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

Add MSA optimized NV12/21 To RGB row functions

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

Performance Gain (vs C auto-vectorized)
NV12ToARGBRow_MSA       - ~1.5x
NV12ToARGBRow_Any_MSA   - ~1.4x
NV12ToRGB565Row_MSA     - ~1.4x
NV12ToRGB565Row_Any_MSA - ~1.4x
NV21ToARGBRow_MSA       - ~1.5x
NV21ToARGBRow_Any_MSA   - ~1.5x
SobelRow_MSA            - ~4.3x
SobelRow_Any_MSA        - ~3.4x
SobelToPlaneRow_MSA     - ~8.0x
SobelToPlaneRow_Any_MSA - ~4.7x
SobelXYRow_MSA          - ~3.0x
SobelXYRow_Any_MSA      - ~2.5x

Performance Gain (vs C non-vectorized)
NV12ToARGBRow_MSA       - ~6.5x
NV12ToARGBRow_Any_MSA   - ~6.5x
NV12ToRGB565Row_MSA     - ~6.2x
NV12ToRGB565Row_Any_MSA - ~6.1x
NV21ToARGBRow_MSA       - ~6.5x
NV21ToARGBRow_Any_MSA   - ~6.5x
SobelRow_MSA            - ~14.5x
SobelRow_Any_MSA        - ~11.3x
SobelToPlaneRow_MSA     - ~34.2x
SobelToPlaneRow_Any_MSA - ~19.4x
SobelXYRow_MSA          - ~11.1x
SobelXYRow_Any_MSA      - ~9.1x

Review-Url: https://codereview.chromium.org/2636483002 .
This commit is contained in:
Manojkumar Bhosale 2017-01-18 09:24:39 +05:30
parent a7c87e19f0
commit 09b8c971b3
5 changed files with 429 additions and 78 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
include/libyuv/row.h
View File

@ -428,6 +428,12 @@ extern "C" {
#define HAS_RGB565TOUVROW_MSA #define HAS_RGB565TOUVROW_MSA
#define HAS_RGB24TOUVROW_MSA #define HAS_RGB24TOUVROW_MSA
#define HAS_RAWTOUVROW_MSA #define HAS_RAWTOUVROW_MSA
#define HAS_NV12TOARGBROW_MSA
#define HAS_NV12TORGB565ROW_MSA
#define HAS_NV21TOARGBROW_MSA
#define HAS_SOBELROW_MSA
#define HAS_SOBELTOPLANEROW_MSA
#define HAS_SOBELXYROW_MSA
#endif #endif
#if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__) #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
@ -807,6 +813,21 @@ void I422ToARGB1555Row_MSA(const uint8* src_y,
uint8* dst_argb1555, uint8* dst_argb1555,
const struct YuvConstants* yuvconstants, const struct YuvConstants* yuvconstants,
int width); int width);
void NV12ToARGBRow_MSA(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void NV12ToRGB565Row_MSA(const uint8* src_y,
const uint8* src_uv,
uint8* dst_rgb565,
const struct YuvConstants* yuvconstants,
int width);
void NV21ToARGBRow_MSA(const uint8* src_y,
const uint8* src_vu,
uint8* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, 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); void ARGBToYRow_Any_AVX2(const uint8* src_argb, uint8* dst_y, int width);
@ -2379,6 +2400,21 @@ void I422ToARGB1555Row_Any_MSA(const uint8* src_y,
uint8* dst_argb1555, uint8* dst_argb1555,
const struct YuvConstants* yuvconstants, const struct YuvConstants* yuvconstants,
int width); int width);
void NV12ToARGBRow_Any_MSA(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void NV12ToRGB565Row_Any_MSA(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void NV21ToARGBRow_Any_MSA(const uint8* src_y,
const uint8* src_vu,
uint8* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width); void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width);
void YUY2ToUVRow_AVX2(const uint8* src_yuy2, void YUY2ToUVRow_AVX2(const uint8* src_yuy2,
@ -2868,6 +2904,10 @@ void SobelRow_NEON(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
int width); int width);
void SobelRow_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_argb,
int width);
void SobelToPlaneRow_C(const uint8* src_sobelx, void SobelToPlaneRow_C(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_y, uint8* dst_y,
@ -2880,6 +2920,10 @@ void SobelToPlaneRow_NEON(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_y, uint8* dst_y,
int width); int width);
void SobelToPlaneRow_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_y,
int width);
void SobelXYRow_C(const uint8* src_sobelx, void SobelXYRow_C(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
@ -2892,6 +2936,10 @@ void SobelXYRow_NEON(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
int width); int width);
