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

Rounding added to scaling upsampler

Browse Source 
Bug: libyuv:872, b/178521093
Change-Id: I86749f73f5e55d5fd8b87ea6938084cbacb1cda7
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/2686945
Reviewed-by: Wan-Teh Chang <wtc@google.com>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
Frank Barchard 2021-02-10 00:22:41 -08:00 committed by Frank Barchard
parent 742791f13a
commit 12a4a2372c
11 changed files with 251 additions and 214 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
source/convert.cc
View File

@ -15,8 +15,8 @@
#include "libyuv/planar_functions.h"
#include "libyuv/rotate.h"
#include "libyuv/row.h"
#include "libyuv/scale.h" // For ScalePlane()
#include "libyuv/scale_uv.h" // For UVScale()
#include "libyuv/scale.h" // For ScalePlane()
#include "libyuv/scale_uv.h" // For UVScale()
#ifdef __cplusplus
namespace libyuv {

32
source/row_gcc.cc
View File

@ -4374,15 +4374,15 @@ void SplitARGBRow_SSSE3(const uint8_t* src_argb,
"lea 8(%1),%1 \n"
"subl $0x8,%5 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_r), // %1
"+r"(dst_g), // %2
"+r"(dst_b), // %3
"+r"(dst_a), // %4
: "+r"(src_argb), // %0
"+r"(dst_r), // %1
"+r"(dst_g), // %2
"+r"(dst_b), // %3
"+r"(dst_a), // %4
#if defined(__i386__)
"+m"(width) // %5
"+m"(width) // %5
#else
"+rm"(width) // %5
"+rm"(width) // %5
#endif
: "m"(kShuffleMaskARGBSplit) // %6
: "memory", "cc", "xmm0", "xmm1", "xmm2");
@ -4465,15 +4465,15 @@ void SplitARGBRow_AVX2(const uint8_t* src_argb,
"lea 16(%1),%1 \n"
"subl $0x10,%5 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_r), // %1
"+r"(dst_g), // %2
"+r"(dst_b), // %3
"+r"(dst_a), // %4
: "+r"(src_argb), // %0
"+r"(dst_r), // %1
"+r"(dst_g), // %2
"+r"(dst_b), // %3
"+r"(dst_a), // %4
#if defined(__i386__)
"+m"(width) // %5
"+m"(width) // %5
#else
"+rm"(width) // %5
"+rm"(width) // %5
#endif
: "m"(kShuffleMaskARGBSplit_AVX2), // %6
"m"(kShuffleMaskARGBPermute_AVX2) // %7
@ -7186,7 +7186,7 @@ void HalfFloatRow_AVX2(const uint16_t* src,
#if defined(__x86_64__)
: "x"(scale) // %3
#else
: "m"(scale) // %3
: "m"(scale) // %3
#endif
: "memory", "cc", "xmm2", "xmm3", "xmm4", "xmm5");
}
@ -7224,7 +7224,7 @@ void HalfFloatRow_F16C(const uint16_t* src,
#if defined(__x86_64__)
: "x"(scale) // %3
#else
: "m"(scale) // %3
: "m"(scale) // %3
#endif
: "memory", "cc", "xmm2", "xmm3", "xmm4");
}

