brw_context.c revision 05da4a7a5e7d5bd988cb31f94ed8e1f053d9ee39
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 ctx->Const.QueryCounterBits.Timestamp = 36; 281 282 ctx->Const.StripTextureBorder = true; 283 284 ctx->Const.MaxDualSourceDrawBuffers = 1; 285 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 286 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 287 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 288 ctx->Const.MaxTextureUnits = 289 MIN2(ctx->Const.MaxTextureCoordUnits, 290 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits); 291 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 292 if (brw->gen >= 7) 293 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 294 else 295 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = 0; 296 ctx->Const.MaxCombinedTextureImageUnits = 297 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits + 298 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits + 299 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits; 300 301 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 302 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 303 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 304 ctx->Const.Max3DTextureLevels = 9; 305 ctx->Const.MaxCubeTextureLevels = 12; 306 307 if (brw->gen >= 7) 308 ctx->Const.MaxArrayTextureLayers = 2048; 309 else 310 ctx->Const.MaxArrayTextureLayers = 512; 311 312 ctx->Const.MaxTextureRectSize = 1 << 12; 313 314 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 315 316 ctx->Const.MaxRenderbufferSize = 8192; 317 318 /* Hardware only supports a limited number of transform feedback buffers. 319 * So we need to override the Mesa default (which is based only on software 320 * limits). 321 */ 322 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 323 324 /* On Gen6, in the worst case, we use up one binding table entry per 325 * transform feedback component (see comments above the definition of 326 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 327 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 328 * BRW_MAX_SOL_BINDINGS. 329 * 330 * In "separate components" mode, we need to divide this value by 331 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 332 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 333 */ 334 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 335 ctx->Const.MaxTransformFeedbackSeparateComponents = 336 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 337 338 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 339 340 int max_samples; 341 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 342 const int clamp_max_samples = 343 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 344 345 if (clamp_max_samples < 0) { 346 max_samples = msaa_modes[0]; 347 } else { 348 /* Select the largest supported MSAA mode that does not exceed 349 * clamp_max_samples. 350 */ 351 max_samples = 0; 352 for (int i = 0; msaa_modes[i] != 0; ++i) { 353 if (msaa_modes[i] <= clamp_max_samples) { 354 max_samples = msaa_modes[i]; 355 break; 356 } 357 } 358 } 359 360 ctx->Const.MaxSamples = max_samples; 361 ctx->Const.MaxColorTextureSamples = max_samples; 362 ctx->Const.MaxDepthTextureSamples = max_samples; 363 ctx->Const.MaxIntegerSamples = max_samples; 364 365 if (brw->gen >= 7) 366 ctx->Const.MaxProgramTextureGatherComponents = 4; 367 368 ctx->Const.MinLineWidth = 1.0; 369 ctx->Const.MinLineWidthAA = 1.0; 370 ctx->Const.MaxLineWidth = 5.0; 371 ctx->Const.MaxLineWidthAA = 5.0; 372 ctx->Const.LineWidthGranularity = 0.5; 373 374 ctx->Const.MinPointSize = 1.0; 375 ctx->Const.MinPointSizeAA = 1.0; 376 ctx->Const.MaxPointSize = 255.0; 377 ctx->Const.MaxPointSizeAA = 255.0; 378 ctx->Const.PointSizeGranularity = 1.0; 379 380 if (brw->gen >= 5 || brw->is_g4x) 381 ctx->Const.MaxClipPlanes = 8; 382 383 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024; 384 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0; 385 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0; 386 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0; 387 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0; 388 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0; 389 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0; 390 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16; 391 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256; 392 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1; 393 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024; 394 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams = 395 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters, 396 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams); 397 398 