r300_state.c revision f86ac27bf9203fdd9b7110dc843263307f475a99
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.h" 40#include "r300_state_inlines.h" 41#include "r300_fs.h" 42#include "r300_texture.h" 43#include "r300_vs.h" 44#include "r300_winsys.h" 45 46/* r300_state: Functions used to intialize state context by translating 47 * Gallium state objects into semi-native r300 state objects. */ 48 49#define UPDATE_STATE(cso, atom) \ 50 if (cso != atom.state) { \ 51 atom.state = cso; \ 52 atom.dirty = TRUE; \ 53 } 54 55static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 56 unsigned dstRGB, unsigned dstA) 57{ 58 /* If the blend equation is ADD or REVERSE_SUBTRACT, 59 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 60 * will not be changed. 61 * Notice that the dst factors are the src factors inverted. */ 62 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 63 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 64 srcRGB == PIPE_BLENDFACTOR_ZERO) && 65 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 66 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 67 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 68 srcA == PIPE_BLENDFACTOR_ZERO) && 69 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 70 dstRGB == PIPE_BLENDFACTOR_ONE) && 71 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 72 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 73 dstA == PIPE_BLENDFACTOR_ONE); 74} 75 76static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 77 unsigned dstRGB, unsigned dstA) 78{ 79 /* If the blend equation is ADD or REVERSE_SUBTRACT, 80 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 81 * will not be changed. 82 * Notice that the dst factors are the src factors inverted. */ 83 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 84 srcRGB == PIPE_BLENDFACTOR_ZERO) && 85 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 86 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 87 srcA == PIPE_BLENDFACTOR_ZERO) && 88 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 89 dstRGB == PIPE_BLENDFACTOR_ONE) && 90 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 91 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 92 dstA == PIPE_BLENDFACTOR_ONE); 93} 94 95static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 96 unsigned dstRGB, unsigned dstA) 97{ 98 /* If the blend equation is ADD or REVERSE_SUBTRACT, 99 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 100 * will not be changed. 101 * Notice that the dst factors are the src factors inverted. */ 102 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 103 srcRGB == PIPE_BLENDFACTOR_ZERO) && 104 (srcA == PIPE_BLENDFACTOR_ZERO) && 105 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 106 dstRGB == PIPE_BLENDFACTOR_ONE) && 107 (dstA == PIPE_BLENDFACTOR_ONE); 108} 109 110static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 111 unsigned dstRGB, unsigned dstA) 112{ 113 /* If the blend equation is ADD or REVERSE_SUBTRACT, 114 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 115 * will not be changed. 116 * Notice that the dst factors are the src factors inverted. */ 117 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 118 srcRGB == PIPE_BLENDFACTOR_ZERO) && 119 (srcA == PIPE_BLENDFACTOR_ZERO) && 120 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 121 dstRGB == PIPE_BLENDFACTOR_ONE) && 122 (dstA == PIPE_BLENDFACTOR_ONE); 123} 124 125static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 126 unsigned dstRGB, unsigned dstA) 127{ 128 /* If the blend equation is ADD or REVERSE_SUBTRACT, 129 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 130 * the colorbuffer will not be changed. 131 * Notice that the dst factors are the src factors inverted. */ 132 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 133 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 134 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 135 srcRGB == PIPE_BLENDFACTOR_ZERO) && 136 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 137 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 138 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 139 srcA == PIPE_BLENDFACTOR_ZERO) && 140 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 141 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 142 dstRGB == PIPE_BLENDFACTOR_ONE) && 143 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 144 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 145 dstA == PIPE_BLENDFACTOR_ONE); 146} 147 148static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 149 unsigned dstRGB, unsigned dstA) 150{ 151 /* If the blend equation is ADD or REVERSE_SUBTRACT, 152 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 153 * the colorbuffer will not be changed. 154 * Notice that the dst factors are the src factors inverted. */ 155 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 156 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 157 srcRGB == PIPE_BLENDFACTOR_ZERO) && 158 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 159 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 160 srcA == PIPE_BLENDFACTOR_ZERO) && 161 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 162 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 163 dstRGB == PIPE_BLENDFACTOR_ONE) && 164 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 165 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 166 dstA == PIPE_BLENDFACTOR_ONE); 167} 168 169static unsigned bgra_cmask(unsigned mask) 170{ 171 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 172 173 return ((mask & PIPE_MASK_R) << 2) | 174 ((mask & PIPE_MASK_B) >> 2) | 175 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 176} 177 178/* Create a new blend state based on the CSO blend state. 