brw_context.c revision 665b8d7b6d8eae03c9dc0ef1a744fe59d9cc6cb6
1/* 2 Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 3 Copyright (C) Intel Corp. 2006. All Rights Reserved. 4 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to 5 develop this 3D driver. 6 7 Permission is hereby granted, free of charge, to any person obtaining 8 a copy of this software and associated documentation files (the 9 "Software"), to deal in the Software without restriction, including 10 without limitation the rights to use, copy, modify, merge, publish, 11 distribute, sublicense, and/or sell copies of the Software, and to 12 permit persons to whom the Software is furnished to do so, subject to 13 the following conditions: 14 15 The above copyright notice and this permission notice (including the 16 next paragraph) shall be included in all copies or substantial 17 portions of the Software. 18 19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 27 **********************************************************************/ 28 /* 29 * Authors: 30 * Keith Whitwell <keith@tungstengraphics.com> 31 */ 32 33 34#include "main/api_exec.h" 35#include "main/context.h" 36#include "main/fbobject.h" 37#include "main/imports.h" 38#include "main/macros.h" 39#include "main/points.h" 40#include "main/version.h" 41#include "main/vtxfmt.h" 42 43#include "vbo/vbo_context.h" 44 45#include "drivers/common/driverfuncs.h" 46#include "drivers/common/meta.h" 47#include "utils.h" 48 49#include "brw_context.h" 50#include "brw_defines.h" 51#include "brw_draw.h" 52#include "brw_state.h" 53 54#include "intel_batchbuffer.h" 55#include "intel_buffer_objects.h" 56#include "intel_buffers.h" 57#include "intel_fbo.h" 58#include "intel_mipmap_tree.h" 59#include "intel_pixel.h" 60#include "intel_regions.h" 61#include "intel_tex.h" 62#include "intel_tex_obj.h" 63 64#include "swrast_setup/swrast_setup.h" 65#include "tnl/tnl.h" 66#include "tnl/t_pipeline.h" 67#include "glsl/ralloc.h" 68 69/*************************************** 70 * Mesa's Driver Functions 71 ***************************************/ 72 73static size_t 74brw_query_samples_for_format(struct gl_context *ctx, GLenum target, 75 GLenum internalFormat, int samples[16]) 76{ 77 struct brw_context *brw = brw_context(ctx); 78 79 (void) target; 80 81 switch (brw->gen) { 82 case 7: 83 samples[0] = 8; 84 samples[1] = 4; 85 return 2; 86 87 case 6: 88 samples[0] = 4; 89 return 1; 90 91 default: 92 samples[0] = 1; 93 return 1; 94 } 95} 96 97const char *const brw_vendor_string = "Intel Open Source Technology Center"; 98 99const char * 100brw_get_renderer_string(unsigned deviceID) 101{ 102 const char *chipset; 103 static char buffer[128]; 104 105 switch (deviceID) { 106#undef CHIPSET 107#define CHIPSET(id, symbol, str) case id: chipset = str; break; 108#include "pci_ids/i965_pci_ids.h" 109 default: 110 chipset = "Unknown Intel Chipset"; 111 break; 112 } 113 114 (void) driGetRendererString(buffer, chipset, 0); 115 return buffer; 116} 117 118static const GLubyte * 119intelGetString(struct gl_context * ctx, GLenum name) 120{ 121 const struct brw_context *const brw = brw_context(ctx); 122 123 switch (name) { 124 case GL_VENDOR: 125 return (GLubyte *) brw_vendor_string; 126 127 case GL_RENDERER: 128 return 129 (GLubyte *) brw_get_renderer_string(brw->intelScreen->deviceID); 130 131 default: 132 return NULL; 133 } 134} 135 136static void 137intel_viewport(struct gl_context *ctx, GLint x, GLint y, GLsizei w, GLsizei h) 138{ 139 struct brw_context *brw = brw_context(ctx); 140 __DRIcontext *driContext = brw->driContext; 141 142 (void) x; 143 (void) y; 144 (void) w; 145 (void) h; 146 147 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) { 148 dri2InvalidateDrawable(driContext->driDrawablePriv); 149 dri2InvalidateDrawable(driContext->driReadablePriv); 150 } 151} 152 153static void 154intelInvalidateState(struct gl_context * ctx, GLuint new_state) 155{ 156 struct brw_context *brw = brw_context(ctx); 157 158 if (ctx->swrast_context) 159 _swrast_InvalidateState(ctx, new_state); 160 _vbo_InvalidateState(ctx, new_state); 161 162 brw->NewGLState |= new_state; 163} 164 165#define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer) 166 167static void 168intel_flush_front(struct gl_context *ctx) 169{ 170 struct brw_context *brw = brw_context(ctx); 171 __DRIcontext *driContext = brw->driContext; 172 __DRIdrawable *driDrawable = driContext->driDrawablePriv; 173 __DRIscreen *const screen = brw->intelScreen->driScrnPriv; 174 175 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) { 176 if (flushFront(screen) && driDrawable && 177 driDrawable->loaderPrivate) { 178 179 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT. 180 * 181 * This potentially resolves both front and back buffer. It 182 * is unnecessary to resolve the back, but harms nothing except 183 * performance. And no one cares about front-buffer render 184 * performance. 185 */ 186 intel_resolve_for_dri2_flush(brw, driDrawable); 187 intel_batchbuffer_flush(brw); 188 189 flushFront(screen)(driDrawable, driDrawable->loaderPrivate); 190 191 /* We set the dirty bit in intel_prepare_render() if we're 192 * front buffer rendering once we get there. 193 */ 194 brw->front_buffer_dirty = false; 195 } 196 } 197} 198 199static void 200intel_glFlush(struct gl_context *ctx) 201{ 202 struct brw_context *brw = brw_context(ctx); 203 204 intel_batchbuffer_flush(brw); 205 intel_flush_front(ctx); 206 if (brw->is_front_buffer_rendering) 207 brw->need_throttle = true; 208} 209 210void 211intelFinish(struct gl_context * ctx) 212{ 213 struct brw_context *brw = brw_context(ctx); 214 215 intel_glFlush(ctx); 216 217 if (brw->batch.last_bo) 218 drm_intel_bo_wait_rendering(brw->batch.last_bo); 219} 220 221static void 222brw_init_driver_functions(struct brw_context *brw, 223 struct dd_function_table *functions) 224{ 225 _mesa_init_driver_functions(functions); 226 227 /* GLX uses DRI2 invalidate events to handle window resizing. 