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024; 399 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024; 400 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024; 401 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024; 402 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12; 403 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256; 404 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0; 405 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024; 406 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams = 407 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters, 408 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams); 409 410 /* Fragment shaders use real, 32-bit twos-complement integers for all 411 * integer types. 412 */ 413 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31; 414 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30; 415 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0; 416 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 417 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 418 419 if (brw->gen >= 7) { 420 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 421 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 422 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 423 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicBuffers = BRW_MAX_ABO; 424 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicBuffers = BRW_MAX_ABO; 425 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicBuffers = BRW_MAX_ABO; 426 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 427 } 428 429 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 430 * but we're not sure how it's actually done for vertex order, 431 * that affect provoking vertex decision. Always use last vertex 432 * convention for quad primitive which works as expected for now. 433 */ 434 if (brw->gen >= 6) 435 ctx->Const.QuadsFollowProvokingVertexConvention = false; 436 437 ctx->Const.NativeIntegers = true; 438 ctx->Const.UniformBooleanTrue = 1; 439 440 /* From the gen4 PRM, volume 4 page 127: 441 * 442 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 443 * the base address of the first element of the surface, computed in 444 * software by adding the surface base address to the byte offset of 445 * the element in the buffer." 446 * 447 * However, unaligned accesses are slower, so enforce buffer alignment. 448 */ 449 ctx->Const.UniformBufferOffsetAlignment = 16; 450 ctx->Const.TextureBufferOffsetAlignment = 16; 451 452 if (brw->gen >= 6) { 453 ctx->Const.MaxVarying = 32; 454 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128; 455 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64; 456 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128; 457 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128; 458 } 459 460 /* We want the GLSL compiler to emit code that uses condition codes */ 461 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 462 ctx->ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 463 ctx->ShaderCompilerOptions[i].EmitCondCodes = true; 464 ctx->ShaderCompilerOptions[i].EmitNoNoise = true; 465 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true; 466 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true; 467 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true; 468 469 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform = 470 (i == MESA_SHADER_FRAGMENT); 471 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp = 472 (i == MESA_SHADER_FRAGMENT); 473 ctx->ShaderCompilerOptions[i].LowerClipDistance = true; 474 } 475 476 ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].PreferDP4 = true; 477} 478 479/** 480 * Process driconf (drirc) options, setting appropriate context flags. 481 * 482 * intelInitExtensions still pokes at optionCache directly, in order to 483 * avoid advertising various extensions. No flags are set, so it makes 484 * sense to continue doing that there. 485 */ 486static void 487brw_process_driconf_options(struct brw_context *brw) 488{ 489 struct gl_context *ctx = &brw->ctx; 490 491 driOptionCache *options = &brw->optionCache; 492 driParseConfigFiles(options, &brw->intelScreen->optionCache, 493 brw->driContext->driScreenPriv->myNum, "i965"); 494 495 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 496 switch (bo_reuse_mode) { 497 case DRI_CONF_BO_REUSE_DISABLED: 498 break; 499 case DRI_CONF_BO_REUSE_ALL: 500 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 501 break; 502 } 503 504 if (!driQueryOptionb(options, "hiz")) { 505 brw->has_hiz = false; 506 /* On gen6, you can only do separate stencil with HIZ. */ 507 if (brw->gen == 6) 508 brw->has_separate_stencil = false; 509 } 510 511 if (driQueryOptionb(options, "always_flush_batch")) { 512 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 513 brw->always_flush_batch = true; 514 } 515 516 if (driQueryOptionb(options, "always_flush_cache")) { 517 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 518 brw->always_flush_cache = true; 519 } 520 521 if (driQueryOptionb(options, "disable_throttling")) { 522 fprintf(stderr, "disabling flush throttling\n"); 523 brw->disable_throttling = true; 524 } 525 526 brw->disable_derivative_optimization = 527 driQueryOptionb(&brw->optionCache, "disable_derivative_optimization"); 528 529 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 530 531 ctx->Const.ForceGLSLExtensionsWarn = 532 driQueryOptionb(options, "force_glsl_extensions_warn"); 533 534 ctx->Const.DisableGLSLLineContinuations = 535 driQueryOptionb(options, "disable_glsl_line_continuations"); 536} 537 538GLboolean 539brwCreateContext(gl_api api, 540 const struct gl_config *mesaVis, 541 __DRIcontext *driContextPriv, 542 unsigned major_version, 543 unsigned minor_version, 544 uint32_t flags, 545 bool notify_reset, 546 unsigned *dri_ctx_error, 547 void *sharedContextPrivate) 548{ 549 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 550 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 551 struct intel_screen *screen = sPriv->driverPrivate; 552 const struct brw_device_info *devinfo = screen->devinfo; 553 struct dd_function_table functions; 554 struct gl_config visual; 555 556 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 557 * provides us with context reset notifications. 558 */ 559 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 560 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 561 562 if (screen->has_context_reset_notification) 563 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 564 565 if (flags & ~allowed_flags) { 566 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 567 return false; 568 } 569 570 struct brw_context *brw = rzalloc(NULL, struct brw_context); 571 if (!brw) { 572 printf("%s: failed to alloc context\n", __FUNCTION__); 573 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 574 return false; 575 } 576 577 driContextPriv->driverPrivate = brw; 578 brw->driContext = driContextPriv; 579 brw->intelScreen = screen; 580 brw->bufmgr = screen->bufmgr; 581 582 brw->gen = devinfo->gen; 583 brw->gt = devinfo->gt; 584 brw->is_g4x = devinfo->is_g4x; 585 brw->is_baytrail = devinfo->is_baytrail; 586 brw->is_haswell = devinfo->is_haswell; 587 brw->has_llc = devinfo->has_llc; 588 brw->has_hiz = devinfo->has_hiz_and_separate_stencil && brw->gen < 8; 589 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 590 brw->has_pln = devinfo->has_pln; 591 brw->has_compr4 = devinfo->has_compr4; 592 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 593 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 594 brw->needs_unlit_centroid_workaround = 595 devinfo->needs_unlit_centroid_workaround; 596 597 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 598 brw->has_swizzling = screen->hw_has_swizzling; 599 600 if (brw->gen >= 7) { 601 gen7_init_vtable_surface_functions(brw); 602 gen7_init_vtable_sampler_functions(brw); 603 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 604 } else { 605 gen4_init_vtable_surface_functions(brw); 606 gen4_init_vtable_sampler_functions(brw); 607 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 608 } 609 610 brw_init_driver_functions(brw, &functions); 611 612 if (notify_reset) 613 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 614 615 struct gl_context *ctx = &brw->ctx; 616 617 if (mesaVis == NULL) { 618 memset(&visual, 0, sizeof visual); 619 mesaVis = &visual; 620 } 621 622 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 623 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 624 printf("%s: failed to init mesa context\n", __FUNCTION__); 625 intelDestroyContext(driContextPriv); 626 return false; 627 } 628 629 driContextSetFlags(ctx, flags); 630 631 /* Initialize the software rasterizer and helper modules. 632 * 633 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 634 * software fallbacks (which we have to support on legacy GL to do weird 635 * glDrawPixels(), glBitmap(), and other functions). 