Skip to content

L
libyuv
Commit 10ce829b authored by Frank Barchard's avatar Frank Barchard
Browse files
Options

Add MSA optimized I422ToRGB565Row_MSA, I422ToARGB4444Row_MSA and I422ToARGB1555Row_MSA functions 

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

Performance Gain (vs C vectorized)
I422ToRGB565Row_MSA             : ~1.5x
I422ToRGB565Row_Any_MSA         : ~1.5x
I422ToARGB4444Row_MSA           : ~1.4x
I422ToARGB4444Row_Any_MSA       : ~1.4x
I422ToARGB1555Row_MSA           : ~1.4x
I422ToARGB1555Row_Any_MSA       : ~1.4x

Performance Gain (vs C non-vectorized)
I422ToRGB565Row_MSA             : ~6.8x
I422ToRGB565Row_Any_MSA         : ~6.8x
I422ToARGB4444Row_MSA           : ~6.6x
I422ToARGB4444Row_Any_MSA       : ~6.6x
I422ToARGB1555Row_MSA           : ~6.6x
I422ToARGB1555Row_Any_MSA       : ~6.6x

Review URL: https://codereview.chromium.org/2445343007 .
parent 532f5708
Show whitespace changes
Inline Side-by-side
Showing with 206 additions and 58 deletions
include/libyuv/row.h
View file @ 10ce829b
......@@ -678,6 +678,24 @@ void I422ToRGB24Row_MSA(const uint8* src_y,
uint8* dst_rgb24,
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);
void I422ToARGB4444Row_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb4444,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGB1555Row_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb1555,
const struct YuvConstants* yuvconstants,
int width);
void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width);
void ARGBToYRow_Any_AVX2(const uint8* src_argb, uint8* dst_y, int width);
......@@ -1681,6 +1699,24 @@ void I422ToRGB24Row_Any_MSA(const uint8* src_y,
uint8* dst_rgb24,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGB565Row_Any_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_rgb565,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGB4444Row_Any_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb4444,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGB1555Row_Any_MSA(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb1555,
const struct YuvConstants* yuvconstants,
int width);
void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width);
void YUY2ToUVRow_AVX2(const uint8* src_yuy2, int stride_yuy2,
......
source/convert_from.cc
View file @ 10ce829b
......@@ -657,6 +657,14 @@ int I420ToARGB1555(const uint8* src_y, int src_stride_y,
}
}
#endif
#if defined(HAS_I422TOARGB1555ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToARGB1555Row = I422ToARGB1555Row_Any_MSA;
if (IS_ALIGNED(width, 8)) {
I422ToARGB1555Row = I422ToARGB1555Row_MSA;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToARGB1555Row(src_y, src_u, src_v, dst_argb1555, &kYuvI601Constants,
......@@ -720,6 +728,14 @@ int I420ToARGB4444(const uint8* src_y, int src_stride_y,
}
}
#endif
#if defined(HAS_I422TOARGB4444ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToARGB4444Row = I422ToARGB4444Row_Any_MSA;
if (IS_ALIGNED(width, 8)) {
I422ToARGB4444Row = I422ToARGB4444Row_MSA;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToARGB4444Row(src_y, src_u, src_v, dst_argb4444, &kYuvI601Constants,
......@@ -782,6 +798,14 @@ int I420ToRGB565(const uint8* src_y, int src_stride_y,
}
}
#endif
#if defined(HAS_I422TORGB565ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToRGB565Row = I422ToRGB565Row_Any_MSA;
if (IS_ALIGNED(width, 8)) {
I422ToRGB565Row = I422ToRGB565Row_MSA;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToRGB565Row(src_y, src_u, src_v, dst_rgb565, &kYuvI601Constants, width);
......
