r300_state.c revision abc2f29dd2baa9c8aa672f07a1500e579364ab29
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_emit.h" 36#include "r300_reg.h" 37#include "r300_screen.h" 38#include "r300_screen_buffer.h" 39#include "r300_state_inlines.h" 40#include "r300_fs.h" 41#include "r300_texture.h" 42#include "r300_vs.h" 43#include "r300_winsys.h" 44 45/* r300_state: Functions used to intialize state context by translating 46 * Gallium state objects into semi-native r300 state objects. */ 47 48#define UPDATE_STATE(cso, atom) \ 49 if (cso != atom.state) { \ 50 atom.state = cso; \ 51 atom.dirty = TRUE; \ 52 } 53 54static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 55 unsigned dstRGB, unsigned dstA) 56{ 57 /* If the blend equation is ADD or REVERSE_SUBTRACT, 58 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 59 * will not be changed. 60 * Notice that the dst factors are the src factors inverted. */ 61 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 62 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 63 srcRGB == PIPE_BLENDFACTOR_ZERO) && 64 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 65 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 66 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 67 srcA == PIPE_BLENDFACTOR_ZERO) && 68 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 69 dstRGB == PIPE_BLENDFACTOR_ONE) && 70 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 71 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 72 dstA == PIPE_BLENDFACTOR_ONE); 73} 74 75static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 76 unsigned dstRGB, unsigned dstA) 77{ 78 /* If the blend equation is ADD or REVERSE_SUBTRACT, 79 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 80 * will not be changed. 81 * Notice that the dst factors are the src factors inverted. */ 82 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 83 srcRGB == PIPE_BLENDFACTOR_ZERO) && 84 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 85 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 86 srcA == PIPE_BLENDFACTOR_ZERO) && 87 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 88 dstRGB == PIPE_BLENDFACTOR_ONE) && 89 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 90 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 91 dstA == PIPE_BLENDFACTOR_ONE); 92} 93 94static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 95 unsigned dstRGB, unsigned dstA) 96{ 97 /* If the blend equation is ADD or REVERSE_SUBTRACT, 98 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 99 * will not be changed. 100 * Notice that the dst factors are the src factors inverted. */ 101 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 102 srcRGB == PIPE_BLENDFACTOR_ZERO) && 103 (srcA == PIPE_BLENDFACTOR_ZERO) && 104 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 105 dstRGB == PIPE_BLENDFACTOR_ONE) && 106 (dstA == PIPE_BLENDFACTOR_ONE); 107} 108 109static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 110 unsigned dstRGB, unsigned dstA) 111{ 112 /* If the blend equation is ADD or REVERSE_SUBTRACT, 113 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 114 * will not be changed. 115 * Notice that the dst factors are the src factors inverted. */ 116 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 117 srcRGB == PIPE_BLENDFACTOR_ZERO) && 118 (srcA == PIPE_BLENDFACTOR_ZERO) && 119 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 120 dstRGB == PIPE_BLENDFACTOR_ONE) && 121 (dstA == PIPE_BLENDFACTOR_ONE); 122} 123 124static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 125 unsigned dstRGB, unsigned dstA) 126{ 127 /* If the blend equation is ADD or REVERSE_SUBTRACT, 128 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 129 * the colorbuffer will not be changed. 130 * Notice that the dst factors are the src factors inverted. */ 131 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 132 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 133 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 134 srcRGB == PIPE_BLENDFACTOR_ZERO) && 135 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 136 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 137 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 138 srcA == PIPE_BLENDFACTOR_ZERO) && 139 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 140 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 141 dstRGB == PIPE_BLENDFACTOR_ONE) && 142 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 143 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 144 dstA == PIPE_BLENDFACTOR_ONE); 145} 146 147static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 148 unsigned dstRGB, unsigned dstA) 149{ 150 /* If the blend equation is ADD or REVERSE_SUBTRACT, 151 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 152 * the colorbuffer will not be changed. 153 * Notice that the dst factors are the src factors inverted. */ 154 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 155 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 156 srcRGB == PIPE_BLENDFACTOR_ZERO) && 157 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 158 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 159 srcA == PIPE_BLENDFACTOR_ZERO) && 160 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 161 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 162 dstRGB == PIPE_BLENDFACTOR_ONE) && 163 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 164 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 165 dstA == PIPE_BLENDFACTOR_ONE); 166} 167 168static unsigned bgra_cmask(unsigned mask) 169{ 170 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 171 172 return ((mask & PIPE_MASK_R) << 2) | 173 ((mask & PIPE_MASK_B) >> 2) | 174 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 175} 176 177/* Create a new blend state based on the CSO blend state. 178 * 179 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 180static void* r300_create_blend_state(struct pipe_context* pipe, 181 const struct pipe_blend_state* state) 182{ 183 struct r300_screen* r300screen = r300_screen(pipe->screen); 184 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 185 186 if (state->rt[0].blend_enable) 187 { 188 unsigned eqRGB = state->rt[0].rgb_func; 189 unsigned srcRGB = state->rt[0].rgb_src_factor; 190 unsigned dstRGB = state->rt[0].rgb_dst_factor; 191 192 unsigned eqA = state->rt[0].alpha_func; 193 unsigned srcA = state->rt[0].alpha_src_factor; 194 unsigned dstA = state->rt[0].alpha_dst_factor; 195 196 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 197 * this is just the crappy D3D naming */ 198 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 199 r300_translate_blend_function(eqRGB) | 200 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 201 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 202 203 /* Optimization: some operations do not require the destination color. 204 * 205 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 206 * otherwise blending gives incorrect results. It seems to be 207 * a hardware bug. */ 208 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 209 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 210 dstRGB != PIPE_BLENDFACTOR_ZERO || 211 dstA != PIPE_BLENDFACTOR_ZERO || 212 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 213 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 214 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 215 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 216 srcA == PIPE_BLENDFACTOR_DST_COLOR || 217 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 218 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 219 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 220 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 221 /* Enable reading from the colorbuffer. */ 222 blend->blend_control |= R300_READ_ENABLE; 223 224 if (r300screen->caps.is_r500) { 225 /* Optimization: Depending on incoming pixels, we can 226 * conditionally disable the reading in hardware... */ 227 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 228 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 229 /* Disable reading if SRC_ALPHA == 0. */ 230 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 231 dstRGB == PIPE_BLENDFACTOR_ZERO) && 232 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 233 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 234 dstA == PIPE_BLENDFACTOR_ZERO)) { 235 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 236 } 237 238 /* Disable reading if SRC_ALPHA == 1. */ 239 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 240 dstRGB == PIPE_BLENDFACTOR_ZERO) && 241 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 242 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 243 dstA == PIPE_BLENDFACTOR_ZERO)) { 244 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 245 } 246 } 247 } 248 } 249 250 /* Optimization: discard pixels which don't change the colorbuffer. 251 * 252 * The code below is non-trivial and some math is involved. 253 * 254 * Discarding pixels must be disabled when FP16 AA is enabled. 255 * This is a hardware bug. Also, this implementation wouldn't work 256 * with FP blending enabled and equation clamping disabled. 257 * 258 * Equations other than ADD are rarely used and therefore won't be 259 * optimized. */ 260 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 261 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 262 /* ADD: X+Y 263 * REVERSE_SUBTRACT: Y-X 264 * 265 * The idea is: 266 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 267 * then CB will not be changed. 268 * 269 * Given the srcFactor and dstFactor variables, we can derive 270 * what src and dst should be equal to and discard appropriate 271 * pixels. 272 */ 273 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 274 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 275 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 276 dstRGB, dstA)) { 277 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 278 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 279 dstRGB, dstA)) { 280 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 281 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 282 dstRGB, dstA)) { 283 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 284 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 285 dstRGB, dstA)) { 286 blend->blend_control |= 287 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 288 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 289 dstRGB, dstA)) { 290 blend->blend_control |= 291 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 292 } 293 } 294 295 /* separate alpha */ 296 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 297 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 298 blend->alpha_blend_control = 299 r300_translate_blend_function(eqA) | 300 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 301 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 302 } 303 } 304 305 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 306 if (state->logicop_enable) { 307 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 308 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 309 } 310 311 /* Color channel masks for all MRTs. */ 312 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); 313 if (r300screen->caps.is_r500 && state->independent_blend_enable) { 314 if (state->rt[1].blend_enable) { 315 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 316 } 317 if (state->rt[2].blend_enable) { 318 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 319 } 320 if (state->rt[3].blend_enable) { 321 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 322 } 323 } 324 325 /* Neither fglrx nor classic r300 ever set this, regardless of dithering 326 * state. Since it's an optional implementation detail, we can leave it 327 * out and never dither. 328 * 329 * This could be revisited if we ever get quality or conformance hints. 330 * 331 if (state->dither) { 332 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 333 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 334 } 335 */ 336 337 return (void*)blend; 338} 339 340/* Bind blend state. */ 341static void r300_bind_blend_state(struct pipe_context* pipe, 342 void* state) 343{ 344 struct r300_context* r300 = r300_context(pipe); 345 346 UPDATE_STATE(state, r300->blend_state); 347} 348 349/* Free blend state. */ 350static void r300_delete_blend_state(struct pipe_context* pipe, 351 void* state) 352{ 353 FREE(state); 354} 355 356/* Convert float to 10bit integer */ 357static unsigned float_to_fixed10(float f) 358{ 359 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 360} 361 362/* Set blend color. 363 * Setup both R300 and R500 registers, figure out later which one to write. */ 364static void r300_set_blend_color(struct pipe_context* pipe, 365 const struct pipe_blend_color* color) 366{ 367 struct r300_context* r300 = r300_context(pipe); 368 struct r300_blend_color_state* state = 369 (struct r300_blend_color_state*)r300->blend_color_state.state; 370 union util_color uc; 371 372 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 373 state->blend_color = uc.ui; 374 375 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 376 state->blend_color_red_alpha = 377 float_to_fixed10(color->color[0]) | 378 (float_to_fixed10(color->color[3]) << 16); 379 state->blend_color_green_blue = 380 float_to_fixed10(color->color[2]) | 381 (float_to_fixed10(color->color[1]) << 16); 382 383 r300->blend_color_state.size = r300->screen->caps.is_r500 ? 3 : 2; 384 r300->blend_color_state.dirty = TRUE; 385} 386 387static void r300_set_clip_state(struct pipe_context* pipe, 388 const struct pipe_clip_state* state) 389{ 390 struct r300_context* r300 = r300_context(pipe); 391 392 r300->clip = *state; 393 394 if (r300->screen->caps.has_tcl) { 395 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 396 r300->clip_state.size = 29; 397 } else { 398 draw_flush(r300->draw); 399 draw_set_clip_state(r300->draw, state); 400 r300->clip_state.size = 2; 401 } 402 403 r300->clip_state.dirty = TRUE; 404} 405 406/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 407 * 408 * This contains the depth buffer, stencil buffer, alpha test, and such. 409 * On the Radeon, depth and stencil buffer setup are intertwined, which is 410 * the reason for some of the strange-looking assignments across registers. */ 411static void* 412 r300_create_dsa_state(struct pipe_context* pipe, 413 const struct pipe_depth_stencil_alpha_state* state) 414{ 415 struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; 416 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 417 418 /* Depth test setup. */ 419 if (state->depth.enabled) { 420 dsa->z_buffer_control |= R300_Z_ENABLE; 421 422 if (state->depth.writemask) { 423 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 424 } 425 426 dsa->z_stencil_control |= 427 (r300_translate_depth_stencil_function(state->depth.func) << 428 R300_Z_FUNC_SHIFT); 429 } 430 431 /* Stencil buffer setup. */ 432 if (state->stencil[0].enabled) { 433 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 434 dsa->z_stencil_control |= 435 (r300_translate_depth_stencil_function(state->stencil[0].func) << 436 R300_S_FRONT_FUNC_SHIFT) | 437 (r300_translate_stencil_op(state->stencil[0].fail_op) << 438 R300_S_FRONT_SFAIL_OP_SHIFT) | 439 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 440 R300_S_FRONT_ZPASS_OP_SHIFT) | 441 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 442 R300_S_FRONT_ZFAIL_OP_SHIFT); 443 444 dsa->stencil_ref_mask = 445 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 446 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 447 448 if (state->stencil[1].enabled) { 449 dsa->two_sided = TRUE; 450 451 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 452 dsa->z_stencil_control |= 453 (r300_translate_depth_stencil_function(state->stencil[1].func) << 454 R300_S_BACK_FUNC_SHIFT) | 455 (r300_translate_stencil_op(state->stencil[1].fail_op) << 456 R300_S_BACK_SFAIL_OP_SHIFT) | 457 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 458 R300_S_BACK_ZPASS_OP_SHIFT) | 459 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 460 R300_S_BACK_ZFAIL_OP_SHIFT); 461 462 dsa->stencil_ref_bf = 463 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | 464 (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); 465 466 if (caps->is_r500) { 467 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 468 } else { 469 dsa->stencil_ref_bf_fallback = 470 (state->stencil[0].valuemask != state->stencil[1].valuemask || 471 state->stencil[0].writemask != state->stencil[1].writemask); 472 } 473 } 474 } 475 476 /* Alpha test setup. */ 477 if (state->alpha.enabled) { 478 dsa->alpha_function = 479 r300_translate_alpha_function(state->alpha.func) | 480 R300_FG_ALPHA_FUNC_ENABLE; 481 482 /* We could use 10bit alpha ref but who needs that? */ 483 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 484 485 if (caps->is_r500) 486 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 487 } 488 489 return (void*)dsa; 490} 491 492static void r300_update_stencil_ref_fallback_status(struct r300_context *r300) 493{ 494 struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; 495 496 if (r300->screen->caps.is_r500) { 497 return; 498 } 499 500 r300->stencil_ref_bf_fallback = 501 dsa->stencil_ref_bf_fallback || 502 (dsa->two_sided && 503 r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]); 504} 505 506/* Bind DSA state. */ 507static void r300_bind_dsa_state(struct pipe_context* pipe, 508 void* state) 509{ 510 struct r300_context* r300 = r300_context(pipe); 511 512 if (!state) { 513 return; 514 } 515 516 UPDATE_STATE(state, r300->dsa_state); 517 518 r300_update_stencil_ref_fallback_status(r300); 519} 520 521/* Free DSA state. */ 522static void r300_delete_dsa_state(struct pipe_context* pipe, 523 void* state) 524{ 525 FREE(state); 526} 527 528static void r300_set_stencil_ref(struct pipe_context* pipe, 529 const struct pipe_stencil_ref* sr) 530{ 531 struct r300_context* r300 = r300_context(pipe); 532 533 r300->stencil_ref = *sr; 534 r300->dsa_state.dirty = TRUE; 535 536 r300_update_stencil_ref_fallback_status(r300); 537} 538 539/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 540static void r300_fb_update_tiling_flags(struct r300_context *r300, 541 const struct pipe_framebuffer_state *old_state, 542 const struct pipe_framebuffer_state *new_state) 543{ 544 struct r300_texture *tex; 545 unsigned i, j, level; 546 547 /* Reset tiling flags for old surfaces to default values. */ 548 for (i = 0; i < old_state->nr_cbufs; i++) { 549 for (j = 0; j < new_state->nr_cbufs; j++) { 550 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { 551 break; 552 } 553 } 554 /* If not binding the surface again... */ 555 if (j != new_state->nr_cbufs) { 556 continue; 557 } 558 559 tex = r300_texture(old_state->cbufs[i]->texture); 560 561 if (tex) { 562 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 563 tex->pitch[0], 564 tex->microtile, 565 tex->macrotile); 566 } 567 } 568 if (old_state->zsbuf && 569 (!new_state->zsbuf || 570 old_state->zsbuf->texture != new_state->zsbuf->texture)) { 571 tex = r300_texture(old_state->zsbuf->texture); 572 573 if (tex) { 574 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 575 tex->pitch[0], 576 tex->microtile, 577 tex->macrotile); 578 } 579 } 580 581 /* Set tiling flags for new surfaces. */ 582 for (i = 0; i < new_state->nr_cbufs; i++) { 583 tex = r300_texture(new_state->cbufs[i]->texture); 584 level = new_state->cbufs[i]->level; 585 586 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 587 tex->pitch[level], 588 tex->microtile, 589 tex->mip_macrotile[level]); 590 } 591 if (new_state->zsbuf) { 592 tex = r300_texture(new_state->zsbuf->texture); 593 level = new_state->zsbuf->level; 594 595 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 596 tex->pitch[level], 597 tex->microtile, 598 tex->mip_macrotile[level]); 599 } 600} 601 602static void 603 r300_set_framebuffer_state(struct pipe_context* pipe, 604 const struct pipe_framebuffer_state* state) 605{ 606 struct r300_context* r300 = r300_context(pipe); 607 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 608 unsigned max_width, max_height; 609 uint32_t zbuffer_bpp = 0; 610 611 if (state->nr_cbufs > 4) { 612 fprintf(stderr, "r300: Implementation error: Too many MRTs in %s, " 613 "refusing to bind framebuffer state!\n", __FUNCTION__); 614 return; 615 } 616 617 if (r300->screen->caps.is_r500) { 618 max_width = max_height = 4096; 619 } else if (r300->screen->caps.is_r400) { 620 max_width = max_height = 4021; 621 } else { 622 max_width = max_height = 2560; 623 } 624 625 if (state->width > max_width || state->height > max_height) { 626 fprintf(stderr, "r300: Implementation error: Render targets are too " 627 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 628 return; 629 } 630 631 if (r300->draw) { 632 draw_flush(r300->draw); 633 } 634 635 r300->fb_state.dirty = TRUE; 636 637 /* If nr_cbufs is changed from zero to non-zero or vice versa... */ 638 if (!!old_state->nr_cbufs != !!state->nr_cbufs) { 639 r300->blend_state.dirty = TRUE; 640 } 641 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 642 if (!!old_state->zsbuf != !!state->zsbuf) { 643 r300->dsa_state.dirty = TRUE; 644 } 645 646 r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); 647 648 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 649 650 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + 651 (state->zsbuf ? 10 : 0) + 11; 652 653 /* Polygon offset depends on the zbuffer bit depth. */ 654 if (state->zsbuf && r300->polygon_offset_enabled) { 655 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 656 case 2: 657 zbuffer_bpp = 16; 658 break; 659 case 4: 660 zbuffer_bpp = 24; 661 break; 662 } 663 664 if (r300->zbuffer_bpp != zbuffer_bpp) { 665 r300->zbuffer_bpp = zbuffer_bpp; 666 r300->rs_state.dirty = TRUE; 667 } 668 } 669} 670 671/* Create fragment shader state. */ 672static void* r300_create_fs_state(struct pipe_context* pipe, 673 const struct pipe_shader_state* shader) 674{ 675 struct r300_fragment_shader* fs = NULL; 676 677 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 678 679 /* Copy state directly into shader. */ 680 fs->state = *shader; 681 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 682 683 return (void*)fs; 684} 685 686static void r300_mark_fs_code_dirty(struct r300_context *r300) 687{ 688 struct r300_fragment_shader* fs = r300_fs(r300); 689 690 r300->fs.dirty = TRUE; 691 r300->fs_rc_constant_state.dirty = TRUE; 692 r300->fs_constants.dirty = TRUE; 693 694 if (r300->screen->caps.is_r500) { 695 r300->fs.size = r500_get_fs_atom_size(r300); 696 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; 697 r300->fs_constants.size = fs->shader->externals_count * 4 + 3; 698 } else { 699 r300->fs.size = r300_get_fs_atom_size(r300); 700 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; 701 r300->fs_constants.size = fs->shader->externals_count * 4 + 1; 702 } 703} 704 705/* Bind fragment shader state. */ 706static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 707{ 708 struct r300_context* r300 = r300_context(pipe); 709 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 710 711 if (fs == NULL) { 712 r300->fs.state = NULL; 713 return; 714 } 715 716 r300->fs.state = fs; 717 r300_pick_fragment_shader(r300); 718 r300_mark_fs_code_dirty(r300); 719 720 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 721 722 if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) { 723 r300->vap_output_state.dirty = TRUE; 724 } 725} 726 727/* Delete fragment shader state. */ 728static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 729{ 730 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 731 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 732 733 while (ptr) { 734 tmp = ptr; 735 ptr = ptr->next; 736 rc_constants_destroy(&tmp->code.constants); 737 FREE(tmp); 738 } 739 FREE((void*)fs->state.tokens); 740 FREE(shader); 741} 742 743static void r300_set_polygon_stipple(struct pipe_context* pipe, 744 const struct pipe_poly_stipple* state) 745{ 746 /* XXX no idea how to set this up, but not terribly important */ 747} 748 749/* Create a new rasterizer state based on the CSO rasterizer state. 