brw_context.c revision 28d9e90428282a5e0a6aa31ad858a5cf514d1264
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 becuase 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} 671 672/* drop when libdrm 2.4.61 is released */ 673#ifndef I915_PARAM_REVISION 674#define I915_PARAM_REVISION 32 675#endif 676 677static int 678brw_get_revision(int fd) 679{ 680 struct drm_i915_getparam gp; 681 int revision; 682 int ret; 683 684 memset(&gp, 0, sizeof(gp)); 685 gp.param = I915_PARAM_REVISION; 686 gp.value = &revision; 687 688 ret = drmCommandWriteRead(fd, DRM_I915_GETPARAM, &gp, sizeof(gp)); 689 if (ret) 690 revision = -1; 691 692 return revision; 693} 694 695GLboolean 696brwCreateContext(gl_api api, 697 const struct gl_config *mesaVis, 698 __DRIcontext *driContextPriv, 699 unsigned major_version, 700 unsigned minor_version, 701 uint32_t flags, 702 bool notify_reset, 703 unsigned *dri_ctx_error, 704 void *sharedContextPrivate) 705{ 706 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 707 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 708 struct intel_screen *screen = sPriv->driverPrivate; 709 const struct brw_device_info *devinfo = screen->devinfo; 710 struct dd_function_table functions; 711 712 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 713 * provides us with context reset notifications. 714 */ 715 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 716 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 717 718 if (screen->has_context_reset_notification) 719 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 720 721 if (flags & ~allowed_flags) { 722 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 723 return false; 724 } 725 726 struct brw_context *brw = rzalloc(NULL, struct brw_context); 727 if (!brw) { 728 fprintf(stderr, "%s: failed to alloc context\n", __func__); 729 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 730 return false; 731 } 732 733 driContextPriv->driverPrivate = brw; 734 brw->driContext = driContextPriv; 735 brw->intelScreen = screen; 736 brw->bufmgr = screen->bufmgr; 737 738 brw->gen = devinfo->gen; 739 brw->gt = devinfo->gt; 740 brw->is_g4x = devinfo->is_g4x; 741 brw->is_baytrail = devinfo->is_baytrail; 742 brw->is_haswell = devinfo->is_haswell; 743 brw->is_cherryview = devinfo->is_cherryview; 744 brw->has_llc = devinfo->has_llc; 745 brw->has_hiz = devinfo->has_hiz_and_separate_stencil; 746 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 747 brw->has_pln = devinfo->has_pln; 748 brw->has_compr4 = devinfo->has_compr4; 749 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 750 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 751 brw->needs_unlit_centroid_workaround = 752 devinfo->needs_unlit_centroid_workaround; 753 brw->revision = brw_get_revision(sPriv->fd); 754 755 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 756 brw->has_swizzling = screen->hw_has_swizzling; 757 758 brw->vs.base.stage = MESA_SHADER_VERTEX; 759 brw->gs.base.stage = MESA_SHADER_GEOMETRY; 760 brw->wm.base.stage = MESA_SHADER_FRAGMENT; 761 if (brw->gen >= 8) { 762 gen8_init_vtable_surface_functions(brw); 763 brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz; 764 } else if (brw->gen >= 7) { 765 gen7_init_vtable_surface_functions(brw); 766 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 767 } else if (brw->gen >= 6) { 768 gen6_init_vtable_surface_functions(brw); 769 brw->vtbl.emit_depth_stencil_hiz = gen6_emit_depth_stencil_hiz; 770 } else { 771 gen4_init_vtable_surface_functions(brw); 772 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 773 } 774 775 brw_init_driver_functions(brw, &functions); 776 777 if (notify_reset) 778 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 779 780 struct gl_context *ctx = &brw->ctx; 781 782 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 783 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 784 fprintf(stderr, "%s: failed to init mesa context\n", __func__); 785 intelDestroyContext(driContextPriv); 786 return false; 787 } 788 789 driContextSetFlags(ctx, flags); 790 791 /* Initialize the software rasterizer and helper modules. 