636 */ 637 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 638 _swrast_CreateContext(ctx); 639 } 640 641 _vbo_CreateContext(ctx); 642 if (ctx->swrast_context) { 643 _tnl_CreateContext(ctx); 644 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 645 _swsetup_CreateContext(ctx); 646 647 /* Configure swrast to match hardware characteristics: */ 648 _swrast_allow_pixel_fog(ctx, false); 649 _swrast_allow_vertex_fog(ctx, true); 650 } 651 652 _mesa_meta_init(ctx); 653 654 brw_process_driconf_options(brw); 655 brw_process_intel_debug_variable(brw); 656 brw_initialize_context_constants(brw); 657 658 ctx->Const.ResetStrategy = notify_reset 659 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 660 661 /* Reinitialize the context point state. It depends on ctx->Const values. */ 662 _mesa_init_point(ctx); 663 664 intel_batchbuffer_init(brw); 665 666 brw_init_state(brw); 667 668 intelInitExtensions(ctx); 669 670 intel_fbo_init(brw); 671 672 if (brw->gen >= 6) { 673 /* Create a new hardware context. Using a hardware context means that 674 * our GPU state will be saved/restored on context switch, allowing us 675 * to assume that the GPU is in the same state we left it in. 676 * 677 * This is required for transform feedback buffer offsets, query objects, 678 * and also allows us to reduce how much state we have to emit. 679 */ 680 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 681 682 if (!brw->hw_ctx) { 683 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 684 intelDestroyContext(driContextPriv); 685 return false; 686 } 687 } 688 689 brw_init_surface_formats(brw); 690 691 if (brw->is_g4x || brw->gen >= 5) { 692 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS; 693 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; 694 } else { 695 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS; 696 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; 697 } 698 699 brw->max_vs_threads = devinfo->max_vs_threads; 700 brw->max_gs_threads = devinfo->max_gs_threads; 701 brw->max_wm_threads = devinfo->max_wm_threads; 702 brw->urb.size = devinfo->urb.size; 703 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 704 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 705 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 706 707 /* Estimate the size of the mappable aperture into the GTT. There's an 708 * ioctl to get the whole GTT size, but not one to get the mappable subset. 709 * It turns out it's basically always 256MB, though some ancient hardware 710 * was smaller. 711 */ 712 uint32_t gtt_size = 256 * 1024 * 1024; 713 714 /* We don't want to map two objects such that a memcpy between them would 715 * just fault one mapping in and then the other over and over forever. So 716 * we would need to divide the GTT size by 2. Additionally, some GTT is 717 * taken up by things like the framebuffer and the ringbuffer and such, so 718 * be more conservative. 719 */ 720 brw->max_gtt_map_object_size = gtt_size / 4; 721 722 if (brw->gen == 6) 723 brw->urb.gen6_gs_previously_active = false; 724 725 brw->prim_restart.in_progress = false; 726 brw->prim_restart.enable_cut_index = false; 727 brw->gs.enabled = false; 728 729 if (brw->gen < 6) { 730 brw->curbe.last_buf = calloc(1, 4096); 731 brw->curbe.next_buf = calloc(1, 4096); 732 } 733 734 ctx->VertexProgram._MaintainTnlProgram = true; 735 ctx->FragmentProgram._MaintainTexEnvProgram = true; 736 737 brw_draw_init( brw ); 738 739 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 740 /* Turn on some extra GL_ARB_debug_output generation. */ 741 brw->perf_debug = true; 742 } 743 744 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 745 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 746 747 brw_fs_alloc_reg_sets(brw); 748 brw_vec4_alloc_reg_set(brw); 749 750 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 751 brw_init_shader_time(brw); 752 753 _mesa_compute_version(ctx); 754 755 _mesa_initialize_dispatch_tables(ctx); 756 _mesa_initialize_vbo_vtxfmt(ctx); 757 758 if (ctx->Extensions.AMD_performance_monitor) { 759 brw_init_performance_monitors(brw); 760 } 761 762 return true; 763} 764 765void 766intelDestroyContext(__DRIcontext * driContextPriv) 767{ 768 struct brw_context *brw = 769 (struct brw_context *) driContextPriv->driverPrivate; 770 struct gl_context *ctx = &brw->ctx; 771 772 assert(brw); /* should never be null */ 773 if (!brw) 774 return; 775 776 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 777 if (INTEL_DEBUG & DEBUG_AUB) { 778 intel_batchbuffer_flush(brw); 779 aub_dump_bmp(&brw->ctx); 780 } 781 782 _mesa_meta_free(&brw->ctx); 783 784 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 785 /* Force a report. */ 786 brw->shader_time.report_time = 0; 787 788 brw_collect_and_report_shader_time(brw); 789 brw_destroy_shader_time(brw); 790 } 791 792 brw_destroy_state(brw); 793 brw_draw_destroy(brw); 794 795 drm_intel_bo_unreference(brw->curbe.curbe_bo); 796 drm_intel_bo_unreference(brw->vs.base.const_bo); 797 drm_intel_bo_unreference(brw->wm.base.const_bo); 798 799 free(brw->curbe.last_buf); 800 free(brw->curbe.next_buf); 801 802 drm_intel_gem_context_destroy(brw->hw_ctx); 803 