Commit d51c3425 authored by fbarchard@google.com's avatar fbarchard@google.com

Rotate ARGB using a scale function that reads with step, combining to form 16 bytes at a time.

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/674007

git-svn-id: http://libyuv.googlecode.com/svn/trunk@294 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 4d2cfd32
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 293
Version: 294
License: BSD
License File: LICENSE
......
......@@ -38,8 +38,7 @@ int I420Rotate(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height,
RotationMode mode);
int src_width, int src_height, RotationMode mode);
// Rotate NV12 input and store in I420
int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
......@@ -47,8 +46,7 @@ int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height,
RotationMode mode);
int src_width, int src_height, RotationMode mode);
#ifdef __cplusplus
} // extern "C"
......
/*
* Copyright (c) 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.
*/
#ifndef INCLUDE_LIBYUV_ROTATE_ARGB_H_
#define INCLUDE_LIBYUV_ROTATE_ARGB_H_
#include "libyuv/basic_types.h"
#include "libyuv/rotate.h" // For RotationMode.
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Rotate ARGB frame
int ARGBRotate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int src_width, int src_height, RotationMode mode);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_ROTATE_ARGB_H_
......@@ -11,7 +11,7 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 293
#define LIBYUV_VERSION 294
#endif // INCLUDE_LIBYUV_VERSION_H_
......@@ -28,20 +28,28 @@
],
},
'sources': [
# includes
'include/libyuv/basic_types.h',
'include/libyuv/convert.h',
'include/libyuv/convert_from.h',
'include/libyuv/planar_functions.h',
'include/libyuv/scale.h',
'include/libyuv/version.h',
'include/libyuv/video_common.h',
# includes.
'include/libyuv.h',
'include/libyuv/basic_types.h',
'include/libyuv/compare.h',
'include/libyuv/convert.h',
'include/libyuv/convert_from.h',
'include/libyuv/cpu_id.h',
'include/libyuv/format_conversion.h',
'include/libyuv/mjpeg_decoder.h',
'include/libyuv/planar_functions.h',
'include/libyuv/rotate.h',
'include/libyuv/rotate_argb.h',
'include/libyuv/scale.h',
'include/libyuv/scale_argb.h',
'include/libyuv/version.h',
'include/libyuv/video_common.h',
# headers
# private includes.
'source/rotate_priv.h',
'source/row.h',
# sources
# sources.
'source/compare.cc',
'source/convert.cc',
'source/convert_from.cc',
......@@ -50,6 +58,7 @@
'source/mjpeg_decoder.cc',
'source/planar_functions.cc',
'source/rotate.cc',
'source/rotate_argb.cc',
'source/rotate_neon.cc',
'source/row_common.cc',
'source/row_neon.cc',
......@@ -60,7 +69,7 @@
'source/video_common.cc',
],
},
], # targets
], # targets.
}
# Local Variables:
......
......@@ -12,27 +12,28 @@
'target_name': 'libyuv_unittest',
'type': 'executable',
'dependencies': [
'libyuv.gyp:libyuv',
# The tests are based on gtest
'testing/gtest.gyp:gtest',
'testing/gtest.gyp:gtest_main',
'libyuv.gyp:libyuv',
# The tests are based on gtest
'testing/gtest.gyp:gtest',
'testing/gtest.gyp:gtest_main',
],
'defines': [
'LIBYUV_SVNREVISION="<!(svnversion -n)"',
],
'sources': [
# headers
'unit_test/unit_test.h',
# headers
'unit_test/unit_test.h',
# sources
'unit_test/compare_test.cc',
'unit_test/cpu_test.cc',
'unit_test/planar_test.cc',
'unit_test/rotate_test.cc',
'unit_test/scale_test.cc',
'unit_test/scale_argb_test.cc',
'unit_test/unit_test.cc',
'unit_test/version_test.cc',
# sources
'unit_test/compare_test.cc',
'unit_test/cpu_test.cc',
'unit_test/planar_test.cc',
'unit_test/rotate_argb_test.cc',
'unit_test/rotate_test.cc',
'unit_test/scale_argb_test.cc',
'unit_test/scale_test.cc',
'unit_test/unit_test.cc',
'unit_test/version_test.cc',
],
'conditions': [
['OS=="linux"', {
......
