r300_state.c revision cba6430524198a1bdcdeada03cbe946a454f3935
1/* 2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> 3 * Copyright 2009 Marek Olšák <maraeo@gmail.com> 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * on the rights to use, copy, modify, merge, publish, distribute, sub 9 * license, and/or sell copies of the Software, and to permit persons to whom 10 * the Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */ 23 24#include "draw/draw_context.h" 25 26#include "util/u_math.h" 27#include "util/u_memory.h" 28#include "util/u_pack_color.h" 29 30#include "tgsi/tgsi_parse.h" 31 32#include "pipe/p_config.h" 33 34#include "r300_context.h" 35#include "r300_reg.h" 36#include "r300_screen.h" 37#include "r300_screen_buffer.h" 38#include "r300_state_inlines.h" 39#include "r300_fs.h" 40#include "r300_vs.h" 41 42#include "radeon_winsys.h" 43 44/* r300_state: Functions used to intialize state context by translating 45 * Gallium state objects into semi-native r300 state objects. */ 46 47#define UPDATE_STATE(cso, atom) \ 48 if (cso != atom.state) { \ 49 atom.state = cso; \ 50 atom.dirty = TRUE; \ 51 } 52 53static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 54 unsigned dstRGB, unsigned dstA) 55{ 56 /* If the blend equation is ADD or REVERSE_SUBTRACT, 57 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 58 * will not be changed. 59 * Notice that the dst factors are the src factors inverted. */ 60 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 61 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 62 srcRGB == PIPE_BLENDFACTOR_ZERO) && 63 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 64 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 65 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 66 srcA == PIPE_BLENDFACTOR_ZERO) && 67 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 68 dstRGB == PIPE_BLENDFACTOR_ONE) && 69 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 70 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 71 dstA == PIPE_BLENDFACTOR_ONE); 72} 73 74static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 75 unsigned dstRGB, unsigned dstA) 76{ 77 /* If the blend equation is ADD or REVERSE_SUBTRACT, 78 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 79 * will not be changed. 80 * Notice that the dst factors are the src factors inverted. */ 81 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 82 srcRGB == PIPE_BLENDFACTOR_ZERO) && 83 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 84 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 85 srcA == PIPE_BLENDFACTOR_ZERO) && 86 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 87 dstRGB == PIPE_BLENDFACTOR_ONE) && 88 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 89 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 90 dstA == PIPE_BLENDFACTOR_ONE); 91} 92 93static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 94 unsigned dstRGB, unsigned dstA) 95{ 96 /* If the blend equation is ADD or REVERSE_SUBTRACT, 97 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 98 * will not be changed. 99 * Notice that the dst factors are the src factors inverted. */ 100 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 101 srcRGB == PIPE_BLENDFACTOR_ZERO) && 102 (srcA == PIPE_BLENDFACTOR_ZERO) && 103 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 104 dstRGB == PIPE_BLENDFACTOR_ONE) && 105 (dstA == PIPE_BLENDFACTOR_ONE); 106} 107 108static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 109 unsigned dstRGB, unsigned dstA) 110{ 111 /* If the blend equation is ADD or REVERSE_SUBTRACT, 112 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 113 * will not be changed. 114 * Notice that the dst factors are the src factors inverted. */ 115 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 116 srcRGB == PIPE_BLENDFACTOR_ZERO) && 117 (srcA == PIPE_BLENDFACTOR_ZERO) && 118 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 119 dstRGB == PIPE_BLENDFACTOR_ONE) && 120 (dstA == PIPE_BLENDFACTOR_ONE); 121} 122 123static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 124 unsigned dstRGB, unsigned dstA) 125{ 126 /* If the blend equation is ADD or REVERSE_SUBTRACT, 127 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 128 * the colorbuffer will not be changed. 129 * Notice that the dst factors are the src factors inverted. */ 130 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 131 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 132 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 133 srcRGB == PIPE_BLENDFACTOR_ZERO) && 134 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 135 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 136 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 137 srcA == PIPE_BLENDFACTOR_ZERO) && 138 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 139 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 140 dstRGB == PIPE_BLENDFACTOR_ONE) && 141 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 142 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 143 dstA == PIPE_BLENDFACTOR_ONE); 144} 145 146static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 147 unsigned dstRGB, unsigned dstA) 148{ 149 /* If the blend equation is ADD or REVERSE_SUBTRACT, 150 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 151 * the colorbuffer will not be changed. 152 * Notice that the dst factors are the src factors inverted. */ 153 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 154 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 155 srcRGB == PIPE_BLENDFACTOR_ZERO) && 156 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 157 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 158 srcA == PIPE_BLENDFACTOR_ZERO) && 159 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 160 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 161 dstRGB == PIPE_BLENDFACTOR_ONE) && 162 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 163 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 164 dstA == PIPE_BLENDFACTOR_ONE); 165} 166 167static unsigned bgra_cmask(unsigned mask) 168{ 169 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 170 171 return ((mask & PIPE_MASK_R) << 2) | 172 ((mask & PIPE_MASK_B) >> 2) | 173 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 174} 175 176/* Create a new blend state based on the CSO blend state. 