r300_state.c revision fc20efe8e511bb9ec15c3d70e28b348ddaa4ad37
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#include "pipe/internal/p_winsys_screen.h" 34 35#include "r300_context.h" 36#include "r300_reg.h" 37#include "r300_screen.h" 38#include "r300_state_inlines.h" 39#include "r300_fs.h" 40#include "r300_vs.h" 41 42/* r300_state: Functions used to intialize state context by translating 43 * Gallium state objects into semi-native r300 state objects. */ 44 45static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 46 unsigned dstRGB, unsigned dstA) 47{ 48 /* If the blend equation is ADD or REVERSE_SUBTRACT, 49 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 50 * will not be changed. 51 * Notice that the dst factors are the src factors inverted. */ 52 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 53 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 54 srcRGB == PIPE_BLENDFACTOR_ZERO) && 55 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 56 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 57 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 58 srcA == PIPE_BLENDFACTOR_ZERO) && 59 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 60 dstRGB == PIPE_BLENDFACTOR_ONE) && 61 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 62 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 63 dstA == PIPE_BLENDFACTOR_ONE); 64} 65 66static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 67 unsigned dstRGB, unsigned dstA) 68{ 69 /* If the blend equation is ADD or REVERSE_SUBTRACT, 70 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 71 * will not be changed. 72 * Notice that the dst factors are the src factors inverted. */ 73 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 74 srcRGB == PIPE_BLENDFACTOR_ZERO) && 75 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 76 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 77 srcA == PIPE_BLENDFACTOR_ZERO) && 78 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 79 dstRGB == PIPE_BLENDFACTOR_ONE) && 80 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 81 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 82 dstA == PIPE_BLENDFACTOR_ONE); 83} 84 85static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 86 unsigned dstRGB, unsigned dstA) 87{ 88 /* If the blend equation is ADD or REVERSE_SUBTRACT, 89 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 90 * will not be changed. 91 * Notice that the dst factors are the src factors inverted. */ 92 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 93 srcRGB == PIPE_BLENDFACTOR_ZERO) && 94 (srcA == PIPE_BLENDFACTOR_ZERO) && 95 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 96 dstRGB == PIPE_BLENDFACTOR_ONE) && 97 (dstA == PIPE_BLENDFACTOR_ONE); 98} 99 100static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 101 unsigned dstRGB, unsigned dstA) 102{ 103 /* If the blend equation is ADD or REVERSE_SUBTRACT, 104 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 105 * will not be changed. 106 * Notice that the dst factors are the src factors inverted. */ 107 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 108 srcRGB == PIPE_BLENDFACTOR_ZERO) && 109 (srcA == PIPE_BLENDFACTOR_ZERO) && 110 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 111 dstRGB == PIPE_BLENDFACTOR_ONE) && 112 (dstA == PIPE_BLENDFACTOR_ONE); 113} 114 115static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 116 unsigned dstRGB, unsigned dstA) 117{ 118 /* If the blend equation is ADD or REVERSE_SUBTRACT, 119 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 120 * the colorbuffer will not be changed. 121 * Notice that the dst factors are the src factors inverted. */ 122 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 123 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 124 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 125 srcRGB == PIPE_BLENDFACTOR_ZERO) && 126 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 127 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 128 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 129 srcA == PIPE_BLENDFACTOR_ZERO) && 130 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 131 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 132 dstRGB == PIPE_BLENDFACTOR_ONE) && 133 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 134 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 135 dstA == PIPE_BLENDFACTOR_ONE); 136} 137 138static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 139 unsigned dstRGB, unsigned dstA) 140{ 141 /* If the blend equation is ADD or REVERSE_SUBTRACT, 142 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 143 * the colorbuffer will not be changed. 