brw_context.c revision bc24590f0c579a2528fd94eb8d40dd4ce12eba29
1/* 2 Copyright 2003 VMware, Inc. 3 Copyright (C) Intel Corp. 2006. All Rights Reserved. 4 Intel funded Tungsten Graphics to 5 develop this 3D driver. 6 7 Permission is hereby granted, free of charge, to any person obtaining 8 a copy of this software and associated documentation files (the 9 "Software"), to deal in the Software without restriction, including 10 without limitation the rights to use, copy, modify, merge, publish, 11 distribute, sublicense, and/or sell copies of the Software, and to 12 permit persons to whom the Software is furnished to do so, subject to 13 the following conditions: 14 15 The above copyright notice and this permission notice (including the 16 next paragraph) shall be included in all copies or substantial 17 portions of the Software. 18 19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 27 **********************************************************************/ 28 /* 29 * Authors: 30 * Keith Whitwell <keithw@vmware.com> 31 */ 32 33 34#include "main/api_exec.h" 35#include "main/context.h" 36#include "main/fbobject.h" 37#include "main/extensions.h" 38#include "main/imports.h" 39#include "main/macros.h" 40#include "main/points.h" 41#include "main/version.h" 42#include "main/vtxfmt.h" 43#include "main/texobj.h" 44#include "main/framebuffer.h" 45 46#include "vbo/vbo_context.h" 47 48#include "drivers/common/driverfuncs.h" 49#include "drivers/common/meta.h" 50#include "utils.h" 51 52#include "brw_context.h" 53#include "brw_defines.h" 54#include "brw_blorp.h" 55#include "brw_compiler.h" 56#include "brw_draw.h" 57#include "brw_state.h" 58 59#include "intel_batchbuffer.h" 60#include "intel_buffer_objects.h" 61#include "intel_buffers.h" 62#include "intel_fbo.h" 63#include "intel_mipmap_tree.h" 64#include "intel_pixel.h" 65#include "intel_image.h" 66#include "intel_tex.h" 67#include "intel_tex_obj.h" 68 69#include "swrast_setup/swrast_setup.h" 70#include "tnl/tnl.h" 71#include "tnl/t_pipeline.h" 72#include "util/ralloc.h" 73#include "util/debug.h" 74#include "isl/isl.h" 75 76/*************************************** 77 * Mesa's Driver Functions 78 ***************************************/ 79 80const char *const brw_vendor_string = "Intel Open Source Technology Center"; 81 82static const char * 83get_bsw_model(const struct intel_screen *screen) 84{ 85 switch (screen->eu_total) { 86 case 16: 87 return "405"; 88 case 12: 89 return "400"; 90 default: 91 return " "; 92 } 93} 94 95const char * 96brw_get_renderer_string(const struct intel_screen *screen) 97{ 98 const char *chipset; 99 static char buffer[128]; 100 char *bsw = NULL; 101 102 switch (screen->deviceID) { 103#undef CHIPSET 104#define CHIPSET(id, symbol, str) case id: chipset = str; break; 105#include "pci_ids/i965_pci_ids.h" 106 default: 107 chipset = "Unknown Intel Chipset"; 108 break; 109 } 110 111 /* Braswell branding is funny, so we have to fix it up here */ 112 if (screen->deviceID == 0x22B1) { 113 bsw = strdup(chipset); 114 char *needle = strstr(bsw, "XXX"); 115 if (needle) { 116 memcpy(needle, get_bsw_model(screen), 3); 117 chipset = bsw; 118 } 119 } 120 121 (void) driGetRendererString(buffer, chipset, 0); 122 free(bsw); 123 return buffer; 124} 125 126static const GLubyte * 127intel_get_string(struct gl_context * ctx, GLenum name) 128{ 129 const struct brw_context *const brw = brw_context(ctx); 130 131 switch (name) { 132 case GL_VENDOR: 133 return (GLubyte *) brw_vendor_string; 134 135 case GL_RENDERER: 136 return 137 (GLubyte *) brw_get_renderer_string(brw->screen); 138 139 default: 140 return NULL; 141 } 142} 143 144static void 145intel_viewport(struct gl_context *ctx) 146{ 147 struct brw_context *brw = brw_context(ctx); 148 __DRIcontext *driContext = brw->driContext; 149 150 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) { 151 if (driContext->driDrawablePriv) 152 dri2InvalidateDrawable(driContext->driDrawablePriv); 153 if (driContext->driReadablePriv) 154 dri2InvalidateDrawable(driContext->driReadablePriv); 155 } 156} 157 158static void 159intel_update_framebuffer(struct gl_context *ctx, 160 struct gl_framebuffer *fb) 161{ 162 struct brw_context *brw = brw_context(ctx); 163 164 /* Quantize the derived default number of samples 165 */ 166 fb->DefaultGeometry._NumSamples = 167 intel_quantize_num_samples(brw->screen, 168 fb->DefaultGeometry.NumSamples); 169} 170 171static bool 172intel_disable_rb_aux_buffer(struct brw_context *brw, const drm_intel_bo *bo) 173{ 174 const struct gl_framebuffer *fb = brw->ctx.DrawBuffer; 175 bool found = false; 176 177 for (unsigned i = 0; i < fb->_NumColorDrawBuffers; i++) { 178 const struct intel_renderbuffer *irb = 179 intel_renderbuffer(fb->_ColorDrawBuffers[i]); 180 181 if (irb && irb->mt->bo == bo) { 182 found = brw->draw_aux_buffer_disabled[i] = true; 183 } 184 } 185 186 return found; 187} 188 189/* On Gen9 color buffers may be compressed by the hardware (lossless 190 * compression). There are, however, format restrictions and care needs to be 191 * taken that the sampler engine is capable for re-interpreting a buffer with 192 * format different the buffer was originally written with. 193 * 194 * For example, SRGB formats are not compressible and the sampler engine isn't 195 * capable of treating RGBA_UNORM as SRGB_ALPHA. In such a case the underlying 196 * color buffer needs to be resolved so that the sampling surface can be 197 * sampled as non-compressed (i.e., without the auxiliary MCS buffer being 198 * set). 199 */ 200static bool 201intel_texture_view_requires_resolve(struct brw_context *brw, 202 struct intel_texture_object *intel_tex) 203{ 204 if (brw->gen < 9 || 205 !intel_miptree_is_lossless_compressed(brw, intel_tex->mt)) 206 return false; 207 208 const uint32_t brw_format = brw_format_for_mesa_format(intel_tex->_Format); 209 210 if (isl_format_supports_lossless_compression(&brw->screen->devinfo, 211 brw_format)) 212 return false; 213 214 perf_debug("Incompatible sampling format (%s) for rbc (%s)\n", 215 _mesa_get_format_name(intel_tex->_Format), 216 _mesa_get_format_name(intel_tex->mt->format)); 217 218 if (intel_disable_rb_aux_buffer(brw, intel_tex->mt->bo)) 219 perf_debug("Sampling renderbuffer with non-compressible format - " 220 "turning off compression"); 221 222 return true; 223} 224 225static void 226intel_update_state(struct gl_context * ctx, GLuint new_state) 227{ 228 struct brw_context *brw = brw_context(ctx); 229 struct intel_texture_object *tex_obj; 230 struct intel_renderbuffer *depth_irb; 231 232 if (ctx->swrast_context) 233 _swrast_InvalidateState(ctx, new_state); 234 _vbo_InvalidateState(ctx, new_state); 235 236 brw->NewGLState |= new_state; 237 238 _mesa_unlock_context_textures(ctx); 239 240 /* Resolve the depth buffer's HiZ buffer. */ 241 depth_irb = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH); 242 if (depth_irb) 243 intel_renderbuffer_resolve_hiz(brw, depth_irb); 244 245 memset(brw->draw_aux_buffer_disabled, 0, 246 sizeof(brw->draw_aux_buffer_disabled)); 247 248 /* Resolve depth buffer and render cache of each enabled texture. */ 249 int maxEnabledUnit = ctx->Texture._MaxEnabledTexImageUnit; 250 for (int i = 0; i <= maxEnabledUnit; i++) { 251 if (!ctx->Texture.Unit[i]._Current) 252 continue; 253 tex_obj = intel_texture_object(ctx->Texture.Unit[i]._Current); 254 if (!tex_obj || !tex_obj->mt) 255 continue; 256 intel_miptree_all_slices_resolve_depth(brw, tex_obj->mt); 257 /* Sampling engine understands lossless compression and resolving 258 * those surfaces should be skipped for performance reasons. 