diff options
| author | Manojkumar Bhosale <manojkumar.bhosale@imgtec.com> | 2017-01-13 15:50:02 +0530 |
|---|---|---|
| committer | Manojkumar Bhosale <manojkumar.bhosale@imgtec.com> | 2017-01-13 15:50:02 +0530 |
| commit | 73a6f100a90af0de5bb16442867ff3720a5817c0 (patch) | |
| tree | 9370558280c980f2499d88fb0fdcf506a8179e8b /source/rotate.cc | |
| parent | 7c64163ff479ec4cdadbcbcf67397080d59117ef (diff) | |
| download | libyuv-73a6f100a90af0de5bb16442867ff3720a5817c0.tar.gz | |
Add MSA optimized rotate functions (used 16x16 transpose)
R=fbarchard@google.com
BUG=libyuv:634
Performance Gain (vs C vectorized)
TransposeWx16_MSA - ~6.0x
TransposeWx16_Any_MSA - ~4.7x
TransposeUVWx16_MSA - ~6.3x
TransposeUVWx16_Any_MSA - ~5.4x
Performance Gain (vs C non-vectorized)
TransposeWx16_MSA - ~6.0x
TransposeWx16_Any_MSA - ~4.8x
TransposeUVWx16_MSA - ~6.3x
TransposeUVWx16_Any_MSA - ~5.4x
Review-Url: https://codereview.chromium.org/2617703002 .
Diffstat (limited to 'source/rotate.cc')
| -rw-r--r-- | source/rotate.cc | 45 |
1 files changed, 39 insertions, 6 deletions
diff --git a/source/rotate.cc b/source/rotate.cc index bfd51421..277c53b2 100644 --- a/source/rotate.cc +++ b/source/rotate.cc @@ -29,8 +29,13 @@ void TransposePlane(const uint8* src, int width, int height) { int i = height; +#if defined(HAS_TRANSPOSEWX16_MSA) + void (*TransposeWx16)(const uint8* src, int src_stride, uint8* dst, + int dst_stride, int width) = TransposeWx16_C; +#else void (*TransposeWx8)(const uint8* src, int src_stride, uint8* dst, int dst_stride, int width) = TransposeWx8_C; +#endif #if defined(HAS_TRANSPOSEWX8_NEON) if (TestCpuFlag(kCpuHasNEON)) { TransposeWx8 = TransposeWx8_NEON; @@ -62,15 +67,24 @@ void TransposePlane(const uint8* src, } } #endif -#if defined(HAS_TRANSPOSEWX8_MSA) +#if defined(HAS_TRANSPOSEWX16_MSA) if (TestCpuFlag(kCpuHasMSA)) { - TransposeWx8 = TransposeWx8_Any_MSA; + TransposeWx16 = TransposeWx16_Any_MSA; if (IS_ALIGNED(width, 16)) { - TransposeWx8 = TransposeWx8_MSA; + TransposeWx16 = TransposeWx16_MSA; } } #endif +#if defined(HAS_TRANSPOSEWX16_MSA) + // 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) { TransposeWx8(src, src_stride, dst, dst_stride, width); @@ -78,6 +92,7 @@ void TransposePlane(const uint8* src, dst += 8; // Move over 8 columns. i -= 8; } +#endif if (i > 0) { TransposeWxH_C(src, src_stride, dst, dst_stride, width, i); @@ -218,9 +233,15 @@ void TransposeUV(const uint8* src, int width, int height) { int i = height; +#if defined(HAS_TRANSPOSEUVWX16_MSA) + void (*TransposeUVWx16)(const uint8* src, int src_stride, uint8* dst_a, + int dst_stride_a, uint8* dst_b, int dst_stride_b, + int width) = TransposeUVWx16_C; +#else void (*TransposeUVWx8)(const uint8* src, int src_stride, uint8* dst_a, int dst_stride_a, uint8* dst_b, int dst_stride_b, int width) = TransposeUVWx8_C; +#endif #if defined(HAS_TRANSPOSEUVWX8_NEON) if (TestCpuFlag(kCpuHasNEON)) { TransposeUVWx8 = TransposeUVWx8_NEON; @@ -240,15 +261,26 @@ void TransposeUV(const uint8* src, TransposeUVWx8 = TransposeUVWx8_DSPR2; } #endif -#if defined(HAS_TRANSPOSEUVWX8_MSA) +#if defined(HAS_TRANSPOSEUVWX16_MSA) if (TestCpuFlag(kCpuHasMSA)) { - TransposeUVWx8 = TransposeUVWx8_Any_MSA; + TransposeUVWx16 = TransposeUVWx16_Any_MSA; if (IS_ALIGNED(width, 8)) { - TransposeUVWx8 = TransposeUVWx8_MSA; + TransposeUVWx16 = TransposeUVWx16_MSA; } } #endif +#if defined(HAS_TRANSPOSEUVWX16_MSA) + // 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) { TransposeUVWx8(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, @@ -258,6 +290,7 @@ void TransposeUV(const uint8* src, dst_b += 8; // Move over 8 columns. i -= 8; } +#endif if (i > 0) { TransposeUVWxH_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, |