mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2025-12-06 08:46:47 +08:00
4f7fd808b7
The existing Neon code only makes use of 64-bit vectors throughout which limits the performance on larger cores. To avoid this, swap the Neon code from a Wx8 implementation to a Wx16 implementation and process blocks of 16 full vectors at a time. The original code also handled widths that were not exact multiples of 16, however this should already be handled by the "any" kernel so it is removed. Finally, avoid duplicating the TransposeWx16_C fallback kernel definition in all architectures that need it, and just put it once in rotate_common.cc instead. Observed speedups for TransposePlane across a range of micro-architectures: Cortex-A53: -40.0% Cortex-A55: -20.7% Cortex-A57: -43.9% Cortex-A510: -43.5% Cortex-A520: -43.9% Cortex-A720: -31.1% Cortex-X2: -38.3% Cortex-X4: -43.6% Change-Id: Ic7c4d5f24eb27091d743ddc00cd95ef178b6984e Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5545459 Reviewed-by: Frank Barchard <fbarchard@chromium.org>
287 lines
10 KiB
C++
287 lines
10 KiB
C++
/*
|
|
* Copyright 2013 The LibYuv Project Authors. All rights reserved.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license
|
|
* that can be found in the LICENSE file in the root of the source
|
|
* tree. An additional intellectual property rights grant can be found
|
|
* in the file PATENTS. All contributing project authors may
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
*/
|
|
|
|
#ifndef INCLUDE_LIBYUV_ROTATE_ROW_H_
|
|
#define INCLUDE_LIBYUV_ROTATE_ROW_H_
|
|
|
|
#include "libyuv/basic_types.h"
|
|
|
|
#ifdef __cplusplus
|
|
namespace libyuv {
|
|
extern "C" {
|
|
#endif
|
|
|
|
#if defined(__pnacl__) || defined(__CLR_VER) || \
|
|
(defined(__native_client__) && defined(__x86_64__)) || \
|
|
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
|
|
#define LIBYUV_DISABLE_X86
|
|
#endif
|
|
#if defined(__native_client__)
|
|
#define LIBYUV_DISABLE_NEON
|
|
#endif
|
|
// MemorySanitizer does not support assembly code yet. http://crbug.com/344505
|
|
#if defined(__has_feature)
|
|
#if __has_feature(memory_sanitizer) && !defined(LIBYUV_DISABLE_NEON)
|
|
#define LIBYUV_DISABLE_NEON
|
|
#endif
|
|
#if __has_feature(memory_sanitizer) && !defined(LIBYUV_DISABLE_X86)
|
|
#define LIBYUV_DISABLE_X86
|
|
#endif
|
|
#endif
|
|
// The following are available for Visual C 32 bit:
|
|
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER) && \
|
|
!defined(__clang__)
|
|
#define HAS_TRANSPOSEWX8_SSSE3
|
|
#define HAS_TRANSPOSEUVWX8_SSE2
|
|
#endif
|
|
|
|
// The following are available for GCC 32 or 64 bit:
|
|
#if !defined(LIBYUV_DISABLE_X86) && (defined(__i386__) || defined(__x86_64__))
|
|
#define HAS_TRANSPOSEWX8_SSSE3
|
|
#define HAS_TRANSPOSE4X4_32_SSE2
|
|
#define HAS_TRANSPOSE4X4_32_AVX2
|
|
#endif
|
|
|
|
// The following are available for 64 bit GCC:
|
|
#if !defined(LIBYUV_DISABLE_X86) && defined(__x86_64__)
|
|
#define HAS_TRANSPOSEWX8_FAST_SSSE3
|
|
#define HAS_TRANSPOSEUVWX8_SSE2
|
|
#endif
|
|
|
|
#if !defined(LIBYUV_DISABLE_NEON) && \
|
|
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
|
|
#if defined(__aarch64__)
|
|
#define HAS_TRANSPOSEWX16_NEON
|
|
#else
|
|
#define HAS_TRANSPOSEWX8_NEON
|
|
#endif
|
|
#define HAS_TRANSPOSEUVWX8_NEON
|
|
#define HAS_TRANSPOSE4X4_32_NEON
|
|
#endif
|
|
|
|
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
|
|
#define HAS_TRANSPOSEWX16_MSA
|
|
#define HAS_TRANSPOSEUVWX16_MSA
|
|
#endif
|
|
|
|
#if !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)
|
|
#define HAS_TRANSPOSEWX16_LSX
|
|
#define HAS_TRANSPOSEUVWX16_LSX
|
|
#endif
|
|
|
|
void TransposeWxH_C(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width,
|
|
int height);
|
|
|
|
void TransposeWx8_C(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_C(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx8_NEON(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_NEON(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx8_SSSE3(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx8_Fast_SSSE3(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_MSA(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_LSX(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
|
|
void TransposeWx8_Any_NEON(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_Any_NEON(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx8_Any_SSSE3(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx8_Fast_Any_SSSE3(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_Any_MSA(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx16_Any_LSX(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
|
|
void TransposeUVWxH_C(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,
|
|
int height);
|
|
|
|
void TransposeUVWx8_C(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);
|
|
void TransposeUVWx16_C(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);
|
|
void TransposeUVWx8_SSE2(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);
|
|
void TransposeUVWx8_NEON(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);
|
|
void TransposeUVWx16_MSA(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);
|
|
void TransposeUVWx16_LSX(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);
|
|
|
|
void TransposeUVWx8_Any_SSE2(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);
|
|
void TransposeUVWx8_Any_NEON(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);
|
|
void TransposeUVWx16_Any_MSA(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);
|
|
void TransposeUVWx16_Any_LSX(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);
|
|
void TransposeWxH_16_C(const uint16_t* src,
|
|
int src_stride,
|
|
uint16_t* dst,
|
|
int dst_stride,
|
|
int width,
|
|
int height);
|
|
|
|
void TransposeWx8_16_C(const uint16_t* src,
|
|
int src_stride,
|
|
uint16_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
void TransposeWx1_16_C(const uint16_t* src,
|
|
int src_stride,
|
|
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_SSE2(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width);
|
|
|
|
void Transpose4x4_32_AVX2(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);
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|
|
|
|
#endif // INCLUDE_LIBYUV_ROTATE_ROW_H_
|