1 /* 2 * Copyright 2008 Corbin Simpson <MostAwesomeDude (at) gmail.com> 3 * Copyright 2009 Marek Olk <maraeo (at) 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_helpers.h" 29 #include "util/u_math.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 58 static 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 79 static 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 98 static 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 } 112 113 static 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 } 127 128 static 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 } 150 151 static 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 } 171 172 static unsigned blend_discard_conditionally(unsigned eqRGB, unsigned eqA, 173 unsigned dstRGB, unsigned dstA, 174 unsigned srcRGB, unsigned srcA) 175 { 176 unsigned blend_control = 0; 177 178 /* Optimization: discard pixels which don't change the colorbuffer. 179 * 180 * The code below is non-trivial and some math is involved. 181 * 182 * Discarding pixels must be disabled when FP16 AA is enabled. 183 * This is a hardware bug. Also, this implementation wouldn't work 184 * with FP blending enabled and equation clamping disabled. 185 * 186 * Equations other than ADD are rarely used and therefore won't be 187 * optimized. */ 188 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 189 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 190 /* ADD: X+Y 191 * REVERSE_SUBTRACT: Y-X 192 * 193 * The idea is: 194 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 195 * then CB will not be changed. 196 * 197 * Given the srcFactor and dstFactor variables, we can derive 198 * what src and dst should be equal to and discard appropriate 199 * pixels. 200 */ 201 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 202 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 203 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 204 dstRGB, dstA)) { 205 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 206 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 207 dstRGB, dstA)) { 208 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 209 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 210 dstRGB, dstA)) { 211 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 212 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 213 dstRGB, dstA)) { 214 blend_control |= 215 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 216 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 217 dstRGB, dstA)) { 218 blend_control |= 219 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 220 } 221 } 222 return blend_control; 223 } 224 225 /* The hardware colormask is clunky a must be swizzled depending on the format. 226 * This was figured out by trial-and-error. */ 227 static unsigned bgra_cmask(unsigned mask) 228 { 229 return ((mask & PIPE_MASK_R) << 2) | 230 ((mask & PIPE_MASK_B) >> 2) | 231 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 232 } 233 234 static unsigned rgba_cmask(unsigned mask) 235 { 236 return mask & PIPE_MASK_RGBA; 237 } 238 239 static unsigned rrrr_cmask(unsigned mask) 240 { 241 return (mask & PIPE_MASK_R) | 242 ((mask & PIPE_MASK_R) << 1) | 243 ((mask & PIPE_MASK_R) << 2) | 244 ((mask & PIPE_MASK_R) << 3); 245 } 246 247 static unsigned aaaa_cmask(unsigned mask) 248 { 249 return ((mask & PIPE_MASK_A) >> 3) | 250 ((mask & PIPE_MASK_A) >> 2) | 251 ((mask & PIPE_MASK_A) >> 1) | 252 (mask & PIPE_MASK_A); 253 } 254 255 static unsigned grrg_cmask(unsigned mask) 256 { 257 return ((mask & PIPE_MASK_R) << 1) | 258 ((mask & PIPE_MASK_R) << 2) | 259 ((mask & PIPE_MASK_G) >> 1) | 260 ((mask & PIPE_MASK_G) << 2); 261 } 262 263 static unsigned arra_cmask(unsigned mask) 264 { 265 return ((mask & PIPE_MASK_R) << 1) | 266 ((mask & PIPE_MASK_R) << 2) | 267 ((mask & PIPE_MASK_A) >> 3) | 268 (mask & PIPE_MASK_A); 269 } 270 271 static unsigned blend_read_enable(unsigned eqRGB, unsigned eqA, 272 unsigned dstRGB, unsigned dstA, 273 unsigned srcRGB, unsigned srcA, 274 boolean src_alpha_optz) 275 { 276 unsigned blend_control = 0; 277 278 /* Optimization: some operations do not require the destination color. 279 * 280 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 281 * otherwise blending gives incorrect results. It seems to be 282 * a hardware bug. */ 283 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 284 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 285 dstRGB != PIPE_BLENDFACTOR_ZERO || 286 dstA != PIPE_BLENDFACTOR_ZERO || 287 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 288 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 289 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 290 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 291 srcA == PIPE_BLENDFACTOR_DST_COLOR || 292 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 293 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 294 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 295 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 296 /* Enable reading from the colorbuffer. */ 297 blend_control |= R300_READ_ENABLE; 298 299 if (src_alpha_optz) { 300 /* Optimization: Depending on incoming pixels, we can 301 * conditionally disable the reading in hardware... */ 302 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 303 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 304 /* Disable reading if SRC_ALPHA == 0. */ 305 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 306 dstRGB == PIPE_BLENDFACTOR_ZERO) && 307 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 308 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 309 dstA == PIPE_BLENDFACTOR_ZERO) && 310 (srcRGB != PIPE_BLENDFACTOR_DST_COLOR && 311 srcRGB != PIPE_BLENDFACTOR_DST_ALPHA && 312 srcRGB != PIPE_BLENDFACTOR_INV_DST_COLOR && 313 srcRGB != PIPE_BLENDFACTOR_INV_DST_ALPHA)) { 314 blend_control |= R500_SRC_ALPHA_0_NO_READ; 315 } 316 317 /* Disable reading if SRC_ALPHA == 1. */ 318 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 319 dstRGB == PIPE_BLENDFACTOR_ZERO) && 320 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 321 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 322 dstA == PIPE_BLENDFACTOR_ZERO) && 323 (srcRGB != PIPE_BLENDFACTOR_DST_COLOR && 324 srcRGB != PIPE_BLENDFACTOR_DST_ALPHA && 325 srcRGB != PIPE_BLENDFACTOR_INV_DST_COLOR && 326 srcRGB != PIPE_BLENDFACTOR_INV_DST_ALPHA)) { 327 blend_control |= R500_SRC_ALPHA_1_NO_READ; 328 } 329 } 330 } 331 } 332 return blend_control; 333 } 334 335 /* Create a new blend state based on the CSO blend state. 336 * 337 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 338 static void* r300_create_blend_state(struct pipe_context* pipe, 339 const struct pipe_blend_state* state) 340 { 341 struct r300_screen* r300screen = r300_screen(pipe->screen); 342 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 343 uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 344 uint32_t blend_control_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 345 uint32_t blend_control_noalpha = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 346 uint32_t blend_control_noalpha_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 347 uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 348 uint32_t alpha_blend_control_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 349 uint32_t alpha_blend_control_noalpha = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 350 uint32_t alpha_blend_control_noalpha_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 351 uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ 352 uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ 353 int i; 354 355 const unsigned eqRGB = state->rt[0].rgb_func; 356 const unsigned srcRGB = state->rt[0].rgb_src_factor; 357 const unsigned dstRGB = state->rt[0].rgb_dst_factor; 358 359 const unsigned eqA = state->rt[0].alpha_func; 360 const unsigned srcA = state->rt[0].alpha_src_factor; 361 const unsigned dstA = state->rt[0].alpha_dst_factor; 362 363 unsigned srcRGBX = srcRGB; 364 unsigned dstRGBX = dstRGB; 365 CB_LOCALS; 366 367 blend->state = *state; 368 369 /* force DST_ALPHA to ONE where we can */ 370 switch (srcRGBX) { 371 case