brw_context.c revision 05e7f7f4388bde882b7ce74124000a4d435affff
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/extensions.h" 38#include "main/imports.h" 39#include "main/macros.h" 40#include "main/points.h" 41#include "main/version.h" 42#include "main/vtxfmt.h" 43#include "main/texobj.h" 44 45#include "vbo/vbo_context.h" 46 47#include "drivers/common/driverfuncs.h" 48#include "drivers/common/meta.h" 49#include "utils.h" 50 51#include "brw_context.h" 52#include "brw_defines.h" 53#include "brw_draw.h" 54#include "brw_state.h" 55 56#include "intel_batchbuffer.h" 57#include "intel_buffer_objects.h" 58#include "intel_buffers.h" 59#include "intel_fbo.h" 60#include "intel_mipmap_tree.h" 61#include "intel_pixel.h" 62#include "intel_image.h" 63#include "intel_tex.h" 64#include "intel_tex_obj.h" 65 66#include "swrast_setup/swrast_setup.h" 67#include "tnl/tnl.h" 68#include "tnl/t_pipeline.h" 69#include "util/ralloc.h" 70 71#include "glsl/nir/nir.h" 72 73/*************************************** 74 * Mesa's Driver Functions 75 ***************************************/ 76 77static size_t 78brw_query_samples_for_format(struct gl_context *ctx, GLenum target, 79 GLenum internalFormat, int samples[16]) 80{ 81 struct brw_context *brw = brw_context(ctx); 82 83 (void) target; 84 85 switch (brw->gen) { 86 case 9: 87 case 8: 88 samples[0] = 8; 89 samples[1] = 4; 90 samples[2] = 2; 91 return 3; 92 93 case 7: 94 samples[0] = 8; 95 samples[1] = 4; 96 return 2; 97 98 case 6: 99 samples[0] = 4; 100 return 1; 101 102 default: 103 assert(brw->gen < 6); 104 samples[0] = 1; 105 return 1; 106 } 107} 108 109const char *const brw_vendor_string = "Intel Open Source Technology Center"; 110 111const char * 112brw_get_renderer_string(unsigned deviceID) 113{ 114 const char *chipset; 115 static char buffer[128]; 116 117 switch (deviceID) { 118#undef CHIPSET 119#define CHIPSET(id, symbol, str) case id: chipset = str; break; 120#include "pci_ids/i965_pci_ids.h" 121 default: 122 chipset = "Unknown Intel Chipset"; 123 break; 124 } 125 126 (void) driGetRendererString(buffer, chipset, 0); 127 return buffer; 128} 129 130static const GLubyte * 131intel_get_string(struct gl_context * ctx, GLenum name) 132{ 133 const struct brw_context *const brw = brw_context(ctx); 134 135 switch (name) { 136 case GL_VENDOR: 137 return (GLubyte *) brw_vendor_string; 138 139 case GL_RENDERER: 140 return 141 (GLubyte *) brw_get_renderer_string(brw->intelScreen->deviceID); 142 143 default: 144 return NULL; 145 } 146} 147 148static void 149intel_viewport(struct gl_context *ctx) 150{ 151 struct brw_context *brw = brw_context(ctx); 152 __DRIcontext *driContext = brw->driContext; 153 154 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) { 155 dri2InvalidateDrawable(driContext->driDrawablePriv); 156 dri2InvalidateDrawable(driContext->driReadablePriv); 157 } 158} 159 160static void 161intel_update_state(struct gl_context * ctx, GLuint new_state) 162{ 163 struct brw_context *brw = brw_context(ctx); 164 struct intel_texture_object *tex_obj; 165 struct intel_renderbuffer *depth_irb; 166 167 if (ctx->swrast_context) 168 _swrast_InvalidateState(ctx, new_state); 169 _vbo_InvalidateState(ctx, new_state); 170 171 brw->NewGLState |= new_state; 172 173 _mesa_unlock_context_textures(ctx); 174 175 /* Resolve the depth buffer's HiZ buffer. */ 176 depth_irb = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH); 177 if (depth_irb) 178 intel_renderbuffer_resolve_hiz(brw, depth_irb); 179 180 /* Resolve depth buffer and render cache of each enabled texture. */ 181 int maxEnabledUnit = ctx->Texture._MaxEnabledTexImageUnit; 182 for (int i = 0; i <= maxEnabledUnit; i++) { 183 if (!ctx->Texture.Unit[i]._Current) 184 continue; 185 tex_obj = intel_texture_object(ctx->Texture.Unit[i]._Current); 186 if (!tex_obj || !tex_obj->mt) 187 continue; 188 intel_miptree_all_slices_resolve_depth(brw, tex_obj->mt); 189 intel_miptree_resolve_color(brw, tex_obj->mt); 190 brw_render_cache_set_check_flush(brw, tex_obj->mt->bo); 191 } 192 193 _mesa_lock_context_textures(ctx); 194} 195 196#define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer) 197 198static void 199intel_flush_front(struct gl_context *ctx) 200{ 201 struct brw_context *brw = brw_context(ctx); 202 __DRIcontext *driContext = brw->driContext; 203 __DRIdrawable *driDrawable = driContext->driDrawablePriv; 204 __DRIscreen *const screen = brw->intelScreen->driScrnPriv; 205 206 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) { 207 if (flushFront(screen) && driDrawable && 208 driDrawable->loaderPrivate) { 209 210 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT. 211 * 212 * This potentially resolves both front and back buffer. It 213 * is unnecessary to resolve the back, but harms nothing except 214 * performance. And no one cares about front-buffer render 215 * performance. 216 */ 217 intel_resolve_for_dri2_flush(brw, driDrawable); 218 intel_batchbuffer_flush(brw); 219 220 flushFront(screen)(driDrawable, driDrawable->loaderPrivate); 221 222 /* We set the dirty bit in intel_prepare_render() if we're 223 * front buffer rendering once we get there. 224 */ 225 brw->front_buffer_dirty = false; 226 } 227 } 228} 229 230static void 231intel_glFlush(struct gl_context *ctx) 232{ 233 struct brw_context *brw = brw_context(ctx); 234 235 intel_batchbuffer_flush(brw); 236 intel_flush_front(ctx); 237 238 brw->need_flush_throttle = true; 239} 240 241static void 242intel_finish(struct gl_context * ctx) 243{ 244 struct brw_context *brw = brw_context(ctx); 245 246 intel_glFlush(ctx); 247 248 if (brw->batch.last_bo) 249 drm_intel_bo_wait_rendering(brw->batch.last_bo); 250} 251 252static void 253brw_init_driver_functions(struct brw_context *brw, 254 struct dd_function_table *functions) 255{ 256 _mesa_init_driver_functions(functions); 257 258 /* GLX uses DRI2 invalidate events to handle window resizing. 259 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib), 260 * which doesn't provide a mechanism for snooping the event queues. 261 * 262 * So EGL still relies on viewport hacks to handle window resizing. 263 * This should go away with DRI3000. 