r300_state.c revision f6c7b911653fb1508256c63518ef0bd15d68186e
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 /* fglrx appears to never set this */ 325 blend->dither = 0; 326 /* blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 327 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; */ 328 } 329 330 return (void*)blend; 331} 332 333/* Bind blend state. */ 334static void r300_bind_blend_state(struct pipe_context* pipe, 335 void* state) 336{ 337 struct r300_context* r300 = r300_context(pipe); 338 339 UPDATE_STATE(state, r300->blend_state); 340} 341 342/* Free blend state. */ 343static void r300_delete_blend_state(struct pipe_context* pipe, 344 void* state) 345{ 346 FREE(state); 347} 348 349/* Convert float to 10bit integer */ 350static unsigned float_to_fixed10(float f) 351{ 352 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 353} 354 355/* Set blend color. 356 * Setup both R300 and R500 registers, figure out later which one to write. */ 357static void r300_set_blend_color(struct pipe_context* pipe, 358 const struct pipe_blend_color* color) 359{ 360 struct r300_context* r300 = r300_context(pipe); 361 struct r300_screen* r300screen = r300_screen(pipe->screen); 362 struct r300_blend_color_state* state = 363 (struct r300_blend_color_state*)r300->blend_color_state.state; 364 union util_color uc; 365 366 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 367 state->blend_color = uc.ui; 368 369 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 370 state->blend_color_red_alpha = 371 float_to_fixed10(color->color[0]) | 372 (float_to_fixed10(color->color[3]) << 16); 373 state->blend_color_green_blue = 374 float_to_fixed10(color->color[2]) | 375 (float_to_fixed10(color->color[1]) << 16); 376 377 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; 378 r300->blend_color_state.dirty = TRUE; 379} 380 381static void r300_set_clip_state(struct pipe_context* pipe, 382 const struct pipe_clip_state* state) 383{ 384 struct r300_context* r300 = r300_context(pipe); 385 386 r300->clip = *state; 387 388 if (r300_screen(pipe->screen)->caps->has_tcl) { 389 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 390 r300->clip_state.size = 29; 391 } else { 392 draw_flush(r300->draw); 393 draw_set_clip_state(r300->draw, state); 394 r300->clip_state.size = 2; 395 } 396 397 r300->clip_state.dirty = TRUE; 398} 399 400/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 401 * 402 * This contains the depth buffer, stencil buffer, alpha test, and such. 403 * On the Radeon, depth and stencil buffer setup are intertwined, which is 404 * the reason for some of the strange-looking assignments across registers. */ 405static void* 406 r300_create_dsa_state(struct pipe_context* pipe, 407 const struct pipe_depth_stencil_alpha_state* state) 408{ 409 struct r300_capabilities *caps = 410 r300_screen(r300_context(pipe)->context.screen)->caps; 411 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 412 413 /* Depth test setup. */ 414 if (state->depth.enabled) { 415 dsa->z_buffer_control |= R300_Z_ENABLE; 416 417 if (state->depth.writemask) { 418 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 419 } 420 421 dsa->z_stencil_control |= 422 (r300_translate_depth_stencil_function(state->depth.func) << 423 R300_Z_FUNC_SHIFT); 424 } 425 426 /* Stencil buffer setup. */ 427 if (state->stencil[0].enabled) { 428 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 429 dsa->z_stencil_control |= 430 (r300_translate_depth_stencil_function(state->stencil[0].func) << 431 R300_S_FRONT_FUNC_SHIFT) | 432 (r300_translate_stencil_op(state->stencil[0].fail_op) << 433 R300_S_FRONT_SFAIL_OP_SHIFT) | 434 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 435 R300_S_FRONT_ZPASS_OP_SHIFT) | 436 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 437 R300_S_FRONT_ZFAIL_OP_SHIFT); 438 439 dsa->stencil_ref_mask = 440 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 441 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 442 443 if (state->stencil[1].enabled) { 444 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 445 dsa->z_stencil_control |= 446 (r300_translate_depth_stencil_function(state->stencil[1].func) << 447 R300_S_BACK_FUNC_SHIFT) | 448 (r300_translate_stencil_op(state->stencil[1].fail_op) << 449 R300_S_BACK_SFAIL_OP_SHIFT) | 450 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 451 R300_S_BACK_ZPASS_OP_SHIFT) | 452 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 453 R300_S_BACK_ZFAIL_OP_SHIFT); 454 455 if (caps->is_r500) 456 { 457 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 458 dsa->stencil_ref_bf = 459 (state->stencil[1].valuemask << 460 R300_STENCILMASK_SHIFT) | 461 (state->stencil[1].writemask << 462 R300_STENCILWRITEMASK_SHIFT); 463 } 464 } 465 } 466 467 /* Alpha test setup. */ 468 if (state->alpha.enabled) { 469 dsa->alpha_function = 470 r300_translate_alpha_function(state->alpha.func) | 471 R300_FG_ALPHA_FUNC_ENABLE; 472 473 /* We could use 10bit alpha ref but who needs that? */ 474 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 475 476 if (caps->is_r500) 477 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 478 } 479 480 return (void*)dsa; 481} 482 483/* Bind DSA state. */ 484static void r300_bind_dsa_state(struct pipe_context* pipe, 485 void* state) 486{ 487 struct r300_context* r300 = r300_context(pipe); 488 489 UPDATE_STATE(state, r300->dsa_state); 490} 491 492/* Free DSA state. */ 493static void r300_delete_dsa_state(struct pipe_context* pipe, 494 void* state) 495{ 496 FREE(state); 497} 498 499static void r300_set_stencil_ref(struct pipe_context* pipe, 500 const struct pipe_stencil_ref* sr) 501{ 502 struct r300_context* r300 = r300_context(pipe); 503 r300->stencil_ref = *sr; 504 r300->dsa_state.dirty = TRUE; 505} 506 507/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 508static void r300_fb_update_tiling_flags(struct r300_context *r300, 509 const struct pipe_framebuffer_state *old_state, 510 const struct pipe_framebuffer_state *new_state) 511{ 512 struct r300_texture *tex; 513 unsigned i, j, level; 514 515 /* Reset tiling flags for old surfaces to default values. */ 516 for (i = 0; i < old_state->nr_cbufs; i++) { 517 for (j = 0; j < new_state->nr_cbufs; j++) { 518 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { 519 break; 520 } 521 } 522 /* If not binding the surface again... */ 523 if (j != new_state->nr_cbufs) { 524 continue; 525 } 526 527 tex = (struct r300_texture*)old_state->cbufs[i]->texture; 528 529 if (tex) { 530 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 531 tex->pitch[0], 532 tex->microtile, 533 tex->macrotile); 534 } 535 } 536 if (old_state->zsbuf && 537 (!new_state->zsbuf || 538 old_state->zsbuf->texture != new_state->zsbuf->texture)) { 539 tex = (struct r300_texture*)old_state->zsbuf->texture; 540 541 if (tex) { 542 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 543 tex->pitch[0], 544 tex->microtile, 545 tex->macrotile); 546 } 547 } 548 549 /* Set tiling flags for new surfaces. */ 550 for (i = 0; i < new_state->nr_cbufs; i++) { 551 tex = (struct r300_texture*)new_state->cbufs[i]->texture; 552 level = new_state->cbufs[i]->level; 553 554 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 555 tex->pitch[level], 556 tex->microtile, 557 tex->mip_macrotile[level]); 558 } 559 if (new_state->zsbuf) { 560 tex = (struct r300_texture*)new_state->zsbuf->texture; 561 level = new_state->zsbuf->level; 562 563 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 564 tex->pitch[level], 565 tex->microtile, 566 tex->mip_macrotile[level]); 567 } 568} 569 570static void 571 r300_set_framebuffer_state(struct pipe_context* pipe, 572 const struct pipe_framebuffer_state* state) 573{ 574 struct r300_context* r300 = r300_context(pipe); 575 struct r300_screen* r300screen = r300_screen(pipe->screen); 576 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 577 unsigned max_width, max_height; 578 uint32_t zbuffer_bpp = 0; 579 580 if (state->nr_cbufs > 4) { 581 fprintf(stderr, "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 fprintf(stderr, "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); 718 719 /* Copy rasterizer state for Draw. */ 720 rs->rs = *state; 721 722#ifdef PIPE_ARCH_LITTLE_ENDIAN 723 rs->vap_control_status = R300_VC_NO_SWAP; 724#else 725 rs->vap_control_status = R300_VC_32BIT_SWAP; 726#endif 727 728 /* If no TCL engine is present, turn off the HW TCL. */ 729 if (!r300screen->caps->has_tcl) { 730 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 731 } 732 733 rs->point_size = pack_float_16_6x(state->point_size) | 734 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 735 736 rs->line_control = pack_float_16_6x(state->line_width) | 737 R300_GA_LINE_CNTL_END_TYPE_COMP; 738 739 /* Enable polygon mode */ 740 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 741 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 742 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 743 } 744 745 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 746 if (state->front_winding == PIPE_WINDING_CW) { 747 rs->cull_mode = R300_FRONT_FACE_CW; 748 749 /* Polygon offset */ 750 if (state->offset_cw) { 751 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 752 } 753 if (state->offset_ccw) { 754 rs->polygon_offset_enable |= R300_BACK_ENABLE; 755 } 756 757 /* Polygon mode */ 758 if (rs->polygon_mode) { 759 rs->polygon_mode |= 760 r300_translate_polygon_mode_front(state->fill_cw); 761 rs->polygon_mode |= 762 r300_translate_polygon_mode_back(state->fill_ccw); 763 } 764 } else { 765 rs->cull_mode = R300_FRONT_FACE_CCW; 766 767 /* Polygon offset */ 768 if (state->offset_ccw) { 769 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 770 } 771 if (state->offset_cw) { 772 rs->polygon_offset_enable |= R300_BACK_ENABLE; 773 } 774 775 /* Polygon mode */ 776 if (rs->polygon_mode) { 777 rs->polygon_mode |= 778 r300_translate_polygon_mode_front(state->fill_ccw); 779 rs->polygon_mode |= 780 r300_translate_polygon_mode_back(state->fill_cw); 781 } 782 } 783 if (state->front_winding & state->cull_mode) { 784 rs->cull_mode |= R300_CULL_FRONT; 785 } 786 if (~(state->front_winding) & state->cull_mode) { 787 rs->cull_mode |= R300_CULL_BACK; 788 } 789 790 if (rs->polygon_offset_enable) { 791 rs->depth_offset = state->offset_units; 792 rs->depth_scale = state->offset_scale; 793 } 794 795 if (state->line_stipple_enable) { 796 rs->line_stipple_config = 797 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 798 (fui((float)state->line_stipple_factor) & 799 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 800 /* XXX this might need to be scaled up */ 801 rs->line_stipple_value = state->line_stipple_pattern; 802 } 803 804 if (state->flatshade) { 805 rs->color_control = R300_SHADE_MODEL_FLAT; 806 } else { 807 rs->color_control = R300_SHADE_MODEL_SMOOTH; 808 } 809 810 return (void*)rs; 811} 812 813/* Bind rasterizer state. */ 814static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 815{ 816 struct r300_context* r300 = r300_context(pipe); 817 struct r300_rs_state* rs = (struct r300_rs_state*)state; 818 boolean scissor_was_enabled = r300->scissor_enabled; 819 820 if (r300->draw) { 821 draw_flush(r300->draw); 822 draw_set_rasterizer_state(r300->draw, &rs->rs); 823 } 824 825 if (rs) { 826 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 827 r300->scissor_enabled = rs->rs.scissor; 828 } else { 829 r300->polygon_offset_enabled = FALSE; 830 r300->scissor_enabled = FALSE; 831 } 832 833 UPDATE_STATE(state, r300->rs_state); 834 r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0); 835 836 if (scissor_was_enabled != r300->scissor_enabled) { 837 r300->scissor_state.dirty = TRUE; 838 } 839} 840 841/* Free rasterizer state. */ 842static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 843{ 844 FREE(state); 845} 846 847static void* 848 r300_create_sampler_state(struct pipe_context* pipe, 849 const struct pipe_sampler_state* state) 850{ 851 struct r300_context* r300 = r300_context(pipe); 852 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 853 boolean is_r500 = r300_screen(pipe->screen)->caps->is_r500; 854 int lod_bias; 855 union util_color uc; 856 857 sampler->state = *state; 858 859 sampler->filter0 |= 860 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 861 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 862 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 863 864 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 865 state->mag_img_filter, 866 state->min_mip_filter, 867 state->max_anisotropy > 0); 868 869 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 870 871 