r300_state.c revision 587c5ef01fad962bcfb087232d0925a8eec86953
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#include "r300_winsys.h" 42 43/* r300_state: Functions used to intialize state context by translating 44 * Gallium state objects into semi-native r300 state objects. */ 45 46#define UPDATE_STATE(cso, atom) \ 47 if (cso != atom.state) { \ 48 atom.state = cso; \ 49 atom.dirty = TRUE; \ 50 } 51 52static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 53 unsigned dstRGB, unsigned dstA) 54{ 55 /* If the blend equation is ADD or REVERSE_SUBTRACT, 56 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 57 * will not be changed. 58 * Notice that the dst factors are the src factors inverted. */ 59 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 60 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 61 srcRGB == PIPE_BLENDFACTOR_ZERO) && 62 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 63 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 64 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 65 srcA == PIPE_BLENDFACTOR_ZERO) && 66 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 67 dstRGB == PIPE_BLENDFACTOR_ONE) && 68 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 69 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 70 dstA == PIPE_BLENDFACTOR_ONE); 71} 72 73static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 74 unsigned dstRGB, unsigned dstA) 75{ 76 /* If the blend equation is ADD or REVERSE_SUBTRACT, 77 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 78 * will not be changed. 79 * Notice that the dst factors are the src factors inverted. */ 80 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 81 srcRGB == PIPE_BLENDFACTOR_ZERO) && 82 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 83 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 84 srcA == PIPE_BLENDFACTOR_ZERO) && 85 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 86 dstRGB == PIPE_BLENDFACTOR_ONE) && 87 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 88 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 89 dstA == PIPE_BLENDFACTOR_ONE); 90} 91 92static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 93 unsigned dstRGB, unsigned dstA) 94{ 95 /* If the blend equation is ADD or REVERSE_SUBTRACT, 96 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 97 * will not be changed. 98 * Notice that the dst factors are the src factors inverted. */ 99 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 100 srcRGB == PIPE_BLENDFACTOR_ZERO) && 101 (srcA == PIPE_BLENDFACTOR_ZERO) && 102 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 103 dstRGB == PIPE_BLENDFACTOR_ONE) && 104 (dstA == PIPE_BLENDFACTOR_ONE); 105} 106 107static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 108 unsigned dstRGB, unsigned dstA) 109{ 110 /* If the blend equation is ADD or REVERSE_SUBTRACT, 111 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 112 * will not be changed. 113 * Notice that the dst factors are the src factors inverted. */ 114 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 115 srcRGB == PIPE_BLENDFACTOR_ZERO) && 116 (srcA == PIPE_BLENDFACTOR_ZERO) && 117 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 118 dstRGB == PIPE_BLENDFACTOR_ONE) && 119 (dstA == PIPE_BLENDFACTOR_ONE); 120} 121 122static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 123 unsigned dstRGB, unsigned dstA) 124{ 125 /* If the blend equation is ADD or REVERSE_SUBTRACT, 126 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 127 * the colorbuffer will not be changed. 128 * Notice that the dst factors are the src factors inverted. */ 129 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 130 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 131 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 132 srcRGB == PIPE_BLENDFACTOR_ZERO) && 133 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 134 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 135 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 136 srcA == PIPE_BLENDFACTOR_ZERO) && 137 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 138 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 139 dstRGB == PIPE_BLENDFACTOR_ONE) && 140 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 141 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 142 dstA == PIPE_BLENDFACTOR_ONE); 143} 144 145static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 146 unsigned dstRGB, unsigned dstA) 147{ 148 /* If the blend equation is ADD or REVERSE_SUBTRACT, 149 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 150 * the colorbuffer will not be changed. 151 * Notice that the dst factors are the src factors inverted. */ 152 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 153 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 154 srcRGB == PIPE_BLENDFACTOR_ZERO) && 155 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 156 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 157 srcA == PIPE_BLENDFACTOR_ZERO) && 158 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 159 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 160 dstRGB == PIPE_BLENDFACTOR_ONE) && 161 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 162 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 163 dstA == PIPE_BLENDFACTOR_ONE); 164} 165 166static unsigned bgra_cmask(unsigned mask) 167{ 168 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 169 170 return ((mask & PIPE_MASK_R) << 2) | 171 ((mask & PIPE_MASK_B) >> 2) | 172 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 173} 174 175/* Create a new blend state based on the CSO blend state. 