r300_state.c revision f03791467090db49e5c3d6111cd8d2a7cbe31d75
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_framebuffer.h" 27#include "util/u_half.h" 28#include "util/u_math.h" 29#include "util/u_mm.h" 30#include "util/u_memory.h" 31#include "util/u_pack_color.h" 32#include "util/u_transfer.h" 33 34#include "tgsi/tgsi_parse.h" 35 36#include "pipe/p_config.h" 37 38#include "r300_cb.h" 39#include "r300_context.h" 40#include "r300_emit.h" 41#include "r300_reg.h" 42#include "r300_screen.h" 43#include "r300_screen_buffer.h" 44#include "r300_state_inlines.h" 45#include "r300_fs.h" 46#include "r300_texture.h" 47#include "r300_vs.h" 48#include "r300_winsys.h" 49 50/* r300_state: Functions used to intialize state context by translating 51 * Gallium state objects into semi-native r300 state objects. */ 52 53#define UPDATE_STATE(cso, atom) \ 54 if (cso != atom.state) { \ 55 atom.state = cso; \ 56 r300_mark_atom_dirty(r300, &(atom)); \ 57 } 58 59static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 60 unsigned dstRGB, unsigned dstA) 61{ 62 /* If the blend equation is ADD or REVERSE_SUBTRACT, 63 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 64 * will not be changed. 65 * Notice that the dst factors are the src factors inverted. */ 66 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 67 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 68 srcRGB == PIPE_BLENDFACTOR_ZERO) && 69 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 70 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 71 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 72 srcA == PIPE_BLENDFACTOR_ZERO) && 73 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 74 dstRGB == PIPE_BLENDFACTOR_ONE) && 75 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 76 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 77 dstA == PIPE_BLENDFACTOR_ONE); 78} 79 80static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 81 unsigned dstRGB, unsigned dstA) 82{ 83 /* If the blend equation is ADD or REVERSE_SUBTRACT, 84 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 85 * will not be changed. 86 * Notice that the dst factors are the src factors inverted. */ 87 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 88 srcRGB == PIPE_BLENDFACTOR_ZERO) && 89 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 90 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 91 srcA == PIPE_BLENDFACTOR_ZERO) && 92 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 93 dstRGB == PIPE_BLENDFACTOR_ONE) && 94 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 95 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 96 dstA == PIPE_BLENDFACTOR_ONE); 97} 98 99static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 100 unsigned dstRGB, unsigned dstA) 101{ 102 /* If the blend equation is ADD or REVERSE_SUBTRACT, 103 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 104 * will not be changed. 105 * Notice that the dst factors are the src factors inverted. */ 106 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 107 srcRGB == PIPE_BLENDFACTOR_ZERO) && 108 (srcA == PIPE_BLENDFACTOR_ZERO) && 109 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 110 dstRGB == PIPE_BLENDFACTOR_ONE) && 111 (dstA == PIPE_BLENDFACTOR_ONE); 112} 113 114static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 115 unsigned dstRGB, unsigned dstA) 116{ 117 /* If the blend equation is ADD or REVERSE_SUBTRACT, 118 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 119 * will not be changed. 120 * Notice that the dst factors are the src factors inverted. */ 121 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 122 srcRGB == PIPE_BLENDFACTOR_ZERO) && 123 (srcA == PIPE_BLENDFACTOR_ZERO) && 124 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 125 dstRGB == PIPE_BLENDFACTOR_ONE) && 126 (dstA == PIPE_BLENDFACTOR_ONE); 127} 128 129static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 130 unsigned dstRGB, unsigned dstA) 131{ 132 /* If the blend equation is ADD or REVERSE_SUBTRACT, 133 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 134 * the colorbuffer will not be changed. 135 * Notice that the dst factors are the src factors inverted. */ 136 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 137 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 138 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 139 srcRGB == PIPE_BLENDFACTOR_ZERO) && 140 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 141 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 142 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 143 srcA == PIPE_BLENDFACTOR_ZERO) && 144 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 145 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 146 dstRGB == PIPE_BLENDFACTOR_ONE) && 147 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 148 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 149 dstA == PIPE_BLENDFACTOR_ONE); 150} 151 152static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 153 unsigned dstRGB, unsigned dstA) 154{ 155 /* If the blend equation is ADD or REVERSE_SUBTRACT, 156 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 157 * the colorbuffer will not be changed. 158 * Notice that the dst factors are the src factors inverted. */ 159 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 160 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 161 srcRGB == PIPE_BLENDFACTOR_ZERO) && 162 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 163 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 164 srcA == PIPE_BLENDFACTOR_ZERO) && 165 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 166 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 167 dstRGB == PIPE_BLENDFACTOR_ONE) && 168 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 169 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 170 dstA == PIPE_BLENDFACTOR_ONE); 171} 172 173static unsigned bgra_cmask(unsigned mask) 174{ 175 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 176 177 return ((mask & PIPE_MASK_R) << 2) | 178 ((mask & PIPE_MASK_B) >> 2) | 179 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 180} 181 182/* Create a new blend state based on the CSO blend state. 183 * 184 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 185static void* r300_create_blend_state(struct pipe_context* pipe, 186 const struct pipe_blend_state* state) 187{ 188 struct r300_screen* r300screen = r300_screen(pipe->screen); 189 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 190 uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 191 uint32_t blend_control_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 192 uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 193 uint32_t alpha_blend_control_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 194 uint32_t color_channel_mask = 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ 195 uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ 196 uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ 197 CB_LOCALS; 198 199 blend->state = *state; 200 201 if (state->rt[0].blend_enable) 202 { 203 unsigned eqRGB = state->rt[0].rgb_func; 204 unsigned srcRGB = state->rt[0].rgb_src_factor; 205 unsigned dstRGB = state->rt[0].rgb_dst_factor; 206 207 unsigned eqA = state->rt[0].alpha_func; 208 unsigned srcA = state->rt[0].alpha_src_factor; 209 unsigned dstA = state->rt[0].alpha_dst_factor; 210 211 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 212 * this is just the crappy D3D naming */ 213 blend_control = blend_control_noclamp = 214 R300_ALPHA_BLEND_ENABLE | 215 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 216 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 217 blend_control |= 218 r300_translate_blend_function(eqRGB, TRUE); 219 blend_control_noclamp |= 220 r300_translate_blend_function(eqRGB, FALSE); 221 222 /* Optimization: some operations do not require the destination color. 