r300_state.c revision fff5be8e7b4557c221f2425dcafc2e7cbbba76ba
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 47static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 48 unsigned dstRGB, unsigned dstA) 49{ 50 /* If the blend equation is ADD or REVERSE_SUBTRACT, 51 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 52 * will not be changed. 53 * Notice that the dst factors are the src factors inverted. */ 54 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 55 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 56 srcRGB == PIPE_BLENDFACTOR_ZERO) && 57 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 58 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 59 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 60 srcA == PIPE_BLENDFACTOR_ZERO) && 61 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 62 dstRGB == PIPE_BLENDFACTOR_ONE) && 63 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 64 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 65 dstA == PIPE_BLENDFACTOR_ONE); 66} 67 68static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 69 unsigned dstRGB, unsigned dstA) 70{ 71 /* If the blend equation is ADD or REVERSE_SUBTRACT, 72 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 73 * will not be changed. 74 * Notice that the dst factors are the src factors inverted. */ 75 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 76 srcRGB == PIPE_BLENDFACTOR_ZERO) && 77 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 78 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 79 srcA == PIPE_BLENDFACTOR_ZERO) && 80 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 81 dstRGB == PIPE_BLENDFACTOR_ONE) && 82 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 83 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 84 dstA == PIPE_BLENDFACTOR_ONE); 85} 86 87static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 88 unsigned dstRGB, unsigned dstA) 89{ 90 /* If the blend equation is ADD or REVERSE_SUBTRACT, 91 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 92 * will not be changed. 93 * Notice that the dst factors are the src factors inverted. */ 94 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 95 srcRGB == PIPE_BLENDFACTOR_ZERO) && 96 (srcA == PIPE_BLENDFACTOR_ZERO) && 97 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 98 dstRGB == PIPE_BLENDFACTOR_ONE) && 99 (dstA == PIPE_BLENDFACTOR_ONE); 100} 101 102static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 103 unsigned dstRGB, unsigned dstA) 104{ 105 /* If the blend equation is ADD or REVERSE_SUBTRACT, 106 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 107 * will not be changed. 108 * Notice that the dst factors are the src factors inverted. */ 109 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 110 srcRGB == PIPE_BLENDFACTOR_ZERO) && 111 (srcA == PIPE_BLENDFACTOR_ZERO) && 112 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 113 dstRGB == PIPE_BLENDFACTOR_ONE) && 114 (dstA == PIPE_BLENDFACTOR_ONE); 115} 116 117static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 118 unsigned dstRGB, unsigned dstA) 119{ 120 /* If the blend equation is ADD or REVERSE_SUBTRACT, 121 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 122 * the colorbuffer will not be changed. 123 * Notice that the dst factors are the src factors inverted. */ 124 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 125 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 126 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 127 srcRGB == PIPE_BLENDFACTOR_ZERO) && 128 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 129 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 130 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 131 srcA == PIPE_BLENDFACTOR_ZERO) && 132 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 133 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 134 dstRGB == PIPE_BLENDFACTOR_ONE) && 135 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 136 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 137 dstA == PIPE_BLENDFACTOR_ONE); 138} 139 140static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 141 unsigned dstRGB, unsigned dstA) 142{ 143 /* If the blend equation is ADD or REVERSE_SUBTRACT, 144 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 145 * the colorbuffer will not be changed. 146 * Notice that the dst factors are the src factors inverted. */ 147 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 148 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 149 srcRGB == PIPE_BLENDFACTOR_ZERO) && 150 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 151 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 152 srcA == PIPE_BLENDFACTOR_ZERO) && 153 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 154 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 155 dstRGB == PIPE_BLENDFACTOR_ONE) && 156 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 157 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 158 dstA == PIPE_BLENDFACTOR_ONE); 159} 160 161static unsigned bgra_cmask(unsigned mask) 162{ 163 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 164 165 return ((mask & PIPE_MASK_R) << 2) | 166 ((mask & PIPE_MASK_B) >> 2) | 167 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 168} 169 170/* Create a new blend state based on the CSO blend state. 