Message ID | 1583394457-21484-1-git-send-email-linjie.fu@intel.com |
---|---|
State | Accepted |
Headers | show |
Series | [FFmpeg-devel,1/4] lavc/x86/hevc_add_res: Fix overflow in ADD_RES_MMX_4_8 | expand |
Context | Check | Description |
---|---|---|
andriy/ffmpeg-patchwork | success | Make fate finished |
Quoting Linjie Fu (2020-03-05 08:47:37) > Fix overflow for coeff -32768 in function ADD_RES_MMX_4_8 with no > performance drop. > > ./checkasm --test=hevc_add_res --bench > > Mainline: > - hevc_add_res.add_residual [OK] > hevc_add_res_4x4_8_mmxext: 15.5 > > Add overflow test case: > - hevc_add_res.add_residual [FAILED] > > After: > - hevc_add_res.add_residual [OK] > hevc_add_res_4x4_8_mmxext: 15.0 > > Signed-off-by: Xu Guangxin <guangxin.xu@intel.com> > Signed-off-by: Linjie Fu <linjie.fu@intel.com> > --- > libavcodec/x86/hevc_add_res.asm | 23 +++++++++++------------ > 1 file changed, 11 insertions(+), 12 deletions(-) Looks ok and even more readable. Thank you, queueing.
diff --git a/libavcodec/x86/hevc_add_res.asm b/libavcodec/x86/hevc_add_res.asm index 36d4d8e..249c607 100644 --- a/libavcodec/x86/hevc_add_res.asm +++ b/libavcodec/x86/hevc_add_res.asm @@ -30,27 +30,26 @@ cextern pw_1023 %macro ADD_RES_MMX_4_8 0 mova m0, [r1] mova m2, [r1+8] - pxor m1, m1 - pxor m3, m3 - psubw m1, m0 - psubw m3, m2 - packuswb m0, m2 - packuswb m1, m3 - movd m2, [r0] + movd m1, [r0] movd m3, [r0+r2] - punpckldq m2, m3 - paddusb m0, m2 - psubusb m0, m1 + punpcklbw m1, m4 + punpcklbw m3, m4 + + paddsw m0, m1 + paddsw m2, m3 + packuswb m0, m4 + packuswb m2, m4 + movd [r0], m0 - psrlq m0, 32 - movd [r0+r2], m0 + movd [r0+r2], m2 %endmacro INIT_MMX mmxext ; void ff_hevc_add_residual_4_8_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride) cglobal hevc_add_residual_4_8, 3, 3, 6 + pxor m4, m4 ADD_RES_MMX_4_8 add r1, 16 lea r0, [r0+r2*2]