r300_emit.c revision 848b8605
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/* r300_emit: Functions for emitting state. */ 25 26#include "util/u_format.h" 27#include "util/u_math.h" 28#include "util/u_mm.h" 29 30#include "r300_context.h" 31#include "r300_cb.h" 32#include "r300_cs.h" 33#include "r300_emit.h" 34#include "r300_fs.h" 35#include "r300_screen.h" 36#include "r300_screen_buffer.h" 37#include "r300_vs.h" 38 39void r300_emit_blend_state(struct r300_context* r300, 40 unsigned size, void* state) 41{ 42 struct r300_blend_state* blend = (struct r300_blend_state*)state; 43 struct pipe_framebuffer_state* fb = 44 (struct pipe_framebuffer_state*)r300->fb_state.state; 45 struct pipe_surface *cb; 46 CS_LOCALS(r300); 47 48 cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL; 49 50 if (cb) { 51 if (cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT) { 52 WRITE_CS_TABLE(blend->cb_noclamp, size); 53 } else if (cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT) { 54 WRITE_CS_TABLE(blend->cb_noclamp_noalpha, size); 55 } else { 56 unsigned swz = r300_surface(cb)->colormask_swizzle; 57 WRITE_CS_TABLE(blend->cb_clamp[swz], size); 58 } 59 } else { 60 WRITE_CS_TABLE(blend->cb_no_readwrite, size); 61 } 62} 63 64void r300_emit_blend_color_state(struct r300_context* r300, 65 unsigned size, void* state) 66{ 67 struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state; 68 CS_LOCALS(r300); 69 70 WRITE_CS_TABLE(bc->cb, size); 71} 72 73void r300_emit_clip_state(struct r300_context* r300, 74 unsigned size, void* state) 75{ 76 struct r300_clip_state* clip = (struct r300_clip_state*)state; 77 CS_LOCALS(r300); 78 79 WRITE_CS_TABLE(clip->cb, size); 80} 81 82void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state) 83{ 84 struct r300_dsa_state* dsa = (struct r300_dsa_state*)state; 85 struct pipe_framebuffer_state* fb = 86 (struct pipe_framebuffer_state*)r300->fb_state.state; 87 boolean is_r500 = r300->screen->caps.is_r500; 88 CS_LOCALS(r300); 89 uint32_t alpha_func = dsa->alpha_function; 90 91 /* Choose the alpha ref value between 8-bit (FG_ALPHA_FUNC.AM_VAL) and 92 * 16-bit (FG_ALPHA_VALUE). */ 93 if (is_r500 && (alpha_func & R300_FG_ALPHA_FUNC_ENABLE)) { 94 struct pipe_surface *cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL; 95 96 if (cb && 97 (cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT || 98 cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT)) { 99 alpha_func |= R500_FG_ALPHA_FUNC_FP16_ENABLE; 100 } else { 101 alpha_func |= R500_FG_ALPHA_FUNC_8BIT; 102 } 103 } 104 105 /* Setup alpha-to-coverage. */ 106 if (r300->alpha_to_coverage && r300->msaa_enable) { 107 /* Always set 3/6, it improves precision even for 2x and 4x MSAA. */ 108 alpha_func |= R300_FG_ALPHA_FUNC_MASK_ENABLE | 109 R300_FG_ALPHA_FUNC_CFG_3_OF_6; 110 } 111 112 BEGIN_CS(size); 113 OUT_CS_REG(R300_FG_ALPHA_FUNC, alpha_func); 114 OUT_CS_TABLE(fb->zsbuf ? &dsa->cb_begin : dsa->cb_zb_no_readwrite, size-2); 115 END_CS; 116} 117 118static void get_rc_constant_state( 119 float vec[4], 120 struct r300_context * r300, 121 struct rc_constant * constant) 122{ 123 struct r300_textures_state* texstate = r300->textures_state.state; 124 struct r300_resource *tex; 125 126 assert(constant->Type == RC_CONSTANT_STATE); 127 128 /* vec should either be (0, 0, 0, 1), which should be a relatively safe 129 * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE 130 * state factors. */ 131 132 switch (constant->u.State[0]) { 133 /* Factor for converting rectangle coords to 134 * normalized coords. Should only show up on non-r500. */ 135 case RC_STATE_R300_TEXRECT_FACTOR: 136 tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture); 137 vec[0] = 1.0 / tex->tex.width0; 138 vec[1] = 1.0 / tex->tex.height0; 139 vec[2] = 0; 140 vec[3] = 1; 141 break; 142 143 case RC_STATE_R300_TEXSCALE_FACTOR: 144 tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture); 145 /* Add a small number to the texture size to work around rounding errors in hw. */ 146 vec[0] = tex->b.b.width0 / (tex->tex.width0 + 0.001f); 147 vec[1] = tex->b.b.height0 / (tex->tex.height0 + 0.001f); 148 vec[2] = tex->b.b.depth0 / (tex->tex.depth0 + 0.001f); 149 vec[3] = 1; 150 break; 151 152 case RC_STATE_R300_VIEWPORT_SCALE: 153 vec[0] = r300->viewport.scale[0]; 154 vec[1] = r300->viewport.scale[1]; 155 vec[2] = r300->viewport.scale[2]; 156 vec[3] = 1; 157 break; 158 159 case RC_STATE_R300_VIEWPORT_OFFSET: 160 vec[0] = r300->viewport.translate[0]; 161 vec[1] = r300->viewport.translate[1]; 162 vec[2] = r300->viewport.translate[2]; 163 vec[3] = 1; 164 break; 165 166 default: 167 fprintf(stderr, "r300: Implementation error: " 168 "Unknown RC_CONSTANT type %d\n", constant->u.State[0]); 169 vec[0] = 0; 170 vec[1] = 0; 171 vec[2] = 0; 172 vec[3] = 1; 173 } 174} 175 176/* Convert a normal single-precision float into the 7.16 format 177 * used by the R300 fragment shader. 