r300_state.c revision 69adebf5945d994485c584c183c148fc2c1373ed
1/* 2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> 3 * Copyright 2009 Marek Olšák <maraeo@gmail.com> 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * on the rights to use, copy, modify, merge, publish, distribute, sub 9 * license, and/or sell copies of the Software, and to permit persons to whom 10 * the Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */ 23 24#include "draw/draw_context.h" 25 26#include "util/u_math.h" 27#include "util/u_memory.h" 28#include "util/u_pack_color.h" 29 30#include "tgsi/tgsi_parse.h" 31 32#include "pipe/p_config.h" 33 34#include "r300_cb.h" 35#include "r300_context.h" 36#include "r300_emit.h" 37#include "r300_reg.h" 38#include "r300_screen.h" 39#include "r300_screen_buffer.h" 40#include "r300_state_inlines.h" 41#include "r300_fs.h" 42#include "r300_texture.h" 43#include "r300_vs.h" 44#include "r300_winsys.h" 45 46/* r300_state: Functions used to intialize state context by translating 47 * Gallium state objects into semi-native r300 state objects. */ 48 49#define UPDATE_STATE(cso, atom) \ 50 if (cso != atom.state) { \ 51 atom.state = cso; \ 52 atom.dirty = TRUE; \ 53 } 54 55static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 56 unsigned dstRGB, unsigned dstA) 57{ 58 /* If the blend equation is ADD or REVERSE_SUBTRACT, 59 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 60 * will not be changed. 61 * Notice that the dst factors are the src factors inverted. */ 62 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 63 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 64 srcRGB == PIPE_BLENDFACTOR_ZERO) && 65 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 66 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 67 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 68 srcA == PIPE_BLENDFACTOR_ZERO) && 69 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 70 dstRGB == PIPE_BLENDFACTOR_ONE) && 71 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 72 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 73 dstA == PIPE_BLENDFACTOR_ONE); 74} 75 76static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 77 unsigned dstRGB, unsigned dstA) 78{ 79 /* If the blend equation is ADD or REVERSE_SUBTRACT, 80 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 81 * will not be changed. 82 * Notice that the dst factors are the src factors inverted. */ 83 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 84 srcRGB == PIPE_BLENDFACTOR_ZERO) && 85 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 86 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 87 srcA == PIPE_BLENDFACTOR_ZERO) && 88 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 89 dstRGB == PIPE_BLENDFACTOR_ONE) && 90 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 91 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 92 dstA == PIPE_BLENDFACTOR_ONE); 93} 94 95static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 96 unsigned dstRGB, unsigned dstA) 97{ 98 /* If the blend equation is ADD or REVERSE_SUBTRACT, 99 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 100 * will not be changed. 101 * Notice that the dst factors are the src factors inverted. */ 102 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 103 srcRGB == PIPE_BLENDFACTOR_ZERO) && 104 (srcA == PIPE_BLENDFACTOR_ZERO) && 105 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 106 dstRGB == PIPE_BLENDFACTOR_ONE) && 107 (dstA == PIPE_BLENDFACTOR_ONE); 108} 109 110static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 111 unsigned dstRGB, unsigned dstA) 112{ 113 /* If the blend equation is ADD or REVERSE_SUBTRACT, 114 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 115 * will not be changed. 116 * Notice that the dst factors are the src factors inverted. */ 117 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 118 srcRGB == PIPE_BLENDFACTOR_ZERO) && 119 (srcA == PIPE_BLENDFACTOR_ZERO) && 120 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 121 dstRGB == PIPE_BLENDFACTOR_ONE) && 122 (dstA == PIPE_BLENDFACTOR_ONE); 123} 124 125static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 126 unsigned dstRGB, unsigned dstA) 127{ 128 /* If the blend equation is ADD or REVERSE_SUBTRACT, 129 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 130 * the colorbuffer will not be changed. 131 * Notice that the dst factors are the src factors inverted. */ 132 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 133 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 134 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 135 srcRGB == PIPE_BLENDFACTOR_ZERO) && 136 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 137 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 138 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 139 srcA == PIPE_BLENDFACTOR_ZERO) && 140 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 141 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 142 dstRGB == PIPE_BLENDFACTOR_ONE) && 143 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 144 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 145 dstA == PIPE_BLENDFACTOR_ONE); 146} 147 148static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 149 unsigned dstRGB, unsigned dstA) 150{ 151 /* If the blend equation is ADD or REVERSE_SUBTRACT, 152 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 153 * the colorbuffer will not be changed. 154 * Notice that the dst factors are the src factors inverted. */ 155 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 156 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 157 srcRGB == PIPE_BLENDFACTOR_ZERO) && 158 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 159 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 160 srcA == PIPE_BLENDFACTOR_ZERO) && 161 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 162 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 163 dstRGB == PIPE_BLENDFACTOR_ONE) && 164 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 165 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 166 dstA == PIPE_BLENDFACTOR_ONE); 167} 168 169static unsigned bgra_cmask(unsigned mask) 170{ 171 /* Gallium uses RGBA color ordering while R300 expects BGRA. */ 172 173 return ((mask & PIPE_MASK_R) << 2) | 174 ((mask & PIPE_MASK_B) >> 2) | 175 (mask & (PIPE_MASK_G | PIPE_MASK_A)); 176} 177 178/* Create a new blend state based on the CSO blend state. 179 * 180 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 181static void* r300_create_blend_state(struct pipe_context* pipe, 182 const struct pipe_blend_state* state) 183{ 184 struct r300_screen* r300screen = r300_screen(pipe->screen); 185 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 186 uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ 187 uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ 188 uint32_t color_channel_mask = 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ 189 uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ 190 uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ 191 CB_LOCALS; 192 193 if (state->rt[0].blend_enable) 194 { 195 unsigned eqRGB = state->rt[0].rgb_func; 196 unsigned srcRGB = state->rt[0].rgb_src_factor; 197 unsigned dstRGB = state->rt[0].rgb_dst_factor; 198 199 unsigned eqA = state->rt[0].alpha_func; 200 unsigned srcA = state->rt[0].alpha_src_factor; 201 unsigned dstA = state->rt[0].alpha_dst_factor; 202 203 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 204 * this is just the crappy D3D naming */ 205 blend_control = R300_ALPHA_BLEND_ENABLE | 206 r300_translate_blend_function(eqRGB) | 207 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 208 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 209 210 /* Optimization: some operations do not require the destination color. 