPE_LEN;
*nbStripesSoFarPtr += nbStripes;
}
+ /* Return end pointer */
+ return input;
}
#ifndef XXH3_STREAM_USE_STACK
-# ifndef __clang__ /* clang doesn't need additional stack space */
+# if XXH_SIZE_OPT <= 0 && !defined(__clang__) /* clang doesn't need additional stack space */
# define XXH3_STREAM_USE_STACK 1
# endif
#endif
@@ -4785,7 +5982,7 @@ XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
XXH_FORCE_INLINE XXH_errorcode
XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
const xxh_u8* XXH_RESTRICT input, size_t len,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble)
{
if (input==NULL) {
@@ -4801,7 +5998,8 @@ XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
* when operating accumulators directly into state.
* Operating into stack space seems to enable proper optimization.
* clang, on the other hand, doesn't seem to need this trick */
- XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; memcpy(acc, state->acc, sizeof(acc));
+ XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8];
+ XXH_memcpy(acc, state->acc, sizeof(acc));
#else
xxh_u64* XXH_RESTRICT const acc = state->acc;
#endif
@@ -4809,7 +6007,7 @@ XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
/* small input : just fill in tmp buffer */
- if (state->bufferedSize + len <= XXH3_INTERNALBUFFER_SIZE) {
+ if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) {
XXH_memcpy(state->buffer + state->bufferedSize, input, len);
state->bufferedSize += (XXH32_hash_t)len;
return XXH_OK;
@@ -4831,57 +6029,20 @@ XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
&state->nbStripesSoFar, state->nbStripesPerBlock,
state->buffer, XXH3_INTERNALBUFFER_STRIPES,
secret, state->secretLimit,
- f_acc512, f_scramble);
+ f_acc, f_scramble);
state->bufferedSize = 0;
}
XXH_ASSERT(input < bEnd);
-
- /* large input to consume : ingest per full block */
- if ((size_t)(bEnd - input) > state->nbStripesPerBlock * XXH_STRIPE_LEN) {
+ if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
- XXH_ASSERT(state->nbStripesPerBlock >= state->nbStripesSoFar);
- /* join to current block's end */
- { size_t const nbStripesToEnd = state->nbStripesPerBlock - state->nbStripesSoFar;
- XXH_ASSERT(nbStripes <= nbStripes);
- XXH3_accumulate(acc, input, secret + state->nbStripesSoFar * XXH_SECRET_CONSUME_RATE, nbStripesToEnd, f_acc512);
- f_scramble(acc, secret + state->secretLimit);
- state->nbStripesSoFar = 0;
- input += nbStripesToEnd * XXH_STRIPE_LEN;
- nbStripes -= nbStripesToEnd;
- }
- /* consume per entire blocks */
- while(nbStripes >= state->nbStripesPerBlock) {
- XXH3_accumulate(acc, input, secret, state->nbStripesPerBlock, f_acc512);
- f_scramble(acc, secret + state->secretLimit);
- input += state->nbStripesPerBlock * XXH_STRIPE_LEN;
- nbStripes -= state->nbStripesPerBlock;
- }
- /* consume last partial block */
- XXH3_accumulate(acc, input, secret, nbStripes, f_acc512);
- input += nbStripes * XXH_STRIPE_LEN;
- XXH_ASSERT(input < bEnd); /* at least some bytes left */
- state->nbStripesSoFar = nbStripes;
- /* buffer predecessor of last partial stripe */
- XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
- XXH_ASSERT(bEnd - input <= XXH_STRIPE_LEN);
- } else {
- /* content to consume <= block size */
- /* Consume input by a multiple of internal buffer size */
- if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
- const xxh_u8* const limit = bEnd - XXH3_INTERNALBUFFER_SIZE;
- do {
- XXH3_consumeStripes(acc,
+ input = XXH3_consumeStripes(acc,
&state->nbStripesSoFar, state->nbStripesPerBlock,
- input, XXH3_INTERNALBUFFER_STRIPES,
- secret, state->secretLimit,
- f_acc512, f_scramble);
- input += XXH3_INTERNALBUFFER_SIZE;
- } while (input<limit);
- /* buffer predecessor of last partial stripe */
- XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
- }
- }
+ input, nbStripes,
+ secret, state->secretLimit,
+ f_acc, f_scramble);
+ XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+ }
/* Some remaining input (always) : buffer it */
XXH_ASSERT(input < bEnd);
XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
@@ -4890,19 +6051,19 @@ XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
state->bufferedSize = (XXH32_hash_t)(bEnd-input);
#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
/* save stack accumulators into state */
- memcpy(state->acc, acc, sizeof(acc));
+ XXH_memcpy(state->acc, acc, sizeof(acc));
#endif
}
return XXH_OK;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_update(XXH3_state_t* state, const void* input, size_t len)
+XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
{
return XXH3_update(state, (const xxh_u8*)input, len,
- XXH3_accumulate_512, XXH3_scrambleAcc);
+ XXH3_accumulate, XXH3_scrambleAcc);
}
@@ -4911,37 +6072,40 @@ XXH3_digest_long (XXH64_hash_t* acc,
const XXH3_state_t* state,
const unsigned char* secret)
{
+ xxh_u8 lastStripe[XXH_STRIPE_LEN];
+ const xxh_u8* lastStripePtr;
+
/*
* Digest on a local copy. This way, the state remains unaltered, and it can
* continue ingesting more input afterwards.
*/
XXH_memcpy(acc, state->acc, sizeof(state->acc));
if (state->bufferedSize >= XXH_STRIPE_LEN) {
+ /* Consume remaining stripes then point to remaining data in buffer */
size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
size_t nbStripesSoFar = state->nbStripesSoFar;
XXH3_consumeStripes(acc,
&nbStripesSoFar, state->nbStripesPerBlock,
state->buffer, nbStripes,
secret, state->secretLimit,
- XXH3_accumulate_512, XXH3_scrambleAcc);
- /* last stripe */
- XXH3_accumulate_512(acc,
- state->buffer + state->bufferedSize - XXH_STRIPE_LEN,
- secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+ XXH3_accumulate, XXH3_scrambleAcc);
+ lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN;
} else { /* bufferedSize < XXH_STRIPE_LEN */
- xxh_u8 lastStripe[XXH_STRIPE_LEN];
+ /* Copy to temp buffer */
size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
XXH_ASSERT(state->bufferedSize > 0); /* there is always some input buffered */
XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
- XXH3_accumulate_512(acc,
- lastStripe,
- secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+ lastStripePtr = lastStripe;
}
+ /* Last stripe */
+ XXH3_accumulate_512(acc,
+ lastStripePtr,
+ secret + state->secretLimit - XXH_SECRET_LASTACC_START);
}
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
{
const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
if (state->totalLen > XXH3_MIDSIZE_MAX) {
@@ -4957,7 +6121,7 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
secret, state->secretLimit + XXH_STRIPE_LEN);
}
-
+#endif /* !XXH_NO_STREAM */
/* ==========================================
@@ -4977,7 +6141,7 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
* fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
*/
-XXH_FORCE_INLINE XXH128_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
/* A doubled version of 1to3_64b with different constants. */
@@ -5006,7 +6170,7 @@ XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_
}
}
-XXH_FORCE_INLINE XXH128_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
@@ -5026,14 +6190,14 @@ XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_
m128.low64 ^= (m128.high64 >> 3);
m128.low64 = XXH_xorshift64(m128.low64, 35);
- m128.low64 *= 0x9FB21C651E98DF25ULL;
+ m128.low64 *= PRIME_MX2;
m128.low64 = XXH_xorshift64(m128.low64, 28);
m128.high64 = XXH3_avalanche(m128.high64);
return m128;
}
}
-XXH_FORCE_INLINE XXH128_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
@@ -5108,7 +6272,7 @@ XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64
/*
* Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
*/
-XXH_FORCE_INLINE XXH128_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(len <= 16);
@@ -5139,7 +6303,7 @@ XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
}
-XXH_FORCE_INLINE XXH128_hash_t
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
@@ -5150,6 +6314,16 @@ XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
{ XXH128_hash_t acc;
acc.low64 = len * XXH_PRIME64_1;
acc.high64 = 0;
+
+#if XXH_SIZE_OPT >= 1
+ {
+ /* Smaller, but slightly slower. */
+ unsigned int i = (unsigned int)(len - 1) / 32;
+ do {
+ acc = XXH128_mix32B(acc, input+16*i, input+len-16*(i+1), secret+32*i, seed);
+ } while (i-- != 0);
+ }
+#else
if (len > 32) {
if (len > 64) {
if (len > 96) {
@@ -5160,6 +6334,7 @@ XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
}
acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
+#endif
{ XXH128_hash_t h128;
h128.low64 = acc.low64 + acc.high64;
h128.high64 = (acc.low64 * XXH_PRIME64_1)
@@ -5172,7 +6347,7 @@ XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
}
}
-XXH_NO_INLINE XXH128_hash_t
+XXH_NO_INLINE XXH_PUREF XXH128_hash_t
XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
@@ -5181,25 +6356,34 @@ XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
{ XXH128_hash_t acc;
- int const nbRounds = (int)len / 32;
- int i;
+ unsigned i;
acc.low64 = len * XXH_PRIME64_1;
acc.high64 = 0;
- for (i=0; i<4; i++) {
+ /*
+ * We set as `i` as offset + 32. We do this so that unchanged
+ * `len` can be used as upper bound. This reaches a sweet spot
+ * where both x86 and aarch64 get simple agen and good codegen
+ * for the loop.
+ */
+ for (i = 32; i < 160; i += 32) {
acc = XXH128_mix32B(acc,
- input + (32 * i),
- input + (32 * i) + 16,
- secret + (32 * i),
+ input + i - 32,
+ input + i - 16,
+ secret + i - 32,
seed);
}
acc.low64 = XXH3_avalanche(acc.low64);
acc.high64 = XXH3_avalanche(acc.high64);
- XXH_ASSERT(nbRounds >= 4);
- for (i=4 ; i < nbRounds; i++) {
+ /*
+ * NB: `i <= len` will duplicate the last 32-bytes if
+ * len % 32 was zero. This is an unfortunate necessity to keep
+ * the hash result stable.
+ */
+ for (i=160; i <= len; i += 32) {
acc = XXH128_mix32B(acc,
- input + (32 * i),
- input + (32 * i) + 16,
- secret + XXH3_MIDSIZE_STARTOFFSET + (32 * (i - 4)),
+ input + i - 32,
+ input + i - 16,
+ secret + XXH3_MIDSIZE_STARTOFFSET + i - 160,
seed);
}
/* last bytes */
@@ -5207,7 +6391,7 @@ XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
input + len - 16,
input + len - 32,
secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
- 0ULL - seed);
+ (XXH64_hash_t)0 - seed);
{ XXH128_hash_t h128;
h128.low64 = acc.low64 + acc.high64;
@@ -5224,12 +6408,12 @@ XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble)
{
XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
- XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc512, f_scramble);
+ XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc, f_scramble);
/* converge into final hash */
XXH_STATIC_ASSERT(sizeof(acc) == 64);
@@ -5247,47 +6431,50 @@ XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
}
/*
- * It's important for performance that XXH3_hashLong is not inlined.
+ * It's important for performance that XXH3_hashLong() is not inlined.
*/
-XXH_NO_INLINE XXH128_hash_t
+XXH_NO_INLINE XXH_PUREF XXH128_hash_t
XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64,
const void* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64; (void)secret; (void)secretLen;
return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
- XXH3_accumulate_512, XXH3_scrambleAcc);
+ XXH3_accumulate, XXH3_scrambleAcc);
}
/*
- * It's important for performance to pass @secretLen (when it's static)
+ * It's important for performance to pass @p secretLen (when it's static)
* to the compiler, so that it can properly optimize the vectorized loop.
+ *
+ * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
+ * breaks -Og, this is XXH_NO_INLINE.
*/
-XXH_FORCE_INLINE XXH128_hash_t
+XXH3_WITH_SECRET_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64,
const void* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64;
return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
- XXH3_accumulate_512, XXH3_scrambleAcc);
+ XXH3_accumulate, XXH3_scrambleAcc);
}
XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64,
- XXH3_f_accumulate_512 f_acc512,
+ XXH3_f_accumulate f_acc,
XXH3_f_scrambleAcc f_scramble,
XXH3_f_initCustomSecret f_initSec)
{
if (seed64 == 0)
return XXH3_hashLong_128b_internal(input, len,
XXH3_kSecret, sizeof(XXH3_kSecret),
- f_acc512, f_scramble);
+ f_acc, f_scramble);
{ XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
f_initSec(secret, seed64);
return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
- f_acc512, f_scramble);
+ f_acc, f_scramble);
}
}
@@ -5300,7 +6487,7 @@ XXH3_hashLong_128b_withSeed(const void* input, size_t len,
{
(void)secret; (void)secretLen;
return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
- XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+ XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
}
typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
@@ -5330,94 +6517,93 @@ XXH3_128bits_internal(const void* input, size_t len,
/* === Public XXH128 API === */
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* input, size_t len)
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* input, size_t len)
{
return XXH3_128bits_internal(input, len, 0,
XXH3_kSecret, sizeof(XXH3_kSecret),
XXH3_hashLong_128b_default);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+XXH3_128bits_withSecret(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize)
{
return XXH3_128bits_internal(input, len, 0,
(const xxh_u8*)secret, secretSize,
XXH3_hashLong_128b_withSecret);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+XXH3_128bits_withSeed(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
{
return XXH3_128bits_internal(input, len, seed,
XXH3_kSecret, sizeof(XXH3_kSecret),
XXH3_hashLong_128b_withSeed);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
{
if (len <= XXH3_MIDSIZE_MAX)
return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
-XXH128(const void* input, size_t len, XXH64_hash_t seed)
+XXH128(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
{
return XXH3_128bits_withSeed(input, len, seed);
}
/* === XXH3 128-bit streaming === */
-
+#ifndef XXH_NO_STREAM
/*
* All initialization and update functions are identical to 64-bit streaming variant.
