r300_fs.c revision 4e52e8f746e3565bf3fe9ba8e4e3744ce0302478
1/* 2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> 3 * Joakim Sindholt <opensource@zhasha.com> 4 * Copyright 2009 Marek Olšák <maraeo@gmail.com> 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the "Software"), 8 * to deal in the Software without restriction, including without limitation 9 * on the rights to use, copy, modify, merge, publish, distribute, sub 10 * license, and/or sell copies of the Software, and to permit persons to whom 11 * the Software is furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice (including the next 14 * paragraph) shall be included in all copies or substantial portions of the 15 * Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 21 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 22 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 23 * USE OR OTHER DEALINGS IN THE SOFTWARE. */ 24 25#include "util/u_math.h" 26#include "util/u_memory.h" 27 28#include "tgsi/tgsi_dump.h" 29#include "tgsi/tgsi_ureg.h" 30 31#include "r300_cb.h" 32#include "r300_context.h" 33#include "r300_emit.h" 34#include "r300_screen.h" 35#include "r300_fs.h" 36#include "r300_reg.h" 37#include "r300_tgsi_to_rc.h" 38 39#include "radeon_code.h" 40#include "radeon_compiler.h" 41 42/* Convert info about FS input semantics to r300_shader_semantics. */ 43void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, 44 struct r300_shader_semantics* fs_inputs) 45{ 46 int i; 47 unsigned index; 48 49 r300_shader_semantics_reset(fs_inputs); 50 51 for (i = 0; i < info->num_inputs; i++) { 52 index = info->input_semantic_index[i]; 53 54 switch (info->input_semantic_name[i]) { 55 case TGSI_SEMANTIC_COLOR: 56 assert(index < ATTR_COLOR_COUNT); 57 fs_inputs->color[index] = i; 58 break; 59 60 case TGSI_SEMANTIC_GENERIC: 61 assert(index < ATTR_GENERIC_COUNT); 62 fs_inputs->generic[index] = i; 63 break; 64 65 case TGSI_SEMANTIC_FOG: 66 assert(index == 0); 67 fs_inputs->fog = i; 68 break; 69 70 case TGSI_SEMANTIC_POSITION: 71 assert(index == 0); 72 fs_inputs->wpos = i; 73 break; 74 75 case TGSI_SEMANTIC_FACE: 76 assert(index == 0); 77 fs_inputs->face = i; 78 break; 79 80 default: 81 fprintf(stderr, "r300: FP: Unknown input semantic: %i\n", 82 info->input_semantic_name[i]); 83 } 84 } 85} 86 87static void find_output_registers(struct r300_fragment_program_compiler * compiler, 88 struct r300_fragment_shader_code *shader) 89{ 90 unsigned i, colorbuf_count = 0; 91 92 /* Mark the outputs as not present initially */ 93 compiler->OutputColor[0] = shader->info.num_outputs; 94 compiler->OutputColor[1] = shader->info.num_outputs; 95 compiler->OutputColor[2] = shader->info.num_outputs; 96 compiler->OutputColor[3] = shader->info.num_outputs; 97 compiler->OutputDepth = shader->info.num_outputs; 98 99 /* Now see where they really are. */ 100 for(i = 0; i < shader->info.num_outputs; ++i) { 101 switch(shader->info.output_semantic_name[i]) { 102 case TGSI_SEMANTIC_COLOR: 103 compiler->OutputColor[colorbuf_count] = i; 104 colorbuf_count++; 105 break; 106 case TGSI_SEMANTIC_POSITION: 107 compiler->OutputDepth = i; 108 break; 109 } 110 } 111} 112 113static void allocate_hardware_inputs( 114 struct r300_fragment_program_compiler * c, 115 void (*allocate)(void * data, unsigned input, unsigned hwreg), 116 void * mydata) 117{ 118 struct r300_shader_semantics* inputs = 119 (struct r300_shader_semantics*)c->UserData; 120 int i, reg = 0; 121 122 /* Allocate input registers. */ 123 for (i = 0; i < ATTR_COLOR_COUNT; i++) { 124 if (inputs->color[i] != ATTR_UNUSED) { 125 allocate(mydata, inputs->color[i], reg++); 126 } 127 } 128 if (inputs->face != ATTR_UNUSED) { 129 allocate(mydata, inputs->face, reg++); 130 } 131 for (i = 0; i < ATTR_GENERIC_COUNT; i++) { 132 if (inputs->generic[i] != ATTR_UNUSED) { 133 allocate(mydata, inputs->generic[i], reg++); 134 } 135 } 136 if (inputs->fog != ATTR_UNUSED) { 137 allocate(mydata, inputs->fog, reg++); 138 } 139 if (inputs->wpos != ATTR_UNUSED) { 140 allocate(mydata, inputs->wpos, reg++); 141 } 142} 143 144static void get_external_state( 145 struct r300_context* r300, 146 struct r300_fragment_program_external_state* state) 147{ 148 struct r300_textures_state *texstate = r300->textures_state.state; 