PIPE_BLENDFACTOR_DST_ALPHA: 372 srcRGBX = PIPE_BLENDFACTOR_ONE; 373 break; 374 case PIPE_BLENDFACTOR_INV_DST_ALPHA: 375 srcRGBX = PIPE_BLENDFACTOR_ZERO; 376 break; 377 } 378 379 switch (dstRGBX) { 380 case PIPE_BLENDFACTOR_DST_ALPHA: 381 dstRGBX = PIPE_BLENDFACTOR_ONE; 382 break; 383 case PIPE_BLENDFACTOR_INV_DST_ALPHA: 384 dstRGBX = PIPE_BLENDFACTOR_ZERO; 385 break; 386 } 387 388 /* Get blending register values. */ 389 if (state->rt[0].blend_enable) { 390 unsigned blend_eq, blend_eq_noclamp; 391 392 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 393 * this is just the crappy D3D naming */ 394 blend_control = blend_control_noclamp = 395 R300_ALPHA_BLEND_ENABLE | 396 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 397 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 398 399 blend_control_noalpha = blend_control_noalpha_noclamp = 400 R300_ALPHA_BLEND_ENABLE | 401 ( r300_translate_blend_factor(srcRGBX) << R300_SRC_BLEND_SHIFT) | 402 ( r300_translate_blend_factor(dstRGBX) << R300_DST_BLEND_SHIFT); 403 404 blend_eq = r300_translate_blend_function(eqRGB, TRUE); 405 blend_eq_noclamp = r300_translate_blend_function(eqRGB, FALSE); 406 407 blend_control |= blend_eq; 408 blend_control_noalpha |= blend_eq; 409 blend_control_noclamp |= blend_eq_noclamp; 410 blend_control_noalpha_noclamp |= blend_eq_noclamp; 411 412 /* Optimization: some operations do not require the destination color. */ 413 blend_control |= blend_read_enable(eqRGB, eqA, dstRGB, dstA, 414 srcRGB, srcA, r300screen->caps.is_r500); 415 blend_control_noclamp |= blend_read_enable(eqRGB, eqA, dstRGB, dstA, 416 srcRGB, srcA, FALSE); 417 blend_control_noalpha |= blend_read_enable(eqRGB, eqA, dstRGBX, dstA, 418 srcRGBX, srcA, r300screen->caps.is_r500); 419 blend_control_noalpha_noclamp |= blend_read_enable(eqRGB, eqA, dstRGBX, dstA, 420 srcRGBX, srcA, FALSE); 421 422 /* Optimization: discard pixels which don't change the colorbuffer. 423 * It cannot be used with FP16 AA. */ 424 blend_control |= blend_discard_conditionally(eqRGB, eqA, dstRGB, dstA, 425 srcRGB, srcA); 426 blend_control_noalpha |= blend_discard_conditionally(eqRGB, eqA, dstRGBX, dstA, 427 srcRGBX, srcA); 428 429 /* separate alpha */ 430 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 431 blend_control |= R300_SEPARATE_ALPHA_ENABLE; 432 blend_control_noclamp |= R300_SEPARATE_ALPHA_ENABLE; 433 434 alpha_blend_control = alpha_blend_control_noclamp = 435 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 436 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 437 alpha_blend_control |= r300_translate_blend_function(eqA, TRUE); 438 alpha_blend_control_noclamp |= r300_translate_blend_function(eqA, FALSE); 439 } 440 if (srcA != srcRGBX || dstA != dstRGBX || eqA != eqRGB) { 441 blend_control_noalpha |= R300_SEPARATE_ALPHA_ENABLE; 442 blend_control_noalpha_noclamp |= R300_SEPARATE_ALPHA_ENABLE; 443 444 alpha_blend_control_noalpha = alpha_blend_control_noalpha_noclamp = 445 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 446 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 447 alpha_blend_control_noalpha |= r300_translate_blend_function(eqA, TRUE); 448 alpha_blend_control_noalpha_noclamp |= r300_translate_blend_function(eqA, FALSE); 449 } 450 } 451 452 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 453 if (state->logicop_enable) { 454 rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 455 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 456 } 457 458 /* Neither fglrx nor classic r300 ever set this, regardless of dithering 459 * state. Since it's an optional implementation detail, we can leave it 460 * out and never dither. 461 * 462 * This could be revisited if we ever get quality or conformance hints. 463 * 464 if (state->dither) { 465 dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 466 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 467 } 468 */ 469 470 /* Build a command buffer. */ 471 { 472 unsigned (*func[COLORMASK_NUM_SWIZZLES])(unsigned) = { 473 bgra_cmask, 474 rgba_cmask, 475 rrrr_cmask, 476 aaaa_cmask, 477 grrg_cmask, 478 arra_cmask, 479 bgra_cmask, 480 rgba_cmask 481 }; 482 483 for (i = 0; i < COLORMASK_NUM_SWIZZLES; i++) { 484 boolean has_alpha = i != COLORMASK_RGBX && i != COLORMASK_BGRX; 485 486 BEGIN_CB(blend->cb_clamp[i], 8); 487 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 488 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 489 OUT_CB(has_alpha ? blend_control : blend_control_noalpha); 490 OUT_CB(has_alpha ? alpha_blend_control : alpha_blend_control_noalpha); 491 OUT_CB(func[i](state->rt[0].colormask)); 492 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 493 END_CB; 494 } 495 } 496 497 /* Build a command buffer (for RGBA16F). */ 498 BEGIN_CB(blend->cb_noclamp, 8); 499 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 500 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 501 OUT_CB(blend_control_noclamp); 502 OUT_CB(alpha_blend_control_noclamp); 503 OUT_CB(rgba_cmask(state->rt[0].colormask)); 504 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 505 END_CB; 506 507 /* Build a command buffer (for RGB16F). */ 508 BEGIN_CB(blend->cb_noclamp_noalpha, 8); 509 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 510 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 511 OUT_CB(blend_control_noalpha_noclamp); 512 OUT_CB(alpha_blend_control_noalpha_noclamp); 513 OUT_CB(rgba_cmask(state->rt[0].colormask)); 514 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 515 END_CB; 516 517 /* The same as above, but with no colorbuffer reads and writes. */ 518 BEGIN_CB(blend->cb_no_readwrite, 8); 519 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 520 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 521 OUT_CB(0); 522 OUT_CB(0); 523 OUT_CB(0); 524 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 525 END_CB; 526 527 return (void*)blend; 528 } 529 530 /* Bind blend state. */ 531 static void r300_bind_blend_state(struct pipe_context* pipe, 532 void* state) 533 { 534 struct r300_context* r300 = r300_context(pipe); 535 struct r300_blend_state *blend = (struct r300_blend_state*)state; 536 boolean last_alpha_to_one = r300->alpha_to_one; 537 boolean last_alpha_to_coverage = r300->alpha_to_coverage; 538 539 UPDATE_STATE(state, r300->blend_state); 540 541 if (!blend) 542 return; 543 544 r300->alpha_to_one = blend->state.alpha_to_one; 545 r300->alpha_to_coverage = blend->state.alpha_to_coverage; 546 547 if (r300->alpha_to_one != last_alpha_to_one && r300->msaa_enable && 548 r300->fs_status == FRAGMENT_SHADER_VALID) { 549 r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; 550 } 551 552 if (r300->alpha_to_coverage != last_alpha_to_coverage && 553 r300->msaa_enable) { 554 r300_mark_atom_dirty(r300, &r300->dsa_state); 555 } 556 } 557 558 /* Free blend state. */ 559 static void r300_delete_blend_state(struct pipe_context* pipe, 560 void* state) 561 { 562 FREE(state); 563 } 564 565 /* Convert float to 10bit integer */ 566 static unsigned float_to_fixed10(float f) 567 { 568 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 569 } 570 571 /* Set blend color. 