| Message ID | DB8P193MB06968DB8968EA73A417B29BCAFD62@DB8P193MB0696.EURP193.PROD.OUTLOOK.COM |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel,v2] libswscale/x86/yuv2rgb: Add missing EMMS | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
Hi, On Wed, Jun 26, 2024 at 8:03 PM Mario Hros <k3x-devel@outlook.com> wrote: > Previous rewrite from inline assembly into nasm (commit e934194) missed the required EMMS instruction to bring the x87 FPU back into usable state. > This needs to be done for 8-byte MMX or Extended MMX only. Sorry I didn't catch this thread earlier. I sent a patch to outright remove the mmx/mmxext code (thread "swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions"): https://lists.ffmpeg.org/pipermail/ffmpeg-devel/2024-June/329785.html The C code should be faster in most cases, or have very similar performance to the mmx/mmxext code. Is this not the case for you? Ramiro
diff --git a/libswscale/x86/yuv_2_rgb.asm b/libswscale/x86/yuv_2_rgb.asm index e3470fd9ad..5926133af8 100644 --- a/libswscale/x86/yuv_2_rgb.asm +++ b/libswscale/x86/yuv_2_rgb.asm @@ -354,6 +354,10 @@ add imageq, 8 * depth * time_num add indexq, 4 * time_num js .loop0 +%if mmsize == 8 +emms +%endif + RET %endmacro
Previous rewrite from inline assembly into nasm (commit e934194) missed the required EMMS instruction to bring the x87 FPU back into usable state. This needs to be done for 8-byte MMX or Extended MMX only. Signed-off-by: Mario Hros <k3x-devel@outlook.com> --- libswscale/x86/yuv_2_rgb.asm | 4 ++++ 1 file changed, 4 insertions(+)