259 */ 260 const int flags = intel_texture_view_requires_resolve(brw, tex_obj) ? 261 0 : INTEL_MIPTREE_IGNORE_CCS_E; 262 intel_miptree_resolve_color(brw, tex_obj->mt, flags); 263 brw_render_cache_set_check_flush(brw, tex_obj->mt->bo); 264 265 if (tex_obj->base.StencilSampling || 266 tex_obj->mt->format == MESA_FORMAT_S_UINT8) { 267 intel_update_r8stencil(brw, tex_obj->mt); 268 } 269 } 270 271 /* Resolve color for each active shader image. */ 272 for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) { 273 const struct gl_linked_shader *shader = 274 ctx->_Shader->CurrentProgram[i] ? 275 ctx->_Shader->CurrentProgram[i]->_LinkedShaders[i] : NULL; 276 277 if (unlikely(shader && shader->NumImages)) { 278 for (unsigned j = 0; j < shader->NumImages; j++) { 279 struct gl_image_unit *u = &ctx->ImageUnits[shader->ImageUnits[j]]; 280 tex_obj = intel_texture_object(u->TexObj); 281 282 if (tex_obj && tex_obj->mt) { 283 /* Access to images is implemented using indirect messages 284 * against data port. Normal render target write understands 285 * lossless compression but unfortunately the typed/untyped 286 * read/write interface doesn't. Therefore even lossless 287 * compressed surfaces need to be resolved prior to accessing 288 * them. Hence skip setting INTEL_MIPTREE_IGNORE_CCS_E. 289 */ 290 intel_miptree_resolve_color(brw, tex_obj->mt, 0); 291 292 if (intel_miptree_is_lossless_compressed(brw, tex_obj->mt) && 293 intel_disable_rb_aux_buffer(brw, tex_obj->mt->bo)) { 294 perf_debug("Using renderbuffer as shader image - turning " 295 "off lossless compression"); 296 } 297 298 brw_render_cache_set_check_flush(brw, tex_obj->mt->bo); 299 } 300 } 301 } 302 } 303 304 /* Resolve color buffers for non-coherent framebuffer fetch. */ 305 if (!ctx->Extensions.MESA_shader_framebuffer_fetch && 306 ctx->FragmentProgram._Current && 307 ctx->FragmentProgram._Current->Base.OutputsRead) { 308 const struct gl_framebuffer *fb = ctx->DrawBuffer; 309 310 for (unsigned i = 0; i < fb->_NumColorDrawBuffers; i++) { 311 const struct intel_renderbuffer *irb = 312 intel_renderbuffer(fb->_ColorDrawBuffers[i]); 313 314 if (irb && 315 intel_miptree_resolve_color(brw, irb->mt, 316 INTEL_MIPTREE_IGNORE_CCS_E)) 317 brw_render_cache_set_check_flush(brw, irb->mt->bo); 318 } 319 } 320 321 /* If FRAMEBUFFER_SRGB is used on Gen9+ then we need to resolve any of the 322 * single-sampled color renderbuffers because the CCS buffer isn't 323 * supported for SRGB formats. This only matters if FRAMEBUFFER_SRGB is 324 * enabled because otherwise the surface state will be programmed with the 325 * linear equivalent format anyway. 326 */ 327 if (brw->gen >= 9 && ctx->Color.sRGBEnabled) { 328 struct gl_framebuffer *fb = ctx->DrawBuffer; 329 for (int i = 0; i < fb->_NumColorDrawBuffers; i++) { 330 struct gl_renderbuffer *rb = fb->_ColorDrawBuffers[i]; 331 332 if (rb == NULL) 333 continue; 334 335 struct intel_renderbuffer *irb = intel_renderbuffer(rb); 336 struct intel_mipmap_tree *mt = irb->mt; 337 338 if (mt == NULL || 339 mt->num_samples > 1 || 340 _mesa_get_srgb_format_linear(mt->format) == mt->format) 341 continue; 342 343 /* Lossless compression is not supported for SRGB formats, it 344 * should be impossible to get here with such surfaces. 345 */ 346 assert(!intel_miptree_is_lossless_compressed(brw, mt)); 347 intel_miptree_resolve_color(brw, mt, 0); 348 brw_render_cache_set_check_flush(brw, mt->bo); 349 } 350 } 351 352 _mesa_lock_context_textures(ctx); 353 354 if (new_state & _NEW_BUFFERS) { 355 intel_update_framebuffer(ctx, ctx->DrawBuffer); 356 if (ctx->DrawBuffer != ctx->ReadBuffer) 357 intel_update_framebuffer(ctx, ctx->ReadBuffer); 358 } 359} 360 361#define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer) 362 363static void 364intel_flush_front(struct gl_context *ctx) 365{ 366 struct brw_context *brw = brw_context(ctx); 367 __DRIcontext *driContext = brw->driContext; 368 __DRIdrawable *driDrawable = driContext->driDrawablePriv; 369 __DRIscreen *const dri_screen = brw->screen->driScrnPriv; 370 371 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) { 372 if (flushFront(dri_screen) && driDrawable && 373 driDrawable->loaderPrivate) { 374 375 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT. 376 * 377 * This potentially resolves both front and back buffer. It 378 * is unnecessary to resolve the back, but harms nothing except 379 * performance. And no one cares about front-buffer render 380 * performance. 381 */ 382 intel_resolve_for_dri2_flush(brw, driDrawable); 383 intel_batchbuffer_flush(brw); 384 385 flushFront(dri_screen)(driDrawable, driDrawable->loaderPrivate); 386 387 /* We set the dirty bit in intel_prepare_render() if we're 388 * front buffer rendering once we get there. 389 */ 390 brw->front_buffer_dirty = false; 391 } 392 } 393} 394 395static void 396intel_glFlush(struct gl_context *ctx) 397{ 398 struct brw_context *brw = brw_context(ctx); 399 400 intel_batchbuffer_flush(brw); 401 intel_flush_front(ctx); 402 403 brw->need_flush_throttle = true; 404} 405 406static void 407intel_finish(struct gl_context * ctx) 408{ 409 struct brw_context *brw = brw_context(ctx); 410 411 intel_glFlush(ctx); 412 413 if (brw->batch.last_bo) 414 drm_intel_bo_wait_rendering(brw->batch.last_bo); 415} 416 417static void 418brw_init_driver_functions(struct brw_context *brw, 419 struct dd_function_table *functions) 420{ 421 _mesa_init_driver_functions(functions); 422 423 /* GLX uses DRI2 invalidate events to handle window resizing. 424 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib), 425 * which doesn't provide a mechanism for snooping the event queues. 426 * 427 * So EGL still relies on viewport hacks to handle window resizing. 428 * This should go away with DRI3000. 429 */ 430 if (!brw->driContext->driScreenPriv->dri2.useInvalidate) 431 functions->Viewport = intel_viewport; 432 433 functions->Flush = intel_glFlush; 434 functions->Finish = intel_finish; 435 functions->GetString = intel_get_string; 436 functions->UpdateState = intel_update_state; 437 438 intelInitTextureFuncs(functions); 439 intelInitTextureImageFuncs(functions); 440 intelInitTextureSubImageFuncs(functions); 441 intelInitTextureCopyImageFuncs(functions); 442 intelInitCopyImageFuncs(functions); 443 intelInitClearFuncs(functions); 444 intelInitBufferFuncs(functions); 445 intelInitPixelFuncs(functions); 446 intelInitBufferObjectFuncs(functions); 447 intel_init_syncobj_functions(functions); 448 brw_init_object_purgeable_functions(functions); 449 450 brwInitFragProgFuncs( functions ); 451 brw_init_common_queryobj_functions(functions); 452 if (brw->gen >= 8 || brw->is_haswell) 453 hsw_init_queryobj_functions(functions); 454 else if (brw->gen >= 6) 455 gen6_init_queryobj_functions(functions); 456 else 457 gen4_init_queryobj_functions(functions); 458 brw_init_compute_functions(functions); 459 if (brw->gen >= 7) 460 brw_init_conditional_render_functions(functions); 461 462 functions->QueryInternalFormat = brw_query_internal_format; 463 464 functions->NewTransformFeedback = brw_new_transform_feedback; 465 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 466 if (brw->screen->has_mi_math_and_lrr) { 467 functions->BeginTransformFeedback = hsw_begin_transform_feedback; 468 functions->EndTransformFeedback = hsw_end_transform_feedback; 469 functions->PauseTransformFeedback = hsw_pause_transform_feedback; 470 functions->ResumeTransformFeedback = hsw_resume_transform_feedback; 471 } else if (brw->gen >= 7) { 472 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 473 functions->EndTransformFeedback = gen7_end_transform_feedback; 