179 * 180 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 181static void* r300_create_blend_state(struct pipe_context* pipe, 182 const struct pipe_blend_state* state) 183{ 184 struct r300_screen* r300screen = r300_screen(pipe->screen); 185 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 186 187 if (state->rt[0].blend_enable) 188 { 189 unsigned eqRGB = state->rt[0].rgb_func; 190 unsigned srcRGB = state->rt[0].rgb_src_factor; 191 unsigned dstRGB = state->rt[0].rgb_dst_factor; 192 193 unsigned eqA = state->rt[0].alpha_func; 194 unsigned srcA = state->rt[0].alpha_src_factor; 195 unsigned dstA = state->rt[0].alpha_dst_factor; 196 197 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 198 * this is just the crappy D3D naming */ 199 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 200 r300_translate_blend_function(eqRGB) | 201 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 202 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 203 204 /* Optimization: some operations do not require the destination color. 205 * 206 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 207 * otherwise blending gives incorrect results. It seems to be 208 * a hardware bug. */ 209 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 210 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 211 dstRGB != PIPE_BLENDFACTOR_ZERO || 212 dstA != PIPE_BLENDFACTOR_ZERO || 213 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 214 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 215 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 216 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 217 srcA == PIPE_BLENDFACTOR_DST_COLOR || 218 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 219 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 220 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 221 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 222 /* Enable reading from the colorbuffer. */ 223 blend->blend_control |= R300_READ_ENABLE; 224 225 if (r300screen->caps.is_r500) { 226 /* Optimization: Depending on incoming pixels, we can 227 * conditionally disable the reading in hardware... */ 228 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 229 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 230 /* Disable reading if SRC_ALPHA == 0. */ 231 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 232 dstRGB == PIPE_BLENDFACTOR_ZERO) && 233 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 234 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 235 dstA == PIPE_BLENDFACTOR_ZERO)) { 236 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 237 } 238 239 /* Disable reading if SRC_ALPHA == 1. */ 240 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 241 dstRGB == PIPE_BLENDFACTOR_ZERO) && 242 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 243 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 244 dstA == PIPE_BLENDFACTOR_ZERO)) { 245 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 246 } 247 } 248 } 249 } 250 251 /* Optimization: discard pixels which don't change the colorbuffer. 252 * 253 * The code below is non-trivial and some math is involved. 254 * 255 * Discarding pixels must be disabled when FP16 AA is enabled. 256 * This is a hardware bug. Also, this implementation wouldn't work 257 * with FP blending enabled and equation clamping disabled. 258 * 259 * Equations other than ADD are rarely used and therefore won't be 260 * optimized. */ 261 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 262 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 263 /* ADD: X+Y 264 * REVERSE_SUBTRACT: Y-X 265 * 266 * The idea is: 267 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 268 * then CB will not be changed. 269 * 270 * Given the srcFactor and dstFactor variables, we can derive 271 * what src and dst should be equal to and discard appropriate 272 * pixels. 273 */ 274 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 275 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 276 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 277 dstRGB, dstA)) { 278 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 279 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 280 dstRGB, dstA)) { 281 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 282 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 283 dstRGB, dstA)) { 284 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 285 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 286 dstRGB, dstA)) { 287 blend->blend_control |= 288 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 289 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 290 dstRGB, dstA)) { 291 blend->blend_control |= 292 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 293 } 294 } 295 296 /* separate alpha */ 297 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 298 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 299 blend->alpha_blend_control = 300 r300_translate_blend_function(eqA) | 301 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 302 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 303 } 304 } 305 306 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 307 if (state->logicop_enable) { 308 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 309 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 310 } 311 312 /* Color channel masks for all MRTs. */ 313 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); 314 if (r300screen->caps.is_r500 && state->independent_blend_enable) { 315 if (state->rt[1].blend_enable) { 316 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 317 } 318 if (state->rt[2].blend_enable) { 319 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 320 } 321 if (state->rt[3].blend_enable) { 322 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 323 } 324 } 325 326 /* Neither fglrx nor classic r300 ever set this, regardless of dithering 327 * state. Since it's an optional implementation detail, we can leave it 328 * out and never dither. 329 * 330 * This could be revisited if we ever get quality or conformance hints. 331 * 332 if (state->dither) { 333 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 334 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 335 } 336 */ 337 338 return (void*)blend; 339} 340 341/* Bind blend state. */ 342static void r300_bind_blend_state(struct pipe_context* pipe, 343 void* state) 344{ 345 struct r300_context* r300 = r300_context(pipe); 346 347 UPDATE_STATE(state, r300->blend_state); 348} 349 350/* Free blend state. */ 351static void r300_delete_blend_state(struct pipe_context* pipe, 352 void* state) 353{ 354 FREE(state); 355} 356 357/* Convert float to 10bit integer */ 358static unsigned float_to_fixed10(float f) 359{ 360 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 361} 362 363/* Set blend color. 