1/* 2 * Copyright 2008 Ben Skeggs 3 * Copyright 2010 Christoph Bumiller 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 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 9 * and/or sell copies of the Software, and to permit persons to whom the 10 * Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice shall be included in 13 * all copies or substantial portions of the Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR 19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 21 * OTHER DEALINGS IN THE SOFTWARE. 22 */ 23 24#include "pipe/p_context.h" 25#include "pipe/p_defines.h" 26#include "pipe/p_state.h" 27#include "util/u_inlines.h" 28 29#include "nv50/nv50_context.h" 30#include "nv50/nv50_query_hw.h" 31 32#include "nv50/nv50_compute.xml.h" 33 34void 35nv50_constbufs_validate(struct nv50_context *nv50) 36{ 37 struct nouveau_pushbuf *push = nv50->base.pushbuf; 38 unsigned s; 39 40 for (s = 0; s < 3; ++s) { 41 unsigned p; 42 43 if (s == PIPE_SHADER_FRAGMENT) 44 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_FRAGMENT; 45 else 46 if (s == PIPE_SHADER_GEOMETRY) 47 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_GEOMETRY; 48 else 49 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_VERTEX; 50 51 while (nv50->constbuf_dirty[s]) { 52 const unsigned i = (unsigned)ffs(nv50->constbuf_dirty[s]) - 1; 53 54 assert(i < NV50_MAX_PIPE_CONSTBUFS); 55 nv50->constbuf_dirty[s] &= ~(1 << i); 56 57 if (nv50->constbuf[s][i].user) { 58 const unsigned b = NV50_CB_PVP + s; 59 unsigned start = 0; 60 unsigned words = nv50->constbuf[s][0].size / 4; 61 if (i) { 62 NOUVEAU_ERR("user constbufs only supported in slot 0\n"); 63 continue; 64 } 65 if (!nv50->state.uniform_buffer_bound[s]) { 66 nv50->state.uniform_buffer_bound[s] = true; 67 BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1); 68 PUSH_DATA (push, (b << 12) | (i << 8) | p | 1); 69 } 70 while (words) { 71 unsigned nr = MIN2(words, NV04_PFIFO_MAX_PACKET_LEN); 72 73 PUSH_SPACE(push, nr + 3); 74 BEGIN_NV04(push, NV50_3D(CB_ADDR), 1); 75 PUSH_DATA (push, (start << 8) | b); 76 BEGIN_NI04(push, NV50_3D(CB_DATA(0)), nr); 77 PUSH_DATAp(push, &nv50->constbuf[s][0].u.data[start * 4], nr); 78 79 start += nr; 80 words -= nr; 81 } 82 } else { 83 struct nv04_resource *res = 84 nv04_resource(nv50->constbuf[s][i].u.buf); 85 if (res) { 86 /* TODO: allocate persistent bindings */ 87 const unsigned b = s * 16 + i; 88 89 assert(nouveau_resource_mapped_by_gpu(&res->base)); 90 91 BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3); 92 PUSH_DATAh(push, res->address + nv50->constbuf[s][i].offset); 93 PUSH_DATA (push, res->address + nv50->constbuf[s][i].offset); 94 PUSH_DATA (push, (b << 16) | 95 (nv50->constbuf[s][i].size & 0xffff)); 96 BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1); 97 PUSH_DATA (push, (b << 12) | (i << 8) | p | 1); 98 99 BCTX_REFN(nv50->bufctx_3d, 3D_CB(s, i), res, RD); 100 101 nv50->cb_dirty = 1; /* Force cache flush for UBO. */ 102 res->cb_bindings[s] |= 1 << i; 103 } else { 104 BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1); 