Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 1 | // Copyright 2011 Google Inc. All Rights Reserved. |
| 2 | // |
Vikas Arora | 0406ce1 | 2013-08-09 15:57:12 -0700 | [diff] [blame] | 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. |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 8 | // ----------------------------------------------------------------------------- |
| 9 | // |
| 10 | // YUV to RGB upsampling functions. |
| 11 | // |
| 12 | // Author: somnath@google.com (Somnath Banerjee) |
| 13 | |
| 14 | #include "./dsp.h" |
| 15 | #include "./yuv.h" |
| 16 | |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 17 | #include <assert.h> |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 18 | |
| 19 | //------------------------------------------------------------------------------ |
| 20 | // Fancy upsampler |
| 21 | |
| 22 | #ifdef FANCY_UPSAMPLING |
| 23 | |
| 24 | // Fancy upsampling functions to convert YUV to RGB |
| 25 | WebPUpsampleLinePairFunc WebPUpsamplers[MODE_LAST]; |
| 26 | |
| 27 | // Given samples laid out in a square as: |
| 28 | // [a b] |
| 29 | // [c d] |
| 30 | // we interpolate u/v as: |
| 31 | // ([9*a + 3*b + 3*c + d 3*a + 9*b + 3*c + d] + [8 8]) / 16 |
| 32 | // ([3*a + b + 9*c + 3*d a + 3*b + 3*c + 9*d] [8 8]) / 16 |
| 33 | |
| 34 | // We process u and v together stashed into 32bit (16bit each). |
Vikas Arora | 0406ce1 | 2013-08-09 15:57:12 -0700 | [diff] [blame] | 35 | #define LOAD_UV(u, v) ((u) | ((v) << 16)) |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 36 | |
| 37 | #define UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \ |
| 38 | static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \ |
| 39 | const uint8_t* top_u, const uint8_t* top_v, \ |
| 40 | const uint8_t* cur_u, const uint8_t* cur_v, \ |
| 41 | uint8_t* top_dst, uint8_t* bottom_dst, int len) { \ |
| 42 | int x; \ |
| 43 | const int last_pixel_pair = (len - 1) >> 1; \ |
| 44 | uint32_t tl_uv = LOAD_UV(top_u[0], top_v[0]); /* top-left sample */ \ |
| 45 | uint32_t l_uv = LOAD_UV(cur_u[0], cur_v[0]); /* left-sample */ \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 46 | assert(top_y != NULL); \ |
| 47 | { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 48 | const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \ |
| 49 | FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst); \ |
| 50 | } \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 51 | if (bottom_y != NULL) { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 52 | const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \ |
| 53 | FUNC(bottom_y[0], uv0 & 0xff, (uv0 >> 16), bottom_dst); \ |
| 54 | } \ |
| 55 | for (x = 1; x <= last_pixel_pair; ++x) { \ |
| 56 | const uint32_t t_uv = LOAD_UV(top_u[x], top_v[x]); /* top sample */ \ |
| 57 | const uint32_t uv = LOAD_UV(cur_u[x], cur_v[x]); /* sample */ \ |
| 58 | /* precompute invariant values associated with first and second diagonals*/\ |
| 59 | const uint32_t avg = tl_uv + t_uv + l_uv + uv + 0x00080008u; \ |
| 60 | const uint32_t diag_12 = (avg + 2 * (t_uv + l_uv)) >> 3; \ |
| 61 | const uint32_t diag_03 = (avg + 2 * (tl_uv + uv)) >> 3; \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 62 | { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 63 | const uint32_t uv0 = (diag_12 + tl_uv) >> 1; \ |
| 64 | const uint32_t uv1 = (diag_03 + t_uv) >> 1; \ |
| 65 | FUNC(top_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \ |
| 66 | top_dst + (2 * x - 1) * XSTEP); \ |
| 67 | FUNC(top_y[2 * x - 0], uv1 & 0xff, (uv1 >> 16), \ |
| 68 | top_dst + (2 * x - 0) * XSTEP); \ |