144 * Notice that the dst factors are the src factors inverted. */ 145 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 146 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 147 srcRGB == PIPE_BLENDFACTOR_ZERO) && 148 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 149 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 150 srcA == PIPE_BLENDFACTOR_ZERO) && 151 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 152 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 153 dstRGB == PIPE_BLENDFACTOR_ONE) && 154 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 155 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 156 dstA == PIPE_BLENDFACTOR_ONE); 157} 158 159/* Create a new blend state based on the CSO blend state. 160 * 161 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 162static void* r300_create_blend_state(struct pipe_context* pipe, 163 const struct pipe_blend_state* state) 164{ 165 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 166 167 if (state->blend_enable) 168 { 169 unsigned eqRGB = state->rgb_func; 170 unsigned srcRGB = state->rgb_src_factor; 171 unsigned dstRGB = state->rgb_dst_factor; 172 173 unsigned eqA = state->alpha_func; 174 unsigned srcA = state->alpha_src_factor; 175 unsigned dstA = state->alpha_dst_factor; 176 177 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 178 * this is just the crappy D3D naming */ 179 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 180 r300_translate_blend_function(eqRGB) | 181 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 182 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 183 184 /* Optimization: some operations do not require the destination color. 185 * 186 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 187 * otherwise blending gives incorrect results. It seems to be 188 * a hardware bug. */ 189 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 190 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 191 dstRGB != PIPE_BLENDFACTOR_ZERO || 192 dstA != PIPE_BLENDFACTOR_ZERO || 193 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 194 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 195 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 196 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 197 srcA == PIPE_BLENDFACTOR_DST_COLOR || 198 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 199 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 200 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 201 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 202 /* Enable reading from the colorbuffer. */ 203 blend->blend_control |= R300_READ_ENABLE; 204 205 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) { 206 /* Optimization: Depending on incoming pixels, we can 207 * conditionally disable the reading in hardware... */ 208 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 209 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 210 /* Disable reading if SRC_ALPHA == 0. */ 211 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 212 dstRGB == PIPE_BLENDFACTOR_ZERO) && 213 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 214 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 215 dstA == PIPE_BLENDFACTOR_ZERO)) { 216 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 217 } 218 219 /* Disable reading if SRC_ALPHA == 1. */ 220 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 221 dstRGB == PIPE_BLENDFACTOR_ZERO) && 222 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 223 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 224 dstA == PIPE_BLENDFACTOR_ZERO)) { 225 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 226 } 227 } 228 } 229 } 230 231 /* Optimization: discard pixels which don't change the colorbuffer. 232 * 233 * The code below is non-trivial and some math is involved. 234 * 235 * Discarding pixels must be disabled when FP16 AA is enabled. 236 * This is a hardware bug. Also, this implementation wouldn't work 237 * with FP blending enabled and equation clamping disabled. 