211 * 212 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 213 * otherwise blending gives incorrect results. It seems to be 214 * a hardware bug. */ 215 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 216 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 217 dstRGB != PIPE_BLENDFACTOR_ZERO || 218 dstA != PIPE_BLENDFACTOR_ZERO || 219 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 220 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 221 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 222 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 223 srcA == PIPE_BLENDFACTOR_DST_COLOR || 224 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 225 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 226 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 227 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 228 /* Enable reading from the colorbuffer. */ 229 blend_control |= R300_READ_ENABLE; 230 231 if (r300screen->caps.is_r500) { 232 /* Optimization: Depending on incoming pixels, we can 233 * conditionally disable the reading in hardware... */ 234 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 235 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 236 /* Disable reading if SRC_ALPHA == 0. */ 237 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 238 dstRGB == PIPE_BLENDFACTOR_ZERO) && 239 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 240 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 241 dstA == PIPE_BLENDFACTOR_ZERO)) { 242 blend_control |= R500_SRC_ALPHA_0_NO_READ; 243 } 244 245 /* Disable reading if SRC_ALPHA == 1. */ 246 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 247 dstRGB == PIPE_BLENDFACTOR_ZERO) && 248 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 249 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 250 dstA == PIPE_BLENDFACTOR_ZERO)) { 251 blend_control |= R500_SRC_ALPHA_1_NO_READ; 252 } 253 } 254 } 255 } 256 257 /* Optimization: discard pixels which don't change the colorbuffer. 258 * 259 * The code below is non-trivial and some math is involved. 260 * 261 * Discarding pixels must be disabled when FP16 AA is enabled. 262 * This is a hardware bug. Also, this implementation wouldn't work 263 * with FP blending enabled and equation clamping disabled. 264 * 265 * Equations other than ADD are rarely used and therefore won't be 266 * optimized. */ 267 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 268 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 269 /* ADD: X+Y 270 * REVERSE_SUBTRACT: Y-X 271 * 272 * The idea is: 273 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 274 * then CB will not be changed. 275 * 276 * Given the srcFactor and dstFactor variables, we can derive 277 * what src and dst should be equal to and discard appropriate 278 * pixels. 279 */ 280 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 281 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 282 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 283 dstRGB, dstA)) { 284 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 285 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 286 dstRGB, dstA)) { 287 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 288 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 289 dstRGB, dstA)) { 290 blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 291 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 292 dstRGB, dstA)) { 293 blend_control |= 294 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 295 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 296 dstRGB, dstA)) { 297 blend_control |= 298 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 299 } 300 } 301 302 /* separate alpha */ 303 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 304 blend_control |= R300_SEPARATE_ALPHA_ENABLE; 305 alpha_blend_control = 306 r300_translate_blend_function(eqA) | 307 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 308 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 309 } 310 } 311 312 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 313 if (state->logicop_enable) { 314 rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 315 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 316 } 317 318 /* Color channel masks for all MRTs. */ 319 color_channel_mask = bgra_cmask(state->rt[0].colormask); 320 if (r300screen->caps.is_r500 && state->independent_blend_enable) { 321 if (state->rt[1].blend_enable) { 322 color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; 323 } 324 if (state->rt[2].blend_enable) { 325 color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; 326 } 327 if (state->rt[3].blend_enable) { 328 color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; 329 } 330 } 331 332 /* Neither fglrx nor classic r300 ever set this, regardless of dithering 333 * state. Since it's an optional implementation detail, we can leave it 334 * out and never dither. 335 * 336 * This could be revisited if we ever get quality or conformance hints. 337 * 338 if (state->dither) { 339 dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 340 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 341 } 342 */ 343 344 /* Build a command buffer. */ 345 BEGIN_CB(blend->cb, 8); 346 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 347 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 348 OUT_CB(blend_control); 349 OUT_CB(alpha_blend_control); 350 OUT_CB(color_channel_mask); 351 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 352 END_CB; 353 354 /* The same as above, but with no colorbuffer reads and writes. */ 355 BEGIN_CB(blend->cb_no_readwrite, 8); 356 OUT_CB_REG(R300_RB3D_ROPCNTL, rop); 357 OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); 358 OUT_CB(0); 359 OUT_CB(0); 360 OUT_CB(0); 361 OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); 362 END_CB; 363 364 return (void*)blend; 365} 366 367/* Bind blend state. */ 368static void r300_bind_blend_state(struct pipe_context* pipe, 369 void* state) 370{ 371 struct r300_context* r300 = r300_context(pipe); 372 373 UPDATE_STATE(state, r300->blend_state); 374} 375 376/* Free blend state. */ 377static void r300_delete_blend_state(struct pipe_context* pipe, 378 void* state) 379{ 380 FREE(state); 381} 382 383/* Convert float to 10bit integer */ 384static unsigned float_to_fixed10(float f) 385{ 386 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 387} 388 389/* Set blend color. 