| 69 | } \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 70 | if (bottom_y != NULL) { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 71 | const uint32_t uv0 = (diag_03 + l_uv) >> 1; \ |
| 72 | const uint32_t uv1 = (diag_12 + uv) >> 1; \ |
| 73 | FUNC(bottom_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \ |
| 74 | bottom_dst + (2 * x - 1) * XSTEP); \ |
| 75 | FUNC(bottom_y[2 * x + 0], uv1 & 0xff, (uv1 >> 16), \ |
| 76 | bottom_dst + (2 * x + 0) * XSTEP); \ |
| 77 | } \ |
| 78 | tl_uv = t_uv; \ |
| 79 | l_uv = uv; \ |
| 80 | } \ |
| 81 | if (!(len & 1)) { \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 82 | { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 83 | const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \ |
| 84 | FUNC(top_y[len - 1], uv0 & 0xff, (uv0 >> 16), \ |
| 85 | top_dst + (len - 1) * XSTEP); \ |
| 86 | } \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 87 | if (bottom_y != NULL) { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 88 | const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \ |
| 89 | FUNC(bottom_y[len - 1], uv0 & 0xff, (uv0 >> 16), \ |
| 90 | bottom_dst + (len - 1) * XSTEP); \ |
| 91 | } \ |
| 92 | } \ |
| 93 | } |
| 94 | |
| 95 | // All variants implemented. |
| 96 | UPSAMPLE_FUNC(UpsampleRgbLinePair, VP8YuvToRgb, 3) |
| 97 | UPSAMPLE_FUNC(UpsampleBgrLinePair, VP8YuvToBgr, 3) |
| 98 | UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4) |
| 99 | UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4) |
| 100 | UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4) |
| 101 | UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2) |
| 102 | UPSAMPLE_FUNC(UpsampleRgb565LinePair, VP8YuvToRgb565, 2) |
| 103 | |
| 104 | #undef LOAD_UV |
| 105 | #undef UPSAMPLE_FUNC |
| 106 | |
| 107 | #endif // FANCY_UPSAMPLING |
| 108 | |
| 109 | //------------------------------------------------------------------------------ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 110 | |
| 111 | #if !defined(FANCY_UPSAMPLING) |
| 112 | #define DUAL_SAMPLE_FUNC(FUNC_NAME, FUNC) \ |
| 113 | static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y, \ |
| 114 | const uint8_t* top_u, const uint8_t* top_v, \ |
| 115 | const uint8_t* bot_u, const uint8_t* bot_v, \ |
| 116 | uint8_t* top_dst, uint8_t* bot_dst, int len) { \ |
| 117 | const int half_len = len >> 1; \ |
| 118 | int x; \ |
Vikas Arora | 8b72022 | 2014-01-02 16:48:02 -0800 | [diff] [blame] | 119 | assert(top_dst != NULL); \ |
| 120 | { \ |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 121 | for (x = 0; x < half_len; ++x) { \ |
| 122 | FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x + 0); \ |
| 123 | FUNC(top_y[2 * x + 1], top_u[x], top_v[x], top_dst + 8 * x + 4); \ |
| 124 | } \ |
| 125 | if (len & 1) FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x); \ |
| 126 | } \ |
| 127 | if (bot_dst != NULL) { \ |
| 128 | for (x = 0; x < half_len; ++x) { \ |
| 129 | FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x + 0); \ |
| 130 | FUNC(bot_y[2 * x + 1], bot_u[x], bot_v[x], bot_dst + 8 * x + 4); \ |
| 131 | } \ |
| 132 | if (len & 1) FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x); \ |
| 133 | } \ |
| 134 | } |
| 135 | |
| 136 | DUAL_SAMPLE_FUNC(DualLineSamplerBGRA, VP8YuvToBgra) |
| 137 | DUAL_SAMPLE_FUNC(DualLineSamplerARGB, VP8YuvToArgb) |
| 138 | #undef DUAL_SAMPLE_FUNC |
| 139 | |
| 140 | #endif // !FANCY_UPSAMPLING |
| 141 | |
| 142 | WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last) { |
| 143 | WebPInitUpsamplers(); |
| 144 | VP8YUVInit(); |
| 145 | #ifdef FANCY_UPSAMPLING |
| 146 | return WebPUpsamplers[alpha_is_last ? MODE_BGRA : MODE_ARGB]; |
| 147 | #else |