364 * Setup both R300 and R500 registers, figure out later which one to write. */ 365static void r300_set_blend_color(struct pipe_context* pipe, 366 const struct pipe_blend_color* color) 367{ 368 struct r300_context* r300 = r300_context(pipe); 369 struct r300_blend_color_state* state = 370 (struct r300_blend_color_state*)r300->blend_color_state.state; 371 union util_color uc; 372 373 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 374 state->blend_color = uc.ui; 375 376 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 377 state->blend_color_red_alpha = 378 float_to_fixed10(color->color[0]) | 379 (float_to_fixed10(color->color[3]) << 16); 380 state->blend_color_green_blue = 381 float_to_fixed10(color->color[2]) | 382 (float_to_fixed10(color->color[1]) << 16); 383 384 r300->blend_color_state.size = r300->screen->caps.is_r500 ? 3 : 2; 385 r300->blend_color_state.dirty = TRUE; 386} 387 388static void r300_set_clip_state(struct pipe_context* pipe, 389 const struct pipe_clip_state* state) 390{ 391 struct r300_context* r300 = r300_context(pipe); 392 393 r300->clip = *state; 394 395 if (r300->screen->caps.has_tcl) { 396 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 397 r300->clip_state.size = 29; 398 399 r300->clip_state.dirty = TRUE; 400 } else { 401 draw_flush(r300->draw); 402 draw_set_clip_state(r300->draw, state); 403 r300->clip_state.size = 2; 404 } 405} 406 407/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 408 * 409 * This contains the depth buffer, stencil buffer, alpha test, and such. 410 * On the Radeon, depth and stencil buffer setup are intertwined, which is 411 * the reason for some of the strange-looking assignments across registers. */ 412static void* 413 r300_create_dsa_state(struct pipe_context* pipe, 414 const struct pipe_depth_stencil_alpha_state* state) 415{ 416 struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; 417 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 418 419 /* Depth test setup. */ 420 if (state->depth.enabled) { 421 dsa->z_buffer_control |= R300_Z_ENABLE; 422 423 if (state->depth.writemask) { 424 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 425 } 426 427 dsa->z_stencil_control |= 428 (r300_translate_depth_stencil_function(state->depth.func) << 429 R300_Z_FUNC_SHIFT); 430 } 431 432 /* Stencil buffer setup. */ 433 if (state->stencil[0].enabled) { 434 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 435 dsa->z_stencil_control |= 436 (r300_translate_depth_stencil_function(state->stencil[0].func) << 437 R300_S_FRONT_FUNC_SHIFT) | 438 (r300_translate_stencil_op(state->stencil[0].fail_op) << 439 R300_S_FRONT_SFAIL_OP_SHIFT) | 440 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 441 R300_S_FRONT_ZPASS_OP_SHIFT) | 442 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 443 R300_S_FRONT_ZFAIL_OP_SHIFT); 444 445 dsa->stencil_ref_mask = 446 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 447 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 448 449 if (state->stencil[1].enabled) { 450 dsa->two_sided = TRUE; 451 452 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 453 dsa->z_stencil_control |= 454 (r300_translate_depth_stencil_function(state->stencil[1].func) << 455 R300_S_BACK_FUNC_SHIFT) | 456 (r300_translate_stencil_op(state->stencil[1].fail_op) << 457 R300_S_BACK_SFAIL_OP_SHIFT) | 458 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 459 R300_S_BACK_ZPASS_OP_SHIFT) | 460 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 461 R300_S_BACK_ZFAIL_OP_SHIFT); 462 463 dsa->stencil_ref_bf = 464 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | 465 (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); 466 467 if (caps->is_r500) { 468 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 469 } else { 470 dsa->stencil_ref_bf_fallback = 471 (state->stencil[0].valuemask != state->stencil[1].valuemask || 472 state->stencil[0].writemask != state->stencil[1].writemask); 473 } 474 } 475 } 476 477 /* Alpha test setup. */ 478 if (state->alpha.enabled) { 479 dsa->alpha_function = 480 r300_translate_alpha_function(state->alpha.func) | 481 R300_FG_ALPHA_FUNC_ENABLE; 482 483 /* We could use 10bit alpha ref but who needs that? */ 484 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 485 486 if (caps->is_r500) 487 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 488 } 489 490 return (void*)dsa; 491} 492 493static void r300_update_stencil_ref_fallback_status(struct r300_context *r300) 494{ 495 struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; 496 497 if (r300->screen->caps.is_r500) { 498 return; 499 } 500 501 r300->stencil_ref_bf_fallback = 502 dsa->stencil_ref_bf_fallback || 503 (dsa->two_sided && 504 r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]); 505} 506 507/* Bind DSA state. */ 508static void r300_bind_dsa_state(struct pipe_context* pipe, 509 void* state) 510{ 511 struct r300_context* r300 = r300_context(pipe); 512 513 if (!state) { 514 return; 515 } 516 517 UPDATE_STATE(state, r300->dsa_state); 518 519 r300_update_stencil_ref_fallback_status(r300); 520} 521 522/* Free DSA state. */ 523static void r300_delete_dsa_state(struct pipe_context* pipe, 524 void* state) 525{ 526 FREE(state); 527} 528 529static void r300_set_stencil_ref(struct pipe_context* pipe, 530 const struct pipe_stencil_ref* sr) 531{ 532 struct r300_context* r300 = r300_context(pipe); 533 534 r300->stencil_ref = *sr; 535 r300->dsa_state.dirty = TRUE; 536 537 r300_update_stencil_ref_fallback_status(r300); 538} 539 540/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 541static void r300_fb_set_tiling_flags(struct r300_context *r300, 542 const struct pipe_framebuffer_state *old_state, 543 const struct pipe_framebuffer_state *new_state) 544{ 545 struct r300_texture *tex; 546 unsigned i, level; 547 548 /* Set tiling flags for new surfaces. */ 549 for (i = 0; i < new_state->nr_cbufs; i++) { 550 tex = r300_texture(new_state->cbufs[i]->texture); 551 level = new_state->cbufs[i]->level; 552 553 