176 * 177 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 178static void* r300_create_blend_state(struct pipe_context* pipe, 179 const struct pipe_blend_state* state) 180{ 181 struct r300_screen* r300screen = r300_screen(pipe->screen); 182 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 183 184 if (state->rt[0].blend_enable) 185 { 186 unsigned eqRGB = state->rt[0].rgb_func; 187 unsigned srcRGB = state->rt[0].rgb_src_factor; 188 unsigned dstRGB = state->rt[0].rgb_dst_factor; 189 190 unsigned eqA = state->rt[0].alpha_func; 191 unsigned srcA = state->rt[0].alpha_src_factor; 192 unsigned dstA = state->rt[0].alpha_dst_factor; 193 194 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 195 * this is just the crappy D3D naming */ 196 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 197 r300_translate_blend_function(eqRGB) | 198 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 199 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 200 201 /* Optimization: some operations do not require the destination color. 202 * 203 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 204 * otherwise blending gives incorrect results. It seems to be 205 * a hardware bug. */ 206 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 207 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 208 dstRGB != PIPE_BLENDFACTOR_ZERO || 209 dstA != PIPE_BLENDFACTOR_ZERO || 210 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 211 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 212 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 213 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 214 srcA == PIPE_BLENDFACTOR_DST_COLOR || 215 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 216 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 217 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 218 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 219 /* Enable reading from the colorbuffer. */ 220 blend->blend_control |= R300_READ_ENABLE; 221 222 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) { 223 /* Optimization: Depending on incoming pixels, we can 224 * conditionally disable the reading in hardware... */ 225 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 226 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 227 /* Disable reading if SRC_ALPHA == 0. */ 228 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 229 dstRGB == PIPE_BLENDFACTOR_ZERO) && 230 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 231 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 232 dstA == PIPE_BLENDFACTOR_ZERO)) { 233 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 234 } 235 236 /* Disable reading if SRC_ALPHA == 1. */ 237 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 238 dstRGB == PIPE_BLENDFACTOR_ZERO) && 239 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 240 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 241 dstA == PIPE_BLENDFACTOR_ZERO)) { 242 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 243 } 244 } 245 } 246 } 247 248 /* Optimization: discard pixels which don't change the colorbuffer. 249 * 250 * The code below is non-trivial and some math is involved. 251 * 252 * Discarding pixels must be disabled when FP16 AA is enabled. 253 * This is a hardware bug. Also, this implementation wouldn't work 254 * with FP blending enabled and equation clamping disabled. 255 * 256 * Equations other than ADD are rarely used and therefore won't be 257 * optimized. */ 258 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 259 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 260 /* ADD: X+Y 261 * REVERSE_SUBTRACT: Y-X 262 * 263 * The idea is: 264 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 265 * then CB will not be changed. 266 * 267 * Given the srcFactor and dstFactor variables, we can derive 268 * what src and dst should be equal to and discard appropriate 269 * pixels. 270 */ 271 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 272 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 273 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 274 dstRGB, dstA)) { 275 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 276 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 277 dstRGB, dstA)) { 278 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 279 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 280 dstRGB, dstA)) { 281 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 282 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 283 dstRGB, dstA)) { 284 blend->blend_control |= 285 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 286 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 287 dstRGB, dstA)) { 288 blend->blend_control |= 289 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 290 } 291 } 292 293 /* separate alpha */ 294 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 295 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 296 blend->alpha_blend_control = 297 r300_translate_blend_function(eqA) | 298 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 299 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 300 } 301 } 302 303 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 304 if (state->logicop_enable) { 305 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 306 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 307 } 308 309 /* Color channel masks for all MRTs. */ 310 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); 311 if (r300screen->caps->is_r500 && state->independent_blend_enable) { 312 if (state->rt[1].blend_enable) { 313 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 