572 * Setup both R300 and R500 registers, figure out later which one to write. */ 573 static void r300_set_blend_color(struct pipe_context* pipe, 574 const struct pipe_blend_color* color) 575 { 576 struct r300_context* r300 = r300_context(pipe); 577 struct pipe_framebuffer_state *fb = r300->fb_state.state; 578 struct r300_blend_color_state *state = 579 (struct r300_blend_color_state*)r300->blend_color_state.state; 580 struct pipe_blend_color c; 581 struct pipe_surface *cb; 582 float tmp; 583 CB_LOCALS; 584 585 state->state = *color; /* Save it, so that we can reuse it in set_fb_state */ 586 c = *color; 587 cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL; 588 589 /* The blend color is dependent on the colorbuffer format. */ 590 if (cb) { 591 switch (cb->format) { 592 case PIPE_FORMAT_R8_UNORM: 593 case PIPE_FORMAT_L8_UNORM: 594 case PIPE_FORMAT_I8_UNORM: 595 c.color[1] = c.color[0]; 596 break; 597 598 case PIPE_FORMAT_A8_UNORM: 599 c.color[1] = c.color[3]; 600 break; 601 602 case PIPE_FORMAT_R8G8_UNORM: 603 c.color[2] = c.color[1]; 604 break; 605 606 case PIPE_FORMAT_L8A8_UNORM: 607 case PIPE_FORMAT_R8A8_UNORM: 608 c.color[2] = c.color[3]; 609 break; 610 611 case PIPE_FORMAT_R8G8B8A8_UNORM: 612 case PIPE_FORMAT_R8G8B8X8_UNORM: 613 tmp = c.color[0]; 614 c.color[0] = c.color[2]; 615 c.color[2] = tmp; 616 break; 617 618 default:; 619 } 620 } 621 622 if (r300->screen->caps.is_r500) { 623 BEGIN_CB(state->cb, 3); 624 OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); 625 626 switch (cb ? cb->format : 0) { 627 case PIPE_FORMAT_R16G16B16A16_FLOAT: 628 case PIPE_FORMAT_R16G16B16X16_FLOAT: 629 OUT_CB(util_float_to_half(c.color[2]) | 630 (util_float_to_half(c.color[3]) << 16)); 631 OUT_CB(util_float_to_half(c.color[0]) | 632 (util_float_to_half(c.color[1]) << 16)); 633 break; 634 635 default: 636 OUT_CB(float_to_fixed10(c.color[0]) | 637 (float_to_fixed10(c.color[3]) << 16)); 638 OUT_CB(float_to_fixed10(c.color[2]) | 639 (float_to_fixed10(c.color[1]) << 16)); 640 } 641 642 END_CB; 643 } else { 644 union util_color uc; 645 util_pack_color(c.color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 646 647 BEGIN_CB(state->cb, 2); 648 OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui[0]); 649 END_CB; 650 } 651 652 r300_mark_atom_dirty(r300, &r300->blend_color_state); 653 } 654 655 static void r300_set_clip_state(struct pipe_context* pipe, 656 const struct pipe_clip_state* state) 657 { 658 struct r300_context* r300 = r300_context(pipe); 659 struct r300_clip_state *clip = 660 (struct r300_clip_state*)r300->clip_state.state; 661 CB_LOCALS; 662 663 if (r300->screen->caps.has_tcl) { 664 BEGIN_CB(clip->cb, r300->clip_state.size); 665 OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, 666 (r300->screen->caps.is_r500 ? 667 R500_PVS_UCP_START : R300_PVS_UCP_START)); 668 OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); 669 OUT_CB_TABLE(state->ucp, 6 * 4); 670 END_CB; 671 672 r300_mark_atom_dirty(r300, &r300->clip_state); 673 } else { 674 draw_set_clip_state(r300->draw, state); 675 } 676 } 677 678 /* Create a new depth, stencil, and alpha state based on the CSO dsa state. 679 * 680 * This contains the depth buffer, stencil buffer, alpha test, and such. 681 * On the Radeon, depth and stencil buffer setup are intertwined, which is 682 * the reason for some of the strange-looking assignments across registers. */ 683 static void* r300_create_dsa_state(struct pipe_context* pipe, 684 const struct pipe_depth_stencil_alpha_state* state) 685 { 686 boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; 687 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 688 CB_LOCALS; 689 uint32_t alpha_value_fp16 = 0; 690 uint32_t z_buffer_control = 0; 691 uint32_t z_stencil_control = 0; 692 uint32_t stencil_ref_mask = 0; 693 uint32_t stencil_ref_bf = 0; 694 695 dsa->dsa = *state; 696 697 /* Depth test setup. - separate write mask depth for decomp flush */ 698 if (state->depth.writemask) { 699 z_buffer_control |= R300_Z_WRITE_ENABLE; 700 } 701 702 if (state->depth.enabled) { 703 z_buffer_control |= R300_Z_ENABLE; 704 705 z_stencil_control |= 706 (r300_translate_depth_stencil_function(state->depth.func) << 707 R300_Z_FUNC_SHIFT); 708 } 709 710 /* Stencil buffer setup. */ 711 if (state->stencil[0].enabled) { 712 z_buffer_control |= R300_STENCIL_ENABLE; 713 z_stencil_control |= 714 (r300_translate_depth_stencil_function(state->stencil[0].func) << 715 R300_S_FRONT_FUNC_SHIFT) | 716 (r300_translate_stencil_op(state->stencil[0].fail_op) << 717 R300_S_FRONT_SFAIL_OP_SHIFT) | 718 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 719 R300_S_FRONT_ZPASS_OP_SHIFT) | 720 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 721 R300_S_FRONT_ZFAIL_OP_SHIFT); 722 723 stencil_ref_mask = 724 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 725 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 726 727 if (state->stencil[1].enabled) { 728 dsa->two_sided = TRUE; 729 730 z_buffer_control |= R300_STENCIL_FRONT_BACK; 731 z_stencil_control |= 732 (r300_translate_depth_stencil_function(state->stencil[1].func) << 733 R300_S_BACK_FUNC_SHIFT) | 734 (r300_translate_stencil_op(state->stencil[1].fail_op) << 735 R300_S_BACK_SFAIL_OP_SHIFT) | 736 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 737 R300_S_BACK_ZPASS_OP_SHIFT) | 738 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 739 R300_S_BACK_ZFAIL_OP_SHIFT); 740 741 stencil_ref_bf = 742 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | 743 (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); 744 745 if (is_r500) { 746 z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 747 } else { 748 dsa->two_sided_stencil_ref = 749 (state->stencil[0].valuemask != state->stencil[1].valuemask || 750 state->stencil[0].writemask != state->stencil[1].writemask); 751 } 752 } 753 } 754 755 /* Alpha test setup. */ 756 if (state->alpha.enabled) { 757 dsa->alpha_function = 758 r300_translate_alpha_function(state->alpha.func) | 759 R300_FG_ALPHA_FUNC_ENABLE; 760 761 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 762 alpha_value_fp16 = util_float_to_half(state->alpha.ref_value); 763 } 764 765 BEGIN_CB(&dsa->cb_begin, 8); 766 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 767 OUT_CB(z_buffer_control); 768 OUT_CB(z_stencil_control); 769 OUT_CB(stencil_ref_mask); 770 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, stencil_ref_bf); 771 OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16); 772 END_CB; 773 774 BEGIN_CB(dsa->cb_zb_no_readwrite, 8); 775 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 776 OUT_CB(0); 777 OUT_CB(0); 778 OUT_CB(0); 779 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); 780 OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16); 781 END_CB; 782 783 return (void*)dsa; 784 } 785 786 static void r300_dsa_inject_stencilref(struct r300_context *r300) 787 { 788 struct r300_dsa_state *dsa = 789 (struct r300_dsa_state*)r300->dsa_state.state; 790 791 if (!dsa) 792 return; 793 794 dsa->stencil_ref_mask = 795 (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) | 796 r300->stencil_ref.ref_value[0]; 797 dsa->stencil_ref_bf = 798 (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) | 799 r300->stencil_ref.ref_value[1]; 800 } 801 802 /* Bind DSA state. */ 803 static void r300_bind_dsa_state(struct pipe_context* pipe, 804 void* state) 805 { 806 struct r300_context* r300 = r300_context(pipe); 807 808 if (!state) { 809 return; 810 } 811 812 UPDATE_STATE(state, r300->dsa_state); 813 814 r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */ 815 r300_dsa_inject_stencilref(r300); 816 } 817 818 /* Free DSA state. */ 819 static void r300_delete_dsa_state(struct pipe_context* pipe, 820 void* state) 821 { 822 FREE(state); 823 } 824 825 static void r300_set_stencil_ref(struct pipe_context* pipe, 826 const struct pipe_stencil_ref* sr) 827 { 828 struct r300_context* r300 = r300_context(pipe); 829 830 r300->stencil_ref = *sr; 831 832 r300_dsa_inject_stencilref(r300); 833 r300_mark_atom_dirty(r300, &r300->dsa_state); 834 } 835 836 static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, 837 const char *binding) 838 { 839 struct pipe_resource *tex = surf->texture; 840 struct r300_resource *rtex = r300_resource(tex); 841 842 fprintf(stderr, 843 "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, " 844 "Lastlayer: %i, Level: %i, Format: %s\n" 845 846 "r300: TEX: Macro: %s, Micro: %s, " 847 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", 848 849 binding, index, surf->width, surf->height, 850 surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level, 851 util_format_short_name(surf->format), 852 853 rtex->tex.macrotile[0] ? "YES" : " NO", 854 rtex->tex.microtile ? "YES" : " NO", 855 tex->width0, tex->height0, tex->depth0, 856 tex->last_level, util_format_short_name(surf->format)); 857 } 858 859 void r300_mark_fb_state_dirty(struct r300_context *r300, 860 enum r300_fb_state_change change) 861 { 862 struct pipe_framebuffer_state *state = r300->fb_state.state; 863 864 r300_mark_atom_dirty(r300, &r300->gpu_flush); 865 r300_mark_atom_dirty(r300, &r300->fb_state); 866 867 /* What is marked as dirty depends on the enum r300_fb_state_change. */ 