228 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib), 229 * which doesn't provide a mechanism for snooping the event queues. 230 * 231 * So EGL still relies on viewport hacks to handle window resizing. 232 * This should go away with DRI3000. 233 */ 234 if (!brw->driContext->driScreenPriv->dri2.useInvalidate) 235 functions->Viewport = intel_viewport; 236 237 functions->Flush = intel_glFlush; 238 functions->Finish = intelFinish; 239 functions->GetString = intelGetString; 240 functions->UpdateState = intelInvalidateState; 241 242 intelInitTextureFuncs(functions); 243 intelInitTextureImageFuncs(functions); 244 intelInitTextureSubImageFuncs(functions); 245 intelInitTextureCopyImageFuncs(functions); 246 intelInitClearFuncs(functions); 247 intelInitBufferFuncs(functions); 248 intelInitPixelFuncs(functions); 249 intelInitBufferObjectFuncs(functions); 250 intel_init_syncobj_functions(functions); 251 brw_init_object_purgeable_functions(functions); 252 253 brwInitFragProgFuncs( functions ); 254 brw_init_common_queryobj_functions(functions); 255 if (brw->gen >= 6) 256 gen6_init_queryobj_functions(functions); 257 else 258 gen4_init_queryobj_functions(functions); 259 260 functions->QuerySamplesForFormat = brw_query_samples_for_format; 261 262 functions->NewTransformFeedback = brw_new_transform_feedback; 263 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 264 functions->GetTransformFeedbackVertexCount = 265 brw_get_transform_feedback_vertex_count; 266 if (brw->gen >= 7) { 267 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 268 functions->EndTransformFeedback = gen7_end_transform_feedback; 269 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 270 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 271 } else { 272 functions->BeginTransformFeedback = brw_begin_transform_feedback; 273 functions->EndTransformFeedback = brw_end_transform_feedback; 274 } 275 276 if (brw->gen >= 6) 277 functions->GetSamplePosition = gen6_get_sample_position; 278} 279 280static void 281brw_initialize_context_constants(struct brw_context *brw) 282{ 283 struct gl_context *ctx = &brw->ctx; 284 285 ctx->Const.QueryCounterBits.Timestamp = 36; 286 287 ctx->Const.StripTextureBorder = true; 288 289 ctx->Const.MaxDualSourceDrawBuffers = 1; 290 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 291 ctx->Const.FragmentProgram.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 292 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 293 ctx->Const.MaxTextureUnits = 294 MIN2(ctx->Const.MaxTextureCoordUnits, 295 ctx->Const.FragmentProgram.MaxTextureImageUnits); 296 ctx->Const.VertexProgram.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 297 if (brw->gen >= 7) 298 ctx->Const.GeometryProgram.MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 299 else 300 ctx->Const.GeometryProgram.MaxTextureImageUnits = 0; 301 ctx->Const.MaxCombinedTextureImageUnits = 302 ctx->Const.VertexProgram.MaxTextureImageUnits + 303 ctx->Const.FragmentProgram.MaxTextureImageUnits + 304 ctx->Const.GeometryProgram.MaxTextureImageUnits; 305 306 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 307 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 308 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 309 ctx->Const.Max3DTextureLevels = 9; 310 ctx->Const.MaxCubeTextureLevels = 12; 311 312 if (brw->gen >= 7) 313 ctx->Const.MaxArrayTextureLayers = 2048; 314 else 315 ctx->Const.MaxArrayTextureLayers = 512; 316 317 ctx->Const.MaxTextureRectSize = 1 << 12; 318 319 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 320 321 ctx->Const.MaxRenderbufferSize = 8192; 322 323 /* Hardware only supports a limited number of transform feedback buffers. 324 * So we need to override the Mesa default (which is based only on software 325 * limits). 326 */ 327 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 328 329 /* On Gen6, in the worst case, we use up one binding table entry per 330 * transform feedback component (see comments above the definition of 331 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 332 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 333 * BRW_MAX_SOL_BINDINGS. 334 * 335 * In "separate components" mode, we need to divide this value by 336 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 337 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 338 */ 339 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 340 ctx->Const.MaxTransformFeedbackSeparateComponents = 341 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 342 343 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 344 345 int max_samples; 346 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 347 const int clamp_max_samples = 348 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 349 350 if (clamp_max_samples < 0) { 351 max_samples = msaa_modes[0]; 352 } else { 353 /* Select the largest supported MSAA mode that does not exceed 354 * clamp_max_samples. 