314 } 315 if (state->rt[2].blend_enable) { 316 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 317 } 318 if (state->rt[3].blend_enable) { 319 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 320 } 321 } 322 323 if (state->dither) { 324 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 325 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 326 } 327 328 return (void*)blend; 329} 330 331/* Bind blend state. */ 332static void r300_bind_blend_state(struct pipe_context* pipe, 333 void* state) 334{ 335 struct r300_context* r300 = r300_context(pipe); 336 337 UPDATE_STATE(state, r300->blend_state); 338} 339 340/* Free blend state. */ 341static void r300_delete_blend_state(struct pipe_context* pipe, 342 void* state) 343{ 344 FREE(state); 345} 346 347/* Convert float to 10bit integer */ 348static unsigned float_to_fixed10(float f) 349{ 350 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 351} 352 353/* Set blend color. 354 * Setup both R300 and R500 registers, figure out later which one to write. */ 355static void r300_set_blend_color(struct pipe_context* pipe, 356 const struct pipe_blend_color* color) 357{ 358 struct r300_context* r300 = r300_context(pipe); 359 struct r300_screen* r300screen = r300_screen(pipe->screen); 360 struct r300_blend_color_state* state = 361 (struct r300_blend_color_state*)r300->blend_color_state.state; 362 union util_color uc; 363 364 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 365 state->blend_color = uc.ui; 366 367 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 368 state->blend_color_red_alpha = 369 float_to_fixed10(color->color[0]) | 370 (float_to_fixed10(color->color[3]) << 16); 371 state->blend_color_green_blue = 372 float_to_fixed10(color->color[2]) | 373 (float_to_fixed10(color->color[1]) << 16); 374 375 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; 376 r300->blend_color_state.dirty = TRUE; 377} 378 379static void r300_set_clip_state(struct pipe_context* pipe, 380 const struct pipe_clip_state* state) 381{ 382 struct r300_context* r300 = r300_context(pipe); 383 384 r300->clip = *state; 385 386 if (r300_screen(pipe->screen)->caps->has_tcl) { 387 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 388 r300->clip_state.size = 29; 389 } else { 390 draw_flush(r300->draw); 391 draw_set_clip_state(r300->draw, state); 392 r300->clip_state.size = 2; 393 } 394 395 r300->clip_state.dirty = TRUE; 396} 397 398/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 399 * 400 * This contains the depth buffer, stencil buffer, alpha test, and such. 401 * On the Radeon, depth and stencil buffer setup are intertwined, which is 402 * the reason for some of the strange-looking assignments across registers. */ 403static void* 404 r300_create_dsa_state(struct pipe_context* pipe, 405 const struct pipe_depth_stencil_alpha_state* state) 406{ 407 struct r300_capabilities *caps = 408 r300_screen(r300_context(pipe)->context.screen)->caps; 409 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 410 411 /* Depth test setup. */ 412 if (state->depth.enabled) { 413 dsa->z_buffer_control |= R300_Z_ENABLE; 414 415 if (state->depth.writemask) { 416 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 417 } 418 419 dsa->z_stencil_control |= 420 (r300_translate_depth_stencil_function(state->depth.func) << 421 R300_Z_FUNC_SHIFT); 422 } 423 424 /* Stencil buffer setup. */ 425 if (state->stencil[0].enabled) { 426 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 427 dsa->z_stencil_control |= 428 (r300_translate_depth_stencil_function(state->stencil[0].func) << 429 R300_S_FRONT_FUNC_SHIFT) | 430 (r300_translate_stencil_op(state->stencil[0].fail_op) << 431 R300_S_FRONT_SFAIL_OP_SHIFT) | 432 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 433 R300_S_FRONT_ZPASS_OP_SHIFT) | 434 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 435 R300_S_FRONT_ZFAIL_OP_SHIFT); 436 437 dsa->stencil_ref_mask = 438 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 439 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 440 441 if (state->stencil[1].enabled) { 442 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 443 dsa->z_stencil_control |= 444 (r300_translate_depth_stencil_function(state->stencil[1].func) << 445 R300_S_BACK_FUNC_SHIFT) | 446 (r300_translate_stencil_op(state->stencil[1].fail_op) << 447 R300_S_BACK_SFAIL_OP_SHIFT) | 448 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 449 R300_S_BACK_ZPASS_OP_SHIFT) | 450 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 451 R300_S_BACK_ZFAIL_OP_SHIFT); 452 453 if (caps->is_r500) 454 { 455 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 456 dsa->stencil_ref_bf = 457 (state->stencil[1].valuemask << 458 R300_STENCILMASK_SHIFT) | 459 (state->stencil[1].writemask << 460 R300_STENCILWRITEMASK_SHIFT); 461 } 462 } 463 } 464 465 /* Alpha test setup. */ 466 if (state->alpha.enabled) { 467 dsa->alpha_function = 468 r300_translate_alpha_function(state->alpha.func) | 469 R300_FG_ALPHA_FUNC_ENABLE; 470 471 /* We could use 10bit alpha ref but who needs that? */ 472 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 473 474 if (caps->is_r500) 475 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 476 } 477 478 return (void*)dsa; 479} 480 481/* Bind DSA state. */ 482static void r300_bind_dsa_state(struct pipe_context* pipe, 483 void* state) 484{ 485 struct r300_context* r300 = r300_context(pipe); 486 487 UPDATE_STATE(state, r300->dsa_state); 488} 489 490/* Free DSA state. */ 491static void r300_delete_dsa_state(struct pipe_context* pipe, 492 void* state) 493{ 494 FREE(state); 495} 496 497static void r300_set_stencil_ref(struct