868 if (change == R300_CHANGED_FB_STATE) { 869 r300_mark_atom_dirty(r300, &r300->aa_state); 870 r300_mark_atom_dirty(r300, &r300->dsa_state); /* for AlphaRef */ 871 r300_set_blend_color(&r300->context, r300->blend_color_state.state); 872 } 873 874 if (change == R300_CHANGED_FB_STATE || 875 change == R300_CHANGED_HYPERZ_FLAG) { 876 r300_mark_atom_dirty(r300, &r300->hyperz_state); 877 } 878 879 if (change == R300_CHANGED_FB_STATE || 880 change == R300_CHANGED_MULTIWRITE) { 881 r300_mark_atom_dirty(r300, &r300->fb_state_pipelined); 882 } 883 884 /* Now compute the fb_state atom size. */ 885 r300->fb_state.size = 2 + (8 * state->nr_cbufs); 886 887 if (r300->cbzb_clear) 888 r300->fb_state.size += 10; 889 else if (state->zsbuf) { 890 r300->fb_state.size += 10; 891 if (r300->hyperz_enabled) 892 r300->fb_state.size += 8; 893 } 894 895 if (r300->cmask_in_use) { 896 r300->fb_state.size += 6; 897 if (r300->screen->caps.is_r500 && r300->screen->info.drm_minor >= 29) { 898 r300->fb_state.size += 3; 899 } 900 } 901 902 /* The size of the rest of atoms stays the same. */ 903 } 904 905 static void 906 r300_set_framebuffer_state(struct pipe_context* pipe, 907 const struct pipe_framebuffer_state* state) 908 { 909 struct r300_context* r300 = r300_context(pipe); 910 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 911 struct pipe_framebuffer_state *current_state = r300->fb_state.state; 912 unsigned max_width, max_height, i; 913 uint32_t zbuffer_bpp = 0; 914 boolean unlock_zbuffer = FALSE; 915 916 if (r300->screen->caps.is_r500) { 917 max_width = max_height = 4096; 918 } else if (r300->screen->caps.is_r400) { 919 max_width = max_height = 4021; 920 } else { 921 max_width = max_height = 2560; 922 } 923 924 if (state->width > max_width || state->height > max_height) { 925 fprintf(stderr, "r300: Implementation error: Render targets are too " 926 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 927 return; 928 } 929 930 if (current_state->zsbuf && r300->zmask_in_use && !r300->locked_zbuffer) { 931 /* There is a zmask in use, what are we gonna do? */ 932 if (state->zsbuf) { 933 if (!pipe_surface_equal(current_state->zsbuf, state->zsbuf)) { 934 /* Decompress the currently bound zbuffer before we bind another one. */ 935 r300_decompress_zmask(r300); 936 r300->hiz_in_use = FALSE; 937 } 938 } else { 939 /* We don't bind another zbuffer, so lock the current one. */ 940 pipe_surface_reference(&r300->locked_zbuffer, current_state->zsbuf); 941 } 942 } else if (r300->locked_zbuffer) { 943 /* We have a locked zbuffer now, what are we gonna do? */ 944 if (state->zsbuf) { 945 if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) { 946 /* We are binding some other zbuffer, so decompress the locked one, 947 * it gets unlocked automatically. */ 948 r300_decompress_zmask_locked_unsafe(r300); 949 r300->hiz_in_use = FALSE; 950 } else { 951 /* We are binding the locked zbuffer again, so unlock it. */ 952 unlock_zbuffer = TRUE; 953 } 954 } 955 } 956 assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use); 957 958 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 959 if (!!current_state->zsbuf != !!state->zsbuf) { 960 r300_mark_atom_dirty(r300, &r300->dsa_state); 961 } 962 963 util_copy_framebuffer_state(r300->fb_state.state, state); 964 965 /* Remove trailing NULL colorbuffers. */ 966 while (current_state->nr_cbufs && !current_state->cbufs[current_state->nr_cbufs-1]) 967 current_state->nr_cbufs--; 968 969 /* Set whether CMASK can be used. */ 970 r300->cmask_in_use = 971 state->nr_cbufs == 1 && state->cbufs[0] && 972 r300->screen->cmask_resource == state->cbufs[0]->texture; 973 974 /* Need to reset clamping or colormask. */ 975 r300_mark_atom_dirty(r300, &r300->blend_state); 976 977 /* Re-swizzle the blend color. */ 978 r300_set_blend_color(pipe, &((struct r300_blend_color_state*)r300->blend_color_state.state)->state); 979 980 if (unlock_zbuffer) { 981 pipe_surface_reference(&r300->locked_zbuffer, NULL); 982 } 983 984 r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); 985 986 if (state->zsbuf) { 987 switch (util_format_get_blocksize(state->zsbuf->format)) { 988 case 2: 989 zbuffer_bpp = 16; 990 break; 991 case 4: 992 zbuffer_bpp = 24; 993 break; 994 } 995 996 /* Polygon offset depends on the zbuffer bit depth. */ 997 if (r300->zbuffer_bpp != zbuffer_bpp) { 998 r300->zbuffer_bpp = zbuffer_bpp; 999 1000 if (r300->polygon_offset_enabled) 1001 r300_mark_atom_dirty(r300, &r300->rs_state); 1002 } 1003 } 1004 1005 r300->num_samples = util_framebuffer_get_num_samples(state); 1006 1007 /* Set up AA config. */ 1008 if (r300->num_samples > 1) { 1009 switch (r300->num_samples) { 1010 case 2: 1011 aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE | 1012 R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; 1013 break; 1014 case 4: 1015 aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE | 1016 R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; 1017 break; 1018 case 6: 1019 aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE | 1020 R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; 1021 break; 1022 } 1023 } else { 1024 aa->aa_config = 0; 1025 } 1026 1027 if (DBG_ON(r300, DBG_FB)) { 1028 fprintf(stderr, "r300: set_framebuffer_state:\n"); 1029 for (i = 0; i < state->nr_cbufs; i++) { 1030 if (state->cbufs[i]) 1031 r300_print_fb_surf_info(state->cbufs[i], i, "CB"); 1032 } 1033 if (state->zsbuf) { 1034 r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); 1035 } 1036 } 1037 } 1038 1039 /* Create fragment shader state. */ 1040 static void* r300_create_fs_state(struct pipe_context* pipe, 1041 const struct pipe_shader_state* shader) 1042 { 1043 struct r300_fragment_shader* fs = NULL; 1044 1045 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 1046 1047 /* Copy state directly into shader. */ 1048 fs->state = *shader; 1049 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 1050 1051 return (void*)fs; 1052 } 1053 1054 void r300_mark_fs_code_dirty(struct r300_context *r300) 1055 { 1056 struct r300_fragment_shader* fs = r300_fs(r300); 1057 1058 r300_mark_atom_dirty(r300, &r300->fs); 1059 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1060 r300_mark_atom_dirty(r300, &r300->fs_constants); 1061 r300->fs.size = fs->shader->cb_code_size; 1062 1063 if (r300->screen->caps.is_r500) { 1064 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; 1065 r300->fs_constants.size = fs->shader->externals_count * 4 + 3; 1066 } else { 1067 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; 1068 r300->fs_constants.size = fs->shader->externals_count * 4 + 1; 1069 } 1070 1071 ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = 1072 fs->shader->code.constants_remap_table; 1073 } 1074 1075 /* Bind fragment shader state. */ 1076 static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 1077 { 1078 struct r300_context* r300 = r300_context(pipe); 1079 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 1080 1081 if (!fs) { 1082 r300->fs.state = NULL; 1083 return; 1084 } 1085 1086 r300->fs.state = fs; 1087 r300->fs_status = FRAGMENT_SHADER_DIRTY; 1088 1089 r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ 1090 } 1091 1092 /* Delete fragment shader state. */ 1093 static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 1094 { 1095 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 1096 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 1097 1098 while (ptr) { 1099 tmp = ptr; 1100 ptr = ptr->next; 1101 rc_constants_destroy(&tmp->code.constants); 1102 FREE(tmp->cb_code); 1103 FREE(tmp); 1104 } 1105 FREE((void*)fs->state.tokens); 1106 FREE(shader); 1107 } 1108 1109 static void r300_set_polygon_stipple(struct pipe_context* pipe, 1110 const struct pipe_poly_stipple* state) 1111 { 1112 } 1113 1114 /* Create a new rasterizer state based on the CSO rasterizer state. 