474 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 475 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 476 functions->GetTransformFeedbackVertexCount = 477 brw_get_transform_feedback_vertex_count; 478 } else { 479 functions->BeginTransformFeedback = brw_begin_transform_feedback; 480 functions->EndTransformFeedback = brw_end_transform_feedback; 481 } 482 483 if (brw->gen >= 6) 484 functions->GetSamplePosition = gen6_get_sample_position; 485} 486 487static void 488brw_initialize_context_constants(struct brw_context *brw) 489{ 490 struct gl_context *ctx = &brw->ctx; 491 const struct brw_compiler *compiler = brw->screen->compiler; 492 493 const bool stage_exists[MESA_SHADER_STAGES] = { 494 [MESA_SHADER_VERTEX] = true, 495 [MESA_SHADER_TESS_CTRL] = brw->gen >= 7, 496 [MESA_SHADER_TESS_EVAL] = brw->gen >= 7, 497 [MESA_SHADER_GEOMETRY] = brw->gen >= 6, 498 [MESA_SHADER_FRAGMENT] = true, 499 [MESA_SHADER_COMPUTE] = 500 (ctx->API == API_OPENGL_CORE && 501 ctx->Const.MaxComputeWorkGroupSize[0] >= 1024) || 502 (ctx->API == API_OPENGLES2 && 503 ctx->Const.MaxComputeWorkGroupSize[0] >= 128) || 504 _mesa_extension_override_enables.ARB_compute_shader, 505 }; 506 507 unsigned num_stages = 0; 508 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 509 if (stage_exists[i]) 510 num_stages++; 511 } 512 513 unsigned max_samplers = 514 brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16; 515 516 ctx->Const.MaxDualSourceDrawBuffers = 1; 517 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 518 ctx->Const.MaxCombinedShaderOutputResources = 519 MAX_IMAGE_UNITS + BRW_MAX_DRAW_BUFFERS; 520 521 ctx->Const.QueryCounterBits.Timestamp = 36; 522 523 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 524 ctx->Const.MaxImageUnits = MAX_IMAGE_UNITS; 525 ctx->Const.MaxRenderbufferSize = 8192; 526 ctx->Const.MaxTextureLevels = MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS); 527 ctx->Const.Max3DTextureLevels = 12; /* 2048 */ 528 ctx->Const.MaxCubeTextureLevels = 14; /* 8192 */ 529 ctx->Const.MaxArrayTextureLayers = brw->gen >= 7 ? 2048 : 512; 530 ctx->Const.MaxTextureMbytes = 1536; 531 ctx->Const.MaxTextureRectSize = 1 << 12; 532 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 533 ctx->Const.StripTextureBorder = true; 534 if (brw->gen >= 7) 535 ctx->Const.MaxProgramTextureGatherComponents = 4; 536 else if (brw->gen == 6) 537 ctx->Const.MaxProgramTextureGatherComponents = 1; 538 539 ctx->Const.MaxUniformBlockSize = 65536; 540 541 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 542 struct gl_program_constants *prog = &ctx->Const.Program[i]; 543 544 if (!stage_exists[i]) 545 continue; 546 547 prog->MaxTextureImageUnits = max_samplers; 548 549 prog->MaxUniformBlocks = BRW_MAX_UBO; 550 prog->MaxCombinedUniformComponents = 551 prog->MaxUniformComponents + 552 ctx->Const.MaxUniformBlockSize / 4 * prog->MaxUniformBlocks; 553 554 prog->MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 555 prog->MaxAtomicBuffers = BRW_MAX_ABO; 556 prog->MaxImageUniforms = compiler->scalar_stage[i] ? BRW_MAX_IMAGES : 0; 557 prog->MaxShaderStorageBlocks = BRW_MAX_SSBO; 558 } 559 560 ctx->Const.MaxTextureUnits = 561 MIN2(ctx->Const.MaxTextureCoordUnits, 562 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits); 563 564 ctx->Const.MaxUniformBufferBindings = num_stages * BRW_MAX_UBO; 565 ctx->Const.MaxCombinedUniformBlocks = num_stages * BRW_MAX_UBO; 566 ctx->Const.MaxCombinedAtomicBuffers = num_stages * BRW_MAX_ABO; 567 ctx->Const.MaxCombinedShaderStorageBlocks = num_stages * BRW_MAX_SSBO; 568 ctx->Const.MaxShaderStorageBufferBindings = num_stages * BRW_MAX_SSBO; 569 ctx->Const.MaxCombinedTextureImageUnits = num_stages * max_samplers; 570 ctx->Const.MaxCombinedImageUniforms = num_stages * BRW_MAX_IMAGES; 571 572 573 /* Hardware only supports a limited number of transform feedback buffers. 574 * So we need to override the Mesa default (which is based only on software 575 * limits). 576 */ 577 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 578 579 /* On Gen6, in the worst case, we use up one binding table entry per 580 * transform feedback component (see comments above the definition of 581 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 582 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 583 * BRW_MAX_SOL_BINDINGS. 584 * 585 * In "separate components" mode, we need to divide this value by 586 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 587 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 588 */ 589 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 590 ctx->Const.MaxTransformFeedbackSeparateComponents = 591 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 592 593 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = 594 !brw->screen->has_mi_math_and_lrr; 595 596 int max_samples; 597 const int *msaa_modes = intel_supported_msaa_modes(brw->screen); 598 const int clamp_max_samples = 599 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 600 601 if (clamp_max_samples < 0) { 602 max_samples = msaa_modes[0]; 603 } else { 604 /* Select the largest supported MSAA mode that does not exceed 605 * clamp_max_samples. 606 */ 607 max_samples = 0; 608 for (int i = 0; msaa_modes[i] != 0; ++i) { 609 if (msaa_modes[i] <= clamp_max_samples) { 610 max_samples = msaa_modes[i]; 611 break; 612 } 613 } 614 } 615 616 ctx->Const.MaxSamples = max_samples; 617 ctx->Const.MaxColorTextureSamples = max_samples; 618 ctx->Const.MaxDepthTextureSamples = max_samples; 619 ctx->Const.MaxIntegerSamples = max_samples; 620 ctx->Const.MaxImageSamples = 0; 621 622 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used 623 * to map indices of rectangular grid to sample numbers within a pixel. 624 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled 625 * extension implementation. For more details see the comment above 626 * gen6_set_sample_maps() definition. 627 */ 628 gen6_set_sample_maps(ctx); 629 630 ctx->Const.MinLineWidth = 1.0; 631 ctx->Const.MinLineWidthAA = 1.0; 632 if (brw->gen >= 6) { 633 ctx->Const.MaxLineWidth = 7.375; 634 ctx->Const.MaxLineWidthAA = 7.375; 635 ctx->Const.LineWidthGranularity = 0.125; 636 } else { 637 ctx->Const.MaxLineWidth = 7.0; 638 ctx->Const.MaxLineWidthAA = 7.0; 639 ctx->Const.LineWidthGranularity = 0.5; 640 } 641 642 /* For non-antialiased lines, we have to round the line width to the 643 * nearest whole number. Make sure that we don't advertise a line 644 * width that, when rounded, will be beyond the actual hardware 645 * maximum. 