hbono@chromium.org | f0c4f33 | 2010-11-01 05:14:55 +0000 | [diff] [blame] | 1 | /* |
| 2 | * jsimd_i386.c |
| 3 | * |
| 4 | * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB |
| 5 | * Copyright 2009 D. R. Commander |
| 6 | * |
| 7 | * Based on the x86 SIMD extension for IJG JPEG library, |
| 8 | * Copyright (C) 1999-2006, MIYASAKA Masaru. |
| 9 | * |
| 10 | * This file contains the interface between the "normal" portions |
| 11 | * of the library and the SIMD implementations when running on a |
| 12 | * 32-bit x86 architecture. |
| 13 | */ |
| 14 | |
| 15 | #define JPEG_INTERNALS |
| 16 | #include "../jinclude.h" |
| 17 | #include "../jpeglib.h" |
| 18 | #include "../jsimd.h" |
| 19 | #include "../jdct.h" |
| 20 | #include "../jsimddct.h" |
| 21 | #include "jsimd.h" |
| 22 | |
| 23 | /* |
| 24 | * In the PIC cases, we have no guarantee that constants will keep |
| 25 | * their alignment. This macro allows us to verify it at runtime. |
| 26 | */ |
| 27 | #define IS_ALIGNED(ptr, order) (((unsigned)ptr & ((1 << order) - 1)) == 0) |
| 28 | |
| 29 | #define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */ |
| 30 | |
| 31 | static unsigned int simd_support = ~0; |
| 32 | |
| 33 | /* |
| 34 | * Check what SIMD accelerations are supported. |
| 35 | * |
| 36 | * FIXME: This code is racy under a multi-threaded environment. |
| 37 | */ |
| 38 | LOCAL(void) |
| 39 | init_simd (void) |
| 40 | { |
| 41 | char *env = NULL; |
| 42 | |
| 43 | if (simd_support != ~0) |
| 44 | return; |
| 45 | |
| 46 | simd_support = jpeg_simd_cpu_support(); |
| 47 | |
| 48 | /* Force different settings through environment variables */ |
| 49 | env = getenv("JSIMD_FORCEMMX"); |
| 50 | if ((env != NULL) && (strcmp(env, "1") == 0)) |
| 51 | simd_support &= JSIMD_MMX; |
| 52 | env = getenv("JSIMD_FORCE3DNOW"); |
| 53 | if ((env != NULL) && (strcmp(env, "1") == 0)) |
| 54 | simd_support &= JSIMD_3DNOW|JSIMD_MMX; |
| 55 | env = getenv("JSIMD_FORCESSE"); |
| 56 | if ((env != NULL) && (strcmp(env, "1") == 0)) |
| 57 | simd_support &= JSIMD_SSE|JSIMD_MMX; |
| 58 | env = getenv("JSIMD_FORCESSE2"); |
| 59 | if ((env != NULL) && (strcmp(env, "1") == 0)) |
| 60 | simd_support &= JSIMD_SSE2; |
| 61 | } |
| 62 | |
| 63 | GLOBAL(int) |
| 64 | jsimd_can_rgb_ycc (void) |
| 65 | { |
| 66 | init_simd(); |
| 67 | |
| 68 | /* The code is optimised for these values only */ |
| 69 | if (BITS_IN_JSAMPLE != 8) |
| 70 | return 0; |
| 71 | if (sizeof(JDIMENSION) != 4) |
| 72 | return 0; |
| 73 | if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) |
| 74 | return 0; |
| 75 | |
| 76 | if ((simd_support & JSIMD_SSE2) && |
| 77 | IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2)) |
| 78 | return 1; |
| 79 | if (simd_support & JSIMD_MMX) |
| 80 | return 1; |
| 81 | |
| 82 | return 0; |
| 83 | } |
| 84 | |
| 85 | GLOBAL(int) |
| 86 | jsimd_can_ycc_rgb (void) |
| 87 | { |
| 88 | init_simd(); |
| 89 | |
| 90 | /* The code is optimised for these values only */ |
| 91 | if (BITS_IN_JSAMPLE != 8) |
| 92 | return 0; |
| 93 | if (sizeof(JDIMENSION) != 4) |
| 94 | return 0; |
| 95 | if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) |
| 96 | return 0; |
| 97 | |
| 98 | if ((simd_support & JSIMD_SSE2) && |
| 99 | IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2)) |
| 100 | return 1; |
| 101 | if (simd_support & JSIMD_MMX) |
| 102 | return 1; |
| 103 | |
| 104 | return 0; |
| 105 | } |
| 106 | |
| 107 | GLOBAL(void) |
| 108 | jsimd_rgb_ycc_convert (j_compress_ptr cinfo, |
| 109 | JSAMPARRAY input_buf, JSAMPIMAGE output_buf, |
| 110 | JDIMENSION output_row, int num_rows) |
| 111 | { |
| 112 | void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); |
| 113 | void (*mmxfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); |
| 114 | |
| 115 | switch(cinfo->in_color_space) |
| 116 | { |
| 117 | case JCS_EXT_RGB: |
