brw_context.c revision eeed49f5f290793870c60b5b635b977a732a1eb4
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/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) 138{ 139 struct brw_context *brw = brw_context(ctx); 140 __DRIcontext *driContext = brw->driContext; 141 142 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) { 143 dri2InvalidateDrawable(driContext->driDrawablePriv); 144 dri2InvalidateDrawable(driContext->driReadablePriv); 145 } 146} 147 148static void 149intelInvalidateState(struct gl_context * ctx, GLuint new_state) 150{ 151 struct brw_context *brw = brw_context(ctx); 152 153 if (ctx->swrast_context) 154 _swrast_InvalidateState(ctx, new_state); 155 _vbo_InvalidateState(ctx, new_state); 156 157 brw->NewGLState |= new_state; 158} 159 160#define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer) 161 162static void 163intel_flush_front(struct gl_context *ctx) 164{ 165 struct brw_context *brw = brw_context(ctx); 166 __DRIcontext *driContext = brw->driContext; 167 __DRIdrawable *driDrawable = driContext->driDrawablePriv; 168 __DRIscreen *const screen = brw->intelScreen->driScrnPriv; 169 170 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) { 171 if (flushFront(screen) && driDrawable && 172 driDrawable->loaderPrivate) { 173 174 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT. 175 * 176 * This potentially resolves both front and back buffer. It 177 * is unnecessary to resolve the back, but harms nothing except 178 * performance. And no one cares about front-buffer render 179 * performance. 180 */ 181 intel_resolve_for_dri2_flush(brw, driDrawable); 182 intel_batchbuffer_flush(brw); 183 184 flushFront(screen)(driDrawable, driDrawable->loaderPrivate); 185 186 /* We set the dirty bit in intel_prepare_render() if we're 187 * front buffer rendering once we get there. 188 */ 189 brw->front_buffer_dirty = false; 190 } 191 } 192} 193 194static void 195intel_glFlush(struct gl_context *ctx) 196{ 197 struct brw_context *brw = brw_context(ctx); 198 199 intel_batchbuffer_flush(brw); 200 intel_flush_front(ctx); 201 if (brw->is_front_buffer_rendering) 202 brw->need_throttle = true; 203} 204 205void 206intelFinish(struct gl_context * ctx) 207{ 208 struct brw_context *brw = brw_context(ctx); 209 210 intel_glFlush(ctx); 211 212 if (brw->batch.last_bo) 213 drm_intel_bo_wait_rendering(brw->batch.last_bo); 214} 215 216static void 217brw_init_driver_functions(struct brw_context *brw, 218 struct dd_function_table *functions) 219{ 220 _mesa_init_driver_functions(functions); 221 222 /* GLX uses DRI2 invalidate events to handle window resizing. 223 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib), 224 * which doesn't provide a mechanism for snooping the event queues. 225 * 226 * So EGL still relies on viewport hacks to handle window resizing. 227 * This should go away with DRI3000. 228 */ 229 if (!brw->driContext->driScreenPriv->dri2.useInvalidate) 230 functions->Viewport = intel_viewport; 231 232 functions->Flush = intel_glFlush; 233 functions->Finish = intelFinish; 234 functions->GetString = intelGetString; 235 functions->UpdateState = intelInvalidateState; 236 237 intelInitTextureFuncs(functions); 238 intelInitTextureImageFuncs(functions); 239 intelInitTextureSubImageFuncs(functions); 240 intelInitTextureCopyImageFuncs(functions); 241 intelInitClearFuncs(functions); 242 intelInitBufferFuncs(functions); 243 intelInitPixelFuncs(functions); 244 intelInitBufferObjectFuncs(functions); 245 intel_init_syncobj_functions(functions); 246 brw_init_object_purgeable_functions(functions); 247 248 brwInitFragProgFuncs( functions ); 249 brw_init_common_queryobj_functions(functions); 250 if (brw->gen >= 6) 251 gen6_init_queryobj_functions(functions); 252 else 253 gen4_init_queryobj_functions(functions); 254 255 functions->QuerySamplesForFormat = brw_query_samples_for_format; 256 257 functions->NewTransformFeedback = brw_new_transform_feedback; 258 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 259 functions->GetTransformFeedbackVertexCount = 260 brw_get_transform_feedback_vertex_count; 261 if (brw->gen >= 7) { 262 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 263 functions->EndTransformFeedback = gen7_end_transform_feedback; 264 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 265 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 266 } else { 267 functions->BeginTransformFeedback = brw_begin_transform_feedback; 268 functions->EndTransformFeedback = brw_end_transform_feedback; 269 } 270 271 if (brw->gen >= 6) 272 functions->GetSamplePosition = gen6_get_sample_position; 273} 274 275static void 276brw_initialize_context_constants(struct brw_context *brw) 277{ 278 struct gl_context *ctx = &brw->ctx; 279 280 unsigned max_samplers = 281 brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16; 282 283 ctx->Const.QueryCounterBits.Timestamp = 36; 284 285 ctx->Const.StripTextureBorder = true; 286 287 ctx->Const.MaxDualSourceDrawBuffers = 1; 288 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 289 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits = max_samplers; 290 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 291 ctx->Const.MaxTextureUnits = 292 MIN2(ctx->Const.MaxTextureCoordUnits, 293 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits); 294 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits = max_samplers; 295 if (brw->gen >= 7) 296 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = max_samplers; 297 else 298 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = 0; 299 ctx->Const.MaxCombinedTextureImageUnits = 300 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits + 301 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits + 302 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits; 303 304 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 305 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 306 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 307 ctx->Const.Max3DTextureLevels = 9; 308 ctx->Const.MaxCubeTextureLevels = 12; 309 310 if (brw->gen >= 7) 311 ctx->Const.MaxArrayTextureLayers = 2048; 312 else 313 ctx->Const.MaxArrayTextureLayers = 512; 314 315 ctx->Const.MaxTextureRectSize = 1 << 12; 316 317 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 318 319 ctx->Const.MaxRenderbufferSize = 8192; 320 321 /* Hardware only supports a limited number of transform feedback buffers. 322 * So we need to override the Mesa default (which is based only on software 323 * limits). 324 */ 325 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 326 327 /* On Gen6, in the worst case, we use up one binding table entry per 328 * transform feedback component (see comments above the definition of 329 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 330 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 331 * BRW_MAX_SOL_BINDINGS. 332 * 333 * In "separate components" mode, we need to divide this value by 334 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 335 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 336 */ 337 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 338 ctx->Const.MaxTransformFeedbackSeparateComponents = 339 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 340 341 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 342 343 int max_samples; 344 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 345 const int clamp_max_samples = 346 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 347 348 if (clamp_max_samples < 0) { 349 max_samples = msaa_modes[0]; 350 } else { 351 /* Select the largest supported MSAA mode that does not exceed 352 * clamp_max_samples. 353 */ 354 max_samples = 0; 355 for (int i = 0; msaa_modes[i] != 0; ++i) { 356 if (msaa_modes[i] <= clamp_max_samples) { 357 max_samples = msaa_modes[i]; 358 break; 359 } 360 } 361 } 362 363 ctx->Const.MaxSamples = max_samples; 364 ctx->Const.MaxColorTextureSamples = max_samples; 365 ctx->Const.MaxDepthTextureSamples = max_samples; 366 ctx->Const.MaxIntegerSamples = max_samples; 367 368 if (brw->gen >= 7) 369 ctx->Const.MaxProgramTextureGatherComponents = 4; 370 371 ctx->Const.MinLineWidth = 1.0; 372 ctx->Const.MinLineWidthAA = 1.0; 373 ctx->Const.MaxLineWidth = 5.0; 374 ctx->Const.MaxLineWidthAA = 5.0; 375 ctx->Const.LineWidthGranularity = 0.5; 376 377 ctx->Const.MinPointSize = 1.0; 378 ctx->Const.MinPointSizeAA = 1.0; 379 ctx->Const.MaxPointSize = 255.0; 380 ctx->Const.MaxPointSizeAA = 255.0; 381 ctx->Const.PointSizeGranularity = 1.0; 382 383 if (brw->gen >= 5 || brw->is_g4x) 384 ctx->Const.MaxClipPlanes = 8; 385 386 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024; 387 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0; 388 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0; 389 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0; 390 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0; 391 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0; 392 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0; 393 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16; 394 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256; 395 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1; 396 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024; 397 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams = 398 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters, 399 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams); 400 401 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024; 402 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024; 403 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024; 404 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024; 405 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12; 406 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256; 407 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0; 408 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024; 409 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams = 410 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters, 411 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams); 412 413 /* Fragment shaders use real, 32-bit twos-complement integers for all 414 * integer types. 