223 * 224 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 225 * otherwise blending gives incorrect results. It seems to be 226 * a hardware bug. */ 227 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 228 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 229 dstRGB != PIPE_BLENDFACTOR_ZERO || 230 dstA != PIPE_BLENDFACTOR_ZERO || 231 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 232 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 233 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 234 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 235 srcA == PIPE_BLENDFACTOR_DST_COLOR || 236 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 237 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 238 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 239 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 240 /* Enable reading from the colorbuffer. */ 241 blend_control |= R300_READ_ENABLE; 242 blend_control_noclamp |= R300_READ_ENABLE; 243 244 if (r300screen->caps.is_r500) { 245 /* Optimization: Depending on incoming pixels, we can 246 * conditionally disable the reading in hardware... */ 247 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 248 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 249 /* Disable reading if SRC_ALPHA == 0. */ 250 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 251 dstRGB == PIPE_BLENDFACTOR_ZERO) && 252 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 253 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 254 dstA == PIPE_BLENDFACTOR_ZERO)) { 255 blend_control |= R500_SRC_ALPHA_0_NO_READ; 256 } 257 258 /* Disable reading if SRC_ALPHA == 1. */ 259 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 260 dstRGB == PIPE_BLENDFACTOR_ZERO) && 261 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 262 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 263 dstA == PIPE_BLENDFACTOR_ZERO)) { 264 blend_control |= R500_SRC_ALPHA_1_NO_READ; 265 } 266 } 267 } 268 } 269 270 /* Optimization: discard pixels which don't change the colorbuffer. 271 * 272 * The code below is non-trivial and some math is involved. 273 * 274 * Discarding pixels must be disabled when FP16 AA is enabled. 275 * This is a hardware bug. Also, this implementation wouldn't work 276 * with FP blending enabled and equation clamping disabled. 277 * 278 * Equations other than ADD are rarely used and therefore won't be 279 * optimized. */ 280 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 281 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 282 /* ADD: X+Y 283 * REVERSE_SUBTRACT: Y-X 284 * 285 * The idea is: 286 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 287 * then CB will not be changed. 288 * 289 * Given the srcFactor and dstFactor variables, we can derive 290 * what src and dst should be equal to and discard appropriate 291 * pixels. 292 */ 293 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 294 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 295 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 296 dstRGB, dstA)) { 297 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 298 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 299 dstRGB, dstA)) { 300 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 301 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 302 dstRGB, dstA)) { 303 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 304 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 305 dstRGB, dstA)) { 306 blend_control |= 307 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 308 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 309 dstRGB, dstA)) { 310 blend_control |= 311 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 312 } 313 } 314 315 /* separate alpha */ 316 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 317 blend_control |= R300_SEPARATE_ALPHA_ENABLE; 318 blend_control_noclamp |= R300_SEPARATE_ALPHA_ENABLE; 319 alpha_blend_control = alpha_blend_control_noclamp = 320 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 321 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 322 alpha_blend_control |= 323 r300_translate_blend_function(eqA, TRUE); 324 alpha_blend_control_noclamp |= 325 r300_translate_blend_function(eqA, FALSE); 326 } 327 } 328 329 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 330 if (state->logicop_enable) { 331 rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 332 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 333 } 334 335 /* Color channel masks for all MRTs. */ 336 color_channel_mask = bgra_cmask(state->rt[0].colormask); 337 if (r300screen->caps.is_r500 && state->independent_blend_enable) { 338 if (state->rt[1].blend_enable) { 339 color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 340 } 341 if (state->rt[2].blend_enable) { 342 color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 343 } 344 if (state->rt[3].blend_enable) { 345 color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 346 } 347 } 348 349 /* Neither fglrx nor classic r300 ever set this, regardless of dithering 350 * state. Since it's an optional implementation detail, we can leave it 351 * out and never dither. 352 * 353 * This could be revisited if we ever get quality or conformance hints. 354 * 355 if (state->dither) { 356 dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 357 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 358 } 359 */ 360 361 /* Build a command buffer. */ 362 BEGIN_CB(blend->cb_clamp, 8); 363 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 364 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 365 OUT_CB(blend_control); 366 OUT_CB(alpha_blend_control); 367 OUT_CB(color_channel_mask); 368 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 369 END_CB; 370 371 /* Build a command buffer. */ 372 BEGIN_CB(blend->cb_noclamp, 8); 373 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 374 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 375 OUT_CB(blend_control_noclamp); 376 OUT_CB(alpha_blend_control_noclamp); 377 OUT_CB(color_channel_mask); 378 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 379 END_CB; 380 381 /* The same as above, but with no colorbuffer reads and writes. */ 382 BEGIN_CB(blend->cb_no_readwrite, 8); 383 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 384 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 385 OUT_CB(0); 386 OUT_CB(0); 387 OUT_CB(0); 388 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 389 END_CB; 390 391 return (void*)blend; 392} 393 394/* Bind blend state. */ 395static void r300_bind_blend_state(struct pipe_context* pipe, 396 void* state) 397{ 398 struct r300_context* r300 = r300_context(pipe); 399 400 UPDATE_STATE(state, r300->blend_state); 401} 402 403/* Free blend state. */ 404static void r300_delete_blend_state(struct pipe_context* pipe, 405 void* state) 406{ 407 FREE(state); 408} 409 410/* Convert float to 10bit integer */ 411static unsigned float_to_fixed10(float f) 412{ 413 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 414} 415 416/* Set blend color. 