| 118 | sse2fct=jsimd_extrgb_ycc_convert_sse2; |
| 119 | mmxfct=jsimd_extrgb_ycc_convert_mmx; |
| 120 | break; |
| 121 | case JCS_EXT_RGBX: |
| 122 | sse2fct=jsimd_extrgbx_ycc_convert_sse2; |
| 123 | mmxfct=jsimd_extrgbx_ycc_convert_mmx; |
| 124 | break; |
| 125 | case JCS_EXT_BGR: |
| 126 | sse2fct=jsimd_extbgr_ycc_convert_sse2; |
| 127 | mmxfct=jsimd_extbgr_ycc_convert_mmx; |
| 128 | break; |
| 129 | case JCS_EXT_BGRX: |
| 130 | sse2fct=jsimd_extbgrx_ycc_convert_sse2; |
| 131 | mmxfct=jsimd_extbgrx_ycc_convert_mmx; |
| 132 | break; |
| 133 | case JCS_EXT_XBGR: |
| 134 | sse2fct=jsimd_extxbgr_ycc_convert_sse2; |
| 135 | mmxfct=jsimd_extxbgr_ycc_convert_mmx; |
| 136 | break; |
| 137 | case JCS_EXT_XRGB: |
| 138 | sse2fct=jsimd_extxrgb_ycc_convert_sse2; |
| 139 | mmxfct=jsimd_extxrgb_ycc_convert_mmx; |
| 140 | break; |
| 141 | default: |
| 142 | sse2fct=jsimd_rgb_ycc_convert_sse2; |
| 143 | mmxfct=jsimd_rgb_ycc_convert_mmx; |
| 144 | break; |
| 145 | } |
| 146 | |
| 147 | if ((simd_support & JSIMD_SSE2) && |
| 148 | IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2)) |
| 149 | sse2fct(cinfo->image_width, input_buf, |
| 150 | output_buf, output_row, num_rows); |
| 151 | else if (simd_support & JSIMD_MMX) |
| 152 | mmxfct(cinfo->image_width, input_buf, |
| 153 | output_buf, output_row, num_rows); |
| 154 | } |
| 155 | |
| 156 | GLOBAL(void) |
| 157 | jsimd_ycc_rgb_convert (j_decompress_ptr cinfo, |
| 158 | JSAMPIMAGE input_buf, JDIMENSION input_row, |
| 159 | JSAMPARRAY output_buf, int num_rows) |
| 160 | { |
| 161 | void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); |
| 162 | void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); |
| 163 | |
| 164 | switch(cinfo->out_color_space) |
| 165 | { |
| 166 | case JCS_EXT_RGB: |
| 167 | sse2fct=jsimd_ycc_extrgb_convert_sse2; |
| 168 | mmxfct=jsimd_ycc_extrgb_convert_mmx; |
| 169 | break; |
| 170 | case JCS_EXT_RGBX: |
| 171 | sse2fct=jsimd_ycc_extrgbx_convert_sse2; |
| 172 | mmxfct=jsimd_ycc_extrgbx_convert_mmx; |
| 173 | break; |
| 174 | case JCS_EXT_BGR: |
| 175 | sse2fct=jsimd_ycc_extbgr_convert_sse2; |
| 176 | mmxfct=jsimd_ycc_extbgr_convert_mmx; |
| 177 | break; |
| 178 | case JCS_EXT_BGRX: |
| 179 | sse2fct=jsimd_ycc_extbgrx_convert_sse2; |
| 180 | mmxfct=jsimd_ycc_extbgrx_convert_mmx; |
| 181 | break; |
| 182 | case JCS_EXT_XBGR: |
| 183 | sse2fct=jsimd_ycc_extxbgr_convert_sse2; |
| 184 | mmxfct=jsimd_ycc_extxbgr_convert_mmx; |
| 185 | break; |
| 186 | case JCS_EXT_XRGB: |
| 187 | sse2fct=jsimd_ycc_extxrgb_convert_sse2; |
| 188 | mmxfct=jsimd_ycc_extxrgb_convert_mmx; |
| 189 | break; |
| 190 | default: |
| 191 | sse2fct=jsimd_ycc_rgb_convert_sse2; |
| 192 | mmxfct=jsimd_ycc_rgb_convert_mmx; |
| 193 | break; |
| 194 | } |
| 195 | |
| 196 | if ((simd_support & JSIMD_SSE2) && |
| 197 | IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2)) |
| 198 | sse2fct(cinfo->output_width, input_buf, |
| 199 | input_row, output_buf, num_rows); |
| 200 | else if (simd_support & JSIMD_MMX) |
| 201 | mmxfct(cinfo->output_width, input_buf, |
| 202 | input_row, output_buf, num_rows); |
| 203 | } |
| 204 | |
| 205 | GLOBAL(int) |
| 206 | jsimd_can_h2v2_downsample (void) |
| 207 | { |
| 208 | init_simd(); |
| 209 | |
| 210 | /* The code is optimised for these values only */ |
| 211 | if (BITS_IN_JSAMPLE != 8) |
| 212 | return 0; |
| 213 | if (sizeof(JDIMENSION) != 4) |
| 214 | return 0; |
| 215 | |
| 216 | if (simd_support & JSIMD_SSE2) |
| 217 | return 1; |
| 218 | if (simd_support & JSIMD_MMX) |
| 219 | return 1; |
| 220 | |
| 221 | return 0; |
| 222 | } |
| 223 | |
| 224 | GLOBAL(int) |
| 225 | jsimd_can_h2v1_downsample (void) |
| 226 | { |
| 227 | init_simd(); |
| 228 | |
| 229 | /* The code is optimised for these values only */ |
| 230 | if (BITS_IN_JSAMPLE != 8) |
| 231 | return 0; |
| 232 | if (sizeof(JDIMENSION) != 4) |
| 233 | return 0; |
| 234 | |