177 * 178 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 179static void* r300_create_blend_state(struct pipe_context* pipe, 180 const struct pipe_blend_state* state) 181{ 182 struct r300_screen* r300screen = r300_screen(pipe->screen); 183 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 184 185 if (state->rt[0].blend_enable) 186 { 187 unsigned eqRGB = state->rt[0].rgb_func; 188 unsigned srcRGB = state->rt[0].rgb_src_factor; 189 unsigned dstRGB = state->rt[0].rgb_dst_factor; 190 191 unsigned eqA = state->rt[0].alpha_func; 192 unsigned srcA = state->rt[0].alpha_src_factor; 193 unsigned dstA = state->rt[0].alpha_dst_factor; 194 195 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 196 * this is just the crappy D3D naming */ 197 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 198 r300_translate_blend_function(eqRGB) | 199 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 200 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 201 202 /* Optimization: some operations do not require the destination color. 203 * 204 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 205 * otherwise blending gives incorrect results. It seems to be 206 * a hardware bug. */ 207 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 208 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 209 dstRGB != PIPE_BLENDFACTOR_ZERO || 210 dstA != PIPE_BLENDFACTOR_ZERO || 211 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 212 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 213 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 214 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 215 srcA == PIPE_BLENDFACTOR_DST_COLOR || 216 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 217 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 218 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 219 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 220 /* Enable reading from the colorbuffer. */ 221 blend->blend_control |= R300_READ_ENABLE; 222 223 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) { 224 /* Optimization: Depending on incoming pixels, we can 225 * conditionally disable the reading in hardware... */ 226 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 227 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 228 /* Disable reading if SRC_ALPHA == 0. */ 229 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 230 dstRGB == PIPE_BLENDFACTOR_ZERO) && 231 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 232 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 233 dstA == PIPE_BLENDFACTOR_ZERO)) { 234 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 235 } 236 237 /* Disable reading if SRC_ALPHA == 1. */ 238 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 239 dstRGB == PIPE_BLENDFACTOR_ZERO) && 240 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 241 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 242 dstA == PIPE_BLENDFACTOR_ZERO)) { 243 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 244 } 245 } 246 } 247 } 248 249 /* Optimization: discard pixels which don't change the colorbuffer. 250 * 251 * The code below is non-trivial and some math is involved. 252 * 253 * Discarding pixels must be disabled when FP16 AA is enabled. 254 * This is a hardware bug. Also, this implementation wouldn't work 255 * with FP blending enabled and equation clamping disabled. 256 * 257 * Equations other than ADD are rarely used and therefore won't be 258 * optimized. */ 259 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 260 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 261 /* ADD: X+Y 262 * REVERSE_SUBTRACT: Y-X 263 * 264 * The idea is: 265 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 266 * then CB will not be changed. 267 * 268 * Given the srcFactor and dstFactor variables, we can derive 269 * what src and dst should be equal to and discard appropriate 270 * pixels. 271 */ 272 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 273 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 274 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 275 dstRGB, dstA)) { 276 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 277 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 278 dstRGB, dstA)) { 279 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 280 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 281 dstRGB, dstA)) { 282 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 283 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 284 dstRGB, dstA)) { 285 blend->blend_control |= 286 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 287 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 288 dstRGB, dstA)) { 289 blend->blend_control |= 290 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 291 } 292 } 293 294 /* separate alpha */ 295 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 296 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 297 blend->alpha_blend_control = 298 r300_translate_blend_function(eqA) | 299 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 300 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 301 } 302 } 303 304 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 305 if (state->logicop_enable) { 306 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 307 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 308 } 309 310 /* Color channel masks for all MRTs. */ 311 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); 312 if (r300screen->caps->is_r500 && state->independent_blend_enable) { 313 if (state->rt[1].blend_enable) { 314 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 315 } 316 if (state->rt[2].blend_enable) { 317 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 318 } 319 if (state->rt[3].blend_enable) { 320 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 321 } 322 } 323 324 if (state->dither) { 325 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 326 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 327 } 328 329 return (void*)blend; 330} 331 332/* Bind blend state. */ 333static void r300_bind_blend_state(struct pipe_context* pipe, 334 void* state) 335{ 336 struct r300_context* r300 = r300_context(pipe); 337 338 UPDATE_STATE(state, r300->blend_state); 339} 340 341/* Free blend state. */ 342static void r300_delete_blend_state(struct pipe_context* pipe, 343 void* state) 344{ 345 FREE(state); 346} 347 348/* Convert float to 10bit integer */ 349static unsigned float_to_fixed10(float f) 350{ 351 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 352} 353 354/* Set blend color. 