void SobelXYRow_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_argb,
int width);
void SobelRow_Any_SSE2(const uint8* src_sobelx, void SobelRow_Any_SSE2(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
@ -2900,6 +2948,10 @@ void SobelRow_Any_NEON(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
int width); int width);
void SobelRow_Any_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_argb,
int width);
void SobelToPlaneRow_Any_SSE2(const uint8* src_sobelx, void SobelToPlaneRow_Any_SSE2(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_y, uint8* dst_y,
@ -2908,6 +2960,10 @@ void SobelToPlaneRow_Any_NEON(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_y, uint8* dst_y,
int width); int width);
void SobelToPlaneRow_Any_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_y,
int width);
void SobelXYRow_Any_SSE2(const uint8* src_sobelx, void SobelXYRow_Any_SSE2(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
@ -2916,6 +2972,10 @@ void SobelXYRow_Any_NEON(const uint8* src_sobelx,
const uint8* src_sobely, const uint8* src_sobely,
uint8* dst_argb, uint8* dst_argb,
int width); int width);
void SobelXYRow_Any_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_argb,
int width);
void ARGBPolynomialRow_C(const uint8* src_argb, void ARGBPolynomialRow_C(const uint8* src_argb,
uint8* dst_argb, uint8* dst_argb,

24
source/convert_argb.cc
View File

@ -1326,6 +1326,14 @@ int NV12ToARGB(const uint8* src_y,
} }
} }
#endif #endif
#if defined(HAS_NV12TOARGBROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
NV12ToARGBRow = NV12ToARGBRow_Any_MSA;
if (IS_ALIGNED(width, 8)) {
NV12ToARGBRow = NV12ToARGBRow_MSA;
}
}
#endif
for (y = 0; y < height; ++y) { for (y = 0; y < height; ++y) {
NV12ToARGBRow(src_y, src_uv, dst_argb, &kYuvI601Constants, width); NV12ToARGBRow(src_y, src_uv, dst_argb, &kYuvI601Constants, width);
@ -1385,6 +1393,14 @@ int NV21ToARGB(const uint8* src_y,
} }
} }
#endif #endif
#if defined(HAS_NV21TOARGBROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
NV21ToARGBRow = NV21ToARGBRow_Any_MSA;
if (IS_ALIGNED(width, 8)) {
NV21ToARGBRow = NV21ToARGBRow_MSA;
}
}
#endif
for (y = 0; y < height; ++y) { for (y = 0; y < height; ++y) {
NV21ToARGBRow(src_y, src_uv, dst_argb, &kYuvI601Constants, width); NV21ToARGBRow(src_y, src_uv, dst_argb, &kYuvI601Constants, width);
@ -1450,6 +1466,14 @@ int M420ToARGB(const uint8* src_m420,
} }
} }
#endif #endif
#if defined(HAS_NV12TOARGBROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
NV12ToARGBRow = NV12ToARGBRow_Any_MSA;
if (IS_ALIGNED(width, 8)) {
NV12ToARGBRow = NV12ToARGBRow_MSA;
}
}
#endif
for (y = 0; y < height - 1; y += 2) { for (y = 0; y < height - 1; y += 2) {
NV12ToARGBRow(src_m420, src_m420 + src_stride_m420 * 2, dst_argb, NV12ToARGBRow(src_m420, src_m420 + src_stride_m420 * 2, dst_argb,

32
source/planar_functions.cc
View File

@ -1443,6 +1443,14 @@ int NV12ToRGB565(const uint8* src_y,
} }
} }
#endif #endif
#if defined(HAS_NV12TORGB565ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
NV12ToRGB565Row = NV12ToRGB565Row_Any_MSA;
if (IS_ALIGNED(width, 8)) {
NV12ToRGB565Row = NV12ToRGB565Row_MSA;
}
}
#endif
for (y = 0; y < height; ++y) { for (y = 0; y < height; ++y) {
NV12ToRGB565Row(src_y, src_uv, dst_rgb565, &kYuvI601Constants, width); NV12ToRGB565Row(src_y, src_uv, dst_rgb565, &kYuvI601Constants, width);
@ -2650,6 +2658,14 @@ int ARGBSobel(const uint8* src_argb,
SobelRow = SobelRow_NEON; SobelRow = SobelRow_NEON;
} }
} }
#endif
#if defined(HAS_SOBELROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
SobelRow = SobelRow_Any_MSA;
if (IS_ALIGNED(width, 16)) {
SobelRow = SobelRow_MSA;
}
}
#endif #endif
return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb, return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height, SobelRow); width, height, SobelRow);
@ -2680,6 +2696,14 @@ int ARGBSobelToPlane(const uint8* src_argb,
SobelToPlaneRow = SobelToPlaneRow_NEON; SobelToPlaneRow = SobelToPlaneRow_NEON;
} }
} }
#endif
#if defined(HAS_SOBELTOPLANEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