646 */ 647 assert(roundf(ctx->Const.MaxLineWidth) <= ctx->Const.MaxLineWidth); 648 649 ctx->Const.MinPointSize = 1.0; 650 ctx->Const.MinPointSizeAA = 1.0; 651 ctx->Const.MaxPointSize = 255.0; 652 ctx->Const.MaxPointSizeAA = 255.0; 653 ctx->Const.PointSizeGranularity = 1.0; 654 655 if (brw->gen >= 5 || brw->is_g4x) 656 ctx->Const.MaxClipPlanes = 8; 657 658 ctx->Const.LowerTessLevel = true; 659 ctx->Const.LowerTCSPatchVerticesIn = brw->gen >= 8; 660 ctx->Const.LowerTESPatchVerticesIn = true; 661 ctx->Const.PrimitiveRestartForPatches = true; 662 663 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024; 664 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0; 665 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0; 666 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0; 667 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0; 668 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0; 669 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0; 670 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16; 671 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256; 672 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1; 673 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024; 674 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams = 675 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters, 676 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams); 677 678 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024; 679 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024; 680 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024; 681 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024; 682 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12; 683 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256; 684 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0; 685 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024; 686 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams = 687 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters, 688 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams); 689 690 /* Fragment shaders use real, 32-bit twos-complement integers for all 691 * integer types. 692 */ 693 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31; 694 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30; 695 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0; 696 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 697 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 698 699 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMin = 31; 700 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMax = 30; 701 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.Precision = 0; 702 ctx->Const.Program[MESA_SHADER_VERTEX].HighInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt; 703 ctx->Const.Program[MESA_SHADER_VERTEX].MediumInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt; 704 705 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 706 * but we're not sure how it's actually done for vertex order, 707 * that affect provoking vertex decision. Always use last vertex 708 * convention for quad primitive which works as expected for now. 709 */ 710 if (brw->gen >= 6) 711 ctx->Const.QuadsFollowProvokingVertexConvention = false; 712 713 ctx->Const.NativeIntegers = true; 714 ctx->Const.VertexID_is_zero_based = true; 715 716 /* Regarding the CMP instruction, the Ivybridge PRM says: 717 * 718 * "For each enabled channel 0b or 1b is assigned to the appropriate flag 719 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord 720 * 0xFFFFFFFF) is assigned to dst." 721 * 722 * but PRMs for earlier generations say 723 * 724 * "In dword format, one GRF may store up to 8 results. When the register 725 * is used later as a vector of Booleans, as only LSB at each channel 726 * contains meaning [sic] data, software should make sure all higher bits 727 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)." 728 * 729 * We select the representation of a true boolean uniform to be ~0, and fix 730 * the results of Gen <= 5 CMP instruction's with -(result & 1). 731 */ 732 ctx->Const.UniformBooleanTrue = ~0; 733 734 /* From the gen4 PRM, volume 4 page 127: 735 * 736 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 737 * the base address of the first element of the surface, computed in 738 * software by adding the surface base address to the byte offset of 739 * the element in the buffer." 740 * 741 * However, unaligned accesses are slower, so enforce buffer alignment. 742 */ 743 ctx->Const.UniformBufferOffsetAlignment = 16; 744 745 /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so 746 * that we can safely have the CPU and GPU writing the same SSBO on 747 * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never 748 * writes, so there's no problem. For an SSBO, the GPU and the CPU can 749 * be updating disjoint regions of the buffer simultaneously and that will 750 * break if the regions overlap the same cacheline. 751 */ 752 ctx->Const.ShaderStorageBufferOffsetAlignment = 64; 753 ctx->Const.TextureBufferOffsetAlignment = 16; 754 ctx->Const.MaxTextureBufferSize = 128 * 1024 * 1024; 755 756 if (brw->gen >= 6) { 757 ctx->Const.MaxVarying = 32; 758 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128; 759 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64; 760 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128; 761 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128; 762 ctx->Const.Program[MESA_SHADER_TESS_CTRL].MaxInputComponents = 128; 763 ctx->Const.Program[MESA_SHADER_TESS_CTRL].MaxOutputComponents = 128; 764 ctx->Const.Program[MESA_SHADER_TESS_EVAL].MaxInputComponents = 128; 765 ctx->Const.Program[MESA_SHADER_TESS_EVAL].MaxOutputComponents = 128; 766 } 767 768 /* We want the GLSL compiler to emit code that uses condition codes */ 769 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 770 ctx->Const.ShaderCompilerOptions[i] = 771 brw->screen->compiler->glsl_compiler_options[i]; 772 } 773 774 if (brw->gen >= 7) { 775 ctx->Const.MaxViewportWidth = 32768; 776 ctx->Const.MaxViewportHeight = 32768; 777 } 778 779 /* ARB_viewport_array */ 780 if (brw->gen >= 6 && ctx->API == API_OPENGL_CORE) { 781 ctx->Const.MaxViewports = GEN6_NUM_VIEWPORTS; 782 ctx->Const.ViewportSubpixelBits = 0; 783 784 /* Cast to float before negating because MaxViewportWidth is unsigned. 785 */ 786 ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth; 787 ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth; 788 } 789 790 /* ARB_gpu_shader5 */ 791 if (brw->gen >= 7) 792 ctx->Const.MaxVertexStreams = MIN2(4, MAX_VERTEX_STREAMS); 793 794 /* ARB_framebuffer_no_attachments */ 795 ctx->Const.MaxFramebufferWidth = 16384; 796 ctx->Const.MaxFramebufferHeight = 16384; 797 ctx->Const.MaxFramebufferLayers = ctx->Const.MaxArrayTextureLayers; 798 ctx->Const.MaxFramebufferSamples = max_samples; 799 800 /* OES_primitive_bounding_box */ 801 ctx->Const.NoPrimitiveBoundingBoxOutput = true; 802} 803 804static void 805brw_initialize_cs_context_constants(struct brw_context *brw) 806{ 807 struct gl_context *ctx = &brw->ctx; 808 const struct intel_screen *screen = brw->screen; 809 const struct gen_device_info *devinfo = &screen->devinfo; 810 811 /* FINISHME: Do this for all platforms that the kernel supports */ 812 if (brw->is_cherryview && 813 screen->subslice_total > 0 && screen->eu_total > 0) { 814 /* Logical CS threads = EUs per subslice * 7 threads per EU */ 815 brw->max_cs_threads = screen->eu_total / screen->subslice_total * 7; 816 817 /* Fuse configurations may give more threads than expected, never less. */ 818 if (brw->max_cs_threads < devinfo->max_cs_threads) 819 brw->max_cs_threads = devinfo->max_cs_threads; 820 } else { 821 brw->max_cs_threads = devinfo->max_cs_threads; 822 } 823 824 /* Maximum number of scalar compute shader invocations that can be run in 825 * parallel in the same subslice assuming SIMD32 dispatch. 826 * 827 * We don't advertise more than 64 threads, because we are limited to 64 by 828 * our usage of thread_width_max in the gpgpu walker command. This only 829 * currently impacts Haswell, which otherwise might be able to advertise 70 830 * threads. With SIMD32 and 64 threads, Haswell still provides twice the 831 * required the number of invocation needed for ARB_compute_shader. 832 */ 833 const unsigned max_threads = MIN2(64, brw->max_cs_threads); 834 const uint32_t max_invocations = 32 * max_threads; 835 ctx->Const.MaxComputeWorkGroupSize[0] = max_invocations; 836 ctx->Const.MaxComputeWorkGroupSize[1] = max_invocations; 837 ctx->Const.MaxComputeWorkGroupSize[2] = max_invocations; 838 ctx->Const.MaxComputeWorkGroupInvocations = max_invocations; 839 ctx->Const.MaxComputeSharedMemorySize = 64 * 1024; 840} 841 842/** 843 * Process driconf (drirc) options, setting appropriate context flags. 