Skip to content

L
libyuv
Commit cb5262db authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

scale by even amounts using Neon 

BUG=156
TEST=./libyuv_unittest --gtest_filter=*ARGBScale*
Review URL: https://webrtc-codereview.appspot.com/930019

git-svn-id: http://libyuv.googlecode.com/svn/trunk@492 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent f08ac6bb
Expand all Show whitespace changes
Inline Side-by-side
Showing with 114 additions and 8 deletions
Android.mk
View file @ cb5262db
...@@ -36,7 +36,8 @@ ifeq ($(TARGET_ARCH_ABI),armeabi-v7a) ...@@ -36,7 +36,8 @@ ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
source/compare_neon.cc.neon \ source/compare_neon.cc.neon \
source/rotate_neon.cc.neon \ source/rotate_neon.cc.neon \
source/row_neon.cc.neon \ source/row_neon.cc.neon \
source/scale_neon.cc.neon source/scale_neon.cc.neon \
source/scale_neon_argb.cc.neon
endif endif
LOCAL_C_INCLUDES += $(LOCAL_PATH)/include LOCAL_C_INCLUDES += $(LOCAL_PATH)/include
......
README.chromium
View file @ cb5262db
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 491 Version: 492
License: BSD License: BSD
License File: LICENSE License File: LICENSE
......
include/libyuv/version.h
View file @ cb5262db
...@@ -11,6 +11,6 @@ ...@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT #ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 491 #define LIBYUV_VERSION 492
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
libyuv.gyp
View file @ cb5262db
...@@ -90,6 +90,7 @@ ...@@ -90,6 +90,7 @@
'source/row_win.cc', 'source/row_win.cc',
'source/scale.cc', 'source/scale.cc',
'source/scale_argb.cc', 'source/scale_argb.cc',
'source/scale_argb_neon.cc',
'source/scale_mips.cc', 'source/scale_mips.cc',
'source/scale_neon.cc', 'source/scale_neon.cc',
'source/video_common.cc', 'source/video_common.cc',
......
source/rotate.cc
View file @ cb5262db
...@@ -41,7 +41,7 @@ extern "C" { ...@@ -41,7 +41,7 @@ extern "C" {
#endif #endif
#endif #endif
#if !defined(YUV_DISABLE_ASM) && defined(__ARM_NEON__) #if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
#define HAS_MIRRORROW_NEON #define HAS_MIRRORROW_NEON
void MirrorRow_NEON(const uint8* src, uint8* dst, int width); void MirrorRow_NEON(const uint8* src, uint8* dst, int width);
#define HAS_MIRRORROW_UV_NEON #define HAS_MIRRORROW_UV_NEON
......
source/rotate_argb.cc
View file @ cb5262db
...@@ -29,6 +29,13 @@ void ScaleARGBRowDownEven_SSE2(const uint8* src_ptr, int src_stride, ...@@ -29,6 +29,13 @@ void ScaleARGBRowDownEven_SSE2(const uint8* src_ptr, int src_stride,
int src_stepx, int src_stepx,
uint8* dst_ptr, int dst_width); uint8* dst_ptr, int dst_width);
#endif #endif
#if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
#define HAS_SCALEARGBROWDOWNEVEN_NEON
void ScaleARGBRowDownEven_NEON(const uint8* src_ptr, int src_stride,
int src_stepx,
uint8* dst_ptr, int dst_width);
#endif
void ScaleARGBRowDownEven_C(const uint8* src_ptr, int, void ScaleARGBRowDownEven_C(const uint8* src_ptr, int,
int src_stepx, int src_stepx,
uint8* dst_ptr, int dst_width); uint8* dst_ptr, int dst_width);
...@@ -36,6 +43,7 @@ void ScaleARGBRowDownEven_C(const uint8* src_ptr, int, ...@@ -36,6 +43,7 @@ void ScaleARGBRowDownEven_C(const uint8* src_ptr, int,
static void ARGBTranspose(const uint8* src, int src_stride, static void ARGBTranspose(const uint8* src, int src_stride,
uint8* dst, int dst_stride, uint8* dst, int dst_stride,
int width, int height) { int width, int height) {
int src_pixel_step = src_stride >> 2;
void (*ScaleARGBRowDownEven)(const uint8* src_ptr, int src_stride, void (*ScaleARGBRowDownEven)(const uint8* src_ptr, int src_stride,
int src_step, uint8* dst_ptr, int dst_width) = ScaleARGBRowDownEven_C; int src_step, uint8* dst_ptr, int dst_width) = ScaleARGBRowDownEven_C;
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2) #if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
...@@ -43,9 +51,13 @@ static void ARGBTranspose(const uint8* src, int src_stride, ...@@ -43,9 +51,13 @@ static void ARGBTranspose(const uint8* src, int src_stride,
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) { IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2; ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2;
} }
#elif defined(HAS_SCALEARGBROWDOWNEVEN_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(height, 4) && // Width of dest.
IS_ALIGNED(src, 4)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_NEON;
}
#endif #endif