SobelToPlaneRow = SobelToPlaneRow_Any_MSA;
if (IS_ALIGNED(width, 32)) {
SobelToPlaneRow = SobelToPlaneRow_MSA;
}
}
#endif #endif
return ARGBSobelize(src_argb, src_stride_argb, dst_y, dst_stride_y, width, return ARGBSobelize(src_argb, src_stride_argb, dst_y, dst_stride_y, width,
height, SobelToPlaneRow); height, SobelToPlaneRow);
@ -2711,6 +2735,14 @@ int ARGBSobelXY(const uint8* src_argb,
SobelXYRow = SobelXYRow_NEON; SobelXYRow = SobelXYRow_NEON;
} }
} }
#endif
#if defined(HAS_SOBELXYROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
SobelXYRow = SobelXYRow_Any_MSA;
if (IS_ALIGNED(width, 16)) {
SobelXYRow = SobelXYRow_MSA;
}
}
#endif #endif
return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb, return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height, SobelXYRow); width, height, SobelXYRow);

18
source/row_any.cc
View File

@ -255,18 +255,27 @@ ANY21(SobelRow_Any_SSE2, SobelRow_SSE2, 0, 1, 1, 4, 15)
#ifdef HAS_SOBELROW_NEON #ifdef HAS_SOBELROW_NEON
ANY21(SobelRow_Any_NEON, SobelRow_NEON, 0, 1, 1, 4, 7) ANY21(SobelRow_Any_NEON, SobelRow_NEON, 0, 1, 1, 4, 7)
#endif #endif
#ifdef HAS_SOBELROW_MSA
ANY21(SobelRow_Any_MSA, SobelRow_MSA, 0, 1, 1, 4, 15)
#endif
#ifdef HAS_SOBELTOPLANEROW_SSE2 #ifdef HAS_SOBELTOPLANEROW_SSE2
ANY21(SobelToPlaneRow_Any_SSE2, SobelToPlaneRow_SSE2, 0, 1, 1, 1, 15) ANY21(SobelToPlaneRow_Any_SSE2, SobelToPlaneRow_SSE2, 0, 1, 1, 1, 15)
#endif #endif
#ifdef HAS_SOBELTOPLANEROW_NEON #ifdef HAS_SOBELTOPLANEROW_NEON
ANY21(SobelToPlaneRow_Any_NEON, SobelToPlaneRow_NEON, 0, 1, 1, 1, 15) ANY21(SobelToPlaneRow_Any_NEON, SobelToPlaneRow_NEON, 0, 1, 1, 1, 15)
#endif #endif
#ifdef HAS_SOBELTOPLANEROW_MSA
ANY21(SobelToPlaneRow_Any_MSA, SobelToPlaneRow_MSA, 0, 1, 1, 1, 31)
#endif
#ifdef HAS_SOBELXYROW_SSE2 #ifdef HAS_SOBELXYROW_SSE2
ANY21(SobelXYRow_Any_SSE2, SobelXYRow_SSE2, 0, 1, 1, 4, 15) ANY21(SobelXYRow_Any_SSE2, SobelXYRow_SSE2, 0, 1, 1, 4, 15)
#endif #endif
#ifdef HAS_SOBELXYROW_NEON #ifdef HAS_SOBELXYROW_NEON
ANY21(SobelXYRow_Any_NEON, SobelXYRow_NEON, 0, 1, 1, 4, 7) ANY21(SobelXYRow_Any_NEON, SobelXYRow_NEON, 0, 1, 1, 4, 7)
#endif #endif
#ifdef HAS_SOBELXYROW_MSA
ANY21(SobelXYRow_Any_MSA, SobelXYRow_MSA, 0, 1, 1, 4, 15)
#endif
#undef ANY21 #undef ANY21
// Any 2 planes to 1 with yuvconstants // Any 2 planes to 1 with yuvconstants
@ -300,6 +309,9 @@ ANY21C(NV12ToARGBRow_Any_NEON, NV12ToARGBRow_NEON, 1, 1, 2, 4, 7)
#ifdef HAS_NV12TOARGBROW_DSPR2 #ifdef HAS_NV12TOARGBROW_DSPR2
ANY21C(NV12ToARGBRow_Any_DSPR2, NV12ToARGBRow_DSPR2, 1, 1, 2, 4, 7) ANY21C(NV12ToARGBRow_Any_DSPR2, NV12ToARGBRow_DSPR2, 1, 1, 2, 4, 7)
#endif #endif
#ifdef HAS_NV12TOARGBROW_MSA
ANY21C(NV12ToARGBRow_Any_MSA, NV12ToARGBRow_MSA, 1, 1, 2, 4, 7)
#endif
#ifdef HAS_NV21TOARGBROW_SSSE3 #ifdef HAS_NV21TOARGBROW_SSSE3
ANY21C(NV21ToARGBRow_Any_SSSE3, NV21ToARGBRow_SSSE3, 1, 1, 2, 4, 7) ANY21C(NV21ToARGBRow_Any_SSSE3, NV21ToARGBRow_SSSE3, 1, 1, 2, 4, 7)
#endif #endif
@ -309,6 +321,9 @@ ANY21C(NV21ToARGBRow_Any_AVX2, NV21ToARGBRow_AVX2, 1, 1, 2, 4, 15)
#ifdef HAS_NV21TOARGBROW_NEON #ifdef HAS_NV21TOARGBROW_NEON
ANY21C(NV21ToARGBRow_Any_NEON, NV21ToARGBRow_NEON, 1, 1, 2, 4, 7) ANY21C(NV21ToARGBRow_Any_NEON, NV21ToARGBRow_NEON, 1, 1, 2, 4, 7)
#endif #endif
#ifdef HAS_NV21TOARGBROW_MSA
ANY21C(NV21ToARGBRow_Any_MSA, NV21ToARGBRow_MSA, 1, 1, 2, 4, 7)
#endif
#ifdef HAS_NV12TORGB565ROW_SSSE3 #ifdef HAS_NV12TORGB565ROW_SSSE3
ANY21C(NV12ToRGB565Row_Any_SSSE3, NV12ToRGB565Row_SSSE3, 1, 1, 2, 2, 7) ANY21C(NV12ToRGB565Row_Any_SSSE3, NV12ToRGB565Row_SSSE3, 1, 1, 2, 2, 7)
#endif #endif
@ -318,6 +333,9 @@ ANY21C(NV12ToRGB565Row_Any_AVX2, NV12ToRGB565Row_AVX2, 1, 1, 2, 2, 15)
#ifdef HAS_NV12TORGB565ROW_NEON #ifdef HAS_NV12TORGB565ROW_NEON
ANY21C(NV12ToRGB565Row_Any_NEON, NV12ToRGB565Row_NEON, 1, 1, 2, 2, 7) ANY21C(NV12ToRGB565Row_Any_NEON, NV12ToRGB565Row_NEON, 1, 1, 2, 2, 7)
#endif #endif
#ifdef HAS_NV12TORGB565ROW_MSA
ANY21C(NV12ToRGB565Row_Any_MSA, NV12ToRGB565Row_MSA, 1, 1, 2, 2, 7)
#endif
#undef ANY21C #undef ANY21C
// Any 1 to 1. // Any 1 to 1.

