brw_context.c revision 019560c127778dc3b08bdc7c3acdc68dbbb80ea0
1/* 2 Copyright 2003 VMware, Inc. 3 Copyright (C) Intel Corp. 2006. All Rights Reserved. 4 Intel funded Tungsten Graphics to 5 develop this 3D driver. 6 7 Permission is hereby granted, free of charge, to any person obtaining 8 a copy of this software and associated documentation files (the 9 "Software"), to deal in the Software without restriction, including 10 without limitation the rights to use, copy, modify, merge, publish, 11 distribute, sublicense, and/or sell copies of the Software, and to 12 permit persons to whom the Software is furnished to do so, subject to 13 the following conditions: 14 15 The above copyright notice and this permission notice (including the 16 next paragraph) shall be included in all copies or substantial 17 portions of the Software. 18 19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 27 **********************************************************************/ 28 /* 29 * Authors: 30 * Keith Whitwell <keithw@vmware.com> 31 */ 32 33 34#include "main/api_exec.h" 35#include "main/context.h" 36#include "main/fbobject.h" 37#include "main/imports.h" 38#include "main/macros.h" 39#include "main/points.h" 40#include "main/version.h" 41#include "main/vtxfmt.h" 42 43#include "vbo/vbo_context.h" 44 45#include "drivers/common/driverfuncs.h" 46#include "drivers/common/meta.h" 47#include "utils.h" 48 49#include "brw_context.h" 50#include "brw_defines.h" 51#include "brw_draw.h" 52#include "brw_state.h" 53 54#include "intel_batchbuffer.h" 55#include "intel_buffer_objects.h" 56#include "intel_buffers.h" 57#include "intel_fbo.h" 58#include "intel_mipmap_tree.h" 59#include "intel_pixel.h" 60#include "intel_regions.h" 61#include "intel_tex.h" 62#include "intel_tex_obj.h" 63 64#include "swrast_setup/swrast_setup.h" 65#include "tnl/tnl.h" 66#include "tnl/t_pipeline.h" 67#include "glsl/ralloc.h" 68 69/*************************************** 70 * Mesa's Driver Functions 71 ***************************************/ 72 73static size_t 74brw_query_samples_for_format(struct gl_context *ctx, GLenum target, 75 GLenum internalFormat, int samples[16]) 76{ 77 struct brw_context *brw = brw_context(ctx); 78 79 (void) target; 80 81 switch (brw->gen) { 82 case 7: 83 samples[0] = 8; 84 samples[1] = 4; 85 return 2; 86 87 case 6: 88 samples[0] = 4; 89 return 1; 90 91 default: 92 samples[0] = 1; 93 return 1; 94 } 95} 96 97const char *const brw_vendor_string = "Intel Open Source Technology Center"; 98 99const char * 100brw_get_renderer_string(unsigned deviceID) 101{ 102 const char *chipset; 103 static char buffer[128]; 104 105 switch (deviceID) { 106#undef CHIPSET 107#define CHIPSET(id, symbol, str) case id: chipset = str; break; 108#include "pci_ids/i965_pci_ids.h" 109 default: 110 chipset = "Unknown Intel Chipset"; 111 break; 112 } 113 114 (void) driGetRendererString(buffer, chipset, 0); 115 return buffer; 116} 117 118static const GLubyte * 119intelGetString(struct gl_context * ctx, GLenum name) 120{ 121 const struct brw_context *const brw = brw_context(ctx); 122 123 switch (name) { 124 case GL_VENDOR: 125 return (GLubyte *) brw_vendor_string; 126 127 case GL_RENDERER: 128 return 129 (GLubyte *) brw_get_renderer_string(brw->intelScreen->deviceID); 130 131 default: 132 return NULL; 133 } 134} 135 136static void 137intel_viewport(struct gl_context *ctx) 138{ 139 struct brw_context *brw = brw_context(ctx); 140 __DRIcontext *driContext = brw->driContext; 141 142 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) { 143 dri2InvalidateDrawable(driContext->driDrawablePriv); 144 dri2InvalidateDrawable(driContext->driReadablePriv); 145 } 146} 147 148static void 149intelInvalidateState(struct gl_context * ctx, GLuint new_state) 150{ 151 struct brw_context *brw = brw_context(ctx); 152 153 if (ctx->swrast_context) 154 _swrast_InvalidateState(ctx, new_state); 155 _vbo_InvalidateState(ctx, new_state); 156 157 brw->NewGLState |= new_state; 158} 159 160#define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer) 161 162static void 163intel_flush_front(struct gl_context *ctx) 164{ 165 struct brw_context *brw = brw_context(ctx); 166 __DRIcontext *driContext = brw->driContext; 167 __DRIdrawable *driDrawable = driContext->driDrawablePriv; 168 __DRIscreen *const screen = brw->intelScreen->driScrnPriv; 169 170 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) { 171 if (flushFront(screen) && driDrawable && 172 driDrawable->loaderPrivate) { 173 174 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT. 175 * 176 * This potentially resolves both front and back buffer. It 177 * is unnecessary to resolve the back, but harms nothing except 178 * performance. And no one cares about front-buffer render 179 * performance. 180 */ 181 intel_resolve_for_dri2_flush(brw, driDrawable); 182 intel_batchbuffer_flush(brw); 183 184 flushFront(screen)(driDrawable, driDrawable->loaderPrivate); 185 186 /* We set the dirty bit in intel_prepare_render() if we're 187 * front buffer rendering once we get there. 188 */ 189 brw->front_buffer_dirty = false; 190 } 191 } 192} 193 194static void 195intel_glFlush(struct gl_context *ctx) 196{ 197 struct brw_context *brw = brw_context(ctx); 198 199 intel_batchbuffer_flush(brw); 200 intel_flush_front(ctx); 201 if (brw->is_front_buffer_rendering) 202 brw->need_throttle = true; 203} 204 205void 206intelFinish(struct gl_context * ctx) 207{ 208 struct brw_context *brw = brw_context(ctx); 209 210 intel_glFlush(ctx); 211 212 if (brw->batch.last_bo) 213 drm_intel_bo_wait_rendering(brw->batch.last_bo); 214} 215 216static void 217brw_init_driver_functions(struct brw_context *brw, 218 struct dd_function_table *functions) 219{ 220 _mesa_init_driver_functions(functions); 221 222 /* GLX uses DRI2 invalidate events to handle window resizing. 223 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib), 224 * which doesn't provide a mechanism for snooping the event queues. 