source/row_any.cc
View file @ 10ce829b
......@@ -172,6 +172,9 @@ ANY31C(I422ToRGB565Row_Any_NEON, I422ToRGB565Row_NEON, 1, 0, 2, 7)
ANY31C(I422ToARGBRow_Any_MSA, I422ToARGBRow_MSA, 1, 0, 4, 7)
ANY31C(I422ToRGBARow_Any_MSA, I422ToRGBARow_MSA, 1, 0, 4, 7)
ANY31C(I422ToRGB24Row_Any_MSA, I422ToRGB24Row_MSA, 1, 0, 3, 15)
ANY31C(I422ToARGB4444Row_Any_MSA, I422ToARGB4444Row_MSA, 1, 0, 2, 7)
ANY31C(I422ToARGB1555Row_Any_MSA, I422ToARGB1555Row_MSA, 1, 0, 2, 7)
ANY31C(I422ToRGB565Row_Any_MSA, I422ToRGB565Row_MSA, 1, 0, 2, 7)
#endif
#undef ANY31C
......
source/row_msa.cc
View file @ 10ce829b
......@@ -19,6 +19,18 @@ namespace libyuv {
extern "C" {
#endif
// Fill YUV -> RGB conversion constants into vectors
#define YUVTORGB_SETUP(yuvconst, ub, vr, ug, vg, bb, bg, br, yg) { \
ub = __msa_fill_w(yuvconst->kUVToB[0]); \
vr = __msa_fill_w(yuvconst->kUVToR[1]); \
ug = __msa_fill_w(yuvconst->kUVToG[0]); \
vg = __msa_fill_w(yuvconst->kUVToG[1]); \
bb = __msa_fill_w(yuvconst->kUVBiasB[0]); \
bg = __msa_fill_w(yuvconst->kUVBiasG[0]); \
br = __msa_fill_w(yuvconst->kUVBiasR[0]); \
yg = __msa_fill_w(yuvconst->kYToRgb[0]); \
}
// Load YUV 422 pixel data
#define READYUV422(psrc_y, psrc_u, psrc_v, out_y, out_u, out_v) { \
uint64 y_m; \
......@@ -33,7 +45,7 @@ extern "C" {
}
// Convert 8 pixels of YUV 420 to RGB.
#define I422TORGB(in_y, in_u, in_v, \
#define YUVTORGB(in_y, in_u, in_v, \
ub, vr, ug, vg, bb, bg, br, yg, \
out_b, out_g, out_r) { \
v8i16 vec0_m; \
......@@ -94,6 +106,17 @@ extern "C" {
out_r = __msa_pckev_h((v8i16) reg5_m, (v8i16) reg4_m); \
}
// Pack and Store 8 ARGB values.
#define STOREARGB(in0, in1, in2, in3, pdst_argb) { \
v8i16 vec0_m, vec1_m; \
v16u8 dst0_m, dst1_m; \
vec0_m = (v8i16) __msa_ilvev_b((v16i8) in1, (v16i8) in0); \
vec1_m = (v8i16) __msa_ilvev_b((v16i8) in3, (v16i8) in2); \
dst0_m = (v16u8) __msa_ilvr_h(vec1_m, vec0_m); \
dst1_m = (v16u8) __msa_ilvl_h(vec1_m, vec0_m); \
ST_UB2(dst0_m, dst1_m, pdst_argb, 16); \
}
void MirrorRow_MSA(const uint8* src, uint8* dst, int width) {
int x;
v16u8 src0, src1, src2, src3;
......@@ -180,29 +203,19 @@ void I422ToARGBRow_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, dst0, dst1;
v16u8 src0, src1, src2;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v16u8 const_255 = (v16u8) __msa_ldi_b(255);
vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]);
vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]);
vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]);
vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]);
vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]);
vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]);
vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]);
vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2);
I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2);
vec0 = (v8i16) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0);
vec1 = (v8i16) __msa_ilvev_b((v16i8) const_255, (v16i8) vec2);
dst0 = (v16u8) __msa_ilvr_h((v8i16) vec1, (v8i16) vec0);
dst1 = (v16u8) __msa_ilvl_h((v8i16) vec1, (v8i16) vec0);
ST_UB2(dst0, dst1, rgb_buf, 16);
STOREARGB(vec0, vec1, vec2, const_255, rgb_buf);
src_y += 8;
src_u += 4;
src_v += 4;
......@@ -210,33 +223,23 @@ void I422ToARGBRow_MSA(const uint8* src_y, const uint8* src_u,
}
}
void I422ToRGBARow_MSA(const uint8* src_y, const uint8* src_u,
