coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 08:46:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

BIT_EXACT for unattenuate and attenuate.

Browse Source 
- reenable Intel SIMD unaffected by BIT_EXACT
- add bit exact version of ARGBAttenuate, which uses ARM version of formula.
- add bit exact version of ARGBUnatenuate, which mimics the AVX code.

Apply clang format to cleanup code.

Bug: libyuv:908, b/202888439
Change-Id: Ie842b1b3956b48f4190858e61c02998caedc2897
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/3224702
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: richard winterton <rrwinterton@gmail.com>
This commit is contained in:
Frank Barchard 2021-10-15 12:12:02 -07:00 committed by libyuv LUCI CQ
parent 11cbf8f976
commit 55b97cb48f
22 changed files with 179 additions and 107 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 1798
Version: 1799
License: BSD
License File: LICENSE

2
include/libyuv/compare_row.h
View File

@ -18,7 +18,7 @@ namespace libyuv {
extern "C" {
#endif
#if defined(LIBYUV_BIT_EXACT) || defined(__pnacl__) || defined(__CLR_VER) || \
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86

2
include/libyuv/planar_functions.h
View File

@ -23,7 +23,7 @@ extern "C" {
#endif
// TODO(fbarchard): Move cpu macros to row.h
#if defined(LIBYUV_BIT_EXACT) || defined(__pnacl__) || defined(__CLR_VER) || \
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86

2
include/libyuv/rotate_row.h
View File

@ -18,7 +18,7 @@ namespace libyuv {
extern "C" {
#endif
#if defined(LIBYUV_BIT_EXACT) || defined(__pnacl__) || defined(__CLR_VER) || \
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86