225 * 226 * So EGL still relies on viewport hacks to handle window resizing. 227 * This should go away with DRI3000. 228 */ 229 if (!brw->driContext->driScreenPriv->dri2.useInvalidate) 230 functions->Viewport = intel_viewport; 231 232 functions->Flush = intel_glFlush; 233 functions->Finish = intelFinish; 234 functions->GetString = intelGetString; 235 functions->UpdateState = intelInvalidateState; 236 237 intelInitTextureFuncs(functions); 238 intelInitTextureImageFuncs(functions); 239 intelInitTextureSubImageFuncs(functions); 240 intelInitTextureCopyImageFuncs(functions); 241 intelInitClearFuncs(functions); 242 intelInitBufferFuncs(functions); 243 intelInitPixelFuncs(functions); 244 intelInitBufferObjectFuncs(functions); 245 intel_init_syncobj_functions(functions); 246 brw_init_object_purgeable_functions(functions); 247 248 brwInitFragProgFuncs( functions ); 249 brw_init_common_queryobj_functions(functions); 250 if (brw->gen >= 6) 251 gen6_init_queryobj_functions(functions); 252 else 253 gen4_init_queryobj_functions(functions); 254 255 functions->QuerySamplesForFormat = brw_query_samples_for_format; 256 257 functions->NewTransformFeedback = brw_new_transform_feedback; 258 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 259 functions->GetTransformFeedbackVertexCount = 260 brw_get_transform_feedback_vertex_count; 261 if (brw->gen >= 7) { 262 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 263 functions->EndTransformFeedback = gen7_end_transform_feedback; 264 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 265 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 266 } else { 267 functions->BeginTransformFeedback = brw_begin_transform_feedback; 268 functions->EndTransformFeedback = brw_end_transform_feedback; 269 } 270 271 if (brw->gen >= 6) 272 functions->GetSamplePosition = gen6_get_sample_position; 273} 274 275static void 276brw_initialize_context_constants(struct brw_context *brw) 277{ 278 struct gl_context *ctx = &brw->ctx; 279 280 unsigned max_samplers = 281 brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16; 282 283 ctx->Const.QueryCounterBits.Timestamp = 36; 284 285 ctx->Const.StripTextureBorder = true; 286 287 ctx->Const.MaxDualSourceDrawBuffers = 1; 288 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 289 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits = max_samplers; 290 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 291 ctx->Const.MaxTextureUnits = 292 MIN2(ctx->Const.MaxTextureCoordUnits, 293 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits); 294 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits = max_samplers; 295 if (brw->gen >= 7) 296 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = max_samplers; 297 else 298 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = 0; 299 if (getenv("INTEL_COMPUTE_SHADER")) { 300 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 301 ctx->Const.MaxUniformBufferBindings += 12; 302 } else { 303 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits = 0; 304 } 305 ctx->Const.MaxCombinedTextureImageUnits = 306 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits + 307 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits + 308 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits + 309 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits; 310 311 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 312 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 313 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 314 ctx->Const.Max3DTextureLevels = 9; 315 ctx->Const.MaxCubeTextureLevels = 12; 316 317 if (brw->gen >= 7) 318 ctx->Const.MaxArrayTextureLayers = 2048; 319 else 320 ctx->Const.MaxArrayTextureLayers = 512; 321 322 ctx->Const.MaxTextureRectSize = 1 << 12; 323 324 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 325 326 ctx->Const.MaxRenderbufferSize = 8192; 327 328 /* Hardware only supports a limited number of transform feedback buffers. 329 * So we need to override the Mesa default (which is based only on software 330 * limits). 331 */ 332 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 333 334 /* On Gen6, in the worst case, we use up one binding table entry per 335 * transform feedback component (see comments above the definition of 336 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 337 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 338 * BRW_MAX_SOL_BINDINGS. 339 * 340 * In "separate components" mode, we need to divide this value by 341 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 342 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 343 */ 344 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 345 ctx->Const.MaxTransformFeedbackSeparateComponents = 346 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 347 348 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 349 350 int max_samples; 351 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 352 const int clamp_max_samples = 353 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 354 355 if (clamp_max_samples < 0) { 356 max_samples = msaa_modes[0]; 357 } else { 358 /* Select the largest supported MSAA mode that does not exceed 359 * clamp_max_samples. 