* The only difference is the finalization routine.
*/
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset(XXH3_state_t* statePtr)
+XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
{
return XXH3_64bits_reset(statePtr);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
{
return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
{
return XXH3_64bits_reset_withSeed(statePtr, seed);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed)
+XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
{
return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_update(XXH3_state_t* state, const void* input, size_t len)
+XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
{
- return XXH3_update(state, (const xxh_u8*)input, len,
- XXH3_accumulate_512, XXH3_scrambleAcc);
+ return XXH3_64bits_update(state, input, len);
}
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* state)
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
{
const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
if (state->totalLen > XXH3_MIDSIZE_MAX) {
@@ -5441,13 +6627,13 @@ XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* state)
return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
secret, state->secretLimit + XXH_STRIPE_LEN);
}
-
+#endif /* !XXH_NO_STREAM */
/* 128-bit utility functions */
#include <string.h> /* memcmp, memcpy */
/* return : 1 is equal, 0 if different */
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
{
/* note : XXH128_hash_t is compact, it has no padding byte */
@@ -5455,11 +6641,11 @@ XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
}
/* This prototype is compatible with stdlib's qsort().
- * return : >0 if *h128_1 > *h128_2
- * <0 if *h128_1 < *h128_2
- * =0 if *h128_1 == *h128_2 */
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2)
+ * @return : >0 if *h128_1 > *h128_2
+ * <0 if *h128_1 < *h128_2
+ * =0 if *h128_1 == *h128_2 */
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2)
{
XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
@@ -5471,9 +6657,9 @@ XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2)
/*====== Canonical representation ======*/
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API void
-XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash)
+XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash)
{
XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
if (XXH_CPU_LITTLE_ENDIAN) {
@@ -5484,9 +6670,9 @@ XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash)
XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH128_hash_t
-XXH128_hashFromCanonical(const XXH128_canonical_t* src)
+XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src)
{
XXH128_hash_t h;
h.high64 = XXH_readBE64(src);
@@ -5502,26 +6688,34 @@ XXH128_hashFromCanonical(const XXH128_canonical_t* src)
*/
#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
-static void XXH3_combine16(void* dst, XXH128_hash_t h128)
+XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
{
XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API XXH_errorcode
-XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize)
+XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize)
{
+#if (XXH_DEBUGLEVEL >= 1)
XXH_ASSERT(secretBuffer != NULL);
- if (secretBuffer == NULL) return XXH_ERROR;
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+#else
+ /* production mode, assert() are disabled */
+ if (secretBuffer == NULL) return XXH_ERROR;
if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+#endif
+
if (customSeedSize == 0) {
customSeed = XXH3_kSecret;
customSeedSize = XXH_SECRET_DEFAULT_SIZE;
}
+#if (XXH_DEBUGLEVEL >= 1)
XXH_ASSERT(customSeed != NULL);
+#else
if (customSeed == NULL) return XXH_ERROR;
+#endif
/* Fill secretBuffer with a copy of customSeed - repeat as needed */
{ size_t pos = 0;
@@ -5545,9 +6739,9 @@ XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSee
return XXH_OK;
}
-/*! @ingroup xxh3_family */
+/*! @ingroup XXH3_family */
XXH_PUBLIC_API void
-XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed)
+XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed)
{
XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
XXH3_initCustomSecret(secret, seed);
@@ -5560,7 +6754,7 @@ XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed)
/* Pop our optimization override from above */
#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
&& defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
- && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+ && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
# pragma GCC pop_options
#endif
@@ -5575,5 +6769,5 @@ XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed)
#if defined (__cplusplus)
-}
+} /* extern "C" */
#endif
diff --git a/dviware/dvisvgm/src/Font.cpp b/dviware/dvisvgm/src/Font.cpp
index 4645d33073..218dd30f82 100644
--- a/dviware/dvisvgm/src/Font.cpp
+++ b/dviware/dvisvgm/src/Font.cpp
@@ -454,14 +454,17 @@ bool PhysicalFont::getExactGlyphBox (int c, GlyphMetrics &metrics, bool vertical
BoundingBox charbox;
if (!getExactGlyphBox(c, charbox, cb))
return false;
- if ((metrics.wl = -charbox.minX()) < 0) metrics.wl=0;
- if ((metrics.wr = charbox.maxX()) < 0) metrics.wr=0;
- if ((metrics.h = charbox.maxY()) < 0) metrics.h=0;
- if ((metrics.d = -charbox.minY()) < 0) metrics.d=0;
+ metrics.wl = -charbox.minX();
+ metrics.wr = charbox.maxX();
+ metrics.h = charbox.maxY();
+ metrics.d = -charbox.minY();
if (vertical) { // vertical text orientation
if (verticalLayout()) { // font designed for vertical layout?
- metrics.wl = metrics.wr = (metrics.wl+metrics.wr)/2;
- metrics.d += metrics.h;
+ double wl = max(0.0, metrics.wl);
+ double wr = max(0.0, metrics.wr);
+ double h = max(0.0, metrics.h);
+ metrics.wl = metrics.wr = (wl+wr)/2;
+ metrics.d += h;
metrics.h = 0;
}
else {
diff --git a/dviware/dvisvgm/src/Makefile.am b/dviware/dvisvgm/src/Makefile.am
index 4ccab654f0..852f5c8f01 100644
--- a/dviware/dvisvgm/src/Makefile.am
+++ b/dviware/dvisvgm/src/Makefile.am
@@ -119,10 +119,9 @@ libdvisvgm_la_SOURCES = \
PreScanDVIReader.hpp PreScanDVIReader.cpp \
Process.hpp Process.cpp \
psdefs.cpp \
- PSFilter.hpp \
PSInterpreter.hpp PSInterpreter.cpp \
PSPattern.hpp PSPattern.cpp \
- PSPreviewFilter.hpp PSPreviewFilter.cpp \
+ PSPreviewHandler.hpp PSPreviewHandler.cpp \
PsSpecialHandler.hpp PsSpecialHandler.cpp \
RangeMap.hpp RangeMap.cpp \
ShadingPatch.hpp ShadingPatch.cpp \
diff --git a/dviware/dvisvgm/src/Makefile.in b/dviware/dvisvgm/src/Makefile.in
index a576da9bc4..c537e635e2 100644
--- a/dviware/dvisvgm/src/Makefile.in
+++ b/dviware/dvisvgm/src/Makefile.in
@@ -149,7 +149,7 @@ am_libdvisvgm_la_OBJECTS = BasicDVIReader.lo Bezier.lo \
PapersizeSpecialHandler.lo PathClipper.lo PDFHandler.lo \
PDFParser.lo PdfSpecialHandler.lo PDFToSVG.lo \
PreScanDVIReader.lo Process.lo psdefs.lo PSInterpreter.lo \
- PSPattern.lo PSPreviewFilter.lo PsSpecialHandler.lo \
+ PSPattern.lo PSPreviewHandler.lo PsSpecialHandler.lo \
RangeMap.lo ShadingPatch.lo SignalHandler.lo SourceInput.lo \
SpecialManager.lo StreamReader.lo StreamWriter.lo Subfont.lo \
SVGCharHandler.lo SVGCharHandlerFactory.lo \
@@ -217,7 +217,7 @@ am__depfiles_remade = ./$(DEPDIR)/BasicDVIReader.Plo \
./$(DEPDIR)/Opacity.Plo ./$(DEPDIR)/PDFHandler.Plo \
./$(DEPDIR)/PDFParser.Plo ./$(DEPDIR)/PDFToSVG.Plo \
./$(DEPDIR)/PSInterpreter.Plo ./$(DEPDIR)/PSPattern.Plo \
- ./$(DEPDIR)/PSPreviewFilter.Plo ./$(DEPDIR)/PageRanges.Plo \
+ ./$(DEPDIR)/PSPreviewHandler.Plo ./$(DEPDIR)/PageRanges.Plo \
./$(DEPDIR)/PageSize.Plo \
./$(DEPDIR)/PapersizeSpecialHandler.Plo \
./$(DEPDIR)/PathClipper.Plo ./$(DEPDIR)/PdfSpecialHandler.Plo \
@@ -602,10 +602,9 @@ libdvisvgm_la_SOURCES = \
PreScanDVIReader.hpp PreScanDVIReader.cpp \
Process.hpp Process.cpp \
psdefs.cpp \
- PSFilter.hpp \
PSInterpreter.hpp PSInterpreter.cpp \
PSPattern.hpp PSPattern.cpp \
- PSPreviewFilter.hpp PSPreviewFilter.cpp \
+ PSPreviewHandler.hpp PSPreviewHandler.cpp \
PsSpecialHandler.hpp PsSpecialHandler.cpp \
RangeMap.hpp RangeMap.cpp \
ShadingPatch.hpp ShadingPatch.cpp \
@@ -829,7 +828,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PDFToSVG.Plo@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PSInterpreter.Plo@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PSPattern.Plo@am__quote@ # am--include-marker
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PSPreviewFilter.Plo@am__quote@ # am--include-marker
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PSPreviewHandler.Plo@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PageRanges.Plo@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PageSize.Plo@am__quote@ # am--include-marker
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/PapersizeSpecialHandler.Plo@am__quote@ # am--include-marker
@@ -1166,7 +1165,7 @@ distclean: distclean-recursive
-rm -f ./$(DEPDIR)/PDFToSVG.Plo
-rm -f ./$(DEPDIR)/PSInterpreter.Plo
-rm -f ./$(DEPDIR)/PSPattern.Plo
- -rm -f ./$(DEPDIR)/PSPreviewFilter.Plo
+ -rm -f ./$(DEPDIR)/PSPreviewHandler.Plo
-rm -f ./$(DEPDIR)/PageRanges.Plo
-rm -f ./$(DEPDIR)/PageSize.Plo
-rm -f ./$(DEPDIR)/PapersizeSpecialHandler.Plo
@@ -1310,7 +1309,7 @@ maintainer-clean: maintainer-clean-recursive
-rm -f ./$(DEPDIR)/PDFToSVG.Plo
-rm -f ./$(DEPDIR)/PSInterpreter.Plo
-rm -f ./$(DEPDIR)/PSPattern.Plo
- -rm -f ./$(DEPDIR)/PSPreviewFilter.Plo
+ -rm -f ./$(DEPDIR)/PSPreviewHandler.Plo
-rm -f ./$(DEPDIR)/PageRanges.Plo
-rm -f ./$(DEPDIR)/PageSize.Plo
-rm -f ./$(DEPDIR)/PapersizeSpecialHandler.Plo
diff --git a/dviware/dvisvgm/src/PSFilter.hpp b/dviware/dvisvgm/src/PSFilter.hpp
deleted file mode 100644
index 17dcf0d93e..0000000000
--- a/dviware/dvisvgm/src/PSFilter.hpp
+++ /dev/null
@@ -1,41 +0,0 @@
-/*************************************************************************
-** PSFilter.hpp **
-** **
-** This file is part of dvisvgm -- a fast DVI to SVG converter **
-** Copyright (C) 2005-2023 Martin Gieseking <martin.gieseking@uos.de> **
-** **
-** This program is free software; you can redistribute it and/or **
-** modify it under the terms of the GNU General Public License as **
-** published by the Free Software Foundation; either version 3 of **
-** the License, or (at your option) any later version. **
-** **
-** This program is distributed in the hope that it will be useful, but **
-** WITHOUT ANY WARRANTY; without even the implied warranty of **
-** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the **
-** GNU General Public License for more details. **
-** **
-** You should have received a copy of the GNU General Public License **
-** along with this program; if not, see <http://www.gnu.org/licenses/>. **
-*************************************************************************/
-
-#ifndef PSFILTER_HPP
-#define PSFILTER_HPP
-
-class PSInterpreter;
-
-class PSFilter
-{
- public:
- explicit PSFilter (PSInterpreter &psi) : _psi(psi) {}
- virtual ~PSFilter () =default;
- virtual void execute (const char *code, size_t len) =0;
- virtual bool active () const =0;
-
- protected:
- PSInterpreter& psInterpreter () {return _psi;}
-
- private:
- PSInterpreter &_psi;
-};
-
-#endif
diff --git a/dviware/dvisvgm/src/PSInterpreter.cpp b/dviware/dvisvgm/src/PSInterpreter.cpp
index 1c20bb6bb4..cd24ec03b9 100644
--- a/dviware/dvisvgm/src/PSInterpreter.cpp
+++ b/dviware/dvisvgm/src/PSInterpreter.cpp
@@ -27,7 +27,6 @@
#include "FileSystem.hpp"
#include "InputReader.hpp"
#include "Message.hpp"
-#include "PSFilter.hpp"
#include "PSInterpreter.hpp"
#include "SignalHandler.hpp"
#include "utility.hpp"
@@ -67,7 +66,8 @@ void PSInterpreter::init () {
// as of GS 9.56.0. Since dvisvgm relies on the old PS-based interpreter for its
// PDF support, we try to disable the new one.