415 */ 416 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31; 417 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30; 418 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0; 419 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 420 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 421 422 if (brw->gen >= 7) { 423 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 424 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 425 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 426 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicBuffers = BRW_MAX_ABO; 427 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicBuffers = BRW_MAX_ABO; 428 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicBuffers = BRW_MAX_ABO; 429 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 430 } 431 432 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 433 * but we're not sure how it's actually done for vertex order, 434 * that affect provoking vertex decision. Always use last vertex 435 * convention for quad primitive which works as expected for now. 436 */ 437 if (brw->gen >= 6) 438 ctx->Const.QuadsFollowProvokingVertexConvention = false; 439 440 ctx->Const.NativeIntegers = true; 441 ctx->Const.UniformBooleanTrue = 1; 442 443 /* From the gen4 PRM, volume 4 page 127: 444 * 445 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 446 * the base address of the first element of the surface, computed in 447 * software by adding the surface base address to the byte offset of 448 * the element in the buffer." 449 * 450 * However, unaligned accesses are slower, so enforce buffer alignment. 451 */ 452 ctx->Const.UniformBufferOffsetAlignment = 16; 453 ctx->Const.TextureBufferOffsetAlignment = 16; 454 455 if (brw->gen >= 6) { 456 ctx->Const.MaxVarying = 32; 457 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128; 458 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64; 459 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128; 460 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128; 461 } 462 463 /* We want the GLSL compiler to emit code that uses condition codes */ 464 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 465 ctx->ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 466 ctx->ShaderCompilerOptions[i].EmitCondCodes = true; 467 ctx->ShaderCompilerOptions[i].EmitNoNoise = true; 468 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true; 469 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true; 470 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true; 471 472 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform = 473 (i == MESA_SHADER_FRAGMENT); 474 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp = 475 (i == MESA_SHADER_FRAGMENT); 476 ctx->ShaderCompilerOptions[i].LowerClipDistance = true; 477 } 478 479 ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = true; 480 ctx->ShaderCompilerOptions[MESA_SHADER_GEOMETRY].OptimizeForAOS = true; 481 482 /* ARB_viewport_array */ 483 if (brw->gen >= 7 && ctx->API == API_OPENGL_CORE) { 484 ctx->Const.MaxViewports = GEN7_NUM_VIEWPORTS; 485 ctx->Const.ViewportSubpixelBits = 0; 486 487 /* Cast to float before negating becuase MaxViewportWidth is unsigned. 488 */ 489 ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth; 490 ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth; 491 } 492} 493 494/** 495 * Process driconf (drirc) options, setting appropriate context flags. 496 * 497 * intelInitExtensions still pokes at optionCache directly, in order to 498 * avoid advertising various extensions. No flags are set, so it makes 499 * sense to continue doing that there. 500 */ 501static void 502brw_process_driconf_options(struct brw_context *brw) 503{ 504 struct gl_context *ctx = &brw->ctx; 505 506 driOptionCache *options = &brw->optionCache; 507 driParseConfigFiles(options, &brw->intelScreen->optionCache, 508 brw->driContext->driScreenPriv->myNum, "i965"); 509 510 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 511 switch (bo_reuse_mode) { 512 case DRI_CONF_BO_REUSE_DISABLED: 513 break; 514 case DRI_CONF_BO_REUSE_ALL: 515 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 516 break; 517 } 518 519 if (!driQueryOptionb(options, "hiz")) { 520 brw->has_hiz = false; 521 /* On gen6, you can only do separate stencil with HIZ. */ 522 if (brw->gen == 6) 523 brw->has_separate_stencil = false; 524 } 525 526 if (driQueryOptionb(options, "always_flush_batch")) { 527 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 528 brw->always_flush_batch = true; 529 } 530 531 if (driQueryOptionb(options, "always_flush_cache")) { 532 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 533 brw->always_flush_cache = true; 534 } 535 536 if (driQueryOptionb(options, "disable_throttling")) { 537 fprintf(stderr, "disabling flush throttling\n"); 538 brw->disable_throttling = true; 539 } 540 541 brw->disable_derivative_optimization = 542 driQueryOptionb(&brw->optionCache, "disable_derivative_optimization"); 543 544 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 545 546 ctx->Const.ForceGLSLExtensionsWarn = 547 driQueryOptionb(options, "force_glsl_extensions_warn"); 548 549 ctx->Const.DisableGLSLLineContinuations = 550 driQueryOptionb(options, "disable_glsl_line_continuations"); 551} 552 553GLboolean 554brwCreateContext(gl_api api, 555 const struct gl_config *mesaVis, 556 __DRIcontext *driContextPriv, 557 unsigned major_version, 558 unsigned minor_version, 559 uint32_t flags, 560 bool notify_reset, 561 unsigned *dri_ctx_error, 562 void *sharedContextPrivate) 563{ 564 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 565 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 566 struct intel_screen *screen = sPriv->driverPrivate; 567 const struct brw_device_info *devinfo = screen->devinfo; 568 struct dd_function_table functions; 569 struct gl_config visual; 570 571 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 572 * provides us with context reset notifications. 