32
include/libyuv/row.h
View File

@ -20,7 +20,7 @@ namespace libyuv {
extern "C" {
#endif
#if defined(LIBYUV_BIT_EXACT) || defined(__pnacl__) || defined(__CLR_VER) || \
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86
@ -74,8 +74,10 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(_M_IX86) || defined(__x86_64__) || defined(__i386__))
// Conversions:
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_ABGRTOUVROW_SSSE3
#define HAS_ABGRTOYROW_SSSE3
#endif
#define HAS_ARGB1555TOARGBROW_SSE2
#define HAS_ARGB4444TOARGBROW_SSE2
#define HAS_ARGBEXTRACTALPHAROW_SSE2
@ -87,13 +89,15 @@ extern "C" {
#define HAS_ARGBTORGB24ROW_SSSE3
#define HAS_ARGBTORGB565DITHERROW_SSE2
#define HAS_ARGBTORGB565ROW_SSE2
#define HAS_ARGBTOYJROW_SSSE3
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_ARGBTOUV444ROW_SSSE3
#define HAS_ARGBTOUVJROW_SSSE3
#define HAS_ARGBTOUVROW_SSSE3
#define HAS_ARGBTOYJROW_SSSE3
#define HAS_ARGBTOYROW_SSSE3
#define HAS_BGRATOUVROW_SSSE3
#define HAS_BGRATOYROW_SSSE3
#endif
#define HAS_COPYROW_ERMS
#define HAS_COPYROW_SSE2
#define HAS_H422TOARGBROW_SSSE3
@ -119,14 +123,16 @@ extern "C" {
#define HAS_NV21TORGB24ROW_SSSE3
#define HAS_RAWTOARGBROW_SSSE3
#define HAS_RAWTORGB24ROW_SSSE3
#define HAS_RAWTOYROW_SSSE3
#define HAS_RGB24TOARGBROW_SSSE3
#define HAS_RGB565TOARGBROW_SSE2
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_RAWTOYROW_SSSE3
#define HAS_RGB24TOYROW_SSSE3
#define HAS_RGB24TOYJROW_SSSE3
#define HAS_RAWTOYJROW_SSSE3
#define HAS_RGB565TOARGBROW_SSE2
#define HAS_RGBATOUVROW_SSSE3
#define HAS_RGBATOYROW_SSSE3
#endif
#define HAS_SETROW_ERMS
#define HAS_SETROW_X86
#define HAS_SPLITUVROW_SSE2
@ -142,7 +148,9 @@ extern "C" {
// Effects:
#define HAS_ARGBADDROW_SSE2
#define HAS_ARGBAFFINEROW_SSE2
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_ARGBATTENUATEROW_SSSE3
#endif
#define HAS_ARGBBLENDROW_SSSE3
#define HAS_ARGBCOLORMATRIXROW_SSSE3
#define HAS_ARGBCOLORTABLEROW_X86
@ -192,12 +200,14 @@ extern "C" {
#define HAS_ARGBPOLYNOMIALROW_AVX2
#define HAS_ARGBSHUFFLEROW_AVX2
#define HAS_ARGBTORGB565DITHERROW_AVX2
#define HAS_ARGBTOYJROW_AVX2
#define HAS_RAWTOYJROW_AVX2
#define HAS_RGB24TOYJROW_AVX2
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_ARGBTOUVJROW_AVX2
#define HAS_ARGBTOUVROW_AVX2
#define HAS_ARGBTOYJROW_AVX2
#define HAS_ARGBTOYROW_AVX2
#define HAS_RGB24TOYJROW_AVX2
#define HAS_RAWTOYJROW_AVX2
#endif
#define HAS_COPYROW_AVX
#define HAS_H422TOARGBROW_AVX2
#define HAS_HALFFLOATROW_AVX2
@ -230,7 +240,9 @@ extern "C" {
// Effects:
#define HAS_ARGBADDROW_AVX2
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_ARGBATTENUATEROW_AVX2
#endif
#define HAS_ARGBMULTIPLYROW_AVX2
#define HAS_ARGBSUBTRACTROW_AVX2
#define HAS_ARGBUNATTENUATEROW_AVX2
@ -297,7 +309,9 @@ extern "C" {
#define HAS_P410TOARGBROW_SSSE3
#define HAS_RAWTORGBAROW_SSSE3
#define HAS_RGB24MIRRORROW_SSSE3
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_RGBATOYJROW_SSSE3
#endif
#define HAS_SPLITARGBROW_SSE2
#define HAS_SPLITARGBROW_SSSE3
#define HAS_SPLITXRGBROW_SSE2
@ -319,8 +333,10 @@ extern "C" {
(defined(__x86_64__) || defined(__i386__)) && \
(defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2))
#define HAS_ABGRTOAR30ROW_AVX2
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_ABGRTOUVROW_AVX2
#define HAS_ABGRTOYROW_AVX2
#endif
#define HAS_ARGBTOAR30ROW_AVX2
#define HAS_ARGBTORAWROW_AVX2
#define HAS_ARGBTORGB24ROW_AVX2
@ -356,7 +372,9 @@ extern "C" {
#define HAS_MERGEUVROW_16_AVX2
#define HAS_MIRRORUVROW_AVX2
#define HAS_MULTIPLYROW_16_AVX2
#if !defined(LIBYUV_BIT_EXACT)
#define HAS_RGBATOYJROW_AVX2
#endif
#define HAS_SPLITARGBROW_AVX2
#define HAS_SPLITXRGBROW_AVX2
#define HAS_SPLITUVROW_16_AVX2

2
include/libyuv/scale_row.h
View File

@ -19,7 +19,7 @@ namespace libyuv {
extern "C" {
#endif
#if defined(LIBYUV_BIT_EXACT) || defined(__pnacl__) || defined(__CLR_VER) || \
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1798
#define LIBYUV_VERSION 1799
#endif // INCLUDE_LIBYUV_VERSION_H_

24
source/convert.cc
View File

@ -2004,16 +2004,22 @@ int RGB24ToJ420(const uint8_t* src_rgb24,
}
}
#endif
#if defined(HAS_ARGBTOYJROW_SSSE3) && defined(HAS_ARGBTOUVJROW_SSSE3)
#if defined(HAS_ARGBTOYJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
ARGBToYJRow = ARGBToYJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
ARGBToYJRow = ARGBToYJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOUVJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOYJROW_AVX2) && defined(HAS_ARGBTOUVJROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
ARGBToUVJRow = ARGBToUVJRow_Any_AVX2;
@ -2347,16 +2353,22 @@ int RAWToJ420(const uint8_t* src_raw,
}
}
#endif
#if defined(HAS_ARGBTOYJROW_SSSE3) && defined(HAS_ARGBTOUVJROW_SSSE3)
#if defined(HAS_ARGBTOYJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
ARGBToYJRow = ARGBToYJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
ARGBToYJRow = ARGBToYJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOUVJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOYJROW_AVX2) && defined(HAS_ARGBTOUVJROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
ARGBToUVJRow = ARGBToUVJRow_Any_AVX2;

