brw_context.c revision a7bdd4cba8ddcab8dff59ecaaa7efbd436c6c307
1/* 2 Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 3 Copyright (C) Intel Corp. 2006. All Rights Reserved. 4 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) 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 <keith@tungstengraphics.com> 31 */ 32 33 34#include "main/api_exec.h" 35#include "main/context.h" 36#include "main/fbobject.h" 37#include "main/imports.h" 38#include "main/macros.h" 39#include "main/points.h" 40#include "main/version.h" 41#include "main/vtxfmt.h" 42 43#include "vbo/vbo_context.h" 44 45#include "drivers/common/driverfuncs.h" 46#include "drivers/common/meta.h" 47#include "utils.h" 48 49#include "brw_context.h" 50#include "brw_defines.h" 51#include "brw_draw.h" 52#include "brw_state.h" 53 54#include "intel_batchbuffer.h" 55#include "intel_buffer_objects.h" 56#include "intel_buffers.h" 57#include "intel_fbo.h" 58#include "intel_mipmap_tree.h" 59#include "intel_pixel.h" 60#include "intel_regions.h" 61#include "intel_tex.h" 62#include "intel_tex_obj.h" 63 64#include "swrast_setup/swrast_setup.h" 65#include "tnl/tnl.h" 66#include "tnl/t_pipeline.h" 67#include "glsl/ralloc.h" 68 69/*************************************** 70 * Mesa's Driver Functions 71 ***************************************/ 72 73static size_t 74brw_query_samples_for_format(struct gl_context *ctx, GLenum target, 75 GLenum internalFormat, int samples[16]) 76{ 77 struct brw_context *brw = brw_context(ctx); 78 79 (void) target; 80 81 switch (brw->gen) { 82 case 7: 83 samples[0] = 8; 84 samples[1] = 4; 85 return 2; 86 87 case 6: 88 samples[0] = 4; 89 return 1; 90 91 default: 92 samples[0] = 1; 93 return 1; 94 } 95} 96 97const char *const brw_vendor_string = "Intel Open Source Technology Center"; 98 99const char * 100brw_get_renderer_string(unsigned deviceID) 101{ 102 const char *chipset; 103 static char buffer[128]; 104 105 switch (deviceID) { 106#undef CHIPSET 107#define CHIPSET(id, symbol, str) case id: chipset = str; break; 108#include "pci_ids/i965_pci_ids.h" 109 default: 110 chipset = "Unknown Intel Chipset"; 111 break; 112 } 113 114 (void) driGetRendererString(buffer, chipset, 0); 115 return buffer; 116} 117 118static const GLubyte * 119intelGetString(struct gl_context * ctx, GLenum name) 120{ 121 const struct brw_context *const brw = brw_context(ctx); 122 123 switch (name) { 124 case GL_VENDOR: 125 return (GLubyte *) brw_vendor_string; 126 127 case GL_RENDERER: 128 return 129 (GLubyte *) brw_get_renderer_string(brw->intelScreen->deviceID); 130 131 default: 132 return NULL; 133 } 134} 135 136static void 137intel_viewport(struct gl_context *ctx, GLint x, GLint y, GLsizei w, GLsizei h) 138{ 139 struct brw_context *brw = brw_context(ctx); 140 __DRIcontext *driContext = brw->driContext; 141 142 (void) x; 143 (void) y; 144 (void) w; 145 (void) h; 146 147 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) { 148 dri2InvalidateDrawable(driContext->driDrawablePriv); 149 dri2InvalidateDrawable(driContext->driReadablePriv); 150 } 151} 152 153static void 154intelInvalidateState(struct gl_context * ctx, GLuint new_state) 155{ 156 struct brw_context *brw = brw_context(ctx); 157 158 if (ctx->swrast_context) 159 _swrast_InvalidateState(ctx, new_state); 160 _vbo_InvalidateState(ctx, new_state); 161 162 brw->NewGLState |= new_state; 163} 164 165#define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer) 166 167static void 168intel_flush_front(struct gl_context *ctx) 169{ 170 struct brw_context *brw = brw_context(ctx); 171 __DRIcontext *driContext = brw->driContext; 172 __DRIdrawable *driDrawable = driContext->driDrawablePriv; 173 __DRIscreen *const screen = brw->intelScreen->driScrnPriv; 174 175 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) { 176 if (flushFront(screen) && driDrawable && 177 driDrawable->loaderPrivate) { 178 179 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT. 180 * 181 * This potentially resolves both front and back buffer. It 182 * is unnecessary to resolve the back, but harms nothing except 183 * performance. And no one cares about front-buffer render 184 * performance. 185 */ 186 intel_resolve_for_dri2_flush(brw, driDrawable); 187 intel_batchbuffer_flush(brw); 188 189 flushFront(screen)(driDrawable, driDrawable->loaderPrivate); 190 191 /* We set the dirty bit in intel_prepare_render() if we're 192 * front buffer rendering once we get there. 193 */ 194 brw->front_buffer_dirty = false; 195 } 196 } 197} 198 199static void 200intel_glFlush(struct gl_context *ctx) 201{ 202 struct brw_context *brw = brw_context(ctx); 203 204 intel_batchbuffer_flush(brw); 205 intel_flush_front(ctx); 206 if (brw->is_front_buffer_rendering) 207 brw->need_throttle = true; 208} 209 210void 211intelFinish(struct gl_context * ctx) 212{ 213 struct brw_context *brw = brw_context(ctx); 214 215 intel_glFlush(ctx); 216 217 if (brw->batch.last_bo) 218 drm_intel_bo_wait_rendering(brw->batch.last_bo); 219} 220 221static void 222brw_init_driver_functions(struct brw_context *brw, 223 struct dd_function_table *functions) 224{ 225 _mesa_init_driver_functions(functions); 226 227 /* GLX uses DRI2 invalidate events to handle window resizing. 228 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib), 229 * which doesn't provide a mechanism for snooping the event queues. 230 * 231 * So EGL still relies on viewport hacks to handle window resizing. 232 * This should go away with DRI3000. 