573 */ 574 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 575 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 576 577 if (screen->has_context_reset_notification) 578 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 579 580 if (flags & ~allowed_flags) { 581 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 582 return false; 583 } 584 585 struct brw_context *brw = rzalloc(NULL, struct brw_context); 586 if (!brw) { 587 printf("%s: failed to alloc context\n", __FUNCTION__); 588 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 589 return false; 590 } 591 592 driContextPriv->driverPrivate = brw; 593 brw->driContext = driContextPriv; 594 brw->intelScreen = screen; 595 brw->bufmgr = screen->bufmgr; 596 597 brw->gen = devinfo->gen; 598 brw->gt = devinfo->gt; 599 brw->is_g4x = devinfo->is_g4x; 600 brw->is_baytrail = devinfo->is_baytrail; 601 brw->is_haswell = devinfo->is_haswell; 602 brw->has_llc = devinfo->has_llc; 603 brw->has_hiz = devinfo->has_hiz_and_separate_stencil && brw->gen < 8; 604 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 605 brw->has_pln = devinfo->has_pln; 606 brw->has_compr4 = devinfo->has_compr4; 607 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 608 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 609 brw->needs_unlit_centroid_workaround = 610 devinfo->needs_unlit_centroid_workaround; 611 612 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 613 brw->has_swizzling = screen->hw_has_swizzling; 614 615 if (brw->gen >= 7) { 616 gen7_init_vtable_surface_functions(brw); 617 gen7_init_vtable_sampler_functions(brw); 618 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 619 } else { 620 gen4_init_vtable_surface_functions(brw); 621 gen4_init_vtable_sampler_functions(brw); 622 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 623 } 624 625 brw_init_driver_functions(brw, &functions); 626 627 if (notify_reset) 628 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 629 630 struct gl_context *ctx = &brw->ctx; 631 632 if (mesaVis == NULL) { 633 memset(&visual, 0, sizeof visual); 634 mesaVis = &visual; 635 } 636 637 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 638 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 639 printf("%s: failed to init mesa context\n", __FUNCTION__); 640 intelDestroyContext(driContextPriv); 641 return false; 642 } 643 644 driContextSetFlags(ctx, flags); 645 646 /* Initialize the software rasterizer and helper modules. 647 * 648 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 649 * software fallbacks (which we have to support on legacy GL to do weird 650 * glDrawPixels(), glBitmap(), and other functions). 651 */ 652 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 653 _swrast_CreateContext(ctx); 654 } 655 656 _vbo_CreateContext(ctx); 657 if (ctx->swrast_context) { 658 _tnl_CreateContext(ctx); 659 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 660 _swsetup_CreateContext(ctx); 661 662 /* Configure swrast to match hardware characteristics: */ 663 _swrast_allow_pixel_fog(ctx, false); 664 _swrast_allow_vertex_fog(ctx, true); 665 } 666 667 _mesa_meta_init(ctx); 668 669 brw_process_driconf_options(brw); 670 brw_process_intel_debug_variable(brw); 671 brw_initialize_context_constants(brw); 672 673 ctx->Const.ResetStrategy = notify_reset 674 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 675 676 /* Reinitialize the context point state. It depends on ctx->Const values. */ 677 _mesa_init_point(ctx); 678 679 intel_batchbuffer_init(brw); 680 681 brw_init_state(brw); 682 683 intelInitExtensions(ctx); 684 685 intel_fbo_init(brw); 686 687 if (brw->gen >= 6) { 688 /* Create a new hardware context. Using a hardware context means that 689 * our GPU state will be saved/restored on context switch, allowing us 690 * to assume that the GPU is in the same state we left it in. 691 * 692 * This is required for transform feedback buffer offsets, query objects, 693 * and also allows us to reduce how much state we have to emit. 694 */ 695 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 696 697 if (!brw->hw_ctx) { 698 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 699 intelDestroyContext(driContextPriv); 700 return false; 701 } 702 } 703 704 brw_init_surface_formats(brw); 705 706 if (brw->is_g4x || brw->gen >= 5) { 707 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS; 708 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; 709 } else { 710 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS; 711 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; 712 } 713 714 brw->max_vs_threads = devinfo->max_vs_threads; 715 brw->max_gs_threads = devinfo->max_gs_threads; 716 brw->max_wm_threads = devinfo->max_wm_threads; 717 brw->urb.size = devinfo->urb.size; 718 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 719 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 720 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 721 722 /* Estimate the size of the mappable aperture into the GTT. There's an 723 * ioctl to get the whole GTT size, but not one to get the mappable subset. 724 * It turns out it's basically always 256MB, though some ancient hardware 725 * was smaller. 