void YUVTORGBARow_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, dst0, dst1;
v16u8 src0, src1, src2;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v16u8 const_255 = (v16u8) __msa_ldi_b(255);
vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]);
vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]);
vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]);
vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]);
vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]);
vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]);
vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]);
vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2);
I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2);
vec0 = (v8i16) __msa_ilvev_b((v16i8) vec0, (v16i8) const_255);
vec1 = (v8i16) __msa_ilvev_b((v16i8) vec2, (v16i8) vec1);
dst0 = (v16u8) __msa_ilvr_h(vec1, vec0);
dst1 = (v16u8) __msa_ilvl_h(vec1, vec0);
ST_UB2(dst0, dst1, rgb_buf, 16);
STOREARGB(const_255, vec0, vec1, vec2, rgb_buf);
src_y += 8;
src_u += 4;
src_v += 4;
......@@ -251,32 +254,22 @@ void I422AlphaToARGBRow_MSA(const uint8* src_y, const uint8* src_u,
int width) {
int x;
int64 data_a;
v16u8 src0, src1, src2, src3, dst0, dst1;
v16u8 src0, src1, src2, src3;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 zero = { 0 };
vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]);
vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]);
vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]);
vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]);
vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]);
vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]);
vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]);
vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 8) {
data_a = LD(src_a);
READYUV422(src_y, src_u, src_v, src0, src1, src2);
src3 = (v16u8) __msa_insert_d((v2i64) zero, 0, data_a);
I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2);
src3 = (v16u8) __msa_ilvr_b((v16i8) src3, (v16i8) src3);
vec0 = (v8i16) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0);
vec1 = (v8i16) __msa_ilvev_b((v16i8) src3, (v16i8) vec2);
dst0 = (v16u8) __msa_ilvr_h((v8i16) vec1, (v8i16) vec0);
dst1 = (v16u8) __msa_ilvl_h((v8i16) vec1, (v8i16) vec0);
ST_UB2(dst0, dst1, rgb_buf, 16);
STOREARGB(vec0, vec1, vec2, src3, rgb_buf);
src_y += 8;
src_u += 4;
src_v += 4;
......@@ -285,7 +278,7 @@ void I422AlphaToARGBRow_MSA(const uint8* src_y, const uint8* src_u,
}
}
void I422ToRGB24Row_MSA(const uint8* src_y, const uint8* src_u,
void YUVTORGB24Row_MSA(const uint8* src_y, const uint8* src_u,
const uint8* src_v, uint8* rgb_buf,
const struct YuvConstants* yuvconstants, int32 width) {
int x;
......@@ -300,14 +293,8 @@ void I422ToRGB24Row_MSA(const uint8* src_y, const uint8* src_u,
v16i8 shuffler2 =
{ 26, 6, 7, 27, 8, 9, 28, 10, 11, 29, 12, 13, 30, 14, 15, 31 };
vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]);
vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]);
vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]);
vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]);
vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]);
vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]);
vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]);
vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 16) {