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 554 tex->pitch[0], 555 tex->microtile, 556 tex->mip_macrotile[level]); 557 } 558 if (new_state->zsbuf) { 559 tex = r300_texture(new_state->zsbuf->texture); 560 level = new_state->zsbuf->level; 561 562 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 563 tex->pitch[0], 564 tex->microtile, 565 tex->mip_macrotile[level]); 566 } 567} 568 569static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, 570 const char *binding) 571{ 572 struct pipe_resource *tex = surf->texture; 573 struct r300_texture *rtex = r300_texture(tex); 574 575 fprintf(stderr, 576 "r300: %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, " 577 "Face: %i, Level: %i, Format: %s\n" 578 579 "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, " 580 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", 581 582 binding, index, surf->width, surf->height, surf->offset, 583 surf->zslice, surf->face, surf->level, 584 util_format_short_name(surf->format), 585 586 rtex->macrotile ? "YES" : " NO", rtex->microtile ? "YES" : " NO", 587 rtex->hwpitch[0], tex->width0, tex->height0, tex->depth0, 588 tex->last_level, util_format_short_name(tex->format)); 589} 590 591static void 592 r300_set_framebuffer_state(struct pipe_context* pipe, 593 const struct pipe_framebuffer_state* state) 594{ 595 struct r300_context* r300 = r300_context(pipe); 596 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 597 unsigned max_width, max_height, i; 598 uint32_t zbuffer_bpp = 0; 599 600 if (state->nr_cbufs > 4) { 601 fprintf(stderr, "r300: Implementation error: Too many MRTs in %s, " 602 "refusing to bind framebuffer state!\n", __FUNCTION__); 603 return; 604 } 605 606 if (r300->screen->caps.is_r500) { 607 max_width = max_height = 4096; 608 } else if (r300->screen->caps.is_r400) { 609 max_width = max_height = 4021; 610 } else { 611 max_width = max_height = 2560; 612 } 613 614 if (state->width > max_width || state->height > max_height) { 615 fprintf(stderr, "r300: Implementation error: Render targets are too " 616 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 617 return; 618 } 619 620 if (r300->draw) { 621 draw_flush(r300->draw); 622 } 623 624 r300->fb_state.dirty = TRUE; 625 626 /* If nr_cbufs is changed from zero to non-zero or vice versa... */ 627 if (!!old_state->nr_cbufs != !!state->nr_cbufs) { 628 r300->blend_state.dirty = TRUE; 629 } 630 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 631 if (!!old_state->zsbuf != !!state->zsbuf) { 632 r300->dsa_state.dirty = TRUE; 633 } 634 635 /* The tiling flags are dependent on the surface miplevel, unfortunately. */ 636 r300_fb_set_tiling_flags(r300, r300->fb_state.state, state); 637 638 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 639 640 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + 641 (state->zsbuf ? 10 : 0) + 9; 642 643 /* Polygon offset depends on the zbuffer bit depth. */ 644 if (state->zsbuf && r300->polygon_offset_enabled) { 645 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 646 case 2: 647 zbuffer_bpp = 16; 648 break; 649 case 4: 650 zbuffer_bpp = 24; 651 break; 652 } 653 654 if (r300->zbuffer_bpp != zbuffer_bpp) { 655 r300->zbuffer_bpp = zbuffer_bpp; 656 r300->rs_state.dirty = TRUE; 657 } 658 } 659 660 if (DBG_ON(r300, DBG_FB)) { 661 fprintf(stderr, "r300: set_framebuffer_state:\n"); 662 for (i = 0; i < state->nr_cbufs; i++) { 663 r300_print_fb_surf_info(state->cbufs[i], i, "CB"); 664 } 665 if (state->zsbuf) { 666 r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); 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 686void 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} 722 723/* Delete fragment shader state. */ 724static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 725{ 726 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 727 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 728 729 while (ptr) { 730 tmp = ptr; 731 ptr = ptr->next; 732 rc_constants_destroy(&tmp->code.constants); 733 FREE(tmp); 734 } 735 FREE((void*)fs->state.tokens); 736 FREE(shader); 737} 738 739static void r300_set_polygon_stipple(struct pipe_context* pipe, 740 const struct pipe_poly_stipple* state) 741{ 742 /* XXX no idea how to set this up, but not terribly important */ 743} 744 745/* Create a new rasterizer state based on the CSO rasterizer state. 746 * 747 * This is a very large chunk of state, and covers most of the graphics 748 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 749 * 750 * In a not entirely unironic sidenote, this state has nearly nothing to do 751 * with the actual block on the Radeon called the rasterizer (RS). */ 752static void* r300_create_rs_state(struct pipe_context* pipe, 753 const struct pipe_rasterizer_state* state) 754{ 755 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 756 int i; 757 float psiz; 758 759 /* Copy rasterizer state for Draw. */ 760 rs->rs = *state; 761 762#ifdef PIPE_ARCH_LITTLE_ENDIAN 763 rs->vap_control_status = R300_VC_NO_SWAP; 764#else 765 rs->vap_control_status = R300_VC_32BIT_SWAP; 766#endif 767 768 /* If no TCL engine is present, turn off the HW TCL. */ 769 if (!r300_screen(pipe->screen)->caps.has_tcl) { 770 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 771 } 772 773 /* Point size width and height. */ 774 rs->point_size = 775 pack_float_16_6x(state->point_size) | 776 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 777 778 /* Point size clamping. */ 779 if (state->point_size_per_vertex) { 780 /* Per-vertex point size. 781 * Clamp to [0, max FB size] */ 782 psiz = pipe->screen->get_paramf(pipe->screen, 783 PIPE_CAP_MAX_POINT_WIDTH); 784 rs->point_minmax = 785 pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; 786 } else { 787 /* We cannot disable the point-size vertex output, 788 * so clamp it. */ 789 psiz = state->point_size; 790 rs->point_minmax = 791 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 792 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 793 } 794 795 /* Line control. */ 796 rs->line_control = pack_float_16_6x(state->line_width) | 797 R300_GA_LINE_CNTL_END_TYPE_COMP; 798 799 /* Enable polygon mode */ 800 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 801 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 