// https://www.ghostscript.com/doc/9.56.0/Use.htm#PDF_switches
- if (gsrev >= 9560)
+ // As of GS 10.02.0, option NEWPDF has been deprecated and has no effect any longer.
+ if (gsrev >= 9560 && gsrev < 10020)
gsargs.emplace_back("-dNEWPDF=false");
}
_gs.init(gsargs.size(), gsargs.data(), this);
@@ -81,16 +81,6 @@ void PSInterpreter::init () {
}
-/** Sets or replaces the filter applied to the PS code.
- * @param[in] filter the new filter being used
- * @return the previous, replaced filter (nullptr if there was none) */
-PSFilter* PSInterpreter::setFilter (PSFilter *filter) {
- PSFilter *prevFilter = _filter;
- _filter = filter;
- return prevFilter;
-}
-
-
PSActions* PSInterpreter::setActions (PSActions *actions) {
PSActions *old_actions = _actions;
_actions = actions;
@@ -136,15 +126,6 @@ bool PSInterpreter::execute (const char *str, size_t len, bool flush) {
complete = true;
}
- if (_filter && _filter->active()) {
- PSFilter *filter = _filter;
- _filter = nullptr; // prevent recursion when filter calls execute()
- filter->execute(str, len);
- if (filter->active()) // filter still active after execution?
- _filter = filter;
- return complete;
- }
-
// feed Ghostscript with code chunks that are not larger than 64KB
// => see documentation of gsapi_run_string_foo()
const char *p=str;
diff --git a/dviware/dvisvgm/src/PSInterpreter.hpp b/dviware/dvisvgm/src/PSInterpreter.hpp
index 6562e3bf98..c71c5af249 100644
--- a/dviware/dvisvgm/src/PSInterpreter.hpp
+++ b/dviware/dvisvgm/src/PSInterpreter.hpp
@@ -108,7 +108,6 @@ class PSInterpreter {
bool executeRaw (const std::string &str, int n);
bool active () const {return _mode != PS_QUIT;}
void limit (size_t max_bytes) {_bytesToRead = max_bytes;}
- PSFilter* setFilter (PSFilter *filter);
PSActions* setActions (PSActions *actions);
int pdfPageCount (const std::string &fname);
BoundingBox pdfPageBox (const std::string &fname, int pageno);
@@ -134,7 +133,6 @@ class PSInterpreter {
Ghostscript _gs;
Mode _mode; ///< current execution mode
PSActions *_actions=nullptr; ///< actions to be performed
- PSFilter *_filter=nullptr; ///< active filter used to process PS code
size_t _bytesToRead=0; ///< if > 0, maximal number of bytes to be processed by following calls of execute()
std::vector<char> _linebuf;
std::string _errorMessage; ///< text of error message
diff --git a/dviware/dvisvgm/src/PSPreviewFilter.cpp b/dviware/dvisvgm/src/PSPreviewHandler.cpp
index 960812a4e0..833890c3eb 100644
--- a/dviware/dvisvgm/src/PSPreviewFilter.cpp
+++ b/dviware/dvisvgm/src/PSPreviewHandler.cpp
@@ -1,5 +1,5 @@
/*************************************************************************
-** PSPreviewFilter.cpp **
+** PSPreviewHandler.cpp **
** **
** This file is part of dvisvgm -- a fast DVI to SVG converter **
** Copyright (C) 2005-2023 Martin Gieseking <martin.gieseking@uos.de> **
@@ -18,66 +18,53 @@
** along with this program; if not, see <http://www.gnu.org/licenses/>. **
*************************************************************************/
-#include <vector>
-#include "InputBuffer.hpp"
-#include "InputReader.hpp"
#include "PSInterpreter.hpp"
-#include "PSPreviewFilter.hpp"
-#include "SpecialActions.hpp"
-
+#include "PSPreviewHandler.hpp"
using namespace std;
+PSPreviewHandler::PSPreviewHandler (PSInterpreter &psi) : _psi(psi) {
+}
+
+
/** Activates this filter so that the PS code will be redirected through it if
* it's hooked into the PSInterpreter. */
-void PSPreviewFilter::activate () {
- if (_tightpage) // reactivate filter?
- _active = true;
- else { // first activation?
- _tightpage = _active = false;
- // try to retrieve version string of preview package set in the PS header section
- if (psInterpreter().executeRaw("SDict begin currentdict/preview@version known{preview@version}{0}ifelse end", 1))
- _version = psInterpreter().rawData()[0];
- // check if tightpage option was set
- if (_version != "0" && psInterpreter().executeRaw("SDict begin preview@tightpage end", 1)) {
- _tightpage = (psInterpreter().rawData()[0] == "true");
- _active = true;
- }
- }
+void PSPreviewHandler::init () {
+ // try to retrieve version string of preview package set in the PS header section
+ if (_psi.executeRaw("SDict begin currentdict/preview@version known{preview@version}{0}ifelse end", 1))
+ _version = _psi.rawData()[0];
+ // check if tightpage option was set
+ if (_version != "0" && _psi.executeRaw("SDict begin preview@tightpage end", 1))
+ _tightpage = (_psi.rawData()[0] == "true");
_boxExtents.clear();
}
-/** Tries to extract the bounding box information from a chunk of PostScript code.
- * @param[in] code pointer to buffer with PS code to filter
- * @param[in] len number of bytes in buffer */
-void PSPreviewFilter::execute (const char *code, size_t len) {
- // If the "tightpage" option was set in the TeX file, 7 integers representing the
- // extent of the bounding box are present at the begin of each page.
+/** Tries to extract the bounding box information from the PS operand stack.
+ @return true on success. */
+bool PSPreviewHandler::readDataFromStack () {
if (!_tightpage)
- psInterpreter().execute(code, len);
- else {
- // Read bounding box information pushed on the operand stack by the preview package.
- // It consists of 7 values in DVI units:
- // adj_left, adj_bottom, adj_right, adj_top, height, depth, width,
- // where the first 4 values are set by \PreviewBorder or \PreviewBbAdjust. They denote
- // the border adjustments to create additional space around the graphics.
- // The baseline of the tight box extends from (0,0) to (tight_width, 0).
- CharInputBuffer ib(code, len);
- BufferInputReader ir(ib);
- ir.skipSpace();
- int val;
- while (ir.parseInt(val) && _boxExtents.size() <= 7) {
- _boxExtents.push_back(val);
- ir.skipSpace();
- }
- }
- _active = false; // no further processing required
+ return false;
+
+ // Read bounding box information pushed on the operand stack by the preview package.
+ // It consists of 7 values in DVI units:
+ // adj_left, adj_bottom, adj_right, adj_top, height, depth, width,
+ // where the first 4 values are set by \PreviewBorder or \PreviewBbAdjust. They denote
+ // the border adjustments to create additional space around the graphics.
+ // The baseline of the tight box extends from (0,0) to (tight_width, 0).
+ const char *getOperands =
+ "count 7 sub neg dup 0 lt{pop 0}if{0}repeat\n" // ensure 7 operands on the stack
+ "7{dup type dup /integertype eq exch /realtype eq or not{pop 0}if cvi 7 1 roll} repeat"; // ensure integer operands
+ _psi.executeRaw(getOperands, 7);
+ _boxExtents.clear();
+ for (const string &str: _psi.rawData())
+ _boxExtents.push_back(stoi(str));
+ return true;
}
/** Returns the bounding box defined by the preview package. */
-bool PSPreviewFilter::getBoundingBox (BoundingBox &bbox) const {
+bool PSPreviewHandler::getBoundingBox (BoundingBox &bbox) const {
if (_boxExtents.size() < 7)
return false;
const double leftX = _boxExtents[0]*_dvi2bp;
@@ -87,18 +74,18 @@ bool PSPreviewFilter::getBoundingBox (BoundingBox &bbox) const {
/** Returns the box height in PS points, or -1 if no data was found or read yet. */
-double PSPreviewFilter::height () const {
+double PSPreviewHandler::height () const {
return _boxExtents.size() > 4 ? (_boxExtents[4]+_boxExtents[3])*_dvi2bp : -1;
}
/** Returns the box depth in PS points, or -1 if no data was found or read yet. */
-double PSPreviewFilter::depth () const {
+double PSPreviewHandler::depth () const {
return _boxExtents.size() > 5 ? (_boxExtents[5]-_boxExtents[1])*_dvi2bp : -1;
}
/** Returns the box width in PS points, or -1 if no data was found or read yet. */
-double PSPreviewFilter::width () const {
+double PSPreviewHandler::width () const {
return _boxExtents.size() > 6 ? (_boxExtents[6]+_boxExtents[2]-_boxExtents[0])*_dvi2bp : -1;
}
diff --git a/dviware/dvisvgm/src/PSPreviewFilter.hpp b/dviware/dvisvgm/src/PSPreviewHandler.hpp
index 24426529bf..4823aca3bc 100644
--- a/dviware/dvisvgm/src/PSPreviewFilter.hpp
+++ b/dviware/dvisvgm/src/PSPreviewHandler.hpp
@@ -1,5 +1,5 @@
/*************************************************************************
-** PSPreviewFilter.hpp **
+** PSPreviewHandler.hpp **
** **
** This file is part of dvisvgm -- a fast DVI to SVG converter **
** Copyright (C) 2005-2023 Martin Gieseking <martin.gieseking@uos.de> **
@@ -18,26 +18,23 @@
** along with this program; if not, see <http://www.gnu.org/licenses/>. **
*************************************************************************/
-#ifndef PSPREVIEWFILTER_HPP
-#define PSPREVIEWFILTER_HPP
+#ifndef PSPREVIEWHANDLER_HPP
+#define PSPREVIEWHANDLER_HPP
#include <string>
#include <vector>
#include "BoundingBox.hpp"
-#include "PSFilter.hpp"
-class SpecialActions;
+class PSInterpreter;
-class PSPreviewFilter : public PSFilter {
+class PSPreviewHandler {
public:
- explicit PSPreviewFilter (PSInterpreter &psi) : PSFilter(psi) {}
- void activate ();
- void execute (const char *code, size_t len) override;
- bool active () const override {return _active;}
- std::string version () const {return _version;}
- bool tightpage () const {return _tightpage;}
+ explicit PSPreviewHandler (PSInterpreter &psi);
+ void init ();
+ std::string version () const {return _version;}
+ bool tightpage () const {return _tightpage;}
+ bool readDataFromStack ();
void setDviScaleFactor (double dvi2bp) {_dvi2bp = dvi2bp;}
- void assignBorders (BoundingBox &bbox) const;
bool getBoundingBox (BoundingBox &bbox) const;
double height () const;
double depth () const;
@@ -45,10 +42,10 @@ class PSPreviewFilter : public PSFilter {
private:
std::string _version; ///< version string of preview package
- bool _active=false; ///< true if filter is active
bool _tightpage=false; ///< true if tightpage option was given
double _dvi2bp=1.0/65536.0; ///< factor to convert dvi units to PS points
std::vector<int> _boxExtents; ///< bounding box data set by the preview package (in DVI units)
+ PSInterpreter &_psi;
};
#endif
diff --git a/dviware/dvisvgm/src/PsSpecialHandler.cpp b/dviware/dvisvgm/src/PsSpecialHandler.cpp
index cb3a8a6bac..a2dc107215 100644
--- a/dviware/dvisvgm/src/PsSpecialHandler.cpp
+++ b/dviware/dvisvgm/src/PsSpecialHandler.cpp
@@ -28,7 +28,7 @@
#include "Message.hpp"
#include "PathClipper.hpp"
#include "PSPattern.hpp"
-#include "PSPreviewFilter.hpp"
+#include "PSPreviewHandler.hpp"
#include "PsSpecialHandler.hpp"
#include "SpecialActions.hpp"
#include "SVGElement.hpp"
@@ -47,7 +47,7 @@ string PsSpecialHandler::BITMAP_FORMAT;
bool PsSpecialHandler::EMBED_BITMAP_DATA = false;
-PsSpecialHandler::PsSpecialHandler () : _psi(this), _previewFilter(_psi)
+PsSpecialHandler::PsSpecialHandler () : _psi(this), _previewHandler(_psi)
{
_psi.setImageDevice(BITMAP_FORMAT);
}
@@ -123,9 +123,8 @@ void PsSpecialHandler::enterBodySection () {
// Check for information generated by preview.sty. If the tightpage options
// was set, don't execute the bop-hook but allow the PS filter to read
// the bbox data present at the beginning of the page.