| 235 | if (simd_support & JSIMD_SSE2) |
| 236 | return 1; |
| 237 | if (simd_support & JSIMD_MMX) |
| 238 | return 1; |
| 239 | |
| 240 | return 0; |
| 241 | } |
| 242 | |
| 243 | GLOBAL(void) |
| 244 | jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr, |
| 245 | JSAMPARRAY input_data, JSAMPARRAY output_data) |
| 246 | { |
| 247 | if (simd_support & JSIMD_SSE2) |
| 248 | jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor, |
| 249 | compptr->v_samp_factor, compptr->width_in_blocks, |
| 250 | input_data, output_data); |
| 251 | else if (simd_support & JSIMD_MMX) |
| 252 | jsimd_h2v2_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor, |
| 253 | compptr->v_samp_factor, compptr->width_in_blocks, |
| 254 | input_data, output_data); |
| 255 | } |
| 256 | |
| 257 | GLOBAL(void) |
| 258 | jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr, |
| 259 | JSAMPARRAY input_data, JSAMPARRAY output_data) |
| 260 | { |
| 261 | if (simd_support & JSIMD_SSE2) |
| 262 | jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor, |
| 263 | compptr->v_samp_factor, compptr->width_in_blocks, |
| 264 | input_data, output_data); |
| 265 | else if (simd_support & JSIMD_MMX) |
| 266 | jsimd_h2v1_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor, |
| 267 | compptr->v_samp_factor, compptr->width_in_blocks, |
| 268 | input_data, output_data); |
| 269 | } |
| 270 | |
| 271 | GLOBAL(int) |
| 272 | jsimd_can_h2v2_upsample (void) |
| 273 | { |
| 274 | init_simd(); |
| 275 | |
| 276 | /* The code is optimised for these values only */ |
| 277 | if (BITS_IN_JSAMPLE != 8) |
| 278 | return 0; |
| 279 | if (sizeof(JDIMENSION) != 4) |
| 280 | return 0; |
| 281 | |
| 282 | if (simd_support & JSIMD_SSE2) |
| 283 | return 1; |
| 284 | if (simd_support & JSIMD_MMX) |
| 285 | return 1; |
| 286 | |
| 287 | return 0; |
| 288 | } |
| 289 | |
| 290 | GLOBAL(int) |
| 291 | jsimd_can_h2v1_upsample (void) |
| 292 | { |
| 293 | init_simd(); |
| 294 | |
| 295 | /* The code is optimised for these values only */ |
| 296 | if (BITS_IN_JSAMPLE != 8) |
| 297 | return 0; |
| 298 | if (sizeof(JDIMENSION) != 4) |
| 299 | return 0; |
| 300 | |
| 301 | if (simd_support & JSIMD_SSE2) |
| 302 | return 1; |
| 303 | if (simd_support & JSIMD_MMX) |
| 304 | return 1; |
| 305 | |
| 306 | return 0; |
| 307 | } |
| 308 | |
| 309 | GLOBAL(void) |
| 310 | jsimd_h2v2_upsample (j_decompress_ptr cinfo, |
| 311 | jpeg_component_info * compptr, |
| 312 | JSAMPARRAY input_data, |
| 313 | JSAMPARRAY * output_data_ptr) |
| 314 | { |
| 315 | if (simd_support & JSIMD_SSE2) |
| 316 | jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, |
| 317 | cinfo->output_width, input_data, output_data_ptr); |
| 318 | else if (simd_support & JSIMD_MMX) |
| 319 | jsimd_h2v2_upsample_mmx(cinfo->max_v_samp_factor, |
| 320 | cinfo->output_width, input_data, output_data_ptr); |
| 321 | } |
| 322 | |
| 323 | GLOBAL(void) |
| 324 | jsimd_h2v1_upsample (j_decompress_ptr cinfo, |
| 325 | jpeg_component_info * compptr, |
| 326 | JSAMPARRAY input_data, |
| 327 | JSAMPARRAY * output_data_ptr) |
| 328 | { |
| 329 | if (simd_support & JSIMD_SSE2) |
| 330 | jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, |
| 331 | cinfo->output_width, input_data, output_data_ptr); |
| 332 | else if (simd_support & JSIMD_MMX) |
| 333 | jsimd_h2v1_upsample_mmx(cinfo->max_v_samp_factor, |
| 334 | cinfo->output_width, input_data, output_data_ptr); |
| 335 | } |
| 336 | |
| 337 | GLOBAL(int) |
| 338 | jsimd_can_h2v2_fancy_upsample (void) |
| 339 | { |
| 340 | init_simd(); |
| 341 | |
| 342 | /* The code is optimised for these values only */ |
| 343 | if (BITS_IN_JSAMPLE != 8) |
| 344 | return 0; |
| 345 | if (sizeof(JDIMENSION) != 4) |
| 346 | return 0; |
| 347 | |
| 348 | if ((simd_support & JSIMD_SSE2) && |
| 349 | IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) |
| 350 | return 1; |
| 351 | if (simd_support & JSIMD_MMX) |