233 */ 234 if (!brw->driContext->driScreenPriv->dri2.useInvalidate) 235 functions->Viewport = intel_viewport; 236 237 functions->Flush = intel_glFlush; 238 functions->Finish = intelFinish; 239 functions->GetString = intelGetString; 240 functions->UpdateState = intelInvalidateState; 241 242 intelInitTextureFuncs(functions); 243 intelInitTextureImageFuncs(functions); 244 intelInitTextureSubImageFuncs(functions); 245 intelInitTextureCopyImageFuncs(functions); 246 intelInitClearFuncs(functions); 247 intelInitBufferFuncs(functions); 248 intelInitPixelFuncs(functions); 249 intelInitBufferObjectFuncs(functions); 250 intel_init_syncobj_functions(functions); 251 brw_init_object_purgeable_functions(functions); 252 253 brwInitFragProgFuncs( functions ); 254 brw_init_common_queryobj_functions(functions); 255 if (brw->gen >= 6) 256 gen6_init_queryobj_functions(functions); 257 else 258 gen4_init_queryobj_functions(functions); 259 260 functions->QuerySamplesForFormat = brw_query_samples_for_format; 261 262 functions->NewTransformFeedback = brw_new_transform_feedback; 263 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 264 functions->GetTransformFeedbackVertexCount = 265 brw_get_transform_feedback_vertex_count; 266 if (brw->gen >= 7) { 267 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 268 functions->EndTransformFeedback = gen7_end_transform_feedback; 269 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 270 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 271 } else { 272 functions->BeginTransformFeedback = brw_begin_transform_feedback; 273 functions->EndTransformFeedback = brw_end_transform_feedback; 274 } 275 276 if (brw->gen >= 6) 277 functions->GetSamplePosition = gen6_get_sample_position; 278} 279 280static void 281brw_initialize_context_constants(struct brw_context *brw) 282{ 283 struct gl_context *ctx = &brw->ctx; 284 285 ctx->Const.QueryCounterBits.Timestamp = 36; 286 287 ctx->Const.StripTextureBorder = true; 288 289 ctx->Const.MaxDualSourceDrawBuffers = 1; 290 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 291 ctx->Const.FragmentProgram.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 292 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 293 ctx->Const.MaxTextureUnits = 294 MIN2(ctx->Const.MaxTextureCoordUnits, 295 ctx->Const.FragmentProgram.MaxTextureImageUnits); 296 ctx->Const.VertexProgram.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 297 if (brw->gen >= 7) 298 ctx->Const.GeometryProgram.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 299 else 300 ctx->Const.GeometryProgram.MaxTextureImageUnits = 0; 301 ctx->Const.MaxCombinedTextureImageUnits = 302 ctx->Const.VertexProgram.MaxTextureImageUnits + 303 ctx->Const.FragmentProgram.MaxTextureImageUnits + 304 ctx->Const.GeometryProgram.MaxTextureImageUnits; 305 306 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 307 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 308 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 309 ctx->Const.Max3DTextureLevels = 9; 310 ctx->Const.MaxCubeTextureLevels = 12; 311 312 if (brw->gen >= 7) 313 ctx->Const.MaxArrayTextureLayers = 2048; 314 else 315 ctx->Const.MaxArrayTextureLayers = 512; 316 317 ctx->Const.MaxTextureRectSize = 1 << 12; 318 319 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 320 321 ctx->Const.MaxRenderbufferSize = 8192; 322 323 /* Hardware only supports a limited number of transform feedback buffers. 324 * So we need to override the Mesa default (which is based only on software 325 * limits). 326 */ 327 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 328 329 /* On Gen6, in the worst case, we use up one binding table entry per 330 * transform feedback component (see comments above the definition of 331 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 332 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 333 * BRW_MAX_SOL_BINDINGS. 334 * 335 * In "separate components" mode, we need to divide this value by 336 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 337 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 338 */ 339 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 340 ctx->Const.MaxTransformFeedbackSeparateComponents = 341 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 342 343 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 344 345 int max_samples; 346 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 347 const int clamp_max_samples = 348 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 349 350 if (clamp_max_samples < 0) { 351 max_samples = msaa_modes[0]; 352 } else { 353 /* Select the largest supported MSAA mode that does not exceed 354 * clamp_max_samples. 355 */ 356 max_samples = 0; 357 for (int i = 0; msaa_modes[i] != 0; ++i) { 358 if (msaa_modes[i] <= clamp_max_samples) { 359 max_samples = msaa_modes[i]; 360 break; 361 } 362 } 363 } 364 365 ctx->Const.MaxSamples = max_samples; 366 ctx->Const.MaxColorTextureSamples = max_samples; 367 ctx->Const.MaxDepthTextureSamples = max_samples; 368 ctx->Const.MaxIntegerSamples = max_samples; 369 370 if (brw->gen >= 7) 371 ctx->Const.MaxProgramTextureGatherComponents = 4; 372 373 ctx->Const.MinLineWidth = 1.0; 374 ctx->Const.MinLineWidthAA = 1.0; 375 ctx->Const.MaxLineWidth = 5.0; 376 ctx->Const.MaxLineWidthAA = 5.0; 377 ctx->Const.LineWidthGranularity = 0.5; 378 379 ctx->Const.MinPointSize = 1.0; 380 ctx->Const.MinPointSizeAA = 1.0; 381 ctx->Const.MaxPointSize = 255.0; 382 ctx->Const.MaxPointSizeAA = 255.0; 383 ctx->Const.PointSizeGranularity = 1.0; 384 385 if (brw->gen >= 5 || brw->is_g4x) 386 