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 872 /* We must pass these to the merge function to clamp them properly. */ 873 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 874 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 875 876 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 877 878 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 879 880 /* This is very high quality anisotropic filtering for R5xx. 881 * It's good for benchmarking the performance of texturing but 882 * in practice we don't want to slow down the driver because it's 883 * a pretty good performance killer. Feel free to play with it. */ 884 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 885 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 886 } 887 888 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 889 sampler->border_color = uc.ui; 890 891 /* R500-specific fixups and optimizations */ 892 if (r300_screen(r300->context.screen)->caps->is_r500) { 893 sampler->filter1 |= R500_BORDER_FIX; 894 } 895 896 return (void*)sampler; 897} 898 899static void r300_bind_sampler_states(struct pipe_context* pipe, 900 unsigned count, 901 void** states) 902{ 903 struct r300_context* r300 = r300_context(pipe); 904 struct r300_textures_state* state = 905 (struct r300_textures_state*)r300->textures_state.state; 906 907 if (count > 8) { 908 return; 909 } 910 911 memcpy(state->sampler_states, states, sizeof(void*) * count); 912 state->sampler_count = count; 913 914 r300->textures_state.dirty = TRUE; 915 916 /* Pick a fragment shader based on the texture compare state. */ 917 if (r300->fs && count) { 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_fragment_sampler_views(struct pipe_context* pipe, 937 unsigned count, 938 struct pipe_sampler_view** views) 939{ 940 struct r300_context* r300 = r300_context(pipe); 941 struct r300_textures_state* state = 942 (struct r300_textures_state*)r300->textures_state.state; 943 struct r300_texture *texture; 944 unsigned i; 945 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; 946 boolean dirty_tex = FALSE; 947 948 /* XXX magic num */ 949 if (count > 8) { 950 return; 951 } 952 953 for (i = 0; i < count; i++) { 954 if (state->fragment_sampler_views[i] != views[i]) { 955 pipe_sampler_view_reference(&state->fragment_sampler_views[i], 956 views[i]); 957 958 if (!views[i]) { 959 continue; 960 } 961 962 /* A new sampler view (= texture)... */ 963 dirty_tex = TRUE; 964 965 /* R300-specific - set the texrect factor in the fragment shader */ 966 texture = (struct r300_texture *)views[i]->texture; 967 if (!is_r500 && texture->is_npot) { 968 /* XXX It would be nice to re-emit just 1 constant, 969 * XXX not all of them */ 970 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 971 } 972 } 973 } 974 975 for (i = count; i < 8; i++) { 976 if (state->fragment_sampler_views[i]) { 977 pipe_sampler_view_reference(&state->fragment_sampler_views[i], 978 NULL); 979 } 980 } 981 982 state->texture_count = count; 983 984 r300->textures_state.dirty = TRUE; 985 986 if (dirty_tex) { 987 r300->texture_cache_inval.dirty = TRUE; 988 } 989} 990 991static struct pipe_sampler_view * 992r300_create_sampler_view(struct pipe_context *pipe, 993 struct pipe_texture *texture, 994 const struct pipe_sampler_view *templ) 995{ 996 struct pipe_sampler_view *view = CALLOC_STRUCT(pipe_sampler_view); 997 998 if (view) { 999 *view = *templ; 1000 view->reference.count = 1; 1001 view->texture = NULL; 1002 pipe_texture_reference(&view->texture, texture); 1003 view->context = pipe; 1004 } 1005 1006 return view; 1007} 1008 1009static void 1010r300_sampler_view_destroy(struct pipe_context *pipe, 1011 struct pipe_sampler_view *view) 1012{ 1013 pipe_texture_reference(&view->texture, NULL); 1014 FREE(view); 1015} 1016 1017static void r300_set_scissor_state(struct pipe_context* pipe, 1018 const struct pipe_scissor_state* state) 1019{ 1020 struct r300_context* r300 = r300_context(pipe); 1021 1022 memcpy(r300->scissor_state.state, state, 1023 sizeof(struct pipe_scissor_state)); 1024 1025 if (r300->scissor_enabled) { 1026 r300->scissor_state.dirty = TRUE; 1027 } 1028} 1029 1030static void r300_set_viewport_state(struct pipe_context* pipe, 1031 const struct pipe_viewport_state* state) 