105 PUSH_DATA (push, (i << 8) | p | 0); 106 } 107 if (i == 0) 108 nv50->state.uniform_buffer_bound[s] = false; 109 } 110 } 111 } 112} 113 114static bool 115nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog) 116{ 117 if (!prog->translated) { 118 prog->translated = nv50_program_translate( 119 prog, nv50->screen->base.device->chipset, &nv50->base.debug); 120 if (!prog->translated) 121 return false; 122 } else 123 if (prog->mem) 124 return true; 125 126 return nv50_program_upload_code(nv50, prog); 127} 128 129static inline void 130nv50_program_update_context_state(struct nv50_context *nv50, 131 struct nv50_program *prog, int stage) 132{ 133 const unsigned flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR; 134 135 if (prog && prog->tls_space) { 136 if (nv50->state.new_tls_space) 137 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_TLS); 138 if (!nv50->state.tls_required || nv50->state.new_tls_space) 139 BCTX_REFN_bo(nv50->bufctx_3d, 3D_TLS, flags, nv50->screen->tls_bo); 140 nv50->state.new_tls_space = false; 141 nv50->state.tls_required |= 1 << stage; 142 } else { 143 if (nv50->state.tls_required == (1 << stage)) 144 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_TLS); 145 nv50->state.tls_required &= ~(1 << stage); 146 } 147} 148 149void 150nv50_vertprog_validate(struct nv50_context *nv50) 151{ 152 struct nouveau_pushbuf *push = nv50->base.pushbuf; 153 struct nv50_program *vp = nv50->vertprog; 154 155 if (!nv50_program_validate(nv50, vp)) 156 return; 157 nv50_program_update_context_state(nv50, vp, 0); 158 159 BEGIN_NV04(push, NV50_3D(VP_ATTR_EN(0)), 2); 160 PUSH_DATA (push, vp->vp.attrs[0]); 161 PUSH_DATA (push, vp->vp.attrs[1]); 162 BEGIN_NV04(push, NV50_3D(VP_REG_ALLOC_RESULT), 1); 163 PUSH_DATA (push, vp->max_out); 164 BEGIN_NV04(push, NV50_3D(VP_REG_ALLOC_TEMP), 1); 165 PUSH_DATA (push, vp->max_gpr); 166 BEGIN_NV04(push, NV50_3D(VP_START_ID), 1); 167 PUSH_DATA (push, vp->code_base); 168} 169 170void 171nv50_fragprog_validate(struct nv50_context *nv50) 172{ 173 struct nouveau_pushbuf *push = nv50->base.pushbuf; 174 struct nv50_program *fp = nv50->fragprog; 175 struct pipe_rasterizer_state *rast = &nv50->rast->pipe; 176 177 if (nv50->zsa && nv50->zsa->pipe.alpha.enabled) { 178 struct pipe_framebuffer_state *fb = &nv50->framebuffer; 179 bool blendable = fb->nr_cbufs == 0 || !fb->cbufs[0] || 180 nv50->screen->base.base.is_format_supported( 181 &nv50->screen->base.base, 182 fb->cbufs[0]->format, 183 fb->cbufs[0]->texture->target, 184 fb->cbufs[0]->texture->nr_samples, 185 PIPE_BIND_BLENDABLE); 186 /* If we already have alphatest code, we have to keep updating 187 * it. However we only have to have different code if the current RT0 is 188 * non-blendable. Otherwise we just set it to always pass and use the 189 * hardware alpha test. 190 */ 191 if (fp->fp.alphatest || !blendable) { 192 uint8_t alphatest = PIPE_FUNC_ALWAYS + 1; 193 if (!blendable) 194 alphatest = nv50->zsa->pipe.alpha.func + 1; 195 if (!fp->fp.alphatest) 196 nv50_program_destroy(nv50, fp); 197 else if (fp->mem && fp->fp.alphatest != alphatest) 198 nouveau_heap_free(&fp->mem); 199 200 fp->fp.alphatest = alphatest; 201 } 202 } else if (fp->fp.alphatest && fp->fp.alphatest != PIPE_FUNC_ALWAYS + 1) { 203 /* Alpha test is disabled but we have a shader where it's filled 204 * in. Make sure to reset the function to 'always', otherwise it'll end 205 * up discarding fragments incorrectly. 