171 * 172 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 173static void* r300_create_blend_state(struct pipe_context* pipe, 174 const struct pipe_blend_state* state) 175{ 176 struct r300_screen* r300screen = r300_screen(pipe->screen); 177 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 178 179 if (state->rt[0].blend_enable) 180 { 181 unsigned eqRGB = state->rt[0].rgb_func; 182 unsigned srcRGB = state->rt[0].rgb_src_factor; 183 unsigned dstRGB = state->rt[0].rgb_dst_factor; 184 185 unsigned eqA = state->rt[0].alpha_func; 186 unsigned srcA = state->rt[0].alpha_src_factor; 187 unsigned dstA = state->rt[0].alpha_dst_factor; 188 189 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 190 * this is just the crappy D3D naming */ 191 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 192 r300_translate_blend_function(eqRGB) | 193 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 194 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 195 196 /* Optimization: some operations do not require the destination color. 197 * 198 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 199 * otherwise blending gives incorrect results. It seems to be 200 * a hardware bug. */ 201 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 202 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 203 dstRGB != PIPE_BLENDFACTOR_ZERO || 204 dstA != PIPE_BLENDFACTOR_ZERO || 205 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 206 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 207 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 208 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 209 srcA == PIPE_BLENDFACTOR_DST_COLOR || 210 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 211 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 212 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 213 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 214 /* Enable reading from the colorbuffer. */ 215 blend->blend_control |= R300_READ_ENABLE; 216 217 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) { 218 /* Optimization: Depending on incoming pixels, we can 219 * conditionally disable the reading in hardware... */ 220 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 221 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 222 /* Disable reading if SRC_ALPHA == 0. */ 223 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 224 dstRGB == PIPE_BLENDFACTOR_ZERO) && 225 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 226 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 227 dstA == PIPE_BLENDFACTOR_ZERO)) { 228 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 229 } 230 231 /* Disable reading if SRC_ALPHA == 1. */ 232 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 233 dstRGB == PIPE_BLENDFACTOR_ZERO) && 234 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 235 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 236 dstA == PIPE_BLENDFACTOR_ZERO)) { 237 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 238 } 239 } 240 } 241 } 242 243 /* Optimization: discard pixels which don't change the colorbuffer. 244 * 245 * The code below is non-trivial and some math is involved. 246 * 247 * Discarding pixels must be disabled when FP16 AA is enabled. 248 * This is a hardware bug. Also, this implementation wouldn't work 249 * with FP blending enabled and equation clamping disabled. 250 * 251 * Equations other than ADD are rarely used and therefore won't be 252 * optimized. */ 253 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 254 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 255 /* ADD: X+Y 256 * REVERSE_SUBTRACT: Y-X 257 * 258 * The idea is: 259 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 260 * then CB will not be changed. 261 * 262 * Given the srcFactor and dstFactor variables, we can derive 263 * what src and dst should be equal to and discard appropriate 264 * pixels. 265 */ 266 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 267 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 268 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 269 dstRGB, dstA)) { 270 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 271 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 272 dstRGB, dstA)) { 273 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 274 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 275 dstRGB, dstA)) { 276 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 277 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 278 dstRGB, dstA)) { 279 blend->blend_control |= 280 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 281 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 282 dstRGB, dstA)) { 283 blend->blend_control |= 284 