r300_state.c revision 371ca689ec5d1d7f301f4a3176c4e0df1c44ec15
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 pipe_framebuffer_state *old_state = r300->fb_state.state; 668 unsigned max_width, max_height, i; 669 uint32_t zbuffer_bpp = 0; 670 671 if (state->nr_cbufs > 4) { 672 fprintf(stderr, "r300: Implementation error: Too many MRTs in %s, " 673 "refusing to bind framebuffer state!\n", __FUNCTION__); 674 return; 675 } 676 677 if (r300->screen->caps.is_r500) { 678 max_width = max_height = 4096; 679 } else if (r300->screen->caps.is_r400) { 680 max_width = max_height = 4021; 681 } else { 682 max_width = max_height = 2560; 683 } 684 685 if (state->width > max_width || state->height > max_height) { 686 fprintf(stderr, "r300: Implementation error: Render targets are too " 687 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); 688 return; 689 } 690 691 if (r300->draw) { 692 draw_flush(r300->draw); 693 } 694 695 r300->fb_state.dirty = TRUE; 696 697 /* If nr_cbufs is changed from zero to non-zero or vice versa... */ 698 if (!!old_state->nr_cbufs != !!state->nr_cbufs) { 699 r300->blend_state.dirty = TRUE; 700 } 701 /* If zsbuf is set from NULL to non-NULL or vice versa.. */ 702 if (!!old_state->zsbuf != !!state->zsbuf) { 703 r300->dsa_state.dirty = TRUE; 704 } 705 706 /* The tiling flags are dependent on the surface miplevel, unfortunately. */ 707 r300_fb_set_tiling_flags(r300, r300->fb_state.state, state); 708 709 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); 710 711 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + 712 (state->zsbuf ? 10 : 0) + 9; 713 714 /* Polygon offset depends on the zbuffer bit depth. */ 715 if (state->zsbuf && r300->polygon_offset_enabled) { 716 switch (util_format_get_blocksize(state->zsbuf->texture->format)) { 717 case 2: 718 zbuffer_bpp = 16; 719 break; 720 case 4: 721 zbuffer_bpp = 24; 722 break; 723 } 724 725 if (r300->zbuffer_bpp != zbuffer_bpp) { 726 r300->zbuffer_bpp = zbuffer_bpp; 727 r300->rs_state.dirty = TRUE; 728 } 729 } 730 731 if (DBG_ON(r300, DBG_FB)) { 732 fprintf(stderr, "r300: set_framebuffer_state:\n"); 733 for (i = 0; i < state->nr_cbufs; i++) { 734 r300_print_fb_surf_info(state->cbufs[i], i, "CB"); 735 } 736 if (state->zsbuf) { 737 r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); 738 } 739 } 740} 741 742/* Create fragment shader state. */ 743static void* r300_create_fs_state(struct pipe_context* pipe, 744 const struct pipe_shader_state* shader) 745{ 746 struct r300_fragment_shader* fs = NULL; 747 748 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader); 749 750 /* Copy state directly into shader. */ 751 fs->state = *shader; 752 fs->state.tokens = tgsi_dup_tokens(shader->tokens); 753 754 return (void*)fs; 755} 756 757void r300_mark_fs_code_dirty(struct r300_context *r300) 758{ 759 struct r300_fragment_shader* fs = r300_fs(r300); 760 761 r300->fs.dirty = TRUE; 762 r300->fs_rc_constant_state.dirty = TRUE; 763 r300->fs_constants.dirty = TRUE; 764 r300->fs.size = fs->shader->cb_code_size; 765 766 if (r300->screen->caps.is_r500) { 767 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; 768 r300->fs_constants.size = fs->shader->externals_count * 4 + 3; 769 } else { 770 r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; 771 r300->fs_constants.size = fs->shader->externals_count * 4 + 1; 772 } 773} 774 775/* Bind fragment shader state. */ 776static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) 777{ 778 struct r300_context* r300 = r300_context(pipe); 779 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 780 781 if (fs == NULL) { 782 r300->fs.state = NULL; 783 return; 784 } 785 786 r300->fs.state = fs; 787 r300_pick_fragment_shader(r300); 788 r300_mark_fs_code_dirty(r300); 789 790 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 791} 792 793/* Delete fragment shader state. */ 794static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) 795{ 796 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; 797 struct r300_fragment_shader_code *tmp, *ptr = fs->first; 798 799 while (ptr) { 800 tmp = ptr; 801 ptr = ptr->next; 802 rc_constants_destroy(&tmp->code.constants); 803 FREE(tmp->cb_code); 804 FREE(tmp); 805 } 806 FREE((void*)fs->state.tokens); 807 FREE(shader); 808} 809 810static void r300_set_polygon_stipple(struct pipe_context* pipe, 811 const struct pipe_poly_stipple* state) 812{ 813 /* XXX no idea how to set this up, but not terribly important */ 814} 815 816/* Create a new rasterizer state based on the CSO rasterizer state. 