[00587dc] | 1 | /********************************************************************* |
---|
| 2 | Blosc - Blocked Suffling and Compression Library |
---|
| 3 | |
---|
| 4 | Author: Francesc Alted <[email protected]> |
---|
| 5 | Creation date: 2009-05-20 |
---|
| 6 | |
---|
| 7 | See LICENSES/BLOSC.txt for details about copyright and rights to use. |
---|
| 8 | **********************************************************************/ |
---|
| 9 | |
---|
| 10 | #include <stdio.h> |
---|
| 11 | #include <string.h> |
---|
| 12 | #include "shuffle.h" |
---|
| 13 | |
---|
| 14 | #if defined(_WIN32) && !defined(__MINGW32__) |
---|
| 15 | #include <windows.h> |
---|
| 16 | #include "win32/stdint-windows.h" |
---|
| 17 | #define __SSE2__ /* Windows does not define this by default */ |
---|
| 18 | #else |
---|
| 19 | #include <stdint.h> |
---|
| 20 | #include <inttypes.h> |
---|
| 21 | #endif /* _WIN32 */ |
---|
| 22 | |
---|
| 23 | |
---|
| 24 | /* The non-SSE2 versions of shuffle and unshuffle */ |
---|
| 25 | |
---|
| 26 | /* Shuffle a block. This can never fail. */ |
---|
| 27 | static void _shuffle(size_t bytesoftype, size_t blocksize, |
---|
| 28 | uint8_t* _src, uint8_t* _dest) |
---|
| 29 | { |
---|
| 30 | size_t i, j, neblock, leftover; |
---|
| 31 | |
---|
| 32 | /* Non-optimized shuffle */ |
---|
| 33 | neblock = blocksize / bytesoftype; /* Number of elements in a block */ |
---|
| 34 | for (j = 0; j < bytesoftype; j++) { |
---|
| 35 | for (i = 0; i < neblock; i++) { |
---|
| 36 | _dest[j*neblock+i] = _src[i*bytesoftype+j]; |
---|
| 37 | } |
---|
| 38 | } |
---|
| 39 | leftover = blocksize % bytesoftype; |
---|
| 40 | memcpy(_dest + neblock*bytesoftype, _src + neblock*bytesoftype, leftover); |
---|
| 41 | } |
---|
| 42 | |
---|
| 43 | /* Unshuffle a block. This can never fail. */ |
---|
| 44 | static void _unshuffle(size_t bytesoftype, size_t blocksize, |
---|
| 45 | uint8_t* _src, uint8_t* _dest) |
---|
| 46 | { |
---|
| 47 | size_t i, j, neblock, leftover; |
---|
| 48 | |
---|
| 49 | /* Non-optimized unshuffle */ |
---|
| 50 | neblock = blocksize / bytesoftype; /* Number of elements in a block */ |
---|
| 51 | for (i = 0; i < neblock; i++) { |
---|
| 52 | for (j = 0; j < bytesoftype; j++) { |
---|
| 53 | _dest[i*bytesoftype+j] = _src[j*neblock+i]; |
---|
| 54 | } |
---|
| 55 | } |
---|
| 56 | leftover = blocksize % bytesoftype; |
---|
| 57 | memcpy(_dest+neblock*bytesoftype, _src+neblock*bytesoftype, leftover); |
---|
| 58 | } |
---|
| 59 | |
---|
| 60 | |
---|
| 61 | #ifdef __SSE2__ |
---|
| 62 | |
---|
| 63 | /* The SSE2 versions of shuffle and unshuffle */ |
---|
| 64 | |
---|
| 65 | #include <emmintrin.h> |
---|
| 66 | |
---|
| 67 | /* The next is useful for debugging purposes */ |