360 */ 361 max_samples = 0; 362 for (int i = 0; msaa_modes[i] != 0; ++i) { 363 if (msaa_modes[i] <= clamp_max_samples) { 364 max_samples = msaa_modes[i]; 365 break; 366 } 367 } 368 } 369 370 ctx->Const.MaxSamples = max_samples; 371 ctx->Const.MaxColorTextureSamples = max_samples; 372 ctx->Const.MaxDepthTextureSamples = max_samples; 373 ctx->Const.MaxIntegerSamples = max_samples; 374 375 if (brw->gen >= 7) 376 ctx->Const.MaxProgramTextureGatherComponents = 4; 377 else if (brw->gen == 6) 378 ctx->Const.MaxProgramTextureGatherComponents = 1; 379 380 ctx->Const.MinLineWidth = 1.0; 381 ctx->Const.MinLineWidthAA = 1.0; 382 ctx->Const.MaxLineWidth = 5.0; 383 ctx->Const.MaxLineWidthAA = 5.0; 384 ctx->Const.LineWidthGranularity = 0.5; 385 386 ctx->Const.MinPointSize = 1.0; 387 ctx->Const.MinPointSizeAA = 1.0; 388 ctx->Const.MaxPointSize = 255.0; 389 ctx->Const.MaxPointSizeAA = 255.0; 390 ctx->Const.PointSizeGranularity = 1.0; 391 392 if (brw->gen >= 5 || brw->is_g4x) 393 ctx->Const.MaxClipPlanes = 8; 394 395 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024; 396 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0; 397 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0; 398 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0; 399 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0; 400 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0; 401 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0; 402 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16; 403 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256; 404 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1; 405 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024; 406 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams = 407 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters, 408 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams); 409 410 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024; 411 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024; 412 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024; 413 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024; 414 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12; 415 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256; 416 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0; 417 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024; 418 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams = 419 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters, 420 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams); 421 422 /* Fragment shaders use real, 32-bit twos-complement integers for all 423 * integer types. 424 */ 425 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31; 426 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30; 427 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0; 428 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 429 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 430 431 if (brw->gen >= 7) { 432 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 433 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 434 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 435 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 436 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicBuffers = BRW_MAX_ABO; 437 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicBuffers = BRW_MAX_ABO; 438 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicBuffers = BRW_MAX_ABO; 439 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxAtomicBuffers = BRW_MAX_ABO; 440 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 441 } 442 443 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 444 * but we're not sure how it's actually done for vertex order, 445 * that affect provoking vertex decision. Always use last vertex 446 * convention for quad primitive which works as expected for now. 447 */ 448 if (brw->gen >= 6) 449 ctx->Const.QuadsFollowProvokingVertexConvention = false; 450 451 ctx->Const.NativeIntegers = true; 452 ctx->Const.UniformBooleanTrue = 1; 453 454 /* From the gen4 PRM, volume 4 page 127: 455 * 456 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 457 * the base address of the first element of the surface, computed in 458 * software by adding the surface base address to the byte offset of 459 * the element in the buffer." 460 * 461 * However, unaligned accesses are slower, so enforce buffer alignment. 462 */ 463 ctx->Const.UniformBufferOffsetAlignment = 16; 464 ctx->Const.TextureBufferOffsetAlignment = 16; 465 466 if (brw->gen >= 6) { 467 ctx->Const.MaxVarying = 32; 468 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128; 469 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64; 470 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128; 471 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128; 472 } 473 474 /* We want the GLSL compiler to emit code that uses condition codes */ 475 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 476 ctx->ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 477 ctx->ShaderCompilerOptions[i].EmitCondCodes = true; 478 ctx->ShaderCompilerOptions[i].EmitNoNoise = true; 479 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true; 480 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true; 481 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true; 482 483 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform = 484 (i == MESA_SHADER_FRAGMENT); 485 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp = 486 (i == MESA_SHADER_FRAGMENT); 487 ctx->ShaderCompilerOptions[i].LowerClipDistance = true; 488 } 489 490 ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = true; 491 ctx->ShaderCompilerOptions[MESA_SHADER_GEOMETRY].OptimizeForAOS = true; 492 493 /* ARB_viewport_array */ 494 if (brw->gen >= 7 && ctx->API == API_OPENGL_CORE) { 495 ctx->Const.MaxViewports = GEN7_NUM_VIEWPORTS; 496 ctx->Const.ViewportSubpixelBits = 0; 497 498 /* Cast to float before negating becuase MaxViewportWidth is unsigned. 499 */ 500 ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth; 501 ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth; 502 } 503} 504 505/** 506 * Process driconf (drirc) options, setting appropriate context flags. 507 * 508 * intelInitExtensions still pokes at optionCache directly, in order to 509 * avoid advertising various extensions. No flags are set, so it makes 510 * sense to continue doing that there. 511 */ 512static void 513brw_process_driconf_options(struct brw_context *brw) 514{ 515 struct gl_context *ctx = &brw->ctx; 516 517 driOptionCache *options = &brw->optionCache; 518 driParseConfigFiles(options, &brw->intelScreen->optionCache, 519 brw->driContext->driScreenPriv->myNum, "i965"); 520 521 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 522 switch (bo_reuse_mode) { 523 case DRI_CONF_BO_REUSE_DISABLED: 524 break; 525 case DRI_CONF_BO_REUSE_ALL: 526 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 527 break; 528 } 529 530 if (!driQueryOptionb(options, "hiz")) { 531 brw->has_hiz = false; 532 /* On gen6, you can only do separate stencil with HIZ. */ 533 if (brw->gen == 6) 534 brw->has_separate_stencil = false; 535 } 536 537 if (driQueryOptionb(options, "always_flush_batch")) { 538 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 539 brw->always_flush_batch = true; 540 } 541 542 if (driQueryOptionb(options, "always_flush_cache")) { 543 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 544 brw->always_flush_cache = true; 545 } 546 547 if (driQueryOptionb(options, "disable_throttling")) { 548 fprintf(stderr, "disabling flush throttling\n"); 549 brw->disable_throttling = true; 550 } 551 552 brw->disable_derivative_optimization = 553 driQueryOptionb(&brw->optionCache, "disable_derivative_optimization"); 554 555 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 556 557 ctx->Const.ForceGLSLExtensionsWarn = 558 driQueryOptionb(options, "force_glsl_extensions_warn"); 559 560 ctx->Const.DisableGLSLLineContinuations = 561 driQueryOptionb(options, "disable_glsl_line_continuations"); 562} 563 564GLboolean 565brwCreateContext(gl_api api, 566 const struct gl_config *mesaVis, 567 __DRIcontext *driContextPriv, 568 unsigned major_version, 569 unsigned minor_version, 570 uint32_t flags, 571 bool notify_reset, 572 unsigned *dri_ctx_error, 573 void *sharedContextPrivate) 574{ 575 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 576 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 577 struct intel_screen *screen = sPriv->driverPrivate; 578 const struct brw_device_info *devinfo = screen->devinfo; 579 struct dd_function_table functions; 580 struct gl_config visual; 581 582 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 583 * provides us with context reset notifications. 584 */ 585 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 586 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 587 588 if (screen->has_context_reset_notification) 589 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 590 591 if (flags & ~allowed_flags) { 592 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 593 return false; 594 } 595 596 struct brw_context *brw = rzalloc(NULL, struct brw_context); 597 if (!brw) { 598 printf("%s: failed to alloc context\n", __FUNCTION__); 599 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 600 return false; 601 } 602 603 driContextPriv->driverPrivate = brw; 604 brw->driContext = driContextPriv; 605 brw->intelScreen = screen; 606 brw->bufmgr = screen->bufmgr; 607 608 brw->gen = devinfo->gen; 609 brw->gt = devinfo->gt; 610 brw->is_g4x = devinfo->is_g4x; 611 brw->is_baytrail = devinfo->is_baytrail; 612 brw->is_haswell = devinfo->is_haswell; 613 brw->has_llc = devinfo->has_llc; 614 brw->has_hiz = devinfo->has_hiz_and_separate_stencil && brw->gen < 8; 615 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 616 brw->has_pln = devinfo->has_pln; 617 brw->has_compr4 = devinfo->has_compr4; 618 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 619 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 620 brw->needs_unlit_centroid_workaround = 621 devinfo->needs_unlit_centroid_workaround; 622 623 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 624 brw->has_swizzling = screen->hw_has_swizzling; 625 626 if (brw->gen >= 8) { 627 gen8_init_vtable_surface_functions(brw); 628 gen7_init_vtable_sampler_functions(brw); 629 brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz; 630 } else if (brw->gen >= 7) { 631 gen7_init_vtable_surface_functions(brw); 632 gen7_init_vtable_sampler_functions(brw); 633 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 634 } else { 635 gen4_init_vtable_surface_functions(brw); 636 gen4_init_vtable_sampler_functions(brw); 637 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 638 } 639 640 brw_init_driver_functions(brw, &functions); 641 642 if (notify_reset) 643 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 644 645 struct gl_context *ctx = &brw->ctx; 646 647 if (mesaVis == NULL) { 648 memset(&visual, 0, sizeof visual); 649 mesaVis = &visual; 650 } 651 652 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 653 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 654 printf("%s: failed to init mesa context\n", __FUNCTION__); 655 intelDestroyContext(driContextPriv); 656 return false; 657 } 658 659 driContextSetFlags(ctx, flags); 660 661 /* Initialize the software rasterizer and helper modules. 662 * 663 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 664 * software fallbacks (which we have to support on legacy GL to do weird 665 * glDrawPixels(), glBitmap(), and other functions). 