726 */ 727 uint32_t gtt_size = 256 * 1024 * 1024; 728 729 /* We don't want to map two objects such that a memcpy between them would 730 * just fault one mapping in and then the other over and over forever. So 731 * we would need to divide the GTT size by 2. Additionally, some GTT is 732 * taken up by things like the framebuffer and the ringbuffer and such, so 733 * be more conservative. 734 */ 735 brw->max_gtt_map_object_size = gtt_size / 4; 736 737 if (brw->gen == 6) 738 brw->urb.gen6_gs_previously_active = false; 739 740 brw->prim_restart.in_progress = false; 741 brw->prim_restart.enable_cut_index = false; 742 brw->gs.enabled = false; 743 744 if (brw->gen < 6) { 745 brw->curbe.last_buf = calloc(1, 4096); 746 brw->curbe.next_buf = calloc(1, 4096); 747 } 748 749 ctx->VertexProgram._MaintainTnlProgram = true; 750 ctx->FragmentProgram._MaintainTexEnvProgram = true; 751 752 brw_draw_init( brw ); 753 754 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 755 /* Turn on some extra GL_ARB_debug_output generation. */ 756 brw->perf_debug = true; 757 } 758 759 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 760 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 761 762 brw_fs_alloc_reg_sets(brw); 763 brw_vec4_alloc_reg_set(brw); 764 765 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 766 brw_init_shader_time(brw); 767 768 _mesa_compute_version(ctx); 769 770 _mesa_initialize_dispatch_tables(ctx); 771 _mesa_initialize_vbo_vtxfmt(ctx); 772 773 if (ctx->Extensions.AMD_performance_monitor) { 774 brw_init_performance_monitors(brw); 775 } 776 777 return true; 778} 779 780void 781intelDestroyContext(__DRIcontext * driContextPriv) 782{ 783 struct brw_context *brw = 784 (struct brw_context *) driContextPriv->driverPrivate; 785 struct gl_context *ctx = &brw->ctx; 786 787 assert(brw); /* should never be null */ 788 if (!brw) 789 return; 790 791 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 792 if (INTEL_DEBUG & DEBUG_AUB) { 793 intel_batchbuffer_flush(brw); 794 aub_dump_bmp(&brw->ctx); 795 } 796 797 _mesa_meta_free(&brw->ctx); 798 799 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 800 /* Force a report. */ 801 brw->shader_time.report_time = 0; 802 803 brw_collect_and_report_shader_time(brw); 804 brw_destroy_shader_time(brw); 805 } 806 807 brw_destroy_state(brw); 808 brw_draw_destroy(brw); 809 810 drm_intel_bo_unreference(brw->curbe.curbe_bo); 811 drm_intel_bo_unreference(brw->vs.base.const_bo); 812 drm_intel_bo_unreference(brw->wm.base.const_bo); 813 814 free(brw->curbe.last_buf); 815 free(brw->curbe.next_buf); 816 817 drm_intel_gem_context_destroy(brw->hw_ctx); 818 819 if (ctx->swrast_context) { 820 _swsetup_DestroyContext(&brw->ctx); 821 _tnl_DestroyContext(&brw->ctx); 822 } 823 _vbo_DestroyContext(&brw->ctx); 824 825 if (ctx->swrast_context) 826 _swrast_DestroyContext(&brw->ctx); 827 828 intel_batchbuffer_free(brw); 829 830 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 831 brw->first_post_swapbuffers_batch = NULL; 832 833 driDestroyOptionCache(&brw->optionCache); 834 835 /* free the Mesa context */ 836 _mesa_free_context_data(&brw->ctx); 837 838 ralloc_free(brw); 839 driContextPriv->driverPrivate = NULL; 840} 841 842GLboolean 843intelUnbindContext(__DRIcontext * driContextPriv) 844{ 845 /* Unset current context and dispath table */ 846 _mesa_make_current(NULL, NULL, NULL); 847 848 return true; 849} 850 851/** 852 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 853 * on window system framebuffers. 854 * 855 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 856 * your renderbuffer can do sRGB encode, and you can flip a switch that does 857 * sRGB encode if the renderbuffer can handle it. You can ask specifically 858 * for a visual where you're guaranteed to be capable, but it turns out that 859 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 860 * incapable ones, becuase there's no difference between the two in resources 861 * used. Applications thus get built that accidentally rely on the default 862 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 863 * great... 864 * 865 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 866 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 867 * So they removed the enable knob and made it "if the renderbuffer is sRGB 868 * capable, do sRGB encode". Then, for your window system renderbuffers, you 869 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 870 * and get no sRGB encode (assuming that both kinds of visual are available). 871 * Thus our choice to support sRGB by default on our visuals for desktop would 872 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 873 * 874 * Unfortunately, renderbuffer setup happens before a context is created. So 875 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 876 * context (without an sRGB visual, though we don't have sRGB visuals exposed 877 * yet), we go turn that back off before anyone finds out. 878 */ 879static void 880intel_gles3_srgb_workaround(struct brw_context *brw, 881 struct gl_framebuffer *fb) 882{ 883 struct gl_context *ctx = &brw->ctx; 884 885 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 886 return; 887 888 /* Some day when we support the sRGB capable bit on visuals available for 889 * GLES, we'll need to respect that and not disable things here. 890 */ 891 fb->Visual.sRGBCapable = false; 892 for (int i = 0; i < BUFFER_COUNT; i++) { 893 if (fb->Attachment[i].Renderbuffer && 894 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) { 895 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM; 896 } 897 } 898} 899 900GLboolean 901intelMakeCurrent(__DRIcontext * driContextPriv, 902 __DRIdrawable * driDrawPriv, 903 __DRIdrawable * driReadPriv) 904{ 905 struct brw_context *brw; 906 GET_CURRENT_CONTEXT(curCtx); 907 908 if (driContextPriv) 909 brw = (struct brw_context *) driContextPriv->driverPrivate; 910 else 911 brw = NULL; 912 913 /* According to the glXMakeCurrent() man page: "Pending commands to 914 * the previous context, if any, are flushed before it is released." 