817 * 818 * This is a very large chunk of state, and covers most of the graphics 819 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks. 820 * 821 * In a not entirely unironic sidenote, this state has nearly nothing to do 822 * with the actual block on the Radeon called the rasterizer (RS). */ 823static void* r300_create_rs_state(struct pipe_context* pipe, 824 const struct pipe_rasterizer_state* state) 825{ 826 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); 827 int i; 828 float psiz; 829 830 /* Copy rasterizer state. */ 831 rs->rs = *state; 832 rs->rs_draw = *state; 833 834 /* Override some states for Draw. */ 835 rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ 836 837#ifdef PIPE_ARCH_LITTLE_ENDIAN 838 rs->vap_control_status = R300_VC_NO_SWAP; 839#else 840 rs->vap_control_status = R300_VC_32BIT_SWAP; 841#endif 842 843 /* If no TCL engine is present, turn off the HW TCL. */ 844 if (!r300_screen(pipe->screen)->caps.has_tcl) { 845 rs->vap_control_status |= R300_VAP_TCL_BYPASS; 846 } 847 848 /* Point size width and height. */ 849 rs->point_size = 850 pack_float_16_6x(state->point_size) | 851 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); 852 853 /* Point size clamping. */ 854 if (state->point_size_per_vertex) { 855 /* Per-vertex point size. 856 * Clamp to [0, max FB size] */ 857 psiz = pipe->screen->get_paramf(pipe->screen, 858 PIPE_CAP_MAX_POINT_WIDTH); 859 rs->point_minmax = 860 pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; 861 } else { 862 /* We cannot disable the point-size vertex output, 863 * so clamp it. */ 864 psiz = state->point_size; 865 rs->point_minmax = 866 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | 867 (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); 868 } 869 870 /* Line control. */ 871 rs->line_control = pack_float_16_6x(state->line_width) | 872 R300_GA_LINE_CNTL_END_TYPE_COMP; 873 874 /* Enable polygon mode */ 875 if (state->fill_front != PIPE_POLYGON_MODE_FILL || 876 state->fill_back != PIPE_POLYGON_MODE_FILL) { 877 rs->polygon_mode = R300_GA_POLY_MODE_DUAL; 878 } 879 880 /* Front face */ 881 if (state->front_ccw) 882 rs->cull_mode = R300_FRONT_FACE_CCW; 883 else 884 rs->cull_mode = R300_FRONT_FACE_CW; 885 886 /* Polygon offset */ 887 if (util_get_offset(state, state->fill_front)) { 888 rs->polygon_offset_enable |= R300_FRONT_ENABLE; 889 } 890 if (util_get_offset(state, state->fill_back)) { 891 rs->polygon_offset_enable |= R300_BACK_ENABLE; 892 } 893 894 /* Polygon mode */ 895 if (rs->polygon_mode) { 896 rs->polygon_mode |= 897 r300_translate_polygon_mode_front(state->fill_front); 898 rs->polygon_mode |= 899 r300_translate_polygon_mode_back(state->fill_back); 900 } 901 902 if (state->cull_face & PIPE_FACE_FRONT) { 903 rs->cull_mode |= R300_CULL_FRONT; 904 } 905 if (state->cull_face & PIPE_FACE_BACK) { 906 rs->cull_mode |= R300_CULL_BACK; 907 } 908 909 if (rs->polygon_offset_enable) { 910 rs->depth_offset = state->offset_units; 911 rs->depth_scale = state->offset_scale; 912 } 913 914 if (state->line_stipple_enable) { 915 rs->line_stipple_config = 916 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | 917 (fui((float)state->line_stipple_factor) & 918 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); 919 /* XXX this might need to be scaled up */ 920 rs->line_stipple_value = state->line_stipple_pattern; 921 } 922 923 if (state->flatshade) { 924 rs->color_control = R300_SHADE_MODEL_FLAT; 925 } else { 926 rs->color_control = R300_SHADE_MODEL_SMOOTH; 927 } 928 929 rs->clip_rule = state->scissor ? 