coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +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

28
source/scale_any.cc
View File

@ -710,8 +710,8 @@ SUH2LANY(ScaleRowUp2_Linear_16_Any_NEON,
const PTYPE* sb = src_ptr + src_stride; \ const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \ PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \ PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0]) >> 2; \ da[0] = (3 * sa[0] + sb[0] + 2) >> 2; \
db[0] = (sa[0] + 3 * sb[0]) >> 2; \ db[0] = (sa[0] + 3 * sb[0] + 2) >> 2; \
if (work_width > 0) { \ if (work_width > 0) { \
if (n != 0) { \ if (n != 0) { \
SIMD(sa, sb - sa, da + 1, db - da, n); \ SIMD(sa, sb - sa, da + 1, db - da, n); \
@ -719,9 +719,9 @@ SUH2LANY(ScaleRowUp2_Linear_16_Any_NEON,
C(sa + (n / 2), sb - sa, da + n + 1, db - da, r); \ C(sa + (n / 2), sb - sa, da + n + 1, db - da, r); \
} \ } \
da[dst_width - 1] = \ da[dst_width - 1] = \
(3 * sa[(dst_width - 1) / 2] + sb[(dst_width - 1) / 2]) >> 2; \ (3 * sa[(dst_width - 1) / 2] + sb[(dst_width - 1) / 2] + 2) >> 2; \
db[dst_width - 1] = \ db[dst_width - 1] = \
(sa[(dst_width - 1) / 2] + 3 * sb[(dst_width - 1) / 2]) >> 2; \ (sa[(dst_width - 1) / 2] + 3 * sb[(dst_width - 1) / 2] + 2) >> 2; \
} }
SU2BLANY(ScaleRowUp2_Bilinear_Any_C, SU2BLANY(ScaleRowUp2_Bilinear_Any_C,
@ -856,10 +856,10 @@ SBUH2LANY(ScaleUVRowUp2_Linear_Any_NEON,
const PTYPE* sb = src_ptr + src_stride; \ const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \ PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \ PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0]) >> 2; \ da[0] = (3 * sa[0] + sb[0] + 2) >> 2; \
db[0] = (sa[0] + 3 * sb[0]) >> 2; \ db[0] = (sa[0] + 3 * sb[0] + 2) >> 2; \
da[1] = (3 * sa[1] + sb[1]) >> 2; \ da[1] = (3 * sa[1] + sb[1] + 2) >> 2; \
db[1] = (sa[1] + 3 * sb[1]) >> 2; \ db[1] = (sa[1] + 3 * sb[1] + 2) >> 2; \
if (work_width > 0) { \ if (work_width > 0) { \
if (n != 0) { \ if (n != 0) { \
SIMD(sa, sb - sa, da + 2, db - da, n); \ 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); \ C(sa + n, sb - sa, da + 2 * n + 2, db - da, r); \
} \ } \
da[2 * dst_width - 2] = (3 * sa[((dst_width + 1) & ~1) - 2] + \ 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] + \ 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] + \ 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] + \ 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, 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)); assert((dst_width % 2 == 0) && (dst_width >= 0));
for (x = 0; x < src_width; ++x) { 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 + 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 + 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 + 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 + 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 + 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 + 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 + 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 + 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;
} }
} }