750 * 751 * This is a very large chunk of state, and covers most of the graphics 752 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 753 * 754 * In a not entirely unironic sidenote, this state has nearly nothing to do 755 * with the actual block on the Radeon called the rasterizer (RS). */ 756static void* r300_create_rs_state(struct pipe_context* pipe, 757 const struct pipe_rasterizer_state* state) 758{ 759 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 760 int i; 761 762 /* Copy rasterizer state for Draw. */ 763 rs->rs = *state; 764 765#ifdef PIPE_ARCH_LITTLE_ENDIAN 766 rs->vap_control_status = R300_VC_NO_SWAP; 767#else 768 rs->vap_control_status = R300_VC_32BIT_SWAP; 769#endif 770 771 /* If no TCL engine is present, turn off the HW TCL. */ 772 if (!r300_screen(pipe->screen)->caps.has_tcl) { 773 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 774 } 775 776 rs->point_size = pack_float_16_6x(state->point_size) | 777 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 778 779 rs->line_control = pack_float_16_6x(state->line_width) | 780 R300_GA_LINE_CNTL_END_TYPE_COMP; 781 782 /* Enable polygon mode */ 783 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 784 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 785 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 786 } 787 788 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 789 if (state->front_winding == PIPE_WINDING_CW) { 790 rs->cull_mode = R300_FRONT_FACE_CW; 791 792 /* Polygon offset */ 793 if (state->offset_cw) { 794 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 795 } 796 if (state->offset_ccw) { 797 rs->polygon_offset_enable |= R300_BACK_ENABLE; 798 } 799 800 /* Polygon mode */ 801 if (rs->polygon_mode) { 802 rs->polygon_mode |= 803 r300_translate_polygon_mode_front(state->fill_cw); 804 rs->polygon_mode |= 805 r300_translate_polygon_mode_back(state->fill_ccw); 806 } 807 } else { 808 rs->cull_mode = R300_FRONT_FACE_CCW; 809 810 /* Polygon offset */ 811 if (state->offset_ccw) { 812 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 813 } 814 if (state->offset_cw) { 815 rs->polygon_offset_enable |= R300_BACK_ENABLE; 816 } 817 818 /* Polygon mode */ 819 if (rs->polygon_mode) { 820 rs->polygon_mode |= 821 r300_translate_polygon_mode_front(state->fill_ccw); 822 rs->polygon_mode |= 823 r300_translate_polygon_mode_back(state->fill_cw); 824 } 825 } 826 if (state->front_winding & state->cull_mode) { 827 rs->cull_mode |= R300_CULL_FRONT; 828 } 829 if (~(state->front_winding) & state->cull_mode) { 830 rs->cull_mode |= R300_CULL_BACK; 831 } 832 833 if (rs->polygon_offset_enable) { 834 rs->depth_offset = state->offset_units; 835 rs->depth_scale = state->offset_scale; 836 } 837 838 if (state->line_stipple_enable) { 839 rs->line_stipple_config = 840 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 841 (fui((float)state->line_stipple_factor) & 842 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 843 /* XXX this might need to be scaled up */ 844 rs->line_stipple_value = state->line_stipple_pattern; 845 } 846 847 if (state->flatshade) { 848 rs->color_control = R300_SHADE_MODEL_FLAT; 849 } else { 850 rs->color_control = R300_SHADE_MODEL_SMOOTH; 851 } 852 853 rs->clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 854 855 /* Point sprites */ 856 if (state->sprite_coord_enable) { 857 rs->stuffing_enable = R300_GB_POINT_STUFF_ENABLE; 858 for (i = 0; i < 8; i++) { 859 if (state->sprite_coord_enable & (1 << i)) 860 rs->stuffing_enable |= 861 R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); 862 } 863 864 rs->point_texcoord_left = 0.0f; 865 rs->point_texcoord_right = 1.0f; 866 867 switch (state->sprite_coord_mode) { 868 case PIPE_SPRITE_COORD_UPPER_LEFT: 869 rs->point_texcoord_top = 0.0f; 870 rs->point_texcoord_bottom = 1.0f; 871 break; 872 case PIPE_SPRITE_COORD_LOWER_LEFT: 873 rs->point_texcoord_top = 1.0f; 874 rs->point_texcoord_bottom = 0.0f; 875 break; 876 } 877 } 878 879 return (void*)rs; 880} 881 882/* Bind rasterizer state. */ 883static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 884{ 885 struct r300_context* r300 = r300_context(pipe); 886 struct r300_rs_state* rs = (struct r300_rs_state*)state; 887 int last_sprite_coord_enable = r300->sprite_coord_enable; 888 889 if (r300->draw) { 890 draw_flush(r300->draw); 891 draw_set_rasterizer_state(r300->draw, &rs->rs); 892 } 893 894 if (rs) { 895 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 896 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 897 } else { 898 r300->polygon_offset_enabled = FALSE; 899 r300->sprite_coord_enable = 0; 900 } 901 902 UPDATE_STATE(state, r300->rs_state); 903 r300->rs_state.size = 26 + (r300->polygon_offset_enabled ? 5 : 0); 904 905 if (last_sprite_coord_enable != r300->sprite_coord_enable) { 906 r300->rs_block_state.dirty = TRUE; 907 } 908} 909 910/* Free rasterizer state. */ 911static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 912{ 913 FREE(state); 914} 915 916static void* 917 r300_create_sampler_state(struct pipe_context* pipe, 918 const struct pipe_sampler_state* state) 919{ 920 struct r300_context* r300 = r300_context(pipe); 921 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 922 boolean is_r500 = r300->screen->caps.is_r500; 923 int lod_bias; 924 union util_color uc; 925 926 sampler->state = *state; 927 928 sampler->filter0 |= 929 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 930 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 931 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 932 933 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 934 state->mag_img_filter, 935 state->min_mip_filter, 936 state->max_anisotropy > 0); 937 938 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 939 940 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 941 /* We