ctx->Const.MaxClipPlanes = 8; 387 388 ctx->Const.VertexProgram.MaxNativeInstructions = 16 * 1024; 389 ctx->Const.VertexProgram.MaxAluInstructions = 0; 390 ctx->Const.VertexProgram.MaxTexInstructions = 0; 391 ctx->Const.VertexProgram.MaxTexIndirections = 0; 392 ctx->Const.VertexProgram.MaxNativeAluInstructions = 0; 393 ctx->Const.VertexProgram.MaxNativeTexInstructions = 0; 394 ctx->Const.VertexProgram.MaxNativeTexIndirections = 0; 395 ctx->Const.VertexProgram.MaxNativeAttribs = 16; 396 ctx->Const.VertexProgram.MaxNativeTemps = 256; 397 ctx->Const.VertexProgram.MaxNativeAddressRegs = 1; 398 ctx->Const.VertexProgram.MaxNativeParameters = 1024; 399 ctx->Const.VertexProgram.MaxEnvParams = 400 MIN2(ctx->Const.VertexProgram.MaxNativeParameters, 401 ctx->Const.VertexProgram.MaxEnvParams); 402 403 ctx->Const.FragmentProgram.MaxNativeInstructions = 1024; 404 ctx->Const.FragmentProgram.MaxNativeAluInstructions = 1024; 405 ctx->Const.FragmentProgram.MaxNativeTexInstructions = 1024; 406 ctx->Const.FragmentProgram.MaxNativeTexIndirections = 1024; 407 ctx->Const.FragmentProgram.MaxNativeAttribs = 12; 408 ctx->Const.FragmentProgram.MaxNativeTemps = 256; 409 ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; 410 ctx->Const.FragmentProgram.MaxNativeParameters = 1024; 411 ctx->Const.FragmentProgram.MaxEnvParams = 412 MIN2(ctx->Const.FragmentProgram.MaxNativeParameters, 413 ctx->Const.FragmentProgram.MaxEnvParams); 414 415 /* Fragment shaders use real, 32-bit twos-complement integers for all 416 * integer types. 417 */ 418 ctx->Const.FragmentProgram.LowInt.RangeMin = 31; 419 ctx->Const.FragmentProgram.LowInt.RangeMax = 30; 420 ctx->Const.FragmentProgram.LowInt.Precision = 0; 421 ctx->Const.FragmentProgram.HighInt = ctx->Const.FragmentProgram.LowInt; 422 ctx->Const.FragmentProgram.MediumInt = ctx->Const.FragmentProgram.LowInt; 423 424 if (brw->gen >= 7) { 425 ctx->Const.FragmentProgram.MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 426 ctx->Const.VertexProgram.MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 427 ctx->Const.GeometryProgram.MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 428 ctx->Const.FragmentProgram.MaxAtomicBuffers = BRW_MAX_ABO; 429 ctx->Const.VertexProgram.MaxAtomicBuffers = BRW_MAX_ABO; 430 ctx->Const.GeometryProgram.MaxAtomicBuffers = BRW_MAX_ABO; 431 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 432 } 433 434 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 435 * but we're not sure how it's actually done for vertex order, 436 * that affect provoking vertex decision. Always use last vertex 437 * convention for quad primitive which works as expected for now. 438 */ 439 if (brw->gen >= 6) 440 ctx->Const.QuadsFollowProvokingVertexConvention = false; 441 442 ctx->Const.NativeIntegers = true; 443 ctx->Const.UniformBooleanTrue = 1; 444 445 /* From the gen4 PRM, volume 4 page 127: 446 * 447 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 448 * the base address of the first element of the surface, computed in 449 * software by adding the surface base address to the byte offset of 450 * the element in the buffer." 451 * 452 * However, unaligned accesses are slower, so enforce buffer alignment. 453 */ 454 ctx->Const.UniformBufferOffsetAlignment = 16; 455 ctx->Const.TextureBufferOffsetAlignment = 16; 456 457 if (brw->gen >= 6) { 458 ctx->Const.MaxVarying = 32; 459 ctx->Const.VertexProgram.MaxOutputComponents = 128; 460 ctx->Const.GeometryProgram.MaxInputComponents = 64; 461 ctx->Const.GeometryProgram.MaxOutputComponents = 128; 462 ctx->Const.FragmentProgram.MaxInputComponents = 128; 463 } 464 465 /* We want the GLSL compiler to emit code that uses condition codes */ 466 for (int i = 0; i < MESA_SHADER_TYPES; i++) { 467 ctx->ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 468 ctx->ShaderCompilerOptions[i].EmitCondCodes = true; 469 ctx->ShaderCompilerOptions[i].EmitNoNoise = true; 470 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true; 471 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true; 472 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true; 473 474 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform = 475 (i == MESA_SHADER_FRAGMENT); 476 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp = 477 (i == MESA_SHADER_FRAGMENT); 478 ctx->ShaderCompilerOptions[i].LowerClipDistance = true; 479 } 480 481 ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].PreferDP4 = true; 482} 483 484/** 485 * Process driconf (drirc) options, setting appropriate context flags. 486 * 487 * intelInitExtensions still pokes at optionCache directly, in order to 488 * avoid advertising various extensions. No flags are set, so it makes 489 * sense to continue doing that there. 