1115 * 1116 * This is a very large chunk of state, and covers most of the graphics 1117 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 1118 * 1119 * In a not entirely unironic sidenote, this state has nearly nothing to do 1120 * with the actual block on the Radeon called the rasterizer (RS). */ 1121 static void* r300_create_rs_state(struct pipe_context* pipe, 1122 const struct pipe_rasterizer_state* state) 1123 { 1124 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 1125 uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ 1126 uint32_t vap_clip_cntl; /* R300_VAP_CLIP_CNTL: 0x221C */ 1127 uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ 1128 uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ 1129 uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ 1130 uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */ 1131 uint32_t cull_mode; /* R300_SU_CULL_MODE: 0x42b8 */ 1132 uint32_t line_stipple_config; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */ 1133 uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ 1134 uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ 1135 uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ 1136 uint32_t round_mode; /* R300_GA_ROUND_MODE: 0x428c */ 1137 1138 /* Point sprites texture coordinates, 0: lower left, 1: upper right */ 1139 float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ 1140 float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ 1141 float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ 1142 float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ 1143 boolean vclamp = !r300_context(pipe)->screen->caps.is_r500; 1144 CB_LOCALS; 1145 1146 /* Copy rasterizer state. */ 1147 rs->rs = *state; 1148 rs->rs_draw = *state; 1149 1150 rs->rs.sprite_coord_enable = state->point_quad_rasterization * 1151 state->sprite_coord_enable; 1152 1153 /* Override some states for Draw. */ 1154 rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ 1155 rs->rs_draw.offset_point = 0; 1156 rs->rs_draw.offset_line = 0; 1157 rs->rs_draw.offset_tri = 0; 1158 rs->rs_draw.offset_clamp = 0; 1159 1160 #ifdef PIPE_ARCH_LITTLE_ENDIAN 1161 vap_control_status = R300_VC_NO_SWAP; 1162 #else 1163 vap_control_status = R300_VC_32BIT_SWAP; 1164 #endif 1165 1166 /* If no TCL engine is present, turn off the HW TCL. */ 1167 if (!r300_screen(pipe->screen)->caps.has_tcl) { 1168 vap_control_status |= R300_VAP_TCL_BYPASS; 1169 } 1170 1171 /* Point size width and height. */ 1172 point_size = 1173 pack_float_16_6x(state->point_size) | 1174 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 1175 1176 /* Point size clamping. */ 1177 if (state->point_size_per_vertex) { 1178 /* Per-vertex point size. 1179 * Clamp to [0, max FB size] */ 1180 float min_psiz = util_get_min_point_size(state); 1181 float max_psiz = pipe->screen->get_paramf(pipe->screen, 1182 PIPE_CAPF_MAX_POINT_WIDTH); 1183 point_minmax = 1184 (pack_float_16_6x(min_psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 1185 (pack_float_16_6x(max_psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 1186 } else { 1187 /* We cannot disable the point-size vertex output, 1188 * so clamp it. */ 1189 float psiz = state->point_size; 1190 point_minmax = 1191 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 1192 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 1193 } 1194 1195 /* Line control. */ 1196 line_control = pack_float_16_6x(state->line_width) | 1197 R300_GA_LINE_CNTL_END_TYPE_COMP; 1198 1199 /* Enable polygon mode */ 1200 polygon_mode = 0; 1201 if (state->fill_front != PIPE_POLYGON_MODE_FILL || 1202 state->fill_back != PIPE_POLYGON_MODE_FILL) { 1203 polygon_mode = R300_GA_POLY_MODE_DUAL; 1204 } 1205 1206 /* Front face */ 1207 if (state->front_ccw) 1208 cull_mode = R300_FRONT_FACE_CCW; 1209 else 1210 cull_mode = R300_FRONT_FACE_CW; 1211 1212 /* Polygon offset */ 1213 polygon_offset_enable = 0; 1214 if (util_get_offset(state, state->fill_front)) { 1215 polygon_offset_enable |= R300_FRONT_ENABLE; 1216 } 1217 if (util_get_offset(state, state->fill_back)) { 1218 polygon_offset_enable |= R300_BACK_ENABLE; 1219 } 1220 1221 rs->polygon_offset_enable = polygon_offset_enable != 0; 1222 1223 /* Polygon mode */ 1224 if (polygon_mode) { 1225 polygon_mode |= 1226 r300_translate_polygon_mode_front(state->fill_front); 1227 polygon_mode |= 1228 r300_translate_polygon_mode_back(state->fill_back); 1229 } 1230 1231 if (state->cull_face & PIPE_FACE_FRONT) { 1232 cull_mode |= R300_CULL_FRONT; 1233 } 1234 if (state->cull_face & PIPE_FACE_BACK) { 1235 cull_mode |= R300_CULL_BACK; 1236 } 1237 1238 if (state->line_stipple_enable) { 1239 line_stipple_config = 1240 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 1241 (fui((float)state->line_stipple_factor) & 1242 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 1243 /* XXX this might need to be scaled up */ 1244 line_stipple_value = state->line_stipple_pattern; 1245 } else { 1246 line_stipple_config = 0; 1247 line_stipple_value = 0; 1248 } 1249 1250 if (state->flatshade) { 1251 rs->color_control = R300_SHADE_MODEL_FLAT; 1252 } else { 1253 rs->color_control = R300_SHADE_MODEL_SMOOTH; 1254 } 1255 1256 clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 1257 1258 /* Point sprites coord mode */ 1259 if (rs->rs.sprite_coord_enable) { 1260 switch (state->sprite_coord_mode) { 1261 case PIPE_SPRITE_COORD_UPPER_LEFT: 1262 point_texcoord_top = 0.0f; 1263 point_texcoord_bottom = 1.0f; 1264 break; 1265 case PIPE_SPRITE_COORD_LOWER_LEFT: 1266 point_texcoord_top = 1.0f; 1267 point_texcoord_bottom = 0.0f; 1268 break; 1269 } 1270 } 1271 1272 if (r300_screen(pipe->screen)->caps.has_tcl) { 1273 vap_clip_cntl = (state->clip_plane_enable & 63) | 1274 R300_PS_UCP_MODE_CLIP_AS_TRIFAN; 1275 } else { 1276 vap_clip_cntl = R300_CLIP_DISABLE; 1277 } 1278 1279 /* Vertex color clamping. FP20 means no clamping. */ 1280 round_mode = 1281 R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST | 1282 (!vclamp ? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20 | 1283 R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20) : 0); 1284 1285 /* Build the main command buffer. */ 1286 BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); 1287 OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); 1288 OUT_CB_REG(R300_VAP_CLIP_CNTL, vap_clip_cntl); 1289 OUT_CB_REG(R300_GA_POINT_SIZE, point_size); 1290 OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2); 1291 OUT_CB(point_minmax); 1292 OUT_CB(line_control); 1293 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2); 1294 OUT_CB(polygon_offset_enable); 1295 rs->cull_mode_index = 11; 1296 OUT_CB(cull_mode); 1297 OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config); 1298 OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); 1299 OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); 1300 OUT_CB_REG(R300_GA_ROUND_MODE, round_mode); 1301 OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); 1302 OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); 1303 OUT_CB_32F(point_texcoord_left); 1304 OUT_CB_32F(point_texcoord_bottom); 1305 OUT_CB_32F(point_texcoord_right); 1306 OUT_CB_32F(point_texcoord_top); 1307 END_CB; 1308 1309 /* Build the two command buffers for polygon offset setup. */ 1310 if (polygon_offset_enable) { 1311 float scale = state->offset_scale * 12; 1312 float offset = state->offset_units * 4; 1313 1314 BEGIN_CB(rs->cb_poly_offset_zb16, 5); 1315 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); 1316 OUT_CB_32F(scale); 1317 OUT_CB_32F(offset); 1318 OUT_CB_32F(scale); 1319 OUT_CB_32F(offset); 1320 END_CB; 1321 1322 offset = state->offset_units * 2; 1323 1324 BEGIN_CB(rs->cb_poly_offset_zb24, 5); 1325 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); 1326 OUT_CB_32F(scale); 1327 OUT_CB_32F(offset); 1328 OUT_CB_32F(scale); 1329 OUT_CB_32F(offset); 1330 END_CB; 1331 } 1332 1333 return (void*)rs; 1334 } 1335 1336 /* Bind rasterizer state. */ 1337 static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 1338 { 1339 struct r300_context* r300 = r300_context(pipe); 1340 struct r300_rs_state* rs = (struct r300_rs_state*)state; 1341 int last_sprite_coord_enable = r300->sprite_coord_enable; 1342 boolean last_two_sided_color = r300->two_sided_color; 1343 boolean last_msaa_enable = r300->msaa_enable; 1344 boolean last_flatshade = r300->flatshade; 1345 boolean last_clip_halfz = r300->clip_halfz; 1346 1347 if (r300->draw && rs) { 1348 draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); 1349 } 1350 1351 if (rs) { 