373
source/row_msa.cc
View File

@ -47,65 +47,66 @@ extern "C" {
} }
// Convert 8 pixels of YUV 420 to RGB. // Convert 8 pixels of YUV 420 to RGB.
#define YUVTORGB(in_y, in_u, in_v, ub, vr, ug, vg, bb, bg, br, yg, out_b, \ #define YUVTORGB(in_y, in_uv, ubvr, ugvg, bb, bg, br, yg, out_b, out_g, out_r) \
out_g, out_r) \ { \
{ \ v8i16 vec0_m, vec1_m; \
v8i16 vec0_m; \ v4i32 reg0_m, reg1_m, reg2_m, reg3_m, reg4_m; \
v4i32 reg0_m, reg1_m, reg2_m, reg3_m, reg4_m; \ v4i32 reg5_m, reg6_m, reg7_m; \
v4i32 reg5_m, reg6_m, reg7_m, reg8_m, reg9_m; \ v4i32 max = __msa_ldi_w(255); \
v4i32 max_val_m = __msa_ldi_w(255); \ v16i8 zero = {0}; \
v8i16 zero_m = {0}; \ \
\ vec0_m = (v8i16)__msa_ilvr_b((v16i8)in_y, (v16i8)in_y); \
in_u = (v16u8)__msa_ilvr_b((v16i8)in_u, (v16i8)in_u); \ vec1_m = (v8i16)__msa_ilvr_b((v16i8)zero, (v16i8)in_uv); \
in_v = (v16u8)__msa_ilvr_b((v16i8)in_v, (v16i8)in_v); \ reg0_m = (v4i32)__msa_ilvr_h((v8i16)zero, (v8i16)vec0_m); \
vec0_m = (v8i16)__msa_ilvr_b((v16i8)in_y, (v16i8)in_y); \ reg1_m = (v4i32)__msa_ilvl_h((v8i16)zero, (v8i16)vec0_m); \
reg0_m = (v4i32)__msa_ilvr_h(zero_m, vec0_m); \ reg2_m = (v4i32)__msa_ilvr_h((v8i16)zero, (v8i16)vec1_m); \
reg1_m = (v4i32)__msa_ilvl_h(zero_m, vec0_m); \ reg3_m = (v4i32)__msa_ilvl_h((v8i16)zero, (v8i16)vec1_m); \
reg0_m *= vec_yg; \ reg0_m *= yg; \
reg1_m *= vec_yg; \ reg1_m *= yg; \
reg0_m = __msa_srai_w(reg0_m, 16); \ reg2_m *= ubvr; \
reg1_m = __msa_srai_w(reg1_m, 16); \ reg3_m *= ubvr; \
reg4_m = reg0_m + br; \ reg0_m = __msa_srai_w(reg0_m, 16); \
reg5_m = reg1_m + br; \ reg1_m = __msa_srai_w(reg1_m, 16); \
reg2_m = reg0_m + bg; \ reg4_m = __msa_dotp_s_w((v8i16)vec1_m, (v8i16)ugvg); \
reg3_m = reg1_m + bg; \ reg5_m = __msa_ilvev_w(reg2_m, reg2_m); \
reg0_m += bb; \ reg6_m = __msa_ilvev_w(reg3_m, reg3_m); \
reg1_m += bb; \ reg7_m = __msa_ilvr_w(reg4_m, reg4_m); \
vec0_m = (v8i16)__msa_ilvr_b((v16i8)zero_m, (v16i8)in_u); \ reg2_m = __msa_ilvod_w(reg2_m, reg2_m); \
reg6_m = (v4i32)__msa_ilvr_h(zero_m, (v8i16)vec0_m); \ reg3_m = __msa_ilvod_w(reg3_m, reg3_m); \
reg7_m = (v4i32)__msa_ilvl_h(zero_m, (v8i16)vec0_m); \ reg4_m = __msa_ilvl_w(reg4_m, reg4_m); \
vec0_m = (v8i16)__msa_ilvr_b((v16i8)zero_m, (v16i8)in_v); \ reg5_m = reg0_m - reg5_m; \
reg8_m = (v4i32)__msa_ilvr_h(zero_m, (v8i16)vec0_m); \ reg6_m = reg1_m - reg6_m; \