804 if (ctx->swrast_context) { 805 _swsetup_DestroyContext(&brw->ctx); 806 _tnl_DestroyContext(&brw->ctx); 807 } 808 _vbo_DestroyContext(&brw->ctx); 809 810 if (ctx->swrast_context) 811 _swrast_DestroyContext(&brw->ctx); 812 813 intel_batchbuffer_free(brw); 814 815 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 816 brw->first_post_swapbuffers_batch = NULL; 817 818 driDestroyOptionCache(&brw->optionCache); 819 820 /* free the Mesa context */ 821 _mesa_free_context_data(&brw->ctx); 822 823 ralloc_free(brw); 824 driContextPriv->driverPrivate = NULL; 825} 826 827GLboolean 828intelUnbindContext(__DRIcontext * driContextPriv) 829{ 830 /* Unset current context and dispath table */ 831 _mesa_make_current(NULL, NULL, NULL); 832 833 return true; 834} 835 836/** 837 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 838 * on window system framebuffers. 839 * 840 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 841 * your renderbuffer can do sRGB encode, and you can flip a switch that does 842 * sRGB encode if the renderbuffer can handle it. You can ask specifically 843 * for a visual where you're guaranteed to be capable, but it turns out that 844 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 845 * incapable ones, becuase there's no difference between the two in resources 846 * used. Applications thus get built that accidentally rely on the default 847 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 848 * great... 849 * 850 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 851 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 852 * So they removed the enable knob and made it "if the renderbuffer is sRGB 853 * capable, do sRGB encode". Then, for your window system renderbuffers, you 854 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 855 * and get no sRGB encode (assuming that both kinds of visual are available). 856 * Thus our choice to support sRGB by default on our visuals for desktop would 857 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 858 * 859 * Unfortunately, renderbuffer setup happens before a context is created. So 860 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 861 * context (without an sRGB visual, though we don't have sRGB visuals exposed 862 * yet), we go turn that back off before anyone finds out. 863 */ 864static void 865intel_gles3_srgb_workaround(struct brw_context *brw, 866 struct gl_framebuffer *fb) 867{ 868 struct gl_context *ctx = &brw->ctx; 869 870 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 871 return; 872 873 /* Some day when we support the sRGB capable bit on visuals available for 874 * GLES, we'll need to respect that and not disable things here. 875 */ 876 fb->Visual.sRGBCapable = false; 877 for (int i = 0; i < BUFFER_COUNT; i++) { 878 if (fb->Attachment[i].Renderbuffer && 879 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_SARGB8) { 880 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_ARGB8888; 881 } 882 } 883} 884 885GLboolean 886intelMakeCurrent(__DRIcontext * driContextPriv, 887 __DRIdrawable * driDrawPriv, 888 __DRIdrawable * driReadPriv) 889{ 890 struct brw_context *brw; 891 GET_CURRENT_CONTEXT(curCtx); 892 893 if (driContextPriv) 894 brw = (struct brw_context *) driContextPriv->driverPrivate; 895 else 896 brw = NULL; 897 898 /* According to the glXMakeCurrent() man page: "Pending commands to 899 * the previous context, if any, are flushed before it is released." 900 * But only flush if we're actually changing contexts. 901 */ 902 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 903 _mesa_flush(curCtx); 904 } 905 906 if (driContextPriv) { 907 struct gl_context *ctx = &brw->ctx; 908 struct gl_framebuffer *fb, *readFb; 909 struct intel_renderbuffer *rb = NULL; 910 911 if (driDrawPriv == NULL && driReadPriv == NULL) { 912 fb = _mesa_get_incomplete_framebuffer(); 913 readFb = _mesa_get_incomplete_framebuffer(); 914 } else { 915 fb = driDrawPriv->driverPrivate; 916 readFb = driReadPriv->driverPrivate; 917 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 918 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 919 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 920 } 921 922 /* The sRGB workaround changes the renderbuffer's format. We must change 923 * the format before the renderbuffer's miptree get's allocated, otherwise 924 * the formats of the renderbuffer and its miptree will differ. 