1352 r300->polygon_offset_enabled = rs->polygon_offset_enable; 1353 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 1354 r300->two_sided_color = rs->rs.light_twoside; 1355 r300->msaa_enable = rs->rs.multisample; 1356 r300->flatshade = rs->rs.flatshade; 1357 r300->clip_halfz = rs->rs.clip_halfz; 1358 } else { 1359 r300->polygon_offset_enabled = FALSE; 1360 r300->sprite_coord_enable = 0; 1361 r300->two_sided_color = FALSE; 1362 r300->msaa_enable = FALSE; 1363 r300->flatshade = FALSE; 1364 r300->clip_halfz = FALSE; 1365 } 1366 1367 UPDATE_STATE(state, r300->rs_state); 1368 r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0); 1369 1370 if (last_sprite_coord_enable != r300->sprite_coord_enable || 1371 last_two_sided_color != r300->two_sided_color || 1372 last_flatshade != r300->flatshade) { 1373 r300_mark_atom_dirty(r300, &r300->rs_block_state); 1374 } 1375 1376 if (last_msaa_enable != r300->msaa_enable) { 1377 if (r300->alpha_to_coverage) { 1378 r300_mark_atom_dirty(r300, &r300->dsa_state); 1379 } 1380 1381 if (r300->alpha_to_one && 1382 r300->fs_status == FRAGMENT_SHADER_VALID) { 1383 r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; 1384 } 1385 } 1386 1387 if (r300->screen->caps.has_tcl && last_clip_halfz != r300->clip_halfz) { 1388 r300_mark_atom_dirty(r300, &r300->vs_state); 1389 } 1390 } 1391 1392 /* Free rasterizer state. */ 1393 static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 1394 { 1395 FREE(state); 1396 } 1397 1398 static void* 1399 r300_create_sampler_state(struct pipe_context* pipe, 1400 const struct pipe_sampler_state* state) 1401 { 1402 struct r300_context* r300 = r300_context(pipe); 1403 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 1404 boolean is_r500 = r300->screen->caps.is_r500; 1405 int lod_bias; 1406 1407 sampler->state = *state; 1408 1409 /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG 1410 * or MIN filter is NEAREST. Since texwrap produces same results 1411 * for CLAMP and CLAMP_TO_EDGE, we use them instead. */ 1412 if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST || 1413 sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) { 1414 /* Wrap S. */ 1415 if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP) 1416 sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1417 else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP) 1418 sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1419 1420 /* Wrap T. */ 1421 if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP) 1422 sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1423 else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP) 1424 sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1425 1426 /* Wrap R. */ 1427 if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP) 1428 sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1429 else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP) 1430 sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1431 } 1432 1433 sampler->filter0 |= 1434 (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) | 1435 (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) | 1436 (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT); 1437 1438 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 1439 state->mag_img_filter, 1440 state->min_mip_filter, 1441 state->max_anisotropy > 1); 1442 1443 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 1444 1445 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 1446 /* We must pass these to the merge function to clamp them properly. */ 1447 sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); 1448 sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); 1449 1450 lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); 1451 1452 sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK; 1453 1454 /* This is very high quality anisotropic filtering for R5xx. 1455 * It's good for benchmarking the performance of texturing but 1456 * in practice we don't want to slow down the driver because it's 1457 * a pretty good performance killer. Feel free to play with it. */ 1458 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 1459 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 1460 } 1461 1462 /* R500-specific fixups and optimizations */ 1463 if (r300->screen->caps.is_r500) { 1464 sampler->filter1 |= R500_BORDER_FIX; 1465 } 1466 1467 return (void*)sampler; 1468 } 1469 1470 static void r300_bind_sampler_states(struct pipe_context* pipe, 1471 enum pipe_shader_type shader, 1472 unsigned start, unsigned count, 1473 void** states) 1474 { 1475 struct r300_context* r300 = r300_context(pipe); 1476 struct r300_textures_state* state = 1477 (struct r300_textures_state*)r300->textures_state.state; 1478 unsigned tex_units = r300->screen->caps.num_tex_units; 1479 1480 assert(start == 0); 1481 1482 if (shader != PIPE_SHADER_FRAGMENT) 1483 return; 1484 1485 if (count > tex_units) 1486 return; 1487 1488 memcpy(state->sampler_states, states, sizeof(void*) * count); 1489 state->sampler_state_count = count; 1490 1491 r300_mark_atom_dirty(r300, &r300->textures_state); 1492 } 1493 1494 static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1495 { 1496 FREE(state); 1497 } 1498 1499 static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num) 1500 { 1501 /* This looks like a hack, but I believe it's suppose to work like 1502 * that. To illustrate how this works, let's assume you have 5 textures. 1503 * From docs, 5 and the successive numbers are: 1504 * 1505 * FOURTH_1 = 5 1506 * FOURTH_2 = 6 1507 * FOURTH_3 = 7 1508 * EIGHTH_0 = 8 1509 * EIGHTH_1 = 9 1510 * 1511 * First 3 textures will get 3/4 of size of the cache, divived evenly 1512 * between them. The last 1/4 of the cache must be divided between 1513 * the last 2 textures, each will therefore get 1/8 of the cache. 1514 * Why not just to use "5 + texture_index" ? 1515 * 1516 * This simple trick works for all "num" <= 16. 1517 */ 1518 if (num <= 1) 1519 return R300_TX_CACHE(R300_TX_CACHE_WHOLE); 1520 else 1521 return R300_TX_CACHE(num + index); 1522 } 1523 1524 static void r300_set_sampler_views(struct pipe_context* pipe, 1525 enum pipe_shader_type shader, 1526 unsigned start, unsigned count, 1527 struct pipe_sampler_view** views) 1528 { 1529 struct r300_context* r300 = r300_context(pipe); 1530 struct r300_textures_state* state = 1531 (struct r300_textures_state*)r300->textures_state.state; 1532 struct r300_resource *texture; 1533 unsigned i, real_num_views = 0, view_index = 0; 1534 unsigned tex_units = r300->screen->caps.num_tex_units; 1535 boolean dirty_tex = FALSE; 1536 1537 if (shader != PIPE_SHADER_FRAGMENT) 1538 return; 1539 1540 assert(start == 0); /* non-zero not handled yet */ 1541 1542 if (count > tex_units) { 1543 return; 1544 } 1545 1546 /* Calculate the real number of views. */ 1547 for (i = 0; i < count; i++) { 1548 if (views[i]) 1549 real_num_views++; 1550 } 1551 1552 for (i = 0; i < count; i++) { 1553 pipe_sampler_view_reference( 1554 (struct pipe_sampler_view**)&state->sampler_views[i], 1555 views[i]); 1556 1557 if (!views[i]) { 1558 continue; 1559 } 1560 1561 /* A new sampler view (= texture)... */ 1562 dirty_tex = TRUE; 1563 1564 /* Set the texrect factor in the fragment shader. 