Fix overflow for coeff -32768 in function ADD_RES_SSE_16_32_8 with no
performance drop.(SSE2/AVX/AVX2)
./checkasm --test=hevc_add_res --bench
Mainline:
- hevc_add_res.add_residual [OK]
hevc_add_res_32x32_8_sse2: 127.5
hevc_add_res_32x32_8_avx: 127.0
hevc_add_res_32x32_8_avx2: 86.5
Add overflow test case:
- hevc_add_res.add_residual [FAILED]
After:
- hevc_add_res.add_residual [OK]
hevc_add_res_32x32_8_sse2: 126.8
hevc_add_res_32x32_8_avx: 128.3
hevc_add_res_32x32_8_avx2: 86.8
Signed-off-by: Xu Guangxin <guangxin...@intel.com>
Signed-off-by: Linjie Fu <linjie...@intel.com>
---
libavcodec/x86/hevc_add_res.asm | 50 +++++++++++++++++++++--------------------
1 file changed, 26 insertions(+), 24 deletions(-)
diff --git a/libavcodec/x86/hevc_add_res.asm b/libavcodec/x86/hevc_add_res.asm
index e5e9f24..4d3aefc 100644
--- a/libavcodec/x86/hevc_add_res.asm
+++ b/libavcodec/x86/hevc_add_res.asm
@@ -84,34 +84,36 @@ cglobal hevc_add_residual_4_8, 3, 3, 6
%endmacro
%macro ADD_RES_SSE_16_32_8 3
- mova xm2, [r1+%1]
- mova xm6, [r1+%1+16]
+ mova m1, [%2]
+ mova m2, m1
+ punpcklbw m1, m0
+ punpckhbw m2, m0
+ mova xm5, [r1+%1]
+ mova xm6, [r1+%1+16]
%if cpuflag(avx2)
- vinserti128 m2, m2, [r1+%1+32], 1
- vinserti128 m6, m6, [r1+%1+48], 1
+ vinserti128 m5, m5, [r1+%1+32], 1
+ vinserti128 m6, m6, [r1+%1+48], 1
%endif
- psubw m1, m0, m2
- psubw m5, m0, m6
- packuswb m2, m6
- packuswb m1, m5
-
- mova xm4, [r1+%1+mmsize*2]
- mova xm6, [r1+%1+mmsize*2+16]
+ paddsw m1, m5
+ paddsw m2, m6
+
+ mova m3, [%3]
+ mova m4, m3
+ punpcklbw m3, m0
+ punpckhbw m4, m0
+ mova xm5, [r1+%1+mmsize*2]
+ mova xm6, [r1+%1+mmsize*2+16]
%if cpuflag(avx2)
- vinserti128 m4, m4, [r1+%1+96 ], 1
- vinserti128 m6, m6, [r1+%1+112], 1
+ vinserti128 m5, m5, [r1+%1+96], 1
+ vinserti128 m6, m6, [r1+%1+112], 1
%endif
- psubw m3, m0, m4
- psubw m5, m0, m6
- packuswb m4, m6
- packuswb m3, m5
-
- paddusb m2, [%2]
- paddusb m4, [%3]
- psubusb m2, m1
- psubusb m4, m3
- mova [%2], m2
- mova [%3], m4
+ paddsw m3, m5
+ paddsw m4, m6
+
+ packuswb m1, m2
+ packuswb m3, m4
+ mova [%2], m1
+ mova [%3], m3
%endmacro
--
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".