925 */ 926 intel_gles3_srgb_workaround(brw, fb); 927 intel_gles3_srgb_workaround(brw, readFb); 928 929 if (rb && !rb->mt) { 930 /* If we don't have buffers for the drawable yet, force a call to 931 * getbuffers here so we can have a default drawable size. */ 932 intel_prepare_render(brw); 933 } 934 935 _mesa_make_current(ctx, fb, readFb); 936 } else { 937 _mesa_make_current(NULL, NULL, NULL); 938 } 939 940 return true; 941} 942 943void 944intel_resolve_for_dri2_flush(struct brw_context *brw, 945 __DRIdrawable *drawable) 946{ 947 if (brw->gen < 6) { 948 /* MSAA and fast color clear are not supported, so don't waste time 949 * checking whether a resolve is needed. 950 */ 951 return; 952 } 953 954 struct gl_framebuffer *fb = drawable->driverPrivate; 955 struct intel_renderbuffer *rb; 956 957 /* Usually, only the back buffer will need to be downsampled. However, 958 * the front buffer will also need it if the user has rendered into it. 959 */ 960 static const gl_buffer_index buffers[2] = { 961 BUFFER_BACK_LEFT, 962 BUFFER_FRONT_LEFT, 963 }; 964 965 for (int i = 0; i < 2; ++i) { 966 rb = intel_get_renderbuffer(fb, buffers[i]); 967 if (rb == NULL || rb->mt == NULL) 968 continue; 969 if (rb->mt->num_samples <= 1) 970 intel_miptree_resolve_color(brw, rb->mt); 971 else 972 intel_miptree_downsample(brw, rb->mt); 973 } 974} 975 976static unsigned 977intel_bits_per_pixel(const struct intel_renderbuffer *rb) 978{ 979 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 980} 981 982static void 983intel_query_dri2_buffers(struct brw_context *brw, 984 __DRIdrawable *drawable, 985 __DRIbuffer **buffers, 986 int *count); 987 988static void 989intel_process_dri2_buffer(struct brw_context *brw, 990 __DRIdrawable *drawable, 991 __DRIbuffer *buffer, 992 struct intel_renderbuffer *rb, 993 const char *buffer_name); 994 995static void 996intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 997 998static void 999intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1000{ 1001 struct gl_framebuffer *fb = drawable->driverPrivate; 1002 struct intel_renderbuffer *rb; 1003 __DRIbuffer *buffers = NULL; 1004 int i, count; 1005 const char *region_name; 1006 1007 /* Set this up front, so that in case our buffers get invalidated 1008 * while we're getting new buffers, we don't clobber the stamp and 1009 * thus ignore the invalidate. */ 1010 drawable->lastStamp = drawable->dri2.stamp; 1011 1012 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1013 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1014 1015 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1016 1017 if (buffers == NULL) 1018 return; 1019 1020 for (i = 0; i < count; i++) { 1021 switch (buffers[i].attachment) { 1022 case __DRI_BUFFER_FRONT_LEFT: 1023 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1024 region_name = "dri2 front buffer"; 1025 break; 1026 1027 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1028 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1029 region_name = "dri2 fake front buffer"; 1030 break; 1031 1032 case __DRI_BUFFER_BACK_LEFT: 1033 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1034 region_name = "dri2 back buffer"; 1035 break; 1036 1037 case __DRI_BUFFER_DEPTH: 1038 case __DRI_BUFFER_HIZ: 1039 case __DRI_BUFFER_DEPTH_STENCIL: 1040 case __DRI_BUFFER_STENCIL: 1041 case __DRI_BUFFER_ACCUM: 1042 default: 1043 fprintf(stderr, 1044 "unhandled buffer attach event, attachment type %d\n", 1045 buffers[i].attachment); 1046 return; 1047 } 1048 1049 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1050 } 1051 1052} 1053 1054void 1055intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1056{ 1057 struct brw_context *brw = context->driverPrivate; 1058 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1059 1060 /* Set this up front, so that in case our buffers get invalidated 1061 * while we're getting new buffers, we don't clobber the stamp and 1062 * thus ignore the invalidate. */ 1063 drawable->lastStamp = drawable->dri2.stamp; 1064 1065 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1066 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1067 1068 if (screen->image.loader) 1069 intel_update_image_buffers(brw, drawable); 1070 else 1071 intel_update_dri2_buffers(brw, drawable); 1072 1073 driUpdateFramebufferSize(&brw->ctx, drawable); 1074} 1075 1076/** 1077 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1078 * state is required. 1079 */ 1080void 1081intel_prepare_render(struct brw_context *brw) 1082{ 1083 __DRIcontext *driContext = brw->driContext; 1084 __DRIdrawable *drawable; 1085 1086 drawable = driContext->driDrawablePriv; 1087 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1088 if (drawable->lastStamp != drawable->dri2.stamp) 1089 intel_update_renderbuffers(driContext, drawable); 1090 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1091 } 1092 1093 drawable = driContext->driReadablePriv; 1094 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1095 if (drawable->lastStamp != drawable->dri2.stamp) 1096 intel_update_renderbuffers(driContext, drawable); 1097 driContext->dri2.read_stamp = drawable->dri2.stamp; 1098 } 1099 1100 /* If we're currently rendering to the front buffer, the rendering 1101 * that will happen next will probably dirty the front buffer. So 1102 * mark it as dirty here. 1103 */ 1104 if (brw->is_front_buffer_rendering) 1105 brw->front_buffer_dirty = true; 1106 1107 /* Wait for the swapbuffers before the one we just emitted, so we 1108 * don't get too many swaps outstanding for apps that are GPU-heavy 1109 * but not CPU-heavy. 