mirror of https://github.com/opencv/opencv.git
Open Source Computer Vision Library
https://opencv.org/
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
158 lines
5.6 KiB
158 lines
5.6 KiB
Enabling SSE support |
|
|
|
Copyright (c) 2016 Google, Inc. |
|
Written by Mike Klein, Matt Sarett |
|
|
|
This INSTALL file written by Glenn Randers-Pehrson, 2016. |
|
|
|
If you have moved intel_init.c and filter_sse2_intrinsics.c to a different |
|
directory, be sure to update the '#include "../../pngpriv.h"' line in both |
|
files if necessary to point to the correct relative location of pngpriv.h |
|
with respect to the new location of those files. |
|
|
|
To enable SSE support in libpng, follow the instructions in I, II, or III, |
|
below: |
|
|
|
I. Using patched "configure" scripts: |
|
|
|
First, apply intel_sse.patch in your build directory. |
|
|
|
patch -i contrib/intel/intel_sse.patch -p1 |
|
|
|
Then, if you are not building in a new GIT clone, e.g., in a tar |
|
distribution, remove any existing pre-built configure scripts: |
|
|
|
./configure --enable-maintainer-mode |
|
make maintainer-clean |
|
./autogen.sh --maintainer --clean |
|
|
|
Finally, configure libpng with -DPNG_INTEL_SSE in CPPFLAGS: |
|
|
|
./autogen.sh --maintainer |
|
CPPFLAGS="-DPNG_INTEL_SSE" ./configure [options] |
|
make CPPFLAGS="-DPNG_INTEL_SSE" [options] |
|
make |
|
|
|
II. Using a custom makefile: |
|
|
|
If you are using a custom makefile makefile, you will have to update it |
|
manually to include contrib/intel/*.o in the dependencies, and to define |
|
PNG_INTEL_SSE. |
|
|
|
III. Using manually updated "configure" scripts: |
|
|
|
If you prefer, manually edit pngpriv.h, configure.ac, and Makefile.am, |
|
following the instructions below, then follow the instructions in |
|
section II of INSTALL in the main libpng directory, then configure libpng |
|
with -DPNG_INTEL_SSE in CPPFLAGS. |
|
|
|
1. Add the following code to configure.ac under HOST SPECIFIC OPTIONS |
|
directly beneath the section for ARM: |
|
|
|
-----------------cut---------------- |
|
# INTEL |
|
# ===== |
|
# |
|
# INTEL SSE (SIMD) support. |
|
|
|
AC_ARG_ENABLE([intel-sse], |
|
AS_HELP_STRING([[[--enable-intel-sse]]], |
|
[Enable Intel SSE optimizations: =no/off, yes/on:] |
|
[no/off: disable the optimizations;] |
|
[yes/on: enable the optimizations.] |
|
[If not specified: determined by the compiler.]), |
|
[case "$enableval" in |
|
no|off) |
|
# disable the default enabling: |
|
AC_DEFINE([PNG_INTEL_SSE_OPT], [0], |
|
[Disable Intel SSE optimizations]) |
|
# Prevent inclusion of the assembler files below: |
|
enable_intel_sse=no;; |
|
yes|on) |
|
AC_DEFINE([PNG_INTEL_SSE_OPT], [1], |
|
[Enable Intel SSE optimizations]);; |
|
*) |
|
AC_MSG_ERROR([--enable-intel-sse=${enable_intel_sse}: invalid value]) |
|
esac]) |
|
|
|
# Add Intel specific files to all builds where the host_cpu is Intel ('x86*') |
|
# or where Intel optimizations were explicitly requested (this allows a |
|
# fallback if a future host CPU does not match 'x86*') |
|
AM_CONDITIONAL([PNG_INTEL_SSE], |
|
[test "$enable_intel_sse" != 'no' && |
|
case "$host_cpu" in |
|
i?86|x86_64) :;; |
|
*) test "$enable_intel_sse" != '';; |
|
esac]) |
|
-----------------cut---------------- |
|
|
|
2. Add the following code to Makefile.am under HOST SPECIFIC OPTIONS |
|
directly beneath the "if PNG_ARM_NEON ... endif" statement: |
|
|
|
-----------------cut---------------- |
|
if PNG_INTEL_SSE |
|
libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += contrib/intel/intel_init.c\ |
|
contrib/intel/filter_sse2_intrinsics.c |
|
endif |
|
-----------------cut---------------- |
|
|
|
3. Add the following lines to pngpriv.h, following the PNG_ARM_NEON_OPT |
|
code: |
|
|
|
-----------------cut---------------- |
|
#ifndef PNG_INTEL_SSE_OPT |
|
# ifdef PNG_INTEL_SSE |
|
/* Only check for SSE if the build configuration has been modified to |
|
* enable SSE optimizations. This means that these optimizations will |
|
* be off by default. See contrib/intel for more details. |
|
*/ |
|
# if defined(__SSE4_1__) || defined(__AVX__) || defined(__SSSE3__) || \ |
|
defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ |
|
(defined(_M_IX86_FP) && _M_IX86_FP >= 2) |
|
# define PNG_INTEL_SSE_OPT 1 |
|
# endif |
|
# endif |
|
#endif |
|
|
|
#if PNG_INTEL_SSE_OPT > 0 |
|
# ifndef PNG_INTEL_SSE_IMPLEMENTATION |
|
# if defined(__SSE4_1__) || defined(__AVX__) |
|
/* We are not actually using AVX, but checking for AVX is the best |
|
way we can detect SSE4.1 and SSSE3 on MSVC. |
|
*/ |
|
# define PNG_INTEL_SSE_IMPLEMENTATION 3 |
|
# elif defined(__SSSE3__) |
|
# define PNG_INTEL_SSE_IMPLEMENTATION 2 |
|
# elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \ |
|
(defined(_M_IX86_FP) && _M_IX86_FP >= 2) |
|
# define PNG_INTEL_SSE_IMPLEMENTATION 1 |
|
# else |
|
# define PNG_INTEL_SSE_IMPLEMENTATION 0 |
|
# endif |
|
# endif |
|
|
|
# if PNG_INTEL_SSE_IMPLEMENTATION > 0 |
|
# define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_sse2 |
|
# endif |
|
#endif |
|
|
|
-----------------cut---------------- |
|
|
|
4. Add the following lines to pngpriv.h, following the prototype for |
|
png_read_filter_row_paeth4_neon: |
|
|
|
-----------------cut---------------- |
|
PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_sse2,(png_row_infop |
|
row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); |
|
PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub4_sse2,(png_row_infop |
|
row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); |
|
PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg3_sse2,(png_row_infop |
|
row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); |
|
PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_sse2,(png_row_infop |
|
row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); |
|
PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_sse2,(png_row_infop |
|
row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); |
|
PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_sse2,(png_row_infop |
|
row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY); |
|
|
|
-----------------cut----------------
|
|
|