pipe_context* pipe, 498 const struct pipe_stencil_ref* sr) 499{ 500 struct r300_context* r300 = r300_context(pipe); 501 r300->stencil_ref = *sr; 502 r300->dsa_state.dirty = TRUE; 503} 504 505/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 506static void r300_fb_update_tiling_flags(struct r300_context *r300, 507 const struct pipe_framebuffer_state *old_state, 508 const struct pipe_framebuffer_state *new_state) 509{ 510 struct r300_texture *tex; 511 unsigned i, j, level; 512 513 /* Reset tiling flags for old surfaces to default values. */ 514 for (i = 0; i < old_state->nr_cbufs; i++) { 515 for (j = 0; j < new_state->nr_cbufs; j++) { 516 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { 517 break; 518 } 519 } 520 /* If not binding the surface again... */ 521 if (j != new_state->nr_cbufs) { 522 continue; 523 } 524 525 tex = (struct r300_texture*)old_state->cbufs[i]->texture; 526 527 if (tex) { 528 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 529 tex->pitch[0], 530 tex->microtile, 531 tex->macrotile); 532 } 533 } 534 if (old_state->zsbuf && 535 (!new_state->zsbuf || 536 old_state->zsbuf->texture != new_state->zsbuf->texture)) { 537 tex = (struct r300_texture*)old_state->zsbuf->texture; 538 539 if (tex) { 540 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 541 tex->pitch[0], 542 tex->microtile, 543 tex->macrotile); 544 } 545 } 546 547 /* Set tiling flags for new surfaces. */ 548 for (i = 0; i < new_state->nr_cbufs; i++) { 549 tex = (struct r300_texture*)new_state->cbufs[i]->texture; 550 level = new_state->cbufs[i]->level; 551 552 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 553 tex->pitch[level], 554 tex->microtile, 555 tex->mip_macrotile[level]); 556 } 557 if (new_state->zsbuf) { 558 tex = (struct r300_texture*)new_state->zsbuf->texture; 559 level = new_state->zsbuf->level; 560 561 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 562 tex->pitch[level], 563 tex->microtile, 564 tex->mip_macrotile[level]); 565 } 566} 567 568static void 569 r300_set_framebuffer_state(struct pipe_context* pipe, 570 const struct pipe_framebuffer_state* state) 571{ 572 struct r300_context* r300 = r300_context(pipe); 573 struct r300_screen* r300screen = r300_screen(pipe->screen); 574 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 575 unsigned max_width, max_height; 576 uint32_t zbuffer_bpp = 0; 577 578 if (state->nr_cbufs > 4) { 579 fprintf(stderr, "r300: Implementation error: Too many MRTs in %s, " 580 "refusing to bind framebuffer state!\n", __FUNCTION__); 581 return; 582 } 583 584 if (r300screen->caps->is_r500) { 585 max_width = max_height = 4096; 586 } else if (r300screen->caps->is_r400) { 587 max_width = max_height = 4021; 588 } else { 589 max_width = max_height = 2560; 590 } 591 592 if (state->width > max_width || state->height > max_height) { 593 fprintf(stderr, "r300: Implementation error: Render targets are too " 594 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 595 return; 596 } 597 598 if (r300->draw) { 599 draw_flush(r300->draw); 600 } 601 602 r300->fb_state.dirty = TRUE; 603 604 /* If nr_cbufs is changed from zero to non-zero or vice versa... */ 605 if (!!old_state->nr_cbufs != !!state->nr_cbufs) { 606 r300->blend_state.dirty = TRUE; 607 } 608 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 609 if (!!old_state->zsbuf != !!state->zsbuf) { 610 r300->dsa_state.dirty = TRUE; 611 } 612 if (!r300->scissor_enabled) { 613 r300->scissor_state.dirty = TRUE; 614 } 615 616 r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); 617 618 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 619 620 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + 621 (state->zsbuf ? 10 : 0) + 8; 622 623 /* Polygon offset depends on the zbuffer bit depth. */ 624 if (state->zsbuf && r300->polygon_offset_enabled) { 625 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 626 case 2: 627 zbuffer_bpp = 16; 628 break; 629 case 4: 630 zbuffer_bpp = 24; 631 break; 632 } 633 634 if (r300->zbuffer_bpp != zbuffer_bpp) { 635 r300->zbuffer_bpp = zbuffer_bpp; 636 r300->rs_state.dirty = TRUE; 637 } 638 } 639} 640 641/* Create fragment shader state. */ 642static void* r300_create_fs_state(struct pipe_context* pipe, 643 const struct pipe_shader_state* shader) 644{ 645 struct r300_fragment_shader* fs = NULL; 646 647 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 648 649 /* Copy state directly into shader. */ 650 fs->state = *shader; 651 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 652 653 tgsi_scan_shader(shader->tokens, &fs->info); 654 r300_shader_read_fs_inputs(&fs->info, &fs->inputs); 655 656 return (void*)fs; 657} 658 659/* Bind fragment shader state. */ 660static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 661{ 662 struct r300_context* r300 = r300_context(pipe); 663 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 664 665 if (fs == NULL) { 666 r300->fs = NULL; 667 return; 668 } 669 670 r300->fs = fs; 671 r300_pick_fragment_shader(r300); 672 673 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 674 675 if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) { 676 r300->vap_output_state.dirty = TRUE; 677 } 678 679 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; 680} 681 682/* Delete fragment shader state. */ 683static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 684{ 685 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 686 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 687 688 while (ptr) { 689 tmp = ptr; 690 ptr = ptr->next; 691 rc_constants_destroy(&tmp->code.constants); 692 FREE(tmp); 693 } 694 FREE((void*)fs->state.tokens); 695 FREE(shader); 696} 697 698static void r300_set_polygon_stipple(struct pipe_context* pipe, 699 const struct pipe_poly_stipple* state) 700{ 701 /* XXX no idea how to set this up, but not terribly important */ 702} 703 704/* Create a new rasterizer state based on the CSO rasterizer state. 