Skip to content

L
libyuv
Commit 4e6c1fd4 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

scale_argb module for point and bilinear scaling of ARGB 

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/549004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@256 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 5bf29b59
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 259 additions and 66 deletions
README.chromium
View file @ 4e6c1fd4
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 255
Version: 256
License: BSD
License File: LICENSE
......
include/libyuv.h
View file @ 4e6c1fd4
......@@ -20,6 +20,7 @@
#include "libyuv/planar_functions.h"
#include "libyuv/rotate.h"
#include "libyuv/scale.h"
#include "libyuv/scale_argb.h"
#include "libyuv/version.h"
#include "libyuv/video_common.h"
......
include/libyuv/scale.h
View file @ 4e6c1fd4
......@@ -66,14 +66,6 @@ int ScaleOffset(const uint8* src, int src_width, int src_height,
uint8* dst, int dst_width, int dst_height, int dst_yoffset,
bool interpolate);
typedef void (*ARGBBlendRow)(const uint8* src_argb0,
const uint8* src_argb1,
uint8* dst_argb, int width);
// Get function to Alpha Blend ARGB pixels and store to destination.
ARGBBlendRow GetARGBBlend(uint8* dst_argb, int dst_stride_argb, int width);
// For testing, allow disabling of optimizations.
void SetUseReferenceImpl(bool use);
......
source/convertfrom.cc → include/libyuv/scale_argb.h
View file @ 4e6c1fd4
......@@ -8,6 +8,32 @@
* be found in the AUTHORS file in the root of the source tree.
*/
// TODO(fbarchard): Remove once builds have switched to convert_from
#include "convert_from.cc"
#ifndef INCLUDE_LIBYUV_SCALE_ARGB_H_
#define INCLUDE_LIBYUV_SCALE_ARGB_H_
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Supported filtering
enum FilterMode {
kFilterNone = 0, // Point sample; Fastest
kFilterBilinear = 1, // Faster than box, but lower quality scaling down.
kFilterBox = 2 // Highest quality
};
int ARGBScale(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
FilterMode filtering);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_SCALE_ARGB_H_
include/libyuv/version.h
View file @ 4e6c1fd4
......@@ -11,7 +11,7 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 255
#define LIBYUV_VERSION 256
#endif // INCLUDE_LIBYUV_VERSION_H_
libyuv.gyp
View file @ 4e6c1fd4
......@@ -56,6 +56,7 @@
'source/row_posix.cc',
'source/row_win.cc',
'source/scale.cc',
'source/scale_argb.cc',
'source/video_common.cc',
],
},
......
libyuv_test.gyp
View file @ 4e6c1fd4
......@@ -27,6 +27,7 @@
'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',
],
'conditions': [
......
source/scale.cc
View file @ 4e6c1fd4
......@@ -23,12 +23,6 @@ namespace libyuv {
extern "C" {
#endif
#if defined(_MSC_VER)
#define ALIGN16(var) __declspec(align(16)) var
#else
#define ALIGN16(var) var __attribute__((aligned(16)))
#endif
// Note: A Neon reference manual
// http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0204j/CJAJIIGG.html
// Note: Some SSE2 reference manuals
......@@ -571,12 +565,12 @@ static void ScaleFilterRows_NEON(uint8* dst_ptr,
(defined(_M_IX86) || defined(__i386__) || defined(__x86_64__))
#if defined(_MSC_VER)
#define TALIGN16(t, var) __declspec(align(16)) t _ ## var
#define TSIMD_ALIGNED(t, var) __declspec(align(16)) t _ ## var
#elif defined(__i386__) && \
(defined(__APPLE__) || defined(__MINGW32__) || defined(__CYGWIN__))
#define TALIGN16(t, var) t var __attribute__((aligned(16)))
#define TSIMD_ALIGNED(t, var) t var __attribute__((aligned(16)))
#else
#define TALIGN16(t, var) t _ ## var __attribute__((aligned(16)))
#define TSIMD_ALIGNED(t, var) t _ ## var __attribute__((aligned(16)))
#endif
#if defined(__APPLE__) && defined(__i386__)
......@@ -598,77 +592,77 @@ static void ScaleFilterRows_NEON(uint8* dst_ptr,
#endif
// Offsets for source bytes 0 to 9
extern "C" TALIGN16(const uint8, shuf0[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shuf0[16]) =
{ 0, 1, 3, 4, 5, 7, 8, 9, 128, 128, 128, 128, 128, 128, 128, 128 };
// Offsets for source bytes 11 to 20 with 8 subtracted = 3 to 12.
