r300_state.c revision bc1c8369384b5e16547c5bf9728aa78f8dfd66cc
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 49/* r300_state: Functions used to intialize state context by translating 50 * Gallium state objects into semi-native r300 state objects. */ 51 52#define UPDATE_STATE(cso, atom) \ 53 if (cso != atom.state) { \ 54 atom.state = cso; \ 55 r300_mark_atom_dirty(r300, &(atom)); \ 56 } 57 58static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 59 unsigned dstRGB, unsigned dstA) 60{ 61 /* If the blend equation is ADD or REVERSE_SUBTRACT, 62 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 63 * will not be changed. 64 * Notice that the dst factors are the src factors inverted. */ 65 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 66 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 67 srcRGB == PIPE_BLENDFACTOR_ZERO) && 68 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 69 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 70 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 71 srcA == PIPE_BLENDFACTOR_ZERO) && 72 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 73 dstRGB == PIPE_BLENDFACTOR_ONE) && 74 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 75 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 76 dstA == PIPE_BLENDFACTOR_ONE); 77} 78 79static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 80 unsigned dstRGB, unsigned dstA) 81{ 82 /* If the blend equation is ADD or REVERSE_SUBTRACT, 83 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 84 * will not be changed. 85 * Notice that the dst factors are the src factors inverted. */ 86 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 87 srcRGB == PIPE_BLENDFACTOR_ZERO) && 88 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 89 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 90 srcA == PIPE_BLENDFACTOR_ZERO) && 91 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 92 dstRGB == PIPE_BLENDFACTOR_ONE) && 93 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 94 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 95 dstA == PIPE_BLENDFACTOR_ONE); 96} 97 98static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 99 unsigned dstRGB, unsigned dstA) 100{ 101 /* If the blend equation is ADD or REVERSE_SUBTRACT, 102 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 103 * will not be changed. 104 * Notice that the dst factors are the src factors inverted. */ 105 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 106 srcRGB == PIPE_BLENDFACTOR_ZERO) && 107 (srcA == PIPE_BLENDFACTOR_ZERO) && 108 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 109 dstRGB == PIPE_BLENDFACTOR_ONE) && 110 (dstA == PIPE_BLENDFACTOR_ONE); 111} 112 113static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 114 unsigned dstRGB, unsigned dstA) 115{ 116 /* If the blend equation is ADD or REVERSE_SUBTRACT, 117 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 118 * will not be changed. 119 * Notice that the dst factors are the src factors inverted. */ 120 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 121 srcRGB == PIPE_BLENDFACTOR_ZERO) && 122 (srcA == PIPE_BLENDFACTOR_ZERO) && 123 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 124 dstRGB == PIPE_BLENDFACTOR_ONE) && 125 (dstA == PIPE_BLENDFACTOR_ONE); 126} 127 128static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 129 unsigned dstRGB, unsigned dstA) 130{ 131 /* If the blend equation is ADD or REVERSE_SUBTRACT, 132 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 133 * the colorbuffer will not be changed. 134 * Notice that the dst factors are the src factors inverted. */ 135 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 136 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 137 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 138 srcRGB == PIPE_BLENDFACTOR_ZERO) && 139 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 140 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 141 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 142 srcA == PIPE_BLENDFACTOR_ZERO) && 143 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 144 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 145 dstRGB == PIPE_BLENDFACTOR_ONE) && 146 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 147 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 148 dstA == PIPE_BLENDFACTOR_ONE); 149} 150 151static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 152 unsigned dstRGB, unsigned dstA) 153{ 154 /* If the blend equation is ADD or REVERSE_SUBTRACT, 155 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 156 * the colorbuffer will not be changed. 157 * Notice that the dst factors are the src factors inverted. */ 158 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 159 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 160 srcRGB == PIPE_BLENDFACTOR_ZERO) && 161 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 162 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 163 srcA == PIPE_BLENDFACTOR_ZERO) && 164 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 165 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 166 dstRGB == PIPE_BLENDFACTOR_ONE) && 167 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 168 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 169 dstA == PIPE_BLENDFACTOR_ONE); 170} 171 172static unsigned bgra_cmask(unsigned mask) 173{ 174 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 175 176 return ((mask & PIPE_MASK_R) << 2) | 177 ((mask & PIPE_MASK_B) >> 2) | 178 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 179} 180 181/* Create a new blend state based on the CSO blend state. 182 * 183 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 184static void* r300_create_blend_state(struct pipe_context* pipe, 185 const struct pipe_blend_state* state) 186{ 187 struct r300_screen* r300screen = r300_screen(pipe->screen); 188 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 189 uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 190 uint32_t blend_control_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 191 uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 192 uint32_t alpha_blend_control_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 193 uint32_t color_channel_mask = 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ 194 uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ 195 uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ 196 CB_LOCALS; 197 198 blend->state = *state; 199 200 if (state->rt[0].blend_enable) 201 { 202 unsigned eqRGB = state->rt[0].rgb_func; 203 unsigned srcRGB = state->rt[0].rgb_src_factor; 204 unsigned dstRGB = state->rt[0].rgb_dst_factor; 205 206 unsigned eqA = state->rt[0].alpha_func; 207 unsigned srcA = state->rt[0].alpha_src_factor; 208 unsigned dstA = state->rt[0].alpha_dst_factor; 209 210 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 211 * this is just the crappy D3D naming */ 212 blend_control = blend_control_noclamp = 213 R300_ALPHA_BLEND_ENABLE | 214 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 215 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 216 blend_control |= 217 r300_translate_blend_function(eqRGB, TRUE); 218 blend_control_noclamp |= 219 r300_translate_blend_function(eqRGB, FALSE); 220 221 /* Optimization: some operations do not require the destination color. 