xorg_exa_tgsi.c revision 2947d1420270476730711892909c3683bb6c5bff
1#include "xorg_exa_tgsi.h" 2 3/*### stupidity defined in X11/extensions/XI.h */ 4#undef Absolute 5 6#include "pipe/p_format.h" 7#include "pipe/p_context.h" 8#include "pipe/p_state.h" 9#include "pipe/p_inlines.h" 10#include "pipe/p_shader_tokens.h" 11 12#include "util/u_memory.h" 13#include "util/u_simple_shaders.h" 14 15#include "tgsi/tgsi_ureg.h" 16 17#include "cso_cache/cso_context.h" 18#include "cso_cache/cso_hash.h" 19 20/* Vertex shader: 21 * IN[0] = vertex pos 22 * IN[1] = src tex coord | solid fill color 23 * IN[2] = mask tex coord 24 * IN[3] = dst tex coord 25 * CONST[0] = (2/dst_width, 2/dst_height, 1, 1) 26 * CONST[1] = (-1, -1, 0, 0) 27 * 28 * OUT[0] = vertex pos 29 * OUT[1] = src tex coord | solid fill color 30 * OUT[2] = mask tex coord 31 * OUT[3] = dst tex coord 32 */ 33 34/* Fragment shader: 35 * SAMP[0] = src 36 * SAMP[1] = mask 37 * SAMP[2] = dst 38 * IN[0] = pos src | solid fill color 39 * IN[1] = pos mask 40 * IN[2] = pos dst 41 * CONST[0] = (0, 0, 0, 1) 42 * 43 * OUT[0] = color 44 */ 45 46struct xorg_shaders { 47 struct xorg_renderer *r; 48 49 struct cso_hash *vs_hash; 50 struct cso_hash *fs_hash; 51}; 52 53static INLINE void 54src_in_mask(struct ureg_program *ureg, 55 struct ureg_dst dst, 56 struct ureg_src src, 57 struct ureg_src mask, 58 int component_alpha) 59{ 60 if (component_alpha == FS_CA_FULL) { 61 ureg_MUL(ureg, dst, src, mask); 62 } else if (component_alpha == FS_CA_SRCALPHA) { 63 ureg_MUL(ureg, dst, 64 ureg_scalar(src, TGSI_SWIZZLE_W), mask); 65 } 66 else { 67 ureg_MUL(ureg, dst, src, 68 ureg_scalar(mask, TGSI_SWIZZLE_X)); 69 } 70} 71 72static struct ureg_src 73vs_normalize_coords(struct ureg_program *ureg, struct ureg_src coords, 74 struct ureg_src const0, struct ureg_src const1) 75{ 76 struct ureg_dst tmp = ureg_DECL_temporary(ureg); 77 struct ureg_src ret; 78 ureg_MAD(ureg, tmp, coords, const0, const1); 79 ret = ureg_src(tmp); 80 ureg_release_temporary(ureg, tmp); 81 return ret; 82} 83 84static void 85linear_gradient(struct ureg_program *ureg, 86 struct ureg_dst out, 87 struct ureg_src pos, 88 struct ureg_src sampler, 89 struct ureg_src coords, 90 struct ureg_src const0124, 91 struct ureg_src matrow0, 92 struct ureg_src matrow1, 93 struct ureg_src matrow2) 94{ 95 struct ureg_dst temp0 = ureg_DECL_temporary(ureg); 96 struct ureg_dst temp1 = ureg_DECL_temporary(ureg); 97 struct ureg_dst temp2 = ureg_DECL_temporary(ureg); 98 struct ureg_dst temp3 = ureg_DECL_temporary(ureg); 99 struct ureg_dst temp4 = ureg_DECL_temporary(ureg); 100 struct ureg_dst temp5 = ureg_DECL_temporary(ureg); 101 102 ureg_MOV(ureg, 103 ureg_writemask(temp0, TGSI_WRITEMASK_XY), pos); 104 ureg_MOV(ureg, 105 ureg_writemask(temp0, TGSI_WRITEMASK_Z), 106 ureg_scalar(const0124, TGSI_SWIZZLE_Y)); 107 108 ureg_DP3(ureg, temp1, matrow0, ureg_src(temp0)); 109 ureg_DP3(ureg, temp2, matrow1, ureg_src(temp0)); 110 ureg_DP3(ureg, temp3, matrow2, ureg_src(temp0)); 111 ureg_RCP(ureg, temp3, ureg_src(temp3)); 112 ureg_MUL(ureg, temp1, ureg_src(temp1), ureg_src(temp3)); 113 ureg_MUL(ureg, temp2, ureg_src(temp2), ureg_src(temp3)); 114 115 ureg_MOV(ureg, ureg_writemask(temp4, TGSI_WRITEMASK_X), 116 