390 * Setup both R300 and R500 registers, figure out later which one to write. */ 391static void r300_set_blend_color(struct pipe_context* pipe, 392 const struct pipe_blend_color* color) 393{ 394 struct r300_context* r300 = r300_context(pipe); 395 struct r300_blend_color_state* state = 396 (struct r300_blend_color_state*)r300->blend_color_state.state; 397 CB_LOCALS; 398 399 if (r300->screen->caps.is_r500) { 400 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 401 BEGIN_CB(state->cb, 3); 402 OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); 403 OUT_CB(float_to_fixed10(color->color[0]) | 404 (float_to_fixed10(color->color[3]) << 16)); 405 OUT_CB(float_to_fixed10(color->color[2]) | 406 (float_to_fixed10(color->color[1]) << 16)); 407 END_CB; 408 } else { 409 union util_color uc; 410 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 411 412 BEGIN_CB(state->cb, 2); 413 OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui); 414 END_CB; 415 } 416 417 r300->blend_color_state.dirty = TRUE; 418} 419 420static void r300_set_clip_state(struct pipe_context* pipe, 421 const struct pipe_clip_state* state) 422{ 423 struct r300_context* r300 = r300_context(pipe); 424 struct r300_clip_state *clip = 425 (struct r300_clip_state*)r300->clip_state.state; 426 CB_LOCALS; 427 428 clip->clip = *state; 429 430 if (r300->screen->caps.has_tcl) { 431 BEGIN_CB(clip->cb, 29); 432 OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, 433 (r300->screen->caps.is_r500 ? 434 R500_PVS_UCP_START : R300_PVS_UCP_START)); 435 OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); 436 OUT_CB_TABLE(state->ucp, 6 * 4); 437 OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) | 438 R300_PS_UCP_MODE_CLIP_AS_TRIFAN); 439 END_CB; 440 441 r300->clip_state.dirty = TRUE; 442 } else { 443 draw_flush(r300->draw); 444 draw_set_clip_state(r300->draw, state); 445 } 446} 447 448static void 449r300_set_sample_mask(struct pipe_context *pipe, 450 unsigned sample_mask) 451{ 452} 453 454 455/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 456 * 457 * This contains the depth buffer, stencil buffer, alpha test, and such. 458 * On the Radeon, depth and stencil buffer setup are intertwined, which is 459 * the reason for some of the strange-looking assignments across registers. */ 460static void* 461 r300_create_dsa_state(struct pipe_context* pipe, 462 const struct pipe_depth_stencil_alpha_state* state) 463{ 464 struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; 465 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 466 CB_LOCALS; 467 468 dsa->dsa = *state; 469 470 /* Depth test setup. */ 471 if (state->depth.enabled) { 472 dsa->z_buffer_control |= R300_Z_ENABLE; 473 474 if (state->depth.writemask) { 475 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 476 } 477 478 dsa->z_stencil_control |= 479 (r300_translate_depth_stencil_function(state->depth.func) << 480 R300_Z_FUNC_SHIFT); 481 } 482 483 /* Stencil buffer setup. */ 484 if (state->stencil[0].enabled) { 485 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 486 dsa->z_stencil_control |= 487 (r300_translate_depth_stencil_function(state->stencil[0].func) << 488 R300_S_FRONT_FUNC_SHIFT) | 489 (r300_translate_stencil_op(state->stencil[0].fail_op) << 490 R300_S_FRONT_SFAIL_OP_SHIFT) | 491 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 492 R300_S_FRONT_ZPASS_OP_SHIFT) | 493 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 494 R300_S_FRONT_ZFAIL_OP_SHIFT); 495 496 dsa->stencil_ref_mask = 497 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 498 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 499 500 if (state->stencil[1].enabled) { 501 dsa->two_sided = TRUE; 502 503 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 504 dsa->z_stencil_control |= 505 (r300_translate_depth_stencil_function(state->stencil[1].func) << 506 R300_S_BACK_FUNC_SHIFT) | 507 (r300_translate_stencil_op(state->stencil[1].fail_op) << 508 R300_S_BACK_SFAIL_OP_SHIFT) | 509 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 510 R300_S_BACK_ZPASS_OP_SHIFT) | 511 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 512 R300_S_BACK_ZFAIL_OP_SHIFT); 513 514 dsa->stencil_ref_bf = 515 (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | 516 (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); 517 518 if (caps->is_r500) { 519 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 520 } else { 521 dsa->two_sided_stencil_ref = 522 (state->stencil[0].valuemask != state->stencil[1].valuemask || 523 state->stencil[0].writemask != state->stencil[1].writemask); 524 } 525 } 526 } 527 528 /* Alpha test setup. */ 529 if (state->alpha.enabled) { 530 dsa->alpha_function = 531 r300_translate_alpha_function(state->alpha.func) | 532 R300_FG_ALPHA_FUNC_ENABLE; 533 534 /* We could use 10bit alpha ref but who needs that? */ 535 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 536 537 if (caps->is_r500) 538 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 539 } 540 541 BEGIN_CB(&dsa->cb_begin, 8); 542 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); 543 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 544 OUT_CB(dsa->z_buffer_control); 545 OUT_CB(dsa->z_stencil_control); 546 OUT_CB(dsa->stencil_ref_mask); 547 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); 548 END_CB; 549 550 BEGIN_CB(dsa->cb_no_readwrite, 8); 551 OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); 552 OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); 553 OUT_CB(0); 554 OUT_CB(0); 555 OUT_CB(0); 556 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); 557 END_CB; 558 559 return (void*)dsa; 560} 561 562static void r300_dsa_inject_stencilref(struct r300_context *r300) 563{ 564 struct r300_dsa_state *dsa = 565 (struct r300_dsa_state*)r300->dsa_state.state; 566 567 if (!dsa) 568 return; 569 570 dsa->stencil_ref_mask = 571 (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) | 572 r300->stencil_ref.ref_value[0]; 573 dsa->stencil_ref_bf = 574 (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) | 575 r300->stencil_ref.ref_value[1]; 576} 577 578/* Bind DSA state. */ 579static void r300_bind_dsa_state(struct pipe_context* pipe, 580 void* state) 581{ 582 struct r300_context* r300 = r300_context(pipe); 583 584 if (!state) { 585 return; 586 } 587 588 UPDATE_STATE(state, r300->dsa_state); 589 590 r300_dsa_inject_stencilref(r300); 591} 592 593/* Free DSA state. */ 594static void r300_delete_dsa_state(struct pipe_context* pipe, 595 void* state) 596{ 597 FREE(state); 598} 599 600static void r300_set_stencil_ref(struct pipe_context* pipe, 601 const struct pipe_stencil_ref* sr) 602{ 603 struct r300_context* r300 = r300_context(pipe); 604 605 r300->stencil_ref = *sr; 606 607 r300_dsa_inject_stencilref(r300); 608 r300->dsa_state.dirty = TRUE; 609} 610 611/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ 612static void r300_fb_set_tiling_flags(struct r300_context *r300, 613 const struct pipe_framebuffer_state *old_state, 614 const struct pipe_framebuffer_state *new_state) 615{ 616 struct r300_texture *tex; 617 unsigned i, level; 618 619 /* Set tiling flags for new surfaces. */ 620 for (i = 0; i < new_state->nr_cbufs; i++) { 621 tex = r300_texture(new_state->cbufs[i]->texture); 622 level = new_state->cbufs[i]->level; 623 624 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 625 tex->pitch[0] * util_format_get_blocksize(tex->b.b.format), 626 