brw_draw_upload.c revision b922999a404322a1034c4f96c21acae860483bb0
1/************************************************************************** 2 * 3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the 8 * "Software"), to deal in the Software without restriction, including 9 * without limitation the rights to use, copy, modify, merge, publish, 10 * distribute, sub license, and/or sell copies of the Software, and to 11 * permit persons to whom the Software is furnished to do so, subject to 12 * the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the 15 * next paragraph) shall be included in all copies or substantial portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28#undef NDEBUG 29 30#include "main/glheader.h" 31#include "main/bufferobj.h" 32#include "main/context.h" 33#include "main/enums.h" 34#include "main/macros.h" 35 36#include "brw_draw.h" 37#include "brw_defines.h" 38#include "brw_context.h" 39#include "brw_state.h" 40 41#include "intel_batchbuffer.h" 42#include "intel_buffer_objects.h" 43 44static GLuint double_types[5] = { 45 0, 46 BRW_SURFACEFORMAT_R64_FLOAT, 47 BRW_SURFACEFORMAT_R64G64_FLOAT, 48 BRW_SURFACEFORMAT_R64G64B64_FLOAT, 49 BRW_SURFACEFORMAT_R64G64B64A64_FLOAT 50}; 51 52static GLuint float_types[5] = { 53 0, 54 BRW_SURFACEFORMAT_R32_FLOAT, 55 BRW_SURFACEFORMAT_R32G32_FLOAT, 56 BRW_SURFACEFORMAT_R32G32B32_FLOAT, 57 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT 58}; 59 60static GLuint half_float_types[5] = { 61 0, 62 BRW_SURFACEFORMAT_R16_FLOAT, 63 BRW_SURFACEFORMAT_R16G16_FLOAT, 64 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT, 65 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT 66}; 67 68static GLuint uint_types_direct[5] = { 69 0, 70 BRW_SURFACEFORMAT_R32_UINT, 71 BRW_SURFACEFORMAT_R32G32_UINT, 72 BRW_SURFACEFORMAT_R32G32B32_UINT, 73 BRW_SURFACEFORMAT_R32G32B32A32_UINT 74}; 75 76static GLuint uint_types_norm[5] = { 77 0, 78 BRW_SURFACEFORMAT_R32_UNORM, 79 BRW_SURFACEFORMAT_R32G32_UNORM, 80 BRW_SURFACEFORMAT_R32G32B32_UNORM, 81 BRW_SURFACEFORMAT_R32G32B32A32_UNORM 82}; 83 84static GLuint uint_types_scale[5] = { 85 0, 86 BRW_SURFACEFORMAT_R32_USCALED, 87 BRW_SURFACEFORMAT_R32G32_USCALED, 88 BRW_SURFACEFORMAT_R32G32B32_USCALED, 89 BRW_SURFACEFORMAT_R32G32B32A32_USCALED 90}; 91 92static GLuint int_types_direct[5] = { 93 0, 94 BRW_SURFACEFORMAT_R32_SINT, 95 BRW_SURFACEFORMAT_R32G32_SINT, 96 BRW_SURFACEFORMAT_R32G32B32_SINT, 97 BRW_SURFACEFORMAT_R32G32B32A32_SINT 98}; 99 100static GLuint int_types_norm[5] = { 101 0, 102 BRW_SURFACEFORMAT_R32_SNORM, 103 BRW_SURFACEFORMAT_R32G32_SNORM, 104 BRW_SURFACEFORMAT_R32G32B32_SNORM, 105 BRW_SURFACEFORMAT_R32G32B32A32_SNORM 106}; 107 108static GLuint int_types_scale[5] = { 109 0, 110 BRW_SURFACEFORMAT_R32_SSCALED, 111 BRW_SURFACEFORMAT_R32G32_SSCALED, 112 BRW_SURFACEFORMAT_R32G32B32_SSCALED, 113 BRW_SURFACEFORMAT_R32G32B32A32_SSCALED 114}; 115 116static GLuint ushort_types_direct[5] = { 117 0, 118 BRW_SURFACEFORMAT_R16_UINT, 119 BRW_SURFACEFORMAT_R16G16_UINT, 120 BRW_SURFACEFORMAT_R16G16B16A16_UINT, 121 BRW_SURFACEFORMAT_R16G16B16A16_UINT 122}; 123 124static GLuint ushort_types_norm[5] = { 125 0, 126 BRW_SURFACEFORMAT_R16_UNORM, 127 BRW_SURFACEFORMAT_R16G16_UNORM, 128 BRW_SURFACEFORMAT_R16G16B16_UNORM, 129 BRW_SURFACEFORMAT_R16G16B16A16_UNORM 130}; 131 132static GLuint ushort_types_scale[5] = { 133 0, 134 BRW_SURFACEFORMAT_R16_USCALED, 135 BRW_SURFACEFORMAT_R16G16_USCALED, 136 BRW_SURFACEFORMAT_R16G16B16_USCALED, 137 BRW_SURFACEFORMAT_R16G16B16A16_USCALED 