1032{ 1033 struct r300_context* r300 = r300_context(pipe); 1034 struct r300_viewport_state* viewport = 1035 (struct r300_viewport_state*)r300->viewport_state.state; 1036 1037 r300->viewport = *state; 1038 1039 /* Do the transform in HW. */ 1040 viewport->vte_control = R300_VTX_W0_FMT; 1041 1042 if (state->scale[0] != 1.0f) { 1043 viewport->xscale = state->scale[0]; 1044 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1045 } 1046 if (state->scale[1] != 1.0f) { 1047 viewport->yscale = state->scale[1]; 1048 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1049 } 1050 if (state->scale[2] != 1.0f) { 1051 viewport->zscale = state->scale[2]; 1052 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1053 } 1054 if (state->translate[0] != 0.0f) { 1055 viewport->xoffset = state->translate[0]; 1056 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1057 } 1058 if (state->translate[1] != 0.0f) { 1059 viewport->yoffset = state->translate[1]; 1060 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1061 } 1062 if (state->translate[2] != 0.0f) { 1063 viewport->zoffset = state->translate[2]; 1064 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1065 } 1066 1067 r300->viewport_state.dirty = TRUE; 1068 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 1069 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1070 } 1071} 1072 1073static void r300_set_vertex_buffers(struct pipe_context* pipe, 1074 unsigned count, 1075 const struct pipe_vertex_buffer* buffers) 1076{ 1077 struct r300_context* r300 = r300_context(pipe); 1078 struct pipe_vertex_buffer *vbo; 1079 unsigned i, max_index = (1 << 24) - 1; 1080 boolean any_user_buffer = FALSE; 1081 1082 if (count == r300->vertex_buffer_count && 1083 memcmp(r300->vertex_buffer, buffers, 1084 sizeof(struct pipe_vertex_buffer) * count) == 0) { 1085 return; 1086 } 1087 1088 /* Check if the stride is aligned to the size of DWORD. */ 1089 for (i = 0; i < count; i++) { 1090 if (buffers[i].buffer) { 1091 if (buffers[i].stride % 4 != 0) { 1092 // XXX Shouldn't we align the buffer? 1093 fprintf(stderr, "r300_set_vertex_buffers: " 1094 "Unaligned buffer stride %i isn't supported.\n", 1095 buffers[i].stride); 1096 assert(0); 1097 abort(); 1098 } 1099 } 1100 } 1101 1102 for (i = 0; i < count; i++) { 1103 /* Why, yes, I AM casting away constness. How did you know? */ 1104 vbo = (struct pipe_vertex_buffer*)&buffers[i]; 1105 1106 /* Reference our buffer. */ 1107 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, vbo->buffer); 1108 1109 /* Skip NULL buffers */ 1110 if (!buffers[i].buffer) { 1111 continue; 1112 } 1113 1114 if (r300_buffer_is_user_buffer(vbo->buffer)) { 1115 any_user_buffer = TRUE; 1116 } 1117 1118 if (vbo->max_index == ~0) { 1119 /* Bogus value from broken state tracker; hax it. */ 1120 vbo->max_index = 1121 (vbo->buffer->size - vbo->buffer_offset) / vbo->stride; 1122 } 1123 1124 max_index = MIN2(vbo->max_index, max_index); 1125 } 1126 1127 for (; i < r300->vertex_buffer_count; i++) { 1128 /* Dereference any old buffers. */ 1129 pipe_buffer_reference(&r300->vertex_buffer[i].buffer, NULL); 1130 } 1131 1132 memcpy(r300->vertex_buffer, buffers, 1133 sizeof(struct pipe_vertex_buffer) * count); 1134 1135 r300->vertex_buffer_count = count; 1136 r300->vertex_buffer_max_index = max_index; 1137 r300->any_user_vbs = any_user_buffer; 1138 1139 if (r300->draw) { 1140 draw_flush(r300->draw); 1141 draw_set_vertex_buffers(r300->draw, count, buffers); 1142 } 1143} 1144 1145/* Update the PSC tables. */ 1146static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1147{ 1148 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1149 uint16_t type, swizzle; 1150 enum pipe_format format; 1151 unsigned i; 1152 1153 assert(velems->count <= 16); 1154 1155 /* Vertex shaders have no semantics on their inputs, 1156 * so PSC should just route stuff based on the vertex elements, 1157 * and not on attrib information. */ 1158 for (i = 0; i < velems->count; i++) { 1159 format = velems->velem[i].src_format; 1160 1161 type = r300_translate_vertex_data_type(format) | 1162 (i << R300_DST_VEC_LOC_SHIFT); 1163 swizzle = r300_translate_vertex_data_swizzle(format); 1164 1165 if (i & 1) { 1166 