From 000d2fa91a487a2430be8ace333f199c51467b5b Mon Sep 17 00:00:00 2001 From: Frank Barchard Date: Wed, 11 Jan 2017 12:19:13 -0800 Subject: [PATCH] Libyuv MIPS DSPR2 optimizations. Optimized functions: I444ToARGBRow_DSPR2 I422ToARGB4444Row_DSPR2 I422ToARGB1555Row_DSPR2 NV12ToARGBRow_DSPR2 BGRAToUVRow_DSPR2 BGRAToYRow_DSPR2 ABGRToUVRow_DSPR2 ARGBToYRow_DSPR2 ABGRToYRow_DSPR2 RGBAToUVRow_DSPR2 RGBAToYRow_DSPR2 ARGBToUVRow_DSPR2 RGB24ToARGBRow_DSPR2 RAWToARGBRow_DSPR2 RGB565ToARGBRow_DSPR2 ARGB1555ToARGBRow_DSPR2 ARGB4444ToARGBRow_DSPR2 ScaleAddRow_DSPR2 Bug-fixes in functions: ScaleRowDown2_DSPR2 ScaleRowDown4_DSPR2 BUG= Review-Url: https://codereview.chromium.org/2626123003 . --- include/libyuv/row.h | 157 +++++ include/libyuv/scale_row.h | 11 +- source/convert.cc | 80 +++ source/convert_argb.cc | 64 ++ source/convert_from.cc | 16 + source/convert_from_argb.cc | 97 +++ source/row_any.cc | 48 ++ source/row_common.cc | 10 +- source/row_mips.cc | 1190 +++++++++++++++++++++++++++++++---- source/row_win.cc | 2 +- source/scale.cc | 8 + source/scale_any.cc | 3 + source/scale_mips.cc | 61 +- unit_test/convert_test.cc | 445 ++++++------- util/psnr_main.cc | 14 +- 15 files changed, 1854 insertions(+), 352 deletions(-) diff --git a/include/libyuv/row.h b/include/libyuv/row.h index e847a728c..2d020d484 100644 --- a/include/libyuv/row.h +++ b/include/libyuv/row.h @@ -364,6 +364,23 @@ extern "C" { #define HAS_MIRRORROW_DSPR2 #define HAS_MIRRORUVROW_DSPR2 #define HAS_SPLITUVROW_DSPR2 +#define HAS_RGB24TOARGBROW_DSPR2 +#define HAS_RAWTOARGBROW_DSPR2 +#define HAS_RGB565TOARGBROW_DSPR2 +#define HAS_ARGB1555TOARGBROW_DSPR2 +#define HAS_ARGB4444TOARGBROW_DSPR2 +#define HAS_I444TOARGBROW_DSPR2 +#define HAS_I422TOARGB4444ROW_DSPR2 +#define HAS_I422TOARGB1555ROW_DSPR2 +#define HAS_NV12TOARGBROW_DSPR2 +#define HAS_BGRATOUVROW_DSPR2 +#define HAS_BGRATOYROW_DSPR2 +#define HAS_ABGRTOUVROW_DSPR2 +#define HAS_ARGBTOYROW_DSPR2 +#define HAS_ABGRTOYROW_DSPR2 +#define HAS_RGBATOUVROW_DSPR2 +#define HAS_RGBATOYROW_DSPR2 +#define HAS_ARGBTOUVROW_DSPR2 #endif #endif @@ -660,6 +677,30 @@ void UYVYToARGBRow_NEON(const uint8* src_uyvy, uint8* dst_argb, const struct YuvConstants* yuvconstants, int width); +void I444ToARGBRow_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); +void I422ToARGB4444Row_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb4444, + const struct YuvConstants* yuvconstants, + int width); +void I422ToARGB1555Row_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb1555, + const struct YuvConstants* yuvconstants, + int width); +void NV12ToARGBRow_DSPR2(const uint8* src_y, + const uint8* src_uv, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); + void I422ToARGBRow_MSA(const uint8* src_y, const uint8* src_u, const uint8* src_v, @@ -789,6 +830,30 @@ void RAWToYRow_NEON(const uint8* src_raw, uint8* dst_y, int width); void RGB565ToYRow_NEON(const uint8* src_rgb565, uint8* dst_y, int width); void ARGB1555ToYRow_NEON(const uint8* src_argb1555, uint8* dst_y, int width); void ARGB4444ToYRow_NEON(const uint8* src_argb4444, uint8* dst_y, int width); +void BGRAToUVRow_DSPR2(const uint8* src_bgra, + int src_stride_bgra, + uint8* dst_u, + uint8* dst_v, + int width); +void BGRAToYRow_DSPR2(const uint8* src_bgra, uint8* dst_y, int width); +void ABGRToUVRow_DSPR2(const uint8* src_abgr, + int src_stride_abgr, + uint8* dst_u, + uint8* dst_v, + int width); +void ARGBToYRow_DSPR2(const uint8* src_argb, uint8* dst_y, int width); +void ABGRToYRow_DSPR2(const uint8* src_abgr, uint8* dst_y, int width); +void RGBAToUVRow_DSPR2(const uint8* src_rgba, + int src_stride_rgba, + uint8* dst_u, + uint8* dst_v, + int width); +void RGBAToYRow_DSPR2(const uint8* src_rgba, uint8* dst_y, int width); +void ARGBToUVRow_DSPR2(const uint8* src_argb, + int src_stride_argb, + uint8* dst_u, + uint8* dst_v, + int width); void ARGBToYRow_C(const uint8* src_argb, uint8* dst_y, int width); void ARGBToYJRow_C(const uint8* src_argb, uint8* dst_y, int width); void BGRAToYRow_C(const uint8* src_bgra, uint8* dst_y, int width); @@ -817,6 +882,10 @@ void RGB565ToYRow_Any_NEON(const uint8* src_rgb565, uint8* dst_y, int width); void ARGB1555ToYRow_Any_NEON(const uint8* src_argb1555, uint8* dst_y, int width); +void BGRAToYRow_Any_DSPR2(const uint8* src_bgra, uint8* dst_y, int width); +void ARGBToYRow_Any_DSPR2(const uint8* src_argb, uint8* dst_y, int width); +void ABGRToYRow_Any_DSPR2(const uint8* src_abgr, uint8* dst_y, int width); +void RGBAToYRow_Any_DSPR2(const uint8* src_rgba, uint8* dst_y, int width); void ARGB4444ToYRow_Any_NEON(const uint8* src_argb4444, uint8* dst_y, int width); @@ -955,6 +1024,36 @@ void ARGB4444ToUVRow_Any_NEON(const uint8* src_argb4444, uint8* dst_u, uint8* dst_v, int width); +void BGRAToUVRow_Any_DSPR2(const uint8* src_bgra, + int src_stride_bgra, + uint8* dst_u, + uint8* dst_v, + int width); +void ABGRToUVRow_Any_DSPR2(const uint8* src_abgr, + int src_stride_abgr, + uint8* dst_u, + uint8* dst_v, + int width); +void RGBAToUVRow_Any_DSPR2(const uint8* src_rgba, + int src_stride_rgba, + uint8* dst_u, + uint8* dst_v, + int width); +void ARGBToUVRow_Any_DSPR2(const uint8* src_argb, + int src_stride_argb, + uint8* dst_u, + uint8* dst_v, + int width); +void ARGBToUVRow_C(const uint8* src_argb, + int src_stride_argb, + uint8* dst_u, + uint8* dst_v, + int width); +void ARGBToUVJRow_C(const uint8* src_argb, + int src_stride_argb, + uint8* dst_u, + uint8* dst_v, + int width); void ARGBToUVRow_C(const uint8* src_argb, int src_stride_argb, uint8* dst_u, @@ -1251,6 +1350,15 @@ void ARGB1555ToARGBRow_NEON(const uint8* src_argb1555, void ARGB4444ToARGBRow_NEON(const uint8* src_argb4444, uint8* dst_argb, int width); +void RGB24ToARGBRow_DSPR2(const uint8* src_rgb24, uint8* dst_argb, int width); +void RAWToARGBRow_DSPR2(const uint8* src_raw, uint8* dst_argb, int width); +void RGB565ToARGBRow_DSPR2(const uint8* src_rgb565, uint8* dst_argb, int width); +void ARGB1555ToARGBRow_DSPR2(const uint8* src_argb1555, + uint8* dst_argb, + int width); +void ARGB4444ToARGBRow_DSPR2(const uint8* src_argb4444, + uint8* dst_argb, + int width); void ARGB4444ToARGBRow_MSA(const uint8* src_argb4444, uint8* dst_argb, int width); @@ -1299,6 +1407,20 @@ void ARGB1555ToARGBRow_Any_NEON(const uint8* src_argb1555, void ARGB4444ToARGBRow_Any_NEON(const uint8* src_argb4444, uint8* dst_argb, int width); +void RGB24ToARGBRow_Any_DSPR2(const uint8* src_rgb24, + uint8* dst_argb, + int width); +void RAWToARGBRow_Any_DSPR2(const uint8* src_raw, uint8* dst_argb, int width); +void RGB565ToARGBRow_Any_DSPR2(const uint8* src_rgb565, + uint8* dst_argb, + int width); +void ARGB1555ToARGBRow_Any_DSPR2(const uint8* src_argb1555, + uint8* dst_argb, + int width); +void ARGB4444ToARGBRow_Any_DSPR2(const uint8* src_argb4444, + uint8* dst_argb, + int width); + void ARGB4444ToARGBRow_Any_MSA(const uint8* src_argb4444, uint8* dst_argb, int width); @@ -2042,12 +2164,47 @@ void UYVYToARGBRow_Any_NEON(const uint8* src_uyvy, uint8* dst_argb, const struct YuvConstants* yuvconstants, int width); +void I444ToARGBRow_Any_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); +void I422ToARGB4444Row_Any_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); +void I422ToARGBRow_Any_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); void I422ToARGBRow_DSPR2(const uint8* src_y, const uint8* src_u, const uint8* src_v, uint8* dst_argb, const struct YuvConstants* yuvconstants, int width); +void I422ToARGB1555Row_Any_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); +void I411ToARGBRow_Any_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); +void NV12ToARGBRow_Any_DSPR2(const uint8* src_y, + const uint8* src_uv, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width); void I422ToARGBRow_DSPR2(const uint8* src_y, const uint8* src_u, const uint8* src_v, diff --git a/include/libyuv/scale_row.h b/include/libyuv/scale_row.h index 77ce96ed5..73796db99 100644 --- a/include/libyuv/scale_row.h +++ b/include/libyuv/scale_row.h @@ -101,6 +101,7 @@ extern "C" { #define HAS_SCALEROWDOWN4_DSPR2 #define HAS_SCALEROWDOWN34_DSPR2 #define HAS_SCALEROWDOWN38_DSPR2 +#define HAS_SCALEADDROW_DSPR2 #endif #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) @@ -846,6 +847,10 @@ void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, uint8* dst_ptr, int dst_width); +void ScaleAddRow_DSPR2(const uint8* src_ptr, uint16* dst_ptr, int src_width); +void ScaleAddRow_Any_DSPR2(const uint8* src_ptr, + uint16* dst_ptr, + int src_width); void ScaleRowDown2_MSA(const uint8_t* src_ptr, ptrdiff_t src_stride, @@ -885,9 +890,9 @@ void ScaleRowDown2_Any_MSA(const uint8_t* src_ptr, uint8_t* dst, int dst_width); void ScaleRowDown2Linear_Any_MSA(const uint8_t* src_ptr, - ptrdiff_t src_stride, - uint8_t* dst, - int dst_width); + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width); void ScaleRowDown2Box_Any_MSA(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst, diff --git a/source/convert.cc b/source/convert.cc index 3779d999d..6bea6472e 100644 --- a/source/convert.cc +++ b/source/convert.cc @@ -579,6 +579,14 @@ int ARGBToI420(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif #if defined(HAS_ARGBTOYROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { ARGBToYRow = ARGBToYRow_Any_MSA; @@ -587,6 +595,14 @@ int ARGBToI420(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToUVRow = ARGBToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_DSPR2; + } + } +#endif #if defined(HAS_ARGBTOUVROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { ARGBToUVRow = ARGBToUVRow_Any_MSA; @@ -664,6 +680,22 @@ int BGRAToI420(const uint8* src_bgra, } } #endif +#if defined(HAS_BGRATOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + BGRAToYRow = BGRAToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + BGRAToYRow = BGRAToYRow_DSPR2; + } + } +#endif +#if defined(HAS_BGRATOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + BGRAToUVRow = BGRAToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + BGRAToUVRow = BGRAToUVRow_DSPR2; + } + } +#endif for (y = 0; y < height - 1; y += 2) { BGRAToUVRow(src_bgra, src_stride_bgra, dst_u, dst_v, width); @@ -733,6 +765,22 @@ int ABGRToI420(const uint8* src_abgr, } } #endif +#if defined(HAS_ABGRTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ABGRToYRow = ABGRToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ABGRToYRow = ABGRToYRow_DSPR2; + } + } +#endif +#if defined(HAS_ABGRTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ABGRToUVRow = ABGRToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ABGRToUVRow = ABGRToUVRow_DSPR2; + } + } +#endif for (y = 0; y < height - 1; y += 2) { ABGRToUVRow(src_abgr, src_stride_abgr, dst_u, dst_v, width); @@ -802,6 +850,22 @@ int RGBAToI420(const uint8* src_rgba, } } #endif +#if defined(HAS_RGBATOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RGBAToYRow = RGBAToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + RGBAToYRow = RGBAToYRow_DSPR2; + } + } +#endif +#if defined(HAS_RGBATOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RGBAToUVRow = RGBAToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + RGBAToUVRow = RGBAToUVRow_DSPR2; + } + } +#endif for (y = 0; y < height - 1; y += 2) { RGBAToUVRow(src_rgba, src_stride_rgba, dst_u, dst_v, width); @@ -1014,6 +1078,14 @@ int RAWToI420(const uint8* src_raw, ARGBToYRow = ARGBToYRow_AVX2; } } +#endif +#if defined(HAS_RAWTOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RAWToARGBRow = RAWToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 4)) { + RAWToARGBRow = RAWToARGBRow_DSPR2; + } + } #endif { // Allocate 2 rows of ARGB. @@ -1142,6 +1214,14 @@ int RGB565ToI420(const uint8* src_rgb565, ARGBToYRow = ARGBToYRow_AVX2; } } +#endif +#if defined(HAS_RGB565TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RGB565ToARGBRow = RGB565ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + RGB565ToARGBRow = RGB565ToARGBRow_DSPR2; + } + } #endif { // Allocate 2 rows of ARGB. diff --git a/source/convert_argb.cc b/source/convert_argb.cc index db3b5d32b..fecd21ff3 100644 --- a/source/convert_argb.cc +++ b/source/convert_argb.cc @@ -485,6 +485,14 @@ static int I444ToARGBMatrix(const uint8* src_y, } } #endif +#if defined(HAS_I444TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + I444ToARGBRow = I444ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + I444ToARGBRow = I444ToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height; ++y) { I444ToARGBRow(src_y, src_u, src_v, dst_argb, yuvconstants, width); @@ -946,6 +954,14 @@ int RGB24ToARGB(const uint8* src_rgb24, } } #endif +#if defined(HAS_RGB24TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RGB24ToARGBRow = RGB24ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + RGB24ToARGBRow = RGB24ToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height; ++y) { RGB24ToARGBRow(src_rgb24, dst_argb, width); @@ -997,6 +1013,14 @@ int RAWToARGB(const uint8* src_raw, } } #endif +#if defined(HAS_RAWTOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RAWToARGBRow = RAWToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + RAWToARGBRow = RAWToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height; ++y) { RAWToARGBRow(src_raw, dst_argb, width); @@ -1056,6 +1080,14 @@ int RGB565ToARGB(const uint8* src_rgb565, } } #endif +#if defined(HAS_RGB565TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + RGB565ToARGBRow = RGB565ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + RGB565ToARGBRow = RGB565ToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height; ++y) { RGB565ToARGBRow(src_rgb565, dst_argb, width); @@ -1115,6 +1147,14 @@ int ARGB1555ToARGB(const uint8* src_argb1555, } } #endif +#if defined(HAS_ARGB1555TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGB1555ToARGBRow = ARGB1555ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 4)) { + ARGB1555ToARGBRow = ARGB1555ToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height; ++y) { ARGB1555ToARGBRow(src_argb1555, dst_argb, width); @@ -1174,6 +1214,14 @@ int ARGB4444ToARGB(const uint8* src_argb4444, } } #endif +#if defined(HAS_ARGB4444TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGB4444ToARGBRow = ARGB4444ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 4)) { + ARGB4444ToARGBRow = ARGB4444ToARGBRow_DSPR2; + } + } +#endif #if defined(HAS_ARGB4444TOARGBROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { ARGB4444ToARGBRow = ARGB4444ToARGBRow_Any_MSA; @@ -1238,6 +1286,14 @@ int NV12ToARGB(const uint8* src_y, } } #endif +#if defined(HAS_NV12TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + NV12ToARGBRow = NV12ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + NV12ToARGBRow = NV12ToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height; ++y) { NV12ToARGBRow(src_y, src_uv, dst_argb, &kYuvI601Constants, width); @@ -1354,6 +1410,14 @@ int M420ToARGB(const uint8* src_m420, } } #endif +#if defined(HAS_NV12TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + NV12ToARGBRow = NV12ToARGBRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + NV12ToARGBRow = NV12ToARGBRow_DSPR2; + } + } +#endif for (y = 0; y < height - 1; y += 2) { NV12ToARGBRow(src_m420, src_m420 + src_stride_m420 * 2, dst_argb, diff --git a/source/convert_from.cc b/source/convert_from.cc index ddfa3b384..e6ff52439 100644 --- a/source/convert_from.cc +++ b/source/convert_from.cc @@ -708,6 +708,14 @@ int I420ToARGB1555(const uint8* src_y, } } #endif +#if defined(HAS_I422TOARGB1555ROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + I422ToARGB1555Row = I422ToARGB1555Row_Any_DSPR2; + if (IS_ALIGNED(width, 4)) { + I422ToARGB1555Row = I422ToARGB1555Row_DSPR2; + } + } +#endif #if defined(HAS_I422TOARGB1555ROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { I422ToARGB1555Row = I422ToARGB1555Row_Any_MSA; @@ -781,6 +789,14 @@ int I420ToARGB4444(const uint8* src_y, } } #endif +#if defined(HAS_I422TOARGB4444ROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + I422ToARGB4444Row = I422ToARGB4444Row_Any_DSPR2; + if (IS_ALIGNED(width, 4)) { + I422ToARGB4444Row = I422ToARGB4444Row_DSPR2; + } + } +#endif #if defined(HAS_I422TOARGB4444ROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { I422ToARGB4444Row = I422ToARGB4444Row_Any_MSA; diff --git a/source/convert_from_argb.cc b/source/convert_from_argb.cc index 81b59ba29..e45bb230f 100644 --- a/source/convert_from_argb.cc +++ b/source/convert_from_argb.cc @@ -100,6 +100,14 @@ int ARGBToI444(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif #if defined(HAS_ARGBTOYROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { ARGBToYRow = ARGBToYRow_Any_MSA; @@ -189,6 +197,23 @@ int ARGBToI422(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif +#if defined(HAS_ARGBTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToUVRow = ARGBToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_DSPR2; + } + } +#endif + #if defined(HAS_ARGBTOYROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { ARGBToYRow = ARGBToYRow_Any_MSA; @@ -318,6 +343,22 @@ int ARGBToNV12(const uint8* src_argb, MergeUVRow_ = MergeUVRow_NEON; } } +#endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif +#if defined(HAS_ARGBTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToUVRow = ARGBToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_DSPR2; + } + } #endif { // Allocate a rows of uv. @@ -445,6 +486,22 @@ int ARGBToNV21(const uint8* src_argb, MergeUVRow_ = MergeUVRow_NEON; } } +#endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif +#if defined(HAS_ARGBTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToUVRow = ARGBToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_DSPR2; + } + } #endif { // Allocate a rows of uv. @@ -570,6 +627,22 @@ int ARGBToYUY2(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif +#if defined(HAS_ARGBTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToUVRow = ARGBToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_DSPR2; + } + } +#endif #if defined(HAS_I422TOYUY2ROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { I422ToYUY2Row = I422ToYUY2Row_Any_MSA; @@ -698,6 +771,22 @@ int ARGBToUYVY(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif +#if defined(HAS_ARGBTOUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToUVRow = ARGBToUVRow_Any_DSPR2; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_DSPR2; + } + } +#endif #if defined(HAS_I422TOUYVYROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { I422ToUYVYRow = I422ToUYVYRow_Any_MSA; @@ -775,6 +864,14 @@ int ARGBToI400(const uint8* src_argb, } } #endif +#if defined(HAS_ARGBTOYROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ARGBToYRow = ARGBToYRow_Any_DSPR2; + if (IS_ALIGNED(width, 8)) { + ARGBToYRow = ARGBToYRow_DSPR2; + } + } +#endif #if defined(HAS_ARGBTOYROW_MSA) if (TestCpuFlag(kCpuHasMSA)) { ARGBToYRow = ARGBToYRow_Any_MSA; diff --git a/source/row_any.cc b/source/row_any.cc index 82bca93ba..ef8dd29d3 100644 --- a/source/row_any.cc +++ b/source/row_any.cc @@ -167,6 +167,12 @@ ANY31C(I422ToARGB4444Row_Any_NEON, I422ToARGB4444Row_NEON, 1, 0, 2, 7) ANY31C(I422ToARGB1555Row_Any_NEON, I422ToARGB1555Row_NEON, 1, 0, 2, 7) ANY31C(I422ToRGB565Row_Any_NEON, I422ToRGB565Row_NEON, 1, 0, 2, 7) #endif +#ifdef HAS_I422TOARGBROW_DSPR2 +ANY31C(I444ToARGBRow_Any_DSPR2, I444ToARGBRow_DSPR2, 0, 0, 4, 7) +ANY31C(I422ToARGBRow_Any_DSPR2, I422ToARGBRow_DSPR2, 1, 0, 4, 7) +ANY31C(I422ToARGB4444Row_Any_DSPR2, I422ToARGB4444Row_DSPR2, 1, 0, 2, 7) +ANY31C(I422ToARGB1555Row_Any_DSPR2, I422ToARGB1555Row_DSPR2, 1, 0, 2, 7) +#endif #ifdef HAS_I422TOARGBROW_MSA ANY31C(I422ToARGBRow_Any_MSA, I422ToARGBRow_MSA, 1, 0, 4, 7) ANY31C(I422ToRGBARow_Any_MSA, I422ToRGBARow_MSA, 1, 0, 4, 7) @@ -291,6 +297,9 @@ ANY21C(NV12ToARGBRow_Any_AVX2, NV12ToARGBRow_AVX2, 1, 1, 2, 4, 15) #ifdef HAS_NV12TOARGBROW_NEON ANY21C(NV12ToARGBRow_Any_NEON, NV12ToARGBRow_NEON, 1, 1, 2, 4, 7) #endif +#ifdef HAS_NV12TOARGBROW_DSPR2 +ANY21C(NV12ToARGBRow_Any_DSPR2, NV12ToARGBRow_DSPR2, 1, 1, 2, 4, 7) +#endif #ifdef HAS_NV21TOARGBROW_SSSE3 ANY21C(NV21ToARGBRow_Any_SSSE3, NV21ToARGBRow_SSSE3, 1, 1, 2, 4, 7) #endif @@ -484,6 +493,33 @@ ANY11(ARGB1555ToARGBRow_Any_NEON, ARGB1555ToARGBRow_NEON, 0, 2, 4, 7) #ifdef HAS_ARGB4444TOARGBROW_NEON ANY11(ARGB4444ToARGBRow_Any_NEON, ARGB4444ToARGBRow_NEON, 0, 2, 4, 7) #endif +#ifdef HAS_RGB24TOARGBROW_DSPR2 +ANY11(RGB24ToARGBRow_Any_DSPR2, RGB24ToARGBRow_DSPR2, 0, 3, 4, 7) +#endif +#ifdef HAS_RAWTOARGBROW_DSPR2 +ANY11(RAWToARGBRow_Any_DSPR2, RAWToARGBRow_DSPR2, 0, 3, 4, 7) +#endif +#ifdef HAS_RGB565TOARGBROW_DSPR2 +ANY11(RGB565ToARGBRow_Any_DSPR2, RGB565ToARGBRow_DSPR2, 0, 2, 4, 7) +#endif +#ifdef HAS_ARGB1555TOARGBROW_DSPR2 +ANY11(ARGB1555ToARGBRow_Any_DSPR2, ARGB1555ToARGBRow_DSPR2, 0, 2, 4, 7) +#endif +#ifdef HAS_ARGB4444TOARGBROW_DSPR2 +ANY11(ARGB4444ToARGBRow_Any_DSPR2, ARGB4444ToARGBRow_DSPR2, 0, 2, 4, 7) +#endif +#ifdef HAS_BGRATOYROW_DSPR2 +ANY11(BGRAToYRow_Any_DSPR2, BGRAToYRow_DSPR2, 0, 4, 1, 7) +#endif +#ifdef HAS_ARGBTOYROW_DSPR2 +ANY11(ARGBToYRow_Any_DSPR2, ARGBToYRow_DSPR2, 0, 4, 1, 7) +#endif +#ifdef HAS_ABGRTOYROW_DSPR2 +ANY11(ABGRToYRow_Any_DSPR2, ABGRToYRow_DSPR2, 0, 4, 1, 7) +#endif +#ifdef HAS_RGBATOYROW_DSPR2 +ANY11(RGBAToYRow_Any_DSPR2, RGBAToYRow_DSPR2, 0, 4, 1, 7) +#endif #ifdef HAS_ARGB4444TOARGBROW_MSA ANY11(ARGB4444ToARGBRow_Any_MSA, ARGB4444ToARGBRow_MSA, 0, 2, 4, 15) #endif @@ -904,6 +940,18 @@ ANY12S(YUY2ToUVRow_Any_NEON, YUY2ToUVRow_NEON, 1, 4, 15) #ifdef HAS_UYVYTOUVROW_NEON ANY12S(UYVYToUVRow_Any_NEON, UYVYToUVRow_NEON, 1, 4, 15) #endif +#ifdef HAS_BGRATOUVROW_DSPR2 +ANY12S(BGRAToUVRow_Any_DSPR2, BGRAToUVRow_DSPR2, 0, 4, 15) +#endif +#ifdef HAS_ABGRTOUVROW_DSPR2 +ANY12S(ABGRToUVRow_Any_DSPR2, ABGRToUVRow_DSPR2, 0, 4, 