138}; 139 140static GLuint short_types_direct[5] = { 141 0, 142 BRW_SURFACEFORMAT_R16_SINT, 143 BRW_SURFACEFORMAT_R16G16_SINT, 144 BRW_SURFACEFORMAT_R16G16B16A16_SINT, 145 BRW_SURFACEFORMAT_R16G16B16A16_SINT 146}; 147 148static GLuint short_types_norm[5] = { 149 0, 150 BRW_SURFACEFORMAT_R16_SNORM, 151 BRW_SURFACEFORMAT_R16G16_SNORM, 152 BRW_SURFACEFORMAT_R16G16B16_SNORM, 153 BRW_SURFACEFORMAT_R16G16B16A16_SNORM 154}; 155 156static GLuint short_types_scale[5] = { 157 0, 158 BRW_SURFACEFORMAT_R16_SSCALED, 159 BRW_SURFACEFORMAT_R16G16_SSCALED, 160 BRW_SURFACEFORMAT_R16G16B16_SSCALED, 161 BRW_SURFACEFORMAT_R16G16B16A16_SSCALED 162}; 163 164static GLuint ubyte_types_direct[5] = { 165 0, 166 BRW_SURFACEFORMAT_R8_UINT, 167 BRW_SURFACEFORMAT_R8G8_UINT, 168 BRW_SURFACEFORMAT_R8G8B8A8_UINT, 169 BRW_SURFACEFORMAT_R8G8B8A8_UINT 170}; 171 172static GLuint ubyte_types_norm[5] = { 173 0, 174 BRW_SURFACEFORMAT_R8_UNORM, 175 BRW_SURFACEFORMAT_R8G8_UNORM, 176 BRW_SURFACEFORMAT_R8G8B8_UNORM, 177 BRW_SURFACEFORMAT_R8G8B8A8_UNORM 178}; 179 180static GLuint ubyte_types_scale[5] = { 181 0, 182 BRW_SURFACEFORMAT_R8_USCALED, 183 BRW_SURFACEFORMAT_R8G8_USCALED, 184 BRW_SURFACEFORMAT_R8G8B8_USCALED, 185 BRW_SURFACEFORMAT_R8G8B8A8_USCALED 186}; 187 188static GLuint byte_types_direct[5] = { 189 0, 190 BRW_SURFACEFORMAT_R8_SINT, 191 BRW_SURFACEFORMAT_R8G8_SINT, 192 BRW_SURFACEFORMAT_R8G8B8A8_SINT, 193 BRW_SURFACEFORMAT_R8G8B8A8_SINT 194}; 195 196static GLuint byte_types_norm[5] = { 197 0, 198 BRW_SURFACEFORMAT_R8_SNORM, 199 BRW_SURFACEFORMAT_R8G8_SNORM, 200 BRW_SURFACEFORMAT_R8G8B8_SNORM, 201 BRW_SURFACEFORMAT_R8G8B8A8_SNORM 202}; 203 204static GLuint byte_types_scale[5] = { 205 0, 206 BRW_SURFACEFORMAT_R8_SSCALED, 207 BRW_SURFACEFORMAT_R8G8_SSCALED, 208 BRW_SURFACEFORMAT_R8G8B8_SSCALED, 209 BRW_SURFACEFORMAT_R8G8B8A8_SSCALED 210}; 211 212 213/** 214 * Given vertex array type/size/format/normalized info, return 215 * the appopriate hardware surface type. 216 * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays. 217 */ 218static GLuint get_surface_type( GLenum type, GLuint size, 219 GLenum format, bool normalized, bool integer ) 220{ 221 if (unlikely(INTEL_DEBUG & DEBUG_VERTS)) 222 printf("type %s size %d normalized %d\n", 223 _mesa_lookup_enum_by_nr(type), size, normalized); 224 225 if (integer) { 226 assert(format == GL_RGBA); /* sanity check */ 227 switch (type) { 228 case GL_INT: return int_types_direct[size]; 229 case GL_SHORT: return short_types_direct[size]; 230 case GL_BYTE: return byte_types_direct[size]; 231 case GL_UNSIGNED_INT: return uint_types_direct[size]; 232 case GL_UNSIGNED_SHORT: return ushort_types_direct[size]; 233 case GL_UNSIGNED_BYTE: return ubyte_types_direct[size]; 234 default: assert(0); return 0; 235 } 236 } else if (normalized) { 237 switch (type) { 238 case GL_DOUBLE: return double_types[size]; 239 case GL_FLOAT: return float_types[size]; 240 case GL_HALF_FLOAT: return half_float_types[size]; 241 case GL_INT: return int_types_norm[size]; 242 case GL_SHORT: return short_types_norm[size]; 243 case GL_BYTE: return byte_types_norm[size]; 244 case GL_UNSIGNED_INT: return uint_types_norm[size]; 245 case GL_UNSIGNED_SHORT: return ushort_types_norm[size]; 246 case GL_UNSIGNED_BYTE: 247 if (format == GL_BGRA) { 248 /* See GL_EXT_vertex_array_bgra */ 249 assert(size == 4); 250 return BRW_SURFACEFORMAT_B8G8R8A8_UNORM; 251 } 252 else { 253 return ubyte_types_norm[size]; 254 } 255 default: assert(0); return 0; 256 } 257 } 258 else { 259 assert(format == GL_RGBA); /* sanity check */ 260 switch (type) { 261 case GL_DOUBLE: return double_types[size]; 262 case GL_FLOAT: return float_types[size]; 263 case GL_HALF_FLOAT: return half_float_types[size]; 264 case GL_INT: return int_types_scale[size]; 265 case GL_SHORT: return short_types_scale[size]; 266 case GL_BYTE: return byte_types_scale[size]; 267 case GL_UNSIGNED_INT: return uint_types_scale[size]; 268 case GL_UNSIGNED_SHORT: return ushort_types_scale[size]; 269 case GL_UNSIGNED_BYTE: return ubyte_types_scale[size]; 270 /* This produces GL_FIXED inputs as values between INT32_MIN and 271 * INT32_MAX, which will be scaled down by 1/65536 by the VS. 272 */ 273 case GL_FIXED: return int_types_scale[size]; 274 default: assert(0); return 0; 275 } 276 } 277} 278 279 280static GLuint get_size( GLenum type ) 281{ 282 switch (type) { 283 case GL_DOUBLE: return sizeof(GLdouble); 284 case GL_FLOAT: return sizeof(GLfloat); 285 case GL_HALF_FLOAT: return sizeof(GLhalfARB); 286 case GL_INT: return sizeof(GLint); 287 case GL_SHORT: return sizeof(GLshort); 288 case GL_BYTE: return sizeof(GLbyte); 289 case GL_UNSIGNED_INT: return sizeof(GLuint); 290 case GL_UNSIGNED_SHORT: return sizeof(GLushort); 291 case GL_UNSIGNED_BYTE: return sizeof(GLubyte); 292 case GL_FIXED: return sizeof(GLuint); 293 default: assert(0); return 0; 294 } 295} 296 297static GLuint get_index_type(GLenum type) 298{ 299 switch (type) { 300 case GL_UNSIGNED_BYTE: return BRW_INDEX_BYTE; 301 case GL_UNSIGNED_SHORT: return BRW_INDEX_WORD; 302 case GL_UNSIGNED_INT: return BRW_INDEX_DWORD; 303 default: assert(0); return 0; 304 } 305} 306 307static void 308copy_array_to_vbo_array(struct brw_context *brw, 309 struct brw_vertex_element *element, 310 int min, int max, 311 struct brw_vertex_buffer *buffer, 312 GLuint dst_stride) 313{ 314 if (min == -1) { 315 /* If we don't have computed min/max bounds, then this must be a use of 316 * the current attribute, which has a 0 stride. Otherwise, we wouldn't 317 * know what data to upload. 318 */ 319 assert(element->glarray->StrideB == 0); 320 321 intel_upload_data(&brw->intel, element->glarray->Ptr, 322 element->element_size, 323 element->element_size, 324 &buffer->bo, &buffer->offset); 325 326 buffer->stride = 0; 327 return; 328 } 329 330 int src_stride = element->glarray->StrideB; 331 const unsigned char *src = element->glarray->Ptr + min * src_stride; 332 int count = max - min + 1; 333 GLuint size = count * dst_stride; 334 335 if (dst_stride == src_stride) { 336 intel_upload_data(&brw->intel, src, size, dst_stride, 337 &buffer->bo, &buffer->offset); 338 } else { 339 char * const map = intel_upload_map(&brw->intel, size, dst_stride); 340 char *dst = map; 341 342 while (count--) { 343 memcpy(dst, src, dst_stride); 344 src += src_stride; 345 dst += dst_stride; 346 } 347 intel_upload_unmap(&brw->intel, map, size, dst_stride, 348 &buffer->bo, &buffer->offset); 349 } 350 buffer->stride = dst_stride; 351} 352 353static void brw_prepare_vertices(struct brw_context *brw) 354{ 355 struct gl_context *ctx = &brw->intel.ctx; 356 struct intel_context *intel = intel_context(ctx); 357 /* CACHE_NEW_VS_PROG */ 358 GLbitfield64 vs_inputs = brw->vs.prog_data->inputs_read; 359 const unsigned char *ptr = NULL; 360 GLuint interleaved = 0; 361 unsigned int min_index = brw->vb.min_index; 362 unsigned int max_index = brw->vb.max_index; 363 int delta, i, j; 364 365 struct brw_vertex_element *upload[VERT_ATTRIB_MAX]; 366 GLuint nr_uploads = 0; 367 368 /* _NEW_POLYGON 369 * 370 * On gen6+, edge flags don't end up in the VUE (either in or out of the 371 * VS). Instead, they're uploaded as the last vertex element, and the data 372 * is passed sideband through the fixed function units. So, we need to 373 * prepare the vertex buffer for it, but it's not present in inputs_read. 374 */ 375 if (intel->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL || 376 ctx->Polygon.BackMode != GL_FILL)) { 377 vs_inputs |= VERT_BIT_EDGEFLAG; 378 } 379 380 /* First build an array of pointers to ve's in vb.inputs_read 381 */ 382 if (0) 383 printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); 384 385 /* Accumulate the list of enabled arrays. */ 386 brw->vb.nr_enabled = 0; 387 while (vs_inputs) { 388 GLuint i = ffsll(vs_inputs) - 1; 389 struct brw_vertex_element *input = &brw->vb.inputs[i]; 390 391 vs_inputs &= ~BITFIELD64_BIT(i); 392 if (input->glarray->Size && get_size(input->glarray->Type)) 393 brw->vb.enabled[brw->vb.nr_enabled++] = input; 394 } 395 396 if (brw->vb.nr_enabled == 0) 397 return; 398 399 if (brw->vb.nr_buffers) 400 goto prepare; 401 402 for (i = j = 0; i < brw->vb.nr_enabled; i++) { 403 struct brw_vertex_element *input = brw->vb.enabled[i]; 404 const struct gl_client_array *glarray = input->glarray; 405 int type_size = get_size(glarray->Type); 406 407 input->element_size = type_size * glarray->Size; 408 409 if (_mesa_is_bufferobj(glarray->BufferObj)) { 410 struct intel_buffer_object *intel_buffer = 411 intel_buffer_object(glarray->BufferObj); 412 int k; 413 414 for (k = 0; k < i; k++) { 415 const struct gl_client_array *other = brw->vb.enabled[k]->glarray; 416 if (glarray->BufferObj == other->BufferObj && 417 glarray->StrideB == other->StrideB && 418 glarray->InstanceDivisor == other->InstanceDivisor && 419 (uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB) 420 { 421 input->buffer = brw->vb.enabled[k]->buffer; 422 input->offset = glarray->Ptr - other->Ptr; 423 break; 424 } 425 } 426 if (k == i) { 427 struct brw_vertex_buffer *buffer = &brw->vb.buffers[j]; 428 429 /* Named buffer object: Just reference its contents directly. */ 430 buffer->bo = intel_bufferobj_source(intel, 431 intel_buffer, type_size, 432 &buffer->offset); 433 drm_intel_bo_reference(buffer->bo); 434 buffer->offset += (uintptr_t)glarray->Ptr; 435 buffer->stride = glarray->StrideB; 436 buffer->step_rate = glarray->InstanceDivisor; 437 438 input->buffer = j++; 439 input->offset = 0; 440 } 441 442 /* This is a common place to reach if the user mistakenly supplies 443 * a pointer in place of a VBO offset. If we just let it go through, 444 * we may end up dereferencing a pointer beyond the bounds of the 445 * GTT. We would hope that the VBO's max_index would save us, but 446 * Mesa appears to hand us min/max values not clipped to the 447 * array object's _MaxElement, and _MaxElement frequently appears 448 * to be wrong anyway. 449 * 450 * The VBO spec allows application termination in this case, and it's 451 * probably a service to the poor programmer to do so rather than 452 * trying to just not render. 