64
source/scale_any.cc
View File

@ -700,28 +700,28 @@ SUH2LANY(ScaleRowUp2_Linear_16_Any_NEON,
// Scale up 2 times using bilinear filter.
// This function produces 2 rows at a time.
#define SU2BLANY(NAME, SIMD, C, MASK, PTYPE) \
void NAME(const PTYPE* src_ptr, ptrdiff_t src_stride, PTYPE* dst_ptr, \
ptrdiff_t dst_stride, int dst_width) { \
int work_width = (dst_width - 1) & ~1; \
int r = work_width & MASK; \
int n = work_width & ~MASK; \
const PTYPE* sa = src_ptr; \
const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0]) >> 2; \
db[0] = (sa[0] + 3 * sb[0]) >> 2; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(sa, sb - sa, da + 1, db - da, n); \
} \
C(sa + (n / 2), sb - sa, da + n + 1, db - da, r); \
} \
da[dst_width - 1] = \
(3 * sa[(dst_width - 1) / 2] + sb[(dst_width - 1) / 2]) >> 2; \
db[dst_width - 1] = \
(sa[(dst_width - 1) / 2] + 3 * sb[(dst_width - 1) / 2]) >> 2; \
#define SU2BLANY(NAME, SIMD, C, MASK, PTYPE) \
void NAME(const PTYPE* src_ptr, ptrdiff_t src_stride, PTYPE* dst_ptr, \
ptrdiff_t dst_stride, int dst_width) { \
int work_width = (dst_width - 1) & ~1; \
int r = work_width & MASK; \
int n = work_width & ~MASK; \
const PTYPE* sa = src_ptr; \
const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0] + 2) >> 2; \
db[0] = (sa[0] + 3 * sb[0] + 2) >> 2; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(sa, sb - sa, da + 1, db - da, n); \
} \
C(sa + (n / 2), sb - sa, da + n + 1, db - da, r); \
} \
da[dst_width - 1] = \
(3 * sa[(dst_width - 1) / 2] + sb[(dst_width - 1) / 2] + 2) >> 2; \
db[dst_width - 1] = \
(sa[(dst_width - 1) / 2] + 3 * sb[(dst_width - 1) / 2] + 2) >> 2; \
}
SU2BLANY(ScaleRowUp2_Bilinear_Any_C,
@ -856,10 +856,10 @@ SBUH2LANY(ScaleUVRowUp2_Linear_Any_NEON,
const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0]) >> 2; \
db[0] = (sa[0] + 3 * sb[0]) >> 2; \
da[1] = (3 * sa[1] + sb[1]) >> 2; \
db[1] = (sa[1] + 3 * sb[1]) >> 2; \
da[0] = (3 * sa[0] + sb[0] + 2) >> 2; \
db[0] = (sa[0] + 3 * sb[0] + 2) >> 2; \
da[1] = (3 * sa[1] + sb[1] + 2) >> 2; \
db[1] = (sa[1] + 3 * sb[1] + 2) >> 2; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(sa, sb - sa, da + 2, db - da, n); \
@ -867,13 +867,17 @@ SBUH2LANY(ScaleUVRowUp2_Linear_Any_NEON,
C(sa + n, sb - sa, da + 2 * n + 2, db - da, r); \
} \
da[2 * dst_width - 2] = (3 * sa[((dst_width + 1) & ~1) - 2] + \
sb[((dst_width + 1) & ~1) - 2]) >> 2; \
sb[((dst_width + 1) & ~1) - 2] + 2) >> \
2; \
db[2 * dst_width - 2] = (sa[((dst_width + 1) & ~1) - 2] + \
3 * sb[((dst_width + 1) & ~1) - 2]) >> 2; \
3 * sb[((dst_width + 1) & ~1) - 2] + 2) >> \
2; \
da[2 * dst_width - 1] = (3 * sa[((dst_width + 1) & ~1) - 1] + \
sb[((dst_width + 1) & ~1) - 1]) >> 2; \
sb[((dst_width + 1) & ~1) - 1] + 2) >> \
2; \
db[2 * dst_width - 1] = (sa[((dst_width + 1) & ~1) - 1] + \
3 * sb[((dst_width + 1) & ~1) - 1]) >> 2; \
3 * sb[((dst_width + 1) & ~1) - 1] + 2) >> \
2; \
}
SBU2BLANY(ScaleUVRowUp2_Bilinear_Any_C,