355 * Setup both R300 and R500 registers, figure out later which one to write. */ 356static void r300_set_blend_color(struct pipe_context* pipe, 357 const struct pipe_blend_color* color) 358{ 359 struct r300_context* r300 = r300_context(pipe); 360 struct r300_screen* r300screen = r300_screen(pipe->screen); 361 struct r300_blend_color_state* state = 362 (struct r300_blend_color_state*)r300->blend_color_state.state; 363 union util_color uc; 364 365 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 366 state->blend_color = uc.ui; 367 368 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 369 state->blend_color_red_alpha = 370 float_to_fixed10(color->color[0]) | 371 (float_to_fixed10(color->color[3]) << 16); 372 state->blend_color_green_blue = 373 float_to_fixed10(color->color[2]) | 374 (float_to_fixed10(color->color[1]) << 16); 375 376 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; 377 r300->blend_color_state.dirty = TRUE; 378} 379 380static void r300_set_clip_state(struct pipe_context* pipe, 381 const struct pipe_clip_state* state) 382{ 383 struct r300_context* r300 = r300_context(pipe); 384 385 r300->clip = *state; 386 387 if (r300_screen(pipe->screen)->caps->has_tcl) { 388 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 389 r300->clip_state.size = 29; 390 } else { 391 draw_flush(r300->draw); 392 draw_set_clip_state(r300->draw, state); 393 r300->clip_state.size = 2; 394 } 395 396 r300->clip_state.dirty = TRUE; 397} 398 399/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 400 * 401 * This contains the depth buffer, stencil buffer, alpha test, and such. 402 * On the Radeon, depth and stencil buffer setup are intertwined, which is 403 * the reason for some of the strange-looking assignments across registers. */ 404static void* 405 r300_create_dsa_state(struct pipe_context* pipe, 406 const struct pipe_depth_stencil_alpha_state* state) 407{ 408 struct r300_capabilities *caps = 409 r300_screen(r300_context(pipe)->context.screen)->caps; 410 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 411 412 /* Depth test setup. */ 413 if (state->depth.enabled) { 414 dsa->z_buffer_control |= R300_Z_ENABLE; 415 416 if (state->depth.writemask) { 417 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 418 } 419 420 dsa->z_stencil_control |= 421 (r300_translate_depth_stencil_function(state->depth.func) << 422 R300_Z_FUNC_SHIFT); 423 } 424 425 /* Stencil buffer setup. */ 426 if (state->stencil[0].enabled) { 427 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 428 dsa->z_stencil_control |= 429 (r300_translate_depth_stencil_function(state->stencil[0].func) << 430 R300_S_FRONT_FUNC_SHIFT) | 431 (r300_translate_stencil_op(state->stencil[0].fail_op) << 432 R300_S_FRONT_SFAIL_OP_SHIFT) | 433 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 434 R300_S_FRONT_ZPASS_OP_SHIFT) | 435 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 436 R300_S_FRONT_ZFAIL_OP_SHIFT); 437 438 dsa->stencil_ref_mask = 439 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 440 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 441 442 if (state->stencil[1].enabled) { 443 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 444 dsa->z_stencil_control |= 445 (r300_translate_depth_stencil_function(state->stencil[1].func) << 446 R300_S_BACK_FUNC_SHIFT) | 447 (r300_translate_stencil_op(state->stencil[1].fail_op) << 448 R300_S_BACK_SFAIL_OP_SHIFT) | 449 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 450 R300_S_BACK_ZPASS_OP_SHIFT) | 451 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 452 R300_S_BACK_ZFAIL_OP_SHIFT); 453 454 if (caps->is_r500) 455 { 456 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 457 dsa->stencil_ref_bf = 458 (state->stencil[1].valuemask << 459 R300_STENCILMASK_SHIFT) | 460 (state->stencil[1].writemask << 461 R300_STENCILWRITEMASK_SHIFT); 462 } 463 } 464 } 465 466 /* Alpha test setup. */ 467 if (state->alpha.enabled) { 468 dsa->alpha_function = 469 r300_translate_alpha_function(state->alpha.func) | 470 R300_FG_ALPHA_FUNC_ENABLE; 471 472 /* We could use 10bit alpha ref but who needs that? */ 473 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 474 475 if (caps->is_r500) 476 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 477 } 478 479 return (void*)dsa; 480} 481 482/* Bind DSA state. */ 483static void r300_bind_dsa_state(struct pipe_context* pipe, 484 void* state) 485{ 486 struct r300_context* r300 = r300_context(pipe); 487 488 UPDATE_STATE(state, r300->dsa_state); 489} 490 491/* Free DSA state. */ 492static void r300_delete_dsa_state(struct pipe_context* pipe, 493 void* state) 494{ 495 FREE(state); 496} 497 498static void r300_set_stencil_ref(struct pipe_context* pipe, 499 const struct pipe_stencil_ref* sr) 500{ 501 struct r300_context* r300 = r300_context(pipe); 502 r300->stencil_ref = *sr; 503 r300->dsa_state.dirty = TRUE; 504} 505 506/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 507static void r300_fb_update_tiling_flags(struct r300_context *r300, 508 const struct pipe_framebuffer_state *old_state, 509 const struct pipe_framebuffer_state *new_state) 510{ 511 struct r300_texture *tex; 512 unsigned i, j, level; 513 514 /* Reset tiling flags for old surfaces to default values. */ 515 for (i = 0; i < old_state->nr_cbufs; i++) { 516 for (j = 0; j < new_state->nr_cbufs; j++) { 517 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { 518 break; 519 } 520 } 521 /* If not binding the surface again... */ 522 if (j != new_state->nr_cbufs) { 523 continue; 524 } 525 526 tex = (struct r300_texture*)old_state->cbufs[i]->texture; 527 528 if (tex) { 529 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 530 tex->pitch[0], 531 tex->microtile != 0, 532 tex->macrotile != 0); 533 } 534 } 535 if (old_state->zsbuf && 536 (!new_state->zsbuf || 537 