- _psi.setFilter(&_previewFilter);
- _previewFilter.activate();
- if (!_previewFilter.tightpage())
+ _previewHandler.init();
+ if (!_previewHandler.tightpage())
_psi.execute("userdict/bop-hook known{bop-hook}if\n", false);
}
}
@@ -155,9 +154,7 @@ void PsSpecialHandler::executeAndSync (istream &is, bool updatePos) {
_actions->getColor().getRGB(r, g, b);
ostringstream oss;
oss << '\n' << r << ' ' << g << ' ' << b << " setrgbcolor ";
- PSFilter *filter = _psi.setFilter(nullptr); // don't apply any filters here
_psi.execute(oss.str(), false);
- _psi.setFilter(filter);
}
_psi.execute(is);
if (updatePos) {
@@ -540,14 +537,15 @@ void PsSpecialHandler::dviBeginPage (unsigned int pageno, SpecialActions &action
void PsSpecialHandler::dviEndPage (unsigned, SpecialActions &actions) {
+ _previewHandler.readDataFromStack();
BoundingBox bbox;
- if (_previewFilter.getBoundingBox(bbox)) { // is there any data written by preview package?
+ if (_previewHandler.getBoundingBox(bbox)) { // is there any data written by preview package?
double w=0, h=0, d=0;
if (actions.getBBoxFormatString() == "preview" || actions.getBBoxFormatString() == "min") {
if (actions.getBBoxFormatString() == "preview") {
- w = max(0.0, _previewFilter.width());
- h = max(0.0, _previewFilter.height());
- d = max(0.0, _previewFilter.depth());
+ w = max(0.0, _previewHandler.width());
+ h = max(0.0, _previewHandler.height());
+ d = max(0.0, _previewHandler.depth());
actions.bbox() = bbox;
Message::mstream() << "\napplying bounding box set by";
}
@@ -557,7 +555,7 @@ void PsSpecialHandler::dviEndPage (unsigned, SpecialActions &actions) {
d = max(0.0, actions.bbox().maxY());
Message::mstream() << "\ncomputing extents based on data set by";
}
- Message::mstream() << " preview package (version " << _previewFilter.version() << ")\n";
+ Message::mstream() << " preview package (version " << _previewHandler.version() << ")\n";
// apply page transformations to box extents
Matrix pagetrans = actions.getPageTransformation();
diff --git a/dviware/dvisvgm/src/PsSpecialHandler.hpp b/dviware/dvisvgm/src/PsSpecialHandler.hpp
index fb44e4bfa4..90285a2b0e 100644
--- a/dviware/dvisvgm/src/PsSpecialHandler.hpp
+++ b/dviware/dvisvgm/src/PsSpecialHandler.hpp
@@ -31,7 +31,7 @@
#include "PSInterpreter.hpp"
#include "Opacity.hpp"
#include "PSPattern.hpp"
-#include "PSPreviewFilter.hpp"
+#include "PSPreviewHandler.hpp"
#include "SpecialHandler.hpp"
class PSPattern;
@@ -97,7 +97,7 @@ class PsSpecialHandler : public SpecialHandler, protected PSActions {
const char* name () const override {return handlerName();}
static const char* handlerName () {return "ps";}
std::vector<const char*> prefixes () const override;
- void setDviScaleFactor (double dvi2bp) override {_previewFilter.setDviScaleFactor(dvi2bp);}
+ void setDviScaleFactor (double dvi2bp) override {_previewHandler.setDviScaleFactor(dvi2bp);}
void enterBodySection ();
PSInterpreter& psInterpreter () {return _psi;}
@@ -180,7 +180,7 @@ class PsSpecialHandler : public SpecialHandler, protected PSActions {
PSInterpreter _psi;
PDFHandler _pdfHandler;
SpecialActions *_actions=nullptr;
- PSPreviewFilter _previewFilter; ///< filter to extract information generated by the preview package
+ PSPreviewHandler _previewHandler; ///< extracts information generated by the preview package
PsSection _psSection=PS_NONE; ///< current section processed (nothing yet, headers, or body specials)
XMLElement *_xmlnode=nullptr; ///< if != 0, created SVG elements are appended to this node
XMLElement *_savenode=nullptr; ///< pointer to temporaryly store _xmlnode
diff --git a/dviware/dvisvgm/src/SVGCharPathHandler.cpp b/dviware/dvisvgm/src/SVGCharPathHandler.cpp
index 2c753d2e5c..dc0c0724bb 100644
--- a/dviware/dvisvgm/src/SVGCharPathHandler.cpp
+++ b/dviware/dvisvgm/src/SVGCharPathHandler.cpp
@@ -95,7 +95,9 @@ void SVGCharPathHandler::appendChar (uint32_t c, double x, double y) {
// the exact location in vertical mode.
GlyphMetrics exact_metrics;
physicalFont->getExactGlyphBox(c, exact_metrics, false, nullptr);
- y += exact_metrics.h + (metrics.d - exact_metrics.h - exact_metrics.d) / 2;
+ double ed = max(0.0, exact_metrics.d);
+ double eh = max(0.0, exact_metrics.h);
+ y += eh + (metrics.d - eh - ed) / 2;
}
}
}
diff --git a/dviware/dvisvgm/src/optimizer/ClipPathReassigner.cpp b/dviware/dvisvgm/src/optimizer/ClipPathReassigner.cpp
index 28d9a64346..2cd59558b0 100644
--- a/dviware/dvisvgm/src/optimizer/ClipPathReassigner.cpp
+++ b/dviware/dvisvgm/src/optimizer/ClipPathReassigner.cpp
@@ -71,6 +71,7 @@ void ClipPathReassigner::execute (XMLElement *defs, XMLElement *context) {
}
// replace clip path references
vector<XMLElement*> descendants;
+ defs->getDescendants(nullptr, "clip-path", descendants);
context->getDescendants(nullptr, "clip-path", descendants);
for (auto &mapEntry : clipPathMap) {
vector<XMLElement*> &identicalClipPathElements = mapEntry.second;
diff --git a/dviware/dvisvgm/src/optimizer/SVGOptimizer.cpp b/dviware/dvisvgm/src/optimizer/SVGOptimizer.cpp
index 889e011571..1554dc9894 100644
--- a/dviware/dvisvgm/src/optimizer/SVGOptimizer.cpp
+++ b/dviware/dvisvgm/src/optimizer/SVGOptimizer.cpp
@@ -21,6 +21,7 @@
#include <algorithm>
#include <array>
#include <map>
+#include <set>
#include "SVGOptimizer.hpp"
#include "../SVGTree.hpp"
@@ -30,7 +31,6 @@
#include "RedundantElementRemover.hpp"
#include "TextSimplifier.hpp"
#include "TransformSimplifier.hpp"
-#include "WSNodeRemover.hpp"
using namespace std;
@@ -51,23 +51,41 @@ SVGOptimizer::SVGOptimizer (SVGTree *svg) : _svg(svg) {
void SVGOptimizer::execute () {
if (!_svg || MODULE_SEQUENCE == "none")
return;
- if (MODULE_SEQUENCE.empty())
- MODULE_SEQUENCE = "remove-clippath"; // default behaviour of previous dvisvgm releases
+
+ vector<string> names = util::split(MODULE_SEQUENCE, ",", true);
+ set<string> removedNames;
+ if (names.empty())
+ names.emplace_back("remove-clippaths"); // default behaviour of previous dvisvgm releases
else {
- if (MODULE_SEQUENCE == "all") {
- for (const auto &entry: _moduleEntries)
- entry.module->execute(_svg->defsNode(), _svg->pageNode());
+ auto it = names.begin();
+ if (*it == "all" || (!it->empty() && it->at(0) == '-')) {
+ if (*it == "all")
+ it = names.erase(it);
+ // add names of all optimizer modules
+ for (const auto &moduleEntry : _moduleEntries) {
+ it = names.insert(it, moduleEntry.modname);
+ ++it;
+ }
+ }
+ }
+ // create sequence of module names to be considered
+ for (auto it=names.begin(); it != names.end();) {
+ if ((*it)[0] == '-') {
+ removedNames.insert(it->substr(1));
+ it = names.erase(it);
}
else {
- vector<string> names = util::split(MODULE_SEQUENCE, ",");
- auto it = find_if(names.begin(), names.end(), [](const string &name) {
- return name == "simplify-transform";
- });
- GroupCollapser::COMBINE_TRANSFORMS = (it != names.end());
- for (const string &name: names) {
- if (OptimizerModule *module = getModule(name))
- module->execute(_svg->defsNode(), _svg->pageNode());
- }
+ if ((*it)[0] == '+')
+ *it = it->substr(1);
+ ++it;
+ }
+ }
+ GroupCollapser::COMBINE_TRANSFORMS = (find(names.begin(), names.end(), "simplify-transform") != names.end());
+ // execute optimizer modules
+ for (const string &name: names) {
+ if (removedNames.find(name) == removedNames.end()) {
+ if (OptimizerModule *module = getModule(name))
+ module->execute(_svg->defsNode(), _svg->pageNode());
}
}
}
@@ -93,11 +111,13 @@ void SVGOptimizer::listModules (ostream &os) const {
* @return true if all names are known */
bool SVGOptimizer::checkModuleString (string &namestr, vector<string> &unknownNames) const {
unknownNames.clear();
- if (namestr.empty() || namestr == "all" || namestr == "none")
+ if (namestr.empty() || namestr == "none" || namestr == "all" || namestr.substr(0,4) == "all,")
return true;
vector<string> givenNames = util::split(namestr, ",");
- for (const string &name : givenNames) {
- if (!getModule(name))
+ for (string name : givenNames) {
+ if (name[0] == '-' || name[0] == '+')
+ name = name.substr(1);
+ if (!name.empty() && !getModule(name))
unknownNames.emplace_back(name);
}
return unknownNames.empty();
diff --git a/dviware/dvisvgm/src/options.xml b/dviware/dvisvgm/src/options.xml
index bd7867970f..70fafe4c37 100644
--- a/dviware/dvisvgm/src/options.xml
+++ b/dviware/dvisvgm/src/options.xml
@@ -21,197 +21,197 @@
<!DOCTYPE cmdline SYSTEM "options.dtd">
<cmdline class="CommandLine">
- <program>
- <usage>[options] dvifile</usage>
- <usage>--eps [options] epsfile</usage>
- <usage>--pdf [options] pdffile</usage>
- <description>This program converts DVI files, as created by TeX/LaTeX, as well as\nEPS and PDF files to the XML-based scalable vector graphics format SVG.</description>
- <copyright>Copyright (C) 2005-2023 Martin Gieseking &lt;martin.gieseking@uos.de></copyright>
- </program>
- <options>
- <section title="Input options">
- <option long="page" short="p">
- <arg type="string" name="ranges" default="1"/>
- <description>choose page(s) to convert</description>
- </option>
- <option long="fontmap" short="m">
- <arg type="string" name="filenames"/>
- <description>evaluate (additional) font map files</description>
- </option>
- <option long="eps" short="E" if="!defined(DISABLE_GS)">
- <description>convert EPS file to SVG</description>
- </option>
- <option long="pdf" short="P" if="!defined(DISABLE_GS)">
- <description>convert PDF file to SVG</description>
- </option>
- <option long="stdin">
- <description>read input file from stdin</description>
- </option>
- </section>
- <section title="SVG output options">
- <option long="bbox" short="b">
- <arg type="string" name="size" default="min"/>
- <description>set size of bounding box</description>
- </option>
- <option long="bitmap-format" short="B" if="!defined(DISABLE_GS)">
- <arg type="string" name="fmt" default="jpeg"/>
- <description>set format used to embed PS/EPS bitmaps</description>
- </option>
- <option long="clipjoin" short="j" if="!defined(DISABLE_GS)">
- <description>compute intersection of clipping paths</description>
- </option>
- <option long="colornames">
- <description>prefer color names to RGB values if possible</description>
- </option>
- <option long="comments">
- <description>add comments with additional information</description>
- </option>
- <option long="currentcolor">
- <arg type="string" name="color" default="#000" optional="yes"/>
- <description>replace given color with 'currentColor'</description>
- </option>
- <option long="embed-bitmaps">
- <description>prevent references to external bitmap files</description>
- </option>
- <option long="font-format" short="f" if="!defined(DISABLE_WOFF)">
- <arg type="string" name="format" default="svg"/>
- <description>set file format of embedded fonts</description>
- </option>
- <option long="grad-overlap" if="!defined(DISABLE_GS)">
- <description>create overlapping color gradient segments</description>
- </option>
- <option long="grad-segments" if="!defined(DISABLE_GS)">
- <arg type="int" name="number" default="20"/>
- <description>number of color gradient segments per row</description>
- </option>
- <option long="grad-simplify" if="!defined(DISABLE_GS)">
- <arg type="double" name="delta" default="0.05"/>
- <description>reduce level of detail for small segments</description>
- </option>
- <option long="linkmark" short="L">
- <arg type="string" name="style" default="box"/>