238 * 239 * Equations other than ADD are rarely used and therefore won't be 240 * optimized. */ 241 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 242 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 243 /* ADD: X+Y 244 * REVERSE_SUBTRACT: Y-X 245 * 246 * The idea is: 247 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 248 * then CB will not be changed. 249 * 250 * Given the srcFactor and dstFactor variables, we can derive 251 * what src and dst should be equal to and discard appropriate 252 * pixels. 253 */ 254 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 255 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 256 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 257 dstRGB, dstA)) { 258 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 259 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 260 dstRGB, dstA)) { 261 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 262 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 263 dstRGB, dstA)) { 264 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 265 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 266 dstRGB, dstA)) { 267 blend->blend_control |= 268 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 269 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 270 dstRGB, dstA)) { 271 blend->blend_control |= 272 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 273 } 274 } 275 276 /* separate alpha */ 277 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 278 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 279 blend->alpha_blend_control = 280 r300_translate_blend_function(eqA) | 281 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 282 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 283 } 284 } 285 286 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 287 if (state->logicop_enable) { 288 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 289 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 290 } 291 292 /* Color Channel Mask */ 293 if (state->colormask & PIPE_MASK_R) { 294 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0; 295 } 296 if (state->colormask & PIPE_MASK_G) { 297 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0; 298 } 299 if (state->colormask & PIPE_MASK_B) { 300 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0; 301 } 302 if (state->colormask & PIPE_MASK_A) { 303 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0; 304 } 305 306 if (state->dither) { 307 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 308 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 309 } 310 311 return (void*)blend; 312} 313 314/* Bind blend state. */ 315static void r300_bind_blend_state(struct pipe_context* pipe, 316 void* state) 317{ 318 struct r300_context* r300 = r300_context(pipe); 319 320 r300->blend_state.state = state; 321 r300->blend_state.dirty = TRUE; 322} 323 324/* Free blend state. */ 325static void r300_delete_blend_state(struct pipe_context* pipe, 326 void* state) 327{ 328 FREE(state); 329} 330 331/* Convert float to 10bit integer */ 332static unsigned float_to_fixed10(float f) 333{ 334 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 335} 336 337/* Set blend color. 338 * Setup both R300 and R500 registers, figure out later which one to write. */ 339static void r300_set_blend_color(struct pipe_context* pipe, 340 const struct pipe_blend_color* color) 341{ 342 struct r300_context* r300 = r300_context(pipe); 343 struct r300_screen* r300screen = r300_screen(pipe->screen); 344 struct r300_blend_color_state* state = 345 (struct r300_blend_color_state*)r300->blend_color_state.state; 346 union util_color uc; 347 348 util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); 349 state->blend_color = uc.ui; 350 351 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 352 state->blend_color_red_alpha = 353 float_to_fixed10(color->color[0]) | 354 (float_to_fixed10(color->color[3]) << 16); 355 state->blend_color_green_blue = 356 float_to_fixed10(color->color[2]) | 357 (float_to_fixed10(color->color[1]) << 16); 358 359 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; 360 r300->blend_color_state.dirty = TRUE; 361} 362 363static void r300_set_clip_state(struct pipe_context* pipe, 364 const struct pipe_clip_state* state) 365{ 366 struct r300_context* r300 = r300_context(pipe); 367 368 if (r300_screen(pipe->screen)->caps->has_tcl) { 369 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 370 r300->clip_state.size = 29; 371 } else { 372 draw_flush(r300->draw); 373 draw_set_clip_state(r300->draw, state); 374 r300->clip_state.size = 2; 375 } 376 377 r300->clip_state.dirty = TRUE; 378} 379 380/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 381 * 382 * This contains the depth buffer, stencil buffer, alpha test, and such. 383 * On the Radeon, depth and stencil buffer setup are intertwined, which is 384 * the reason for some of the strange-looking assignments across registers. */ 385static void* 386 r300_create_dsa_state(struct pipe_context* pipe, 387 const struct pipe_depth_stencil_alpha_state* state) 388{ 389 struct r300_capabilities *caps = 390 r300_screen(r300_context(pipe)->context.screen)->caps; 391 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 392 393 /* Depth test setup. */ 394 if (state->depth.enabled) { 395 dsa->z_buffer_control |= R300_Z_ENABLE; 396 397 if (state->depth.writemask) { 398 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 399 } 400 401 dsa->z_stencil_control |= 402 (r300_translate_depth_stencil_function(state->depth.func) << 403 R300_Z_FUNC_SHIFT); 404 } 405 406 /* Stencil buffer setup. */ 407 if (state->stencil[0].enabled) { 408 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 409 dsa->z_stencil_control |= 410 (r300_translate_depth_stencil_function(state->stencil[0].func) << 411 R300_S_FRONT_FUNC_SHIFT) | 412 (r300_translate_stencil_op(state->stencil[0].fail_op) << 413 R300_S_FRONT_SFAIL_OP_SHIFT) | 414 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 415 R300_S_FRONT_ZPASS_OP_SHIFT) | 416 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 417 R300_S_FRONT_ZFAIL_OP_SHIFT); 418 419 dsa->stencil_ref_mask = (state->stencil[0].ref_value) | 420 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 421 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 422 423 if (state->stencil[1].enabled) { 424 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 425 dsa->z_stencil_control |= 426 (r300_translate_depth_stencil_function(state->stencil[1].func) << 427 R300_S_BACK_FUNC_SHIFT) | 428 (r300_translate_stencil_op(state->stencil[1].fail_op) << 429 R300_S_BACK_SFAIL_OP_SHIFT) | 430 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 431 R300_S_BACK_ZPASS_OP_SHIFT) | 432 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 433 R300_S_BACK_ZFAIL_OP_SHIFT); 434 435 /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */ 436 if (caps->is_r500) 437 { 438 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 439 dsa->stencil_ref_bf = (state->stencil[1].ref_value) | 440 (state->stencil[1].valuemask << 441 R300_STENCILMASK_SHIFT) | 442 (state->stencil[1].writemask << 443 R300_STENCILWRITEMASK_SHIFT); 444 } 445 } 446 } 447 448 /* Alpha test setup. */ 449 if (state->alpha.enabled) { 450 dsa->alpha_function = 451 r300_translate_alpha_function(state->alpha.func) | 452 R300_FG_ALPHA_FUNC_ENABLE; 453 454 /* XXX figure out why emitting 10bit alpha ref causes CS to dump */ 455 /* always use 8bit alpha ref */ 456 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 457 458 if (caps->is_r500) 459 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 460 } 461 462 return (void*)dsa; 463} 464 465/* Bind DSA state. */ 466static void r300_bind_dsa_state(struct pipe_context* pipe, 467 void* state) 468{ 469 struct r300_context* r300 = r300_context(pipe); 470 struct r300_screen* r300screen = r300_screen(pipe->screen); 471 472 r300->dsa_state.state = state; 473 r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; 474 r300->dsa_state.dirty = TRUE; 475} 476 477/* Free DSA state. */ 478static void r300_delete_dsa_state(struct pipe_context* pipe, 479 void* state) 480{ 481 FREE(state); 482} 483 484static void 485 r300_set_framebuffer_state(struct pipe_context* pipe, 486 const struct pipe_framebuffer_state* state) 487{ 488 struct r300_context* r300 = r300_context(pipe); 489 uint32_t zbuffer_bpp = 0; 490 491 if (r300->draw) { 492 draw_flush(r300->draw); 493 } 494 495 r300->framebuffer_state = *state; 496 497 /* Don't rely on the order of states being set for the first time. */ 498 r300->dirty_state |= R300_NEW_FRAMEBUFFERS; 499 500 r300->blend_state.dirty = TRUE; 501 r300->dsa_state.dirty = TRUE; 502 r300->scissor_state.dirty = TRUE; 503 504 /* Polyfon offset depends on the zbuffer bit depth. */ 505 if (state->zsbuf && r300->polygon_offset_enabled) { 506 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 507 case 2: 508 zbuffer_bpp = 16; 509 break; 510 case 4: 511 zbuffer_bpp = 24; 512 break; 513 } 514 515 if (r300->zbuffer_bpp != zbuffer_bpp) { 