int src_pixel_step = src_stride >> 2;
for (int i = 0; i < width; ++i) { // column of source to row of dest. for (int i = 0; i < width; ++i) { // column of source to row of dest.
ScaleARGBRowDownEven(src, 0, src_pixel_step, dst, height); ScaleARGBRowDownEven(src, 0, src_pixel_step, dst, height);
dst += dst_stride; dst += dst_stride;
......
source/row_any.cc
View file @ cb5262db
...@@ -315,7 +315,8 @@ SplitUVRowANY(SplitUVRow_Any_AVX2, SplitUVRow_Unaligned_AVX2, SplitUVRow_C, 31) ...@@ -315,7 +315,8 @@ SplitUVRowANY(SplitUVRow_Any_AVX2, SplitUVRow_Unaligned_AVX2, SplitUVRow_C, 31)
SplitUVRowANY(SplitUVRow_Any_NEON, SplitUVRow_Unaligned_NEON, SplitUVRow_C, 15) SplitUVRowANY(SplitUVRow_Any_NEON, SplitUVRow_Unaligned_NEON, SplitUVRow_C, 15)
#endif #endif
#ifdef HAS_SPLITUVROW_MIPS_DSPR2 #ifdef HAS_SPLITUVROW_MIPS_DSPR2
SplitUVRowANY(SplitUVRow_Any_MIPS_DSPR2, SplitUVRow_Unaligned_MIPS_DSPR2, SplitUVRow_C, 15) SplitUVRowANY(SplitUVRow_Any_MIPS_DSPR2, SplitUVRow_Unaligned_MIPS_DSPR2,
SplitUVRow_C, 15)
#endif #endif
#undef SplitUVRowANY #undef SplitUVRowANY
......
source/row_common.cc
View file @ cb5262db
...@@ -722,7 +722,7 @@ static __inline void YuvPixel2(uint8 y, uint8 u, uint8 v, ...@@ -722,7 +722,7 @@ static __inline void YuvPixel2(uint8 y, uint8 u, uint8 v,
*r = Clip(static_cast<int32>((u * UR + v * VR) - (BR) + y1) >> 6); *r = Clip(static_cast<int32>((u * UR + v * VR) - (BR) + y1) >> 6);
} }
#if defined(__ARM_NEON__) #if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
// C mimic assembly. // C mimic assembly.
// TODO(fbarchard): Remove subsampling from Neon. // TODO(fbarchard): Remove subsampling from Neon.
void I444ToARGBRow_C(const uint8* src_y, void I444ToARGBRow_C(const uint8* src_y,
......
source/scale.cc
View file @ cb5262db
...@@ -50,7 +50,7 @@ void SetUseReferenceImpl(bool use) { ...@@ -50,7 +50,7 @@ void SetUseReferenceImpl(bool use) {
* *
*/ */
#if !defined(YUV_DISABLE_ASM) && defined(__ARM_NEON__) #if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
#define HAS_SCALEROWDOWN2_NEON #define HAS_SCALEROWDOWN2_NEON
// Note - not static due to reuse in convert for 444 to 420. // Note - not static due to reuse in convert for 444 to 420.
void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t /* src_stride */, void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t /* src_stride */,
......
source/scale_argb.cc
View file @ cb5262db
This diff is collapsed.
source/scale_argb_neon.cc 0 → 100644
View file @ cb5262db
/*
* Copyright 2012 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.
*/
#include "libyuv/basic_types.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for GCC Neon
#if !defined(YUV_DISABLE_ASM) && defined(__ARM_NEON__)
// Reads 4 pixels at a time.
// Alignment requirement: src_argb 4 byte aligned.
void ScaleARGBRowDownEven_NEON(const uint8* src_argb, ptrdiff_t,
int src_stepx,
uint8* dst_argb, int dst_width) {
asm volatile (
"mov r12, %3, lsl #2 \n"
".p2align 2 \n"
"1: \n"
"vld1.32 {d0[0]}, [%0], r12 \n"
"vld1.32 {d0[1]}, [%0], r12 \n"
"vld1.32 {d1[0]}, [%0], r12 \n"
"vld1.32 {d1[1]}, [%0], r12 \n"
"subs %2, #4 \n" // 4 pixels per loop.
"vst1.8 {q0}, [%1]! \n"
"bgt 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
: "r"(src_stepx) // %3
: "memory", "cc", "r12", "q0"
);
}
// Reads 4 pixels at a time.
// Alignment requirement: src_argb 4 byte aligned.
void ScaleARGBRowDownEvenInt_NEON(const uint8* src_argb, ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb, int dst_width) {
asm volatile (
"mov r12, %4, lsl #2 \n"
"add %1, %0 \n"
".p2align 2 \n"
"1: \n"
"vld1.8 {d0}, [%0], r12 \n" // Read 4 2x2 blocks -> 2x1
"vld1.8 {d1}, [%1], r12 \n"
"vld1.8 {d2}, [%0], r12 \n"
"vld1.8 {d3}, [%1], r12 \n"
"vld1.8 {d4}, [%0], r12 \n"
"vld1.8 {d5}, [%1], r12 \n"
"vld1.8 {d6}, [%0], r12 \n"
"vld1.8 {d7}, [%1], r12 \n"
"vaddl.u8 q0, d0, d1 \n"
"vaddl.u8 q1, d2, d3 \n"
"vaddl.u8 q2, d4, d5 \n"
"vaddl.u8 q3, d6, d7 \n"
"vswp.8 d1, d2 \n" // ab_cd -> ac_bd
"vswp.8 d5, d6 \n" // ef_gh -> eg_fh
"vadd.u16 q0, q0, q1 \n" // (a+b)_(c+d)
"vadd.u16 q2, q2, q3 \n" // (e+f)_(g+h)
"vrshrn.u16 d0, q0, #2 \n" // first 2 pixels.
"vrshrn.u16 d1, q2, #2 \n" // next 2 pixels.
"subs %3, #4 \n" // 4 pixels per loop.
"vst1.8 {q0}, [%2]! \n"
"bgt 1b \n"
: "+r"(src_argb), // %0
"+r"(src_stride), // %1
"+r"(dst_argb), // %2
"+r"(dst_width) // %3
: "r"(src_stepx) // %4
: "memory", "cc", "r12", "q0", "q1", "q2", "q3"
);
}
#endif // __ARM_NEON__
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment