r300_state.c revision 9b346f83a7b672e913a7bb6a089d5dbd7fbdce06
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_context.h" 35#include "r300_reg.h" 36#include "r300_screen.h" 37#include "r300_state_inlines.h" 38#include "r300_fs.h" 39#include "r300_vs.h" 40 41/* r300_state: Functions used to intialize state context by translating 42 * Gallium state objects into semi-native r300 state objects. */ 43 44static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, 45 unsigned dstRGB, unsigned dstA) 46{ 47 /* If the blend equation is ADD or REVERSE_SUBTRACT, 48 * SRC_ALPHA == 0, and the following state is set, the colorbuffer 49 * will not be changed. 50 * Notice that the dst factors are the src factors inverted. */ 51 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 52 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 53 srcRGB == PIPE_BLENDFACTOR_ZERO) && 54 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 55 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 56 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 57 srcA == PIPE_BLENDFACTOR_ZERO) && 58 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 59 dstRGB == PIPE_BLENDFACTOR_ONE) && 60 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 61 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 62 dstA == PIPE_BLENDFACTOR_ONE); 63} 64 65static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, 66 unsigned dstRGB, unsigned dstA) 67{ 68 /* If the blend equation is ADD or REVERSE_SUBTRACT, 69 * SRC_ALPHA == 1, and the following state is set, the colorbuffer 70 * will not be changed. 71 * Notice that the dst factors are the src factors inverted. */ 72 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 73 srcRGB == PIPE_BLENDFACTOR_ZERO) && 74 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 75 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 76 srcA == PIPE_BLENDFACTOR_ZERO) && 77 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 78 dstRGB == PIPE_BLENDFACTOR_ONE) && 79 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 80 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 81 dstA == PIPE_BLENDFACTOR_ONE); 82} 83 84static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, 85 unsigned dstRGB, unsigned dstA) 86{ 87 /* If the blend equation is ADD or REVERSE_SUBTRACT, 88 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer 89 * will not be changed. 90 * Notice that the dst factors are the src factors inverted. */ 91 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 92 srcRGB == PIPE_BLENDFACTOR_ZERO) && 93 (srcA == PIPE_BLENDFACTOR_ZERO) && 94 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 95 dstRGB == PIPE_BLENDFACTOR_ONE) && 96 (dstA == PIPE_BLENDFACTOR_ONE); 97} 98 99static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, 100 unsigned dstRGB, unsigned dstA) 101{ 102 /* If the blend equation is ADD or REVERSE_SUBTRACT, 103 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer 104 * will not be changed. 105 * Notice that the dst factors are the src factors inverted. */ 106 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 107 srcRGB == PIPE_BLENDFACTOR_ZERO) && 108 (srcA == PIPE_BLENDFACTOR_ZERO) && 109 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 110 dstRGB == PIPE_BLENDFACTOR_ONE) && 111 (dstA == PIPE_BLENDFACTOR_ONE); 112} 113 114static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, 115 unsigned dstRGB, unsigned dstA) 116{ 117 /* If the blend equation is ADD or REVERSE_SUBTRACT, 118 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, 119 * the colorbuffer will not be changed. 120 * Notice that the dst factors are the src factors inverted. */ 121 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || 122 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 123 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 124 srcRGB == PIPE_BLENDFACTOR_ZERO) && 125 (srcA == PIPE_BLENDFACTOR_SRC_COLOR || 126 srcA == PIPE_BLENDFACTOR_SRC_ALPHA || 127 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || 128 srcA == PIPE_BLENDFACTOR_ZERO) && 129 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 130 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 131 dstRGB == PIPE_BLENDFACTOR_ONE) && 132 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 133 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 134 dstA == PIPE_BLENDFACTOR_ONE); 135} 136 137static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, 138 unsigned dstRGB, unsigned dstA) 139{ 140 /* If the blend equation is ADD or REVERSE_SUBTRACT, 141 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, 142 * the colorbuffer will not be changed. 