264 */ 265 if (!brw->driContext->driScreenPriv->dri2.useInvalidate) 266 functions->Viewport = intel_viewport; 267 268 functions->Flush = intel_glFlush; 269 functions->Finish = intel_finish; 270 functions->GetString = intel_get_string; 271 functions->UpdateState = intel_update_state; 272 273 intelInitTextureFuncs(functions); 274 intelInitTextureImageFuncs(functions); 275 intelInitTextureSubImageFuncs(functions); 276 intelInitTextureCopyImageFuncs(functions); 277 intelInitCopyImageFuncs(functions); 278 intelInitClearFuncs(functions); 279 intelInitBufferFuncs(functions); 280 intelInitPixelFuncs(functions); 281 intelInitBufferObjectFuncs(functions); 282 intel_init_syncobj_functions(functions); 283 brw_init_object_purgeable_functions(functions); 284 285 brwInitFragProgFuncs( functions ); 286 brw_init_common_queryobj_functions(functions); 287 if (brw->gen >= 6) 288 gen6_init_queryobj_functions(functions); 289 else 290 gen4_init_queryobj_functions(functions); 291 292 functions->QuerySamplesForFormat = brw_query_samples_for_format; 293 294 functions->NewTransformFeedback = brw_new_transform_feedback; 295 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 296 functions->GetTransformFeedbackVertexCount = 297 brw_get_transform_feedback_vertex_count; 298 if (brw->gen >= 7) { 299 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 300 functions->EndTransformFeedback = gen7_end_transform_feedback; 301 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 302 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 303 } else { 304 functions->BeginTransformFeedback = brw_begin_transform_feedback; 305 functions->EndTransformFeedback = brw_end_transform_feedback; 306 } 307 308 if (brw->gen >= 6) 309 functions->GetSamplePosition = gen6_get_sample_position; 310} 311 312static void 313brw_initialize_context_constants(struct brw_context *brw) 314{ 315 struct gl_context *ctx = &brw->ctx; 316 317 unsigned max_samplers = 318 brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16; 319 320 ctx->Const.QueryCounterBits.Timestamp = 36; 321 322 ctx->Const.StripTextureBorder = true; 323 324 ctx->Const.MaxDualSourceDrawBuffers = 1; 325 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 326 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits = max_samplers; 327 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 328 ctx->Const.MaxTextureUnits = 329 MIN2(ctx->Const.MaxTextureCoordUnits, 330 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits); 331 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits = max_samplers; 332 if (brw->gen >= 6) 333 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = max_samplers; 334 else 335 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = 0; 336 if (_mesa_extension_override_enables.ARB_compute_shader) { 337 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 338 ctx->Const.MaxUniformBufferBindings += 12; 339 } else { 340 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits = 0; 341 } 342 ctx->Const.MaxCombinedTextureImageUnits = 343 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits + 344 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits + 345 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits + 346 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits; 347 348 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 349 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 350 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 351 ctx->Const.Max3DTextureLevels = 12; /* 2048 */ 352 ctx->Const.MaxCubeTextureLevels = 14; /* 8192 */ 353 ctx->Const.MaxTextureMbytes = 1536; 354 355 if (brw->gen >= 7) 356 ctx->Const.MaxArrayTextureLayers = 2048; 357 else 358 ctx->Const.MaxArrayTextureLayers = 512; 359 360 ctx->Const.MaxTextureRectSize = 1 << 12; 361 362 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 363 364 ctx->Const.MaxRenderbufferSize = 8192; 365 366 /* Hardware only supports a limited number of transform feedback buffers. 367 * So we need to override the Mesa default (which is based only on software 368 * limits). 369 */ 370 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 371 372 /* On Gen6, in the worst case, we use up one binding table entry per 373 * transform feedback component (see comments above the definition of 374 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 375 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 376 * BRW_MAX_SOL_BINDINGS. 377 * 378 * In "separate components" mode, we need to divide this value by 379 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 380 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 381 */ 382 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 383 ctx->Const.MaxTransformFeedbackSeparateComponents = 384 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 385 386 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 387 388 int max_samples; 389 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 390 const int clamp_max_samples = 391 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 392 393 if (clamp_max_samples < 0) { 394 max_samples = msaa_modes[0]; 395 } else { 396 /* Select the largest supported MSAA mode that does not exceed 397 * clamp_max_samples. 398 */ 399 max_samples = 0; 400 for (int i = 0; msaa_modes[i] != 0; ++i) { 401 if (msaa_modes[i] <= clamp_max_samples) { 402 max_samples = msaa_modes[i]; 403 break; 404 } 405 } 406 } 407 408 ctx->Const.MaxSamples = max_samples; 409 ctx->Const.MaxColorTextureSamples = max_samples; 410 ctx->Const.MaxDepthTextureSamples = max_samples; 411 ctx->Const.MaxIntegerSamples = max_samples; 412 413 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used 414 * to map indices of rectangular grid to sample numbers within a pixel. 415 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled 416 * extension implementation. For more details see the comment above 417 * gen6_set_sample_maps() definition. 