24
source/convert_from_argb.cc
View File

@ -1830,16 +1830,22 @@ int ARGBToJ420(const uint8_t* src_argb,
src_argb = src_argb + (height - 1) * src_stride_argb;
src_stride_argb = -src_stride_argb;
}
#if defined(HAS_ARGBTOYJROW_SSSE3) && defined(HAS_ARGBTOUVJROW_SSSE3)
#if defined(HAS_ARGBTOYJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
ARGBToYJRow = ARGBToYJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
ARGBToYJRow = ARGBToYJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOUVJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOYJROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
ARGBToYJRow = ARGBToYJRow_Any_AVX2;
@ -1939,16 +1945,22 @@ int ARGBToJ422(const uint8_t* src_argb,
height = 1;
src_stride_argb = dst_stride_yj = dst_stride_u = dst_stride_v = 0;
}
#if defined(HAS_ARGBTOYJROW_SSSE3) && defined(HAS_ARGBTOUVJROW_SSSE3)
#if defined(HAS_ARGBTOYJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
ARGBToYJRow = ARGBToYJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
ARGBToYJRow = ARGBToYJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOUVJROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
if (IS_ALIGNED(width, 16)) {
ARGBToUVJRow = ARGBToUVJRow_SSSE3;
}
}
#endif
#if defined(HAS_ARGBTOYJROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
ARGBToYJRow = ARGBToYJRow_Any_AVX2;

2
source/row_any.cc
View File

@ -974,6 +974,8 @@ ANY11(ARGBToYRow_Any_SSSE3, ARGBToYRow_SSSE3, 0, 4, 1, 15)
ANY11(BGRAToYRow_Any_SSSE3, BGRAToYRow_SSSE3, 0, 4, 1, 15)
ANY11(ABGRToYRow_Any_SSSE3, ABGRToYRow_SSSE3, 0, 4, 1, 15)
ANY11(RGBAToYRow_Any_SSSE3, RGBAToYRow_SSSE3, 0, 4, 1, 15)
#endif
#ifdef HAS_YUY2TOYROW_SSE2
ANY11(YUY2ToYRow_Any_SSE2, YUY2ToYRow_SSE2, 1, 4, 1, 15)
ANY11(UYVYToYRow_Any_SSE2, UYVYToYRow_SSE2, 1, 4, 1, 15)
#endif