705 * 706 * This is a very large chunk of state, and covers most of the graphics 707 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 708 * 709 * In a not entirely unironic sidenote, this state has nearly nothing to do 710 * with the actual block on the Radeon called the rasterizer (RS). */ 711static void* r300_create_rs_state(struct pipe_context* pipe, 712 const struct pipe_rasterizer_state* state) 713{ 714 struct r300_screen* r300screen = r300_screen(pipe->screen); 715 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 716 717 /* Copy rasterizer state for Draw. */ 718 rs->rs = *state; 719 720#ifdef PIPE_ARCH_LITTLE_ENDIAN 721 rs->vap_control_status = R300_VC_NO_SWAP; 722#else 723 rs->vap_control_status = R300_VC_32BIT_SWAP; 724#endif 725 726 /* If no TCL engine is present, turn off the HW TCL. */ 727 if (!r300screen->caps->has_tcl) { 728 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 729 } 730 731 rs->point_size = pack_float_16_6x(state->point_size) | 732 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 733 734 rs->line_control = pack_float_16_6x(state->line_width) | 735 R300_GA_LINE_CNTL_END_TYPE_COMP; 736 737 /* Enable polygon mode */ 738 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 739 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 740 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 741 } 742 743 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 744 if (state->front_winding == PIPE_WINDING_CW) { 745 rs->cull_mode = R300_FRONT_FACE_CW; 746 747 /* Polygon offset */ 748 if (state->offset_cw) { 749 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 750 } 751 if (state->offset_ccw) { 752 rs->polygon_offset_enable |= R300_BACK_ENABLE; 753 } 754 755 /* Polygon mode */ 756 if (rs->polygon_mode) { 757 rs->polygon_mode |= 758 r300_translate_polygon_mode_front(state->fill_cw); 759 rs->polygon_mode |= 760 r300_translate_polygon_mode_back(state->fill_ccw); 761 } 762 } else { 763 rs->cull_mode = R300_FRONT_FACE_CCW; 764 765 /* Polygon offset */ 766 if (state->offset_ccw) { 767 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 768 } 769 if (state->offset_cw) { 770 rs->polygon_offset_enable |= R300_BACK_ENABLE; 771 } 772 773 /* Polygon mode */ 774 if (rs->polygon_mode) { 775 rs->polygon_mode |= 776 r300_translate_polygon_mode_front(state->fill_ccw); 777 rs->polygon_mode |= 778 r300_translate_polygon_mode_back(state->fill_cw); 779 } 780 } 781 if (state->front_winding & state->cull_mode) { 782 rs->cull_mode |= R300_CULL_FRONT; 783 } 784 if (~(state->front_winding) & state->cull_mode) { 785 rs->cull_mode |= R300_CULL_BACK; 786 } 787 788 if (rs->polygon_offset_enable) { 789 rs->depth_offset = state->offset_units; 790 rs->depth_scale = state->offset_scale; 791 } 792 793 if (state->line_stipple_enable) { 794 rs->line_stipple_config = 795 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 796 (fui((float)state->line_stipple_factor) & 797 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 798 /* XXX this might need to be scaled up */ 799 rs->line_stipple_value = state->line_stipple_pattern; 800 } 801 802 if (state->flatshade) { 803 rs->color_control = R300_SHADE_MODEL_FLAT; 804 } else { 805 rs->color_control = R300_SHADE_MODEL_SMOOTH; 806 } 807 808 return (void*)rs; 809} 810 811/* Bind rasterizer state. */ 812static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 813{ 814 struct r300_context* r300 = r300_context(pipe); 815 struct r300_rs_state* rs = (struct r300_rs_state*)state; 816 boolean scissor_was_enabled = r300->scissor_enabled; 817 818 if (r300->draw) { 819 draw_flush(r300->draw); 820 draw_set_rasterizer_state(r300->draw, &rs->rs); 821 } 822 823 if (rs) { 824 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 825 r300->scissor_enabled = rs->rs.scissor; 826 } else { 827 r300->polygon_offset_enabled = FALSE; 828 r300->scissor_enabled = FALSE; 829 } 830 831 UPDATE_STATE(state, r300->rs_state); 832 r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0); 833 834 if (scissor_was_enabled != r300->scissor_enabled) { 835 r300->scissor_state.dirty = TRUE; 836 } 837} 838 839/* Free rasterizer state. */ 840static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 841{ 842 FREE(state); 843} 844 845static void* 846 r300_create_sampler_state(struct pipe_context* pipe, 847 const struct pipe_sampler_state* state) 848{ 849 struct r300_context* r300 = r300_context(pipe); 850 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 851 boolean is_r500 = r300_screen(pipe->screen)->caps->is_r500; 852 int lod_bias; 853 union util_color uc; 854 855 sampler->state = *state; 856 857 sampler->filter0 |= 858 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 859 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 860 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 861 862 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 863 state->mag_img_filter, 864 state->min_mip_filter, 865 state->max_anisotropy > 0); 866 867 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 868 869 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 870 /* We must pass these to the merge function to clamp them properly. */ 871 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 872 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 873 874 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 875 876 