915 * But only flush if we're actually changing contexts. 916 */ 917 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 918 _mesa_flush(curCtx); 919 } 920 921 if (driContextPriv) { 922 struct gl_context *ctx = &brw->ctx; 923 struct gl_framebuffer *fb, *readFb; 924 925 if (driDrawPriv == NULL && driReadPriv == NULL) { 926 fb = _mesa_get_incomplete_framebuffer(); 927 readFb = _mesa_get_incomplete_framebuffer(); 928 } else { 929 fb = driDrawPriv->driverPrivate; 930 readFb = driReadPriv->driverPrivate; 931 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 932 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 933 } 934 935 /* The sRGB workaround changes the renderbuffer's format. We must change 936 * the format before the renderbuffer's miptree get's allocated, otherwise 937 * the formats of the renderbuffer and its miptree will differ. 938 */ 939 intel_gles3_srgb_workaround(brw, fb); 940 intel_gles3_srgb_workaround(brw, readFb); 941 942 /* If the context viewport hasn't been initialized, force a call out to 943 * the loader to get buffers so we have a drawable size for the initial 944 * viewport. */ 945 if (!brw->ctx.ViewportInitialized) 946 intel_prepare_render(brw); 947 948 _mesa_make_current(ctx, fb, readFb); 949 } else { 950 _mesa_make_current(NULL, NULL, NULL); 951 } 952 953 return true; 954} 955 956void 957intel_resolve_for_dri2_flush(struct brw_context *brw, 958 __DRIdrawable *drawable) 959{ 960 if (brw->gen < 6) { 961 /* MSAA and fast color clear are not supported, so don't waste time 962 * checking whether a resolve is needed. 963 */ 964 return; 965 } 966 967 struct gl_framebuffer *fb = drawable->driverPrivate; 968 struct intel_renderbuffer *rb; 969 970 /* Usually, only the back buffer will need to be downsampled. However, 971 * the front buffer will also need it if the user has rendered into it. 972 */ 973 static const gl_buffer_index buffers[2] = { 974 BUFFER_BACK_LEFT, 975 BUFFER_FRONT_LEFT, 976 }; 977 978 for (int i = 0; i < 2; ++i) { 979 rb = intel_get_renderbuffer(fb, buffers[i]); 980 if (rb == NULL || rb->mt == NULL) 981 continue; 982 if (rb->mt->num_samples <= 1) 983 intel_miptree_resolve_color(brw, rb->mt); 984 else 985 intel_miptree_downsample(brw, rb->mt); 986 } 987} 988 989static unsigned 990intel_bits_per_pixel(const struct intel_renderbuffer *rb) 991{ 992 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 993} 994 995static void 996intel_query_dri2_buffers(struct brw_context *brw, 997 __DRIdrawable *drawable, 998 __DRIbuffer **buffers, 999 int *count); 1000 1001static void 1002intel_process_dri2_buffer(struct brw_context *brw, 1003 __DRIdrawable *drawable, 1004 __DRIbuffer *buffer, 1005 struct intel_renderbuffer *rb, 1006 const char *buffer_name); 1007 1008static void 1009intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 1010 1011static void 1012intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1013{ 1014 struct gl_framebuffer *fb = drawable->driverPrivate; 1015 struct intel_renderbuffer *rb; 1016 __DRIbuffer *buffers = NULL; 1017 int i, count; 1018 const char *region_name; 1019 1020 /* Set this up front, so that in case our buffers get invalidated 1021 * while we're getting new buffers, we don't clobber the stamp and 1022 * thus ignore the invalidate. */ 1023 drawable->lastStamp = drawable->dri2.stamp; 1024 1025 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1026 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1027 1028 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1029 1030 if (buffers == NULL) 1031 return; 1032 1033 for (i = 0; i < count; i++) { 1034 switch (buffers[i].attachment) { 1035 case __DRI_BUFFER_FRONT_LEFT: 1036 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1037 region_name = "dri2 front buffer"; 1038 break; 1039 1040 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1041 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1042 region_name = "dri2 fake front buffer"; 1043 break; 1044 1045 case __DRI_BUFFER_BACK_LEFT: 1046 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1047 region_name = "dri2 back buffer"; 1048 break; 1049 1050 case __DRI_BUFFER_DEPTH: 1051 case __DRI_BUFFER_HIZ: 1052 case __DRI_BUFFER_DEPTH_STENCIL: 1053 case __DRI_BUFFER_STENCIL: 1054 case __DRI_BUFFER_ACCUM: 1055 default: 1056 fprintf(stderr, 1057 "unhandled buffer attach event, attachment type %d\n", 1058 buffers[i].attachment); 1059 return; 1060 } 1061 1062 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1063 } 1064 1065} 1066 1067void 1068intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1069{ 1070 struct brw_context *brw = context->driverPrivate; 1071 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1072 1073 /* Set this up front, so that in case our buffers get invalidated 1074 * while we're getting new buffers, we don't clobber the stamp and 1075 * thus ignore the invalidate. */ 1076 drawable->lastStamp = drawable->dri2.stamp; 1077 1078 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1079 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1080 1081 if (screen->image.loader) 1082 intel_update_image_buffers(brw, drawable); 1083 else 1084 intel_update_dri2_buffers(brw, drawable); 1085 1086 driUpdateFramebufferSize(&brw->ctx, drawable); 1087} 1088 1089/** 1090 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1091 * state is required. 