222 * 223 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 224 * otherwise blending gives incorrect results. It seems to be 225 * a hardware bug. */ 226 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 227 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 228 dstRGB != PIPE_BLENDFACTOR_ZERO || 229 dstA != PIPE_BLENDFACTOR_ZERO || 230 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 231 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 232 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 233 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 234 srcA == PIPE_BLENDFACTOR_DST_COLOR || 235 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 236 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 237 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 238 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 239 /* Enable reading from the colorbuffer. */ 240 blend_control |= R300_READ_ENABLE; 241 blend_control_noclamp |= R300_READ_ENABLE; 242 243 if (r300screen->caps.is_r500) { 244 /* Optimization: Depending on incoming pixels, we can 245 * conditionally disable the reading in hardware... */ 246 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 247 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 248 /* Disable reading if SRC_ALPHA == 0. */ 249 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 250 dstRGB == PIPE_BLENDFACTOR_ZERO) && 251 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 252 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 253 dstA == PIPE_BLENDFACTOR_ZERO)) { 254 blend_control |= R500_SRC_ALPHA_0_NO_READ; 255 } 256 257 /* Disable reading if SRC_ALPHA == 1. */ 258 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 259 dstRGB == PIPE_BLENDFACTOR_ZERO) && 260 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 261 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 262 dstA == PIPE_BLENDFACTOR_ZERO)) { 263 blend_control |= R500_SRC_ALPHA_1_NO_READ; 264 } 265 } 266 } 267 } 268 269 /* Optimization: discard pixels which don't change the colorbuffer. 270 * 271 * The code below is non-trivial and some math is involved. 272 * 273 * Discarding pixels must be disabled when FP16 AA is enabled. 274 * This is a hardware bug. Also, this implementation wouldn't work 275 * with FP blending enabled and equation clamping disabled. 276 * 277 * Equations other than ADD are rarely used and therefore won't be 278 * optimized. */ 279 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 280 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 281 /* ADD: X+Y 282 * REVERSE_SUBTRACT: Y-X 283 * 284 * The idea is: 285 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 286 * then CB will not be changed. 287 * 288 * Given the srcFactor and dstFactor variables, we can derive 289 * what src and dst should be equal to and discard appropriate 290 * pixels. 291 */ 292 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 293 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 294 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 295 dstRGB, dstA)) { 296 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 297 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 298 dstRGB, dstA)) { 299 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 300 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 301 dstRGB, dstA)) { 302 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 303 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 304 dstRGB, dstA)) { 305 blend_control |= 306 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 307 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 308 dstRGB, dstA)) { 309 blend_control |= 310 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 311 } 312 } 313 314 /* separate alpha */ 315 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 316 blend_control |= R300_SEPARATE_ALPHA_ENABLE; 317 blend_control_noclamp |= R300_SEPARATE_ALPHA_ENABLE; 318 alpha_blend_control = alpha_blend_control_noclamp = 319 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 320 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 321 alpha_blend_control |= 322 r300_translate_blend_function(eqA, TRUE); 323 alpha_blend_control_noclamp |= 324 r300_translate_blend_function(eqA, FALSE); 325 } 326 } 327 328 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 329 if (state->logicop_enable) { 330 rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 331 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 332 } 333 334 /* Color channel masks for all MRTs. */ 335 color_channel_mask = bgra_cmask(state->rt[0].colormask); 336 if (r300screen->caps.is_r500 && state->independent_blend_enable) { 337 if (state->rt[1].blend_enable) { 338 color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 339 } 340 if (state->rt[2].blend_enable) { 341 color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 342 } 343 if (state->rt[3].blend_enable) { 344 color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 345 } 346 } 347 348 /* Neither fglrx nor classic r300 ever set this, regardless of dithering 349 * state. Since it's an optional implementation detail, we can leave it 350 * out and never dither. 351 * 352 * This could be revisited if we ever get quality or conformance hints. 353 * 354 if (state->dither) { 355 dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 356 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 357 } 358 */ 359 360 /* Build a command buffer. */ 361 BEGIN_CB(blend->cb_clamp, 8); 362 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 363 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 364 OUT_CB(blend_control); 365 OUT_CB(alpha_blend_control); 366 OUT_CB(color_channel_mask); 367 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 368 END_CB; 369 370 /* Build a command buffer. */ 371 BEGIN_CB(blend->cb_noclamp, 8); 372 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 373 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 374 OUT_CB(blend_control_noclamp); 375 OUT_CB(alpha_blend_control_noclamp); 376 OUT_CB(color_channel_mask); 377 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 378 END_CB; 379 380 /* The same as above, but with no colorbuffer reads and writes. */ 381 BEGIN_CB(blend->cb_no_readwrite, 8); 382 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 383 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 384 OUT_CB(0); 385 OUT_CB(0); 386 OUT_CB(0); 387 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 388 END_CB; 389 390 return (void*)blend; 391} 392 393/* Bind blend state. */ 394static void r300_bind_blend_state(struct pipe_context* pipe, 395 void* state) 396{ 397 struct r300_context* r300 = r300_context(pipe); 398 399 UPDATE_STATE(state, r300->blend_state); 400} 401 402/* Free blend state. */ 403static void r300_delete_blend_state(struct pipe_context* pipe, 404 void* state) 405{ 406 FREE(state); 407} 408 409/* Convert float to 10bit integer */ 410static unsigned float_to_fixed10(float f) 411{ 412 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 413} 414 415/* Set blend color. 416 * Setup both R300 and R500 registers, figure out later which one to write. */ 417static void r300_set_blend_color(struct pipe_context* pipe, 418 const struct pipe_blend_color* color) 419{ 420 struct r300_context* r300 = r300_context(pipe); 421 struct pipe_framebuffer_state *fb = r300->fb_state.state; 422 struct r300_blend_color_state *state = 423 (struct r300_blend_color_state*)r300->blend_color_state.state; 424 struct pipe_blend_color c; 425 enum pipe_format format = fb->nr_cbufs ? fb->cbufs[0]->format : 0; 426 CB_LOCALS; 427 428 state->state = *color; /* Save it, so that we can reuse it in set_fb_state */ 429 c = *color; 430 431 /* The blend color is dependent on the colorbuffer format. */ 432 if (fb->nr_cbufs) { 433 switch (format) { 434 case PIPE_FORMAT_R8_UNORM: 435 case PIPE_FORMAT_L8_UNORM: 436 case PIPE_FORMAT_I8_UNORM: 437 c.color[1] = c.color[0]; 438 break; 439 440 case PIPE_FORMAT_A8_UNORM: 441 c.color[1] = c.color[3]; 442 break; 443 444 case PIPE_FORMAT_R8G8_UNORM: 445 c.color[2] = c.color[1]; 446 break; 447 448 case PIPE_FORMAT_L8A8_UNORM: 449 c.color[2] = c.color[3]; 450 break; 451 452 default:; 453 } 454 } 455 456 if (r300->screen->caps.is_r500) { 457 BEGIN_CB(state->cb, 3); 458 OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); 459 460 switch (format) { 461 case PIPE_FORMAT_R16G16B16A16_FLOAT: 462 OUT_CB(util_float_to_half(c.color[2]) | 463 (util_float_to_half(c.color[3]) << 16)); 464 OUT_CB(util_float_to_half(c.color[0]) | 465 (util_float_to_half(c.color[1]) << 16)); 466 break; 467 468 default: 469 OUT_CB(float_to_fixed10(c.color[0]) | 470 (float_to_fixed10(c.color[3]) << 16)); 471 OUT_CB(float_to_fixed10(c.color[2]) | 472 (float_to_fixed10(c.color[1]) << 16)); 473 } 474 475 END_CB; 476 } else { 477 union util_color uc; 478 util_pack_color(c.color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 479 480 BEGIN_CB(state->cb, 2); 481 OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui); 482 END_CB; 483 } 484 485 r300_mark_atom_dirty(r300, &r300->blend_color_state); 486} 487 488static void r300_set_clip_state(struct pipe_context* pipe, 489 const struct pipe_clip_state* state) 490{ 491 struct r300_context* r300 = r300_context(pipe); 492 struct r300_clip_state *clip = 493 (struct r300_clip_state*)r300->clip_state.state; 494 CB_LOCALS; 495 496 if (r300->screen->caps.has_tcl) { 497 BEGIN_CB(clip->cb, r300->clip_state.size); 498 OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, 499 (r300->screen->caps.is_r500 ? 