24
source/scale_common.cc
View File

@ -1232,21 +1232,29 @@ void ScaleUVRowUp2_Bilinear_C(const uint8_t* src_ptr,
assert((dst_width % 2 == 0) && (dst_width >= 0));
for (x = 0; x < src_width; ++x) {
d[4 * x + 0] = (s[2 * x + 0] * 9 + s[2 * x + 2] * 3 + t[2 * x + 0] * 3 +
t[2 * x + 2] * 1 + 8) >> 4;
t[2 * x + 2] * 1 + 8) >>
4;
d[4 * x + 1] = (s[2 * x + 1] * 9 + s[2 * x + 3] * 3 + t[2 * x + 1] * 3 +
t[2 * x + 3] * 1 + 8) >> 4;
t[2 * x + 3] * 1 + 8) >>
4;
d[4 * x + 2] = (s[2 * x + 0] * 3 + s[2 * x + 2] * 9 + t[2 * x + 0] * 1 +
t[2 * x + 2] * 3 + 8) >> 4;
t[2 * x + 2] * 3 + 8) >>
4;
d[4 * x + 3] = (s[2 * x + 1] * 3 + s[2 * x + 3] * 9 + t[2 * x + 1] * 1 +
t[2 * x + 3] * 3 + 8) >> 4;
t[2 * x + 3] * 3 + 8) >>
4;
e[4 * x + 0] = (s[2 * x + 0] * 3 + s[2 * x + 2] * 1 + t[2 * x + 0] * 9 +
t[2 * x + 2] * 3 + 8) >> 4;
t[2 * x + 2] * 3 + 8) >>
4;
e[4 * x + 1] = (s[2 * x + 1] * 3 + s[2 * x + 3] * 1 + t[2 * x + 1] * 9 +
t[2 * x + 3] * 3 + 8) >> 4;
t[2 * x + 3] * 3 + 8) >>
4;
e[4 * x + 2] = (s[2 * x + 0] * 1 + s[2 * x + 2] * 3 + t[2 * x + 0] * 3 +
t[2 * x + 2] * 9 + 8) >> 4;
t[2 * x + 2] * 9 + 8) >>
4;
e[4 * x + 3] = (s[2 * x + 1] * 1 + s[2 * x + 3] * 3 + t[2 * x + 1] * 3 +
t[2 * x + 3] * 9 + 8) >> 4;
t[2 * x + 3] * 9 + 8) >>
4;
}
}