---|
| 68 | #if 0 |
---|
| 69 | static void printxmm(__m128i xmm0) |
---|
| 70 | { |
---|
| 71 | uint8_t buf[16]; |
---|
| 72 | |
---|
| 73 | ((__m128i *)buf)[0] = xmm0; |
---|
| 74 | printf("%x,%x,%x,%x,%x,%x,%x,%x,%x,%x,%x,%x,%x,%x,%x,%x\n", |
---|
| 75 | buf[0], buf[1], buf[2], buf[3], |
---|
| 76 | buf[4], buf[5], buf[6], buf[7], |
---|
| 77 | buf[8], buf[9], buf[10], buf[11], |
---|
| 78 | buf[12], buf[13], buf[14], buf[15]); |
---|
| 79 | } |
---|
| 80 | #endif |
---|
| 81 | |
---|
| 82 | |
---|
| 83 | /* Routine optimized for shuffling a buffer for a type size of 2 bytes. */ |
---|
| 84 | static void |
---|
| 85 | shuffle2(uint8_t* dest, uint8_t* src, size_t size) |
---|
| 86 | { |
---|
| 87 | size_t i, j, k; |
---|
| 88 | size_t numof16belem; |
---|
| 89 | __m128i xmm0[2], xmm1[2]; |
---|
| 90 | |
---|
| 91 | numof16belem = size / (16*2); |
---|
| 92 | for (i = 0, j = 0; i < numof16belem; i++, j += 16*2) { |
---|
| 93 | /* Fetch and transpose bytes, words and double words in groups of |
---|
| 94 | 32 bytes */ |
---|
| 95 | for (k = 0; k < 2; k++) { |
---|
| 96 | xmm0[k] = _mm_loadu_si128((__m128i*)(src+j+k*16)); |
---|
| 97 | xmm0[k] = _mm_shufflelo_epi16(xmm0[k], 0xd8); |
---|
| 98 | xmm0[k] = _mm_shufflehi_epi16(xmm0[k], 0xd8); |
---|
| 99 | xmm0[k] = _mm_shuffle_epi32(xmm0[k], 0xd8); |
---|
| 100 | xmm1[k] = _mm_shuffle_epi32(xmm0[k], 0x4e); |
---|
| 101 | xmm0[k] = _mm_unpacklo_epi8(xmm0[k], xmm1[k]); |
---|
| 102 | xmm0[k] = _mm_shuffle_epi32(xmm0[k], 0xd8); |
---|
| 103 | xmm1[k] = _mm_shuffle_epi32(xmm0[k], 0x4e); |
---|
| 104 | xmm0[k] = _mm_unpacklo_epi16(xmm0[k], xmm1[k]); |
---|
| 105 | xmm0[k] = _mm_shuffle_epi32(xmm0[k], 0xd8); |
---|
| 106 | } |
---|
| 107 | /* Transpose quad words */ |
---|
| 108 | for (k = 0; k < 1; k++) { |
---|
| 109 | xmm1[k*2] = _mm_unpacklo_epi64(xmm0[k], xmm0[k+1]); |
---|
| 110 | xmm1[k*2+1] = _mm_unpackhi_epi64(xmm0[k], xmm0[k+1]); |
---|
| 111 | } |
---|
| 112 | /* Store the result vectors */ |
---|
| 113 | for (k = 0; k < 2; k++) { |
---|
| 114 | ((__m128i *)dest)[k*numof16belem+i] = xmm1[k]; |
---|
| 115 | } |
---|
| 116 | } |
---|
| 117 | } |
---|
| 118 | |
---|
| 119 | |
---|
| 120 | /* Routine optimized for shuffling a buffer for a type size of 4 bytes. */ |
---|
| 121 | static void |
---|
| 122 | shuffle4(uint8_t* dest, uint8_t* src, size_t size) |
---|
| 123 | { |
---|
| 124 | size_t i, j, k; |
---|
| 125 | size_t numof16belem; |
---|
| 126 | __m128i xmm0[4], xmm1[4]; |
---|
| 127 | |
---|
| 128 | numof16belem = size / (16*4); |
---|
| 129 | for (i = 0, j = 0; i < numof16belem; i++, j += 16*4) { |
---|
| 130 | /* Fetch and transpose bytes and words in groups of 64 bytes */ |
---|