......@@ -766,9 +766,11 @@ static void TransposeWx8_C(const uint8* src, int src_stride,
static void TransposeWxH_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
for (int i = 0; i < width; ++i)
for (int j = 0; j < height; ++j)
for (int i = 0; i < width; ++i) {
for (int j = 0; j < height; ++j) {
dst[i * dst_stride + j] = src[j * src_stride + i];
}
}
}
void TransposePlane(const uint8* src, int src_stride,
......@@ -859,16 +861,17 @@ void RotatePlane180(const uint8* src, int src_stride,
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 64)) {
CopyRow = CopyRow_NEON;
}
#elif defined(HAS_COPYROW_X86)
if (IS_ALIGNED(width, 4)) {
#endif
#if defined(HAS_COPYROW_X86)
if (TestCpuFlag(kCpuHasX86) && IS_ALIGNED(width, 4)) {
CopyRow = CopyRow_X86;
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) &&
IS_ALIGNED(width, 32) &&
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
CopyRow = CopyRow_SSE2;
}
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32) &&
IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) &&
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
CopyRow = CopyRow_SSE2;
}
#endif
if (width > kMaxStride) {
......@@ -1034,6 +1037,10 @@ int I420Rotate(const uint8* src_y, int src_stride_y,
uint8* dst_v, int dst_stride_v,
int width, int height,
RotationMode mode) {
if (!src_y || !src_u || !src_v || width <= 0 || height == 0 ||
!dst_y || !dst_u || !dst_v) {
return -1;
}
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
......@@ -1105,6 +1112,10 @@ int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
uint8* dst_v, int dst_stride_v,
int width, int height,
RotationMode mode) {
if (!src_y || !src_uv || width <= 0 || height == 0 ||
!dst_y || !dst_u || !dst_v) {
return -1;
}
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
......
/*
* Copyright (c) 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/rotate.h"
#include "libyuv/cpu_id.h"
#include "libyuv/convert.h"
#include "libyuv/planar_functions.h"
#include "source/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// ARGBScale has a function to copy pixels to a row, striding each source
// pixel by a constant.
#if !defined(YUV_DISABLE_ASM) && (defined(_M_IX86) || \
defined(__x86_64__) || defined(__i386__))
#define HAS_SCALEARGBROWDOWNEVEN_SSE2
void ScaleARGBRowDownEven_SSE2(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,
int src_stepx,
uint8* dst_ptr, int dst_width);
static void ARGBTranspose(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void (*ScaleARGBRowDownEven)(const uint8* src_ptr, int src_stride,
int src_step, uint8* dst_ptr, int dst_width) = ScaleARGBRowDownEven_C;
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
if (TestCpuFlag(kCpuHasSSE2) &&
IS_ALIGNED(height, 4) && // width of dest.
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2;
}
#endif
int src_pixel_step = src_stride / 4;
for (int i = 0; i < width; ++i) { // column of source to row of dest.
ScaleARGBRowDownEven(src, 0, src_pixel_step, dst, height);
dst += dst_stride;
src += 4;
}
}
void ARGBRotate90(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
// Rotate by 90 is a ARGBTranspose with the source read
// from bottom to top. So set the source pointer to the end
// of the buffer and flip the sign of the source stride.
src += src_stride * (height - 1);
src_stride = -src_stride;
ARGBTranspose(src, src_stride, dst, dst_stride, width, height);
}
void ARGBRotate270(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
// Rotate by 270 is a ARGBTranspose with the destination written
// from bottom to top. So set the destination pointer to the end
// of the buffer and flip the sign of the destination stride.