188
source/scale_gcc.cc
View File

@ -196,8 +196,7 @@ void ScaleRowDown2_AVX2(const uint8_t* src_ptr,
uint8_t* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"vmovdqu (%0),%%ymm0 \n"
"vmovdqu 0x20(%0),%%ymm1 \n"
@ -211,11 +210,11 @@ void ScaleRowDown2_AVX2(const uint8_t* src_ptr,
"sub $0x20,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
}
void ScaleRowDown2Linear_AVX2(const uint8_t* src_ptr,
@ -483,8 +482,7 @@ void ScaleRowDown34_SSSE3(const uint8_t* src_ptr,
"m"(kShuf1), // %1
"m"(kShuf2) // %2
);
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm2 \n"
@ -500,11 +498,11 @@ void ScaleRowDown34_SSSE3(const uint8_t* src_ptr,
"lea 0x18(%1),%1 \n"
"sub $0x18,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5");
}
void ScaleRowDown34_1_Box_SSSE3(const uint8_t* src_ptr,
@ -529,8 +527,7 @@ void ScaleRowDown34_1_Box_SSSE3(const uint8_t* src_ptr,
"m"(kMadd11), // %1
"m"(kRound34) // %2
);
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm6 \n"
"movdqu 0x00(%0,%3,1),%%xmm7 \n"
@ -563,13 +560,13 @@ void ScaleRowDown34_1_Box_SSSE3(const uint8_t* src_ptr,
"lea 0x18(%1),%1 \n"
"sub $0x18,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3
"m"(kMadd21) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3
"m"(kMadd21) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm6", "xmm7");
}
void ScaleRowDown34_0_Box_SSSE3(const uint8_t* src_ptr,
@ -595,8 +592,7 @@ void ScaleRowDown34_0_Box_SSSE3(const uint8_t* src_ptr,
"m"(kRound34) // %2
);
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm6 \n"
"movdqu 0x00(%0,%3,1),%%xmm7 \n"
@ -632,13 +628,13 @@ void ScaleRowDown34_0_Box_SSSE3(const uint8_t* src_ptr,
"lea 0x18(%1),%1 \n"
"sub $0x18,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3
"m"(kMadd21) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3
"m"(kMadd21) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm6", "xmm7");
}
void ScaleRowDown38_SSSE3(const uint8_t* src_ptr,
@ -687,8 +683,7 @@ void ScaleRowDown38_2_Box_SSSE3(const uint8_t* src_ptr,
"m"(kShufAb2), // %2
"m"(kScaleAb2) // %3
);
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x00(%0,%3,1),%%xmm1 \n"
@ -709,11 +704,12 @@ void ScaleRowDown38_2_Box_SSSE3(const uint8_t* src_ptr,
"lea 0x6(%1),%1 \n"
"sub $0x6,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm6");
}
void ScaleRowDown38_3_Box_SSSE3(const uint8_t* src_ptr,
@ -730,8 +726,7 @@ void ScaleRowDown38_3_Box_SSSE3(const uint8_t* src_ptr,
"m"(kShufAc3), // %1
"m"(kScaleAc33) // %2
);
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x00(%0,%3,1),%%xmm6 \n"
@ -771,12 +766,12 @@ void ScaleRowDown38_3_Box_SSSE3(const uint8_t* src_ptr,
"lea 0x6(%1),%1 \n"
"sub $0x6,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm6", "xmm7");
}
#ifdef HAS_SCALEROWUP2LINEAR_SSE2
@ -1601,11 +1596,10 @@ void ScaleRowUp2_Bilinear_16_AVX2_Full(const uint16_t* src_ptr,
void ScaleAddRow_SSE2(const uint8_t* src_ptr,
uint16_t* dst_ptr,
int src_width) {
asm volatile(
"pxor %%xmm5,%%xmm5 \n"
asm volatile("pxor %%xmm5,%%xmm5 \n"
// 16 pixel loop.
LABELALIGN
// 16 pixel loop.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm3 \n"
"lea 0x10(%0),%0 \n" // src_ptr += 16
@ -1621,11 +1615,11 @@ void ScaleAddRow_SSE2(const uint8_t* src_ptr,
"lea 0x20(%1),%1 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(src_width) // %2
:
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(src_width) // %2
:
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
}
#ifdef HAS_SCALEADDROW_AVX2
@ -1633,10 +1627,9 @@ void ScaleAddRow_SSE2(const uint8_t* src_ptr,
void ScaleAddRow_AVX2(const uint8_t* src_ptr,
uint16_t* dst_ptr,
int src_width) {
asm volatile(
"vpxor %%ymm5,%%ymm5,%%ymm5 \n"
asm volatile("vpxor %%ymm5,%%ymm5,%%ymm5 \n"
LABELALIGN
LABELALIGN
"1: \n"
"vmovdqu (%0),%%ymm3 \n"
"lea 0x20(%0),%0 \n" // src_ptr += 32
@ -1651,11 +1644,11 @@ void ScaleAddRow_AVX2(const uint8_t* src_ptr,
"sub $0x20,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(src_width) // %2
:
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(src_width) // %2
:
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
}
#endif // HAS_SCALEADDROW_AVX2
@ -1772,8 +1765,7 @@ void ScaleColsUp2_SSE2(uint8_t* dst_ptr,
int dx) {
(void)x;
(void)dx;
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%1),%%xmm0 \n"
"lea 0x10(%1),%1 \n"
@ -1786,11 +1778,11 @@ void ScaleColsUp2_SSE2(uint8_t* dst_ptr,
"sub $0x20,%2 \n"
"jg 1b \n"
: "+r"(dst_ptr), // %0
"+r"(src_ptr), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
: "+r"(dst_ptr), // %0
"+r"(src_ptr), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
}
void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb,
@ -1798,8 +1790,7 @@ void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb,
uint8_t* dst_argb,
int dst_width) {
(void)src_stride;
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n"
@ -1809,11 +1800,11 @@ void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb,
"lea 0x10(%1),%1 \n"
"sub $0x4,%2 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
}
void ScaleARGBRowDown2Linear_SSE2(const uint8_t* src_argb,
@ -1821,8 +1812,7 @@ void ScaleARGBRowDown2Linear_SSE2(const uint8_t* src_argb,
uint8_t* dst_argb,
int dst_width) {
(void)src_stride;
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n"
@ -1835,19 +1825,18 @@ void ScaleARGBRowDown2Linear_SSE2(const uint8_t* src_argb,
"lea 0x10(%1),%1 \n"
"sub $0x4,%2 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
}
void ScaleARGBRowDown2Box_SSE2(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n"
@ -1864,11 +1853,11 @@ void ScaleARGBRowDown2Box_SSE2(const uint8_t* src_argb,
"lea 0x10(%1),%1 \n"
"sub $0x4,%2 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3");
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3");
}
// Reads 4 pixels at a time.
@ -2032,8 +2021,7 @@ void ScaleARGBColsUp2_SSE2(uint8_t* dst_argb,
int dx) {
(void)x;
(void)dx;
asm volatile(
LABELALIGN
asm volatile(LABELALIGN
"1: \n"
"movdqu (%1),%%xmm0 \n"
"lea 0x10(%1),%1 \n"
@ -2046,11 +2034,11 @@ void ScaleARGBColsUp2_SSE2(uint8_t* dst_argb,
"sub $0x8,%2 \n"
"jg 1b \n"
: "+r"(dst_argb), // %0
"+r"(src_argb), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
: "+r"(dst_argb), // %0
"+r"(src_argb), // %1
"+r"(dst_width) // %2
::"memory",
"cc", "xmm0", "xmm1");
}
// Shuffle table for arranging 2 pixels into pairs for pmaddubsw
@ -2381,7 +2369,7 @@ void ScaleUVRowUp2_Bilinear_SSSE3(const uint8_t* src_ptr,
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)), // %4
"m"(kUVLinearMadd31_SSSE3) // %5
"m"(kUVLinearMadd31_SSSE3) // %5
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
}