| 131 | for (k = 0; k < 4; k++) { |
---|
| 132 | xmm0[k] = _mm_loadu_si128((__m128i*)(src+j+k*16)); |
---|
| 133 | xmm1[k] = _mm_shuffle_epi32(xmm0[k], 0xd8); |
---|
| 134 | xmm0[k] = _mm_shuffle_epi32(xmm0[k], 0x8d); |
---|
| 135 | xmm0[k] = _mm_unpacklo_epi8(xmm1[k], xmm0[k]); |
---|
| 136 | xmm1[k] = _mm_shuffle_epi32(xmm0[k], 0x04e); |
---|
| 137 | xmm0[k] = _mm_unpacklo_epi16(xmm0[k], xmm1[k]); |
---|
| 138 | } |
---|
| 139 | /* Transpose double words */ |
---|
| 140 | for (k = 0; k < 2; k++) { |
---|
| 141 | xmm1[k*2] = _mm_unpacklo_epi32(xmm0[k*2], xmm0[k*2+1]); |
---|
| 142 | xmm1[k*2+1] = _mm_unpackhi_epi32(xmm0[k*2], xmm0[k*2+1]); |
---|
| 143 | } |
---|
| 144 | /* Transpose quad words */ |
---|
| 145 | for (k = 0; k < 2; k++) { |
---|
| 146 | xmm0[k*2] = _mm_unpacklo_epi64(xmm1[k], xmm1[k+2]); |
---|
| 147 | xmm0[k*2+1] = _mm_unpackhi_epi64(xmm1[k], xmm1[k+2]); |
---|
| 148 | } |
---|
| 149 | /* Store the result vectors */ |
---|
| 150 | for (k = 0; k < 4; k++) { |
---|
| 151 | ((__m128i *)dest)[k*numof16belem+i] = xmm0[k]; |
---|
| 152 | } |
---|
| 153 | } |
---|
| 154 | } |
---|
| 155 | |
---|
| 156 | |
---|
| 157 | /* Routine optimized for shuffling a buffer for a type size of 8 bytes. */ |
---|
| 158 | static void |
---|
| 159 | shuffle8(uint8_t* dest, uint8_t* src, size_t size) |
---|
| 160 | { |
---|
| 161 | size_t i, j, k, l; |
---|
| 162 | size_t numof16belem; |
---|
| 163 | __m128i xmm0[8], xmm1[8]; |
---|
| 164 | |
---|
| 165 | numof16belem = size / (16*8); |
---|
| 166 | for (i = 0, j = 0; i < numof16belem; i++, j += 16*8) { |
---|
| 167 | /* Fetch and transpose bytes in groups of 128 bytes */ |
---|
| 168 | for (k = 0; k < 8; k++) { |
---|
| 169 | xmm0[k] = _mm_loadu_si128((__m128i*)(src+j+k*16)); |
---|
| 170 | xmm1[k] = _mm_shuffle_epi32(xmm0[k], 0x4e); |
---|
| 171 | xmm1[k] = _mm_unpacklo_epi8(xmm0[k], xmm1[k]); |
---|
| 172 | } |
---|
| 173 | /* Transpose words */ |
---|
| 174 | for (k = 0, l = 0; k < 4; k++, l +=2) { |
---|
| 175 | xmm0[k*2] = _mm_unpacklo_epi16(xmm1[l], xmm1[l+1]); |
---|
| 176 | xmm0[k*2+1] = _mm_unpackhi_epi16(xmm1[l], xmm1[l+1]); |
---|
| 177 | } |
---|
| 178 | /* Transpose double words */ |
---|
| 179 | for (k = 0, l = 0; k < 4; k++, l++) { |
---|
| 180 | if (k == 2) l += 2; |
---|
| 181 | xmm1[k*2] = _mm_unpacklo_epi32(xmm0[l], xmm0[l+2]); |
---|
| 182 | xmm1[k*2+1] = _mm_unpackhi_epi32(xmm0[l], xmm0[l+2]); |
---|
| 183 | } |
---|
| 184 | /* Transpose quad words */ |
---|
| 185 | for (k = 0; k < 4; k++) { |
---|
| 186 | xmm0[k*2] = _mm_unpacklo_epi64(xmm1[k], xmm1[k+4]); |
---|
| 187 | xmm0[k*2+1] = _mm_unpackhi_epi64(xmm1[k], xmm1[k+4]); |
---|
| 188 | } |
---|
| 189 | /* Store the result vectors */ |
---|
| 190 | for (k = 0; k < 8; k++) { |