418 */ 419 gen6_set_sample_maps(ctx); 420 421 if (brw->gen >= 7) 422 ctx->Const.MaxProgramTextureGatherComponents = 4; 423 else if (brw->gen == 6) 424 ctx->Const.MaxProgramTextureGatherComponents = 1; 425 426 ctx->Const.MinLineWidth = 1.0; 427 ctx->Const.MinLineWidthAA = 1.0; 428 if (brw->gen >= 9 || brw->is_cherryview) { 429 ctx->Const.MaxLineWidth = 40.0; 430 ctx->Const.MaxLineWidthAA = 40.0; 431 ctx->Const.LineWidthGranularity = 0.125; 432 } else if (brw->gen >= 6) { 433 ctx->Const.MaxLineWidth = 7.375; 434 ctx->Const.MaxLineWidthAA = 7.375; 435 ctx->Const.LineWidthGranularity = 0.125; 436 } else { 437 ctx->Const.MaxLineWidth = 7.0; 438 ctx->Const.MaxLineWidthAA = 7.0; 439 ctx->Const.LineWidthGranularity = 0.5; 440 } 441 442 ctx->Const.MinPointSize = 1.0; 443 ctx->Const.MinPointSizeAA = 1.0; 444 ctx->Const.MaxPointSize = 255.0; 445 ctx->Const.MaxPointSizeAA = 255.0; 446 ctx->Const.PointSizeGranularity = 1.0; 447 448 if (brw->gen >= 5 || brw->is_g4x) 449 ctx->Const.MaxClipPlanes = 8; 450 451 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024; 452 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0; 453 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0; 454 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0; 455 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0; 456 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0; 457 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0; 458 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16; 459 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256; 460 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1; 461 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024; 462 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams = 463 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters, 464 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams); 465 466 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024; 467 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024; 468 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024; 469 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024; 470 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12; 471 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256; 472 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0; 473 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024; 474 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams = 475 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters, 476 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams); 477 478 /* Fragment shaders use real, 32-bit twos-complement integers for all 479 * integer types. 480 */ 481 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31; 482 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30; 483 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0; 484 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 485 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 486 487 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMin = 31; 488 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMax = 30; 489 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.Precision = 0; 490 ctx->Const.Program[MESA_SHADER_VERTEX].HighInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt; 491 ctx->Const.Program[MESA_SHADER_VERTEX].MediumInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt; 492 493 if (brw->gen >= 7) { 494 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 495 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 496 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 497 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 498 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicBuffers = BRW_MAX_ABO; 499 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicBuffers = BRW_MAX_ABO; 500 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicBuffers = BRW_MAX_ABO; 501 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxAtomicBuffers = BRW_MAX_ABO; 502 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 503 } 504 505 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 506 * but we're not sure how it's actually done for vertex order, 507 * that affect provoking vertex decision. Always use last vertex 508 * convention for quad primitive which works as expected for now. 509 */ 510 if (brw->gen >= 6) 511 ctx->Const.QuadsFollowProvokingVertexConvention = false; 512 513 ctx->Const.NativeIntegers = true; 514 ctx->Const.VertexID_is_zero_based = true; 515 516 /* Regarding the CMP instruction, the Ivybridge PRM says: 517 * 518 * "For each enabled channel 0b or 1b is assigned to the appropriate flag 519 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord 520 * 0xFFFFFFFF) is assigned to dst." 521 * 522 * but PRMs for earlier generations say 523 * 524 * "In dword format, one GRF may store up to 8 results. When the register 525 * is used later as a vector of Booleans, as only LSB at each channel 526 * contains meaning [sic] data, software should make sure all higher bits 527 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)." 528 * 529 * We select the representation of a true boolean uniform to be ~0, and fix 530 * the results of Gen <= 5 CMP instruction's with -(result & 1). 531 */ 532 ctx->Const.UniformBooleanTrue = ~0; 533 534 /* From the gen4 PRM, volume 4 page 127: 535 * 536 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 537 * the base address of the first element of the surface, computed in 538 * software by adding the surface base address to the byte offset of 539 * the element in the buffer." 540 * 541 * However, unaligned accesses are slower, so enforce buffer alignment. 542 */ 543 ctx->Const.UniformBufferOffsetAlignment = 16; 544 ctx->Const.TextureBufferOffsetAlignment = 16; 545 546 if (brw->gen >= 6) { 547 ctx->Const.MaxVarying = 32; 548 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128; 549 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64; 550 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128; 551 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128; 552 } 553 554 static const nir_shader_compiler_options nir_options = { 555 .native_integers = true, 556 /* In order to help allow for better CSE at the NIR level we tell NIR 557 * to split all ffma instructions during opt_algebraic and we then 558 * re-combine them as a later step. 559 */ 560 .lower_ffma = true, 561 .lower_sub = true, 562 }; 563 564 /* We want the GLSL compiler to emit code that uses condition codes */ 565 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 566 ctx->Const.ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 567 ctx->Const.ShaderCompilerOptions[i].EmitCondCodes = true; 568 ctx->Const.ShaderCompilerOptions[i].EmitNoNoise = true; 569 ctx->Const.ShaderCompilerOptions[i].EmitNoMainReturn = true; 570 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectInput = true; 571 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectOutput = 572 (i == MESA_SHADER_FRAGMENT); 573 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectTemp = 574 (i == MESA_SHADER_FRAGMENT); 575 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectUniform = false; 576 ctx->Const.ShaderCompilerOptions[i].LowerClipDistance = true; 577 } 578 579 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = true; 580 ctx->Const.ShaderCompilerOptions[MESA_SHADER_GEOMETRY].OptimizeForAOS = true; 581 582 if (brw->scalar_vs) { 583 /* If we're using the scalar backend for vertex shaders, we need to 584 * configure these accordingly. 585 */ 586 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].EmitNoIndirectOutput = true; 587 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].EmitNoIndirectTemp = true; 588 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = false; 589 590 if (brw_env_var_as_boolean("INTEL_USE_NIR", false)) 591 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].NirOptions = &nir_options; 592 } 593 594 if (brw_env_var_as_boolean("INTEL_USE_NIR", true)) 595 ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT].NirOptions = &nir_options; 596 597 /* ARB_viewport_array */ 598 if (brw->gen >= 7 && ctx->API == API_OPENGL_CORE) { 599 ctx->Const.MaxViewports = GEN7_NUM_VIEWPORTS; 600 ctx->Const.ViewportSubpixelBits = 0; 601 602 /* Cast to float before negating because MaxViewportWidth is unsigned. 603 */ 604 ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth; 605 ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth; 606 } 607 608 /* ARB_gpu_shader5 */ 609 if (brw->gen >= 7) 610 ctx->Const.MaxVertexStreams = MIN2(4, MAX_VERTEX_STREAMS); 611} 612 613/** 614 * Process driconf (drirc) options, setting appropriate context flags. 615 * 616 * intelInitExtensions still pokes at optionCache directly, in order to 617 * avoid advertising various extensions. No flags are set, so it makes 618 * sense to continue doing that there. 619 */ 620static void 621brw_process_driconf_options(struct brw_context *brw) 622{ 623 struct gl_context *ctx = &brw->ctx; 624 625 driOptionCache *options = &brw->optionCache; 626 driParseConfigFiles(options, &brw->intelScreen->optionCache, 627 brw->driContext->driScreenPriv->myNum, "i965"); 628 629 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 630 switch (bo_reuse_mode) { 631 case DRI_CONF_BO_REUSE_DISABLED: 632 break; 633 case DRI_CONF_BO_REUSE_ALL: 634 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 635 break; 636 } 637 638 if (!driQueryOptionb(options, "hiz")) { 639 brw->has_hiz = false; 640 /* On gen6, you can only do separate stencil with HIZ. */ 641 if (brw->gen == 6) 642 brw->has_separate_stencil = false; 643 } 644 645 if (driQueryOptionb(options, "always_flush_batch")) { 646 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 647 brw->always_flush_batch = true; 648 } 649 650 if (driQueryOptionb(options, "always_flush_cache")) { 651 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 652 brw->always_flush_cache = true; 653 } 654 655 if (driQueryOptionb(options, "disable_throttling")) { 656 fprintf(stderr, "disabling flush throttling\n"); 657 brw->disable_throttling = true; 658 } 659 660 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 661 662 ctx->Const.ForceGLSLExtensionsWarn = 663 driQueryOptionb(options, "force_glsl_extensions_warn"); 664 665 ctx->Const.DisableGLSLLineContinuations = 666 driQueryOptionb(options, "disable_glsl_line_continuations"); 667 668 ctx->Const.AllowGLSLExtensionDirectiveMidShader = 669 driQueryOptionb(options, "allow_glsl_extension_directive_midshader"); 670} 671 672GLboolean 673brwCreateContext(gl_api api, 674 const struct gl_config *mesaVis, 675 __DRIcontext *driContextPriv, 676 unsigned major_version, 677 unsigned minor_version, 678 uint32_t flags, 679 bool notify_reset, 680 unsigned *dri_ctx_error, 681 void *sharedContextPrivate) 682{ 683 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 684 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 685 struct intel_screen *screen = sPriv->driverPrivate; 686 const struct brw_device_info *devinfo = screen->devinfo; 687 struct dd_function_table functions; 688 689 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 690 * provides us with context reset notifications. 691 */ 692 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 693 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 694 695 if (screen->has_context_reset_notification) 696 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 697 698 if (flags & ~allowed_flags) { 699 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 700 return false; 701 } 702 703 struct brw_context *brw = rzalloc(NULL, struct brw_context); 704 if (!brw) { 705 fprintf(stderr, "%s: failed to alloc context\n", __func__); 706 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 707 return false; 708 } 709 710 driContextPriv->driverPrivate = brw; 711 brw->driContext = driContextPriv; 712 brw->intelScreen = screen; 713 brw->bufmgr = screen->bufmgr; 714 715 brw->gen = devinfo->gen; 716 brw->gt = devinfo->gt; 717 brw->is_g4x = devinfo->is_g4x; 718 brw->is_baytrail = devinfo->is_baytrail; 719 brw->is_haswell = devinfo->is_haswell; 720 brw->is_cherryview = devinfo->is_cherryview; 721 brw->has_llc = devinfo->has_llc; 722 brw->has_hiz = devinfo->has_hiz_and_separate_stencil; 723 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 724 brw->has_pln = devinfo->has_pln; 725 brw->has_compr4 = devinfo->has_compr4; 726 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 727 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 728 brw->needs_unlit_centroid_workaround = 729 devinfo->needs_unlit_centroid_workaround; 730 731 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 732 brw->has_swizzling = screen->hw_has_swizzling; 733 734 brw->vs.base.stage = MESA_SHADER_VERTEX; 735 brw->gs.base.stage = MESA_SHADER_GEOMETRY; 736 brw->wm.base.stage = MESA_SHADER_FRAGMENT; 737 if (brw->gen >= 8) { 738 gen8_init_vtable_surface_functions(brw); 739 brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz; 740 } else if (brw->gen >= 7) { 741 gen7_init_vtable_surface_functions(brw); 742 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 743 } else if (brw->gen >= 6) { 744 gen6_init_vtable_surface_functions(brw); 745 brw->vtbl.emit_depth_stencil_hiz = gen6_emit_depth_stencil_hiz; 746 } else { 747 gen4_init_vtable_surface_functions(brw); 748 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 749 } 750 751 brw_init_driver_functions(brw, &functions); 752 753 if (notify_reset) 754 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 755 756 struct gl_context *ctx = &brw->ctx; 757 758 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 759 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 760 fprintf(stderr, "%s: failed to init mesa context\n", __func__); 761 intelDestroyContext(driContextPriv); 762 return false; 763 } 764 765 driContextSetFlags(ctx, flags); 766 767 /* Initialize the software rasterizer and helper modules. 768 * 769 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 770 * software fallbacks (which we have to support on legacy GL to do weird 771 * glDrawPixels(), glBitmap(), and other functions). 772 */ 773 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 774 _swrast_CreateContext(ctx); 775 } 776 777 _vbo_CreateContext(ctx); 778 if (ctx->swrast_context) { 779 _tnl_CreateContext(ctx); 780 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 781 _swsetup_CreateContext(ctx); 782 783 /* Configure swrast to match hardware characteristics: */ 784 _swrast_allow_pixel_fog(ctx, false); 785 _swrast_allow_vertex_fog(ctx, true); 786 } 787 788 _mesa_meta_init(ctx); 789 790 brw_process_driconf_options(brw); 791 brw_process_intel_debug_variable(brw); 792 793 if (brw->gen >= 8 && !(INTEL_DEBUG & DEBUG_VEC4VS)) 794 brw->scalar_vs = true; 795 796 brw_initialize_context_constants(brw); 797 798 ctx->Const.ResetStrategy = notify_reset 799 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 800 801 /* Reinitialize the context point state. It depends on ctx->Const values. */ 802 _mesa_init_point(ctx); 803 804 intel_fbo_init(brw); 805 806 intel_batchbuffer_init(brw); 807 808 if (brw->gen >= 6) { 809 /* Create a new hardware context. Using a hardware context means that 810 * our GPU state will be saved/restored on context switch, allowing us 811 * to assume that the GPU is in the same state we left it in. 812 * 813 * This is required for transform feedback buffer offsets, query objects, 814 * and also allows us to reduce how much state we have to emit. 815 */ 816 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 817 818 if (!brw->hw_ctx) { 819 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 820 intelDestroyContext(driContextPriv); 821 return false; 822 } 823 } 824 825 brw_init_state(brw); 826 827 intelInitExtensions(ctx); 828 829 brw_init_surface_formats(brw); 830 831 brw->max_vs_threads = devinfo->max_vs_threads; 832 brw->max_hs_threads = devinfo->max_hs_threads; 833 brw->max_ds_threads = devinfo->max_ds_threads; 834 brw->max_gs_threads = devinfo->max_gs_threads; 835 brw->max_wm_threads = devinfo->max_wm_threads; 836 brw->urb.size = devinfo->urb.size; 837 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 838 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 839 brw->urb.max_hs_entries = devinfo->urb.max_hs_entries; 840 brw->urb.max_ds_entries = devinfo->urb.max_ds_entries; 841 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 842 843 /* Estimate the size of the mappable aperture into the GTT. There's an 844 * ioctl to get the whole GTT size, but not one to get the mappable subset. 845 * It turns out it's basically always 256MB, though some ancient hardware 846 * was smaller. 847 */ 848 uint32_t gtt_size = 256 * 1024 * 1024; 849 850 /* We don't want to map two objects such that a memcpy between them would 851 * just fault one mapping in and then the other over and over forever. So 852 * we would need to divide the GTT size by 2. Additionally, some GTT is 853 * taken up by things like the framebuffer and the ringbuffer and such, so 854 * be more conservative. 855 */ 856 brw->max_gtt_map_object_size = gtt_size / 4; 857 858 if (brw->gen == 6) 859 brw->urb.gs_present = false; 860 861 brw->prim_restart.in_progress = false; 862 brw->prim_restart.enable_cut_index = false; 863 brw->gs.enabled = false; 864 brw->sf.viewport_transform_enable = true; 865 866 ctx->VertexProgram._MaintainTnlProgram = true; 867 ctx->FragmentProgram._MaintainTexEnvProgram = true; 868 869 brw_draw_init( brw ); 870 871 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 872 /* Turn on some extra GL_ARB_debug_output generation. */ 873 brw->perf_debug = true; 874 } 875 876 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 877 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 878 879 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 880 brw_init_shader_time(brw); 881 882 _mesa_compute_version(ctx); 883 884 _mesa_initialize_dispatch_tables(ctx); 885 _mesa_initialize_vbo_vtxfmt(ctx); 886 887 if (ctx->Extensions.AMD_performance_monitor) { 888 brw_init_performance_monitors(brw); 889 } 890 891 vbo_use_buffer_objects(ctx); 892 vbo_always_unmap_buffers(ctx); 893 894 return true; 895} 896 897void 898intelDestroyContext(__DRIcontext * driContextPriv) 899{ 900 struct brw_context *brw = 901 (struct brw_context *) driContextPriv->driverPrivate; 902 struct gl_context *ctx = &brw->ctx; 903 904 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 905 if (INTEL_DEBUG & DEBUG_AUB) { 906 intel_batchbuffer_flush(brw); 907 aub_dump_bmp(&brw->ctx); 908 } 909 910 _mesa_meta_free(&brw->ctx); 911 brw_meta_fast_clear_free(brw); 912 913 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 914 /* Force a report. */ 915 brw->shader_time.report_time = 0; 916 917 brw_collect_and_report_shader_time(brw); 918 brw_destroy_shader_time(brw); 919 } 920 921 brw_destroy_state(brw); 922 brw_draw_destroy(brw); 923 924 drm_intel_bo_unreference(brw->curbe.curbe_bo); 925 if (brw->vs.base.scratch_bo) 926 drm_intel_bo_unreference(brw->vs.base.scratch_bo); 927 if (brw->gs.base.scratch_bo) 928 drm_intel_bo_unreference(brw->gs.base.scratch_bo); 929 if (brw->wm.base.scratch_bo) 930 drm_intel_bo_unreference(brw->wm.base.scratch_bo); 