54
source/scale_gcc.cc
View File

@ -196,8 +196,7 @@ void ScaleRowDown2_AVX2(const uint8_t* src_ptr,
uint8_t* dst_ptr, uint8_t* dst_ptr,
int dst_width) { int dst_width) {
(void)src_stride; (void)src_stride;
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"vmovdqu (%0),%%ymm0 \n" "vmovdqu (%0),%%ymm0 \n"
"vmovdqu 0x20(%0),%%ymm1 \n" "vmovdqu 0x20(%0),%%ymm1 \n"
@ -483,8 +482,7 @@ void ScaleRowDown34_SSSE3(const uint8_t* src_ptr,
"m"(kShuf1), // %1 "m"(kShuf1), // %1
"m"(kShuf2) // %2 "m"(kShuf2) // %2
); );
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm0 \n" "movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm2 \n" "movdqu 0x10(%0),%%xmm2 \n"
@ -529,8 +527,7 @@ void ScaleRowDown34_1_Box_SSSE3(const uint8_t* src_ptr,
"m"(kMadd11), // %1 "m"(kMadd11), // %1
"m"(kRound34) // %2 "m"(kRound34) // %2
); );
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm6 \n" "movdqu (%0),%%xmm6 \n"
"movdqu 0x00(%0,%3,1),%%xmm7 \n" "movdqu 0x00(%0,%3,1),%%xmm7 \n"
@ -568,8 +565,8 @@ void ScaleRowDown34_1_Box_SSSE3(const uint8_t* src_ptr,
"+r"(dst_width) // %2 "+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3 : "r"((intptr_t)(src_stride)), // %3
"m"(kMadd21) // %4 "m"(kMadd21) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm7"); "xmm6", "xmm7");
} }
void ScaleRowDown34_0_Box_SSSE3(const uint8_t* src_ptr, 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 "m"(kRound34) // %2
); );
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm6 \n" "movdqu (%0),%%xmm6 \n"
"movdqu 0x00(%0,%3,1),%%xmm7 \n" "movdqu 0x00(%0,%3,1),%%xmm7 \n"
@ -637,8 +633,8 @@ void ScaleRowDown34_0_Box_SSSE3(const uint8_t* src_ptr,
"+r"(dst_width) // %2 "+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3 : "r"((intptr_t)(src_stride)), // %3
"m"(kMadd21) // %4 "m"(kMadd21) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm7"); "xmm6", "xmm7");
} }
void ScaleRowDown38_SSSE3(const uint8_t* src_ptr, 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"(kShufAb2), // %2
"m"(kScaleAb2) // %3 "m"(kScaleAb2) // %3
); );
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm0 \n" "movdqu (%0),%%xmm0 \n"
"movdqu 0x00(%0,%3,1),%%xmm1 \n" "movdqu 0x00(%0,%3,1),%%xmm1 \n"
@ -713,7 +708,8 @@ void ScaleRowDown38_2_Box_SSSE3(const uint8_t* src_ptr,
"+r"(dst_ptr), // %1 "+r"(dst_ptr), // %1
"+r"(dst_width) // %2 "+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3 : "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"); : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm6");
} }
void ScaleRowDown38_3_Box_SSSE3(const uint8_t* src_ptr, 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"(kShufAc3), // %1
"m"(kScaleAc33) // %2 "m"(kScaleAc33) // %2
); );
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm0 \n" "movdqu (%0),%%xmm0 \n"
"movdqu 0x00(%0,%3,1),%%xmm6 \n" "movdqu 0x00(%0,%3,1),%%xmm6 \n"
@ -775,8 +770,8 @@ void ScaleRowDown38_3_Box_SSSE3(const uint8_t* src_ptr,
"+r"(dst_ptr), // %1 "+r"(dst_ptr), // %1
"+r"(dst_width) // %2 "+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)) // %3 : "r"((intptr_t)(src_stride)) // %3
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5",
"xmm7"); "xmm6", "xmm7");
} }
#ifdef HAS_SCALEROWUP2LINEAR_SSE2 #ifdef HAS_SCALEROWUP2LINEAR_SSE2
@ -1601,8 +1596,7 @@ void ScaleRowUp2_Bilinear_16_AVX2_Full(const uint16_t* src_ptr,
void ScaleAddRow_SSE2(const uint8_t* src_ptr, void ScaleAddRow_SSE2(const uint8_t* src_ptr,
uint16_t* dst_ptr, uint16_t* dst_ptr,
int src_width) { int src_width) {
asm volatile( asm volatile("pxor %%xmm5,%%xmm5 \n"
"pxor %%xmm5,%%xmm5 \n"
// 16 pixel loop. // 16 pixel loop.
LABELALIGN LABELALIGN
@ -1633,8 +1627,7 @@ void ScaleAddRow_SSE2(const uint8_t* src_ptr,
void ScaleAddRow_AVX2(const uint8_t* src_ptr, void ScaleAddRow_AVX2(const uint8_t* src_ptr,
uint16_t* dst_ptr, uint16_t* dst_ptr,
int src_width) { int src_width) {
asm volatile( asm volatile("vpxor %%ymm5,%%ymm5,%%ymm5 \n"
"vpxor %%ymm5,%%ymm5,%%ymm5 \n"
LABELALIGN LABELALIGN
"1: \n" "1: \n"
@ -1772,8 +1765,7 @@ void ScaleColsUp2_SSE2(uint8_t* dst_ptr,
int dx) { int dx) {
(void)x; (void)x;
(void)dx; (void)dx;
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%1),%%xmm0 \n" "movdqu (%1),%%xmm0 \n"
"lea 0x10(%1),%1 \n" "lea 0x10(%1),%1 \n"
@ -1798,8 +1790,7 @@ void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb,
uint8_t* dst_argb, uint8_t* dst_argb,
int dst_width) { int dst_width) {
(void)src_stride; (void)src_stride;
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm0 \n" "movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n" "movdqu 0x10(%0),%%xmm1 \n"
@ -1821,8 +1812,7 @@ void ScaleARGBRowDown2Linear_SSE2(const uint8_t* src_argb,
uint8_t* dst_argb, uint8_t* dst_argb,
int dst_width) { int dst_width) {
(void)src_stride; (void)src_stride;
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm0 \n" "movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n" "movdqu 0x10(%0),%%xmm1 \n"
@ -1846,8 +1836,7 @@ void ScaleARGBRowDown2Box_SSE2(const uint8_t* src_argb,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_argb, uint8_t* dst_argb,
int dst_width) { int dst_width) {
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm0 \n" "movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n" "movdqu 0x10(%0),%%xmm1 \n"
@ -2032,8 +2021,7 @@ void ScaleARGBColsUp2_SSE2(uint8_t* dst_argb,
int dx) { int dx) {
(void)x; (void)x;
(void)dx; (void)dx;
asm volatile( asm volatile(LABELALIGN
LABELALIGN
"1: \n" "1: \n"
"movdqu (%1),%%xmm0 \n" "movdqu (%1),%%xmm0 \n"
"lea 0x10(%1),%1 \n" "lea 0x10(%1),%1 \n"