500 R500_PVS_UCP_START : R300_PVS_UCP_START)); 501 OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); 502 OUT_CB_TABLE(state->ucp, 6 * 4); 503 END_CB; 504 505 r300_mark_atom_dirty(r300, &r300->clip_state); 506 } else { 507 draw_set_clip_state(r300->draw, state); 508 } 509} 510 511static void 512r300_set_sample_mask(struct pipe_context *pipe, 513 unsigned sample_mask) 514{ 515} 516 517 518/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 519 * 520 * This contains the depth buffer, stencil buffer, alpha test, and such. 521 * On the Radeon, depth and stencil buffer setup are intertwined, which is 522 * the reason for some of the strange-looking assignments across registers. */ 523static void* 524 r300_create_dsa_state(struct pipe_context* pipe, 525 const struct pipe_depth_stencil_alpha_state* state) 526{ 527 struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; 528 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 529 CB_LOCALS; 530 531 dsa->dsa = *state; 532 533 /* Depth test setup. - separate write mask depth for decomp flush */ 534 if (state->depth.writemask) { 535 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 536 } 537 538 if (state->depth.enabled) { 539 dsa->z_buffer_control |= R300_Z_ENABLE; 540 541 dsa->z_stencil_control |= 542 (r300_translate_depth_stencil_function(state->depth.func) << 543 R300_Z_FUNC_SHIFT); 544 } else { 545 /* We must enable depth test, otherwise occlusion queries won't work. */ 546 dsa->z_buffer_control |= R300_Z_ENABLE; 547 dsa->z_stencil_control |= R300_ZS_ALWAYS; 548 } 549 550 /* Stencil buffer setup. */ 551 if (state->stencil[0].enabled) { 552 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 553 dsa->z_stencil_control |= 554 (r300_translate_depth_stencil_function(state->stencil[0].func) << 555 R300_S_FRONT_FUNC_SHIFT) | 556 (r300_translate_stencil_op(state->stencil[0].fail_op) << 557 R300_S_FRONT_SFAIL_OP_SHIFT) | 558 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 559 R300_S_FRONT_ZPASS_OP_SHIFT) | 560 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 561 R300_S_FRONT_ZFAIL_OP_SHIFT); 562 563 dsa->stencil_ref_mask = 564 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 565 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 566 567 if (state->stencil[1].enabled) { 568 dsa->two_sided = TRUE; 569 570 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 571 dsa->z_stencil_control |= 572 (r300_translate_depth_stencil_function(state->stencil[1].func) << 573 R300_S_BACK_FUNC_SHIFT) | 574 (r300_translate_stencil_op(state->stencil[1].fail_op) << 575 R300_S_BACK_SFAIL_OP_SHIFT) | 576 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 577 R300_S_BACK_ZPASS_OP_SHIFT) | 578 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 579 R300_S_BACK_ZFAIL_OP_SHIFT); 580 581 dsa->stencil_ref_bf = 582 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | 583 (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); 584 585 if (caps->is_r500) { 586 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 587 } else { 588 dsa->two_sided_stencil_ref = 589 (state->stencil[0].valuemask != state->stencil[1].valuemask || 590 state->stencil[0].writemask != state->stencil[1].writemask); 591 } 592 } 593 } 594 595 /* Alpha test setup. */ 596 if (state->alpha.enabled) { 597 dsa->alpha_function = 598 r300_translate_alpha_function(state->alpha.func) | 599 R300_FG_ALPHA_FUNC_ENABLE; 600 601 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 602 dsa->alpha_value = util_float_to_half(state->alpha.ref_value); 603 604 if (caps->is_r500) { 605 dsa->alpha_function_fp16 = dsa->alpha_function | 606 R500_FG_ALPHA_FUNC_FP16_ENABLE; 607 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 608 } 609 } 610 611 BEGIN_CB(&dsa->cb_begin, 10); 612 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); 613 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 614 OUT_CB(dsa->z_buffer_control); 615 OUT_CB(dsa->z_stencil_control); 616 OUT_CB(dsa->stencil_ref_mask); 617 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); 618 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 619 END_CB; 620 621 BEGIN_CB(&dsa->cb_begin_fp16, 10); 622 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16); 623 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 624 OUT_CB(dsa->z_buffer_control); 625 OUT_CB(dsa->z_stencil_control); 626 OUT_CB(dsa->stencil_ref_mask); 627 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); 628 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 629 END_CB; 630 631 /* We must enable depth test, otherwise occlusion queries won't work. 632 * We setup a dummy zbuffer to silent the CS checker, see emit_fb_state. */ 633 BEGIN_CB(dsa->cb_zb_no_readwrite, 10); 634 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); 635 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 636 OUT_CB(R300_Z_ENABLE); 637 OUT_CB(R300_ZS_ALWAYS); 638 OUT_CB(0); 639 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); 640 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 641 END_CB; 642 643 BEGIN_CB(dsa->cb_fp16_zb_no_readwrite, 10); 644 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16); 645 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 646 OUT_CB(R300_Z_ENABLE); 647 OUT_CB(R300_ZS_ALWAYS); 648 OUT_CB(0); 649 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); 650 OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); 651 END_CB; 652 653 return (void*)dsa; 654} 655 656static void r300_dsa_inject_stencilref(struct r300_context *r300) 657{ 658 struct r300_dsa_state *dsa = 659 (struct r300_dsa_state*)r300->dsa_state.state; 660 661 if (!dsa) 662 return; 663 664 dsa->stencil_ref_mask = 665 (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) | 666 r300->stencil_ref.ref_value[0]; 667 dsa->stencil_ref_bf = 668 (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) | 669 r300->stencil_ref.ref_value[1]; 670} 671 672/* Bind DSA state. */ 673static void r300_bind_dsa_state(struct pipe_context* pipe, 674 void* state) 675{ 676 struct r300_context* r300 = r300_context(pipe); 677 678 if (!state) { 679 return; 680 } 681 682 UPDATE_STATE(state, r300->dsa_state); 683 684 r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */ 685 r300_dsa_inject_stencilref(r300); 686} 687 688/* Free DSA state. */ 689static void r300_delete_dsa_state(struct pipe_context* pipe, 690 void* state) 691{ 692 FREE(state); 693} 694 695static void r300_set_stencil_ref(struct pipe_context* pipe, 696 const struct pipe_stencil_ref* sr) 697{ 698 struct r300_context* r300 = r300_context(pipe); 699 700 r300->stencil_ref = *sr; 701 702 r300_dsa_inject_stencilref(r300); 703 r300_mark_atom_dirty(r300, &r300->dsa_state); 704} 705 706static void r300_tex_set_tiling_flags(struct r300_context *r300, 707 struct r300_resource *tex, 708 unsigned level) 709{ 710 /* Check if the macrotile flag needs to be changed. 