931 932 drm_intel_gem_context_destroy(brw->hw_ctx); 933 934 if (ctx->swrast_context) { 935 _swsetup_DestroyContext(&brw->ctx); 936 _tnl_DestroyContext(&brw->ctx); 937 } 938 _vbo_DestroyContext(&brw->ctx); 939 940 if (ctx->swrast_context) 941 _swrast_DestroyContext(&brw->ctx); 942 943 intel_batchbuffer_free(brw); 944 945 drm_intel_bo_unreference(brw->throttle_batch[1]); 946 drm_intel_bo_unreference(brw->throttle_batch[0]); 947 brw->throttle_batch[1] = NULL; 948 brw->throttle_batch[0] = NULL; 949 950 driDestroyOptionCache(&brw->optionCache); 951 952 /* free the Mesa context */ 953 _mesa_free_context_data(&brw->ctx); 954 955 ralloc_free(brw); 956 driContextPriv->driverPrivate = NULL; 957} 958 959GLboolean 960intelUnbindContext(__DRIcontext * driContextPriv) 961{ 962 /* Unset current context and dispath table */ 963 _mesa_make_current(NULL, NULL, NULL); 964 965 return true; 966} 967 968/** 969 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 970 * on window system framebuffers. 971 * 972 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 973 * your renderbuffer can do sRGB encode, and you can flip a switch that does 974 * sRGB encode if the renderbuffer can handle it. You can ask specifically 975 * for a visual where you're guaranteed to be capable, but it turns out that 976 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 977 * incapable ones, because there's no difference between the two in resources 978 * used. Applications thus get built that accidentally rely on the default 979 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 980 * great... 981 * 982 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 983 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 984 * So they removed the enable knob and made it "if the renderbuffer is sRGB 985 * capable, do sRGB encode". Then, for your window system renderbuffers, you 986 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 987 * and get no sRGB encode (assuming that both kinds of visual are available). 988 * Thus our choice to support sRGB by default on our visuals for desktop would 989 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 990 * 991 * Unfortunately, renderbuffer setup happens before a context is created. So 992 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 993 * context (without an sRGB visual, though we don't have sRGB visuals exposed 994 * yet), we go turn that back off before anyone finds out. 995 */ 996static void 997intel_gles3_srgb_workaround(struct brw_context *brw, 998 struct gl_framebuffer *fb) 999{ 1000 struct gl_context *ctx = &brw->ctx; 1001 1002 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 1003 return; 1004 1005 /* Some day when we support the sRGB capable bit on visuals available for 1006 * GLES, we'll need to respect that and not disable things here. 1007 */ 1008 fb->Visual.sRGBCapable = false; 1009 for (int i = 0; i < BUFFER_COUNT; i++) { 1010 if (fb->Attachment[i].Renderbuffer && 1011 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) { 1012 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM; 1013 } 1014 } 1015} 1016 1017GLboolean 1018intelMakeCurrent(__DRIcontext * driContextPriv, 1019 __DRIdrawable * driDrawPriv, 1020 __DRIdrawable * driReadPriv) 1021{ 1022 struct brw_context *brw; 1023 GET_CURRENT_CONTEXT(curCtx); 1024 1025 if (driContextPriv) 1026 brw = (struct brw_context *) driContextPriv->driverPrivate; 1027 else 1028 brw = NULL; 1029 1030 /* According to the glXMakeCurrent() man page: "Pending commands to 1031 * the previous context, if any, are flushed before it is released." 1032 * But only flush if we're actually changing contexts. 1033 */ 1034 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 1035 _mesa_flush(curCtx); 1036 } 1037 1038 if (driContextPriv) { 1039 struct gl_context *ctx = &brw->ctx; 1040 struct gl_framebuffer *fb, *readFb; 1041 1042 if (driDrawPriv == NULL) { 1043 fb = _mesa_get_incomplete_framebuffer(); 1044 } else { 1045 fb = driDrawPriv->driverPrivate; 1046 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 1047 } 1048 1049 if (driReadPriv == NULL) { 1050 readFb = _mesa_get_incomplete_framebuffer(); 1051 } else { 1052 readFb = driReadPriv->driverPrivate; 1053 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 1054 } 1055 1056 /* The sRGB workaround changes the renderbuffer's format. We must change 1057 * the format before the renderbuffer's miptree get's allocated, otherwise 1058 * the formats of the renderbuffer and its miptree will differ. 1059 */ 1060 intel_gles3_srgb_workaround(brw, fb); 1061 intel_gles3_srgb_workaround(brw, readFb); 1062 1063 /* If the context viewport hasn't been initialized, force a call out to 1064 * the loader to get buffers so we have a drawable size for the initial 1065 * viewport. */ 1066 if (!brw->ctx.ViewportInitialized) 1067 intel_prepare_render(brw); 1068 1069 _mesa_make_current(ctx, fb, readFb); 1070 } else { 1071 _mesa_make_current(NULL, NULL, NULL); 1072 } 1073 1074 return true; 1075} 1076 1077void 1078intel_resolve_for_dri2_flush(struct brw_context *brw, 1079 __DRIdrawable *drawable) 1080{ 1081 if (brw->gen < 6) { 1082 /* MSAA and fast color clear are not supported, so don't waste time 1083 * checking whether a resolve is needed. 1084 */ 1085 return; 1086 } 1087 1088 struct gl_framebuffer *fb = drawable->driverPrivate; 1089 struct intel_renderbuffer *rb; 1090 1091 /* Usually, only the back buffer will need to be downsampled. However, 1092 * the front buffer will also need it if the user has rendered into it. 1093 */ 1094 static const gl_buffer_index buffers[2] = { 1095 BUFFER_BACK_LEFT, 1096 BUFFER_FRONT_LEFT, 1097 }; 1098 1099 for (int i = 0; i < 2; ++i) { 1100 rb = intel_get_renderbuffer(fb, buffers[i]); 1101 if (rb == NULL || rb->mt == NULL) 1102 continue; 1103 if (rb->mt->num_samples <= 1) 1104 intel_miptree_resolve_color(brw, rb->mt); 1105 else 1106 intel_renderbuffer_downsample(brw, rb); 1107 } 1108} 1109 1110static unsigned 1111intel_bits_per_pixel(const struct intel_renderbuffer *rb) 1112{ 1113 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 1114} 1115 1116static void 1117intel_query_dri2_buffers(struct brw_context *brw, 1118 __DRIdrawable *drawable, 1119 __DRIbuffer **buffers, 1120 int *count); 1121 1122static void 1123intel_process_dri2_buffer(struct brw_context *brw, 1124 __DRIdrawable *drawable, 1125 __DRIbuffer *buffer, 1126 struct intel_renderbuffer *rb, 1127 const char *buffer_name); 1128 1129static void 1130intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 1131 1132static void 1133intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1134{ 1135 struct gl_framebuffer *fb = drawable->driverPrivate; 1136 struct intel_renderbuffer *rb; 1137 __DRIbuffer *buffers = NULL; 1138 int i, count; 1139 const char *region_name; 1140 1141 /* Set this up front, so that in case our buffers get invalidated 1142 * while we're getting new buffers, we don't clobber the stamp and 1143 * thus ignore the invalidate. */ 1144 drawable->lastStamp = drawable->dri2.stamp; 1145 1146 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1147 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1148 1149 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1150 1151 if (buffers == NULL) 1152 return; 1153 1154 for (i = 0; i < count; i++) { 1155 switch (buffers[i].attachment) { 1156 case __DRI_BUFFER_FRONT_LEFT: 1157 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1158 region_name = "dri2 front buffer"; 1159 break; 1160 1161 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1162 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1163 region_name = "dri2 fake front buffer"; 1164 break; 1165 1166 case __DRI_BUFFER_BACK_LEFT: 1167 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1168 region_name = "dri2 back buffer"; 1169 break; 1170 1171 case __DRI_BUFFER_DEPTH: 1172 case __DRI_BUFFER_HIZ: 1173 case __DRI_BUFFER_DEPTH_STENCIL: 1174 case __DRI_BUFFER_STENCIL: 1175 case __DRI_BUFFER_ACCUM: 1176 default: 1177 fprintf(stderr, 1178 "unhandled buffer attach event, attachment type %d\n", 1179 buffers[i].attachment); 1180 return; 1181 } 1182 1183 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1184 } 1185 1186} 1187 1188void 1189intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1190{ 1191 struct brw_context *brw = context->driverPrivate; 1192 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1193 1194 /* Set this up front, so that in case our buffers get invalidated 1195 * while we're getting new buffers, we don't clobber the stamp and 1196 * thus ignore the invalidate. */ 1197 drawable->lastStamp = drawable->dri2.stamp; 1198 1199 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1200 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1201 1202 if (screen->image.loader) 1203 intel_update_image_buffers(brw, drawable); 1204 else 1205 intel_update_dri2_buffers(brw, drawable); 1206 1207 driUpdateFramebufferSize(&brw->ctx, drawable); 1208} 1209 1210/** 1211 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1212 * state is required. 1213 */ 1214void 1215intel_prepare_render(struct brw_context *brw) 1216{ 1217 struct gl_context *ctx = &brw->ctx; 1218 __DRIcontext *driContext = brw->driContext; 1219 __DRIdrawable *drawable; 1220 1221 drawable = driContext->driDrawablePriv; 1222 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1223 if (drawable->lastStamp != drawable->dri2.stamp) 1224 intel_update_renderbuffers(driContext, drawable); 1225 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1226 } 1227 1228 drawable = driContext->driReadablePriv; 1229 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1230 if (drawable->lastStamp != drawable->dri2.stamp) 1231 intel_update_renderbuffers(driContext, drawable); 1232 driContext->dri2.read_stamp = drawable->dri2.stamp; 1233 } 1234 1235 /* If we're currently rendering to the front buffer, the rendering 1236 * that will happen next will probably dirty the front buffer. So 1237 * mark it as dirty here. 1238 */ 1239 if (brw_is_front_buffer_drawing(ctx->DrawBuffer)) 1240 brw->front_buffer_dirty = true; 1241} 1242 1243/** 1244 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1245 * 1246 * To determine which DRI buffers to request, examine the renderbuffers 1247 * attached to the drawable's framebuffer. Then request the buffers with 1248 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1249 * 1250 * This is called from intel_update_renderbuffers(). 1251 * 1252 * \param drawable Drawable whose buffers are queried. 1253 * \param buffers [out] List of buffers returned by DRI2 query. 1254 * \param buffer_count [out] Number of buffers returned. 1255 * 1256 * \see intel_update_renderbuffers() 1257 * \see DRI2GetBuffers() 1258 * \see DRI2GetBuffersWithFormat() 1259 */ 1260static void 1261intel_query_dri2_buffers(struct brw_context *brw, 1262 __DRIdrawable *drawable, 1263 __DRIbuffer **buffers, 1264 int *buffer_count) 1265{ 1266 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1267 struct gl_framebuffer *fb = drawable->driverPrivate; 1268 int i = 0; 1269 unsigned attachments[8]; 1270 1271 struct intel_renderbuffer *front_rb; 1272 struct intel_renderbuffer *back_rb; 1273 1274 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1275 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1276 1277 memset(attachments, 0, sizeof(attachments)); 1278 if ((brw_is_front_buffer_drawing(fb) || 1279 brw_is_front_buffer_reading(fb) || 1280 !back_rb) && front_rb) { 1281 /* If a fake front buffer is in use, then querying for 1282 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1283 * the real front buffer to the fake front buffer. So before doing the 1284 * query, we need to make sure all the pending drawing has landed in the 1285 * real front buffer. 1286 */ 1287 intel_batchbuffer_flush(brw); 1288 intel_flush_front(&brw->ctx); 1289 1290 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1291 attachments[i++] = intel_bits_per_pixel(front_rb); 1292 } else if (front_rb && brw->front_buffer_dirty) { 1293 /* We have pending front buffer rendering, but we aren't querying for a 1294 * front buffer. If the front buffer we have is a fake front buffer, 1295 * the X server is going to throw it away when it processes the query. 1296 * So before doing the query, make sure all the pending drawing has 1297 * landed in the real front buffer. 