ureg_src(temp1)); 117 ureg_MOV(ureg, ureg_writemask(temp4, TGSI_WRITEMASK_Y), 118 ureg_src(temp2)); 119 120 ureg_MUL(ureg, temp0, 121 ureg_scalar(coords, TGSI_SWIZZLE_Y), 122 ureg_scalar(ureg_src(temp4), TGSI_SWIZZLE_Y)); 123 ureg_MAD(ureg, temp1, 124 ureg_scalar(coords, TGSI_SWIZZLE_X), 125 ureg_scalar(ureg_src(temp4), TGSI_SWIZZLE_X), 126 ureg_src(temp0)); 127 128 ureg_MUL(ureg, temp2, 129 ureg_src(temp1), 130 ureg_scalar(coords, TGSI_SWIZZLE_Z)); 131 132 ureg_TEX(ureg, out, 133 TGSI_TEXTURE_1D, ureg_src(temp2), sampler); 134 135 ureg_release_temporary(ureg, temp0); 136 ureg_release_temporary(ureg, temp1); 137 ureg_release_temporary(ureg, temp2); 138 ureg_release_temporary(ureg, temp3); 139 ureg_release_temporary(ureg, temp4); 140 ureg_release_temporary(ureg, temp5); 141} 142 143 144static void 145radial_gradient(struct ureg_program *ureg, 146 struct ureg_dst out, 147 struct ureg_src pos, 148 struct ureg_src sampler, 149 struct ureg_src coords, 150 struct ureg_src const0124, 151 struct ureg_src matrow0, 152 struct ureg_src matrow1, 153 struct ureg_src matrow2) 154{ 155 struct ureg_dst temp0 = ureg_DECL_temporary(ureg); 156 struct ureg_dst temp1 = ureg_DECL_temporary(ureg); 157 struct ureg_dst temp2 = ureg_DECL_temporary(ureg); 158 struct ureg_dst temp3 = ureg_DECL_temporary(ureg); 159 struct ureg_dst temp4 = ureg_DECL_temporary(ureg); 160 struct ureg_dst temp5 = ureg_DECL_temporary(ureg); 161 162 ureg_MOV(ureg, 163 ureg_writemask(temp0, TGSI_WRITEMASK_XY), 164 pos); 165 ureg_MOV(ureg, 166 ureg_writemask(temp0, TGSI_WRITEMASK_Z), 167 ureg_scalar(const0124, TGSI_SWIZZLE_Y)); 168 169 ureg_DP3(ureg, temp1, matrow0, ureg_src(temp0)); 170 ureg_DP3(ureg, temp2, matrow1, ureg_src(temp0)); 171 ureg_DP3(ureg, temp3, matrow2, ureg_src(temp0)); 172 ureg_RCP(ureg, temp3, ureg_src(temp3)); 173 ureg_MUL(ureg, temp1, ureg_src(temp1), ureg_src(temp3)); 174 ureg_MUL(ureg, temp2, ureg_src(temp2), ureg_src(temp3)); 175 176 ureg_MOV(ureg, ureg_writemask(temp5, TGSI_WRITEMASK_X), 177 ureg_src(temp1)); 178 ureg_MOV(ureg, ureg_writemask(temp5, TGSI_WRITEMASK_Y), 179 ureg_src(temp2)); 180 181 ureg_MUL(ureg, temp0, ureg_scalar(coords, TGSI_SWIZZLE_Y), 182 ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y)); 183 ureg_MAD(ureg, temp1, 184 ureg_scalar(coords, TGSI_SWIZZLE_X), 185 ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), 186 ureg_src(temp0)); 187 ureg_ADD(ureg, temp1, 188 ureg_src(temp1), ureg_src(temp1)); 189 ureg_MUL(ureg, temp3, 190 ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y), 191 ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y)); 192 ureg_MAD(ureg, temp4, 193 ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), 194 ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), 195 ureg_src(temp3)); 196 ureg_MOV(ureg, temp4, ureg_negate(ureg_src(temp4))); 197 ureg_MUL(ureg, temp2, 198 ureg_scalar(coords, TGSI_SWIZZLE_Z), 199 ureg_src(temp4)); 200 ureg_MUL(ureg, temp0, 201 ureg_scalar(const0124, TGSI_SWIZZLE_W), 202 ureg_src(temp2)); 203 ureg_MUL(ureg, temp3, 204 ureg_src(temp1), ureg_src(temp1)); 205 ureg_SUB(ureg, temp2, 206 ureg_src(temp3), ureg_src(temp0)); 207 ureg_RSQ(ureg, temp2, ureg_abs(ureg_src(temp2))); 208 ureg_RCP(ureg, temp2, ureg_src(temp2)); 