36
source/row_common.cc
View File

@ -28,14 +28,20 @@ extern "C" {
// The following macro from row_win makes the C code match the row_win code,
// which is 7 bit fixed point for ARGBToI420:
#if !defined(LIBYUV_BIT_EXACT) && !defined(LIBYUV_DISABLE_X86) && defined(_MSC_VER) && \
!defined(__clang__) && (defined(_M_IX86) || defined(_M_X64))
#if !defined(LIBYUV_BIT_EXACT) && !defined(LIBYUV_DISABLE_X86) && \
defined(_MSC_VER) && !defined(__clang__) && \
(defined(_M_IX86) || defined(_M_X64))
#define LIBYUV_RGB7 1
#endif
#if !defined(LIBYUV_BIT_EXACT) && (defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86))
#if !defined(LIBYUV_BIT_EXACT) && (defined(__x86_64__) || defined(_M_X64) || \
defined(__i386__) || defined(_M_IX86))
#define LIBYUV_ARGBTOUV_PAVGB 1
#define LIBYUV_RGBTOU_TRUNCATE 1
#define LIBYUV_ATTENUATE_DUP 1
#endif
#if defined(LIBYUV_BIT_EXACT)
#define LIBYUV_UNATTENUATE_DUP 1
#endif
// llvm x86 is poor at ternary operator, so use branchless min/max.
@ -3151,11 +3157,11 @@ void BlendPlaneRow_C(const uint8_t* src0,
}
#undef UBLEND
#if defined(__aarch64__) || defined(__arm__)
#define ATTENUATE(f, a) (f * a + 128) >> 8
#else
#if LIBYUV_ATTENUATE_DUP
// This code mimics the SSSE3 version for better testability.
#define ATTENUATE(f, a) (a | (a << 8)) * (f | (f << 8)) >> 24
#else
#define ATTENUATE(f, a) (f * a + 128) >> 8
#endif
// Multiply source RGB by alpha and store to destination.
@ -3242,6 +3248,14 @@ const uint32_t fixed_invtbl8[256] = {
T(0xfc), T(0xfd), T(0xfe), 0x01000100};
#undef T
#if LIBYUV_UNATTENUATE_DUP
// This code mimics the Intel SIMD version for better testability.
#define UNATTENUATE(f, ia) clamp255(((f | (f << 8)) * ia) >> 16)
#else
#define UNATTENUATE(f, ia) clamp255((f * ia) >> 8)
#endif
// mimics the Intel SIMD code for exactness.
void ARGBUnattenuateRow_C(const uint8_t* src_argb,
uint8_t* dst_argb,
int width) {
@ -3252,13 +3266,11 @@ void ARGBUnattenuateRow_C(const uint8_t* src_argb,
uint32_t r = src_argb[2];
const uint32_t a = src_argb[3];
const uint32_t ia = fixed_invtbl8[a] & 0xffff; // 8.8 fixed point
b = (b * ia) >> 8;
g = (g * ia) >> 8;
r = (r * ia) >> 8;
// Clamping should not be necessary but is free in assembly.
dst_argb[0] = clamp255(b);
dst_argb[1] = clamp255(g);
dst_argb[2] = clamp255(r);
dst_argb[0] = UNATTENUATE(b, ia);
dst_argb[1] = UNATTENUATE(g, ia);
dst_argb[2] = UNATTENUATE(r, ia);
dst_argb[3] = a;
src_argb += 4;
dst_argb += 4;

5
source/row_gcc.cc
View File

@ -1417,9 +1417,12 @@ void RGBAToYJRow_SSSE3(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
}
#endif // HAS_RGBATOYJROW_SSSE3
#ifdef HAS_ARGBTOYROW_AVX2
#if defined(HAS_ARGBTOYROW_AVX2) || defined(HAS_ARGBEXTRACTALPHAROW_AVX2)
// vpermd for vphaddw + vpackuswb vpermd.
static const lvec32 kPermdARGBToY_AVX = {0, 4, 1, 5, 2, 6, 3, 7};
#endif
#ifdef HAS_ARGBTOYROW_AVX2
// Convert 32 ARGB pixels (128 bytes) to 32 Y values.
void ARGBToYRow_AVX2(const uint8_t* src_argb, uint8_t* dst_y, int width) {

7
source/row_neon64.cc
View File

@ -1373,10 +1373,10 @@ void ARGBToRGB24Row_NEON(const uint8_t* src_argb,
int width) {
asm volatile(
"1: \n"
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64\n" // load 16 ARGB
"subs %w2, %w2, #16 \n" // 16 pixels per loop.
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16 ARGB
"subs %w2, %w2, #16 \n" // 16 pixels per loop.
"prfm pldl1keep, [%0, 448] \n"
"st3 {v0.16b,v1.16b,v2.16b}, [%1], #48\n" // store 8 RGB24
"st3 {v0.16b,v1.16b,v2.16b}, [%1], #48 \n" // store 8 RGB24
"b.gt 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_rgb24), // %1
@ -1683,7 +1683,6 @@ void ARGBToARGB4444Row_NEON(const uint8_t* src_argb,
: "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v23");
}
void ARGBToAR64Row_NEON(const uint8_t* src_argb,
uint16_t* dst_ar64,
int width) {

6
source/scale.cc
View File

@ -1829,7 +1829,8 @@ static void ScalePlaneSimple(int src_width,
}
for (i = 0; i < dst_height; ++i) {
ScaleCols(dst_ptr, src_ptr + (y >> 16) * (int64_t)src_stride, dst_width, x, dx);
ScaleCols(dst_ptr, src_ptr + (y >> 16) * (int64_t)src_stride, dst_width, x,
dx);
dst_ptr += dst_stride;
y += dy;
}
@ -1870,7 +1871,8 @@ static void ScalePlaneSimple_16(int src_width,
}
for (i = 0; i < dst_height; ++i) {
ScaleCols(dst_ptr, src_ptr + (y >> 16) * (int64_t)src_stride, dst_width, x, dx);
ScaleCols(dst_ptr, src_ptr + (y >> 16) * (int64_t)src_stride, dst_width, x,
dx);
dst_ptr += dst_stride;
y += dy;
}