844 * 845 * intelInitExtensions still pokes at optionCache directly, in order to 846 * avoid advertising various extensions. No flags are set, so it makes 847 * sense to continue doing that there. 848 */ 849static void 850brw_process_driconf_options(struct brw_context *brw) 851{ 852 struct gl_context *ctx = &brw->ctx; 853 854 driOptionCache *options = &brw->optionCache; 855 driParseConfigFiles(options, &brw->screen->optionCache, 856 brw->driContext->driScreenPriv->myNum, "i965"); 857 858 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 859 switch (bo_reuse_mode) { 860 case DRI_CONF_BO_REUSE_DISABLED: 861 break; 862 case DRI_CONF_BO_REUSE_ALL: 863 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 864 break; 865 } 866 867 if (!driQueryOptionb(options, "hiz")) { 868 brw->has_hiz = false; 869 /* On gen6, you can only do separate stencil with HIZ. */ 870 if (brw->gen == 6) 871 brw->has_separate_stencil = false; 872 } 873 874 if (driQueryOptionb(options, "always_flush_batch")) { 875 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 876 brw->always_flush_batch = true; 877 } 878 879 if (driQueryOptionb(options, "always_flush_cache")) { 880 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 881 brw->always_flush_cache = true; 882 } 883 884 if (driQueryOptionb(options, "disable_throttling")) { 885 fprintf(stderr, "disabling flush throttling\n"); 886 brw->disable_throttling = true; 887 } 888 889 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 890 891 if (driQueryOptionb(&brw->optionCache, "precise_trig")) 892 brw->screen->compiler->precise_trig = true; 893 894 ctx->Const.ForceGLSLExtensionsWarn = 895 driQueryOptionb(options, "force_glsl_extensions_warn"); 896 897 ctx->Const.DisableGLSLLineContinuations = 898 driQueryOptionb(options, "disable_glsl_line_continuations"); 899 900 ctx->Const.AllowGLSLExtensionDirectiveMidShader = 901 driQueryOptionb(options, "allow_glsl_extension_directive_midshader"); 902 903 ctx->Const.GLSLZeroInit = driQueryOptionb(options, "glsl_zero_init"); 904 905 brw->dual_color_blend_by_location = 906 driQueryOptionb(options, "dual_color_blend_by_location"); 907} 908 909GLboolean 910brwCreateContext(gl_api api, 911 const struct gl_config *mesaVis, 912 __DRIcontext *driContextPriv, 913 unsigned major_version, 914 unsigned minor_version, 915 uint32_t flags, 916 bool notify_reset, 917 unsigned *dri_ctx_error, 918 void *sharedContextPrivate) 919{ 920 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 921 struct intel_screen *screen = driContextPriv->driScreenPriv->driverPrivate; 922 const struct gen_device_info *devinfo = &screen->devinfo; 923 struct dd_function_table functions; 924 925 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 926 * provides us with context reset notifications. 927 */ 928 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 929 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 930 931 if (screen->has_context_reset_notification) 932 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 933 934 if (flags & ~allowed_flags) { 935 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 936 return false; 937 } 938 939 struct brw_context *brw = rzalloc(NULL, struct brw_context); 940 if (!brw) { 941 fprintf(stderr, "%s: failed to alloc context\n", __func__); 942 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 943 return false; 944 } 945 946 driContextPriv->driverPrivate = brw; 947 brw->driContext = driContextPriv; 948 brw->screen = screen; 949 brw->bufmgr = screen->bufmgr; 950 951 brw->gen = devinfo->gen; 952 brw->gt = devinfo->gt; 953 brw->is_g4x = devinfo->is_g4x; 954 brw->is_baytrail = devinfo->is_baytrail; 955 brw->is_haswell = devinfo->is_haswell; 956 brw->is_cherryview = devinfo->is_cherryview; 957 brw->is_broxton = devinfo->is_broxton; 958 brw->has_llc = devinfo->has_llc; 959 brw->has_hiz = devinfo->has_hiz_and_separate_stencil; 960 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 961 brw->has_pln = devinfo->has_pln; 962 brw->has_compr4 = devinfo->has_compr4; 963 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 964 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 965 brw->needs_unlit_centroid_workaround = 966 devinfo->needs_unlit_centroid_workaround; 967 968 brw->must_use_separate_stencil = devinfo->must_use_separate_stencil; 969 brw->has_swizzling = screen->hw_has_swizzling; 970 971 isl_device_init(&brw->isl_dev, devinfo, screen->hw_has_swizzling); 972 973 brw->vs.base.stage = MESA_SHADER_VERTEX; 974 brw->tcs.base.stage = MESA_SHADER_TESS_CTRL; 975 brw->tes.base.stage = MESA_SHADER_TESS_EVAL; 976 brw->gs.base.stage = MESA_SHADER_GEOMETRY; 977 brw->wm.base.stage = MESA_SHADER_FRAGMENT; 978 if (brw->gen >= 8) { 979 gen8_init_vtable_surface_functions(brw); 980 brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz; 981 } else if (brw->gen >= 7) { 982 gen7_init_vtable_surface_functions(brw); 983 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 984 } else if (brw->gen >= 6) { 985 gen6_init_vtable_surface_functions(brw); 986 brw->vtbl.emit_depth_stencil_hiz = gen6_emit_depth_stencil_hiz; 987 } else { 988 gen4_init_vtable_surface_functions(brw); 989 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 990 } 991 992 brw_init_driver_functions(brw, &functions); 993 994 if (notify_reset) 995 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 996 997 struct gl_context *ctx = &brw->ctx; 998 999 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 1000 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 1001 fprintf(stderr, "%s: failed to init mesa context\n", __func__); 1002 intelDestroyContext(driContextPriv); 1003 return false; 1004 } 1005 1006 driContextSetFlags(ctx, flags); 1007 1008 /* Initialize the software rasterizer and helper modules. 1009 * 1010 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 1011 * software fallbacks (which we have to support on legacy GL to do weird 1012 * glDrawPixels(), glBitmap(), and other functions). 1013 */ 1014 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 1015 _swrast_CreateContext(ctx); 1016 } 1017 1018 _vbo_CreateContext(ctx); 1019 if (ctx->swrast_context) { 1020 _tnl_CreateContext(ctx); 1021 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 1022 _swsetup_CreateContext(ctx); 1023 1024 /* Configure swrast to match hardware characteristics: */ 1025 _swrast_allow_pixel_fog(ctx, false); 1026 _swrast_allow_vertex_fog(ctx, true); 1027 } 1028 1029 _mesa_meta_init(ctx); 1030 1031 brw_process_driconf_options(brw); 1032 1033 if (INTEL_DEBUG & DEBUG_PERF) 1034 brw->perf_debug = true; 1035 1036 brw_initialize_cs_context_constants(brw); 1037 brw_initialize_context_constants(brw); 1038 1039 ctx->Const.ResetStrategy = notify_reset 1040 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 1041 1042 /* Reinitialize the context point state. It depends on ctx->Const values. */ 1043 _mesa_init_point(ctx); 1044 1045 intel_fbo_init(brw); 1046 1047 intel_batchbuffer_init(brw); 1048 1049 if (brw->gen >= 6) { 1050 /* Create a new hardware context. Using a hardware context means that 1051 * our GPU state will be saved/restored on context switch, allowing us 1052 * to assume that the GPU is in the same state we left it in. 1053 * 1054 * This is required for transform feedback buffer offsets, query objects, 1055 * and also allows us to reduce how much state we have to emit. 1056 */ 1057 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 