453 */ 454 assert(input->offset < brw->vb.buffers[input->buffer].bo->size); 455 } else { 456 /* Queue the buffer object up to be uploaded in the next pass, 457 * when we've decided if we're doing interleaved or not. 458 */ 459 if (nr_uploads == 0) { 460 interleaved = glarray->StrideB; 461 ptr = glarray->Ptr; 462 } 463 else if (interleaved != glarray->StrideB || 464 (uintptr_t)(glarray->Ptr - ptr) > interleaved) 465 { 466 interleaved = 0; 467 } 468 else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1)) 469 { 470 /* enforce natural alignment (for doubles) */ 471 interleaved = 0; 472 } 473 474 upload[nr_uploads++] = input; 475 } 476 } 477 478 /* If we need to upload all the arrays, then we can trim those arrays to 479 * only the used elements [min_index, max_index] so long as we adjust all 480 * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias. 481 */ 482 brw->vb.start_vertex_bias = 0; 483 delta = min_index; 484 if (nr_uploads == brw->vb.nr_enabled) { 485 brw->vb.start_vertex_bias = -delta; 486 delta = 0; 487 } 488 if (delta && !brw->intel.intelScreen->relaxed_relocations) 489 min_index = delta = 0; 490 491 /* Handle any arrays to be uploaded. */ 492 if (nr_uploads > 1) { 493 if (interleaved) { 494 struct brw_vertex_buffer *buffer = &brw->vb.buffers[j]; 495 /* All uploads are interleaved, so upload the arrays together as 496 * interleaved. First, upload the contents and set up upload[0]. 497 */ 498 copy_array_to_vbo_array(brw, upload[0], min_index, max_index, 499 buffer, interleaved); 500 buffer->offset -= delta * interleaved; 501 502 for (i = 0; i < nr_uploads; i++) { 503 /* Then, just point upload[i] at upload[0]'s buffer. */ 504 upload[i]->offset = 505 ((const unsigned char *)upload[i]->glarray->Ptr - ptr); 506 upload[i]->buffer = j; 507 } 508 j++; 509 510 nr_uploads = 0; 511 } 512 } 513 /* Upload non-interleaved arrays */ 514 for (i = 0; i < nr_uploads; i++) { 515 struct brw_vertex_buffer *buffer = &brw->vb.buffers[j]; 516 if (upload[i]->glarray->InstanceDivisor == 0) { 517 copy_array_to_vbo_array(brw, upload[i], min_index, max_index, 518 buffer, upload[i]->element_size); 519 } else { 520 /* This is an instanced attribute, since its InstanceDivisor 521 * is not zero. Therefore, its data will be stepped after the 522 * instanced draw has been run InstanceDivisor times. 523 */ 524 uint32_t instanced_attr_max_index = 525 (brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor; 526 copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index, 527 buffer, upload[i]->element_size); 528 } 529 buffer->offset -= delta * buffer->stride; 530 buffer->step_rate = upload[i]->glarray->InstanceDivisor; 531 upload[i]->buffer = j++; 532 upload[i]->offset = 0; 533 } 534 535 /* can we simply extend the current vb? */ 536 if (j == brw->vb.nr_current_buffers) { 537 int delta = 0; 538 for (i = 0; i < j; i++) { 539 int d; 540 541 if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle || 542 brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride || 543 brw->vb.current_buffers[i].step_rate != brw->vb.buffers[i].step_rate) 544 break; 545 546 d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset; 547 if (d < 0) 548 break; 549 if (i == 0) 550 delta = d / brw->vb.current_buffers[i].stride; 551 if (delta * brw->vb.current_buffers[i].stride != d) 552 break; 553 } 554 555 if (i == j) { 556 brw->vb.start_vertex_bias += delta; 557 while (--j >= 0) 558 drm_intel_bo_unreference(brw->vb.buffers[j].bo); 559 j = 0; 560 } 561 } 562 563 brw->vb.nr_buffers = j; 564 565prepare: 566 brw_prepare_query_begin(brw); 567} 568 569static void brw_emit_vertices(struct brw_context *brw) 570{ 571 struct gl_context *ctx = &brw->intel.ctx; 572 struct intel_context *intel = intel_context(ctx); 573 GLuint i, nr_elements; 574 575 brw_prepare_vertices(brw); 576 577 brw_emit_query_begin(brw); 578 579 /* If the VS doesn't read any inputs (calculating vertex position from 580 * a state variable for some reason, for