516 r300->zbuffer_bpp = zbuffer_bpp; 517 r300->rs_state.dirty = TRUE; 518 } 519 } 520} 521 522/* Create fragment shader state. */ 523static void* r300_create_fs_state(struct pipe_context* pipe, 524 const struct pipe_shader_state* shader) 525{ 526 struct r300_fragment_shader* fs = NULL; 527 528 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 529 530 /* Copy state directly into shader. */ 531 fs->state = *shader; 532 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 533 534 tgsi_scan_shader(shader->tokens, &fs->info); 535 r300_shader_read_fs_inputs(&fs->info, &fs->inputs); 536 537 return (void*)fs; 538} 539 540/* Bind fragment shader state. */ 541static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 542{ 543 struct r300_context* r300 = r300_context(pipe); 544 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 545 546 if (fs == NULL) { 547 r300->fs = NULL; 548 return; 549 } 550 551 r300->fs = fs; 552 r300_pick_fragment_shader(r300); 553 554 if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { 555 r300->vertex_format_state.dirty = TRUE; 556 } 557 558 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; 559} 560 561/* Delete fragment shader state. */ 562static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 563{ 564 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 565 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 566 567 while (ptr) { 568 tmp = ptr; 569 ptr = ptr->next; 570 rc_constants_destroy(&tmp->code.constants); 571 FREE(tmp); 572 } 573 FREE((void*)fs->state.tokens); 574 FREE(shader); 575} 576 577static void r300_set_polygon_stipple(struct pipe_context* pipe, 578 const struct pipe_poly_stipple* state) 579{ 580 /* XXX no idea how to set this up, but not terribly important */ 581} 582 583/* Create a new rasterizer state based on the CSO rasterizer state. 584 * 585 * This is a very large chunk of state, and covers most of the graphics 586 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 587 * 588 * In a not entirely unironic sidenote, this state has nearly nothing to do 589 * with the actual block on the Radeon called the rasterizer (RS). */ 590static void* r300_create_rs_state(struct pipe_context* pipe, 591 const struct pipe_rasterizer_state* state) 592{ 593 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 594 595 /* Copy rasterizer state for Draw. */ 596 rs->rs = *state; 597 598#ifdef PIPE_ARCH_LITTLE_ENDIAN 599 rs->vap_control_status = R300_VC_NO_SWAP; 600#else 601 rs->vap_control_status = R300_VC_32BIT_SWAP; 602#endif 603 604 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. 605 * Else, enable HW TCL and force Draw's TCL off. */ 606 if (state->bypass_vs_clip_and_viewport || 607 !r300_screen(pipe->screen)->caps->has_tcl) { 608 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 609 } 610 611 rs->point_size = pack_float_16_6x(state->point_size) | 612 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 613 614 rs->point_minmax = 615 ((int)(state->point_size_min * 6.0) << 616 R300_GA_POINT_MINMAX_MIN_SHIFT) | 617 ((int)(state->point_size_max * 6.0) << 618 R300_GA_POINT_MINMAX_MAX_SHIFT); 619 620 rs->line_control = pack_float_16_6x(state->line_width) | 621 R300_GA_LINE_CNTL_END_TYPE_COMP; 622 623 /* Enable polygon mode */ 624 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 625 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 626 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 627 } 628 629 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 630 if (state->front_winding == PIPE_WINDING_CW) { 631 rs->cull_mode = R300_FRONT_FACE_CW; 632 633 /* Polygon offset */ 634 if (state->offset_cw) { 635 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 636 } 637 if (state->offset_ccw) { 638 rs->polygon_offset_enable |= R300_BACK_ENABLE; 639 } 640 641 /* Polygon mode */ 642 if (rs->polygon_mode) { 643 rs->polygon_mode |= 644 r300_translate_polygon_mode_front(state->fill_cw); 645 rs->polygon_mode |= 646 r300_translate_polygon_mode_back(state->fill_ccw); 647 } 648 } else { 649 rs->cull_mode = R300_FRONT_FACE_CCW; 650 651 /* Polygon offset */ 652 if (state->offset_ccw) { 653 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 654 } 655 if (state->offset_cw) { 656 rs->polygon_offset_enable |= R300_BACK_ENABLE; 657 } 658 659 /* Polygon mode */ 660 if (rs->polygon_mode) { 661 rs->polygon_mode |= 662 r300_translate_polygon_mode_front(state->fill_ccw); 663 rs->polygon_mode |= 664 