355 */ 356 max_samples = 0; 357 for (int i = 0; msaa_modes[i] != 0; ++i) { 358 if (msaa_modes[i] <= clamp_max_samples) { 359 max_samples = msaa_modes[i]; 360 break; 361 } 362 } 363 } 364 365 ctx->Const.MaxSamples = max_samples; 366 ctx->Const.MaxColorTextureSamples = max_samples; 367 ctx->Const.MaxDepthTextureSamples = max_samples; 368 ctx->Const.MaxIntegerSamples = max_samples; 369 370 if (brw->gen >= 7) 371 ctx->Const.MaxProgramTextureGatherComponents = 4; 372 373 ctx->Const.MinLineWidth = 1.0; 374 ctx->Const.MinLineWidthAA = 1.0; 375 ctx->Const.MaxLineWidth = 5.0; 376 ctx->Const.MaxLineWidthAA = 5.0; 377 ctx->Const.LineWidthGranularity = 0.5; 378 379 ctx->Const.MinPointSize = 1.0; 380 ctx->Const.MinPointSizeAA = 1.0; 381 ctx->Const.MaxPointSize = 255.0; 382 ctx->Const.MaxPointSizeAA = 255.0; 383 ctx->Const.PointSizeGranularity = 1.0; 384 385 if (brw->gen >= 5 || brw->is_g4x) 386 ctx->Const.MaxClipPlanes = 8; 387 388 ctx->Const.VertexProgram.MaxNativeInstructions = 16 * 1024; 389 ctx->Const.VertexProgram.MaxAluInstructions = 0; 390 ctx->Const.VertexProgram.MaxTexInstructions = 0; 391 ctx->Const.VertexProgram.MaxTexIndirections = 0; 392 ctx->Const.VertexProgram.MaxNativeAluInstructions = 0; 393 ctx->Const.VertexProgram.MaxNativeTexInstructions = 0; 394 ctx->Const.VertexProgram.MaxNativeTexIndirections = 0; 395 ctx->Const.VertexProgram.MaxNativeAttribs = 16; 396 ctx->Const.VertexProgram.MaxNativeTemps = 256; 397 ctx->Const.VertexProgram.MaxNativeAddressRegs = 1; 398 ctx->Const.VertexProgram.MaxNativeParameters = 1024; 399 ctx->Const.VertexProgram.MaxEnvParams = 400 MIN2(ctx->Const.VertexProgram.MaxNativeParameters, 401 ctx->Const.VertexProgram.MaxEnvParams); 402 403 ctx->Const.FragmentProgram.MaxNativeInstructions = 1024; 404 ctx->Const.FragmentProgram.MaxNativeAluInstructions = 1024; 405 ctx->Const.FragmentProgram.MaxNativeTexInstructions = 1024; 406 ctx->Const.FragmentProgram.MaxNativeTexIndirections = 1024; 407 ctx->Const.FragmentProgram.MaxNativeAttribs = 12; 408 ctx->Const.FragmentProgram.MaxNativeTemps = 256; 409 ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; 410 ctx->Const.FragmentProgram.MaxNativeParameters = 1024; 411 ctx->Const.FragmentProgram.MaxEnvParams = 412 MIN2(ctx->Const.FragmentProgram.MaxNativeParameters, 413 ctx->Const.FragmentProgram.MaxEnvParams); 414 415 /* Fragment shaders use real, 32-bit twos-complement integers for all 416 * integer types. 417 */ 418 ctx->Const.FragmentProgram.LowInt.RangeMin = 31; 419 ctx->Const.FragmentProgram.LowInt.RangeMax = 30; 420 ctx->Const.FragmentProgram.LowInt.Precision = 0; 421 ctx->Const.FragmentProgram.HighInt = ctx->Const.FragmentProgram.LowInt; 422 ctx->Const.FragmentProgram.MediumInt = ctx->Const.FragmentProgram.LowInt; 423 424 if (brw->gen >= 7) { 425 ctx->Const.FragmentProgram.MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 426 ctx->Const.VertexProgram.MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 427 ctx->Const.GeometryProgram.MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 428 ctx->Const.FragmentProgram.MaxAtomicBuffers = BRW_MAX_ABO; 429 ctx->Const.VertexProgram.MaxAtomicBuffers = BRW_MAX_ABO; 430 ctx->Const.GeometryProgram.MaxAtomicBuffers = BRW_MAX_ABO; 431 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 432 } 433 434 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 435 * but we're not sure how it's actually done for vertex order, 436 * that affect provoking vertex decision. Always use last vertex 437 * convention for quad primitive which works as expected for now. 438 */ 439 if (brw->gen >= 6) 440 ctx->Const.QuadsFollowProvokingVertexConvention = false; 441 442 ctx->Const.NativeIntegers = true; 443 ctx->Const.UniformBooleanTrue = 1; 444 445 /* From the gen4 PRM, volume 4 page 127: 446 * 447 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 448 * the base address of the first element of the surface, computed in 449 * software by adding the surface base address to the byte offset of 450 * the element in the buffer." 451 * 452 * However, unaligned accesses are slower, so enforce buffer alignment. 453 */ 454 ctx->Const.UniformBufferOffsetAlignment = 16; 455 ctx->Const.TextureBufferOffsetAlignment = 16; 456 457 if (brw->gen >= 6) { 458 ctx->Const.MaxVarying = 32; 459 ctx->Const.VertexProgram.MaxOutputComponents = 128; 460 ctx->Const.GeometryProgram.MaxInputComponents = 64; 461 ctx->Const.GeometryProgram.MaxOutputComponents = 128; 462 ctx->Const.FragmentProgram.MaxInputComponents = 128; 463 } 464 465 /* We want the GLSL compiler to emit code that uses condition codes */ 466 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 467 ctx->ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 468 ctx->ShaderCompilerOptions[i].EmitCondCodes = true; 469 ctx->ShaderCompilerOptions[i].EmitNoNoise = true; 470 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true; 471 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true; 472 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true; 473 474 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform = 475 (i == MESA_SHADER_FRAGMENT); 476 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp = 477 (i == MESA_SHADER_FRAGMENT); 478 ctx->ShaderCompilerOptions[i].LowerClipDistance = true; 479 } 480 481 ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].PreferDP4 = true; 482} 483 484/** 485 * Process driconf (drirc) options, setting appropriate context flags. 486 * 487 * intelInitExtensions still pokes at optionCache directly, in order to 488 * avoid advertising various extensions. No flags are set, so it makes 489 * sense to continue doing that there. 