792 * 793 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 794 * software fallbacks (which we have to support on legacy GL to do weird 795 * glDrawPixels(), glBitmap(), and other functions). 796 */ 797 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 798 _swrast_CreateContext(ctx); 799 } 800 801 _vbo_CreateContext(ctx); 802 if (ctx->swrast_context) { 803 _tnl_CreateContext(ctx); 804 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 805 _swsetup_CreateContext(ctx); 806 807 /* Configure swrast to match hardware characteristics: */ 808 _swrast_allow_pixel_fog(ctx, false); 809 _swrast_allow_vertex_fog(ctx, true); 810 } 811 812 _mesa_meta_init(ctx); 813 814 brw_process_driconf_options(brw); 815 brw_process_intel_debug_variable(brw); 816 817 if (brw->gen >= 8 && !(INTEL_DEBUG & DEBUG_VEC4VS)) 818 brw->scalar_vs = true; 819 820 brw_initialize_context_constants(brw); 821 822 ctx->Const.ResetStrategy = notify_reset 823 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 824 825 /* Reinitialize the context point state. It depends on ctx->Const values. */ 826 _mesa_init_point(ctx); 827 828 intel_fbo_init(brw); 829 830 intel_batchbuffer_init(brw); 831 832 if (brw->gen >= 6) { 833 /* Create a new hardware context. Using a hardware context means that 834 * our GPU state will be saved/restored on context switch, allowing us 835 * to assume that the GPU is in the same state we left it in. 836 * 837 * This is required for transform feedback buffer offsets, query objects, 838 * and also allows us to reduce how much state we have to emit. 839 */ 840 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 841 842 if (!brw->hw_ctx) { 843 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 844 intelDestroyContext(driContextPriv); 845 return false; 846 } 847 } 848 849 brw_init_state(brw); 850 851 intelInitExtensions(ctx); 852 853 brw_init_surface_formats(brw); 854 855 brw->max_vs_threads = devinfo->max_vs_threads; 856 brw->max_hs_threads = devinfo->max_hs_threads; 857 brw->max_ds_threads = devinfo->max_ds_threads; 858 brw->max_gs_threads = devinfo->max_gs_threads; 859 brw->max_wm_threads = devinfo->max_wm_threads; 860 brw->urb.size = devinfo->urb.size; 861 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 862 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 863 brw->urb.max_hs_entries = devinfo->urb.max_hs_entries; 864 brw->urb.max_ds_entries = devinfo->urb.max_ds_entries; 865 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 866 867 /* Estimate the size of the mappable aperture into the GTT. There's an 868 * ioctl to get the whole GTT size, but not one to get the mappable subset. 869 * It turns out it's basically always 256MB, though some ancient hardware 870 * was smaller. 871 */ 872 uint32_t gtt_size = 256 * 1024 * 1024; 873 874 /* We don't want to map two objects such that a memcpy between them would 875 * just fault one mapping in and then the other over and over forever. So 876 * we would need to divide the GTT size by 2. Additionally, some GTT is 877 * taken up by things like the framebuffer and the ringbuffer and such, so 878 * be more conservative. 879 */ 880 brw->max_gtt_map_object_size = gtt_size / 4; 881 882 if (brw->gen == 6) 883 brw->urb.gs_present = false; 884 885 brw->prim_restart.in_progress = false; 886 brw->prim_restart.enable_cut_index = false; 887 brw->gs.enabled = false; 888 brw->sf.viewport_transform_enable = true; 889 890 ctx->VertexProgram._MaintainTnlProgram = true; 891 ctx->FragmentProgram._MaintainTexEnvProgram = true; 892 893 brw_draw_init( brw ); 894 895 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 896 /* Turn on some extra GL_ARB_debug_output generation. */ 897 brw->perf_debug = true; 898 } 899 900 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 901 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 902 903 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 904 brw_init_shader_time(brw); 905 906 _mesa_compute_version(ctx); 907 908 _mesa_initialize_dispatch_tables(ctx); 909 _mesa_initialize_vbo_vtxfmt(ctx); 910 911 if (ctx->Extensions.AMD_performance_monitor) { 912 brw_init_performance_monitors(brw); 913 } 914 915 vbo_use_buffer_objects(ctx); 916 vbo_always_unmap_buffers(ctx); 917 918 return true; 919} 920 921void 922intelDestroyContext(__DRIcontext * driContextPriv) 923{ 924 struct brw_context *brw = 925 (struct brw_context *) driContextPriv->driverPrivate; 926 struct gl_context *ctx = &brw->ctx; 927 928 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 929 if (INTEL_DEBUG & DEBUG_AUB) { 930 intel_batchbuffer_flush(brw); 931 aub_dump_bmp(&brw->ctx); 932 } 933 934 _mesa_meta_free(&brw->ctx); 935 brw_meta_fast_clear_free(brw); 936 937 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 938 /* Force a report. */ 939 brw->shader_time.report_time = 0; 940 941 brw_collect_and_report_shader_time(brw); 942 brw_destroy_shader_time(brw); 943 } 944 945 brw_destroy_state(brw); 946 brw_draw_destroy(brw); 947 948 drm_intel_bo_unreference(brw->curbe.curbe_bo); 949 if (brw->vs.base.scratch_bo) 950 drm_intel_bo_unreference(brw->vs.base.scratch_bo); 951 if (brw->gs.base.scratch_bo) 952 drm_intel_bo_unreference(brw->gs.base.scratch_bo); 953 if (brw->wm.base.scratch_bo) 954 drm_intel_bo_unreference(brw->wm.base.scratch_bo); 955 956 drm_intel_gem_context_destroy(brw->hw_ctx); 957 958 if (ctx->swrast_context) { 959 _swsetup_DestroyContext(&brw->ctx); 960 _tnl_DestroyContext(&brw->ctx); 961 } 962 _vbo_DestroyContext(&brw->ctx); 963 964 if (ctx->swrast_context) 965 _swrast_DestroyContext(&brw->ctx); 966 967 intel_batchbuffer_free(brw); 968 969 drm_intel_bo_unreference(brw->throttle_batch[1]); 970 drm_intel_bo_unreference(brw->throttle_batch[0]); 971 brw->throttle_batch[1] = NULL; 972 brw->throttle_batch[0] = NULL; 973 974 driDestroyOptionCache(&brw->optionCache); 975 976 /* free the Mesa context */ 977 _mesa_free_context_data(&brw->ctx); 978 979 ralloc_free(brw); 980 driContextPriv->driverPrivate = NULL; 981} 982 983GLboolean 984intelUnbindContext(__DRIcontext * driContextPriv) 985{ 986 /* Unset current context and dispath table */ 987 _mesa_make_current(NULL, NULL, NULL); 988 989 return true; 990} 991 992/** 993 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 994 * on window system framebuffers. 995 * 996 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 997 * your renderbuffer can do sRGB encode, and you can flip a switch that does 998 * sRGB encode if the renderbuffer can handle it. You can ask specifically 999 * for a visual where you're guaranteed to be capable, but it turns out that 1000 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 1001 * incapable ones, becuase there's no difference between the two in resources 1002 * used. Applications thus get built that accidentally rely on the default 1003 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 1004 * great... 1005 * 1006 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 1007 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 1008 * So they removed the enable knob and made it "if the renderbuffer is sRGB 1009 * capable, do sRGB encode". Then, for your window system renderbuffers, you 1010 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 1011 * and get no sRGB encode (assuming that both kinds of visual are available). 1012 * Thus our choice to support sRGB by default on our visuals for desktop would 1013 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 1014 * 1015 * Unfortunately, renderbuffer setup happens before a context is created. So 1016 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 1017 * context (without an sRGB visual, though we don't have sRGB visuals exposed 1018 * yet), we go turn that back off before anyone finds out. 1019 */ 1020static void 1021intel_gles3_srgb_workaround(struct brw_context *brw, 1022 struct gl_framebuffer *fb) 1023{ 1024 struct gl_context *ctx = &brw->ctx; 1025 1026 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 1027 return; 1028 1029 /* Some day when we support the sRGB