417 * Setup both R300 and R500 registers, figure out later which one to write. */ 418static void r300_set_blend_color(struct pipe_context* pipe, 419 const struct pipe_blend_color* color) 420{ 421 struct r300_context* r300 = r300_context(pipe); 422 struct pipe_framebuffer_state *fb = r300->fb_state.state; 423 struct r300_blend_color_state *state = 424 (struct r300_blend_color_state*)r300->blend_color_state.state; 425 struct pipe_blend_color c; 426 enum pipe_format format = fb->nr_cbufs ? fb->cbufs[0]->format : 0; 427 CB_LOCALS; 428 429 state->state = *color; /* Save it, so that we can reuse it in set_fb_state */ 430 c = *color; 431 432 /* The blend color is dependent on the colorbuffer format. */ 433 if (fb->nr_cbufs) { 434 switch (format) { 435 case PIPE_FORMAT_R8_UNORM: 436 case PIPE_FORMAT_L8_UNORM: 437 case PIPE_FORMAT_I8_UNORM: 438 c.color[1] = c.color[0]; 439 break; 440 441 case PIPE_FORMAT_A8_UNORM: 442 c.color[1] = c.color[3]; 443 break; 444 445 case PIPE_FORMAT_R8G8_UNORM: 446 c.color[2] = c.color[1]; 447 break; 448 449 case PIPE_FORMAT_L8A8_UNORM: 450 c.color[2] = c.color[3]; 451 break; 452 453 default:; 454 } 455 } 456 457 if (r300->screen->caps.is_r500) { 458 BEGIN_CB(state->cb, 3); 459 OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); 460 461 switch (format) { 462 case PIPE_FORMAT_R16G16B16A16_FLOAT: 463 OUT_CB(util_float_to_half(c.color[2]) | 464 (util_float_to_half(c.color[3]) << 16)); 465 OUT_CB(util_float_to_half(c.color[0]) | 466 (util_float_to_half(c.color[1]) << 16)); 467 break; 468 469 default: 470 OUT_CB(float_to_fixed10(c.color[0]) | 471 (float_to_fixed10(c.color[3]) << 16)); 472 OUT_CB(float_to_fixed10(c.color[2]) | 473 (float_to_fixed10(c.color[1]) << 16)); 474 } 475 476 END_CB; 477 } else { 478 union util_color uc; 479 util_pack_color(c.color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 480 481 BEGIN_CB(state->cb, 2); 482 OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui); 483 END_CB; 484 } 485 486 r300_mark_atom_dirty(r300, &r300->blend_color_state); 487} 488 489static void r300_set_clip_state(struct pipe_context* pipe, 490 const struct pipe_clip_state* state) 491{ 492 struct r300_context* r300 = r300_context(pipe); 493 struct r300_clip_state *clip = 494 (struct r300_clip_state*)r300->clip_state.state; 495 CB_LOCALS; 496 497 clip->clip = *state; 498 499 if (r300->screen->caps.has_tcl) { 500 r300->clip_state.size = 2 + !!state->nr * 3 + state->nr * 4; 501 502 BEGIN_CB(clip->cb, r300->clip_state.size); 503 if (state->nr) { 504 OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, 505 (r300->screen->caps.is_r500 ? 506 R500_PVS_UCP_START : R300_PVS_UCP_START)); 507 OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, state->nr * 4); 508 OUT_CB_TABLE(state->ucp, state->nr * 4); 509 } 510 OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) | 511 R300_PS_UCP_MODE_CLIP_AS_TRIFAN); 512 END_CB; 513 514 r300_mark_atom_dirty(r300, &r300->clip_state); 515 } else { 516 draw_set_clip_state(r300->draw, state); 517 } 518} 519 520static void 521r300_set_sample_mask(struct pipe_context *pipe, 522 unsigned sample_mask) 523{ 524} 525 526 527/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 528 * 529 * This contains the depth buffer, stencil buffer, alpha test, and such. 530 * On the Radeon, depth and stencil buffer setup are intertwined, which is 531 * the reason for some of the strange-looking assignments across registers. */ 532static void* 533 r300_create_dsa_state(struct pipe_context* pipe, 534 const struct pipe_depth_stencil_alpha_state* state) 535{ 536 struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; 537 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 538 CB_LOCALS; 539 540 dsa->dsa = *state; 541 542 /* Depth test setup. - separate write mask depth for decomp flush */ 543 if (state->depth.writemask) { 544 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 545 } 546 547 if (state->depth.enabled) { 548 dsa->z_buffer_control |= R300_Z_ENABLE; 549 550 dsa->z_stencil_control |= 551 (r300_translate_depth_stencil_function(state->depth.func) << 552 R300_Z_FUNC_SHIFT); 553 } 554 555 /* Stencil buffer setup. */ 556 if (state->stencil[0].enabled) { 557 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 558 dsa->z_stencil_control |= 559 (r300_translate_depth_stencil_function(state->stencil[0].func) << 560 R300_S_FRONT_FUNC_SHIFT) | 561 (r300_translate_stencil_op(state->stencil[0].fail_op) << 562 R300_S_FRONT_SFAIL_OP_SHIFT) | 563 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 564 R300_S_FRONT_ZPASS_OP_SHIFT) | 565 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 566 R300_S_FRONT_ZFAIL_OP_SHIFT); 567 568 dsa->stencil_ref_mask = 569 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 570 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 571 572 if (state->stencil[1].enabled) { 573 dsa->two_sided = TRUE; 574 575 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 576 dsa->z_stencil_control |= 577 (r300_translate_depth_stencil_function(state->stencil[1].func) << 578 R300_S_BACK_FUNC_SHIFT) | 579 (r300_translate_stencil_op(state->stencil[1].fail_op) << 580 R300_S_BACK_SFAIL_OP_SHIFT) | 581 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 582 R300_S_BACK_ZPASS_OP_SHIFT) | 583 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 584 R300_S_BACK_ZFAIL_OP_SHIFT); 585 586 dsa->stencil_ref_bf = 587 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | 588 (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); 589 590 if (caps->is_r500) { 591 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 592 } else { 593 dsa->two_sided_stencil_ref = 594 (state->stencil[0].valuemask != state->stencil[1].valuemask || 595 state->stencil[0].writemask != state->stencil[1].writemask); 596 } 597 } 598 } 599 600 /* Alpha test setup. */ 601 if (state->alpha.enabled) { 602 dsa->alpha_function = 603 r300_translate_alpha_function(state->alpha.func) | 604 R300_FG_ALPHA_FUNC_ENABLE; 605 606 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 607 dsa->alpha_value = util_float_to_half(state->alpha.ref_value); 608 609 if (caps->is_r500) { 610 dsa->alpha_function_fp16 = dsa->alpha_function | 611 R500_FG_ALPHA_FUNC_FP16_ENABLE; 612 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 613 } 614 } 615 616 BEGIN_CB(&dsa->cb_begin, 10); 617 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); 618 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 619 OUT_CB(dsa->z_buffer_control); 620 OUT_CB(dsa->z_stencil_control); 621 OUT_CB(dsa->stencil_ref_mask); 622 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); 623 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 624 END_CB; 625 626 BEGIN_CB(&dsa->cb_begin_fp16, 10); 627 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16); 628 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 629 OUT_CB(dsa->z_buffer_control); 630 OUT_CB(dsa->z_stencil_control); 631 OUT_CB(dsa->stencil_ref_mask); 632 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); 633 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 634 END_CB; 635 636 BEGIN_CB(dsa->cb_zb_no_readwrite, 10); 637 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); 638 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 639 OUT_CB(0); 640 OUT_CB(0); 641 OUT_CB(0); 642 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); 643 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 644 END_CB; 645 646 BEGIN_CB(dsa->cb_fp16_zb_no_readwrite, 10); 647 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16); 648 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 649 OUT_CB(0); 650 OUT_CB(0); 651 OUT_CB(0); 652 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); 653 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 654 END_CB; 655 656 return (void*)dsa; 657} 658 659static void r300_dsa_inject_stencilref(struct r300_context *r300) 660{ 661 struct r300_dsa_state *dsa = 662 (struct r300_dsa_state*)r300->dsa_state.state; 663 664 if (!dsa) 665 return; 666 667 dsa->stencil_ref_mask = 668 (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) | 669 r300->stencil_ref.ref_value[0]; 670 dsa->stencil_ref_bf = 671 (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) | 672 r300->stencil_ref.ref_value[1]; 673} 674 675/* Bind DSA state. */ 676static void r300_bind_dsa_state(struct pipe_context* pipe, 677 void* state) 678{ 679 struct r300_context* r300 = r300_context(pipe); 680 681 if (!state) { 682 return; 683 } 684 685 UPDATE_STATE(state, r300->dsa_state); 686 687 r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */ 688 r300_dsa_inject_stencilref(r300); 689} 690 691/* Free DSA state. */ 692static void r300_delete_dsa_state(struct pipe_context* pipe, 693 void* state) 694{ 695 FREE(state); 696} 697 698static void r300_set_stencil_ref(struct pipe_context* pipe, 699 const struct pipe_stencil_ref* sr) 700{ 701 struct r300_context* r300 = r300_context(pipe); 702 703 r300->stencil_ref = *sr; 704 705 r300_dsa_inject_stencilref(r300); 706 r300_mark_atom_dirty(r300, &r300->dsa_state); 707} 708 709static void r300_tex_set_tiling_flags(struct r300_context *r300, 710 struct r300_resource *tex, 711 unsigned level) 712{ 713 /* Check if the macrotile flag needs to be changed. 