---|
| 191 | ((__m128i *)dest)[k*numof16belem+i] = xmm0[k]; |
---|
| 192 | } |
---|
| 193 | } |
---|
| 194 | } |
---|
| 195 | |
---|
| 196 | |
---|
| 197 | /* Routine optimized for shuffling a buffer for a type size of 16 bytes. */ |
---|
| 198 | static void |
---|
| 199 | shuffle16(uint8_t* dest, uint8_t* src, size_t size) |
---|
| 200 | { |
---|
| 201 | size_t i, j, k, l; |
---|
| 202 | size_t numof16belem; |
---|
| 203 | __m128i xmm0[16], xmm1[16]; |
---|
| 204 | |
---|
| 205 | numof16belem = size / (16*16); |
---|
| 206 | for (i = 0, j = 0; i < numof16belem; i++, j += 16*16) { |
---|
| 207 | /* Fetch elements in groups of 256 bytes */ |
---|
| 208 | for (k = 0; k < 16; k++) { |
---|
| 209 | xmm0[k] = _mm_loadu_si128((__m128i*)(src+j+k*16)); |
---|
| 210 | } |
---|
| 211 | /* Transpose bytes */ |
---|
| 212 | for (k = 0, l = 0; k < 8; k++, l +=2) { |
---|
| 213 | xmm1[k*2] = _mm_unpacklo_epi8(xmm0[l], xmm0[l+1]); |
---|
| 214 | xmm1[k*2+1] = _mm_unpackhi_epi8(xmm0[l], xmm0[l+1]); |
---|
| 215 | } |
---|
| 216 | /* Transpose words */ |
---|
| 217 | for (k = 0, l = -2; k < 8; k++, l++) { |
---|
| 218 | if ((k%2) == 0) l += 2; |
---|
| 219 | xmm0[k*2] = _mm_unpacklo_epi16(xmm1[l], xmm1[l+2]); |
---|
| 220 | xmm0[k*2+1] = _mm_unpackhi_epi16(xmm1[l], xmm1[l+2]); |
---|
| 221 | } |
---|
| 222 | /* Transpose double words */ |
---|
| 223 | for (k = 0, l = -4; k < 8; k++, l++) { |
---|
| 224 | if ((k%4) == 0) l += 4; |
---|
| 225 | xmm1[k*2] = _mm_unpacklo_epi32(xmm0[l], xmm0[l+4]); |
---|
| 226 | xmm1[k*2+1] = _mm_unpackhi_epi32(xmm0[l], xmm0[l+4]); |
---|
| 227 | } |
---|
| 228 | /* Transpose quad words */ |
---|
| 229 | for (k = 0; k < 8; k++) { |
---|
| 230 | xmm0[k*2] = _mm_unpacklo_epi64(xmm1[k], xmm1[k+8]); |
---|
| 231 | xmm0[k*2+1] = _mm_unpackhi_epi64(xmm1[k], xmm1[k+8]); |
---|
| 232 | } |
---|
| 233 | /* Store the result vectors */ |
---|
| 234 | for (k = 0; k < 16; k++) { |
---|
| 235 | ((__m128i *)dest)[k*numof16belem+i] = xmm0[k]; |
---|
| 236 | } |
---|
| 237 | } |
---|
| 238 | } |
---|
| 239 | |
---|
| 240 | |
---|
| 241 | /* Shuffle a block. This can never fail. */ |
---|
| 242 | void shuffle(size_t bytesoftype, size_t blocksize, |
---|
| 243 | uint8_t* _src, uint8_t* _dest) { |
---|
| 244 | int unaligned_dest = (int)((uintptr_t)_dest % 16); |
---|
| 245 | int power_of_two = (blocksize & (blocksize - 1)) == 0; |
---|
| 246 | int too_small = (blocksize < 256); |
---|
| 247 | |
---|
| 248 | if (unaligned_dest || !power_of_two || too_small) { |
---|
| 249 | /* _dest buffer is not aligned, not a power of two or is too |
---|
| 250 | small. Call the non-sse2 version. */ |
---|
| 251 | _shuffle(bytesoftype, blocksize, _src, _dest); |
---|