1058 1059 if (!brw->hw_ctx) { 1060 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 1061 intelDestroyContext(driContextPriv); 1062 return false; 1063 } 1064 } 1065 1066 if (brw_init_pipe_control(brw, devinfo)) { 1067 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 1068 intelDestroyContext(driContextPriv); 1069 return false; 1070 } 1071 1072 brw_init_state(brw); 1073 1074 intelInitExtensions(ctx); 1075 1076 brw_init_surface_formats(brw); 1077 1078 if (brw->gen >= 6) 1079 brw_blorp_init(brw); 1080 1081 brw->max_vs_threads = devinfo->max_vs_threads; 1082 brw->max_hs_threads = devinfo->max_hs_threads; 1083 brw->max_ds_threads = devinfo->max_ds_threads; 1084 brw->max_gs_threads = devinfo->max_gs_threads; 1085 brw->max_wm_threads = devinfo->max_wm_threads; 1086 brw->urb.size = devinfo->urb.size; 1087 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 1088 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 1089 brw->urb.max_hs_entries = devinfo->urb.max_hs_entries; 1090 brw->urb.max_ds_entries = devinfo->urb.max_ds_entries; 1091 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 1092 1093 if (brw->gen == 6) 1094 brw->urb.gs_present = false; 1095 1096 brw->prim_restart.in_progress = false; 1097 brw->prim_restart.enable_cut_index = false; 1098 brw->gs.enabled = false; 1099 brw->sf.viewport_transform_enable = true; 1100 1101 brw->predicate.state = BRW_PREDICATE_STATE_RENDER; 1102 1103 brw->max_gtt_map_object_size = screen->max_gtt_map_object_size; 1104 1105 brw->use_resource_streamer = screen->has_resource_streamer && 1106 (env_var_as_boolean("INTEL_USE_HW_BT", false) || 1107 env_var_as_boolean("INTEL_USE_GATHER", false)); 1108 1109 ctx->VertexProgram._MaintainTnlProgram = true; 1110 ctx->FragmentProgram._MaintainTexEnvProgram = true; 1111 1112 brw_draw_init( brw ); 1113 1114 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 1115 /* Turn on some extra GL_ARB_debug_output generation. */ 1116 brw->perf_debug = true; 1117 } 1118 1119 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 1120 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 1121 1122 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 1123 brw_init_shader_time(brw); 1124 1125 _mesa_compute_version(ctx); 1126 1127 _mesa_initialize_dispatch_tables(ctx); 1128 _mesa_initialize_vbo_vtxfmt(ctx); 1129 1130 if (ctx->Extensions.AMD_performance_monitor) { 1131 brw_init_performance_monitors(brw); 1132 } 1133 1134 vbo_use_buffer_objects(ctx); 1135 vbo_always_unmap_buffers(ctx); 1136 1137 return true; 1138} 1139 1140void 1141intelDestroyContext(__DRIcontext * driContextPriv) 1142{ 1143 struct brw_context *brw = 1144 (struct brw_context *) driContextPriv->driverPrivate; 1145 struct gl_context *ctx = &brw->ctx; 1146 1147 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 1148 if (INTEL_DEBUG & DEBUG_AUB) { 1149 intel_batchbuffer_flush(brw); 1150 aub_dump_bmp(&brw->ctx); 1151 } 1152 1153 _mesa_meta_free(&brw->ctx); 1154 1155 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 1156 /* Force a report. */ 1157 brw->shader_time.report_time = 0; 1158 1159 brw_collect_and_report_shader_time(brw); 1160 brw_destroy_shader_time(brw); 1161 } 1162 1163 if (brw->gen >= 6) 1164 blorp_finish(&brw->blorp); 1165 1166 brw_destroy_state(brw); 1167 brw_draw_destroy(brw); 1168 1169 drm_intel_bo_unreference(brw->curbe.curbe_bo); 1170 if (brw->vs.base.scratch_bo) 1171 drm_intel_bo_unreference(brw->vs.base.scratch_bo); 1172 if (brw->tcs.base.scratch_bo) 1173 drm_intel_bo_unreference(brw->tcs.base.scratch_bo); 1174 if (brw->tes.base.scratch_bo) 1175 drm_intel_bo_unreference(brw->tes.base.scratch_bo); 1176 if (brw->gs.base.scratch_bo) 1177 drm_intel_bo_unreference(brw->gs.base.scratch_bo); 1178 if (brw->wm.base.scratch_bo) 1179 drm_intel_bo_unreference(brw->wm.base.scratch_bo); 1180 1181 gen7_reset_hw_bt_pool_offsets(brw); 1182 drm_intel_bo_unreference(brw->hw_bt_pool.bo); 1183 brw->hw_bt_pool.bo = NULL; 1184 1185 drm_intel_gem_context_destroy(brw->hw_ctx); 1186 1187 if (ctx->swrast_context) { 1188 _swsetup_DestroyContext(&brw->ctx); 1189 _tnl_DestroyContext(&brw->ctx); 1190 } 1191 _vbo_DestroyContext(&brw->ctx); 1192 1193 if (ctx->swrast_context) 1194 _swrast_DestroyContext(&brw->ctx); 1195 1196 brw_fini_pipe_control(brw); 1197 intel_batchbuffer_free(brw); 1198 1199 drm_intel_bo_unreference(brw->throttle_batch[1]); 1200 drm_intel_bo_unreference(brw->throttle_batch[0]); 1201 brw->throttle_batch[1] = NULL; 1202 brw->throttle_batch[0] = NULL; 1203 1204 driDestroyOptionCache(&brw->optionCache); 1205 1206 /* free the Mesa context */ 1207 _mesa_free_context_data(&brw->ctx); 1208 1209 ralloc_free(brw); 1210 driContextPriv->driverPrivate = NULL; 1211} 1212 1213GLboolean 1214intelUnbindContext(__DRIcontext * driContextPriv) 1215{ 1216 /* Unset current context and dispath table */ 1217 _mesa_make_current(NULL, NULL, NULL); 1218 1219 return true; 1220} 1221 1222/** 1223 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 1224 * on window system framebuffers. 1225 * 1226 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 1227 * your renderbuffer can do sRGB encode, and you can flip a switch that does 1228 * sRGB encode if the renderbuffer can handle it. You can ask specifically 1229 * for a visual where you're guaranteed to be capable, but it turns out that 1230 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 1231 * incapable ones, because there's no difference between the two in resources 1232 * used. Applications thus get built that accidentally rely on the default 1233 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 1234 * great... 1235 * 1236 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 1237 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 1238 * So they removed the enable knob and made it "if the renderbuffer is sRGB 1239 * capable, do sRGB encode". Then, for your window system renderbuffers, you 1240 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 1241 * and get no sRGB encode (assuming that both kinds of visual are available). 1242 * Thus our choice to support sRGB by default on our visuals for desktop would 1243 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 1244 * 1245 * Unfortunately, renderbuffer setup happens before a context is created. So 1246 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 1247 * context (without an sRGB visual, though we don't have sRGB visuals exposed 1248 * yet), we go turn that back off before anyone finds out. 1249 */ 1250static void 1251intel_gles3_srgb_workaround(struct brw_context *brw, 1252 struct gl_framebuffer *fb) 1253{ 1254 struct gl_context *ctx = &brw->ctx; 1255 1256 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 1257 return; 1258 1259 /* Some day when we support the sRGB capable bit on visuals available for 1260 * GLES, we'll need to respect that and not disable things here. 1261 */ 1262 fb->Visual.sRGBCapable = false; 1263 for (int i = 0; i < BUFFER_COUNT; i++) { 1264 struct gl_renderbuffer *rb = fb->Attachment[i].Renderbuffer; 1265 if (rb) 1266 rb->Format = _mesa_get_srgb_format_linear(rb->Format); 1267 } 1268} 1269 1270GLboolean 1271intelMakeCurrent(__DRIcontext * driContextPriv, 1272 __DRIdrawable * driDrawPriv, 1273 __DRIdrawable * driReadPriv) 1274{ 1275 struct brw_context *brw; 1276 GET_CURRENT_CONTEXT(curCtx); 1277 1278 if (driContextPriv) 1279 brw = (struct brw_context *) driContextPriv->driverPrivate; 1280 else 1281 brw = NULL; 1282 1283 /* According to the glXMakeCurrent() man page: "Pending commands to 1284 * the previous context, if any, are flushed before it is released." 