0xAAAA : 0xFFFF; 930 931 /* Point sprites */ 932 if (state->sprite_coord_enable) { 933 rs->stuffing_enable = R300_GB_POINT_STUFF_ENABLE; 934 for (i = 0; i < 8; i++) { 935 if (state->sprite_coord_enable & (1 << i)) 936 rs->stuffing_enable |= 937 R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); 938 } 939 940 rs->point_texcoord_left = 0.0f; 941 rs->point_texcoord_right = 1.0f; 942 943 switch (state->sprite_coord_mode) { 944 case PIPE_SPRITE_COORD_UPPER_LEFT: 945 rs->point_texcoord_top = 0.0f; 946 rs->point_texcoord_bottom = 1.0f; 947 break; 948 case PIPE_SPRITE_COORD_LOWER_LEFT: 949 rs->point_texcoord_top = 1.0f; 950 rs->point_texcoord_bottom = 0.0f; 951 break; 952 } 953 } 954 955 if (state->gl_rasterization_rules) { 956 rs->multisample_position_0 = 0x66666666; 957 rs->multisample_position_1 = 0x6666666; 958 } 959 960 return (void*)rs; 961} 962 963/* Bind rasterizer state. */ 964static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 965{ 966 struct r300_context* r300 = r300_context(pipe); 967 struct r300_rs_state* rs = (struct r300_rs_state*)state; 968 int last_sprite_coord_enable = r300->sprite_coord_enable; 969 boolean last_two_sided_color = r300->two_sided_color; 970 971 if (r300->draw && rs) { 972 draw_flush(r300->draw); 973 draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); 974 } 975 976 if (rs) { 977 r300->polygon_offset_enabled = (rs->rs.offset_point || 978 rs->rs.offset_line || 979 rs->rs.offset_tri); 980 r300->sprite_coord_enable = rs->rs.sprite_coord_enable; 981 r300->two_sided_color = rs->rs.light_twoside; 982 } else { 983 r300->polygon_offset_enabled = FALSE; 984 r300->sprite_coord_enable = 0; 985 r300->two_sided_color = FALSE; 986 } 987 988 UPDATE_STATE(state, r300->rs_state); 989 r300->rs_state.size = 25 + (r300->polygon_offset_enabled ? 5 : 0) + 990 (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0) ? 5 : 0); 991 992 if (last_sprite_coord_enable != r300->sprite_coord_enable || 993 last_two_sided_color != r300->two_sided_color) { 994 r300->rs_block_state.dirty = TRUE; 995 } 996} 997 998/* Free rasterizer state. */ 999static void r300_delete_rs_state(struct pipe_context* pipe, void* state) 1000{ 1001 FREE(state); 1002} 1003 1004static void* 1005 r300_create_sampler_state(struct pipe_context* pipe, 1006 const struct pipe_sampler_state* state) 1007{ 1008 struct r300_context* r300 = r300_context(pipe); 1009 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); 1010 boolean is_r500 = r300->screen->caps.is_r500; 1011 int lod_bias; 1012 union util_color uc; 1013 1014 sampler->state = *state; 1015 1016 sampler->filter0 |= 1017 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | 1018 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | 1019 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); 1020 1021 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, 1022 state->mag_img_filter, 1023 state->min_mip_filter, 1024 state->max_anisotropy > 0); 1025 1026 sampler->filter0 |= r300_anisotropy(state->max_anisotropy); 1027 1028 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ 1029 /* We must pass these to the merge function to clamp them properly. */ 1030 sampler->min_lod = MAX2((unsigned)state->min_lod, 0); 1031 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); 1032 1033 lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); 1034 1035 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; 1036 1037 /* This is very high quality anisotropic filtering for R5xx. 1038 * It's good for benchmarking the performance of texturing but 1039 * in practice we don't want to slow down the driver because it's 1040 * a pretty good performance killer. Feel free to play with it. */ 1041 if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { 1042 sampler->filter1 |= r500_anisotropy(state->max_anisotropy); 1043 } 1044 1045 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); 1046 sampler->border_color = uc.ui; 1047 1048 /* R500-specific fixups and optimizations */ 1049 if (r300->screen->caps.is_r500) { 1050 sampler->filter1 |= R500_BORDER_FIX; 1051 } 1052 1053 return (void*)sampler; 1054} 1055 1056static void r300_bind_sampler_states(struct pipe_context* pipe, 1057 unsigned count, 1058 void** states) 1059{ 1060 struct r300_context* r300 = r300_context(pipe); 1061 struct r300_textures_state* state = 1062 (struct r300_textures_state*)r300->textures_state.state; 1063 unsigned tex_units = r300->screen->caps.num_tex_units; 1064 1065 if (count > tex_units) { 1066 return; 1067 } 1068 1069 memcpy(state->sampler_states, states, sizeof(void*) * count); 1070 state->sampler_state_count = count; 1071 1072 r300->textures_state.dirty = TRUE; 1073} 1074 1075static void r300_lacks_vertex_textures(struct pipe_context* pipe, 1076 unsigned count, 1077 void** states) 1078{ 1079} 1080 1081static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) 1082{ 1083 FREE(state); 1084} 1085 1086static void r300_set_fragment_sampler_views(struct pipe_context* pipe, 1087 unsigned count, 1088 struct pipe_sampler_view** views) 1089{ 1090 struct r300_context* r300 = r300_context(pipe); 1091 struct r300_textures_state* state = 1092 (struct r300_textures_state*)r300->textures_state.state; 1093 struct r300_texture *texture; 1094 unsigned i; 1095 unsigned tex_units = r300->screen->caps.num_tex_units; 1096 boolean dirty_tex = FALSE; 1097 1098 if (count > tex_units) { 1099 return; 1100 } 1101 1102 for (i = 0; i < count; i++) { 1103 if (&state->sampler_views[i]->base != views[i]) { 1104 pipe_sampler_view_reference( 1105 (struct pipe_sampler_view**)&state->sampler_views[i], 1106 views[i]); 1107 1108 if (!views[i]) { 1109 continue; 1110 } 1111 1112 /* A new sampler view (= texture)... */ 1113 dirty_tex = TRUE; 1114 1115 /* Set the texrect factor in the fragment shader. 1116 * Needed for RECT and NPOT fallback. */ 1117 texture = r300_texture(views[i]->texture); 1118 if (texture->uses_pitch) { 1119 r300->fs_rc_constant_state.dirty = TRUE; 1120 } 1121 } 1122 } 1123 1124 for (i = count; i < tex_units; i++) { 1125 if (state->sampler_views[i]) { 1126 pipe_sampler_view_reference( 1127 (struct pipe_sampler_view**)&state->sampler_views[i], 1128 NULL); 1129 } 1130 } 1131 1132 state->sampler_view_count = count; 1133 1134 r300->textures_state.dirty = TRUE; 1135 1136 if (dirty_tex) { 1137 r300->texture_cache_inval.dirty = TRUE; 1138 } 1139} 1140 1141static struct pipe_sampler_view * 1142r300_create_sampler_view(struct pipe_context *pipe, 1143 struct pipe_resource *texture, 1144 const struct pipe_sampler_view *templ) 1145{ 1146 struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); 1147 struct r300_texture *tex = r300_texture(texture); 1148 1149 if (view) { 1150 view->base = *templ; 1151 view->base.reference.count = 1; 1152 view->base.context = pipe; 1153 view->base.texture = NULL; 1154 pipe_resource_reference(&view->base.texture, texture); 1155 1156 view->swizzle[0] = templ->swizzle_r; 1157 view->swizzle[1] = templ->swizzle_g; 1158 view->swizzle[2] = templ->swizzle_b; 1159 view->swizzle[3] = templ->swizzle_a; 1160 1161 view->format = tex->tx_format; 1162 view->format.format1 |= r300_translate_texformat(templ->format, 1163 view->swizzle); 1164 if (r300_screen(pipe->screen)->caps.is_r500) { 1165 view->format.format2 |= r500_tx_format_msb_bit(templ->format); 1166 } 1167 } 1168 1169 return (struct pipe_sampler_view*)view; 1170} 1171 1172static void 1173r300_sampler_view_destroy(struct pipe_context *pipe, 1174 struct pipe_sampler_view *view) 1175{ 1176 pipe_resource_reference(&view->texture, NULL); 1177 FREE(view); 1178} 1179 1180static void r300_set_scissor_state(struct pipe_context* pipe, 1181 const struct pipe_scissor_state* state) 1182{ 1183 struct r300_context* r300 = r300_context(pipe); 1184 1185 memcpy(r300->scissor_state.state, state, 1186 sizeof(struct pipe_scissor_state)); 1187 1188 r300->scissor_state.dirty = TRUE; 1189} 1190 1191static void r300_set_viewport_state(struct pipe_context* pipe, 1192 const struct pipe_viewport_state* state) 