1110 * 1111 * We're using intelDRI2Flush (called from the loader before 1112 * swapbuffer) and glFlush (for front buffer rendering) as the 1113 * indicator that a frame is done and then throttle when we get 1114 * here as we prepare to render the next frame. At this point for 1115 * round trips for swap/copy and getting new buffers are done and 1116 * we'll spend less time waiting on the GPU. 1117 * 1118 * Unfortunately, we don't have a handle to the batch containing 1119 * the swap, and getting our hands on that doesn't seem worth it, 1120 * so we just us the first batch we emitted after the last swap. 1121 */ 1122 if (brw->need_throttle && brw->first_post_swapbuffers_batch) { 1123 if (!brw->disable_throttling) 1124 drm_intel_bo_wait_rendering(brw->first_post_swapbuffers_batch); 1125 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 1126 brw->first_post_swapbuffers_batch = NULL; 1127 brw->need_throttle = false; 1128 } 1129} 1130 1131/** 1132 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1133 * 1134 * To determine which DRI buffers to request, examine the renderbuffers 1135 * attached to the drawable's framebuffer. Then request the buffers with 1136 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1137 * 1138 * This is called from intel_update_renderbuffers(). 1139 * 1140 * \param drawable Drawable whose buffers are queried. 1141 * \param buffers [out] List of buffers returned by DRI2 query. 1142 * \param buffer_count [out] Number of buffers returned. 1143 * 1144 * \see intel_update_renderbuffers() 1145 * \see DRI2GetBuffers() 1146 * \see DRI2GetBuffersWithFormat() 1147 */ 1148static void 1149intel_query_dri2_buffers(struct brw_context *brw, 1150 __DRIdrawable *drawable, 1151 __DRIbuffer **buffers, 1152 int *buffer_count) 1153{ 1154 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1155 struct gl_framebuffer *fb = drawable->driverPrivate; 1156 int i = 0; 1157 unsigned attachments[8]; 1158 1159 struct intel_renderbuffer *front_rb; 1160 struct intel_renderbuffer *back_rb; 1161 1162 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1163 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1164 1165 memset(attachments, 0, sizeof(attachments)); 1166 if ((brw->is_front_buffer_rendering || 1167 brw->is_front_buffer_reading || 1168 !back_rb) && front_rb) { 1169 /* If a fake front buffer is in use, then querying for 1170 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1171 * the real front buffer to the fake front buffer. So before doing the 1172 * query, we need to make sure all the pending drawing has landed in the 1173 * real front buffer. 1174 */ 1175 intel_batchbuffer_flush(brw); 1176 intel_flush_front(&brw->ctx); 1177 1178 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1179 attachments[i++] = intel_bits_per_pixel(front_rb); 1180 } else if (front_rb && brw->front_buffer_dirty) { 1181 /* We have pending front buffer rendering, but we aren't querying for a 1182 * front buffer. If the front buffer we have is a fake front buffer, 1183 * the X server is going to throw it away when it processes the query. 1184 * So before doing the query, make sure all the pending drawing has 1185 * landed in the real front buffer. 1186 */ 1187 intel_batchbuffer_flush(brw); 1188 intel_flush_front(&brw->ctx); 1189 } 1190 1191 if (back_rb) { 1192 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1193 attachments[i++] = intel_bits_per_pixel(back_rb); 1194 } 1195 1196 assert(i <= ARRAY_SIZE(attachments)); 1197 1198 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1199 &drawable->w, 1200 &drawable->h, 1201 attachments, i / 2, 1202 buffer_count, 1203 drawable->loaderPrivate); 1204} 1205 1206/** 1207 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1208 * 1209 * This is called from intel_update_renderbuffers(). 1210 * 1211 * \par Note: 1212 * DRI buffers whose attachment point is DRI2BufferStencil or 1213 * DRI2BufferDepthStencil are handled as special cases. 1214 * 1215 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1216 * that is passed to intel_region_alloc_for_handle(). 