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 877 878 /* This is very high quality anisotropic filtering for R5xx. 879 * It's good for benchmarking the performance of texturing but 880 * in practice we don't want to slow down the driver because it's 881 * a pretty good performance killer. Feel free to play with it. */ 882 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 883 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 884 } 885 886 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 887 sampler->border_color = uc.ui; 888 889 /* R500-specific fixups and optimizations */ 890 if (r300_screen(r300->context.screen)->caps->is_r500) { 891 sampler->filter1 |= R500_BORDER_FIX; 892 } 893 894 return (void*)sampler; 895} 896 897static void r300_bind_sampler_states(struct pipe_context* pipe, 898 unsigned count, 899 void** states) 900{ 901 struct r300_context* r300 = r300_context(pipe); 902 struct r300_textures_state* state = 903 (struct r300_textures_state*)r300->textures_state.state; 904 905 if (count > 8) { 906 return; 907 } 908 909 memcpy(state->sampler_states, states, sizeof(void*) * count); 910 state->sampler_count = count; 911 912 r300->textures_state.dirty = TRUE; 913 914 /* Pick a fragment shader based on the texture compare state. */ 915 if (r300->fs && count) { 916 if (r300_pick_fragment_shader(r300)) { 917 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | 918 R300_NEW_FRAGMENT_SHADER_CONSTANTS; 919 } 920 } 921} 922 923static void r300_lacks_vertex_textures(struct pipe_context* pipe, 924 unsigned count, 925 void** states) 926{ 927} 928 929static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 930{ 931 FREE(state); 932} 933 934static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 935 unsigned count, 936 struct pipe_sampler_view** views) 937{ 938 struct r300_context* r300 = r300_context(pipe); 939 struct r300_textures_state* state = 940 (struct r300_textures_state*)r300->textures_state.state; 941 struct r300_texture *texture; 942 unsigned i; 943 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; 944 boolean dirty_tex = FALSE; 945 946 /* XXX magic num */ 947 if (count > 8) { 948 return; 949 } 950 951 for (i = 0; i < count; i++) { 952 if (state->fragment_sampler_views[i] != views[i]) { 953 pipe_sampler_view_reference(&state->fragment_sampler_views[i], 954 views[i]); 955 956 if (!views[i]) { 957 continue; 958 } 959 960 /* A new sampler view (= texture)... */ 961 dirty_tex = TRUE; 962 963 /* R300-specific - set the texrect factor in the fragment shader */ 964 texture = (struct r300_texture *)views[i]->texture; 965 if (!is_r500 && texture->is_npot) { 966 /* XXX It would be nice to re-emit just 1 constant, 967 * XXX not all of them */ 968 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 969 } 970 } 971 } 972 973 for (i = count; i < 8; i++) { 974 if (state->fragment_sampler_views[i]) { 975 pipe_sampler_view_reference(&state->fragment_sampler_views[i], 976 NULL); 977 } 978 } 979 980 state->texture_count = count; 981 982 r300->textures_state.dirty = TRUE; 983 984 if (dirty_tex) { 985 r300->texture_cache_inval.dirty = TRUE; 986 } 987} 988 989static struct pipe_sampler_view * 990r300_create_sampler_view(struct pipe_context *pipe, 991 struct pipe_texture *texture, 992 const struct pipe_sampler_view *templ) 993{ 994 struct pipe_sampler_view *view = CALLOC_STRUCT(pipe_sampler_view); 995 996 if (view) { 997 *view = *templ; 998 view->reference.count = 1; 999 view->texture = NULL; 1000 pipe_texture_reference(&view->texture, texture); 1001 view->context = pipe; 1002 } 1003 1004 return view; 1005} 1006 1007static void 1008r300_sampler_view_destroy(struct pipe_context *pipe, 1009 struct pipe_sampler_view *view) 1010{ 1011 pipe_texture_reference(&view->texture, NULL); 1012 FREE(view); 1013} 1014 1015static void r300_set_scissor_state(struct pipe_context* pipe, 1016 const struct pipe_scissor_state* state) 1017{ 1018 struct r300_context* r300 = r300_context(pipe); 1019 1020 memcpy(r300->scissor_state.state, state, 1021 sizeof(struct pipe_scissor_state)); 1022 1023 if (r300->scissor_enabled) { 1024 r300->scissor_state.dirty = TRUE; 1025 } 1026} 1027 1028static void r300_set_viewport_state(struct pipe_context* pipe, 1029 const struct pipe_viewport_state* state) 1030{ 1031 struct r300_context* r300 = r300_context(pipe); 1032 struct r300_viewport_state* viewport = 1033 (struct r300_viewport_state*)r300->viewport_state.state; 1034 1035 r300->viewport = *state; 1036 1037 /* Do the transform in HW. */ 1038 viewport->vte_control = R300_VTX_W0_FMT; 1039 1040 if (state->scale[0] != 1.0f) { 1041 viewport->xscale = state->scale[0]; 1042 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1043 } 1044 if (state->scale[1] != 1.0f) { 1045 viewport->yscale = state->scale[1]; 1046 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1047 } 1048 if (state->scale[2] != 1.0f) { 1049 viewport->zscale = state->scale[2]; 1050 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1051 } 1052 if (state->translate[0] != 0.0f) { 1053 viewport->xoffset = state->translate[0]; 1054 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1055 } 1056 if (state->translate[1] != 0.0f) { 1057 viewport->yoffset = state->translate[1]; 1058 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1059 } 1060 if (state->translate[2] != 0.0f) { 1061 viewport->zoffset = state->translate[2]; 1062 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1063 } 1064 1065 r300->viewport_state.dirty = TRUE; 1066 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 1067 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1068 } 1069} 1070 1071static void r300_set_vertex_buffers(struct pipe_context* pipe, 1072 unsigned count, 1073 const struct pipe_vertex_buffer* buffers) 1074{ 1075 struct r300_context* r300 = r300_context(pipe); 1076 struct pipe_vertex_buffer *vbo; 1077 unsigned i, max_index = (1 << 24) - 1; 1078 boolean any_user_buffer = FALSE; 1079 1080 if (count == r300->vertex_buffer_count && 1081 memcmp(r300->vertex_buffer, buffers, 1082 sizeof(struct pipe_vertex_buffer) * count) == 0) { 1083 return; 1084 } 1085 1086 /* Check if the stride is aligned to the size of DWORD. */ 1087 for (i = 0; i < count; i++) { 1088 if (buffers[i].buffer) { 1089 if (buffers[i].stride % 4 != 0) { 1090 // XXX Shouldn't we align the buffer? 