1298 */ 1299 intel_batchbuffer_flush(brw); 1300 intel_flush_front(&brw->ctx); 1301 } 1302 1303 if (back_rb) { 1304 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1305 attachments[i++] = intel_bits_per_pixel(back_rb); 1306 } 1307 1308 assert(i <= ARRAY_SIZE(attachments)); 1309 1310 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1311 &drawable->w, 1312 &drawable->h, 1313 attachments, i / 2, 1314 buffer_count, 1315 drawable->loaderPrivate); 1316} 1317 1318/** 1319 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1320 * 1321 * This is called from intel_update_renderbuffers(). 1322 * 1323 * \par Note: 1324 * DRI buffers whose attachment point is DRI2BufferStencil or 1325 * DRI2BufferDepthStencil are handled as special cases. 1326 * 1327 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1328 * that is passed to drm_intel_bo_gem_create_from_name(). 1329 * 1330 * \see intel_update_renderbuffers() 1331 */ 1332static void 1333intel_process_dri2_buffer(struct brw_context *brw, 1334 __DRIdrawable *drawable, 1335 __DRIbuffer *buffer, 1336 struct intel_renderbuffer *rb, 1337 const char *buffer_name) 1338{ 1339 struct gl_framebuffer *fb = drawable->driverPrivate; 1340 drm_intel_bo *bo; 1341 1342 if (!rb) 1343 return; 1344 1345 unsigned num_samples = rb->Base.Base.NumSamples; 1346 1347 /* We try to avoid closing and reopening the same BO name, because the first 1348 * use of a mapping of the buffer involves a bunch of page faulting which is 1349 * moderately expensive. 1350 */ 1351 struct intel_mipmap_tree *last_mt; 1352 if (num_samples == 0) 1353 last_mt = rb->mt; 1354 else 1355 last_mt = rb->singlesample_mt; 1356 1357 uint32_t old_name = 0; 1358 if (last_mt) { 1359 /* The bo already has a name because the miptree was created by a 1360 * previous call to intel_process_dri2_buffer(). If a bo already has a 1361 * name, then drm_intel_bo_flink() is a low-cost getter. It does not 1362 * create a new name. 1363 */ 1364 drm_intel_bo_flink(last_mt->bo, &old_name); 1365 } 1366 1367 if (old_name == buffer->name) 1368 return; 1369 1370 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1371 fprintf(stderr, 1372 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1373 buffer->name, buffer->attachment, 1374 buffer->cpp, buffer->pitch); 1375 } 1376 1377 intel_miptree_release(&rb->mt); 1378 bo = drm_intel_bo_gem_create_from_name(brw->bufmgr, buffer_name, 1379 buffer->name); 1380 if (!bo) { 1381 fprintf(stderr, 1382 "Failed to open BO for returned DRI2 buffer " 1383 "(%dx%d, %s, named %d).\n" 1384 "This is likely a bug in the X Server that will lead to a " 1385 "crash soon.\n", 1386 drawable->w, drawable->h, buffer_name, buffer->name); 1387 return; 1388 } 1389 1390 intel_update_winsys_renderbuffer_miptree(brw, rb, bo, 1391 drawable->w, drawable->h, 1392 buffer->pitch); 1393 1394 if (brw_is_front_buffer_drawing(fb) && 1395 (buffer->attachment == __DRI_BUFFER_FRONT_LEFT || 1396 buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) && 1397 rb->Base.Base.NumSamples > 1) { 1398 intel_renderbuffer_upsample(brw, rb); 1399 } 1400 1401 assert(rb->mt); 1402 1403 drm_intel_bo_unreference(bo); 1404} 1405 1406/** 1407 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1408 * 1409 * To determine which DRI buffers to request, examine the renderbuffers 1410 * attached to the drawable's framebuffer. Then request the buffers from 1411 * the image loader 1412 * 1413 * This is called from intel_update_renderbuffers(). 1414 * 1415 * \param drawable Drawable whose buffers are queried. 1416 * \param buffers [out] List of buffers returned by DRI2 query. 1417 * \param buffer_count [out] Number of buffers returned. 1418 * 1419 * \see intel_update_renderbuffers() 1420 */ 1421 1422static void 1423intel_update_image_buffer(struct brw_context *intel, 1424 __DRIdrawable *drawable, 1425 struct intel_renderbuffer *rb, 1426 __DRIimage *buffer, 1427 enum __DRIimageBufferMask buffer_type) 1428{ 1429 struct gl_framebuffer *fb = drawable->driverPrivate; 1430 1431 if (!rb || !buffer->bo) 1432 return; 1433 1434 unsigned num_samples = rb->Base.Base.NumSamples; 1435 1436 /* Check and see if we're already bound to the right 1437 * buffer object 1438 */ 1439 struct intel_mipmap_tree *last_mt; 1440 if (num_samples == 0) 1441 last_mt = rb->mt; 1442 else 1443 last_mt = rb->singlesample_mt; 1444 1445 if (last_mt && last_mt->bo == buffer->bo) 1446 return; 1447 1448 intel_update_winsys_renderbuffer_miptree(intel, rb, buffer->bo, 1449 buffer->width, buffer->height, 1450 buffer->pitch); 1451 1452 if (brw_is_front_buffer_drawing(fb) && 1453 buffer_type == __DRI_IMAGE_BUFFER_FRONT && 1454 rb->Base.Base.NumSamples > 1) { 1455 intel_renderbuffer_upsample(intel, rb); 1456 } 1457} 1458 1459static void 1460intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1461{ 1462 struct gl_framebuffer *fb = drawable->driverPrivate; 1463 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1464 struct intel_renderbuffer *front_rb; 1465 struct intel_renderbuffer *back_rb; 1466 struct __DRIimageList images; 1467 unsigned int format; 1468 uint32_t buffer_mask = 0; 1469 1470 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1471 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1472 1473 if (back_rb) 1474 format = intel_rb_format(back_rb); 1475 else if (front_rb) 1476 format = intel_rb_format(front_rb); 1477 else 1478 return; 1479 1480 if (front_rb && (brw_is_front_buffer_drawing(fb) || 1481 brw_is_front_buffer_reading(fb) || !back_rb)) { 1482 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1483 } 1484 1485 if (back_rb) 1486 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1487 1488 (*screen->image.loader->getBuffers) (drawable, 1489 driGLFormatToImageFormat(format), 1490 &drawable->dri2.stamp, 1491 drawable->loaderPrivate, 1492 buffer_mask, 1493 &images); 1494 1495 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1496 drawable->w = images.front->width; 1497 drawable->h = images.front->height; 1498 intel_update_image_buffer(brw, 1499 drawable, 1500 front_rb, 1501 images.front, 1502 __DRI_IMAGE_BUFFER_FRONT); 1503 } 1504 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1505 drawable->w = images.back->width; 1506 drawable->h = images.back->height; 1507 intel_update_image_buffer(brw, 1508 drawable, 1509 back_rb, 1510 images.back, 1511 __DRI_IMAGE_BUFFER_BACK); 1512 } 1513} 1514