| 352 | return 1; |
| 353 | |
| 354 | return 0; |
| 355 | } |
| 356 | |
| 357 | GLOBAL(int) |
| 358 | jsimd_can_h2v1_fancy_upsample (void) |
| 359 | { |
| 360 | init_simd(); |
| 361 | |
| 362 | /* The code is optimised for these values only */ |
| 363 | if (BITS_IN_JSAMPLE != 8) |
| 364 | return 0; |
| 365 | if (sizeof(JDIMENSION) != 4) |
| 366 | return 0; |
| 367 | |
| 368 | if ((simd_support & JSIMD_SSE2) && |
| 369 | IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) |
| 370 | return 1; |
| 371 | if (simd_support & JSIMD_MMX) |
| 372 | return 1; |
| 373 | |
| 374 | return 0; |
| 375 | } |
| 376 | |
| 377 | GLOBAL(void) |
| 378 | jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo, |
| 379 | jpeg_component_info * compptr, |
| 380 | JSAMPARRAY input_data, |
| 381 | JSAMPARRAY * output_data_ptr) |
| 382 | { |
| 383 | if ((simd_support & JSIMD_SSE2) && |
| 384 | IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) |
| 385 | jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor, |
| 386 | compptr->downsampled_width, input_data, output_data_ptr); |
| 387 | else if (simd_support & JSIMD_MMX) |
| 388 | jsimd_h2v2_fancy_upsample_mmx(cinfo->max_v_samp_factor, |
| 389 | compptr->downsampled_width, input_data, output_data_ptr); |
| 390 | } |
| 391 | |
| 392 | GLOBAL(void) |
| 393 | jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo, |
| 394 | jpeg_component_info * compptr, |
| 395 | JSAMPARRAY input_data, |
| 396 | JSAMPARRAY * output_data_ptr) |
| 397 | { |
| 398 | if ((simd_support & JSIMD_SSE2) && |
| 399 | IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) |
| 400 | jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor, |
| 401 | compptr->downsampled_width, input_data, output_data_ptr); |
| 402 | else if (simd_support & JSIMD_MMX) |
| 403 | jsimd_h2v1_fancy_upsample_mmx(cinfo->max_v_samp_factor, |
| 404 | compptr->downsampled_width, input_data, output_data_ptr); |
| 405 | } |
| 406 | |
| 407 | GLOBAL(int) |
| 408 | jsimd_can_h2v2_merged_upsample (void) |
| 409 | { |
| 410 | init_simd(); |
| 411 | |
| 412 | /* The code is optimised for these values only */ |
| 413 | if (BITS_IN_JSAMPLE != 8) |
| 414 | return 0; |
| 415 | if (sizeof(JDIMENSION) != 4) |
| 416 | return 0; |
| 417 | |
| 418 | if ((simd_support & JSIMD_SSE2) && |
| 419 | IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) |
| 420 | return 1; |
| 421 | if (simd_support & JSIMD_MMX) |
| 422 | return 1; |
| 423 | |
| 424 | return 0; |
| 425 | } |
| 426 | |
| 427 | GLOBAL(int) |
| 428 | jsimd_can_h2v1_merged_upsample (void) |
| 429 | { |
| 430 | init_simd(); |
| 431 | |
| 432 | /* The code is optimised for these values only */ |
| 433 | if (BITS_IN_JSAMPLE != 8) |
| 434 | return 0; |
| 435 | if (sizeof(JDIMENSION) != 4) |
| 436 | return 0; |
| 437 | |
| 438 | if ((simd_support & JSIMD_SSE2) && |
| 439 | IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) |
| 440 | return 1; |
| 441 | if (simd_support & JSIMD_MMX) |
| 442 | return 1; |
| 443 | |
| 444 | return 0; |
| 445 | } |
| 446 | |
| 447 | GLOBAL(void) |
| 448 | jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo, |
| 449 | JSAMPIMAGE input_buf, |
| 450 | JDIMENSION in_row_group_ctr, |
| 451 | JSAMPARRAY output_buf) |
| 452 | { |
| 453 | void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); |
| 454 | void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); |
| 455 | |
| 456 | switch(cinfo->out_color_space) |
| 457 | { |
| 458 | case JCS_EXT_RGB: |
| 459 | sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2; |
| 460 | mmxfct=jsimd_h2v2_extrgb_merged_upsample_mmx; |
| 461 | break; |
| 462 | case JCS_EXT_RGBX: |
| 463 | sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2; |
| 464 | mmxfct=jsimd_h2v2_extrgbx_merged_upsample_mmx; |
| 465 | break; |
| 466 | case JCS_EXT_BGR: |
| 467 | sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2; |