714 * Skip changing the flags otherwise. */ 715 if (tex->tex.macrotile[tex->surface_level] != 716 tex->tex.macrotile[level]) { 717 r300->rws->buffer_set_tiling(tex->buf, r300->cs, 718 tex->tex.microtile, tex->tex.macrotile[level], 719 tex->tex.stride_in_bytes[0]); 720 721 tex->surface_level = level; 722 } 723} 724 725/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 726static void r300_fb_set_tiling_flags(struct r300_context *r300, 727 const struct pipe_framebuffer_state *state) 728{ 729 unsigned i; 730 731 /* Set tiling flags for new surfaces. */ 732 for (i = 0; i < state->nr_cbufs; i++) { 733 r300_tex_set_tiling_flags(r300, 734 r300_resource(state->cbufs[i]->texture), 735 state->cbufs[i]->u.tex.level); 736 } 737 if (state->zsbuf) { 738 r300_tex_set_tiling_flags(r300, 739 r300_resource(state->zsbuf->texture), 740 state->zsbuf->u.tex.level); 741 } 742} 743 744static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, 745 const char *binding) 746{ 747 struct pipe_resource *tex = surf->texture; 748 struct r300_resource *rtex = r300_resource(tex); 749 750 fprintf(stderr, 751 "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, " 752 "Lastlayer: %i, Level: %i, Format: %s\n" 753 754 "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, " 755 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", 756 757 binding, index, surf->width, surf->height, 758 surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level, 759 util_format_short_name(surf->format), 760 761 rtex->tex.macrotile[0] ? "YES" : " NO", 762 rtex->tex.microtile ? "YES" : " NO", 763 rtex->tex.stride_in_pixels[0], 764 tex->width0, tex->height0, tex->depth0, 765 tex->last_level, util_format_short_name(tex->format)); 766} 767 768void r300_mark_fb_state_dirty(struct r300_context *r300, 769 enum r300_fb_state_change change) 770{ 771 struct pipe_framebuffer_state *state = r300->fb_state.state; 772 boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); 773 774 r300_mark_atom_dirty(r300, &r300->gpu_flush); 775 r300_mark_atom_dirty(r300, &r300->fb_state); 776 777 /* What is marked as dirty depends on the enum r300_fb_state_change. */ 778 if (change == R300_CHANGED_FB_STATE) { 779 r300_mark_atom_dirty(r300, &r300->aa_state); 780 r300_mark_atom_dirty(r300, &r300->dsa_state); /* for AlphaRef */ 781 r300_set_blend_color(&r300->context, r300->blend_color_state.state); 782 } 783 784 if (change == R300_CHANGED_FB_STATE || 785 change == R300_CHANGED_HYPERZ_FLAG) { 786 r300_mark_atom_dirty(r300, &r300->hyperz_state); 787 } 788 789 if (change == R300_CHANGED_FB_STATE || 790 change == R300_CHANGED_MULTIWRITE) { 791 r300_mark_atom_dirty(r300, &r300->fb_state_pipelined); 792 } 793 794 /* Now compute the fb_state atom size. */ 795 r300->fb_state.size = 2 + (8 * state->nr_cbufs); 796 797 if (r300->cbzb_clear) 798 r300->fb_state.size += 10; 799 else if (state->zsbuf) { 800 r300->fb_state.size += 10; 801 if (can_hyperz) 802 r300->fb_state.size += 8; 803 } 804 805 /* The size of the rest of atoms stays the same. */ 806} 807 808static void 809r300_set_framebuffer_state(struct pipe_context* pipe, 810 const struct pipe_framebuffer_state* state) 811{ 812 struct r300_context* r300 = r300_context(pipe); 813 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 814 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 815 unsigned max_width, max_height, i; 816 uint32_t zbuffer_bpp = 0; 817 818 if (r300->screen->caps.is_r500) { 819 max_width = max_height = 4096; 820 } else if (r300->screen->caps.is_r400) { 821 max_width = max_height = 4021; 822 } else { 823 max_width = max_height = 2560; 824 } 825 826 if (state->width > max_width || state->height > max_height) { 827 fprintf(stderr, "r300: Implementation error: Render targets are too " 828 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 829 return; 830 } 831 832 if (old_state->zsbuf && r300->zmask_in_use && !r300->hyperz_locked) { 833 /* There is a zmask in use, what are we gonna do? */ 834 if (state->zsbuf) { 835 if (!pipe_surface_equal(old_state->zsbuf, state->zsbuf)) { 836 /* Decompress the currently bound zbuffer before we bind another one. */ 837 r300_decompress_zmask(r300); 838 r300->hiz_in_use = FALSE; 839 } 840 } else { 841 /* We don't bind another zbuffer, so lock the current one. */ 842 r300->hyperz_locked = TRUE; 843 pipe_surface_reference(&r300->locked_zbuffer, old_state->zsbuf); 844 } 845 } else if (r300->hyperz_locked && r300->locked_zbuffer) { 846 /* We have a locked zbuffer now, what are we gonna do? */ 847 if (state->zsbuf) { 848 if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) { 849 /* We are binding some other zbuffer, so decompress the locked one, 850 * it gets unlocked automatically. */ 851 r300_decompress_zmask_locked_unsafe(r300); 852 r300->hiz_in_use = FALSE; 853 } else { 854 /* We are binding the locked zbuffer again, so unlock it. */ 855 r300->hyperz_locked = FALSE; 856 } 857 } 858 } 859 860 /* Need to reset clamping or colormask. */ 861 r300_mark_atom_dirty(r300, &r300->blend_state); 862 863 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 864 if (!!old_state->zsbuf != !!state->zsbuf) { 865 r300_mark_atom_dirty(r300, &r300->dsa_state); 866 } 867 868 /* The tiling flags are dependent on the surface miplevel, unfortunately. */ 869 r300_fb_set_tiling_flags(r300, state); 870 871 util_copy_framebuffer_state(r300->fb_state.state, state); 872 873 if (!r300->hyperz_locked) { 874 pipe_surface_reference(&r300->locked_zbuffer, NULL); 875 } 876 877 r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); 878 879 if (state->zsbuf) { 880 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 881 case 2: 882 zbuffer_bpp = 16; 883 break; 884 case 4: 885 zbuffer_bpp = 24; 886 break; 887 } 888 889 /* Polygon offset depends on the zbuffer bit depth. */ 890 if (r300->zbuffer_bpp != zbuffer_bpp) { 891 r300->zbuffer_bpp = zbuffer_bpp; 892 893 if (r300->polygon_offset_enabled) 894 r300_mark_atom_dirty(r300, &r300->rs_state); 895 } 896 } 897 898 /* Set up AA config. */ 899 if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) { 900 aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE; 901 902 switch (state->cbufs[0]->texture->nr_samples) { 903 case 2: 904 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; 905 break; 906 case 3: 907 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3; 908 break; 909 case 4: 910 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; 911 break; 912 case 6: 913 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; 914 break; 915 } 916 } else { 917 aa->aa_config = 0; 918 } 919 920 if (DBG_ON(r300, DBG_FB)) { 921 fprintf(stderr, "r300: set_framebuffer_state:\n"); 922 for (i = 0; i < state->nr_cbufs; i++) { 923 r300_print_fb_surf_info(state->cbufs[i], i, "CB"); 924 } 925 if (state->zsbuf) { 926 r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); 927 } 928 } 929} 930 931/* Create fragment shader state. */ 932static void* r300_create_fs_state(struct pipe_context* pipe, 933 const struct pipe_shader_state* shader) 934{ 935 struct r300_fragment_shader* fs = NULL; 936 937 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 938 939 /* Copy state directly into shader. */ 940 fs->state = *shader; 941 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 942 943 return (void*)fs; 944} 945 946void r300_mark_fs_code_dirty(struct r300_context *r300) 947{ 948 struct r300_fragment_shader* fs = r300_fs(r300); 949 950 r300_mark_atom_dirty(r300, &r300->fs); 951 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 952 r300_mark_atom_dirty(r300, &r300->fs_constants); 953 r300->fs.size = fs->shader->cb_code_size; 954 955 if (r300->screen->caps.is_r500) { 956 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; 957 r300->fs_constants.size = fs->shader->externals_count * 4 + 3; 958 } else { 959 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; 960 r300->fs_constants.size = fs->shader->externals_count * 4 + 1; 961 } 962 