1285 * But only flush if we're actually changing contexts. 1286 */ 1287 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 1288 _mesa_flush(curCtx); 1289 } 1290 1291 if (driContextPriv) { 1292 struct gl_context *ctx = &brw->ctx; 1293 struct gl_framebuffer *fb, *readFb; 1294 1295 if (driDrawPriv == NULL) { 1296 fb = _mesa_get_incomplete_framebuffer(); 1297 } else { 1298 fb = driDrawPriv->driverPrivate; 1299 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 1300 } 1301 1302 if (driReadPriv == NULL) { 1303 readFb = _mesa_get_incomplete_framebuffer(); 1304 } else { 1305 readFb = driReadPriv->driverPrivate; 1306 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 1307 } 1308 1309 /* The sRGB workaround changes the renderbuffer's format. We must change 1310 * the format before the renderbuffer's miptree get's allocated, otherwise 1311 * the formats of the renderbuffer and its miptree will differ. 1312 */ 1313 intel_gles3_srgb_workaround(brw, fb); 1314 intel_gles3_srgb_workaround(brw, readFb); 1315 1316 /* If the context viewport hasn't been initialized, force a call out to 1317 * the loader to get buffers so we have a drawable size for the initial 1318 * viewport. */ 1319 if (!brw->ctx.ViewportInitialized) 1320 intel_prepare_render(brw); 1321 1322 _mesa_make_current(ctx, fb, readFb); 1323 } else { 1324 _mesa_make_current(NULL, NULL, NULL); 1325 } 1326 1327 return true; 1328} 1329 1330void 1331intel_resolve_for_dri2_flush(struct brw_context *brw, 1332 __DRIdrawable *drawable) 1333{ 1334 if (brw->gen < 6) { 1335 /* MSAA and fast color clear are not supported, so don't waste time 1336 * checking whether a resolve is needed. 1337 */ 1338 return; 1339 } 1340 1341 struct gl_framebuffer *fb = drawable->driverPrivate; 1342 struct intel_renderbuffer *rb; 1343 1344 /* Usually, only the back buffer will need to be downsampled. However, 1345 * the front buffer will also need it if the user has rendered into it. 1346 */ 1347 static const gl_buffer_index buffers[2] = { 1348 BUFFER_BACK_LEFT, 1349 BUFFER_FRONT_LEFT, 1350 }; 1351 1352 for (int i = 0; i < 2; ++i) { 1353 rb = intel_get_renderbuffer(fb, buffers[i]); 1354 if (rb == NULL || rb->mt == NULL) 1355 continue; 1356 if (rb->mt->num_samples <= 1) 1357 intel_miptree_resolve_color(brw, rb->mt, 0); 1358 else 1359 intel_renderbuffer_downsample(brw, rb); 1360 } 1361} 1362 1363static unsigned 1364intel_bits_per_pixel(const struct intel_renderbuffer *rb) 1365{ 1366 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 1367} 1368 1369static void 1370intel_query_dri2_buffers(struct brw_context *brw, 1371 __DRIdrawable *drawable, 1372 __DRIbuffer **buffers, 1373 int *count); 1374 1375static void 1376intel_process_dri2_buffer(struct brw_context *brw, 1377 __DRIdrawable *drawable, 1378 __DRIbuffer *buffer, 1379 struct intel_renderbuffer *rb, 1380 const char *buffer_name); 1381 1382static void 1383intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 1384 1385static void 1386intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1387{ 1388 struct gl_framebuffer *fb = drawable->driverPrivate; 1389 struct intel_renderbuffer *rb; 1390 __DRIbuffer *buffers = NULL; 1391 int i, count; 1392 const char *region_name; 1393 1394 /* Set this up front, so that in case our buffers get invalidated 1395 * while we're getting new buffers, we don't clobber the stamp and 1396 * thus ignore the invalidate. */ 1397 drawable->lastStamp = drawable->dri2.stamp; 1398 1399 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1400 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1401 1402 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1403 1404 if (buffers == NULL) 1405 return; 1406 1407 for (i = 0; i < count; i++) { 1408 switch (buffers[i].attachment) { 1409 case __DRI_BUFFER_FRONT_LEFT: 1410 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1411 region_name = "dri2 front buffer"; 1412 break; 1413 1414 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1415 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1416 region_name = "dri2 fake front buffer"; 1417 break; 1418 1419 case __DRI_BUFFER_BACK_LEFT: 1420 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1421 region_name = "dri2 back buffer"; 1422 break; 1423 1424 case __DRI_BUFFER_DEPTH: 1425 case __DRI_BUFFER_HIZ: 1426 case __DRI_BUFFER_DEPTH_STENCIL: 1427 case __DRI_BUFFER_STENCIL: 1428 case __DRI_BUFFER_ACCUM: 1429 default: 1430 fprintf(stderr, 1431 "unhandled buffer attach event, attachment type %d\n", 1432 buffers[i].attachment); 1433 return; 1434 } 1435 1436 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1437 } 1438 1439} 1440 1441void 1442intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1443{ 1444 struct brw_context *brw = context->driverPrivate; 1445 __DRIscreen *dri_screen = brw->screen->driScrnPriv; 1446 1447 /* Set this up front, so that in case our buffers get invalidated 1448 * while we're getting new buffers, we don't clobber the stamp and 1449 * thus ignore the invalidate. */ 1450 drawable->lastStamp = drawable->dri2.stamp; 1451 1452 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1453 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1454 1455 if (dri_screen->image.loader) 1456 intel_update_image_buffers(brw, drawable); 1457 else 1458 intel_update_dri2_buffers(brw, drawable); 1459 1460 driUpdateFramebufferSize(&brw->ctx, drawable); 1461} 1462 1463/** 1464 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1465 * state is required. 1466 */ 1467void 1468intel_prepare_render(struct brw_context *brw) 1469{ 1470 struct gl_context *ctx = &brw->ctx; 1471 __DRIcontext *driContext = brw->driContext; 1472 __DRIdrawable *drawable; 1473 1474 drawable = driContext->driDrawablePriv; 1475 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1476 if (drawable->lastStamp != drawable->dri2.stamp) 1477 intel_update_renderbuffers(driContext, drawable); 1478 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1479 } 1480 1481 drawable = driContext->driReadablePriv; 1482 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1483 if (drawable->lastStamp != drawable->dri2.stamp) 1484 intel_update_renderbuffers(driContext, drawable); 1485 driContext->dri2.read_stamp = drawable->dri2.stamp; 1486 } 1487 1488 /* If we're currently rendering to the front buffer, the rendering 1489 * that will happen next will probably dirty the front buffer. So 1490 * mark it as dirty here. 1491 */ 1492 if (_mesa_is_front_buffer_drawing(ctx->DrawBuffer)) 1493 brw->front_buffer_dirty = true; 1494} 1495 1496/** 1497 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1498 * 1499 * To determine which DRI buffers to request, examine the renderbuffers 1500 * attached to the drawable's framebuffer. Then request the buffers with 1501 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1502 * 1503 * This is called from intel_update_renderbuffers(). 1504 * 1505 * \param drawable Drawable whose buffers are queried. 1506 * \param buffers [out] List of buffers returned by DRI2 query. 1507 * \param buffer_count [out] Number of buffers returned. 