reg9_m = (v4i32)__msa_ilvl_h(zero_m, (v8i16)vec0_m); \ reg2_m = reg0_m - reg2_m; \
reg0_m -= reg6_m * ub; \ reg3_m = reg1_m - reg3_m; \
reg1_m -= reg7_m * ub; \ reg7_m = reg0_m - reg7_m; \
reg2_m -= reg6_m * ug; \ reg4_m = reg1_m - reg4_m; \
reg3_m -= reg7_m * ug; \ reg5_m += bb; \
reg4_m -= reg8_m * vr; \ reg6_m += bb; \
reg5_m -= reg9_m * vr; \ reg7_m += bg; \
reg2_m -= reg8_m * vg; \ reg4_m += bg; \
reg3_m -= reg9_m * vg; \ reg2_m += br; \
reg0_m = __msa_srai_w(reg0_m, 6); \ reg3_m += br; \
reg1_m = __msa_srai_w(reg1_m, 6); \ reg5_m = __msa_srai_w(reg5_m, 6); \
reg2_m = __msa_srai_w(reg2_m, 6); \ reg6_m = __msa_srai_w(reg6_m, 6); \
reg3_m = __msa_srai_w(reg3_m, 6); \ reg7_m = __msa_srai_w(reg7_m, 6); \
reg4_m = __msa_srai_w(reg4_m, 6); \ reg4_m = __msa_srai_w(reg4_m, 6); \
reg5_m = __msa_srai_w(reg5_m, 6); \ reg2_m = __msa_srai_w(reg2_m, 6); \
reg0_m = __msa_maxi_s_w(reg0_m, 0); \ reg3_m = __msa_srai_w(reg3_m, 6); \
reg1_m = __msa_maxi_s_w(reg1_m, 0); \ reg5_m = __msa_maxi_s_w(reg5_m, 0); \
reg2_m = __msa_maxi_s_w(reg2_m, 0); \ reg6_m = __msa_maxi_s_w(reg6_m, 0); \
reg3_m = __msa_maxi_s_w(reg3_m, 0); \ reg7_m = __msa_maxi_s_w(reg7_m, 0); \
reg4_m = __msa_maxi_s_w(reg4_m, 0); \ reg4_m = __msa_maxi_s_w(reg4_m, 0); \
reg5_m = __msa_maxi_s_w(reg5_m, 0); \ reg2_m = __msa_maxi_s_w(reg2_m, 0); \
reg0_m = __msa_min_s_w(reg0_m, max_val_m); \ reg3_m = __msa_maxi_s_w(reg3_m, 0); \
reg1_m = __msa_min_s_w(reg1_m, max_val_m); \ reg5_m = __msa_min_s_w(max, reg5_m); \
reg2_m = __msa_min_s_w(reg2_m, max_val_m); \ reg6_m = __msa_min_s_w(max, reg6_m); \
reg3_m = __msa_min_s_w(reg3_m, max_val_m); \ reg7_m = __msa_min_s_w(max, reg7_m); \
reg4_m = __msa_min_s_w(reg4_m, max_val_m); \ reg4_m = __msa_min_s_w(max, reg4_m); \
reg5_m = __msa_min_s_w(reg5_m, max_val_m); \ reg2_m = __msa_min_s_w(max, reg2_m); \
out_b = __msa_pckev_h((v8i16)reg1_m, (v8i16)reg0_m); \ reg3_m = __msa_min_s_w(max, reg3_m); \
out_g = __msa_pckev_h((v8i16)reg3_m, (v8i16)reg2_m); \ out_b = __msa_pckev_h((v8i16)reg6_m, (v8i16)reg5_m); \
out_r = __msa_pckev_h((v8i16)reg5_m, (v8i16)reg4_m); \ out_g = __msa_pckev_h((v8i16)reg4_m, (v8i16)reg7_m); \
out_r = __msa_pckev_h((v8i16)reg3_m, (v8i16)reg2_m); \
} }
// Pack and Store 8 ARGB values. // Pack and Store 8 ARGB values.