tex->microtile, 627 tex->mip_macrotile[level]); 628 } 629 if (new_state->zsbuf) { 630 tex = r300_texture(new_state->zsbuf->texture); 631 level = new_state->zsbuf->level; 632 633 r300->rws->buffer_set_tiling(r300->rws, tex->buffer, 634 tex->pitch[0] * util_format_get_blocksize(tex->b.b.format), 635 tex->microtile, 636 tex->mip_macrotile[level]); 637 } 638} 639 640static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, 641 const char *binding) 642{ 643 struct pipe_resource *tex = surf->texture; 644 struct r300_texture *rtex = r300_texture(tex); 645 646 fprintf(stderr, 647 "r300: %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, " 648 "Face: %i, Level: %i, Format: %s\n" 649 650 "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, " 651 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", 652 653 binding, index, surf->width, surf->height, surf->offset, 654 surf->zslice, surf->face, surf->level, 655 util_format_short_name(surf->format), 656 657 rtex->macrotile ? "YES" : " NO", rtex->microtile ? "YES" : " NO", 658 rtex->hwpitch[0], tex->width0, tex->height0, tex->depth0, 659 tex->last_level, util_format_short_name(tex->format)); 660} 661 662static void 663 r300_set_framebuffer_state(struct pipe_context* pipe, 664 const struct pipe_framebuffer_state* state) 665{ 666 struct r300_context* r300 = r300_context(pipe); 667 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 668 struct pipe_framebuffer_state *old_state = r300->fb_state.state; 669 unsigned max_width, max_height, i; 670 uint32_t zbuffer_bpp = 0; 671 672 if (r300->screen->caps.is_r500) { 673 max_width = max_height = 4096; 674 } else if (r300->screen->caps.is_r400) { 675 max_width = max_height = 4021; 676 } else { 677 max_width = max_height = 2560; 678 } 679 680 if (state->width > max_width || state->height > max_height) { 681 fprintf(stderr, "r300: Implementation error: Render targets are too " 682 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 683 return; 684 } 685 686 if (r300->draw) { 687 draw_flush(r300->draw); 688 } 689 690 r300->gpu_flush.dirty = TRUE; 691 r300->aa_state.dirty = TRUE; 692 r300->fb_state.dirty = TRUE; 693 694 /* If nr_cbufs is changed from zero to non-zero or vice versa... */ 695 if (!!old_state->nr_cbufs != !!state->nr_cbufs) { 696 r300->blend_state.dirty = TRUE; 697 } 698 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 699 if (!!old_state->zsbuf != !!state->zsbuf) { 700 r300->dsa_state.dirty = TRUE; 701 } 702 703 /* The tiling flags are dependent on the surface miplevel, unfortunately. */ 704 r300_fb_set_tiling_flags(r300, r300->fb_state.state, state); 705 706 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 707 708 r300->fb_state.size = 709 7 + 710 (8 * state->nr_cbufs) + 711 (state->zsbuf ? (r300->screen->caps.has_hiz ? 22 : 18) : 0); 712 713 /* Polygon offset depends on the zbuffer bit depth. */ 714 if (state->zsbuf && r300->polygon_offset_enabled) { 715 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 716 case 2: 717 zbuffer_bpp = 16; 718 break; 719 case 4: 720 zbuffer_bpp = 24; 721 break; 722 } 723 724 if (r300->zbuffer_bpp != zbuffer_bpp) { 725 r300->zbuffer_bpp = zbuffer_bpp; 726 r300->rs_state.dirty = TRUE; 727 } 728 } 729 730 /* Set up AA config. */ 731 if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { 732 if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) { 733 aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE; 734 735 switch (state->cbufs[0]->texture->nr_samples) { 736 case 2: 737 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; 738 break; 739 case 3: 740 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3; 741 break; 742 case 4: 743 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; 744 break; 745 case 6: 746 aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; 747 break; 748 } 749 } else { 750 aa->aa_config = 0; 751 } 752 } 753 754 if (DBG_ON(r300, DBG_FB)) { 755 fprintf(stderr, "r300: set_framebuffer_state:\n"); 756 for (i = 0; i < state->nr_cbufs; i++) { 757 r300_print_fb_surf_info(state->cbufs[i], i, "CB"); 758 } 759 if (state->zsbuf) { 760 r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); 761 } 762 } 763} 764 765/* Create fragment shader state. */ 766static void* r300_create_fs_state(struct pipe_context* pipe, 767 const struct pipe_shader_state* shader) 768{ 769 struct r300_fragment_shader* fs = NULL; 770 771 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 772 773 /* Copy state directly into shader. */ 774 fs->state = *shader; 775 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 776 777 return (void*)fs; 778} 779 780void r300_mark_fs_code_dirty(struct r300_context *r300) 781{ 782 struct r300_fragment_shader* fs = r300_fs(r300); 783 784 r300->fs.dirty = TRUE; 785 r300->fs_rc_constant_state.dirty = TRUE; 786 r300->fs_constants.dirty = TRUE; 787 r300->fs.size = fs->shader->cb_code_size; 788 789 if (r300->screen->caps.is_r500) { 790 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; 791 r300->fs_constants.size = fs->shader->externals_count * 4 + 3; 792 } else { 793 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; 794 r300->fs_constants.size = fs->shader->externals_count * 4 + 1; 795 } 796} 797 798/* Bind fragment shader state. */ 799static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 800{ 801 struct r300_context* r300 = r300_context(pipe); 802 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 803 804 if (fs == NULL) { 805 r300->fs.state = NULL; 806 return; 807 } 808 809 r300->fs.state = fs; 810 r300_pick_fragment_shader(r300); 811 r300_mark_fs_code_dirty(r300); 812 813 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 814} 815 816/* Delete fragment shader state. */ 817static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 818{ 819 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 820 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 821 822 while (ptr) { 823 tmp = ptr; 824 ptr = ptr->next; 825 rc_constants_destroy(&tmp->code.constants); 826 FREE(tmp->cb_code); 827 FREE(tmp); 828 } 829 FREE((void*)fs->state.tokens); 830 FREE(shader); 831} 832 833static void r300_set_polygon_stipple(struct pipe_context* pipe, 834 const struct pipe_poly_stipple* state) 835{ 836 /* XXX no idea how to set this up, but not terribly important */ 837} 838 839/* Create a new rasterizer state based on the CSO rasterizer state. 