14
source/scale_neon.cc
View File

@ -194,21 +194,21 @@ void ScaleRowDown34_0_Box_NEON(const uint8_t* src_ptr,
"vmlal.u8 q10, d2, d24 \n"
"vmlal.u8 q11, d3, d24 \n"
// (3 * line_0 + line_1) >> 2
// (3 * line_0 + line_1 + 2) >> 2
"vqrshrn.u16 d0, q8, #2 \n"
"vqrshrn.u16 d1, q9, #2 \n"
"vqrshrn.u16 d2, q10, #2 \n"
"vqrshrn.u16 d3, q11, #2 \n"
// a0 = (src[0] * 3 + s[1] * 1) >> 2
// a0 = (src[0] * 3 + s[1] * 1 + 2) >> 2
"vmovl.u8 q8, d1 \n"
"vmlal.u8 q8, d0, d24 \n"
"vqrshrn.u16 d0, q8, #2 \n"
// a1 = (src[1] * 1 + s[2] * 1) >> 1
// a1 = (src[1] * 1 + s[2] * 1 + 1) >> 1
"vrhadd.u8 d1, d1, d2 \n"
// a2 = (src[2] * 1 + s[3] * 3) >> 2
// a2 = (src[2] * 1 + s[3] * 3 + 2) >> 2
"vmovl.u8 q8, d2 \n"
"vmlal.u8 q8, d3, d24 \n"
"vqrshrn.u16 d2, q8, #2 \n"
@ -240,15 +240,15 @@ void ScaleRowDown34_1_Box_NEON(const uint8_t* src_ptr,
"vrhadd.u8 q0, q0, q2 \n"
"vrhadd.u8 q1, q1, q3 \n"
// a0 = (src[0] * 3 + s[1] * 1) >> 2
// a0 = (src[0] * 3 + s[1] * 1 + 2) >> 2
"vmovl.u8 q3, d1 \n"
"vmlal.u8 q3, d0, d24 \n"
"vqrshrn.u16 d0, q3, #2 \n"
// a1 = (src[1] * 1 + s[2] * 1) >> 1
// a1 = (src[1] * 1 + s[2] * 1 + 1) >> 1
"vrhadd.u8 d1, d1, d2 \n"
// a2 = (src[2] * 1 + s[3] * 3) >> 2
// a2 = (src[2] * 1 + s[3] * 3 + 2) >> 2
"vmovl.u8 q3, d2 \n"
"vmlal.u8 q3, d3, d24 \n"
"vqrshrn.u16 d2, q3, #2 \n"