- <description>select how to mark hyperlinked areas</description>
- </option>
- <option long="optimize" short="O">
- <arg name="modules" type="string" default="all" optional="yes"/>
- <description>perform several SVG optimizations</description>
- </option>
- <option long="output" short="o">
- <arg type="string" name="pattern"/>
- <description>set name pattern of output files</description>
- </option>
- <option long="precision" short="d">
- <arg type="int" name="number" default="0"/>
- <description>set number of decimal points (0-6)</description>
- </option>
- <option long="relative" short="R">
- <description>create relative path commands</description>
- </option>
- <option long="stdout" short="s">
- <description>write SVG output to stdout</description>
- </option>
- <option long="tmpdir">
- <arg type="string" name="path" optional="yes"/>
- <description>set/print the directory for temporary files</description>
- </option>
- <option long="no-fonts" short="n">
- <arg type="int" name="variant" default="0" optional="yes"/>
- <description>draw glyphs by using path elements</description>
- </option>
- <option long="no-merge">
- <description>don't merge adjacent text elements</description>
- </option>
- <option long="no-styles">
- <description>don't use CSS styles to reference fonts</description>
- </option>
- <option long="zip" short="z">
- <arg type="int" name="level" default="9" optional="yes"/>
- <description>create compressed .svgz file</description>
- </option>
- </section>
- <section title="SVG transformations">
- <option long="rotate" short="r">
- <arg type="double" name="angle"/>
- <description>rotate page content clockwise</description>
- </option>
- <option long="scale" short="c">
- <arg type="string" name="sx[,sy]"/>
- <description>scale page content</description>
- </option>
- <option long="translate" short="t">
- <arg type="string" name="tx[,ty]"/>
- <description>shift page content</description>
- </option>
- <option long="transform" short="T">
- <arg type="string" name="commands"/>
- <description>transform page content</description>
- </option>
- <option long="zoom" short="Z">
- <arg type="double" name="factor" default="1.0"/>
- <description>zoom page content</description>
- </option>
- </section>
- <section title="Processing options">
- <option long="cache" short="C">
- <arg type="string" name="dir" optional="yes"/>
- <description>set/print path of cache directory</description>
- </option>
- <option long="debug-glyphs" if="defined(TTFDEBUG)">
- <description>create PS files for all glyphs converted to TTF</description>
- </option>
- <option long="exact-bbox" short="e">
- <description>compute exact glyph bounding boxes</description>
- </option>
- <option long="keep">
- <description>keep temporary files</description>
- </option>
- <option long="libgs" if="!defined(HAVE_LIBGS) &amp;&amp; !defined(DISABLE_GS)">
- <arg name="filename" type="string"/>
- <description>set name of Ghostscript shared library</description>
- </option>
- <option long="mag" short="M">
- <arg type="double" name="factor" default="4"/>
- <description>magnification of Metafont output</description>
- </option>
- <option long="no-mktexmf">
- <description>don't try to create missing fonts</description>
- </option>
- <option long="no-specials" short="S">
- <arg type="string" name="prefixes" optional="yes"/>
- <description>don't process [selected] specials</description>
- </option>
- <option long="page-hashes" short="H">
- <arg type="string" name="params" optional="yes" default="xxh64"/>
- <description>activate usage of page hashes</description>
- </option>
- <option long="trace-all" short="a">
- <arg name="retrace" type="bool" optional="yes" default="false"/>
- <description>trace all glyphs of bitmap fonts</description>
- </option>
- </section>
- <section title="Message options">
- <option long="color">
- <description>colorize messages</description>
- </option>
- <option long="help" short="h">
- <arg name="mode" type="int" optional="yes" default="0"/>
- <description>print this summary of options and exit</description>
- </option>
- <option long="list-specials" short="l">
- <description>print supported special sets and exit</description>
- </option>
- <option long="progress">
- <arg name="delay" type="double" optional="yes" default="0.5"/>
- <description>enable progress indicator</description>
- </option>
- <option long="verbosity" short="v">
- <arg type="unsigned" name="level" default="7"/>
- <description>set verbosity level (0-7)</description>
- </option>
- <option long="version" short="V">
- <arg type="bool" name="extended" optional="yes" default="false"/>
- <description>print version and exit</description>
- </option>
- </section>
- </options>
+ <program>
+ <usage>[options] dvifile</usage>
+ <usage>--eps [options] epsfile</usage>
+ <usage>--pdf [options] pdffile</usage>
+ <description>This program converts DVI files, as created by TeX/LaTeX, as well as\nEPS and PDF files to the XML-based scalable vector graphics format SVG.</description>
+ <copyright>Copyright (C) 2005-2023 Martin Gieseking &lt;martin.gieseking@uos.de></copyright>
+ </program>
+ <options>
+ <section title="Input options">
+ <option long="page" short="p">
+ <arg type="string" name="ranges" default="1"/>
+ <description>choose page(s) to convert</description>
+ </option>
+ <option long="fontmap" short="m">
+ <arg type="string" name="filenames"/>
+ <description>evaluate (additional) font map files</description>
+ </option>
+ <option long="eps" short="E" if="!defined(DISABLE_GS)">
+ <description>convert EPS file to SVG</description>
+ </option>
+ <option long="pdf" short="P" if="!defined(DISABLE_GS)">
+ <description>convert PDF file to SVG</description>
+ </option>
+ <option long="stdin">
+ <description>read input file from stdin</description>
+ </option>
+ </section>
+ <section title="SVG output options">
+ <option long="bbox" short="b">
+ <arg type="string" name="size" default="min"/>
+ <description>set size of bounding box</description>
+ </option>
+ <option long="bitmap-format" short="B" if="!defined(DISABLE_GS)">
+ <arg type="string" name="fmt" default="jpeg"/>
+ <description>set format used to embed PS/EPS bitmaps</description>
+ </option>
+ <option long="clipjoin" short="j" if="!defined(DISABLE_GS)">
+ <description>compute intersection of clipping paths</description>
+ </option>
+ <option long="colornames">
+ <description>prefer color names to RGB values if possible</description>
+ </option>
+ <option long="comments">
+ <description>add comments with additional information</description>
+ </option>
+ <option long="currentcolor">
+ <arg type="string" name="color" default="#000" optional="yes"/>
+ <description>replace given color with 'currentColor'</description>
+ </option>
+ <option long="embed-bitmaps">
+ <description>prevent references to external bitmap files</description>
+ </option>
+ <option long="font-format" short="f" if="!defined(DISABLE_WOFF)">
+ <arg type="string" name="format" default="svg"/>
+ <description>set file format of embedded fonts</description>
+ </option>
+ <option long="grad-overlap" if="!defined(DISABLE_GS)">
+ <description>create overlapping color gradient segments</description>
+ </option>
+ <option long="grad-segments" if="!defined(DISABLE_GS)">
+ <arg type="int" name="number" default="20"/>
+ <description>number of color gradient segments per row</description>
+ </option>
+ <option long="grad-simplify" if="!defined(DISABLE_GS)">
+ <arg type="double" name="delta" default="0.05"/>
+ <description>reduce level of detail for small segments</description>
+ </option>
+ <option long="linkmark" short="L">
+ <arg type="string" name="style" default="box"/>
+ <description>select how to mark hyperlinked areas</description>
+ </option>
+ <option long="optimize" short="O">
+ <arg name="modules" type="string" default="all" optional="yes"/>
+ <description>perform several SVG optimizations</description>
+ </option>
+ <option long="output" short="o">
+ <arg type="string" name="pattern"/>
+ <description>set name pattern of output files</description>
+ </option>
+ <option long="precision" short="d">
+ <arg type="int" name="number" default="0"/>
+ <description>set number of decimal points (0-6)</description>
+ </option>
+ <option long="relative" short="R">
+ <description>create relative path commands</description>
+ </option>
+ <option long="stdout" short="s">
+ <description>write SVG output to stdout</description>
+ </option>
+ <option long="tmpdir">
+ <arg type="string" name="path" optional="yes"/>
+ <description>set/print the directory for temporary files</description>
+ </option>
+ <option long="no-fonts" short="n">
+ <arg type="int" name="variant" default="0" optional="yes"/>
+ <description>draw glyphs by using path elements</description>
+ </option>
+ <option long="no-merge">
+ <description>don't merge adjacent text elements</description>
+ </option>
+ <option long="no-styles">
+ <description>don't use CSS styles to reference fonts</description>
+ </option>
+ <option long="zip" short="z">
+ <arg type="int" name="level" default="9" optional="yes"/>
+ <description>create compressed .svgz file</description>
+ </option>
+ </section>
+ <section title="SVG transformations">
+ <option long="rotate" short="r">
+ <arg type="double" name="angle"/>
+ <description>rotate page content clockwise</description>
+ </option>
+ <option long="scale" short="c">
+ <arg type="string" name="sx[,sy]"/>
+ <description>scale page content</description>
+ </option>
+ <option long="translate" short="t">
+ <arg type="string" name="tx[,ty]"/>
+ <description>shift page content</description>
+ </option>
+ <option long="transform" short="T">
+ <arg type="string" name="commands"/>
+ <description>transform page content</description>
+ </option>
+ <option long="zoom" short="Z">
+ <arg type="double" name="factor" default="1.0"/>
+ <description>zoom page content</description>
+ </option>
+ </section>
+ <section title="Processing options">
+ <option long="cache" short="C">
+ <arg type="string" name="dir" optional="yes"/>
+ <description>set/print path of cache directory</description>
+ </option>
+ <option long="debug-glyphs" if="defined(TTFDEBUG)">
+ <description>create PS files for all glyphs converted to TTF</description>
+ </option>
+ <option long="exact-bbox" short="e">
+ <description>compute exact glyph bounding boxes</description>
+ </option>
+ <option long="keep">
+ <description>keep temporary files</description>
+ </option>
+ <option long="libgs" if="!defined(HAVE_LIBGS) &amp;&amp; !defined(DISABLE_GS)">
+ <arg name="filename" type="string"/>
+ <description>set name of Ghostscript shared library</description>
+ </option>
+ <option long="mag" short="M">
+ <arg type="double" name="factor" default="4"/>
+ <description>magnification of Metafont output</description>
+ </option>
+ <option long="no-mktexmf">
+ <description>don't try to create missing fonts</description>
+ </option>
+ <option long="no-specials" short="S">
+ <arg type="string" name="prefixes" optional="yes"/>
+ <description>don't process [selected] specials</description>
+ </option>
+ <option long="page-hashes" short="H">
+ <arg type="string" name="params" optional="yes" default="xxh64"/>
+ <description>activate usage of page hashes</description>
+ </option>
+ <option long="trace-all" short="a">
+ <arg name="retrace" type="bool" optional="yes" default="false"/>