711 * Skip changing the flags otherwise. */ 712 if (tex->tex.macrotile[tex->surface_level] != 713 tex->tex.macrotile[level]) { 714 r300->rws->buffer_set_tiling(tex->buf, r300->cs, 715 tex->tex.microtile, tex->tex.macrotile[level], 716 tex->tex.stride_in_bytes[0]); 717 718 tex->surface_level = level; 719 } 720} 721 722/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 723static void r300_fb_set_tiling_flags(struct r300_context *r300, 724 const struct pipe_framebuffer_state *state) 725{ 726 unsigned i; 727 728 /* Set tiling flags for new surfaces. */ 729 for (i = 0; i < state->nr_cbufs; i++) { 730 r300_tex_set_tiling_flags(r300, 731 r300_resource(state->cbufs[i]->texture), 732 state->cbufs[i]->u.tex.level); 733 } 734 if (state->zsbuf) { 735 r300_tex_set_tiling_flags(r300, 736 r300_resource(state->zsbuf->texture), 737 state->zsbuf->u.tex.level); 738 } 739} 740 741static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, 742 const char *binding) 743{ 744 struct pipe_resource *tex = surf->texture; 745 struct r300_resource *rtex = r300_resource(tex); 746 747 fprintf(stderr, 748 "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, " 749 "Lastlayer: %i, Level: %i, Format: %s\n" 750 751 "r300: TEX: Macro: %s, Micro: %s, " 752 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", 753 754 binding, index, surf->width, surf->height, 755 surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level, 756 util_format_short_name(surf->format), 757 758 rtex->tex.macrotile[0] ? "YES" : " NO", 759 rtex->tex.microtile ? "YES" : " NO", 760 tex->width0, tex->height0, tex->depth0, 761 tex->last_level, util_format_short_name(surf->format)); 762} 763 764void r300_mark_fb_state_dirty(struct r300_context *r300, 765 enum r300_fb_state_change change) 766{ 767 struct pipe_framebuffer_state *state = r300->fb_state.state; 768 769 r300_mark_atom_dirty(r300, &r300->gpu_flush); 770 r300_mark_atom_dirty(r300, &r300->fb_state); 771 772 /* What is marked as dirty depends on the enum r300_fb_state_change. */ 773 if (change == R300_CHANGED_FB_STATE) { 774 r300_mark_atom_dirty(r300, &r300->aa_state); 775 r300_mark_atom_dirty(r300, &r300->dsa_state); /* for AlphaRef */ 776 r300_set_blend_color(&r300->context, r300->blend_color_state.state); 777 } 778 779 if (change == R300_CHANGED_FB_STATE || 780 change == R300_CHANGED_HYPERZ_FLAG) { 781 r300_mark_atom_dirty(r300, &r300->hyperz_state); 782 } 783 784 if (change == R300_CHANGED_FB_STATE || 785 change == R300_CHANGED_MULTIWRITE) { 786 r300_mark_atom_dirty(r300, &r300->fb_state_pipelined); 787 } 788 789 /* Now compute the fb_state atom size. */ 790 r300->fb_state.size = 2 + (8 * state->nr_cbufs); 791 792 if (r300->cbzb_clear) { 793 r300->fb_state.size += 10; 794 } else if (state->zsbuf) { 795 r300->fb_state.size += 10; 796 if (r300->hyperz_enabled) 797 r300->fb_state.size += 8; 798 } else if (state->nr_cbufs) { 799 r300->fb_state.size += 10; 800 } 801 802 /* The size of the rest of atoms stays the same. */ 803} 804 805static void 806r300_set_framebuffer_state(struct pipe_context* pipe, 807 const struct pipe_framebuffer_state* state) 808{ 809 struct r300_context* r300 = r300_context(pipe); 810 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 811 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 812 unsigned max_width, max_height, i; 813 uint32_t zbuffer_bpp = 0; 814 boolean unlock_zbuffer = FALSE; 815 816 if (r300->screen->caps.is_r500) { 817 max_width = max_height = 4096; 818 } else if (r300->screen->caps.is_r400) { 819 max_width = max_height = 4021; 820 } else { 821 max_width = max_height = 2560; 822 } 823 824 if (state->width > max_width || state->height > max_height) { 825 fprintf(stderr, "r300: Implementation error: Render targets are too " 826 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 827 return; 828 } 829 830 if (old_state->zsbuf && r300->zmask_in_use && !r300->locked_zbuffer) { 831 /* There is a zmask in use, what are we gonna do? */ 832 if (state->zsbuf) { 833 if (!pipe_surface_equal(old_state->zsbuf, state->zsbuf)) { 834 /* Decompress the currently bound zbuffer before we bind another one. */ 835 r300_decompress_zmask(r300); 836 r300->hiz_in_use = FALSE; 837 } 838 } else { 839 /* We don't bind another zbuffer, so lock the current one. */ 840 pipe_surface_reference(&r300->locked_zbuffer, old_state->zsbuf); 841 } 842 } else if (r300->locked_zbuffer) { 843 /* We have a locked zbuffer now, what are we gonna do? */ 844 if (state->zsbuf) { 845 if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) { 846 /* We are binding some other zbuffer, so decompress the locked one, 847 * it gets unlocked automatically. */ 848 r300_decompress_zmask_locked_unsafe(r300); 849 r300->hiz_in_use = FALSE; 850 } else { 851 /* We are binding the locked zbuffer again, so unlock it. */ 852 unlock_zbuffer = TRUE; 853 } 854 } 855 } 856 assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use); 857 858 /* Need to reset clamping or colormask. */ 859 r300_mark_atom_dirty(r300, &r300->blend_state); 860 861 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 862 if (!!old_state->zsbuf != !!state->zsbuf) { 863 r300_mark_atom_dirty(r300, &r300->dsa_state); 864 } 865 866 if (r300->screen->info.drm_minor < 12) { 867 /* The tiling flags are dependent on the surface miplevel, unfortunately. 868 * This workarounds a bad design decision in old kernels which were 869 * rewriting tile fields in registers. */ 870 r300_fb_set_tiling_flags(r300, state); 871 } 872 873 util_copy_framebuffer_state(r300->fb_state.state, state); 874 875 if (unlock_zbuffer) { 876 pipe_surface_reference(&r300->locked_zbuffer, NULL); 877 } 878 879 r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); 880 881 if (state->zsbuf) { 882 switch (util_format_get_blocksize(state->zsbuf->format)) { 883 case 2: 884 zbuffer_bpp = 16; 885 break; 886 case 4: 887 zbuffer_bpp = 24; 888 break; 889 } 890 891 /* Polygon offset depends on the zbuffer bit depth. */ 892 if (r300->zbuffer_bpp != zbuffer_bpp) { 893 r300->zbuffer_bpp = zbuffer_bpp; 894 895 if (r300->polygon_offset_enabled) 896 r300_mark_atom_dirty(r300, &r300->rs_state); 897 } 898 } 899 900 /* Set up AA config. */ 901 if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) { 902 aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE; 903 904 switch (state->cbufs[0]->texture->nr_samples) { 905 case 2: 906 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; 907 break; 908 case 3: 909 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3; 910 break; 911 case 4: 912 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; 913 break; 914 case 6: 915 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; 916 break; 917 } 918 } else { 919 aa->aa_config = 0; 920 } 921 922 if (DBG_ON(r300, DBG_FB)) { 923 fprintf(stderr, "r300: set_framebuffer_state:\n"); 924 for (i = 0; i < state->nr_cbufs; i++) { 925 r300_print_fb_surf_info(state->cbufs[i], i, "CB"); 926 } 927 if (state->zsbuf) { 928 r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); 929 } 930 } 931} 932 933/* Create fragment shader state. */ 934static void* r300_create_fs_state(struct pipe_context* pipe, 935 const struct pipe_shader_state* shader) 936{ 937 struct r300_fragment_shader* fs = NULL; 938 939 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 940 941 /* Copy state directly into shader. */ 942 fs->state = *shader; 943 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 944 945 return (void*)fs; 946} 947 948void r300_mark_fs_code_dirty(struct r300_context *r300) 949{ 950 struct r300_fragment_shader* fs = r300_fs(r300); 951 952 r300_mark_atom_dirty(r300, &r300->fs); 953 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 954 r300_mark_atom_dirty(r300, &r300->fs_constants); 955 r300->fs.size = fs->shader->cb_code_size; 956 957 if (r300->screen->caps.is_r500) { 958 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; 959 r300->fs_constants.size = fs->shader->externals_count * 4 + 3; 960 } else { 961 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; 962 r300->fs_constants.size = fs->shader->externals_count * 4 + 1; 963 } 964 965 ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = 966 fs->shader->code.constants_remap_table; 967} 968 969/* Bind fragment shader state. */ 970static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 971{ 972 struct r300_context* r300 = r300_context(pipe); 973 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 974 975 if (fs == NULL) { 976 r300->fs.state = NULL; 977 return; 978 } 979 980 r300->fs.state = fs; 981 r300->fs_status = FRAGMENT_SHADER_DIRTY; 982 983 r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ 984} 985 986/* Delete fragment shader state. */ 987static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 988{ 989 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 990 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 991 992 while (ptr) { 993 tmp = ptr; 994 ptr = ptr->next; 995 rc_constants_destroy(&tmp->code.constants); 996 FREE(tmp->cb_code); 997 FREE(tmp); 998 } 999 FREE((void*)fs->state.tokens); 1000 FREE(shader); 1001} 1002 1003static void r300_set_polygon_stipple(struct pipe_context* pipe, 1004 const struct pipe_poly_stipple* state) 1005{ 1006 /* XXX no idea how to set this up, but not terribly important */ 1007} 1008 1009/* Create a new rasterizer state based on the CSO rasterizer state. 