[FFmpeg-devel] [PATCH] enable runtime-cpudetect in configure

[Ramiro Polla]

Hi,

On Sun, Apr 5, 2009 at 10:10 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sun, Apr 05, 2009 at 09:34:41PM -0300, Ramiro Polla wrote:
>> On Sun, Apr 5, 2009 at 9:08 PM, Ramiro Polla <ramiro.polla at gmail.com> wrote:
>> > 2009/4/5 M?ns Rullg?rd <mans at mansr.com>:
>> >> Ramiro Polla <ramiro.polla at gmail.com> writes:
>> >>> Enable runtime-cpudetection in configure so that other projects don't
>> >>> have to use the --extra-cflags=-DRUNTIME_CPUDETECT hack.
>> >>> Also applies it to swscale.
>> >>>
>> >>> This can then be used, like Michael suggested, in dsputils as well to
>> >>> save on binary size.
>> >>>
>> >>> And then there's the issue that with runtime cpu detection, swscale
>> >>> expects the use to pass the appropriate sws_flags. Shouldn't ffmpeg.c
>> >>> use mm_support() and set those flags?
>> >>>
>> >>> Ramiro Polla
>> >>>
>> >>>
>> >>> From f9b99c86386b21984c9b409654f8fa4f363d9149 Mon Sep 17 00:00:00 2001
>> >>> From: ramiro <ramiro at ramiro-q6600.(none)>
>> >>> Date: Sun, 5 Apr 2009 01:34:26 -0300
>> >>> Subject: [PATCH] enable runtime-cpudetect in configure
>> >>>
>> >>> ---
>> >>> ?configure ? ? ? ? ? ? ? ? ? ? ?| ? ?3 +++
>> >>> ?libavcodec/ppc/check_altivec.c | ? ?2 +-
>> >>> ?libpostproc/postprocess.c ? ? ?| ? 14 +++++++-------
>> >>> ?3 files changed, 11 insertions(+), 8 deletions(-)
>> >>>
>> >>> diff --git a/configure b/configure
>> >>> index a9cc2fd..eae40c6 100755
>> >>> --- a/configure
>> >>> +++ b/configure
>> >>> @@ -100,6 +100,7 @@ show_help(){
>> >>> ? ?echo " ?--disable-golomb ? ? ? ? disable Golomb code"
>> >>> ? ?echo " ?--disable-mdct ? ? ? ? ? disable MDCT code"
>> >>> ? ?echo " ?--disable-rdft ? ? ? ? ? disable RDFT code"
>> >>> + ?echo " ?--enable-runtime-cpudetect detect cpu capabilities at runtime (bigger binary)"
>> >>> ? ?echo " ?--enable-hardcoded-tables use hardcoded tables instead of runtime generation"
>> >>> ? ?echo " ?--enable-memalign-hack ? emulate memalign, interferes with memory debuggers"
>> >>> ? ?echo " ?--enable-beos-netserver ?enable BeOS netserver"
>> >>> @@ -806,6 +807,7 @@ CONFIG_LIST="
>> >>> ? ? ?postproc
>> >>> ? ? ?powerpc_perf
>> >>> ? ? ?rdft
>> >>> + ? ?runtime_cpudetect
>> >>> ? ? ?shared
>> >>> ? ? ?small
>> >>> ? ? ?static
>> >>> @@ -2282,6 +2284,7 @@ if test "$extra_version" != ""; then
>> >>> ? ? ?echo "version string suffix ? ? $extra_version"
>> >>> ?fi
>> >>> ?echo "big-endian ? ? ? ? ? ? ? ?${bigendian-no}"
>> >>> +echo "runtime cpu detection ? ? ${runtime_cpudetect-no}"
>> >>> ?if enabled x86; then
>> >>> ? ? ?echo "yasm ? ? ? ? ? ? ? ? ? ? ?${yasm-no}"
>> >>> ? ? ?echo "MMX enabled ? ? ? ? ? ? ? ${mmx-no}"
>> >>
>> >> This parts looks OK.
>> >>
>> >> The renaming throughout the code should be safe, but those files are
>> >
>> >> not mine. ?Did you make sure it still works with both settings?
>> >
>> > Yes. Regression tests always pass and there is no difference in the
>> > object files.
>>
>> And here's an updated patch for swscale, since the previously attached
>> one depends on patches that got rejected/will take longer to be
>> accepted.
>
> ok, also note this might need changes in mplayers configure

Fixed mplayer and applied both patches.