490 */ 491static void 492brw_process_driconf_options(struct brw_context *brw) 493{ 494 struct gl_context *ctx = &brw->ctx; 495 496 driOptionCache *options = &brw->optionCache; 497 driParseConfigFiles(options, &brw->intelScreen->optionCache, 498 brw->driContext->driScreenPriv->myNum, "i965"); 499 500 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 501 switch (bo_reuse_mode) { 502 case DRI_CONF_BO_REUSE_DISABLED: 503 break; 504 case DRI_CONF_BO_REUSE_ALL: 505 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 506 break; 507 } 508 509 if (!driQueryOptionb(options, "hiz")) { 510 brw->has_hiz = false; 511 /* On gen6, you can only do separate stencil with HIZ. */ 512 if (brw->gen == 6) 513 brw->has_separate_stencil = false; 514 } 515 516 if (driQueryOptionb(options, "always_flush_batch")) { 517 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 518 brw->always_flush_batch = true; 519 } 520 521 if (driQueryOptionb(options, "always_flush_cache")) { 522 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 523 brw->always_flush_cache = true; 524 } 525 526 if (driQueryOptionb(options, "disable_throttling")) { 527 fprintf(stderr, "disabling flush throttling\n"); 528 brw->disable_throttling = true; 529 } 530 531 brw->disable_derivative_optimization = 532 driQueryOptionb(&brw->optionCache, "disable_derivative_optimization"); 533 534 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 535 536 ctx->Const.ForceGLSLExtensionsWarn = 537 driQueryOptionb(options, "force_glsl_extensions_warn"); 538 539 ctx->Const.DisableGLSLLineContinuations = 540 driQueryOptionb(options, "disable_glsl_line_continuations"); 541} 542 543GLboolean 544brwCreateContext(gl_api api, 545 const struct gl_config *mesaVis, 546 __DRIcontext *driContextPriv, 547 unsigned major_version, 548 unsigned minor_version, 549 uint32_t flags, 550 bool notify_reset, 551 unsigned *dri_ctx_error, 552 void *sharedContextPrivate) 553{ 554 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 555 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 556 struct intel_screen *screen = sPriv->driverPrivate; 557 const struct brw_device_info *devinfo = screen->devinfo; 558 struct dd_function_table functions; 559 struct gl_config visual; 560 561 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 562 * provides us with context reset notifications. 563 */ 564 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 565 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 566 567 if (screen->has_context_reset_notification) 568 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 569 570 if (flags & ~allowed_flags) { 571 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 572 return false; 573 } 574 575 struct brw_context *brw = rzalloc(NULL, struct brw_context); 576 if (!brw) { 577 printf("%s: failed to alloc context\n", __FUNCTION__); 578 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 579 return false; 580 } 581 582 driContextPriv->driverPrivate = brw; 583 brw->driContext = driContextPriv; 584 brw->intelScreen = screen; 585 brw->bufmgr = screen->bufmgr; 586 587 brw->gen = devinfo->gen; 588 brw->gt = devinfo->gt; 589 brw->is_g4x = devinfo->is_g4x; 590 brw->is_baytrail = devinfo->is_baytrail; 591 brw->is_haswell = devinfo->is_haswell; 592 brw->has_llc = devinfo->has_llc; 593 brw->has_hiz = devinfo->has_hiz_and_separate_stencil && brw->gen < 8; 594 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 595 brw->has_pln = devinfo->has_pln; 596 brw->has_compr4 = devinfo->has_compr4; 597 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 598 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 599 brw->needs_unlit_centroid_workaround = 600 devinfo->needs_unlit_centroid_workaround; 601 602 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 603 brw->has_swizzling = screen->hw_has_swizzling; 604 605 if (brw->gen >= 7) { 606 gen7_init_vtable_surface_functions(brw); 607 gen7_init_vtable_sampler_functions(brw); 608 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 609 } else { 610 gen4_init_vtable_surface_functions(brw); 611 gen4_init_vtable_sampler_functions(brw); 612 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 613 } 614 615 brw_init_driver_functions(brw, &functions); 616 617 if (notify_reset) 618 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 619 620 struct gl_context *ctx = &brw->ctx; 621 622 if (mesaVis == NULL) { 623 memset(&visual, 0, sizeof visual); 624 mesaVis = &visual; 625 } 626 627 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 628 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 629 printf("%s: failed to init mesa context\n", __FUNCTION__); 630 intelDestroyContext(driContextPriv); 631 return false; 632 } 633 634 driContextSetFlags(ctx, flags); 635 636 /* Initialize the software rasterizer and helper modules. 637 * 638 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 639 * software fallbacks (which we have to support on legacy GL to do weird 640 * glDrawPixels(), glBitmap(), and other functions). 641 */ 642 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 643 _swrast_CreateContext(ctx); 644 } 645 646 _vbo_CreateContext(ctx); 647 if (ctx->swrast_context) { 648 _tnl_CreateContext(ctx); 649 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 650 _swsetup_CreateContext(ctx); 651 652 /* Configure swrast to match hardware characteristics: */ 653 _swrast_allow_pixel_fog(ctx, false); 654 _swrast_allow_vertex_fog(ctx, true); 655 } 656 657 _mesa_meta_init(ctx); 658 659 brw_process_driconf_options(brw); 660 brw_process_intel_debug_variable(brw); 661 brw_initialize_context_constants(brw); 662 663 ctx->Const.ResetStrategy = notify_reset 664 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 665 666 /* Reinitialize the context point state. It depends on ctx->Const values. */ 667 _mesa_init_point(ctx); 668 669 intel_batchbuffer_init(brw); 670 671 brw_init_state(brw); 672 673 intelInitExtensions(ctx); 674 675 intel_fbo_init(brw); 676 677 if (brw->gen >= 6) { 678 /* Create a new hardware context. Using a hardware context means that 679 * our GPU state will be saved/restored