1193{ 1194 struct r300_context* r300 = r300_context(pipe); 1195 struct r300_viewport_state* viewport = 1196 (struct r300_viewport_state*)r300->viewport_state.state; 1197 1198 r300->viewport = *state; 1199 1200 if (r300->draw) { 1201 draw_flush(r300->draw); 1202 draw_set_viewport_state(r300->draw, state); 1203 viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; 1204 return; 1205 } 1206 1207 /* Do the transform in HW. */ 1208 viewport->vte_control = R300_VTX_W0_FMT; 1209 1210 if (state->scale[0] != 1.0f) { 1211 viewport->xscale = state->scale[0]; 1212 viewport->vte_control |= R300_VPORT_X_SCALE_ENA; 1213 } 1214 if (state->scale[1] != 1.0f) { 1215 viewport->yscale = state->scale[1]; 1216 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; 1217 } 1218 if (state->scale[2] != 1.0f) { 1219 viewport->zscale = state->scale[2]; 1220 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; 1221 } 1222 if (state->translate[0] != 0.0f) { 1223 viewport->xoffset = state->translate[0]; 1224 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; 1225 } 1226 if (state->translate[1] != 0.0f) { 1227 viewport->yoffset = state->translate[1]; 1228 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; 1229 } 1230 if (state->translate[2] != 0.0f) { 1231 viewport->zoffset = state->translate[2]; 1232 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; 1233 } 1234 1235 r300->viewport_state.dirty = TRUE; 1236 if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { 1237 r300->fs_rc_constant_state.dirty = TRUE; 1238 } 1239} 1240 1241static void r300_set_vertex_buffers(struct pipe_context* pipe, 1242 unsigned count, 1243 const struct pipe_vertex_buffer* buffers) 1244{ 1245 struct r300_context* r300 = r300_context(pipe); 1246 struct pipe_vertex_buffer *vbo; 1247 unsigned i, max_index = (1 << 24) - 1; 1248 boolean any_user_buffer = FALSE; 1249 1250 if (count == r300->vertex_buffer_count && 1251 memcmp(r300->vertex_buffer, buffers, 1252 sizeof(struct pipe_vertex_buffer) * count) == 0) { 1253 return; 1254 } 1255 1256 if (r300->screen->caps.has_tcl) { 1257 /* HW TCL. */ 1258 r300->incompatible_vb_layout = FALSE; 1259 1260 /* Check if the strides and offsets are aligned to the size of DWORD. */ 1261 for (i = 0; i < count; i++) { 1262 if (buffers[i].buffer) { 1263 if (buffers[i].stride % 4 != 0 || 1264 buffers[i].buffer_offset % 4 != 0) { 1265 r300->incompatible_vb_layout = TRUE; 1266 break; 1267 } 1268 } 1269 } 1270 1271 for (i = 0; i < count; i++) { 1272 /* Why, yes, I AM casting away constness. How did you know? */ 1273 vbo = (struct pipe_vertex_buffer*)&buffers[i]; 1274 1275 /* Skip NULL buffers */ 1276 if (!buffers[i].buffer) { 1277 continue; 1278 } 1279 1280 if (r300_buffer_is_user_buffer(vbo->buffer)) { 1281 any_user_buffer = TRUE; 1282 } 1283 1284 if (vbo->max_index == ~0) { 1285 /* if no VBO stride then only one vertex value so max index is 1 */ 1286 /* should think about converting to VS constants like svga does */ 1287 if (!vbo->stride) 1288 vbo->max_index = 1; 1289 else 1290 vbo->max_index = 1291 (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; 1292 } 1293 1294 max_index = MIN2(vbo->max_index, max_index); 1295 } 1296 1297 r300->any_user_vbs = any_user_buffer; 1298 r300->vertex_buffer_max_index = max_index; 1299 1300 } else { 1301 /* SW TCL. */ 1302 draw_flush(r300->draw); 1303 draw_set_vertex_buffers(r300->draw, count, buffers); 1304 } 1305 1306 /* Common code. */ 1307 for (i = 0; i < count; i++) { 1308 /* Reference our buffer. */ 1309 pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); 1310 } 1311 for (; i < r300->vertex_buffer_count; i++) { 1312 /* Dereference any old buffers. */ 1313 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); 1314 } 1315 1316 memcpy(r300->vertex_buffer, buffers, 1317 sizeof(struct pipe_vertex_buffer) * count); 1318 r300->vertex_buffer_count = count; 1319} 1320 1321/* Initialize the PSC tables. */ 1322static void r300_vertex_psc(struct r300_vertex_element_state *velems) 1323{ 1324 struct r300_vertex_stream_state *vstream = &velems->vertex_stream; 1325 uint16_t type, swizzle; 1326 enum pipe_format format; 1327 unsigned i; 1328 1329 if (velems->count > 16) { 1330 fprintf(stderr, "r300: More than 16 vertex elements are not supported," 1331 " requested %i, using 16.