802 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 803 } 804 805 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 806 if (state->front_winding == PIPE_WINDING_CW) { 807 rs->cull_mode = R300_FRONT_FACE_CW; 808 809 /* Polygon offset */ 810 if (state->offset_cw) { 811 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 812 } 813 if (state->offset_ccw) { 814 rs->polygon_offset_enable |= R300_BACK_ENABLE; 815 } 816 817 /* Polygon mode */ 818 if (rs->polygon_mode) { 819 rs->polygon_mode |= 820 r300_translate_polygon_mode_front(state->fill_cw); 821 rs->polygon_mode |= 822 r300_translate_polygon_mode_back(state->fill_ccw); 823 } 824 } else { 825 rs->cull_mode = R300_FRONT_FACE_CCW; 826 827 /* Polygon offset */ 828 if (state->offset_ccw) { 829 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 830 } 831 if (state->offset_cw) { 832 rs->polygon_offset_enable |= R300_BACK_ENABLE; 833 } 834 835 /* Polygon mode */ 836 if (rs->polygon_mode) { 837 rs->polygon_mode |= 838 r300_translate_polygon_mode_front(state->fill_ccw); 839 rs->polygon_mode |= 840 r300_translate_polygon_mode_back(state->fill_cw); 841 } 842 } 843 if (state->front_winding & state->cull_mode) { 844 rs->cull_mode |= R300_CULL_FRONT; 845 } 846 if (~(state->front_winding) & state->cull_mode) { 847 rs->cull_mode |= R300_CULL_BACK; 848 } 849 850 if (rs->polygon_offset_enable) { 851 rs->depth_offset = state->offset_units; 852 rs->depth_scale = state->offset_scale; 853 } 854 855 if (state->line_stipple_enable) { 856 rs->line_stipple_config = 857 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 858 (fui((float)state->line_stipple_factor) & 859 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 860 /* XXX this might need to be scaled up */ 861 rs->line_stipple_value = state->line_stipple_pattern; 862 } 863 864 if (state->flatshade) { 865 rs->color_control = R300_SHADE_MODEL_FLAT; 866 } else { 867 rs->color_control = R300_SHADE_MODEL_SMOOTH; 868 } 869 870 rs->clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 871 872 /* Point sprites */ 873 if (state->sprite_coord_enable) { 874 rs->stuffing_enable = R300_GB_POINT_STUFF_ENABLE; 875 for (i = 0; i < 8; i++) { 876 if (state->sprite_coord_enable & (1 << i)) 877 rs->stuffing_enable |= 878 R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); 879 } 880 881 rs->point_texcoord_left = 0.0f; 882 rs->point_texcoord_right = 1.0f; 883 884 switch (state->sprite_coord_mode) { 885 case PIPE_SPRITE_COORD_UPPER_LEFT: 886 rs->point_texcoord_top = 0.0f; 887 rs->point_texcoord_bottom = 1.0f; 888 break; 889 case PIPE_SPRITE_COORD_LOWER_LEFT: 890 rs->point_texcoord_top = 1.0f; 891 rs->point_texcoord_bottom = 0.0f; 892 break; 893 } 894 } 895 896 return (void*)rs; 897} 898 899/* Bind rasterizer state. */ 900static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 901{ 902 struct r300_context* r300 = r300_context(pipe); 903 struct r300_rs_state* rs = (struct r300_rs_state*)state; 904 int last_sprite_coord_enable = r300->sprite_coord_enable; 905 boolean last_two_sided_color = r300->two_sided_color; 906 907 if (r300->draw) { 908 draw_flush(r300->draw); 909 draw_set_rasterizer_state(r300->draw, &rs->rs, state); 910 } 911 912 if (rs) { 913 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 914 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 915 r300->two_sided_color = rs->rs.light_twoside; 916 } else { 917 r300->polygon_offset_enabled = FALSE; 918 r300->sprite_coord_enable = 0; 919 r300->two_sided_color = FALSE; 920 } 921 922 UPDATE_STATE(state, r300->rs_state); 923 r300->rs_state.size = 27 + (r300->polygon_offset_enabled ? 5 : 0); 924 925 if (last_sprite_coord_enable != r300->sprite_coord_enable || 926 last_two_sided_color != r300->two_sided_color) { 927 r300->rs_block_state.dirty = TRUE; 928 } 929} 930 931/* Free rasterizer state. */ 932static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 933{ 934 FREE(state); 935} 936 937static void* 938 r300_create_sampler_state(struct pipe_context* pipe, 939 const struct pipe_sampler_state* state) 940{ 941 struct r300_context* r300 = r300_context(pipe); 942 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 943 boolean is_r500 = r300->screen->caps.is_r500; 944 int lod_bias; 945 union util_color uc; 946 947 sampler->state = *state; 948 949 sampler->filter0 |= 950 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 951 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 952 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 953 954 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 955 state->mag_img_filter, 956 state->min_mip_filter, 957 state->max_anisotropy > 0); 958 959 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 960 961 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 962 /* We must pass these to the merge function to clamp them properly. */ 963 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 964 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 965 966 lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); 967 968 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 969 970 /* This is very high quality anisotropic filtering for R5xx. 971 * It's good for benchmarking the performance of texturing but 972 * in practice we don't want to slow down the driver because it's 973 * a pretty good performance killer. Feel free to play with it. */ 974 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 975 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 976 } 977 978 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 979 sampler->border_color = uc.ui; 980 981 /* R500-specific fixups and optimizations */ 982 if (r300->screen->caps.is_r500) { 983 sampler->filter1 |= R500_BORDER_FIX; 984 } 985 986 return (void*)sampler; 987} 988 989static void r300_bind_sampler_states(struct pipe_context* pipe, 990 unsigned count, 991 void** states) 992{ 993 struct r300_context* r300 = r300_context(pipe); 994 struct r300_textures_state* state = 995 (struct r300_textures_state*)r300->textures_state.state; 996 unsigned tex_units = r300->screen->caps.num_tex_units; 997 998 if (count > tex_units) { 999 return; 1000 } 1001 1002 memcpy(state->sampler_states, states, sizeof(void*) * count); 