209 ureg_SUB(ureg, temp1, 210 ureg_src(temp2), ureg_src(temp1)); 211 ureg_ADD(ureg, temp0, 212 ureg_scalar(coords, TGSI_SWIZZLE_Z), 213 ureg_scalar(coords, TGSI_SWIZZLE_Z)); 214 ureg_RCP(ureg, temp0, ureg_src(temp0)); 215 ureg_MUL(ureg, temp2, 216 ureg_src(temp1), ureg_src(temp0)); 217 ureg_TEX(ureg, out, TGSI_TEXTURE_1D, 218 ureg_src(temp2), sampler); 219 220 ureg_release_temporary(ureg, temp0); 221 ureg_release_temporary(ureg, temp1); 222 ureg_release_temporary(ureg, temp2); 223 ureg_release_temporary(ureg, temp3); 224 ureg_release_temporary(ureg, temp4); 225 ureg_release_temporary(ureg, temp5); 226} 227 228static void * 229create_vs(struct pipe_context *pipe, 230 unsigned vs_traits) 231{ 232 struct ureg_program *ureg; 233 struct ureg_src src; 234 struct ureg_dst dst; 235 struct ureg_src const0, const1; 236 boolean is_fill = vs_traits & VS_FILL; 237 boolean is_composite = vs_traits & VS_COMPOSITE; 238 boolean has_mask = vs_traits & VS_MASK; 239 boolean is_yuv = vs_traits & VS_YUV; 240 unsigned input_slot = 0; 241 242 ureg = ureg_create(TGSI_PROCESSOR_VERTEX); 243 if (ureg == NULL) 244 return 0; 245 246 const0 = ureg_DECL_constant(ureg, 0); 247 const1 = ureg_DECL_constant(ureg, 1); 248 249 /* it has to be either a fill or a composite op */ 250 debug_assert((is_fill ^ is_composite) ^ is_yuv); 251 252 src = ureg_DECL_vs_input(ureg, input_slot++); 253 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); 254 src = vs_normalize_coords(ureg, src, 255 const0, const1); 256 ureg_MOV(ureg, dst, src); 257 258 if (is_yuv) { 259 src = ureg_DECL_vs_input(ureg, input_slot++); 260 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 0); 261 ureg_MOV(ureg, dst, src); 262 } 263 264 if (is_composite) { 265 src = ureg_DECL_vs_input(ureg, input_slot++); 266 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 0); 267 ureg_MOV(ureg, dst, src); 268 } 269 270 if (is_fill) { 271 src = ureg_DECL_vs_input(ureg, input_slot++); 272 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0); 273 ureg_MOV(ureg, dst, src); 274 } 275 276 if (has_mask) { 277 src = ureg_DECL_vs_input(ureg, input_slot++); 278 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 1); 279 ureg_MOV(ureg, dst, src); 280 } 281 282 ureg_END(ureg); 283 284 return ureg_create_shader_and_destroy(ureg, pipe); 285} 286 287static void * 288create_yuv_shader(struct pipe_context *pipe, struct ureg_program *ureg) 289{ 290 struct ureg_src y_sampler, u_sampler, v_sampler; 291 struct ureg_src pos; 292 struct ureg_src matrow0, matrow1, matrow2; 293 struct ureg_dst y, u, v, rgb; 294 struct ureg_dst out = ureg_DECL_output(ureg, 295 TGSI_SEMANTIC_COLOR, 296 0); 297 298 pos = ureg_DECL_fs_input(ureg, 299 TGSI_SEMANTIC_GENERIC, 300 0, 301 TGSI_INTERPOLATE_PERSPECTIVE); 302 303 rgb = ureg_DECL_temporary(ureg); 304 y = ureg_DECL_temporary(ureg); 305 u = ureg_DECL_temporary(ureg); 306 v = ureg_DECL_temporary(ureg); 307 308 y_sampler = ureg_DECL_sampler(ureg, 0); 309 u_sampler = ureg_DECL_sampler(ureg, 1); 310 v_sampler = ureg_DECL_sampler(ureg, 2); 311 312 matrow0 = ureg_DECL_constant(ureg, 0); 313 matrow1 = ureg_DECL_constant(ureg, 1); 314 matrow2 = ureg_DECL_constant(ureg, 2); 315 316 ureg_TEX(ureg, y, 317 