\n", velems->count); 1332 velems->count = 16; 1333 } 1334 1335 /* Vertex shaders have no semantics on their inputs, 1336 * so PSC should just route stuff based on the vertex elements, 1337 * and not on attrib information. */ 1338 for (i = 0; i < velems->count; i++) { 1339 format = velems->hw_format[i]; 1340 1341 type = r300_translate_vertex_data_type(format); 1342 if (type == R300_INVALID_FORMAT) { 1343 fprintf(stderr, "r300: Bad vertex format %s.\n", 1344 util_format_short_name(format)); 1345 assert(0); 1346 abort(); 1347 } 1348 1349 type |= i << R300_DST_VEC_LOC_SHIFT; 1350 swizzle = r300_translate_vertex_data_swizzle(format); 1351 1352 if (i & 1) { 1353 vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; 1354 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; 1355 } else { 1356 vstream->vap_prog_stream_cntl[i >> 1] |= type; 1357 vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; 1358 } 1359 } 1360 1361 /* Set the last vector in the PSC. */ 1362 if (i) { 1363 i -= 1; 1364 } 1365 vstream->vap_prog_stream_cntl[i >> 1] |= 1366 (R300_LAST_VEC << (i & 1 ? 16 : 0)); 1367 1368 vstream->count = (i >> 1) + 1; 1369} 1370 1371#define FORMAT_REPLACE(what, withwhat) \ 1372 case PIPE_FORMAT_##what: *format = PIPE_FORMAT_##withwhat; break 1373 1374static void* r300_create_vertex_elements_state(struct pipe_context* pipe, 1375 unsigned count, 1376 const struct pipe_vertex_element* attribs) 1377{ 1378 struct r300_vertex_element_state *velems; 1379 unsigned i; 1380 enum pipe_format *format; 1381 1382 assert(count <= PIPE_MAX_ATTRIBS); 1383 velems = CALLOC_STRUCT(r300_vertex_element_state); 1384 if (velems != NULL) { 1385 velems->count = count; 1386 memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); 1387 1388 if (r300_screen(pipe->screen)->caps.has_tcl) { 1389 /* Set the best hw format in case the original format is not 1390 * supported by hw. */ 1391 for (i = 0; i < count; i++) { 1392 velems->hw_format[i] = velems->velem[i].src_format; 1393 format = &velems->hw_format[i]; 1394 1395 /* This is basically the list of unsupported formats. 1396 * For now we don't care about the alignment, that's going to 1397 * be sorted out after the PSC setup. */ 1398 switch (*format) { 1399 FORMAT_REPLACE(R64_FLOAT, R32_FLOAT); 1400 FORMAT_REPLACE(R64G64_FLOAT, R32G32_FLOAT); 1401 FORMAT_REPLACE(R64G64B64_FLOAT, R32G32B32_FLOAT); 1402 FORMAT_REPLACE(R64G64B64A64_FLOAT, R32G32B32A32_FLOAT); 1403 1404 FORMAT_REPLACE(R32_UNORM, R32_FLOAT); 1405 FORMAT_REPLACE(R32G32_UNORM, R32G32_FLOAT); 1406 FORMAT_REPLACE(R32G32B32_UNORM, R32G32B32_FLOAT); 1407 FORMAT_REPLACE(R32G32B32A32_UNORM, R32G32B32A32_FLOAT); 1408 1409 FORMAT_REPLACE(R32_USCALED, R32_FLOAT); 1410 FORMAT_REPLACE(R32G32_USCALED, R32G32_FLOAT); 1411 FORMAT_REPLACE(R32G32B32_USCALED, R32G32B32_FLOAT); 1412 FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT); 1413 1414 FORMAT_REPLACE(R32_SNORM, R32_FLOAT); 1415 FORMAT_REPLACE(R32G32_SNORM, R32G32_FLOAT); 1416 FORMAT_REPLACE(R32G32B32_SNORM, R32G32B32_FLOAT); 1417 FORMAT_REPLACE(R32G32B32A32_SNORM, R32G32B32A32_FLOAT); 1418 1419 FORMAT_REPLACE(R32_SSCALED, R32_FLOAT); 1420 FORMAT_REPLACE(R32G32_SSCALED, R32G32_FLOAT); 1421 FORMAT_REPLACE(R32G32B32_SSCALED, R32G32B32_FLOAT); 1422 FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT); 1423 1424 FORMAT_REPLACE(R32_FIXED, R32_FLOAT); 1425 FORMAT_REPLACE(R32G32_FIXED, R32G32_FLOAT); 1426 FORMAT_REPLACE(R32G32B32_FIXED, R32G32B32_FLOAT); 1427 FORMAT_REPLACE(R32G32B32A32_FIXED, R32G32B32A32_FLOAT); 1428 1429 default:; 1430 } 1431 1432 velems->incompatible_layout = 1433 velems->incompatible_layout || 1434 velems->velem[i].src_format != velems->hw_format[i] || 1435 velems->velem[i].src_offset % 4 != 0; 1436 } 1437 1438 /* Now setup PSC. 1439 * The unused components will be replaced by (..., 0, 1). */ 1440 r300_vertex_psc(velems); 1441 1442 /* Align the formats to the size of DWORD. 