@ -212,15 +213,19 @@ void I422ToARGBRow_MSA(const uint8* src_y,
v16u8 src0, src1, src2; v16u8 src0, src1, src2;
v8i16 vec0, vec1, vec2; v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v16u8 const_255 = (v16u8)__msa_ldi_b(255); v16u8 const_255 = (v16u8)__msa_ldi_b(255);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) { for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2); 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, src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
vec_br, vec_yg, vec0, vec1, vec2); YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
STOREARGB(vec0, vec1, vec2, const_255, rgb_buf); STOREARGB(vec0, vec1, vec2, const_255, rgb_buf);
src_y += 8; src_y += 8;
src_u += 4; src_u += 4;
@ -239,15 +244,19 @@ void I422ToRGBARow_MSA(const uint8* src_y,
v16u8 src0, src1, src2; v16u8 src0, src1, src2;
v8i16 vec0, vec1, vec2; v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v16u8 const_255 = (v16u8)__msa_ldi_b(255); v16u8 const_255 = (v16u8)__msa_ldi_b(255);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) { for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2); 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, src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
vec_br, vec_yg, vec0, vec1, vec2); YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
STOREARGB(const_255, vec0, vec1, vec2, rgb_buf); STOREARGB(const_255, vec0, vec1, vec2, rgb_buf);
src_y += 8; src_y += 8;
src_u += 4; src_u += 4;
@ -268,17 +277,21 @@ void I422AlphaToARGBRow_MSA(const uint8* src_y,
v16u8 src0, src1, src2, src3; v16u8 src0, src1, src2, src3;
v8i16 vec0, vec1, vec2; v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v4i32 zero = {0}; v4i32 zero = {0};
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) { for (x = 0; x < width; x += 8) {
data_a = LD(src_a); data_a = LD(src_a);
READYUV422(src_y, src_u, src_v, src0, src1, src2); READYUV422(src_y, src_u, src_v, src0, src1, src2);
src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
src3 = (v16u8)__msa_insert_d((v2i64)zero, 0, data_a); src3 = (v16u8)__msa_insert_d((v2i64)zero, 0, data_a);
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec_br, vec_yg, vec0, vec1, vec2); vec0, vec1, vec2);
src3 = (v16u8)__msa_ilvr_b((v16i8)src3, (v16i8)src3); src3 = (v16u8)__msa_ilvr_b((v16i8)src3, (v16i8)src3);
STOREARGB(vec0, vec1, vec2, src3, rgb_buf); STOREARGB(vec0, vec1, vec2, src3, rgb_buf);
src_y += 8; src_y += 8;
@ -297,9 +310,10 @@ void I422ToRGB24Row_MSA(const uint8* src_y,
int32 width) { int32 width) {
int x; int x;
int64 data_u, data_v; int64 data_u, data_v;
v16u8 src0, src1, src2, src3, src4, src5, dst0, dst1, dst2; v16u8 src0, src1, src2, src3, src4, dst0, dst1, dst2;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5; v8i16 vec0, vec1, vec2, vec3, vec4, vec5;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v16u8 reg0, reg1, reg2, reg3; v16u8 reg0, reg1, reg2, reg3;
v2i64 zero = {0}; v2i64 zero = {0};
v16i8 shuffler0 = {0, 1, 16, 2, 3, 17, 4, 5, 18, 6, 7, 19, 8, 9, 20, 10}; v16i8 shuffler0 = {0, 1, 16, 2, 3, 17, 4, 5, 18, 6, 7, 19, 8, 9, 20, 10};
@ -309,6 +323,8 @@ void I422ToRGB24Row_MSA(const uint8* src_y,
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 16) { for (x = 0; x < width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16u8*)src_y, 0); src0 = (v16u8)__msa_ld_b((v16u8*)src_y, 0);
@ -316,13 +332,13 @@ void I422ToRGB24Row_MSA(const uint8* src_y,
data_v = LD(src_v); data_v = LD(src_v);
src1 = (v16u8)__msa_insert_d(zero, 0, data_u); src1 = (v16u8)__msa_insert_d(zero, 0, data_u);
src2 = (v16u8)__msa_insert_d(zero, 0, data_v); src2 = (v16u8)__msa_insert_d(zero, 0, data_v);
src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
src3 = (v16u8)__msa_sldi_b((v16i8)src0, (v16i8)src0, 8); src3 = (v16u8)__msa_sldi_b((v16i8)src0, (v16i8)src0, 8);
src4 = (v16u8)__msa_sldi_b((v16i8)src1, (v16i8)src1, 4); src4 = (v16u8)__msa_sldi_b((v16i8)src1, (v16i8)src1, 8);
src5 = (v16u8)__msa_sldi_b((v16i8)src2, (v16i8)src2, 4); YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
YUVTORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec0, vec1, vec2);