| 252 | return; |
---|
| 253 | } |
---|
| 254 | |
---|
| 255 | /* Optimized shuffle */ |
---|
| 256 | /* The buffer must be aligned on a 16 bytes boundary, have a power */ |
---|
| 257 | /* of 2 size and be larger or equal than 256 bytes. */ |
---|
| 258 | if (bytesoftype == 4) { |
---|
| 259 | shuffle4(_dest, _src, blocksize); |
---|
| 260 | } |
---|
| 261 | else if (bytesoftype == 8) { |
---|
| 262 | shuffle8(_dest, _src, blocksize); |
---|
| 263 | } |
---|
| 264 | else if (bytesoftype == 16) { |
---|
| 265 | shuffle16(_dest, _src, blocksize); |
---|
| 266 | } |
---|
| 267 | else if (bytesoftype == 2) { |
---|
| 268 | shuffle2(_dest, _src, blocksize); |
---|
| 269 | } |
---|
| 270 | else { |
---|
| 271 | /* Non-optimized shuffle */ |
---|
| 272 | _shuffle(bytesoftype, blocksize, _src, _dest); |
---|
| 273 | } |
---|
| 274 | } |
---|
| 275 | |
---|
| 276 | |
---|
| 277 | /* Routine optimized for unshuffling a buffer for a type size of 2 bytes. */ |
---|
| 278 | static void |
---|
| 279 | unshuffle2(uint8_t* dest, uint8_t* orig, size_t size) |
---|
| 280 | { |
---|
| 281 | size_t i, k; |
---|
| 282 | size_t neblock, numof16belem; |
---|
| 283 | __m128i xmm1[2], xmm2[2]; |
---|
| 284 | |
---|
| 285 | neblock = size / 2; |
---|
| 286 | numof16belem = neblock / 16; |
---|
| 287 | for (i = 0, k = 0; i < numof16belem; i++, k += 2) { |
---|
| 288 | /* Load the first 32 bytes in 2 XMM registrers */ |
---|
| 289 | xmm1[0] = ((__m128i *)orig)[0*numof16belem+i]; |
---|
| 290 | xmm1[1] = ((__m128i *)orig)[1*numof16belem+i]; |
---|
| 291 | /* Shuffle bytes */ |
---|
| 292 | /* Compute the low 32 bytes */ |
---|
| 293 | xmm2[0] = _mm_unpacklo_epi8(xmm1[0], xmm1[1]); |
---|
| 294 | /* Compute the hi 32 bytes */ |
---|
| 295 | xmm2[1] = _mm_unpackhi_epi8(xmm1[0], xmm1[1]); |
---|
| 296 | /* Store the result vectors in proper order */ |
---|
| 297 | ((__m128i *)dest)[k+0] = xmm2[0]; |
---|
| 298 | ((__m128i *)dest)[k+1] = xmm2[1]; |
---|
| 299 | } |
---|
| 300 | } |
---|
| 301 | |
---|
| 302 | |
---|
| 303 | /* Routine optimized for unshuffling a buffer for a type size of 4 bytes. */ |
---|
| 304 | static void |
---|
| 305 | unshuffle4(uint8_t* dest, uint8_t* orig, size_t size) |
---|
| 306 | { |
---|
| 307 | size_t i, j, k; |
---|
| 308 | size_t neblock, numof16belem; |
---|
| 309 | __m128i xmm0[4], xmm1[4]; |
---|
| 310 | |
---|
| 311 | neblock = size / 4; |
---|
| 312 | numof16belem = neblock / 16; |
---|
| 313 | for (i = 0, k = 0; i < numof16belem; i++, k += 4) { |
---|
| 314 | /* Load the first 64 bytes in 4 XMM registrers */ |
---|
| 315 | for (j = 0; j < 4; j++) { |
---|
| 316 | xmm0[j] = ((__m128i *)orig)[j*numof16belem+i]; |
---|
| 317 | } |
---|
| 318 | /* Shuffle bytes */ |
---|