1443 * We only care about the blocksizes of the formats since 1444 * swizzles are already set up. */ 1445 for (i = 0; i < count; i++) { 1446 /* This is OK because we check for aligned strides too. */ 1447 velems->hw_format_size[i] = 1448 align(util_format_get_blocksize(velems->hw_format[i]), 4); 1449 } 1450 } 1451 } 1452 return velems; 1453} 1454 1455static void r300_bind_vertex_elements_state(struct pipe_context *pipe, 1456 void *state) 1457{ 1458 struct r300_context *r300 = r300_context(pipe); 1459 struct r300_vertex_element_state *velems = state; 1460 1461 if (velems == NULL) { 1462 return; 1463 } 1464 1465 r300->velems = velems; 1466 1467 if (r300->draw) { 1468 draw_flush(r300->draw); 1469 draw_set_vertex_elements(r300->draw, velems->count, velems->velem); 1470 return; 1471 } 1472 1473 UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); 1474 r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; 1475} 1476 1477static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) 1478{ 1479 FREE(state); 1480} 1481 1482static void* r300_create_vs_state(struct pipe_context* pipe, 1483 const struct pipe_shader_state* shader) 1484{ 1485 struct r300_context* r300 = r300_context(pipe); 1486 1487 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); 1488 1489 /* Copy state directly into shader. */ 1490 vs->state = *shader; 1491 vs->state.tokens = tgsi_dup_tokens(shader->tokens); 1492 1493 if (r300->screen->caps.has_tcl) { 1494 r300_init_vs_outputs(vs); 1495 r300_translate_vertex_shader(r300, vs); 1496 } else { 1497 r300_draw_init_vertex_shader(r300->draw, vs); 1498 } 1499 1500 return vs; 1501} 1502 1503static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) 1504{ 1505 struct r300_context* r300 = r300_context(pipe); 1506 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1507 1508 if (vs == NULL) { 1509 r300->vs_state.state = NULL; 1510 return; 1511 } 1512 if (vs == r300->vs_state.state) { 1513 return; 1514 } 1515 r300->vs_state.state = vs; 1516 1517 /* The majority of the RS block bits is dependent on the vertex shader. */ 1518 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ 1519 1520 if (r300->screen->caps.has_tcl) { 1521 r300->vs_state.dirty = TRUE; 1522 r300->vs_state.size = 1523 vs->code.length + 9 + 1524 (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); 1525 1526 if (vs->externals_count) { 1527 r300->vs_constants.dirty = TRUE; 1528 r300->vs_constants.size = vs->externals_count * 4 + 3; 1529 } else { 1530 r300->vs_constants.size = 0; 1531 } 1532 1533 r300->pvs_flush.dirty = TRUE; 1534 } else { 1535 draw_flush(r300->draw); 1536 draw_bind_vertex_shader(r300->draw, 1537 (struct draw_vertex_shader*)vs->draw_vs); 1538 } 1539} 1540 1541static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) 1542{ 1543 struct r300_context* r300 = r300_context(pipe); 1544 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; 1545 1546 if (r300->screen->caps.has_tcl) { 1547 rc_constants_destroy(&vs->code.constants); 1548 } else { 1549 draw_delete_vertex_shader(r300->draw, 1550 (struct draw_vertex_shader*)vs->draw_vs); 1551 } 1552 1553 FREE((void*)vs->state.tokens); 1554 FREE(shader); 1555} 1556 1557static void r300_set_constant_buffer(struct pipe_context *pipe, 1558 uint shader, uint index, 1559 struct pipe_resource *buf) 1560{ 1561 struct r300_context* r300 = r300_context(pipe); 1562 struct r300_constant_buffer *cbuf; 1563 struct pipe_transfer *tr; 1564 float *mapped; 1565 int max_size = 0, max_size_bytes = 0, clamped_size = 0; 1566 1567 switch (shader) { 1568 case PIPE_SHADER_VERTEX: 1569 cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; 1570 max_size = 256; 1571 break; 1572 case PIPE_SHADER_FRAGMENT: 1573 cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; 1574 if (r300->screen->caps.is_r500) { 1575 max_size = 256; 1576 } else { 1577 max_size = 32; 