178 */ 179uint32_t pack_float24(float f) 180{ 181 union { 182 float fl; 183 uint32_t u; 184 } u; 185 float mantissa; 186 int exponent; 187 uint32_t float24 = 0; 188 189 if (f == 0.0) 190 return 0; 191 192 u.fl = f; 193 194 mantissa = frexpf(f, &exponent); 195 196 /* Handle -ve */ 197 if (mantissa < 0) { 198 float24 |= (1 << 23); 199 mantissa = mantissa * -1.0; 200 } 201 /* Handle exponent, bias of 63 */ 202 exponent += 62; 203 float24 |= (exponent << 16); 204 /* Kill 7 LSB of mantissa */ 205 float24 |= (u.u & 0x7FFFFF) >> 7; 206 207 return float24; 208} 209 210void r300_emit_fs(struct r300_context* r300, unsigned size, void *state) 211{ 212 struct r300_fragment_shader *fs = r300_fs(r300); 213 CS_LOCALS(r300); 214 215 WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size); 216} 217 218void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state) 219{ 220 struct r300_fragment_shader *fs = r300_fs(r300); 221 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; 222 unsigned count = fs->shader->externals_count; 223 unsigned i, j; 224 CS_LOCALS(r300); 225 226 if (count == 0) 227 return; 228 229 BEGIN_CS(size); 230 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4); 231 if (buf->remap_table){ 232 for (i = 0; i < count; i++) { 233 float *data = (float*)&buf->ptr[buf->remap_table[i]*4]; 234 for (j = 0; j < 4; j++) 235 OUT_CS(pack_float24(data[j])); 236 } 237 } else { 238 for (i = 0; i < count; i++) 239 for (j = 0; j < 4; j++) 240 OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j])); 241 } 242 243 END_CS; 244} 245 246void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state) 247{ 248 struct r300_fragment_shader *fs = r300_fs(r300); 249 struct rc_constant_list *constants = &fs->shader->code.constants; 250 unsigned i; 251 unsigned count = fs->shader->rc_state_count; 252 unsigned first = fs->shader->externals_count; 253 unsigned end = constants->Count; 254 unsigned j; 255 CS_LOCALS(r300); 256 257 if (count == 0) 258 return; 259 260 BEGIN_CS(size); 261 for(i = first; i < end; ++i) { 262 if (constants->Constants[i].Type == RC_CONSTANT_STATE) { 263 float data[4]; 264 265 get_rc_constant_state(data, r300, &constants->Constants[i]); 266 267 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); 268 for (j = 0; j < 4; j++) 269 OUT_CS(pack_float24(data[j])); 270 } 271 } 272 END_CS; 273} 274 275void r500_emit_fs(struct r300_context* r300, unsigned size, void *state) 276{ 277 struct r300_fragment_shader *fs = r300_fs(r300); 278 CS_LOCALS(r300); 279 280 WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size); 281} 282 283void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state) 284{ 285 struct r300_fragment_shader *fs = r300_fs(r300); 286 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; 287 unsigned count = fs->shader->externals_count; 288 CS_LOCALS(r300); 289 290 if (count == 0) 291 return; 292 293 BEGIN_CS(size); 294 OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); 295 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4); 296 if (buf->remap_table){ 297 for (unsigned i = 0; i < count; i++) { 298 uint32_t *data = &buf->ptr[buf->remap_table[i]*4]; 299 OUT_CS_TABLE(data, 4); 300 } 301 } else { 302 OUT_CS_TABLE(buf->ptr, count * 4); 303 } 304 END_CS; 305} 306 307void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state) 308{ 309 struct r300_fragment_shader *fs = r300_fs(r300); 310 struct rc_constant_list *constants = &fs->shader->code.constants; 311 unsigned i; 312 unsigned count = fs->shader->rc_state_count; 313 unsigned first = fs->shader->externals_count; 314 unsigned end = constants->Count; 315 CS_LOCALS(r300); 316 317 if (count == 0) 318 return; 319 320 BEGIN_CS(size); 321 for(i = first; i < end; ++i) { 322 if (constants->Constants[i].Type == RC_CONSTANT_STATE) { 323 float data[4]; 324 325 get_rc_constant_state(data, r300, &constants->Constants[i]); 326 327 OUT_CS_REG(R500_GA_US_VECTOR_INDEX, 328 R500_GA_US_VECTOR_INDEX_TYPE_CONST | 329 (i & R500_GA_US_VECTOR_INDEX_MASK)); 330 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4); 331 OUT_CS_TABLE(data, 4); 332 } 333 } 334 END_CS; 335} 336 337void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state) 338{ 339 struct r300_gpu_flush *gpuflush = (struct r300_gpu_flush*)state; 340 struct pipe_framebuffer_state* fb = 341 (struct pipe_framebuffer_state*)r300->fb_state.state; 342 uint32_t height = fb->height; 343 uint32_t width = fb->width; 344 CS_LOCALS(r300); 345 346 if (r300->cbzb_clear) { 347 struct r300_surface *surf = r300_surface(fb->cbufs[0]); 348 349 height = surf->cbzb_height; 350 width = surf->cbzb_width; 351 } 352 353 DBG(r300, DBG_SCISSOR, 354 "r300: Scissor width: %i, height: %i, CBZB clear: %s\n", 355 width, height, r300->cbzb_clear ? "YES" : "NO"); 356 357 BEGIN_CS(size); 358 359 /* Set up scissors. 360 * By writing to the SC registers, SC & US assert idle. */ 361 OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); 362 if (r300->screen->caps.is_r500) { 363 OUT_CS(0); 364 OUT_CS(((width - 1) << R300_SCISSORS_X_SHIFT) | 365 ((height - 1) << R300_SCISSORS_Y_SHIFT)); 366 } else { 367 OUT_CS((1440 << R300_SCISSORS_X_SHIFT) | 368 (1440 << R300_SCISSORS_Y_SHIFT)); 369 OUT_CS(((width + 1440-1) << R300_SCISSORS_X_SHIFT) | 370 ((height + 1440-1) << R300_SCISSORS_Y_SHIFT)); 371 } 372 373 /* Flush CB & ZB caches and wait until the 3D engine is idle and clean. */ 374 OUT_CS_TABLE(gpuflush->cb_flush_clean, 6); 375 END_CS; 376} 377 378void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state) 379{ 380 struct r300_aa_state *aa = (struct r300_aa_state*)state; 381 CS_LOCALS(r300); 382 383 BEGIN_CS(size); 384 OUT_CS_REG(R300_GB_AA_CONFIG, aa->aa_config); 385 386 if (aa->dest) { 387 OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_OFFSET, 3); 388 OUT_CS(aa->dest->offset); 389 OUT_CS(aa->dest->pitch & R300_RB3D_AARESOLVE_PITCH_MASK); 390 OUT_CS(R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE | 391 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE); 392 OUT_CS_RELOC(aa->dest); 393 } else { 394 OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0); 395 } 396 397 END_CS; 398} 399 400void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state) 401{ 402 struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state; 