206 */ 207 if (fp->mem) 208 nouveau_heap_free(&fp->mem); 209 210 fp->fp.alphatest = PIPE_FUNC_ALWAYS + 1; 211 } 212 213 if (fp->fp.force_persample_interp != rast->force_persample_interp) { 214 /* Force the program to be reuploaded, which will trigger interp fixups 215 * to get applied 216 */ 217 if (fp->mem) 218 nouveau_heap_free(&fp->mem); 219 220 fp->fp.force_persample_interp = rast->force_persample_interp; 221 } 222 223 if (fp->mem && !(nv50->dirty_3d & (NV50_NEW_3D_FRAGPROG | NV50_NEW_3D_MIN_SAMPLES))) 224 return; 225 226 if (!nv50_program_validate(nv50, fp)) 227 return; 228 nv50_program_update_context_state(nv50, fp, 1); 229 230 BEGIN_NV04(push, NV50_3D(FP_REG_ALLOC_TEMP), 1); 231 PUSH_DATA (push, fp->max_gpr); 232 BEGIN_NV04(push, NV50_3D(FP_RESULT_COUNT), 1); 233 PUSH_DATA (push, fp->max_out); 234 BEGIN_NV04(push, NV50_3D(FP_CONTROL), 1); 235 PUSH_DATA (push, fp->fp.flags[0]); 236 BEGIN_NV04(push, NV50_3D(FP_CTRL_UNK196C), 1); 237 PUSH_DATA (push, fp->fp.flags[1]); 238 BEGIN_NV04(push, NV50_3D(FP_START_ID), 1); 239 PUSH_DATA (push, fp->code_base); 240 241 if (nv50->screen->tesla->oclass >= NVA3_3D_CLASS) { 242 BEGIN_NV04(push, SUBC_3D(NVA3_3D_FP_MULTISAMPLE), 1); 243 if (nv50->min_samples > 1 || fp->fp.has_samplemask) 244 PUSH_DATA(push, 245 NVA3_3D_FP_MULTISAMPLE_FORCE_PER_SAMPLE | 246 (NVA3_3D_FP_MULTISAMPLE_EXPORT_SAMPLE_MASK * 247 fp->fp.has_samplemask)); 248 else 249 PUSH_DATA(push, 0); 250 } 251} 252 253void 254nv50_gmtyprog_validate(struct nv50_context *nv50) 255{ 256 struct nouveau_pushbuf *push = nv50->base.pushbuf; 257 struct nv50_program *gp = nv50->gmtyprog; 258 259 if (gp) { 260 if (!nv50_program_validate(nv50, gp)) 261 return; 262 BEGIN_NV04(push, NV50_3D(GP_REG_ALLOC_TEMP), 1); 263 PUSH_DATA (push, gp->max_gpr); 264 BEGIN_NV04(push, NV50_3D(GP_REG_ALLOC_RESULT), 1); 265 PUSH_DATA (push, gp->max_out); 266 BEGIN_NV04(push, NV50_3D(GP_OUTPUT_PRIMITIVE_TYPE), 1); 267 PUSH_DATA (push, gp->gp.prim_type); 268 BEGIN_NV04(push, NV50_3D(GP_VERTEX_OUTPUT_COUNT), 1); 269 PUSH_DATA (push, gp->gp.vert_count); 270 BEGIN_NV04(push, NV50_3D(GP_START_ID), 1); 271 PUSH_DATA (push, gp->code_base); 272 273 nv50->state.prim_size = gp->gp.prim_type; /* enum matches vertex count */ 274 } 275 nv50_program_update_context_state(nv50, gp, 2); 276 277 /* GP_ENABLE is updated in linkage validation */ 278} 279 280void 281nv50_compprog_validate(struct nv50_context *nv50) 282{ 283 struct nouveau_pushbuf *push = nv50->base.pushbuf; 284 struct nv50_program *cp = nv50->compprog; 285 286 if (cp && !nv50_program_validate(nv50, cp)) 287 return; 288 289 BEGIN_NV04(push, NV50_CP(CODE_CB_FLUSH), 1); 290 PUSH_DATA (push, 0); 291} 292 293static