diff options
| author | Hao Chen <chenhao@loongson.cn> | 2020-07-29 18:30:43 +0800 |
|---|---|---|
| committer | Commit Bot <commit-bot@chromium.org> | 2020-08-12 18:12:19 +0000 |
| commit | 0de9bf3b1855fa38310fb981ebe1eaee6b5e7d85 (patch) | |
| tree | 40c81cd806f5df9cb52f58ab734f9b7644f775a2 | |
| parent | 5c6cdd0747a7bb991e58021f6e977a7f14b28719 (diff) | |
| download | libyuv-0de9bf3b1855fa38310fb981ebe1eaee6b5e7d85.tar.gz | |
Fix two failed case after enabling msa optimization.
Failed case: LibYUVConvertTest.TestI400 and LibYUVPlanarTest.ARGBBlend_Unattenuated.
This patch updates the I400ToARGBRow_MSA and ARGBBlendRow_MSA functions in the row_msa.cc file.
Signed-off-by: Shiyou Yin <yinshiyou-hf@loongson.cn>
Change-Id: Iec1a647af79be3ca1f2724802f6698deab60eac8
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/2330807
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
| -rw-r--r-- | source/row_msa.cc | 23 |
1 files changed, 15 insertions, 8 deletions
diff --git a/source/row_msa.cc b/source/row_msa.cc index 27fa4460..0a0cf10f 100644 --- a/source/row_msa.cc +++ b/source/row_msa.cc @@ -2834,14 +2834,21 @@ void I444ToARGBRow_MSA(const uint8_t* src_y, // TODO - respect YuvConstants void I400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, - const struct YuvConstants*, + const struct YuvConstants* yuvconstants, int width) { int x; +#if defined(__aarch64__) || defined(__arm__) + int ygb = yuvconstants->kUVBiasBGR[3]; + int yg = yuvconstants->kYToRgb[1]; +#else + int ygb = yuvconstants->kYBiasToRgb[0]; + int yg = yuvconstants->kYToRgb[0]; +#endif v16u8 src0, res0, res1, res2, res3, res4, dst0, dst1, dst2, dst3; v8i16 vec0, vec1; v4i32 reg0, reg1, reg2, reg3; - v4i32 vec_yg = __msa_fill_w(0x4A35); - v8i16 vec_ygb = __msa_fill_h(0xFB78); + v4i32 vec_yg = __msa_fill_w(yg); + v8i16 vec_ygb = __msa_fill_h(ygb); v16u8 alpha = (v16u8)__msa_ldi_b(ALPHA_VAL); v8i16 max = __msa_ldi_h(0xFF); v8i16 zero = {0}; @@ -3107,7 +3114,7 @@ void ARGBBlendRow_MSA(const uint8_t* src_argb0, uint8_t* dst_argb, int width) { int x; - v16u8 src0, src1, src2, src3, dst0, dst1; + v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3; v8u16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; v8u16 vec8, vec9, vec10, vec11, vec12, vec13; v8u16 const_256 = (v8u16)__msa_ldi_h(256); @@ -3152,12 +3159,12 @@ void ARGBBlendRow_MSA(const uint8_t* src_argb0, vec9 = (v8u16)__msa_srai_h((v8i16)vec9, 8); vec10 = (v8u16)__msa_srai_h((v8i16)vec10, 8); vec11 = (v8u16)__msa_srai_h((v8i16)vec11, 8); - vec0 += vec8; - vec1 += vec9; - vec2 += vec10; - vec3 += vec11; dst0 = (v16u8)__msa_pckev_b((v16i8)vec1, (v16i8)vec0); dst1 = (v16u8)__msa_pckev_b((v16i8)vec3, (v16i8)vec2); + dst2 = (v16u8)__msa_pckev_b((v16i8)vec9, (v16i8)vec8); + dst3 = (v16u8)__msa_pckev_b((v16i8)vec11, (v16i8)vec10); + dst0 = (v16u8)__msa_adds_u_b(dst0,dst2); + dst1 = (v16u8)__msa_adds_u_b(dst1,dst3); dst0 = __msa_bmnz_v(dst0, const_255, mask); dst1 = __msa_bmnz_v(dst1, const_255, mask); ST_UB2(dst0, dst1, dst_argb, 16); |