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...@intel.com>
Signed-off-by: Linjie Fu <linjie...@intel.com>
---
libavcodec/x86/hevc_add_res.asm | 23 +++++++++++------------
1 file changed, 11 insertions(+), 12 deletions(-)
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]
--
2.7.4
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