840 * 841 * This is a very large chunk of state, and covers most of the graphics 842 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 843 * 844 * In a not entirely unironic sidenote, this state has nearly nothing to do 845 * with the actual block on the Radeon called the rasterizer (RS). */ 846static void* r300_create_rs_state(struct pipe_context* pipe, 847 const struct pipe_rasterizer_state* state) 848{ 849 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 850 int i; 851 float psiz; 852 853 /* Copy rasterizer state. */ 854 rs->rs = *state; 855 rs->rs_draw = *state; 856 857 /* Override some states for Draw. */ 858 rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ 859 860#ifdef PIPE_ARCH_LITTLE_ENDIAN 861 rs->vap_control_status = R300_VC_NO_SWAP; 862#else 863 rs->vap_control_status = R300_VC_32BIT_SWAP; 864#endif 865 866 /* If no TCL engine is present, turn off the HW TCL. */ 867 if (!r300_screen(pipe->screen)->caps.has_tcl) { 868 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 869 } 870 871 /* Point size width and height. */ 872 rs->point_size = 873 pack_float_16_6x(state->point_size) | 874 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 875 876 /* Point size clamping. */ 877 if (state->point_size_per_vertex) { 878 /* Per-vertex point size. 879 * Clamp to [0, max FB size] */ 880 psiz = pipe->screen->get_paramf(pipe->screen, 881 PIPE_CAP_MAX_POINT_WIDTH); 882 rs->point_minmax = 883 pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; 884 } else { 885 /* We cannot disable the point-size vertex output, 886 * so clamp it. */ 887 psiz = state->point_size; 888 rs->point_minmax = 889 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 890 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 891 } 892 893 /* Line control. */ 894 rs->line_control = pack_float_16_6x(state->line_width) | 895 R300_GA_LINE_CNTL_END_TYPE_COMP; 896 897 /* Enable polygon mode */ 898 if (state->fill_front != PIPE_POLYGON_MODE_FILL || 899 state->fill_back != PIPE_POLYGON_MODE_FILL) { 900 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 901 } 902 903 /* Front face */ 904 if (state->front_ccw) 905 rs->cull_mode = R300_FRONT_FACE_CCW; 906 else 907 rs->cull_mode = R300_FRONT_FACE_CW; 908 909 /* Polygon offset */ 910 if (util_get_offset(state, state->fill_front)) { 911 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 912 } 913 if (util_get_offset(state, state->fill_back)) { 914 rs->polygon_offset_enable |= R300_BACK_ENABLE; 915 } 916 917 /* Polygon mode */ 918 if (rs->polygon_mode) { 919 rs->polygon_mode |= 920 r300_translate_polygon_mode_front(state->fill_front); 921 rs->polygon_mode |= 922 r300_translate_polygon_mode_back(state->fill_back); 923 } 924 925 if (state->cull_face & PIPE_FACE_FRONT) { 926 rs->cull_mode |= R300_CULL_FRONT; 927 } 928 if (state->cull_face & PIPE_FACE_BACK) { 929 rs->cull_mode |= R300_CULL_BACK; 930 } 931 932 if (rs->polygon_offset_enable) { 933 rs->depth_offset = state->offset_units; 934 rs->depth_scale = state->offset_scale; 935 } 936 937 if (state->line_stipple_enable) { 938 rs->line_stipple_config = 939 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 940 (fui((float)state->line_stipple_factor) & 941 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 942 /* XXX this might need to be scaled up */ 943 rs->line_stipple_value = state->line_stipple_pattern; 944 } 945 946 if (state->flatshade) { 947 rs->color_control = R300_SHADE_MODEL_FLAT; 948 } else { 949 rs->color_control = R300_SHADE_MODEL_SMOOTH; 950 } 951 952 rs->clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 953 954 /* Point sprites */ 955 if (state->sprite_coord_enable) { 956 rs->stuffing_enable = R300_GB_POINT_STUFF_ENABLE; 957 for (i = 0; i < 8; i++) { 958 if (state->sprite_coord_enable & (1 << i)) 959 rs->stuffing_enable |= 960 R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); 961 } 962 963 rs->point_texcoord_left = 0.0f; 964 rs->point_texcoord_right = 1.0f; 965 966 switch (state->sprite_coord_mode) { 967 case PIPE_SPRITE_COORD_UPPER_LEFT: 968 rs->point_texcoord_top = 0.0f; 969 rs->point_texcoord_bottom = 1.0f; 970 break; 971 case PIPE_SPRITE_COORD_LOWER_LEFT: 972 rs->point_texcoord_top = 1.0f; 973 rs->point_texcoord_bottom = 0.0f; 974 break; 975 } 976 } 977 978 if (state->gl_rasterization_rules) { 979 rs->multisample_position_0 = 0x66666666; 980 rs->multisample_position_1 = 0x6666666; 981 } 982 983 return (void*)rs; 984} 985 986/* Bind rasterizer state. */ 987static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 988{ 989 struct r300_context* r300 = r300_context(pipe); 990 struct r300_rs_state* rs = (struct r300_rs_state*)state; 991 int last_sprite_coord_enable = r300->sprite_coord_enable; 992 boolean last_two_sided_color = r300->two_sided_color; 993 994 if (r300->draw && rs) { 995 draw_flush(r300->draw); 996 draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); 997 } 998 999 if (rs) { 1000 r300->polygon_offset_enabled = (rs->rs.offset_point || 1001 rs->rs.offset_line || 1002 rs->rs.offset_tri); 1003 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 1004 r300->two_sided_color = rs->rs.light_twoside; 1005 } else { 1006 r300->polygon_offset_enabled = FALSE; 1007 r300->sprite_coord_enable = 0; 1008 r300->two_sided_color = FALSE; 1009 } 1010 1011 UPDATE_STATE(state, r300->rs_state); 1012 r300->rs_state.size = 25 + (r300->polygon_offset_enabled ? 5 : 0) + 1013 (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0) ? 3 : 0); 1014 1015 if (last_sprite_coord_enable != r300->sprite_coord_enable || 1016 last_two_sided_color != r300->two_sided_color) { 1017 r300->rs_block_state.dirty = TRUE; 1018 } 1019} 1020 1021/* Free rasterizer state. */ 1022static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 1023{ 1024 FREE(state); 1025} 1026 1027static void* 1028 r300_create_sampler_state(struct pipe_context* pipe, 1029 const struct pipe_sampler_state* state) 1030{ 1031 struct r300_context* r300 = r300_context(pipe); 1032 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 1033 boolean is_r500 = r300->screen->caps.is_r500; 1034 int lod_bias; 1035 union util_color uc; 1036 1037 sampler->state = *state; 1038 1039 /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG 1040 * or MIN filter is NEAREST. Since texwrap produces same results 1041 * for CLAMP and CLAMP_TO_EDGE, we use them instead. */ 1042 if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST || 1043 sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) { 1044 /* Wrap S. */ 1045 if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP) 1046 sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1047 else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP) 1048 sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1049 1050 /* Wrap T. */ 1051 if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP) 1052 sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1053 else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP) 1054 sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1055 1056 /* Wrap R. */ 1057 if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP) 1058 sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 1059 else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP) 1060 sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; 1061 } 1062 1063 sampler->filter0 |= 1064 (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) | 1065 (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) | 1066 (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT); 1067 1068 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 1069 state->mag_img_filter, 1070 state->min_mip_filter, 1071 state->max_anisotropy > 0); 1072 1073 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 1074 1075 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 1076 /* We must pass these to the merge function to clamp them properly. */ 1077 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 1078 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 1079 1080 lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); 1081 1082 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 