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1167 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1168 } else { 1169 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1170 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1171 } 1172 } 1173 1174 /* Set the last vector in the PSC. */ 1175 if (i) { 1176 i -= 1; 1177 } 1178 vstream->vap_prog_stream_cntl[i >> 1] |= 1179 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1180 1181 vstream->count = (i >> 1) + 1; 1182} 1183 1184static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1185 unsigned count, 1186 const struct pipe_vertex_element* attribs) 1187{ 1188 struct r300_screen* r300screen = r300_screen(pipe->screen); 1189 struct r300_vertex_element_state *velems; 1190 unsigned i, size; 1191 1192 assert(count <= PIPE_MAX_ATTRIBS); 1193 velems = CALLOC_STRUCT(r300_vertex_element_state); 1194 if (velems != NULL) { 1195 velems->count = count; 1196 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1197 1198 if (r300screen->caps->has_tcl) { 1199 /* Check if the format is aligned to the size of DWORD. */ 1200 for (i = 0; i < count; i++) { 1201 size = util_format_get_blocksize(attribs[i].src_format); 1202 1203 if (size % 4 != 0) { 1204 /* XXX Shouldn't we align the format? */ 1205 fprintf(stderr, "r300_create_vertex_elements_state: " 1206 "Unaligned format %s:%i isn't supported\n", 1207 util_format_name(attribs[i].src_format), size); 1208 assert(0); 1209 abort(); 1210 } 1211 } 1212 1213 r300_vertex_psc(velems); 1214 } 1215 } 1216 return velems; 1217} 1218 1219static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1220 void *state) 1221{ 1222 struct r300_context *r300 = r300_context(pipe); 1223 struct r300_vertex_element_state *velems = state; 1224 1225 if (velems == NULL) { 1226 return; 1227 } 1228 1229 r300->velems = velems; 1230 1231 if (r300->draw) { 1232 draw_flush(r300->draw); 1233 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1234 } 1235 1236 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1237 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1238} 1239 1240static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1241{ 1242 FREE(state); 1243} 1244 1245static void* r300_create_vs_state(struct pipe_context* pipe, 1246 const struct pipe_shader_state* shader) 1247{ 1248 struct r300_context* r300 = r300_context(pipe); 1249 1250 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1251 r300_vertex_shader_common_init(vs, shader); 1252 1253 if (r300_screen(pipe->screen)->caps->has_tcl) { 1254 r300_translate_vertex_shader(r300, vs); 1255 } else { 1256 vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); 1257 } 1258 1259 return vs; 1260} 1261 1262static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1263{ 1264 struct r300_context* r300 = r300_context(pipe); 1265 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1266 1267 if (vs == NULL) { 1268 r300->vs_state.state = NULL; 1269 return; 1270 } 1271 if (vs == r300->vs_state.state) { 1272 return; 1273 } 1274 r300->vs_state.state = vs; 1275 1276 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block 1277 if (r300->fs) { 1278 r300_vertex_shader_setup_wpos(r300); 1279 } 1280 memcpy(r300->vap_output_state.state, &vs->vap_out, 1281 sizeof(struct r300_vap_output_state)); 1282 r300->vap_output_state.dirty = TRUE; 1283 1284 /* The majority of the RS block bits is dependent on the vertex shader. */ 1285 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1286 1287 if (r300_screen(pipe->screen)->caps->has_tcl) { 1288 r300->vs_state.dirty = TRUE; 1289 r300->vs_state.size = vs->code.length + 9; 1290 1291 r300->pvs_flush.dirty = TRUE; 1292 1293 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1294 } else { 1295 draw_flush(r300->draw); 1296 draw_bind_vertex_shader(r300->draw, 1297 (struct draw_vertex_shader*)vs->draw_vs); 1298 } 1299} 1300 1301static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1302{ 1303 struct r300_context* r300 = r300_context(pipe); 1304 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1305 1306 if (r300_screen(pipe->screen)->caps->has_tcl) { 1307 rc_constants_destroy(&vs->code.constants); 1308 } else { 1309 draw_delete_vertex_shader(r300->draw, 1310 (struct