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 285 } 286 } 287 288 /* separate alpha */ 289 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 290 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 291 blend->alpha_blend_control = 292 r300_translate_blend_function(eqA) | 293 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 294 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 295 } 296 } 297 298 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 299 if (state->logicop_enable) { 300 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 301 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 302 } 303 304 /* Color channel masks for all MRTs. */ 305 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); 306 if (r300screen->caps->is_r500 && state->independent_blend_enable) { 307 if (state->rt[1].blend_enable) { 308 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 309 } 310 if (state->rt[2].blend_enable) { 311 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 312 } 313 if (state->rt[3].blend_enable) { 314 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 315 } 316 } 317 318 if (state->dither) { 319 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 320 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 321 } 322 323 return (void*)blend; 324} 325 326/* Bind blend state. */ 327static void r300_bind_blend_state(struct pipe_context* pipe, 328 void* state) 329{ 330 struct r300_context* r300 = r300_context(pipe); 331 332 r300->blend_state.state = state; 333 r300->blend_state.dirty = TRUE; 334} 335 336/* Free blend state. */ 337static void r300_delete_blend_state(struct pipe_context* pipe, 338 void* state) 339{ 340 FREE(state); 341} 342 343/* Convert float to 10bit integer */ 344static unsigned float_to_fixed10(float f) 345{ 346 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 347} 348 349/* Set blend color. 350 * Setup both R300 and R500 registers, figure out later which one to write. */ 351static void r300_set_blend_color(struct pipe_context* pipe, 352 const struct pipe_blend_color* color) 353{ 354 struct r300_context* r300 = r300_context(pipe); 355 struct r300_screen* r300screen = r300_screen(pipe->screen); 356 struct r300_blend_color_state* state = 357 (struct r300_blend_color_state*)r300->blend_color_state.state; 358 union util_color uc; 359 360 util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); 361 state->blend_color = uc.ui; 362 363 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 364 state->blend_color_red_alpha = 365 float_to_fixed10(color->color[0]) | 366 (float_to_fixed10(color->color[3]) << 16); 367 state->blend_color_green_blue = 368 float_to_fixed10(color->color[2]) | 369 (float_to_fixed10(color->color[1]) << 16); 370 371 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; 372 r300->blend_color_state.dirty = TRUE; 373} 374 375static void r300_set_clip_state(struct pipe_context* pipe, 376 const struct pipe_clip_state* state) 377{ 378 struct r300_context* r300 = r300_context(pipe); 379 380 if (r300_screen(pipe->screen)->caps->has_tcl) { 381 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 382 r300->clip_state.size = 29; 383 } else { 384 draw_flush(r300->draw); 385 draw_set_clip_state(r300->draw, state); 386 r300->clip_state.size = 2; 387 } 388 389 r300->clip_state.dirty = TRUE; 390} 391 392/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 393 * 394 * This contains the depth buffer, stencil buffer, alpha test, and such. 395 * On the Radeon, depth and stencil buffer setup are intertwined, which is 396 * the reason for some of the strange-looking assignments across registers. */ 397static void* 398 r300_create_dsa_state(struct pipe_context* pipe, 399 const struct pipe_depth_stencil_alpha_state* state) 400{ 401 struct r300_capabilities *caps = 402 r300_screen(r300_context(pipe)->context.screen)->caps; 403 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 404 405 /* Depth test setup. */ 406 if (state->depth.enabled) { 407 dsa->z_buffer_control |= R300_Z_ENABLE; 408 409 if (state->depth.writemask) { 410 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 411 } 412 413 dsa->z_stencil_control |= 414 (r300_translate_depth_stencil_function(state->depth.func) << 415 R300_Z_FUNC_SHIFT); 416 } 417 418 /* Stencil buffer setup. */ 419 if (state->stencil[0].enabled) { 420 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 421 dsa->z_stencil_control |= 422 (r300_translate_depth_stencil_function(state->stencil[0].func) << 423 R300_S_FRONT_FUNC_SHIFT) | 424 (r300_translate_stencil_op(state->stencil[0].fail_op) << 425 R300_S_FRONT_SFAIL_OP_SHIFT) | 426 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 427 R300_S_FRONT_ZPASS_OP_SHIFT) | 428 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 429 R300_S_FRONT_ZFAIL_OP_SHIFT); 430 431 dsa->stencil_ref_mask = 432 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 433 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 434 435 if (state->stencil[1].enabled) { 436 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 