1003 state->sampler_state_count = count; 1004 1005 r300->textures_state.dirty = TRUE; 1006} 1007 1008static void r300_lacks_vertex_textures(struct pipe_context* pipe, 1009 unsigned count, 1010 void** states) 1011{ 1012} 1013 1014static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1015{ 1016 FREE(state); 1017} 1018 1019static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 1020 unsigned count, 1021 struct pipe_sampler_view** views) 1022{ 1023 struct r300_context* r300 = r300_context(pipe); 1024 struct r300_textures_state* state = 1025 (struct r300_textures_state*)r300->textures_state.state; 1026 struct r300_texture *texture; 1027 unsigned i; 1028 unsigned tex_units = r300->screen->caps.num_tex_units; 1029 boolean dirty_tex = FALSE; 1030 1031 if (count > tex_units) { 1032 return; 1033 } 1034 1035 for (i = 0; i < count; i++) { 1036 if (&state->sampler_views[i]->base != views[i]) { 1037 pipe_sampler_view_reference( 1038 (struct pipe_sampler_view**)&state->sampler_views[i], 1039 views[i]); 1040 1041 if (!views[i]) { 1042 continue; 1043 } 1044 1045 /* A new sampler view (= texture)... */ 1046 dirty_tex = TRUE; 1047 1048 /* Set the texrect factor in the fragment shader. 1049 * Needed for RECT and NPOT fallback. */ 1050 texture = r300_texture(views[i]->texture); 1051 if (texture->uses_pitch) { 1052 r300->fs_rc_constant_state.dirty = TRUE; 1053 } 1054 } 1055 } 1056 1057 for (i = count; i < tex_units; i++) { 1058 if (state->sampler_views[i]) { 1059 pipe_sampler_view_reference( 1060 (struct pipe_sampler_view**)&state->sampler_views[i], 1061 NULL); 1062 } 1063 } 1064 1065 state->sampler_view_count = count; 1066 1067 r300->textures_state.dirty = TRUE; 1068 1069 if (dirty_tex) { 1070 r300->texture_cache_inval.dirty = TRUE; 1071 } 1072} 1073 1074static struct pipe_sampler_view * 1075r300_create_sampler_view(struct pipe_context *pipe, 1076 struct pipe_resource *texture, 1077 const struct pipe_sampler_view *templ) 1078{ 1079 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1080 struct r300_texture *tex = r300_texture(texture); 1081 1082 if (view) { 1083 view->base = *templ; 1084 view->base.reference.count = 1; 1085 view->base.context = pipe; 1086 view->base.texture = NULL; 1087 pipe_resource_reference(&view->base.texture, texture); 1088 1089 view->swizzle[0] = templ->swizzle_r; 1090 view->swizzle[1] = templ->swizzle_g; 1091 view->swizzle[2] = templ->swizzle_b; 1092 view->swizzle[3] = templ->swizzle_a; 1093 1094 view->format = tex->tx_format; 1095 view->format.format1 |= r300_translate_texformat(templ->format, 1096 view->swizzle); 1097 if (r300_screen(pipe->screen)->caps.is_r500) { 1098 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1099 } 1100 } 1101 1102 return (struct pipe_sampler_view*)view; 1103} 1104 1105static void 1106r300_sampler_view_destroy(struct pipe_context *pipe, 1107 struct pipe_sampler_view *view) 1108{ 1109 pipe_resource_reference(&view->texture, NULL); 1110 FREE(view); 1111} 1112 1113static void r300_set_scissor_state(struct pipe_context* pipe, 1114 const struct pipe_scissor_state* state) 1115{ 1116 struct r300_context* r300 = r300_context(pipe); 1117 1118 memcpy(r300->scissor_state.state, state, 1119 sizeof(struct pipe_scissor_state)); 1120 1121 r300->scissor_state.dirty = TRUE; 1122} 1123 1124static void r300_set_viewport_state(struct pipe_context* pipe, 1125 const struct pipe_viewport_state* state) 1126{ 1127 struct r300_context* r300 = r300_context(pipe); 1128 struct r300_viewport_state* viewport = 1129 (struct r300_viewport_state*)r300->viewport_state.state; 1130 1131 r300->viewport = *state; 1132 1133 if (r300->draw) { 1134 draw_flush(r300->draw); 1135 draw_set_viewport_state(r300->draw, state); 1136 viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; 1137 return; 1138 } 1139 1140 /* Do the transform in HW. */ 1141 viewport->vte_control = R300_VTX_W0_FMT; 1142 1143 if (state->scale[0] != 1.0f) { 1144 viewport->xscale = state->scale[0]; 1145 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1146 } 1147 if (state->scale[1] != 1.0f) { 1148 viewport->yscale = state->scale[1]; 1149 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1150 } 1151 if (state->scale[2] != 1.0f) { 1152 viewport->zscale = state->scale[2]; 1153 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1154 } 1155 if (state->translate[0] != 0.0f) { 1156 viewport->xoffset = state->translate[0]; 1157 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1158 } 1159 if (state->translate[1] != 0.0f) { 1160 viewport->yoffset = state->translate[1]; 1161 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1162 } 1163 if (state->translate[2] != 0.0f) { 1164 viewport->zoffset = state->translate[2]; 1165 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1166 } 1167 1168 r300->viewport_state.dirty = TRUE; 1169 if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1170 r300->fs_rc_constant_state.dirty = TRUE; 1171 } 1172} 1173 1174static void r300_set_vertex_buffers(struct pipe_context* pipe, 1175 unsigned count, 1176 const struct pipe_vertex_buffer* buffers) 1177{ 1178 struct r300_context* r300 = r300_context(pipe); 1179 struct pipe_vertex_buffer *vbo; 1180 unsigned i, max_index = (1 << 24) - 1; 1181 boolean any_user_buffer = FALSE; 1182 1183 if (count == r300->vertex_buffer_count && 1184 memcmp(r300->vertex_buffer, buffers, 1185 sizeof(struct pipe_vertex_buffer) * count) == 0) { 1186 return; 1187 } 1188 1189 /* Check if the stride is aligned to the size of DWORD. */ 1190 for (i = 0; i < count; i++) { 1191 if (buffers[i].buffer) { 1192 if (buffers[i].stride % 4 != 0) { 1193 // XXX Shouldn't we align the buffer? 1194 fprintf(stderr, "r300: set_vertex_buffers: " 1195 "Unaligned buffer stride %i isn't supported.