| 319 | for (j = 0; j < 2; j++) { |
---|
| 320 | /* Compute the low 32 bytes */ |
---|
| 321 | xmm1[j] = _mm_unpacklo_epi8(xmm0[j*2], xmm0[j*2+1]); |
---|
| 322 | /* Compute the hi 32 bytes */ |
---|
| 323 | xmm1[2+j] = _mm_unpackhi_epi8(xmm0[j*2], xmm0[j*2+1]); |
---|
| 324 | } |
---|
| 325 | /* Shuffle 2-byte words */ |
---|
| 326 | for (j = 0; j < 2; j++) { |
---|
| 327 | /* Compute the low 32 bytes */ |
---|
| 328 | xmm0[j] = _mm_unpacklo_epi16(xmm1[j*2], xmm1[j*2+1]); |
---|
| 329 | /* Compute the hi 32 bytes */ |
---|
| 330 | xmm0[2+j] = _mm_unpackhi_epi16(xmm1[j*2], xmm1[j*2+1]); |
---|
| 331 | } |
---|
| 332 | /* Store the result vectors in proper order */ |
---|
| 333 | ((__m128i *)dest)[k+0] = xmm0[0]; |
---|
| 334 | ((__m128i *)dest)[k+1] = xmm0[2]; |
---|
| 335 | ((__m128i *)dest)[k+2] = xmm0[1]; |
---|
| 336 | ((__m128i *)dest)[k+3] = xmm0[3]; |
---|
| 337 | } |
---|
| 338 | } |
---|
| 339 | |
---|
| 340 | |
---|
| 341 | /* Routine optimized for unshuffling a buffer for a type size of 8 bytes. */ |
---|
| 342 | static void |
---|
| 343 | unshuffle8(uint8_t* dest, uint8_t* orig, size_t size) |
---|
| 344 | { |
---|
| 345 | size_t i, j, k; |
---|
| 346 | size_t neblock, numof16belem; |
---|
| 347 | __m128i xmm0[8], xmm1[8]; |
---|
| 348 | |
---|
| 349 | neblock = size / 8; |
---|
| 350 | numof16belem = neblock / 16; |
---|
| 351 | for (i = 0, k = 0; i < numof16belem; i++, k += 8) { |
---|
| 352 | /* Load the first 64 bytes in 8 XMM registrers */ |
---|
| 353 | for (j = 0; j < 8; j++) { |
---|
| 354 | xmm0[j] = ((__m128i *)orig)[j*numof16belem+i]; |
---|
| 355 | } |
---|
| 356 | /* Shuffle bytes */ |
---|
| 357 | for (j = 0; j < 4; j++) { |
---|
| 358 | /* Compute the low 32 bytes */ |
---|
| 359 | xmm1[j] = _mm_unpacklo_epi8(xmm0[j*2], xmm0[j*2+1]); |
---|
| 360 | /* Compute the hi 32 bytes */ |
---|
| 361 | xmm1[4+j] = _mm_unpackhi_epi8(xmm0[j*2], xmm0[j*2+1]); |
---|
| 362 | } |
---|
| 363 | /* Shuffle 2-byte words */ |
---|
| 364 | for (j = 0; j < 4; j++) { |
---|
| 365 | /* Compute the low 32 bytes */ |
---|
| 366 | xmm0[j] = _mm_unpacklo_epi16(xmm1[j*2], xmm1[j*2+1]); |
---|
| 367 | /* Compute the hi 32 bytes */ |
---|
| 368 | xmm0[4+j] = _mm_unpackhi_epi16(xmm1[j*2], xmm1[j*2+1]); |
---|
| 369 | } |
---|
| 370 | /* Shuffle 4-byte dwords */ |
---|
| 371 | for (j = 0; j < 4; j++) { |
---|
| 372 | /* Compute the low 32 bytes */ |
---|
| 373 | xmm1[j] = _mm_unpacklo_epi32(xmm0[j*2], xmm0[j*2+1]); |
---|
| 374 | /* Compute the hi 32 bytes */ |
---|
| 375 | xmm1[4+j] = _mm_unpackhi_epi32(xmm0[j*2], xmm0[j*2+1]); |
---|
| 376 | } |
---|
| 377 | /* Store the result vectors in proper order */ |
---|
| 378 | ((__m128i *)dest)[k+0] = xmm1[0]; |
---|
