1 Enabling SSE support 2 3 Copyright (c) 2016 Google, Inc. 4 Written by Mike Klein, Matt Sarett 5 6 This INSTALL file written by Glenn Randers-Pehrson, 2016. 7 8 If you have moved intel_init.c and filter_sse2_intrinsics.c to a different 9 directory, be sure to update the '#include "../../pngpriv.h"' line in both 10 files if necessary to point to the correct relative location of pngpriv.h 11 with respect to the new location of those files. 12 13 To enable SSE support in libpng, follow the instructions in I, II, or III, 14 below: 15 16 I. Using patched "configure" scripts: 17 18 First, apply intel_sse.patch in your build directory. 19 20 patch -i contrib/intel/intel_sse.patch -p1 21 22 Then, if you are not building in a new GIT clone, e.g., in a tar 23 distribution, remove any existing pre-built configure scripts: 24 25 ./configure --enable-maintainer-mode 26 make maintainer-clean 27 ./autogen.sh --maintainer --clean 28 29 Finally, configure libpng with -DPNG_INTEL_SSE in CPPFLAGS: 30 31 ./autogen.sh --maintainer 32 CPPFLAGS="-DPNG_INTEL_SSE" ./configure [options] 33 make CPPFLAGS="-DPNG_INTEL_SSE" [options] 34 make 35 36 II. Using a custom makefile: 37 38 If you are using a custom makefile makefile, you will have to update it 39 manually to include contrib/intel/*.o in the dependencies, and to define 40 PNG_INTEL_SSE. 41 42 III. Using manually updated "configure" scripts: 43 44 If you prefer, manually edit pngpriv.h, configure.ac, and Makefile.am, 45 following the instructions below, then follow the instructions in 46 section II of INSTALL in the main libpng directory, then configure libpng 47 with -DPNG_INTEL_SSE in CPPFLAGS. 48 49 1. Insert the following lines above the copyright line near the top of 50 configure.ac: 51 52 -----------------cut---------------- 53 # Copyright (c) 2016 Google, Inc. 54 # Written by Mike Klein and Matt Sarett 55 # Derived from the ARM supporting code in libpng/configure.ac, which was 56 -----------------cut---------------- 57 58 2. Add the following code to configure.ac under HOST SPECIFIC OPTIONS 59 directly beneath the section for ARM: 60 61 -----------------cut---------------- 62 # INTEL 63 # ===== 64 # 65 # INTEL SSE (SIMD) support. 66 67 AC_ARG_ENABLE([intel-sse], 68 AS_HELP_STRING([[[--enable-intel-sse]]], 69 [Enable Intel SSE optimizations: =no/off, yes/on:] 70 [no/off: disable the optimizations;] 71 [yes/on: enable the optimizations.] 72 [If not specified: determined by the compiler.]), 73 [case "$enableval" in 74 no|off) 75 # disable the default enabling: 76 AC_DEFINE([PNG_INTEL_SSE_OPT], [0], 77 [Disable Intel SSE optimizations]) 78 # Prevent inclusion of the assembler files below: 79 enable_intel_sse=no;; 80 yes|on) 81 AC_DEFINE([PNG_INTEL_SSE_OPT], [1], 82 [Enable Intel SSE optimizations]);; 83 *) 84 AC_MSG_ERROR([--enable-intel-sse=${enable_intel_sse}: invalid value]) 85 esac]) 86 87 # Add Intel specific files to all builds where the host_cpu is Intel ('x86*') 88 # or where Intel optimizations were explicitly requested (this allows a 89 # fallback if a future host CPU does not match 'x86*') 90 AM_CONDITIONAL([PNG_INTEL_SSE], 91 [test "$enable_intel_sse" != 'no' && 92 case "$host_cpu" in 93 i?86|x86_64) :;; 94 *) test "$enable_intel_sse" != '';; 95 esac]) 96 -----------------cut---------------- 97 98 3. Insert the following lines above the copyright line near the top of 99 Makefile.am: 100 101 -----------------cut---------------- 102 # Copyright (c) 2016 Google, Inc. 103 # Written by Mike Klein and Matt Sarett 104 # Derived from the ARM supporting code in libpng/configure.ac, which was 105 -----------------cut---------------- 106 107 4. Add the following code to Makefile.am under HOST SPECIFIC OPTIONS 108 directly beneath the "if PNG_ARM_NEON ... endif" statement: 109 110 -----------------cut---------------- 111 if PNG_INTEL_SSE 112 libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += contrib/intel/intel_init.c\ 113 contrib/intel/filter_sse2_intrinsics.c 114 endif 115 -----------------cut---------------- 116 117 5. Add the following lines to pngpriv.h, following the PNG_ARM_NEON_OPT 118 code: 119 120 -----------------cut---------------- 121 #ifndef PNG_INTEL_SSE_OPT 122 # ifdef PNG_INTEL_SSE 123 /* Only check for SSE if the build configuration has been modified to 124 * enable SSE optimizations. This means that these optimizations will 125 * be off by default. See contrib/intel for more details. 126 */ 127 # if defined(__SSE4_1__) || defined(__AVX__) || defined(__SSSE3__) || \ 128 defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ 129 (defined(_M_IX86_FP) && _M_IX86_FP >= 2) 130 # define PNG_INTEL_SSE_OPT 1 131 # endif 132 # endif 133 #endif 134 135 #if PNG_INTEL_SSE_OPT > 0 136 # ifndef PNG_INTEL_SSE_IMPLEMENTATION 137 # if defined(__SSE4_1__) || defined(__AVX__) 138 /* We are not actually using AVX, but checking for AVX is the best 139 way we can detect SSE4.1 and SSSE3 on MSVC. 140 */ 141 # define PNG_INTEL_SSE_IMPLEMENTATION 3 142 # elif defined(__SSSE3__) 143 # define PNG_INTEL_SSE_IMPLEMENTATION 2 144 # elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ 145 (defined(_M_IX86_FP) && _M_IX86_FP >= 2) 146 # define PNG_INTEL_SSE_IMPLEMENTATION 1 147 # else 148 # define PNG_INTEL_SSE_IMPLEMENTATION 0 149 # endif 150 # endif 151 152 # if PNG_INTEL_SSE_IMPLEMENTATION > 0 153 # define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_sse2 154 # endif 155 #endif 156 157 -----------------cut---------------- 158 159 5. Add the following lines to pngpriv.h, following the prototype for 160 png_read_filter_row_paeth4_neon: 161 162 -----------------cut---------------- 163 PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_sse2,(png_row_infop 164 row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); 165 PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub4_sse2,(png_row_infop 166 row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); 167 PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg3_sse2,(png_row_infop 168 row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); 169 PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_sse2,(png_row_infop 170 row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); 171 PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_sse2,(png_row_infop 172 row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); 173 PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_sse2,(png_row_infop 174 row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); 175 176 -----------------cut---------------- 177