1092 */ 1093void 1094intel_prepare_render(struct brw_context *brw) 1095{ 1096 __DRIcontext *driContext = brw->driContext; 1097 __DRIdrawable *drawable; 1098 1099 drawable = driContext->driDrawablePriv; 1100 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1101 if (drawable->lastStamp != drawable->dri2.stamp) 1102 intel_update_renderbuffers(driContext, drawable); 1103 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1104 } 1105 1106 drawable = driContext->driReadablePriv; 1107 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1108 if (drawable->lastStamp != drawable->dri2.stamp) 1109 intel_update_renderbuffers(driContext, drawable); 1110 driContext->dri2.read_stamp = drawable->dri2.stamp; 1111 } 1112 1113 /* If we're currently rendering to the front buffer, the rendering 1114 * that will happen next will probably dirty the front buffer. So 1115 * mark it as dirty here. 1116 */ 1117 if (brw->is_front_buffer_rendering) 1118 brw->front_buffer_dirty = true; 1119 1120 /* Wait for the swapbuffers before the one we just emitted, so we 1121 * don't get too many swaps outstanding for apps that are GPU-heavy 1122 * but not CPU-heavy. 1123 * 1124 * We're using intelDRI2Flush (called from the loader before 1125 * swapbuffer) and glFlush (for front buffer rendering) as the 1126 * indicator that a frame is done and then throttle when we get 1127 * here as we prepare to render the next frame. At this point for 1128 * round trips for swap/copy and getting new buffers are done and 1129 * we'll spend less time waiting on the GPU. 1130 * 1131 * Unfortunately, we don't have a handle to the batch containing 1132 * the swap, and getting our hands on that doesn't seem worth it, 1133 * so we just us the first batch we emitted after the last swap. 1134 */ 1135 if (brw->need_throttle && brw->first_post_swapbuffers_batch) { 1136 if (!brw->disable_throttling) 1137 drm_intel_bo_wait_rendering(brw->first_post_swapbuffers_batch); 1138 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 1139 brw->first_post_swapbuffers_batch = NULL; 1140 brw->need_throttle = false; 1141 } 1142} 1143 1144/** 1145 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1146 * 1147 * To determine which DRI buffers to request, examine the renderbuffers 1148 * attached to the drawable's framebuffer. Then request the buffers with 1149 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1150 * 1151 * This is called from intel_update_renderbuffers(). 1152 * 1153 * \param drawable Drawable whose buffers are queried. 1154 * \param buffers [out] List of buffers returned by DRI2 query. 1155 * \param buffer_count [out] Number of buffers returned. 1156 * 1157 * \see intel_update_renderbuffers() 1158 * \see DRI2GetBuffers() 1159 * \see DRI2GetBuffersWithFormat() 1160 */ 1161static void 1162intel_query_dri2_buffers(struct brw_context *brw, 1163 __DRIdrawable *drawable, 1164 __DRIbuffer **buffers, 1165 int *buffer_count) 1166{ 1167 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1168 struct gl_framebuffer *fb = drawable->driverPrivate; 1169 int i = 0; 1170 unsigned attachments[8]; 1171 1172 struct intel_renderbuffer *front_rb; 1173 struct intel_renderbuffer *back_rb; 1174 1175 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1176 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1177 1178 memset(attachments, 0, sizeof(attachments)); 1179 if ((brw->is_front_buffer_rendering || 1180 brw->is_front_buffer_reading || 1181 !back_rb) && front_rb) { 1182 /* If a fake front buffer is in use, then querying for 1183 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1184 * the real front buffer to the fake front buffer. So before doing the 1185 * query, we need to make sure all the pending drawing has landed in the 1186 * real front buffer. 1187 */ 1188 intel_batchbuffer_flush(brw); 1189 intel_flush_front(&brw->ctx); 1190 1191 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1192 attachments[i++] = intel_bits_per_pixel(front_rb); 1193 } else if (front_rb && brw->front_buffer_dirty) { 1194 /* We have pending front buffer rendering, but we aren't querying for a 1195 * front buffer. If the front buffer we have is a fake front buffer, 1196 * the X server is going to throw it away when it processes the query. 1197 * So before doing the query, make sure all the pending drawing has 1198 * landed in the real front buffer. 