149 unsigned i; 150 unsigned char *swizzle; 151 152 for (i = 0; i < texstate->sampler_state_count; i++) { 153 struct r300_sampler_state *s = texstate->sampler_states[i]; 154 struct r300_sampler_view *v = texstate->sampler_views[i]; 155 struct r300_texture *t; 156 157 if (!s || !v) { 158 continue; 159 } 160 161 t = r300_texture(texstate->sampler_views[i]->base.texture); 162 163 if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { 164 state->unit[i].compare_mode_enabled = 1; 165 166 /* Pass depth texture swizzling to the compiler. */ 167 if (texstate->sampler_views[i]) { 168 swizzle = texstate->sampler_views[i]->swizzle; 169 170 state->unit[i].depth_texture_swizzle = 171 RC_MAKE_SWIZZLE(swizzle[0], swizzle[1], 172 swizzle[2], swizzle[3]); 173 } else { 174 state->unit[i].depth_texture_swizzle = RC_SWIZZLE_XYZW; 175 } 176 177 /* Fortunately, no need to translate this. */ 178 state->unit[i].texture_compare_func = s->state.compare_func; 179 } 180 181 state->unit[i].non_normalized_coords = !s->state.normalized_coords; 182 183 /* XXX this should probably take into account STR, not just S. */ 184 if (t->desc.is_npot) { 185 switch (s->state.wrap_s) { 186 case PIPE_TEX_WRAP_REPEAT: 187 state->unit[i].wrap_mode = RC_WRAP_REPEAT; 188 break; 189 190 case PIPE_TEX_WRAP_MIRROR_REPEAT: 191 state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT; 192 break; 193 194 case PIPE_TEX_WRAP_MIRROR_CLAMP: 195 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: 196 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: 197 state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP; 198 break; 199 200 default: 201 state->unit[i].wrap_mode = RC_WRAP_NONE; 202 } 203 204 if (t->desc.b.b.target == PIPE_TEXTURE_3D) 205 state->unit[i].clamp_and_scale_before_fetch = TRUE; 206 } 207 } 208} 209 210static void r300_translate_fragment_shader( 211 struct r300_context* r300, 212 struct r300_fragment_shader_code* shader, 213 const struct tgsi_token *tokens); 214 215static void r300_dummy_fragment_shader( 216 struct r300_context* r300, 217 struct r300_fragment_shader_code* shader) 218{ 219 struct pipe_shader_state state; 220 struct ureg_program *ureg; 221 struct ureg_dst out; 222 struct ureg_src imm; 223 224 /* Make a simple fragment shader which outputs (0, 0, 0, 1) */ 225 ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT); 226 out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0); 227 imm = ureg_imm4f(ureg, 0, 0, 0, 1); 228 229 ureg_MOV(ureg, out, imm); 230 ureg_END(ureg); 231 232 state.tokens = ureg_finalize(ureg); 233 234 shader->dummy = TRUE; 235 r300_translate_fragment_shader(r300, shader, state.tokens); 236 237 ureg_destroy(ureg); 238} 239 240static void r300_emit_fs_code_to_buffer( 241 struct r300_context *r300, 242 struct r300_fragment_shader_code *shader) 243{ 244 struct rX00_fragment_program_code *generic_code = &shader->code; 245 unsigned imm_count = shader->immediates_count; 246 unsigned imm_first = shader->externals_count; 247 unsigned imm_end = generic_code->constants.Count; 248 struct rc_constant *constants = generic_code->constants.Constants; 249 unsigned i; 250 CB_LOCALS; 251 252 if (r300->screen->caps.is_r500) { 253 struct r500_fragment_program_code *code = &generic_code->code.r500; 254 255 shader->cb_code_size = 19 + 256 ((code->inst_end + 1) * 6) + 257 imm_count * 7 + 258 code->int_constant_count * 2; 259 260 NEW_CB(shader->cb_code, shader->cb_code_size); 261 OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); 262 OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx); 263 OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl); 264 for(i = 0; i < code->int_constant_count; i++){ 265 OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4), 266 code->int_constants[i]); 267 } 268 OUT_CB_REG(R500_US_CODE_RANGE, 269 R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); 270 OUT_CB_REG(R500_US_CODE_OFFSET, 0); 271 OUT_CB_REG(R500_US_CODE_ADDR, 272 R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); 273 274 OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); 275 OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); 276 for (i = 0; i <= code->inst_end; i++) { 277 OUT_CB(code->inst[i].inst0); 278 OUT_CB(code->inst[i].inst1); 279 