1083 1084 /* This is very high quality anisotropic filtering for R5xx. 1085 * It's good for benchmarking the performance of texturing but 1086 * in practice we don't want to slow down the driver because it's 1087 * a pretty good performance killer. Feel free to play with it. */ 1088 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 1089 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 1090 } 1091 1092 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 1093 sampler->border_color = uc.ui; 1094 1095 /* R500-specific fixups and optimizations */ 1096 if (r300->screen->caps.is_r500) { 1097 sampler->filter1 |= R500_BORDER_FIX; 1098 } 1099 1100 return (void*)sampler; 1101} 1102 1103static void r300_bind_sampler_states(struct pipe_context* pipe, 1104 unsigned count, 1105 void** states) 1106{ 1107 struct r300_context* r300 = r300_context(pipe); 1108 struct r300_textures_state* state = 1109 (struct r300_textures_state*)r300->textures_state.state; 1110 unsigned tex_units = r300->screen->caps.num_tex_units; 1111 1112 if (count > tex_units) { 1113 return; 1114 } 1115 1116 memcpy(state->sampler_states, states, sizeof(void*) * count); 1117 state->sampler_state_count = count; 1118 1119 r300->textures_state.dirty = TRUE; 1120} 1121 1122static void r300_lacks_vertex_textures(struct pipe_context* pipe, 1123 unsigned count, 1124 void** states) 1125{ 1126} 1127 1128static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1129{ 1130 FREE(state); 1131} 1132 1133static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num) 1134{ 1135 /* This looks like a hack, but I believe it's suppose to work like 1136 * that. To illustrate how this works, let's assume you have 5 textures. 1137 * From docs, 5 and the successive numbers are: 1138 * 1139 * FOURTH_1 = 5 1140 * FOURTH_2 = 6 1141 * FOURTH_3 = 7 1142 * EIGHTH_0 = 8 1143 * EIGHTH_1 = 9 1144 * 1145 * First 3 textures will get 3/4 of size of the cache, divived evenly 1146 * between them. The last 1/4 of the cache must be divided between 1147 * the last 2 textures, each will therefore get 1/8 of the cache. 1148 * Why not just to use "5 + texture_index" ? 1149 * 1150 * This simple trick works for all "num" <= 16. 1151 */ 1152 if (num <= 1) 1153 return R300_TX_CACHE(R300_TX_CACHE_WHOLE); 1154 else 1155 return R300_TX_CACHE(num + index); 1156} 1157 1158static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 1159 unsigned count, 1160 struct pipe_sampler_view** views) 1161{ 1162 struct r300_context* r300 = r300_context(pipe); 1163 struct r300_textures_state* state = 1164 (struct r300_textures_state*)r300->textures_state.state; 1165 struct r300_texture *texture; 1166 unsigned i, real_num_views = 0, view_index = 0; 1167 unsigned tex_units = r300->screen->caps.num_tex_units; 1168 boolean dirty_tex = FALSE; 1169 1170 if (count > tex_units) { 1171 return; 1172 } 1173 1174 /* Calculate the real number of views. */ 1175 for (i = 0; i < count; i++) { 1176 if (views[i]) 1177 real_num_views++; 1178 } 1179 1180 for (i = 0; i < count; i++) { 1181 if (&state->sampler_views[i]->base != views[i]) { 1182 pipe_sampler_view_reference( 1183 (struct pipe_sampler_view**)&state->sampler_views[i], 1184 views[i]); 1185 1186 if (!views[i]) { 1187 continue; 1188 } 1189 1190 /* A new sampler view (= texture)... */ 1191 dirty_tex = TRUE; 1192 1193 /* Set the texrect factor in the fragment shader. 1194 * Needed for RECT and NPOT fallback. */ 1195 texture = r300_texture(views[i]->texture); 1196 if (texture->uses_pitch) { 1197 r300->fs_rc_constant_state.dirty = TRUE; 1198 } 1199 1200 state->sampler_views[i]->texcache_region = 1201 r300_assign_texture_cache_region(view_index, real_num_views); 1202 view_index++; 1203 } 1204 } 1205 1206 for (i = count; i < tex_units; i++) { 1207 if (state->sampler_views[i]) { 1208 pipe_sampler_view_reference( 1209 (struct pipe_sampler_view**)&state->sampler_views[i], 1210 NULL); 1211 } 1212 } 1213 1214 state->sampler_view_count = count; 1215 1216 r300->textures_state.dirty = TRUE; 1217 1218 if (dirty_tex) { 1219 r300->texture_cache_inval.dirty = TRUE; 1220 } 1221} 1222 1223static struct pipe_sampler_view * 1224r300_create_sampler_view(struct pipe_context *pipe, 1225 struct pipe_resource *texture, 1226 const struct pipe_sampler_view *templ) 1227{ 1228 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1229 struct r300_texture *tex = r300_texture(texture); 1230 1231 if (view) { 1232 view->base = *templ; 1233 view->base.reference.count = 1; 1234 view->base.context = pipe; 1235 view->base.texture = NULL; 1236 pipe_resource_reference(&view->base.texture, texture); 1237 1238 view->swizzle[0] = templ->swizzle_r; 1239 view->swizzle[1] = templ->swizzle_g; 1240 view->swizzle[2] = templ->swizzle_b; 1241 view->swizzle[3] = templ->swizzle_a; 1242 1243 view->format = tex->tx_format; 1244 view->format.format1 |= r300_translate_texformat(templ->format, 1245 view->swizzle); 1246 if (r300_screen(pipe->screen)->caps.is_r500) { 1247 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1248 } 1249 } 1250 1251 return (struct pipe_sampler_view*)view; 1252} 1253 1254static void 1255r300_sampler_view_destroy(struct pipe_context *pipe, 1256 struct pipe_sampler_view *view) 1257{ 1258 pipe_resource_reference(&view->texture, NULL); 1259 FREE(view); 1260} 1261 1262static void r300_set_scissor_state(struct pipe_context* pipe, 1263 const struct pipe_scissor_state* state) 1264{ 1265 struct r300_context* r300 = r300_context(pipe); 1266 1267 memcpy(r300->scissor_state.state, state, 1268 sizeof(struct pipe_scissor_state)); 1269 1270 r300->scissor_state.dirty = TRUE; 1271} 1272 1273static void r300_set_viewport_state(struct pipe_context* pipe, 1274 const struct pipe_viewport_state* state) 1275{ 1276 struct r300_context* r300 = r300_context(pipe); 1277 struct r300_viewport_state* viewport = 1278 (struct r300_viewport_state*)r300->viewport_state.state; 1279 1280 r300->viewport = *state; 1281 1282 if (r300->draw) { 1283 draw_flush(r300->draw); 1284 draw_set_viewport_state(r300->draw, state); 1285 viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; 1286 return; 1287 } 1288 1289 /* Do the transform in HW. */ 1290 viewport->vte_control = R300_VTX_W0_FMT; 1291 1292 if (state->scale[0] != 1.0f) { 1293 viewport->xscale = state->scale[0]; 1294 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1295 } 1296 if (state->scale[1] != 1.0f) { 1297 viewport->yscale = state->scale[1]; 1298 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1299 } 1300 if (state->scale[2] != 1.0f) { 1301 viewport->zscale = state->scale[2]; 1302 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1303 } 1304 if (state->translate[0] != 0.0f) { 1305 viewport->xoffset = state->translate[0]; 1306 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1307 } 1308 if (state->translate[1] != 0.0f) { 1309 viewport->yoffset = state->translate[1]; 1310 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1311 } 1312 if (state->translate[2] != 0.0f) { 1313 viewport->zoffset = state->translate[2]; 1314 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1315 } 1316 1317 r300->viewport_state.dirty = TRUE; 1318 if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1319 r300->fs_rc_constant_state.dirty = TRUE; 1320 } 1321} 1322 1323static void r300_set_vertex_buffers(struct pipe_context* pipe, 1324 unsigned count, 1325 const struct pipe_vertex_buffer* buffers) 1326{ 1327 struct r300_context* r300 = r300_context(pipe); 1328 struct pipe_vertex_buffer *vbo; 1329 unsigned i, max_index = (1 << 24) - 1; 1330 boolean any_user_buffer = FALSE; 1331 1332 if (count == r300->vertex_buffer_count && 1333 memcmp(r300->vertex_buffer, buffers, 1334 sizeof(struct pipe_vertex_buffer) * count) == 0) { 1335 return; 1336 } 1337 1338 if (r300->screen->caps.has_tcl) { 1339 /* HW TCL. */ 1340 r300->incompatible_vb_layout = FALSE; 1341 1342 /* Check if the strides and offsets are aligned to the size of DWORD. */ 1343 for (i = 0; i < count; i++) { 1344 if (buffers[i].buffer) { 1345 if (buffers[i].stride % 4 != 0 || 1346 buffers[i].buffer_offset % 4 != 0) { 1347 r300->incompatible_vb_layout = TRUE; 1348 break; 1349 } 1350 } 1351 } 1352 1353 for (i = 0; i < count; i++) { 1354 /* Why, yes, I AM casting away constness. How did you know? */ 1355 vbo = (struct pipe_vertex_buffer*)&buffers[i]; 1356 1357 /* Skip NULL buffers */ 1358 if (!buffers[i].buffer) { 1359 continue; 1360 } 1361 1362 if (r300_buffer_is_user_buffer(vbo->buffer)) { 1363 any_user_buffer = TRUE; 1364 } 1365 1366 if (vbo->max_index == ~0) { 1367 /* if no VBO stride then only one vertex value so max index is 1 */ 1368 /* should think about converting to VS constants like svga does */ 1369 if (!vbo->stride) 1370 vbo->max_index = 1; 1371 else 1372 vbo->max_index = 1373 (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; 1374 } 1375 1376 max_index = MIN2(vbo->max_index, max_index); 1377 } 1378 1379 r300->any_user_vbs = any_user_buffer; 1380 r300->vertex_buffer_max_index = max_index; 1381 1382 } else { 1383 /* SW TCL. */ 1384 draw_flush(r300->draw); 1385 draw_set_vertex_buffers(r300->draw, count, buffers); 1386 } 1387 1388 /* Common code. */ 1389 for (i = 0; i < count; i++) { 1390 /* Reference our buffer. */ 1391 pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); 1392 } 1393 for (; i < r300->vertex_buffer_count; i++) { 1394 /* Dereference any old buffers. */ 1395 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); 1396 } 1397 1398 memcpy(r300->vertex_buffer, buffers, 1399 sizeof(struct pipe_vertex_buffer) * count); 1400 r300->vertex_buffer_count = count; 1401} 1402 1403/* Initialize the PSC tables. */ 1404static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1405{ 1406 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1407 uint16_t type, swizzle; 1408 enum pipe_format format; 1409 unsigned i; 1410 1411 if (velems->count > 16) { 1412 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1413 " requested %i, using 16.\n", velems->count); 1414 velems->count = 16; 1415 } 1416 1417 /* Vertex shaders have no semantics on their inputs, 1418 * so PSC should just route stuff based on the vertex elements, 1419 * and not on attrib information. */ 1420 for (i = 0; i < velems->count; i++) { 1421 format = velems->hw_format[i]; 1422 1423 type = r300_translate_vertex_data_type(format); 1424 if (type == R300_INVALID_FORMAT) { 1425 fprintf(stderr, "r300: Bad vertex format %s.\n", 1426 util_format_short_name(format)); 1427 assert(0); 1428 abort(); 1429 } 1430 1431 type |= i << R300_DST_VEC_LOC_SHIFT; 1432 swizzle = r300_translate_vertex_data_swizzle(format); 1433 1434 if (i & 1) { 1435 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1436 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1437 } else { 1438 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1439 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1440 } 1441 } 1442 1443 /* Set the last vector in the PSC. */ 1444 if (i) { 1445 i -= 1; 1446 } 1447 vstream->vap_prog_stream_cntl[i >> 1] |= 1448 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1449 1450 vstream->count = (i >> 1) + 1; 1451} 1452 1453#define FORMAT_REPLACE(what, withwhat) \ 1454 case PIPE_FORMAT_##what: *format = PIPE_FORMAT_##withwhat; break 1455 1456static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1457 unsigned count, 1458 const struct pipe_vertex_element* attribs) 1459{ 1460 struct r300_vertex_element_state *velems; 1461 unsigned i; 1462 enum pipe_format *format; 1463 1464 assert(count <= PIPE_MAX_ATTRIBS); 1465 velems = CALLOC_STRUCT(r300_vertex_element_state); 1466 if (velems != NULL) { 1467 velems->count = count; 1468 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1469 1470 if (r300_screen(pipe->screen)->caps.has_tcl) { 1471 /* Set the best hw format in case the original format is not 1472 * supported by hw. */ 1473 for (i = 0; i < count; i++) { 1474 velems->hw_format[i] = velems->velem[i].src_format; 1475 format = &velems->hw_format[i]; 1476 1477 /* This is basically the list of unsupported formats. 1478 * For now we don't care about the alignment, that's going to 1479 * be sorted out after the PSC setup. */ 1480 switch (*format) { 1481 FORMAT_REPLACE(R64_FLOAT, R32_FLOAT); 1482 FORMAT_REPLACE(R64G64_FLOAT, R32G32_FLOAT); 1483 FORMAT_REPLACE(R64G64B64_FLOAT, R32G32B32_FLOAT); 1484 FORMAT_REPLACE(R64G64B64A64_FLOAT, R32G32B32A32_FLOAT); 1485 1486 FORMAT_REPLACE(R32_UNORM, R32_FLOAT); 1487 FORMAT_REPLACE(R32G32_UNORM, R32G32_FLOAT); 1488 FORMAT_REPLACE(R32G32B32_UNORM, R32G32B32_FLOAT); 1489 FORMAT_REPLACE(R32G32B32A32_UNORM, R32G32B32A32_FLOAT); 1490 1491 FORMAT_REPLACE(R32_USCALED, R32_FLOAT); 1492 FORMAT_REPLACE(R32G32_USCALED, R32G32_FLOAT); 1493 FORMAT_REPLACE(R32G32B32_USCALED, R32G32B32_FLOAT); 1494 FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT); 1495 1496 FORMAT_REPLACE(R32_SNORM, R32_FLOAT); 1497 FORMAT_REPLACE(R32G32_SNORM, R32G32_FLOAT); 1498 FORMAT_REPLACE(R32G32B32_SNORM, R32G32B32_FLOAT); 1499 FORMAT_REPLACE(R32G32B32A32_SNORM, R32G32B32A32_FLOAT); 1500 1501 FORMAT_REPLACE(R32_SSCALED, R32_FLOAT); 1502 FORMAT_REPLACE(R32G32_SSCALED, R32G32_FLOAT); 1503 FORMAT_REPLACE(R32G32B32_SSCALED, R32G32B32_FLOAT); 1504 FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT); 1505 1506 FORMAT_REPLACE(R32_FIXED, R32_FLOAT); 1507 FORMAT_REPLACE(R32G32_FIXED, R32G32_FLOAT); 1508 FORMAT_REPLACE(R32G32B32_FIXED, R32G32B32_FLOAT); 1509 FORMAT_REPLACE(R32G32B32A32_FIXED, R32G32B32A32_FLOAT); 1510 1511 default:; 1512 } 1513 1514 velems->incompatible_layout = 1515 velems->incompatible_layout || 1516 velems->velem[i].src_format != velems->hw_format[i] || 1517 velems->velem[i].src_offset % 4 != 0; 1518 } 1519 1520 /* Now setup PSC. 1521 * The unused components will be replaced by (..., 0, 1). */ 1522 r300_vertex_psc(velems); 1523 1524 /* Align the formats to the size of DWORD. 