extern "C" TALIGN16(const uint8, shuf1[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shuf1[16]) =
{ 3, 4, 5, 7, 8, 9, 11, 12, 128, 128, 128, 128, 128, 128, 128, 128 };
// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
extern "C" TALIGN16(const uint8, shuf2[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shuf2[16]) =
{ 5, 7, 8, 9, 11, 12, 13, 15, 128, 128, 128, 128, 128, 128, 128, 128 };
// Offsets for source bytes 0 to 10
extern "C" TALIGN16(const uint8, shuf01[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shuf01[16]) =
{ 0, 1, 1, 2, 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10 };
// Offsets for source bytes 10 to 21 with 8 subtracted = 3 to 13.
extern "C" TALIGN16(const uint8, shuf11[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shuf11[16]) =
{ 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13 };
// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
extern "C" TALIGN16(const uint8, shuf21[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shuf21[16]) =
{ 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13, 13, 14, 14, 15 };
// Coefficients for source bytes 0 to 10
extern "C" TALIGN16(const uint8, madd01[16]) =
extern "C" TSIMD_ALIGNED(const uint8, madd01[16]) =
{ 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2 };
// Coefficients for source bytes 10 to 21
extern "C" TALIGN16(const uint8, madd11[16]) =
extern "C" TSIMD_ALIGNED(const uint8, madd11[16]) =
{ 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1 };
// Coefficients for source bytes 21 to 31
extern "C" TALIGN16(const uint8, madd21[16]) =
extern "C" TSIMD_ALIGNED(const uint8, madd21[16]) =
{ 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3 };
// Coefficients for source bytes 21 to 31
extern "C" TALIGN16(const int16, round34[8]) =
extern "C" TSIMD_ALIGNED(const int16, round34[8]) =
{ 2, 2, 2, 2, 2, 2, 2, 2 };
extern "C" TALIGN16(const uint8, kShuf38a[16]) =
extern "C" TSIMD_ALIGNED(const uint8, kShuf38a[16]) =
{ 0, 3, 6, 8, 11, 14, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
extern "C" TALIGN16(const uint8, kShuf38b[16]) =
extern "C" TSIMD_ALIGNED(const uint8, kShuf38b[16]) =
{ 128, 128, 128, 128, 128, 128, 0, 3, 6, 8, 11, 14, 128, 128, 128, 128 };
// Arrange words 0,3,6 into 0,1,2
extern "C" TALIGN16(const uint8, shufac0[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shufac0[16]) =
{ 0, 1, 6, 7, 12, 13, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
// Arrange words 0,3,6 into 3,4,5
extern "C" TALIGN16(const uint8, shufac3[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shufac3[16]) =
{ 128, 128, 128, 128, 128, 128, 0, 1, 6, 7, 12, 13, 128, 128, 128, 128 };
// Scaling values for boxes of 3x3 and 2x3
extern "C" TALIGN16(const uint16, scaleac3[8]) =
extern "C" TSIMD_ALIGNED(const uint16, scaleac3[8]) =
{ 65536 / 9, 65536 / 9, 65536 / 6, 65536 / 9, 65536 / 9, 65536 / 6, 0, 0 };
// Arrange first value for pixels 0,1,2,3,4,5
extern "C" TALIGN16(const uint8, shufab0[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shufab0[16]) =
{ 0, 128, 3, 128, 6, 128, 8, 128, 11, 128, 14, 128, 128, 128, 128, 128 };
// Arrange second value for pixels 0,1,2,3,4,5
extern "C" TALIGN16(const uint8, shufab1[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shufab1[16]) =
{ 1, 128, 4, 128, 7, 128, 9, 128, 12, 128, 15, 128, 128, 128, 128, 128 };
// Arrange third value for pixels 0,1,2,3,4,5
extern "C" TALIGN16(const uint8, shufab2[16]) =
extern "C" TSIMD_ALIGNED(const uint8, shufab2[16]) =
{ 2, 128, 5, 128, 128, 128, 10, 128, 13, 128, 128, 128, 128, 128, 128, 128 };
// Scaling values for boxes of 3x2 and 2x2
extern "C" TALIGN16(const uint16, scaleab2[8]) =
extern "C" TSIMD_ALIGNED(const uint16, scaleab2[8]) =
{ 65536 / 3, 65536 / 3, 65536 / 2, 65536 / 3, 65536 / 3, 65536 / 2, 0, 0 };
#endif
......@@ -3052,7 +3046,7 @@ static void ScaleRowDown8_C(const uint8* src_ptr, int,
// uses ScaleRowDown8_C instead.