| 148 | return (alpha_is_last ? DualLineSamplerBGRA : DualLineSamplerARGB); |
| 149 | #endif |
| 150 | } |
| 151 | |
| 152 | //------------------------------------------------------------------------------ |
| 153 | // YUV444 converter |
| 154 | |
| 155 | #define YUV444_FUNC(FUNC_NAME, FUNC, XSTEP) \ |
| 156 | static void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \ |
| 157 | uint8_t* dst, int len) { \ |
| 158 | int i; \ |
| 159 | for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * XSTEP]); \ |
| 160 | } |
| 161 | |
| 162 | YUV444_FUNC(Yuv444ToRgb, VP8YuvToRgb, 3) |
| 163 | YUV444_FUNC(Yuv444ToBgr, VP8YuvToBgr, 3) |
| 164 | YUV444_FUNC(Yuv444ToRgba, VP8YuvToRgba, 4) |
| 165 | YUV444_FUNC(Yuv444ToBgra, VP8YuvToBgra, 4) |
| 166 | YUV444_FUNC(Yuv444ToArgb, VP8YuvToArgb, 4) |
| 167 | YUV444_FUNC(Yuv444ToRgba4444, VP8YuvToRgba4444, 2) |
| 168 | YUV444_FUNC(Yuv444ToRgb565, VP8YuvToRgb565, 2) |
| 169 | |
| 170 | #undef YUV444_FUNC |
| 171 | |
| 172 | const WebPYUV444Converter WebPYUV444Converters[MODE_LAST] = { |
| 173 | Yuv444ToRgb, // MODE_RGB |
| 174 | Yuv444ToRgba, // MODE_RGBA |
| 175 | Yuv444ToBgr, // MODE_BGR |
| 176 | Yuv444ToBgra, // MODE_BGRA |
| 177 | Yuv444ToArgb, // MODE_ARGB |
| 178 | Yuv444ToRgba4444, // MODE_RGBA_4444 |
| 179 | Yuv444ToRgb565, // MODE_RGB_565 |
| 180 | Yuv444ToRgba, // MODE_rgbA |
| 181 | Yuv444ToBgra, // MODE_bgrA |
| 182 | Yuv444ToArgb, // MODE_Argb |
| 183 | Yuv444ToRgba4444 // MODE_rgbA_4444 |
| 184 | }; |
| 185 | |
| 186 | //------------------------------------------------------------------------------ |
Vikas Arora | af51b94 | 2014-08-28 10:51:12 -0700 | [diff] [blame] | 187 | // Main calls |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 188 | |
Vikas Arora | af51b94 | 2014-08-28 10:51:12 -0700 | [diff] [blame] | 189 | extern void WebPInitUpsamplersSSE2(void); |
| 190 | extern void WebPInitUpsamplersNEON(void); |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 191 | |
| 192 | void WebPInitUpsamplers(void) { |
| 193 | #ifdef FANCY_UPSAMPLING |
| 194 | WebPUpsamplers[MODE_RGB] = UpsampleRgbLinePair; |
| 195 | WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair; |
| 196 | WebPUpsamplers[MODE_BGR] = UpsampleBgrLinePair; |
| 197 | WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair; |
| 198 | WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair; |
| 199 | WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair; |
| 200 | WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair; |
Vikas Arora | af51b94 | 2014-08-28 10:51:12 -0700 | [diff] [blame] | 201 | WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair; |
| 202 | WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair; |
| 203 | WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair; |
| 204 | WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair; |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 205 | |
| 206 | // If defined, use CPUInfo() to overwrite some pointers with faster versions. |
| 207 | if (VP8GetCPUInfo != NULL) { |
| 208 | #if defined(WEBP_USE_SSE2) |
| 209 | if (VP8GetCPUInfo(kSSE2)) { |
| 210 | WebPInitUpsamplersSSE2(); |
| 211 | } |
| 212 | #endif |
Vikas Arora | 1e7bf88 | 2013-03-13 16:43:18 -0700 | [diff] [blame] | 213 | #if defined(WEBP_USE_NEON) |
| 214 | if (VP8GetCPUInfo(kNEON)) { |
| 215 | WebPInitUpsamplersNEON(); |
| 216 | } |
| 217 | #endif |
Vikas Arora | a241572 | 2012-08-09 16:18:58 -0700 | [diff] [blame] | 218 | } |
| 219 | #endif // FANCY_UPSAMPLING |
| 220 | } |
| 221 | |
Vikas Arora | af51b94 | 2014-08-28 10:51:12 -0700 | [diff] [blame] | 222 | //------------------------------------------------------------------------------ |