TGSI_TEXTURE_2D, pos, y_sampler); 318 ureg_TEX(ureg, u, 319 TGSI_TEXTURE_2D, pos, u_sampler); 320 ureg_TEX(ureg, v, 321 TGSI_TEXTURE_2D, pos, v_sampler); 322 323 ureg_SUB(ureg, u, ureg_src(u), 324 ureg_scalar(matrow0, TGSI_SWIZZLE_W)); 325 ureg_SUB(ureg, v, ureg_src(v), 326 ureg_scalar(matrow0, TGSI_SWIZZLE_W)); 327 328 ureg_MUL(ureg, rgb, 329 ureg_scalar(ureg_src(y), TGSI_SWIZZLE_X), 330 matrow0); 331 ureg_MAD(ureg, rgb, 332 ureg_scalar(ureg_src(u), TGSI_SWIZZLE_X), 333 matrow1, 334 ureg_src(rgb)); 335 ureg_MAD(ureg, rgb, 336 ureg_scalar(ureg_src(v), TGSI_SWIZZLE_X), 337 matrow2, 338 ureg_src(rgb)); 339 340 /* rgb.a = 1; */ 341 ureg_MOV(ureg, ureg_writemask(rgb, TGSI_WRITEMASK_W), 342 ureg_scalar(matrow0, TGSI_SWIZZLE_X)); 343 344 ureg_MOV(ureg, out, ureg_src(rgb)); 345 346 ureg_release_temporary(ureg, rgb); 347 ureg_release_temporary(ureg, y); 348 ureg_release_temporary(ureg, u); 349 ureg_release_temporary(ureg, v); 350 351 ureg_END(ureg); 352 353 return ureg_create_shader_and_destroy(ureg, pipe); 354} 355 356static void * 357create_fs(struct pipe_context *pipe, 358 unsigned fs_traits) 359{ 360 struct ureg_program *ureg; 361 struct ureg_src /*dst_sampler,*/ src_sampler, mask_sampler; 362 struct ureg_src /*dst_pos,*/ src_input, mask_pos; 363 struct ureg_dst src, mask; 364 struct ureg_dst out; 365 boolean has_mask = fs_traits & FS_MASK; 366 boolean is_fill = fs_traits & FS_FILL; 367 boolean is_composite = fs_traits & FS_COMPOSITE; 368 boolean is_solid = fs_traits & FS_SOLID_FILL; 369 boolean is_lingrad = fs_traits & FS_LINGRAD_FILL; 370 boolean is_radgrad = fs_traits & FS_RADGRAD_FILL; 371 unsigned comp_alpha = fs_traits & FS_COMPONENT_ALPHA; 372 boolean is_yuv = fs_traits & FS_YUV; 373 374 ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT); 375 if (ureg == NULL) 376 return 0; 377 378 /* it has to be either a fill, a composite op or a yuv conversion */ 379 debug_assert((is_fill ^ is_composite) ^ is_yuv); 380 381 out = ureg_DECL_output(ureg, 382 TGSI_SEMANTIC_COLOR, 383 0); 384 385 if (is_composite) { 386 src_sampler = ureg_DECL_sampler(ureg, 0); 387 src_input = ureg_DECL_fs_input(ureg, 388 TGSI_SEMANTIC_GENERIC, 389 0, 390 TGSI_INTERPOLATE_PERSPECTIVE); 391 } else if (is_fill) { 392 if (is_solid) 393 src_input = ureg_DECL_fs_input(ureg, 394 TGSI_SEMANTIC_COLOR, 395 0, 396 TGSI_INTERPOLATE_PERSPECTIVE); 397 else 398 src_input = ureg_DECL_fs_input(ureg, 399 TGSI_SEMANTIC_POSITION, 400 0, 401 TGSI_INTERPOLATE_PERSPECTIVE); 402 } else { 403 debug_assert(is_yuv); 404 return create_yuv_shader(pipe, ureg); 405 } 406 407 if (has_mask) { 408 mask_sampler = ureg_DECL_sampler(ureg, 1); 409 mask_pos = ureg_DECL_fs_input(ureg, 410 TGSI_SEMANTIC_GENERIC, 411 1, 412 TGSI_INTERPOLATE_PERSPECTIVE); 413 } 414 415#if 0 /* unused right now */ 416 dst_sampler = ureg_DECL_sampler(ureg, 2); 417 dst_pos = ureg_DECL_fs_input(ureg, 418 TGSI_SEMANTIC_POSITION, 419 2, 420 TGSI_INTERPOLATE_PERSPECTIVE); 421#endif 422 423 if (is_composite) { 424 if (has_mask) 425 src = ureg_DECL_temporary(ureg); 426 else 427 src = out; 428 ureg_TEX(ureg, src, 429 TGSI_TEXTURE_2D, src_input, src_sampler); 430 } else if (is_fill) { 431 if (is_solid) { 432 if (has_mask) 433 src = ureg_dst(src_input); 434 