143 * Notice that the dst factors are the src factors inverted. */ 144 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || 145 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 146 srcRGB == PIPE_BLENDFACTOR_ZERO) && 147 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 148 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 149 srcA == PIPE_BLENDFACTOR_ZERO) && 150 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || 151 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 152 dstRGB == PIPE_BLENDFACTOR_ONE) && 153 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 154 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 155 dstA == PIPE_BLENDFACTOR_ONE); 156} 157 158/* Create a new blend state based on the CSO blend state. 159 * 160 * This encompasses alpha blending, logic/raster ops, and blend dithering. */ 161static void* r300_create_blend_state(struct pipe_context* pipe, 162 const struct pipe_blend_state* state) 163{ 164 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); 165 166 if (state->rt[0].blend_enable) 167 { 168 unsigned eqRGB = state->rt[0].rgb_func; 169 unsigned srcRGB = state->rt[0].rgb_src_factor; 170 unsigned dstRGB = state->rt[0].rgb_dst_factor; 171 172 unsigned eqA = state->rt[0].alpha_func; 173 unsigned srcA = state->rt[0].alpha_src_factor; 174 unsigned dstA = state->rt[0].alpha_dst_factor; 175 176 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, 177 * this is just the crappy D3D naming */ 178 blend->blend_control = R300_ALPHA_BLEND_ENABLE | 179 r300_translate_blend_function(eqRGB) | 180 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | 181 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); 182 183 /* Optimization: some operations do not require the destination color. 184 * 185 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, 186 * otherwise blending gives incorrect results. It seems to be 187 * a hardware bug. */ 188 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || 189 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || 190 dstRGB != PIPE_BLENDFACTOR_ZERO || 191 dstA != PIPE_BLENDFACTOR_ZERO || 192 srcRGB == PIPE_BLENDFACTOR_DST_COLOR || 193 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || 194 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || 195 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || 196 srcA == PIPE_BLENDFACTOR_DST_COLOR || 197 srcA == PIPE_BLENDFACTOR_DST_ALPHA || 198 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || 199 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || 200 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { 201 /* Enable reading from the colorbuffer. */ 202 blend->blend_control |= R300_READ_ENABLE; 203 204 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) { 205 /* Optimization: Depending on incoming pixels, we can 206 * conditionally disable the reading in hardware... */ 207 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && 208 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { 209 /* Disable reading if SRC_ALPHA == 0. */ 210 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || 211 dstRGB == PIPE_BLENDFACTOR_ZERO) && 212 (dstA == PIPE_BLENDFACTOR_SRC_COLOR || 213 dstA == PIPE_BLENDFACTOR_SRC_ALPHA || 214 dstA == PIPE_BLENDFACTOR_ZERO)) { 215 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; 216 } 217 218 /* Disable reading if SRC_ALPHA == 1. */ 219 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 220 dstRGB == PIPE_BLENDFACTOR_ZERO) && 221 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || 222 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || 223 dstA == PIPE_BLENDFACTOR_ZERO)) { 224 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; 225 } 226 } 227 } 228 } 229 230 /* Optimization: discard pixels which don't change the colorbuffer. 231 * 232 * The code below is non-trivial and some math is involved. 233 * 234 * Discarding pixels must be disabled when FP16 AA is enabled. 235 * This is a hardware bug. Also, this implementation wouldn't work 236 * with FP blending enabled and equation clamping disabled. 237 * 238 * Equations other than ADD are rarely used and therefore won't be 239 * optimized. */ 240 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && 241 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { 242 /* ADD: X+Y 243 * REVERSE_SUBTRACT: Y-X 244 * 245 * The idea is: 246 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, 247 * then CB will not be changed. 