1010 * 1011 * This is a very large chunk of state, and covers most of the graphics 1012 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 1013 * 1014 * In a not entirely unironic sidenote, this state has nearly nothing to do 1015 * with the actual block on the Radeon called the rasterizer (RS). */ 1016static void* r300_create_rs_state(struct pipe_context* pipe, 1017 const struct pipe_rasterizer_state* state) 1018{ 1019 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 1020 float psiz; 1021 uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ 1022 uint32_t vap_clip_cntl; /* R300_VAP_CLIP_CNTL: 0x221C */ 1023 uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ 1024 uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ 1025 uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ 1026 uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */ 1027 uint32_t cull_mode; /* R300_SU_CULL_MODE: 0x42b8 */ 1028 uint32_t line_stipple_config; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */ 1029 uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ 1030 uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ 1031 uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ 1032 uint32_t round_mode; /* R300_GA_ROUND_MODE: 0x428c */ 1033 1034 /* Point sprites texture coordinates, 0: lower left, 1: upper right */ 1035 float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ 1036 float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ 1037 float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ 1038 float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ 1039 boolean vclamp = state->clamp_vertex_color || 1040 !r300_context(pipe)->screen->caps.is_r500; 1041 CB_LOCALS; 1042 1043 /* Copy rasterizer state. */ 1044 rs->rs = *state; 1045 rs->rs_draw = *state; 1046 1047 rs->rs.sprite_coord_enable = state->point_quad_rasterization * 1048 state->sprite_coord_enable; 1049 1050 /* Override some states for Draw. */ 1051 rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ 1052 1053#ifdef PIPE_ARCH_LITTLE_ENDIAN 1054 vap_control_status = R300_VC_NO_SWAP; 1055#else 1056 vap_control_status = R300_VC_32BIT_SWAP; 1057#endif 1058 1059 /* If no TCL engine is present, turn off the HW TCL. */ 1060 if (!r300_screen(pipe->screen)->caps.has_tcl) { 1061 vap_control_status |= R300_VAP_TCL_BYPASS; 1062 } 1063 1064 /* Point size width and height. */ 1065 point_size = 1066 pack_float_16_6x(state->point_size) | 1067 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 1068 1069 /* Point size clamping. */ 1070 if (state->point_size_per_vertex) { 1071 /* Per-vertex point size. 1072 * Clamp to [0, max FB size] */ 1073 psiz = pipe->screen->get_paramf(pipe->screen, 1074 PIPE_CAPF_MAX_POINT_WIDTH); 1075 point_minmax = 1076 pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; 1077 } else { 1078 /* We cannot disable the point-size vertex output, 1079 * so clamp it. */ 1080 psiz = state->point_size; 1081 point_minmax = 1082 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 1083 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 1084 } 1085 1086 /* Line control. */ 1087 line_control = pack_float_16_6x(state->line_width) | 1088 R300_GA_LINE_CNTL_END_TYPE_COMP; 1089 1090 /* Enable polygon mode */ 1091 polygon_mode = 0; 1092 if (state->fill_front != PIPE_POLYGON_MODE_FILL || 1093 state->fill_back != PIPE_POLYGON_MODE_FILL) { 1094 polygon_mode = R300_GA_POLY_MODE_DUAL; 1095 } 1096 1097 /* Front face */ 1098 if (state->front_ccw) 1099 cull_mode = R300_FRONT_FACE_CCW; 1100 else 1101 cull_mode = R300_FRONT_FACE_CW; 1102 1103 /* Polygon offset */ 1104 polygon_offset_enable = 0; 1105 if (util_get_offset(state, state->fill_front)) { 1106 polygon_offset_enable |= R300_FRONT_ENABLE; 1107 } 1108 if (util_get_offset(state, state->fill_back)) { 1109 polygon_offset_enable |= R300_BACK_ENABLE; 1110 } 1111 1112 rs->polygon_offset_enable = polygon_offset_enable != 0; 1113 1114 /* Polygon mode */ 1115 if (polygon_mode) { 1116 polygon_mode |= 1117 r300_translate_polygon_mode_front(state->fill_front); 1118 polygon_mode |= 1119 r300_translate_polygon_mode_back(state->fill_back); 1120 } 1121 1122 if (state->cull_face & PIPE_FACE_FRONT) { 1123 cull_mode |= R300_CULL_FRONT; 1124 } 1125 if (state->cull_face & PIPE_FACE_BACK) { 1126 cull_mode |= R300_CULL_BACK; 1127 } 1128 1129 if (state->line_stipple_enable) { 1130 line_stipple_config = 1131 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 1132 (fui((float)state->line_stipple_factor) & 1133 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 1134 /* XXX this might need to be scaled up */ 1135 line_stipple_value = state->line_stipple_pattern; 1136 } else { 1137 line_stipple_config = 0; 1138 line_stipple_value = 0; 1139 } 1140 1141 if (state->flatshade) { 1142 rs->color_control = R300_SHADE_MODEL_FLAT; 1143 } else { 1144 rs->color_control = R300_SHADE_MODEL_SMOOTH; 1145 } 1146 1147 clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 1148 1149 /* Point sprites coord mode */ 1150 if (rs->rs.sprite_coord_enable) { 1151 switch (state->sprite_coord_mode) { 1152 case PIPE_SPRITE_COORD_UPPER_LEFT: 1153 point_texcoord_top = 0.0f; 1154 point_texcoord_bottom = 1.0f; 1155 break; 1156 case PIPE_SPRITE_COORD_LOWER_LEFT: 1157 point_texcoord_top = 1.0f; 1158 point_texcoord_bottom = 0.0f; 1159 break; 1160 } 1161 } 1162 1163 if (r300_screen(pipe->screen)->caps.has_tcl) { 1164 vap_clip_cntl = (state->clip_plane_enable & 63) | 1165 R300_PS_UCP_MODE_CLIP_AS_TRIFAN; 1166 } else { 1167 vap_clip_cntl = R300_CLIP_DISABLE; 1168 } 1169 1170 /* Vertex color clamping. FP20 means no clamping. */ 1171 round_mode = 1172 R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST | 1173 (!vclamp ? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20 | 1174 R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20) : 0); 1175 1176 /* Build the main command buffer. */ 1177 BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); 1178 OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); 1179 OUT_CB_REG(R300_VAP_CLIP_CNTL, vap_clip_cntl); 1180 OUT_CB_REG(R300_GA_POINT_SIZE, point_size); 1181 OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2); 1182 OUT_CB(point_minmax); 1183 OUT_CB(line_control); 1184 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2); 1185 OUT_CB(polygon_offset_enable); 1186 rs->cull_mode_index = 11; 1187 OUT_CB(cull_mode); 1188 OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config); 1189 OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); 1190 OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); 1191 OUT_CB_REG(R300_GA_ROUND_MODE, round_mode); 1192 OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); 1193 OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); 1194 OUT_CB_32F(point_texcoord_left); 1195 OUT_CB_32F(point_texcoord_bottom); 1196 OUT_CB_32F(point_texcoord_right); 1197 OUT_CB_32F(point_texcoord_top); 1198 END_CB; 1199 1200 /* Build the two command buffers for polygon offset setup. */ 1201 if (polygon_offset_enable) { 1202 float scale = state->offset_scale * 12; 1203 float offset = state->offset_units * 4; 1204 1205 BEGIN_CB(rs->cb_poly_offset_zb16, 5); 1206 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); 1207 OUT_CB_32F(scale); 