666 */ 667 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 668 _swrast_CreateContext(ctx); 669 } 670 671 _vbo_CreateContext(ctx); 672 if (ctx->swrast_context) { 673 _tnl_CreateContext(ctx); 674 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 675 _swsetup_CreateContext(ctx); 676 677 /* Configure swrast to match hardware characteristics: */ 678 _swrast_allow_pixel_fog(ctx, false); 679 _swrast_allow_vertex_fog(ctx, true); 680 } 681 682 _mesa_meta_init(ctx); 683 684 brw_process_driconf_options(brw); 685 brw_process_intel_debug_variable(brw); 686 brw_initialize_context_constants(brw); 687 688 ctx->Const.ResetStrategy = notify_reset 689 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 690 691 /* Reinitialize the context point state. It depends on ctx->Const values. */ 692 _mesa_init_point(ctx); 693 694 intel_batchbuffer_init(brw); 695 696 brw_init_state(brw); 697 698 intelInitExtensions(ctx); 699 700 intel_fbo_init(brw); 701 702 if (brw->gen >= 6) { 703 /* Create a new hardware context. Using a hardware context means that 704 * our GPU state will be saved/restored on context switch, allowing us 705 * to assume that the GPU is in the same state we left it in. 706 * 707 * This is required for transform feedback buffer offsets, query objects, 708 * and also allows us to reduce how much state we have to emit. 709 */ 710 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 711 712 if (!brw->hw_ctx) { 713 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 714 intelDestroyContext(driContextPriv); 715 return false; 716 } 717 } 718 719 brw_init_surface_formats(brw); 720 721 if (brw->is_g4x || brw->gen >= 5) { 722 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS; 723 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; 724 } else { 725 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS; 726 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; 727 } 728 729 brw->max_vs_threads = devinfo->max_vs_threads; 730 brw->max_gs_threads = devinfo->max_gs_threads; 731 brw->max_wm_threads = devinfo->max_wm_threads; 732 brw->urb.size = devinfo->urb.size; 733 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 734 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 735 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 736 737 /* Estimate the size of the mappable aperture into the GTT. There's an 738 * ioctl to get the whole GTT size, but not one to get the mappable subset. 739 * It turns out it's basically always 256MB, though some ancient hardware 740 * was smaller. 741 */ 742 uint32_t gtt_size = 256 * 1024 * 1024; 743 744 /* We don't want to map two objects such that a memcpy between them would 745 * just fault one mapping in and then the other over and over forever. So 746 * we would need to divide the GTT size by 2. Additionally, some GTT is 747 * taken up by things like the framebuffer and the ringbuffer and such, so 748 * be more conservative. 749 */ 750 brw->max_gtt_map_object_size = gtt_size / 4; 751 752 if (brw->gen == 6) 753 brw->urb.gen6_gs_previously_active = false; 754 755 brw->prim_restart.in_progress = false; 756 brw->prim_restart.enable_cut_index = false; 757 brw->gs.enabled = false; 758 759 if (brw->gen < 6) { 760 brw->curbe.last_buf = calloc(1, 4096); 761 brw->curbe.next_buf = calloc(1, 4096); 762 } 763 764 ctx->VertexProgram._MaintainTnlProgram = true; 765 ctx->FragmentProgram._MaintainTexEnvProgram = true; 766 767 brw_draw_init( brw ); 768 769 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 770 /* Turn on some extra GL_ARB_debug_output generation. */ 771 brw->perf_debug = true; 772 } 773 774 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 775 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 776 777 brw_fs_alloc_reg_sets(brw); 778 brw_vec4_alloc_reg_set(brw); 779 780 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 781 brw_init_shader_time(brw); 782 783 _mesa_compute_version(ctx); 784 785 _mesa_initialize_dispatch_tables(ctx); 786 _mesa_initialize_vbo_vtxfmt(ctx); 787 788 if (ctx->Extensions.AMD_performance_monitor) { 789 brw_init_performance_monitors(brw); 790 } 791 792 return true; 793} 794 795void 796intelDestroyContext(__DRIcontext * driContextPriv) 797{ 798 struct brw_context *brw = 799 (struct brw_context *) driContextPriv->driverPrivate; 800 struct gl_context *ctx = &brw->ctx; 801 802 assert(brw); /* should never be null */ 803 if (!brw) 804 return; 805 806 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 807 if (INTEL_DEBUG & DEBUG_AUB) { 808 intel_batchbuffer_flush(brw); 809 aub_dump_bmp(&brw->ctx); 810 } 811 812 _mesa_meta_free(&brw->ctx); 813 814 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 815 /* Force a report. */ 816 brw->shader_time.report_time = 0; 817 818 brw_collect_and_report_shader_time(brw); 819 brw_destroy_shader_time(brw); 820 } 821 822 brw_destroy_state(brw); 823 brw_draw_destroy(brw); 824 825 drm_intel_bo_unreference(brw->curbe.curbe_bo); 826 drm_intel_bo_unreference(brw->vs.base.const_bo); 827 drm_intel_bo_unreference(brw->wm.base.const_bo); 828 829 free(brw->curbe.last_buf); 830 free(brw->curbe.next_buf); 831 832 drm_intel_gem_context_destroy(brw->hw_ctx); 833 834 if (ctx->swrast_context) { 835 _swsetup_DestroyContext(&brw->ctx); 836 _tnl_DestroyContext(&brw->ctx); 837 } 838 _vbo_DestroyContext(&brw->ctx); 839 840 if (ctx->swrast_context) 841 _swrast_DestroyContext(&brw->ctx); 842 843 intel_batchbuffer_free(brw); 844 845 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 846 brw->first_post_swapbuffers_batch = NULL; 847 848 driDestroyOptionCache(&brw->optionCache); 849 850 /* free the Mesa context */ 851 _mesa_free_context_data(&brw->ctx); 852 853 ralloc_free(brw); 854 driContextPriv->driverPrivate = NULL; 855} 856 857GLboolean 858intelUnbindContext(__DRIcontext * driContextPriv) 859{ 860 /* Unset current context and dispath table */ 861 _mesa_make_current(NULL, NULL, NULL); 862 863 return true; 864} 865 866/** 867 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 868 * on window system framebuffers. 869 * 870 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 871 * your renderbuffer can do sRGB encode, and you can flip a switch that does 872 * sRGB encode if the renderbuffer can handle it. You can ask specifically 873 * for a visual where you're guaranteed to be capable, but it turns out that 874 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 875 * incapable ones, becuase there's no difference between the two in resources 876 * used. Applications thus get built that accidentally rely on the default 877 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 878 * great... 