490 */ 491static void 492brw_process_driconf_options(struct brw_context *brw) 493{ 494 struct gl_context *ctx = &brw->ctx; 495 496 driOptionCache *options = &brw->optionCache; 497 driParseConfigFiles(options, &brw->intelScreen->optionCache, 498 brw->driContext->driScreenPriv->myNum, "i965"); 499 500 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 501 switch (bo_reuse_mode) { 502 case DRI_CONF_BO_REUSE_DISABLED: 503 break; 504 case DRI_CONF_BO_REUSE_ALL: 505 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 506 break; 507 } 508 509 if (!driQueryOptionb(options, "hiz")) { 510 brw->has_hiz = false; 511 /* On gen6, you can only do separate stencil with HIZ. */ 512 if (brw->gen == 6) 513 brw->has_separate_stencil = false; 514 } 515 516 if (driQueryOptionb(options, "always_flush_batch")) { 517 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 518 brw->always_flush_batch = true; 519 } 520 521 if (driQueryOptionb(options, "always_flush_cache")) { 522 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 523 brw->always_flush_cache = true; 524 } 525 526 if (driQueryOptionb(options, "disable_throttling")) { 527 fprintf(stderr, "disabling flush throttling\n"); 528 brw->disable_throttling = true; 529 } 530 531 brw->disable_derivative_optimization = 532 driQueryOptionb(&brw->optionCache, "disable_derivative_optimization"); 533 534 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 535 536 ctx->Const.ForceGLSLExtensionsWarn = 537 driQueryOptionb(options, "force_glsl_extensions_warn"); 538 539 ctx->Const.DisableGLSLLineContinuations = 540 driQueryOptionb(options, "disable_glsl_line_continuations"); 541} 542 543GLboolean 544brwCreateContext(gl_api api, 545 const struct gl_config *mesaVis, 546 __DRIcontext *driContextPriv, 547 unsigned major_version, 548 unsigned minor_version, 549 uint32_t flags, 550 bool notify_reset, 551 unsigned *dri_ctx_error, 552 void *sharedContextPrivate) 553{ 554 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 555 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 556 struct intel_screen *screen = sPriv->driverPrivate; 557 const struct brw_device_info *devinfo = screen->devinfo; 558 struct dd_function_table functions; 559 struct gl_config visual; 560 561 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 562 * provides us with context reset notifications. 563 */ 564 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 565 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 566 567 if (screen->has_context_reset_notification) 568 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 569 570 if (flags & ~allowed_flags) { 571 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 572 return false; 573 } 574 575 struct brw_context *brw = rzalloc(NULL, struct brw_context); 576 if (!brw) { 577 printf("%s: failed to alloc context\n", __FUNCTION__); 578 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 579 return false; 580 } 581 582 driContextPriv->driverPrivate = brw; 583 brw->driContext = driContextPriv; 584 brw->intelScreen = screen; 585 brw->bufmgr = screen->bufmgr; 586 587 brw->gen = devinfo->gen; 588 brw->gt = devinfo->gt; 589 brw->is_g4x = devinfo->is_g4x; 590 brw->is_baytrail = devinfo->is_baytrail; 591 brw->is_haswell = devinfo->is_haswell; 592 brw->has_llc = devinfo->has_llc; 593 brw->has_hiz = devinfo->has_hiz_and_separate_stencil && brw->gen < 8; 594 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 595 brw->has_pln = devinfo->has_pln; 596 brw->has_compr4 = devinfo->has_compr4; 597 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 598 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 599 brw->needs_unlit_centroid_workaround = 600 devinfo->needs_unlit_centroid_workaround; 601 602 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 603 brw->has_swizzling = screen->hw_has_swizzling; 604 605 if (brw->gen >= 7) { 606 gen7_init_vtable_surface_functions(brw); 607 gen7_init_vtable_sampler_functions(brw); 608 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 609 } else { 610 gen4_init_vtable_surface_functions(brw); 611 gen4_init_vtable_sampler_functions(brw); 612 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 613 } 614 615 brw_init_driver_functions(brw, &functions); 616 617 if (notify_reset) 618 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 619 620 struct gl_context *ctx = &brw->ctx; 621 622 if (mesaVis == NULL) { 623 memset(&visual, 0, sizeof visual); 624 mesaVis = &visual; 625 } 626 627 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 628 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 629 printf("%s: failed to init mesa context\n", __FUNCTION__); 630 intelDestroyContext(driContextPriv); 631 return false; 632 } 633 634 /* Initialize the software rasterizer and helper modules. 635 * 636 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 637 * software fallbacks (which we have to support on legacy GL to do weird 638 * glDrawPixels(), glBitmap(), and other functions). 639 */ 640 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 641 _swrast_CreateContext(ctx); 642 } 643 644 _vbo_CreateContext(ctx); 645 if (ctx->swrast_context) { 646 _tnl_CreateContext(ctx); 647 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 648 _swsetup_CreateContext(ctx); 649 650 /* Configure swrast to match hardware characteristics: */ 651 _swrast_allow_pixel_fog(ctx, false); 652 _swrast_allow_vertex_fog(ctx, true); 653 } 654 655 _mesa_meta_init(ctx); 656 657 brw_process_driconf_options(brw); 658 brw_process_intel_debug_variable(brw); 659 brw_initialize_context_constants(brw); 660 661 ctx->Const.ResetStrategy = notify_reset 662 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 663 664 /* Reinitialize the context point state. It depends on ctx->Const values. */ 665 _mesa_init_point(ctx); 666 667 intel_batchbuffer_init(brw); 668 669 brw_init_state(brw); 670 671 intelInitExtensions(ctx); 672 673 intel_fbo_init(brw); 674 675 if (brw->gen >= 6) { 676 /* Create a new hardware context. Using a hardware context means that 677 * our GPU state will be saved/restored on context switch, allowing us 678 * to assume that the GPU is in the same state we left it in. 679 * 680 * This is required for transform feedback buffer offsets, query objects, 681 * and also allows us to reduce how much state we have to emit. 682 */ 683 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 684 685 if (!brw->hw_ctx) { 686 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 687 intelDestroyContext(driContextPriv); 688 return false; 689 } 690 } 691 692 brw_init_surface_formats(brw); 693 694 if (brw->is_g4x || brw->gen >= 5) { 695 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS; 696 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; 697 } else { 698 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS; 699 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; 700 } 701 702 brw->max_vs_threads = devinfo->max_vs_threads; 703 brw->max_gs_threads = devinfo->max_gs_threads; 704 brw->max_wm_threads = devinfo->max_wm_threads; 705 brw->urb.size = devinfo->urb.size; 706 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 707 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 708 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 709 710 /* Estimate the size of the mappable aperture into the GTT. There's an 711 * ioctl to get the whole GTT size, but not one to get the mappable subset. 712 * It turns out it's basically always 256MB, though some ancient hardware 713 * was smaller. 714 */ 715 uint32_t gtt_size = 256 * 1024 * 1024; 716 717 /* We don't want to map two objects such that a memcpy between them would 718 * just fault one mapping in and then the other over and over forever. So 719 * we would need to divide the GTT size by 2. Additionally, some GTT is 720 * taken up by things like the framebuffer and the ringbuffer and such, so 721 * be more conservative. 722 */ 723 brw->max_gtt_map_object_size = gtt_size / 4; 724 725 if (brw->gen == 6) 726 brw->urb.gen6_gs_previously_active = false; 727 728 brw->prim_restart.in_progress = false; 729 brw->prim_restart.enable_cut_index = false; 730 brw->gs.enabled = false; 731 732 if (brw->gen < 6) { 733 brw->curbe.last_buf = calloc(1, 4096); 734 brw->curbe.next_buf = calloc(1, 4096); 735 } 736 737 ctx->VertexProgram._MaintainTnlProgram = true; 738 ctx->FragmentProgram._MaintainTexEnvProgram = true; 739 740 brw_draw_init( brw ); 741 742 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 743 /* Turn on some extra GL_ARB_debug_output generation. */ 744 brw->perf_debug = true; 745 } 746 747 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 748 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 749 750 brw_fs_alloc_reg_sets(brw); 751 brw_vec4_alloc_reg_set(brw); 752 753 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 754 brw_init_shader_time(brw); 755 756 _mesa_compute_version(ctx); 757 758 _mesa_initialize_dispatch_tables(ctx); 759 _mesa_initialize_vbo_vtxfmt(ctx); 760 761 if (ctx->Extensions.AMD_performance_monitor) { 762 brw_init_performance_monitors(brw); 763 } 764 765 return true; 766} 767 768void 769intelDestroyContext(__DRIcontext * driContextPriv) 770{ 771 struct brw_context *brw = 772 (struct brw_context *) driContextPriv->driverPrivate; 773 struct gl_context *ctx = &brw->ctx; 774 775 assert(brw); /* should never be null */ 776 if (!brw) 777 return; 778 779 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 780 if (INTEL_DEBUG & DEBUG_AUB) { 781 intel_batchbuffer_flush(brw); 782 aub_dump_bmp(&brw->ctx); 783 } 784 785 _mesa_meta_free(&brw->ctx); 786 787 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 788 /* Force a report. */ 789 brw->shader_time.report_time = 0; 790 791 brw_collect_and_report_shader_time(brw); 792 brw_destroy_shader_time(brw); 793 } 794 795 brw_destroy_state(brw); 796 brw_draw_destroy(brw); 797 798 drm_intel_bo_unreference(brw->curbe.curbe_bo); 799 drm_intel_bo_unreference(brw->vs.base.const_bo); 800 drm_intel_bo_unreference(brw->wm.base.const_bo); 801 802 free(brw->curbe.last_buf); 803 free(brw->curbe.next_buf); 804 805 drm_intel_gem_context_destroy(brw->hw_ctx); 806 807 if (ctx->swrast_context) { 808 _swsetup_DestroyContext(&brw->ctx); 809 _tnl_DestroyContext(&brw->ctx); 810 } 811 _vbo_DestroyContext(&brw->ctx); 812 813 if (ctx->swrast_context) 814 _swrast_DestroyContext(&brw->ctx); 815 816 intel_batchbuffer_free(brw); 817 818 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 819 brw->first_post_swapbuffers_batch = NULL; 820 821 driDestroyOptionCache(&brw->optionCache); 822 823 /* free the Mesa context */ 824 _mesa_free_context_data(&brw->ctx); 825 826 ralloc_free(brw); 827 driContextPriv->driverPrivate = NULL; 828} 829 830GLboolean 831intelUnbindContext(__DRIcontext * driContextPriv) 832{ 833 /* Unset current context and dispath table */ 834 _mesa_make_current(NULL, NULL, NULL); 835 836 return true; 837} 838 839/** 840 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 841 * on window system framebuffers. 842 * 843 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 844 * your renderbuffer can do sRGB encode, and you can flip a switch that does 845 * sRGB encode if the renderbuffer can handle it. You can ask specifically 846 * for a visual where you're guaranteed to be capable, but it turns out that 847 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 848 * incapable ones, becuase there's no difference between the two in resources 849 * used. Applications thus get built that accidentally rely on the default 850 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 851 * great... 852 * 853 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 854 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 855 * So they removed the enable knob and made it "if the renderbuffer is sRGB 856 * capable, do sRGB encode". Then, for your window system renderbuffers, you 857 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 858 * and get no sRGB encode (assuming that both kinds of visual are available). 