on context switch, allowing us 680 * to assume that the GPU is in the same state we left it in. 681 * 682 * This is required for transform feedback buffer offsets, query objects, 683 * and also allows us to reduce how much state we have to emit. 684 */ 685 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 686 687 if (!brw->hw_ctx) { 688 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 689 intelDestroyContext(driContextPriv); 690 return false; 691 } 692 } 693 694 brw_init_surface_formats(brw); 695 696 if (brw->is_g4x || brw->gen >= 5) { 697 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS; 698 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; 699 } else { 700 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS; 701 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; 702 } 703 704 brw->max_vs_threads = devinfo->max_vs_threads; 705 brw->max_gs_threads = devinfo->max_gs_threads; 706 brw->max_wm_threads = devinfo->max_wm_threads; 707 brw->urb.size = devinfo->urb.size; 708 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 709 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 710 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 711 712 /* Estimate the size of the mappable aperture into the GTT. There's an 713 * ioctl to get the whole GTT size, but not one to get the mappable subset. 714 * It turns out it's basically always 256MB, though some ancient hardware 715 * was smaller. 716 */ 717 uint32_t gtt_size = 256 * 1024 * 1024; 718 719 /* We don't want to map two objects such that a memcpy between them would 720 * just fault one mapping in and then the other over and over forever. So 721 * we would need to divide the GTT size by 2. Additionally, some GTT is 722 * taken up by things like the framebuffer and the ringbuffer and such, so 723 * be more conservative. 724 */ 725 brw->max_gtt_map_object_size = gtt_size / 4; 726 727 if (brw->gen == 6) 728 brw->urb.gen6_gs_previously_active = false; 729 730 brw->prim_restart.in_progress = false; 731 brw->prim_restart.enable_cut_index = false; 732 brw->gs.enabled = false; 733 734 if (brw->gen < 6) { 735 brw->curbe.last_buf = calloc(1, 4096); 736 brw->curbe.next_buf = calloc(1, 4096); 737 } 738 739 ctx->VertexProgram._MaintainTnlProgram = true; 740 ctx->FragmentProgram._MaintainTexEnvProgram = true; 741 742 brw_draw_init( brw ); 743 744 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 745 /* Turn on some extra GL_ARB_debug_output generation. */ 746 brw->perf_debug = true; 747 } 748 749 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 750 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 751 752 brw_fs_alloc_reg_sets(brw); 753 brw_vec4_alloc_reg_set(brw); 754 755 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 756 brw_init_shader_time(brw); 757 758 _mesa_compute_version(ctx); 759 760 _mesa_initialize_dispatch_tables(ctx); 761 _mesa_initialize_vbo_vtxfmt(ctx); 762 763 if (ctx->Extensions.AMD_performance_monitor) { 764 brw_init_performance_monitors(brw); 765 } 766 767 return true; 768} 769 770void 771intelDestroyContext(__DRIcontext * driContextPriv) 772{ 773 struct brw_context *brw = 774 (struct brw_context *) driContextPriv->driverPrivate; 775 struct gl_context *ctx = &brw->ctx; 776 777 assert(brw); /* should never be null */ 778 if (!brw) 779 return; 780 781 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 782 if (INTEL_DEBUG & DEBUG_AUB) { 783 intel_batchbuffer_flush(brw); 784 aub_dump_bmp(&brw->ctx); 785 } 786 787 _mesa_meta_free(&brw->ctx); 788 789 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 790 /* Force a report. */ 791 brw->shader_time.report_time = 0; 792 793 brw_collect_and_report_shader_time(brw); 794 brw_destroy_shader_time(brw); 795 } 796 797 brw_destroy_state(brw); 798 brw_draw_destroy(brw); 799 800 drm_intel_bo_unreference(brw->curbe.curbe_bo); 801 drm_intel_bo_unreference(brw->vs.base.const_bo); 802 drm_intel_bo_unreference(brw->wm.base.const_bo); 803 804 free(brw->curbe.last_buf); 805 free(brw->curbe.next_buf); 806 807 drm_intel_gem_context_destroy(brw->hw_ctx); 808 809 if (ctx->swrast_context) { 810 _swsetup_DestroyContext(&brw->ctx); 811 _tnl_DestroyContext(&brw->ctx); 812 } 813 _vbo_DestroyContext(&brw->ctx); 814 815 if (ctx->swrast_context) 816 _swrast_DestroyContext(&brw->ctx); 817 818 intel_batchbuffer_free(brw); 819 820 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 821 brw->first_post_swapbuffers_batch = NULL; 822 823 driDestroyOptionCache(&brw->optionCache); 824 825 /* free the Mesa context */ 826 _mesa_free_context_data(&brw->ctx); 827 828 ralloc_free(brw); 829 driContextPriv->driverPrivate = NULL; 830} 831 832GLboolean 833intelUnbindContext(__DRIcontext * driContextPriv) 834{ 835 /* Unset current context and dispath table */ 836 _mesa_make_current(NULL, NULL, NULL); 837 838 return true; 839} 840 841/** 842 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 843 * on window system framebuffers. 844 * 845 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 846 * your renderbuffer can do sRGB encode, and you can flip a switch that does 847 * sRGB encode if the renderbuffer can handle it. You can ask specifically 848 * for a visual where you're guaranteed to be capable, but it turns out that 849 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 850 * incapable ones, becuase there's no difference between the two in resources 851 * used. Applications thus get built that accidentally rely on the default 852 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 853 * great... 