| OLD | NEW |
| 1 /* | 1 /* |
| 2 * Copyright 2014 | 2 * Copyright 2014 |
| 3 * | 3 * |
| 4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
| 5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
| 6 */ | 6 */ |
| 7 | 7 |
| 8 #include "SkTextureCompressor.h" | 8 #include "SkTextureCompressor.h" |
| 9 #include "SkTextureCompression_opts.h" | 9 #include "SkTextureCompression_opts.h" |
| 10 | 10 |
| (...skipping 129 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 140 #if defined (SK_CPU_BENDIAN) | 140 #if defined (SK_CPU_BENDIAN) |
| 141 static inline uint64x2_t fix_endianness(uint64x2_t x) { | 141 static inline uint64x2_t fix_endianness(uint64x2_t x) { |
| 142 return x; | 142 return x; |
| 143 } | 143 } |
| 144 #else | 144 #else |
| 145 static inline uint64x2_t fix_endianness(uint64x2_t x) { | 145 static inline uint64x2_t fix_endianness(uint64x2_t x) { |
| 146 return vreinterpretq_u64_u8(vrev64q_u8(vreinterpretq_u8_u64(x))); | 146 return vreinterpretq_u64_u8(vrev64q_u8(vreinterpretq_u8_u64(x))); |
| 147 } | 147 } |
| 148 #endif | 148 #endif |
| 149 | 149 |
| 150 static void compress_r11eac_blocks(uint64_t* dst, const uint8_t* src, int rowByt
es) { | 150 static void compress_r11eac_blocks(uint64_t* dst, const uint8_t* src, size_t row
Bytes) { |
| 151 | 151 |
| 152 // Try to avoid switching between vector and non-vector ops... | 152 // Try to avoid switching between vector and non-vector ops... |
| 153 const uint8_t *const src1 = src; | 153 const uint8_t *const src1 = src; |
| 154 const uint8_t *const src2 = src + rowBytes; | 154 const uint8_t *const src2 = src + rowBytes; |
| 155 const uint8_t *const src3 = src + 2*rowBytes; | 155 const uint8_t *const src3 = src + 2*rowBytes; |
| 156 const uint8_t *const src4 = src + 3*rowBytes; | 156 const uint8_t *const src4 = src + 3*rowBytes; |
| 157 uint64_t *const dst1 = dst; | 157 uint64_t *const dst1 = dst; |
| 158 uint64_t *const dst2 = dst + 2; | 158 uint64_t *const dst2 = dst + 2; |
| 159 | 159 |
| 160 const uint8x16_t alphaRow1 = vld1q_u8(src1); | 160 const uint8x16_t alphaRow1 = vld1q_u8(src1); |
| (...skipping 40 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 201 const uint64x2_t indicesLeft = fix_endianness(pack_indices(blockIndicesLeft)
); | 201 const uint64x2_t indicesLeft = fix_endianness(pack_indices(blockIndicesLeft)
); |
| 202 const uint64x2_t indicesRight = fix_endianness(pack_indices(blockIndicesRigh
t)); | 202 const uint64x2_t indicesRight = fix_endianness(pack_indices(blockIndicesRigh
t)); |
| 203 | 203 |
| 204 const uint64x2_t d1 = vcombine_u64(vget_low_u64(indicesLeft), vget_low_u64(i
ndicesRight)); | 204 const uint64x2_t d1 = vcombine_u64(vget_low_u64(indicesLeft), vget_low_u64(i
ndicesRight)); |
| 205 const uint64x2_t d2 = vcombine_u64(vget_high_u64(indicesLeft), vget_high_u64
(indicesRight)); | 205 const uint64x2_t d2 = vcombine_u64(vget_high_u64(indicesLeft), vget_high_u64
(indicesRight)); |
| 206 vst1q_u64(dst1, d1); | 206 vst1q_u64(dst1, d1); |
| 207 vst1q_u64(dst2, d2); | 207 vst1q_u64(dst2, d2); |
| 208 } | 208 } |
| 209 | 209 |
| 210 bool CompressA8toR11EAC_NEON(uint8_t* dst, const uint8_t* src, | 210 bool CompressA8toR11EAC_NEON(uint8_t* dst, const uint8_t* src, |
| 211 int width, int height, int rowBytes) { | 211 int width, int height, size_t rowBytes) { |
| 212 | 212 |
| 213 // Since we're going to operate on 4 blocks at a time, the src width | 213 // Since we're going to operate on 4 blocks at a time, the src width |
| 214 // must be a multiple of 16. However, the height only needs to be a | 214 // must be a multiple of 16. However, the height only needs to be a |
| 215 // multiple of 4 | 215 // multiple of 4 |
| 216 if (0 == width || 0 == height || (width % 16) != 0 || (height % 4) != 0) { | 216 if (0 == width || 0 == height || (width % 16) != 0 || (height % 4) != 0) { |
| 217 return SkTextureCompressor::CompressBufferToFormat( | 217 return SkTextureCompressor::CompressBufferToFormat( |
| 218 dst, src, | 218 dst, src, |
| 219 kAlpha_8_SkColorType, | 219 kAlpha_8_SkColorType, |
| 220 width, height, rowBytes, | 220 width, height, rowBytes, |
| 221 SkTextureCompressor::kR11_EAC_Format, false); | 221 SkTextureCompressor::kR11_EAC_Format, false); |
| 222 } | 222 } |
| 223 | 223 |
| 224 const int blocksX = width >> 2; | 224 const int blocksX = width >> 2; |
| 225 const int blocksY = height >> 2; | 225 const int blocksY = height >> 2; |
| 226 | 226 |
| 227 SkASSERT((blocksX % 4) == 0); | 227 SkASSERT((blocksX % 4) == 0); |
| 228 | 228 |
| 229 uint64_t* encPtr = reinterpret_cast<uint64_t*>(dst); | 229 uint64_t* encPtr = reinterpret_cast<uint64_t*>(dst); |
| 230 for (int y = 0; y < blocksY; ++y) { | 230 for (int y = 0; y < blocksY; ++y) { |
| 231 for (int x = 0; x < blocksX; x+=4) { | 231 for (int x = 0; x < blocksX; x+=4) { |
| 232 // Compress it | 232 // Compress it |
| 233 compress_r11eac_blocks(encPtr, src + 4*x, rowBytes); | 233 compress_r11eac_blocks(encPtr, src + 4*x, rowBytes); |
| 234 encPtr += 4; | 234 encPtr += 4; |
| 235 } | 235 } |
| 236 src += 4 * rowBytes; | 236 src += 4 * rowBytes; |
| 237 } | 237 } |
| 238 return true; | 238 return true; |
| 239 } | 239 } |
| OLD | NEW |