| Message ID | 20210217164106.6370-1-jamrial@gmail.com |
|---|---|
| State | Accepted |
| Commit | 2b4da1cb8c2984b37e5c912e103a1b8b734e7c1f |
| Headers | show |
| Series | [FFmpeg-devel,1/3] x86/vf_gblur: fix postscale_slice prologue | expand |
| Context | Check | Description |
|---|---|---|
| andriy/x86_make | success | Make finished |
| andriy/x86_make_fate | success | Make fate finished |
| andriy/PPC64_make | warning | Make failed |
lgtm
On Wed, Feb 17, 2021 at 01:41:04PM -0300, James Almer wrote: > x86_32 ABI does not pass float arguments directly on xmm regs, and the Win64 > ABI uses only the first four regs for this purpose. > > Signed-off-by: James Almer <jamrial@gmail.com> > --- > libavfilter/vf_gblur.c | 3 +-- > libavfilter/x86/vf_gblur.asm | 29 +++++++++++++---------------- > 2 files changed, 14 insertions(+), 18 deletions(-) > > diff --git a/libavfilter/vf_gblur.c b/libavfilter/vf_gblur.c > index 109a7a95f9..40956e122d 100644 > --- a/libavfilter/vf_gblur.c > +++ b/libavfilter/vf_gblur.c > @@ -234,8 +234,7 @@ void ff_gblur_init(GBlurContext *s) > { > s->horiz_slice = horiz_slice_c; > s->postscale_slice = postscale_c; > - if (ARCH_X86_64) > - ff_gblur_init_x86(s); > + ff_gblur_init_x86(s); > } fails on arm LD ffmpeg_g libavfilter/libavfilter.a(vf_gblur.o): In function `ff_gblur_init': arm/src/libavfilter/vf_gblur.c:237: undefined reference to `ff_gblur_init_x86' arm/src/libavfilter/vf_gblur.c:237: undefined reference to `ff_gblur_init_x86' collect2: error: ld returned 1 exit status Makefile:124: recipe for target 'ffmpeg_g' failed make: *** [ffmpeg_g] Error 1 [...]
On 2/17/2021 3:34 PM, Michael Niedermayer wrote: > On Wed, Feb 17, 2021 at 01:41:04PM -0300, James Almer wrote: >> x86_32 ABI does not pass float arguments directly on xmm regs, and the Win64 >> ABI uses only the first four regs for this purpose. >> >> Signed-off-by: James Almer <jamrial@gmail.com> >> --- >> libavfilter/vf_gblur.c | 3 +-- >> libavfilter/x86/vf_gblur.asm | 29 +++++++++++++---------------- >> 2 files changed, 14 insertions(+), 18 deletions(-) >> >> diff --git a/libavfilter/vf_gblur.c b/libavfilter/vf_gblur.c >> index 109a7a95f9..40956e122d 100644 >> --- a/libavfilter/vf_gblur.c >> +++ b/libavfilter/vf_gblur.c >> @@ -234,8 +234,7 @@ void ff_gblur_init(GBlurContext *s) >> { >> s->horiz_slice = horiz_slice_c; >> s->postscale_slice = postscale_c; >> - if (ARCH_X86_64) >> - ff_gblur_init_x86(s); >> + ff_gblur_init_x86(s); >> } > > fails on arm > > LD ffmpeg_g > libavfilter/libavfilter.a(vf_gblur.o): In function `ff_gblur_init': > arm/src/libavfilter/vf_gblur.c:237: undefined reference to `ff_gblur_init_x86' > arm/src/libavfilter/vf_gblur.c:237: undefined reference to `ff_gblur_init_x86' > collect2: error: ld returned 1 exit status > Makefile:124: recipe for target 'ffmpeg_g' failed > make: *** [ffmpeg_g] Error 1 Fixed, sorry about that.
diff --git a/libavfilter/vf_gblur.c b/libavfilter/vf_gblur.c index 109a7a95f9..40956e122d 100644 --- a/libavfilter/vf_gblur.c +++ b/libavfilter/vf_gblur.c @@ -234,8 +234,7 @@ void ff_gblur_init(GBlurContext *s) { s->horiz_slice = horiz_slice_c; s->postscale_slice = postscale_c; - if (ARCH_X86_64) - ff_gblur_init_x86(s); + ff_gblur_init_x86(s); } static int config_input(AVFilterLink *inlink) diff --git a/libavfilter/x86/vf_gblur.asm b/libavfilter/x86/vf_gblur.asm index c29ecba889..c2b2998202 100644 --- a/libavfilter/x86/vf_gblur.asm +++ b/libavfilter/x86/vf_gblur.asm @@ -185,27 +185,24 @@ HORIZ_SLICE %endif %macro POSTSCALE_SLICE 0 -%if UNIX64 -cglobal postscale_slice, 2, 2, 4, ptr, length -%else -cglobal postscale_slice, 5, 5, 4, ptr, length, postscale, min, max -%endif +cglobal postscale_slice, 2, 2, 4, ptr, length, postscale, min, max shl lengthd, 2 add ptrq, lengthq neg lengthq -%if WIN64 +%if ARCH_X86_32 + VBROADCASTSS m0, postscalem + VBROADCASTSS m1, minm + VBROADCASTSS m2, maxm +%elif WIN64 SWAP 0, 2 SWAP 1, 3 - SWAP 2, 4 -%endif -%if cpuflag(avx2) - vbroadcastss m0, xm0 - vbroadcastss m1, xm1 - vbroadcastss m2, xm2 -%else - shufps xm0, xm0, 0 - shufps xm1, xm1, 0 - shufps xm2, xm2, 0 + VBROADCASTSS m0, xm0 + VBROADCASTSS m1, xm1 + VBROADCASTSS m2, maxm +%else ; UNIX64 + VBROADCASTSS m0, xm0 + VBROADCASTSS m1, xm1 + VBROADCASTSS m2, xm3 %endif .loop:
x86_32 ABI does not pass float arguments directly on xmm regs, and the Win64 ABI uses only the first four regs for this purpose. Signed-off-by: James Almer <jamrial@gmail.com> --- libavfilter/vf_gblur.c | 3 +-- libavfilter/x86/vf_gblur.asm | 29 +++++++++++++---------------- 2 files changed, 14 insertions(+), 18 deletions(-)