static void ScaleRowDown8Int_C(const uint8* src_ptr, int src_stride,
uint8* dst, int dst_width) {
ALIGN16(uint8 src_row[kMaxRow12 * 2]);
SIMD_ALIGNED(uint8 src_row[kMaxRow12 * 2]);
assert(dst_width <= kMaxOutputWidth);
ScaleRowDown4Int_C(src_ptr, src_stride, src_row, dst_width * 2);
ScaleRowDown4Int_C(src_ptr + src_stride * 4, src_stride,
......@@ -3171,7 +3165,7 @@ static const int kMaxInputWidth = 2560;
static void ScaleRowDown34_0_Int_SSE2(const uint8* src_ptr, int src_stride,
uint8* dst_ptr, int dst_width) {
assert((dst_width % 3 == 0) && (dst_width > 0));
ALIGN16(uint8 row[kMaxInputWidth]);
SIMD_ALIGNED(uint8 row[kMaxInputWidth]);
ScaleFilterRows_SSE2(row, src_ptr, src_stride, dst_width * 4 / 3, 256 / 4);
ScaleFilterCols34_C(dst_ptr, row, dst_width);
}
......@@ -3180,7 +3174,7 @@ static void ScaleRowDown34_0_Int_SSE2(const uint8* src_ptr, int src_stride,
static void ScaleRowDown34_1_Int_SSE2(const uint8* src_ptr, int src_stride,
uint8* dst_ptr, int dst_width) {
assert((dst_width % 3 == 0) && (dst_width > 0));
ALIGN16(uint8 row[kMaxInputWidth]);
SIMD_ALIGNED(uint8 row[kMaxInputWidth]);
ScaleFilterRows_SSE2(row, src_ptr, src_stride, dst_width * 4 / 3, 256 / 2);
ScaleFilterCols34_C(dst_ptr, row, dst_width);
}
......@@ -3648,7 +3642,7 @@ static void ScalePlaneBox(int src_width, int src_height,
dst += dst_stride;
}
} else {
ALIGN16(uint16 row[kMaxInputWidth]);
SIMD_ALIGNED(uint16 row[kMaxInputWidth]);
void (*ScaleAddRows)(const uint8* src_ptr, int src_stride,
uint16* dst_ptr, int src_width, int src_height)=
ScaleAddRows_C;
......@@ -3737,7 +3731,7 @@ void ScalePlaneBilinear(int src_width, int src_height,
src_stride, dst_stride, src_ptr, dst_ptr);
} else {
ALIGN16(uint8 row[kMaxInputWidth + 1]);
SIMD_ALIGNED(uint8 row[kMaxInputWidth + 1]);
void (*ScaleFilterRows)(uint8* dst_ptr, const uint8* src_ptr,
int src_stride,
int dst_width, int source_y_fraction) =
......
source/scale_argb.cc 0 → 100644
View file @ 4e6c1fd4
This diff is collapsed.
unit_test/scale_argb_test.cc 0 → 100644
View file @ 4e6c1fd4
/*
* Copyright (c) 2011 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/scale_argb.h"
#include "unit_test/unit_test.h"
namespace libyuv {
static int ARGBTestFilter(int src_width, int src_height,
int dst_width, int dst_height,
FilterMode f) {
int b = 128;
int src_y_plane_size = (src_width + (2 * b)) * (src_height + (2 * b)) * 4;
int src_stride_y = (2 * b + src_width) * 4;
align_buffer_16(src_y, src_y_plane_size)
int dst_y_plane_size = (dst_width + (2 * b)) * (dst_height + (2 * b)) * 4;
int dst_stride_y = (2 * b + 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_y[(i * src_stride_y) + j] = (random() & 0xff);
}
}
const int runs = 1000;
align_buffer_16(dst_y_c, dst_y_plane_size)
align_buffer_16(dst_y_opt, dst_y_plane_size)
MaskCpuFlags(kCpuInitialized);
double c_time = get_time();
for (i = 0; i < runs; ++i)
ARGBScale(src_y + (src_stride_y * b) + b * 4, src_stride_y,
src_width, src_height,
dst_y_c + (dst_stride_y * b) + b * 4, dst_stride_y,
dst_width, dst_height, f);
c_time = (get_time() - c_time) / runs;
MaskCpuFlags(-1);
double opt_time = get_time();
for (i = 0; i < runs; ++i)
ARGBScale(src_y + (src_stride_y * b) + b * 4, src_stride_y,
src_width, src_height,
dst_y_opt + (dst_stride_y * b) + b * 4, dst_stride_y,
dst_width, dst_height, f);
opt_time = (get_time() - opt_time) / runs;
printf ("filter %d - %8d us c - %8d us opt\n",
f, (int)(c_time*1e6), (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 err = 0;
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_y_c[(i * dst_stride_y) + j] -
dst_y_opt[(i * dst_stride_y) + j]);
if (abs_diff > max_diff)
max_diff = abs_diff;
}
}
if (max_diff > 2)
err++;
free_aligned_buffer_16(dst_y_c)
free_aligned_buffer_16(dst_y_opt)
free_aligned_buffer_16(src_y)
return err;
}
TEST_F(libyuvTest, ARGBScaleDownBy2) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = src_width / 2;
const int dst_height = src_height / 2;
int err = 0;
for (int f = 0; f < 1; ++f) {
err += ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
}
EXPECT_EQ(0, err);
}
TEST_F(libyuvTest, ARGBScaleDownBy4) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = src_width / 4;
const int dst_height = src_height / 4;
int err = 0;
for (int f = 0; f < 1; ++f) {
err += ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
}
EXPECT_EQ(0, err);
}
TEST_F(libyuvTest, ARGBScaleDownBy34) {
const int src_width = 1280;
const int src_height = 720;
const int dst_width = src_width * 3 / 4;
const int dst_height = src_height * 3 / 4;
int err = 0;
for (int f = 0; f < 1; ++f) {
err += ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
}
EXPECT_EQ(0, err);
}
TEST_F(libyuvTest, ARGBScaleDownBy38) {
int src_width = 1280;
int src_height = 720;
int dst_width = src_width * 3 / 8;
int dst_height = src_height * 3 / 8;
int err = 0;
for (int f = 0; f < 1; ++f) {
err += ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
}
EXPECT_EQ(0, err);
}
TEST_F(libyuvTest, ARGBScalePlaneBilinear) {
int src_width = 1280;
int src_height = 720;
int dst_width = 1366;
int dst_height = 768;
int err = 0;
for (int f = 0; f < 1; ++f) {
err += ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
}
EXPECT_EQ(0, err);
}
} // namespace libyuv
unit_test/scale_test.cc
View file @ 4e6c1fd4
......@@ -61,7 +61,7 @@ static int TestFilter(int src_width, int src_height,
}
}
const int runs = 128;
const int runs = 1000;
align_buffer_16(dst_y_c, dst_y_plane_size)
align_buffer_16(dst_u_c, dst_uv_plane_size)
align_buffer_16(dst_v_c, dst_uv_plane_size)
......@@ -111,7 +111,7 @@ static int TestFilter(int src_width, int src_height,
for (i = b; i < (dst_height + b); ++i) {
for (j = b; j < (dst_width + b); ++j) {
int abs_diff = abs(dst_y_c[(i * dst_stride_y) + j] -
dst_y_opt[(i * dst_stride_y) + j]);
dst_y_opt[(i * dst_stride_y) + j]);
if (abs_diff > max_diff)
max_diff = abs_diff;
}
......@@ -120,11 +120,11 @@ static int TestFilter(int src_width, int src_height,
for (i = b; i < (dst_height_uv + b); ++i) {
for (j = b; j < (dst_width_uv + b); ++j) {
int abs_diff = abs(dst_u_c[(i * dst_stride_uv) + j] -
dst_u_opt[(i * dst_stride_uv) + j]);
dst_u_opt[(i * dst_stride_uv) + j]);
if (abs_diff > max_diff)
max_diff = abs_diff;
abs_diff = abs(dst_v_c[(i * dst_stride_uv) + j] -
dst_v_opt[(i * dst_stride_uv) + j]);
dst_v_opt[(i * dst_stride_uv) + j]);
if (abs_diff > max_diff)
max_diff = abs_diff;
......@@ -157,9 +157,9 @@ TEST_F(libyuvTest, ScaleDownBy2) {
int err = 0;
for (int f = 0; f < 3; ++f)
err += TestFilter (src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
err += TestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
EXPECT_EQ(0, err);
}
......@@ -173,9 +173,9 @@ TEST_F(libyuvTest, ScaleDownBy4) {
int err = 0;
for (int f = 0; f < 3; ++f)
err += TestFilter (src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
err += TestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
EXPECT_EQ(0, err);
}
......@@ -189,9 +189,9 @@ TEST_F(libyuvTest, ScaleDownBy34) {
int err = 0;
for (int f = 0; f < 3; ++f)
err += TestFilter (src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
err += TestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
EXPECT_EQ(0, err);
}
......@@ -201,13 +201,12 @@ TEST_F(libyuvTest, ScaleDownBy38) {
int src_height = 720;
int dst_width = src_width * 3 / 8;
int dst_height = src_height * 3 / 8;
int err = 0;
for (int f = 0; f < 3; ++f)
err += TestFilter (src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
err += TestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
EXPECT_EQ(0, err);
}
......@@ -217,13 +216,12 @@ TEST_F(libyuvTest, ScalePlaneBilinear) {
int src_height = 720;
int dst_width = 1366;
int dst_height = 768;
int err = 0;
for (int f = 0; f < 3; ++f)
err += TestFilter (src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
err += TestFilter(src_width, src_height,
dst_width, dst_height,
static_cast<FilterMode>(f));
EXPECT_EQ(0, err);
}
......
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