draw_vertex_shader*)vs->draw_vs); 1311 } 1312 1313 FREE((void*)vs->state.tokens); 1314 FREE(shader); 1315} 1316 1317static void r300_set_constant_buffer(struct pipe_context *pipe, 1318 uint shader, uint index, 1319 struct pipe_buffer *buf) 1320{ 1321 struct r300_context* r300 = r300_context(pipe); 1322 struct r300_screen *r300screen = r300_screen(pipe->screen); 1323 void *mapped; 1324 int max_size = 0; 1325 1326 if (buf == NULL || buf->size == 0 || 1327 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) 1328 { 1329 r300->shader_constants[shader].count = 0; 1330 return; 1331 } 1332 1333 assert((buf->size % 4 * sizeof(float)) == 0); 1334 1335 /* Check the size of the constant buffer. */ 1336 switch (shader) { 1337 case PIPE_SHADER_VERTEX: 1338 max_size = 256; 1339 break; 1340 case PIPE_SHADER_FRAGMENT: 1341 if (r300screen->caps->is_r500) { 1342 max_size = 256; 1343 /* XXX Implement emission of r400's extended constant buffer. */ 1344 /*} else if (r300screen->caps->is_r400) { 1345 max_size = 64;*/ 1346 } else { 1347 max_size = 32; 1348 } 1349 break; 1350 default: 1351 assert(0); 1352 } 1353 1354 /* XXX Subtract immediates and RC_STATE_* variables. */ 1355 if (buf->size > (sizeof(float) * 4 * max_size)) { 1356 fprintf(stderr, "r300: Max size of the constant buffer is " 1357 "%i*4 floats.\n", max_size); 1358 abort(); 1359 } 1360 1361 memcpy(r300->shader_constants[shader].constants, mapped, buf->size); 1362 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); 1363 pipe_buffer_unmap(pipe->screen, buf); 1364 1365 if (shader == PIPE_SHADER_VERTEX) { 1366 if (r300screen->caps->has_tcl) { 1367 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1368 r300->pvs_flush.dirty = TRUE; 1369 } else if (r300->draw) { 1370 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1371 0, r300->shader_constants[PIPE_SHADER_VERTEX].constants, 1372 buf->size); 1373 } 1374 } else if (shader == PIPE_SHADER_FRAGMENT) { 1375 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1376 } 1377} 1378 1379void r300_init_state_functions(struct r300_context* r300) 1380{ 1381 r300->context.create_blend_state = r300_create_blend_state; 1382 r300->context.bind_blend_state = r300_bind_blend_state; 1383 r300->context.delete_blend_state = r300_delete_blend_state; 1384 1385 r300->context.set_blend_color = r300_set_blend_color; 1386 1387 r300->context.set_clip_state = r300_set_clip_state; 1388 1389 r300->context.set_constant_buffer = r300_set_constant_buffer; 1390 1391 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1392 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1393 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1394 1395 r300->context.set_stencil_ref = r300_set_stencil_ref; 1396 1397 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1398 1399 r300->context.create_fs_state = r300_create_fs_state; 1400 r300->context.bind_fs_state = r300_bind_fs_state; 1401 r300->context.delete_fs_state = r300_delete_fs_state; 1402 1403 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1404 1405 r300->context.create_rasterizer_state = r300_create_rs_state; 1406 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1407 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1408 1409 r300->context.create_sampler_state = r300_create_sampler_state; 1410 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1411 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1412 r300->context.delete_sampler_state = r300_delete_sampler_state; 1413 1414 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1415 r300->context.create_sampler_view = r300_create_sampler_view; 1416 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1417 1418 r300->context.set_scissor_state = r300_set_scissor_state; 1419 1420 r300->context.set_viewport_state = r300_set_viewport_state; 1421 1422 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1423 1424 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1425 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1426 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1427 1428 r300->context.create_vs_state = r300_create_vs_state; 1429 r300->context.bind_vs_state = r300_bind_vs_state; 1430 r300->context.delete_vs_state = r300_delete_vs_state; 1431} 1432