\n", 1196 buffers[i].stride); 1197 abort(); 1198 } 1199 } 1200 } 1201 1202 for (i = 0; i < count; i++) { 1203 /* Why, yes, I AM casting away constness. How did you know? */ 1204 vbo = (struct pipe_vertex_buffer*)&buffers[i]; 1205 1206 /* Reference our buffer. */ 1207 pipe_resource_reference(&r300->vertex_buffer[i].buffer, vbo->buffer); 1208 1209 /* Skip NULL buffers */ 1210 if (!buffers[i].buffer) { 1211 continue; 1212 } 1213 1214 if (r300_buffer_is_user_buffer(vbo->buffer)) { 1215 any_user_buffer = TRUE; 1216 } 1217 1218 if (vbo->max_index == ~0) { 1219 /* if no VBO stride then only one vertex value so max index is 1 */ 1220 /* should think about converting to VS constants like svga does */ 1221 if (!vbo->stride) 1222 vbo->max_index = 1; 1223 else 1224 vbo->max_index = 1225 (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; 1226 } 1227 1228 max_index = MIN2(vbo->max_index, max_index); 1229 } 1230 1231 for (; i < r300->vertex_buffer_count; i++) { 1232 /* Dereference any old buffers. */ 1233 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); 1234 } 1235 1236 memcpy(r300->vertex_buffer, buffers, 1237 sizeof(struct pipe_vertex_buffer) * count); 1238 1239 r300->vertex_buffer_count = count; 1240 r300->vertex_buffer_max_index = max_index; 1241 r300->any_user_vbs = any_user_buffer; 1242 1243 if (r300->draw) { 1244 draw_flush(r300->draw); 1245 draw_set_vertex_buffers(r300->draw, count, buffers); 1246 } 1247} 1248 1249/* Initialize the PSC tables. */ 1250static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1251{ 1252 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1253 uint16_t type, swizzle; 1254 enum pipe_format format; 1255 unsigned i; 1256 1257 if (velems->count > 16) { 1258 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1259 " requested %i, using 16.\n", velems->count); 1260 velems->count = 16; 1261 } 1262 1263 /* Vertex shaders have no semantics on their inputs, 1264 * so PSC should just route stuff based on the vertex elements, 1265 * and not on attrib information. */ 1266 for (i = 0; i < velems->count; i++) { 1267 format = velems->velem[i].src_format; 1268 1269 type = r300_translate_vertex_data_type(format) | 1270 (i << R300_DST_VEC_LOC_SHIFT); 1271 swizzle = r300_translate_vertex_data_swizzle(format); 1272 1273 if (i & 1) { 1274 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1275 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1276 } else { 1277 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1278 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1279 } 1280 } 1281 1282 /* Set the last vector in the PSC. */ 1283 if (i) { 1284 i -= 1; 1285 } 1286 vstream->vap_prog_stream_cntl[i >> 1] |= 1287 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1288 1289 vstream->count = (i >> 1) + 1; 1290} 1291 1292static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1293 unsigned count, 1294 const struct pipe_vertex_element* attribs) 1295{ 1296 struct r300_vertex_element_state *velems; 1297 unsigned i, size; 1298 enum pipe_format *format; 1299 1300 assert(count <= PIPE_MAX_ATTRIBS); 1301 velems = CALLOC_STRUCT(r300_vertex_element_state); 1302 if (velems != NULL) { 1303 velems->count = count; 1304 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1305 1306 if (r300_screen(pipe->screen)->caps.has_tcl) { 1307 r300_vertex_psc(velems); 1308 1309 /* Check if the format is aligned to the size of DWORD. 1310 * We only care about the blocksizes of the formats since 1311 * swizzles are already set up. */ 1312 for (i = 0; i < count; i++) { 1313 format = &velems->velem[i].src_format; 1314 1315 /* Replace some formats with their aligned counterparts, 1316 * this is OK because we check for aligned strides too. */ 1317 switch (*format) { 1318 /* Align to RGBA8. */ 1319 case PIPE_FORMAT_R8_UNORM: 1320 case PIPE_FORMAT_R8G8_UNORM: 1321 case PIPE_FORMAT_R8G8B8_UNORM: 1322 *format = PIPE_FORMAT_R8G8B8A8_UNORM; 1323 continue; 1324 case PIPE_FORMAT_R8_SNORM: 1325 case PIPE_FORMAT_R8G8_SNORM: 1326 case PIPE_FORMAT_R8G8B8_SNORM: 1327 *format = PIPE_FORMAT_R8G8B8A8_SNORM; 1328 continue; 1329 case PIPE_FORMAT_R8_USCALED: 1330 case PIPE_FORMAT_R8G8_USCALED: 1331 case PIPE_FORMAT_R8G8B8_USCALED: 1332 *format = PIPE_FORMAT_R8G8B8A8_USCALED; 1333 continue; 1334 case PIPE_FORMAT_R8_SSCALED: 1335 case PIPE_FORMAT_R8G8_SSCALED: 1336 case PIPE_FORMAT_R8G8B8_SSCALED: 1337 *format = PIPE_FORMAT_R8G8B8A8_SSCALED; 1338 continue; 1339 1340 /* Align to RG16. */ 1341 case PIPE_FORMAT_R16_UNORM: 1342 *format = PIPE_FORMAT_R16G16_UNORM; 1343 continue; 1344 case PIPE_FORMAT_R16_SNORM: 1345 *format = PIPE_FORMAT_R16G16_SNORM; 1346 continue; 1347 case PIPE_FORMAT_R16_USCALED: 1348 *format = PIPE_FORMAT_R16G16_USCALED; 1349 continue; 1350 case PIPE_FORMAT_R16_SSCALED: 1351 *format = PIPE_FORMAT_R16G16_SSCALED; 1352 continue; 1353 case PIPE_FORMAT_R16_FLOAT: 1354 *format = PIPE_FORMAT_R16G16_FLOAT; 1355 continue; 1356 1357 /* Align to RGBA16. */ 1358 case PIPE_FORMAT_R16G16B16_UNORM: 1359 *format = PIPE_FORMAT_R16G16B16A16_UNORM; 1360 continue; 1361 case PIPE_FORMAT_R16G16B16_SNORM: 1362 *format = PIPE_FORMAT_R16G16B16A16_SNORM; 1363 continue; 1364 case PIPE_FORMAT_R16G16B16_USCALED: 1365 *format = PIPE_FORMAT_R16G16B16A16_USCALED; 1366 continue; 1367 case PIPE_FORMAT_R16G16B16_SSCALED: 1368 *format = PIPE_FORMAT_R16G16B16A16_SSCALED; 1369 continue; 1370 case PIPE_FORMAT_R16G16B16_FLOAT: 1371 *format = PIPE_FORMAT_R16G16B16A16_FLOAT; 1372 continue; 1373 1374 default:; 1375 } 1376 1377 size = util_format_get_blocksize(*format); 1378 1379 if (size % 4 != 0) { 1380 /* XXX Shouldn't we align the format? */ 1381 fprintf(stderr, "r300_create_vertex_elements_state: " 1382 "Unaligned format %s:%i isn't supported\n", 1383 util_format_short_name(*format), size); 1384 assert(0); 1385 abort(); 1386 } 1387 } 1388 } 1389 } 1390 return velems; 1391} 1392 1393static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1394 void *state) 1395{ 1396 struct r300_context *r300 = r300_context(pipe); 1397 struct r300_vertex_element_state *velems = state; 1398 1399 if (velems == NULL) { 1400 return; 1401 } 1402 1403 r300->velems = velems; 1404 1405 if (r300->draw) { 1406 draw_flush(r300->draw); 1407 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1408 return; 1409 } 1410 1411 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1412 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1413} 1414 1415static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1416{ 1417 FREE(state); 1418} 1419 1420static void* r300_create_vs_state(struct pipe_context* pipe, 1421 const struct pipe_shader_state* shader) 1422{ 1423 struct r300_context* r300 = r300_context(pipe); 1424 