14
source/scale_neon64.cc
View File

@ -201,22 +201,22 @@ void ScaleRowDown34_0_Box_NEON(const uint8_t* src_ptr,
"umlal v19.8h, v3.8b, v20.8b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead
// (3 * line_0 + line_1) >> 2
// (3 * line_0 + line_1 + 2) >> 2
"uqrshrn v0.8b, v16.8h, #2 \n"
"uqrshrn v1.8b, v17.8h, #2 \n"
"uqrshrn v2.8b, v18.8h, #2 \n"
"uqrshrn v3.8b, v19.8h, #2 \n"
"prfm pldl1keep, [%3, 448] \n"
// a0 = (src[0] * 3 + s[1] * 1) >> 2
// a0 = (src[0] * 3 + s[1] * 1 + 2) >> 2
"ushll v16.8h, v1.8b, #0 \n"
"umlal v16.8h, v0.8b, v20.8b \n"
"uqrshrn v0.8b, v16.8h, #2 \n"
// a1 = (src[1] * 1 + s[2] * 1) >> 1
// a1 = (src[1] * 1 + s[2] * 1 + 1) >> 1
"urhadd v1.8b, v1.8b, v2.8b \n"
// a2 = (src[2] * 1 + s[3] * 3) >> 2
// a2 = (src[2] * 1 + s[3] * 3 + 2) >> 2
"ushll v16.8h, v2.8b, #0 \n"
"umlal v16.8h, v3.8b, v20.8b \n"
"uqrshrn v2.8b, v16.8h, #2 \n"
@ -251,16 +251,16 @@ void ScaleRowDown34_1_Box_NEON(const uint8_t* src_ptr,
"urhadd v3.8b, v3.8b, v7.8b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead
// a0 = (src[0] * 3 + s[1] * 1) >> 2
// a0 = (src[0] * 3 + s[1] * 1 + 2) >> 2
"ushll v4.8h, v1.8b, #0 \n"
"umlal v4.8h, v0.8b, v20.8b \n"
"uqrshrn v0.8b, v4.8h, #2 \n"
"prfm pldl1keep, [%3, 448] \n"
// a1 = (src[1] * 1 + s[2] * 1) >> 1
// a1 = (src[1] * 1 + s[2] * 1 + 1) >> 1
"urhadd v1.8b, v1.8b, v2.8b \n"
// a2 = (src[2] * 1 + s[3] * 3) >> 2
// a2 = (src[2] * 1 + s[3] * 3 + 2) >> 2
"ushll v4.8h, v2.8b, #0 \n"
"umlal v4.8h, v3.8b, v20.8b \n"
"uqrshrn v2.8b, v4.8h, #2 \n"

