diff options
| author | Hao Chen <chenhao@loongson.cn> | 2021-12-20 20:20:26 +0800 |
|---|---|---|
| committer | Frank Barchard <fbarchard@chromium.org> | 2022-01-21 01:34:38 +0000 |
| commit | f8e2da48aed24a7b2608172aa5e59421f1f802d4 (patch) | |
| tree | 9dc87821993ff629500bf1fc320d195efdaa6dbe /source/rotate.cc | |
| parent | dfe046d27255cff06fc4cfe42c6d373fd83bc2aa (diff) | |
| download | libyuv-f8e2da48aed24a7b2608172aa5e59421f1f802d4.tar.gz | |
Add optimization functions in rotate_lsx.cc file.
Optimize two functions in source/rotate_lsx.cc file.
All test cases passed on loongarch platform.
Bug: libyuv:913
Change-Id: Idf670a1bc078f6284a499a292e0cb795f5b603b4
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/3351468
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Diffstat (limited to 'source/rotate.cc')
| -rw-r--r-- | source/rotate.cc | 47 |
1 files changed, 47 insertions, 0 deletions
diff --git a/source/rotate.cc b/source/rotate.cc index ddfaf5bb..e2601f1d 100644 --- a/source/rotate.cc +++ b/source/rotate.cc @@ -32,6 +32,9 @@ void TransposePlane(const uint8_t* src, #if defined(HAS_TRANSPOSEWX16_MSA) void (*TransposeWx16)(const uint8_t* src, int src_stride, uint8_t* dst, int dst_stride, int width) = TransposeWx16_C; +#elif defined(HAS_TRANSPOSEWX16_LSX) + void (*TransposeWx16)(const uint8_t* src, int src_stride, uint8_t* dst, + int dst_stride, int width) = TransposeWx16_C; #else void (*TransposeWx8)(const uint8_t* src, int src_stride, uint8_t* dst, int dst_stride, int width) = TransposeWx8_C; @@ -44,6 +47,13 @@ void TransposePlane(const uint8_t* src, TransposeWx16 = TransposeWx16_MSA; } } +#elif defined(HAS_TRANSPOSEWX16_LSX) + if (TestCpuFlag(kCpuHasLSX)) { + TransposeWx16 = TransposeWx16_Any_LSX; + if (IS_ALIGNED(width, 16)) { + TransposeWx16 = TransposeWx16_LSX; + } + } #else #if defined(HAS_TRANSPOSEWX8_NEON) if (TestCpuFlag(kCpuHasNEON)) { @@ -81,6 +91,14 @@ void TransposePlane(const uint8_t* src, dst += 16; // Move over 16 columns. i -= 16; } +#elif defined(HAS_TRANSPOSEWX16_LSX) + // Work across the source in 16x16 tiles + while (i >= 16) { + TransposeWx16(src, src_stride, dst, dst_stride, width); + src += 16 * src_stride; // Go down 16 rows. + dst += 16; // Move over 16 columns. + i -= 16; + } #else // Work across the source in 8x8 tiles while (i >= 8) { @@ -181,6 +199,14 @@ void RotatePlane180(const uint8_t* src, } } #endif +#if defined(HAS_MIRRORROW_LASX) + if (TestCpuFlag(kCpuHasLASX)) { + MirrorRow = MirrorRow_Any_LASX; + if (IS_ALIGNED(width, 64)) { + MirrorRow = MirrorRow_LASX; + } + } +#endif #if defined(HAS_COPYROW_SSE2) if (TestCpuFlag(kCpuHasSSE2)) { CopyRow = IS_ALIGNED(width, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2; @@ -234,6 +260,10 @@ void SplitTransposeUV(const uint8_t* src, void (*TransposeUVWx16)(const uint8_t* src, int src_stride, uint8_t* dst_a, int dst_stride_a, uint8_t* dst_b, int dst_stride_b, int width) = TransposeUVWx16_C; +#elif defined(HAS_TRANSPOSEUVWX16_LSX) + void (*TransposeUVWx16)(const uint8_t* src, int src_stride, uint8_t* dst_a, + int dst_stride_a, uint8_t* dst_b, int dst_stride_b, + int width) = TransposeUVWx16_C; #else void (*TransposeUVWx8)(const uint8_t* src, int src_stride, uint8_t* dst_a, int dst_stride_a, uint8_t* dst_b, int dst_stride_b, @@ -247,6 +277,13 @@ void SplitTransposeUV(const uint8_t* src, TransposeUVWx16 = TransposeUVWx16_MSA; } } +#elif defined(HAS_TRANSPOSEUVWX16_LSX) + if (TestCpuFlag(kCpuHasLSX)) { + TransposeUVWx16 = TransposeUVWx16_Any_LSX; + if (IS_ALIGNED(width, 8)) { + TransposeUVWx16 = TransposeUVWx16_LSX; + } + } #else #if defined(HAS_TRANSPOSEUVWX8_NEON) if (TestCpuFlag(kCpuHasNEON)) { @@ -281,6 +318,16 @@ void SplitTransposeUV(const uint8_t* src, dst_b += 16; // Move over 8 columns. i -= 16; } +#elif defined(HAS_TRANSPOSEUVWX16_LSX) + // Work through the source in 8x8 tiles. + while (i >= 16) { + TransposeUVWx16(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, + width); + src += 16 * src_stride; // Go down 16 rows. + dst_a += 16; // Move over 8 columns. + dst_b += 16; // Move over 8 columns. + i -= 16; + } #else // Work through the source in 8x8 tiles. while (i >= 8) { |