From dd2d512e5afad7536e3a010c0193ca1b43c14985 Mon Sep 17 00:00:00 2001 From: "fbarchard@google.com" Date: Wed, 7 Nov 2012 00:53:52 +0000 Subject: [PATCH] 420 subsampler BUG=none TEST=none Review URL: https://webrtc-codereview.appspot.com/935012 git-svn-id: http://libyuv.googlecode.com/svn/trunk@478 16f28f9a-4ce2-e073-06de-1de4eb20be90 --- README.chromium | 2 +- include/libyuv/row.h | 3 ++ include/libyuv/version.h | 2 +- source/convert.cc | 18 ++++++++++++ source/convert_from_argb.cc | 12 ++++++++ source/format_conversion.cc | 3 ++ source/row_neon.cc | 55 +++++++++++++++++-------------------- unit_test/convert_test.cc | 34 +++++++++++------------ 8 files changed, 80 insertions(+), 49 deletions(-) diff --git a/README.chromium b/README.chromium index 319c2c401..e3d5fe7da 100644 --- a/README.chromium +++ b/README.chromium @@ -1,6 +1,6 @@ Name: libyuv URL: http://code.google.com/p/libyuv/ -Version: 476 +Version: 478 License: BSD License File: LICENSE diff --git a/include/libyuv/row.h b/include/libyuv/row.h index 0a82aad31..f4d4b3c40 100644 --- a/include/libyuv/row.h +++ b/include/libyuv/row.h @@ -196,6 +196,7 @@ extern "C" { #define HAS_ARGBTOUV444ROW_NEON #define HAS_ARGBTOUV422ROW_NEON #define HAS_ARGBTOUV411ROW_NEON +#define HAS_ARGBTOUVROW_NEON #define HAS_BGRATOYROW_NEON #define HAS_ABGRTOYROW_NEON #define HAS_RGBATOYROW_NEON @@ -351,6 +352,8 @@ void ARGBToUV422Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int pix); void ARGBToUV411Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int pix); +void ARGBToUVRow_NEON(const uint8* src_argb, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int pix); void BGRAToYRow_NEON(const uint8* src_bgra, uint8* dst_y, int pix); void ABGRToYRow_NEON(const uint8* src_abgr, uint8* dst_y, int pix); void RGBAToYRow_NEON(const uint8* src_rgba, uint8* dst_y, int pix); diff --git a/include/libyuv/version.h b/include/libyuv/version.h index aca0d25f9..d9aec1491 100644 --- a/include/libyuv/version.h +++ b/include/libyuv/version.h @@ -11,6 +11,6 @@ #ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT #define INCLUDE_LIBYUV_VERSION_H_ -#define LIBYUV_VERSION 476 +#define LIBYUV_VERSION 478 #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT diff --git a/source/convert.cc b/source/convert.cc index e2d4870f3..ce55e05e0 100644 --- a/source/convert.cc +++ b/source/convert.cc @@ -943,6 +943,9 @@ int ARGBToI420(const uint8* src_argb, int src_stride_argb, ARGBToYRow = ARGBToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGBToYRow = ARGBToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #endif @@ -1204,6 +1207,9 @@ int RGB24ToI420(const uint8* src_rgb24, int src_stride_rgb24, RGB24ToYRow = RGB24ToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { RGB24ToYRow = RGB24ToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #else @@ -1306,6 +1312,9 @@ int RAWToI420(const uint8* src_raw, int src_stride_raw, RAWToYRow = RAWToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { RAWToYRow = RAWToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #else @@ -1407,6 +1416,9 @@ int RGB565ToI420(const uint8* src_rgb565, int src_stride_rgb565, RGB565ToYRow = RGB565ToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { RGB565ToYRow = RGB565ToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #else @@ -1508,6 +1520,9 @@ int ARGB1555ToI420(const uint8* src_argb1555, int src_stride_argb1555, ARGB1555ToYRow = ARGB1555ToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGB1555ToYRow = ARGB1555ToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #else @@ -1609,6 +1624,9 @@ int ARGB4444ToI420(const uint8* src_argb4444, int src_stride_argb4444, ARGB4444ToYRow = ARGB4444ToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGB4444ToYRow = ARGB4444ToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #else diff --git a/source/convert_from_argb.cc b/source/convert_from_argb.cc index af9e97eb1..98b00ecc4 100644 --- a/source/convert_from_argb.cc +++ b/source/convert_from_argb.cc @@ -232,6 +232,9 @@ int ARGBToNV12(const uint8* src_argb, int src_stride_argb, ARGBToYRow = ARGBToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGBToYRow = ARGBToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #endif @@ -335,6 +338,9 @@ int ARGBToNV21(const uint8* src_argb, int src_stride_argb, ARGBToYRow = ARGBToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGBToYRow = ARGBToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #endif @@ -435,6 +441,9 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, ARGBToYRow = ARGBToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGBToYRow = ARGBToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #endif @@ -511,6 +520,9 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, ARGBToYRow = ARGBToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGBToYRow = ARGBToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #endif diff --git a/source/format_conversion.cc b/source/format_conversion.cc index e1ae5c4c4..0cb745f39 100644 --- a/source/format_conversion.cc +++ b/source/format_conversion.cc @@ -321,6 +321,9 @@ int BayerToI420(const uint8* src_bayer, int src_stride_bayer, ARGBToYRow = ARGBToYRow_Any_NEON; if (IS_ALIGNED(width, 8)) { ARGBToYRow = ARGBToYRow_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } } } #endif diff --git a/source/row_neon.cc b/source/row_neon.cc index 2cb39ca7f..95869e1b1 100644 --- a/source/row_neon.cc +++ b/source/row_neon.cc @@ -1350,7 +1350,7 @@ void UYVYToUV422Row_NEON(const uint8* src_uyvy, uint8* dst_u, uint8* dst_v, void YUY2ToUVRow_NEON(const uint8* src_yuy2, int stride_yuy2, uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( - "adds %1, %0, %1 \n" // stride + src_yuy2 + "add %1, %0, %1 \n" // stride + src_yuy2 ".p2align 2 \n" "1: \n" "vld4.8 {d0, d1, d2, d3}, [%0]! \n" // load 16 pixels of YUY2. @@ -1376,7 +1376,7 @@ void YUY2ToUVRow_NEON(const uint8* src_yuy2, int stride_yuy2, void UYVYToUVRow_NEON(const uint8* src_uyvy, int stride_uyvy, uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( - "adds %1, %0, %1 \n" // stride + src_uyvy + "add %1, %0, %1 \n" // stride + src_uyvy ".p2align 2 \n" "1: \n" "vld4.8 {d0, d1, d2, d3}, [%0]! \n" // load 16 pixels of UYVY. @@ -1715,11 +1715,12 @@ void ARGBToUV411Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, } #endif // HAS_ARGBTOUV411ROW_NEON -// 32x1 pixels -> 8x1. pix is number of argb pixels. e.g. 32. -#ifdef HAS_ARGBTOUV411ROW_NEON_ALT -void ARGBToUV411Row_Alt_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int pix) { +// 16x2 pixels -> 8x1. pix is number of argb pixels. e.g. 16. +#ifdef HAS_ARGBTOUVROW_NEON +void ARGBToUVRow_NEON(const uint8* src_argb, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( + "add %1, %0, %1 \n" // src_stride + src_argb "vmov.s16 q10, #112 / 4 \n" // UB / VR 0.875 coefficient "vmov.s16 q11, #74 / 4 \n" // UG -0.5781 coefficient "vmov.s16 q12, #38 / 4 \n" // UR -0.2969 coefficient @@ -1728,47 +1729,41 @@ void ARGBToUV411Row_Alt_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, "vmov.u16 q15, #0x8080 \n" // 128.5 ".p2align 2 \n" "1: \n" - "vld4.32 {d0, d2, d4, d6}, [%0]! \n" // load 8 ARGB pixels. - "vld4.32 {d1, d3, d5, d7}, [%0]! \n" // load next 8 ARGB pixels. - "vaddl.u8 q0, q0, q1 \n" - "vaddl.u8 q2, q2, q3 \n" - "vadd.u16 q4, q0, q2 \n" // 4 pixels <- 16 - "vld4.32 {d0, d2, d4, d6}, [%0]! \n" // load 8 ARGB pixels. - "vld4.32 {d1, d3, d5, d7}, [%0]! \n" // load next 8 ARGB pixels. - "vaddl.u8 q0, q0, q1 \n" - "vaddl.u8 q2, q2, q3 \n" - "vadd.u16 q3, q0, q2 \n" // 4 more pixels <- 16 - "vtbl.u16 q0, {q3, q4}, q5 \n" // REQUIRES SETUP - "vtbl.u16 q1, {q3, q4}, q6 \n" - "vtbl.u16 q2, {q3, q4}, q7 \n" // 8 pixels - - "subs %3, %3, #16 \n" // 32 processed per loop. + "vld4.8 {d0, d2, d4, d6}, [%0]! \n" // load 8 ARGB pixels. + "vld4.8 {d1, d3, d5, d7}, [%0]! \n" // load next 8 ARGB pixels. + "vpaddl.u8 q0, q0 \n" // B 16 bytes -> 8 shorts. + "vpaddl.u8 q1, q1 \n" // G 16 bytes -> 8 shorts. + "vpaddl.u8 q2, q2 \n" // R 16 bytes -> 8 shorts. + "vld4.8 {d8, d10, d12, d14}, [%1]! \n" // load 8 more ARGB pixels. + "vld4.8 {d9, d11, d13, d15}, [%1]! \n" // load last 8 ARGB pixels. + "vpadal.u8 q0, q4 \n" // B 16 bytes -> 8 shorts. + "vpadal.u8 q1, q5 \n" // G 16 bytes -> 8 shorts. + "vpadal.u8 q2, q6 \n" // R 16 bytes -> 8 shorts. + "subs %4, %4, #16 \n" // 32 processed per loop. "vmul.s16 q8, q0, q10 \n" // B "vmls.s16 q8, q1, q11 \n" // G "vmls.s16 q8, q2, q12 \n" // R "vadd.u16 q8, q8, q15 \n" // +128 -> unsigned - "vmul.s16 q9, q2, q10 \n" // R "vmls.s16 q9, q1, q14 \n" // G "vmls.s16 q9, q0, q13 \n" // B "vadd.u16 q9, q9, q15 \n" // +128 -> unsigned - "vqshrn.u16 d0, q8, #8 \n" // 16 bit to 8 bit U "vqshrn.u16 d1, q9, #8 \n" // 16 bit to 8 bit V - - "vst1.8 {d0}, [%1]! \n" // store 8 pixels U. - "vst1.8 {d1}, [%2]! \n" // store 8 pixels V. + "vst1.8 {d0}, [%2]! \n" // store 8 pixels U. + "vst1.8 {d1}, [%3]! \n" // store 8 pixels V. "bgt 1b \n" : "+r"(src_argb), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(pix) // %3 + "+r"(src_stride_argb), // %1 + "+r"(dst_u), // %2 + "+r"(dst_v), // %3 + "+r"(pix) // %4 : : "memory", "cc", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15" ); } -#endif // HAS_ARGBTOUV411ROW_NEON_ALTERNATIVE +#endif // HAS_ARGBTOUVROW_NEON #ifdef HAS_RGB565TOYROW_NEON void RGB565ToYRow_NEON(const uint8* src_rgb565, uint8* dst_y, int pix) { diff --git a/unit_test/convert_test.cc b/unit_test/convert_test.cc index 82ff8d063..f1a76f90e 100644 --- a/unit_test/convert_test.cc +++ b/unit_test/convert_test.cc @@ -605,15 +605,15 @@ TEST_F(libyuvTest, FMT_A##To##FMT_PLANAR##N) { \ TESTATOPLANARI(FMT_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, \ benchmark_width_, DIFF, _Opt, +, 0) -TESTATOPLANAR(ARGB, 4, I420, 2, 2, 2) -TESTATOPLANAR(BGRA, 4, I420, 2, 2, 2) -TESTATOPLANAR(ABGR, 4, I420, 2, 2, 2) -TESTATOPLANAR(RGBA, 4, I420, 2, 2, 2) -TESTATOPLANAR(RAW, 3, I420, 2, 2, 2) -TESTATOPLANAR(RGB24, 3, I420, 2, 2, 2) -TESTATOPLANAR(RGB565, 2, I420, 2, 2, 2) -TESTATOPLANAR(ARGB1555, 2, I420, 2, 2, 2) -TESTATOPLANAR(ARGB4444, 2, I420, 2, 2, 2) +TESTATOPLANAR(ARGB, 4, I420, 2, 2, 4) +TESTATOPLANAR(BGRA, 4, I420, 2, 2, 4) +TESTATOPLANAR(ABGR, 4, I420, 2, 2, 4) +TESTATOPLANAR(RGBA, 4, I420, 2, 2, 4) +TESTATOPLANAR(RAW, 3, I420, 2, 2, 4) +TESTATOPLANAR(RGB24, 3, I420, 2, 2, 4) +TESTATOPLANAR(RGB565, 2, I420, 2, 2, 4) +TESTATOPLANAR(ARGB1555, 2, I420, 2, 2, 4) +TESTATOPLANAR(ARGB4444, 2, I420, 2, 2, 4) TESTATOPLANAR(ARGB, 4, I411, 4, 1, 4) TESTATOPLANAR(ARGB, 4, I422, 2, 1, 2) TESTATOPLANAR(ARGB, 4, I444, 1, 1, 2) @@ -623,10 +623,10 @@ TESTATOPLANAR(UYVY, 2, I420, 2, 2, 2) TESTATOPLANAR(YUY2, 2, I422, 2, 1, 2) TESTATOPLANAR(UYVY, 2, I422, 2, 1, 2) TESTATOPLANAR(I400, 1, I420, 2, 2, 2) -TESTATOPLANAR(BayerBGGR, 1, I420, 2, 2, 2) -TESTATOPLANAR(BayerRGGB, 1, I420, 2, 2, 2) -TESTATOPLANAR(BayerGBRG, 1, I420, 2, 2, 2) -TESTATOPLANAR(BayerGRBG, 1, I420, 2, 2, 2) +TESTATOPLANAR(BayerBGGR, 1, I420, 2, 2, 4) +TESTATOPLANAR(BayerRGGB, 1, I420, 2, 2, 4) +TESTATOPLANAR(BayerGBRG, 1, I420, 2, 2, 4) +TESTATOPLANAR(BayerGRBG, 1, I420, 2, 2, 4) #define TESTATOBIPLANARI(FMT_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, \ W1280, N, NEG, OFF) \ @@ -666,7 +666,7 @@ TEST_F(libyuvTest, FMT_A##To##FMT_PLANAR##N) { \ } \ } \ } \ - EXPECT_LE(max_diff, 2); \ + EXPECT_LE(max_diff, 4); \ for (int i = 0; i < kHeight / SUBSAMP_Y; ++i) { \ for (int j = 0; j < kWidth / SUBSAMP_X * 2; ++j) { \ int abs_diff = \ @@ -677,7 +677,7 @@ TEST_F(libyuvTest, FMT_A##To##FMT_PLANAR##N) { \ } \ } \ } \ - EXPECT_LE(max_diff, 2); \ + EXPECT_LE(max_diff, 4); \ free_aligned_buffer_16(dst_y_c) \ free_aligned_buffer_16(dst_uv_c) \ free_aligned_buffer_16(dst_y_opt) \ @@ -811,8 +811,8 @@ TESTATOB(ARGB, 4, 4, 1, BayerBGGR, 1, 2, 2, 0) TESTATOB(ARGB, 4, 4, 1, BayerRGGB, 1, 2, 2, 0) TESTATOB(ARGB, 4, 4, 1, BayerGBRG, 1, 2, 2, 0) TESTATOB(ARGB, 4, 4, 1, BayerGRBG, 1, 2, 2, 0) -TESTATOB(ARGB, 4, 4, 1, YUY2, 2, 4, 1, 2) -TESTATOB(ARGB, 4, 4, 1, UYVY, 2, 4, 1, 2) +TESTATOB(ARGB, 4, 4, 1, YUY2, 2, 4, 1, 4) +TESTATOB(ARGB, 4, 4, 1, UYVY, 2, 4, 1, 4) TESTATOB(ARGB, 4, 4, 1, I400, 1, 1, 1, 2) TESTATOB(BGRA, 4, 4, 1, ARGB, 4, 4, 1, 0) TESTATOB(ABGR, 4, 4, 1, ARGB, 4, 4, 1, 0)