15) +#endif +#ifdef HAS_RGBATOUVROW_DSPR2 +ANY12S(RGBAToUVRow_Any_DSPR2, RGBAToUVRow_DSPR2, 0, 4, 15) +#endif +#ifdef HAS_ARGBTOUVROW_DSPR2 +ANY12S(ARGBToUVRow_Any_DSPR2, ARGBToUVRow_DSPR2, 0, 4, 15) +#endif #ifdef HAS_YUY2TOUVROW_MSA ANY12S(YUY2ToUVRow_Any_MSA, YUY2ToUVRow_MSA, 1, 4, 31) #endif diff --git a/source/row_common.cc b/source/row_common.cc index 8368024a8..b8e14199d 100644 --- a/source/row_common.cc +++ b/source/row_common.cc @@ -202,8 +202,9 @@ void ARGBToRGB565Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { uint8 b1 = src_argb[4] >> 3; uint8 g1 = src_argb[5] >> 2; uint8 r1 = src_argb[6] >> 3; - WRITEWORD(dst_rgb, b0 | (g0 << 5) | (r0 << 11) | (b1 << 16) | (g1 << 21) | - (r1 << 27)); + WRITEWORD( + dst_rgb, + b0 | (g0 << 5) | (r0 << 11) | (b1 << 16) | (g1 << 21) | (r1 << 27)); dst_rgb += 4; src_argb += 8; } @@ -237,8 +238,9 @@ void ARGBToRGB565DitherRow_C(const uint8* src_argb, uint8 b1 = clamp255(src_argb[4] + dither1) >> 3; uint8 g1 = clamp255(src_argb[5] + dither1) >> 2; uint8 r1 = clamp255(src_argb[6] + dither1) >> 3; - WRITEWORD(dst_rgb, b0 | (g0 << 5) | (r0 << 11) | (b1 << 16) | (g1 << 21) | - (r1 << 27)); + WRITEWORD( + dst_rgb, + b0 | (g0 << 5) | (r0 << 11) | (b1 << 16) | (g1 << 21) | (r1 << 27)); dst_rgb += 4; src_argb += 8; } diff --git a/source/row_mips.cc b/source/row_mips.cc index ce30fc7e1..ff58e8c1c 100644 --- a/source/row_mips.cc +++ b/source/row_mips.cc @@ -585,126 +585,89 @@ void MirrorUVRow_DSPR2(const uint8* src_uv, : "t0", "t1", "t2", "t3", "t4", "t5", "t7", "t8", "t9"); } -// Convert (4 Y and 2 VU) I422 and arrange RGB values into -// t5 = | 0 | B0 | 0 | b0 | -// t4 = | 0 | B1 | 0 | b1 | -// t9 = | 0 | G0 | 0 | g0 | -// t8 = | 0 | G1 | 0 | g1 | -// t2 = | 0 | R0 | 0 | r0 | -// t1 = | 0 | R1 | 0 | r1 | -#define YUVTORGB \ - "lw $t0, 0(%[y_buf]) \n" \ - "lhu $t1, 0(%[u_buf]) \n" \ - "lhu $t2, 0(%[v_buf]) \n" \ - "preceu.ph.qbr $t1, $t1 \n" \ - "preceu.ph.qbr $t2, $t2 \n" \ - "preceu.ph.qbra $t3, $t0 \n" \ - "preceu.ph.qbla $t0, $t0 \n" \ - "subu.ph $t1, $t1, $s5 \n" \ - "subu.ph $t2, $t2, $s5 \n" \ - "subu.ph $t3, $t3, $s4 \n" \ - "subu.ph $t0, $t0, $s4 \n" \ - "mul.ph $t3, $t3, $s0 \n" \ - "mul.ph $t0, $t0, $s0 \n" \ - "shll.ph $t4, $t1, 0x7 \n" \ - "subu.ph $t4, $t4, $t1 \n" \ - "mul.ph $t6, $t1, $s1 \n" \ - "mul.ph $t1, $t2, $s2 \n" \ - "addq_s.ph $t5, $t4, $t3 \n" \ - "addq_s.ph $t4, $t4, $t0 \n" \ - "shra.ph $t5, $t5, 6 \n" \ - "shra.ph $t4, $t4, 6 \n" \ - "addiu %[u_buf], 2 \n" \ - "addiu %[v_buf], 2 \n" \ - "addu.ph $t6, $t6, $t1 \n" \ - "mul.ph $t1, $t2, $s3 \n" \ - "addu.ph $t9, $t6, $t3 \n" \ - "addu.ph $t8, $t6, $t0 \n" \ - "shra.ph $t9, $t9, 6 \n" \ - "shra.ph $t8, $t8, 6 \n" \ - "addu.ph $t2, $t1, $t3 \n" \ - "addu.ph $t1, $t1, $t0 \n" \ - "shra.ph $t2, $t2, 6 \n" \ - "shra.ph $t1, $t1, 6 \n" \ - "subu.ph $t5, $t5, $s5 \n" \ - "subu.ph $t4, $t4, $s5 \n" \ - "subu.ph $t9, $t9, $s5 \n" \ - "subu.ph $t8, $t8, $s5 \n" \ - "subu.ph $t2, $t2, $s5 \n" \ - "subu.ph $t1, $t1, $s5 \n" \ - "shll_s.ph $t5, $t5, 8 \n" \ - "shll_s.ph $t4, $t4, 8 \n" \ - "shll_s.ph $t9, $t9, 8 \n" \ - "shll_s.ph $t8, $t8, 8 \n" \ - "shll_s.ph $t2, $t2, 8 \n" \ - "shll_s.ph $t1, $t1, 8 \n" \ - "shra.ph $t5, $t5, 8 \n" \ - "shra.ph $t4, $t4, 8 \n" \ - "shra.ph $t9, $t9, 8 \n" \ - "shra.ph $t8, $t8, 8 \n" \ - "shra.ph $t2, $t2, 8 \n" \ - "shra.ph $t1, $t1, 8 \n" \ - "addu.ph $t5, $t5, $s5 \n" \ - "addu.ph $t4, $t4, $s5 \n" \ - "addu.ph $t9, $t9, $s5 \n" \ - "addu.ph $t8, $t8, $s5 \n" \ - "addu.ph $t2, $t2, $s5 \n" \ - "addu.ph $t1, $t1, $s5 \n" - -// TODO(fbarchard): accept yuv conversion constants. -void I422ToARGBRow_DSPR2(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void I422ToARGBRow_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, uint8* rgb_buf, const struct YuvConstants* yuvconstants, int width) { - __asm__ __volatile__( - ".set push \n" - ".set noreorder \n" - "beqz %[width], 2f \n" - " repl.ph $s0, 74 \n" // |YG|YG| = |74|74| - "repl.ph $s1, -25 \n" // |UG|UG| = |-25|-25| - "repl.ph $s2, -52 \n" // |VG|VG| = |-52|-52| - "repl.ph $s3, 102 \n" // |VR|VR| = |102|102| - "repl.ph $s4, 16 \n" // |0|16|0|16| - "repl.ph $s5, 128 \n" // |128|128| // clipping - "lui $s6, 0xff00 \n" - "ori $s6, 0xff00 \n" // |ff|00|ff|00|ff| + int x; + uint32 tmp_ub = yuvconstants->kUVToB[0]; + uint32 tmp_ug = yuvconstants->kUVToG[0]; + uint32 tmp_vg = yuvconstants->kUVToG[1]; + uint32 tmp_vr = yuvconstants->kUVToR[1]; + uint32 tmp_bb = yuvconstants->kUVBiasB[0]; + uint32 tmp_bg = yuvconstants->kUVBiasG[0]; + uint32 tmp_br = yuvconstants->kUVBiasR[0]; + uint32 yg = yuvconstants->kYToRgb[0]; + uint32 tmp_yg; + uint32 tmp_mask = 0x7fff7fff; + tmp_bb = ((uint)(tmp_bb & 0xffff) << 16) | (tmp_bb & 0xffff); + tmp_bg = ((uint)(tmp_bg & 0xffff) << 16) | (tmp_bg & 0xffff); + tmp_br = ((uint)(tmp_br & 0xffff) << 16) | (tmp_br & 0xffff); + tmp_yg = ((uint)(yg & 0xffff) << 16) | (yg & 0xffff); + tmp_ub = ~(((uint)(tmp_ub & 0xffff) << 16) | (tmp_ub & 0xffff)) + 0x00010001; + tmp_ug = ((uint)(tmp_ug & 0xffff) << 16) | (tmp_ug & 0xffff); + tmp_vg = ((uint)(tmp_vg & 0xffff) << 16) | (tmp_vg & 0xffff); + tmp_vr = ~(((uint)(tmp_vr & 0xffff) << 16) | (tmp_vr & 0xffff)) + 0x00010001; + yg = yg * 0x0101; - "1: \n" YUVTORGB - // Arranging into argb format - "precr.qb.ph $t4, $t8, $t4 \n" // |G1|g1|B1|b1| - "precr.qb.ph $t5, $t9, $t5 \n" // |G0|g0|B0|b0| - "addiu %[width], -4 \n" - "precrq.qb.ph $t8, $t4, $t5 \n" // |G1|B1|G0|B0| - "precr.qb.ph $t9, $t4, $t5 \n" // |g1|b1|g0|b0| - "precr.qb.ph $t2, $t1, $t2 \n" // |R1|r1|R0|r0| - - "addiu %[y_buf], 4 \n" - "preceu.ph.qbla $t1, $t2 \n" // |0 |R1|0 |R0| - "preceu.ph.qbra $t2, $t2 \n" // |0 |r1|0 |r0| - "or $t1, $t1, $s6 \n" // |ff|R1|ff|R0| - "or $t2, $t2, $s6 \n" // |ff|r1|ff|r0| - "precrq.ph.w $t0, $t2, $t9 \n" // |ff|r1|g1|b1| - "precrq.ph.w $t3, $t1, $t8 \n" // |ff|R1|G1|B1| - "sll $t9, $t9, 16 \n" - "sll $t8, $t8, 16 \n" - "packrl.ph $t2, $t2, $t9 \n" // |ff|r0|g0|b0| - "packrl.ph $t1, $t1, $t8 \n" // |ff|R0|G0|B0| - // Store results. - "sw $t2, 0(%[rgb_buf]) \n" - "sw $t0, 4(%[rgb_buf]) \n" - "sw $t1, 8(%[rgb_buf]) \n" - "sw $t3, 12(%[rgb_buf]) \n" - "bnez %[width], 1b \n" - " addiu %[rgb_buf], 16 \n" - "2: \n" - ".set pop \n" - : [y_buf] "+r"(y_buf), [u_buf] "+r"(u_buf), [v_buf] "+r"(v_buf), - [width] "+r"(width), [rgb_buf] "+r"(rgb_buf) - : - : "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9", "s0", "s1", - "s2", "s3", "s4", "s5", "s6"); + for (x = 0; x < width - 1; x += 2) { + uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + uint32 tmp_t6, tmp_t7, tmp_t8, tmp_t9; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lbu %[tmp_t7], 0(%[src_y]) \n" + "lbu %[tmp_t1], 1(%[src_y]) \n" + "mul %[tmp_t7], %[tmp_t7], %[yg] \n" + "mul %[tmp_t1], %[tmp_t1], %[yg] \n" + "lbu %[tmp_t2], 0(%[src_u]) \n" + "lbu %[tmp_t3], 0(%[src_v]) \n" + "replv.ph %[tmp_t2], %[tmp_t2] \n" + "replv.ph %[tmp_t3], %[tmp_t3] \n" + "mul.ph %[tmp_t4], %[tmp_t2], %[tmp_ub] \n" + "mul.ph %[tmp_t5], %[tmp_t2], %[tmp_ug] \n" + "mul.ph %[tmp_t6], %[tmp_t3], %[tmp_vr] \n" + "mul.ph %[tmp_t3], %[tmp_t3], %[tmp_vg] \n" + "srl %[tmp_t7], %[tmp_t7], 16 \n" + "ins %[tmp_t1], %[tmp_t7], 0, 16 \n" + "addq_s.ph %[tmp_t7], %[tmp_t1], %[tmp_bb] \n" + "addq_s.ph %[tmp_t8], %[tmp_t1], %[tmp_bg] \n" + "addq_s.ph %[tmp_t9], %[tmp_t1], %[tmp_br] \n" + "addq_s.ph %[tmp_t5], %[tmp_t5], %[tmp_t3] \n" + "addq_s.ph %[tmp_t7], %[tmp_t7], %[tmp_t4] \n" + "subq_s.ph %[tmp_t8], %[tmp_t8], %[tmp_t5] \n" + "addq_s.ph %[tmp_t9], %[tmp_t9], %[tmp_t6] \n" + "shra.ph %[tmp_t7], %[tmp_t7], 6 \n" + "shra.ph %[tmp_t8], %[tmp_t8], 6 \n" + "shra.ph %[tmp_t9], %[tmp_t9], 6 \n" + "shll_s.ph %[tmp_t7], %[tmp_t7], 7 \n" + "shll_s.ph %[tmp_t8], %[tmp_t8], 7 \n" + "shll_s.ph %[tmp_t9], %[tmp_t9], 7 \n" + "precrqu_s.qb.ph %[tmp_t8], %[tmp_mask], %[tmp_t8] \n" + "precrqu_s.qb.ph %[tmp_t7], %[tmp_t9], %[tmp_t7] \n" + "precrq.ph.w %[tmp_t9], %[tmp_t8], %[tmp_t7] \n" + "ins %[tmp_t7], %[tmp_t8], 16, 16 \n" + "precr.qb.ph %[tmp_t8], %[tmp_t9], %[tmp_t7] \n" + "precrq.qb.ph %[tmp_t7], %[tmp_t9], %[tmp_t7] \n" + "sw %[tmp_t8], 0(%[rgb_buf]) \n" + "sw %[tmp_t7], 4(%[rgb_buf]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [tmp_t9] "=&r"(tmp_t9) + : [src_y] "r"(src_y), [src_u] "r"(src_u), [src_v] "r"(src_v), + [tmp_ub] "r"(tmp_ub), [tmp_ug] "r"(tmp_ug), [yg] "r"(yg), + [tmp_vg] "r"(tmp_vg), [tmp_vr] "r"(tmp_vr), [tmp_bb] "r"(tmp_bb), + [tmp_bg] "r"(tmp_bg), [tmp_br] "r"(tmp_br), [tmp_yg] "r"(tmp_yg), + [rgb_buf] "r"(rgb_buf), [tmp_mask] "r"(tmp_mask)); + src_y += 2; + src_u += 1; + src_v += 1; + rgb_buf += 8; // Advance 4 pixels. + } } // Bilinear filter 8x2 -> 8x1 @@ -740,10 +703,10 @@ void InterpolateRow_DSPR2(uint8* dst_ptr, "addq.ph $t7, $t7, $t9 \n" "addq.ph $t2, $t2, $t4 \n" "addq.ph $t3, $t3, $t5 \n" - "shra.ph $t6, $t6, 8 \n" - "shra.ph $t7, $t7, 8 \n" - "shra.ph $t2, $t2, 8 \n" - "shra.ph $t3, $t3, 8 \n" + "shra_r.ph $t6, $t6, 8 \n" + "shra_r.ph $t7, $t7, 8 \n" + "shra_r.ph $t2, $t2, 8 \n" + "shra_r.ph $t3, $t3, 8 \n" "precr.qb.ph $t6, $t6, $t7 \n" "precr.qb.ph $t2, $t2, $t3 \n" "addiu %[src_ptr], %[src_ptr], 8 \n" @@ -761,6 +724,993 @@ void InterpolateRow_DSPR2(uint8* dst_ptr, [y0_fraction] "r"(y0_fraction), [src_stride] "r"(src_stride) : "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9"); } +#include +void RGB24ToARGBRow_DSPR2(const uint8* src_rgb24, uint8* dst_argb, int width) { + int x; + uint32 tmp_mask = 0xff; + uint32 tmp_t1; + for (x = 0; x < (width - 1); ++x) { + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "ulw %[tmp_t1], 0(%[src_rgb24]) \n" + "addiu %[dst_argb], %[dst_argb], 4 \n" + "addiu %[src_rgb24], %[src_rgb24], 3 \n" + "ins %[tmp_t1], %[tmp_mask], 24, 8 \n" + "sw %[tmp_t1], -4(%[dst_argb]) \n" + ".set pop \n" + : [src_rgb24] "+r"(src_rgb24), [dst_argb] "+r"(dst_argb), + [tmp_t1] "=&r"(tmp_t1) + : [tmp_mask] "r"(tmp_mask) + : "memory"); + } + uint8 b = src_rgb24[0]; + uint8 g = src_rgb24[1]; + uint8 r = src_rgb24[2]; + dst_argb[0] = b; + dst_argb[1] = g; + dst_argb[2] = r; + dst_argb[3] = 255u; +} + +void RAWToARGBRow_DSPR2(const uint8* src_raw, uint8* dst_argb, int width) { + int x; + uint32 tmp_mask = 0xff; + uint32 tmp_t1, tmp_t2; + for (x = 0; x < (width - 1); ++x) { + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "ulw %[tmp_t1], 0(%[src_raw]) \n" + "addiu %[dst_argb], %[dst_argb], 4 \n" + "addiu %[src_raw], %[src_raw], 3 \n" + "srl %[tmp_t2], %[tmp_t1], 16 \n" + "ins %[tmp_t1], %[tmp_mask], 24, 8 \n" + "ins %[tmp_t1], %[tmp_t1], 16, 8 \n" + "ins %[tmp_t1], %[tmp_t2], 0, 8 \n" + "sw %[tmp_t1], -4(%[dst_argb]) \n" + ".set pop \n" + : [src_raw] "+r"(src_raw), [dst_argb] "+r"(dst_argb), + [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2) + : [tmp_mask] "r"(tmp_mask) + : "memory"); + } + uint8 r = src_raw[0]; + uint8 g = src_raw[1]; + uint8 b = src_raw[2]; + dst_argb[0] = b; + dst_argb[1] = g; + dst_argb[2] = r; + dst_argb[3] = 255u; +} + +void RGB565ToARGBRow_DSPR2(const uint8* src_rgb565, + uint8* dst_argb, + int width) { + int x; + uint32 tmp_mask = 0xff; + uint32 tmp_t1, tmp_t2, tmp_t3; + for (x = 0; x < width; ++x) { + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lhu %[tmp_t1], 0(%[src_rgb565]) \n" + "addiu %[dst_argb], %[dst_argb], 4 \n" + "addiu %[src_rgb565], %[src_rgb565], 2 \n" + "sll %[tmp_t2], %[tmp_t1], 8 \n" + "ins %[tmp_t2], %[tmp_mask], 24,8 \n" + "ins %[tmp_t2], %[tmp_t1], 3, 16 \n" + "ins %[tmp_t2], %[tmp_t1], 5, 11 \n" + "srl %[tmp_t3], %[tmp_t1], 9 \n" + "ins %[tmp_t2], %[tmp_t3], 8, 2 \n" + "ins %[tmp_t2], %[tmp_t1], 3, 5 \n" + "srl %[tmp_t3], %[tmp_t1], 2 \n" + "ins %[tmp_t2], %[tmp_t3], 0, 3 \n" + "sw %[tmp_t2], -4(%[dst_argb]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [src_rgb565] "+r"(src_rgb565), + [dst_argb] "+r"(dst_argb) + : [tmp_mask] "r"(tmp_mask)); + } +} + +void ARGB1555ToARGBRow_DSPR2(const uint8* src_argb1555, + uint8* dst_argb, + int width) { + int x; + uint32 tmp_t1, tmp_t2, tmp_t3; + for (x = 0; x < width; ++x) { + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lh %[tmp_t1], 0(%[src_argb1555]) \n" + "addiu %[dst_argb], %[dst_argb], 4 \n" + "addiu %[src_argb1555], %[src_argb1555], 2 \n" + "sll %[tmp_t2], %[tmp_t1], 9 \n" + "ins %[tmp_t2], %[tmp_t1], 4, 15 \n" + "ins %[tmp_t2], %[tmp_t1], 6, 10 \n" + "srl %[tmp_t3], %[tmp_t1], 7 \n" + "ins %[tmp_t2], %[tmp_t3], 8, 3 \n" + "ins %[tmp_t2], %[tmp_t1], 3, 5 \n" + "srl %[tmp_t3], %[tmp_t1], 2 \n" + "ins %[tmp_t2], %[tmp_t3], 0, 3 \n" + "sw %[tmp_t2], -4(%[dst_argb]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [src_argb1555] "+r"(src_argb1555), + [dst_argb] "+r"(dst_argb) + :); + } +} + +void ARGB4444ToARGBRow_DSPR2(const uint8* src_argb4444, + uint8* dst_argb, + int width) { + int x; + uint32 tmp_t1; + for (x = 0; x < width; ++x) { + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lh %[tmp_t1], 0(%[src_argb4444]) \n" + "addiu %[dst_argb], %[dst_argb], 4 \n" + "addiu %[src_argb4444], %[src_argb4444], 2 \n" + "ins %[tmp_t1], %[tmp_t1], 16, 16 \n" + "ins %[tmp_t1], %[tmp_t1], 12, 16 \n" + "ins %[tmp_t1], %[tmp_t1], 8, 12 \n" + "ins %[tmp_t1], %[tmp_t1], 4, 8 \n" + "sw %[tmp_t1], -4(%[dst_argb]) \n" + ".set pop \n" + : [src_argb4444] "+r"(src_argb4444), [dst_argb] "+r"(dst_argb), + [tmp_t1] "=&r"(tmp_t1)); + } +} + +void I444ToARGBRow_DSPR2(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + const struct YuvConstants* yuvconstants, + int width) { + int x; + uint32 tmp_ub = yuvconstants->kUVToB[0]; + uint32 tmp_ug = yuvconstants->kUVToG[0]; + uint32 tmp_vg = yuvconstants->kUVToG[1]; + uint32 tmp_vr = yuvconstants->kUVToR[1]; + uint32 tmp_bb = yuvconstants->kUVBiasB[0]; + uint32 tmp_bg = yuvconstants->kUVBiasG[0]; + uint32 tmp_br = yuvconstants->kUVBiasR[0]; + uint32 yg = yuvconstants->kYToRgb[0]; + uint32 tmp_mask = 0x7fff7fff; + uint32 tmp_yg; + + tmp_bb = ((uint)(tmp_bb & 0xffff) << 16) | (tmp_bb & 0xffff); + tmp_bg = ((uint)(tmp_bg & 0xffff) << 16) | (tmp_bg & 0xffff); + tmp_br = ((uint)(tmp_br & 0xffff) << 16) | (tmp_br & 0xffff); + tmp_yg = ((uint)(yg & 0xffff) << 16) | (yg & 0xffff); + tmp_ub = ~(((uint)(tmp_ub & 0xffff) << 16) | (tmp_ub & 0xffff)) + 0x00010001; + tmp_ug = ((uint)(tmp_ug & 0xffff) << 16) | (tmp_ug & 0xffff); + tmp_vg = ((uint)(tmp_vg & 0xffff) << 16) | (tmp_vg & 0xffff); + tmp_vr = ~(((uint)(tmp_vr & 0xffff) << 16) | (tmp_vr & 0xffff)) + 0x00010001; + yg = yg * 0x0101; + + for (x = 0; x < width - 1; x += 2) { + uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + uint32 tmp_t6, tmp_t7, tmp_t8, tmp_t9; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lbu %[tmp_t7], 0(%[y_buf]) \n" + "lbu %[tmp_t1], 1(%[y_buf]) \n" + "mul %[tmp_t7], %[tmp_t7], %[yg] \n" + "mul %[tmp_t1], %[tmp_t1], %[yg] \n" + "lh %[tmp_t2], 0(%[u_buf]) \n" + "lh %[tmp_t3], 0(%[v_buf]) \n" + "preceu.ph.qbr %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t3], %[tmp_t3] \n" + "mul.ph %[tmp_t4], %[tmp_t2], %[tmp_ub] \n" + "mul.ph %[tmp_t5], %[tmp_t2], %[tmp_ug] \n" + "mul.ph %[tmp_t6], %[tmp_t3], %[tmp_vr] \n" + "mul.ph %[tmp_t3], %[tmp_t3], %[tmp_vg] \n" + "srl %[tmp_t7], %[tmp_t7], 16 \n" + "ins %[tmp_t1], %[tmp_t7], 0, 16 \n" + "addq_s.ph %[tmp_t7], %[tmp_t1], %[tmp_bb] \n" + "addq_s.ph %[tmp_t8], %[tmp_t1], %[tmp_bg] \n" + "addq_s.ph %[tmp_t9], %[tmp_t1], %[tmp_br] \n" + "addq_s.ph %[tmp_t5], %[tmp_t5], %[tmp_t3] \n" + "addq_s.ph %[tmp_t7], %[tmp_t7], %[tmp_t4] \n" + "subq_s.ph %[tmp_t8], %[tmp_t8], %[tmp_t5] \n" + "addq_s.ph %[tmp_t9], %[tmp_t9], %[tmp_t6] \n" + "shra.ph %[tmp_t7], %[tmp_t7], 6 \n" + "shra.ph %[tmp_t8], %[tmp_t8], 6 \n" + "shra.ph %[tmp_t9], %[tmp_t9], 6 \n" + "shll_s.ph %[tmp_t7], %[tmp_t7], 7 \n" + "shll_s.ph %[tmp_t8], %[tmp_t8], 7 \n" + "shll_s.ph %[tmp_t9], %[tmp_t9], 7 \n" + "precrqu_s.qb.ph %[tmp_t8], %[tmp_mask], %[tmp_t8] \n" + "precrqu_s.qb.ph %[tmp_t7], %[tmp_t9], %[tmp_t7] \n" + "precrq.ph.w %[tmp_t2], %[tmp_t8], %[tmp_t7] \n" + "ins %[tmp_t7], %[tmp_t8], 16, 16 \n" + "precr.qb.ph %[tmp_t8], %[tmp_t2], %[tmp_t7] \n" + "precrq.qb.ph %[tmp_t7], %[tmp_t2], %[tmp_t7] \n" + "sw %[tmp_t8], 0(%[rgb_buf]) \n" + "sw %[tmp_t7], 4(%[rgb_buf]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [tmp_t9] "=&r"(tmp_t9) + : [y_buf] "r"(y_buf), [yg] "r"(yg), [u_buf] "r"(u_buf), + [v_buf] "r"(v_buf), [tmp_ub] "r"(tmp_ub), [tmp_ug] "r"(tmp_ug), + [tmp_vg] "r"(tmp_vg), [tmp_vr] "r"(tmp_vr), [tmp_bb] "r"(tmp_bb), + [tmp_bg] "r"(tmp_bg), [tmp_br] "r"(tmp_br), [tmp_yg] "r"(tmp_yg), + [rgb_buf] "r"(rgb_buf), [tmp_mask] "r"(tmp_mask)); + y_buf += 2; + u_buf += 2; + v_buf += 2; + rgb_buf += 8; // Advance 1 pixel. + } +} + +void I422ToARGB4444Row_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb4444, + const struct YuvConstants* yuvconstants, + int width) { + int x; + uint32 tmp_ub = yuvconstants->kUVToB[0]; + uint32 tmp_ug = yuvconstants->kUVToG[0]; + uint32 tmp_vg = yuvconstants->kUVToG[1]; + uint32 tmp_vr = yuvconstants->kUVToR[1]; + uint32 tmp_bb = yuvconstants->kUVBiasB[0]; + uint32 tmp_bg = yuvconstants->kUVBiasG[0]; + uint32 tmp_br = yuvconstants->kUVBiasR[0]; + uint32 yg = yuvconstants->kYToRgb[0]; + uint32 tmp_yg; + uint32 tmp_mask = 0x7fff7fff; + tmp_bb = ((uint)(tmp_bb & 0xffff) << 16) | (tmp_bb & 0xffff); + tmp_bg = ((uint)(tmp_bg & 0xffff) << 16) | (tmp_bg & 0xffff); + tmp_br = ((uint)(tmp_br & 0xffff) << 16) | (tmp_br & 0xffff); + tmp_yg = ((uint)(yg & 0xffff) << 16) | (yg & 0xffff); + tmp_ub = ~(((uint)(tmp_ub & 0xffff) << 16) | (tmp_ub & 0xffff)) + 0x00010001; + tmp_ug = ((uint)(tmp_ug & 0xffff) << 16) | (tmp_ug & 0xffff); + tmp_vg = ((uint)(tmp_vg & 0xffff) << 16) | (tmp_vg & 0xffff); + tmp_vr = ~(((uint)(tmp_vr & 0xffff) << 16) | (tmp_vr & 0xffff)) + 0x00010001; + yg = yg * 0x0101; + + for (x = 0; x < width - 1; x += 2) { + uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + uint32 tmp_t6, tmp_t7, tmp_t8, tmp_t9; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lbu %[tmp_t7], 0(%[src_y]) \n" + "lbu %[tmp_t1], 1(%[src_y]) \n" + "mul %[tmp_t7], %[tmp_t7], %[yg] \n" + "mul %[tmp_t1], %[tmp_t1], %[yg] \n" + "lbu %[tmp_t2], 0(%[src_u]) \n" + "lbu %[tmp_t3], 0(%[src_v]) \n" + "replv.ph %[tmp_t2], %[tmp_t2] \n" + "replv.ph %[tmp_t3], %[tmp_t3] \n" + "mul.ph %[tmp_t4], %[tmp_t2], %[tmp_ub] \n" + "mul.ph %[tmp_t5], %[tmp_t2], %[tmp_ug] \n" + "mul.ph %[tmp_t6], %[tmp_t3], %[tmp_vr] \n" + "mul.ph %[tmp_t3], %[tmp_t3], %[tmp_vg] \n" + "srl %[tmp_t7], %[tmp_t7], 16 \n" + "ins %[tmp_t1], %[tmp_t7], 0, 16 \n" + "addq_s.ph %[tmp_t7], %[tmp_t1], %[tmp_bb] \n" + "addq_s.ph %[tmp_t8], %[tmp_t1], %[tmp_bg] \n" + "addq_s.ph %[tmp_t9], %[tmp_t1], %[tmp_br] \n" + "addq_s.ph %[tmp_t5], %[tmp_t5], %[tmp_t3] \n" + "addq_s.ph %[tmp_t7], %[tmp_t7], %[tmp_t4] \n" + "subq_s.ph %[tmp_t8], %[tmp_t8], %[tmp_t5] \n" + "addq_s.ph %[tmp_t9], %[tmp_t9], %[tmp_t6] \n" + "shra.ph %[tmp_t7], %[tmp_t7], 6 \n" + "shra.ph %[tmp_t8], %[tmp_t8], 6 \n" + "shra.ph %[tmp_t9], %[tmp_t9], 6 \n" + "shll_s.ph %[tmp_t7], %[tmp_t7], 7 \n" + "shll_s.ph %[tmp_t8], %[tmp_t8], 7 \n" + "shll_s.ph %[tmp_t9], %[tmp_t9], 7 \n" + "precrqu_s.qb.ph %[tmp_t8], %[tmp_mask], %[tmp_t8] \n" + "precrqu_s.qb.ph %[tmp_t7], %[tmp_t9], %[tmp_t7] \n" + "precrq.ph.w %[tmp_t2], %[tmp_t8], %[tmp_t7] \n" + "ins %[tmp_t7], %[tmp_t8], 16, 16 \n" + "precr.qb.ph %[tmp_t8], %[tmp_t2], %[tmp_t7] \n" + "precrq.qb.ph %[tmp_t7], %[tmp_t2], %[tmp_t7] \n" + "shrl.qb %[tmp_t1], %[tmp_t8], 4 \n" + "shrl.qb %[tmp_t2], %[tmp_t7], 4 \n" + "shrl.ph %[tmp_t8], %[tmp_t1], 4 \n" + "shrl.ph %[tmp_t7], %[tmp_t2], 4 \n" + "or %[tmp_t8], %[tmp_t8], %[tmp_t1] \n" + "or %[tmp_t7], %[tmp_t7], %[tmp_t2] \n" + "precr.qb.ph %[tmp_t8], %[tmp_t7], %[tmp_t8] \n" + "sw %[tmp_t8], 0(%[dst_argb4444]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [tmp_t9] "=&r"(tmp_t9) + : [dst_argb4444] "r"(dst_argb4444), [yg] "r"(yg), [src_u] "r"(src_u), + [src_v] "r"(src_v), [src_y] "r"(src_y), [tmp_ub] "r"(tmp_ub), + [tmp_ug] "r"(tmp_ug), [tmp_vg] "r"(tmp_vg), [tmp_vr] "r"(tmp_vr), + [tmp_bb] "r"(tmp_bb), [tmp_bg] "r"(tmp_bg), [tmp_br] "r"(tmp_br), + [tmp_yg] "r"(tmp_yg), [tmp_mask] "r"(tmp_mask)); + src_y += 2; + src_u += 1; + src_v += 1; + dst_argb4444 += 4; // Advance 2 pixels. + } +} + +void I422ToARGB1555Row_DSPR2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb1555, + const struct YuvConstants* yuvconstants, + int width) { + int x; + uint32 tmp_ub = yuvconstants->kUVToB[0]; + uint32 tmp_ug = yuvconstants->kUVToG[0]; + uint32 tmp_vg = yuvconstants->kUVToG[1]; + uint32 tmp_vr = yuvconstants->kUVToR[1]; + uint32 tmp_bb = yuvconstants->kUVBiasB[0]; + uint32 tmp_bg = yuvconstants->kUVBiasG[0]; + uint32 tmp_br = yuvconstants->kUVBiasR[0]; + uint32 yg = yuvconstants->kYToRgb[0]; + uint32 tmp_yg; + uint32 tmp_mask = 0x80008000; + tmp_bb = ((uint)(tmp_bb & 0xffff) << 16) | (tmp_bb & 0xffff); + tmp_bg = ((uint)(tmp_bg & 0xffff) << 16) | (tmp_bg & 0xffff); + tmp_br = ((uint)(tmp_br & 0xffff) << 16) | (tmp_br & 0xffff); + tmp_yg = ((uint)(yg & 0xffff) << 16) | (yg & 0xffff); + tmp_ub = ~(((uint)(tmp_ub & 0xffff) << 16) | (tmp_ub & 0xffff)) + 0x00010001; + tmp_ug = ((uint)(tmp_ug & 0xffff) << 16) | (tmp_ug & 0xffff); + tmp_vg = ((uint)(tmp_vg & 0xffff) << 16) | (tmp_vg & 0xffff); + tmp_vr = ~(((uint)(tmp_vr & 0xffff) << 16) | (tmp_vr & 0xffff)) + 0x00010001; + yg = yg * 0x0101; + + for (x = 0; x < width - 1; x += 2) { + uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + uint32 tmp_t6, tmp_t7, tmp_t8, tmp_t9; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lbu %[tmp_t7], 0(%[src_y]) \n" + "lbu %[tmp_t1], 1(%[src_y]) \n" + "mul %[tmp_t7], %[tmp_t7], %[yg] \n" + "mul %[tmp_t1], %[tmp_t1], %[yg] \n" + "lbu %[tmp_t2], 0(%[src_u]) \n" + "lbu %[tmp_t3], 0(%[src_v]) \n" + "replv.ph %[tmp_t2], %[tmp_t2] \n" + "replv.ph %[tmp_t3], %[tmp_t3] \n" + "mul.ph %[tmp_t4], %[tmp_t2], %[tmp_ub] \n" + "mul.ph %[tmp_t5], %[tmp_t2], %[tmp_ug] \n" + "mul.ph %[tmp_t6], %[tmp_t3], %[tmp_vr] \n" + "mul.ph %[tmp_t3], %[tmp_t3], %[tmp_vg] \n" + "srl %[tmp_t7], %[tmp_t7], 16 \n" + "ins %[tmp_t1], %[tmp_t7], 0, 16 \n" + "addq_s.ph %[tmp_t7], %[tmp_t1], %[tmp_bb] \n" + "addq_s.ph %[tmp_t8], %[tmp_t1], %[tmp_bg] \n" + "addq_s.ph %[tmp_t9], %[tmp_t1], %[tmp_br] \n" + "addq_s.ph %[tmp_t5], %[tmp_t5], %[tmp_t3] \n" + "addq_s.ph %[tmp_t7], %[tmp_t7], %[tmp_t4] \n" + "subq_s.ph %[tmp_t8], %[tmp_t8], %[tmp_t5] \n" + "addq_s.ph %[tmp_t9], %[tmp_t9], %[tmp_t6] \n" + "shra.ph %[tmp_t7], %[tmp_t7], 6 \n" + "shra.ph %[tmp_t8], %[tmp_t8], 6 \n" + "shra.ph %[tmp_t9], %[tmp_t9], 6 \n" + "shll_s.ph %[tmp_t7], %[tmp_t7], 7 \n" + "shll_s.ph %[tmp_t8], %[tmp_t8], 7 \n" + "shll_s.ph %[tmp_t9], %[tmp_t9], 7 \n" + "precrqu_s.qb.ph %[tmp_t8], %[tmp_mask], %[tmp_t8] \n" + "precrqu_s.qb.ph %[tmp_t7], %[tmp_t9], %[tmp_t7] \n" + "precrq.ph.w %[tmp_t2], %[tmp_t8], %[tmp_t7] \n" + "ins %[tmp_t7], %[tmp_t8], 16, 16 \n" + "precr.qb.ph %[tmp_t8], %[tmp_t2], %[tmp_t7] \n" + "precrq.qb.ph %[tmp_t7], %[tmp_t2], %[tmp_t7] \n" + "ins %[tmp_t3], %[tmp_t8], 7, 24 \n" + "ins %[tmp_t3], %[tmp_t8], 10, 16 \n" + "ins %[tmp_t3], %[tmp_t8], 13, 8 \n" + "ins %[tmp_t4], %[tmp_t7], 7, 24 \n" + "ins %[tmp_t4], %[tmp_t7], 10, 16 \n" + "ins %[tmp_t4], %[tmp_t7], 13, 8 \n" + "precrq.ph.w %[tmp_t8], %[tmp_t4], %[tmp_t3] \n" + "or %[tmp_t8], %[tmp_t8], %[tmp_mask]\n" + "sw %[tmp_t8], 0(%[dst_argb1555]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [tmp_t9] "=&r"(tmp_t9) + : [dst_argb1555] "r"(dst_argb1555), [yg] "r"(yg), [src_u] "r"(src_u), + [src_v] "r"(src_v), [src_y] "r"(src_y), [tmp_ub] "r"(tmp_ub), + [tmp_ug] "r"(tmp_ug), [tmp_vg] "r"(tmp_vg), [tmp_vr] "r"(tmp_vr), + [tmp_bb] "r"(tmp_bb), [tmp_bg] "r"(tmp_bg), [tmp_br] "r"(tmp_br), + [tmp_yg] "r"(tmp_yg), [tmp_mask] "r"(tmp_mask)); + src_y += 2; + src_u += 1; + src_v += 1; + dst_argb1555 += 4; // Advance 2 pixels. + } +} + +void NV12ToARGBRow_DSPR2(const uint8* src_y, + const uint8* src_uv, + uint8* rgb_buf, + const struct YuvConstants* yuvconstants, + int width) { + int x; + uint32 tmp_ub = yuvconstants->kUVToB[0]; + uint32 tmp_ug = yuvconstants->kUVToG[0]; + uint32 tmp_vg = yuvconstants->kUVToG[1]; + uint32 tmp_vr = yuvconstants->kUVToR[1]; + uint32 tmp_bb = yuvconstants->kUVBiasB[0]; + uint32 tmp_bg = yuvconstants->kUVBiasG[0]; + uint32 tmp_br = yuvconstants->kUVBiasR[0]; + uint32 yg = yuvconstants->kYToRgb[0]; + uint32 tmp_mask = 0x7fff7fff; + uint32 tmp_yg; + tmp_bb = ((uint)(tmp_bb & 0xffff) << 16) | (tmp_bb & 0xffff); + tmp_bg = ((uint)(tmp_bg & 0xffff) << 16) | (tmp_bg & 0xffff); + tmp_br = ((uint)(tmp_br & 0xffff) << 16) | (tmp_br & 0xffff); + tmp_yg = ((uint)(yg & 0xffff) << 16) | (yg & 0xffff); + tmp_ub = ~(((uint)(tmp_ub & 0xffff) << 16) | (tmp_ub & 0xffff)) + 0x00010001; + tmp_ug = ((uint)(tmp_ug & 0xffff) << 16) | (tmp_ug & 0xffff); + tmp_vg = ((uint)(tmp_vg & 0xffff) << 16) | (tmp_vg & 0xffff); + tmp_vr = ~(((uint)(tmp_vr & 0xffff) << 16) | (tmp_vr & 0xffff)) + 0x00010001; + yg = yg * 0x0101; + + for (x = 0; x < width - 1; x += 2) { + uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + uint32 tmp_t6, tmp_t7, tmp_t8, tmp_t9; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lbu %[tmp_t7], 0(%[src_y]) \n" + "lbu %[tmp_t1], 1(%[src_y]) \n" + "mul %[tmp_t7], %[tmp_t7], %[yg] \n" + "mul %[tmp_t1], %[tmp_t1], %[yg] \n" + "lbu %[tmp_t2], 0(%[src_uv]) \n" + "lbu %[tmp_t3], 1(%[src_uv]) \n" + "replv.ph %[tmp_t2], %[tmp_t2] \n" + "replv.ph %[tmp_t3], %[tmp_t3] \n" + "mul.ph %[tmp_t4], %[tmp_t2], %[tmp_ub] \n" + "mul.ph %[tmp_t5], %[tmp_t2], %[tmp_ug] \n" + "mul.ph %[tmp_t6], %[tmp_t3], %[tmp_vr] \n" + "mul.ph %[tmp_t3], %[tmp_t3], %[tmp_vg] \n" + "srl %[tmp_t7], %[tmp_t7], 16 \n" + "ins %[tmp_t1], %[tmp_t7], 0, 16 \n" + "addq_s.ph %[tmp_t7], %[tmp_t1], %[tmp_bb] \n" + "addq_s.ph %[tmp_t8], %[tmp_t1], %[tmp_bg] \n" + "addq_s.ph %[tmp_t9], %[tmp_t1], %[tmp_br] \n" + "addq_s.ph %[tmp_t5], %[tmp_t5], %[tmp_t3] \n" + "addq_s.ph %[tmp_t7], %[tmp_t7], %[tmp_t4] \n" + "subq_s.ph %[tmp_t8], %[tmp_t8], %[tmp_t5] \n" + "addq_s.ph %[tmp_t9], %[tmp_t9], %[tmp_t6] \n" + "shra.ph %[tmp_t7], %[tmp_t7], 6 \n" + "shra.ph %[tmp_t8], %[tmp_t8], 6 \n" + "shra.ph %[tmp_t9], %[tmp_t9], 6 \n" + "shll_s.ph %[tmp_t7], %[tmp_t7], 7 \n" + "shll_s.ph %[tmp_t8], %[tmp_t8], 7 \n" + "shll_s.ph %[tmp_t9], %[tmp_t9], 7 \n" + "precrqu_s.qb.ph %[tmp_t8], %[tmp_mask], %[tmp_t8] \n" + "precrqu_s.qb.ph %[tmp_t7], %[tmp_t9], %[tmp_t7] \n" + "precrq.ph.w %[tmp_t2], %[tmp_t8], %[tmp_t7] \n" + "ins %[tmp_t7], %[tmp_t8], 16, 16 \n" + "precr.qb.ph %[tmp_t8], %[tmp_t2], %[tmp_t7] \n" + "precrq.qb.ph %[tmp_t7], %[tmp_t2], %[tmp_t7] \n" + "sw %[tmp_t8], 0(%[rgb_buf]) \n" + "sw %[tmp_t7], 4(%[rgb_buf]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [tmp_t9] "=&r"(tmp_t9) + : [src_y] "r"(src_y), [src_uv] "r"(src_uv), [yg] "r"(yg), + [tmp_ub] "r"(tmp_ub), [tmp_ug] "r"(tmp_ug), [tmp_vg] "r"(tmp_vg), + [tmp_vr] "r"(tmp_vr), [tmp_bb] "r"(tmp_bb), [tmp_bg] "r"(tmp_bg), + [tmp_br] "r"(tmp_br), [tmp_yg] "r"(tmp_yg), [rgb_buf] "r"(rgb_buf), + [tmp_mask] "r"(tmp_mask)); + + src_y += 2; + src_uv += 2; + rgb_buf += 8; // Advance 2 pixels. + } +} + +void BGRAToUVRow_DSPR2(const uint8* src_rgb0, + int src_stride_rgb, + uint8* dst_u, + uint8* dst_v, + int width) { + const uint8* src_rgb1 = src_rgb0 + src_stride_rgb; + int x; + int const1 = 0xffda0000; + int const2 = 0x0070ffb6; + int const3 = 0x00700000; + int const4 = 0xffeeffa2; + int const5 = 0x100; + for (x = 0; x < width - 1; x += 2) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_rgb0]) \n" + "lw %[tmp_t2], 4(%[src_rgb0]) \n" + "lw %[tmp_t3], 0(%[src_rgb1]) \n" + "lw %[tmp_t4], 4(%[src_rgb1]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t6] \n" + "addu.ph %[tmp_t7], %[tmp_t7], %[tmp_t8] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t2] \n" + "addu.ph %[tmp_t3], %[tmp_t3], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t7] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t3] \n" + "shrl.ph %[tmp_t5], %[tmp_t5], 2 \n" + "shrl.ph %[tmp_t1], %[tmp_t1], 2 \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "dpaq_s.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpaq_s.w.ph $ac1, %[tmp_t5], %[const3] \n" + "dpaq_s.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpaq_s.w.ph $ac1, %[tmp_t1], %[const4] \n" + "extr_r.w %[tmp_t7], $ac0, 9 \n" + "extr_r.w %[tmp_t8], $ac1, 9 \n" + "addiu %[dst_u], %[dst_u], 1 \n" + "addiu %[dst_v], %[dst_v], 1 \n" + "addiu %[src_rgb0], %[src_rgb0], 8 \n" + "addiu %[src_rgb1], %[src_rgb1], 8 \n" + "sb %[tmp_t7], -1(%[dst_u]) \n" + "sb %[tmp_t8], -1(%[dst_v]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_rgb0] "+r"(src_rgb0), [src_rgb1] "+r"(src_rgb1), + [dst_u] "+r"(dst_u), [dst_v] "+r"(dst_v) + : [const1] "r"(const1), [const2] "r"(const2), [const3] "r"(const3), + [const4] "r"(const4), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi"); + } +} + +void BGRAToYRow_DSPR2(const uint8* src_argb0, uint8* dst_y, int width) { + int x; + int const1 = 0x00420000; + int const2 = 0x00190081; + int const5 = 0x40; + for (x = 0; x < width; x += 4) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_argb0]) \n" + "lw %[tmp_t2], 4(%[src_argb0]) \n" + "lw %[tmp_t3], 8(%[src_argb0]) \n" + "lw %[tmp_t4], 12(%[src_argb0]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "mult $ac2, %[const5], %[const5] \n" + "mult $ac3, %[const5], %[const5] \n" + "dpa.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpa.w.ph $ac1, %[tmp_t6], %[const1] \n" + "dpa.w.ph $ac2, %[tmp_t7], %[const1] \n" + "dpa.w.ph $ac3, %[tmp_t8], %[const1] \n" + "dpa.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpa.w.ph $ac1, %[tmp_t2], %[const2] \n" + "dpa.w.ph $ac2, %[tmp_t3], %[const2] \n" + "dpa.w.ph $ac3, %[tmp_t4], %[const2] \n" + "extr_r.w %[tmp_t1], $ac0, 8 \n" + "extr_r.w %[tmp_t2], $ac1, 8 \n" + "extr_r.w %[tmp_t3], $ac2, 8 \n" + "extr_r.w %[tmp_t4], $ac3, 8 \n" + "addiu %[src_argb0],%[src_argb0], 16 \n" + "addiu %[dst_y], %[dst_y], 4 \n" + "sb %[tmp_t1], -4(%[dst_y]) \n" + "sb %[tmp_t2], -3(%[dst_y]) \n" + "sb %[tmp_t3], -2(%[dst_y]) \n" + "sb %[tmp_t4], -1(%[dst_y]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_argb0] "+r"(src_argb0), [dst_y] "+r"(dst_y) + : [const1] "r"(const1), [const2] "r"(const2), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi", "$ac2lo", "$ac2hi", "$ac3lo", + "$ac3hi"); + } +} + +void ABGRToUVRow_DSPR2(const uint8* src_rgb0, + int src_stride_rgb, + uint8* dst_u, + uint8* dst_v, + int width) { + const uint8* src_rgb1 = src_rgb0 + src_stride_rgb; + int x; + int const1 = 0xffb6ffda; + int const2 = 0x00000070; + int const3 = 0xffa20070; + int const4 = 0x0000ffee; + int const5 = 0x100; + + for (x = 0; x < width - 1; x += 2) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_rgb0]) \n" + "lw %[tmp_t2], 4(%[src_rgb0]) \n" + "lw %[tmp_t3], 0(%[src_rgb1]) \n" + "lw %[tmp_t4], 4(%[src_rgb1]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t6] \n" + "addu.ph %[tmp_t7], %[tmp_t7], %[tmp_t8] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t2] \n" + "addu.ph %[tmp_t3], %[tmp_t3], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t7] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t3] \n" + "shrl.ph %[tmp_t5], %[tmp_t5], 2 \n" + "shrl.ph %[tmp_t1], %[tmp_t1], 2 \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "dpaq_s.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpaq_s.w.ph $ac1, %[tmp_t5], %[const3] \n" + "dpaq_s.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpaq_s.w.ph $ac1, %[tmp_t1], %[const4] \n" + "extr_r.w %[tmp_t7], $ac0, 9 \n" + "extr_r.w %[tmp_t8], $ac1, 9 \n" + "addiu %[dst_u], %[dst_u], 1 \n" + "addiu %[dst_v], %[dst_v], 1 \n" + "addiu %[src_rgb0], %[src_rgb0], 8 \n" + "addiu %[src_rgb1], %[src_rgb1], 8 \n" + "sb %[tmp_t7], -1(%[dst_u]) \n" + "sb %[tmp_t8], -1(%[dst_v]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_rgb0] "+r"(src_rgb0), [src_rgb1] "+r"(src_rgb1), + [dst_u] "+r"(dst_u), [dst_v] "+r"(dst_v) + : [const1] "r"(const1), [const2] "r"(const2), [const3] "r"(const3), + [const4] "r"(const4), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi"); + } +} + +void ARGBToYRow_DSPR2(const uint8* src_argb0, uint8* dst_y, int width) { + int x; + int const1 = 0x00810019; + int const2 = 0x00000042; + int const5 = 0x40; + for (x = 0; x < width; x += 4) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_argb0]) \n" + "lw %[tmp_t2], 4(%[src_argb0]) \n" + "lw %[tmp_t3], 8(%[src_argb0]) \n" + "lw %[tmp_t4], 12(%[src_argb0]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "mult $ac2, %[const5], %[const5] \n" + "mult $ac3, %[const5], %[const5] \n" + "dpa.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpa.w.ph $ac1, %[tmp_t6], %[const1] \n" + "dpa.w.ph $ac2, %[tmp_t7], %[const1] \n" + "dpa.w.ph $ac3, %[tmp_t8], %[const1] \n" + "dpa.