248 * 249 * Given the srcFactor and dstFactor variables, we can derive 250 * what src and dst should be equal to and discard appropriate 251 * pixels. 252 */ 253 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { 254 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; 255 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, 256 dstRGB, dstA)) { 257 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; 258 } else if (blend_discard_if_src_color_0(srcRGB, srcA, 259 dstRGB, dstA)) { 260 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; 261 } else if (blend_discard_if_src_color_1(srcRGB, srcA, 262 dstRGB, dstA)) { 263 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; 264 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, 265 dstRGB, dstA)) { 266 blend->blend_control |= 267 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; 268 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, 269 dstRGB, dstA)) { 270 blend->blend_control |= 271 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; 272 } 273 } 274 275 /* separate alpha */ 276 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 277 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; 278 blend->alpha_blend_control = 279 r300_translate_blend_function(eqA) | 280 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | 281 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); 282 } 283 } 284 285 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ 286 if (state->logicop_enable) { 287 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | 288 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; 289 } 290 291 /* Color Channel Mask */ 292 if (state->rt[0].colormask & PIPE_MASK_R) { 293 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0; 294 } 295 if (state->rt[0].colormask & PIPE_MASK_G) { 296 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0; 297 } 298 if (state->rt[0].colormask & PIPE_MASK_B) { 299 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0; 300 } 301 if (state->rt[0].colormask & PIPE_MASK_A) { 302 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0; 303 } 304 305 if (state->dither) { 306 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | 307 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; 308 } 309 310 return (void*)blend; 311} 312 313/* Bind blend state. */ 314static void r300_bind_blend_state(struct pipe_context* pipe, 315 void* state) 316{ 317 struct r300_context* r300 = r300_context(pipe); 318 319 r300->blend_state.state = state; 320 r300->blend_state.dirty = TRUE; 321} 322 323/* Free blend state. */ 324static void r300_delete_blend_state(struct pipe_context* pipe, 325 void* state) 326{ 327 FREE(state); 328} 329 330/* Convert float to 10bit integer */ 331static unsigned float_to_fixed10(float f) 332{ 333 return CLAMP((unsigned)(f * 1023.9f), 0, 1023); 334} 335 336/* Set blend color. 337 * Setup both R300 and R500 registers, figure out later which one to write. */ 338static void r300_set_blend_color(struct pipe_context* pipe, 339 const struct pipe_blend_color* color) 340{ 341 struct r300_context* r300 = r300_context(pipe); 342 struct r300_screen* r300screen = r300_screen(pipe->screen); 343 struct r300_blend_color_state* state = 344 (struct r300_blend_color_state*)r300->blend_color_state.state; 345 union util_color uc; 346 347 util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); 348 state->blend_color = uc.ui; 349 350 /* XXX if FP16 blending is enabled, we should use the FP16 format */ 351 state->blend_color_red_alpha = 352 float_to_fixed10(color->color[0]) | 353 (float_to_fixed10(color->color[3]) << 16); 354 state->blend_color_green_blue = 355 float_to_fixed10(color->color[2]) | 356 (float_to_fixed10(color->color[1]) << 16); 357 358 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; 359 r300->blend_color_state.dirty = TRUE; 360} 361 362static void r300_set_clip_state(struct pipe_context* pipe, 363 const struct pipe_clip_state* state) 364{ 365 struct r300_context* r300 = r300_context(pipe); 366 367 if (r300_screen(pipe->screen)->caps->has_tcl) { 368 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); 369 r300->clip_state.size = 29; 370 } else { 371 draw_flush(r300->draw); 372 draw_set_clip_state(r300->draw, state); 373 r300->clip_state.size = 2; 374 } 375 376 r300->clip_state.dirty = TRUE; 377} 378 379/* Create a new depth, stencil, and alpha state based on the CSO dsa state. 