437 dsa->z_stencil_control |= 438 (r300_translate_depth_stencil_function(state->stencil[1].func) << 439 R300_S_BACK_FUNC_SHIFT) | 440 (r300_translate_stencil_op(state->stencil[1].fail_op) << 441 R300_S_BACK_SFAIL_OP_SHIFT) | 442 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 443 R300_S_BACK_ZPASS_OP_SHIFT) | 444 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 445 R300_S_BACK_ZFAIL_OP_SHIFT); 446 447 if (caps->is_r500) 448 { 449 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 450 dsa->stencil_ref_bf = 451 (state->stencil[1].valuemask << 452 R300_STENCILMASK_SHIFT) | 453 (state->stencil[1].writemask << 454 R300_STENCILWRITEMASK_SHIFT); 455 } 456 } 457 } 458 459 /* Alpha test setup. */ 460 if (state->alpha.enabled) { 461 dsa->alpha_function = 462 r300_translate_alpha_function(state->alpha.func) | 463 R300_FG_ALPHA_FUNC_ENABLE; 464 465 /* We could use 10bit alpha ref but who needs that? */ 466 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 467 468 if (caps->is_r500) 469 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 470 } 471 472 return (void*)dsa; 473} 474 475/* Bind DSA state. */ 476static void r300_bind_dsa_state(struct pipe_context* pipe, 477 void* state) 478{ 479 struct r300_context* r300 = r300_context(pipe); 480 struct r300_screen* r300screen = r300_screen(pipe->screen); 481 482 r300->dsa_state.state = state; 483 r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; 484 r300->dsa_state.dirty = TRUE; 485} 486 487/* Free DSA state. */ 488static void r300_delete_dsa_state(struct pipe_context* pipe, 489 void* state) 490{ 491 FREE(state); 492} 493 494static void r300_set_stencil_ref(struct pipe_context* pipe, 495 const struct pipe_stencil_ref* sr) 496{ 497 struct r300_context* r300 = r300_context(pipe); 498 r300->stencil_ref = *sr; 499 r300->dsa_state.dirty = TRUE; 500} 501 502/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 503static void r300_fb_update_tiling_flags(struct r300_context *r300, 504 const struct pipe_framebuffer_state *old_state, 505 const struct pipe_framebuffer_state *new_state) 506{ 507 struct r300_texture *tex; 508 unsigned i, j, level; 509 510 /* Reset tiling flags for old surfaces to default values. */ 511 for (i = 0; i < old_state->nr_cbufs; i++) { 512 for (j = 0; j < new_state->nr_cbufs; j++) { 513 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { 514 break; 515 } 516 } 517 /* If not binding the surface again... */ 518 if (j != new_state->nr_cbufs) { 519 continue; 520 } 521 522 tex = (struct r300_texture*)old_state->cbufs[i]->texture; 523 524 if (tex) { 525 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 526 tex->pitch[0], 527 tex->microtile != 0, 528 tex->macrotile != 0); 529 } 530 } 531 if (old_state->zsbuf && 532 (!new_state->zsbuf || 533 old_state->zsbuf->texture != new_state->zsbuf->texture)) { 534 tex = (struct r300_texture*)old_state->zsbuf->texture; 535 536 if (tex) { 537 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 538 tex->pitch[0], 539 tex->microtile != 0, 540 tex->macrotile != 0); 541 } 542 } 543 544 /* Set tiling flags for new surfaces. */ 545 for (i = 0; i < new_state->nr_cbufs; i++) { 546 tex = (struct r300_texture*)new_state->cbufs[i]->texture; 547 level = new_state->cbufs[i]->level; 548 549 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 550 tex->pitch[level], 551 tex->microtile != 0, 552 tex->mip_macrotile[level] != 0); 553 } 554 if (new_state->zsbuf) { 555 tex = (struct r300_texture*)new_state->zsbuf->texture; 556 level = new_state->zsbuf->level; 557 558 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 559 tex->pitch[level], 560 tex->microtile != 0, 561 tex->mip_macrotile[level] != 0); 562 } 563} 564 565static void 566 r300_set_framebuffer_state(struct pipe_context* pipe, 567 const struct pipe_framebuffer_state* state) 568{ 569 struct r300_context* r300 = r300_context(pipe); 570 struct r300_screen* r300screen = r300_screen(pipe->screen); 571 unsigned max_width, max_height; 572 uint32_t zbuffer_bpp = 0; 573 574 575 if (state->nr_cbufs > 4) { 576 debug_printf("r300: Implementation error: Too many MRTs in %s, " 577 "refusing to bind framebuffer state!\n", __FUNCTION__); 578 return; 579 } 580 581 if (r300screen->caps->is_r500) { 582 max_width = max_height = 4096; 583 } else if (r300screen->caps->is_r400) { 584 max_width = max_height = 4021; 585 } else { 586 max_width = max_height = 2560; 587 } 588 589 if (state->width > max_width || state->height > max_height) { 590 debug_printf("r300: Implementation error: Render targets are too " 591 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 592 return; 593 } 594 595 596 if (r300->draw) { 597 draw_flush(r300->draw); 598 } 599 600 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 601 602 r300->fb_state.size = (10 * state->nr_cbufs) + (state->zsbuf ? 