must pass these to the merge function to clamp them properly. */ 942 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 943 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 944 945 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 946 947 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 948 949 /* This is very high quality anisotropic filtering for R5xx. 950 * It's good for benchmarking the performance of texturing but 951 * in practice we don't want to slow down the driver because it's 952 * a pretty good performance killer. Feel free to play with it. */ 953 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 954 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 955 } 956 957 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 958 sampler->border_color = uc.ui; 959 960 /* R500-specific fixups and optimizations */ 961 if (r300->screen->caps.is_r500) { 962 sampler->filter1 |= R500_BORDER_FIX; 963 } 964 965 return (void*)sampler; 966} 967 968static void r300_bind_sampler_states(struct pipe_context* pipe, 969 unsigned count, 970 void** states) 971{ 972 struct r300_context* r300 = r300_context(pipe); 973 struct r300_textures_state* state = 974 (struct r300_textures_state*)r300->textures_state.state; 975 unsigned tex_units = r300->screen->caps.num_tex_units; 976 977 if (count > tex_units) { 978 return; 979 } 980 981 memcpy(state->sampler_states, states, sizeof(void*) * count); 982 state->sampler_state_count = count; 983 984 r300->textures_state.dirty = TRUE; 985 986 /* Pick a fragment shader based on the texture compare state. */ 987 if (r300->fs.state && count) { 988 if (r300_pick_fragment_shader(r300)) { 989 r300_mark_fs_code_dirty(r300); 990 } 991 } 992} 993 994static void r300_lacks_vertex_textures(struct pipe_context* pipe, 995 unsigned count, 996 void** states) 997{ 998} 999 1000static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1001{ 1002 FREE(state); 1003} 1004 1005static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 1006 unsigned count, 1007 struct pipe_sampler_view** views) 1008{ 1009 struct r300_context* r300 = r300_context(pipe); 1010 struct r300_textures_state* state = 1011 (struct r300_textures_state*)r300->textures_state.state; 1012 struct r300_texture *texture; 1013 unsigned i; 1014 unsigned tex_units = r300->screen->caps.num_tex_units; 1015 boolean is_r500 = r300->screen->caps.is_r500; 1016 boolean dirty_tex = FALSE; 1017 1018 if (count > tex_units) { 1019 return; 1020 } 1021 1022 for (i = 0; i < count; i++) { 1023 if (&state->sampler_views[i]->base != views[i]) { 1024 pipe_sampler_view_reference( 1025 (struct pipe_sampler_view**)&state->sampler_views[i], 1026 views[i]); 1027 1028 if (!views[i]) { 1029 continue; 1030 } 1031 1032 /* A new sampler view (= texture)... */ 1033 dirty_tex = TRUE; 1034 1035 /* R300-specific - set the texrect factor in the fragment shader */ 1036 texture = r300_texture(views[i]->texture); 1037 if (!is_r500 && texture->uses_pitch) { 1038 r300->fs_rc_constant_state.dirty = TRUE; 1039 } 1040 } 1041 } 1042 1043 for (i = count; i < tex_units; i++) { 1044 if (state->sampler_views[i]) { 1045 pipe_sampler_view_reference( 1046 (struct pipe_sampler_view**)&state->sampler_views[i], 1047 NULL); 1048 } 1049 } 1050 1051 state->sampler_view_count = count; 1052 1053 r300->textures_state.dirty = TRUE; 1054 1055 if (dirty_tex) { 1056 r300->texture_cache_inval.dirty = TRUE; 1057 } 1058} 1059 1060static struct pipe_sampler_view * 1061r300_create_sampler_view(struct pipe_context *pipe, 1062 struct pipe_resource *texture, 1063 const struct pipe_sampler_view *templ) 1064{ 1065 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1066 struct r300_texture *tex = r300_texture(texture); 1067 unsigned char swizzle[4]; 1068 1069 if (view) { 1070 view->base = *templ; 1071 view->base.reference.count = 1; 1072 view->base.context = pipe; 1073 view->base.texture = NULL; 1074 pipe_resource_reference(&view->base.texture, texture); 1075 1076 swizzle[0] = templ->swizzle_r; 1077 swizzle[1] = templ->swizzle_g; 1078 swizzle[2] = templ->swizzle_b; 1079 swizzle[3] = templ->swizzle_a; 1080 1081 /* XXX Enable swizzles when they become supported. Now we get RGBA 1082 * everywhere. And do testing! */ 1083 view->format = tex->tx_format; 1084 view->format.format1 |= r300_translate_texformat(templ->format, 1085 0); /*swizzle);*/ 1086 if (r300_screen(pipe->screen)->caps.is_r500) { 1087 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1088 } 1089 } 1090 1091 return (struct pipe_sampler_view*)view; 1092} 1093 1094static void 1095r300_sampler_view_destroy(struct pipe_context *pipe, 1096 struct pipe_sampler_view *view) 1097{ 1098 pipe_resource_reference(&view->texture, NULL); 1099 FREE(view); 1100} 1101 1102static void r300_set_scissor_state(struct pipe_context* pipe, 1103 const struct pipe_scissor_state* state) 1104{ 1105 struct r300_context* r300 = r300_context(pipe); 1106 1107 memcpy(r300->scissor_state.state, state, 1108 sizeof(struct pipe_scissor_state)); 1109 1110 r300->scissor_state.dirty = TRUE; 1111} 1112 1113static void r300_set_viewport_state(struct pipe_context* pipe, 1114 const struct pipe_viewport_state* state) 1115{ 1116 struct r300_context* r300 = r300_context(pipe); 1117 struct r300_viewport_state* viewport = 1118 (struct r300_viewport_state*)r300->viewport_state.state; 1119 1120 r300->viewport = *state; 1121 1122 /* Do the transform in HW. */ 1123 viewport->vte_control = R300_VTX_W0_FMT; 1124 1125 if (state->scale[0] != 1.0f) { 1126 viewport->xscale = state->scale[0]; 1127 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1128 } 1129 if (state->scale[1] != 1.0f) { 1130 viewport->yscale = state->scale[1]; 1131 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1132 } 1133 if (state->scale[2] != 1.0f) { 1134 viewport->zscale = state->scale[2]; 1135 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1136 } 1137 if (state->translate[0] != 0.0f) { 1138 viewport->xoffset = state->translate[0]; 1139 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1140 } 1141 if (state->translate[1] != 0.0f) { 1142 viewport->yoffset = state->translate[1]; 1143 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1144 } 1145 if (state->translate[2] != 0.0f) { 1146 viewport->zoffset = state->translate[2]; 1147 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1148 } 1149 1150 r300->viewport_state.dirty = TRUE; 1151 if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1152 r300->fs_rc_constant_state.dirty = TRUE; 1153 } 1154} 1155 1156static void r300_set_vertex_buffers(struct pipe_context* pipe, 1157 unsigned count, 1158 const struct pipe_vertex_buffer* buffers) 1159{ 1160 struct r300_context* r300 = r300_context(pipe); 1161 struct pipe_vertex_buffer *vbo; 1162 unsigned i, max_index = (1 << 24) - 1; 1163 boolean any_user_buffer = FALSE; 1164 1165 if (count == r300->vertex_buffer_count && 1166 memcmp(r300->vertex_buffer, buffers, 1167 sizeof(struct pipe_vertex_buffer) * count) == 0) { 1168 return; 1169 } 1170 1171 /* Check if the stride is aligned to the size of DWORD. */ 1172 for (i = 0; i < count; i++) { 1173 if (buffers[i].buffer) { 1174 if (buffers[i].stride % 4 != 0) { 1175 // XXX Shouldn't we align the buffer? 1176 fprintf(stderr, "r300: set_vertex_buffers: " 1177 "Unaligned buffer stride %i isn't supported.\n", 1178 buffers[i].stride); 1179 abort(); 1180 } 1181 } 1182 } 1183 1184 for (i = 0; i < count; i++) { 1185 /* Why, yes, I AM casting away constness. How did you know? */ 1186 vbo = (struct pipe_vertex_buffer*)&buffers[i]; 1187 1188 /* Reference our buffer. */ 1189 pipe_resource_reference(&r300->vertex_buffer[i].buffer, vbo->buffer); 1190 1191 /* Skip NULL buffers */ 1192 if (!buffers[i].buffer) { 1193 continue; 1194 } 1195 1196 if (r300_buffer_is_user_buffer(vbo->buffer)) { 1197 any_user_buffer = TRUE; 1198 } 1199 1200 if (vbo->max_index == ~0) { 1201 /* Bogus value from broken state tracker; hax it. */ 1202 /* TODO - more hax - fixes doom3 from almos on irc */ 1203 if (!vbo->stride) { 1204 fprintf(stderr, "r300: got a VBO with stride 0 fixing up to stide 4\n"); 1205 vbo->stride = 4; 1206 } 1207 vbo->max_index = 1208 (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; 1209 } 1210 1211 max_index = MIN2(vbo->max_index, max_index); 1212 } 1213 1214 for (; i < r300->vertex_buffer_count; i++) { 1215 /* Dereference any old buffers. */ 1216 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); 1217 } 1218 1219 memcpy(r300->vertex_buffer, buffers, 1220 sizeof(struct pipe_vertex_buffer) * count); 1221 1222 r300->vertex_buffer_count = count; 1223 r300->vertex_buffer_max_index = max_index; 1224 r300->any_user_vbs = any_user_buffer; 1225 1226 if (r300->draw) { 1227 draw_flush(r300->draw); 1228 draw_set_vertex_buffers(r300->draw, count, buffers); 1229 } 1230} 1231 1232/* Update the PSC tables. */ 1233static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1234{ 1235 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1236 uint16_t type, swizzle; 1237 enum pipe_format format; 1238 unsigned i; 1239 1240 if (velems->count > 16) { 1241 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1242 " requested %i, using 16.\n", velems->count); 1243 velems->count = 16; 1244 } 1245 1246 /* Vertex shaders have no semantics on their inputs, 1247 * so PSC should just route stuff based on the vertex elements, 1248 * and not on attrib information. */ 1249 for (i = 0; i < velems->count; i++) { 1250 format = velems->velem[i].src_format; 1251 1252 type = r300_translate_vertex_data_type(format) | 1253 (i << R300_DST_VEC_LOC_SHIFT); 1254 swizzle = r300_translate_vertex_data_swizzle(format); 1255 1256 if (i & 1) { 1257 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1258 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1259 } else { 1260 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1261 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1262 } 1263 } 1264 1265 /* Set the last vector in the PSC. */ 1266 if (i) { 1267 i -= 1; 1268 } 1269 vstream->vap_prog_stream_cntl[i >> 1] |= 1270 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1271 1272 vstream->count = (i >> 1) + 1; 1273} 1274 1275static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1276 unsigned count, 1277 const struct pipe_vertex_element* attribs) 1278{ 1279 struct r300_vertex_element_state *velems; 1280 unsigned i, size; 1281 1282 assert(count <= PIPE_MAX_ATTRIBS); 1283 velems = CALLOC_STRUCT(r300_vertex_element_state); 1284 if (velems != NULL) { 1285 velems->count = count; 1286 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1287 1288 if (r300_screen(pipe->screen)->caps.has_tcl) { 1289 /* Check if the format is aligned to the size of DWORD. */ 1290 for (i = 0; i < count; i++) { 1291 size = util_format_get_blocksize(attribs[i].src_format); 1292 1293 if (size % 4 != 0) { 1294 /* XXX Shouldn't we align the format? */ 1295 fprintf(stderr, "r300_create_vertex_elements_state: " 1296 "Unaligned format %s:%i isn't supported\n", 1297 util_format_name(attribs[i].src_format), size); 1298 assert(0); 1299 abort(); 1300 } 1301 } 1302 1303 r300_vertex_psc(velems); 1304 } 1305 } 1306 return velems; 1307} 1308 1309static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1310 void *state) 1311{ 1312 struct r300_context *r300 = r300_context(pipe); 1313 struct r300_vertex_element_state *velems = state; 1314 1315 if (velems == NULL) { 1316 return; 1317 } 1318 1319 r300->velems = velems; 1320 1321 if (r300->draw) { 1322 draw_flush(r300->draw); 1323 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1324 } 1325 1326 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1327 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1328} 1329 1330static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1331{ 1332 FREE(state); 1333} 1334 1335static void* r300_create_vs_state(struct pipe_context* pipe, 1336 const struct pipe_shader_state* shader) 1337{ 1338 struct r300_context* r300 = r300_context(pipe); 1339 1340 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1341 1342 /* Copy state