OUT_CB(code->inst[i].inst2); 280 OUT_CB(code->inst[i].inst3); 281 OUT_CB(code->inst[i].inst4); 282 OUT_CB(code->inst[i].inst5); 283 } 284 285 /* Emit immediates. */ 286 if (imm_count) { 287 for(i = imm_first; i < imm_end; ++i) { 288 if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { 289 const float *data = constants[i].u.Immediate; 290 291 OUT_CB_REG(R500_GA_US_VECTOR_INDEX, 292 R500_GA_US_VECTOR_INDEX_TYPE_CONST | 293 (i & R500_GA_US_VECTOR_INDEX_MASK)); 294 OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4); 295 OUT_CB_TABLE(data, 4); 296 } 297 } 298 } 299 } else { /* r300 */ 300 struct r300_fragment_program_code *code = &generic_code->code.r300; 301 302 shader->cb_code_size = 19 + 303 (r300->screen->caps.is_r400 ? 2 : 0) + 304 code->alu.length * 4 + 305 (code->tex.length ? (1 + code->tex.length) : 0) + 306 imm_count * 5; 307 308 NEW_CB(shader->cb_code, shader->cb_code_size); 309 310 if (r300->screen->caps.is_r400) 311 OUT_CB_REG(R400_US_CODE_BANK, 0); 312 313 OUT_CB_REG(R300_US_CONFIG, code->config); 314 OUT_CB_REG(R300_US_PIXSIZE, code->pixsize); 315 OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset); 316 317 OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4); 318 OUT_CB_TABLE(code->code_addr, 4); 319 320 OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length); 321 for (i = 0; i < code->alu.length; i++) 322 OUT_CB(code->alu.inst[i].rgb_inst); 323 324 OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length); 325 for (i = 0; i < code->alu.length; i++) 326 OUT_CB(code->alu.inst[i].rgb_addr); 327 328 OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length); 329 for (i = 0; i < code->alu.length; i++) 330 OUT_CB(code->alu.inst[i].alpha_inst); 331 332 OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length); 333 for (i = 0; i < code->alu.length; i++) 334 OUT_CB(code->alu.inst[i].alpha_addr); 335 336 if (code->tex.length) { 337 OUT_CB_REG_SEQ(R300_US_TEX_INST_0, code->tex.length); 338 OUT_CB_TABLE(code->tex.inst, code->tex.length); 339 } 340 341 /* Emit immediates. */ 342 if (imm_count) { 343 for(i = imm_first; i < imm_end; ++i) { 344 if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { 345 const float *data = constants[i].u.Immediate; 346 347 OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); 348 OUT_CB(pack_float24(data[0])); 349 OUT_CB(pack_float24(data[1])); 350 OUT_CB(pack_float24(data[2])); 351 OUT_CB(pack_float24(data[3])); 352 } 353 } 354 } 355 } 356 357 OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src); 358 OUT_CB_REG(R300_US_W_FMT, shader->us_out_w); 359 END_CB; 360} 361 362static void r300_translate_fragment_shader( 363 struct r300_context* r300, 364 struct r300_fragment_shader_code* shader, 365 const struct tgsi_token *tokens) 366{ 367 struct r300_fragment_program_compiler compiler; 368 struct tgsi_to_rc ttr; 369 int wpos, face; 370 unsigned i; 371 372 tgsi_scan_shader(tokens, &shader->info); 373 r300_shader_read_fs_inputs(&shader->info, &shader->inputs); 374 375 wpos = shader->inputs.wpos; 376 face = shader->inputs.face; 377 378 /* Setup the compiler. */ 379 memset(&compiler, 0, sizeof(compiler)); 380 rc_init(&compiler.Base); 381 DBG_ON(r300, DBG_FP) ? compiler.Base.Debug |= RC_DBG_LOG : 0; 382 DBG_ON(r300, DBG_P_STAT) ? compiler.Base.Debug |= RC_DBG_STATS : 0; 383 384 compiler.code = &shader->code; 385 compiler.state = shader->compare_state; 386 compiler.Base.is_r500 = r300->screen->caps.is_r500; 387 compiler.Base.disable_optimizations = DBG_ON(r300, DBG_NO_OPT); 388 compiler.Base.has_half_swizzles = TRUE; 389 compiler.Base.has_presub = TRUE; 390 compiler.Base.max_temp_regs = compiler.Base.is_r500 ? 