| 468 | mmxfct=jsimd_h2v2_extbgr_merged_upsample_mmx; |
| 469 | break; |
| 470 | case JCS_EXT_BGRX: |
| 471 | sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2; |
| 472 | mmxfct=jsimd_h2v2_extbgrx_merged_upsample_mmx; |
| 473 | break; |
| 474 | case JCS_EXT_XBGR: |
| 475 | sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2; |
| 476 | mmxfct=jsimd_h2v2_extxbgr_merged_upsample_mmx; |
| 477 | break; |
| 478 | case JCS_EXT_XRGB: |
| 479 | sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2; |
| 480 | mmxfct=jsimd_h2v2_extxrgb_merged_upsample_mmx; |
| 481 | break; |
| 482 | default: |
| 483 | sse2fct=jsimd_h2v2_merged_upsample_sse2; |
| 484 | mmxfct=jsimd_h2v2_merged_upsample_mmx; |
| 485 | break; |
| 486 | } |
| 487 | |
| 488 | if ((simd_support & JSIMD_SSE2) && |
| 489 | IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) |
| 490 | sse2fct(cinfo->output_width, input_buf, |
| 491 | in_row_group_ctr, output_buf); |
| 492 | else if (simd_support & JSIMD_MMX) |
| 493 | mmxfct(cinfo->output_width, input_buf, |
| 494 | in_row_group_ctr, output_buf); |
| 495 | } |
| 496 | |
| 497 | GLOBAL(void) |
| 498 | jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo, |
| 499 | JSAMPIMAGE input_buf, |
| 500 | JDIMENSION in_row_group_ctr, |
| 501 | JSAMPARRAY output_buf) |
| 502 | { |
| 503 | void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); |
| 504 | void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); |
| 505 | |
| 506 | switch(cinfo->out_color_space) |
| 507 | { |
| 508 | case JCS_EXT_RGB: |
| 509 | sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2; |
| 510 | mmxfct=jsimd_h2v1_extrgb_merged_upsample_mmx; |
| 511 | break; |
| 512 | case JCS_EXT_RGBX: |
| 513 | sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2; |
| 514 | mmxfct=jsimd_h2v1_extrgbx_merged_upsample_mmx; |
| 515 | break; |
| 516 | case JCS_EXT_BGR: |
| 517 | sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2; |
| 518 | mmxfct=jsimd_h2v1_extbgr_merged_upsample_mmx; |
| 519 | break; |
| 520 | case JCS_EXT_BGRX: |
| 521 | sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2; |
| 522 | mmxfct=jsimd_h2v1_extbgrx_merged_upsample_mmx; |
| 523 | break; |
| 524 | case JCS_EXT_XBGR: |
| 525 | sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2; |
| 526 | mmxfct=jsimd_h2v1_extxbgr_merged_upsample_mmx; |
| 527 | break; |
| 528 | case JCS_EXT_XRGB: |
| 529 | sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2; |
| 530 | mmxfct=jsimd_h2v1_extxrgb_merged_upsample_mmx; |
| 531 | break; |
| 532 | default: |
| 533 | sse2fct=jsimd_h2v1_merged_upsample_sse2; |
| 534 | mmxfct=jsimd_h2v1_merged_upsample_mmx; |
| 535 | break; |
| 536 | } |
| 537 | |
| 538 | if ((simd_support & JSIMD_SSE2) && |
| 539 | IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) |
| 540 | sse2fct(cinfo->output_width, input_buf, |
| 541 | in_row_group_ctr, output_buf); |
| 542 | else if (simd_support & JSIMD_MMX) |
| 543 | mmxfct(cinfo->output_width, input_buf, |
| 544 | in_row_group_ctr, output_buf); |
| 545 | } |
| 546 | |
| 547 | GLOBAL(int) |
| 548 | jsimd_can_convsamp (void) |
| 549 | { |
| 550 | init_simd(); |
| 551 | |
| 552 | /* The code is optimised for these values only */ |
| 553 | if (DCTSIZE != 8) |
| 554 | return 0; |
| 555 | if (BITS_IN_JSAMPLE != 8) |
| 556 | return 0; |
| 557 | if (sizeof(JDIMENSION) != 4) |
| 558 | return 0; |
| 559 | if (sizeof(DCTELEM) != 2) |
| 560 | return 0; |
| 561 | |
| 562 | if (simd_support & JSIMD_SSE2) |
| 563 | return 1; |
| 564 | if (simd_support & JSIMD_MMX) |
| 565 | return 1; |
| 566 | |
| 567 | return 0; |
| 568 | } |
| 569 | |
| 570 | GLOBAL(int) |
| 571 | jsimd_can_convsamp_float (void) |
| 572 | { |
| 573 | init_simd(); |
| 574 | |
| 575 | /* The code is optimised for these values only */ |
| 576 | if (DCTSIZE != 8) |
| 577 | return 0; |
| 578 | if (BITS_IN_JSAMPLE != 8) |