example), emit a single pad 581 * VERTEX_ELEMENT struct and bail. 582 * 583 * The stale VB state stays in place, but they don't do anything unless 584 * a VE loads from them. 585 */ 586 if (brw->vb.nr_enabled == 0) { 587 BEGIN_BATCH(3); 588 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | 1); 589 if (intel->gen >= 6) { 590 OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) | 591 GEN6_VE0_VALID | 592 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) | 593 (0 << BRW_VE0_SRC_OFFSET_SHIFT)); 594 } else { 595 OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) | 596 BRW_VE0_VALID | 597 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) | 598 (0 << BRW_VE0_SRC_OFFSET_SHIFT)); 599 } 600 OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) | 601 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) | 602 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) | 603 (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT)); 604 CACHED_BATCH(); 605 return; 606 } 607 608 /* Now emit VB and VEP state packets. 609 */ 610 611 if (brw->vb.nr_buffers) { 612 if (intel->gen >= 6) { 613 assert(brw->vb.nr_buffers <= 33); 614 } else { 615 assert(brw->vb.nr_buffers <= 17); 616 } 617 618 BEGIN_BATCH(1 + 4*brw->vb.nr_buffers); 619 OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (4*brw->vb.nr_buffers - 1)); 620 for (i = 0; i < brw->vb.nr_buffers; i++) { 621 struct brw_vertex_buffer *buffer = &brw->vb.buffers[i]; 622 uint32_t dw0; 623 624 if (intel->gen >= 6) { 625 dw0 = buffer->step_rate 626 ? GEN6_VB0_ACCESS_INSTANCEDATA 627 : GEN6_VB0_ACCESS_VERTEXDATA; 628 dw0 |= i << GEN6_VB0_INDEX_SHIFT; 629 } else { 630 dw0 = buffer->step_rate 631 ? BRW_VB0_ACCESS_INSTANCEDATA 632 : BRW_VB0_ACCESS_VERTEXDATA; 633 dw0 |= i << BRW_VB0_INDEX_SHIFT; 634 } 635 636 if (intel->gen >= 7) 637 dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE; 638 639 OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT)); 640 OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset); 641 if (intel->gen >= 5) { 642 OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1); 643 } else 644 OUT_BATCH(0); 645 OUT_BATCH(buffer->step_rate); 646 647 brw->vb.current_buffers[i].handle = buffer->bo->handle; 648 brw->vb.current_buffers[i].offset = buffer->offset; 649 brw->vb.current_buffers[i].stride = buffer->stride; 650 brw->vb.current_buffers[i].step_rate = buffer->step_rate; 651 } 652 brw->vb.nr_current_buffers = i; 653 ADVANCE_BATCH(); 654 } 655 656 nr_elements = brw->vb.nr_enabled + brw->vs.prog_data->uses_vertexid; 657 658 /* The hardware allows one more VERTEX_ELEMENTS than VERTEX_BUFFERS, presumably 659 * for VertexID/InstanceID. 660 */ 661 if (intel->gen >= 6) { 662 assert(nr_elements <= 34); 663 } else { 664 assert(nr_elements <= 18); 665 } 666 667 struct brw_vertex_element *gen6_edgeflag_input = NULL; 668 669 BEGIN_BATCH(1 + nr_elements * 2); 670 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (2 * nr_elements - 1)); 671 for (i = 0; i < brw->vb.nr_enabled; i++) { 672 struct brw_vertex_element *input = brw->vb.enabled[i]; 673 uint32_t format = get_surface_type(input->glarray->Type, 674 input->glarray->Size, 675 input->glarray->Format, 676 input->glarray->Normalized, 677 input->glarray->Integer); 678 uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC; 679 uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC; 680 uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC; 681 uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC; 682 683 /* The gen4 driver expects edgeflag to come in as a float, and passes 684 * that float on to the tests in the clipper. Mesa's current vertex 685 * attribute value for EdgeFlag is stored as a float, which works out. 686 * glEdgeFlagPointer, on the other hand, gives us an unnormalized 687 * integer ubyte. Just rewrite that to convert to a float. 