void 294nv50_sprite_coords_validate(struct nv50_context *nv50) 295{ 296 struct nouveau_pushbuf *push = nv50->base.pushbuf; 297 uint32_t pntc[8], mode; 298 struct nv50_program *fp = nv50->fragprog; 299 unsigned i, c; 300 unsigned m = (nv50->state.interpolant_ctrl >> 8) & 0xff; 301 302 if (!nv50->rast->pipe.point_quad_rasterization) { 303 if (nv50->state.point_sprite) { 304 BEGIN_NV04(push, NV50_3D(POINT_COORD_REPLACE_MAP(0)), 8); 305 for (i = 0; i < 8; ++i) 306 PUSH_DATA(push, 0); 307 308 nv50->state.point_sprite = false; 309 } 310 return; 311 } else { 312 nv50->state.point_sprite = true; 313 } 314 315 memset(pntc, 0, sizeof(pntc)); 316 317 for (i = 0; i < fp->in_nr; i++) { 318 unsigned n = util_bitcount(fp->in[i].mask); 319 320 if (fp->in[i].sn != TGSI_SEMANTIC_GENERIC) { 321 m += n; 322 continue; 323 } 324 if (!(nv50->rast->pipe.sprite_coord_enable & (1 << fp->in[i].si))) { 325 m += n; 326 continue; 327 } 328 329 for (c = 0; c < 4; ++c) { 330 if (fp->in[i].mask & (1 << c)) { 331 pntc[m / 8] |= (c + 1) << ((m % 8) * 4); 332 ++m; 333 } 334 } 335 } 336 337 if (nv50->rast->pipe.sprite_coord_mode == PIPE_SPRITE_COORD_LOWER_LEFT) 338 mode = 0x00; 339 else 340 mode = 0x10; 341 342 BEGIN_NV04(push, NV50_3D(POINT_SPRITE_CTRL), 1); 343 PUSH_DATA (push, mode); 344 345 BEGIN_NV04(push, NV50_3D(POINT_COORD_REPLACE_MAP(0)), 8); 346 PUSH_DATAp(push, pntc, 8); 347} 348 349/* Validate state derived from shaders and the rasterizer cso. */ 350void 351nv50_validate_derived_rs(struct nv50_context *nv50) 352{ 353 struct nouveau_pushbuf *push = nv50->base.pushbuf; 354 uint32_t color, psize; 355 356 nv50_sprite_coords_validate(nv50); 357 358 if (nv50->state.rasterizer_discard != nv50->rast->pipe.rasterizer_discard) { 359 nv50->state.rasterizer_discard = nv50->rast->pipe.rasterizer_discard; 360 BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1); 361 PUSH_DATA (push, !nv50->rast->pipe.rasterizer_discard); 362 } 363 364 if (nv50->dirty_3d & NV50_NEW_3D_FRAGPROG) 365 return; 366 psize = nv50->state.semantic_psize & ~NV50_3D_SEMANTIC_PTSZ_PTSZ_EN__MASK; 367 color = nv50->state.semantic_color & ~NV50_3D_SEMANTIC_COLOR_CLMP_EN; 368 369 if (nv50->rast->pipe.clamp_vertex_color) 370 color |= NV50_3D_SEMANTIC_COLOR_CLMP_EN; 371 372 if (color != nv50->state.semantic_color) { 373 nv50->state.semantic_color = color; 374 BEGIN_NV04(push, NV50_3D(SEMANTIC_COLOR), 1); 375 PUSH_DATA (push, color); 376 } 377 378 if (nv50->rast->pipe.point_size_per_vertex) 379 psize |= NV50_3D_SEMANTIC_PTSZ_PTSZ_EN__MASK; 380 381 if (psize != nv50->state.semantic_psize) { 382 nv50->state.semantic_psize = psize; 383 BEGIN_NV04(push, NV50_3D(SEMANTIC_PTSZ), 1); 384 PUSH_DATA (push, psize); 385 } 386} 387 388static int 389nv50_vec4_map(uint8_t *map, int mid, uint32_t lin[4], 390 struct nv50_varying *in, struct nv50_varying *out) 391{ 392 int c; 393 uint8_t mv = out->mask, mf = in->mask, oid = out->hw; 394 395 for (c = 0; c < 4; ++c) { 396 if (mf & 1) { 397 if (in->linear) 398 lin[mid / 32] |= 1 << (mid % 32); 399 if (mv & 1) 400 map[mid] = oid; 401 else 402 