403 struct r300_surface* surf; 404 unsigned i; 405 uint32_t rb3d_cctl = 0; 406 407 CS_LOCALS(r300); 408 409 BEGIN_CS(size); 410 411 if (r300->screen->caps.is_r500) { 412 rb3d_cctl = R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE; 413 } 414 /* NUM_MULTIWRITES replicates COLOR[0] to all colorbuffers. */ 415 if (fb->nr_cbufs && r300->fb_multiwrite) { 416 rb3d_cctl |= R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs); 417 } 418 if (r300->cmask_in_use) { 419 rb3d_cctl |= R300_RB3D_CCTL_AA_COMPRESSION_ENABLE | 420 R300_RB3D_CCTL_CMASK_ENABLE; 421 } 422 423 OUT_CS_REG(R300_RB3D_CCTL, rb3d_cctl); 424 425 /* Set up colorbuffers. */ 426 for (i = 0; i < fb->nr_cbufs; i++) { 427 surf = r300_surface(r300_get_nonnull_cb(fb, i)); 428 429 OUT_CS_REG(R300_RB3D_COLOROFFSET0 + (4 * i), surf->offset); 430 OUT_CS_RELOC(surf); 431 432 OUT_CS_REG(R300_RB3D_COLORPITCH0 + (4 * i), surf->pitch); 433 OUT_CS_RELOC(surf); 434 435 if (r300->cmask_in_use && i == 0) { 436 OUT_CS_REG(R300_RB3D_CMASK_OFFSET0, 0); 437 OUT_CS_REG(R300_RB3D_CMASK_PITCH0, surf->pitch_cmask); 438 OUT_CS_REG(R300_RB3D_COLOR_CLEAR_VALUE, r300->color_clear_value); 439 if (r300->screen->caps.is_r500 && r300->screen->info.drm_minor >= 29) { 440 OUT_CS_REG_SEQ(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2); 441 OUT_CS(r300->color_clear_value_ar); 442 OUT_CS(r300->color_clear_value_gb); 443 } 444 } 445 } 446 447 /* Set up the ZB part of the CBZB clear. */ 448 if (r300->cbzb_clear) { 449 surf = r300_surface(fb->cbufs[0]); 450 451 OUT_CS_REG(R300_ZB_FORMAT, surf->cbzb_format); 452 453 OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->cbzb_midpoint_offset); 454 OUT_CS_RELOC(surf); 455 456 OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->cbzb_pitch); 457 OUT_CS_RELOC(surf); 458 459 DBG(r300, DBG_CBZB, 460 "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format, 461 surf->cbzb_pitch); 462 } 463 /* Set up a zbuffer. */ 464 else if (fb->zsbuf) { 465 surf = r300_surface(fb->zsbuf); 466 467 OUT_CS_REG(R300_ZB_FORMAT, surf->format); 468 469 OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->offset); 470 OUT_CS_RELOC(surf); 471 472 OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->pitch); 473 OUT_CS_RELOC(surf); 474 475 if (r300->hyperz_enabled) { 476 /* HiZ RAM. */ 477 OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0); 478 OUT_CS_REG(R300_ZB_HIZ_PITCH, surf->pitch_hiz); 479 /* Z Mask RAM. (compressed zbuffer) */ 480 OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0); 481 OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf->pitch_zmask); 482 } 483 } 484 485 END_CS; 486} 487 488void r300_emit_hyperz_state(struct r300_context *r300, 489 unsigned size, void *state) 490{ 491 struct r300_hyperz_state *z = state; 492 CS_LOCALS(r300); 493 494 if (z->flush) 495 WRITE_CS_TABLE(&z->cb_flush_begin, size); 496 else 497 WRITE_CS_TABLE(&z->cb_begin, size - 2); 498} 499 500void r300_emit_hyperz_end(struct r300_context *r300) 501{ 502 struct r300_hyperz_state z = 503 *(struct r300_hyperz_state*)r300->hyperz_state.state; 504 505 z.flush = 1; 506 z.zb_bw_cntl = 0; 507 z.zb_depthclearvalue = 0; 508 z.sc_hyperz = R300_SC_HYPERZ_ADJ_2; 509 z.gb_z_peq_config = 0; 510 511 r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z); 512} 513 514#define R300_NIBBLES(x0, y0, x1, y1, x2, y2, d0y, d0x) \ 515 (((x0) & 0xf) | (((y0) & 0xf) << 4) | \ 516 (((x1) & 0xf) << 8) | (((y1) & 0xf) << 12) | \ 517 (((x2) & 0xf) << 16) | (((y2) & 0xf) << 20) | \ 518 (((d0y) & 0xf) << 24) | (((d0x) & 0xf) << 28)) 519 520static unsigned r300_get_mspos(int index, unsigned *p) 521{ 522 unsigned reg, i, distx, disty, dist; 523 524 if (index == 0) { 525 /* MSPOS0 contains positions for samples 0,1,2 as (X,Y) pairs of nibbles, 526 * followed by a (Y,X) pair containing the minimum distance from the pixel 527 * edge: 528 * X0, Y0, X1, Y1, X2, Y2, D0_Y, D0_X 529 * 530 * There is a quirk when setting D0_X. The value represents the distance 531 * from the left edge of the pixel quad to the first sample in subpixels. 532 * All values less than eight should use the actual value, but „7‟ should 533 * be used for the distance „8‟. The hardware will convert 7 into 8 internally. 534 */ 535 distx = 11; 536 for (i = 0; i < 12; i += 2) { 537 if (p[i] < distx) 538 distx = p[i]; 539 } 540 541 disty = 11; 542 for (i = 1; i < 12; i += 2) { 543 if (p[i] < disty) 544 disty = p[i]; 545 } 546 547 if (distx == 8) 548 distx = 7; 549 550 reg = R300_NIBBLES(p[0], p[1], p[2], p[3], p[4], p[5], disty, distx); 551 } else { 552 /* MSPOS1 contains positions for samples 3,4,5 as (X,Y) pairs of nibbles, 553 * followed by the minimum distance from the pixel edge (not sure if X or Y): 554 * X3, Y3, X4, Y4, X5, Y5, D1 555 */ 556 dist = 11; 557 for (i = 0; i < 12; i++) { 558 if (p[i] < dist) 559 dist = p[i]; 560 } 561 562 reg = R300_NIBBLES(p[6], p[7], p[8], p[9], p[10], p[11], dist, 0); 563 } 564 return reg; 565} 566 567void r300_emit_fb_state_pipelined(struct r300_context *r300, 568 unsigned size, void *state) 569{ 570 /* The sample coordinates are in the range [0,11], because 571 * GB_TILE_CONFIG.SUBPIXEL is set to the 1/12 subpixel precision. 