[Buildroot] [PATCH] gst-ffmpeg: work-around bogus configure logic on SPARC

[Buildroot] [PATCH] gst-ffmpeg: work-around bogus configure logic on SPARC

[Thomas Petazzoni]

The libav version built into the gst-ffmpeg code produces a bogus
binary on SPARC, which causes the following error of the
check-bin-arch script:

  ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstffmpeg.so is Sparc v8+, should be Sparc
  ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstpostproc.so is Sparc v8+, should be Sparc
  ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstffmpegscale.so is Sparc v8+, should be Sparc

The problem is the following bit of code in
gst-lib/ext/libav/configure:

elif enabled sparc; then

    enabled vis && check_asm vis '"pdist %f0, %f0, %f0"' -mcpu=ultrasparc &&
        add_cflags -mcpu=ultrasparc -mtune=ultrasparc

I.e, it checks if the architecture supports the pdist
instruction... but forces -mcpu to ultrasparc while doing so. So it's
like "let's see if this Ultrasparc instruction exists when I force the
compiler to think I'm using Ultrasparc", which is non-sensical. This
has been fixed later on in libav upstream:

  https://git.libav.org/?p=libav.git;a=commit;h=6aa93689abe8c095cec9fa828c2dee3131008995

However, this commit cannot be backported as-is since the shell
function check_inline_asm did not exist in the old libav version
bundled in gst-ffmpeg.

Therefore, we take the simpler route of disabling the VIS
optimizations on SPARCv8 and Leon3.

Fixes:

  http://autobuild.buildroot.net/results/e82d179c3d4f92ad7423693a4b1d42379a3f5411/

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
---
 package/gstreamer/gst-ffmpeg/gst-ffmpeg.mk | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/package/gstreamer/gst-ffmpeg/gst-ffmpeg.mk b/package/gstreamer/gst-ffmpeg/gst-ffmpeg.mk
index b1d8f8b..3992fdf 100644
--- a/package/gstreamer/gst-ffmpeg/gst-ffmpeg.mk
+++ b/package/gstreamer/gst-ffmpeg/gst-ffmpeg.mk
@@ -80,6 +80,12 @@ else
 GST_FFMPEG_CONF_EXTRA_OPTS += --disable-altivec
 endif
 
+# libav configure script misdetects the VIS optimizations as being
+# available, so forcefully disable them.
+ifeq ($(BR2_sparc_v8)$(BR2_sparc_leon3),y)
+GST_FFMPEG_CONF_EXTRA_OPTS += --disable-vis
+endif
+
 ifeq ($(BR2_STATIC_LIBS),)
 GST_FFMPEG_CONF_EXTRA_OPTS += --enable-pic
 endif
-- 
2.7.4

[Buildroot] [PATCH] gst-ffmpeg: work-around bogus configure logic on SPARC

[Thomas Petazzoni]

Hello,

On Thu, 23 Mar 2017 00:04:56 +0100, Thomas Petazzoni wrote:
> The libav version built into the gst-ffmpeg code produces a bogus
> binary on SPARC, which causes the following error of the
> check-bin-arch script:
> 
>   ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstffmpeg.so is Sparc v8+, should be Sparc
>   ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstpostproc.so is Sparc v8+, should be Sparc
>   ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstffmpegscale.so is Sparc v8+, should be Sparc
> 
> The problem is the following bit of code in
> gst-lib/ext/libav/configure:
> 
> elif enabled sparc; then
> 
>     enabled vis && check_asm vis '"pdist %f0, %f0, %f0"' -mcpu=ultrasparc &&
>         add_cflags -mcpu=ultrasparc -mtune=ultrasparc
> 
> I.e, it checks if the architecture supports the pdist
> instruction... but forces -mcpu to ultrasparc while doing so. So it's
> like "let's see if this Ultrasparc instruction exists when I force the
> compiler to think I'm using Ultrasparc", which is non-sensical. This
> has been fixed later on in libav upstream:
> 
>   https://git.libav.org/?p=libav.git;a=commit;h=6aa93689abe8c095cec9fa828c2dee3131008995
> 
> However, this commit cannot be backported as-is since the shell
> function check_inline_asm did not exist in the old libav version
> bundled in gst-ffmpeg.
> 
> Therefore, we take the simpler route of disabling the VIS
> optimizations on SPARCv8 and Leon3.
> 
> Fixes:
> 
>   http://autobuild.buildroot.net/results/e82d179c3d4f92ad7423693a4b1d42379a3f5411/
> 
> Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
> ---
>  package/gstreamer/gst-ffmpeg/gst-ffmpeg.mk | 6 ++++++
>  1 file changed, 6 insertions(+)

Applied to master, thanks.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH] gst-ffmpeg: work-around bogus configure logic on SPARC

[Peter Korsgaard]

>>>>> "Thomas" == Thomas Petazzoni <thomas.petazzoni@free-electrons.com> writes:

 > The libav version built into the gst-ffmpeg code produces a bogus
 > binary on SPARC, which causes the following error of the
 > check-bin-arch script:

 >   ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstffmpeg.so is Sparc v8+, should be Sparc
 >   ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstpostproc.so is Sparc v8+, should be Sparc
 >   ERROR: architecture for ./usr/lib/gstreamer-0.10/libgstffmpegscale.so is Sparc v8+, should be Sparc

 > The problem is the following bit of code in
 > gst-lib/ext/libav/configure:

 > elif enabled sparc; then

 >     enabled vis && check_asm vis '"pdist %f0, %f0, %f0"' -mcpu=ultrasparc &&
 >         add_cflags -mcpu=ultrasparc -mtune=ultrasparc

 > I.e, it checks if the architecture supports the pdist
 > instruction... but forces -mcpu to ultrasparc while doing so. So it's
 > like "let's see if this Ultrasparc instruction exists when I force the
 > compiler to think I'm using Ultrasparc", which is non-sensical. This
 > has been fixed later on in libav upstream:

 >   https://git.libav.org/?p=libav.git;a=commit;h=6aa93689abe8c095cec9fa828c2dee3131008995

 > However, this commit cannot be backported as-is since the shell
 > function check_inline_asm did not exist in the old libav version
 > bundled in gst-ffmpeg.

 > Therefore, we take the simpler route of disabling the VIS
 > optimizations on SPARCv8 and Leon3.

 > Fixes:

 >   http://autobuild.buildroot.net/results/e82d179c3d4f92ad7423693a4b1d42379a3f5411/

 > Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

I take these VIS instructions cause runtime errors with the older sparc
cores?

Committed to 2017.02.x, thanks.

-- 
Bye, Peter Korsgaard