3
source/scale_uv.cc
View File

@ -690,8 +690,7 @@ void ScaleUVLinearUp2(int src_width,
#endif
if (dst_height == 1) {
ScaleRowUp(src_uv + ((src_height - 1) / 2) * src_stride, dst_uv,
dst_width);
ScaleRowUp(src_uv + ((src_height - 1) / 2) * src_stride, dst_uv, dst_width);
} else {
dy = FixedDiv(src_height - 1, dst_height - 1);
y = (1 << 15) - 1;

74
unit_test/color_test.cc
View File

@ -470,7 +470,7 @@ static void YUVUToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
// BT.2020 full range YUV to RGB reference
static void YUVVToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
*r = RoundToByte(y + (v - 128) * 1.474600);
*r = RoundToByte(y + (v - 128) * 1.474600);
*g = RoundToByte(y - (u - 128) * 0.164553 - (v - 128) * 0.571353);
*b = RoundToByte(y + (u - 128) * 1.881400);
}
@ -609,9 +609,15 @@ static void PrintHistogram(int rh[256], int gh[256], int bh[256]) {
// BT.601 limited range.
TEST_F(LibYUVColorTest, TestFullYUV) {
int rh[256] = { 0, };
int gh[256] = { 0, };
int bh[256] = { 0, };
int rh[256] = {
0,
};
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -633,9 +639,15 @@ TEST_F(LibYUVColorTest, TestFullYUV) {
// BT.601 full range.
TEST_F(LibYUVColorTest, TestFullYUVJ) {
int rh[256] = { 0, };
int gh[256] = { 0, };
int bh[256] = { 0, };
int rh[256] = {
0,
};
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -657,9 +669,15 @@ TEST_F(LibYUVColorTest, TestFullYUVJ) {
// BT.709 limited range.
TEST_F(LibYUVColorTest, TestFullYUVH) {
int rh[256] = { 0, };
int gh[256] = { 0, };
int bh[256] = { 0, };
int rh[256] = {
0,
};
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -682,9 +700,15 @@ TEST_F(LibYUVColorTest, TestFullYUVH) {
// BT.709 full range.
TEST_F(LibYUVColorTest, TestFullYUVF) {
int rh[256] = { 0, };
int gh[256] = { 0, };
int bh[256] = { 0, };
int rh[256] = {
0,
};
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -706,9 +730,15 @@ TEST_F(LibYUVColorTest, TestFullYUVF) {
// BT.2020 limited range.
TEST_F(LibYUVColorTest, TestFullYUVU) {
int rh[256] = { 0, };
int gh[256] = { 0, };
int bh[256] = { 0, };
int rh[256] = {
0,
};
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -731,9 +761,15 @@ TEST_F(LibYUVColorTest, TestFullYUVU) {
// BT.2020 full range.
TEST_F(LibYUVColorTest, TestFullYUVV) {
int rh[256] = { 0, };
int gh[256] = { 0, };
int bh[256] = { 0, };
int rh[256] = {
0,
};
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {

12
unit_test/convert_test.cc
View File

@ -794,10 +794,10 @@ TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
#define U420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
l, m)
#define V420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
#define V420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I420AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
l, m)
#define V420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
#define V420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
l, m)
#define J422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
@ -824,10 +824,10 @@ TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
#define U422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
l, m)
#define V422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
#define V422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I422AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
l, m)
#define V422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
#define V422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
l, m)
#define J444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
@ -854,10 +854,10 @@ TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
#define U444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
l, m)
#define V444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
#define V444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I444AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
l, m)
#define V444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
#define V444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m) \
I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
l, m)