572 * 573 * Some sample coordinates reach to neighboring pixels and should not be used. 574 * (e.g. Y=11) 575 * 576 * The unused samples must be set to the positions of other valid samples. */ 577 static unsigned sample_locs_1x[12] = { 578 6,6, 6,6, 6,6, 6,6, 6,6, 6,6 579 }; 580 static unsigned sample_locs_2x[12] = { 581 3,9, 9,3, 9,3, 9,3, 9,3, 9,3 582 }; 583 static unsigned sample_locs_4x[12] = { 584 4,4, 8,8, 2,10, 10,2, 10,2, 10,2 585 }; 586 static unsigned sample_locs_6x[12] = { 587 3,1, 7,3, 11,5, 1,7, 5,9, 9,10 588 }; 589 590 struct pipe_framebuffer_state* fb = 591 (struct pipe_framebuffer_state*)r300->fb_state.state; 592 unsigned i, num_cbufs = fb->nr_cbufs; 593 unsigned mspos0, mspos1; 594 CS_LOCALS(r300); 595 596 /* If we use the multiwrite feature, the colorbuffers 2,3,4 must be 597 * marked as UNUSED in the US block. */ 598 if (r300->fb_multiwrite) { 599 num_cbufs = MIN2(num_cbufs, 1); 600 } 601 602 BEGIN_CS(size); 603 604 /* Colorbuffer format in the US block. 605 * (must be written after unpipelined regs) */ 606 OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4); 607 for (i = 0; i < num_cbufs; i++) { 608 OUT_CS(r300_surface(r300_get_nonnull_cb(fb, i))->format); 609 } 610 for (; i < 1; i++) { 611 OUT_CS(R300_US_OUT_FMT_C4_8 | 612 R300_C0_SEL_B | R300_C1_SEL_G | 613 R300_C2_SEL_R | R300_C3_SEL_A); 614 } 615 for (; i < 4; i++) { 616 OUT_CS(R300_US_OUT_FMT_UNUSED); 617 } 618 619 /* Set sample positions. It depends on the framebuffer sample count. 620 * These are pipelined regs and as such cannot be moved to the AA state. 621 */ 622 switch (r300->num_samples) { 623 default: 624 mspos0 = r300_get_mspos(0, sample_locs_1x); 625 mspos1 = r300_get_mspos(1, sample_locs_1x); 626 break; 627 case 2: 628 mspos0 = r300_get_mspos(0, sample_locs_2x); 629 mspos1 = r300_get_mspos(1, sample_locs_2x); 630 break; 631 case 4: 632 mspos0 = r300_get_mspos(0, sample_locs_4x); 633 mspos1 = r300_get_mspos(1, sample_locs_4x); 634 break; 635 case 6: 636 mspos0 = r300_get_mspos(0, sample_locs_6x); 637 mspos1 = r300_get_mspos(1, sample_locs_6x); 638 break; 639 } 640 641 OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2); 642 OUT_CS(mspos0); 643 OUT_CS(mspos1); 644 END_CS; 645} 646 647void r300_emit_query_start(struct r300_context *r300, unsigned size, void*state) 648{ 649 struct r300_query *query = r300->query_current; 650 CS_LOCALS(r300); 651 652 if (!query) 653 return; 654 655 BEGIN_CS(size); 656 if (r300->screen->caps.family == CHIP_RV530) { 657 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); 658 } else { 659 OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL); 660 } 661 OUT_CS_REG(R300_ZB_ZPASS_DATA, 0); 662 END_CS; 663 query->begin_emitted = TRUE; 664} 665 666static void r300_emit_query_end_frag_pipes(struct r300_context *r300, 667 struct r300_query *query) 668{ 669 struct r300_capabilities* caps = &r300->screen->caps; 670 uint32_t gb_pipes = r300->screen->info.r300_num_gb_pipes; 671 CS_LOCALS(r300); 672 673 assert(gb_pipes); 674 675 BEGIN_CS(6 * gb_pipes + 2); 676 /* I'm not so sure I like this switch, but it's hard to be elegant 677 * when there's so many special cases... 678 * 679 * So here's the basic idea. For each pipe, enable writes to it only, 680 * then put out the relocation for ZPASS_ADDR, taking into account a 681 * 4-byte offset for each pipe. RV380 and older are special; they have 682 * only two pipes, and the second pipe's enable is on bit 3, not bit 1, 683 * so there's a chipset cap for that. */ 684 switch (gb_pipes) { 685 case 4: 686 /* pipe 3 only */ 687 OUT_CS_REG(R300_SU_REG_DEST, 1 << 3); 688 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 3) * 4); 689 OUT_CS_RELOC(r300->query_current); 690 case 3: 691 /* pipe 2 only */ 692 OUT_CS_REG(R300_SU_REG_DEST, 1 << 2); 693 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 2) * 4); 694 OUT_CS_RELOC(r300->query_current); 695 case 2: 696 /* pipe 1 only */ 697 /* As mentioned above, accomodate RV380 and older. */ 698 OUT_CS_REG(R300_SU_REG_DEST, 699 1 << (caps->high_second_pipe ? 3 : 1)); 700 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4); 701 OUT_CS_RELOC(r300->query_current); 702 case 1: 703 /* pipe 0 only */ 704 OUT_CS_REG(R300_SU_REG_DEST, 1 << 0); 705 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4); 706 OUT_CS_RELOC(r300->query_current); 707 break; 708 default: 709 fprintf(stderr, "r300: Implementation error: Chipset reports %d" 710 " pixel pipes!\n", gb_pipes); 711 abort(); 712 } 713 714 /* And, finally, reset it to normal... */ 715 OUT_CS_REG(R300_SU_REG_DEST, 0xF); 716 END_CS; 717} 718 719static void rv530_emit_query_end_single_z(struct r300_context *r300, 720 struct r300_query *query) 721{ 722 CS_LOCALS(r300); 723 724 BEGIN_CS(8); 725 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); 726 OUT_CS_REG(R300_ZB_ZPASS_ADDR, query->num_results * 4); 727 OUT_CS_RELOC(r300->query_current); 728 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); 729 END_CS; 730} 731 732static void rv530_emit_query_end_double_z(struct r300_context *r300, 733 struct r300_query *query) 734{ 735 CS_LOCALS(r300); 736 737 BEGIN_CS(14); 738 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); 739 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4); 740 OUT_CS_RELOC(r300->query_current); 741 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1); 742 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4); 743 OUT_CS_RELOC(r300->query_current); 744 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); 745 END_CS; 746} 747 748void r300_emit_query_end(struct r300_context* r300) 749{ 750 struct r300_capabilities *caps = &r300->screen->caps; 751 struct r300_query *query = r300->query_current; 752 753 if (!query) 754 return; 755 756 if (query->begin_emitted == FALSE) 757 return; 758 759 if (caps->family == CHIP_RV530) { 760 if (r300->screen->info.r300_num_z_pipes == 2) 761 rv530_emit_query_end_double_z(r300, query); 762 else 763 rv530_emit_query_end_single_z(r300, query); 764 } else 765 r300_emit_query_end_frag_pipes(r300, query); 766 767 query->begin_emitted = FALSE; 768 query->num_results += query->num_pipes; 769 770 /* XXX grab all the results and reset the counter. */ 771 if (query->num_results >= query->buf->size / 4 - 4) { 772 query->num_results = (query->buf->size / 4) / 2; 773 fprintf(stderr, "r300: Rewinding OQBO...