if (c == 3) 403 map[mid] |= 1; 404 ++mid; 405 } 406 407 oid += mv & 1; 408 mf >>= 1; 409 mv >>= 1; 410 } 411 412 return mid; 413} 414 415void 416nv50_fp_linkage_validate(struct nv50_context *nv50) 417{ 418 struct nouveau_pushbuf *push = nv50->base.pushbuf; 419 struct nv50_program *vp = nv50->gmtyprog ? nv50->gmtyprog : nv50->vertprog; 420 struct nv50_program *fp = nv50->fragprog; 421 struct nv50_varying dummy; 422 int i, n, c, m; 423 uint32_t primid = 0; 424 uint32_t layerid = 0; 425 uint32_t viewportid = 0; 426 uint32_t psiz = 0x000; 427 uint32_t interp = fp->fp.interp; 428 uint32_t colors = fp->fp.colors; 429 uint32_t clpd_nr = util_last_bit(vp->vp.clip_enable | vp->vp.cull_enable); 430 uint32_t lin[4]; 431 uint8_t map[64]; 432 uint8_t so_map[64]; 433 434 if (!(nv50->dirty_3d & (NV50_NEW_3D_VERTPROG | 435 NV50_NEW_3D_FRAGPROG | 436 NV50_NEW_3D_GMTYPROG))) { 437 uint8_t bfc, ffc; 438 ffc = (nv50->state.semantic_color & NV50_3D_SEMANTIC_COLOR_FFC0_ID__MASK); 439 bfc = (nv50->state.semantic_color & NV50_3D_SEMANTIC_COLOR_BFC0_ID__MASK) 440 >> 8; 441 if (nv50->rast->pipe.light_twoside == ((ffc == bfc) ? 0 : 1)) 442 return; 443 } 444 445 memset(lin, 0x00, sizeof(lin)); 446 447 /* XXX: in buggy-endian mode, is the first element of map (u32)0x000000xx 448 * or is it the first byte ? 449 */ 450 memset(map, nv50->gmtyprog ? 0x80 : 0x40, sizeof(map)); 451 452 dummy.mask = 0xf; /* map all components of HPOS */ 453 dummy.linear = 0; 454 m = nv50_vec4_map(map, 0, lin, &dummy, &vp->out[0]); 455 456 for (c = 0; c < clpd_nr; ++c) 457 map[m++] = vp->vp.clpd[c / 4] + (c % 4); 458 459 colors |= m << 8; /* adjust BFC0 id */ 460 461 dummy.mask = 0x0; 462 463 /* if light_twoside is active, FFC0_ID == BFC0_ID is invalid */ 464 if (nv50->rast->pipe.light_twoside) { 465 for (i = 0; i < 2; ++i) { 466 n = vp->vp.bfc[i]; 467 if (fp->vp.bfc[i] >= fp->in_nr) 468 continue; 469 m = nv50_vec4_map(map, m, lin, &fp->in[fp->vp.bfc[i]], 470 (n < vp->out_nr) ? &vp->out[n] : &dummy); 471 } 472 } 473 colors += m - 4; /* adjust FFC0 id */ 474 interp |= m << 8; /* set map id where 'normal' FP inputs start */ 475 476 for (i = 0; i < fp->in_nr; ++i) { 477 for (n = 0; n < vp->out_nr; ++n) 478 if (vp->out[n].sn == fp->in[i].sn && 479 vp->out[n].si == fp->in[i].si) 480 break; 481 switch (fp->in[i].sn) { 482 case TGSI_SEMANTIC_PRIMID: 483 primid = m; 484 break; 485 case TGSI_SEMANTIC_LAYER: 486 layerid = m; 487 break; 488 case TGSI_SEMANTIC_VIEWPORT_INDEX: 489 viewportid = m; 490 break; 491 } 492 m = nv50_vec4_map(map, m, lin, 493 &fp->in[i], (n < vp->out_nr) ? &vp->out[n] : &dummy); 494 } 495 496 if (vp->gp.has_layer && !layerid) { 497 layerid = m; 498 map[m++] = vp->gp.layerid; 499 } 500 501 if (vp->gp.has_viewport && !viewportid) { 502 viewportid = m; 503 map[m++] = vp->gp.viewportid; 504 } 505 506 if (nv50->rast->pipe.point_size_per_vertex) { 507 psiz = (m << 4) | 1; 508 map[m++] = vp->vp.psiz; 509 } 510 511 if (nv50->rast->pipe.clamp_vertex_color) 512 colors |= NV50_3D_SEMANTIC_COLOR_CLMP_EN; 513 514 if (unlikely(vp->so)) { 515 /* Slot i in STRMOUT_MAP specifies the offset where slot i in RESULT_MAP 516 * gets written. 