128 : 32; 391 compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32; 392 compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64; 393 compiler.AllocateHwInputs = &allocate_hardware_inputs; 394 compiler.UserData = &shader->inputs; 395 396 find_output_registers(&compiler, shader); 397 398 shader->write_all = FALSE; 399 for (i = 0; i < shader->info.num_properties; i++) { 400 if (shader->info.properties[i].name == TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS) { 401 shader->write_all = TRUE; 402 } 403 } 404 405 if (compiler.Base.Debug & RC_DBG_LOG) { 406 DBG(r300, DBG_FP, "r300: Initial fragment program\n"); 407 tgsi_dump(tokens, 0); 408 } 409 410 /* Translate TGSI to our internal representation */ 411 ttr.compiler = &compiler.Base; 412 ttr.info = &shader->info; 413 ttr.use_half_swizzles = TRUE; 414 415 r300_tgsi_to_rc(&ttr, tokens); 416 417 if (!r300->screen->caps.is_r500 || 418 compiler.Base.Program.Constants.Count > 200) { 419 compiler.Base.remove_unused_constants = TRUE; 420 } 421 422 /** 423 * Transform the program to support WPOS. 424 * 425 * Introduce a small fragment at the start of the program that will be 426 * the only code that directly reads the WPOS input. 427 * All other code pieces that reference that input will be rewritten 428 * to read from a newly allocated temporary. */ 429 if (wpos != ATTR_UNUSED) { 430 /* Moving the input to some other reg is not really necessary. */ 431 rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE); 432 } 433 434 if (face != ATTR_UNUSED) { 435 rc_transform_fragment_face(&compiler.Base, face); 436 } 437 438 /* Invoke the compiler */ 439 r3xx_compile_fragment_program(&compiler); 440 441 if (compiler.Base.Error) { 442 fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader" 443 " instead.\n", compiler.Base.ErrorMsg); 444 445 if (shader->dummy) { 446 fprintf(stderr, "r300 FP: Cannot compile the dummy shader! " 447 "Giving up...\n"); 448 abort(); 449 } 450 451 rc_destroy(&compiler.Base); 452 r300_dummy_fragment_shader(r300, shader); 453 return; 454 } 455 456 /* Shaders with zero instructions are invalid, 457 * use the dummy shader instead. */ 458 if (shader->code.code.r500.inst_end == -1) { 459 rc_destroy(&compiler.Base); 460 r300_dummy_fragment_shader(r300, shader); 461 return; 462 } 463 464 /* Initialize numbers of constants for each type. */ 465 shader->externals_count = 0; 466 for (i = 0; 467 i < shader->code.constants.Count && 468 shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { 469 shader->externals_count = i+1; 470 } 471 shader->immediates_count = 0; 472 shader->rc_state_count = 0; 473 474 for (i = shader->externals_count; i < shader->code.constants.Count; i++) { 475 switch (shader->code.constants.Constants[i].Type) { 476 case RC_CONSTANT_IMMEDIATE: 477 ++shader->immediates_count; 478 break; 479 case RC_CONSTANT_STATE: 480 ++shader->rc_state_count; 481 break; 482 default: 483 assert(0); 484 } 485 } 486 487 /* Setup shader depth output. */ 488 if (shader->code.writes_depth) { 489 shader->fg_depth_src = R300_FG_DEPTH_SRC_SHADER; 490 shader->us_out_w = R300_W_FMT_W24 | R300_W_SRC_US; 491 } else { 492 shader->fg_depth_src = R300_FG_DEPTH_SRC_SCAN; 493 shader->us_out_w = R300_W_FMT_W0 | R300_W_SRC_US; 494 } 495 496 /* And, finally... */ 497 rc_destroy(&compiler.Base); 498 499 /* Build the command buffer. */ 500 r300_emit_fs_code_to_buffer(r300, shader); 501} 502 503boolean r300_pick_fragment_shader(struct r300_context* r300) 504{ 505 struct r300_fragment_shader* fs = r300_fs(r300); 506 struct r300_fragment_program_external_state state = {{{ 0 }}}; 507 struct r300_fragment_shader_code* ptr; 508 509 get_external_state(r300, &state); 510 511 if (!fs->first) { 512 /* Build the fragment shader for the first time. */ 513 fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code); 514 515 memcpy(&fs->shader->compare_state, &state, 516 sizeof(struct r300_fragment_program_external_state)); 517 r300_translate_fragment_shader(r300, fs->shader, fs->state.tokens); 518 return TRUE; 519 520 } else { 521 /* Check if the currently-bound shader has been compiled 522 * with the texture-compare state we need. */ 523 if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) { 524 /* Search for the right shader. */ 525 ptr = fs->first; 526 while (ptr) { 527 if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) { 528 if (fs->shader != ptr) { 529 fs->shader = ptr; 530 return TRUE; 531 } 532 /* The currently-bound one is OK. */ 533 return FALSE; 534 } 535 ptr = ptr->next; 536 } 537 538 /* Not found, gotta compile a new one. */ 539 ptr = CALLOC_STRUCT(r300_fragment_shader_code); 540 ptr->next = fs->first; 541 fs->first = fs->shader = ptr; 542 543 ptr->compare_state = state; 544 r300_translate_fragment_shader(r300, ptr, fs->state.tokens); 545 return TRUE; 546 } 547 } 548 549 return FALSE; 550} 551