| 579 | return 0; |
| 580 | if (sizeof(JDIMENSION) != 4) |
| 581 | return 0; |
| 582 | if (sizeof(FAST_FLOAT) != 4) |
| 583 | return 0; |
| 584 | |
| 585 | if (simd_support & JSIMD_SSE2) |
| 586 | return 1; |
| 587 | if (simd_support & JSIMD_SSE) |
| 588 | return 1; |
| 589 | if (simd_support & JSIMD_3DNOW) |
| 590 | return 1; |
| 591 | |
| 592 | return 0; |
| 593 | } |
| 594 | |
| 595 | GLOBAL(void) |
| 596 | jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col, |
| 597 | DCTELEM * workspace) |
| 598 | { |
| 599 | if (simd_support & JSIMD_SSE2) |
| 600 | jsimd_convsamp_sse2(sample_data, start_col, workspace); |
| 601 | else if (simd_support & JSIMD_MMX) |
| 602 | jsimd_convsamp_mmx(sample_data, start_col, workspace); |
| 603 | } |
| 604 | |
| 605 | GLOBAL(void) |
| 606 | jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col, |
| 607 | FAST_FLOAT * workspace) |
| 608 | { |
| 609 | if (simd_support & JSIMD_SSE2) |
| 610 | jsimd_convsamp_float_sse2(sample_data, start_col, workspace); |
| 611 | else if (simd_support & JSIMD_SSE) |
| 612 | jsimd_convsamp_float_sse(sample_data, start_col, workspace); |
| 613 | else if (simd_support & JSIMD_3DNOW) |
| 614 | jsimd_convsamp_float_3dnow(sample_data, start_col, workspace); |
| 615 | } |
| 616 | |
| 617 | GLOBAL(int) |
| 618 | jsimd_can_fdct_islow (void) |
| 619 | { |
| 620 | init_simd(); |
| 621 | |
| 622 | /* The code is optimised for these values only */ |
| 623 | if (DCTSIZE != 8) |
| 624 | return 0; |
| 625 | if (sizeof(DCTELEM) != 2) |
| 626 | return 0; |
| 627 | |
| 628 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2)) |
| 629 | return 1; |
| 630 | if (simd_support & JSIMD_MMX) |
| 631 | return 1; |
| 632 | |
| 633 | return 0; |
| 634 | } |
| 635 | |
| 636 | GLOBAL(int) |
| 637 | jsimd_can_fdct_ifast (void) |
| 638 | { |
| 639 | init_simd(); |
| 640 | |
| 641 | /* The code is optimised for these values only */ |
| 642 | if (DCTSIZE != 8) |
| 643 | return 0; |
| 644 | if (sizeof(DCTELEM) != 2) |
| 645 | return 0; |
| 646 | |
| 647 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_ifast_sse2)) |
| 648 | return 1; |
| 649 | if (simd_support & JSIMD_MMX) |
| 650 | return 1; |
| 651 | |
| 652 | return 0; |
| 653 | } |
| 654 | |
| 655 | GLOBAL(int) |
| 656 | jsimd_can_fdct_float (void) |
| 657 | { |
| 658 | init_simd(); |
| 659 | |
| 660 | /* The code is optimised for these values only */ |
| 661 | if (DCTSIZE != 8) |
| 662 | return 0; |
| 663 | if (sizeof(FAST_FLOAT) != 4) |
| 664 | return 0; |
| 665 | |
| 666 | if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse)) |
| 667 | return 1; |
| 668 | if (simd_support & JSIMD_3DNOW) |
| 669 | return 1; |
| 670 | |
| 671 | return 0; |
| 672 | } |
| 673 | |
| 674 | GLOBAL(void) |
| 675 | jsimd_fdct_islow (DCTELEM * data) |
| 676 | { |
| 677 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2)) |
| 678 | jsimd_fdct_islow_sse2(data); |
| 679 | else if (simd_support & JSIMD_MMX) |
| 680 | jsimd_fdct_islow_mmx(data); |
| 681 | } |
| 682 | |
| 683 | GLOBAL(void) |
| 684 | jsimd_fdct_ifast (DCTELEM * data) |
| 685 | { |
| 686 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2)) |
| 687 | jsimd_fdct_ifast_sse2(data); |
| 688 | else if (simd_support & JSIMD_MMX) |
| 689 | jsimd_fdct_ifast_mmx(data); |
| 690 | } |
| 691 | |
| 692 | GLOBAL(void) |
| 693 | jsimd_fdct_float (FAST_FLOAT * data) |
| 694 | { |
| 695 | if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse)) |
| 696 | jsimd_fdct_float_sse(data); |
| 697 | else if (simd_support & JSIMD_3DNOW) |
| 698 | jsimd_fdct_float_3dnow(data); |
| 699 | } |
| 700 | |
| 701 | GLOBAL(int) |
| 702 | jsimd_can_quantize (void) |
| 703 | { |
| 704 | init_simd(); |
| 705 | |
| 706 | /* The code is optimised for these values only */ |
| 707 | if (DCTSIZE != 8) |
| 708 | return 0; |
| 709 | if (sizeof(JCOEF) != 2) |
| 710 | return 0; |