380 * 381 * This contains the depth buffer, stencil buffer, alpha test, and such. 382 * On the Radeon, depth and stencil buffer setup are intertwined, which is 383 * the reason for some of the strange-looking assignments across registers. */ 384static void* 385 r300_create_dsa_state(struct pipe_context* pipe, 386 const struct pipe_depth_stencil_alpha_state* state) 387{ 388 struct r300_capabilities *caps = 389 r300_screen(r300_context(pipe)->context.screen)->caps; 390 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); 391 392 /* Depth test setup. */ 393 if (state->depth.enabled) { 394 dsa->z_buffer_control |= R300_Z_ENABLE; 395 396 if (state->depth.writemask) { 397 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; 398 } 399 400 dsa->z_stencil_control |= 401 (r300_translate_depth_stencil_function(state->depth.func) << 402 R300_Z_FUNC_SHIFT); 403 } 404 405 /* Stencil buffer setup. */ 406 if (state->stencil[0].enabled) { 407 dsa->z_buffer_control |= R300_STENCIL_ENABLE; 408 dsa->z_stencil_control |= 409 (r300_translate_depth_stencil_function(state->stencil[0].func) << 410 R300_S_FRONT_FUNC_SHIFT) | 411 (r300_translate_stencil_op(state->stencil[0].fail_op) << 412 R300_S_FRONT_SFAIL_OP_SHIFT) | 413 (r300_translate_stencil_op(state->stencil[0].zpass_op) << 414 R300_S_FRONT_ZPASS_OP_SHIFT) | 415 (r300_translate_stencil_op(state->stencil[0].zfail_op) << 416 R300_S_FRONT_ZFAIL_OP_SHIFT); 417 418 dsa->stencil_ref_mask = (state->stencil[0].ref_value) | 419 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | 420 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); 421 422 if (state->stencil[1].enabled) { 423 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; 424 dsa->z_stencil_control |= 425 (r300_translate_depth_stencil_function(state->stencil[1].func) << 426 R300_S_BACK_FUNC_SHIFT) | 427 (r300_translate_stencil_op(state->stencil[1].fail_op) << 428 R300_S_BACK_SFAIL_OP_SHIFT) | 429 (r300_translate_stencil_op(state->stencil[1].zpass_op) << 430 R300_S_BACK_ZPASS_OP_SHIFT) | 431 (r300_translate_stencil_op(state->stencil[1].zfail_op) << 432 R300_S_BACK_ZFAIL_OP_SHIFT); 433 434 /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */ 435 if (caps->is_r500) 436 { 437 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; 438 dsa->stencil_ref_bf = (state->stencil[1].ref_value) | 439 (state->stencil[1].valuemask << 440 R300_STENCILMASK_SHIFT) | 441 (state->stencil[1].writemask << 442 R300_STENCILWRITEMASK_SHIFT); 443 } 444 } 445 } 446 447 /* Alpha test setup. */ 448 if (state->alpha.enabled) { 449 dsa->alpha_function = 450 r300_translate_alpha_function(state->alpha.func) | 451 R300_FG_ALPHA_FUNC_ENABLE; 452 453 /* XXX figure out why emitting 10bit alpha ref causes CS to dump */ 454 /* always use 8bit alpha ref */ 455 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); 456 457 if (caps->is_r500) 458 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; 459 } 460 461 return (void*)dsa; 462} 463 464/* Bind DSA state. */ 465static void r300_bind_dsa_state(struct pipe_context* pipe, 466 void* state) 467{ 468 struct r300_context* r300 = r300_context(pipe); 469 struct r300_screen* r300screen = r300_screen(pipe->screen); 470 471 r300->dsa_state.state = state; 472 r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; 473 r300->dsa_state.dirty = TRUE; 474} 475 476/* Free DSA state. */ 477static void r300_delete_dsa_state(struct pipe_context* pipe, 478 void* state) 479{ 480 FREE(state); 481} 482 483static void 484 r300_set_framebuffer_state(struct pipe_context* pipe, 485 const struct pipe_framebuffer_state* state) 486{ 487 struct r300_context* r300 = r300_context(pipe); 488 uint32_t zbuffer_bpp = 0; 489 490 r300->fb_state.size = (10 * state->nr_cbufs) + 491 (2 * (4 - state->nr_cbufs)) + 492 (state->zsbuf ? 10 : 0) + 6; 493 494 if (state->nr_cbufs > 4) { 495 debug_printf("r300: Implementation error: Too many MRTs in %s, " 496 "refusing to bind framebuffer state!