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpa.w.ph $ac1, %[tmp_t2], %[const2] \n" + "dpa.w.ph $ac2, %[tmp_t3], %[const2] \n" + "dpa.w.ph $ac3, %[tmp_t4], %[const2] \n" + "extr_r.w %[tmp_t1], $ac0, 8 \n" + "extr_r.w %[tmp_t2], $ac1, 8 \n" + "extr_r.w %[tmp_t3], $ac2, 8 \n" + "extr_r.w %[tmp_t4], $ac3, 8 \n" + "addiu %[dst_y], %[dst_y], 4 \n" + "addiu %[src_argb0],%[src_argb0], 16 \n" + "sb %[tmp_t1], -4(%[dst_y]) \n" + "sb %[tmp_t2], -3(%[dst_y]) \n" + "sb %[tmp_t3], -2(%[dst_y]) \n" + "sb %[tmp_t4], -1(%[dst_y]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_argb0] "+r"(src_argb0), [dst_y] "+r"(dst_y) + : [const1] "r"(const1), [const2] "r"(const2), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi", "$ac2lo", "$ac2hi", "$ac3lo", + "$ac3hi"); + } +} + +void ABGRToYRow_DSPR2(const uint8* src_argb0, uint8* dst_y, int width) { + int x; + int const1 = 0x00810042; + int const2 = 0x00000019; + int const5 = 0x40; + for (x = 0; x < width; x += 4) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_argb0]) \n" + "lw %[tmp_t2], 4(%[src_argb0]) \n" + "lw %[tmp_t3], 8(%[src_argb0]) \n" + "lw %[tmp_t4], 12(%[src_argb0]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "mult $ac2, %[const5], %[const5] \n" + "mult $ac3, %[const5], %[const5] \n" + "dpa.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpa.w.ph $ac1, %[tmp_t6], %[const1] \n" + "dpa.w.ph $ac2, %[tmp_t7], %[const1] \n" + "dpa.w.ph $ac3, %[tmp_t8], %[const1] \n" + "dpa.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpa.w.ph $ac1, %[tmp_t2], %[const2] \n" + "dpa.w.ph $ac2, %[tmp_t3], %[const2] \n" + "dpa.w.ph $ac3, %[tmp_t4], %[const2] \n" + "extr_r.w %[tmp_t1], $ac0, 8 \n" + "extr_r.w %[tmp_t2], $ac1, 8 \n" + "extr_r.w %[tmp_t3], $ac2, 8 \n" + "extr_r.w %[tmp_t4], $ac3, 8 \n" + "addiu %[src_argb0],%[src_argb0], 16 \n" + "addiu %[dst_y], %[dst_y], 4 \n" + "sb %[tmp_t1], -4(%[dst_y]) \n" + "sb %[tmp_t2], -3(%[dst_y]) \n" + "sb %[tmp_t3], -2(%[dst_y]) \n" + "sb %[tmp_t4], -1(%[dst_y]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_argb0] "+r"(src_argb0), [dst_y] "+r"(dst_y) + : [const1] "r"(const1), [const2] "r"(const2), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi", "$ac2lo", "$ac2hi", "$ac3lo", + "$ac3hi"); + } +} + +void RGBAToUVRow_DSPR2(const uint8* src_rgb0, + int src_stride_rgb, + uint8* dst_u, + uint8* dst_v, + int width) { + const uint8* src_rgb1 = src_rgb0 + src_stride_rgb; + int x; + int const1 = 0xffb60070; + int const2 = 0x0000ffda; + int const3 = 0xffa2ffee; + int const4 = 0x00000070; + int const5 = 0x100; + + for (x = 0; x < width - 1; x += 2) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "ulw %[tmp_t1], 0+1(%[src_rgb0]) \n" + "ulw %[tmp_t2], 4+1(%[src_rgb0]) \n" + "ulw %[tmp_t3], 0+1(%[src_rgb1]) \n" + "ulw %[tmp_t4], 4+1(%[src_rgb1]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t6] \n" + "addu.ph %[tmp_t7], %[tmp_t7], %[tmp_t8] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t2] \n" + "addu.ph %[tmp_t3], %[tmp_t3], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t7] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t3] \n" + "shrl.ph %[tmp_t5], %[tmp_t5], 2 \n" + "shrl.ph %[tmp_t1], %[tmp_t1], 2 \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "dpaq_s.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpaq_s.w.ph $ac1, %[tmp_t5], %[const3] \n" + "dpaq_s.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpaq_s.w.ph $ac1, %[tmp_t1], %[const4] \n" + "extr_r.w %[tmp_t7], $ac0, 9 \n" + "extr_r.w %[tmp_t8], $ac1, 9 \n" + "addiu %[src_rgb0], %[src_rgb0], 8 \n" + "addiu %[src_rgb1], %[src_rgb1], 8 \n" + "addiu %[dst_u], %[dst_u], 1 \n" + "addiu %[dst_v], %[dst_v], 1 \n" + "sb %[tmp_t7], -1(%[dst_u]) \n" + "sb %[tmp_t8], -1(%[dst_v]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_rgb0] "+r"(src_rgb0), [src_rgb1] "+r"(src_rgb1), + [dst_u] "+r"(dst_u), [dst_v] "+r"(dst_v) + : [const1] "r"(const1), [const2] "r"(const2), [const3] "r"(const3), + [const4] "r"(const4), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi"); + } +} + +void RGBAToYRow_DSPR2(const uint8* src_argb0, uint8* dst_y, int width) { + int x; + int const1 = 0x00420081; + int const2 = 0x00190000; + int const5 = 0x40; + for (x = 0; x < width; x += 4) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_argb0]) \n" + "lw %[tmp_t2], 4(%[src_argb0]) \n" + "lw %[tmp_t3], 8(%[src_argb0]) \n" + "lw %[tmp_t4], 12(%[src_argb0]) \n" + "preceu.ph.qbl %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbr %[tmp_t4], %[tmp_t4] \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "mult $ac2, %[const5], %[const5] \n" + "mult $ac3, %[const5], %[const5] \n" + "dpa.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpa.w.ph $ac1, %[tmp_t6], %[const1] \n" + "dpa.w.ph $ac2, %[tmp_t7], %[const1] \n" + "dpa.w.ph $ac3, %[tmp_t8], %[const1] \n" + "dpa.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpa.w.ph $ac1, %[tmp_t2], %[const2] \n" + "dpa.w.ph $ac2, %[tmp_t3], %[const2] \n" + "dpa.w.ph $ac3, %[tmp_t4], %[const2] \n" + "extr_r.w %[tmp_t1], $ac0, 8 \n" + "extr_r.w %[tmp_t2], $ac1, 8 \n" + "extr_r.w %[tmp_t3], $ac2, 8 \n" + "extr_r.w %[tmp_t4], $ac3, 8 \n" + "addiu %[dst_y], %[dst_y], 4 \n" + "addiu %[src_argb0],%[src_argb0], 16 \n" + "sb %[tmp_t1], -4(%[dst_y]) \n" + "sb %[tmp_t2], -3(%[dst_y]) \n" + "sb %[tmp_t3], -2(%[dst_y]) \n" + "sb %[tmp_t4], -1(%[dst_y]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_argb0] "+r"(src_argb0), [dst_y] "+r"(dst_y) + : [const1] "r"(const1), [const2] "r"(const2), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi", "$ac2lo", "$ac2hi", "$ac3lo", + "$ac3hi"); + } +} + +void ARGBToUVRow_DSPR2(const uint8* src_rgb0, + int src_stride_rgb, + uint8* dst_u, + uint8* dst_v, + int width) { + const uint8* src_rgb1 = src_rgb0 + src_stride_rgb; + int x; + int const1 = 0xffb60070; + int const2 = 0x0000ffda; + int const3 = 0xffa2ffee; + int const4 = 0x00000070; + int const5 = 0x100; + + for (x = 0; x < width - 1; x += 2) { + int tmp_t1, tmp_t2, tmp_t3, tmp_t4, tmp_t5; + int tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t1], 0(%[src_rgb0]) \n" + "lw %[tmp_t2], 4(%[src_rgb0]) \n" + "lw %[tmp_t3], 0(%[src_rgb1]) \n" + "lw %[tmp_t4], 4(%[src_rgb1]) \n" + "preceu.ph.qbr %[tmp_t5], %[tmp_t1] \n" + "preceu.ph.qbl %[tmp_t1], %[tmp_t1] \n" + "preceu.ph.qbr %[tmp_t6], %[tmp_t2] \n" + "preceu.ph.qbl %[tmp_t2], %[tmp_t2] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t3] \n" + "preceu.ph.qbl %[tmp_t3], %[tmp_t3] \n" + "preceu.ph.qbr %[tmp_t8], %[tmp_t4] \n" + "preceu.ph.qbl %[tmp_t4], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t6] \n" + "addu.ph %[tmp_t7], %[tmp_t7], %[tmp_t8] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t2] \n" + "addu.ph %[tmp_t3], %[tmp_t3], %[tmp_t4] \n" + "addu.ph %[tmp_t5], %[tmp_t5], %[tmp_t7] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t3] \n" + "shrl.ph %[tmp_t5], %[tmp_t5], 2 \n" + "shrl.ph %[tmp_t1], %[tmp_t1], 2 \n" + "mult $ac0, %[const5], %[const5] \n" + "mult $ac1, %[const5], %[const5] \n" + "dpaq_s.w.ph $ac0, %[tmp_t5], %[const1] \n" + "dpaq_s.w.ph $ac1, %[tmp_t5], %[const3] \n" + "dpaq_s.w.ph $ac0, %[tmp_t1], %[const2] \n" + "dpaq_s.w.ph $ac1, %[tmp_t1], %[const4] \n" + "extr_r.w %[tmp_t7], $ac0, 9 \n" + "extr_r.w %[tmp_t8], $ac1, 9 \n" + "addiu %[src_rgb0], %[src_rgb0], 8 \n" + "addiu %[src_rgb1], %[src_rgb1], 8 \n" + "addiu %[dst_u], %[dst_u], 1 \n" + "addiu %[dst_v], %[dst_v], 1 \n" + "sb %[tmp_t7], -1(%[dst_u]) \n" + "sb %[tmp_t8], -1(%[dst_v]) \n" + ".set pop \n" + : [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), + [tmp_t3] "=&r"(tmp_t3), [tmp_t4] "=&r"(tmp_t4), + [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), + [src_rgb0] "+r"(src_rgb0), [src_rgb1] "+r"(src_rgb1), + [dst_u] "+r"(dst_u), [dst_v] "+r"(dst_v) + : [const1] "r"(const1), [const2] "r"(const2), [const3] "r"(const3), + [const4] "r"(const4), [const5] "r"(const5) + : "hi", "lo", "$ac1lo", "$ac1hi"); + } +} + #endif // __mips_dsp_rev >= 2 #endif // defined(__mips__) diff --git a/source/row_win.cc b/source/row_win.cc index a1df4319b..028f7bf92 100644 --- a/source/row_win.cc +++ b/source/row_win.cc @@ -2032,7 +2032,7 @@ __declspec(naked) void RGBAToUVRow_SSSE3(const uint8* src_argb0, __asm vpsraw ymm2, ymm2, 6 \ __asm vpackuswb ymm0, ymm0, ymm0 /* B */ \ __asm vpackuswb ymm1, ymm1, ymm1 /* G */ \ - __asm vpackuswb ymm2, ymm2, ymm2 /* R */ \ + __asm vpackuswb ymm2, ymm2, ymm2 /* R */ \ } // Store 16 ARGB values. diff --git a/source/scale.cc b/source/scale.cc index e66f90a0d..65df1f09e 100644 --- a/source/scale.cc +++ b/source/scale.cc @@ -894,6 +894,14 @@ static void ScalePlaneBox(int src_width, } } #endif +#if defined(HAS_SCALEADDROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + ScaleAddRow = ScaleAddRow_Any_DSPR2; + if (IS_ALIGNED(src_width, 16)) { + ScaleAddRow = ScaleAddRow_DSPR2; + } + } +#endif for (j = 0; j < dst_height; ++j) { int boxheight; diff --git a/source/scale_any.cc b/source/scale_any.cc index 9068e92d8..d64ba7a9d 100644 --- a/source/scale_any.cc +++ b/source/scale_any.cc @@ -421,6 +421,9 @@ SAANY(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, ScaleAddRow_C, 15) #ifdef HAS_SCALEADDROW_MSA SAANY(ScaleAddRow_Any_MSA, ScaleAddRow_MSA, ScaleAddRow_C, 15) #endif +#ifdef HAS_SCALEADDROW_DSPR2 +SAANY(ScaleAddRow_Any_DSPR2, ScaleAddRow_DSPR2, ScaleAddRow_C, 15) +#endif #undef SAANY #ifdef __cplusplus diff --git a/source/scale_mips.cc b/source/scale_mips.cc index cb7f46d26..d7dd20344 100644 --- a/source/scale_mips.cc +++ b/source/scale_mips.cc @@ -42,10 +42,10 @@ void ScaleRowDown2_DSPR2(const uint8* src_ptr, "lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24| "lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28| // TODO(fbarchard): Use odd pixels instead of even. - "precr.qb.ph $t8, $t1, $t0 \n" // |6|4|2|0| - "precr.qb.ph $t0, $t3, $t2 \n" // |14|12|10|8| - "precr.qb.ph $t1, $t5, $t4 \n" // |22|20|18|16| - "precr.qb.ph $t2, $t7, $t6 \n" // |30|28|26|24| + "precrq.qb.ph $t8, $t1, $t0 \n" // |7|5|3|1| + "precrq.qb.ph $t0, $t3, $t2 \n" // |15|13|11|9| + "precrq.qb.ph $t1, $t5, $t4 \n" // |23|21|19|17| + "precrq.qb.ph $t2, $t7, $t6 \n" // |31|29|27|25| "addiu %[src_ptr], %[src_ptr], 32 \n" "addiu $t9, $t9, -1 \n" "sw $t8, 0(%[dst]) \n" @@ -61,7 +61,7 @@ void ScaleRowDown2_DSPR2(const uint8* src_ptr, " nop \n" "21: \n" - "lbu $t0, 0(%[src_ptr]) \n" + "lbu $t0, 1(%[src_ptr]) \n" "addiu %[src_ptr], %[src_ptr], 2 \n" "addiu $t9, $t9, -1 \n" "sb $t0, 0(%[dst]) \n" @@ -198,8 +198,8 @@ void ScaleRowDown4_DSPR2(const uint8* src_ptr, "precr.qb.ph $t2, $t4, $t3 \n" // |14|12|10|8| "precr.qb.ph $t5, $t6, $t5 \n" // |22|20|18|16| "precr.qb.ph $t6, $t8, $t7 \n" // |30|28|26|24| - "precr.qb.ph $t1, $t2, $t1 \n" // |12|8|4|0| - "precr.qb.ph $t5, $t6, $t5 \n" // |28|24|20|16| + "precrq.qb.ph $t1, $t2, $t1 \n" // |14|10|6|2| + "precrq.qb.ph $t5, $t6, $t5 \n" // |30|26|22|18| "addiu %[src_ptr], %[src_ptr], 32 \n" "addiu $t9, $t9, -1 \n" "sw $t1, 0(%[dst]) \n" @@ -213,7 +213,7 @@ void ScaleRowDown4_DSPR2(const uint8* src_ptr, " nop \n" "21: \n" - "lbu $t1, 0(%[src_ptr]) \n" + "lbu $t1, 2(%[src_ptr]) \n" "addiu %[src_ptr], %[src_ptr], 4 \n" "addiu $t9, $t9, -1 \n" "sb $t1, 0(%[dst]) \n" @@ -615,6 +615,51 @@ void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr, : "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8"); } +void ScaleAddRow_DSPR2(const uint8* src_ptr, uint16* dst_ptr, int src_width) { + int x; + for (x = 0; x < ((src_width - 1)); x += 8) { + uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4; + uint32 tmp_t5, tmp_t6, tmp_t7, tmp_t8; + __asm__ __volatile__( + ".set push \n" + ".set noreorder \n" + "lw %[tmp_t5], 0(%[src_ptr]) \n" + "lw %[tmp_t6], 4(%[src_ptr]) \n" + "lw %[tmp_t1], 0(%[dst_ptr]) \n" + "lw %[tmp_t2], 4(%[dst_ptr]) \n" + "lw %[tmp_t3], 8(%[dst_ptr]) \n" + "lw %[tmp_t4], 12(%[dst_ptr]) \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t5] \n" + "preceu.ph.qbl %[tmp_t8], %[tmp_t5] \n" + "addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t7] \n" + "addu.ph %[tmp_t2], %[tmp_t2], %[tmp_t8] \n" + "preceu.ph.qbr %[tmp_t7], %[tmp_t6] \n" + "preceu.ph.qbl %[tmp_t8], %[tmp_t6] \n" + "addu.ph %[tmp_t3], %[tmp_t3], %[tmp_t7] \n" + "addu.ph %[tmp_t4], %[tmp_t4], %[tmp_t8] \n" + "sw %[tmp_t1], 0(%[dst_ptr]) \n" + "sw %[tmp_t2], 4(%[dst_ptr]) \n" + "sw %[tmp_t3], 8(%[dst_ptr]) \n" + "sw %[tmp_t4], 12(%[dst_ptr]) \n" + ".set pop \n" + : + [tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), [tmp_t3] "=&r"(tmp_t3), + [tmp_t4] "=&r"(tmp_t4), [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6), + [tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [src_ptr] "+r"(src_ptr) + : [dst_ptr] "r"(dst_ptr)); + src_ptr += 8; + dst_ptr += 8; + } + + if ((src_width)&7) { + for (x = 0; x < ((src_width - 1) & 7); x += 1) { + dst_ptr[0] += src_ptr[0]; + src_ptr += 1; + dst_ptr += 1; + } + } +} + #endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2) #ifdef __cplusplus diff --git a/unit_test/convert_test.cc b/unit_test/convert_test.cc index 50d2569b2..cd0b93984 100644 --- a/unit_test/convert_test.cc +++ b/unit_test/convert_test.cc @@ -30,107 +30,113 @@ namespace libyuv { #define SUBSAMPLE(v, a) ((((v) + (a)-1)) / (a)) -#define TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \ - FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF) \ - TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) { \ - const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ - const int kHeight = benchmark_height_; \ - align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ - align_buffer_page_end(src_u, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ - OFF); \ - align_buffer_page_end(src_v, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ - OFF); \ - align_buffer_page_end(dst_y_c, kWidth* kHeight); \ - align_buffer_page_end(dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_v_c, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_y_opt, kWidth* kHeight); \ - align_buffer_page_end(dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - for (int i = 0; i < kHeight; ++i) \ - for (int j = 0; j < kWidth; ++j) \ - src_y[i * kWidth + j + OFF] = (fastrand() & 0xff); \ - for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) { \ - for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) { \ - src_u[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ - (fastrand() & 0xff); \ - src_v[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ - (fastrand() & 0xff); \ - } \ - } \ - memset(dst_y_c, 1, kWidth* kHeight); \ - memset(dst_u_c, 2, \ - SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - memset(dst_v_c, 3, \ - SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - memset(dst_y_opt, 101, kWidth* kHeight); \ - memset(dst_u_opt, 102, \ - SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - memset(dst_v_opt, 103, \ - SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - MaskCpuFlags(disable_cpu_flags_); \ - SRC_FMT_PLANAR##To##FMT_PLANAR( \ - src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ - src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_c, kWidth, \ - dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_c, \ - SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight); \ - MaskCpuFlags(benchmark_cpu_info_); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - SRC_FMT_PLANAR##To##FMT_PLANAR( \ - src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ - src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_opt, kWidth, \ - dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_opt, \ - SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - for (int i = 0; i < kHeight; ++i) { \ - for (int j = 0; j < kWidth; ++j) { \ - int abs_diff = abs(static_cast(dst_y_c[i * kWidth + j]) - \ - static_cast(dst_y_opt[i * kWidth + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_EQ(0, max_diff); \ - for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \ - for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) { \ - int abs_diff = abs( \ - static_cast(dst_u_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) - \ - static_cast( \ - dst_u_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 3); \ - for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \ - for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) { \ - int abs_diff = abs( \ - static_cast(dst_v_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) - \ - static_cast( \ - dst_v_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 3); \ - free_aligned_buffer_page_end(dst_y_c); \ - free_aligned_buffer_page_end(dst_u_c); \ - free_aligned_buffer_page_end(dst_v_c); \ - free_aligned_buffer_page_end(dst_y_opt); \ - free_aligned_buffer_page_end(dst_u_opt); \ - free_aligned_buffer_page_end(dst_v_opt); \ - free_aligned_buffer_page_end(src_y); \ - free_aligned_buffer_page_end(src_u); \ - free_aligned_buffer_page_end(src_v); \ +#define TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \ + FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF) \ + TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) { \ + const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ + const int kHeight = benchmark_height_; \ + align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ + align_buffer_page_end( \ + src_u, \ + SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ + OFF); \ + align_buffer_page_end( \ + src_v, \ + SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ + OFF); \ + align_buffer_page_end(dst_y_c, kWidth* kHeight); \ + align_buffer_page_end( \ + dst_u_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_v_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end(dst_y_opt, kWidth* kHeight); \ + align_buffer_page_end( \ + dst_u_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_v_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + for (int i = 0; i < kHeight; ++i) \ + for (int j = 0; j < kWidth; ++j) \ + src_y[i * kWidth + j + OFF] = (fastrand() & 0xff); \ + for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) { \ + for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) { \ + src_u[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ + (fastrand() & 0xff); \ + src_v[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ + (fastrand() & 0xff); \ + } \ + } \ + memset(dst_y_c, 1, kWidth* kHeight); \ + memset(dst_u_c, 2, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + memset(dst_v_c, 3, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + memset(dst_y_opt, 101, kWidth* kHeight); \ + memset(dst_u_opt, 102, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + memset(dst_v_opt, 103, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + MaskCpuFlags(disable_cpu_flags_); \ + SRC_FMT_PLANAR##To##FMT_PLANAR( \ + src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ + src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_c, kWidth, \ + dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight); \ + MaskCpuFlags(benchmark_cpu_info_); \ + for (int i = 0; i < benchmark_iterations_; ++i) { \ + SRC_FMT_PLANAR##To##FMT_PLANAR( \ + src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ + src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_opt, kWidth, \ + dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight); \ + } \ + int max_diff = 0; \ + for (int i = 0; i < kHeight; ++i) { \ + for (int j = 0; j < kWidth; ++j) { \ + int abs_diff = abs(static_cast(dst_y_c[i * kWidth + j]) - \ + static_cast(dst_y_opt[i * kWidth + j])); \ + if (abs_diff > max_diff) { \ + max_diff = abs_diff; \ + } \ + } \ + } \ + EXPECT_EQ(0, max_diff); \ + for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \ + for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) { \ + int abs_diff = abs( \ + static_cast(dst_u_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) - \ + static_cast( \ + dst_u_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j])); \ + if (abs_diff > max_diff) { \ + max_diff = abs_diff; \ + } \ + } \ + } \ + EXPECT_LE(max_diff, 3); \ + for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \ + for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) { \ + int abs_diff = abs( \ + static_cast(dst_v_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) - \ + static_cast( \ + dst_v_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j])); \ + if (abs_diff > max_diff) { \ + max_diff = abs_diff; \ + } \ + } \ + } \ + EXPECT_LE(max_diff, 3); \ + free_aligned_buffer_page_end(dst_y_c); \ + free_aligned_buffer_page_end(dst_u_c); \ + free_aligned_buffer_page_end(dst_v_c); \ + free_aligned_buffer_page_end(dst_y_opt); \ + free_aligned_buffer_page_end(dst_u_opt); \ + free_aligned_buffer_page_end(dst_v_opt); \ + free_aligned_buffer_page_end(src_y); \ + free_aligned_buffer_page_end(src_u); \ + free_aligned_buffer_page_end(src_v); \ } #define TESTPLANARTOP(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \ @@ -166,15 +172,19 @@ TESTPLANARTOP(I444, 1, 1, I444, 1, 1) align_buffer_page_end(src_uv, \ kSizeUV*((PIXEL_STRIDE == 3) ? 