859 * Thus our choice to support sRGB by default on our visuals for desktop would 860 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 861 * 862 * Unfortunately, renderbuffer setup happens before a context is created. So 863 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 864 * context (without an sRGB visual, though we don't have sRGB visuals exposed 865 * yet), we go turn that back off before anyone finds out. 866 */ 867static void 868intel_gles3_srgb_workaround(struct brw_context *brw, 869 struct gl_framebuffer *fb) 870{ 871 struct gl_context *ctx = &brw->ctx; 872 873 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 874 return; 875 876 /* Some day when we support the sRGB capable bit on visuals available for 877 * GLES, we'll need to respect that and not disable things here. 878 */ 879 fb->Visual.sRGBCapable = false; 880 for (int i = 0; i < BUFFER_COUNT; i++) { 881 if (fb->Attachment[i].Renderbuffer && 882 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_SARGB8) { 883 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_ARGB8888; 884 } 885 } 886} 887 888GLboolean 889intelMakeCurrent(__DRIcontext * driContextPriv, 890 __DRIdrawable * driDrawPriv, 891 __DRIdrawable * driReadPriv) 892{ 893 struct brw_context *brw; 894 GET_CURRENT_CONTEXT(curCtx); 895 896 if (driContextPriv) 897 brw = (struct brw_context *) driContextPriv->driverPrivate; 898 else 899 brw = NULL; 900 901 /* According to the glXMakeCurrent() man page: "Pending commands to 902 * the previous context, if any, are flushed before it is released." 903 * But only flush if we're actually changing contexts. 904 */ 905 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 906 _mesa_flush(curCtx); 907 } 908 909 if (driContextPriv) { 910 struct gl_context *ctx = &brw->ctx; 911 struct gl_framebuffer *fb, *readFb; 912 913 if (driDrawPriv == NULL && driReadPriv == NULL) { 914 fb = _mesa_get_incomplete_framebuffer(); 915 readFb = _mesa_get_incomplete_framebuffer(); 916 } else { 917 fb = driDrawPriv->driverPrivate; 918 readFb = driReadPriv->driverPrivate; 919 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 920 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 921 } 922 923 /* The sRGB workaround changes the renderbuffer's format. We must change 924 * the format before the renderbuffer's miptree get's allocated, otherwise 925 * the formats of the renderbuffer and its miptree will differ. 926 */ 927 intel_gles3_srgb_workaround(brw, fb); 928 intel_gles3_srgb_workaround(brw, readFb); 929 930 intel_prepare_render(brw); 931 _mesa_make_current(ctx, fb, readFb); 932 } else { 933 _mesa_make_current(NULL, NULL, NULL); 934 } 935 936 return true; 937} 938 939void 940intel_resolve_for_dri2_flush(struct brw_context *brw, 941 __DRIdrawable *drawable) 942{ 943 if (brw->gen < 6) { 944 /* MSAA and fast color clear are not supported, so don't waste time 945 * checking whether a resolve is needed. 946 */ 947 return; 948 } 949 950 struct gl_framebuffer *fb = drawable->driverPrivate; 951 struct intel_renderbuffer *rb; 952 953 /* Usually, only the back buffer will need to be downsampled. However, 954 * the front buffer will also need it if the user has rendered into it. 955 */ 956 static const gl_buffer_index buffers[2] = { 957 BUFFER_BACK_LEFT, 958 BUFFER_FRONT_LEFT, 959 }; 960 961 for (int i = 0; i < 2; ++i) { 962 rb = intel_get_renderbuffer(fb, buffers[i]); 963 if (rb == NULL || rb->mt == NULL) 964 continue; 965 if (rb->mt->num_samples <= 1) 966 intel_miptree_resolve_color(brw, rb->mt); 967 else 968 intel_miptree_downsample(brw, rb->mt); 969 } 970} 971 972static unsigned 973intel_bits_per_pixel(const struct intel_renderbuffer *rb) 974{ 975 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 976} 977 978static void 979intel_query_dri2_buffers(struct brw_context *brw, 980 __DRIdrawable *drawable, 981 __DRIbuffer **buffers, 982 int *count); 983 984static void 985intel_process_dri2_buffer(struct brw_context *brw, 986 __DRIdrawable *drawable, 987 __DRIbuffer *buffer, 988 struct intel_renderbuffer *rb, 989 const char *buffer_name); 990 991static void 992intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 993 994static void 995intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 996{ 997 struct gl_framebuffer *fb = drawable->driverPrivate; 998 struct intel_renderbuffer *rb; 999 __DRIbuffer *buffers = NULL; 1000 int i, count; 1001 const char *region_name; 1002 1003 /* Set this up front, so that in case our buffers get invalidated 1004 * while we're getting new buffers, we don't clobber the stamp and 1005 * thus ignore the invalidate. */ 1006 drawable->lastStamp = drawable->dri2.stamp; 1007 1008 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1009 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1010 1011 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1012 1013 if (buffers == NULL) 1014 return; 1015 1016 for (i = 0; i < count; i++) { 1017 switch (buffers[i].attachment) { 1018 case __DRI_BUFFER_FRONT_LEFT: 1019 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1020 region_name = "dri2 front buffer"; 1021 break; 1022 1023 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1024 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1025 region_name = "dri2 fake front buffer"; 1026 break; 1027 1028 case __DRI_BUFFER_BACK_LEFT: 1029 