\n", __FUNCTION__); 497 return; 498 } 499 500 if (r300->draw) { 501 draw_flush(r300->draw); 502 } 503 504 r300->fb_state.state = state; 505 506 /* Don't rely on the order of states being set for the first time. */ 507 /* XXX wait what */ 508 r300->blend_state.dirty = TRUE; 509 r300->dsa_state.dirty = TRUE; 510 r300->fb_state.dirty = TRUE; 511 r300->scissor_state.dirty = TRUE; 512 513 /* Polygon offset depends on the zbuffer bit depth. */ 514 if (state->zsbuf && r300->polygon_offset_enabled) { 515 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 516 case 2: 517 zbuffer_bpp = 16; 518 break; 519 case 4: 520 zbuffer_bpp = 24; 521 break; 522 } 523 524 if (r300->zbuffer_bpp != zbuffer_bpp) { 525 r300->zbuffer_bpp = zbuffer_bpp; 526 r300->rs_state.dirty = TRUE; 527 } 528 } 529} 530 531/* Create fragment shader state. */ 532static void* r300_create_fs_state(struct pipe_context* pipe, 533 const struct pipe_shader_state* shader) 534{ 535 struct r300_fragment_shader* fs = NULL; 536 537 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 538 539 /* Copy state directly into shader. */ 540 fs->state = *shader; 541 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 542 543 tgsi_scan_shader(shader->tokens, &fs->info); 544 r300_shader_read_fs_inputs(&fs->info, &fs->inputs); 545 546 return (void*)fs; 547} 548 549/* Bind fragment shader state. */ 550static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 551{ 552 struct r300_context* r300 = r300_context(pipe); 553 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 554 555 if (fs == NULL) { 556 r300->fs = NULL; 557 return; 558 } 559 560 r300->fs = fs; 561 r300_pick_fragment_shader(r300); 562 563 if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { 564 r300->vertex_format_state.dirty = TRUE; 565 } 566 567 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; 568} 569 570/* Delete fragment shader state. */ 571static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 572{ 573 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 574 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 575 576 while (ptr) { 577 tmp = ptr; 578 ptr = ptr->next; 579 rc_constants_destroy(&tmp->code.constants); 580 FREE(tmp); 581 } 582 FREE((void*)fs->state.tokens); 583 FREE(shader); 584} 585 586static void r300_set_polygon_stipple(struct pipe_context* pipe, 587 const struct pipe_poly_stipple* state) 588{ 589 /* XXX no idea how to set this up, but not terribly important */ 590} 591 592/* Create a new rasterizer state based on the CSO rasterizer state. 593 * 594 * This is a very large chunk of state, and covers most of the graphics 595 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 596 * 597 * In a not entirely unironic sidenote, this state has nearly nothing to do 598 * with the actual block on the Radeon called the rasterizer (RS). */ 599static void* r300_create_rs_state(struct pipe_context* pipe, 600 const struct pipe_rasterizer_state* state) 601{ 602 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 603 604 /* Copy rasterizer state for Draw. */ 605 rs->rs = *state; 606 607#ifdef PIPE_ARCH_LITTLE_ENDIAN 608 rs->vap_control_status = R300_VC_NO_SWAP; 609#else 610 rs->vap_control_status = R300_VC_32BIT_SWAP; 611#endif 612 613 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. 614 * Else, enable HW TCL and force Draw's TCL off. */ 615 if (state->bypass_vs_clip_and_viewport || 616 !r300_screen(pipe->screen)->caps->has_tcl) { 617 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 618 } 619 620 rs->point_size = pack_float_16_6x(state->point_size) | 621 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 622 623 rs->point_minmax = 624 ((int)(state->point_size_min * 6.0) << 625 R300_GA_POINT_MINMAX_MIN_SHIFT) | 626 ((int)(state->point_size_max * 6.0) << 627 R300_GA_POINT_MINMAX_MAX_SHIFT); 628 629 rs->line_control = pack_float_16_6x(state->line_width) | 630 R300_GA_LINE_CNTL_END_TYPE_COMP; 631 632 /* Enable polygon mode */ 633 if (state->fill_cw != PIPE_POLYGON_MODE_FILL || 634 state->fill_ccw != PIPE_POLYGON_MODE_FILL) { 635 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 636 } 637 638 /* Radeons don't think in "CW/CCW", they think in "front/back". */ 639 if (state->front_winding == PIPE_WINDING_CW) { 640 rs->cull_mode = R300_FRONT_FACE_CW; 641 642 /* Polygon offset */ 643 if (state->offset_cw) { 644 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 645 } 646 if (state->offset_ccw) { 647 rs->polygon_offset_enable |= R300_BACK_ENABLE; 648 } 649 650 /* Polygon mode */ 651 if (rs->polygon_mode) { 652 rs->polygon_mode |= 653 r300_translate_polygon_mode_front(state->fill_cw); 654 rs->polygon_mode |= 655 r300_translate_polygon_mode_back(state->fill_ccw); 656 } 657 } else { 658 rs->cull_mode = R300_FRONT_FACE_CCW; 659 660 /* Polygon offset */ 661 if (state->offset_ccw) { 662 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 663 } 664 if (state->offset_cw) { 665 rs->polygon_offset_enable |= R300_BACK_ENABLE; 666 } 667 668 /* Polygon mode */ 669 if (rs->polygon_mode) { 670 rs->polygon_mode |= 671 r300_translate_polygon_mode_front(state->fill_ccw); 672 rs->polygon_mode |= 673 r300_translate_polygon_mode_back(state->fill_cw); 674 } 675 } 676 if (state->front_winding & state->cull_mode) { 677 rs->cull_mode |= R300_CULL_FRONT; 678 } 679 if (~(state->front_winding) & state->cull_mode) { 680 rs->cull_mode |= R300_CULL_BACK; 681 } 682 683 if (rs->polygon_offset_enable) { 684 rs->depth_offset = state->offset_units; 685 rs->depth_scale = state->offset_scale; 686 } 687 688 if (state->line_stipple_enable) { 689 rs->line_stipple_config = 690 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 691 (fui((float)state->line_stipple_factor) & 692 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 693 /* XXX this might need to be scaled up */ 694 rs->line_stipple_value = state->line_stipple_pattern; 695 } 696 697 if (state->flatshade) { 698 rs->color_control = R300_SHADE_MODEL_FLAT; 699 } else { 700 rs->color_control = R300_SHADE_MODEL_SMOOTH; 701 } 702 703 return (void*)rs; 704} 705 706/* Bind rasterizer state. */ 707static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 708{ 709 struct r300_context* r300 = r300_context(pipe); 710 struct r300_rs_state* rs = (struct r300_rs_state*)state; 711 712 if (r300->draw) { 713 draw_flush(r300->draw); 714 draw_set_rasterizer_state(r300->draw, &rs->rs); 715 } 716 717 if (rs) { 718 r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; 719 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; 720 } else { 721 r300->tcl_bypass = FALSE; 722 r300->polygon_offset_enabled = FALSE; 723 } 724 725 r300->rs_state.state = rs; 726 r300->rs_state.dirty = TRUE; 727 /* XXX Why is this still needed, dammit!? */ 728 r300->scissor_state.dirty = TRUE; 729 r300->viewport_state.dirty = TRUE; 730 731 /* XXX Clean these up when we move to atom emits */ 732 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 733 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 734 } 735} 736 737/* Free rasterizer state. */ 738static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 739{ 740 FREE(state); 741} 742 743static void* 744 r300_create_sampler_state(struct pipe_context* pipe, 745 const struct pipe_sampler_state* state) 746{ 747 struct r300_context* r300 = r300_context(pipe); 748 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 749 int lod_bias; 750 union util_color uc; 751 752 sampler->state = *state; 753 754 sampler->filter0 |= 755 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 756 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 757 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 758 759 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 760 state->mag_img_filter, 761 state->min_mip_filter, 762 state->max_anisotropy > 1.0); 763 764 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 765 /* We must pass these to the emit function to clamp them properly. */ 766 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 767 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 768 769 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); 770 771 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 772 773 sampler->filter1 |= r300_anisotropy(state->max_anisotropy); 774 775 util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); 776 sampler->border_color = uc.ui; 777 778 /* R500-specific fixups and optimizations */ 779 if (r300_screen(r300->context.screen)->caps->is_r500) { 780 sampler->filter1 |= R500_BORDER_FIX; 781 } 782 783 return (void*)sampler; 784} 785 786static void r300_bind_sampler_states(struct pipe_context* pipe, 787 unsigned count, 788 void** states) 789{ 790 struct r300_context* r300 = r300_context(pipe); 791 int i; 792 793 if (count > 8) { 794 return; 795 } 796 797 for (i = 0; i < count; i++) { 798 if (r300->sampler_states[i] != states[i]) { 799 r300->sampler_states[i] = (struct r300_sampler_state*)states[i]; 800 r300->dirty_state |= (R300_NEW_SAMPLER << i); 801 } 802 } 803 804 r300->sampler_count = count; 805 806 /* Pick a fragment shader based on the texture compare state. */ 807 if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) { 808 if (r300_pick_fragment_shader(r300)) { 809 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | 810 R300_NEW_FRAGMENT_SHADER_CONSTANTS; 811 } 812 } 813} 814 815static void r300_lacks_vertex_textures(struct pipe_context* pipe, 816 unsigned count, 817 void** states) 818{ 819} 820 821static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 822{ 823 