517 * 518 * TODO: 519 * Inverting vp->so->map (output -> offset) would probably speed this up. 520 */ 521 memset(so_map, 0, sizeof(so_map)); 522 for (i = 0; i < vp->so->map_size; ++i) { 523 if (vp->so->map[i] == 0xff) 524 continue; 525 for (c = 0; c < m; ++c) 526 if (map[c] == vp->so->map[i] && !so_map[c]) 527 break; 528 if (c == m) { 529 c = m; 530 map[m++] = vp->so->map[i]; 531 } 532 so_map[c] = 0x80 | i; 533 } 534 for (c = m; c & 3; ++c) 535 so_map[c] = 0; 536 } 537 538 n = (m + 3) / 4; 539 assert(m <= 64); 540 541 if (unlikely(nv50->gmtyprog)) { 542 BEGIN_NV04(push, NV50_3D(GP_RESULT_MAP_SIZE), 1); 543 PUSH_DATA (push, m); 544 BEGIN_NV04(push, NV50_3D(GP_RESULT_MAP(0)), n); 545 PUSH_DATAp(push, map, n); 546 } else { 547 BEGIN_NV04(push, NV50_3D(VP_GP_BUILTIN_ATTR_EN), 1); 548 PUSH_DATA (push, vp->vp.attrs[2] | fp->vp.attrs[2]); 549 550 BEGIN_NV04(push, NV50_3D(SEMANTIC_PRIM_ID), 1); 551 PUSH_DATA (push, primid); 552 553 assert(m > 0); 554 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP_SIZE), 1); 555 PUSH_DATA (push, m); 556 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP(0)), n); 557 PUSH_DATAp(push, map, n); 558 } 559 560 BEGIN_NV04(push, NV50_3D(GP_VIEWPORT_ID_ENABLE), 5); 561 PUSH_DATA (push, vp->gp.has_viewport); 562 PUSH_DATA (push, colors); 563 PUSH_DATA (push, (clpd_nr << 8) | 4); 564 PUSH_DATA (push, layerid); 565 PUSH_DATA (push, psiz); 566 567 BEGIN_NV04(push, NV50_3D(SEMANTIC_VIEWPORT), 1); 568 PUSH_DATA (push, viewportid); 569 570 BEGIN_NV04(push, NV50_3D(LAYER), 1); 571 PUSH_DATA (push, vp->gp.has_layer << 16); 572 573 BEGIN_NV04(push, NV50_3D(FP_INTERPOLANT_CTRL), 1); 574 PUSH_DATA (push, interp); 575 576 nv50->state.interpolant_ctrl = interp; 577 578 nv50->state.semantic_color = colors; 579 nv50->state.semantic_psize = psiz; 580 581 BEGIN_NV04(push, NV50_3D(NOPERSPECTIVE_BITMAP(0)), 4); 582 PUSH_DATAp(push, lin, 4); 583 584 BEGIN_NV04(push, NV50_3D(GP_ENABLE), 1); 585 PUSH_DATA (push, nv50->gmtyprog ? 1 : 0); 586 587 if (vp->so) { 588 BEGIN_NV04(push, NV50_3D(STRMOUT_MAP(0)), n); 589 PUSH_DATAp(push, so_map, n); 590 } 591} 592 593static int 594nv50_vp_gp_mapping(uint8_t *map, int m, 595 struct nv50_program *vp, struct nv50_program *gp) 596{ 597 int i, j, c; 598 599 for (i = 0; i < gp->in_nr; ++i) { 600 uint8_t oid = 0, mv = 0, mg = gp->in[i].mask; 601 602 for (j = 0; j < vp->out_nr; ++j) { 603 if (vp->out[j].sn == gp->in[i].sn && 604 vp->out[j].si == gp->in[i].si) { 605 mv = vp->out[j].mask; 606 oid = vp->out[j].hw; 607 break; 608 } 609 } 610 611 for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) { 612 if (mg & mv & 1) 613 map[m++] = oid; 614 else 615 if (mg & 1) 616 map[m++] = (c == 3) ? 