r300_translate_polygon_mode_back(state->fill_cw); 665 } 666 } 667 if (state->front_winding & state->cull_mode) { 668 rs->cull_mode |= R300_CULL_FRONT; 669 } 670 if (~(state->front_winding) & state->cull_mode) { 671 rs->cull_mode |= R300_CULL_BACK; 672 } 673 674 if (rs->polygon_offset_enable) { 675 rs->depth_offset = state->offset_units; 676 rs->depth_scale = state->offset_scale; 677 } 678 679 if (state->line_stipple_enable) { 680 rs->line_stipple_config = 681 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 682 (fui((float)state->line_stipple_factor) & 683 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 684 /* XXX this might need to be scaled up */ 685 rs->line_stipple_value = state->line_stipple_pattern; 686 } 687 688 if (state->flatshade) { 689 rs->color_control = R300_SHADE_MODEL_FLAT; 690 } else { 691 rs->color_control = R300_SHADE_MODEL_SMOOTH; 692 } 693 694 return (void*)rs; 695} 696 697/* Bind rasterizer state. */ 698static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 699{ 700 struct r300_context* r300 = r300_context(pipe); 701 struct r300_rs_state* rs = (struct r300_rs_state*)state; 702 703 if (r300->draw) { 704 draw_flush(r300->draw); 705 draw_set_rasterizer_state(r300->draw, &rs->rs); 706 } 707 708 if (rs) { 709 r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; 710 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 711 } else { 712 r300->tcl_bypass = FALSE; 713 r300->polygon_offset_enabled = FALSE; 714 } 715 716 r300->rs_state.state = rs; 717 r300->rs_state.dirty = TRUE; 718 /* XXX Why is this still needed, dammit!? */ 719 r300->scissor_state.dirty = TRUE; 720 r300->viewport_state.dirty = TRUE; 721 722 /* XXX Clean these up when we move to atom emits */ 723 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 724 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 725 } 726} 727 728/* Free rasterizer state. */ 729static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 730{ 731 FREE(state); 732} 733 734static void* 735 r300_create_sampler_state(struct pipe_context* pipe, 736 const struct pipe_sampler_state* state) 737{ 738 struct r300_context* r300 = r300_context(pipe); 739 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 740 int lod_bias; 741 union util_color uc; 742 743 sampler->state = *state; 744 745 sampler->filter0 |= 746 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 747 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 748 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 749 750 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 751 state->mag_img_filter, 752 state->min_mip_filter, 753 state->max_anisotropy > 1.0); 754 755 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 756 /* We must pass these to the emit function to clamp them properly. */ 757 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 758 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 759 760 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 761 762 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 763 764 sampler->filter1 |= r300_anisotropy(state->max_anisotropy); 765 766 util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); 767 sampler->border_color = uc.ui; 768 769 /* R500-specific fixups and optimizations */ 770 if (r300_screen(r300->context.screen)->caps->is_r500) { 771 sampler->filter1 |= R500_BORDER_FIX; 772 } 773 774 return (void*)sampler; 775} 776 777static void r300_bind_sampler_states(struct pipe_context* pipe, 778 unsigned count, 779 void** states) 780{ 781 struct r300_context* r300 = r300_context(pipe); 782 int i; 783 784 if (count > 8) { 785 return; 786 } 787 788 for (i = 0; i < count; i++) { 789 if (r300->sampler_states[i] != states[i]) { 790 r300->sampler_states[i] = (struct r300_sampler_state*)states[i]; 791 r300->dirty_state |= (R300_NEW_SAMPLER << i); 792 } 793 } 794 795 r300->sampler_count = count; 796 797 /* Pick a fragment shader based on the texture compare state. */ 798 if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) { 799 if (r300_pick_fragment_shader(r300)) { 800 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | 801 R300_NEW_FRAGMENT_SHADER_CONSTANTS; 802 } 803 } 804} 805 806static void r300_lacks_vertex_textures(struct pipe_context* pipe, 807 unsigned count, 808 void** states) 809{ 810} 811 812static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 813{ 814 FREE(state); 815} 816 817static void r300_set_sampler_textures(struct pipe_context* pipe, 