FREE(state); 824} 825 826static void r300_set_sampler_textures(struct pipe_context* pipe, 827 unsigned count, 828 struct pipe_texture** texture) 829{ 830 struct r300_context* r300 = r300_context(pipe); 831 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; 832 int i; 833 834 /* XXX magic num */ 835 if (count > 8) { 836 return; 837 } 838 839 for (i = 0; i < count; i++) { 840 if (r300->textures[i] != (struct r300_texture*)texture[i]) { 841 pipe_texture_reference((struct pipe_texture**)&r300->textures[i], 842 texture[i]); 843 r300->dirty_state |= (R300_NEW_TEXTURE << i); 844 845 /* R300-specific - set the texrect factor in a fragment shader */ 846 if (!is_r500 && r300->textures[i]->is_npot) { 847 /* XXX It would be nice to re-emit just 1 constant, 848 * XXX not all of them */ 849 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 850 } 851 } 852 } 853 854 for (i = count; i < 8; i++) { 855 if (r300->textures[i]) { 856 pipe_texture_reference((struct pipe_texture**)&r300->textures[i], 857 NULL); 858 r300->dirty_state |= (R300_NEW_TEXTURE << i); 859 } 860 } 861 862 r300->texture_count = count; 863} 864 865static void r300_set_scissor_state(struct pipe_context* pipe, 866 const struct pipe_scissor_state* state) 867{ 868 struct r300_context* r300 = r300_context(pipe); 869 870 memcpy(r300->scissor_state.state, state, 871 sizeof(struct pipe_scissor_state)); 872 873 r300->scissor_state.dirty = TRUE; 874} 875 876static void r300_set_viewport_state(struct pipe_context* pipe, 877 const struct pipe_viewport_state* state) 878{ 879 struct r300_context* r300 = r300_context(pipe); 880 struct r300_viewport_state* viewport = 881 (struct r300_viewport_state*)r300->viewport_state.state; 882 883 /* Do the transform in HW. */ 884 viewport->vte_control = R300_VTX_W0_FMT; 885 886 if (state->scale[0] != 1.0f) { 887 viewport->xscale = state->scale[0]; 888 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 889 } 890 if (state->scale[1] != 1.0f) { 891 viewport->yscale = state->scale[1]; 892 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 893 } 894 if (state->scale[2] != 1.0f) { 895 viewport->zscale = state->scale[2]; 896 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 897 } 898 if (state->translate[0] != 0.0f) { 899 viewport->xoffset = state->translate[0]; 900 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 901 } 902 if (state->translate[1] != 0.0f) { 903 viewport->yoffset = state->translate[1]; 904 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 905 } 906 if (state->translate[2] != 0.0f) { 907 viewport->zoffset = state->translate[2]; 908 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 909 } 910 911 r300->viewport_state.dirty = TRUE; 912 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { 913 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 914 } 915} 916 917static void r300_set_vertex_buffers(struct pipe_context* pipe, 918 unsigned count, 919 const struct pipe_vertex_buffer* buffers) 920{ 921 struct r300_context* r300 = r300_context(pipe); 922 923 memcpy(r300->vertex_buffer, buffers, 924 sizeof(struct pipe_vertex_buffer) * count); 925 r300->vertex_buffer_count = count; 926 927 if (r300->draw) { 928 draw_flush(r300->draw); 929 draw_set_vertex_buffers(r300->draw, count, buffers); 930 } 931 932 r300->vertex_format_state.dirty = TRUE; 933} 934 935static boolean r300_validate_aos(struct r300_context *r300) 936{ 937 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; 938 struct pipe_vertex_element *velem = r300->vertex_element; 939 int i; 940 941 /* Check if formats and strides are aligned to the size of DWORD. */ 942 for (i = 0; i < r300->vertex_element_count; i++) { 943 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || 944 util_format_get_blocksize(velem[i].src_format) % 4 != 0) { 945 return FALSE; 946 } 947 } 948 return TRUE; 949} 950 951static void r300_set_vertex_elements(struct pipe_context* pipe, 952 unsigned count, 953 const struct pipe_vertex_element* elements) 954{ 955 struct r300_context* r300 = r300_context(pipe); 956 957 memcpy(r300->vertex_element, 958 elements, 959 sizeof(struct pipe_vertex_element) * count); 960 r300->vertex_element_count = count; 961 962 if (r300->draw) { 963 draw_flush(r300->draw); 964 draw_set_vertex_elements(r300->draw, count, elements); 965 } 966 967 if (!r300_validate_aos(r300)) { 968 /* XXX We should fallback using draw. */ 969 assert(0); 970 abort(); 971 } 972} 973 974static void* r300_create_vs_state(struct pipe_context* pipe, 975 const