diff options
| author | Frank Barchard <fbarchard@google.com> | 2023-02-13 10:52:58 -0800 |
|---|---|---|
| committer | libyuv LUCI CQ <libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com> | 2023-02-13 20:14:57 +0000 |
| commit | 2bdc210be9eb11ded16bf3ef1f6cadb0d4dcb0c2 (patch) | |
| tree | c446d71538c965d0e5391ef77cd49b45ba51463d /include | |
| parent | b2528b0be934de1918e20c85fc170d809eeb49ab (diff) | |
| download | libyuv-2bdc210be9eb11ded16bf3ef1f6cadb0d4dcb0c2.tar.gz | |
MergeUV_AVX512BW for I420ToNV12
On Skylake Xeon 640x360 100000 iterations
AVX512 MergeUVPlane_Opt (1196 ms)
AVX2 MergeUVPlane_Opt (1565 ms)
SSE2 MergeUVPlane_Opt (1780 ms)
Pixel 7 MergeUVPlane_Opt (1177 ms)
Bug: None
Change-Id: If47d4fa957cf27781bba5fd6a2f0bf554101a5c6
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4242247
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Diffstat (limited to 'include')
| -rw-r--r-- | include/libyuv/rotate_row.h | 21 | ||||
| -rw-r--r-- | include/libyuv/row.h | 12 | ||||
| -rw-r--r-- | include/libyuv/version.h | 2 |
3 files changed, 33 insertions, 2 deletions
diff --git a/include/libyuv/rotate_row.h b/include/libyuv/rotate_row.h index b773f886..64d0b59f 100644 --- a/include/libyuv/rotate_row.h +++ b/include/libyuv/rotate_row.h @@ -232,6 +232,27 @@ void TransposeWx1_16_C(const uint16_t* src, uint16_t* dst, int dst_stride, int width); + +// Transpose 32 bit values (ARGB) +void Transpose4x4_32_NEON(const uint8_t* src, + int src_stride, + uint8_t* dst, + int dst_stride, + int width); + +void Transpose4x4_32_C(const uint8_t* src, + int src_stride, + uint8_t* dst, + int dst_stride, + int width); + +// Transpose 32 bit values (ARGB) +void Transpose8x8_32_NEON(const uint8_t* src, + int src_stride, + uint8_t* dst, + int dst_stride, + int width); + #ifdef __cplusplus } // extern "C" } // namespace libyuv diff --git a/include/libyuv/row.h b/include/libyuv/row.h index 8d998727..e1837169 100644 --- a/include/libyuv/row.h +++ b/include/libyuv/row.h @@ -402,9 +402,11 @@ extern "C" { // The following are available for AVX512 clang x86 platforms: // TODO(fbarchard): Port to GCC and Visual C // TODO(fbarchard): re-enable HAS_ARGBTORGB24ROW_AVX512VBMI. Issue libyuv:789 +// TODO(fbarchard): Port MERGEUV to assembly #if !defined(LIBYUV_DISABLE_X86) && \ - (defined(__x86_64__) || defined(__i386__)) && (defined(CLANG_HAS_AVX512)) + (defined(__x86_64__) || defined(__i386__)) && (defined(CLANG_HAS_AVX512) && !defined(_MSC_VER)) #define HAS_ARGBTORGB24ROW_AVX512VBMI +#define HAS_MERGEUVROW_AVX512BW #endif // The following are available for AVX512 clang x64 platforms: @@ -2184,6 +2186,10 @@ void MergeUVRow_AVX2(const uint8_t* src_u, const uint8_t* src_v, uint8_t* dst_uv, int width); +void MergeUVRow_AVX512BW(const uint8_t* src_u, + const uint8_t* src_v, + uint8_t* dst_uv, + int width); void MergeUVRow_NEON(const uint8_t* src_u, const uint8_t* src_v, uint8_t* dst_uv, @@ -2204,6 +2210,10 @@ void MergeUVRow_Any_AVX2(const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* dst_ptr, int width); +void MergeUVRow_Any_AVX512BW(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + int width); void MergeUVRow_Any_NEON(const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* dst_ptr, diff --git a/include/libyuv/version.h b/include/libyuv/version.h index adc49c4e..3e877f38 100644 --- a/include/libyuv/version.h +++ b/include/libyuv/version.h @@ -11,6 +11,6 @@ #ifndef INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_ -#define LIBYUV_VERSION 1857 +#define LIBYUV_VERSION 1860 #endif // INCLUDE_LIBYUV_VERSION_H_ |