10 : 0) + 6; 603 604 r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); 605 606 /* XXX wait what */ 607 r300->blend_state.dirty = TRUE; 608 r300->dsa_state.dirty = TRUE; 609 r300->fb_state.dirty = TRUE; 610 r300->scissor_state.dirty = TRUE; 611 612 /* Polygon offset depends on the zbuffer bit depth. */ 613 if (state->zsbuf && r300->polygon_offset_enabled) { 614 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 615 case 2: 616 zbuffer_bpp = 16; 617 break; 618 case 4: 619 zbuffer_bpp = 24; 620 break; 621 } 622 623 if (r300->zbuffer_bpp != zbuffer_bpp) { 624 r300->zbuffer_bpp = zbuffer_bpp; 625 r300->rs_state.dirty = TRUE; 626 } 627 } 628} 629 630/* Create fragment shader state. */ 631static void* r300_create_fs_state(struct pipe_context* pipe, 632 const struct pipe_shader_state* shader) 633{ 634 struct r300_fragment_shader* fs = NULL; 635 636 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 637 638 /* Copy state directly into shader. */ 639 fs->state = *shader; 640 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 641 642 tgsi_scan_shader(shader->tokens, &fs->info); 643 r300_shader_read_fs_inputs(&fs->info, &fs->inputs); 644 645 return (void*)fs; 646} 647 648/* Bind fragment shader state. */ 649static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 650{ 651 struct r300_context* r300 = r300_context(pipe); 652 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 653 654 if (fs == NULL) { 655 r300->fs = NULL; 656 return; 657 } 658 659 r300->fs = fs; 660 r300_pick_fragment_shader(r300); 661 662 if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { 663 r300->vertex_format_state.dirty = TRUE; 664 } 665 666 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; 667} 668 669/* Delete fragment shader state. */ 670static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 671{ 672 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 673 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 674 675 while (ptr) { 676 tmp = ptr; 677 ptr = ptr->next; 678 rc_constants_destroy(&tmp->code.constants); 679 FREE(tmp); 680 } 681 FREE((void*)fs->state.tokens); 682 FREE(shader); 683} 684 685static void r300_set_polygon_stipple(struct pipe_context* pipe, 686 const struct pipe_poly_stipple* state) 687{ 688 /* XXX no idea how to set this up, but not terribly important */ 689} 690 691/* Create a new rasterizer state based on the CSO rasterizer state. 692 * 693 * This is a very large chunk of state, and covers most of the graphics 694 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 695 * 696 * In a not entirely unironic sidenote, this state has nearly nothing to do 697 * with the actual block on the Radeon called the rasterizer (RS). */ 698static void* r300_create_rs_state(struct pipe_context* pipe, 699 const struct pipe_rasterizer_state* state) 700{ 701 struct r300_screen* r300screen = r300_screen(pipe->screen); 702 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 703 704 /* Copy rasterizer state for Draw. */ 705 rs->rs = *state; 706 707#ifdef PIPE_ARCH_LITTLE_ENDIAN 708 rs->vap_control_status = R300_VC_NO_SWAP; 709#else 710 rs->vap_control_status = R300_VC_32BIT_SWAP; 711#endif 712 713 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. 714 * Else, enable HW TCL and force Draw's TCL off. */ 715 if (state->bypass_vs_clip_and_viewport || 716 !r300screen->caps->has_tcl) { 717 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 718 } 719 720 rs->point_size = pack_float_16_6x(state->point_size) | 721 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 722 723 /* Point minimum and maximum sizes. This register has to be emitted, 724 * and it'd be a step backwards to put it in invariant state. */ 725 if (r300screen->caps->is_r500) { 726 rs->point_minmax = 727 ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 728 ((int)(4096.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); 729 } else if (r300screen->caps->is_r400) { 730 rs->point_minmax = 731 ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 732 ((int)(4021.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); 733 } else { 734 rs->point_minmax = 735 ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 736 ((int)(2560.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); 737 } 738 739 rs->line_control = pack_float_16_6x(state->line_width) | 740 R300_GA_LINE_CNTL_END_TYPE_COMP; 741 742 /* Enable polygon mode */ 743 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 744 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 745 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 746 } 747 748 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 749 if (state->front_winding == PIPE_WINDING_CW) { 750 rs->cull_mode = R300_FRONT_FACE_CW; 751 752 /* Polygon offset */ 753 if (state->offset_cw) { 754 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 755 } 756 if (state->offset_ccw) { 757 rs->polygon_offset_enable |= R300_BACK_ENABLE; 758 } 759 760 /* Polygon mode */ 761 if (rs->polygon_mode) { 762 rs->polygon_mode |= 763 r300_translate_polygon_mode_front(state->fill_cw); 764 rs->polygon_mode |= 765 r300_translate_polygon_mode_back(state->fill_ccw); 766 } 767 } else { 768 rs->cull_mode = R300_FRONT_FACE_CCW; 769 770 /* Polygon offset */ 771 if (state->offset_ccw) { 772 