dst += dst_stride * (width - 1);
dst_stride = -dst_stride;
ARGBTranspose(src, src_stride, dst, dst_stride, width, height);
}
void ARGBRotate180(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void (*ARGBMirrorRow)(const uint8* src, uint8* dst, int width) =
ARGBMirrorRow_C;
#if defined(HAS_ARGBMIRRORROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4) &&
IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) &&
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
ARGBMirrorRow = ARGBMirrorRow_SSSE3;
}
#endif
void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
#if defined(HAS_COPYROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width * 4, 64)) {
CopyRow = CopyRow_NEON;
}
#endif
#if defined(HAS_COPYROW_X86)
if (TestCpuFlag(kCpuHasX86)) {
CopyRow = CopyRow_X86;
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width * 4, 32) &&
IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) &&
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
CopyRow = CopyRow_SSE2;
}
#endif
if (width * 4 > kMaxStride) {
return;
}
// Swap first and last row and mirror the content. Uses a temporary row.
SIMD_ALIGNED(uint8 row[kMaxStride]);
const uint8* src_bot = src + src_stride * (height - 1);
uint8* dst_bot = dst + dst_stride * (height - 1);
int half_height = (height + 1) >> 1;
// Odd height will harmlessly mirror the middle row twice.
for (int y = 0; y < half_height; ++y) {
ARGBMirrorRow(src, row, width); // Mirror first row into a buffer
src += src_stride;
ARGBMirrorRow(src_bot, dst, width); // Mirror last row into first row
dst += dst_stride;
CopyRow(row, dst_bot, width * 4); // Copy first mirrored row into last
src_bot -= src_stride;
dst_bot -= dst_stride;
}
}
int ARGBRotate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height,
RotationMode mode) {
if (!src_argb || width <= 0 || height == 0 || !dst_argb) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
src_argb = src_argb + (height - 1) * src_stride_argb;
src_stride_argb = -src_stride_argb;
}
switch (mode) {
case kRotate0:
// copy frame
return ARGBCopy(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
case kRotate90:
ARGBRotate90(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
return 0;
case kRotate270:
ARGBRotate270(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
return 0;
case kRotate180:
ARGBRotate180(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
return 0;
default:
break;
}
return -1;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
/*
* Copyright (c) 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 <stdlib.h>
#include <time.h>
#include "libyuv/cpu_id.h"
#include "libyuv/rotate_argb.h"
#include "unit_test/unit_test.h"
namespace libyuv {
static int ARGBTestRotate(int src_width, int src_height,
int dst_width, int dst_height,
libyuv::RotationMode mode) {
const int b = 128;
int src_argb_plane_size = (src_width + b * 2) * (src_height + b * 2) * 4;
int src_stride_argb = (b * 2 + src_width) * 4;
align_buffer_16(src_argb, src_argb_plane_size)
memset(src_argb, 1, src_argb_plane_size);
int dst_argb_plane_size = (dst_width + b * 2) * (dst_height + b * 2) * 4;
int dst_stride_argb = (b * 2 + dst_width) * 4;
srandom(time(NULL));
int i, j;
for (i = b; i < (src_height + b); ++i) {
for (j = b; j < (src_width + b) * 4; ++j) {
src_argb[(i * src_stride_argb) + j] = (random() & 0xff);
}
}
const int runs = 1000;
align_buffer_16(dst_argb_c, dst_argb_plane_size)
align_buffer_16(dst_argb_opt, dst_argb_plane_size)
memset(dst_argb_c, 2, dst_argb_plane_size);
memset(dst_argb_opt, 3, dst_argb_plane_size);
// Warm up both versions for consistent benchmarks.
MaskCpuFlags(0); // Disable all CPU optimization.
ARGBRotate(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
dst_argb_c + (dst_stride_argb * b) + b * 4, dst_stride_argb,
src_width, src_height, mode);
MaskCpuFlags(-1); // Enable all CPU optimization.