| 379 | ((__m128i *)dest)[k+1] = xmm1[4]; |
---|
| 380 | ((__m128i *)dest)[k+2] = xmm1[2]; |
---|
| 381 | ((__m128i *)dest)[k+3] = xmm1[6]; |
---|
| 382 | ((__m128i *)dest)[k+4] = xmm1[1]; |
---|
| 383 | ((__m128i *)dest)[k+5] = xmm1[5]; |
---|
| 384 | ((__m128i *)dest)[k+6] = xmm1[3]; |
---|
| 385 | ((__m128i *)dest)[k+7] = xmm1[7]; |
---|
| 386 | } |
---|
| 387 | } |
---|
| 388 | |
---|
| 389 | |
---|
| 390 | /* Routine optimized for unshuffling a buffer for a type size of 16 bytes. */ |
---|
| 391 | static void |
---|
| 392 | unshuffle16(uint8_t* dest, uint8_t* orig, size_t size) |
---|
| 393 | { |
---|
| 394 | size_t i, j, k; |
---|
| 395 | size_t neblock, numof16belem; |
---|
| 396 | __m128i xmm1[16], xmm2[16]; |
---|
| 397 | |
---|
| 398 | neblock = size / 16; |
---|
| 399 | numof16belem = neblock / 16; |
---|
| 400 | for (i = 0, k = 0; i < numof16belem; i++, k += 16) { |
---|
| 401 | /* Load the first 128 bytes in 16 XMM registrers */ |
---|
| 402 | for (j = 0; j < 16; j++) { |
---|
| 403 | xmm1[j] = ((__m128i *)orig)[j*numof16belem+i]; |
---|
| 404 | } |
---|
| 405 | /* Shuffle bytes */ |
---|
| 406 | for (j = 0; j < 8; j++) { |
---|
| 407 | /* Compute the low 32 bytes */ |
---|
| 408 | xmm2[j] = _mm_unpacklo_epi8(xmm1[j*2], xmm1[j*2+1]); |
---|
| 409 | /* Compute the hi 32 bytes */ |
---|
| 410 | xmm2[8+j] = _mm_unpackhi_epi8(xmm1[j*2], xmm1[j*2+1]); |
---|
| 411 | } |
---|
| 412 | /* Shuffle 2-byte words */ |
---|
| 413 | for (j = 0; j < 8; j++) { |
---|
| 414 | /* Compute the low 32 bytes */ |
---|
| 415 | xmm1[j] = _mm_unpacklo_epi16(xmm2[j*2], xmm2[j*2+1]); |
---|
| 416 | /* Compute the hi 32 bytes */ |
---|
| 417 | xmm1[8+j] = _mm_unpackhi_epi16(xmm2[j*2], xmm2[j*2+1]); |
---|
| 418 | } |
---|
| 419 | /* Shuffle 4-byte dwords */ |
---|
| 420 | for (j = 0; j < 8; j++) { |
---|
| 421 | /* Compute the low 32 bytes */ |
---|
| 422 | xmm2[j] = _mm_unpacklo_epi32(xmm1[j*2], xmm1[j*2+1]); |
---|
| 423 | /* Compute the hi 32 bytes */ |
---|
| 424 | xmm2[8+j] = _mm_unpackhi_epi32(xmm1[j*2], xmm1[j*2+1]); |
---|
| 425 | } |
---|
| 426 | /* Shuffle 8-byte qwords */ |
---|
| 427 | for (j = 0; j < 8; j++) { |
---|
| 428 | /* Compute the low 32 bytes */ |
---|
| 429 | xmm1[j] = _mm_unpacklo_epi64(xmm2[j*2], xmm2[j*2+1]); |
---|
| 430 | /* Compute the hi 32 bytes */ |
---|
| 431 | xmm1[8+j] = _mm_unpackhi_epi64(xmm2[j*2], xmm2[j*2+1]); |
---|
| 432 | } |
---|
| 433 | /* Store the result vectors in proper order */ |
---|
| 434 | ((__m128i *)dest)[k+0] = xmm1[0]; |
---|
| 435 | ((__m128i *)dest)[k+1] = xmm1[8]; |
---|
| 436 | ((__m128i *)dest)[k+2] = xmm1[4]; |
---|
| 437 | ((__m128i *)dest)[k+3] = xmm1[12]; |
---|
| 438 | ((__m128i *)dest)[k+4] = xmm1[2]; |
---|