8
source/scale_argb.cc
View File

@ -888,8 +888,8 @@ static void ScaleARGBSimple(int src_width,
}
for (j = 0; j < dst_height; ++j) {
ScaleARGBCols(dst_argb, src_argb + (y >> 16) * (int64_t)src_stride, dst_width, x,
dx);
ScaleARGBCols(dst_argb, src_argb + (y >> 16) * (int64_t)src_stride,
dst_width, x, dx);
dst_argb += dst_stride;
y += dy;
}
@ -973,8 +973,8 @@ static void ScaleARGB(const uint8_t* src,
filtering = kFilterNone;
if (dx == 0x10000 && dy == 0x10000) {
// Straight copy.
ARGBCopy(src + (y >> 16) * (int64_t)src_stride + (x >> 16) * 4, src_stride,
dst, dst_stride, clip_width, clip_height);
ARGBCopy(src + (y >> 16) * (int64_t)src_stride + (x >> 16) * 4,
src_stride, dst, dst_stride, clip_width, clip_height);
return;
}
}

22
source/scale_uv.cc
View File

@ -690,7 +690,8 @@ void ScaleUVLinearUp2(int src_width,
#endif
if (dst_height == 1) {
ScaleRowUp(src_uv + ((src_height - 1) / 2) * (int64_t)src_stride, dst_uv, dst_width);
ScaleRowUp(src_uv + ((src_height - 1) / 2) * (int64_t)src_stride, dst_uv,
dst_width);
} else {
dy = FixedDiv(src_height - 1, dst_height - 1);
y = (1 << 15) - 1;
@ -796,7 +797,8 @@ void ScaleUVLinearUp2_16(int src_width,
#endif
if (dst_height == 1) {
ScaleRowUp(src_uv + ((src_height - 1) / 2) * (int64_t)src_stride, dst_uv, dst_width);
ScaleRowUp(src_uv + ((src_height - 1) / 2) * (int64_t)src_stride, dst_uv,
dst_width);
} else {
dy = FixedDiv(src_height - 1, dst_height - 1);
y = (1 << 15) - 1;
@ -927,7 +929,8 @@ static void ScaleUVSimple(int src_width,
}
for (j = 0; j < dst_height; ++j) {
ScaleUVCols(dst_uv, src_uv + (y >> 16) * (int64_t)src_stride, dst_width, x, dx);
ScaleUVCols(dst_uv, src_uv + (y >> 16) * (int64_t)src_stride, dst_width, x,
dx);
dst_uv += dst_stride;
y += dy;
}
@ -1061,8 +1064,8 @@ static void ScaleUV(const uint8_t* src,
#ifdef HAS_UVCOPY
if (dx == 0x10000 && dy == 0x10000) {
// Straight copy.
UVCopy(src + (y >> 16) * (int64_t)src_stride + (x >> 16) * 2, src_stride, dst,
dst_stride, clip_width, clip_height);
UVCopy(src + (y >> 16) * (int64_t)src_stride + (x >> 16) * 2,
src_stride, dst, dst_stride, clip_width, clip_height);
return;
}
#endif
@ -1163,12 +1166,13 @@ int UVScale_16(const uint16_t* src_uv,
#ifdef HAS_UVCOPY
if (!filtering && src_width == dst_width && (src_height % dst_height == 0)) {
if (dst_height == 1) {
UVCopy_16(src_uv + ((src_height - 1) / 2) * (int64_t)src_stride_uv, src_stride_uv,
dst_uv, dst_stride_uv, dst_width, dst_height);
UVCopy_16(src_uv + ((src_height - 1) / 2) * (int64_t)src_stride_uv,
src_stride_uv, dst_uv, dst_stride_uv, dst_width, dst_height);
} else {
dy = src_height / dst_height;
UVCopy_16(src_uv + ((dy - 1) / 2) * (int64_t)src_stride_uv, dy * (int64_t)src_stride_uv,
dst_uv, dst_stride_uv, dst_width, dst_height);
UVCopy_16(src_uv + ((dy - 1) / 2) * (int64_t)src_stride_uv,
dy * (int64_t)src_stride_uv, dst_uv, dst_stride_uv, dst_width,
dst_height);
}
return 0;