directly into shader. */ 1343 vs->state = *shader; 1344 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1345 1346 if (r300->screen->caps.has_tcl) { 1347 r300_translate_vertex_shader(r300, vs, vs->state.tokens); 1348 } else { 1349 vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); 1350 } 1351 1352 return vs; 1353} 1354 1355static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1356{ 1357 struct r300_context* r300 = r300_context(pipe); 1358 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1359 1360 if (vs == NULL) { 1361 r300->vs_state.state = NULL; 1362 return; 1363 } 1364 if (vs == r300->vs_state.state) { 1365 return; 1366 } 1367 r300->vs_state.state = vs; 1368 1369 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block 1370 if (r300->fs.state) { 1371 r300_vertex_shader_setup_wpos(r300); 1372 } 1373 memcpy(r300->vap_output_state.state, &vs->vap_out, 1374 sizeof(struct r300_vap_output_state)); 1375 r300->vap_output_state.dirty = TRUE; 1376 1377 /* The majority of the RS block bits is dependent on the vertex shader. */ 1378 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1379 1380 if (r300->screen->caps.has_tcl) { 1381 r300->vs_state.dirty = TRUE; 1382 r300->vs_state.size = 1383 vs->code.length + 9 + 1384 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1385 1386 r300->pvs_flush.dirty = TRUE; 1387 1388 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1389 } else { 1390 draw_flush(r300->draw); 1391 draw_bind_vertex_shader(r300->draw, 1392 (struct draw_vertex_shader*)vs->draw_vs); 1393 } 1394} 1395 1396static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1397{ 1398 struct r300_context* r300 = r300_context(pipe); 1399 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1400 1401 if (r300->screen->caps.has_tcl) { 1402 rc_constants_destroy(&vs->code.constants); 1403 } else { 1404 draw_delete_vertex_shader(r300->draw, 1405 (struct draw_vertex_shader*)vs->draw_vs); 1406 } 1407 1408 FREE((void*)vs->state.tokens); 1409 FREE(shader); 1410} 1411 1412static void r300_set_constant_buffer(struct pipe_context *pipe, 1413 uint shader, uint index, 1414 struct pipe_resource *buf) 1415{ 1416 struct r300_context* r300 = r300_context(pipe); 1417 struct r300_constant_buffer *cbuf; 1418 struct pipe_transfer *tr; 1419 void *mapped; 1420 int max_size = 0; 1421 1422 switch (shader) { 1423 case PIPE_SHADER_VERTEX: 1424 cbuf = &r300->shader_constants[PIPE_SHADER_VERTEX]; 1425 max_size = 256; 1426 break; 1427 case PIPE_SHADER_FRAGMENT: 1428 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1429 if (r300->screen->caps.is_r500) { 1430 max_size = 256; 1431 } else { 1432 max_size = 32; 1433 } 1434 break; 1435 default: 1436 assert(0); 1437 cbuf = NULL; 1438 } 1439 1440 if (buf == NULL || buf->width0 == 0 || 1441 (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL) 1442 { 1443 cbuf->count = 0; 1444 return; 1445 } 1446 1447 assert((buf->width0 % 4 * sizeof(float)) == 0); 1448 1449 /* Check the size of the constant buffer. */ 1450 /* XXX Subtract immediates and RC_STATE_* variables. */ 1451 if (buf->width0 > (sizeof(float) * 4 * max_size)) { 1452 fprintf(stderr, "r300: Max size of the constant buffer is " 1453 "%i*4 floats.\n", max_size); 1454 abort(); 1455 } 1456 1457 memcpy(cbuf->constants, mapped, buf->width0); 1458 cbuf->count = buf->width0 / (4 * sizeof(float)); 1459 pipe_buffer_unmap(pipe, buf, tr); 1460 1461 if (shader == PIPE_SHADER_VERTEX) { 1462 if (r300->screen->caps.has_tcl) { 1463 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1464 r300->pvs_flush.dirty = TRUE; 1465 } else if (r300->draw) { 1466 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1467 0, cbuf->constants, 1468 buf->width0); 1469 } 1470 } else if (shader == PIPE_SHADER_FRAGMENT) { 1471 r300->fs_constants.dirty = TRUE; 1472 } 1473} 1474 1475void r300_init_state_functions(struct r300_context* r300) 1476{ 1477 r300->context.create_blend_state = r300_create_blend_state; 1478 r300->context.bind_blend_state = r300_bind_blend_state; 1479 r300->context.delete_blend_state = r300_delete_blend_state; 1480 1481 r300->context.set_blend_color = r300_set_blend_color; 1482 1483 r300->context.set_clip_state = r300_set_clip_state; 1484 1485 r300->context.set_constant_buffer = r300_set_constant_buffer; 1486 1487 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1488 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1489 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1490 1491 r300->context.set_stencil_ref = r300_set_stencil_ref; 1492 1493 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1494 1495 r300->context.create_fs_state = r300_create_fs_state; 1496 r300->context.bind_fs_state = r300_bind_fs_state; 1497 r300->context.delete_fs_state = r300_delete_fs_state; 1498 1499 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1500 1501 r300->context.create_rasterizer_state = r300_create_rs_state; 1502 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1503 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1504 1505 r300->context.create_sampler_state = r300_create_sampler_state; 1506 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1507 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1508 r300->context.delete_sampler_state = r300_delete_sampler_state; 1509 1510 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1511 r300->context.create_sampler_view = r300_create_sampler_view; 1512 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1513 1514 r300->context.set_scissor_state = r300_set_scissor_state; 1515 1516 r300->context.set_viewport_state = r300_set_viewport_state; 1517 1518 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1519 1520 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1521 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1522 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1523 1524 r300->context.create_vs_state = r300_create_vs_state; 1525 r300->context.bind_vs_state = r300_bind_vs_state; 1526 r300->context.delete_vs_state = r300_delete_vs_state; 1527} 1528