854 * 855 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 856 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 857 * So they removed the enable knob and made it "if the renderbuffer is sRGB 858 * capable, do sRGB encode". Then, for your window system renderbuffers, you 859 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 860 * and get no sRGB encode (assuming that both kinds of visual are available). 861 * Thus our choice to support sRGB by default on our visuals for desktop would 862 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 863 * 864 * Unfortunately, renderbuffer setup happens before a context is created. So 865 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 866 * context (without an sRGB visual, though we don't have sRGB visuals exposed 867 * yet), we go turn that back off before anyone finds out. 868 */ 869static void 870intel_gles3_srgb_workaround(struct brw_context *brw, 871 struct gl_framebuffer *fb) 872{ 873 struct gl_context *ctx = &brw->ctx; 874 875 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 876 return; 877 878 /* Some day when we support the sRGB capable bit on visuals available for 879 * GLES, we'll need to respect that and not disable things here. 880 */ 881 fb->Visual.sRGBCapable = false; 882 for (int i = 0; i < BUFFER_COUNT; i++) { 883 if (fb->Attachment[i].Renderbuffer && 884 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_SARGB8) { 885 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_ARGB8888; 886 } 887 } 888} 889 890GLboolean 891intelMakeCurrent(__DRIcontext * driContextPriv, 892 __DRIdrawable * driDrawPriv, 893 __DRIdrawable * driReadPriv) 894{ 895 struct brw_context *brw; 896 GET_CURRENT_CONTEXT(curCtx); 897 898 if (driContextPriv) 899 brw = (struct brw_context *) driContextPriv->driverPrivate; 900 else 901 brw = NULL; 902 903 /* According to the glXMakeCurrent() man page: "Pending commands to 904 * the previous context, if any, are flushed before it is released." 905 * But only flush if we're actually changing contexts. 906 */ 907 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 908 _mesa_flush(curCtx); 909 } 910 911 if (driContextPriv) { 912 struct gl_context *ctx = &brw->ctx; 913 struct gl_framebuffer *fb, *readFb; 914 915 if (driDrawPriv == NULL && driReadPriv == NULL) { 916 fb = _mesa_get_incomplete_framebuffer(); 917 readFb = _mesa_get_incomplete_framebuffer(); 918 } else { 919 fb = driDrawPriv->driverPrivate; 920 readFb = driReadPriv->driverPrivate; 921 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 922 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 923 } 924 925 /* The sRGB workaround changes the renderbuffer's format. We must change 926 * the format before the renderbuffer's miptree get's allocated, otherwise 927 * the formats of the renderbuffer and its miptree will differ. 928 */ 929 intel_gles3_srgb_workaround(brw, fb); 930 intel_gles3_srgb_workaround(brw, readFb); 931 932 intel_prepare_render(brw); 933 _mesa_make_current(ctx, fb, readFb); 934 } else { 935 _mesa_make_current(NULL, NULL, NULL); 936 } 937 938 return true; 939} 940 941void 942intel_resolve_for_dri2_flush(struct brw_context *brw, 943 __DRIdrawable *drawable) 944{ 945 if (brw->gen < 6) { 946 /* MSAA and fast color clear are not supported, so don't waste time 947 * checking whether a resolve is needed. 948 */ 949 return; 950 } 951 952 struct gl_framebuffer *fb = drawable->driverPrivate; 953 struct intel_renderbuffer *rb; 954 955 /* Usually, only the back buffer will need to be downsampled. However, 956 * the front buffer will also need it if the user has rendered into it. 957 */ 958 static const gl_buffer_index buffers[2] = { 959 BUFFER_BACK_LEFT, 960 BUFFER_FRONT_LEFT, 961 }; 962 963 for (int i = 0; i < 2; ++i) { 964 rb = intel_get_renderbuffer(fb, buffers[i]); 965 if (rb == NULL || rb->mt == NULL) 966 continue; 967 if (rb->mt->num_samples <= 1) 968 intel_miptree_resolve_color(brw, rb->mt); 969 else 970 intel_miptree_downsample(brw, rb->mt); 971 } 972} 973 974static unsigned 975intel_bits_per_pixel(const struct intel_renderbuffer *rb) 976{ 977 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 978} 979 980static void 981intel_query_dri2_buffers(struct brw_context *brw, 982 __DRIdrawable *drawable, 983 __DRIbuffer **buffers, 984 int *count); 985 986static void 987intel_process_dri2_buffer(struct brw_context *brw, 988 __DRIdrawable *drawable, 989 __DRIbuffer *buffer, 990 struct intel_renderbuffer *rb, 991 const char *buffer_name); 992 993static void 994intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 995 996static void 997intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 998{ 999 struct gl_framebuffer *fb = drawable->driverPrivate; 1000 struct intel_renderbuffer *rb; 1001 __DRIbuffer *buffers = NULL; 1002 int i, count; 1003 const char *region_name; 1004 1005 /* Set this up front, so that in case our buffers get invalidated 1006 * while we're getting new buffers, we don't clobber the stamp and 1007 * thus ignore the invalidate. */ 1008 drawable->lastStamp = drawable->dri2.stamp; 1009 1010 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1011 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1012 1013 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1014 1015 if (buffers == NULL) 1016 return; 1017 1018 for (i = 0; i < count; i++) { 1019 switch (buffers[i].attachment) { 1020 case __DRI_BUFFER_FRONT_LEFT: 1021 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1022 region_name = "dri2 front buffer"; 1023 break; 1024 1025 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1026 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1027 region_name = "dri2 fake front buffer"; 1028 break; 1029 1030 case __DRI_BUFFER_BACK_LEFT: 1031 