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 q10, d2, d24 \n"
"vmlal.u8 q11, d3, 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 d0, q8, #2 \n"
"vqrshrn.u16 d1, q9, #2 \n" "vqrshrn.u16 d1, q9, #2 \n"
"vqrshrn.u16 d2, q10, #2 \n" "vqrshrn.u16 d2, q10, #2 \n"
"vqrshrn.u16 d3, q11, #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" "vmovl.u8 q8, d1 \n"
"vmlal.u8 q8, d0, d24 \n" "vmlal.u8 q8, d0, d24 \n"
"vqrshrn.u16 d0, q8, #2 \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" "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" "vmovl.u8 q8, d2 \n"
"vmlal.u8 q8, d3, d24 \n" "vmlal.u8 q8, d3, d24 \n"
"vqrshrn.u16 d2, q8, #2 \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 q0, q0, q2 \n"
"vrhadd.u8 q1, q1, q3 \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" "vmovl.u8 q3, d1 \n"
"vmlal.u8 q3, d0, d24 \n" "vmlal.u8 q3, d0, d24 \n"
"vqrshrn.u16 d0, q3, #2 \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" "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" "vmovl.u8 q3, d2 \n"
"vmlal.u8 q3, d3, d24 \n" "vmlal.u8 q3, d3, d24 \n"
"vqrshrn.u16 d2, q3, #2 \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" "umlal v19.8h, v3.8b, v20.8b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead "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 v0.8b, v16.8h, #2 \n"
"uqrshrn v1.8b, v17.8h, #2 \n" "uqrshrn v1.8b, v17.8h, #2 \n"
"uqrshrn v2.8b, v18.8h, #2 \n" "uqrshrn v2.8b, v18.8h, #2 \n"
"uqrshrn v3.8b, v19.8h, #2 \n" "uqrshrn v3.8b, v19.8h, #2 \n"
"prfm pldl1keep, [%3, 448] \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" "ushll v16.8h, v1.8b, #0 \n"
"umlal v16.8h, v0.8b, v20.8b \n" "umlal v16.8h, v0.8b, v20.8b \n"
"uqrshrn v0.8b, v16.8h, #2 \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" "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" "ushll v16.8h, v2.8b, #0 \n"
"umlal v16.8h, v3.8b, v20.8b \n" "umlal v16.8h, v3.8b, v20.8b \n"
"uqrshrn v2.8b, v16.8h, #2 \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" "urhadd v3.8b, v3.8b, v7.8b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead "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" "ushll v4.8h, v1.8b, #0 \n"
"umlal v4.8h, v0.8b, v20.8b \n" "umlal v4.8h, v0.8b, v20.8b \n"
"uqrshrn v0.8b, v4.8h, #2 \n" "uqrshrn v0.8b, v4.8h, #2 \n"
"prfm pldl1keep, [%3, 448] \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" "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" "ushll v4.8h, v2.8b, #0 \n"
"umlal v4.8h, v3.8b, v20.8b \n" "umlal v4.8h, v3.8b, v20.8b \n"
"uqrshrn v2.8b, v4.8h, #2 \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 #endif
if (dst_height == 1) { if (dst_height == 1) {
ScaleRowUp(src_uv + ((src_height - 1) / 2) * src_stride, dst_uv, ScaleRowUp(src_uv + ((src_height - 1) / 2) * src_stride, dst_uv, dst_width);
dst_width);
} else { } else {
dy = FixedDiv(src_height - 1, dst_height - 1); dy = FixedDiv(src_height - 1, dst_height - 1);
y = (1 << 15) - 1; y = (1 << 15) - 1;