879 * 880 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 881 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 882 * So they removed the enable knob and made it "if the renderbuffer is sRGB 883 * capable, do sRGB encode". Then, for your window system renderbuffers, you 884 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 885 * and get no sRGB encode (assuming that both kinds of visual are available). 886 * Thus our choice to support sRGB by default on our visuals for desktop would 887 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 888 * 889 * Unfortunately, renderbuffer setup happens before a context is created. So 890 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 891 * context (without an sRGB visual, though we don't have sRGB visuals exposed 892 * yet), we go turn that back off before anyone finds out. 893 */ 894static void 895intel_gles3_srgb_workaround(struct brw_context *brw, 896 struct gl_framebuffer *fb) 897{ 898 struct gl_context *ctx = &brw->ctx; 899 900 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 901 return; 902 903 /* Some day when we support the sRGB capable bit on visuals available for 904 * GLES, we'll need to respect that and not disable things here. 905 */ 906 fb->Visual.sRGBCapable = false; 907 for (int i = 0; i < BUFFER_COUNT; i++) { 908 if (fb->Attachment[i].Renderbuffer && 909 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) { 910 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM; 911 } 912 } 913} 914 915GLboolean 916intelMakeCurrent(__DRIcontext * driContextPriv, 917 __DRIdrawable * driDrawPriv, 918 __DRIdrawable * driReadPriv) 919{ 920 struct brw_context *brw; 921 GET_CURRENT_CONTEXT(curCtx); 922 923 if (driContextPriv) 924 brw = (struct brw_context *) driContextPriv->driverPrivate; 925 else 926 brw = NULL; 927 928 /* According to the glXMakeCurrent() man page: "Pending commands to 929 * the previous context, if any, are flushed before it is released." 930 * But only flush if we're actually changing contexts. 931 */ 932 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 933 _mesa_flush(curCtx); 934 } 935 936 if (driContextPriv) { 937 struct gl_context *ctx = &brw->ctx; 938 struct gl_framebuffer *fb, *readFb; 939 940 if (driDrawPriv == NULL && driReadPriv == NULL) { 941 fb = _mesa_get_incomplete_framebuffer(); 942 readFb = _mesa_get_incomplete_framebuffer(); 943 } else { 944 fb = driDrawPriv->driverPrivate; 945 readFb = driReadPriv->driverPrivate; 946 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 947 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 948 } 949 950 /* The sRGB workaround changes the renderbuffer's format. We must change 951 * the format before the renderbuffer's miptree get's allocated, otherwise 952 * the formats of the renderbuffer and its miptree will differ. 953 */ 954 intel_gles3_srgb_workaround(brw, fb); 955 intel_gles3_srgb_workaround(brw, readFb); 956 957 /* If the context viewport hasn't been initialized, force a call out to 958 * the loader to get buffers so we have a drawable size for the initial 959 * viewport. */ 960 if (!brw->ctx.ViewportInitialized) 961 intel_prepare_render(brw); 962 963 _mesa_make_current(ctx, fb, readFb); 964 } else { 965 _mesa_make_current(NULL, NULL, NULL); 966 } 967 968 return true; 969} 970 971void 972intel_resolve_for_dri2_flush(struct brw_context *brw, 973 __DRIdrawable *drawable) 974{ 975 if (brw->gen < 6) { 976 /* MSAA and fast color clear are not supported, so don't waste time 977 * checking whether a resolve is needed. 978 */ 979 return; 980 } 981 982 struct gl_framebuffer *fb = drawable->driverPrivate; 983 struct intel_renderbuffer *rb; 984 985 /* Usually, only the back buffer will need to be downsampled. However, 986 * the front buffer will also need it if the user has rendered into it. 987 */ 988 static const gl_buffer_index buffers[2] = { 989 BUFFER_BACK_LEFT, 990 BUFFER_FRONT_LEFT, 991 }; 992 993 for (int i = 0; i < 2; ++i) { 994 rb = intel_get_renderbuffer(fb, buffers[i]); 995 if (rb == NULL || rb->mt == NULL) 996 continue; 997 if (rb->mt->num_samples <= 1) 998 intel_miptree_resolve_color(brw, rb->mt); 999 else 1000 intel_miptree_downsample(brw, rb->mt); 1001 } 1002} 1003 1004static unsigned 1005intel_bits_per_pixel(const struct intel_renderbuffer *rb) 1006{ 1007 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 1008} 1009 1010static void 1011intel_query_dri2_buffers(struct brw_context *brw, 1012 __DRIdrawable *drawable, 1013 __DRIbuffer **buffers, 1014 int *count); 1015 1016static void 1017intel_process_dri2_buffer(struct brw_context *brw, 1018 __DRIdrawable *drawable, 1019 __DRIbuffer *buffer, 1020 struct intel_renderbuffer *rb, 1021 const char *buffer_name); 1022 1023static void 1024intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 1025 1026static void 1027intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1028{ 1029 struct gl_framebuffer *fb = drawable->driverPrivate; 1030 struct intel_renderbuffer *rb; 1031 __DRIbuffer *buffers = NULL; 1032 int i, count; 1033 const char *region_name; 1034 1035 /* Set this up front, so that in case our buffers get invalidated 1036 * while we're getting new buffers, we don't clobber the stamp and 1037 * thus ignore the invalidate. */ 1038 drawable->lastStamp = drawable->dri2.stamp; 1039 1040 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1041 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1042 1043 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1044 1045 if (buffers == NULL) 1046 return; 1047 1048 for (i = 0; i < count; i++) { 1049 switch (buffers[i].attachment) { 1050 case __DRI_BUFFER_FRONT_LEFT: 1051 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1052 region_name = "dri2 front buffer"; 1053 break; 1054 1055 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1056 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1057 region_name = "dri2 fake front buffer"; 1058 break; 1059 1060 case __DRI_BUFFER_BACK_LEFT: 1061 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1062 