1091 fprintf(stderr, "r300_set_vertex_buffers: " 1092 "Unaligned buffer stride %i isn't supported.\n", 1093 buffers[i].stride); 1094 assert(0); 1095 abort(); 1096 } 1097 } 1098 } 1099 1100 for (i = 0; i < count; i++) { 1101 /* Why, yes, I AM casting away constness. How did you know? */ 1102 vbo = (struct pipe_vertex_buffer*)&buffers[i]; 1103 1104 /* Reference our buffer. */ 1105 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, vbo->buffer); 1106 1107 /* Skip NULL buffers */ 1108 if (!buffers[i].buffer) { 1109 continue; 1110 } 1111 1112 if (r300_buffer_is_user_buffer(vbo->buffer)) { 1113 any_user_buffer = TRUE; 1114 } 1115 1116 if (vbo->max_index == ~0) { 1117 /* Bogus value from broken state tracker; hax it. */ 1118 vbo->max_index = 1119 (vbo->buffer->size - vbo->buffer_offset) / vbo->stride; 1120 } 1121 1122 max_index = MIN2(vbo->max_index, max_index); 1123 } 1124 1125 for (; i < r300->vertex_buffer_count; i++) { 1126 /* Dereference any old buffers. */ 1127 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, NULL); 1128 } 1129 1130 memcpy(r300->vertex_buffer, buffers, 1131 sizeof(struct pipe_vertex_buffer) * count); 1132 1133 r300->vertex_buffer_count = count; 1134 r300->vertex_buffer_max_index = max_index; 1135 r300->any_user_vbs = any_user_buffer; 1136 1137 if (r300->draw) { 1138 draw_flush(r300->draw); 1139 draw_set_vertex_buffers(r300->draw, count, buffers); 1140 } 1141} 1142 1143/* Update the PSC tables. */ 1144static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1145{ 1146 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1147 uint16_t type, swizzle; 1148 enum pipe_format format; 1149 unsigned i; 1150 1151 assert(velems->count <= 16); 1152 1153 /* Vertex shaders have no semantics on their inputs, 1154 * so PSC should just route stuff based on the vertex elements, 1155 * and not on attrib information. */ 1156 for (i = 0; i < velems->count; i++) { 1157 format = velems->velem[i].src_format; 1158 1159 type = r300_translate_vertex_data_type(format) | 1160 (i << R300_DST_VEC_LOC_SHIFT); 1161 swizzle = r300_translate_vertex_data_swizzle(format); 1162 1163 if (i & 1) { 1164 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1165 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1166 } else { 1167 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1168 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1169 } 1170 } 1171 1172 /* Set the last vector in the PSC. */ 1173 if (i) { 1174 i -= 1; 1175 } 1176 vstream->vap_prog_stream_cntl[i >> 1] |= 1177 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1178 1179 vstream->count = (i >> 1) + 1; 1180} 1181 1182static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1183 unsigned count, 1184 const struct pipe_vertex_element* attribs) 1185{ 1186 struct r300_context *r300 = r300_context(pipe); 1187 struct r300_screen* r300screen = r300_screen(pipe->screen); 1188 struct r300_vertex_element_state *velems; 1189 unsigned i, size; 1190 1191 assert(count <= PIPE_MAX_ATTRIBS); 1192 velems = CALLOC_STRUCT(r300_vertex_element_state); 1193 if (velems != NULL) { 1194 velems->count = count; 1195 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1196 1197 if (r300screen->caps->has_tcl) { 1198 /* Check if the format is aligned to the size of DWORD. */ 1199 for (i = 0; i < count; i++) { 1200 size = util_format_get_blocksize(attribs[i].src_format); 1201 1202 if (size % 4 != 0) { 1203 /* XXX Shouldn't we align the format? */ 1204 fprintf(stderr, "r300_create_vertex_elements_state: " 1205 "Unaligned format %s:%i isn't supported\n", 1206 util_format_name(attribs[i].src_format), size); 1207 assert(0); 1208 abort(); 1209 } 1210 } 1211 1212 r300_vertex_psc(velems); 1213 } 1214 } 1215 return velems; 1216} 1217 1218static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1219 void *state) 1220{ 1221 struct r300_context *r300 = r300_context(pipe); 1222 struct r300_vertex_element_state *velems = state; 1223 1224 if (velems == NULL) { 1225 return; 1226 } 1227 1228 r300->velems = velems; 1229 1230 if (r300->draw) { 1231 draw_flush(r300->draw); 1232 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1233 } 1234 1235 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1236 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1237} 1238 1239static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1240{ 1241 FREE(state); 1242} 1243 1244static void* r300_create_vs_state(struct pipe_context* pipe, 1245 const struct pipe_shader_state* shader) 1246{ 1247 struct r300_context* r300 = r300_context(pipe); 1248 1249 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1250 r300_vertex_shader_common_init(vs, shader); 1251 1252 if (r300_screen(pipe->screen)->caps->has_tcl) { 1253 r300_translate_vertex_shader(r300, vs); 1254 } else { 1255 vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); 1256 } 1257 1258 return vs; 1259} 1260 1261static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1262{ 1263 struct r300_context* r300 = r300_context(pipe); 