src0 = (v16u8) __msa_ld_b((v16u8*) src_y, 0);
......@@ -318,9 +305,9 @@ void I422ToRGB24Row_MSA(const uint8* src_y, const uint8* src_u,
src3 = (v16u8) __msa_sldi_b((v16i8) src0, (v16i8) src0, 8);
src4 = (v16u8) __msa_sldi_b((v16i8) src1, (v16i8) src1, 4);
src5 = (v16u8) __msa_sldi_b((v16i8) src2, (v16i8) src2, 4);
I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2);
I422TORGB(src3, src4, src5, vec_ub, vec_vr, vec_ug, vec_vg,
YUVTORGB(src3, src4, src5, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec3, vec4, vec5);
reg0 = (v16u8) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0);
reg2 = (v16u8) __msa_ilvev_b((v16i8) vec4, (v16i8) vec3);
......@@ -338,6 +325,104 @@ void I422ToRGB24Row_MSA(const uint8* src_y, const uint8* src_u,
}
}
// 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) {
int x;
v16u8 src0, src1, src2, dst0;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2);
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec2, vec1);
vec0 = __msa_srai_h(vec0, 3);
vec1 = __msa_srai_h(vec1, 3);
vec2 = __msa_srai_h(vec2, 2);
vec1 = __msa_slli_h(vec1, 11);
vec2 = __msa_slli_h(vec2, 5);
vec0 |= vec1;
dst0 = (v16u8) (vec2 | vec0);
ST_UB(dst0, dst_rgb565);
src_y += 8;
src_u += 4;
src_v += 4;
dst_rgb565 += 16;
}
}
// TODO(fbarchard): Consider AND instead of shift to isolate 4 upper bits of G.
void I422ToARGB4444Row_MSA(const uint8* src_y, const uint8* src_u,
const uint8* src_v, uint8* dst_argb4444,
const struct YuvConstants* yuvconstants, int width) {
int x;
v16u8 src0, src1, src2, dst0;
v8i16 vec0, vec1, vec2;
v8u16 reg0, reg1, reg2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v8u16 const_0xF000 = (v8u16) __msa_fill_h(0xF000);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2);
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2);
reg0 = (v8u16) __msa_srai_h(vec0, 4);
reg1 = (v8u16) __msa_srai_h(vec1, 4);
reg2 = (v8u16) __msa_srai_h(vec2, 4);
reg1 = (v8u16) __msa_slli_h((v8i16) reg1, 4);
reg2 = (v8u16) __msa_slli_h((v8i16) reg2, 8);
reg1 |= const_0xF000;
reg0 |= reg2;
dst0 = (v16u8) (reg1 | reg0);
ST_UB(dst0, dst_argb4444);
src_y += 8;
src_u += 4;
src_v += 4;
dst_argb4444 += 16;
}
}
void I422ToARGB1555Row_MSA(const uint8* src_y, const uint8* src_u,
const uint8* src_v, uint8* dst_argb1555,
const struct YuvConstants* yuvconstants, int width) {
int x;
v16u8 src0, src1, src2, dst0;
v8i16 vec0, vec1, vec2;
v8u16 reg0, reg1, reg2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v8u16 const_0x8000 = (v8u16) __msa_fill_h(0x8000);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug,
vec_vg, vec_bb, vec_bg, vec_br, vec_yg);
for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2);
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg,
vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2);
reg0 = (v8u16) __msa_srai_h(vec0, 3);
reg1 = (v8u16) __msa_srai_h(vec1, 3);
reg2 = (v8u16) __msa_srai_h(vec2, 3);
reg1 = (v8u16) __msa_slli_h((v8i16) reg1, 5);
reg2 = (v8u16) __msa_slli_h((v8i16) reg2, 10);
reg1 |= const_0x8000;
reg0 |= reg2;
dst0 = (v16u8) (reg1 | reg0);
ST_UB(dst0, dst_argb1555);
src_y += 8;
src_u += 4;
src_v += 4;
dst_argb1555 += 16;
}
}
void YUY2ToYRow_MSA(const uint8* src_yuy2, uint8* dst_y, int width) {
int x;
v16u8 src0, src1, src2, src3, dst0, dst1;
......
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