1565 * Needed for RECT and NPOT fallback. */ 1566 texture = r300_resource(views[i]->texture); 1567 if (texture->tex.is_npot) { 1568 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1569 } 1570 1571 state->sampler_views[i]->texcache_region = 1572 r300_assign_texture_cache_region(view_index, real_num_views); 1573 view_index++; 1574 } 1575 1576 for (i = count; i < tex_units; i++) { 1577 if (state->sampler_views[i]) { 1578 pipe_sampler_view_reference( 1579 (struct pipe_sampler_view**)&state->sampler_views[i], 1580 NULL); 1581 } 1582 } 1583 1584 state->sampler_view_count = count; 1585 1586 r300_mark_atom_dirty(r300, &r300->textures_state); 1587 1588 if (dirty_tex) { 1589 r300_mark_atom_dirty(r300, &r300->texture_cache_inval); 1590 } 1591 } 1592 1593 struct pipe_sampler_view * 1594 r300_create_sampler_view_custom(struct pipe_context *pipe, 1595 struct pipe_resource *texture, 1596 const struct pipe_sampler_view *templ, 1597 unsigned width0_override, 1598 unsigned height0_override) 1599 { 1600 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1601 struct r300_resource *tex = r300_resource(texture); 1602 boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; 1603 boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle; 1604 1605 if (view) { 1606 unsigned hwformat; 1607 1608 view->base = *templ; 1609 view->base.reference.count = 1; 1610 view->base.context = pipe; 1611 view->base.texture = NULL; 1612 pipe_resource_reference(&view->base.texture, texture); 1613 1614 view->width0_override = width0_override; 1615 view->height0_override = height0_override; 1616 view->swizzle[0] = templ->swizzle_r; 1617 view->swizzle[1] = templ->swizzle_g; 1618 view->swizzle[2] = templ->swizzle_b; 1619 view->swizzle[3] = templ->swizzle_a; 1620 1621 hwformat = r300_translate_texformat(templ->format, 1622 view->swizzle, 1623 is_r500, 1624 dxtc_swizzle); 1625 1626 if (hwformat == ~0) { 1627 fprintf(stderr, "r300: Ooops. Got unsupported format %s in %s.\n", 1628 util_format_short_name(templ->format), __func__); 1629 } 1630 assert(hwformat != ~0); 1631 1632 r300_texture_setup_format_state(r300_screen(pipe->screen), tex, 1633 templ->format, 0, 1634 width0_override, height0_override, 1635 &view->format); 1636 view->format.format1 |= hwformat; 1637 if (is_r500) { 1638 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1639 } 1640 } 1641 1642 return (struct pipe_sampler_view*)view; 1643 } 1644 1645 static struct pipe_sampler_view * 1646 r300_create_sampler_view(struct pipe_context *pipe, 1647 struct pipe_resource *texture, 1648 const struct pipe_sampler_view *templ) 1649 { 1650 return r300_create_sampler_view_custom(pipe, texture, templ, 1651 r300_resource(texture)->tex.width0, 1652 r300_resource(texture)->tex.height0); 1653 } 1654 1655 1656 static void 1657 r300_sampler_view_destroy(struct pipe_context *pipe, 1658 struct pipe_sampler_view *view) 1659 { 1660 pipe_resource_reference(&view->texture, NULL); 1661 FREE(view); 1662 } 1663 1664 static void r300_set_sample_mask(struct pipe_context *pipe, 1665 unsigned mask) 1666 { 1667 struct r300_context* r300 = r300_context(pipe); 1668 1669 *((unsigned*)r300->sample_mask.state) = mask; 1670 1671 r300_mark_atom_dirty(r300, &r300->sample_mask); 1672 } 1673 1674 static void r300_set_scissor_states(struct pipe_context* pipe, 1675 unsigned start_slot, 1676 unsigned num_scissors, 1677 const struct pipe_scissor_state* state) 1678 { 1679 struct r300_context* r300 = r300_context(pipe); 1680 1681 memcpy(r300->scissor_state.state, state, 1682 sizeof(struct pipe_scissor_state)); 1683 1684 r300_mark_atom_dirty(r300, &r300->scissor_state); 1685 } 1686 1687 static void r300_set_viewport_states(struct pipe_context* pipe, 1688 unsigned start_slot, 1689 unsigned num_viewports, 1690 const struct pipe_viewport_state* state) 1691 { 1692 struct r300_context* r300 = r300_context(pipe); 1693 struct r300_viewport_state* viewport = 1694 (struct r300_viewport_state*)r300->viewport_state.state; 1695 1696 r300->viewport = *state; 1697 1698 if (r300->draw) { 1699 draw_set_viewport_states(r300->draw, start_slot, num_viewports, state); 1700 viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; 1701 return; 1702 } 1703 1704 /* Do the transform in HW. */ 1705 viewport->vte_control = R300_VTX_W0_FMT; 1706 1707 if (state->scale[0] != 1.0f) { 1708 viewport->xscale = state->scale[0]; 1709 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1710 } 1711 if (state->scale[1] != 1.0f) { 1712 viewport->yscale = state->scale[1]; 1713 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1714 } 1715 if (state->scale[2] != 1.0f) { 1716 viewport->zscale = state->scale[2]; 1717 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1718 } 1719 if (state->translate[0] != 0.0f) { 1720 viewport->xoffset = state->translate[0]; 1721 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1722 } 1723 if (state->translate[1] != 0.0f) { 1724 viewport->yoffset = state->translate[1]; 1725 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1726 } 1727 if (state->translate[2] != 0.0f) { 1728 viewport->zoffset = state->translate[2]; 1729 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1730 } 1731 1732 r300_mark_atom_dirty(r300, &r300->viewport_state); 1733 if (r300->fs.state && r300_fs(r300)->shader && 1734 r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1735 r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); 1736 } 1737 } 1738 1739 static void r300_set_vertex_buffers_hwtcl(struct pipe_context* pipe, 1740 unsigned start_slot, unsigned count, 1741 const struct pipe_vertex_buffer* buffers) 1742 { 1743 struct r300_context* r300 = r300_context(pipe); 1744 1745 util_set_vertex_buffers_count(r300->vertex_buffer, 1746 &r300->nr_vertex_buffers, 1747 buffers, start_slot, count); 1748 1749 /* There must be at least one vertex buffer set, otherwise it locks up. */ 1750 if (!r300->nr_vertex_buffers) { 1751 util_set_vertex_buffers_count(r300->vertex_buffer, 1752 &r300->nr_vertex_buffers, 1753 &r300->dummy_vb, 0, 1); 1754 } 1755 1756 r300->vertex_arrays_dirty = TRUE; 1757 } 1758 1759 static void r300_set_vertex_buffers_swtcl(struct pipe_context* pipe, 1760 unsigned start_slot, unsigned count, 1761 const struct pipe_vertex_buffer* buffers) 1762 { 1763 struct r300_context* r300 = r300_context(pipe); 1764 unsigned i; 1765 1766 util_set_vertex_buffers_count(r300->vertex_buffer, 1767 &r300->nr_vertex_buffers, 1768 buffers, start_slot, count); 1769 draw_set_vertex_buffers(r300->draw, start_slot, count, buffers); 1770 1771 if (!buffers) 1772 return; 1773 1774 for (i = 0; i < count; i++) { 1775 if (buffers[i].is_user_buffer) { 1776 draw_set_mapped_vertex_buffer(r300->draw, start_slot + i, 1777 buffers[i].buffer.user, ~0); 1778 } else if (buffers[i].buffer.resource) { 1779 draw_set_mapped_vertex_buffer(r300->draw, start_slot + i, 1780 r300_resource(buffers[i].buffer.resource)->malloced_buffer, ~0); 1781 } 1782 } 1783 } 1784 1785 /* Initialize the PSC tables. */ 1786 static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1787 { 1788 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1789 uint16_t type, swizzle; 1790 enum pipe_format format; 1791 unsigned i; 1792 1793 /* Vertex shaders have no semantics on their inputs, 1794 * so PSC should just route stuff based on the vertex elements, 1795 * and not on attrib information. */ 1796 for (i = 0; i < velems->count; i++) { 1797 format = velems->velem[i].src_format; 1798 1799 type = r300_translate_vertex_data_type(format); 1800 if (type == R300_INVALID_FORMAT) { 1801 fprintf(stderr, "r300: Bad vertex format %s.\n", 1802 util_format_short_name(format)); 1803 assert(0); 1804 abort(); 1805 } 1806 1807 type |= i << R300_DST_VEC_LOC_SHIFT; 1808 swizzle = r300_translate_vertex_data_swizzle(format); 1809 1810 if (i & 1) { 1811 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1812 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1813 } else { 1814 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1815 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1816 } 1817 } 1818 1819 /* Set the last vector in the PSC. */ 1820 if (i) { 1821 i -= 1; 1822 } 1823 vstream->vap_prog_stream_cntl[i >> 1] |= 1824 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1825 1826 vstream->count = (i >> 1) + 1; 1827 } 1828 1829 static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1830 unsigned count, 1831 const struct pipe_vertex_element* attribs) 1832 { 1833 struct r300_vertex_element_state *velems; 1834 unsigned i; 1835 struct pipe_vertex_element dummy_attrib = {0}; 1836 1837 /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */ 1838 if (!count) { 1839 dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM; 1840 attribs = &dummy_attrib; 1841 count = 1; 1842 } else if (count > 16) { 1843 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1844 " requested %i, using 16.