36
util/color.cc
View File

@ -18,11 +18,15 @@
// For those MCs that can be represented as kr and kb:
// Full range
// float M[3][3] {{1,0,2*(1-kr)},{1,-((2*kb)/((2-kb)*(1-kb-kr))),-((2*kr)/((2-kr)*(1-kb-kr)))},{1,2*(1-kb),0}};
// float B[3] {1+(256*(1-kr))/255,1-(256*kb)/(255*(2-kb)*(1-kb-kr))-(256*kr)/(255*(2-kr)*(1-kb-kr)),1+(256*(1-kb))/255};
// float M[3][3]
// {{1,0,2*(1-kr)},{1,-((2*kb)/((2-kb)*(1-kb-kr))),-((2*kr)/((2-kr)*(1-kb-kr)))},{1,2*(1-kb),0}};
// float B[3]
// {1+(256*(1-kr))/255,1-(256*kb)/(255*(2-kb)*(1-kb-kr))-(256*kr)/(255*(2-kr)*(1-kb-kr)),1+(256*(1-kb))/255};
// Limited range
// float M[3][3] {{85/73,0,255/112-(255*kr)/112},{85/73,-((255*kb)/(112*(2-kb)*(1-kb-kr))),-((255*kr)/(112*(2-kr)*(1-kb-kr)))},{85/73,255/112-(255*kb)/112,0}};
// float B[3] {77662/43435-(1537*kr)/1785,203/219-(1537*kb)/(1785*(2-kb)*(1-kb-kr))-(1537*kr)/(1785*(2-kr)*(1-kb-kr)),77662/43435-(1537*kb)/1785};
// float M[3][3]
// {{85/73,0,255/112-(255*kr)/112},{85/73,-((255*kb)/(112*(2-kb)*(1-kb-kr))),-((255*kr)/(112*(2-kr)*(1-kb-kr)))},{85/73,255/112-(255*kb)/112,0}};
// float B[3]
// {77662/43435-(1537*kr)/1785,203/219-(1537*kb)/(1785*(2-kb)*(1-kb-kr))-(1537*kr)/(1785*(2-kr)*(1-kb-kr)),77662/43435-(1537*kb)/1785};
// mc bt
// 1 bt.709 KR = 0.2126; KB = 0.0722
@ -56,11 +60,10 @@
// #define BR (-VR * 128 + YB)
int round(float v) {
return (int) (v + 0.5);
return (int)(v + 0.5);
}
int main(int argc, const char* argv[]) {
if (argc < 2) {
printf("color kr kb\n");
return -1;
@ -81,11 +84,11 @@ int main(int argc, const char* argv[]) {
printf("KR = %4f; ", kr);
printf("KB = %4f\n", kb);
// printf("KG = %4f\n", kg);
// #define YG 16320 /* round(1.000 * 64 * 256 * 256 / 257) */
// #define YB 32 /* 64 / 2 */
//
// // U and V contributions to R,G,B.
// printf("KG = %4f\n", kg);
// #define YG 16320 /* round(1.000 * 64 * 256 * 256 / 257) */
// #define YB 32 /* 64 / 2 */
//
// // U and V contributions to R,G,B.
printf("UB %-3d /* round(%f * 64) */\n", round(ub * 64), ub);
printf("UG %-3d /* round(%f * 64) */\n", round(ug * 64), ug);
@ -102,11 +105,11 @@ int main(int argc, const char* argv[]) {
printf("G = (Y - 16) * 1.164 - U * %6f - V * %6f\n", ug, vg);
printf("B = (Y - 16) * 1.164 + U * %5f\n", ub);
// printf("KG = %4f\n", kg);
// #define YG 16320 /* round(1.000 * 64 * 256 * 256 / 257) */
// #define YB 32 /* 64 / 2 */
//
// // U and V contributions to R,G,B.
// printf("KG = %4f\n", kg);
// #define YG 16320 /* round(1.000 * 64 * 256 * 256 / 257) */
// #define YB 32 /* 64 / 2 */
//
// // U and V contributions to R,G,B.
printf("UB %-3d /* round(%f * 64) */\n", round(ub * 64), ub);
printf("UG %-3d /* round(%f * 64) */\n", round(ug * 64), ug);
@ -115,4 +118,3 @@ int main(int argc, const char* argv[]) {
return 0;
}