\n"); 774 } 775} 776 777void r300_emit_invariant_state(struct r300_context *r300, 778 unsigned size, void *state) 779{ 780 CS_LOCALS(r300); 781 WRITE_CS_TABLE(state, size); 782} 783 784void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state) 785{ 786 struct r300_rs_state* rs = state; 787 CS_LOCALS(r300); 788 789 BEGIN_CS(size); 790 OUT_CS_TABLE(rs->cb_main, RS_STATE_MAIN_SIZE); 791 if (rs->polygon_offset_enable) { 792 if (r300->zbuffer_bpp == 16) { 793 OUT_CS_TABLE(rs->cb_poly_offset_zb16, 5); 794 } else { 795 OUT_CS_TABLE(rs->cb_poly_offset_zb24, 5); 796 } 797 } 798 END_CS; 799} 800 801void r300_emit_rs_block_state(struct r300_context* r300, 802 unsigned size, void* state) 803{ 804 struct r300_rs_block* rs = (struct r300_rs_block*)state; 805 unsigned i; 806 /* It's the same for both INST and IP tables */ 807 unsigned count = (rs->inst_count & R300_RS_INST_COUNT_MASK) + 1; 808 CS_LOCALS(r300); 809 810 if (DBG_ON(r300, DBG_RS_BLOCK)) { 811 r500_dump_rs_block(rs); 812 813 fprintf(stderr, "r300: RS emit:\n"); 814 815 for (i = 0; i < count; i++) 816 fprintf(stderr, " : ip %d: 0x%08x\n", i, rs->ip[i]); 817 818 for (i = 0; i < count; i++) 819 fprintf(stderr, " : inst %d: 0x%08x\n", i, rs->inst[i]); 820 821 fprintf(stderr, " : count: 0x%08x inst_count: 0x%08x\n", 822 rs->count, rs->inst_count); 823 } 824 825 BEGIN_CS(size); 826 OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2); 827 OUT_CS(rs->vap_vtx_state_cntl); 828 OUT_CS(rs->vap_vsm_vtx_assm); 829 OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); 830 OUT_CS(rs->vap_out_vtx_fmt[0]); 831 OUT_CS(rs->vap_out_vtx_fmt[1]); 832 OUT_CS_REG_SEQ(R300_GB_ENABLE, 1); 833 OUT_CS(rs->gb_enable); 834 835 if (r300->screen->caps.is_r500) { 836 OUT_CS_REG_SEQ(R500_RS_IP_0, count); 837 } else { 838 OUT_CS_REG_SEQ(R300_RS_IP_0, count); 839 } 840 OUT_CS_TABLE(rs->ip, count); 841 842 OUT_CS_REG_SEQ(R300_RS_COUNT, 2); 843 OUT_CS(rs->count); 844 OUT_CS(rs->inst_count); 845 846 if (r300->screen->caps.is_r500) { 847 OUT_CS_REG_SEQ(R500_RS_INST_0, count); 848 } else { 849 OUT_CS_REG_SEQ(R300_RS_INST_0, count); 850 } 851 OUT_CS_TABLE(rs->inst, count); 852 END_CS; 853} 854 855void r300_emit_sample_mask(struct r300_context *r300, 856 unsigned size, void *state) 857{ 858 unsigned mask = (*(unsigned*)state) & ((1 << 6)-1); 859 CS_LOCALS(r300); 860 861 BEGIN_CS(size); 862 OUT_CS_REG(R300_SC_SCREENDOOR, 863 mask | (mask << 6) | (mask << 12) | (mask << 18)); 864 END_CS; 865} 866 867void r300_emit_scissor_state(struct r300_context* r300, 868 unsigned size, void* state) 869{ 870 struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state; 871 CS_LOCALS(r300); 872 873 BEGIN_CS(size); 874 OUT_CS_REG_SEQ(R300_SC_CLIPRECT_TL_0, 2); 875 if (r300->screen->caps.is_r500) { 876 OUT_CS((scissor->minx << R300_CLIPRECT_X_SHIFT) | 877 (scissor->miny << R300_CLIPRECT_Y_SHIFT)); 878 OUT_CS(((scissor->maxx - 1) << R300_CLIPRECT_X_SHIFT) | 879 ((scissor->maxy - 1) << R300_CLIPRECT_Y_SHIFT)); 880 } else { 881 OUT_CS(((scissor->minx + 1440) << R300_CLIPRECT_X_SHIFT) | 882 ((scissor->miny + 1440) << R300_CLIPRECT_Y_SHIFT)); 883 OUT_CS(((scissor->maxx + 1440-1) << R300_CLIPRECT_X_SHIFT) | 884 ((scissor->maxy + 1440-1) << R300_CLIPRECT_Y_SHIFT)); 885 } 886 END_CS; 887} 888 889void r300_emit_textures_state(struct r300_context *r300, 890 unsigned size, void *state) 891{ 892 struct r300_textures_state *allstate = (struct r300_textures_state*)state; 893 struct r300_texture_sampler_state *texstate; 894 struct r300_resource *tex; 895 unsigned i; 896 boolean has_us_format = r300->screen->caps.has_us_format; 897 CS_LOCALS(r300); 898 899 BEGIN_CS(size); 900 OUT_CS_REG(R300_TX_ENABLE, allstate->tx_enable); 901 902 for (i = 0; i < allstate->count; i++) { 903 if ((1 << i) & allstate->tx_enable) { 904 texstate = &allstate->regs[i]; 905 tex = r300_resource(allstate->sampler_views[i]->base.texture); 906 907 OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter0); 908 OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter1); 909 OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (i * 4), 910 texstate->border_color); 911 912 OUT_CS_REG(R300_TX_FORMAT0_0 + (i * 4), texstate->format.format0); 913 OUT_CS_REG(R300_TX_FORMAT1_0 + (i * 4), texstate->format.format1); 914 OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format.format2); 915 916 OUT_CS_REG(R300_TX_OFFSET_0 + (i * 4), texstate->format.tile_config); 917 OUT_CS_RELOC(tex); 918 919 if (has_us_format) { 920 OUT_CS_REG(R500_US_FORMAT0_0 + (i * 4), 921 texstate->format.us_format0); 922 } 923 } 924 } 925 END_CS; 926} 927 928void r300_emit_vertex_arrays(struct r300_context* r300, int offset, 929 boolean indexed, int instance_id) 930{ 931 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; 932 struct pipe_vertex_element *velem = r300->velems->velem; 933 struct r300_resource *buf; 934 int i; 935 unsigned vertex_array_count = r300->velems->count; 936 unsigned packet_size = (vertex_array_count * 3 + 1) / 2; 937 struct pipe_vertex_buffer *vb1, *vb2; 938 unsigned *hw_format_size = r300->velems->format_size; 939 unsigned size1, size2, offset1, offset2, stride1, stride2; 940 CS_LOCALS(r300); 941 942 BEGIN_CS(2 + packet_size + vertex_array_count * 2); 