1578 } 1579 break; 1580 default: 1581 assert(0); 1582 return; 1583 } 1584 max_size_bytes = max_size * 4 * sizeof(float); 1585 1586 if (buf == NULL || buf->width0 == 0 || 1587 (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL) 1588 { 1589 cbuf->count = 0; 1590 return; 1591 } 1592 1593 if (shader == PIPE_SHADER_FRAGMENT || 1594 (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { 1595 assert((buf->width0 % (4 * sizeof(float))) == 0); 1596 1597 /* Check the size of the constant buffer. */ 1598 /* XXX Subtract immediates and RC_STATE_* variables. */ 1599 if (buf->width0 > max_size_bytes) { 1600 fprintf(stderr, "r300: Max size of the constant buffer is " 1601 "%i*4 floats.\n", max_size); 1602 } 1603 1604 clamped_size = MIN2(buf->width0, max_size_bytes); 1605 cbuf->count = clamped_size / (4 * sizeof(float)); 1606 1607 if (shader == PIPE_SHADER_FRAGMENT && !r300->screen->caps.is_r500) { 1608 unsigned i,j; 1609 1610 /* Convert constants to float24. */ 1611 for (i = 0; i < cbuf->count; i++) 1612 for (j = 0; j < 4; j++) 1613 cbuf->constants[i][j] = pack_float24(mapped[i*4+j]); 1614 } else { 1615 memcpy(cbuf->constants, mapped, clamped_size); 1616 } 1617 } 1618 1619 if (shader == PIPE_SHADER_VERTEX) { 1620 if (r300->screen->caps.has_tcl) { 1621 if (r300->vs_constants.size) { 1622 r300->vs_constants.dirty = TRUE; 1623 } 1624 r300->pvs_flush.dirty = TRUE; 1625 } else if (r300->draw) { 1626 draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 1627 0, mapped, buf->width0); 1628 } 1629 } else if (shader == PIPE_SHADER_FRAGMENT) { 1630 r300->fs_constants.dirty = TRUE; 1631 } 1632 1633 pipe_buffer_unmap(pipe, buf, tr); 1634} 1635 1636void r300_init_state_functions(struct r300_context* r300) 1637{ 1638 r300->context.create_blend_state = r300_create_blend_state; 1639 r300->context.bind_blend_state = r300_bind_blend_state; 1640 r300->context.delete_blend_state = r300_delete_blend_state; 1641 1642 r300->context.set_blend_color = r300_set_blend_color; 1643 1644 r300->context.set_clip_state = r300_set_clip_state; 1645 r300->context.set_sample_mask = r300_set_sample_mask; 1646 1647 r300->context.set_constant_buffer = r300_set_constant_buffer; 1648 1649 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state; 1650 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; 1651 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; 1652 1653 r300->context.set_stencil_ref = r300_set_stencil_ref; 1654 1655 r300->context.set_framebuffer_state = r300_set_framebuffer_state; 1656 1657 r300->context.create_fs_state = r300_create_fs_state; 1658 r300->context.bind_fs_state = r300_bind_fs_state; 1659 r300->context.delete_fs_state = r300_delete_fs_state; 1660 1661 r300->context.set_polygon_stipple = r300_set_polygon_stipple; 1662 1663 r300->context.create_rasterizer_state = r300_create_rs_state; 1664 r300->context.bind_rasterizer_state = r300_bind_rs_state; 1665 r300->context.delete_rasterizer_state = r300_delete_rs_state; 1666 1667 r300->context.create_sampler_state = r300_create_sampler_state; 1668 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; 1669 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; 1670 r300->context.delete_sampler_state = r300_delete_sampler_state; 1671 1672 r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; 1673 r300->context.create_sampler_view = r300_create_sampler_view; 1674 r300->context.sampler_view_destroy = r300_sampler_view_destroy; 1675 1676 r300->context.set_scissor_state = r300_set_scissor_state; 1677 1678 r300->context.set_viewport_state = r300_set_viewport_state; 1679 1680 r300->context.set_vertex_buffers = r300_set_vertex_buffers; 1681 1682 r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; 1683 r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; 1684 r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; 1685 1686 r300->context.create_vs_state = r300_create_vs_state; 1687 r300->context.bind_vs_state = r300_bind_vs_state; 1688 r300->context.delete_vs_state = r300_delete_vs_state; 1689} 1690