3 : 2) + OFF); \ align_buffer_page_end(dst_y_c, kWidth* kHeight); \ - align_buffer_page_end(dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_v_c, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_u_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_v_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ align_buffer_page_end(dst_y_opt, kWidth* kHeight); \ - align_buffer_page_end(dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_u_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_v_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ uint8* src_u = src_uv + OFF_U; \ uint8* src_v = src_uv + (PIXEL_STRIDE == 1 ? kSizeUV : OFF_V); \ int src_stride_uv = SUBSAMPLE(kWidth, SUBSAMP_X) * PIXEL_STRIDE; \ @@ -278,84 +288,88 @@ TESTAPLANARTOP(Android420, I420, 1, 0, 0, 2, 2, I420, 2, 2) TESTAPLANARTOP(Android420, NV12, 2, 0, 1, 2, 2, I420, 2, 2) TESTAPLANARTOP(Android420, NV21, 2, 1, 0, 2, 2, I420, 2, 2) -#define TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \ - FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF) \ - TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) { \ - const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ - const int kHeight = benchmark_height_; \ - align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ - align_buffer_page_end(src_u, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ - OFF); \ - align_buffer_page_end(src_v, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ - OFF); \ - align_buffer_page_end(dst_y_c, kWidth* kHeight); \ - align_buffer_page_end(dst_uv_c, SUBSAMPLE(kWidth * 2, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_y_opt, kWidth* kHeight); \ - align_buffer_page_end(dst_uv_opt, SUBSAMPLE(kWidth * 2, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - for (int i = 0; i < kHeight; ++i) \ - for (int j = 0; j < kWidth; ++j) \ - src_y[i * kWidth + j + OFF] = (fastrand() & 0xff); \ - for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) { \ - for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) { \ - src_u[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ - (fastrand() & 0xff); \ - src_v[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ - (fastrand() & 0xff); \ - } \ - } \ - memset(dst_y_c, 1, kWidth* kHeight); \ - memset(dst_uv_c, 2, \ - SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - memset(dst_y_opt, 101, kWidth* kHeight); \ - memset(dst_uv_opt, 102, \ - SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - MaskCpuFlags(disable_cpu_flags_); \ - SRC_FMT_PLANAR##To##FMT_PLANAR( \ - src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ - src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_c, kWidth, \ - dst_uv_c, SUBSAMPLE(kWidth * 2, SUBSAMP_X), kWidth, NEG kHeight); \ - MaskCpuFlags(benchmark_cpu_info_); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - SRC_FMT_PLANAR##To##FMT_PLANAR( \ - src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ - src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_opt, kWidth, \ - dst_uv_opt, SUBSAMPLE(kWidth * 2, SUBSAMP_X), kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - for (int i = 0; i < kHeight; ++i) { \ - for (int j = 0; j < kWidth; ++j) { \ - int abs_diff = abs(static_cast(dst_y_c[i * kWidth + j]) - \ - static_cast(dst_y_opt[i * kWidth + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 1); \ - for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \ - for (int j = 0; j < SUBSAMPLE(kWidth * 2, SUBSAMP_X); ++j) { \ - int abs_diff = \ - abs(static_cast( \ - dst_uv_c[i * SUBSAMPLE(kWidth * 2, SUBSAMP_X) + j]) - \ - static_cast( \ - dst_uv_opt[i * SUBSAMPLE(kWidth * 2, SUBSAMP_X) + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 1); \ - free_aligned_buffer_page_end(dst_y_c); \ - free_aligned_buffer_page_end(dst_uv_c); \ - free_aligned_buffer_page_end(dst_y_opt); \ - free_aligned_buffer_page_end(dst_uv_opt); \ - free_aligned_buffer_page_end(src_y); \ - free_aligned_buffer_page_end(src_u); \ - free_aligned_buffer_page_end(src_v); \ +#define TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \ + FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF) \ + TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) { \ + const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ + const int kHeight = benchmark_height_; \ + align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ + align_buffer_page_end( \ + src_u, \ + SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ + OFF); \ + align_buffer_page_end( \ + src_v, \ + SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ + OFF); \ + align_buffer_page_end(dst_y_c, kWidth* kHeight); \ + align_buffer_page_end( \ + dst_uv_c, \ + SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end(dst_y_opt, kWidth* kHeight); \ + align_buffer_page_end( \ + dst_uv_opt, \ + SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + for (int i = 0; i < kHeight; ++i) \ + for (int j = 0; j < kWidth; ++j) \ + src_y[i * kWidth + j + OFF] = (fastrand() & 0xff); \ + for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) { \ + for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) { \ + src_u[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ + (fastrand() & 0xff); \ + src_v[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] = \ + (fastrand() & 0xff); \ + } \ + } \ + memset(dst_y_c, 1, kWidth* kHeight); \ + memset(dst_uv_c, 2, \ + SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + memset(dst_y_opt, 101, kWidth* kHeight); \ + memset(dst_uv_opt, 102, \ + SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + MaskCpuFlags(disable_cpu_flags_); \ + SRC_FMT_PLANAR##To##FMT_PLANAR( \ + src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ + src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_c, kWidth, \ + dst_uv_c, SUBSAMPLE(kWidth * 2, SUBSAMP_X), kWidth, NEG kHeight); \ + MaskCpuFlags(benchmark_cpu_info_); \ + for (int i = 0; i < benchmark_iterations_; ++i) { \ + SRC_FMT_PLANAR##To##FMT_PLANAR( \ + src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \ + src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_opt, kWidth, \ + dst_uv_opt, SUBSAMPLE(kWidth * 2, SUBSAMP_X), kWidth, NEG kHeight); \ + } \ + int max_diff = 0; \ + for (int i = 0; i < kHeight; ++i) { \ + for (int j = 0; j < kWidth; ++j) { \ + int abs_diff = abs(static_cast(dst_y_c[i * kWidth + j]) - \ + static_cast(dst_y_opt[i * kWidth + j])); \ + if (abs_diff > max_diff) { \ + max_diff = abs_diff; \ + } \ + } \ + } \ + EXPECT_LE(max_diff, 1); \ + for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \ + for (int j = 0; j < SUBSAMPLE(kWidth * 2, SUBSAMP_X); ++j) { \ + int abs_diff = \ + abs(static_cast( \ + dst_uv_c[i * SUBSAMPLE(kWidth * 2, SUBSAMP_X) + j]) - \ + static_cast( \ + dst_uv_opt[i * SUBSAMPLE(kWidth * 2, SUBSAMP_X) + j])); \ + if (abs_diff > max_diff) { \ + max_diff = abs_diff; \ + } \ + } \ + } \ + EXPECT_LE(max_diff, 1); \ + free_aligned_buffer_page_end(dst_y_c); \ + free_aligned_buffer_page_end(dst_uv_c); \ + free_aligned_buffer_page_end(dst_y_opt); \ + free_aligned_buffer_page_end(dst_uv_opt); \ + free_aligned_buffer_page_end(src_y); \ + free_aligned_buffer_page_end(src_u); \ + free_aligned_buffer_page_end(src_v); \ } #define TESTPLANARTOBP(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \ @@ -379,19 +393,24 @@ TESTPLANARTOBP(I420, 2, 2, NV21, 2, 2) const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ const int kHeight = benchmark_height_; \ align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ - align_buffer_page_end(src_uv, 2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ - OFF); \ + align_buffer_page_end(src_uv, \ + 2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * \ + SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \ + OFF); \ align_buffer_page_end(dst_y_c, kWidth* kHeight); \ - align_buffer_page_end(dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_v_c, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_u_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_v_c, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ align_buffer_page_end(dst_y_opt, kWidth* kHeight); \ - align_buffer_page_end(dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ - align_buffer_page_end(dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * \ - SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_u_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ + align_buffer_page_end( \ + dst_v_opt, \ + SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \ for (int i = 0; i < kHeight; ++i) \ for (int j = 0; j < kWidth; ++j) \ src_y[i * kWidth + j + OFF] = (fastrand() & 0xff); \ @@ -1369,10 +1388,12 @@ TEST_F(LibYUVConvertTest, MJPGToI420) { const int kSize = kImageSize + kOff; align_buffer_page_end(orig_pixels, kSize); align_buffer_page_end(dst_y_opt, benchmark_width_ * benchmark_height_); - align_buffer_page_end(dst_u_opt, SUBSAMPLE(benchmark_width_, 2) * - SUBSAMPLE(benchmark_height_, 2)); - align_buffer_page_end(dst_v_opt, SUBSAMPLE(benchmark_width_, 2) * - SUBSAMPLE(benchmark_height_, 2)); + align_buffer_page_end( + dst_u_opt, + SUBSAMPLE(benchmark_width_, 2) * SUBSAMPLE(benchmark_height_, 2)); + align_buffer_page_end( + dst_v_opt, + SUBSAMPLE(benchmark_width_, 2) * SUBSAMPLE(benchmark_height_, 2)); // EOI, SOI to make MJPG appear valid. memset(orig_pixels, 0, kSize); @@ -1444,16 +1465,20 @@ TEST_F(LibYUVConvertTest, NV12Crop) { uint8* src_uv = src_y + kWidth * kHeight; align_buffer_page_end(dst_y, kDestWidth * kDestHeight); - align_buffer_page_end(dst_u, SUBSAMPLE(kDestWidth, SUBSAMP_X) * - SUBSAMPLE(kDestHeight, SUBSAMP_Y)); - align_buffer_page_end(dst_v, SUBSAMPLE(kDestWidth, SUBSAMP_X) * - SUBSAMPLE(kDestHeight, SUBSAMP_Y)); + align_buffer_page_end( + dst_u, + SUBSAMPLE(kDestWidth, SUBSAMP_X) * SUBSAMPLE(kDestHeight, SUBSAMP_Y)); + align_buffer_page_end( + dst_v, + SUBSAMPLE(kDestWidth, SUBSAMP_X) * SUBSAMPLE(kDestHeight, SUBSAMP_Y)); align_buffer_page_end(dst_y_2, kDestWidth * kDestHeight); - align_buffer_page_end(dst_u_2, SUBSAMPLE(kDestWidth, SUBSAMP_X) * - SUBSAMPLE(kDestHeight, SUBSAMP_Y)); - align_buffer_page_end(dst_v_2, SUBSAMPLE(kDestWidth, SUBSAMP_X) * - SUBSAMPLE(kDestHeight, SUBSAMP_Y)); + align_buffer_page_end( + dst_u_2, + SUBSAMPLE(kDestWidth, SUBSAMP_X) * SUBSAMPLE(kDestHeight, SUBSAMP_Y)); + align_buffer_page_end( + dst_v_2, + SUBSAMPLE(kDestWidth, SUBSAMP_X) * SUBSAMPLE(kDestHeight, SUBSAMP_Y)); for (int i = 0; i < kHeight * kWidth; ++i) { src_y[i] = (fastrand() & 0xff); diff --git a/util/psnr_main.cc b/util/psnr_main.cc index 01e877746..4d930be4a 100644 --- a/util/psnr_main.cc +++ b/util/psnr_main.cc @@ -356,17 +356,19 @@ int main(int argc, const char* argv[]) { const int uv_size = ((image_width + 1) / 2) * ((image_height + 1) / 2); const size_t total_size = y_size + 2 * uv_size; // NOLINT #if defined(_MSC_VER) - _fseeki64(file_org, static_cast<__int64>(num_skip_org) * - static_cast<__int64>(total_size), - SEEK_SET); + _fseeki64( + file_org, + static_cast<__int64>(num_skip_org) * static_cast<__int64>(total_size), + SEEK_SET); #else fseek(file_org, num_skip_org * total_size, SEEK_SET); #endif for (int cur_rec = 0; cur_rec < num_rec; ++cur_rec) { #if defined(_MSC_VER) - _fseeki64(file_rec[cur_rec], static_cast<__int64>(num_skip_rec) * - static_cast<__int64>(total_size), - SEEK_SET); + _fseeki64( + file_rec[cur_rec], + static_cast<__int64>(num_skip_rec) * static_cast<__int64>(total_size), + SEEK_SET); #else fseek(file_rec[cur_rec], num_skip_rec * total_size, SEEK_SET); #endif