\n", count); 1845 count = 16; 1846 } 1847 1848 velems = CALLOC_STRUCT(r300_vertex_element_state); 1849 if (!velems) 1850 return NULL; 1851 1852 velems->count = count; 1853 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1854 1855 if (r300_screen(pipe->screen)->caps.has_tcl) { 1856 /* Setup PSC. 1857 * The unused components will be replaced by (..., 0, 1). */ 1858 r300_vertex_psc(velems); 1859 1860 for (i = 0; i < count; i++) { 1861 velems->format_size[i] = 1862 align(util_format_get_blocksize(velems->velem[i].src_format), 4); 1863 velems->vertex_size_dwords += velems->format_size[i] / 4; 1864 } 1865 } 1866 1867 return velems; 1868 } 1869 1870 static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1871 void *state) 1872 { 1873 struct r300_context *r300 = r300_context(pipe); 1874 struct r300_vertex_element_state *velems = state; 1875 1876 if (!velems) { 1877 return; 1878 } 1879 1880 r300->velems = velems; 1881 1882 if (r300->draw) { 1883 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1884 return; 1885 } 1886 1887 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1888 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1889 r300->vertex_arrays_dirty = TRUE; 1890 } 1891 1892 static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1893 { 1894 FREE(state); 1895 } 1896 1897 static void* r300_create_vs_state(struct pipe_context* pipe, 1898 const struct pipe_shader_state* shader) 1899 { 1900 struct r300_context* r300 = r300_context(pipe); 1901 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1902 1903 /* Copy state directly into shader. */ 1904 vs->state = *shader; 1905 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1906 1907 if (r300->screen->caps.has_tcl) { 1908 r300_init_vs_outputs(r300, vs); 1909 r300_translate_vertex_shader(r300, vs); 1910 } else { 1911 r300_draw_init_vertex_shader(r300, vs); 1912 } 1913 1914 return vs; 1915 } 1916 1917 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1918 { 1919 struct r300_context* r300 = r300_context(pipe); 1920 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1921 1922 if (!vs) { 1923 r300->vs_state.state = NULL; 1924 return; 1925 } 1926 if (vs == r300->vs_state.state) { 1927 return; 1928 } 1929 r300->vs_state.state = vs; 1930 1931 /* The majority of the RS block bits is dependent on the vertex shader. */ 1932 r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ 1933 1934 if (r300->screen->caps.has_tcl) { 1935 unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2; 1936 r300_mark_atom_dirty(r300, &r300->vs_state); 1937 r300->vs_state.size = vs->code.length + 9 + 1938 (R300_VS_MAX_FC_OPS * fc_op_dwords + 4); 1939 1940 r300_mark_atom_dirty(r300, &r300->vs_constants); 1941 r300->vs_constants.size = 1942 2 + 1943 (vs->externals_count ? vs->externals_count * 4 + 3 : 0) + 1944 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1945 1946 ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = 1947 vs->code.constants_remap_table; 1948 1949 r300_mark_atom_dirty(r300, &r300->pvs_flush); 1950 } else { 1951 draw_bind_vertex_shader(r300->draw, 1952 (struct draw_vertex_shader*)vs->draw_vs); 1953 } 1954 } 1955 1956 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1957 { 1958 struct r300_context* r300 = r300_context(pipe); 1959 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1960 1961 if (r300->screen->caps.has_tcl) { 1962 rc_constants_destroy(&vs->code.constants); 1963 FREE(vs->code.constants_remap_table); 1964 } else { 1965 draw_delete_vertex_shader(r300->draw, 1966 (struct draw_vertex_shader*)vs->draw_vs); 1967 } 1968 1969 FREE((void*)vs->state.tokens); 1970 FREE(shader); 1971 } 1972 1973 static void r300_set_constant_buffer(struct pipe_context *pipe, 1974 enum pipe_shader_type shader, uint index, 1975 const struct pipe_constant_buffer *cb) 1976 { 1977 struct r300_context* r300 = r300_context(pipe); 1978 struct r300_constant_buffer *cbuf; 1979 uint32_t *mapped; 1980 1981 if (!cb || (!cb->buffer && !cb->user_buffer)) 1982 return; 1983 1984 switch (shader) { 1985 case PIPE_SHADER_VERTEX: 1986 cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; 1987 break; 1988 case PIPE_SHADER_FRAGMENT: 1989 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1990 break; 1991 default: 1992 return; 1993 } 1994 1995 1996 if (cb->user_buffer) 1997 mapped = (uint32_t*)cb->user_buffer; 1998 else { 1999 struct r300_resource *rbuf = r300_resource(cb->buffer); 2000 2001 if (rbuf && rbuf->malloced_buffer) 2002 mapped = (uint32_t*)rbuf->malloced_buffer; 2003 else 2004 return; 2005 } 2006 2007 if (shader == PIPE_SHADER_FRAGMENT || 2008 (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { 2009 cbuf->ptr = mapped; 2010 } 2011 2012 if (shader == PIPE_SHADER_VERTEX) { 2013 if (r300->screen->caps.has_tcl) { 2014 struct r300_vertex_shader *vs = 2015 (struct r300_vertex_shader*)r300->vs_state.state; 2016 2017 if (!vs) { 2018 cbuf->buffer_base = 0; 2019 return; 2020 } 2021 2022 cbuf->buffer_base = r300->vs_const_base; 2023 r300->vs_const_base += vs->code.constants.Count; 2024 if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) { 2025 r300->vs_const_base = vs->code.constants.Count; 2026 cbuf->buffer_base = 0; 2027 r300_mark_atom_dirty(r300, &r300->pvs_flush); 2028 } 2029 r300_mark_atom_dirty(r300, &r300->vs_constants); 2030 } else if (r300->draw) { 2031 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 2032 0, mapped, cb->buffer_size); 2033 } 2034 } else if (shader == PIPE_SHADER_FRAGMENT) { 2035 r300_mark_atom_dirty(r300, &r300->fs_constants); 2036 } 2037 } 2038 2039 static void r300_texture_barrier(struct pipe_context *pipe, unsigned flags) 2040 { 2041 struct r300_context *r300 = r300_context(pipe); 2042 2043 r300_mark_atom_dirty(r300, &r300->gpu_flush); 2044 r300_mark_atom_dirty(r300, &r300->texture_cache_inval); 2045 } 2046 2047 static void r300_memory_barrier(struct pipe_context *pipe, unsigned flags) 2048 { 2049 } 2050 2051 void r300_init_state_functions(struct r300_context* r300) 2052 { 2053 r300->context.create_blend_state = r300_create_blend_state; 2054 r300->context.bind_blend_state = r300_bind_blend_state; 2055 r300->context.delete_blend_state = r300_delete_blend_state; 2056 2057 r300->context.set_blend_color = r300_set_blend_color; 2058 2059 r300->context.set_clip_state = r300_set_clip_state; 2060 r300->context.set_sample_mask = r300_set_sample_mask; 2061 2062 r300->context.set_constant_buffer = r300_set_constant_buffer; 2063 2064 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 2065 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 2066 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 2067 2068 r300->context.set_stencil_ref = r300_set_stencil_ref; 2069 2070 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 2071 2072 r300->context.create_fs_state = r300_create_fs_state; 2073 r300->context.bind_fs_state = r300_bind_fs_state; 2074 r300->context.delete_fs_state = r300_delete_fs_state; 2075 2076 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 2077 2078 r300->context.create_rasterizer_state = r300_create_rs_state; 2079 r300->context.bind_rasterizer_state = r300_bind_rs_state; 2080 r300->context.delete_rasterizer_state = r300_delete_rs_state; 2081 2082 r300->context.create_sampler_state = r300_create_sampler_state; 2083 r300->context.bind_sampler_states = r300_bind_sampler_states; 2084 r300->context.delete_sampler_state = r300_delete_sampler_state; 2085 2086 r300->context.set_sampler_views = r300_set_sampler_views; 2087 r300->context.create_sampler_view = r300_create_sampler_view; 2088 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 2089 2090 r300->context.set_scissor_states = r300_set_scissor_states; 2091 2092 r300->context.set_viewport_states = r300_set_viewport_states; 2093 2094 if (r300->screen->caps.has_tcl) { 2095 r300->context.set_vertex_buffers = r300_set_vertex_buffers_hwtcl; 2096 } else { 2097 r300->context.set_vertex_buffers = r300_set_vertex_buffers_swtcl; 2098 } 2099 2100 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 2101 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 2102 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 2103 2104 r300->context.create_vs_state = r300_create_vs_state; 2105 r300->context.bind_vs_state = r300_bind_vs_state; 2106 r300->context.delete_vs_state = r300_delete_vs_state; 2107 2108 r300->context.texture_barrier = r300_texture_barrier; 2109 r300->context.memory_barrier = r300_memory_barrier; 2110 } 2111