old_state->zsbuf->texture != new_state->zsbuf->texture)) { 538 tex = (struct r300_texture*)old_state->zsbuf->texture; 539 540 if (tex) { 541 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 542 tex->pitch[0], 543 tex->microtile != 0, 544 tex->macrotile != 0); 545 } 546 } 547 548 /* Set tiling flags for new surfaces. */ 549 for (i = 0; i < new_state->nr_cbufs; i++) { 550 tex = (struct r300_texture*)new_state->cbufs[i]->texture; 551 level = new_state->cbufs[i]->level; 552 553 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 554 tex->pitch[level], 555 tex->microtile != 0, 556 tex->mip_macrotile[level] != 0); 557 } 558 if (new_state->zsbuf) { 559 tex = (struct r300_texture*)new_state->zsbuf->texture; 560 level = new_state->zsbuf->level; 561 562 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 563 tex->pitch[level], 564 tex->microtile != 0, 565 tex->mip_macrotile[level] != 0); 566 } 567} 568 569static void 570 r300_set_framebuffer_state(struct pipe_context* pipe, 571 const struct pipe_framebuffer_state* state) 572{ 573 struct r300_context* r300 = r300_context(pipe); 574 struct r300_screen* r300screen = r300_screen(pipe->screen); 575 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 576 unsigned max_width, max_height; 577 uint32_t zbuffer_bpp = 0; 578 579 580 if (state->nr_cbufs > 4) { 581 debug_printf("r300: Implementation error: Too many MRTs in %s, " 582 "refusing to bind framebuffer state!\n", __FUNCTION__); 583 return; 584 } 585 586 if (r300screen->caps->is_r500) { 587 max_width = max_height = 4096; 588 } else if (r300screen->caps->is_r400) { 589 max_width = max_height = 4021; 590 } else { 591 max_width = max_height = 2560; 592 } 593 594 if (state->width > max_width || state->height > max_height) { 595 debug_printf("r300: Implementation error: Render targets are too " 596 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 597 return; 598 } 599 600 if (r300->draw) { 601 draw_flush(r300->draw); 602 } 603 604 r300->fb_state.dirty = TRUE; 605 606 /* If nr_cbufs is changed from zero to non-zero or vice versa... */ 607 if (!!old_state->nr_cbufs != !!state->nr_cbufs) { 608 r300->blend_state.dirty = TRUE; 609 } 610 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 611 if (!!old_state->zsbuf != !!state->zsbuf) { 612 r300->dsa_state.dirty = TRUE; 613 } 614 if (!r300->scissor_enabled) { 615 r300->scissor_state.dirty = TRUE; 616 } 617 618 r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); 619 620 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 621 622 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + 623 (state->zsbuf ? 10 : 0) + 8; 624 625 /* Polygon offset depends on the zbuffer bit depth. */ 626 if (state->zsbuf && r300->polygon_offset_enabled) { 627 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 628 case 2: 629 zbuffer_bpp = 16; 630 break; 631 case 4: 632 zbuffer_bpp = 24; 633 break; 634 } 635 636 if (r300->zbuffer_bpp != zbuffer_bpp) { 637 r300->zbuffer_bpp = zbuffer_bpp; 638 r300->rs_state.dirty = TRUE; 639 } 640 } 641} 642 643/* Create fragment shader state. */ 644static void* r300_create_fs_state(struct pipe_context* pipe, 645 const struct pipe_shader_state* shader) 646{ 647 struct r300_fragment_shader* fs = NULL; 648 649 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 650 651 /* Copy state directly into shader. */ 652 fs->state = *shader; 653 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 654 655 tgsi_scan_shader(shader->tokens, &fs->info); 656 r300_shader_read_fs_inputs(&fs->info, &fs->inputs); 657 658 return (void*)fs; 659} 660 661/* Bind fragment shader state. */ 662static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 663{ 664 struct r300_context* r300 = r300_context(pipe); 665 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 666 667 if (fs == NULL) { 668 r300->fs = NULL; 669 return; 670 } 671 672 r300->fs = fs; 673 r300_pick_fragment_shader(r300); 674 675 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 676 677 if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) { 678 r300->vap_output_state.dirty = TRUE; 679 } 680 681 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; 682} 683 684/* Delete fragment shader state. */ 685static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 686{ 687 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 688 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 689 690 while (ptr) { 691 tmp = ptr; 692 ptr = ptr->next; 693 rc_constants_destroy(&tmp->code.constants); 694 FREE(tmp); 695 } 696 FREE((void*)fs->state.tokens); 697 FREE(shader); 698} 699 700static void r300_set_polygon_stipple(struct pipe_context* pipe, 701 const struct pipe_poly_stipple* state) 702{ 703 /* XXX no idea how to set this up, but not terribly important */ 704} 705 706/* Create a new rasterizer state based on the CSO rasterizer state. 707 * 708 * This is a very large chunk of state, and covers most of the graphics 709 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 710 * 711 * In a not entirely unironic sidenote, this state has nearly nothing to do 712 * with the actual block on the Radeon called the rasterizer (RS). */ 713static void* r300_create_rs_state(struct pipe_context* pipe, 714 const struct pipe_rasterizer_state* state) 715{ 716 struct r300_screen* r300screen = r300_screen(pipe->screen); 717 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 718 unsigned coord_index; 719 720 /* Copy rasterizer state for Draw. */ 721 rs->rs = *state; 722 723#ifdef PIPE_ARCH_LITTLE_ENDIAN 724 rs->vap_control_status = R300_VC_NO_SWAP; 725#else 726 rs->vap_control_status = R300_VC_32BIT_SWAP; 727#endif 728 729 /* If no TCL engine is present, turn off the HW TCL. */ 730 if (!r300screen->caps->has_tcl) { 731 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 732 } 733 734 rs->point_size = pack_float_16_6x(state->point_size) | 735 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 736 737 rs->line_control = pack_float_16_6x(state->line_width) | 738 R300_GA_LINE_CNTL_END_TYPE_COMP; 739 740 /* Enable polygon mode */ 741 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 742 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 743 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 744 } 745 746 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 747 if (state->front_winding == PIPE_WINDING_CW) { 748 rs->cull_mode = R300_FRONT_FACE_CW; 749 750 /* Polygon offset */ 751 if (state->offset_cw) { 752 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 753 } 754 if (state->offset_ccw) { 755 rs->polygon_offset_enable |= R300_BACK_ENABLE; 756 } 757 758 /* Polygon mode */ 759 if (rs->polygon_mode) { 760 rs->polygon_mode |= 761 r300_translate_polygon_mode_front(state->fill_cw); 762 rs->polygon_mode |= 763 r300_translate_polygon_mode_back(state->fill_ccw); 764 } 765 } else { 766 rs->cull_mode = R300_FRONT_FACE_CCW; 767 768 /* Polygon offset */ 769 if (state->offset_ccw) { 770 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 771 } 772 if (state->offset_cw) { 773 rs->polygon_offset_enable |= R300_BACK_ENABLE; 774 } 775 776 /* Polygon mode */ 777 if (rs->polygon_mode) { 778 rs->polygon_mode |= 779 r300_translate_polygon_mode_front(state->fill_ccw); 780 rs->polygon_mode |= 781 r300_translate_polygon_mode_back(state->fill_cw); 782 } 783 } 784 if (state->front_winding & state->cull_mode) { 785 rs->cull_mode |= R300_CULL_FRONT; 786 } 787 if (~(state->front_winding) & state->cull_mode) { 788 rs->cull_mode |= R300_CULL_BACK; 789 } 790 791 if (rs->polygon_offset_enable) { 792 rs->depth_offset = state->offset_units; 793 rs->depth_scale = state->offset_scale; 794 } 795 796 if (state->line_stipple_enable) { 797 rs->line_stipple_config = 798 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 799 (fui((float)state->line_stipple_factor) & 800 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 801 /* XXX this might need to be scaled up */ 802 rs->line_stipple_value = state->line_stipple_pattern; 803 } 804 805 if (state->flatshade) { 806 rs->color_control = R300_SHADE_MODEL_FLAT; 807 } else { 808 rs->color_control = R300_SHADE_MODEL_SMOOTH; 809 } 810 811 /* Point sprites */ 812 if (state->sprite_coord_enable) { 813 coord_index = ffs(state->sprite_coord_enable)-1; 814 815 SCREEN_DBG(r300screen, DBG_DRAW, 816 "r300: point sprite: shader coord=%d\n", coord_index); 817 818 rs->stuffing_enable = 819 R300_GB_POINT_STUFF_ENABLE | 820 R300_GB_TEX_ST << (R300_GB_TEX0_SOURCE_SHIFT + (coord_index*2)); 821 822 rs->point_texcoord_left = 0.0f; 823 rs->point_texcoord_right = 1.0f; 824 825 switch (state->sprite_coord_mode) { 826 case PIPE_SPRITE_COORD_UPPER_LEFT: 827 rs->point_texcoord_top = 0.0f; 828 rs->point_texcoord_bottom = 1.0f; 829 break; 830 case PIPE_SPRITE_COORD_LOWER_LEFT: 831 rs->point_texcoord_top = 1.0f; 832 rs->point_texcoord_bottom = 0.0f; 833 break; 834 } 835 } 836 837 return (void*)rs; 838} 839 840/* Bind rasterizer state. */ 841static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 842{ 843 struct r300_context* r300 = r300_context(pipe); 844 struct r300_rs_state* rs = (struct r300_rs_state*)state; 845 boolean scissor_was_enabled = r300->scissor_enabled; 846 int last_sprite_coord_index = r300->sprite_coord_index; 847 848 if (r300->draw) { 849 draw_flush(r300->draw); 850 draw_set_rasterizer_state(r300->draw, &rs->rs); 851 } 852 853 if (rs) { 854 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 855 r300->scissor_enabled = rs->rs.scissor; 856 r300->sprite_coord_index = ffs(rs->rs.sprite_coord_enable)-1; 857 } else { 858 r300->polygon_offset_enabled = FALSE; 859 r300->scissor_enabled = FALSE; 860 r300->sprite_coord_index = -1; 861 } 862 863 UPDATE_STATE(state, r300->rs_state); 864 r300->rs_state.size = 24 + (r300->polygon_offset_enabled ? 5 : 0); 865 866 if (scissor_was_enabled != r300->scissor_enabled) { 867 r300->scissor_state.dirty = TRUE; 868 } 869 if (last_sprite_coord_index != r300->sprite_coord_index) { 870 r300->rs_block_state.dirty = TRUE; 871 } 872} 873 874/* Free rasterizer state. */ 875static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 876{ 877 FREE(state); 878} 879 880static void* 881 r300_create_sampler_state(struct pipe_context* pipe, 882 const struct pipe_sampler_state* state) 883{ 884 struct r300_context* r300 = r300_context(pipe); 885 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 886 boolean is_r500 = r300_screen(pipe->screen)->caps->is_r500; 887 int lod_bias; 888 union util_color uc; 889 890 sampler->state = *state; 891 892 sampler->filter0 |= 893 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 894 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 895 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 896 897 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 898 state->mag_img_filter, 899 state->min_mip_filter, 900 state->max_anisotropy > 0); 901 902 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 903 904 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 905 /* We must pass these to the merge function to clamp them properly. */ 906 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 907 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 908 909 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 910 911 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 912 913 /* This is very high quality anisotropic filtering for R5xx. 914 * It's good for benchmarking the performance of texturing but 915 * in practice we don't want to slow down the driver because it's 916 * a pretty good performance killer. Feel free to play with it. */ 917 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 918 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 919 } 920 921 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 922 sampler->border_color = uc.ui; 923 924 /* R500-specific fixups and optimizations */ 925 if (r300_screen(r300->context.screen)->caps->is_r500) { 926 sampler->filter1 |= R500_BORDER_FIX; 927 } 928 929 return (void*)sampler; 930} 931 932static void r300_bind_sampler_states(struct pipe_context* pipe, 933 unsigned count, 934 void** states) 935{ 936 struct r300_context* r300 = r300_context(pipe); 937 struct r300_textures_state* state = 938 (struct r300_textures_state*)r300->textures_state.state; 939 940 if (count > 8) { 941 return; 942 } 943 944 memcpy(state->sampler_states, states, sizeof(void*) * count); 945 state->sampler_count = count; 946 947 r300->textures_state.dirty = TRUE; 948 949 /* Pick a fragment shader based on the texture compare state. */ 950 if (r300->fs && count) { 951 if (r300_pick_fragment_shader(r300)) { 952 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | 953 R300_NEW_FRAGMENT_SHADER_CONSTANTS; 954 } 955 } 956} 957 958static void r300_lacks_vertex_textures(struct pipe_context* pipe, 959 unsigned count, 960 void** states) 961{ 962} 963 964static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 965{ 966 FREE(state); 967} 968 969static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 970 unsigned count, 971 struct pipe_sampler_view** views) 972{ 973 struct r300_context* r300 = r300_context(pipe); 974 struct r300_textures_state* state = 975 (struct r300_textures_state*)r300->textures_state.state; 976 unsigned i; 977 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; 978 boolean dirty_tex = FALSE; 979 980 /* XXX magic num */ 981 if (count > 8) { 982 return; 983 } 984 985 for (i = 0; i < count; i++) { 986 if (state->fragment_sampler_views[i] != views[i]) { 987 struct r300_texture *texture; 988 989 pipe_sampler_view_reference(&state->fragment_sampler_views[i], 990 views[i]); 991 dirty_tex = TRUE; 992 993 texture = (struct r300_texture *)views[i]->texture; 994 995 /* R300-specific - set the texrect factor in the fragment shader */ 996 if (!is_r500 && texture->is_npot) { 997 /* XXX It would be nice to re-emit just 1 constant, 998 * XXX not all of them */ 999 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1000 } 1001 } 1002 } 1003 1004 for (i = count; i < 8; i++) { 1005 if (state->fragment_sampler_views[i]) { 1006 pipe_sampler_view_reference(&state->fragment_sampler_views[i], 1007 NULL); 1008 } 1009 } 1010 1011 state->texture_count = count; 1012 1013 r300->textures_state.dirty = TRUE; 1014 1015 if (dirty_tex) { 1016 r300->texture_cache_inval.dirty = TRUE; 1017 } 1018} 1019 1020static struct pipe_sampler_view * 1021r300_create_sampler_view(struct pipe_context *pipe, 1022 struct pipe_texture *texture, 1023 const struct pipe_sampler_view *templ) 1024{ 1025 struct pipe_sampler_view *view = CALLOC_STRUCT(pipe_sampler_view); 1026 1027 if (view) { 1028 *view = *templ; 1029 view->reference.count = 1; 1030 view->texture = NULL; 1031 pipe_texture_reference(&view->texture, texture); 1032 view->context = pipe; 1033 } 1034 1035 return view; 1036} 1037 1038 1039static void 1040r300_sampler_view_destroy(struct pipe_context *pipe, 1041 struct pipe_sampler_view *view) 1042{ 1043 pipe_texture_reference(&view->texture, NULL); 1044 FREE(view); 1045} 1046 1047static void r300_set_scissor_state(struct pipe_context* pipe, 1048 const struct pipe_scissor_state* state) 1049{ 1050 struct r300_context* r300 = r300_context(pipe); 1051 1052 memcpy(r300->scissor_state.state, state, 1053 sizeof(struct pipe_scissor_state)); 1054 1055 if (r300->scissor_enabled) { 1056 r300->scissor_state.dirty = TRUE; 1057 } 1058} 1059 1060static void r300_set_viewport_state(struct pipe_context* pipe, 1061 const struct pipe_viewport_state* state) 1062{ 1063 struct r300_context* r300 = r300_context(pipe); 1064 struct r300_viewport_state* viewport = 1065 (struct r300_viewport_state*)r300->viewport_state.state; 1066 1067 r300->viewport = *state; 1068 1069 /* Do the transform in HW. */ 1070 viewport->vte_control = R300_VTX_W0_FMT; 1071 1072 if (state->scale[0] != 1.0f) { 1073 viewport->xscale = state->scale[0]; 1074 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1075 } 1076 if (state->scale[1] != 1.0f) { 1077 viewport->yscale = state->scale[1]; 1078 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1079 } 1080 if (state->scale[2] != 1.0f) { 1081 viewport->zscale = state->scale[2]; 1082 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1083 } 1084 if (state->translate[0] != 0.0f) { 1085 viewport->xoffset = state->translate[0]; 1086 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1087 } 1088 if (state->translate[1] != 0.0f) { 1089 viewport->yoffset = state->translate[1]; 1090 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1091 } 1092 if (state->translate[2] != 0.0f) { 1093 viewport->zoffset = state->translate[2]; 1094 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1095 } 1096 1097 r300->viewport_state.dirty = TRUE; 1098 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 1099 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1100 } 1101} 1102 1103static void r300_set_vertex_buffers(struct pipe_context* pipe, 1104 unsigned count, 1105 const struct pipe_vertex_buffer* buffers) 1106{ 1107 struct r300_context* r300 = r300_context(pipe); 1108 int i; 1109 unsigned max_index = (1 << 24) - 1; 1110 boolean any_user_buffer = false; 1111 1112 if (count == r300->vertex_buffer_count && 1113 memcmp(r300->vertex_buffer, buffers, count * sizeof(buffers[0])) == 0) 1114 return; 1115 1116 for (i = 0; i < count; i++) { 1117 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); 1118 if (r300_buffer_is_user_buffer(buffers[i].buffer)) 1119 any_user_buffer = true; 1120 max_index = MIN2(buffers[i].max_index, max_index); 1121 } 1122 1123 for ( ; i < r300->vertex_buffer_count; i++) 1124 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, NULL); 1125 1126 memcpy(r300->vertex_buffer, buffers, 1127 sizeof(struct pipe_vertex_buffer) * count); 1128 1129 r300->vertex_buffer_count = count; 1130 r300->vertex_buffer_max_index = max_index; 1131 r300->any_user_vbs = any_user_buffer; 1132 1133 if (r300->draw) { 1134 draw_flush(r300->draw); 1135 draw_set_vertex_buffers(r300->draw, count, buffers); 1136 } 1137} 1138 1139static boolean r300_validate_aos(struct r300_context *r300) 1140{ 1141 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; 1142 struct pipe_vertex_element *velem = r300->velems->velem; 1143 int i; 1144 1145 /* Check if formats and strides are aligned to the size of DWORD. */ 1146 for (i = 0; i < r300->velems->count; i++) { 1147 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || 1148 util_format_get_blocksize(velem[i].src_format) % 4 != 0) { 1149 return FALSE; 1150 } 1151 } 1152 return TRUE; 1153} 1154 1155static void r300_draw_emit_attrib(struct r300_context* r300, 1156 enum attrib_emit emit, 1157 enum interp_mode interp, 1158 int index) 1159{ 1160 struct r300_vertex_shader* vs = r300->vs_state.state; 1161 struct tgsi_shader_info* info = &vs->info; 1162 int output; 1163 1164 output = draw_find_shader_output(r300->draw, 1165 