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1030 region_name = "dri2 back buffer"; 1031 break; 1032 1033 case __DRI_BUFFER_DEPTH: 1034 case __DRI_BUFFER_HIZ: 1035 case __DRI_BUFFER_DEPTH_STENCIL: 1036 case __DRI_BUFFER_STENCIL: 1037 case __DRI_BUFFER_ACCUM: 1038 default: 1039 fprintf(stderr, 1040 "unhandled buffer attach event, attachment type %d\n", 1041 buffers[i].attachment); 1042 return; 1043 } 1044 1045 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1046 } 1047 1048} 1049 1050void 1051intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1052{ 1053 struct brw_context *brw = context->driverPrivate; 1054 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1055 1056 /* Set this up front, so that in case our buffers get invalidated 1057 * while we're getting new buffers, we don't clobber the stamp and 1058 * thus ignore the invalidate. */ 1059 drawable->lastStamp = drawable->dri2.stamp; 1060 1061 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1062 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1063 1064 if (screen->image.loader) 1065 intel_update_image_buffers(brw, drawable); 1066 else 1067 intel_update_dri2_buffers(brw, drawable); 1068 1069 driUpdateFramebufferSize(&brw->ctx, drawable); 1070} 1071 1072/** 1073 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1074 * state is required. 1075 */ 1076void 1077intel_prepare_render(struct brw_context *brw) 1078{ 1079 __DRIcontext *driContext = brw->driContext; 1080 __DRIdrawable *drawable; 1081 1082 drawable = driContext->driDrawablePriv; 1083 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1084 if (drawable->lastStamp != drawable->dri2.stamp) 1085 intel_update_renderbuffers(driContext, drawable); 1086 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1087 } 1088 1089 drawable = driContext->driReadablePriv; 1090 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1091 if (drawable->lastStamp != drawable->dri2.stamp) 1092 intel_update_renderbuffers(driContext, drawable); 1093 driContext->dri2.read_stamp = drawable->dri2.stamp; 1094 } 1095 1096 /* If we're currently rendering to the front buffer, the rendering 1097 * that will happen next will probably dirty the front buffer. So 1098 * mark it as dirty here. 1099 */ 1100 if (brw->is_front_buffer_rendering) 1101 brw->front_buffer_dirty = true; 1102 1103 /* Wait for the swapbuffers before the one we just emitted, so we 1104 * don't get too many swaps outstanding for apps that are GPU-heavy 1105 * but not CPU-heavy. 1106 * 1107 * We're using intelDRI2Flush (called from the loader before 1108 * swapbuffer) and glFlush (for front buffer rendering) as the 1109 * indicator that a frame is done and then throttle when we get 1110 * here as we prepare to render the next frame. At this point for 1111 * round trips for swap/copy and getting new buffers are done and 1112 * we'll spend less time waiting on the GPU. 1113 * 1114 * Unfortunately, we don't have a handle to the batch containing 1115 * the swap, and getting our hands on that doesn't seem worth it, 1116 * so we just us the first batch we emitted after the last swap. 1117 */ 1118 if (brw->need_throttle && brw->first_post_swapbuffers_batch) { 1119 if (!brw->disable_throttling) 1120 drm_intel_bo_wait_rendering(brw->first_post_swapbuffers_batch); 1121 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 1122 brw->first_post_swapbuffers_batch = NULL; 1123 brw->need_throttle = false; 1124 } 1125} 1126 1127/** 1128 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1129 * 1130 * To determine which DRI buffers to request, examine the renderbuffers 1131 * attached to the drawable's framebuffer. Then request the buffers with 1132 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1133 * 1134 * This is called from intel_update_renderbuffers(). 1135 * 1136 * \param drawable Drawable whose buffers are queried. 1137 * \param buffers [out] List of buffers returned by DRI2 query. 1138 * \param buffer_count [out] Number of buffers returned. 1139 * 1140 * \see intel_update_renderbuffers() 1141 * \see DRI2GetBuffers() 1142 * \see DRI2GetBuffersWithFormat() 1143 */ 1144static void 1145intel_query_dri2_buffers(struct brw_context *brw, 1146 __DRIdrawable *drawable, 1147 __DRIbuffer **buffers, 1148 int *buffer_count) 1149{ 1150 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1151 struct gl_framebuffer *fb = drawable->driverPrivate; 1152 int i = 0; 1153 unsigned attachments[8]; 1154 1155 struct intel_renderbuffer *front_rb; 1156 struct intel_renderbuffer *back_rb; 1157 1158 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1159 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1160 1161 memset(attachments, 0, sizeof(attachments)); 1162 if ((brw->is_front_buffer_rendering || 1163 brw->is_front_buffer_reading || 1164 !back_rb) && front_rb) { 1165 /* If a fake front buffer is in use, then querying for 1166 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1167 * the real front buffer to the fake front buffer. So before doing the 1168 * query, we need to make sure all the pending drawing has landed in the 1169 * real front buffer. 1170 */ 1171 intel_batchbuffer_flush(brw); 1172 intel_flush_front(&brw->ctx); 1173 1174 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1175 attachments[i++] = intel_bits_per_pixel(front_rb); 1176 } else if (front_rb && brw->front_buffer_dirty) { 1177 /* We have pending front buffer rendering, but we aren't querying for a 1178 * front buffer. If the front buffer we have is a fake front buffer, 1179 * the X server is going to throw it away when it processes the query. 