3
unit_test/color_test.cc
View File

@ -22,7 +22,8 @@ namespace libyuv {
// TODO(fbarchard): clang x86 has a higher accuracy YUV to RGB.
// Port to Visual C and other CPUs
#if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__))
#if !defined(LIBYUV_BIT_EXACT) && !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || defined(__i386__))
#define ERROR_FULL 5
#define ERROR_J420 4
#else

22
unit_test/convert_test.cc
View File

@ -1532,7 +1532,7 @@ TESTATOBD(ARGB, 4, 4, 1, RGB565, 2, 2, 1)
// e.g. endian swap twice.
#define TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, W1280, N, NEG, \
OFF) \
TEST_F(LibYUVConvertTest, FMT_ATOB##_Endswap##N) { \
TEST_F(LibYUVConvertTest, FMT_ATOB##_Endswap##N) { \
const int kWidth = W1280; \
const int kHeight = benchmark_height_; \
const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A; \
@ -1580,8 +1580,8 @@ TESTATOBD(ARGB, 4, 4, 1, RGB565, 2, 2, 1)
TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_, \
_Opt, +, 0)
#else
#define TESTEND(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A) \
TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_, \
#define TESTEND(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A) \
TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_, \
_Opt, +, 0)
#endif
@ -3509,9 +3509,9 @@ TESTPLANAR16TOB(I012, 2, 2, 0xfff, AR30, 4, 4, 1)
TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
YALIGN, benchmark_width_, _Premult, +, 0, 1, S_DEPTH)
#else
#define TESTQPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, \
ALIGN, YALIGN, S_DEPTH) \
TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
#define TESTQPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, \
ALIGN, YALIGN, S_DEPTH) \
TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
YALIGN, benchmark_width_, _Opt, +, 0, 0, S_DEPTH)
#endif
@ -4140,17 +4140,17 @@ TEST_F(LibYUVConvertTest, TestRGB24ToJ420) {
const int kSize = 256;
align_buffer_page_end(orig_rgb24, kSize * 3 * 2); // 2 rows of RGB24
align_buffer_page_end(dest_j420, kSize * 3 / 2 * 2);
int iterations256 = (benchmark_width_ * benchmark_height_ + (kSize * 2 - 1)) / (kSize * 2) * benchmark_iterations_;
int iterations256 = (benchmark_width_ * benchmark_height_ + (kSize * 2 - 1)) /
(kSize * 2) * benchmark_iterations_;
for (int i = 0; i < kSize * 3 * 2; ++i) {
orig_rgb24[i] = i;
}
for (int i = 0; i < iterations256; ++i) {
RGB24ToJ420(orig_rgb24, kSize * 3,
dest_j420, kSize, // Y plane
dest_j420 + kSize * 2, kSize / 2, // U plane
dest_j420 + kSize * 5 / 2, kSize / 2, // V plane
RGB24ToJ420(orig_rgb24, kSize * 3, dest_j420, kSize, // Y plane
dest_j420 + kSize * 2, kSize / 2, // U plane
dest_j420 + kSize * 5 / 2, kSize / 2, // V plane
kSize, 2);
}