else 435 ureg_MOV(ureg, out, src_input); 436 } else if (is_lingrad || is_radgrad) { 437 struct ureg_src coords, const0124, 438 matrow0, matrow1, matrow2; 439 440 if (has_mask) 441 src = ureg_DECL_temporary(ureg); 442 else 443 src = out; 444 445 coords = ureg_DECL_constant(ureg, 0); 446 const0124 = ureg_DECL_constant(ureg, 1); 447 matrow0 = ureg_DECL_constant(ureg, 2); 448 matrow1 = ureg_DECL_constant(ureg, 3); 449 matrow2 = ureg_DECL_constant(ureg, 4); 450 451 if (is_lingrad) { 452 linear_gradient(ureg, src, 453 src_input, src_sampler, 454 coords, const0124, 455 matrow0, matrow1, matrow2); 456 } else if (is_radgrad) { 457 radial_gradient(ureg, src, 458 src_input, src_sampler, 459 coords, const0124, 460 matrow0, matrow1, matrow2); 461 } 462 } else 463 debug_assert(!"Unknown fill type!"); 464 } 465 466 if (has_mask) { 467 mask = ureg_DECL_temporary(ureg); 468 ureg_TEX(ureg, mask, 469 TGSI_TEXTURE_2D, mask_pos, mask_sampler); 470 /* src IN mask */ 471 src_in_mask(ureg, out, ureg_src(src), ureg_src(mask), comp_alpha); 472 ureg_release_temporary(ureg, mask); 473 } 474 475 ureg_END(ureg); 476 477 return ureg_create_shader_and_destroy(ureg, pipe); 478} 479 480struct xorg_shaders * xorg_shaders_create(struct xorg_renderer *r) 481{ 482 struct xorg_shaders *sc = CALLOC_STRUCT(xorg_shaders); 483 484 sc->r = r; 485 sc->vs_hash = cso_hash_create(); 486 sc->fs_hash = cso_hash_create(); 487 488 return sc; 489} 490 491static void 492cache_destroy(struct cso_context *cso, 493 struct cso_hash *hash, 494 unsigned processor) 495{ 496 struct cso_hash_iter iter = cso_hash_first_node(hash); 497 while (!cso_hash_iter_is_null(iter)) { 498 void *shader = (void *)cso_hash_iter_data(iter); 499 if (processor == PIPE_SHADER_FRAGMENT) { 500 cso_delete_fragment_shader(cso, shader); 501 } else if (processor == PIPE_SHADER_VERTEX) { 502 cso_delete_vertex_shader(cso, shader); 503 } 504 iter = cso_hash_erase(hash, iter); 505 } 506 cso_hash_delete(hash); 507} 508 509void xorg_shaders_destroy(struct xorg_shaders *sc) 510{ 511 cache_destroy(sc->r->cso, sc->vs_hash, 512 PIPE_SHADER_VERTEX); 513 cache_destroy(sc->r->cso, sc->fs_hash, 514 PIPE_SHADER_FRAGMENT); 515 516 free(sc); 517} 518 519static INLINE void * 520shader_from_cache(struct pipe_context *pipe, 521 unsigned type, 522 struct cso_hash *hash, 523 unsigned key) 524{ 525 void *shader = 0; 526 527 struct cso_hash_iter iter = cso_hash_find(hash, key); 528 529 if (cso_hash_iter_is_null(iter)) { 530 if (type == PIPE_SHADER_VERTEX) 531 shader = create_vs(pipe, key); 532 else 533 shader = create_fs(pipe, key); 534 cso_hash_insert(hash, key, shader); 535 } else 536 shader = (void *)cso_hash_iter_data(iter); 537 538 return shader; 539} 540 541struct xorg_shader xorg_shaders_get(struct xorg_shaders *sc, 542 unsigned vs_traits, 543 unsigned fs_traits) 544{ 545 struct xorg_shader shader = { NULL, NULL }; 546 void *vs, *fs; 547 548 vs = shader_from_cache(sc->r->pipe, PIPE_SHADER_VERTEX, 549 sc->vs_hash, vs_traits); 550 fs = shader_from_cache(sc->r->pipe, PIPE_SHADER_FRAGMENT, 551 sc->fs_hash, fs_traits); 552 553 debug_assert(vs && fs); 554 if (!vs || !fs) 555 return shader; 556 557 shader.vs = vs; 558 shader.fs = fs; 559 560 return shader; 561} 562