coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-07 17:26:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add MSA optimized ARGBToRGB24Row_MSA and ARGBToRAWRow_MSA functions

Browse Source 
R=fbarchard@google.com
BUG=libyuv:634

Review URL: https://codereview.chromium.org/2487913004 .
This commit is contained in:
Frank Barchard 2016-11-18 15:05:10 -08:00
parent 97fb18b846
commit b1504a8e48
4 changed files with 92 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
include/libyuv/row.h
View File

@ -384,6 +384,8 @@ extern "C" {
#define HAS_I422TORGBAROW_MSA
#define HAS_I422ALPHATOARGBROW_MSA
#define HAS_I422TORGB24ROW_MSA
#define HAS_ARGBTORGB24ROW_MSA
#define HAS_ARGBTORAWROW_MSA
#endif
#if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
@ -1304,6 +1306,8 @@ void ARGBToRGB565DitherRow_NEON(const uint8* src_argb,
uint8* dst_rgb,
const uint32 dither4,
int width);
void ARGBToRGB24Row_MSA(const uint8* src_argb, uint8* dst_rgb, int width);
void ARGBToRAWRow_MSA(const uint8* src_argb, uint8* dst_rgb, int width);
void ARGBToRGBARow_C(const uint8* src_argb, uint8* dst_rgb, int width);
void ARGBToRGB24Row_C(const uint8* src_argb, uint8* dst_rgb, int width);
@ -1892,6 +1896,8 @@ void ARGBToRGB565DitherRow_Any_NEON(const uint8* src_argb,
uint8* dst_rgb,
const uint32 dither4,
int width);
void ARGBToRGB24Row_Any_MSA(const uint8* src_argb, uint8* dst_rgb, int width);
void ARGBToRAWRow_Any_MSA(const uint8* src_argb, uint8* dst_rgb, int width);
void I444ToARGBRow_Any_NEON(const uint8* src_y,
const uint8* src_u,

16
source/convert_from_argb.cc
View File

@ -841,6 +841,14 @@ int ARGBToRGB24(const uint8* src_argb,
}
}
#endif
#if defined(HAS_ARGBTORGB24ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ARGBToRGB24Row = ARGBToRGB24Row_Any_MSA;
if (IS_ALIGNED(width, 16)) {
ARGBToRGB24Row = ARGBToRGB24Row_MSA;
}
}
#endif
for (y = 0; y < height; ++y) {
ARGBToRGB24Row(src_argb, dst_rgb24, width);
@ -891,6 +899,14 @@ int ARGBToRAW(const uint8* src_argb,
}
}
#endif
#if defined(HAS_ARGBTORAWROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ARGBToRAWRow = ARGBToRAWRow_Any_MSA;
if (IS_ALIGNED(width, 16)) {
ARGBToRAWRow = ARGBToRAWRow_MSA;
}
}
#endif
for (y = 0; y < height; ++y) {
ARGBToRAWRow(src_argb, dst_raw, width);

4
source/row_any.cc
View File

@ -380,6 +380,10 @@ ANY11(ARGBToARGB4444Row_Any_NEON, ARGBToARGB4444Row_NEON, 0, 4, 2, 7)
ANY11(J400ToARGBRow_Any_NEON, J400ToARGBRow_NEON, 0, 1, 4, 7)
ANY11(I400ToARGBRow_Any_NEON, I400ToARGBRow_NEON, 0, 1, 4, 7)
#endif
#if defined(HAS_ARGBTORGB24ROW_MSA)
ANY11(ARGBToRGB24Row_Any_MSA, ARGBToRGB24Row_MSA, 0, 4, 3, 15)
ANY11(ARGBToRAWRow_Any_MSA, ARGBToRAWRow_MSA, 0, 4, 3, 15)
#endif
#if defined(HAS_RAWTORGB24ROW_NEON)
ANY11(RAWToRGB24Row_Any_NEON, RAWToRGB24Row_NEON, 0, 3, 3, 7)
#endif