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 773 } 774 if (state->offset_cw) { 775 rs->polygon_offset_enable |= R300_BACK_ENABLE; 776 } 777 778 /* Polygon mode */ 779 if (rs->polygon_mode) { 780 rs->polygon_mode |= 781 r300_translate_polygon_mode_front(state->fill_ccw); 782 rs->polygon_mode |= 783 r300_translate_polygon_mode_back(state->fill_cw); 784 } 785 } 786 if (state->front_winding & state->cull_mode) { 787 rs->cull_mode |= R300_CULL_FRONT; 788 } 789 if (~(state->front_winding) & state->cull_mode) { 790 rs->cull_mode |= R300_CULL_BACK; 791 } 792 793 if (rs->polygon_offset_enable) { 794 rs->depth_offset = state->offset_units; 795 rs->depth_scale = state->offset_scale; 796 } 797 798 if (state->line_stipple_enable) { 799 rs->line_stipple_config = 800 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 801 (fui((float)state->line_stipple_factor) & 802 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 803 /* XXX this might need to be scaled up */ 804 rs->line_stipple_value = state->line_stipple_pattern; 805 } 806 807 if (state->flatshade) { 808 rs->color_control = R300_SHADE_MODEL_FLAT; 809 } else { 810 rs->color_control = R300_SHADE_MODEL_SMOOTH; 811 } 812 813 return (void*)rs; 814} 815 816/* Bind rasterizer state. */ 817static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 818{ 819 struct r300_context* r300 = r300_context(pipe); 820 struct r300_rs_state* rs = (struct r300_rs_state*)state; 821 822 if (r300->draw) { 823 draw_flush(r300->draw); 824 draw_set_rasterizer_state(r300->draw, &rs->rs); 825 } 826 827 if (rs) { 828 r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; 829 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 830 } else { 831 r300->tcl_bypass = FALSE; 832 r300->polygon_offset_enabled = FALSE; 833 } 834 835 r300->rs_state.state = rs; 836 r300->rs_state.dirty = TRUE; 837 /* XXX Why is this still needed, dammit!? */ 838 r300->scissor_state.dirty = TRUE; 839 r300->viewport_state.dirty = TRUE; 840 841 /* XXX Clean these up when we move to atom emits */ 842 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 843 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 844 } 845} 846 847/* Free rasterizer state. */ 848static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 849{ 850 FREE(state); 851} 852 853static void* 854 r300_create_sampler_state(struct pipe_context* pipe, 855 const struct pipe_sampler_state* state) 856{ 857 struct r300_context* r300 = r300_context(pipe); 858 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 859 int lod_bias; 860 union util_color uc; 861 862 sampler->state = *state; 863 864 sampler->filter0 |= 865 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 866 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 867 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 868 869 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 870 state->mag_img_filter, 871 state->min_mip_filter, 872 state->max_anisotropy > 0); 873 874 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 875 /* We must pass these to the emit function to clamp them properly. */ 876 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 877 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 878 879 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 880 881 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 882 883 sampler->filter1 |= r300_anisotropy(state->max_anisotropy); 884 885 util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); 886 sampler->border_color = uc.ui; 887 888 /* R500-specific fixups and optimizations */ 889 if (r300_screen(r300->context.screen)->caps->is_r500) { 890 sampler->filter1 |= R500_BORDER_FIX; 891 } 892 893 return (void*)sampler; 894} 895 896static void r300_bind_sampler_states(struct pipe_context* pipe, 897 unsigned count, 898 void** states) 899{ 900 struct r300_context* r300 = r300_context(pipe); 901 int i; 902 903 if (count > 8) { 904 return; 905 } 906 907 for (i = 0; i < count; i++) { 908 if (r300->sampler_states[i] != states[i]) { 909 r300->sampler_states[i] = (struct r300_sampler_state*)states[i]; 910 r300->dirty_state |= (R300_NEW_SAMPLER << i); 911 } 912 } 913 914 r300->sampler_count = count; 915 916 /* Pick a fragment shader based on the texture compare state. */ 917 if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) { 918 if (r300_pick_fragment_shader(r300)) { 919 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | 920 R300_NEW_FRAGMENT_SHADER_CONSTANTS; 921 } 922 } 923} 924 925static void r300_lacks_vertex_textures(struct pipe_context* pipe, 926 unsigned count, 927 void** states) 928{ 929} 930 931static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 932{ 933 FREE(state); 934} 935 936static void r300_set_sampler_textures(struct pipe_context* pipe, 937 unsigned count, 938 struct pipe_texture** texture) 939{ 940 struct r300_context* r300 = r300_context(pipe); 941 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; 942 int i; 943 944 /* XXX magic num */ 945 if (count > 8) { 946 return; 947 } 948 949 for (i = 0; i < count; i++) { 950 if (r300->textures[i] != (struct r300_texture*)texture[i]) { 951 pipe_texture_reference((struct pipe_texture**)&r300->textures[i], 952 texture[i]); 953 r300->dirty_state |= (R300_NEW_TEXTURE << i); 954 955 /* R300-specific - set the texrect factor in a fragment shader */ 956 if (!is_r500 && r300->textures[i]->is_npot) { 957 /* XXX It would be nice to re-emit just 1 constant, 958 * XXX not all of them */ 959 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 960 } 961 } 962 } 963 964 for (i = count; i < 8; i++) { 965 if (r300->textures[i]) { 966 pipe_texture_reference((struct pipe_texture**)&r300->textures[i], 967 NULL); 968 r300->dirty_state |= (R300_NEW_TEXTURE << i); 969 } 970 } 971 972 r300->texture_count = count; 973} 974 975static void r300_set_scissor_state(struct pipe_context* pipe, 976 const struct pipe_scissor_state* state) 977{ 978 struct r300_context* r300 = r300_context(pipe); 979 980 memcpy(r300->scissor_state.state, state, 981 sizeof(struct pipe_scissor_state)); 982 983 r300->scissor_state.dirty = TRUE; 984} 985 986static void r300_set_viewport_state(struct pipe_context* pipe, 987 const struct pipe_viewport_state* state) 988{ 989 struct r300_context* r300 = r300_context(pipe); 990 struct r300_viewport_state* viewport = 991 (struct r300_viewport_state*)r300->viewport_state.state; 992 993 /* Do the transform in HW. */ 994 viewport->vte_control = R300_VTX_W0_FMT; 995 996 if (state->scale[0] != 1.0f) { 997 viewport->xscale = state->scale[0]; 998 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 999 } 1000 if (state->scale[1] != 1.0f) { 1001 viewport->yscale = state->scale[1]; 1002 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1003 } 1004 if (state->scale[2] != 1.0f) { 1005 viewport->zscale = state->scale[2]; 1006 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1007 } 1008 if (state->translate[0] != 0.0f) { 1009 viewport->xoffset = state->translate[0]; 1010 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1011 } 1012 if (state->translate[1] != 0.0f) { 1013 viewport->yoffset = state->translate[1]; 1014 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1015 } 1016 if (state->translate[2] != 0.0f) { 1017 viewport->zoffset = state->translate[2]; 1018 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1019 } 1020 1021 r300->viewport_state.dirty = TRUE; 1022 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 1023 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1024 } 1025} 1026 1027static void r300_set_vertex_buffers(struct pipe_context* pipe, 1028 unsigned count, 1029 const struct pipe_vertex_buffer* buffers) 1030{ 1031 struct r300_context* r300 = r300_context(pipe); 1032 boolean any_user_buffer = false; 1033 int i; 1034 1035 if (count == r300->vertex_buffer_count && 1036 memcmp(r300->vertex_buffer, buffers, count * sizeof(buffers[0])) == 0) 1037 return; 1038 1039 for (i = 0; i < count; i++) { 1040 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); 1041 if (r300_buffer_is_user_buffer(buffers[i].buffer)) 1042 any_user_buffer = true; 1043 } 1044 1045 for ( ; i < r300->vertex_buffer_count; i++) 1046 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, NULL); 1047 1048 memcpy(r300->vertex_buffer, buffers, 1049 sizeof(struct pipe_vertex_buffer) * count); 1050 r300->vertex_buffer_count = count; 1051 r300->any_user_vbs = any_user_buffer; 1052 1053 if (r300->draw) { 1054 draw_flush(r300->draw); 1055 draw_set_vertex_buffers(r300->draw, count, buffers); 1056 } 1057 1058 r300->vertex_format_state.dirty = TRUE; 1059} 1060 1061static boolean r300_validate_aos(struct r300_context *r300) 1062{ 1063 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; 1064 struct pipe_vertex_element *velem = r300->vertex_element; 1065 int i; 1066 1067 /* Check if formats and strides are aligned to the size of DWORD. */ 1068 for (i = 0; i < r300->vertex_element_count; i++) { 1069 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || 1070 util_format_get_blocksize(velem[i].src_format) % 4 != 0) { 1071 return FALSE; 1072 } 1073 } 1074 return TRUE; 1075} 1076 1077static void r300_set_vertex_elements(struct pipe_context* pipe, 1078 unsigned count, 1079 const struct pipe_vertex_element* elements) 1080{ 1081 struct r300_context* r300 = r300_context(pipe); 1082 1083 memcpy(r300->vertex_element, 1084 elements, 1085 sizeof(struct pipe_vertex_element) * count); 1086 r300->vertex_element_count = count; 1087 1088 if (r300->draw) { 1089 draw_flush(r300->draw); 1090 draw_set_vertex_elements(r300->draw, count, elements); 1091 } 1092 1093 if (!r300_validate_aos(r300)) { 1094 /* XXX We should fallback using draw. */ 1095 assert(0); 1096 abort(); 1097 } 1098} 1099 1100static void* r300_create_vs_state(struct pipe_context* pipe, 1101 const struct pipe_shader_state* shader) 1102{ 1103 struct r300_context* r300 = r300_context(pipe); 1104 1105 if (r300_screen(pipe->screen)->caps->has_tcl) { 1106 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1107 /* Copy state directly into shader. */ 1108 vs->state = *shader; 1109 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1110 