1217 * 1218 * \see intel_update_renderbuffers() 1219 * \see intel_region_alloc_for_handle() 1220 */ 1221static void 1222intel_process_dri2_buffer(struct brw_context *brw, 1223 __DRIdrawable *drawable, 1224 __DRIbuffer *buffer, 1225 struct intel_renderbuffer *rb, 1226 const char *buffer_name) 1227{ 1228 struct intel_region *region = NULL; 1229 1230 if (!rb) 1231 return; 1232 1233 unsigned num_samples = rb->Base.Base.NumSamples; 1234 1235 /* We try to avoid closing and reopening the same BO name, because the first 1236 * use of a mapping of the buffer involves a bunch of page faulting which is 1237 * moderately expensive. 1238 */ 1239 if (num_samples == 0) { 1240 if (rb->mt && 1241 rb->mt->region && 1242 rb->mt->region->name == buffer->name) 1243 return; 1244 } else { 1245 if (rb->mt && 1246 rb->mt->singlesample_mt && 1247 rb->mt->singlesample_mt->region && 1248 rb->mt->singlesample_mt->region->name == buffer->name) 1249 return; 1250 } 1251 1252 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1253 fprintf(stderr, 1254 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1255 buffer->name, buffer->attachment, 1256 buffer->cpp, buffer->pitch); 1257 } 1258 1259 intel_miptree_release(&rb->mt); 1260 region = intel_region_alloc_for_handle(brw->intelScreen, 1261 buffer->cpp, 1262 drawable->w, 1263 drawable->h, 1264 buffer->pitch, 1265 buffer->name, 1266 buffer_name); 1267 if (!region) 1268 return; 1269 1270 rb->mt = intel_miptree_create_for_dri2_buffer(brw, 1271 buffer->attachment, 1272 intel_rb_format(rb), 1273 num_samples, 1274 region); 1275 intel_region_release(®ion); 1276} 1277 1278/** 1279 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1280 * 1281 * To determine which DRI buffers to request, examine the renderbuffers 1282 * attached to the drawable's framebuffer. Then request the buffers from 1283 * the image loader 1284 * 1285 * This is called from intel_update_renderbuffers(). 1286 * 1287 * \param drawable Drawable whose buffers are queried. 1288 * \param buffers [out] List of buffers returned by DRI2 query. 1289 * \param buffer_count [out] Number of buffers returned. 1290 * 1291 * \see intel_update_renderbuffers() 1292 */ 1293 1294static void 1295intel_update_image_buffer(struct brw_context *intel, 1296 __DRIdrawable *drawable, 1297 struct intel_renderbuffer *rb, 1298 __DRIimage *buffer, 1299 enum __DRIimageBufferMask buffer_type) 1300{ 1301 struct intel_region *region = buffer->region; 1302 1303 if (!rb || !region) 1304 return; 1305 1306 unsigned num_samples = rb->Base.Base.NumSamples; 1307 1308 /* Check and see if we're already bound to the right 1309 * buffer object 1310 */ 1311 if (num_samples == 0) { 1312 if (rb->mt && 1313 rb->mt->region && 1314 rb->mt->region->bo == region->bo) 1315 return; 1316 } else { 1317 if (rb->mt && 1318 rb->mt->singlesample_mt && 1319 rb->mt->singlesample_mt->region && 1320 rb->mt->singlesample_mt->region->bo == region->bo) 1321 return; 1322 } 1323 1324 intel_miptree_release(&rb->mt); 1325 rb->mt = intel_miptree_create_for_image_buffer(intel, 1326 buffer_type, 1327 intel_rb_format(rb), 1328 num_samples, 1329 region); 1330} 1331 1332static void 1333intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1334{ 1335 struct gl_framebuffer *fb = drawable->driverPrivate; 1336 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1337 struct intel_renderbuffer *front_rb; 1338 struct intel_renderbuffer *back_rb; 1339 struct __DRIimageList images; 1340 unsigned int format; 1341 uint32_t buffer_mask = 0; 1342 1343 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1344 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1345 1346 if (back_rb) 1347 format = intel_rb_format(back_rb); 1348 else if (front_rb) 1349 format = intel_rb_format(front_rb); 1350 else 1351 return; 1352 1353 if ((brw->is_front_buffer_rendering || brw->is_front_buffer_reading || !back_rb) && front_rb) 1354 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1355 1356 if (back_rb) 1357 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1358 1359 (*screen->image.loader->getBuffers) (drawable, 1360 driGLFormatToImageFormat(format), 1361 &drawable->dri2.stamp, 1362 drawable->loaderPrivate, 1363 buffer_mask, 1364 &images); 1365 1366 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1367 drawable->w = images.front->width; 1368 drawable->h = images.front->height; 1369 intel_update_image_buffer(brw, 1370 drawable, 1371 front_rb, 1372 images.front, 1373 __DRI_IMAGE_BUFFER_FRONT); 1374 } 1375 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1376 drawable->w = images.back->width; 1377 drawable->h = images.back->height; 1378 intel_update_image_buffer(brw, 1379 drawable, 1380 back_rb, 1381 images.back, 1382 __DRI_IMAGE_BUFFER_BACK); 1383 } 1384} 1385