info->output_semantic_name[index], 1166 info->output_semantic_index[index]); 1167 draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output); 1168} 1169 1170static void r300_draw_emit_all_attribs(struct r300_context* r300) 1171{ 1172 struct r300_vertex_shader* vs = r300->vs_state.state; 1173 struct r300_shader_semantics* vs_outputs = &vs->outputs; 1174 int i, gen_count; 1175 1176 /* Position. */ 1177 if (vs_outputs->pos != ATTR_UNUSED) { 1178 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, 1179 vs_outputs->pos); 1180 } else { 1181 assert(0); 1182 } 1183 1184 /* Point size. */ 1185 if (vs_outputs->psize != ATTR_UNUSED) { 1186 r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS, 1187 vs_outputs->psize); 1188 } 1189 1190 /* Colors. */ 1191 for (i = 0; i < ATTR_COLOR_COUNT; i++) { 1192 if (vs_outputs->color[i] != ATTR_UNUSED) { 1193 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR, 1194 vs_outputs->color[i]); 1195 } 1196 } 1197 1198 /* XXX Back-face colors. */ 1199 1200 /* Texture coordinates. */ 1201 gen_count = 0; 1202 for (i = 0; i < ATTR_GENERIC_COUNT; i++) { 1203 if (vs_outputs->generic[i] != ATTR_UNUSED) { 1204 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, 1205 vs_outputs->generic[i]); 1206 gen_count++; 1207 } 1208 } 1209 1210 /* Fog coordinates. */ 1211 if (vs_outputs->fog != ATTR_UNUSED) { 1212 r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, 1213 vs_outputs->fog); 1214 gen_count++; 1215 } 1216 1217 /* XXX magic */ 1218 assert(gen_count <= 8); 1219} 1220 1221/* Update the PSC tables. */ 1222static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1223{ 1224 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1225 uint16_t type, swizzle; 1226 enum pipe_format format; 1227 unsigned i; 1228 1229 assert(velems->count <= 16); 1230 1231 /* Vertex shaders have no semantics on their inputs, 1232 * so PSC should just route stuff based on the vertex elements, 1233 * and not on attrib information. */ 1234 for (i = 0; i < velems->count; i++) { 1235 format = velems->velem[i].src_format; 1236 1237 type = r300_translate_vertex_data_type(format) | 1238 (i << R300_DST_VEC_LOC_SHIFT); 1239 swizzle = r300_translate_vertex_data_swizzle(format); 1240 1241 if (i & 1) { 1242 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1243 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1244 } else { 1245 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1246 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1247 } 1248 } 1249 1250 /* Set the last vector in the PSC. */ 1251 if (i) { 1252 i -= 1; 1253 } 1254 vstream->vap_prog_stream_cntl[i >> 1] |= 1255 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1256 1257 vstream->count = (i >> 1) + 1; 1258} 1259 1260/* Update the PSC tables for SW TCL, using Draw. */ 1261static void r300_swtcl_vertex_psc(struct r300_context *r300, 1262 struct r300_vertex_element_state *velems) 1263{ 1264 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1265 struct r300_vertex_shader* vs = r300->vs_state.state; 1266 struct vertex_info* vinfo = &r300->vertex_info; 1267 uint16_t type, swizzle; 1268 enum pipe_format format; 1269 unsigned i, attrib_count; 1270 int* vs_output_tab = vs->stream_loc_notcl; 1271 1272 /* For each Draw attribute, route it to the fragment shader according 1273 * to the vs_output_tab. */ 1274 attrib_count = vinfo->num_attribs; 1275 DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); 1276 for (i = 0; i < attrib_count; i++) { 1277 DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," 1278 " vs_output_tab %d\n", vinfo->attrib[i].src_index, 1279 vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, 1280 vs_output_tab[i]); 1281 } 1282 1283 for (i = 0; i < attrib_count; i++) { 1284 /* Make sure we have a proper destination for our attribute. */ 1285 assert(vs_output_tab[i] != -1); 1286 1287 format = draw_translate_vinfo_format(vinfo->attrib[i].emit); 1288 1289 /* Obtain the type of data in this attribute. */ 1290 type = r300_translate_vertex_data_type(format) | 1291 vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; 1292 1293 /* Obtain the swizzle for this attribute. Note that the default 1294 * swizzle in the hardware is not XYZW! */ 1295 swizzle = r300_translate_vertex_data_swizzle(format); 1296 1297 /* Add the attribute to the PSC table. */ 1298 if (i & 1) { 1299 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1300 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1301 } else { 1302 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1303 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1304 } 1305 } 1306 1307 /* Set the last vector in the PSC. */ 1308 if (i) { 1309 i -= 1; 1310 } 1311 vstream->vap_prog_stream_cntl[i >> 1] |= 1312 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1313 1314 vstream->count = (i >> 1) + 1; 1315} 1316 1317static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1318 unsigned count, 1319 const struct pipe_vertex_element* attribs) 1320{ 1321 struct r300_context *r300 = r300_context(pipe); 1322 struct r300_screen* r300screen = r300_screen(pipe->screen); 1323 struct r300_vertex_element_state *velems; 1324 1325 assert(count <= PIPE_MAX_ATTRIBS); 1326 velems = CALLOC_STRUCT(r300_vertex_element_state); 1327 if (velems != NULL) { 1328 velems->count = count; 1329 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1330 1331 if (r300screen->caps->has_tcl) { 1332 r300_vertex_psc(velems); 1333 } else { 1334 memset(&r300->vertex_info, 0, sizeof(struct vertex_info)); 1335 r300_draw_emit_all_attribs(r300); 1336 draw_compute_vertex_size(&r300->vertex_info); 1337 r300_swtcl_vertex_psc(r300, velems); 1338 } 1339 } 1340 return velems; 1341} 1342 1343static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1344 void *state) 1345{ 1346 struct r300_context *r300 = r300_context(pipe); 1347 struct r300_vertex_element_state *velems = state; 1348 1349 if (velems == NULL) { 1350 return; 1351 } 1352 1353 r300->velems = velems; 1354 1355 if (r300->draw) { 1356 draw_flush(r300->draw); 1357 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1358 } 1359 1360 if (!r300_validate_aos(r300)) { 1361 /* XXX We should fallback using draw. */ 1362 assert(0); 1363 abort(); 1364 } 1365 1366 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1367 