diff options
author | Darren Hsieh <darren.hsieh@sifive.com> | 2023-06-09 05:51:48 -0700 |
---|---|---|
committer | libyuv LUCI CQ <libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com> | 2023-07-26 18:05:50 +0000 |
commit | 10de943a12737be193e4f51df9e37b555ea09710 (patch) | |
tree | f31b2b0c07e99a30356fc8b10102123ef45f279e /source/convert_argb.cc | |
parent | a04fbafc5d50d585ee9526a2bed633a38ae69328 (diff) | |
download | libyuv-10de943a12737be193e4f51df9e37b555ea09710.tar.gz |
[RVV] Enable ScaleRowUp2_(Bi)linear_RVV/ScaleUVRowUp2_(Bi)linear_RVV
ScaleUVRowUp2_(Bi)linear_RVV function is equal to other platforms' ScaleRowUp2_(Bi)linear_Any_XXX.
We process entire row in this function.
Other platforms only implement non-edge part of image and process edge with scalar.
ScaleRowUp2_(Bi)linear_Any_XXX: Combine ScaleRowUp2_(Bi)linear_XXX(non-edge) + ScaleRowUp2_(Bi)linear_C(edge) by SBUH2LANY/SU2BLANY.
* Run on SiFive internal FPGA:
Test case RVV function Speedup
I444ScaleFrom640x360_Bilinear ScaleRowUp2_Bilinear_RVV 8.21
I444ScaleFrom640x360_Linear ScaleRowUp2_Linear_RVV 8.08
UVScaleFrom640x360_Bilinear ScaleUVRowUp2_Bilinear_RVV 7.80
UVScaleFrom640x360_Linear ScaleUVRowUp2_Linear_RVV 7.03
Change-Id: I539245ce51858f077506a78f0e7e82377ac6a95d
Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com>
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4666062
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Diffstat (limited to 'source/convert_argb.cc')
-rw-r--r-- | source/convert_argb.cc | 33 |
1 files changed, 33 insertions, 0 deletions
diff --git a/source/convert_argb.cc b/source/convert_argb.cc index f3a1f913..f6ab0784 100644 --- a/source/convert_argb.cc +++ b/source/convert_argb.cc @@ -6040,6 +6040,12 @@ static int I420ToARGBMatrixBilinear(const uint8_t* src_y, ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_NEON; } #endif +#if defined(HAS_SCALEROWUP2_BILINEAR_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + Scale2RowUp_Bilinear = ScaleRowUp2_Bilinear_RVV; + ScaleRowUp2_Linear = ScaleRowUp2_Linear_RVV; + } +#endif // alloc 4 lines temp const int row_size = (width + 31) & ~31; @@ -6171,6 +6177,11 @@ static int I422ToARGBMatrixLinear(const uint8_t* src_y, ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_NEON; } #endif +#if defined(HAS_SCALEROWUP2_LINEAR_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + ScaleRowUp2_Linear = ScaleRowUp2_Linear_RVV; + } +#endif // alloc 2 lines temp const int row_size = (width + 31) & ~31; @@ -6296,6 +6307,12 @@ static int I420ToRGB24MatrixBilinear(const uint8_t* src_y, ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_NEON; } #endif +#if defined(HAS_SCALEROWUP2_BILINEAR_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + Scale2RowUp_Bilinear = ScaleRowUp2_Bilinear_RVV; + ScaleRowUp2_Linear = ScaleRowUp2_Linear_RVV; + } +#endif // alloc 4 lines temp const int row_size = (width + 31) & ~31; @@ -6857,6 +6874,12 @@ static int I420AlphaToARGBMatrixBilinear( ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_NEON; } #endif +#if defined(HAS_SCALEROWUP2_BILINEAR_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + Scale2RowUp_Bilinear = ScaleRowUp2_Bilinear_RVV; + ScaleRowUp2_Linear = ScaleRowUp2_Linear_RVV; + } +#endif // alloc 4 lines temp const int row_size = (width + 31) & ~31; @@ -7052,6 +7075,11 @@ static int I422AlphaToARGBMatrixLinear(const uint8_t* src_y, ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_NEON; } #endif +#if defined(HAS_SCALEROWUP2_LINEAR_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + ScaleRowUp2_Linear = ScaleRowUp2_Linear_RVV; + } +#endif // alloc 2 lines temp const int row_size = (width + 31) & ~31; @@ -7790,6 +7818,11 @@ static int I422ToRGB24MatrixLinear(const uint8_t* src_y, ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_NEON; } #endif +#if defined(HAS_SCALEROWUP2_LINEAR_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + ScaleRowUp2_Linear = ScaleRowUp2_Linear_RVV; + } +#endif // alloc 2 lines temp const int row_size = (width + 31) & ~31; |