943 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size); 944 OUT_CS(vertex_array_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0)); 945 946 if (instance_id == -1) { 947 /* Non-instanced arrays. This ignores instance_divisor and instance_id. */ 948 for (i = 0; i < vertex_array_count - 1; i += 2) { 949 vb1 = &vbuf[velem[i].vertex_buffer_index]; 950 vb2 = &vbuf[velem[i+1].vertex_buffer_index]; 951 size1 = hw_format_size[i]; 952 size2 = hw_format_size[i+1]; 953 954 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) | 955 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride)); 956 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride); 957 OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride); 958 } 959 960 if (vertex_array_count & 1) { 961 vb1 = &vbuf[velem[i].vertex_buffer_index]; 962 size1 = hw_format_size[i]; 963 964 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride)); 965 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride); 966 } 967 968 for (i = 0; i < vertex_array_count; i++) { 969 buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer); 970 OUT_CS_RELOC(buf); 971 } 972 } else { 973 /* Instanced arrays. */ 974 for (i = 0; i < vertex_array_count - 1; i += 2) { 975 vb1 = &vbuf[velem[i].vertex_buffer_index]; 976 vb2 = &vbuf[velem[i+1].vertex_buffer_index]; 977 size1 = hw_format_size[i]; 978 size2 = hw_format_size[i+1]; 979 980 if (velem[i].instance_divisor) { 981 stride1 = 0; 982 offset1 = vb1->buffer_offset + velem[i].src_offset + 983 (instance_id / velem[i].instance_divisor) * vb1->stride; 984 } else { 985 stride1 = vb1->stride; 986 offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride; 987 } 988 if (velem[i+1].instance_divisor) { 989 stride2 = 0; 990 offset2 = vb2->buffer_offset + velem[i+1].src_offset + 991 (instance_id / velem[i+1].instance_divisor) * vb2->stride; 992 } else { 993 stride2 = vb2->stride; 994 offset2 = vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride; 995 } 996 997 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1) | 998 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(stride2)); 999 OUT_CS(offset1); 1000 OUT_CS(offset2); 1001 } 1002 1003 if (vertex_array_count & 1) { 1004 vb1 = &vbuf[velem[i].vertex_buffer_index]; 1005 size1 = hw_format_size[i]; 1006 1007 if (velem[i].instance_divisor) { 1008 stride1 = 0; 1009 offset1 = vb1->buffer_offset + velem[i].src_offset + 1010 (instance_id / velem[i].instance_divisor) * vb1->stride; 1011 } else { 1012 stride1 = vb1->stride; 1013 offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride; 1014 } 1015 1016 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1)); 1017 OUT_CS(offset1); 1018 } 1019 1020 for (i = 0; i < vertex_array_count; i++) { 1021 buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer); 1022 OUT_CS_RELOC(buf); 1023 } 1024 } 1025 END_CS; 1026} 1027 1028void r300_emit_vertex_arrays_swtcl(struct r300_context *r300, boolean indexed) 1029{ 1030 CS_LOCALS(r300); 1031 1032 DBG(r300, DBG_SWTCL, "r300: Preparing vertex buffer %p for render, " 1033 "vertex size %d\n", r300->vbo, 1034 r300->vertex_info.size); 1035 /* Set the pointer to our vertex buffer. The emitted values are this: 1036 * PACKET3 [3D_LOAD_VBPNTR] 1037 * COUNT [1] 1038 * FORMAT [size | stride << 8] 1039 * OFFSET [offset into BO] 1040 * VBPNTR [relocated BO] 1041 */ 1042 BEGIN_CS(7); 1043 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3); 1044 OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0)); 1045 OUT_CS(r300->vertex_info.size | 1046 (r300->vertex_info.size << 8)); 1047 OUT_CS(r300->draw_vbo_offset); 1048 OUT_CS(0); 1049 1050 assert(r300->vbo_cs); 1051 OUT_CS(0xc0001000); /* PKT3_NOP */ 1052 OUT_CS(r300->rws->cs_get_reloc(r300->cs, r300->vbo_cs) * 4); 1053 END_CS; 1054} 1055 1056void r300_emit_vertex_stream_state(struct r300_context* r300, 1057 unsigned size, void* state) 1058{ 1059 struct r300_vertex_stream_state *streams = 1060 (struct r300_vertex_stream_state*)state; 1061 unsigned i; 1062 CS_LOCALS(r300); 1063 1064 if (DBG_ON(r300, DBG_PSC)) { 1065 fprintf(stderr, "r300: PSC emit:\n"); 1066 1067 for (i = 0; i < streams->count; i++) { 1068 fprintf(stderr, " : prog_stream_cntl%d: 0x%08x\n", i, 1069 streams->vap_prog_stream_cntl[i]); 1070 } 1071 1072 for (i = 0; i < streams->count; i++) { 1073 fprintf(stderr, " : prog_stream_cntl_ext%d: 0x%08x\n", i, 1074 streams->vap_prog_stream_cntl_ext[i]); 1075 } 1076 } 1077 1078 BEGIN_CS(size); 1079 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count); 1080 OUT_CS_TABLE(streams->vap_prog_stream_cntl, streams->count); 1081 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count); 1082 OUT_CS_TABLE(streams->vap_prog_stream_cntl_ext, streams->count); 1083 END_CS; 1084} 1085 1086void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state) 1087{ 1088 CS_LOCALS(r300); 1089 1090 BEGIN_CS(size); 1091 OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); 1092 END_CS; 1093} 1094 1095void r300_emit_vap_invariant_state(struct r300_context *r300, 1096 unsigned size, void *state) 1097{ 1098 CS_LOCALS(r300); 1099 WRITE_CS_TABLE(state, size); 1100} 1101 1102void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) 1103{ 1104 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)state; 1105 struct r300_vertex_program_code* code = &vs->code; 1106 struct r300_screen* r300screen = r300->screen; 1107 unsigned instruction_count = code->length / 4; 1108 1109 unsigned vtx_mem_size = r300screen->caps.is_r500 ? 