+ <description>trace all glyphs of bitmap fonts</description>
+ </option>
+ </section>
+ <section title="Message options">
+ <option long="color">
+ <description>colorize messages</description>
+ </option>
+ <option long="help" short="h">
+ <arg name="mode" type="int" optional="yes" default="0"/>
+ <description>print this summary of options and exit</description>
+ </option>
+ <option long="list-specials" short="l">
+ <description>print supported special sets and exit</description>
+ </option>
+ <option long="progress">
+ <arg name="delay" type="double" optional="yes" default="0.5"/>
+ <description>enable progress indicator</description>
+ </option>
+ <option long="verbosity" short="v">
+ <arg type="unsigned" name="level" default="7"/>
+ <description>set verbosity level (0-7)</description>
+ </option>
+ <option long="version" short="V">
+ <arg type="bool" name="extended" optional="yes" default="false"/>
+ <description>print version and exit</description>
+ </option>
+ </section>
+ </options>
</cmdline>
diff --git a/dviware/dvisvgm/src/psdefs.cpp b/dviware/dvisvgm/src/psdefs.cpp
index e3a8372671..fdbdabba41 100644
--- a/dviware/dvisvgm/src/psdefs.cpp
+++ b/dviware/dvisvgm/src/psdefs.cpp
@@ -35,131 +35,130 @@ const char *PSInterpreter::PSDEFS =
"tshapealpha known not{@SD/.setshapealpha{pop}put}if @SD/.setblendmode known no"
"t{@SD/.setblendmode{pop}put}if @SD/prseq{-1 1{-1 roll =only( )print}for(\\n)pr"
"int}put @SD/prcmd{( )exch(\\ndvi.)3{print}repeat prseq}put @SD/cvxall{{cvx}for"
-"all}put @SD/defpr{[exch[/copy @SD]cvxall 5 -1 roll dup 6 1 roll[/get/exec]cvxa"
-"ll 6 -1 roll dup 7 1 roll 4 -1 roll dup 5 1 roll dup length string cvs/prcmd c"
-"vx]cvx def}put @SD/querypos{{currentpoint}stopped{$error/newerror false put}{2"
-"(querypos)prcmd}ifelse}put @SD/applyscalevals{1 0 dtransform exch dup mul exch"
-" dup mul add sqrt 0 1 dtransform exch dup mul exch dup mul add sqrt 1 0 dtrans"
-"form dup mul exch dup dup mul 3 -1 roll add dup 0 eq{pop}{sqrt div}ifelse 3(ap"
-"plyscalevals)prcmd}put @SD/prpath{{2(moveto)prcmd}{2(lineto)prcmd}{6(curveto)p"
-"rcmd}{0(closepath)prcmd}pathforall}put @SD/nulldevice{@GD/@nulldev true put :n"
-"ulldevice 1 1(setnulldevice)prcmd}put @SD/charpath{/@dodraw false store :charp"
-"ath/@dodraw true store}put @SD/stringwidth{/@dodraw false store :stringwidth/@"
-"dodraw true store}put @SD/show{@dodraw @GD/@nulldev get not and{dup :gsave cur"
-"rentpoint 2{50 mul exch}repeat :newpath moveto 50 50/scale sysexec true charpa"
-"th fill :grestore/@dodraw false store :show/@dodraw true store}{:show}ifelse}p"
-"ut @SD/varxyshow{dup 0 ge{<</chr 3 -1 roll string/prc 5 -1 roll/arr 7 -1 roll/"
-"str 9 -1 roll/idx 0>>begin 0 chr length str length 1 sub{str exch chr length g"
-"etinterval/chr exch store :gsave chr show :grestore currentpoint prc moveto/id"
-"x idx 1 add store}for end}{pop pop show}ifelse}put @SD/xyshow{dup dup type/arr"
-"aytype eq exch length 0 gt and{dup length 2 idiv 2 index length exch idiv}{-1}"
-"ifelse{exch arr idx 2 mul get add exch arr idx 2 mul 1 add get add}exch varxys"
-"how}put @SD/xshow{dup dup type/arraytype eq exch length 0 gt and{dup length 2 "
-"index length exch idiv}{-1}ifelse{exch arr idx get add exch}exch varxyshow}put"
-" @SD/yshow{dup dup type/arraytype eq exch length 0 gt and{dup length 2 index l"
-"ength exch idiv}{-1}ifelse{arr idx get add}exch varxyshow}put @SD/awidthshow{{"
-"1 string dup 0 5 index put :gsave show :grestore pop 0 rmoveto 3 index eq{4 in"
-"dex 4 index rmoveto}if 1 index 1 index rmoveto}exch cshow 5{pop}repeat}put @SD"
-"/widthshow{0 0 3 -1 roll awidthshow}put @SD/ashow{0 0 0 6 3 roll awidthshow}pu"
-"t @SD/newpath{:newpath 1 1(newpath)prcmd}put @SD/stroke{@dodraw @GD/@nulldev g"
-"et not and{prcolor 0 1(newpath)prcmd prpath 0(stroke)prcmd :newpath}{:stroke}i"
-"felse}put @SD/fill{@dodraw @GD/@nulldev get not and{prcolor 0 1(newpath)prcmd "
-"prpath 0(fill)prcmd :newpath}{:fill}ifelse}put @SD/eofill{@dodraw @GD/@nulldev"
-" get not and{prcolor 0 1(newpath)prcmd prpath 0(eofill)prcmd :newpath}{:eofill"
-"}ifelse}put/.fillstroke{:gsave fill :grestore .swapcolors stroke .swapcolors}b"
-"ind def/.eofillstroke{:gsave eofill :grestore .swapcolors stroke .swapcolors}b"
-"ind def @SD/clip{:clip @GD/@nulldev get not{0 1(newpath)prcmd prpath 0(clip)pr"
-"cmd}if}put @SD/eoclip{:eoclip @GD/@nulldev get not{0 1(newpath)prcmd prpath 0("
-"eoclip)prcmd}if}put @SD/shfill{begin currentdict/ShadingType known currentdict"
-"/ColorSpace known and currentdict/DataSource known and currentdict/Function kn"
-"own not and ShadingType 4 ge{DataSource type/arraytype eq{<</DeviceGray 1/Devi"
-"ceRGB 3/DeviceCMYK 4/bgknown currentdict/Background known/bbknown currentdict/"
-"BBox known>>begin currentdict ColorSpace known{ShadingType ColorSpace load bgk"
-"nown{1 Background aload pop}{0}ifelse bbknown{1 BBox aload pop}{0}ifelse Shadi"
-"ngType 5 eq{VerticesPerRow}if DataSource aload length 4 add bgknown{ColorSpace"
-" load add}if bbknown{4 add}if ShadingType 5 eq{1 add}if(shfill)prcmd}if end}if"
-"}if end}put @SD/image{dup type/dicttype eq{dup}{<</Width 6 index/Height 7 inde"
-"x/colorimg false>>}ifelse @execimg}put @SD/colorimage{<<2 index{/Width 2 index"
-" 8 add index/Height 4 index 9 add index}{/Width 8 index/Height 9 index}ifelse/"
-"colorimg true>>@execimg}put/@imgbase(./)def/@imgdevice(jpeg)def/@execimg{@GD/@"
-"imgcnt 2 copy .knownget{1 add}{1}ifelse put begin<</imgdev null/imgid @GD/@img"
-"cnt get/ispng @imgdevice 0 3 getinterval(png)eq dup/suffix exch{(.png)}{(.jpg)"
-"}ifelse/colorimg currentdict/colorimg .knownget dup{pop}if/colordev 1 index cu"
-"rrentcolorspace dup length 1 ne exch 0 get/DeviceGray ne or or>>begin @imgdevi"
-"ce(png)ne @imgdevice(jpeg)ne and{@imgdevice cvn}{colordev{ispng{/png16m}{/jpeg"
-"}ifelse}{ispng{/pnggray}{/jpeggray}ifelse}ifelse}ifelse dup devicedict exch kn"
-"own{:gsave/imgdev exch finddevice def mark/OutputFile @imgbase imgid 20 string"
-" cvs strconcat suffix strconcat/PageSize[Width Height]/UseFastColor true ispng"
-"{@imgdevice(pngmonod)eq{/MinFeatureSize where{pop/MinFeatureSize MinFeatureSiz"
-"e}if}if}{/JPEGQ where{pop/JPEGQ JPEGQ}if}ifelse imgdev putdeviceprops setdevic"
-"e[Width 0 0 Height neg 0 Height]/setmatrix sysexec colorimg{:colorimage}{:imag"
-"e}ifelse/copypage sysexec mark/OutputFile()imgdev putdeviceprops pop :grestore"
-" imgid Width Height 3(image)prcmd}{pop colorimg{:colorimage}{:image}ifelse}ife"
-"lse end end}def/@rect{4 -2 roll moveto exch dup 0 rlineto exch 0 exch rlineto "
-"neg 0 rlineto closepath}bind def/@rectcc{4 -2 roll moveto 2 copy 0 lt exch 0 l"
-"t xor{dup 0 exch rlineto exch 0 rlineto neg 0 exch rlineto}{exch dup 0 rlineto"
-" exch 0 exch rlineto neg 0 rlineto}ifelse closepath}bind def @SD/rectclip{:new"
-"path dup type/arraytype eq{aload length 4 idiv{@rectcc}repeat}{@rectcc}ifelse "
-"clip :newpath}put @SD/rectfill{:gsave :newpath dup type/arraytype eq{aload len"
-"gth 4 idiv{@rectcc}repeat}{@rectcc}ifelse fill :grestore}put @SD/rectstroke{gs"
-"ave :newpath dup type/arraytype eq{aload length 4 idiv{@rect}repeat}{@rect}ife"
-"lse stroke grestore}put false setglobal @SD readonly pop/initclip 0 defpr/clip"
-"path 0 defpr/sysexec{@SD exch get exec}def/adddot{dup length 1 add string dup "
-"0 46 put dup 3 -1 roll 1 exch putinterval}def/setlinewidth{dup/setlinewidth sy"
-"sexec 1(setlinewidth)prcmd}def/setlinecap 1 defpr/setlinejoin 1 defpr/setmiter"
-"limit 1 defpr/setdash{mark 3 1 roll 2 copy/setdash sysexec exch aload length 1"
-" add -1 roll counttomark(setdash)prcmd pop}def/@setpagedevice{pop<<>>/setpaged"
-"evice sysexec matrix setmatrix newpath 0(setpagedevice)prcmd}def/@checknulldev"
-"{@GD/@nulldev get{currentpagedevice maxlength 0 ne{@GD/@nulldev false put 0 1("
-"setnulldevice)prcmd}if}if}def/prcolor{currentcolorspace @setcolorspace current"
-"rgbcolor 3(setrgbcolor)prcmd}def/printgstate{@dodraw @GD/@nulldev get not and{"
-"matrix currentmatrix aload pop 6(setmatrix)prcmd applyscalevals currentlinewid"
-"th 1(setlinewidth)prcmd currentlinecap 1(setlinecap)prcmd currentlinejoin 1(se"
-"tlinejoin)prcmd currentmiterlimit 1(setmiterlimit)prcmd revision dup 952 lt{po"
-"p}{.currentblendmode .setblendmode 952 eq{.currentopacityalpha .setopacityalph"
-"a .currentshapealpha .setshapealpha}{.currentalphaisshape{1}{0}ifelse 1(setalp"
-"haisshape)prcmd .currentstrokeconstantalpha 1(setstrokeconstantalpha)prcmd .cu"
-"rrentfillconstantalpha 1(setfillconstantalpha)prcmd}ifelse}ifelse prcolor curr"
-"entdash mark 3 1 roll exch aload length 1 add -1 roll counttomark(setdash)prcm"
-"d pop}if}def/strconcat{exch dup length 2 index length add string dup dup 4 2 r"
-"oll copy length 4 -1 roll putinterval}def/setgstate{/setgstate sysexec printgs"
-"tate}def/save{@UD begin/@saveID vmstatus pop pop def end :save @saveID 1(save)"
-"prcmd}def/restore{:restore @checknulldev printgstate @UD/@saveID known{@UD beg"
-"in @saveID end}{0}ifelse 1(restore)prcmd}def/gsave 0 defpr/grestore{:grestore "
-"@checknulldev printgstate 0(grestore)prcmd}def/grestoreall{:grestoreall @check"
-"nulldev setstate 0(grestoreall)prcmd}def/rotate{dup type/arraytype ne @dodraw "
-"and{dup 1(rotate)prcmd}if/rotate sysexec applyscalevals}def/scale{dup type/arr"
-"aytype ne @dodraw and{2 copy 2(scale)prcmd}if/scale sysexec applyscalevals}def"
-"/translate{dup type/arraytype ne @dodraw and{2 copy 2(translate)prcmd}if/trans"
-"late sysexec}def/setmatrix{dup/setmatrix sysexec @dodraw{aload pop 6(setmatrix"
-")prcmd applyscalevals}{pop}ifelse}def/initmatrix{matrix setmatrix}def/concat{m"
-"atrix currentmatrix matrix concatmatrix setmatrix}def/makepattern{gsave<</mx 3"
-" -1 roll>>begin<</XUID[1000000 @patcnt]>>copy mx/makepattern sysexec dup begin"
-" PatternType 2 lt{PatternType @patcnt BBox aload pop XStep YStep PaintType mx "
-"aload pop 15(makepattern)prcmd :newpath matrix setmatrix dup PaintProc 0 1(mak"
-"epattern)prcmd @GD/@patcnt @patcnt 1 add put}if end end grestore}def/setpatter"
-"n{dup begin PatternType end 1 eq{begin PaintType 1 eq{XUID aload pop exch pop "
-"1}{:gsave[currentcolorspace aload length -1 roll pop]/setcolorspace sysexec/se"
-"tcolor sysexec XUID aload pop exch pop currentrgbcolor :grestore 4}ifelse(setp"
-"attern)prcmd currentcolorspace 0 get/Pattern ne{[/Pattern currentcolorspace]/s"
-"etcolorspace sysexec}if currentcolorspace @setcolorspace end}{/setpattern syse"
-"xec}ifelse}def/setcolor{dup type/dicttype eq{setpattern}{/setcolor sysexec/cur"
-"rentrgbcolor sysexec setrgbcolor}ifelse}def/setcolorspace{dup/setcolorspace sy"
-"sexec @setcolorspace}def/@setcolorspace{dup type/arraytype eq{0 get}if/Pattern"
-" eq{1}{0}ifelse 1(setcolorspace)prcmd}def/setgray 1 defpr/setcmykcolor 4 defpr"
-"/sethsbcolor 3 defpr/setrgbcolor 3 defpr/.setalphaisshape{@SD/.setalphaisshape"
-" known{dup/.setalphaisshape sysexec}if{1}{0}ifelse 1(setalphaisshape)prcmd}bin"
-"d def/.setfillconstantalpha{@SD/.setfillconstantalpha known{dup/.setfillconsta"
-"ntalpha sysexec}if 1(setfillconstantalpha)prcmd}bind def/.setstrokeconstantalp"
-"ha{@SD/.setstrokeconstantalpha known{dup/.setstrokeconstantalpha sysexec}if 1("
-"setstrokeconstantalpha)prcmd}bind def/.setopacityalpha{false .setalphaisshape "
-"dup .setfillconstantalpha .setstrokeconstantalpha}bind def/.setshapealpha{true"
-" .setalphaisshape dup .setfillconstantalpha .setstrokeconstantalpha}bind def/."