84
source/row_msa.cc
View File

@ -229,12 +229,12 @@ void I422ToARGBRow_MSA(const uint8* src_y,
}
}
void YUVTORGBARow_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
void I422ToRGBARow_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
int x;
v16u8 src0, src1, src2;
v8i16 vec0, vec1, vec2;
@ -289,12 +289,12 @@ void I422AlphaToARGBRow_MSA(const uint8* src_y,
}
}
void YUVTORGB24Row_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int32 width) {
void I422ToRGB24Row_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int32 width) {
int x;
int64 data_u, data_v;
v16u8 src0, src1, src2, src3, src4, src5, dst0, dst1, dst2;
@ -340,12 +340,12 @@ void YUVTORGB24Row_MSA(const uint8* src_y,
}
// TODO(fbarchard): Consider AND instead of shift to isolate 5 upper bits of R.
void YUVTORGB565Row_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_rgb565,
const struct YuvConstants* yuvconstants,
int width) {
void I422ToRGB565Row_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_rgb565,
const struct YuvConstants* yuvconstants,
int width) {
int x;
v16u8 src0, src1, src2, dst0;
v8i16 vec0, vec1, vec2;
@ -740,6 +740,54 @@ void ARGBToUVRow_MSA(const uint8* src_argb0,
}
}
void ARGBToRGB24Row_MSA(const uint8* src_argb, uint8* dst_rgb, int width) {
int x;
v16u8 src0, src1, src2, src3, dst0, dst1, dst2;
v16i8 shuffler0 = {0, 1, 2, 4, 5, 6, 8, 9, 10, 12, 13, 14, 16, 17, 18, 20};
v16i8 shuffler1 = {5, 6, 8, 9, 10, 12, 13, 14,
16, 17, 18, 20, 21, 22, 24, 25};
v16i8 shuffler2 = {10, 12, 13, 14, 16, 17, 18, 20,
21, 22, 24, 25, 26, 28, 29, 30};
for (x = 0; x < width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)src_argb, 48);
dst0 = (v16u8)__msa_vshf_b(shuffler0, (v16i8)src1, (v16i8)src0);
dst1 = (v16u8)__msa_vshf_b(shuffler1, (v16i8)src2, (v16i8)src1);
dst2 = (v16u8)__msa_vshf_b(shuffler2, (v16i8)src3, (v16i8)src2);
ST_UB2(dst0, dst1, dst_rgb, 16);
ST_UB(dst2, (dst_rgb + 32));
src_argb += 64;
dst_rgb += 48;
}
}
void ARGBToRAWRow_MSA(const uint8* src_argb, uint8* dst_rgb, int width) {
int x;
v16u8 src0, src1, src2, src3, dst0, dst1, dst2;
v16i8 shuffler0 = {2, 1, 0, 6, 5, 4, 10, 9, 8, 14, 13, 12, 18, 17, 16, 22};
v16i8 shuffler1 = {5, 4, 10, 9, 8, 14, 13, 12,
18, 17, 16, 22, 21, 20, 26, 25};
v16i8 shuffler2 = {8, 14, 13, 12, 18, 17, 16, 22,
21, 20, 26, 25, 24, 30, 29, 28};
for (x = 0; x < width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_argb, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)src_argb, 48);
dst0 = (v16u8)__msa_vshf_b(shuffler0, (v16i8)src1, (v16i8)src0);
dst1 = (v16u8)__msa_vshf_b(shuffler1, (v16i8)src2, (v16i8)src1);
dst2 = (v16u8)__msa_vshf_b(shuffler2, (v16i8)src3, (v16i8)src2);
ST_UB2(dst0, dst1, dst_rgb, 16);
ST_UB(dst2, (dst_rgb + 32));
src_argb += 64;
dst_rgb += 48;
}
}
void ARGB4444ToARGBRow_MSA(const uint8* src_argb4444,
uint8* dst_argb,
int width) {