128 : 72; 1110 unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1); 1111 unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1); 1112 unsigned temp_count = MAX2(code->num_temporaries, 1); 1113 1114 unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count, 1115 vtx_mem_size / output_count, 10); 1116 unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5); 1117 1118 CS_LOCALS(r300); 1119 1120 BEGIN_CS(size); 1121 1122 /* R300_VAP_PVS_CODE_CNTL_0 1123 * R300_VAP_PVS_CONST_CNTL 1124 * R300_VAP_PVS_CODE_CNTL_1 1125 * See the r5xx docs for instructions on how to use these. */ 1126 OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) | 1127 R300_PVS_XYZW_VALID_INST(instruction_count - 1) | 1128 R300_PVS_LAST_INST(instruction_count - 1)); 1129 OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, instruction_count - 1); 1130 1131 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0); 1132 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length); 1133 OUT_CS_TABLE(code->body.d, code->length); 1134 1135 OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) | 1136 R300_PVS_NUM_CNTLRS(pvs_num_controllers) | 1137 R300_PVS_NUM_FPUS(r300screen->caps.num_vert_fpus) | 1138 R300_PVS_VF_MAX_VTX_NUM(12) | 1139 (r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0)); 1140 1141 /* Emit flow control instructions. Even if there are no fc instructions, 1142 * we still need to write the registers to make sure they are cleared. */ 1143 OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops); 1144 if (r300screen->caps.is_r500) { 1145 OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, R300_VS_MAX_FC_OPS * 2); 1146 OUT_CS_TABLE(code->fc_op_addrs.r500, R300_VS_MAX_FC_OPS * 2); 1147 } else { 1148 OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, R300_VS_MAX_FC_OPS); 1149 OUT_CS_TABLE(code->fc_op_addrs.r300, R300_VS_MAX_FC_OPS); 1150 } 1151 OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, R300_VS_MAX_FC_OPS); 1152 OUT_CS_TABLE(code->fc_loop_index, R300_VS_MAX_FC_OPS); 1153 1154 END_CS; 1155} 1156 1157void r300_emit_vs_constants(struct r300_context* r300, 1158 unsigned size, void *state) 1159{ 1160 unsigned count = 1161 ((struct r300_vertex_shader*)r300->vs_state.state)->externals_count; 1162 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; 1163 struct r300_vertex_shader *vs = (struct r300_vertex_shader*)r300->vs_state.state; 1164 unsigned i; 1165 int imm_first = vs->externals_count; 1166 int imm_end = vs->code.constants.Count; 1167 int imm_count = vs->immediates_count; 1168 CS_LOCALS(r300); 1169 1170 BEGIN_CS(size); 1171 OUT_CS_REG(R300_VAP_PVS_CONST_CNTL, 1172 R300_PVS_CONST_BASE_OFFSET(buf->buffer_base) | 1173 R300_PVS_MAX_CONST_ADDR(MAX2(imm_end - 1, 0))); 1174 if (vs->externals_count) { 1175 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 1176 (r300->screen->caps.is_r500 ? 1177 R500_PVS_CONST_START : R300_PVS_CONST_START) + buf->buffer_base); 1178 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4); 1179 if (buf->remap_table){ 1180 for (i = 0; i < count; i++) { 1181 uint32_t *data = &buf->ptr[buf->remap_table[i]*4]; 1182 OUT_CS_TABLE(data, 4); 1183 } 1184 } else { 1185 OUT_CS_TABLE(buf->ptr, count * 4); 1186 } 1187 } 1188 1189 /* Emit immediates. */ 1190 if (imm_count) { 1191 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 1192 (r300->screen->caps.is_r500 ? 1193 R500_PVS_CONST_START : R300_PVS_CONST_START) + 1194 buf->buffer_base + imm_first); 1195 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4); 1196 for (i = imm_first; i < imm_end; i++) { 1197 const float *data = vs->code.constants.Constants[i].u.Immediate; 1198 OUT_CS_TABLE(data, 4); 1199 } 1200 } 1201 END_CS; 1202} 1203 1204void r300_emit_viewport_state(struct r300_context* r300, 1205 unsigned size, void* state) 1206{ 1207 struct r300_viewport_state* viewport = (struct r300_viewport_state*)state; 1208 CS_LOCALS(r300); 1209 1210 BEGIN_CS(size); 1211 OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); 1212 OUT_CS_TABLE(&viewport->xscale, 6); 1213 OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); 1214 END_CS; 1215} 1216 1217void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state) 1218{ 1219 struct pipe_framebuffer_state *fb = 1220 (struct pipe_framebuffer_state*)r300->fb_state.state; 1221 struct r300_resource* tex; 1222 CS_LOCALS(r300); 1223 1224 tex = r300_resource(fb->zsbuf->texture); 1225 1226 BEGIN_CS(size); 1227 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2); 1228 OUT_CS(0); 1229 OUT_CS(tex->tex.hiz_dwords[fb->zsbuf->u.tex.level]); 1230 OUT_CS(r300->hiz_clear_value); 1231 END_CS; 1232 1233 /* Mark the current zbuffer's hiz ram as in use. */ 1234 r300->hiz_in_use = TRUE; 1235 r300->hiz_func = HIZ_FUNC_NONE; 1236 r300_mark_atom_dirty(r300, &r300->hyperz_state); 1237} 1238 1239void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state) 1240{ 1241 struct pipe_framebuffer_state *fb = 1242 (struct pipe_framebuffer_state*)r300->fb_state.state; 1243 struct r300_resource *tex; 1244 CS_LOCALS(r300); 1245 1246 tex = r300_resource(fb->zsbuf->texture); 1247 1248 BEGIN_CS(size); 1249 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2); 1250 OUT_CS(0); 1251 OUT_CS(tex->tex.zmask_dwords[fb->zsbuf->u.tex.level]); 1252 OUT_CS(0); 1253 END_CS; 1254 1255 /* Mark the current zbuffer's zmask as in use. */ 1256 r300->zmask_in_use = TRUE; 1257 r300_mark_atom_dirty(r300, &r300->hyperz_state); 1258} 1259 1260void r300_emit_cmask_clear(struct r300_context *r300, unsigned size, void *state) 1261{ 1262 struct pipe_framebuffer_state *fb = 1263 (struct pipe_framebuffer_state*)r300->fb_state.state; 1264 struct r300_resource *tex; 1265 