region_name = "dri2 back buffer"; 1063 break; 1064 1065 case __DRI_BUFFER_DEPTH: 1066 case __DRI_BUFFER_HIZ: 1067 case __DRI_BUFFER_DEPTH_STENCIL: 1068 case __DRI_BUFFER_STENCIL: 1069 case __DRI_BUFFER_ACCUM: 1070 default: 1071 fprintf(stderr, 1072 "unhandled buffer attach event, attachment type %d\n", 1073 buffers[i].attachment); 1074 return; 1075 } 1076 1077 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1078 } 1079 1080} 1081 1082void 1083intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1084{ 1085 struct brw_context *brw = context->driverPrivate; 1086 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1087 1088 /* Set this up front, so that in case our buffers get invalidated 1089 * while we're getting new buffers, we don't clobber the stamp and 1090 * thus ignore the invalidate. */ 1091 drawable->lastStamp = drawable->dri2.stamp; 1092 1093 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1094 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1095 1096 if (screen->image.loader) 1097 intel_update_image_buffers(brw, drawable); 1098 else 1099 intel_update_dri2_buffers(brw, drawable); 1100 1101 driUpdateFramebufferSize(&brw->ctx, drawable); 1102} 1103 1104/** 1105 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1106 * state is required. 1107 */ 1108void 1109intel_prepare_render(struct brw_context *brw) 1110{ 1111 __DRIcontext *driContext = brw->driContext; 1112 __DRIdrawable *drawable; 1113 1114 drawable = driContext->driDrawablePriv; 1115 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1116 if (drawable->lastStamp != drawable->dri2.stamp) 1117 intel_update_renderbuffers(driContext, drawable); 1118 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1119 } 1120 1121 drawable = driContext->driReadablePriv; 1122 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1123 if (drawable->lastStamp != drawable->dri2.stamp) 1124 intel_update_renderbuffers(driContext, drawable); 1125 driContext->dri2.read_stamp = drawable->dri2.stamp; 1126 } 1127 1128 /* If we're currently rendering to the front buffer, the rendering 1129 * that will happen next will probably dirty the front buffer. So 1130 * mark it as dirty here. 1131 */ 1132 if (brw->is_front_buffer_rendering) 1133 brw->front_buffer_dirty = true; 1134 1135 /* Wait for the swapbuffers before the one we just emitted, so we 1136 * don't get too many swaps outstanding for apps that are GPU-heavy 1137 * but not CPU-heavy. 1138 * 1139 * We're using intelDRI2Flush (called from the loader before 1140 * swapbuffer) and glFlush (for front buffer rendering) as the 1141 * indicator that a frame is done and then throttle when we get 1142 * here as we prepare to render the next frame. At this point for 1143 * round trips for swap/copy and getting new buffers are done and 1144 * we'll spend less time waiting on the GPU. 1145 * 1146 * Unfortunately, we don't have a handle to the batch containing 1147 * the swap, and getting our hands on that doesn't seem worth it, 1148 * so we just us the first batch we emitted after the last swap. 1149 */ 1150 if (brw->need_throttle && brw->first_post_swapbuffers_batch) { 1151 if (!brw->disable_throttling) 1152 drm_intel_bo_wait_rendering(brw->first_post_swapbuffers_batch); 1153 drm_intel_bo_unreference(brw->first_post_swapbuffers_batch); 1154 brw->first_post_swapbuffers_batch = NULL; 1155 brw->need_throttle = false; 1156 } 1157} 1158 1159/** 1160 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1161 * 1162 * To determine which DRI buffers to request, examine the renderbuffers 1163 * attached to the drawable's framebuffer. Then request the buffers with 1164 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1165 * 1166 * This is called from intel_update_renderbuffers(). 1167 * 1168 * \param drawable Drawable whose buffers are queried. 1169 * \param buffers [out] List of buffers returned by DRI2 query. 1170 * \param buffer_count [out] Number of buffers returned. 1171 * 1172 * \see intel_update_renderbuffers() 1173 * \see DRI2GetBuffers() 1174 * \see DRI2GetBuffersWithFormat() 1175 */ 1176static void 1177intel_query_dri2_buffers(struct brw_context *brw, 1178 __DRIdrawable *drawable, 1179 __DRIbuffer **buffers, 1180 int *buffer_count) 1181{ 1182 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1183 struct gl_framebuffer *fb = drawable->driverPrivate; 1184 int i = 0; 1185 unsigned attachments[8]; 1186 1187 struct intel_renderbuffer *front_rb; 1188 struct intel_renderbuffer *back_rb; 1189 1190 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1191 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1192 1193 memset(attachments, 0, sizeof(attachments)); 1194 if ((brw->is_front_buffer_rendering || 1195 brw->is_front_buffer_reading || 1196 !back_rb) && front_rb) { 1197 /* If a fake front buffer is in use, then querying for 1198 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1199 * the real front buffer to the fake front buffer. So before doing the 1200 * query, we need to make sure all the pending drawing has landed in the 1201 * real front buffer. 1202 */ 1203 intel_batchbuffer_flush(brw); 1204 intel_flush_front(&brw->ctx); 1205 1206 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1207 attachments[i++] = intel_bits_per_pixel(front_rb); 1208 } else if (front_rb && brw->front_buffer_dirty) { 1209 /* We have pending front buffer rendering, but we aren't querying for a 1210 * front buffer. If the front buffer we have is a fake front buffer, 1211 * the X server is going to throw it away when it processes the query. 1212 * So before doing the query, make sure all the pending drawing has 1213 * landed in the real front buffer. 