29
unit_test/planar_test.cc
View File

@ -29,6 +29,12 @@
#include "libyuv/row.h" /* For ScaleSumSamples_Neon */
#endif
#if defined(LIBYUV_BIT_EXACT)
#define EXPECTED_ATTENUATE_DIFF 0
#else
#define EXPECTED_ATTENUATE_DIFF 2
#endif
namespace libyuv {
TEST_F(LibYUVPlanarTest, TestAttenuate) {
@ -100,9 +106,9 @@ TEST_F(LibYUVPlanarTest, TestAttenuate) {
EXPECT_EQ(32, atten_pixels[128 * 4 + 1]);
EXPECT_EQ(21, atten_pixels[128 * 4 + 2]);
EXPECT_EQ(128, atten_pixels[128 * 4 + 3]);
EXPECT_NEAR(255, atten_pixels[255 * 4 + 0], 1);
EXPECT_NEAR(127, atten_pixels[255 * 4 + 1], 1);
EXPECT_NEAR(85, atten_pixels[255 * 4 + 2], 1);
EXPECT_NEAR(254, atten_pixels[255 * 4 + 0], EXPECTED_ATTENUATE_DIFF);
EXPECT_NEAR(127, atten_pixels[255 * 4 + 1], EXPECTED_ATTENUATE_DIFF);
EXPECT_NEAR(85, atten_pixels[255 * 4 + 2], EXPECTED_ATTENUATE_DIFF);
EXPECT_EQ(255, atten_pixels[255 * 4 + 3]);
free_aligned_buffer_page_end(atten2_pixels);
@ -158,28 +164,29 @@ TEST_F(LibYUVPlanarTest, ARGBAttenuate_Any) {
int max_diff = TestAttenuateI(benchmark_width_ + 1, benchmark_height_,
benchmark_iterations_, disable_cpu_flags_,
benchmark_cpu_info_, +1, 0);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, ARGBAttenuate_Unaligned) {
int max_diff =
TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, ARGBAttenuate_Invert) {
int max_diff =
TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, ARGBAttenuate_Opt) {
int max_diff =
TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
static int TestUnattenuateI(int width,
@ -231,28 +238,28 @@ TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Any) {
int max_diff = TestUnattenuateI(benchmark_width_ + 1, benchmark_height_,
benchmark_iterations_, disable_cpu_flags_,
benchmark_cpu_info_, +1, 0);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Unaligned) {
int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
benchmark_iterations_, disable_cpu_flags_,
benchmark_cpu_info_, +1, 1);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Invert) {
int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
benchmark_iterations_, disable_cpu_flags_,
benchmark_cpu_info_, -1, 0);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Opt) {
int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
benchmark_iterations_, disable_cpu_flags_,
benchmark_cpu_info_, +1, 0);
EXPECT_LE(max_diff, 2);
EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
}
TEST_F(LibYUVPlanarTest, TestARGBComputeCumulativeSum) {

30
unit_test/scale_argb_test.cc
View File

@ -258,20 +258,20 @@ static int ARGBClipTestFilter(int src_width,
// Test a scale factor with all 4 filters. Expect unfiltered to be exact, but
// filtering is different fixed point implementations for SSSE3, Neon and C.
#ifndef DISABLE_SLOW_TESTS
#define TEST_FACTOR(name, nom, denom) \
#define TEST_FACTOR(name, nom, denom) \
TEST_FACTOR1(, name, None, nom, denom, 0) \
TEST_FACTOR1(, name, Linear, nom, denom, 3) \
TEST_FACTOR1(, name, Bilinear, nom, denom, 3) \
TEST_FACTOR1(, name, Box, nom, denom, 3)
#else
#if defined(ENABLE_FULL_TESTS)
#define TEST_FACTOR(name, nom, denom) \
#define TEST_FACTOR(name, nom, denom) \
TEST_FACTOR1(DISABLED_, name, None, nom, denom, 0) \
TEST_FACTOR1(DISABLED_, name, Linear, nom, denom, 3) \
TEST_FACTOR1(DISABLED_, name, Bilinear, nom, denom, 3) \
TEST_FACTOR1(DISABLED_, name, Box, nom, denom, 3)
#else
#define TEST_FACTOR(name, nom, denom) \
#define TEST_FACTOR(name, nom, denom) \
TEST_FACTOR1(DISABLED_, name, Bilinear, nom, denom, 3)
#endif
#endif
@ -320,24 +320,24 @@ TEST_FACTOR(3, 1, 3)
#ifndef DISABLE_SLOW_TESTS
// Test scale to a specified size with all 4 filters.
#define TEST_SCALETO(name, width, height) \
TEST_SCALETO1(, name, width, height, None, 0) \
TEST_SCALETO1(, name, width, height, Linear, 3) \
#define TEST_SCALETO(name, width, height) \
TEST_SCALETO1(, name, width, height, None, 0) \
TEST_SCALETO1(, name, width, height, Linear, 3) \
TEST_SCALETO1(, name, width, height, Bilinear, 3)
#else
#if defined(ENABLE_FULL_TESTS)
#define TEST_SCALETO(name, width, height) \
TEST_SCALETO1(DISABLED_, name, width, height, None, 0) \
TEST_SCALETO1(DISABLED_, name, width, height, Linear, 3) \
#define TEST_SCALETO(name, width, height) \
TEST_SCALETO1(DISABLED_, name, width, height, None, 0) \
TEST_SCALETO1(DISABLED_, name, width, height, Linear, 3) \
TEST_SCALETO1(DISABLED_, name, width, height, Bilinear, 3)
#else
#define TEST_SCALETO(name, width, height) \
#define TEST_SCALETO(name, width, height) \
TEST_SCALETO1(DISABLED_, name, width, height, Bilinear, 3)
#endif
#endif
TEST_SCALETO(ARGBScale, 1, 1)
//TEST_SCALETO(ARGBScale, 256, 144) /* 128x72 * 2 */
// TEST_SCALETO(ARGBScale, 256, 144) /* 128x72 * 2 */
TEST_SCALETO(ARGBScale, 320, 240)
TEST_SCALETO(ARGBScale, 569, 480)
TEST_SCALETO(ARGBScale, 640, 360)
@ -524,8 +524,8 @@ TEST_F(LibYUVScaleTest, ARGBTest3x) {
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 =
(benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 * benchmark_iterations_;
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
ARGBScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);
@ -561,8 +561,8 @@ TEST_F(LibYUVScaleTest, ARGBTest4x) {
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 =
(benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 * benchmark_iterations_;
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
ARGBScale(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);