CS_LOCALS(r300); 1266 1267 tex = r300_resource(fb->cbufs[0]->texture); 1268 1269 BEGIN_CS(size); 1270 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_CMASK, 2); 1271 OUT_CS(0); 1272 OUT_CS(tex->tex.cmask_dwords); 1273 OUT_CS(0); 1274 END_CS; 1275 1276 /* Mark the current zbuffer's zmask as in use. */ 1277 r300->cmask_in_use = TRUE; 1278 r300_mark_fb_state_dirty(r300, R300_CHANGED_CMASK_ENABLE); 1279} 1280 1281void r300_emit_ztop_state(struct r300_context* r300, 1282 unsigned size, void* state) 1283{ 1284 struct r300_ztop_state* ztop = (struct r300_ztop_state*)state; 1285 CS_LOCALS(r300); 1286 1287 BEGIN_CS(size); 1288 OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top); 1289 END_CS; 1290} 1291 1292void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state) 1293{ 1294 CS_LOCALS(r300); 1295 1296 BEGIN_CS(size); 1297 OUT_CS_REG(R300_TX_INVALTAGS, 0); 1298 END_CS; 1299} 1300 1301boolean r300_emit_buffer_validate(struct r300_context *r300, 1302 boolean do_validate_vertex_buffers, 1303 struct pipe_resource *index_buffer) 1304{ 1305 struct pipe_framebuffer_state *fb = 1306 (struct pipe_framebuffer_state*)r300->fb_state.state; 1307 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 1308 struct r300_textures_state *texstate = 1309 (struct r300_textures_state*)r300->textures_state.state; 1310 struct r300_resource *tex; 1311 unsigned i; 1312 boolean flushed = FALSE; 1313 1314validate: 1315 if (r300->fb_state.dirty) { 1316 /* Color buffers... */ 1317 for (i = 0; i < fb->nr_cbufs; i++) { 1318 if (!fb->cbufs[i]) 1319 continue; 1320 tex = r300_resource(fb->cbufs[i]->texture); 1321 assert(tex && tex->buf && "cbuf is marked, but NULL!"); 1322 r300->rws->cs_add_reloc(r300->cs, tex->cs_buf, 1323 RADEON_USAGE_READWRITE, 1324 r300_surface(fb->cbufs[i])->domain, 1325 tex->b.b.nr_samples > 1 ? 1326 RADEON_PRIO_COLOR_BUFFER_MSAA : 1327 RADEON_PRIO_COLOR_BUFFER); 1328 } 1329 /* ...depth buffer... */ 1330 if (fb->zsbuf) { 1331 tex = r300_resource(fb->zsbuf->texture); 1332 assert(tex && tex->buf && "zsbuf is marked, but NULL!"); 1333 r300->rws->cs_add_reloc(r300->cs, tex->cs_buf, 1334 RADEON_USAGE_READWRITE, 1335 r300_surface(fb->zsbuf)->domain, 1336 tex->b.b.nr_samples > 1 ? 1337 RADEON_PRIO_DEPTH_BUFFER_MSAA : 1338 RADEON_PRIO_DEPTH_BUFFER); 1339 } 1340 } 1341 /* The AA resolve buffer. */ 1342 if (r300->aa_state.dirty) { 1343 if (aa->dest) { 1344 r300->rws->cs_add_reloc(r300->cs, aa->dest->cs_buf, 1345 RADEON_USAGE_WRITE, 1346 aa->dest->domain, 1347 RADEON_PRIO_COLOR_BUFFER); 1348 } 1349 } 1350 if (r300->textures_state.dirty) { 1351 /* ...textures... */ 1352 for (i = 0; i < texstate->count; i++) { 1353 if (!(texstate->tx_enable & (1 << i))) { 1354 continue; 1355 } 1356 1357 tex = r300_resource(texstate->sampler_views[i]->base.texture); 1358 r300->rws->cs_add_reloc(r300->cs, tex->cs_buf, RADEON_USAGE_READ, 1359 tex->domain, RADEON_PRIO_SHADER_TEXTURE_RO); 1360 } 1361 } 1362 /* ...occlusion query buffer... */ 1363 if (r300->query_current) 1364 r300->rws->cs_add_reloc(r300->cs, r300->query_current->cs_buf, 1365 RADEON_USAGE_WRITE, RADEON_DOMAIN_GTT, 1366 RADEON_PRIO_MIN); 1367 /* ...vertex buffer for SWTCL path... */ 1368 if (r300->vbo_cs) 1369 r300->rws->cs_add_reloc(r300->cs, r300->vbo_cs, 1370 RADEON_USAGE_READ, RADEON_DOMAIN_GTT, 1371 RADEON_PRIO_MIN); 1372 /* ...vertex buffers for HWTCL path... */ 1373 if (do_validate_vertex_buffers && r300->vertex_arrays_dirty) { 1374 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; 1375 struct pipe_vertex_buffer *last = r300->vertex_buffer + 1376 r300->nr_vertex_buffers; 1377 struct pipe_resource *buf; 1378 1379 for (; vbuf != last; vbuf++) { 1380 buf = vbuf->buffer; 1381 if (!buf) 1382 continue; 1383 1384 r300->rws->cs_add_reloc(r300->cs, r300_resource(buf)->cs_buf, 1385 RADEON_USAGE_READ, 1386 r300_resource(buf)->domain, 1387 RADEON_PRIO_SHADER_BUFFER_RO); 1388 } 1389 } 1390 /* ...and index buffer for HWTCL path. */ 1391 if (index_buffer) 1392 r300->rws->cs_add_reloc(r300->cs, r300_resource(index_buffer)->cs_buf, 1393 RADEON_USAGE_READ, 1394 r300_resource(index_buffer)->domain, 1395 RADEON_PRIO_MIN); 1396 1397 /* Now do the validation (flush is called inside cs_validate on failure). */ 1398 if (!r300->rws->cs_validate(r300->cs)) { 1399 /* Ooops, an infinite loop, give up. */ 1400 if (flushed) 1401 return FALSE; 1402 1403 flushed = TRUE; 1404 goto validate; 1405 } 1406 1407 return TRUE; 1408} 1409 1410unsigned r300_get_num_dirty_dwords(struct r300_context *r300) 1411{ 1412 struct r300_atom* atom; 1413 unsigned dwords = 0; 1414 1415 foreach_dirty_atom(r300, atom) { 1416 if (atom->dirty) { 1417 dwords += atom->size; 1418 } 1419 } 1420 1421 /* let's reserve some more, just in case */ 1422 dwords += 32; 1423 1424 return dwords; 1425} 1426 1427unsigned r300_get_num_cs_end_dwords(struct r300_context *r300) 1428{ 1429 unsigned dwords = 0; 1430 1431 /* Emitted in flush. */ 1432 dwords += 26; /* emit_query_end */ 1433 dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */ 1434 if (r300->screen->caps.is_r500) 1435 dwords += 2; /* emit_index_bias */ 1436 if (r300->screen->info.drm_minor >= 6) 1437 dwords += 3; /* MSPOS */ 1438 1439 return dwords; 1440} 1441 1442/* Emit all dirty state. */ 1443void r300_emit_dirty_state(struct r300_context* r300) 1444{ 1445 struct r300_atom *atom; 1446 1447 foreach_dirty_atom(r300, atom) { 1448 if (atom->dirty) { 1449 atom->emit(r300, atom->size, atom->state); 1450 atom->dirty = FALSE; 1451 } 1452 } 1453 1454 r300->first_dirty = NULL; 1455 r300->last_dirty = NULL; 1456 r300->dirty_hw++; 1457} 1458