688 */ 689 if (input->attrib == VERT_ATTRIB_EDGEFLAG) { 690 /* Gen6+ passes edgeflag as sideband along with the vertex, instead 691 * of in the VUE. We have to upload it sideband as the last vertex 692 * element according to the B-Spec. 693 */ 694 if (intel->gen >= 6) { 695 gen6_edgeflag_input = input; 696 continue; 697 } 698 699 if (format == BRW_SURFACEFORMAT_R8_UINT) 700 format = BRW_SURFACEFORMAT_R8_SSCALED; 701 } 702 703 switch (input->glarray->Size) { 704 case 0: comp0 = BRW_VE1_COMPONENT_STORE_0; 705 case 1: comp1 = BRW_VE1_COMPONENT_STORE_0; 706 case 2: comp2 = BRW_VE1_COMPONENT_STORE_0; 707 case 3: comp3 = input->glarray->Integer ? BRW_VE1_COMPONENT_STORE_1_INT 708 : BRW_VE1_COMPONENT_STORE_1_FLT; 709 break; 710 } 711 712 if (intel->gen >= 6) { 713 OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) | 714 GEN6_VE0_VALID | 715 (format << BRW_VE0_FORMAT_SHIFT) | 716 (input->offset << BRW_VE0_SRC_OFFSET_SHIFT)); 717 } else { 718 OUT_BATCH((input->buffer << BRW_VE0_INDEX_SHIFT) | 719 BRW_VE0_VALID | 720 (format << BRW_VE0_FORMAT_SHIFT) | 721 (input->offset << BRW_VE0_SRC_OFFSET_SHIFT)); 722 } 723 724 if (intel->gen >= 5) 725 OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) | 726 (comp1 << BRW_VE1_COMPONENT_1_SHIFT) | 727 (comp2 << BRW_VE1_COMPONENT_2_SHIFT) | 728 (comp3 << BRW_VE1_COMPONENT_3_SHIFT)); 729 else 730 OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) | 731 (comp1 << BRW_VE1_COMPONENT_1_SHIFT) | 732 (comp2 << BRW_VE1_COMPONENT_2_SHIFT) | 733 (comp3 << BRW_VE1_COMPONENT_3_SHIFT) | 734 ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT)); 735 } 736 737 if (intel->gen >= 6 && gen6_edgeflag_input) { 738 uint32_t format = get_surface_type(gen6_edgeflag_input->glarray->Type, 739 gen6_edgeflag_input->glarray->Size, 740 gen6_edgeflag_input->glarray->Format, 741 gen6_edgeflag_input->glarray->Normalized, 742 gen6_edgeflag_input->glarray->Integer); 743 744 OUT_BATCH((gen6_edgeflag_input->buffer << GEN6_VE0_INDEX_SHIFT) | 745 GEN6_VE0_VALID | 746 GEN6_VE0_EDGE_FLAG_ENABLE | 747 (format << BRW_VE0_FORMAT_SHIFT) | 748 (gen6_edgeflag_input->offset << BRW_VE0_SRC_OFFSET_SHIFT)); 749 OUT_BATCH((BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT) | 750 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) | 751 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) | 752 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_3_SHIFT)); 753 } 754 755 if (brw->vs.prog_data->uses_vertexid) { 756 uint32_t dw0 = 0, dw1 = 0; 757 758 dw1 = ((BRW_VE1_COMPONENT_STORE_VID << BRW_VE1_COMPONENT_0_SHIFT) | 759 (BRW_VE1_COMPONENT_STORE_IID << BRW_VE1_COMPONENT_1_SHIFT) | 760 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) | 761 (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_3_SHIFT)); 762 763 if (intel->gen >= 6) { 764 dw0 |= GEN6_VE0_VALID; 765 } else { 766 dw0 |= BRW_VE0_VALID; 767 dw1 |= (i * 4) << BRW_VE1_DST_OFFSET_SHIFT; 768 } 769 770 /* Note that for gl_VertexID, gl_InstanceID, and gl_PrimitiveID values, 771 * the format is ignored and the value is always int. 772 */ 773 774 OUT_BATCH(dw0); 775 OUT_BATCH(dw1); 776 } 777 778 CACHED_BATCH(); 779} 780 781const struct brw_tracked_state brw_vertices = { 782 .dirty = { 783 .mesa = _NEW_POLYGON, 784 .