1525 * We only care about the blocksizes of the formats since 1526 * swizzles are already set up. 1527 * Also compute the vertex size. */ 1528 for (i = 0; i < count; i++) { 1529 /* This is OK because we check for aligned strides too. */ 1530 velems->hw_format_size[i] = 1531 align(util_format_get_blocksize(velems->hw_format[i]), 4); 1532 velems->vertex_size_dwords += velems->hw_format_size[i] / 4; 1533 } 1534 } 1535 } 1536 return velems; 1537} 1538 1539static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1540 void *state) 1541{ 1542 struct r300_context *r300 = r300_context(pipe); 1543 struct r300_vertex_element_state *velems = state; 1544 1545 if (velems == NULL) { 1546 return; 1547 } 1548 1549 r300->velems = velems; 1550 1551 if (r300->draw) { 1552 draw_flush(r300->draw); 1553 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1554 return; 1555 } 1556 1557 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1558 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1559} 1560 1561static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1562{ 1563 FREE(state); 1564} 1565 1566static void* r300_create_vs_state(struct pipe_context* pipe, 1567 const struct pipe_shader_state* shader) 1568{ 1569 struct r300_context* r300 = r300_context(pipe); 1570 1571 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1572 1573 /* Copy state directly into shader. */ 1574 vs->state = *shader; 1575 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1576 1577 if (r300->screen->caps.has_tcl) { 1578 r300_init_vs_outputs(vs); 1579 r300_translate_vertex_shader(r300, vs); 1580 } else { 1581 r300_draw_init_vertex_shader(r300->draw, vs); 1582 } 1583 1584 return vs; 1585} 1586 1587static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1588{ 1589 struct r300_context* r300 = r300_context(pipe); 1590 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1591 1592 if (vs == NULL) { 1593 r300->vs_state.state = NULL; 1594 return; 1595 } 1596 if (vs == r300->vs_state.state) { 1597 return; 1598 } 1599 r300->vs_state.state = vs; 1600 1601 /* The majority of the RS block bits is dependent on the vertex shader. */ 1602 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1603 1604 if (r300->screen->caps.has_tcl) { 1605 r300->vs_state.dirty = TRUE; 1606 r300->vs_state.size = 1607 vs->code.length + 18 + 1608 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1609 1610 if (vs->externals_count) { 1611 r300->vs_constants.dirty = TRUE; 1612 r300->vs_constants.size = vs->externals_count * 4 + 3; 1613 } else { 1614 r300->vs_constants.size = 0; 1615 } 1616 1617 r300->pvs_flush.dirty = TRUE; 1618 } else { 1619 draw_flush(r300->draw); 1620 draw_bind_vertex_shader(r300->draw, 1621 (struct draw_vertex_shader*)vs->draw_vs); 1622 } 1623} 1624 1625static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1626{ 1627 struct r300_context* r300 = r300_context(pipe); 1628 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1629 1630 if (r300->screen->caps.has_tcl) { 1631 rc_constants_destroy(&vs->code.constants); 1632 } else { 1633 draw_delete_vertex_shader(r300->draw, 1634 (struct draw_vertex_shader*)vs->draw_vs); 1635 } 1636 1637 FREE((void*)vs->state.tokens); 1638 FREE(shader); 1639} 1640 1641static void r300_set_constant_buffer(struct pipe_context *pipe, 1642 uint shader, uint index, 1643 struct pipe_resource *buf) 1644{ 1645 struct r300_context* r300 = r300_context(pipe); 1646 struct r300_constant_buffer *cbuf; 1647 struct pipe_transfer *tr; 1648 float *mapped; 1649 int max_size = 0, max_size_bytes = 0, clamped_size = 0; 1650 1651 switch (shader) { 1652 case PIPE_SHADER_VERTEX: 1653 cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; 1654 max_size = 256; 1655 break; 1656 case PIPE_SHADER_FRAGMENT: 1657 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1658 if (r300->screen->caps.is_r500) { 1659 max_size = 256; 1660 } else { 1661 max_size = 32; 1662 } 1663 break; 1664 default: 1665 assert(0); 1666 return; 1667 } 1668 max_size_bytes = max_size * 4 * sizeof(float); 1669 1670 if (buf == NULL || buf->width0 == 0 || 1671 (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL) 1672 { 1673 cbuf->count = 0; 1674 return; 1675 } 1676 1677 if (shader == PIPE_SHADER_FRAGMENT || 1678 (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { 1679 assert((buf->width0 % (4 * sizeof(float))) == 0); 1680 1681 /* Check the size of the constant buffer. */ 1682 /* XXX Subtract immediates and RC_STATE_* variables. */ 1683 if (buf->width0 > max_size_bytes) { 1684 fprintf(stderr, "r300: Max size of the constant buffer is " 1685 "%i*4 floats.\n", max_size); 1686 } 1687 1688 clamped_size = MIN2(buf->width0, max_size_bytes); 1689 cbuf->count = clamped_size / (4 * sizeof(float)); 1690 1691 if (shader == PIPE_SHADER_FRAGMENT && !r300->screen->caps.is_r500) { 1692 unsigned i,j; 1693 1694 /* Convert constants to float24. */ 1695 for (i = 0; i < cbuf->count; i++) 1696 for (j = 0; j < 4; j++) 1697 cbuf->constants[i][j] = pack_float24(mapped[i*4+j]); 1698 } else { 1699 memcpy(cbuf->constants, mapped, clamped_size); 1700 } 1701 } 1702 1703 if (shader == PIPE_SHADER_VERTEX) { 1704 if (r300->screen->caps.has_tcl) { 1705 if (r300->vs_constants.size) { 1706 r300->vs_constants.dirty = TRUE; 1707 } 1708 r300->pvs_flush.dirty = TRUE; 1709 } else if (r300->draw) { 1710 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1711 0, mapped, buf->width0); 1712 } 1713 } else if (shader == PIPE_SHADER_FRAGMENT) { 1714 r300->fs_constants.dirty = TRUE; 1715 } 1716 1717 pipe_buffer_unmap(pipe, buf, tr); 1718} 1719 1720void r300_init_state_functions(struct r300_context* r300) 1721{ 1722 r300->context.create_blend_state = r300_create_blend_state; 1723 r300->context.bind_blend_state = r300_bind_blend_state; 1724 r300->context.delete_blend_state = r300_delete_blend_state; 1725 1726 r300->context.set_blend_color = r300_set_blend_color; 1727 1728 r300->context.set_clip_state = r300_set_clip_state; 1729 r300->context.set_sample_mask = r300_set_sample_mask; 1730 1731 r300->context.set_constant_buffer = r300_set_constant_buffer; 1732 1733 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1734 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1735 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1736 1737 r300->context.set_stencil_ref = r300_set_stencil_ref; 1738 1739 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1740 1741 r300->context.create_fs_state = r300_create_fs_state; 1742 r300->context.bind_fs_state = r300_bind_fs_state; 1743 r300->context.delete_fs_state = r300_delete_fs_state; 1744 1745 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1746 1747 r300->context.create_rasterizer_state = r300_create_rs_state; 1748 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1749 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1750 1751 r300->context.create_sampler_state = r300_create_sampler_state; 1752 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1753 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1754 r300->context.delete_sampler_state = r300_delete_sampler_state; 1755 1756 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1757 r300->context.create_sampler_view = r300_create_sampler_view; 1758 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1759 1760 r300->context.set_scissor_state = r300_set_scissor_state; 1761 1762 r300->context.set_viewport_state = r300_set_viewport_state; 1763 1764 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1765 1766 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1767 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1768 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1769 1770 r300->context.create_vs_state = r300_create_vs_state; 1771 r300->context.bind_vs_state = r300_bind_vs_state; 1772 r300->context.delete_vs_state = r300_delete_vs_state; 1773} 1774