963 ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = 964 fs->shader->code.constants_remap_table; 965} 966 967/* Bind fragment shader state. */ 968static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 969{ 970 struct r300_context* r300 = r300_context(pipe); 971 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 972 struct pipe_framebuffer_state *fb = r300->fb_state.state; 973 boolean last_multi_write; 974 975 if (fs == NULL) { 976 r300->fs.state = NULL; 977 return; 978 } 979 980 last_multi_write = r300_fragment_shader_writes_all(r300_fs(r300)); 981 982 r300->fs.state = fs; 983 r300_pick_fragment_shader(r300); 984 r300_mark_fs_code_dirty(r300); 985 986 if (fb->nr_cbufs > 1 && 987 last_multi_write != r300_fragment_shader_writes_all(fs)) { 988 r300_mark_fb_state_dirty(r300, R300_CHANGED_MULTIWRITE); 989 } 990 991 r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ 992} 993 994/* Delete fragment shader state. */ 995static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 996{ 997 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 998 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 999 1000 while (ptr) { 1001 tmp = ptr; 1002 ptr = ptr->next; 1003 rc_constants_destroy(&tmp->code.constants); 1004 FREE(tmp->cb_code); 1005 FREE(tmp); 1006 } 1007 FREE((void*)fs->state.tokens); 1008 FREE(shader); 1009} 1010 1011static void r300_set_polygon_stipple(struct pipe_context* pipe, 1012 const struct pipe_poly_stipple* state) 1013{ 1014 /* XXX no idea how to set this up, but not terribly important */ 1015} 1016 1017/* Create a new rasterizer state based on the CSO rasterizer state. 1018 * 1019 * This is a very large chunk of state, and covers most of the graphics 1020 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 1021 * 1022 * In a not entirely unironic sidenote, this state has nearly nothing to do 1023 * with the actual block on the Radeon called the rasterizer (RS). */ 1024static void* r300_create_rs_state(struct pipe_context* pipe, 1025 const struct pipe_rasterizer_state* state) 1026{ 1027 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 1028 float psiz; 1029 uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ 1030 uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ 1031 uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ 1032 uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ 1033 uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */ 1034 uint32_t cull_mode; /* R300_SU_CULL_MODE: 0x42b8 */ 1035 uint32_t line_stipple_config; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */ 1036 uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ 1037 uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ 1038 uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ 1039 uint32_t round_mode; /* R300_GA_ROUND_MODE: 0x428c */ 1040 1041 /* Point sprites texture coordinates, 0: lower left, 1: upper right */ 1042 float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ 1043 float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ 1044 float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ 1045 float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ 1046 boolean vclamp = state->clamp_vertex_color; 1047 CB_LOCALS; 1048 1049 /* Copy rasterizer state. */ 1050 rs->rs = *state; 1051 rs->rs_draw = *state; 1052 1053 rs->rs.sprite_coord_enable = state->point_quad_rasterization * 1054 state->sprite_coord_enable; 1055 1056 /* Override some states for Draw. */ 1057 rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ 1058 1059#ifdef PIPE_ARCH_LITTLE_ENDIAN 1060 vap_control_status = R300_VC_NO_SWAP; 1061#else 1062 vap_control_status = R300_VC_32BIT_SWAP; 1063#endif 1064 1065 /* If no TCL engine is present, turn off the HW TCL. */ 1066 if (!r300_screen(pipe->screen)->caps.has_tcl) { 1067 vap_control_status |= R300_VAP_TCL_BYPASS; 1068 } 1069 1070 /* Point size width and height. */ 1071 point_size = 1072 pack_float_16_6x(state->point_size) | 1073 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 1074 1075 /* Point size clamping. */ 1076 if (state->point_size_per_vertex) { 1077 /* Per-vertex point size. 1078 * Clamp to [0, max FB size] */ 1079 psiz = pipe->screen->get_paramf(pipe->screen, 1080 PIPE_CAP_MAX_POINT_WIDTH); 1081 point_minmax = 1082 pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; 1083 } else { 1084 /* We cannot disable the point-size vertex output, 1085 * so clamp it. */ 1086 psiz = state->point_size; 1087 point_minmax = 1088 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 1089 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 1090 } 1091 1092 /* Line control. */ 1093 line_control = pack_float_16_6x(state->line_width) | 1094 R300_GA_LINE_CNTL_END_TYPE_COMP; 1095 1096 /* Enable polygon mode */ 1097 polygon_mode = 0; 1098 if (state->fill_front != PIPE_POLYGON_MODE_FILL || 1099 state->fill_back != PIPE_POLYGON_MODE_FILL) { 1100 polygon_mode = R300_GA_POLY_MODE_DUAL; 1101 } 1102 1103 /* Front face */ 1104 if (state->front_ccw) 1105 cull_mode = R300_FRONT_FACE_CCW; 1106 else 1107 cull_mode = R300_FRONT_FACE_CW; 1108 1109 /* Polygon offset */ 1110 polygon_offset_enable = 0; 1111 if (util_get_offset(state, state->fill_front)) { 1112 polygon_offset_enable |= R300_FRONT_ENABLE; 1113 } 1114 if (util_get_offset(state, state->fill_back)) { 1115 polygon_offset_enable |= R300_BACK_ENABLE; 1116 } 1117 1118 rs->polygon_offset_enable = polygon_offset_enable != 0; 1119 1120 /* Polygon mode */ 1121 if (polygon_mode) { 1122 polygon_mode |= 1123 r300_translate_polygon_mode_front(state->fill_front); 1124 polygon_mode |= 1125 r300_translate_polygon_mode_back(state->fill_back); 1126 } 1127 1128 if (state->cull_face & PIPE_FACE_FRONT) { 1129 cull_mode |= R300_CULL_FRONT; 1130 } 1131 if (state->cull_face & PIPE_FACE_BACK) { 1132 cull_mode |= R300_CULL_BACK; 1133 } 1134 1135 if (state->line_stipple_enable) { 1136 line_stipple_config = 1137 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 1138 (fui((float)state->line_stipple_factor) & 1139 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 1140 /* XXX this might need to be scaled up */ 1141 line_stipple_value = state->line_stipple_pattern; 1142 } else { 1143 line_stipple_config = 0; 1144 line_stipple_value = 0; 1145 } 1146 1147 if (state->flatshade) { 1148 rs->color_control = R300_SHADE_MODEL_FLAT; 1149 } else { 1150 rs->color_control = R300_SHADE_MODEL_SMOOTH; 1151 } 1152 1153 clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 1154 1155 /* Point sprites coord mode */ 1156 if (rs->rs.sprite_coord_enable) { 1157 switch (state->sprite_coord_mode) { 1158 case PIPE_SPRITE_COORD_UPPER_LEFT: 1159 point_texcoord_top = 0.0f; 1160 point_texcoord_bottom = 1.0f; 1161 break; 1162 case PIPE_SPRITE_COORD_LOWER_LEFT: 1163 point_texcoord_top = 1.0f; 1164 point_texcoord_bottom = 0.0f; 1165 break; 1166 } 1167 } 1168 1169 /* Vertex color clamping. FP20 means no clamping. */ 1170 round_mode = 1171 R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST | 1172 (!vclamp ? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20 | 1173 R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20) : 0); 1174 1175 /* Build the main command buffer. */ 1176 BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); 1177 OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); 1178 OUT_CB_REG(R300_GA_POINT_SIZE, point_size); 1179 OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2); 1180 OUT_CB(point_minmax); 1181 OUT_CB(line_control); 1182 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2); 1183 OUT_CB(polygon_offset_enable); 1184 rs->cull_mode_index = 9; 1185 OUT_CB(cull_mode); 1186 OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config); 1187 OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); 1188 OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); 1189 OUT_CB_REG(R300_GA_ROUND_MODE, round_mode); 1190 OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); 1191 OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); 1192 OUT_CB_32F(point_texcoord_left); 1193 OUT_CB_32F(point_texcoord_bottom); 1194 OUT_CB_32F(point_texcoord_right); 1195 OUT_CB_32F(point_texcoord_top); 1196 END_CB; 1197 1198 /* Build the two command buffers for polygon offset setup. */ 1199 if (polygon_offset_enable) { 1200 float scale = state->offset_scale * 12; 1201 float offset = state->offset_units * 4; 1202 1203 BEGIN_CB(rs->cb_poly_offset_zb16, 5); 1204 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); 1205 OUT_CB_32F(scale); 1206 OUT_CB_32F(offset); 1207 OUT_CB_32F(scale); 1208 OUT_CB_32F(offset); 1209 END_CB; 1210 1211 offset = state->offset_units * 2; 1212 1213 BEGIN_CB(rs->cb_poly_offset_zb24, 5); 1214 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); 1215 OUT_CB_32F(scale); 1216 OUT_CB_32F(offset); 1217 OUT_CB_32F(scale); 1218 OUT_CB_32F(offset); 1219 END_CB; 1220 } 1221 1222 return (void*)rs; 1223} 1224 1225/* Bind rasterizer state. */ 1226static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 1227{ 1228 struct r300_context* r300 = r300_context(pipe); 1229 struct r300_rs_state* rs = (struct r300_rs_state*)state; 1230 int last_sprite_coord_enable = r300->sprite_coord_enable; 1231 boolean last_two_sided_color = r300->two_sided_color; 1232 boolean last_frag_clamp = r300->frag_clamp; 1233 1234 if (r300->draw && rs) { 1235 draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); 1236 } 1237 1238 if (rs) { 1239 r300->polygon_offset_enabled = rs->polygon_offset_enable; 1240 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 1241 r300->two_sided_color = rs->rs.light_twoside; 1242 r300->frag_clamp = rs->rs.clamp_fragment_color; 1243 } else { 1244 r300->polygon_offset_enabled = FALSE; 1245 r300->sprite_coord_enable = 0; 1246 r300->two_sided_color = FALSE; 1247 r300->frag_clamp = FALSE; 1248 } 1249 1250 UPDATE_STATE(state, r300->rs_state); 1251 r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0); 1252 1253 if (last_sprite_coord_enable != r300->sprite_coord_enable || 1254 last_two_sided_color != r300->two_sided_color) { 1255 r300_mark_atom_dirty(r300, &r300->rs_block_state); 1256 } 1257 1258 if (last_frag_clamp != r300->frag_clamp && 1259 r300->fs.state && r300_pick_fragment_shader(r300)) { 1260 r300_mark_fs_code_dirty(r300); 1261 } 1262} 1263 1264/* Free rasterizer state. */ 1265static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 1266{ 1267 FREE(state); 1268} 1269 1270static void* 1271 r300_create_sampler_state(struct pipe_context* pipe, 1272 const struct pipe_sampler_state* state) 1273{ 1274 struct r300_context* r300 = r300_context(pipe); 1275 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 1276 boolean is_r500 = r300->screen->caps.is_r500; 1277 int lod_bias; 1278 1279 sampler->state = *state; 1280 1281 /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG 1282 * or MIN filter is NEAREST. Since texwrap produces same results 1283 * for CLAMP and CLAMP_TO_EDGE, we use them instead. */ 1284 if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST || 1285 sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) { 1286 /* Wrap S. */ 1287 if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP) 1288 sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1289 else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP) 1290 sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1291 1292 /* Wrap T. */ 1293 if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP) 1294 sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1295 else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP) 1296 sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1297 1298 /* Wrap R. */ 1299 if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP) 1300 sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1301 else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP) 1302 sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1303 } 1304 1305 sampler->filter0 |= 1306 (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) | 1307 (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) | 1308 (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT); 1309 1310 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 1311 state->mag_img_filter, 1312 state->min_mip_filter, 1313 state->max_anisotropy > 0); 1314 1315 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 1316 1317 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 1318 /* We must pass these to the merge function to clamp them properly. */ 1319 sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); 1320 sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); 1321 1322 lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); 1323 1324 sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK; 1325 1326 /* This is very high quality anisotropic filtering for R5xx. 1327 * It's good for benchmarking the performance of texturing but 1328 * in practice we don't want to slow down the driver because it's 1329 * a pretty good performance killer. Feel free to play with it. */ 1330 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 1331 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 1332 } 1333 1334 /* R500-specific fixups and optimizations */ 1335 if (r300->screen->caps.is_r500) { 1336 sampler->filter1 |= R500_BORDER_FIX; 1337 } 1338 1339 return (void*)sampler; 1340} 1341 1342static void r300_bind_sampler_states(struct pipe_context* pipe, 1343 unsigned count, 1344 void** states) 1345{ 1346 struct r300_context* r300 = r300_context(pipe); 1347 struct r300_textures_state* state = 1348 (struct r300_textures_state*)r300->textures_state.state; 1349 unsigned tex_units = r300->screen->caps.num_tex_units; 1350 1351 if (count > tex_units) { 1352 return; 1353 } 1354 1355 memcpy(state->sampler_states, states, sizeof(void*) * count); 1356 state->sampler_state_count = count; 1357 1358 r300_mark_atom_dirty(r300, &r300->textures_state); 1359} 1360 1361static void r300_lacks_vertex_textures(struct pipe_context* pipe, 1362 unsigned count, 1363 void** states) 1364{ 1365} 1366 1367static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1368{ 1369 FREE(state); 1370} 1371 1372static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num) 1373{ 1374 /* This looks like a hack, but I believe it's suppose to work like 1375 * that. To illustrate how this works, let's assume you have 5 textures. 1376 * From docs, 5 and the successive numbers are: 1377 * 1378 * FOURTH_1 = 5 1379 * FOURTH_2 = 6 1380 * FOURTH_3 = 7 1381 * EIGHTH_0 = 8 1382 * EIGHTH_1 = 9 1383 * 1384 * First 3 textures will get 3/4 of size of the cache, divived evenly 1385 * between them. The last 1/4 of the cache must be divided between 1386 * the last 2 textures, each will therefore get 1/8 of the cache. 1387 * Why not just to use "5 + texture_index" ? 1388 * 1389 * This simple trick works for all "num" <= 16. 1390 */ 1391 if (num <= 1) 1392 return R300_TX_CACHE(R300_TX_CACHE_WHOLE); 1393 else 1394 return R300_TX_CACHE(num + index); 1395} 1396 1397static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 1398 unsigned count, 1399 struct pipe_sampler_view** views) 1400{ 1401 struct r300_context* r300 = r300_context(pipe); 1402 struct r300_textures_state* state = 1403 (struct r300_textures_state*)r300->textures_state.state; 1404 struct r300_resource *texture; 1405 unsigned i, real_num_views = 0, view_index = 0; 1406 unsigned tex_units = r300->screen->caps.num_tex_units; 1407 boolean dirty_tex = FALSE; 1408 1409 if (count > tex_units) { 1410 return; 1411 } 1412 1413 /* Calculate the real number of views. */ 1414 for (i = 0; i < count; i++) { 1415 if (views[i]) 1416 real_num_views++; 1417 } 1418 1419 for (i = 0; i < count; i++) { 1420 pipe_sampler_view_reference( 1421 (struct pipe_sampler_view**)&state->sampler_views[i], 1422 views[i]); 1423 1424 if (!views[i]) { 1425 continue; 1426 } 1427 1428 /* A new sampler view (= texture)... */ 1429 dirty_tex = TRUE; 1430 1431 /* Set the texrect factor in the fragment shader. 