72
unit_test/color_test.cc
View File

@ -609,9 +609,15 @@ static void PrintHistogram(int rh[256], int gh[256], int bh[256]) {
// BT.601 limited range. // BT.601 limited range.
TEST_F(LibYUVColorTest, TestFullYUV) { TEST_F(LibYUVColorTest, TestFullYUV) {
int rh[256] = { 0, }; int rh[256] = {
int gh[256] = { 0, }; 0,
int bh[256] = { 0, }; };
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) { for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) { for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) { for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -633,9 +639,15 @@ TEST_F(LibYUVColorTest, TestFullYUV) {
// BT.601 full range. // BT.601 full range.
TEST_F(LibYUVColorTest, TestFullYUVJ) { TEST_F(LibYUVColorTest, TestFullYUVJ) {
int rh[256] = { 0, }; int rh[256] = {
int gh[256] = { 0, }; 0,
int bh[256] = { 0, }; };
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) { for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) { for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) { for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -657,9 +669,15 @@ TEST_F(LibYUVColorTest, TestFullYUVJ) {
// BT.709 limited range. // BT.709 limited range.
TEST_F(LibYUVColorTest, TestFullYUVH) { TEST_F(LibYUVColorTest, TestFullYUVH) {
int rh[256] = { 0, }; int rh[256] = {
int gh[256] = { 0, }; 0,
int bh[256] = { 0, }; };
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) { for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) { for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) { for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -682,9 +700,15 @@ TEST_F(LibYUVColorTest, TestFullYUVH) {
// BT.709 full range. // BT.709 full range.
TEST_F(LibYUVColorTest, TestFullYUVF) { TEST_F(LibYUVColorTest, TestFullYUVF) {
int rh[256] = { 0, }; int rh[256] = {
int gh[256] = { 0, }; 0,
int bh[256] = { 0, }; };
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) { for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) { for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) { for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -706,9 +730,15 @@ TEST_F(LibYUVColorTest, TestFullYUVF) {
// BT.2020 limited range. // BT.2020 limited range.
TEST_F(LibYUVColorTest, TestFullYUVU) { TEST_F(LibYUVColorTest, TestFullYUVU) {
int rh[256] = { 0, }; int rh[256] = {
int gh[256] = { 0, }; 0,
int bh[256] = { 0, }; };
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) { for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) { for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) { for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
@ -731,9 +761,15 @@ TEST_F(LibYUVColorTest, TestFullYUVU) {
// BT.2020 full range. // BT.2020 full range.
TEST_F(LibYUVColorTest, TestFullYUVV) { TEST_F(LibYUVColorTest, TestFullYUVV) {
int rh[256] = { 0, }; int rh[256] = {
int gh[256] = { 0, }; 0,
int bh[256] = { 0, }; };
int gh[256] = {
0,
};
int bh[256] = {
0,
};
for (int u = 0; u < 256; ++u) { for (int u = 0; u < 256; ++u) {
for (int v = 0; v < 256; ++v) { for (int v = 0; v < 256; ++v) {
for (int y2 = 0; y2 < 256; y2 += FASTSTEP) { for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {

14
util/color.cc
View File

@ -18,11 +18,15 @@
// For those MCs that can be represented as kr and kb: // For those MCs that can be represented as kr and kb:
// Full range // 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 M[3][3]
// 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}; // {{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 // 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 M[3][3]
// 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}; // {{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 // mc bt
// 1 bt.709 KR = 0.2126; KB = 0.0722 // 1 bt.709 KR = 0.2126; KB = 0.0722
@ -60,7 +64,6 @@ int round(float v) {
} }
int main(int argc, const char* argv[]) { int main(int argc, const char* argv[]) {
if (argc < 2) { if (argc < 2) {
printf("color kr kb\n"); printf("color kr kb\n");
return -1; return -1;
@ -115,4 +118,3 @@ int main(int argc, const char* argv[]) {
return 0; return 0;
} }