1425 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1426 1427 /* Copy state directly into shader. */ 1428 vs->state = *shader; 1429 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1430 1431 r300_init_vs_outputs(vs); 1432 1433 if (r300->screen->caps.has_tcl) { 1434 r300_translate_vertex_shader(r300, vs); 1435 } else { 1436 vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); 1437 } 1438 1439 return vs; 1440} 1441 1442static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1443{ 1444 struct r300_context* r300 = r300_context(pipe); 1445 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1446 1447 if (vs == NULL) { 1448 r300->vs_state.state = NULL; 1449 return; 1450 } 1451 if (vs == r300->vs_state.state) { 1452 return; 1453 } 1454 r300->vs_state.state = vs; 1455 1456 /* The majority of the RS block bits is dependent on the vertex shader. */ 1457 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1458 1459 if (r300->screen->caps.has_tcl) { 1460 r300->vs_state.dirty = TRUE; 1461 r300->vs_state.size = 1462 vs->code.length + 9 + 1463 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1464 1465 if (vs->externals_count) { 1466 r300->vs_constants.dirty = TRUE; 1467 r300->vs_constants.size = vs->externals_count * 4 + 3; 1468 } else { 1469 r300->vs_constants.size = 0; 1470 } 1471 1472 r300->pvs_flush.dirty = TRUE; 1473 } else { 1474 draw_flush(r300->draw); 1475 draw_bind_vertex_shader(r300->draw, 1476 (struct draw_vertex_shader*)vs->draw_vs); 1477 } 1478} 1479 1480static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1481{ 1482 struct r300_context* r300 = r300_context(pipe); 1483 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1484 1485 if (r300->screen->caps.has_tcl) { 1486 rc_constants_destroy(&vs->code.constants); 1487 } else { 1488 draw_delete_vertex_shader(r300->draw, 1489 (struct draw_vertex_shader*)vs->draw_vs); 1490 } 1491 1492 FREE((void*)vs->state.tokens); 1493 FREE(shader); 1494} 1495 1496static void r300_set_constant_buffer(struct pipe_context *pipe, 1497 uint shader, uint index, 1498 struct pipe_resource *buf) 1499{ 1500 struct r300_context* r300 = r300_context(pipe); 1501 struct r300_constant_buffer *cbuf; 1502 struct pipe_transfer *tr; 1503 void *mapped; 1504 int max_size = 0, max_size_bytes = 0, clamped_size = 0; 1505 1506 switch (shader) { 1507 case PIPE_SHADER_VERTEX: 1508 cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; 1509 max_size = 256; 1510 break; 1511 case PIPE_SHADER_FRAGMENT: 1512 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1513 if (r300->screen->caps.is_r500) { 1514 max_size = 256; 1515 } else { 1516 max_size = 32; 1517 } 1518 break; 1519 default: 1520 assert(0); 1521 return; 1522 } 1523 max_size_bytes = max_size * 4 * sizeof(float); 1524 1525 if (buf == NULL || buf->width0 == 0 || 1526 (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL) 1527 { 1528 cbuf->count = 0; 1529 return; 1530 } 1531 1532 if (shader == PIPE_SHADER_FRAGMENT || 1533 (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { 1534 assert((buf->width0 % (4 * sizeof(float))) == 0); 1535 1536 /* Check the size of the constant buffer. */ 1537 /* XXX Subtract immediates and RC_STATE_* variables. */ 1538 if (buf->width0 > max_size_bytes) { 1539 fprintf(stderr, "r300: Max size of the constant buffer is " 1540 "%i*4 floats.\n", max_size); 1541 } 1542 clamped_size = MIN2(buf->width0, max_size_bytes); 1543 1544 memcpy(cbuf->constants, mapped, clamped_size); 1545 cbuf->count = clamped_size / (4 * sizeof(float)); 1546 } 1547 1548 if (shader == PIPE_SHADER_VERTEX) { 1549 if (r300->screen->caps.has_tcl) { 1550 if (r300->vs_constants.size) { 1551 r300->vs_constants.dirty = TRUE; 1552 } 1553 r300->pvs_flush.dirty = TRUE; 1554 } else if (r300->draw) { 1555 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1556 0, mapped, buf->width0); 1557 } 1558 } else if (shader == PIPE_SHADER_FRAGMENT) { 1559 r300->fs_constants.dirty = TRUE; 1560 } 1561 1562 pipe_buffer_unmap(pipe, buf, tr); 1563} 1564 1565void r300_init_state_functions(struct r300_context* r300) 1566{ 1567 r300->context.create_blend_state = r300_create_blend_state; 1568 r300->context.bind_blend_state = r300_bind_blend_state; 1569 r300->context.delete_blend_state = r300_delete_blend_state; 1570 1571 r300->context.set_blend_color = r300_set_blend_color; 1572 1573 r300->context.set_clip_state = r300_set_clip_state; 1574 1575 r300->context.set_constant_buffer = r300_set_constant_buffer; 1576 1577 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1578 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1579 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1580 1581 r300->context.set_stencil_ref = r300_set_stencil_ref; 1582 1583 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1584 1585 r300->context.create_fs_state = r300_create_fs_state; 1586 r300->context.bind_fs_state = r300_bind_fs_state; 1587 r300->context.delete_fs_state = r300_delete_fs_state; 1588 1589 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1590 1591 r300->context.create_rasterizer_state = r300_create_rs_state; 1592 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1593 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1594 1595 r300->context.create_sampler_state = r300_create_sampler_state; 1596 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1597 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1598 r300->context.delete_sampler_state = r300_delete_sampler_state; 1599 1600 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1601 r300->context.create_sampler_view = r300_create_sampler_view; 1602 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1603 1604 r300->context.set_scissor_state = r300_set_scissor_state; 1605 1606 r300->context.set_viewport_state = r300_set_viewport_state; 1607 1608 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1609 1610 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1611 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1612 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1613 1614 r300->context.create_vs_state = r300_create_vs_state; 1615 r300->context.bind_vs_state = r300_bind_vs_state; 1616 r300->context.delete_vs_state = r300_delete_vs_state; 1617} 1618