115 lines
3.7 KiB
Plaintext
115 lines
3.7 KiB
Plaintext
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
|
|
|
|
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 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. Insert the following lines above the copyright line near the top of
|
|
configure.ac:
|
|
|
|
-----------------cut----------------
|
|
# Copyright (c) 2016 Google, Inc.
|
|
# Written by Mike Klein and Matt Sarett
|
|
# Derived from the ARM supporting code in libpng/configure.ac, which was
|
|
-----------------cut----------------
|
|
|
|
2. 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----------------
|
|
|
|
3. Insert the following lines above the copyright line near the top of
|
|
Makefile.am:
|
|
|
|
-----------------cut----------------
|
|
# Copyright (c) 2016 Google, Inc.
|
|
# Written by Mike Klein and Matt Sarett
|
|
# Derived from the ARM supporting code in libpng/configure.ac, which was
|
|
-----------------cut----------------
|
|
|
|
4. 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----------------
|