1208 OUT_CB_32F(offset); 1209 OUT_CB_32F(scale); 1210 OUT_CB_32F(offset); 1211 END_CB; 1212 1213 offset = state->offset_units * 2; 1214 1215 BEGIN_CB(rs->cb_poly_offset_zb24, 5); 1216 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); 1217 OUT_CB_32F(scale); 1218 OUT_CB_32F(offset); 1219 OUT_CB_32F(scale); 1220 OUT_CB_32F(offset); 1221 END_CB; 1222 } 1223 1224 return (void*)rs; 1225} 1226 1227/* Bind rasterizer state. */ 1228static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 1229{ 1230 struct r300_context* r300 = r300_context(pipe); 1231 struct r300_rs_state* rs = (struct r300_rs_state*)state; 1232 int last_sprite_coord_enable = r300->sprite_coord_enable; 1233 boolean last_two_sided_color = r300->two_sided_color; 1234 boolean last_frag_clamp = r300->frag_clamp; 1235 1236 if (r300->draw && rs) { 1237 draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); 1238 } 1239 1240 if (rs) { 1241 r300->polygon_offset_enabled = rs->polygon_offset_enable; 1242 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 1243 r300->two_sided_color = rs->rs.light_twoside; 1244 r300->frag_clamp = rs->rs.clamp_fragment_color; 1245 } else { 1246 r300->polygon_offset_enabled = FALSE; 1247 r300->sprite_coord_enable = 0; 1248 r300->two_sided_color = FALSE; 1249 r300->frag_clamp = FALSE; 1250 } 1251 1252 UPDATE_STATE(state, r300->rs_state); 1253 r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0); 1254 1255 if (last_sprite_coord_enable != r300->sprite_coord_enable || 1256 last_two_sided_color != r300->two_sided_color) { 1257 r300_mark_atom_dirty(r300, &r300->rs_block_state); 1258 } 1259 1260 if (last_frag_clamp != r300->frag_clamp && 1261 r300->fs_status == FRAGMENT_SHADER_VALID) { 1262 r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; 1263 } 1264} 1265 1266/* Free rasterizer state. */ 1267static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 1268{ 1269 FREE(state); 1270} 1271 1272static void* 1273 r300_create_sampler_state(struct pipe_context* pipe, 1274 const struct pipe_sampler_state* state) 1275{ 1276 struct r300_context* r300 = r300_context(pipe); 1277 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 1278 boolean is_r500 = r300->screen->caps.is_r500; 1279 int lod_bias; 1280 1281 sampler->state = *state; 1282 1283 /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG 1284 * or MIN filter is NEAREST. Since texwrap produces same results 1285 * for CLAMP and CLAMP_TO_EDGE, we use them instead. */ 1286 if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST || 1287 sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) { 1288 /* Wrap S. */ 1289 if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP) 1290 sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1291 else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP) 1292 sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1293 1294 /* Wrap T. */ 1295 if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP) 1296 sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1297 else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP) 1298 sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1299 1300 /* Wrap R. */ 1301 if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP) 1302 sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1303 else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP) 1304 sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1305 } 1306 1307 sampler->filter0 |= 1308 (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) | 1309 (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) | 1310 (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT); 1311 1312 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 1313 state->mag_img_filter, 1314 state->min_mip_filter, 1315 state->max_anisotropy > 1); 1316 1317 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 1318 1319 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 1320 /* We must pass these to the merge function to clamp them properly. */ 1321 sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); 1322 sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); 1323 1324 lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); 1325 1326 sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK; 1327 1328 /* This is very high quality anisotropic filtering for R5xx. 1329 * It's good for benchmarking the performance of texturing but 1330 * in practice we don't want to slow down the driver because it's 1331 * a pretty good performance killer. Feel free to play with it. */ 1332 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 1333 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 1334 } 1335 1336 /* R500-specific fixups and optimizations */ 1337 if (r300->screen->caps.is_r500) { 1338 sampler->filter1 |= R500_BORDER_FIX; 1339 } 1340 1341 return (void*)sampler; 1342} 1343 1344static void r300_bind_sampler_states(struct pipe_context* pipe, 1345 unsigned count, 1346 void** states) 1347{ 1348 struct r300_context* r300 = r300_context(pipe); 1349 struct r300_textures_state* state = 1350 (struct r300_textures_state*)r300->textures_state.state; 1351 unsigned tex_units = r300->screen->caps.num_tex_units; 1352 1353 if (count > tex_units) { 1354 return; 1355 } 1356 1357 memcpy(state->sampler_states, states, sizeof(void*) * count); 1358 state->sampler_state_count = count; 1359 1360 r300_mark_atom_dirty(r300, &r300->textures_state); 1361} 1362 1363static void r300_lacks_vertex_textures(struct pipe_context* pipe, 1364 unsigned count, 1365 void** states) 1366{ 1367} 1368 1369static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1370{ 1371 FREE(state); 1372} 1373 1374static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num) 1375{ 1376 /* This looks like a hack, but I believe it's suppose to work like 1377 * that. To illustrate how this works, let's assume you have 5 textures. 1378 * From docs, 5 and the successive numbers are: 1379 * 1380 * FOURTH_1 = 5 1381 * FOURTH_2 = 6 1382 * FOURTH_3 = 7 1383 * EIGHTH_0 = 8 1384 * EIGHTH_1 = 9 1385 * 1386 * First 3 textures will get 3/4 of size of the cache, divived evenly 1387 * between them. The last 1/4 of the cache must be divided between 1388 * the last 2 textures, each will therefore get 1/8 of the cache. 1389 * Why not just to use "5 + texture_index" ? 1390 * 1391 * This simple trick works for all "num" <= 16. 1392 */ 1393 if (num <= 1) 1394 return R300_TX_CACHE(R300_TX_CACHE_WHOLE); 1395 else 1396 return R300_TX_CACHE(num + index); 1397} 1398 1399static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 1400 unsigned count, 1401 struct pipe_sampler_view** views) 1402{ 1403 struct r300_context* r300 = r300_context(pipe); 1404 struct r300_textures_state* state = 1405 (struct r300_textures_state*)r300->textures_state.state; 1406 struct r300_resource *texture; 1407 unsigned i, real_num_views = 0, view_index = 0; 1408 unsigned tex_units = r300->screen->caps.num_tex_units; 1409 boolean dirty_tex = FALSE; 1410 1411 if (count > tex_units) { 1412 return; 1413 } 1414 1415 /* Calculate the real number of views. */ 1416 for (i = 0; i < count; i++) { 1417 if (views[i]) 1418 real_num_views++; 1419 } 1420 1421 for (i = 0; i < count; i++) { 1422 pipe_sampler_view_reference( 1423 (struct pipe_sampler_view**)&state->sampler_views[i], 1424 views[i]); 1425 1426 if (!views[i]) { 1427 continue; 1428 } 1429 1430 /* A new sampler view (= texture)... */ 1431 dirty_tex = TRUE; 1432 1433 /* Set the texrect factor in the fragment shader. 