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1032 region_name = "dri2 back buffer"; 1033 break; 1034 1035 case __DRI_BUFFER_DEPTH: 1036 case __DRI_BUFFER_HIZ: 1037 case __DRI_BUFFER_DEPTH_STENCIL: 1038 case __DRI_BUFFER_STENCIL: 1039 case __DRI_BUFFER_ACCUM: 1040 default: 1041 fprintf(stderr, 1042 "unhandled buffer attach event, attachment type %d\n", 1043 buffers[i].attachment); 1044 return; 1045 } 1046 1047 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1048 } 1049 1050} 1051 1052void 1053intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1054{ 1055 struct brw_context *brw = context->driverPrivate; 1056 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1057 1058 /* Set this up front, so that in case our buffers get invalidated 1059 * while we're getting new buffers, we don't clobber the stamp and 1060 * thus ignore the invalidate. */ 1061 drawable->lastStamp = drawable->dri2.stamp; 1062 1063 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1064 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1065 1066 if (screen->image.loader) 1067 intel_update_image_buffers(brw, drawable); 1068 else 1069 intel_update_dri2_buffers(brw, drawable); 1070 1071 driUpdateFramebufferSize(&brw->ctx, drawable); 1072} 1073 1074/** 1075 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1076 * state is required. 1077 */ 1078void 1079intel_prepare_render(struct brw_context *brw) 1080{ 1081 __DRIcontext *driContext = brw->driContext; 1082 __DRIdrawable *drawable; 1083 1084 drawable = driContext->driDrawablePriv; 1085 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1086 if (drawable->lastStamp != drawable->dri2.stamp) 1087 intel_update_renderbuffers(driContext, drawable); 1088 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1089 } 1090 1091 drawable = driContext->driReadablePriv; 1092 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1093 if (drawable->lastStamp != drawable->dri2.stamp) 1094 intel_update_renderbuffers(driContext, drawable); 1095 driContext->dri2.read_stamp = drawable->dri2.stamp; 1096 } 1097 1098 /* If we're currently rendering to the front buffer, the rendering 1099 * that will happen next will probably dirty the front buffer. So 1100 * mark it as dirty here. 1101 */ 1102 if (brw->is_front_buffer_rendering) 1103 brw->front_buffer_dirty = true; 1104 1105 /* Wait for the swapbuffers before the one we just emitted, so we 1106 * don't get too many swaps outstanding for apps that are GPU-heavy 1107 * but not CPU-heavy. 1108 * 1109 * We're using intelDRI2Flush (called from the loader before 1110 * swapbuffer) and glFlush (for front buffer rendering) as the 1111 * indicator that a frame is done and then throttle when we get 1112 * here as we prepare to render the next frame. At this point for 1113 * round trips for swap/copy and getting new buffers are done and 1114 * we'll spend less time waiting on the GPU. 1115 * 1116 * Unfortunately, we don't have a handle to the batch containing 1117 * the swap, and getting our hands on that doesn't seem worth it, 1118 * so we just us the first batch we emitted after the last swap. 1119 */ 1120 if (brw->need_throttle && brw->first_post_swapbuffers_batch) { 1121 if (!brw->disable_throttling) 1122 drm_intel_bo_wait_rendering(brw->first_post_swapbuffers_batch); 1123 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 1124 brw->first_post_swapbuffers_batch = NULL; 1125 brw->need_throttle = false; 1126 } 1127} 1128 1129/** 1130 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1131 * 1132 * To determine which DRI buffers to request, examine the renderbuffers 1133 * attached to the drawable's framebuffer. Then request the buffers with 1134 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1135 * 1136 * This is called from intel_update_renderbuffers(). 1137 * 1138 * \param drawable Drawable whose buffers are queried. 1139 * \param buffers [out] List of buffers returned by DRI2 query. 1140 * \param buffer_count [out] Number of buffers returned. 1141 * 1142 * \see intel_update_renderbuffers() 1143 * \see DRI2GetBuffers() 1144 * \see DRI2GetBuffersWithFormat() 1145 */ 1146static void 1147intel_query_dri2_buffers(struct brw_context *brw, 1148 __DRIdrawable *drawable, 1149 __DRIbuffer **buffers, 1150 int *buffer_count) 1151{ 1152 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1153 struct gl_framebuffer *fb = drawable->driverPrivate; 1154 int i = 0; 1155 unsigned attachments[8]; 1156 1157 struct intel_renderbuffer *front_rb; 1158 struct intel_renderbuffer *back_rb; 1159 1160 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1161 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1162 1163 memset(attachments, 0, sizeof(attachments)); 1164 if ((brw->is_front_buffer_rendering || 1165 brw->is_front_buffer_reading || 1166 !back_rb) && front_rb) { 1167 /* If a fake front buffer is in use, then querying for 1168 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1169 * the real front buffer to the fake front buffer. So before doing the 1170 * query, we need to make sure all the pending drawing has landed in the 1171 * real front buffer. 1172 */ 1173 intel_batchbuffer_flush(brw); 1174 intel_flush_front(&brw->ctx); 1175 1176 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1177 attachments[i++] = intel_bits_per_pixel(front_rb); 1178 } else if (front_rb && brw->front_buffer_dirty) { 1179 /* We have pending front buffer rendering, but we aren't querying for a 1180 * front buffer. If the front buffer we have is a fake front buffer, 1181 * the X server is going to throw it away when it processes the query. 1182 * So before doing the query, make sure all the pending drawing has 1183 * landed in the real front buffer. 