818 unsigned count, 819 struct pipe_texture** texture) 820{ 821 struct r300_context* r300 = r300_context(pipe); 822 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; 823 int i; 824 825 /* XXX magic num */ 826 if (count > 8) { 827 return; 828 } 829 830 for (i = 0; i < count; i++) { 831 if (r300->textures[i] != (struct r300_texture*)texture[i]) { 832 pipe_texture_reference((struct pipe_texture**)&r300->textures[i], 833 texture[i]); 834 r300->dirty_state |= (R300_NEW_TEXTURE << i); 835 836 /* R300-specific - set the texrect factor in a fragment shader */ 837 if (!is_r500 && r300->textures[i]->is_npot) { 838 /* XXX It would be nice to re-emit just 1 constant, 839 * XXX not all of them */ 840 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 841 } 842 } 843 } 844 845 for (i = count; i < 8; i++) { 846 if (r300->textures[i]) { 847 pipe_texture_reference((struct pipe_texture**)&r300->textures[i], 848 NULL); 849 r300->dirty_state |= (R300_NEW_TEXTURE << i); 850 } 851 } 852 853 r300->texture_count = count; 854} 855 856static void r300_set_scissor_state(struct pipe_context* pipe, 857 const struct pipe_scissor_state* state) 858{ 859 struct r300_context* r300 = r300_context(pipe); 860 861 memcpy(r300->scissor_state.state, state, 862 sizeof(struct pipe_scissor_state)); 863 864 r300->scissor_state.dirty = TRUE; 865} 866 867static void r300_set_viewport_state(struct pipe_context* pipe, 868 const struct pipe_viewport_state* state) 869{ 870 struct r300_context* r300 = r300_context(pipe); 871 struct r300_viewport_state* viewport = 872 (struct r300_viewport_state*)r300->viewport_state.state; 873 874 /* Do the transform in HW. */ 875 viewport->vte_control = R300_VTX_W0_FMT; 876 877 if (state->scale[0] != 1.0f) { 878 viewport->xscale = state->scale[0]; 879 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 880 } 881 if (state->scale[1] != 1.0f) { 882 viewport->yscale = state->scale[1]; 883 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 884 } 885 if (state->scale[2] != 1.0f) { 886 viewport->zscale = state->scale[2]; 887 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 888 } 889 if (state->translate[0] != 0.0f) { 890 viewport->xoffset = state->translate[0]; 891 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 892 } 893 if (state->translate[1] != 0.0f) { 894 viewport->yoffset = state->translate[1]; 895 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 896 } 897 if (state->translate[2] != 0.0f) { 898 viewport->zoffset = state->translate[2]; 899 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 900 } 901 902 r300->viewport_state.dirty = TRUE; 903 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 904 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 905 } 906} 907 908static void r300_set_vertex_buffers(struct pipe_context* pipe, 909 unsigned count, 910 const struct pipe_vertex_buffer* buffers) 911{ 912 struct r300_context* r300 = r300_context(pipe); 913 914 memcpy(r300->vertex_buffer, buffers, 915 sizeof(struct pipe_vertex_buffer) * count); 916 r300->vertex_buffer_count = count; 917 918 if (r300->draw) { 919 draw_flush(r300->draw); 920 draw_set_vertex_buffers(r300->draw, count, buffers); 921 } 922 923 r300->vertex_format_state.dirty = TRUE; 924} 925 926static boolean r300_validate_aos(struct r300_context *r300) 927{ 928 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; 929 struct pipe_vertex_element *velem = r300->vertex_element; 930 int i; 931 932 /* Check if formats and strides are aligned to the size of DWORD. */ 933 for (i = 0; i < r300->vertex_element_count; i++) { 934 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || 935 util_format_get_blocksize(velem[i].src_format) % 4 != 0) { 936 return FALSE; 937 } 938 } 939 return TRUE; 940} 941 942static void r300_set_vertex_elements(struct pipe_context* pipe, 943 unsigned count, 944 const struct pipe_vertex_element* elements) 945{ 946 struct r300_context* r300 = r300_context(pipe); 947 948 memcpy(r300->vertex_element, 949 elements, 950 sizeof(struct pipe_vertex_element) * count); 951 r300->vertex_element_count = count; 952 953 if (r300->draw) { 954 draw_flush(r300->draw); 955 draw_set_vertex_elements(r300->draw, count, elements); 956 } 957 958 if (!r300_validate_aos(r300)) { 959 /* XXX We should fallback using draw. */ 960 assert(0); 961 abort(); 962 } 963} 964 965static void* r300_create_vs_state(struct pipe_context* pipe, 966 const struct pipe_shader_state* shader) 