1111 tgsi_scan_shader(shader->tokens, &vs->info); 1112 1113 return (void*)vs; 1114 } else { 1115 return draw_create_vertex_shader(r300->draw, shader); 1116 } 1117} 1118 1119static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1120{ 1121 struct r300_context* r300 = r300_context(pipe); 1122 1123 if (r300_screen(pipe->screen)->caps->has_tcl) { 1124 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1125 1126 if (vs == NULL) { 1127 r300->vs = NULL; 1128 return; 1129 } else if (!vs->translated) { 1130 r300_translate_vertex_shader(r300, vs); 1131 } 1132 1133 r300->vs = vs; 1134 if (r300->fs) { 1135 r300_vertex_shader_setup_wpos(r300); 1136 } 1137 1138 r300->vertex_format_state.dirty = TRUE; 1139 1140 r300->dirty_state |= 1141 R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; 1142 } else { 1143 draw_flush(r300->draw); 1144 draw_bind_vertex_shader(r300->draw, 1145 (struct draw_vertex_shader*)shader); 1146 } 1147} 1148 1149static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1150{ 1151 struct r300_context* r300 = r300_context(pipe); 1152 1153 if (r300_screen(pipe->screen)->caps->has_tcl) { 1154 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1155 1156 rc_constants_destroy(&vs->code.constants); 1157 FREE((void*)vs->state.tokens); 1158 FREE(shader); 1159 } else { 1160 draw_delete_vertex_shader(r300->draw, 1161 (struct draw_vertex_shader*)shader); 1162 } 1163} 1164 1165static void r300_set_constant_buffer(struct pipe_context *pipe, 1166 uint shader, uint index, 1167 struct pipe_buffer *buf) 1168{ 1169 struct r300_context* r300 = r300_context(pipe); 1170 struct r300_screen *r300screen = r300_screen(pipe->screen); 1171 void *mapped; 1172 int max_size = 0; 1173 1174 if (buf == NULL || buf->size == 0 || 1175 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) 1176 { 1177 r300->shader_constants[shader].count = 0; 1178 return; 1179 } 1180 1181 assert((buf->size % 4 * sizeof(float)) == 0); 1182 1183 /* Check the size of the constant buffer. */ 1184 switch (shader) { 1185 case PIPE_SHADER_VERTEX: 1186 max_size = 256; 1187 break; 1188 case PIPE_SHADER_FRAGMENT: 1189 if (r300screen->caps->is_r500) { 1190 max_size = 256; 1191 /* XXX Implement emission of r400's extended constant buffer. */ 1192 /*} else if (r300screen->caps->is_r400) { 1193 max_size = 64;*/ 1194 } else { 1195 max_size = 32; 1196 } 1197 break; 1198 default: 1199 assert(0); 1200 } 1201 1202 /* XXX Subtract immediates and RC_STATE_* variables. */ 1203 if (buf->size > (sizeof(float) * 4 * max_size)) { 1204 debug_printf("r300: Max size of the constant buffer is " 1205 "%i*4 floats.\n", max_size); 1206 abort(); 1207 } 1208 1209 memcpy(r300->shader_constants[shader].constants, mapped, buf->size); 1210 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); 1211 pipe_buffer_unmap(pipe->screen, buf); 1212 1213 if (shader == PIPE_SHADER_VERTEX) 1214 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1215 else if (shader == PIPE_SHADER_FRAGMENT) 1216 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1217} 1218 1219void r300_init_state_functions(struct r300_context* r300) 1220{ 1221 r300->context.create_blend_state = r300_create_blend_state; 1222 r300->context.bind_blend_state = r300_bind_blend_state; 1223 r300->context.delete_blend_state = r300_delete_blend_state; 1224 1225 r300->context.set_blend_color = r300_set_blend_color; 1226 1227 r300->context.set_clip_state = r300_set_clip_state; 1228 1229 r300->context.set_constant_buffer = r300_set_constant_buffer; 1230 1231 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1232 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1233 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1234 1235 r300->context.set_stencil_ref = r300_set_stencil_ref; 1236 1237 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1238 1239 r300->context.create_fs_state = r300_create_fs_state; 1240 r300->context.bind_fs_state = r300_bind_fs_state; 1241 r300->context.delete_fs_state = r300_delete_fs_state; 1242 1243 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1244 1245 r300->context.create_rasterizer_state = r300_create_rs_state; 1246 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1247 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1248 1249 r300->context.create_sampler_state = r300_create_sampler_state; 1250 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1251 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1252 r300->context.delete_sampler_state = r300_delete_sampler_state; 1253 1254 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures; 1255 1256 r300->context.set_scissor_state = r300_set_scissor_state; 1257 1258 r300->context.set_viewport_state = r300_set_viewport_state; 1259 1260 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1261 r300->context.set_vertex_elements = r300_set_vertex_elements; 1262 1263 r300->context.create_vs_state = r300_create_vs_state; 1264 r300->context.bind_vs_state = r300_bind_vs_state; 1265 r300->context.delete_vs_state = r300_delete_vs_state; 1266} 1267