-"setblendmode{dup/.setblendmode sysexec<</Normal 0/Compatible 0/Multiply 1/Scre"
-"en 2/Overlay 3/SoftLight 4/HardLight 5/ColorDodge 6/ColorBurn 7/Darken 8/Light"
-"en 9/Difference 10/Exclusion 11/Hue 12/Saturation 13/Color 14/Luminosity 15/Co"
-"mpatibleOverprint 16>>exch get 1(setblendmode)prcmd}def/@pdfpagecount{(r)file "
-"runpdfbegin pdfpagecount runpdfend}def/@pdfpagebox{(r)file runpdfbegin dup dup"
-" 1 lt exch pdfpagecount gt or{pop}{pdfgetpage/MediaBox pget pop aload pop}ifel"
-"se runpdfend}def DELAYBIND{.bindnow}if ";
+"all}put @SD/defpr{[exch/copy cvx @SD 4 index[/get/exec]cvxall 5 index 3 index "
+"dup length string cvs/prcmd cvx]cvx bind def}put @SD/querypos{{currentpoint}st"
+"opped{$error/newerror false put}{2(querypos)prcmd}ifelse}put @SD/applyscaleval"
+"s{1 0 dtransform exch dup mul exch dup mul add sqrt 0 1 dtransform exch dup mu"
+"l exch dup mul add sqrt 1 0 dtransform dup mul exch dup dup mul 3 -1 roll add "
+"dup 0 eq{pop}{sqrt div}ifelse 3(applyscalevals)prcmd}put @SD/prpath{{2(moveto)"
+"prcmd}{2(lineto)prcmd}{6(curveto)prcmd}{0(closepath)prcmd}pathforall}put @SD/n"
+"ulldevice{@GD/@nulldev true put :nulldevice 1 1(setnulldevice)prcmd}put @SD/ch"
+"arpath{/@dodraw false store :charpath/@dodraw true store}put @SD/stringwidth{/"
+"@dodraw false store :stringwidth/@dodraw true store}put @SD/show{@dodraw @GD/@"
+"nulldev get not and{dup :gsave currentpoint 2{50 mul exch}repeat :newpath move"
+"to 50 50/scale sysexec true charpath fill :grestore/@dodraw false store :show/"
+"@dodraw true store}{:show}ifelse}put @SD/varxyshow{dup 0 ge{<</chr 3 -1 roll s"
+"tring/prc 5 -1 roll/arr 7 -1 roll/str 9 -1 roll/idx 0>>begin 0 chr length str "
+"length 1 sub{str exch chr length getinterval/chr exch store :gsave chr show :g"
+"restore currentpoint prc moveto/idx idx 1 add store}for end}{pop pop show}ifel"
+"se}put @SD/xyshow{dup dup type/arraytype eq exch length 0 gt and{dup length 2 "
+"idiv 2 index length exch idiv}{-1}ifelse{exch arr idx 2 mul get add exch arr i"
+"dx 2 mul 1 add get add}exch varxyshow}put @SD/xshow{dup dup type/arraytype eq "
+"exch length 0 gt and{dup length 2 index length exch idiv}{-1}ifelse{exch arr i"
+"dx get add exch}exch varxyshow}put @SD/yshow{dup dup type/arraytype eq exch le"
+"ngth 0 gt and{dup length 2 index length exch idiv}{-1}ifelse{arr idx get add}e"
+"xch varxyshow}put @SD/awidthshow{{1 string dup 0 5 index put :gsave show :gres"
+"tore pop 0 rmoveto 3 index eq{4 index 4 index rmoveto}if 1 index 1 index rmove"
+"to}exch cshow 5{pop}repeat}put @SD/widthshow{0 0 3 -1 roll awidthshow}put @SD/"
+"ashow{0 0 0 6 3 roll awidthshow}put @SD/newpath{:newpath 1 1(newpath)prcmd}put"
+" @SD/stroke{@dodraw @GD/@nulldev get not and{prcolor 0 1(newpath)prcmd prpath "
+"0(stroke)prcmd :newpath}{:stroke}ifelse}put @SD/fill{@dodraw @GD/@nulldev get "
+"not and{prcolor 0 1(newpath)prcmd prpath 0(fill)prcmd :newpath}{:fill}ifelse}p"
+"ut @SD/eofill{@dodraw @GD/@nulldev get not and{prcolor 0 1(newpath)prcmd prpat"
+"h 0(eofill)prcmd :newpath}{:eofill}ifelse}put/.fillstroke{:gsave fill :grestor"
+"e .swapcolors stroke .swapcolors}bind def/.eofillstroke{:gsave eofill :grestor"
+"e .swapcolors stroke .swapcolors}bind def @SD/clip{:clip @GD/@nulldev get not{"
+"0 1(newpath)prcmd prpath 0(clip)prcmd}if}put @SD/eoclip{:eoclip @GD/@nulldev g"
+"et not{0 1(newpath)prcmd prpath 0(eoclip)prcmd}if}put @SD/shfill{begin current"
+"dict/ShadingType known currentdict/ColorSpace known and currentdict/DataSource"
+" known and currentdict/Function known not and ShadingType 4 ge{DataSource type"
+"/arraytype eq{<</DeviceGray 1/DeviceRGB 3/DeviceCMYK 4/bgknown currentdict/Bac"
+"kground known/bbknown currentdict/BBox known>>begin currentdict ColorSpace kno"
+"wn{ShadingType ColorSpace load bgknown{1 Background aload pop}{0}ifelse bbknow"
+"n{1 BBox aload pop}{0}ifelse ShadingType 5 eq{VerticesPerRow}if DataSource alo"
+"ad length 4 add bgknown{ColorSpace load add}if bbknown{4 add}if ShadingType 5 "
+"eq{1 add}if(shfill)prcmd}if end}if}if end}put @SD/image{dup type/dicttype eq{d"
+"up}{<</Width 6 index/Height 7 index/colorimg false>>}ifelse @execimg}put @SD/c"
+"olorimage{<<2 index{/Width 2 index 8 add index/Height 4 index 9 add index}{/Wi"
+"dth 8 index/Height 9 index}ifelse/colorimg true>>@execimg}put/@imgbase(./)def/"
+"@imgdevice(jpeg)def/@execimg{@GD/@imgcnt 2 copy .knownget{1 add}{1}ifelse put "
+"begin<</imgdev null/imgid @GD/@imgcnt get/ispng @imgdevice 0 3 getinterval(png"
+")eq dup/suffix exch{(.png)}{(.jpg)}ifelse/colorimg currentdict/colorimg .known"
+"get dup{pop}if/colordev 1 index currentcolorspace dup length 1 ne exch 0 get/D"
+"eviceGray ne or or>>begin @imgdevice(png)ne @imgdevice(jpeg)ne and{@imgdevice "
+"cvn}{colordev{ispng{/png16m}{/jpeg}ifelse}{ispng{/pnggray}{/jpeggray}ifelse}if"
+"else}ifelse dup devicedict exch known{:gsave/imgdev exch finddevice def mark/O"
+"utputFile @imgbase imgid 20 string cvs strconcat suffix strconcat/PageSize[Wid"
+"th Height]/UseFastColor true ispng{@imgdevice(pngmonod)eq{/MinFeatureSize wher"
+"e{pop/MinFeatureSize MinFeatureSize}if}if}{/JPEGQ where{pop/JPEGQ JPEGQ}if}ife"
+"lse imgdev putdeviceprops setdevice[Width 0 0 Height neg 0 Height]/setmatrix s"
+"ysexec colorimg{:colorimage}{:image}ifelse/copypage sysexec mark/OutputFile()i"
+"mgdev putdeviceprops pop :grestore imgid Width Height 3(image)prcmd}{pop color"
+"img{:colorimage}{:image}ifelse}ifelse end end}def/@rect{4 -2 roll moveto exch "
+"dup 0 rlineto exch 0 exch rlineto neg 0 rlineto closepath}bind def/@rectcc{4 -"
+"2 roll moveto 2 copy 0 lt exch 0 lt xor{dup 0 exch rlineto exch 0 rlineto neg "
+"0 exch rlineto}{exch dup 0 rlineto exch 0 exch rlineto neg 0 rlineto}ifelse cl"
+"osepath}bind def @SD/rectclip{:newpath dup type/arraytype eq{aload length 4 id"
+"iv{@rectcc}repeat}{@rectcc}ifelse clip :newpath}put @SD/rectfill{:gsave :newpa"
+"th dup type/arraytype eq{aload length 4 idiv{@rectcc}repeat}{@rectcc}ifelse fi"
+"ll :grestore}put @SD/rectstroke{gsave :newpath dup type/arraytype eq{aload len"
+"gth 4 idiv{@rect}repeat}{@rect}ifelse stroke grestore}put false setglobal @SD "
+"readonly pop/initclip 0 defpr/clippath 0 defpr/sysexec{@SD exch get exec}def/a"
+"dddot{dup length 1 add string dup 0 46 put dup 3 -1 roll 1 exch putinterval}de"
+"f/setlinewidth{dup/setlinewidth sysexec 1(setlinewidth)prcmd}def/setlinecap 1 "
+"defpr/setlinejoin 1 defpr/setmiterlimit 1 defpr/setdash{mark 3 1 roll 2 copy/s"
+"etdash sysexec exch aload length 1 add -1 roll counttomark(setdash)prcmd pop}d"
+"ef/@setpagedevice{pop<<>>/setpagedevice sysexec matrix setmatrix newpath 0(set"
+"pagedevice)prcmd}def/@checknulldev{@GD/@nulldev get{currentpagedevice maxlengt"
+"h 0 ne{@GD/@nulldev false put 0 1(setnulldevice)prcmd}if}if}def/prcolor{curren"
+"tcolorspace @setcolorspace currentrgbcolor 3(setrgbcolor)prcmd}def/printgstate"
+"{@dodraw @GD/@nulldev get not and{matrix currentmatrix aload pop 6(setmatrix)p"
+"rcmd applyscalevals currentlinewidth 1(setlinewidth)prcmd currentlinecap 1(set"
+"linecap)prcmd currentlinejoin 1(setlinejoin)prcmd currentmiterlimit 1(setmiter"
+"limit)prcmd revision dup 952 lt{pop}{.currentblendmode .setblendmode 952 eq{.c"
+"urrentopacityalpha .setopacityalpha .currentshapealpha .setshapealpha}{.curren"
+"talphaisshape{1}{0}ifelse 1(setalphaisshape)prcmd .currentstrokeconstantalpha "
+"1(setstrokeconstantalpha)prcmd .currentfillconstantalpha 1(setfillconstantalph"
+"a)prcmd}ifelse}ifelse prcolor currentdash mark 3 1 roll exch aload length 1 ad"
+"d -1 roll counttomark(setdash)prcmd pop}if}def/strconcat{exch dup length 2 ind"
+"ex length add string dup dup 4 2 roll copy length 4 -1 roll putinterval}def/se"
+"tgstate{/setgstate sysexec printgstate}def/save{@UD begin/@saveID vmstatus pop"
+" pop def end :save @saveID 1(save)prcmd}def/restore{:restore @checknulldev pri"
+"ntgstate @UD/@saveID known{@UD begin @saveID end}{0}ifelse 1(restore)prcmd}def"
+"/gsave 0 defpr/grestore{:grestore @checknulldev printgstate 0(grestore)prcmd}d"
+"ef/grestoreall{:grestoreall @checknulldev setstate 0(grestoreall)prcmd}def/rot"
+"ate{dup type/arraytype ne @dodraw and{dup 1(rotate)prcmd}if/rotate sysexec app"
+"lyscalevals}def/scale{dup type/arraytype ne @dodraw and{2 copy 2(scale)prcmd}i"
+"f/scale sysexec applyscalevals}def/translate{dup type/arraytype ne @dodraw and"
+"{2 copy 2(translate)prcmd}if/translate sysexec}def/setmatrix{dup/setmatrix sys"
+"exec @dodraw{aload pop 6(setmatrix)prcmd applyscalevals}{pop}ifelse}def/initma"
+"trix{matrix setmatrix}def/concat{matrix currentmatrix matrix concatmatrix setm"
+"atrix}def/makepattern{gsave<</mx 3 -1 roll>>begin<</XUID[1000000 @patcnt]>>cop"
+"y mx/makepattern sysexec dup begin PatternType 2 lt{PatternType @patcnt BBox a"
+"load pop XStep YStep PaintType mx aload pop 15(makepattern)prcmd :newpath matr"
+"ix setmatrix dup PaintProc 0 1(makepattern)prcmd @GD/@patcnt @patcnt 1 add put"
+"}if end end grestore}def/setpattern{dup begin PatternType end 1 eq{begin Paint"
+"Type 1 eq{XUID aload pop exch pop 1}{:gsave[currentcolorspace aload length -1 "
+"roll pop]/setcolorspace sysexec/setcolor sysexec XUID aload pop exch pop curre"
+"ntrgbcolor :grestore 4}ifelse(setpattern)prcmd currentcolorspace 0 get/Pattern"
+" ne{[/Pattern currentcolorspace]/setcolorspace sysexec}if currentcolorspace @s"
+"etcolorspace end}{/setpattern sysexec}ifelse}def/setcolor{dup type/dicttype eq"
+"{setpattern}{/setcolor sysexec/currentrgbcolor sysexec setrgbcolor}ifelse}def/"
+"setcolorspace{dup/setcolorspace sysexec @setcolorspace}def/@setcolorspace{dup "
+"type/arraytype eq{0 get}if/Pattern eq{1}{0}ifelse 1(setcolorspace)prcmd}def/se"
+"tgray 1 defpr/setcmykcolor 4 defpr/sethsbcolor 3 defpr/setrgbcolor 3 defpr/.se"
+"talphaisshape{@SD/.setalphaisshape known{dup/.setalphaisshape sysexec}if{1}{0}"
+"ifelse 1(setalphaisshape)prcmd}bind def/.setfillconstantalpha{@SD/.setfillcons"
+"tantalpha known{dup/.setfillconstantalpha sysexec}if 1(setfillconstantalpha)pr"
+"cmd}bind def/.setstrokeconstantalpha{@SD/.setstrokeconstantalpha known{dup/.se"
+"tstrokeconstantalpha sysexec}if 1(setstrokeconstantalpha)prcmd}bind def/.setop"
+"acityalpha{false .setalphaisshape dup .setfillconstantalpha .setstrokeconstant"
+"alpha}bind def/.setshapealpha{true .setalphaisshape dup .setfillconstantalpha "
+".setstrokeconstantalpha}bind def/.setblendmode{dup/.setblendmode sysexec<</Nor"
+"mal 0/Compatible 0/Multiply 1/Screen 2/Overlay 3/SoftLight 4/HardLight 5/Color"
+"Dodge 6/ColorBurn 7/Darken 8/Lighten 9/Difference 10/Exclusion 11/Hue 12/Satur"
+"ation 13/Color 14/Luminosity 15/CompatibleOverprint 16>>exch get 1(setblendmod"
+"e)prcmd}def/@pdfpagecount{(r)file runpdfbegin pdfpagecount runpdfend}def/@pdfp"
+"agebox{(r)file runpdfbegin dup dup 1 lt exch pdfpagecount gt or{pop}{pdfgetpag"
+"e/MediaBox pget pop aload pop}ifelse runpdfend}def DELAYBIND{.bindnow}if ";
diff --git a/dviware/dvisvgm/src/utility.cpp b/dviware/dvisvgm/src/utility.cpp
index 8492cbdeed..3ffa376435 100644
--- a/dviware/dvisvgm/src/utility.cpp
+++ b/dviware/dvisvgm/src/utility.cpp
@@ -150,12 +150,13 @@ string util::replace (string str, const string &find, const string &repl) {
* returns the substrings.