| 711 | if (sizeof(DCTELEM) != 2) |
| 712 | return 0; |
| 713 | |
| 714 | if (simd_support & JSIMD_SSE2) |
| 715 | return 1; |
| 716 | if (simd_support & JSIMD_MMX) |
| 717 | return 1; |
| 718 | |
| 719 | return 0; |
| 720 | } |
| 721 | |
| 722 | GLOBAL(int) |
| 723 | jsimd_can_quantize_float (void) |
| 724 | { |
| 725 | init_simd(); |
| 726 | |
| 727 | /* The code is optimised for these values only */ |
| 728 | if (DCTSIZE != 8) |
| 729 | return 0; |
| 730 | if (sizeof(JCOEF) != 2) |
| 731 | return 0; |
| 732 | if (sizeof(FAST_FLOAT) != 4) |
| 733 | return 0; |
| 734 | |
| 735 | if (simd_support & JSIMD_SSE2) |
| 736 | return 1; |
| 737 | if (simd_support & JSIMD_SSE) |
| 738 | return 1; |
| 739 | if (simd_support & JSIMD_3DNOW) |
| 740 | return 1; |
| 741 | |
| 742 | return 0; |
| 743 | } |
| 744 | |
| 745 | GLOBAL(void) |
| 746 | jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors, |
| 747 | DCTELEM * workspace) |
| 748 | { |
| 749 | if (simd_support & JSIMD_SSE2) |
| 750 | jsimd_quantize_sse2(coef_block, divisors, workspace); |
| 751 | else if (simd_support & JSIMD_MMX) |
| 752 | jsimd_quantize_mmx(coef_block, divisors, workspace); |
| 753 | } |
| 754 | |
| 755 | GLOBAL(void) |
| 756 | jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors, |
| 757 | FAST_FLOAT * workspace) |
| 758 | { |
| 759 | if (simd_support & JSIMD_SSE2) |
| 760 | jsimd_quantize_float_sse2(coef_block, divisors, workspace); |
| 761 | else if (simd_support & JSIMD_SSE) |
| 762 | jsimd_quantize_float_sse(coef_block, divisors, workspace); |
| 763 | else if (simd_support & JSIMD_3DNOW) |
| 764 | jsimd_quantize_float_3dnow(coef_block, divisors, workspace); |
| 765 | } |
| 766 | |
| 767 | GLOBAL(int) |
| 768 | jsimd_can_idct_2x2 (void) |
| 769 | { |
| 770 | init_simd(); |
| 771 | |
| 772 | /* The code is optimised for these values only */ |
| 773 | if (DCTSIZE != 8) |
| 774 | return 0; |
| 775 | if (sizeof(JCOEF) != 2) |
| 776 | return 0; |
| 777 | if (BITS_IN_JSAMPLE != 8) |
| 778 | return 0; |
| 779 | if (sizeof(JDIMENSION) != 4) |
| 780 | return 0; |
| 781 | if (sizeof(ISLOW_MULT_TYPE) != 2) |
| 782 | return 0; |
| 783 | |
| 784 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) |
| 785 | return 1; |
| 786 | if (simd_support & JSIMD_MMX) |
| 787 | return 1; |
| 788 | |
| 789 | return 0; |
| 790 | } |
| 791 | |
| 792 | GLOBAL(int) |
| 793 | jsimd_can_idct_4x4 (void) |
| 794 | { |
| 795 | init_simd(); |
| 796 | |
| 797 | /* The code is optimised for these values only */ |
| 798 | if (DCTSIZE != 8) |
| 799 | return 0; |
| 800 | if (sizeof(JCOEF) != 2) |
| 801 | return 0; |
| 802 | if (BITS_IN_JSAMPLE != 8) |
| 803 | return 0; |
| 804 | if (sizeof(JDIMENSION) != 4) |
| 805 | return 0; |
| 806 | if (sizeof(ISLOW_MULT_TYPE) != 2) |
| 807 | return 0; |
| 808 | |
| 809 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) |
| 810 | return 1; |
| 811 | if (simd_support & JSIMD_MMX) |
| 812 | return 1; |
| 813 | |
| 814 | return 0; |
| 815 | } |
| 816 | |
| 817 | GLOBAL(void) |
| 818 | jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| 819 | JCOEFPTR coef_block, JSAMPARRAY output_buf, |
| 820 | JDIMENSION output_col) |
| 821 | { |
| 822 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) |
| 823 | jsimd_idct_2x2_sse2(compptr->dct_table, coef_block, output_buf, output_col); |
| 824 | else if (simd_support & JSIMD_MMX) |
| 825 | jsimd_idct_2x2_mmx(compptr->dct_table, coef_block, output_buf, output_col); |
| 826 | } |
| 827 | |
| 828 | GLOBAL(void) |
| 829 | jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| 830 | JCOEFPTR coef_block, JSAMPARRAY output_buf, |
| 831 | JDIMENSION output_col) |
| 832 | { |
| 833 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) |
| 834 | jsimd_idct_4x4_sse2(compptr->dct_table, coef_block, output_buf, output_col); |
| 835 | else if (simd_support & JSIMD_MMX) |