| 439 | ((__m128i *)dest)[k+5] = xmm1[10]; |
---|
| 440 | ((__m128i *)dest)[k+6] = xmm1[6]; |
---|
| 441 | ((__m128i *)dest)[k+7] = xmm1[14]; |
---|
| 442 | ((__m128i *)dest)[k+8] = xmm1[1]; |
---|
| 443 | ((__m128i *)dest)[k+9] = xmm1[9]; |
---|
| 444 | ((__m128i *)dest)[k+10] = xmm1[5]; |
---|
| 445 | ((__m128i *)dest)[k+11] = xmm1[13]; |
---|
| 446 | ((__m128i *)dest)[k+12] = xmm1[3]; |
---|
| 447 | ((__m128i *)dest)[k+13] = xmm1[11]; |
---|
| 448 | ((__m128i *)dest)[k+14] = xmm1[7]; |
---|
| 449 | ((__m128i *)dest)[k+15] = xmm1[15]; |
---|
| 450 | } |
---|
| 451 | } |
---|
| 452 | |
---|
| 453 | |
---|
| 454 | /* Unshuffle a block. This can never fail. */ |
---|
| 455 | void unshuffle(size_t bytesoftype, size_t blocksize, |
---|
| 456 | uint8_t* _src, uint8_t* _dest) { |
---|
| 457 | int unaligned_src = (int)((uintptr_t)_src % 16); |
---|
| 458 | int unaligned_dest = (int)((uintptr_t)_dest % 16); |
---|
| 459 | int power_of_two = (blocksize & (blocksize - 1)) == 0; |
---|
| 460 | int too_small = (blocksize < 256); |
---|
| 461 | |
---|
| 462 | if (unaligned_src || unaligned_dest || !power_of_two || too_small) { |
---|
| 463 | /* _src or _dest buffer is not aligned, not a power of two or is |
---|
| 464 | too small. Call the non-sse2 version. */ |
---|
| 465 | _unshuffle(bytesoftype, blocksize, _src, _dest); |
---|
| 466 | return; |
---|
| 467 | } |
---|
| 468 | |
---|
| 469 | /* Optimized unshuffle */ |
---|
| 470 | /* The buffers must be aligned on a 16 bytes boundary, have a power */ |
---|
| 471 | /* of 2 size and be larger or equal than 256 bytes. */ |
---|
| 472 | if (bytesoftype == 4) { |
---|
| 473 | unshuffle4(_dest, _src, blocksize); |
---|
| 474 | } |
---|
| 475 | else if (bytesoftype == 8) { |
---|
| 476 | unshuffle8(_dest, _src, blocksize); |
---|
| 477 | } |
---|
| 478 | else if (bytesoftype == 16) { |
---|
| 479 | unshuffle16(_dest, _src, blocksize); |
---|
| 480 | } |
---|
| 481 | else if (bytesoftype == 2) { |
---|
| 482 | unshuffle2(_dest, _src, blocksize); |
---|
| 483 | } |
---|
| 484 | else { |
---|
| 485 | /* Non-optimized unshuffle */ |
---|
| 486 | _unshuffle(bytesoftype, blocksize, _src, _dest); |
---|
| 487 | } |
---|
| 488 | } |
---|
| 489 | |
---|
| 490 | #else /* no __SSE2__ available */ |
---|
| 491 | |
---|
| 492 | void shuffle(size_t bytesoftype, size_t blocksize, |
---|
| 493 | uint8_t* _src, uint8_t* _dest) { |
---|
| 494 | _shuffle(bytesoftype, blocksize, _src, _dest); |
---|
| 495 | } |
---|
| 496 | |
---|
| 497 | void unshuffle(size_t bytesoftype, size_t blocksize, |
---|
| 498 | uint8_t* _src, uint8_t* _dest) { |
---|
| 499 | _unshuffle(bytesoftype, blocksize, _src, _dest); |
---|
| 500 | } |
---|
| 501 | |
---|
| 502 | #endif /* __SSE2__ */ |
---|