vec_br, vec_yg, vec0, vec1, vec2); YUVTORGB(src3, src4, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
YUVTORGB(src3, src4, src5, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec3, vec4, vec5);
vec_br, vec_yg, vec3, vec4, vec5);
reg0 = (v16u8)__msa_ilvev_b((v16i8)vec1, (v16i8)vec0); reg0 = (v16u8)__msa_ilvev_b((v16i8)vec1, (v16i8)vec0);
reg2 = (v16u8)__msa_ilvev_b((v16i8)vec4, (v16i8)vec3); reg2 = (v16u8)__msa_ilvev_b((v16i8)vec4, (v16i8)vec3);
reg3 = (v16u8)__msa_pckev_b((v16i8)vec5, (v16i8)vec2); reg3 = (v16u8)__msa_pckev_b((v16i8)vec5, (v16i8)vec2);
@ -350,14 +366,18 @@ void I422ToRGB565Row_MSA(const uint8* src_y,
v16u8 src0, src1, src2, dst0; v16u8 src0, src1, src2, dst0;
v8i16 vec0, vec1, vec2; v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) { for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2); 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, src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
vec_br, vec_yg, vec0, vec2, vec1); YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec2, vec1);
vec0 = __msa_srai_h(vec0, 3); vec0 = __msa_srai_h(vec0, 3);
vec1 = __msa_srai_h(vec1, 3); vec1 = __msa_srai_h(vec1, 3);
vec2 = __msa_srai_h(vec2, 2); vec2 = __msa_srai_h(vec2, 2);
@ -385,15 +405,19 @@ void I422ToARGB4444Row_MSA(const uint8* src_y,
v8i16 vec0, vec1, vec2; v8i16 vec0, vec1, vec2;
v8u16 reg0, reg1, reg2; v8u16 reg0, reg1, reg2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v8u16 const_0xF000 = (v8u16)__msa_fill_h(0xF000); v8u16 const_0xF000 = (v8u16)__msa_fill_h(0xF000);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) { for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2); 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, src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
vec_br, vec_yg, vec0, vec1, vec2); YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
reg0 = (v8u16)__msa_srai_h(vec0, 4); reg0 = (v8u16)__msa_srai_h(vec0, 4);
reg1 = (v8u16)__msa_srai_h(vec1, 4); reg1 = (v8u16)__msa_srai_h(vec1, 4);
reg2 = (v8u16)__msa_srai_h(vec2, 4); reg2 = (v8u16)__msa_srai_h(vec2, 4);
@ -421,15 +445,19 @@ void I422ToARGB1555Row_MSA(const uint8* src_y,
v8i16 vec0, vec1, vec2; v8i16 vec0, vec1, vec2;
v8u16 reg0, reg1, reg2; v8u16 reg0, reg1, reg2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v8u16 const_0x8000 = (v8u16)__msa_fill_h(0x8000); v8u16 const_0x8000 = (v8u16)__msa_fill_h(0x8000);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg); vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) { for (x = 0; x < width; x += 8) {
READYUV422(src_y, src_u, src_v, src0, src1, src2); 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, src1 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src1);
vec_br, vec_yg, vec0, vec1, vec2); YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
reg0 = (v8u16)__msa_srai_h(vec0, 3); reg0 = (v8u16)__msa_srai_h(vec0, 3);
reg1 = (v8u16)__msa_srai_h(vec1, 3); reg1 = (v8u16)__msa_srai_h(vec1, 3);
reg2 = (v8u16)__msa_srai_h(vec2, 3); reg2 = (v8u16)__msa_srai_h(vec2, 3);
@ -2023,6 +2051,195 @@ void RAWToUVRow_MSA(const uint8* src_rgb0,
} }
} }
void NV12ToARGBRow_MSA(const uint8* src_y,
const uint8* src_uv,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
int x;
uint64 val0, val1;
v16u8 src0, src1, res0, res1, dst0, dst1;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v16u8 zero = {0};
v16u8 const_255 = (v16u8)__msa_ldi_b(255);
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) {
val0 = LD(src_y);
val1 = LD(src_uv);
src0 = (v16u8)__msa_insert_d((v2i64)zero, 0, val0);
src1 = (v16u8)__msa_insert_d((v2i64)zero, 0, val1);
YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
res0 = (v16u8)__msa_ilvev_b((v16i8)vec2, (v16i8)vec0);
res1 = (v16u8)__msa_ilvev_b((v16i8)const_255, (v16i8)vec1);
dst0 = (v16u8)__msa_ilvr_b((v16i8)res1, (v16i8)res0);
dst1 = (v16u8)__msa_ilvl_b((v16i8)res1, (v16i8)res0);
ST_UB2(dst0, dst1, rgb_buf, 16);
src_y += 8;
src_uv += 8;
rgb_buf += 32;
}
}
void NV12ToRGB565Row_MSA(const uint8* src_y,
const uint8* src_uv,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
int x;
uint64 val0, val1;