1508 * 1509 * \see intel_update_renderbuffers() 1510 * \see DRI2GetBuffers() 1511 * \see DRI2GetBuffersWithFormat() 1512 */ 1513static void 1514intel_query_dri2_buffers(struct brw_context *brw, 1515 __DRIdrawable *drawable, 1516 __DRIbuffer **buffers, 1517 int *buffer_count) 1518{ 1519 __DRIscreen *dri_screen = brw->screen->driScrnPriv; 1520 struct gl_framebuffer *fb = drawable->driverPrivate; 1521 int i = 0; 1522 unsigned attachments[8]; 1523 1524 struct intel_renderbuffer *front_rb; 1525 struct intel_renderbuffer *back_rb; 1526 1527 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1528 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1529 1530 memset(attachments, 0, sizeof(attachments)); 1531 if ((_mesa_is_front_buffer_drawing(fb) || 1532 _mesa_is_front_buffer_reading(fb) || 1533 !back_rb) && front_rb) { 1534 /* If a fake front buffer is in use, then querying for 1535 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1536 * the real front buffer to the fake front buffer. So before doing the 1537 * query, we need to make sure all the pending drawing has landed in the 1538 * real front buffer. 1539 */ 1540 intel_batchbuffer_flush(brw); 1541 intel_flush_front(&brw->ctx); 1542 1543 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1544 attachments[i++] = intel_bits_per_pixel(front_rb); 1545 } else if (front_rb && brw->front_buffer_dirty) { 1546 /* We have pending front buffer rendering, but we aren't querying for a 1547 * front buffer. If the front buffer we have is a fake front buffer, 1548 * the X server is going to throw it away when it processes the query. 1549 * So before doing the query, make sure all the pending drawing has 1550 * landed in the real front buffer. 1551 */ 1552 intel_batchbuffer_flush(brw); 1553 intel_flush_front(&brw->ctx); 1554 } 1555 1556 if (back_rb) { 1557 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1558 attachments[i++] = intel_bits_per_pixel(back_rb); 1559 } 1560 1561 assert(i <= ARRAY_SIZE(attachments)); 1562 1563 *buffers = 1564 dri_screen->dri2.loader->getBuffersWithFormat(drawable, 1565 &drawable->w, 1566 &drawable->h, 1567 attachments, i / 2, 1568 buffer_count, 1569 drawable->loaderPrivate); 1570} 1571 1572/** 1573 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1574 * 1575 * This is called from intel_update_renderbuffers(). 1576 * 1577 * \par Note: 1578 * DRI buffers whose attachment point is DRI2BufferStencil or 1579 * DRI2BufferDepthStencil are handled as special cases. 1580 * 1581 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1582 * that is passed to drm_intel_bo_gem_create_from_name(). 1583 * 1584 * \see intel_update_renderbuffers() 1585 */ 1586static void 1587intel_process_dri2_buffer(struct brw_context *brw, 1588 __DRIdrawable *drawable, 1589 __DRIbuffer *buffer, 1590 struct intel_renderbuffer *rb, 1591 const char *buffer_name) 1592{ 1593 struct gl_framebuffer *fb = drawable->driverPrivate; 1594 drm_intel_bo *bo; 1595 1596 if (!rb) 1597 return; 1598 1599 unsigned num_samples = rb->Base.Base.NumSamples; 1600 1601 /* We try to avoid closing and reopening the same BO name, because the first 1602 * use of a mapping of the buffer involves a bunch of page faulting which is 1603 * moderately expensive. 1604 */ 1605 struct intel_mipmap_tree *last_mt; 1606 if (num_samples == 0) 1607 last_mt = rb->mt; 1608 else 1609 last_mt = rb->singlesample_mt; 1610 1611 uint32_t old_name = 0; 1612 if (last_mt) { 1613 /* The bo already has a name because the miptree was created by a 1614 * previous call to intel_process_dri2_buffer(). If a bo already has a 1615 * name, then drm_intel_bo_flink() is a low-cost getter. It does not 1616 * create a new name. 1617 */ 1618 drm_intel_bo_flink(last_mt->bo, &old_name); 1619 } 1620 1621 if (old_name == buffer->name) 1622 return; 1623 1624 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1625 fprintf(stderr, 1626 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1627 buffer->name, buffer->attachment, 1628 buffer->cpp, buffer->pitch); 1629 } 1630 1631 bo = drm_intel_bo_gem_create_from_name(brw->bufmgr, buffer_name, 1632 buffer->name); 1633 if (!bo) { 1634 fprintf(stderr, 1635 "Failed to open BO for returned DRI2 buffer " 1636 "(%dx%d, %s, named %d).\n" 1637 "This is likely a bug in the X Server that will lead to a " 1638 "crash soon.\n", 1639 drawable->w, drawable->h, buffer_name, buffer->name); 1640 return; 1641 } 1642 1643 intel_update_winsys_renderbuffer_miptree(brw, rb, bo, 1644 drawable->w, drawable->h, 1645 buffer->pitch); 1646 1647 if (_mesa_is_front_buffer_drawing(fb) && 1648 (buffer->attachment == __DRI_BUFFER_FRONT_LEFT || 1649 buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) && 1650 rb->Base.Base.NumSamples > 1) { 1651 intel_renderbuffer_upsample(brw, rb); 1652 } 1653 1654 assert(rb->mt); 1655 1656 drm_intel_bo_unreference(bo); 1657} 1658 1659/** 1660 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1661 * 1662 * To determine which DRI buffers to request, examine the renderbuffers 1663 * attached to the drawable's framebuffer. Then request the buffers from 1664 * the image loader 1665 * 1666 * This is called from intel_update_renderbuffers(). 1667 * 1668 * \param drawable Drawable whose buffers are queried. 1669 * \param buffers [out] List of buffers returned by DRI2 query. 1670 * \param buffer_count [out] Number of buffers returned. 1671 * 1672 * \see intel_update_renderbuffers() 1673 */ 1674 1675static void 1676intel_update_image_buffer(struct brw_context *intel, 1677 __DRIdrawable *drawable, 1678 struct intel_renderbuffer *rb, 1679 __DRIimage *buffer, 1680 enum __DRIimageBufferMask buffer_type) 1681{ 1682 struct gl_framebuffer *fb = drawable->driverPrivate; 1683 1684 if (!rb || !buffer->bo) 1685 return; 1686 1687 unsigned num_samples = rb->Base.Base.NumSamples; 1688 1689 /* Check and see if we're already bound to the right 1690 * buffer object 1691 */ 1692 struct intel_mipmap_tree *last_mt; 1693 if (num_samples == 0) 1694 last_mt = rb->mt; 1695 else 1696 last_mt = rb->singlesample_mt; 1697 1698 if (last_mt && last_mt->bo == buffer->bo) 1699 return; 1700 1701 intel_update_winsys_renderbuffer_miptree(intel, rb, buffer->bo, 1702 buffer->width, buffer->height, 1703 buffer->pitch); 1704 1705 if (_mesa_is_front_buffer_drawing(fb) && 1706 buffer_type == __DRI_IMAGE_BUFFER_FRONT && 1707 rb->Base.Base.NumSamples > 1) { 1708 intel_renderbuffer_upsample(intel, rb); 1709 } 1710} 1711 1712static void 1713intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1714{ 1715 struct gl_framebuffer *fb = drawable->driverPrivate; 1716 __DRIscreen *dri_screen = brw->screen->driScrnPriv; 1717 struct intel_renderbuffer *front_rb; 1718 struct intel_renderbuffer *back_rb; 1719 struct __DRIimageList images; 1720 unsigned int format; 1721 uint32_t buffer_mask = 0; 1722 int ret; 1723 1724 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1725 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1726 1727 if (back_rb) 1728 format = intel_rb_format(back_rb); 1729 else if (front_rb) 1730 format = intel_rb_format(front_rb); 1731 else 1732 return; 1733 1734 if (front_rb && (_mesa_is_front_buffer_drawing(fb) || 1735 _mesa_is_front_buffer_reading(fb) || !back_rb)) { 1736 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1737 } 1738 1739 if (back_rb) 1740 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1741 1742 ret = dri_screen->image.loader->getBuffers(drawable, 1743 driGLFormatToImageFormat(format), 1744 &drawable->dri2.stamp, 1745 drawable->loaderPrivate, 1746 buffer_mask, 1747 &images); 1748 if (!ret) 1749 return; 1750 1751 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1752 drawable->w = images.front->width; 1753 drawable->h = images.front->height; 1754 intel_update_image_buffer(brw, 1755 drawable, 1756 front_rb, 1757 images.front, 1758 __DRI_IMAGE_BUFFER_FRONT); 1759 } 1760 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1761 drawable->w = images.back->width; 1762 drawable->h = images.back->height; 1763 intel_update_image_buffer(brw, 1764 drawable, 1765 back_rb, 1766 images.back, 1767 __DRI_IMAGE_BUFFER_BACK); 1768 } 1769} 1770