1184 */ 1185 intel_batchbuffer_flush(brw); 1186 intel_flush_front(&brw->ctx); 1187 } 1188 1189 if (back_rb) { 1190 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1191 attachments[i++] = intel_bits_per_pixel(back_rb); 1192 } 1193 1194 assert(i <= ARRAY_SIZE(attachments)); 1195 1196 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1197 &drawable->w, 1198 &drawable->h, 1199 attachments, i / 2, 1200 buffer_count, 1201 drawable->loaderPrivate); 1202} 1203 1204/** 1205 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1206 * 1207 * This is called from intel_update_renderbuffers(). 1208 * 1209 * \par Note: 1210 * DRI buffers whose attachment point is DRI2BufferStencil or 1211 * DRI2BufferDepthStencil are handled as special cases. 1212 * 1213 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1214 * that is passed to intel_region_alloc_for_handle(). 1215 * 1216 * \see intel_update_renderbuffers() 1217 * \see intel_region_alloc_for_handle() 1218 */ 1219static void 1220intel_process_dri2_buffer(struct brw_context *brw, 1221 __DRIdrawable *drawable, 1222 __DRIbuffer *buffer, 1223 struct intel_renderbuffer *rb, 1224 const char *buffer_name) 1225{ 1226 struct intel_region *region = NULL; 1227 1228 if (!rb) 1229 return; 1230 1231 unsigned num_samples = rb->Base.Base.NumSamples; 1232 1233 /* We try to avoid closing and reopening the same BO name, because the first 1234 * use of a mapping of the buffer involves a bunch of page faulting which is 1235 * moderately expensive. 1236 */ 1237 if (num_samples == 0) { 1238 if (rb->mt && 1239 rb->mt->region && 1240 rb->mt->region->name == buffer->name) 1241 return; 1242 } else { 1243 if (rb->mt && 1244 rb->mt->singlesample_mt && 1245 rb->mt->singlesample_mt->region && 1246 rb->mt->singlesample_mt->region->name == buffer->name) 1247 return; 1248 } 1249 1250 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1251 fprintf(stderr, 1252 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1253 buffer->name, buffer->attachment, 1254 buffer->cpp, buffer->pitch); 1255 } 1256 1257 intel_miptree_release(&rb->mt); 1258 region = intel_region_alloc_for_handle(brw->intelScreen, 1259 buffer->cpp, 1260 drawable->w, 1261 drawable->h, 1262 buffer->pitch, 1263 buffer->name, 1264 buffer_name); 1265 if (!region) 1266 return; 1267 1268 rb->mt = intel_miptree_create_for_dri2_buffer(brw, 1269 buffer->attachment, 1270 intel_rb_format(rb), 1271 num_samples, 1272 region); 1273 intel_region_release(®ion); 1274} 1275 1276/** 1277 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1278 * 1279 * To determine which DRI buffers to request, examine the renderbuffers 1280 * attached to the drawable's framebuffer. Then request the buffers from 1281 * the image loader 1282 * 1283 * This is called from intel_update_renderbuffers(). 1284 * 1285 * \param drawable Drawable whose buffers are queried. 1286 * \param buffers [out] List of buffers returned by DRI2 query. 1287 * \param buffer_count [out] Number of buffers returned. 1288 * 1289 * \see intel_update_renderbuffers() 1290 */ 1291 1292static void 1293intel_update_image_buffer(struct brw_context *intel, 1294 __DRIdrawable *drawable, 1295 struct intel_renderbuffer *rb, 1296 __DRIimage *buffer, 1297 enum __DRIimageBufferMask buffer_type) 1298{ 1299 struct intel_region *region = buffer->region; 1300 1301 if (!rb || !region) 1302 return; 1303 1304 unsigned num_samples = rb->Base.Base.NumSamples; 1305 1306 /* Check and see if we're already bound to the right 1307 * buffer object 1308 */ 1309 if (num_samples == 0) { 1310 if (rb->mt && 1311 rb->mt->region && 1312 rb->mt->region->bo == region->bo) 1313 return; 1314 } else { 1315 if (rb->mt && 1316 rb->mt->singlesample_mt && 1317 rb->mt->singlesample_mt->region && 1318 rb->mt->singlesample_mt->region->bo == region->bo) 1319 return; 1320 } 1321 1322 intel_miptree_release(&rb->mt); 1323 rb->mt = intel_miptree_create_for_image_buffer(intel, 1324 buffer_type, 1325 intel_rb_format(rb), 1326 num_samples, 1327 region); 1328} 1329 1330static void 1331intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1332{ 1333 struct gl_framebuffer *fb = drawable->driverPrivate; 1334 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1335 struct intel_renderbuffer *front_rb; 1336 struct intel_renderbuffer *back_rb; 1337 struct __DRIimageList images; 1338 unsigned int format; 1339 uint32_t buffer_mask = 0; 1340 1341 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1342 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1343 1344 if (back_rb) 1345 format = intel_rb_format(back_rb); 1346 else if (front_rb) 1347 format = intel_rb_format(front_rb); 1348 else 1349 return; 1350 1351 if ((brw->is_front_buffer_rendering || brw->is_front_buffer_reading || !back_rb) && front_rb) 1352 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1353 1354 if (back_rb) 1355 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1356 1357 (*screen->image.loader->getBuffers) (drawable, 1358 driGLFormatToImageFormat(format), 1359 &drawable->dri2.stamp, 1360 drawable->loaderPrivate, 1361 buffer_mask, 1362 &images); 1363 1364 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1365 drawable->w = images.front->width; 1366 drawable->h = images.front->height; 1367 intel_update_image_buffer(brw, 1368 drawable, 1369 front_rb, 1370 images.front, 1371 __DRI_IMAGE_BUFFER_FRONT); 1372 } 1373 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1374 drawable->w = images.back->width; 1375 drawable->h = images.back->height; 1376 intel_update_image_buffer(brw, 1377 drawable, 1378 back_rb, 1379 images.back, 1380 __DRI_IMAGE_BUFFER_BACK); 1381 } 1382} 1383