v16u8 src0, src1, dst0;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v16u8 zero = {0};
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) {
val0 = LD(src_y);
val1 = LD(src_uv);
src0 = (v16u8)__msa_insert_d((v2i64)zero, 0, val0);
src1 = (v16u8)__msa_insert_d((v2i64)zero, 0, val1);
YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
vec0 = vec0 >> 3;
vec1 = (vec1 >> 2) << 5;
vec2 = (vec2 >> 3) << 11;
dst0 = (v16u8)(vec0 | vec1 | vec2);
ST_UB(dst0, rgb_buf);
src_y += 8;
src_uv += 8;
rgb_buf += 16;
}
}
void NV21ToARGBRow_MSA(const uint8* src_y,
const uint8* src_vu,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
int x;
uint64 val0, val1;
v16u8 src0, src1, res0, res1, dst0, dst1;
v8i16 vec0, vec1, vec2;
v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg;
v4i32 vec_ubvr, vec_ugvg;
v16u8 const_255 = (v16u8)__msa_ldi_b(255);
v16u8 zero = {0};
v16i8 shuffler = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14};
YUVTORGB_SETUP(yuvconstants, vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg,
vec_br, vec_yg);
vec_ubvr = __msa_ilvr_w(vec_vr, vec_ub);
vec_ugvg = (v4i32)__msa_ilvev_h((v8i16)vec_vg, (v8i16)vec_ug);
for (x = 0; x < width; x += 8) {
val0 = LD(src_y);
val1 = LD(src_vu);
src0 = (v16u8)__msa_insert_d((v2i64)zero, 0, val0);
src1 = (v16u8)__msa_insert_d((v2i64)zero, 0, val1);
src1 = (v16u8)__msa_vshf_b(shuffler, (v16i8)src1, (v16i8)src1);
YUVTORGB(src0, src1, vec_ubvr, vec_ugvg, vec_bb, vec_bg, vec_br, vec_yg,
vec0, vec1, vec2);
res0 = (v16u8)__msa_ilvev_b((v16i8)vec2, (v16i8)vec0);
res1 = (v16u8)__msa_ilvev_b((v16i8)const_255, (v16i8)vec1);
dst0 = (v16u8)__msa_ilvr_b((v16i8)res1, (v16i8)res0);
dst1 = (v16u8)__msa_ilvl_b((v16i8)res1, (v16i8)res0);
ST_UB2(dst0, dst1, rgb_buf, 16);
src_y += 8;
src_vu += 8;
rgb_buf += 32;
}
}
void SobelRow_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_argb,
int width) {
int x;
v16u8 src0, src1, vec0, dst0, dst1, dst2, dst3;
v16i8 mask0 = {0, 0, 0, 16, 1, 1, 1, 16, 2, 2, 2, 16, 3, 3, 3, 16};
v16i8 const_0x4 = __msa_ldi_b(0x4);
v16i8 mask1 = mask0 + const_0x4;
v16i8 mask2 = mask1 + const_0x4;
v16i8 mask3 = mask2 + const_0x4;
v16u8 const_0xFF = (v16u8)__msa_ldi_b(0xFF);
for (x = 0; x < width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 0);
vec0 = __msa_adds_u_b(src0, src1);
dst0 = (v16u8)__msa_vshf_b(mask0, (v16i8)const_0xFF, (v16i8)vec0);
dst1 = (v16u8)__msa_vshf_b(mask1, (v16i8)const_0xFF, (v16i8)vec0);
dst2 = (v16u8)__msa_vshf_b(mask2, (v16i8)const_0xFF, (v16i8)vec0);
dst3 = (v16u8)__msa_vshf_b(mask3, (v16i8)const_0xFF, (v16i8)vec0);
ST_UB4(dst0, dst1, dst2, dst3, dst_argb, 16);
src_sobelx += 16;
src_sobely += 16;
dst_argb += 64;
}
}
void SobelToPlaneRow_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_y,
int width) {
int x;
v16u8 src0, src1, src2, src3, dst0, dst1;
for (x = 0; x < width; x += 32) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 0);
src3 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 16);
dst0 = __msa_adds_u_b(src0, src2);
dst1 = __msa_adds_u_b(src1, src3);
ST_UB2(dst0, dst1, dst_y, 16);
src_sobelx += 32;
src_sobely += 32;
dst_y += 32;
}
}
void SobelXYRow_MSA(const uint8* src_sobelx,
const uint8* src_sobely,
uint8* dst_argb,
int width) {
int x;
v16u8 src0, src1, vec0, vec1, vec2;
v16u8 reg0, reg1, dst0, dst1, dst2, dst3;
v16u8 const_0xFF = (v16u8)__msa_ldi_b(0xFF);
for (x = 0; x < width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_sobelx, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_sobely, 0);
vec0 = __msa_adds_u_b(src0, src1);
vec1 = (v16u8)__msa_ilvr_b((v16i8)src0, (v16i8)src1);
vec2 = (v16u8)__msa_ilvl_b((v16i8)src0, (v16i8)src1);
reg0 = (v16u8)__msa_ilvr_b((v16i8)const_0xFF, (v16i8)vec0);
reg1 = (v16u8)__msa_ilvl_b((v16i8)const_0xFF, (v16i8)vec0);
dst0 = (v16u8)__msa_ilvr_b((v16i8)reg0, (v16i8)vec1);
dst1 = (v16u8)__msa_ilvl_b((v16i8)reg0, (v16i8)vec1);
dst2 = (v16u8)__msa_ilvr_b((v16i8)reg1, (v16i8)vec2);
dst3 = (v16u8)__msa_ilvl_b((v16i8)reg1, (v16i8)vec2);
ST_UB4(dst0, dst1, dst2, dst3, dst_argb, 16);
src_sobelx += 16;
src_sobely += 16;
dst_argb += 64;
}
}
#ifdef __cplusplus #ifdef __cplusplus
} // extern "C" } // extern "C"
} // namespace libyuv } // namespace libyuv