1432 * Needed for RECT and NPOT fallback. */ 1433 texture = r300_resource(views[i]->texture); 1434 if (texture->tex.is_npot) { 1435 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1436 } 1437 1438 state->sampler_views[i]->texcache_region = 1439 r300_assign_texture_cache_region(view_index, real_num_views); 1440 view_index++; 1441 } 1442 1443 for (i = count; i < tex_units; i++) { 1444 if (state->sampler_views[i]) { 1445 pipe_sampler_view_reference( 1446 (struct pipe_sampler_view**)&state->sampler_views[i], 1447 NULL); 1448 } 1449 } 1450 1451 state->sampler_view_count = count; 1452 1453 r300_mark_atom_dirty(r300, &r300->textures_state); 1454 1455 if (dirty_tex) { 1456 r300_mark_atom_dirty(r300, &r300->texture_cache_inval); 1457 } 1458} 1459 1460static struct pipe_sampler_view * 1461r300_create_sampler_view(struct pipe_context *pipe, 1462 struct pipe_resource *texture, 1463 const struct pipe_sampler_view *templ) 1464{ 1465 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1466 struct r300_resource *tex = r300_resource(texture); 1467 boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; 1468 boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle; 1469 1470 if (view) { 1471 view->base = *templ; 1472 view->base.reference.count = 1; 1473 view->base.context = pipe; 1474 view->base.texture = NULL; 1475 pipe_resource_reference(&view->base.texture, texture); 1476 1477 view->swizzle[0] = templ->swizzle_r; 1478 view->swizzle[1] = templ->swizzle_g; 1479 view->swizzle[2] = templ->swizzle_b; 1480 view->swizzle[3] = templ->swizzle_a; 1481 1482 view->format = tex->tx_format; 1483 view->format.format1 |= r300_translate_texformat(templ->format, 1484 view->swizzle, 1485 is_r500, 1486 dxtc_swizzle); 1487 if (is_r500) { 1488 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1489 } 1490 } 1491 1492 return (struct pipe_sampler_view*)view; 1493} 1494 1495static void 1496r300_sampler_view_destroy(struct pipe_context *pipe, 1497 struct pipe_sampler_view *view) 1498{ 1499 pipe_resource_reference(&view->texture, NULL); 1500 FREE(view); 1501} 1502 1503static void r300_set_scissor_state(struct pipe_context* pipe, 1504 const struct pipe_scissor_state* state) 1505{ 1506 struct r300_context* r300 = r300_context(pipe); 1507 1508 memcpy(r300->scissor_state.state, state, 1509 sizeof(struct pipe_scissor_state)); 1510 1511 r300_mark_atom_dirty(r300, &r300->scissor_state); 1512} 1513 1514static void r300_set_viewport_state(struct pipe_context* pipe, 1515 const struct pipe_viewport_state* state) 1516{ 1517 struct r300_context* r300 = r300_context(pipe); 1518 struct r300_viewport_state* viewport = 1519 (struct r300_viewport_state*)r300->viewport_state.state; 1520 1521 r300->viewport = *state; 1522 1523 if (r300->draw) { 1524 draw_set_viewport_state(r300->draw, state); 1525 viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; 1526 return; 1527 } 1528 1529 /* Do the transform in HW. */ 1530 viewport->vte_control = R300_VTX_W0_FMT; 1531 1532 if (state->scale[0] != 1.0f) { 1533 viewport->xscale = state->scale[0]; 1534 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1535 } 1536 if (state->scale[1] != 1.0f) { 1537 viewport->yscale = state->scale[1]; 1538 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1539 } 1540 if (state->scale[2] != 1.0f) { 1541 viewport->zscale = state->scale[2]; 1542 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1543 } 1544 if (state->translate[0] != 0.0f) { 1545 viewport->xoffset = state->translate[0]; 1546 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1547 } 1548 if (state->translate[1] != 0.0f) { 1549 viewport->yoffset = state->translate[1]; 1550 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1551 } 1552 if (state->translate[2] != 0.0f) { 1553 viewport->zoffset = state->translate[2]; 1554 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1555 } 1556 1557 r300_mark_atom_dirty(r300, &r300->viewport_state); 1558 if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1559 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1560 } 1561} 1562 1563static void r300_set_vertex_buffers(struct pipe_context* pipe, 1564 unsigned count, 1565 const struct pipe_vertex_buffer* buffers) 1566{ 1567 struct r300_context* r300 = r300_context(pipe); 1568 unsigned i; 1569 struct pipe_vertex_buffer dummy_vb = {0}; 1570 1571 /* There must be at least one vertex buffer set, otherwise it locks up. */ 1572 if (!count) { 1573 dummy_vb.buffer = r300->dummy_vb; 1574 buffers = &dummy_vb; 1575 count = 1; 1576 } 1577 1578 u_vbuf_mgr_set_vertex_buffers(r300->vbuf_mgr, count, buffers); 1579 1580 if (r300->screen->caps.has_tcl) { 1581 /* HW TCL. */ 1582 for (i = 0; i < count; i++) { 1583 if (buffers[i].buffer && 1584 !r300_resource(buffers[i].buffer)->b.user_ptr) { 1585 } 1586 } 1587 r300->vertex_arrays_dirty = TRUE; 1588 } else { 1589 /* SW TCL. */ 1590 draw_set_vertex_buffers(r300->draw, count, buffers); 1591 } 1592} 1593 1594static void r300_set_index_buffer(struct pipe_context* pipe, 1595 const struct pipe_index_buffer *ib) 1596{ 1597 struct r300_context* r300 = r300_context(pipe); 1598 1599 if (ib && ib->buffer) { 1600 assert(ib->offset % ib->index_size == 0); 1601 1602 pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer); 1603 memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer)); 1604 r300->index_buffer.offset /= r300->index_buffer.index_size; 1605 } 1606 else { 1607 pipe_resource_reference(&r300->index_buffer.buffer, NULL); 1608 memset(&r300->index_buffer, 0, sizeof(r300->index_buffer)); 1609 } 1610 1611 if (!r300->screen->caps.has_tcl) { 1612 draw_set_index_buffer(r300->draw, ib); 1613 } 1614} 1615 1616/* Initialize the PSC tables. */ 1617static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1618{ 1619 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1620 uint16_t type, swizzle; 1621 enum pipe_format format; 1622 unsigned i; 1623 1624 /* Vertex shaders have no semantics on their inputs, 1625 * so PSC should just route stuff based on the vertex elements, 1626 * and not on attrib information. */ 1627 for (i = 0; i < velems->count; i++) { 1628 format = velems->velem[i].src_format; 1629 1630 type = r300_translate_vertex_data_type(format); 1631 if (type == R300_INVALID_FORMAT) { 1632 fprintf(stderr, "r300: Bad vertex format %s.\n", 1633 util_format_short_name(format)); 1634 assert(0); 1635 abort(); 1636 } 1637 1638 type |= i << R300_DST_VEC_LOC_SHIFT; 1639 swizzle = r300_translate_vertex_data_swizzle(format); 1640 1641 if (i & 1) { 1642 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1643 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1644 } else { 1645 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1646 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1647 } 1648 } 1649 1650 /* Set the last vector in the PSC. */ 1651 if (i) { 1652 i -= 1; 1653 } 1654 vstream->vap_prog_stream_cntl[i >> 1] |= 1655 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1656 1657 vstream->count = (i >> 1) + 1; 1658} 1659 1660static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1661 unsigned count, 1662 const struct pipe_vertex_element* attribs) 1663{ 1664 struct r300_context *r300 = r300_context(pipe); 1665 struct r300_vertex_element_state *velems; 1666 unsigned i; 1667 struct pipe_vertex_element dummy_attrib = {0}; 1668 1669 /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */ 1670 if (!count) { 1671 dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM; 1672 attribs = &dummy_attrib; 1673 count = 1; 1674 } else if (count > 16) { 1675 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1676 " requested %i, using 16.