* @param[in] str string to split
* @param[in] sep separator to look for
+ * @param[in] allowEmptyResult if true, the result vector is empty if str is empty; otherwise an empty string is inserted
* @return the substrings between the separators */
-vector<string> util::split (const string &str, const string &sep) {
+vector<string> util::split (const string &str, const string &sep, bool allowEmptyResult) {
vector<string> parts;
- if (str.empty() || sep.empty())
+ if ((str.empty() && !allowEmptyResult) || sep.empty())
parts.push_back(str);
- else {
+ else if (!str.empty()) {
string::size_type left=0;
while (left <= str.length()) {
auto right = str.find(sep, left);
diff --git a/dviware/dvisvgm/src/utility.hpp b/dviware/dvisvgm/src/utility.hpp
index ee272e22ec..9058baaa58 100644
--- a/dviware/dvisvgm/src/utility.hpp
+++ b/dviware/dvisvgm/src/utility.hpp
@@ -69,7 +69,7 @@ std::string replace (std::string str, const std::string &find, const std::string
std::string to_string (double val);
std::string mimetype (const std::string &fname);
-std::vector<std::string> split (const std::string &str, const std::string &sep);
+std::vector<std::string> split (const std::string &str, const std::string &sep, bool allowEmptyResult=false);
int ilog10 (int n);
bool read_double (std::istream &is, double &value);
diff --git a/dviware/dvisvgm/tests/UtilityTest.cpp b/dviware/dvisvgm/tests/UtilityTest.cpp
index f5fda4a35b..c118aee8f9 100644
--- a/dviware/dvisvgm/tests/UtilityTest.cpp
+++ b/dviware/dvisvgm/tests/UtilityTest.cpp
@@ -80,6 +80,9 @@ TEST(UtilityTest, split) {
ASSERT_EQ(parts.size(), 1u);
EXPECT_EQ(parts[0], "");
+ parts = split("", ",", true);
+ ASSERT_EQ(parts.size(), 0u);
+
parts = split("abcde", "");
ASSERT_EQ(parts.size(), 1u);
EXPECT_EQ(parts[0], "abcde");
diff --git a/dviware/dvisvgm/tests/normalize.xsl b/dviware/dvisvgm/tests/normalize.xsl
index 89df7ea746..b521173e6a 100644
--- a/dviware/dvisvgm/tests/normalize.xsl
+++ b/dviware/dvisvgm/tests/normalize.xsl
@@ -21,115 +21,115 @@
** of the GNU general public license version 3 or later. **
***********************************************************************-->
<xsl:stylesheet version="1.0"
- xmlns="http://www.w3.org/2000/svg"
- xmlns:svg="http://www.w3.org/2000/svg"
- xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
- xmlns:xlink="http://www.w3.org/1999/xlink"
- xmlns:exsl="http://exslt.org/common"
- xmlns:str="http://exslt.org/strings"
- xmlns:mg="my-namespace"
- extension-element-prefixes="exsl str"
- exclude-result-prefixes="svg xlink exsl str mg">
-
- <xsl:key name="path-by-id" match="/svg:defs/svg:path" use="@id"/>
-
-
- <xsl:variable name="styles-rtf">
- <xsl:for-each select="str:tokenize(/*/svg:style[@type='text/css'], '&#10;')">
- <xsl:sort select="substring-after(., ' ')"/>
- <mg:style new-id="f{position()}" id="{substring-before(substring-after(., '.'), ' ')}">
- <xsl:value-of select="substring-after(., ' ')"/>
- </mg:style>
- </xsl:for-each>
- </xsl:variable>
-
- <xsl:variable name="styles" select="exsl:node-set($styles-rtf)/mg:style"/>
-
- <xsl:variable name="num-paths" select="count(/*/svg:defs/svg:path)"/>
-
- <xsl:variable name="defs-rtf">
- <xsl:for-each select="/*/svg:defs/svg:path">
- <xsl:sort select="@d"/>
- <mg:path id="{@id}" new-id="g{position()}"/>
- </xsl:for-each>
- <xsl:for-each select="/*/svg:defs/svg:use">
- <xsl:sort select="concat(key('path-by-id', substring(@xlink:href, 2))/@d, @transform)"/>
- <mg:use id="{@id}" new-id="g{position()+$num-paths}"/>
- </xsl:for-each>
- </xsl:variable>
-
- <xsl:variable name="defs" select="exsl:node-set($defs-rtf)"/>
-
-
- <xsl:template match="*|@*">
- <xsl:copy>
- <xsl:apply-templates select="@*|node()"/>
- </xsl:copy>
- </xsl:template>
-
-
- <xsl:template match="svg:defs[svg:path]">
- <xsl:copy>
- <xsl:apply-templates select="svg:path">
- <xsl:sort select="@d"/>
- </xsl:apply-templates>
- <xsl:apply-templates select="svg:use">
- <xsl:sort select="$defs/mg:use[@id=current()/@id]/@new-id"/>
- </xsl:apply-templates>
- </xsl:copy>
- </xsl:template>
-
-
- <xsl:template match="svg:defs[svg:font]">
- <xsl:copy>
- <xsl:apply-templates>
- <xsl:sort select="@id"/>
- </xsl:apply-templates>
- </xsl:copy>
- </xsl:template>
-
-
- <xsl:template match="svg:defs/svg:path">
- <path id="{$defs/mg:path[@id=current()/@id]/@new-id}" d="{@d}"/>
- </xsl:template>
-
-
- <xsl:template match="svg:defs/text()"/>
-
-
- <xsl:template match="svg:use">
- <xsl:variable name="href" select="substring(@xlink:href, 2)"/>
- <xsl:copy>
- <xsl:apply-templates select="@*"/>
- <xsl:if test="ancestor::svg:defs">
- <xsl:attribute name="id">
- <xsl:value-of select="$defs/mg:use[@id=current()/@id]/@new-id"/>
- </xsl:attribute>
- </xsl:if>
- <xsl:attribute name="xlink:href">
- <xsl:value-of select="concat('#', $defs/*[@id=$href]/@new-id)"/>
- </xsl:attribute>
- </xsl:copy>
- </xsl:template>
-
-
- <xsl:template match="svg:style[@type='text/css']">
- <style type="text/css">
- <xsl:for-each select="$styles">
- <xsl:value-of select="concat('text.f', position(), ' ', ., '&#10;')"/>
- </xsl:for-each>
- </style>
- </xsl:template>
-
-
- <xsl:template match="svg:text">
- <xsl:copy>
- <xsl:copy-of select="@*"/>
- <xsl:attribute name="class">
- <xsl:value-of select="$styles[@id=current()/@class]/@new-id"/>
- </xsl:attribute>
- <xsl:apply-templates/>
- </xsl:copy>
- </xsl:template>
+ xmlns="http://www.w3.org/2000/svg"
+ xmlns:svg="http://www.w3.org/2000/svg"
+ xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
+ xmlns:xlink="http://www.w3.org/1999/xlink"
+ xmlns:exsl="http://exslt.org/common"
+ xmlns:str="http://exslt.org/strings"
+ xmlns:mg="my-namespace"
+ extension-element-prefixes="exsl str"
+ exclude-result-prefixes="svg xlink exsl str mg">
+
+ <xsl:key name="path-by-id" match="/svg:defs/svg:path" use="@id"/>
+
+
+ <xsl:variable name="styles-rtf">
+ <xsl:for-each select="str:tokenize(/*/svg:style[@type='text/css'], '&#10;')">
+ <xsl:sort select="substring-after(., ' ')"/>
+ <mg:style new-id="f{position()}" id="{substring-before(substring-after(., '.'), ' ')}">
+ <xsl:value-of select="substring-after(., ' ')"/>
+ </mg:style>
+ </xsl:for-each>
+ </xsl:variable>
+
+ <xsl:variable name="styles" select="exsl:node-set($styles-rtf)/mg:style"/>
+
+ <xsl:variable name="num-paths" select="count(/*/svg:defs/svg:path)"/>
+
+ <xsl:variable name="defs-rtf">
+ <xsl:for-each select="/*/svg:defs/svg:path">
+ <xsl:sort select="@d"/>
+ <mg:path id="{@id}" new-id="g{position()}"/>
+ </xsl:for-each>
+ <xsl:for-each select="/*/svg:defs/svg:use">
+ <xsl:sort select="concat(key('path-by-id', substring(@xlink:href, 2))/@d, @transform)"/>
+ <mg:use id="{@id}" new-id="g{position()+$num-paths}"/>
+ </xsl:for-each>
+ </xsl:variable>
+
+ <xsl:variable name="defs" select="exsl:node-set($defs-rtf)"/>
+
+
+ <xsl:template match="*|@*">
+ <xsl:copy>
+ <xsl:apply-templates select="@*|node()"/>
+ </xsl:copy>
+ </xsl:template>
+
+
+ <xsl:template match="svg:defs[svg:path]">
+ <xsl:copy>
+ <xsl:apply-templates select="svg:path">
+ <xsl:sort select="@d"/>
+ </xsl:apply-templates>
+ <xsl:apply-templates select="svg:use">
+ <xsl:sort select="$defs/mg:use[@id=current()/@id]/@new-id"/>
+ </xsl:apply-templates>
+ </xsl:copy>
+ </xsl:template>
+
+
+ <xsl:template match="svg:defs[svg:font]">
+ <xsl:copy>
+ <xsl:apply-templates>
+ <xsl:sort select="@id"/>
+ </xsl:apply-templates>
+ </xsl:copy>
+ </xsl:template>
+
+
+ <xsl:template match="svg:defs/svg:path">
+ <path id="{$defs/mg:path[@id=current()/@id]/@new-id}" d="{@d}"/>
+ </xsl:template>
+
+
+ <xsl:template match="svg:defs/text()"/>
+
+
+ <xsl:template match="svg:use">
+ <xsl:variable name="href" select="substring(@xlink:href, 2)"/>
+ <xsl:copy>
+ <xsl:apply-templates select="@*"/>
+ <xsl:if test="ancestor::svg:defs">
+ <xsl:attribute name="id">
+ <xsl:value-of select="$defs/mg:use[@id=current()/@id]/@new-id"/>
+ </xsl:attribute>
+ </xsl:if>
+ <xsl:attribute name="xlink:href">
+ <xsl:value-of select="concat('#', $defs/*[@id=$href]/@new-id)"/>
+ </xsl:attribute>
+ </xsl:copy>
+ </xsl:template>
+
+
+ <xsl:template match="svg:style[@type='text/css']">
+ <style type="text/css">
+ <xsl:for-each select="$styles">
+ <xsl:value-of select="concat('text.f', position(), ' ', ., '&#10;')"/>
+ </xsl:for-each>
+ </style>
+ </xsl:template>
+
+
+ <xsl:template match="svg:text">
+ <xsl:copy>
+ <xsl:copy-of select="@*"/>
+ <xsl:attribute name="class">
+ <xsl:value-of select="$styles[@id=current()/@class]/@new-id"/>
+ </xsl:attribute>
+ <xsl:apply-templates/>
+ </xsl:copy>
+ </xsl:template>
</xsl:stylesheet>