967{ 968 struct r300_context* r300 = r300_context(pipe); 969 970 if (r300_screen(pipe->screen)->caps->has_tcl) { 971 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 972 /* Copy state directly into shader. */ 973 vs->state = *shader; 974 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 975 976 tgsi_scan_shader(shader->tokens, &vs->info); 977 978 return (void*)vs; 979 } else { 980 return draw_create_vertex_shader(r300->draw, shader); 981 } 982} 983 984static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 985{ 986 struct r300_context* r300 = r300_context(pipe); 987 988 if (r300_screen(pipe->screen)->caps->has_tcl) { 989 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 990 991 if (vs == NULL) { 992 r300->vs = NULL; 993 return; 994 } else if (!vs->translated) { 995 r300_translate_vertex_shader(r300, vs); 996 } 997 998 r300->vs = vs; 999 if (r300->fs) { 1000 r300_vertex_shader_setup_wpos(r300); 1001 } 1002 1003 r300->vertex_format_state.dirty = TRUE; 1004 1005 r300->dirty_state |= 1006 R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; 1007 } else { 1008 draw_flush(r300->draw); 1009 draw_bind_vertex_shader(r300->draw, 1010 (struct draw_vertex_shader*)shader); 1011 } 1012} 1013 1014static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1015{ 1016 struct r300_context* r300 = r300_context(pipe); 1017 1018 if (r300_screen(pipe->screen)->caps->has_tcl) { 1019 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1020 1021 rc_constants_destroy(&vs->code.constants); 1022 FREE((void*)vs->state.tokens); 1023 FREE(shader); 1024 } else { 1025 draw_delete_vertex_shader(r300->draw, 1026 (struct draw_vertex_shader*)shader); 1027 } 1028} 1029 1030static void r300_set_constant_buffer(struct pipe_context *pipe, 1031 uint shader, uint index, 1032 struct pipe_buffer *buf) 1033{ 1034 struct r300_context* r300 = r300_context(pipe); 1035 void *mapped; 1036 1037 if (buf == NULL || buf->size == 0 || 1038 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) 1039 { 1040 r300->shader_constants[shader].count = 0; 1041 return; 1042 } 1043 1044 assert((buf->size % 4 * sizeof(float)) == 0); 1045 memcpy(r300->shader_constants[shader].constants, mapped, buf->size); 1046 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); 1047 pipe_buffer_unmap(pipe->screen, buf); 1048 1049 if (shader == PIPE_SHADER_VERTEX) 1050 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1051 else if (shader == PIPE_SHADER_FRAGMENT) 1052 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1053} 1054 1055void r300_init_state_functions(struct r300_context* r300) 1056{ 1057 r300->context.create_blend_state = r300_create_blend_state; 1058 r300->context.bind_blend_state = r300_bind_blend_state; 1059 r300->context.delete_blend_state = r300_delete_blend_state; 1060 1061 r300->context.set_blend_color = r300_set_blend_color; 1062 1063 r300->context.set_clip_state = r300_set_clip_state; 1064 1065 r300->context.set_constant_buffer = r300_set_constant_buffer; 1066 1067 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1068 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1069 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1070 1071 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1072 1073 r300->context.create_fs_state = r300_create_fs_state; 1074 r300->context.bind_fs_state = r300_bind_fs_state; 1075 r300->context.delete_fs_state = r300_delete_fs_state; 1076 1077 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1078 1079 r300->context.create_rasterizer_state = r300_create_rs_state; 1080 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1081 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1082 1083 r300->context.create_sampler_state = r300_create_sampler_state; 1084 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1085 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1086 r300->context.delete_sampler_state = r300_delete_sampler_state; 1087 1088 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures; 1089 1090 r300->context.set_scissor_state = r300_set_scissor_state; 1091 1092 r300->context.set_viewport_state = r300_set_viewport_state; 1093 1094 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1095 r300->context.set_vertex_elements = r300_set_vertex_elements; 1096 1097 r300->context.create_vs_state = r300_create_vs_state; 1098 r300->context.bind_vs_state = r300_bind_vs_state; 1099 r300->context.delete_vs_state = r300_delete_vs_state; 1100} 1101