1264 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1265 1266 if (vs == NULL) { 1267 r300->vs_state.state = NULL; 1268 return; 1269 } 1270 if (vs == r300->vs_state.state) { 1271 return; 1272 } 1273 r300->vs_state.state = vs; 1274 1275 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block 1276 if (r300->fs) { 1277 r300_vertex_shader_setup_wpos(r300); 1278 } 1279 memcpy(r300->vap_output_state.state, &vs->vap_out, 1280 sizeof(struct r300_vap_output_state)); 1281 r300->vap_output_state.dirty = TRUE; 1282 1283 /* The majority of the RS block bits is dependent on the vertex shader. */ 1284 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1285 1286 if (r300_screen(pipe->screen)->caps->has_tcl) { 1287 r300->vs_state.dirty = TRUE; 1288 r300->vs_state.size = vs->code.length + 9; 1289 1290 r300->pvs_flush.dirty = TRUE; 1291 1292 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1293 } else { 1294 draw_flush(r300->draw); 1295 draw_bind_vertex_shader(r300->draw, 1296 (struct draw_vertex_shader*)vs->draw_vs); 1297 } 1298} 1299 1300static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1301{ 1302 struct r300_context* r300 = r300_context(pipe); 1303 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1304 1305 if (r300_screen(pipe->screen)->caps->has_tcl) { 1306 rc_constants_destroy(&vs->code.constants); 1307 } else { 1308 draw_delete_vertex_shader(r300->draw, 1309 (struct draw_vertex_shader*)vs->draw_vs); 1310 } 1311 1312 FREE((void*)vs->state.tokens); 1313 FREE(shader); 1314} 1315 1316static void r300_set_constant_buffer(struct pipe_context *pipe, 1317 uint shader, uint index, 1318 struct pipe_buffer *buf) 1319{ 1320 struct r300_context* r300 = r300_context(pipe); 1321 struct r300_screen *r300screen = r300_screen(pipe->screen); 1322 void *mapped; 1323 int max_size = 0; 1324 1325 if (buf == NULL || buf->size == 0 || 1326 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) 1327 { 1328 r300->shader_constants[shader].count = 0; 1329 return; 1330 } 1331 1332 assert((buf->size % 4 * sizeof(float)) == 0); 1333 1334 /* Check the size of the constant buffer. */ 1335 switch (shader) { 1336 case PIPE_SHADER_VERTEX: 1337 max_size = 256; 1338 break; 1339 case PIPE_SHADER_FRAGMENT: 1340 if (r300screen->caps->is_r500) { 1341 max_size = 256; 1342 /* XXX Implement emission of r400's extended constant buffer. */ 1343 /*} else if (r300screen->caps->is_r400) { 1344 max_size = 64;*/ 1345 } else { 1346 max_size = 32; 1347 } 1348 break; 1349 default: 1350 assert(0); 1351 } 1352 1353 /* XXX Subtract immediates and RC_STATE_* variables. */ 1354 if (buf->size > (sizeof(float) * 4 * max_size)) { 1355 fprintf(stderr, "r300: Max size of the constant buffer is " 1356 "%i*4 floats.\n", max_size); 1357 abort(); 1358 } 1359 1360 memcpy(r300->shader_constants[shader].constants, mapped, buf->size); 1361 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); 1362 pipe_buffer_unmap(pipe->screen, buf); 1363 1364 if (shader == PIPE_SHADER_VERTEX) { 1365 if (r300screen->caps->has_tcl) { 1366 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1367 r300->pvs_flush.dirty = TRUE; 1368 } else if (r300->draw) { 1369 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1370 0, r300->shader_constants[PIPE_SHADER_VERTEX].constants, 1371 buf->size); 1372 } 1373 } else if (shader == PIPE_SHADER_FRAGMENT) { 1374 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1375 } 1376} 1377 1378void r300_init_state_functions(struct r300_context* r300) 1379{ 1380 r300->context.create_blend_state = r300_create_blend_state; 1381 r300->context.bind_blend_state = r300_bind_blend_state; 1382 r300->context.delete_blend_state = r300_delete_blend_state; 1383 1384 r300->context.set_blend_color = r300_set_blend_color; 1385 1386 r300->context.set_clip_state = r300_set_clip_state; 1387 1388 r300->context.set_constant_buffer = r300_set_constant_buffer; 1389 1390 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1391 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1392 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1393 1394 r300->context.set_stencil_ref = r300_set_stencil_ref; 1395 1396 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1397 1398 r300->context.create_fs_state = r300_create_fs_state; 1399 r300->context.bind_fs_state = r300_bind_fs_state; 1400 r300->context.delete_fs_state = r300_delete_fs_state; 1401 1402 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1403 1404 r300->context.create_rasterizer_state = r300_create_rs_state; 1405 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1406 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1407 1408 r300->context.create_sampler_state = r300_create_sampler_state; 1409 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1410 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1411 r300->context.delete_sampler_state = r300_delete_sampler_state; 1412 1413 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1414 r300->context.create_sampler_view = r300_create_sampler_view; 1415 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1416 1417 r300->context.set_scissor_state = r300_set_scissor_state; 1418 1419 r300->context.set_viewport_state = r300_set_viewport_state; 1420 1421 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1422 1423 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1424 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1425 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1426 1427 r300->context.create_vs_state = r300_create_vs_state; 1428 r300->context.bind_vs_state = r300_bind_vs_state; 1429 r300->context.delete_vs_state = r300_delete_vs_state; 1430} 1431