github.com/cellofellow/gopkg@v0.0.0-20140722061823-eec0544a62ad/image/webp/libwebp/src/dsp/yuv.c (about) 1 // Copyright 2010 Google Inc. All Rights Reserved. 2 // 3 // Use of this source code is governed by a BSD-style license 4 // that can be found in the COPYING file in the root of the source 5 // tree. An additional intellectual property rights grant can be found 6 // in the file PATENTS. All contributing project authors may 7 // be found in the AUTHORS file in the root of the source tree. 8 // ----------------------------------------------------------------------------- 9 // 10 // YUV->RGB conversion function 11 // 12 // Author: Skal (pascal.massimino@gmail.com) 13 14 #include "./yuv.h" 15 16 17 #if defined(WEBP_YUV_USE_TABLE) 18 19 static int done = 0; 20 21 static WEBP_INLINE uint8_t clip(int v, int max_value) { 22 return v < 0 ? 0 : v > max_value ? max_value : v; 23 } 24 25 int16_t VP8kVToR[256], VP8kUToB[256]; 26 int32_t VP8kVToG[256], VP8kUToG[256]; 27 uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN]; 28 uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN]; 29 30 void VP8YUVInit(void) { 31 int i; 32 if (done) { 33 return; 34 } 35 #ifndef USE_YUVj 36 for (i = 0; i < 256; ++i) { 37 VP8kVToR[i] = (89858 * (i - 128) + YUV_HALF) >> YUV_FIX; 38 VP8kUToG[i] = -22014 * (i - 128) + YUV_HALF; 39 VP8kVToG[i] = -45773 * (i - 128); 40 VP8kUToB[i] = (113618 * (i - 128) + YUV_HALF) >> YUV_FIX; 41 } 42 for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) { 43 const int k = ((i - 16) * 76283 + YUV_HALF) >> YUV_FIX; 44 VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255); 45 VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15); 46 } 47 #else 48 for (i = 0; i < 256; ++i) { 49 VP8kVToR[i] = (91881 * (i - 128) + YUV_HALF) >> YUV_FIX; 50 VP8kUToG[i] = -22554 * (i - 128) + YUV_HALF; 51 VP8kVToG[i] = -46802 * (i - 128); 52 VP8kUToB[i] = (116130 * (i - 128) + YUV_HALF) >> YUV_FIX; 53 } 54 for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) { 55 const int k = i; 56 VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255); 57 VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15); 58 } 59 #endif 60 61 done = 1; 62 } 63 64 #else 65 66 void VP8YUVInit(void) {} 67 68 #endif // WEBP_YUV_USE_TABLE 69 70 //----------------------------------------------------------------------------- 71 // SSE2 extras 72 73 #if defined(WEBP_USE_SSE2) 74 75 #ifdef FANCY_UPSAMPLING 76 77 #include <emmintrin.h> 78 #include <string.h> // for memcpy 79 80 typedef union { // handy struct for converting SSE2 registers 81 int32_t i32[4]; 82 uint8_t u8[16]; 83 __m128i m; 84 } VP8kCstSSE2; 85 86 static int done_sse2 = 0; 87 static VP8kCstSSE2 VP8kUtoRGBA[256], VP8kVtoRGBA[256], VP8kYtoRGBA[256]; 88 89 void VP8YUVInitSSE2(void) { 90 if (!done_sse2) { 91 int i; 92 for (i = 0; i < 256; ++i) { 93 VP8kYtoRGBA[i].i32[0] = 94 VP8kYtoRGBA[i].i32[1] = 95 VP8kYtoRGBA[i].i32[2] = (i - 16) * kYScale + YUV_HALF2; 96 VP8kYtoRGBA[i].i32[3] = 0xff << YUV_FIX2; 97 98 VP8kUtoRGBA[i].i32[0] = 0; 99 VP8kUtoRGBA[i].i32[1] = -kUToG * (i - 128); 100 VP8kUtoRGBA[i].i32[2] = kUToB * (i - 128); 101 VP8kUtoRGBA[i].i32[3] = 0; 102 103 VP8kVtoRGBA[i].i32[0] = kVToR * (i - 128); 104 VP8kVtoRGBA[i].i32[1] = -kVToG * (i - 128); 105 VP8kVtoRGBA[i].i32[2] = 0; 106 VP8kVtoRGBA[i].i32[3] = 0; 107 } 108 done_sse2 = 1; 109 } 110 } 111 112 static WEBP_INLINE __m128i VP8GetRGBA32b(int y, int u, int v) { 113 const __m128i u_part = _mm_loadu_si128(&VP8kUtoRGBA[u].m); 114 const __m128i v_part = _mm_loadu_si128(&VP8kVtoRGBA[v].m); 115 const __m128i y_part = _mm_loadu_si128(&VP8kYtoRGBA[y].m); 116 const __m128i uv_part = _mm_add_epi32(u_part, v_part); 117 const __m128i rgba1 = _mm_add_epi32(y_part, uv_part); 118 const __m128i rgba2 = _mm_srai_epi32(rgba1, YUV_FIX2); 119 return rgba2; 120 } 121 122 static WEBP_INLINE void VP8YuvToRgbSSE2(uint8_t y, uint8_t u, uint8_t v, 123 uint8_t* const rgb) { 124 const __m128i tmp0 = VP8GetRGBA32b(y, u, v); 125 const __m128i tmp1 = _mm_packs_epi32(tmp0, tmp0); 126 const __m128i tmp2 = _mm_packus_epi16(tmp1, tmp1); 127 // Note: we store 8 bytes at a time, not 3 bytes! -> memory stomp 128 _mm_storel_epi64((__m128i*)rgb, tmp2); 129 } 130 131 static WEBP_INLINE void VP8YuvToBgrSSE2(uint8_t y, uint8_t u, uint8_t v, 132 uint8_t* const bgr) { 133 const __m128i tmp0 = VP8GetRGBA32b(y, u, v); 134 const __m128i tmp1 = _mm_shuffle_epi32(tmp0, _MM_SHUFFLE(3, 0, 1, 2)); 135 const __m128i tmp2 = _mm_packs_epi32(tmp1, tmp1); 136 const __m128i tmp3 = _mm_packus_epi16(tmp2, tmp2); 137 // Note: we store 8 bytes at a time, not 3 bytes! -> memory stomp 138 _mm_storel_epi64((__m128i*)bgr, tmp3); 139 } 140 141 void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v, 142 uint8_t* dst) { 143 int n; 144 for (n = 0; n < 32; n += 4) { 145 const __m128i tmp0_1 = VP8GetRGBA32b(y[n + 0], u[n + 0], v[n + 0]); 146 const __m128i tmp0_2 = VP8GetRGBA32b(y[n + 1], u[n + 1], v[n + 1]); 147 const __m128i tmp0_3 = VP8GetRGBA32b(y[n + 2], u[n + 2], v[n + 2]); 148 const __m128i tmp0_4 = VP8GetRGBA32b(y[n + 3], u[n + 3], v[n + 3]); 149 const __m128i tmp1_1 = _mm_packs_epi32(tmp0_1, tmp0_2); 150 const __m128i tmp1_2 = _mm_packs_epi32(tmp0_3, tmp0_4); 151 const __m128i tmp2 = _mm_packus_epi16(tmp1_1, tmp1_2); 152 _mm_storeu_si128((__m128i*)dst, tmp2); 153 dst += 4 * 4; 154 } 155 } 156 157 void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v, 158 uint8_t* dst) { 159 int n; 160 for (n = 0; n < 32; n += 2) { 161 const __m128i tmp0_1 = VP8GetRGBA32b(y[n + 0], u[n + 0], v[n + 0]); 162 const __m128i tmp0_2 = VP8GetRGBA32b(y[n + 1], u[n + 1], v[n + 1]); 163 const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(3, 0, 1, 2)); 164 const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(3, 0, 1, 2)); 165 const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2); 166 const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1); 167 _mm_storel_epi64((__m128i*)dst, tmp3); 168 dst += 4 * 2; 169 } 170 } 171 172 void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v, 173 uint8_t* dst) { 174 int n; 175 uint8_t tmp0[2 * 3 + 5 + 15]; 176 uint8_t* const tmp = (uint8_t*)((uintptr_t)(tmp0 + 15) & ~15); // align 177 for (n = 0; n < 30; ++n) { // we directly stomp the *dst memory 178 VP8YuvToRgbSSE2(y[n], u[n], v[n], dst + n * 3); 179 } 180 // Last two pixels are special: we write in a tmp buffer before sending 181 // to dst. 182 VP8YuvToRgbSSE2(y[n + 0], u[n + 0], v[n + 0], tmp + 0); 183 VP8YuvToRgbSSE2(y[n + 1], u[n + 1], v[n + 1], tmp + 3); 184 memcpy(dst + n * 3, tmp, 2 * 3); 185 } 186 187 void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v, 188 uint8_t* dst) { 189 int n; 190 uint8_t tmp0[2 * 3 + 5 + 15]; 191 uint8_t* const tmp = (uint8_t*)((uintptr_t)(tmp0 + 15) & ~15); // align 192 for (n = 0; n < 30; ++n) { 193 VP8YuvToBgrSSE2(y[n], u[n], v[n], dst + n * 3); 194 } 195 VP8YuvToBgrSSE2(y[n + 0], u[n + 0], v[n + 0], tmp + 0); 196 VP8YuvToBgrSSE2(y[n + 1], u[n + 1], v[n + 1], tmp + 3); 197 memcpy(dst + n * 3, tmp, 2 * 3); 198 } 199 200 #else 201 202 void VP8YUVInitSSE2(void) {} 203 204 #endif // FANCY_UPSAMPLING 205 206 #endif // WEBP_USE_SSE2 207