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1368} 1369 1370static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1371{ 1372 FREE(state); 1373} 1374 1375static void* r300_create_vs_state(struct pipe_context* pipe, 1376 const struct pipe_shader_state* shader) 1377{ 1378 struct r300_context* r300 = r300_context(pipe); 1379 1380 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1381 r300_vertex_shader_common_init(vs, shader); 1382 1383 if (r300_screen(pipe->screen)->caps->has_tcl) { 1384 r300_translate_vertex_shader(r300, vs); 1385 } else { 1386 vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); 1387 } 1388 1389 return vs; 1390} 1391 1392static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1393{ 1394 struct r300_context* r300 = r300_context(pipe); 1395 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1396 1397 if (vs == NULL) { 1398 r300->vs_state.state = NULL; 1399 return; 1400 } 1401 if (vs == r300->vs_state.state) { 1402 return; 1403 } 1404 r300->vs_state.state = vs; 1405 1406 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block 1407 if (r300->fs) { 1408 r300_vertex_shader_setup_wpos(r300); 1409 } 1410 memcpy(r300->vap_output_state.state, &vs->vap_out, 1411 sizeof(struct r300_vap_output_state)); 1412 r300->vap_output_state.dirty = TRUE; 1413 1414 /* The majority of the RS block bits is dependent on the vertex shader. */ 1415 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1416 1417 if (r300_screen(pipe->screen)->caps->has_tcl) { 1418 r300->vs_state.dirty = TRUE; 1419 r300->vs_state.size = vs->code.length + 9; 1420 1421 r300->pvs_flush.dirty = TRUE; 1422 1423 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1424 } else { 1425 draw_flush(r300->draw); 1426 draw_bind_vertex_shader(r300->draw, 1427 (struct draw_vertex_shader*)vs->draw_vs); 1428 } 1429} 1430 1431static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1432{ 1433 struct r300_context* r300 = r300_context(pipe); 1434 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1435 1436 if (r300_screen(pipe->screen)->caps->has_tcl) { 1437 rc_constants_destroy(&vs->code.constants); 1438 } else { 1439 draw_delete_vertex_shader(r300->draw, 1440 (struct draw_vertex_shader*)vs->draw_vs); 1441 } 1442 1443 FREE((void*)vs->state.tokens); 1444 FREE(shader); 1445} 1446 1447static void r300_set_constant_buffer(struct pipe_context *pipe, 1448 uint shader, uint index, 1449 struct pipe_buffer *buf) 1450{ 1451 struct r300_context* r300 = r300_context(pipe); 1452 struct r300_screen *r300screen = r300_screen(pipe->screen); 1453 void *mapped; 1454 int max_size = 0; 1455 1456 if (buf == NULL || buf->size == 0 || 1457 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) 1458 { 1459 r300->shader_constants[shader].count = 0; 1460 return; 1461 } 1462 1463 assert((buf->size % 4 * sizeof(float)) == 0); 1464 1465 /* Check the size of the constant buffer. */ 1466 switch (shader) { 1467 case PIPE_SHADER_VERTEX: 1468 max_size = 256; 1469 break; 1470 case PIPE_SHADER_FRAGMENT: 1471 if (r300screen->caps->is_r500) { 1472 max_size = 256; 1473 /* XXX Implement emission of r400's extended constant buffer. */ 1474 /*} else if (r300screen->caps->is_r400) { 1475 max_size = 64;*/ 1476 } else { 1477 max_size = 32; 1478 } 1479 break; 1480 default: 1481 assert(0); 1482 } 1483 1484 /* XXX Subtract immediates and RC_STATE_* variables. */ 1485 if (buf->size > (sizeof(float) * 4 * max_size)) { 1486 debug_printf("r300: Max size of the constant buffer is " 1487 "%i*4 floats.\n", max_size); 1488 abort(); 1489 } 1490 1491 memcpy(r300->shader_constants[shader].constants, mapped, buf->size); 1492 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); 1493 pipe_buffer_unmap(pipe->screen, buf); 1494 1495 if (shader == PIPE_SHADER_VERTEX) { 1496 if (r300screen->caps->has_tcl) { 1497 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1498 r300->pvs_flush.dirty = TRUE; 1499 } 1500 } 1501 else if (shader == PIPE_SHADER_FRAGMENT) 1502 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1503} 1504 1505void r300_init_state_functions(struct r300_context* r300) 1506{ 1507 r300->context.create_blend_state = r300_create_blend_state; 1508 r300->context.bind_blend_state = r300_bind_blend_state; 1509 r300->context.delete_blend_state = r300_delete_blend_state; 1510 1511 r300->context.set_blend_color = r300_set_blend_color; 1512 1513 r300->context.set_clip_state = r300_set_clip_state; 1514 1515 r300->context.set_constant_buffer = r300_set_constant_buffer; 1516 1517 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1518 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1519 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1520 1521 r300->context.set_stencil_ref = r300_set_stencil_ref; 1522 1523 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1524 1525 r300->context.create_fs_state = r300_create_fs_state; 1526 r300->context.bind_fs_state = r300_bind_fs_state; 1527 r300->context.delete_fs_state = r300_delete_fs_state; 1528 1529 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1530 1531 r300->context.create_rasterizer_state = r300_create_rs_state; 1532 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1533 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1534 1535 r300->context.create_sampler_state = r300_create_sampler_state; 1536 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1537 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1538 r300->context.delete_sampler_state = r300_delete_sampler_state; 1539 1540 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1541 r300->context.create_sampler_view = r300_create_sampler_view; 1542 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1543 1544 r300->context.set_scissor_state = r300_set_scissor_state; 1545 1546 r300->context.set_viewport_state = r300_set_viewport_state; 1547 1548 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1549 1550 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1551 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1552 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1553 1554 r300->context.create_vs_state = r300_create_vs_state; 1555 r300->context.bind_vs_state = r300_bind_vs_state; 1556 r300->context.delete_vs_state = r300_delete_vs_state; 1557} 1558