ARGBRotate(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
dst_argb_opt + (dst_stride_argb * b) + b * 4, dst_stride_argb,
src_width, src_height, mode);
MaskCpuFlags(0); // Disable all CPU optimization.
double c_time = get_time();
for (i = 0; i < runs; ++i) {
ARGBRotate(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
dst_argb_c + (dst_stride_argb * b) + b * 4, dst_stride_argb,
src_width, src_height, mode);
}
c_time = (get_time() - c_time) / runs;
MaskCpuFlags(-1); // Enable all CPU optimization.
double opt_time = get_time();
for (i = 0; i < runs; ++i) {
ARGBRotate(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
dst_argb_opt + (dst_stride_argb * b) + b * 4, dst_stride_argb,
src_width, src_height, mode);
}
opt_time = (get_time() - opt_time) / runs;
// Report performance of C vs OPT
printf("filter %d - %8d us C - %8d us OPT\n",
mode, static_cast<int>(c_time*1e6), static_cast<int>(opt_time*1e6));
// C version may be a little off from the optimized. Order of
// operations may introduce rounding somewhere. So do a difference
// of the buffers and look to see that the max difference isn't
// over 2.
int max_diff = 0;
for (i = b; i < (dst_height + b); ++i) {
for (j = b * 4; j < (dst_width + b) * 4; ++j) {
int abs_diff = abs(dst_argb_c[(i * dst_stride_argb) + j] -
dst_argb_opt[(i * dst_stride_argb) + j]);
if (abs_diff > max_diff)
max_diff = abs_diff;
}
}
free_aligned_buffer_16(dst_argb_c)
free_aligned_buffer_16(dst_argb_opt)
free_aligned_buffer_16(src_argb)
return max_diff;
}
TEST_F(libyuvTest, ARGBRotate0) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = 1280;
const int dst_height = 720;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate0);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate90) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = 720;
const int dst_height = 1280;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate90);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate180) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = 1280;
const int dst_height = 720;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate180);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate270) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = 720;
const int dst_height = 1280;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate270);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate0_Odd) {
const int src_width = 1277;
const int src_height = 719;
const int dst_width = 1277;
const int dst_height = 719;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate0);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate90_Odd) {
const int src_width = 1277;
const int src_height = 719;
const int dst_width = 719;
const int dst_height = 1277;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate90);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate180_Odd) {
const int src_width = 1277;
const int src_height = 719;
const int dst_width = 1277;
const int dst_height = 719;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate180);
EXPECT_GE(1, err);
}
TEST_F(libyuvTest, ARGBRotate270_Odd) {
const int src_width = 1277;
const int src_height = 719;
const int dst_width = 719;
const int dst_height = 1277;
int err = ARGBTestRotate(src_width, src_height,
dst_width, dst_height, kRotate270);
EXPECT_GE(1, err);
}
} // namespace libyuv
......@@ -213,6 +213,21 @@ TEST_F(libyuvTest, ARGBScaleTo1366) {
}
}
TEST_F(libyuvTest, ARGBScaleTo4074) {
int src_width = 2880 * 2;
int src_height = 1800;
int dst_width = 4074;
int dst_height = 1272;
for (int f = 0; f < 2; ++f) {
int err = ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
EXPECT_GE(1, err);
}
}
TEST_F(libyuvTest, ARGBScaleTo853) {
int src_width = 1280;
int src_height = 720;
......
......@@ -275,6 +275,20 @@ TEST_F(libyuvTest, ScaleTo1366) {
}
}
TEST_F(libyuvTest, ScaleTo4074) {
int src_width = 2880 * 2;
int src_height = 1800;
int dst_width = 4074;
int dst_height = 1272;
for (int f = 0; f < 3; ++f) {
int err = TestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f), 1);
EXPECT_GE(1, err);
}
}
TEST_F(libyuvTest, ScaleTo853) {
int src_width = 1280;
int src_height = 720;
......
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