1199 */ 1200 intel_batchbuffer_flush(brw); 1201 intel_flush_front(&brw->ctx); 1202 } 1203 1204 if (back_rb) { 1205 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1206 attachments[i++] = intel_bits_per_pixel(back_rb); 1207 } 1208 1209 assert(i <= ARRAY_SIZE(attachments)); 1210 1211 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1212 &drawable->w, 1213 &drawable->h, 1214 attachments, i / 2, 1215 buffer_count, 1216 drawable->loaderPrivate); 1217} 1218 1219/** 1220 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1221 * 1222 * This is called from intel_update_renderbuffers(). 1223 * 1224 * \par Note: 1225 * DRI buffers whose attachment point is DRI2BufferStencil or 1226 * DRI2BufferDepthStencil are handled as special cases. 1227 * 1228 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1229 * that is passed to intel_region_alloc_for_handle(). 1230 * 1231 * \see intel_update_renderbuffers() 1232 * \see intel_region_alloc_for_handle() 1233 */ 1234static void 1235intel_process_dri2_buffer(struct brw_context *brw, 1236 __DRIdrawable *drawable, 1237 __DRIbuffer *buffer, 1238 struct intel_renderbuffer *rb, 1239 const char *buffer_name) 1240{ 1241 struct intel_region *region = NULL; 1242 1243 if (!rb) 1244 return; 1245 1246 unsigned num_samples = rb->Base.Base.NumSamples; 1247 1248 /* We try to avoid closing and reopening the same BO name, because the first 1249 * use of a mapping of the buffer involves a bunch of page faulting which is 1250 * moderately expensive. 1251 */ 1252 if (num_samples == 0) { 1253 if (rb->mt && 1254 rb->mt->region && 1255 rb->mt->region->name == buffer->name) 1256 return; 1257 } else { 1258 if (rb->mt && 1259 rb->mt->singlesample_mt && 1260 rb->mt->singlesample_mt->region && 1261 rb->mt->singlesample_mt->region->name == buffer->name) 1262 return; 1263 } 1264 1265 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1266 fprintf(stderr, 1267 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1268 buffer->name, buffer->attachment, 1269 buffer->cpp, buffer->pitch); 1270 } 1271 1272 intel_miptree_release(&rb->mt); 1273 region = intel_region_alloc_for_handle(brw->intelScreen, 1274 buffer->cpp, 1275 drawable->w, 1276 drawable->h, 1277 buffer->pitch, 1278 buffer->name, 1279 buffer_name); 1280 if (!region) 1281 return; 1282 1283 rb->mt = intel_miptree_create_for_dri2_buffer(brw, 1284 buffer->attachment, 1285 intel_rb_format(rb), 1286 num_samples, 1287 region); 1288 intel_region_release(®ion); 1289} 1290 1291/** 1292 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1293 * 1294 * To determine which DRI buffers to request, examine the renderbuffers 1295 * attached to the drawable's framebuffer. Then request the buffers from 1296 * the image loader 1297 * 1298 * This is called from intel_update_renderbuffers(). 1299 * 1300 * \param drawable Drawable whose buffers are queried. 1301 * \param buffers [out] List of buffers returned by DRI2 query. 1302 * \param buffer_count [out] Number of buffers returned. 1303 * 1304 * \see intel_update_renderbuffers() 1305 */ 1306 1307static void 1308intel_update_image_buffer(struct brw_context *intel, 1309 __DRIdrawable *drawable, 1310 struct intel_renderbuffer *rb, 1311 __DRIimage *buffer, 1312 enum __DRIimageBufferMask buffer_type) 1313{ 1314 struct intel_region *region = buffer->region; 1315 1316 if (!rb || !region) 1317 return; 1318 1319 unsigned num_samples = rb->Base.Base.NumSamples; 1320 1321 /* Check and see if we're already bound to the right 1322 * buffer object 1323 */ 1324 if (num_samples == 0) { 1325 if (rb->mt && 1326 rb->mt->region && 1327 rb->mt->region->bo == region->bo) 1328 return; 1329 } else { 1330 if (rb->mt && 1331 rb->mt->singlesample_mt && 1332 rb->mt->singlesample_mt->region && 1333 rb->mt->singlesample_mt->region->bo == region->bo) 1334 return; 1335 } 1336 1337 intel_miptree_release(&rb->mt); 1338 rb->mt = intel_miptree_create_for_image_buffer(intel, 1339 buffer_type, 1340 intel_rb_format(rb), 1341 num_samples, 1342 region); 1343} 1344 1345static void 1346intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1347{ 1348 struct gl_framebuffer *fb = drawable->driverPrivate; 1349 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1350 struct intel_renderbuffer *front_rb; 1351 struct intel_renderbuffer *back_rb; 1352 struct __DRIimageList images; 1353 unsigned int format; 1354 uint32_t buffer_mask = 0; 1355 1356 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1357 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1358 1359 if (back_rb) 1360 format = intel_rb_format(back_rb); 1361 else if (front_rb) 1362 format = intel_rb_format(front_rb); 1363 else 1364 return; 1365 1366 if ((brw->is_front_buffer_rendering || brw->is_front_buffer_reading || !back_rb) && front_rb) 1367 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1368 1369 if (back_rb) 1370 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1371 1372 (*screen->image.loader->getBuffers) (drawable, 1373 driGLFormatToImageFormat(format), 1374 &drawable->dri2.stamp, 1375 drawable->loaderPrivate, 1376 buffer_mask, 1377 &images); 1378 1379 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1380 drawable->w = images.front->width; 1381 drawable->h = images.front->height; 1382 intel_update_image_buffer(brw, 1383 drawable, 1384 front_rb, 1385 images.front, 1386 __DRI_IMAGE_BUFFER_FRONT); 1387 } 1388 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1389 drawable->w = images.back->width; 1390 drawable->h = images.back->height; 1391 intel_update_image_buffer(brw, 1392 drawable, 1393 back_rb, 1394 images.back, 1395 __DRI_IMAGE_BUFFER_BACK); 1396 } 1397} 1398