\n", count); 1677 count = 16; 1678 } 1679 1680 velems = CALLOC_STRUCT(r300_vertex_element_state); 1681 if (!velems) 1682 return NULL; 1683 1684 velems->count = count; 1685 velems->vmgr_elements = 1686 u_vbuf_mgr_create_vertex_elements(r300->vbuf_mgr, count, attribs, 1687 velems->velem); 1688 1689 if (r300_screen(pipe->screen)->caps.has_tcl) { 1690 /* Setup PSC. 1691 * The unused components will be replaced by (..., 0, 1). */ 1692 r300_vertex_psc(velems); 1693 1694 for (i = 0; i < count; i++) { 1695 velems->format_size[i] = 1696 align(util_format_get_blocksize(velems->velem[i].src_format), 4); 1697 velems->vertex_size_dwords += velems->format_size[i] / 4; 1698 } 1699 } 1700 1701 return velems; 1702} 1703 1704static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1705 void *state) 1706{ 1707 struct r300_context *r300 = r300_context(pipe); 1708 struct r300_vertex_element_state *velems = state; 1709 1710 if (velems == NULL) { 1711 return; 1712 } 1713 1714 r300->velems = velems; 1715 1716 u_vbuf_mgr_bind_vertex_elements(r300->vbuf_mgr, state, velems->vmgr_elements); 1717 1718 if (r300->draw) { 1719 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1720 return; 1721 } 1722 1723 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1724 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1725 r300->vertex_arrays_dirty = TRUE; 1726} 1727 1728static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1729{ 1730 struct r300_context *r300 = r300_context(pipe); 1731 struct r300_vertex_element_state *velems = state; 1732 1733 u_vbuf_mgr_destroy_vertex_elements(r300->vbuf_mgr, velems->vmgr_elements); 1734 FREE(state); 1735} 1736 1737static void* r300_create_vs_state(struct pipe_context* pipe, 1738 const struct pipe_shader_state* shader) 1739{ 1740 struct r300_context* r300 = r300_context(pipe); 1741 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1742 1743 /* Copy state directly into shader. */ 1744 vs->state = *shader; 1745 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1746 1747 if (r300->screen->caps.has_tcl) { 1748 r300_init_vs_outputs(vs); 1749 r300_translate_vertex_shader(r300, vs); 1750 } else { 1751 r300_draw_init_vertex_shader(r300->draw, vs); 1752 } 1753 1754 return vs; 1755} 1756 1757static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1758{ 1759 struct r300_context* r300 = r300_context(pipe); 1760 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1761 1762 if (vs == NULL) { 1763 r300->vs_state.state = NULL; 1764 return; 1765 } 1766 if (vs == r300->vs_state.state) { 1767 return; 1768 } 1769 r300->vs_state.state = vs; 1770 1771 /* The majority of the RS block bits is dependent on the vertex shader. */ 1772 r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ 1773 1774 if (r300->screen->caps.has_tcl) { 1775 unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2; 1776 r300_mark_atom_dirty(r300, &r300->vs_state); 1777 r300->vs_state.size = 1778 vs->code.length + 9 + 1779 (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0); 1780 1781 r300_mark_atom_dirty(r300, &r300->vs_constants); 1782 r300->vs_constants.size = 1783 2 + 1784 (vs->externals_count ? vs->externals_count * 4 + 3 : 0) + 1785 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1786 1787 ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = 1788 vs->code.constants_remap_table; 1789 1790 r300_mark_atom_dirty(r300, &r300->pvs_flush); 1791 } else { 1792 draw_bind_vertex_shader(r300->draw, 1793 (struct draw_vertex_shader*)vs->draw_vs); 1794 } 1795} 1796 1797static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1798{ 1799 struct r300_context* r300 = r300_context(pipe); 1800 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1801 1802 if (r300->screen->caps.has_tcl) { 1803 rc_constants_destroy(&vs->code.constants); 1804 if (vs->code.constants_remap_table) 1805 FREE(vs->code.constants_remap_table); 1806 } else { 1807 draw_delete_vertex_shader(r300->draw, 1808 (struct draw_vertex_shader*)vs->draw_vs); 1809 } 1810 1811 FREE((void*)vs->state.tokens); 1812 FREE(shader); 1813} 1814 1815static void r300_set_constant_buffer(struct pipe_context *pipe, 1816 uint shader, uint index, 1817 struct pipe_resource *buf) 1818{ 1819 struct r300_context* r300 = r300_context(pipe); 1820 struct r300_constant_buffer *cbuf; 1821 struct r300_resource *rbuf = r300_resource(buf); 1822 uint32_t *mapped; 1823 1824 switch (shader) { 1825 case PIPE_SHADER_VERTEX: 1826 cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; 1827 break; 1828 case PIPE_SHADER_FRAGMENT: 1829 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1830 break; 1831 default: 1832 return; 1833 } 1834 1835 if (buf == NULL || buf->width0 == 0) 1836 return; 1837 1838 if (rbuf->b.user_ptr) 1839 mapped = (uint32_t*)rbuf->b.user_ptr; 1840 else if (rbuf->constant_buffer) 1841 mapped = (uint32_t*)rbuf->constant_buffer; 1842 else 1843 return; 1844 1845 if (shader == PIPE_SHADER_FRAGMENT || 1846 (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { 1847 cbuf->ptr = mapped; 1848 } 1849 1850 if (shader == PIPE_SHADER_VERTEX) { 1851 if (r300->screen->caps.has_tcl) { 1852 struct r300_vertex_shader *vs = 1853 (struct r300_vertex_shader*)r300->vs_state.state; 1854 1855 if (!vs) { 1856 cbuf->buffer_base = 0; 1857 return; 1858 } 1859 1860 cbuf->buffer_base = r300->vs_const_base; 1861 r300->vs_const_base += vs->code.constants.Count; 1862 if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) { 1863 r300->vs_const_base = vs->code.constants.Count; 1864 cbuf->buffer_base = 0; 1865 r300_mark_atom_dirty(r300, &r300->pvs_flush); 1866 } 1867 r300_mark_atom_dirty(r300, &r300->vs_constants); 1868 } else if (r300->draw) { 1869 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1870 0, mapped, buf->width0); 1871 } 1872 } else if (shader == PIPE_SHADER_FRAGMENT) { 1873 r300_mark_atom_dirty(r300, &r300->fs_constants); 1874 } 1875} 1876 1877static void r300_texture_barrier(struct pipe_context *pipe) 1878{ 1879 struct r300_context *r300 = r300_context(pipe); 1880 1881 r300_mark_atom_dirty(r300, &r300->gpu_flush); 1882 r300_mark_atom_dirty(r300, &r300->texture_cache_inval); 1883} 1884 1885void r300_init_state_functions(struct r300_context* r300) 1886{ 1887 r300->context.create_blend_state = r300_create_blend_state; 1888 r300->context.bind_blend_state = r300_bind_blend_state; 1889 r300->context.delete_blend_state = r300_delete_blend_state; 1890 1891 r300->context.set_blend_color = r300_set_blend_color; 1892 1893 r300->context.set_clip_state = r300_set_clip_state; 1894 r300->context.set_sample_mask = r300_set_sample_mask; 1895 1896 r300->context.set_constant_buffer = r300_set_constant_buffer; 1897 1898 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1899 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1900 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1901 1902 r300->context.set_stencil_ref = r300_set_stencil_ref; 1903 1904 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1905 1906 r300->context.create_fs_state = r300_create_fs_state; 1907 r300->context.bind_fs_state = r300_bind_fs_state; 1908 r300->context.delete_fs_state = r300_delete_fs_state; 1909 1910 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1911 1912 r300->context.create_rasterizer_state = r300_create_rs_state; 1913 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1914 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1915 1916 r300->context.create_sampler_state = r300_create_sampler_state; 1917 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1918 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1919 r300->context.delete_sampler_state = r300_delete_sampler_state; 1920 1921 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1922 r300->context.create_sampler_view = r300_create_sampler_view; 1923 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1924 1925 r300->context.set_scissor_state = r300_set_scissor_state; 1926 1927 r300->context.set_viewport_state = r300_set_viewport_state; 1928 1929 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1930 r300->context.set_index_buffer = r300_set_index_buffer; 1931 r300->context.redefine_user_buffer = u_default_redefine_user_buffer; 1932 1933 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1934 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1935 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1936 1937 r300->context.create_vs_state = r300_create_vs_state; 1938 r300->context.bind_vs_state = r300_bind_vs_state; 1939 r300->context.delete_vs_state = r300_delete_vs_state; 1940 1941 r300->context.texture_barrier = r300_texture_barrier; 1942} 1943