struct pipe_shader_state* shader) 976{ 977 struct r300_context* r300 = r300_context(pipe); 978 979 if (r300_screen(pipe->screen)->caps->has_tcl) { 980 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 981 /* Copy state directly into shader. */ 982 vs->state = *shader; 983 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 984 985 tgsi_scan_shader(shader->tokens, &vs->info); 986 987 return (void*)vs; 988 } else { 989 return draw_create_vertex_shader(r300->draw, shader); 990 } 991} 992 993static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 994{ 995 struct r300_context* r300 = r300_context(pipe); 996 997 if (r300_screen(pipe->screen)->caps->has_tcl) { 998 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 999 1000 if (vs == NULL) { 1001 r300->vs = NULL; 1002 return; 1003 } else if (!vs->translated) { 1004 r300_translate_vertex_shader(r300, vs); 1005 } 1006 1007 r300->vs = vs; 1008 if (r300->fs) { 1009 r300_vertex_shader_setup_wpos(r300); 1010 } 1011 1012 r300->vertex_format_state.dirty = TRUE; 1013 1014 r300->dirty_state |= 1015 R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; 1016 } else { 1017 draw_flush(r300->draw); 1018 draw_bind_vertex_shader(r300->draw, 1019 (struct draw_vertex_shader*)shader); 1020 } 1021} 1022 1023static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1024{ 1025 struct r300_context* r300 = r300_context(pipe); 1026 1027 if (r300_screen(pipe->screen)->caps->has_tcl) { 1028 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1029 1030 rc_constants_destroy(&vs->code.constants); 1031 FREE((void*)vs->state.tokens); 1032 FREE(shader); 1033 } else { 1034 draw_delete_vertex_shader(r300->draw, 1035 (struct draw_vertex_shader*)shader); 1036 } 1037} 1038 1039static void r300_set_constant_buffer(struct pipe_context *pipe, 1040 uint shader, uint index, 1041 struct pipe_buffer *buf) 1042{ 1043 struct r300_context* r300 = r300_context(pipe); 1044 void *mapped; 1045 1046 if (buf == NULL || buf->size == 0 || 1047 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) 1048 { 1049 r300->shader_constants[shader].count = 0; 1050 return; 1051 } 1052 1053 assert((buf->size % 4 * sizeof(float)) == 0); 1054 memcpy(r300->shader_constants[shader].constants, mapped, buf->size); 1055 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); 1056 pipe_buffer_unmap(pipe->screen, buf); 1057 1058 if (shader == PIPE_SHADER_VERTEX) 1059 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; 1060 else if (shader == PIPE_SHADER_FRAGMENT) 1061 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; 1062} 1063 1064void r300_init_state_functions(struct r300_context* r300) 1065{ 1066 r300->context.create_blend_state = r300_create_blend_state; 1067 r300->context.bind_blend_state = r300_bind_blend_state; 1068 r300->context.delete_blend_state = r300_delete_blend_state; 1069 1070 r300->context.set_blend_color = r300_set_blend_color; 1071 1072 r300->context.set_clip_state = r300_set_clip_state; 1073 1074 r300->context.set_constant_buffer = r300_set_constant_buffer; 1075 1076 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1077 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1078 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1079 1080 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1081 1082 r300->context.create_fs_state = r300_create_fs_state; 1083 r300->context.bind_fs_state = r300_bind_fs_state; 1084 r300->context.delete_fs_state = r300_delete_fs_state; 1085 1086 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1087 1088 r300->context.create_rasterizer_state = r300_create_rs_state; 1089 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1090 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1091 1092 r300->context.create_sampler_state = r300_create_sampler_state; 1093 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1094 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1095 r300->context.delete_sampler_state = r300_delete_sampler_state; 1096 1097 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures; 1098 1099 r300->context.set_scissor_state = r300_set_scissor_state; 1100 1101 r300->context.set_viewport_state = r300_set_viewport_state; 1102 1103 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1104 r300->context.set_vertex_elements = r300_set_vertex_elements; 1105 1106 r300->context.create_vs_state = r300_create_vs_state; 1107 r300->context.bind_vs_state = r300_bind_vs_state; 1108 r300->context.delete_vs_state = r300_delete_vs_state; 1109} 1110