1214 */ 1215 intel_batchbuffer_flush(brw); 1216 intel_flush_front(&brw->ctx); 1217 } 1218 1219 if (back_rb) { 1220 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1221 attachments[i++] = intel_bits_per_pixel(back_rb); 1222 } 1223 1224 assert(i <= ARRAY_SIZE(attachments)); 1225 1226 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1227 &drawable->w, 1228 &drawable->h, 1229 attachments, i / 2, 1230 buffer_count, 1231 drawable->loaderPrivate); 1232} 1233 1234/** 1235 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1236 * 1237 * This is called from intel_update_renderbuffers(). 1238 * 1239 * \par Note: 1240 * DRI buffers whose attachment point is DRI2BufferStencil or 1241 * DRI2BufferDepthStencil are handled as special cases. 1242 * 1243 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1244 * that is passed to intel_region_alloc_for_handle(). 1245 * 1246 * \see intel_update_renderbuffers() 1247 * \see intel_region_alloc_for_handle() 1248 */ 1249static void 1250intel_process_dri2_buffer(struct brw_context *brw, 1251 __DRIdrawable *drawable, 1252 __DRIbuffer *buffer, 1253 struct intel_renderbuffer *rb, 1254 const char *buffer_name) 1255{ 1256 struct intel_region *region = NULL; 1257 1258 if (!rb) 1259 return; 1260 1261 unsigned num_samples = rb->Base.Base.NumSamples; 1262 1263 /* We try to avoid closing and reopening the same BO name, because the first 1264 * use of a mapping of the buffer involves a bunch of page faulting which is 1265 * moderately expensive. 1266 */ 1267 if (num_samples == 0) { 1268 if (rb->mt && 1269 rb->mt->region && 1270 rb->mt->region->name == buffer->name) 1271 return; 1272 } else { 1273 if (rb->mt && 1274 rb->mt->singlesample_mt && 1275 rb->mt->singlesample_mt->region && 1276 rb->mt->singlesample_mt->region->name == buffer->name) 1277 return; 1278 } 1279 1280 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1281 fprintf(stderr, 1282 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1283 buffer->name, buffer->attachment, 1284 buffer->cpp, buffer->pitch); 1285 } 1286 1287 intel_miptree_release(&rb->mt); 1288 region = intel_region_alloc_for_handle(brw->intelScreen, 1289 buffer->cpp, 1290 drawable->w, 1291 drawable->h, 1292 buffer->pitch, 1293 buffer->name, 1294 buffer_name); 1295 if (!region) { 1296 fprintf(stderr, 1297 "Failed to make region for returned DRI2 buffer " 1298 "(%dx%d, named %d).\n" 1299 "This is likely a bug in the X Server that will lead to a " 1300 "crash soon.\n", 1301 drawable->w, drawable->h, buffer->name); 1302 return; 1303 } 1304 1305 intel_update_winsys_renderbuffer_miptree(brw, rb, region); 1306 1307 if (brw->is_front_buffer_rendering && 1308 (buffer->attachment == __DRI_BUFFER_FRONT_LEFT || 1309 buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) && 1310 rb->Base.Base.NumSamples > 1) { 1311 intel_miptree_upsample(brw, rb->mt); 1312 } 1313 1314 assert(rb->mt); 1315 1316 intel_region_release(®ion); 1317} 1318 1319/** 1320 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1321 * 1322 * To determine which DRI buffers to request, examine the renderbuffers 1323 * attached to the drawable's framebuffer. Then request the buffers from 1324 * the image loader 1325 * 1326 * This is called from intel_update_renderbuffers(). 1327 * 1328 * \param drawable Drawable whose buffers are queried. 1329 * \param buffers [out] List of buffers returned by DRI2 query. 1330 * \param buffer_count [out] Number of buffers returned. 1331 * 1332 * \see intel_update_renderbuffers() 1333 */ 1334 1335static void 1336intel_update_image_buffer(struct brw_context *intel, 1337 __DRIdrawable *drawable, 1338 struct intel_renderbuffer *rb, 1339 __DRIimage *buffer, 1340 enum __DRIimageBufferMask buffer_type) 1341{ 1342 struct intel_region *region = buffer->region; 1343 1344 if (!rb || !region) 1345 return; 1346 1347 unsigned num_samples = rb->Base.Base.NumSamples; 1348 1349 /* Check and see if we're already bound to the right 1350 * buffer object 1351 */ 1352 if (num_samples == 0) { 1353 if (rb->mt && 1354 rb->mt->region && 1355 rb->mt->region->bo == region->bo) 1356 return; 1357 } else { 1358 if (rb->mt && 1359 rb->mt->singlesample_mt && 1360 rb->mt->singlesample_mt->region && 1361 rb->mt->singlesample_mt->region->bo == region->bo) 1362 return; 1363 } 1364 1365 intel_update_winsys_renderbuffer_miptree(intel, rb, region); 1366 1367 if (intel->is_front_buffer_rendering && 1368 buffer_type == __DRI_IMAGE_BUFFER_FRONT && 1369 rb->Base.Base.NumSamples > 1) { 1370 intel_miptree_upsample(intel, rb->mt); 1371 } 1372} 1373 1374static void 1375intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1376{ 1377 struct gl_framebuffer *fb = drawable->driverPrivate; 1378 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1379 struct intel_renderbuffer *front_rb; 1380 struct intel_renderbuffer *back_rb; 1381 struct __DRIimageList images; 1382 unsigned int format; 1383 uint32_t buffer_mask = 0; 1384 1385 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1386 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1387 1388 if (back_rb) 1389 format = intel_rb_format(back_rb); 1390 else if (front_rb) 1391 format = intel_rb_format(front_rb); 1392 else 1393 return; 1394 1395 if ((brw->is_front_buffer_rendering || brw->is_front_buffer_reading || !back_rb) && front_rb) 1396 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1397 1398 if (back_rb) 1399 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1400 1401 (*screen->image.loader->getBuffers) (drawable, 1402 driGLFormatToImageFormat(format), 1403 &drawable->dri2.stamp, 1404 drawable->loaderPrivate, 1405 buffer_mask, 1406 &images); 1407 1408 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1409 drawable->w = images.front->width; 1410 drawable->h = images.front->height; 1411 intel_update_image_buffer(brw, 1412 drawable, 1413 front_rb, 1414 images.front, 1415 __DRI_IMAGE_BUFFER_FRONT); 1416 } 1417 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1418 drawable->w = images.back->width; 1419 drawable->h = images.back->height; 1420 intel_update_image_buffer(brw, 1421 drawable, 1422 back_rb, 1423 images.back, 1424 __DRI_IMAGE_BUFFER_BACK); 1425 } 1426} 1427