0x41 : 0x40; 617 oid += mv & 1; 618 } 619 } 620 if (!m) 621 map[m++] = 0; 622 return m; 623} 624 625void 626nv50_gp_linkage_validate(struct nv50_context *nv50) 627{ 628 struct nouveau_pushbuf *push = nv50->base.pushbuf; 629 struct nv50_program *vp = nv50->vertprog; 630 struct nv50_program *gp = nv50->gmtyprog; 631 int m = 0; 632 int n; 633 uint8_t map[64]; 634 635 if (!gp) 636 return; 637 memset(map, 0, sizeof(map)); 638 639 m = nv50_vp_gp_mapping(map, m, vp, gp); 640 641 n = (m + 3) / 4; 642 643 BEGIN_NV04(push, NV50_3D(VP_GP_BUILTIN_ATTR_EN), 1); 644 PUSH_DATA (push, vp->vp.attrs[2] | gp->vp.attrs[2]); 645 646 assert(m > 0); 647 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP_SIZE), 1); 648 PUSH_DATA (push, m); 649 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP(0)), n); 650 PUSH_DATAp(push, map, n); 651} 652 653void 654nv50_stream_output_validate(struct nv50_context *nv50) 655{ 656 struct nouveau_pushbuf *push = nv50->base.pushbuf; 657 struct nv50_stream_output_state *so; 658 uint32_t ctrl; 659 unsigned i; 660 unsigned prims = ~0; 661 662 so = nv50->gmtyprog ? nv50->gmtyprog->so : nv50->vertprog->so; 663 664 BEGIN_NV04(push, NV50_3D(STRMOUT_ENABLE), 1); 665 PUSH_DATA (push, 0); 666 if (!so || !nv50->num_so_targets) { 667 if (nv50->screen->base.class_3d < NVA0_3D_CLASS) { 668 BEGIN_NV04(push, NV50_3D(STRMOUT_PRIMITIVE_LIMIT), 1); 669 PUSH_DATA (push, 0); 670 } 671 BEGIN_NV04(push, NV50_3D(STRMOUT_PARAMS_LATCH), 1); 672 PUSH_DATA (push, 1); 673 return; 674 } 675 676 /* previous TFB needs to complete */ 677 if (nv50->screen->base.class_3d < NVA0_3D_CLASS) { 678 BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1); 679 PUSH_DATA (push, 0); 680 } 681 682 ctrl = so->ctrl; 683 if (nv50->screen->base.class_3d >= NVA0_3D_CLASS) 684 ctrl |= NVA0_3D_STRMOUT_BUFFERS_CTRL_LIMIT_MODE_OFFSET; 685 686 BEGIN_NV04(push, NV50_3D(STRMOUT_BUFFERS_CTRL), 1); 687 PUSH_DATA (push, ctrl); 688 689 for (i = 0; i < nv50->num_so_targets; ++i) { 690 struct nv50_so_target *targ = nv50_so_target(nv50->so_target[i]); 691 struct nv04_resource *buf = nv04_resource(targ->pipe.buffer); 692 693 const unsigned n = nv50->screen->base.class_3d >= NVA0_3D_CLASS ? 4 : 3; 694 695 if (n == 4 && !targ->clean) 696 nv84_hw_query_fifo_wait(push, nv50_query(targ->pq)); 697 BEGIN_NV04(push, NV50_3D(STRMOUT_ADDRESS_HIGH(i)), n); 698 PUSH_DATAh(push, buf->address + targ->pipe.buffer_offset); 699 PUSH_DATA (push, buf->address + targ->pipe.buffer_offset); 700 PUSH_DATA (push, so->num_attribs[i]); 701 if (n == 4) { 702 PUSH_DATA(push, targ->pipe.buffer_size); 703 if (!targ->clean) { 704 assert(targ->pq); 705 nv50_hw_query_pushbuf_submit(push, NVA0_3D_STRMOUT_OFFSET(i), 706 nv50_query(targ->pq), 0x4); 707 } else { 708 BEGIN_NV04(push, NVA0_3D(STRMOUT_OFFSET(i)), 1); 709 PUSH_DATA(push, 0); 710 targ->clean = false; 711 } 712 } else { 713 const unsigned limit = targ->pipe.buffer_size / 714 (so->stride[i] * nv50->state.prim_size); 715 prims = MIN2(prims, limit); 716 } 717 targ->stride = so->stride[i]; 718 BCTX_REFN(nv50->bufctx_3d, 3D_SO, buf, WR); 719 } 720 if (prims != ~0) { 721 BEGIN_NV04(push, NV50_3D(STRMOUT_PRIMITIVE_LIMIT), 1); 722 PUSH_DATA (push, prims); 723 } 724 BEGIN_NV04(push, NV50_3D(STRMOUT_PARAMS_LATCH), 1); 725 PUSH_DATA (push, 1); 726 BEGIN_NV04(push, NV50_3D(STRMOUT_ENABLE), 1); 727 PUSH_DATA (push, 1); 728} 729