1434 * Needed for RECT and NPOT fallback. */ 1435 texture = r300_resource(views[i]->texture); 1436 if (texture->tex.is_npot) { 1437 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1438 } 1439 1440 state->sampler_views[i]->texcache_region = 1441 r300_assign_texture_cache_region(view_index, real_num_views); 1442 view_index++; 1443 } 1444 1445 for (i = count; i < tex_units; i++) { 1446 if (state->sampler_views[i]) { 1447 pipe_sampler_view_reference( 1448 (struct pipe_sampler_view**)&state->sampler_views[i], 1449 NULL); 1450 } 1451 } 1452 1453 state->sampler_view_count = count; 1454 1455 r300_mark_atom_dirty(r300, &r300->textures_state); 1456 1457 if (dirty_tex) { 1458 r300_mark_atom_dirty(r300, &r300->texture_cache_inval); 1459 } 1460} 1461 1462struct pipe_sampler_view * 1463r300_create_sampler_view_custom(struct pipe_context *pipe, 1464 struct pipe_resource *texture, 1465 const struct pipe_sampler_view *templ, 1466 unsigned width0_override, 1467 unsigned height0_override) 1468{ 1469 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1470 struct r300_resource *tex = r300_resource(texture); 1471 boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; 1472 boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle; 1473 1474 if (view) { 1475 unsigned hwformat; 1476 1477 view->base = *templ; 1478 view->base.reference.count = 1; 1479 view->base.context = pipe; 1480 view->base.texture = NULL; 1481 pipe_resource_reference(&view->base.texture, texture); 1482 1483 view->width0_override = width0_override; 1484 view->height0_override = height0_override; 1485 view->swizzle[0] = templ->swizzle_r; 1486 view->swizzle[1] = templ->swizzle_g; 1487 view->swizzle[2] = templ->swizzle_b; 1488 view->swizzle[3] = templ->swizzle_a; 1489 1490 hwformat = r300_translate_texformat(templ->format, 1491 view->swizzle, 1492 is_r500, 1493 dxtc_swizzle); 1494 1495 if (hwformat == ~0) { 1496 fprintf(stderr, "r300: Ooops. Got unsupported format %s in %s.\n", 1497 util_format_short_name(templ->format), __func__); 1498 } 1499 assert(hwformat != ~0); 1500 1501 r300_texture_setup_format_state(r300_screen(pipe->screen), tex, 1502 templ->format, 0, 1503 width0_override, height0_override, 1504 &view->format); 1505 view->format.format1 |= hwformat; 1506 if (is_r500) { 1507 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1508 } 1509 } 1510 1511 return (struct pipe_sampler_view*)view; 1512} 1513 1514static struct pipe_sampler_view * 1515r300_create_sampler_view(struct pipe_context *pipe, 1516 struct pipe_resource *texture, 1517 const struct pipe_sampler_view *templ) 1518{ 1519 return r300_create_sampler_view_custom(pipe, texture, templ, 1520 r300_resource(texture)->tex.width0, 1521 r300_resource(texture)->tex.height0); 1522} 1523 1524 1525static void 1526r300_sampler_view_destroy(struct pipe_context *pipe, 1527 struct pipe_sampler_view *view) 1528{ 1529 pipe_resource_reference(&view->texture, NULL); 1530 FREE(view); 1531} 1532 1533static void r300_set_scissor_state(struct pipe_context* pipe, 1534 const struct pipe_scissor_state* state) 1535{ 1536 struct r300_context* r300 = r300_context(pipe); 1537 1538 memcpy(r300->scissor_state.state, state, 1539 sizeof(struct pipe_scissor_state)); 1540 1541 r300_mark_atom_dirty(r300, &r300->scissor_state); 1542} 1543 1544static void r300_set_viewport_state(struct pipe_context* pipe, 1545 const struct pipe_viewport_state* state) 1546{ 1547 struct r300_context* r300 = r300_context(pipe); 1548 struct r300_viewport_state* viewport = 1549 (struct r300_viewport_state*)r300->viewport_state.state; 1550 1551 r300->viewport = *state; 1552 1553 if (r300->draw) { 1554 draw_set_viewport_state(r300->draw, state); 1555 viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; 1556 return; 1557 } 1558 1559 /* Do the transform in HW. */ 1560 viewport->vte_control = R300_VTX_W0_FMT; 1561 1562 if (state->scale[0] != 1.0f) { 1563 viewport->xscale = state->scale[0]; 1564 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1565 } 1566 if (state->scale[1] != 1.0f) { 1567 viewport->yscale = state->scale[1]; 1568 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1569 } 1570 if (state->scale[2] != 1.0f) { 1571 viewport->zscale = state->scale[2]; 1572 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1573 } 1574 if (state->translate[0] != 0.0f) { 1575 viewport->xoffset = state->translate[0]; 1576 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1577 } 1578 if (state->translate[1] != 0.0f) { 1579 viewport->yoffset = state->translate[1]; 1580 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1581 } 1582 if (state->translate[2] != 0.0f) { 1583 viewport->zoffset = state->translate[2]; 1584 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1585 } 1586 1587 r300_mark_atom_dirty(r300, &r300->viewport_state); 1588 if (r300->fs.state && r300_fs(r300)->shader && 1589 r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1590 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1591 } 1592} 1593 1594static void r300_set_vertex_buffers(struct pipe_context* pipe, 1595 unsigned count, 1596 const struct pipe_vertex_buffer* buffers) 1597{ 1598 struct r300_context* r300 = r300_context(pipe); 1599 unsigned i; 1600 struct pipe_vertex_buffer dummy_vb = {0}; 1601 1602 /* There must be at least one vertex buffer set, otherwise it locks up. */ 1603 if (!count) { 1604 dummy_vb.buffer = r300->dummy_vb; 1605 buffers = &dummy_vb; 1606 count = 1; 1607 } 1608 1609 u_vbuf_set_vertex_buffers(r300->vbuf_mgr, count, buffers); 1610 1611 if (r300->screen->caps.has_tcl) { 1612 /* HW TCL. */ 1613 for (i = 0; i < count; i++) { 1614 if (buffers[i].buffer && 1615 !r300_resource(buffers[i].buffer)->b.user_ptr) { 1616 } 1617 } 1618 r300->vertex_arrays_dirty = TRUE; 1619 } else { 1620 /* SW TCL. */ 1621 draw_set_vertex_buffers(r300->draw, count, buffers); 1622 } 1623} 1624 1625static void r300_set_index_buffer(struct pipe_context* pipe, 1626 const struct pipe_index_buffer *ib) 1627{ 1628 struct r300_context* r300 = r300_context(pipe); 1629 1630 u_vbuf_set_index_buffer(r300->vbuf_mgr, ib); 1631 1632 if (!r300->screen->caps.has_tcl) { 1633 draw_set_index_buffer(r300->draw, ib); 1634 } 1635} 1636 1637/* Initialize the PSC tables. */ 1638static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1639{ 1640 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1641 uint16_t type, swizzle; 1642 enum pipe_format format; 1643 unsigned i; 1644 1645 /* Vertex shaders have no semantics on their inputs, 1646 * so PSC should just route stuff based on the vertex elements, 1647 * and not on attrib information. */ 1648 for (i = 0; i < velems->count; i++) { 1649 format = velems->velem[i].src_format; 1650 1651 type = r300_translate_vertex_data_type(format); 1652 if (type == R300_INVALID_FORMAT) { 1653 fprintf(stderr, "r300: Bad vertex format %s.\n", 1654 util_format_short_name(format)); 1655 assert(0); 1656 abort(); 1657 } 1658 1659 type |= i << R300_DST_VEC_LOC_SHIFT; 1660 swizzle = r300_translate_vertex_data_swizzle(format); 1661 1662 if (i & 1) { 1663 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1664 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1665 } else { 1666 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1667 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1668 } 1669 } 1670 1671 /* Set the last vector in the PSC. */ 1672 if (i) { 1673 i -= 1; 1674 } 1675 vstream->vap_prog_stream_cntl[i >> 1] |= 1676 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1677 1678 vstream->count = (i >> 1) + 1; 1679} 1680 1681static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1682 unsigned count, 1683 const struct pipe_vertex_element* attribs) 1684{ 1685 struct r300_context *r300 = r300_context(pipe); 1686 struct r300_vertex_element_state *velems; 1687 unsigned i; 1688 struct pipe_vertex_element dummy_attrib = {0}; 1689 1690 /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */ 1691 if (!count) { 1692 dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM; 1693 attribs = &dummy_attrib; 1694 count = 1; 1695 } else if (count > 16) { 1696 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1697 " requested %i, using 16.