10
unit_test/scale_test.cc
View File

@ -1044,7 +1044,7 @@ TEST_FACTOR(3, 1, 3, 0)
#endif
TEST_SCALETO(Scale, 1, 1)
//TEST_SCALETO(Scale, 256, 144) /* 128x72 * 2 */
// TEST_SCALETO(Scale, 256, 144) /* 128x72 * 2 */
TEST_SCALETO(Scale, 320, 240)
TEST_SCALETO(Scale, 569, 480)
TEST_SCALETO(Scale, 640, 360)
@ -1418,8 +1418,8 @@ TEST_F(LibYUVScaleTest, PlaneTest3x) {
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 =
(benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 * benchmark_iterations_;
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
ScalePlane(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);
@ -1446,8 +1446,8 @@ TEST_F(LibYUVScaleTest, PlaneTest4x) {
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 =
(benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 * benchmark_iterations_;
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
ScalePlane(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);

14
unit_test/scale_uv_test.cc
View File

@ -141,7 +141,7 @@ static int UVTestFilter(int src_width,
TEST_FACTOR1(name, Box, nom, denom, 3)
#else
// Test a scale factor with Bilinear.
#define TEST_FACTOR(name, nom, denom) \
#define TEST_FACTOR(name, nom, denom) \
TEST_FACTOR1(name, Bilinear, nom, denom, 3)
#endif
@ -178,12 +178,12 @@ TEST_FACTOR(3, 1, 3)
TEST_SCALETO1(name, width, height, Linear, 3) \
TEST_SCALETO1(name, width, height, Bilinear, 3)
#else
#define TEST_SCALETO(name, width, height) \
#define TEST_SCALETO(name, width, height) \
TEST_SCALETO1(name, width, height, Bilinear, 3)
#endif
TEST_SCALETO(UVScale, 1, 1)
//TEST_SCALETO(UVScale, 256, 144) /* 128x72 * 2 */
// TEST_SCALETO(UVScale, 256, 144) /* 128x72 * 2 */
TEST_SCALETO(UVScale, 320, 240)
TEST_SCALETO(UVScale, 569, 480)
TEST_SCALETO(UVScale, 640, 360)
@ -224,8 +224,8 @@ TEST_F(LibYUVScaleTest, UVTest3x) {
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 =
(benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 * benchmark_iterations_;
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
UVScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);
@ -255,8 +255,8 @@ TEST_F(LibYUVScaleTest, UVTest4x) {
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 =
(benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 * benchmark_iterations_;
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
UVScale(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);