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES, 785 .cache = CACHE_NEW_VS_PROG, 786 }, 787 .emit = brw_emit_vertices, 788}; 789 790static void brw_upload_indices(struct brw_context *brw) 791{ 792 struct gl_context *ctx = &brw->intel.ctx; 793 struct intel_context *intel = &brw->intel; 794 const struct _mesa_index_buffer *index_buffer = brw->ib.ib; 795 GLuint ib_size; 796 drm_intel_bo *bo = NULL; 797 struct gl_buffer_object *bufferobj; 798 GLuint offset; 799 GLuint ib_type_size; 800 801 if (index_buffer == NULL) 802 return; 803 804 ib_type_size = get_size(index_buffer->type); 805 ib_size = ib_type_size * index_buffer->count; 806 bufferobj = index_buffer->obj; 807 808 /* Turn into a proper VBO: 809 */ 810 if (!_mesa_is_bufferobj(bufferobj)) { 811 812 /* Get new bufferobj, offset: 813 */ 814 intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size, 815 &bo, &offset); 816 brw->ib.start_vertex_offset = offset / ib_type_size; 817 } else { 818 offset = (GLuint) (unsigned long) index_buffer->ptr; 819 820 /* If the index buffer isn't aligned to its element size, we have to 821 * rebase it into a temporary. 822 */ 823 if ((get_size(index_buffer->type) - 1) & offset) { 824 GLubyte *map = ctx->Driver.MapBufferRange(ctx, 825 offset, 826 ib_size, 827 GL_MAP_WRITE_BIT, 828 bufferobj); 829 830 intel_upload_data(&brw->intel, map, ib_size, ib_type_size, 831 &bo, &offset); 832 brw->ib.start_vertex_offset = offset / ib_type_size; 833 834 ctx->Driver.UnmapBuffer(ctx, bufferobj); 835 } else { 836 /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading 837 * the index buffer state when we're just moving the start index 838 * of our drawing. 839 */ 840 brw->ib.start_vertex_offset = offset / ib_type_size; 841 842 bo = intel_bufferobj_source(intel, 843 intel_buffer_object(bufferobj), 844 ib_type_size, 845 &offset); 846 drm_intel_bo_reference(bo); 847 848 brw->ib.start_vertex_offset += offset / ib_type_size; 849 } 850 } 851 852 if (brw->ib.bo != bo) { 853 drm_intel_bo_unreference(brw->ib.bo); 854 brw->ib.bo = bo; 855 856 brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER; 857 } else { 858 drm_intel_bo_unreference(bo); 859 } 860 861 if (index_buffer->type != brw->ib.type) { 862 brw->ib.type = index_buffer->type; 863 brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER; 864 } 865} 866 867const struct brw_tracked_state brw_indices = { 868 .dirty = { 869 .mesa = 0, 870 .brw = BRW_NEW_INDICES, 871 .cache = 0, 872 }, 873 .emit = brw_upload_indices, 874}; 875 876static void brw_emit_index_buffer(struct brw_context *brw) 877{ 878 struct intel_context *intel = &brw->intel; 879 const struct _mesa_index_buffer *index_buffer = brw->ib.ib; 880 GLuint cut_index_setting; 881 882 if (index_buffer == NULL) 883 return; 884 885 if (brw->prim_restart.enable_cut_index && !intel->is_haswell) { 886 cut_index_setting = BRW_CUT_INDEX_ENABLE; 887 } else { 888 cut_index_setting = 0; 889 } 890 891 BEGIN_BATCH(3); 892 OUT_BATCH(CMD_INDEX_BUFFER << 16 | 893 cut_index_setting | 894 get_index_type(index_buffer->type) << 8 | 895 1); 896 OUT_RELOC(brw->ib.bo, 897 I915_GEM_DOMAIN_VERTEX, 0, 898 0); 899 OUT_RELOC(brw->ib.bo, 900 I915_GEM_DOMAIN_VERTEX, 0, 901 brw->ib.bo->size - 1); 902 ADVANCE_BATCH(); 903} 904 905const struct brw_tracked_state brw_index_buffer = { 906 .dirty = { 907 .mesa = 0, 908 .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER, 909 .cache = 0, 910 }, 911 .emit = brw_emit_index_buffer, 912}; 913