\n", count); 1698 count = 16; 1699 } 1700 1701 velems = CALLOC_STRUCT(r300_vertex_element_state); 1702 if (!velems) 1703 return NULL; 1704 1705 velems->count = count; 1706 velems->vmgr_elements = 1707 u_vbuf_create_vertex_elements(r300->vbuf_mgr, count, attribs, 1708 velems->velem); 1709 1710 if (r300_screen(pipe->screen)->caps.has_tcl) { 1711 /* Setup PSC. 1712 * The unused components will be replaced by (..., 0, 1). */ 1713 r300_vertex_psc(velems); 1714 1715 for (i = 0; i < count; i++) { 1716 velems->format_size[i] = 1717 align(util_format_get_blocksize(velems->velem[i].src_format), 4); 1718 velems->vertex_size_dwords += velems->format_size[i] / 4; 1719 } 1720 } 1721 1722 return velems; 1723} 1724 1725static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1726 void *state) 1727{ 1728 struct r300_context *r300 = r300_context(pipe); 1729 struct r300_vertex_element_state *velems = state; 1730 1731 if (velems == NULL) { 1732 return; 1733 } 1734 1735 r300->velems = velems; 1736 1737 u_vbuf_bind_vertex_elements(r300->vbuf_mgr, state, velems->vmgr_elements); 1738 1739 if (r300->draw) { 1740 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1741 return; 1742 } 1743 1744 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1745 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1746 r300->vertex_arrays_dirty = TRUE; 1747} 1748 1749static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1750{ 1751 struct r300_context *r300 = r300_context(pipe); 1752 struct r300_vertex_element_state *velems = state; 1753 1754 u_vbuf_destroy_vertex_elements(r300->vbuf_mgr, velems->vmgr_elements); 1755 FREE(state); 1756} 1757 1758static void* r300_create_vs_state(struct pipe_context* pipe, 1759 const struct pipe_shader_state* shader) 1760{ 1761 struct r300_context* r300 = r300_context(pipe); 1762 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1763 1764 /* Copy state directly into shader. */ 1765 vs->state = *shader; 1766 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1767 1768 if (r300->screen->caps.has_tcl) { 1769 r300_init_vs_outputs(vs); 1770 r300_translate_vertex_shader(r300, vs); 1771 } else { 1772 r300_draw_init_vertex_shader(r300->draw, vs); 1773 } 1774 1775 return vs; 1776} 1777 1778static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1779{ 1780 struct r300_context* r300 = r300_context(pipe); 1781 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1782 1783 if (vs == NULL) { 1784 r300->vs_state.state = NULL; 1785 return; 1786 } 1787 if (vs == r300->vs_state.state) { 1788 return; 1789 } 1790 r300->vs_state.state = vs; 1791 1792 /* The majority of the RS block bits is dependent on the vertex shader. */ 1793 r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ 1794 1795 if (r300->screen->caps.has_tcl) { 1796 unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2; 1797 r300_mark_atom_dirty(r300, &r300->vs_state); 1798 r300->vs_state.size = 1799 vs->code.length + 9 + 1800 (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0); 1801 1802 r300_mark_atom_dirty(r300, &r300->vs_constants); 1803 r300->vs_constants.size = 1804 2 + 1805 (vs->externals_count ? vs->externals_count * 4 + 3 : 0) + 1806 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1807 1808 ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = 1809 vs->code.constants_remap_table; 1810 1811 r300_mark_atom_dirty(r300, &r300->pvs_flush); 1812 } else { 1813 draw_bind_vertex_shader(r300->draw, 1814 (struct draw_vertex_shader*)vs->draw_vs); 1815 } 1816} 1817 1818static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1819{ 1820 struct r300_context* r300 = r300_context(pipe); 1821 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1822 1823 if (r300->screen->caps.has_tcl) { 1824 rc_constants_destroy(&vs->code.constants); 1825 if (vs->code.constants_remap_table) 1826 FREE(vs->code.constants_remap_table); 1827 } else { 1828 draw_delete_vertex_shader(r300->draw, 1829 (struct draw_vertex_shader*)vs->draw_vs); 1830 } 1831 1832 FREE((void*)vs->state.tokens); 1833 FREE(shader); 1834} 1835 1836static void r300_set_constant_buffer(struct pipe_context *pipe, 1837 uint shader, uint index, 1838 struct pipe_resource *buf) 1839{ 1840 struct r300_context* r300 = r300_context(pipe); 1841 struct r300_constant_buffer *cbuf; 1842 struct r300_resource *rbuf = r300_resource(buf); 1843 uint32_t *mapped; 1844 1845 switch (shader) { 1846 case PIPE_SHADER_VERTEX: 1847 cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; 1848 break; 1849 case PIPE_SHADER_FRAGMENT: 1850 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1851 break; 1852 default: 1853 return; 1854 } 1855 1856 if (buf == NULL || buf->width0 == 0) 1857 return; 1858 1859 if (rbuf->b.user_ptr) 1860 mapped = (uint32_t*)rbuf->b.user_ptr; 1861 else if (rbuf->constant_buffer) 1862 mapped = (uint32_t*)rbuf->constant_buffer; 1863 else 1864 return; 1865 1866 if (shader == PIPE_SHADER_FRAGMENT || 1867 (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { 1868 cbuf->ptr = mapped; 1869 } 1870 1871 if (shader == PIPE_SHADER_VERTEX) { 1872 if (r300->screen->caps.has_tcl) { 1873 struct r300_vertex_shader *vs = 1874 (struct r300_vertex_shader*)r300->vs_state.state; 1875 1876 if (!vs) { 1877 cbuf->buffer_base = 0; 1878 return; 1879 } 1880 1881 cbuf->buffer_base = r300->vs_const_base; 1882 r300->vs_const_base += vs->code.constants.Count; 1883 if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) { 1884 r300->vs_const_base = vs->code.constants.Count; 1885 cbuf->buffer_base = 0; 1886 r300_mark_atom_dirty(r300, &r300->pvs_flush); 1887 } 1888 r300_mark_atom_dirty(r300, &r300->vs_constants); 1889 } else if (r300->draw) { 1890 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1891 0, mapped, buf->width0); 1892 } 1893 } else if (shader == PIPE_SHADER_FRAGMENT) { 1894 r300_mark_atom_dirty(r300, &r300->fs_constants); 1895 } 1896} 1897 1898static void r300_texture_barrier(struct pipe_context *pipe) 1899{ 1900 struct r300_context *r300 = r300_context(pipe); 1901 1902 r300_mark_atom_dirty(r300, &r300->gpu_flush); 1903 r300_mark_atom_dirty(r300, &r300->texture_cache_inval); 1904} 1905 1906void r300_init_state_functions(struct r300_context* r300) 1907{ 1908 r300->context.create_blend_state = r300_create_blend_state; 1909 r300->context.bind_blend_state = r300_bind_blend_state; 1910 r300->context.delete_blend_state = r300_delete_blend_state; 1911 1912 r300->context.set_blend_color = r300_set_blend_color; 1913 1914 r300->context.set_clip_state = r300_set_clip_state; 1915 r300->context.set_sample_mask = r300_set_sample_mask; 1916 1917 r300->context.set_constant_buffer = r300_set_constant_buffer; 1918 1919 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1920 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1921 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1922 1923 r300->context.set_stencil_ref = r300_set_stencil_ref; 1924 1925 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1926 1927 r300->context.create_fs_state = r300_create_fs_state; 1928 r300->context.bind_fs_state = r300_bind_fs_state; 1929 r300->context.delete_fs_state = r300_delete_fs_state; 1930 1931 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1932 1933 r300->context.create_rasterizer_state = r300_create_rs_state; 1934 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1935 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1936 1937 r300->context.create_sampler_state = r300_create_sampler_state; 1938 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1939 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1940 r300->context.delete_sampler_state = r300_delete_sampler_state; 1941 1942 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1943 r300->context.create_sampler_view = r300_create_sampler_view; 1944 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1945 1946 r300->context.set_scissor_state = r300_set_scissor_state; 1947 1948 r300->context.set_viewport_state = r300_set_viewport_state; 1949 1950 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1951 r300->context.set_index_buffer = r300_set_index_buffer; 1952 r300->context.redefine_user_buffer = u_default_redefine_user_buffer; 1953 1954 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1955 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1956 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1957 1958 r300->context.create_vs_state = r300_create_vs_state; 1959 r300->context.bind_vs_state = r300_bind_vs_state; 1960 r300->context.delete_vs_state = r300_delete_vs_state; 1961 1962 r300->context.texture_barrier = r300_texture_barrier; 1963} 1964