1180 * So before doing the query, make sure all the pending drawing has 1181 * landed in the real front buffer. 1182 */ 1183 intel_batchbuffer_flush(brw); 1184 intel_flush_front(&brw->ctx); 1185 } 1186 1187 if (back_rb) { 1188 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1189 attachments[i++] = intel_bits_per_pixel(back_rb); 1190 } 1191 1192 assert(i <= ARRAY_SIZE(attachments)); 1193 1194 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1195 &drawable->w, 1196 &drawable->h, 1197 attachments, i / 2, 1198 buffer_count, 1199 drawable->loaderPrivate); 1200} 1201 1202/** 1203 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1204 * 1205 * This is called from intel_update_renderbuffers(). 1206 * 1207 * \par Note: 1208 * DRI buffers whose attachment point is DRI2BufferStencil or 1209 * DRI2BufferDepthStencil are handled as special cases. 1210 * 1211 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1212 * that is passed to intel_region_alloc_for_handle(). 1213 * 1214 * \see intel_update_renderbuffers() 1215 * \see intel_region_alloc_for_handle() 1216 */ 1217static void 1218intel_process_dri2_buffer(struct brw_context *brw, 1219 __DRIdrawable *drawable, 1220 __DRIbuffer *buffer, 1221 struct intel_renderbuffer *rb, 1222 const char *buffer_name) 1223{ 1224 struct intel_region *region = NULL; 1225 1226 if (!rb) 1227 return; 1228 1229 unsigned num_samples = rb->Base.Base.NumSamples; 1230 1231 /* We try to avoid closing and reopening the same BO name, because the first 1232 * use of a mapping of the buffer involves a bunch of page faulting which is 1233 * moderately expensive. 1234 */ 1235 if (num_samples == 0) { 1236 if (rb->mt && 1237 rb->mt->region && 1238 rb->mt->region->name == buffer->name) 1239 return; 1240 } else { 1241 if (rb->mt && 1242 rb->mt->singlesample_mt && 1243 rb->mt->singlesample_mt->region && 1244 rb->mt->singlesample_mt->region->name == buffer->name) 1245 return; 1246 } 1247 1248 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1249 fprintf(stderr, 1250 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1251 buffer->name, buffer->attachment, 1252 buffer->cpp, buffer->pitch); 1253 } 1254 1255 intel_miptree_release(&rb->mt); 1256 region = intel_region_alloc_for_handle(brw->intelScreen, 1257 buffer->cpp, 1258 drawable->w, 1259 drawable->h, 1260 buffer->pitch, 1261 buffer->name, 1262 buffer_name); 1263 if (!region) 1264 return; 1265 1266 rb->mt = intel_miptree_create_for_dri2_buffer(brw, 1267 buffer->attachment, 1268 intel_rb_format(rb), 1269 num_samples, 1270 region); 1271 intel_region_release(®ion); 1272} 1273 1274/** 1275 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1276 * 1277 * To determine which DRI buffers to request, examine the renderbuffers 1278 * attached to the drawable's framebuffer. Then request the buffers from 1279 * the image loader 1280 * 1281 * This is called from intel_update_renderbuffers(). 1282 * 1283 * \param drawable Drawable whose buffers are queried. 1284 * \param buffers [out] List of buffers returned by DRI2 query. 1285 * \param buffer_count [out] Number of buffers returned. 1286 * 1287 * \see intel_update_renderbuffers() 1288 */ 1289 1290static void 1291intel_update_image_buffer(struct brw_context *intel, 1292 __DRIdrawable *drawable, 1293 struct intel_renderbuffer *rb, 1294 __DRIimage *buffer, 1295 enum __DRIimageBufferMask buffer_type) 1296{ 1297 struct intel_region *region = buffer->region; 1298 1299 if (!rb || !region) 1300 return; 1301 1302 unsigned num_samples = rb->Base.Base.NumSamples; 1303 1304 if (rb->mt && 1305 rb->mt->region && 1306 rb->mt->region == region) 1307 return; 1308 1309 intel_miptree_release(&rb->mt); 1310 rb->mt = intel_miptree_create_for_image_buffer(intel, 1311 buffer_type, 1312 intel_rb_format(rb), 1313 num_samples, 1314 region); 1315} 1316 1317static void 1318intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1319{ 1320 struct gl_framebuffer *fb = drawable->driverPrivate; 1321 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1322 struct intel_renderbuffer *front_rb; 1323 struct intel_renderbuffer *back_rb; 1324 struct __DRIimageList images; 1325 unsigned int format; 1326 uint32_t buffer_mask = 0; 1327 1328 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1329 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1330 1331 if (back_rb) 1332 format = intel_rb_format(back_rb); 1333 else if (front_rb) 1334 format = intel_rb_format(front_rb); 1335 else 1336 return; 1337 1338 if ((brw->is_front_buffer_rendering || brw->is_front_buffer_reading || !back_rb) && front_rb) 1339 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1340 1341 if (back_rb) 1342 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1343 1344 (*screen->image.loader->getBuffers) (drawable, 1345 driGLFormatToImageFormat(format), 1346 &drawable->dri2.stamp, 1347 drawable->loaderPrivate, 1348 buffer_mask, 1349 &images); 1350 1351 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1352 drawable->w = images.front->width; 1353 drawable->h = images.front->height; 1354 intel_update_image_buffer(brw, 1355 drawable, 1356 front_rb, 1357 images.front, 1358 __DRI_IMAGE_BUFFER_FRONT); 1359 } 1360 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1361 drawable->w = images.back->width; 1362 drawable->h = images.back->height; 1363 intel_update_image_buffer(brw, 1364 drawable, 1365 back_rb, 1366 images.back, 1367 __DRI_IMAGE_BUFFER_BACK); 1368 } 1369} 1370