| 836 | jsimd_idct_4x4_mmx(compptr->dct_table, coef_block, output_buf, output_col); |
| 837 | } |
| 838 | |
| 839 | GLOBAL(int) |
| 840 | jsimd_can_idct_islow (void) |
| 841 | { |
| 842 | init_simd(); |
| 843 | |
| 844 | /* The code is optimised for these values only */ |
| 845 | if (DCTSIZE != 8) |
| 846 | return 0; |
| 847 | if (sizeof(JCOEF) != 2) |
| 848 | return 0; |
| 849 | if (BITS_IN_JSAMPLE != 8) |
| 850 | return 0; |
| 851 | if (sizeof(JDIMENSION) != 4) |
| 852 | return 0; |
| 853 | if (sizeof(ISLOW_MULT_TYPE) != 2) |
| 854 | return 0; |
| 855 | |
| 856 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2)) |
| 857 | return 1; |
| 858 | if (simd_support & JSIMD_MMX) |
| 859 | return 1; |
| 860 | |
| 861 | return 0; |
| 862 | } |
| 863 | |
| 864 | GLOBAL(int) |
| 865 | jsimd_can_idct_ifast (void) |
| 866 | { |
| 867 | init_simd(); |
| 868 | |
| 869 | /* The code is optimised for these values only */ |
| 870 | if (DCTSIZE != 8) |
| 871 | return 0; |
| 872 | if (sizeof(JCOEF) != 2) |
| 873 | return 0; |
| 874 | if (BITS_IN_JSAMPLE != 8) |
| 875 | return 0; |
| 876 | if (sizeof(JDIMENSION) != 4) |
| 877 | return 0; |
| 878 | if (sizeof(IFAST_MULT_TYPE) != 2) |
| 879 | return 0; |
| 880 | if (IFAST_SCALE_BITS != 2) |
| 881 | return 0; |
| 882 | |
| 883 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2)) |
| 884 | return 1; |
| 885 | if (simd_support & JSIMD_MMX) |
| 886 | return 1; |
| 887 | |
| 888 | return 0; |
| 889 | } |
| 890 | |
| 891 | GLOBAL(int) |
| 892 | jsimd_can_idct_float (void) |
| 893 | { |
| 894 | init_simd(); |
| 895 | |
| 896 | if (DCTSIZE != 8) |
| 897 | return 0; |
| 898 | if (sizeof(JCOEF) != 2) |
| 899 | return 0; |
| 900 | if (BITS_IN_JSAMPLE != 8) |
| 901 | return 0; |
| 902 | if (sizeof(JDIMENSION) != 4) |
| 903 | return 0; |
| 904 | if (sizeof(FAST_FLOAT) != 4) |
| 905 | return 0; |
| 906 | if (sizeof(FLOAT_MULT_TYPE) != 4) |
| 907 | return 0; |
| 908 | |
| 909 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2)) |
| 910 | return 1; |
| 911 | if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse)) |
| 912 | return 1; |
| 913 | if (simd_support & JSIMD_3DNOW) |
| 914 | return 1; |
| 915 | |
| 916 | return 0; |
| 917 | } |
| 918 | |
| 919 | GLOBAL(void) |
| 920 | jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| 921 | JCOEFPTR coef_block, JSAMPARRAY output_buf, |
| 922 | JDIMENSION output_col) |
| 923 | { |
| 924 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2)) |
| 925 | jsimd_idct_islow_sse2(compptr->dct_table, coef_block, output_buf, output_col); |
| 926 | else if (simd_support & JSIMD_MMX) |
| 927 | jsimd_idct_islow_mmx(compptr->dct_table, coef_block, output_buf, output_col); |
| 928 | } |
| 929 | |
| 930 | GLOBAL(void) |
| 931 | jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| 932 | JCOEFPTR coef_block, JSAMPARRAY output_buf, |
| 933 | JDIMENSION output_col) |
| 934 | { |
| 935 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2)) |
| 936 | jsimd_idct_ifast_sse2(compptr->dct_table, coef_block, output_buf, output_col); |
| 937 | else if (simd_support & JSIMD_MMX) |
| 938 | jsimd_idct_ifast_mmx(compptr->dct_table, coef_block, output_buf, output_col); |
| 939 | } |
| 940 | |
| 941 | GLOBAL(void) |
| 942 | jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| 943 | JCOEFPTR coef_block, JSAMPARRAY output_buf, |
| 944 | JDIMENSION output_col) |
| 945 | { |
| 946 | if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2)) |
| 947 | jsimd_idct_float_sse2(compptr->dct_table, coef_block, |
| 948 | output_buf, output_col); |
| 949 | else if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse)) |
| 950 | jsimd_idct_float_sse(compptr->dct_table, coef_block, |
| 951 | output_buf, output_col); |
| 952 | else if (simd_support & JSIMD_3DNOW) |
| 953 | jsimd_idct_float_3dnow(compptr->dct_table, coef_block, |
| 954 | output_buf, output_col); |
| 955 | } |
| 956 | |