capable bit on visuals available for 1030 * GLES, we'll need to respect that and not disable things here. 1031 */ 1032 fb->Visual.sRGBCapable = false; 1033 for (int i = 0; i < BUFFER_COUNT; i++) { 1034 if (fb->Attachment[i].Renderbuffer && 1035 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) { 1036 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM; 1037 } 1038 } 1039} 1040 1041GLboolean 1042intelMakeCurrent(__DRIcontext * driContextPriv, 1043 __DRIdrawable * driDrawPriv, 1044 __DRIdrawable * driReadPriv) 1045{ 1046 struct brw_context *brw; 1047 GET_CURRENT_CONTEXT(curCtx); 1048 1049 if (driContextPriv) 1050 brw = (struct brw_context *) driContextPriv->driverPrivate; 1051 else 1052 brw = NULL; 1053 1054 /* According to the glXMakeCurrent() man page: "Pending commands to 1055 * the previous context, if any, are flushed before it is released." 1056 * But only flush if we're actually changing contexts. 1057 */ 1058 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 1059 _mesa_flush(curCtx); 1060 } 1061 1062 if (driContextPriv) { 1063 struct gl_context *ctx = &brw->ctx; 1064 struct gl_framebuffer *fb, *readFb; 1065 1066 if (driDrawPriv == NULL) { 1067 fb = _mesa_get_incomplete_framebuffer(); 1068 } else { 1069 fb = driDrawPriv->driverPrivate; 1070 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 1071 } 1072 1073 if (driReadPriv == NULL) { 1074 readFb = _mesa_get_incomplete_framebuffer(); 1075 } else { 1076 readFb = driReadPriv->driverPrivate; 1077 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 1078 } 1079 1080 /* The sRGB workaround changes the renderbuffer's format. We must change 1081 * the format before the renderbuffer's miptree get's allocated, otherwise 1082 * the formats of the renderbuffer and its miptree will differ. 1083 */ 1084 intel_gles3_srgb_workaround(brw, fb); 1085 intel_gles3_srgb_workaround(brw, readFb); 1086 1087 /* If the context viewport hasn't been initialized, force a call out to 1088 * the loader to get buffers so we have a drawable size for the initial 1089 * viewport. */ 1090 if (!brw->ctx.ViewportInitialized) 1091 intel_prepare_render(brw); 1092 1093 _mesa_make_current(ctx, fb, readFb); 1094 } else { 1095 _mesa_make_current(NULL, NULL, NULL); 1096 } 1097 1098 return true; 1099} 1100 1101void 1102intel_resolve_for_dri2_flush(struct brw_context *brw, 1103 __DRIdrawable *drawable) 1104{ 1105 if (brw->gen < 6) { 1106 /* MSAA and fast color clear are not supported, so don't waste time 1107 * checking whether a resolve is needed. 1108 */ 1109 return; 1110 } 1111 1112 struct gl_framebuffer *fb = drawable->driverPrivate; 1113 struct intel_renderbuffer *rb; 1114 1115 /* Usually, only the back buffer will need to be downsampled. However, 1116 * the front buffer will also need it if the user has rendered into it. 1117 */ 1118 static const gl_buffer_index buffers[2] = { 1119 BUFFER_BACK_LEFT, 1120 BUFFER_FRONT_LEFT, 1121 }; 1122 1123 for (int i = 0; i < 2; ++i) { 1124 rb = intel_get_renderbuffer(fb, buffers[i]); 1125 if (rb == NULL || rb->mt == NULL) 1126 continue; 1127 if (rb->mt->num_samples <= 1) 1128 intel_miptree_resolve_color(brw, rb->mt); 1129 else 1130 intel_renderbuffer_downsample(brw, rb); 1131 } 1132} 1133 1134static unsigned 1135intel_bits_per_pixel(const struct intel_renderbuffer *rb) 1136{ 1137 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 1138} 1139 1140static void 1141intel_query_dri2_buffers(struct brw_context *brw, 1142 __DRIdrawable *drawable, 1143 __DRIbuffer **buffers, 1144 int *count); 1145 1146static void 1147intel_process_dri2_buffer(struct brw_context *brw, 1148 __DRIdrawable *drawable, 1149 __DRIbuffer *buffer, 1150 struct intel_renderbuffer *rb, 1151 const char *buffer_name); 1152 1153static void 1154intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 1155 1156static void 1157intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1158{ 1159 struct gl_framebuffer *fb = drawable->driverPrivate; 1160 struct intel_renderbuffer *rb; 1161 __DRIbuffer *buffers = NULL; 1162 int i, count; 1163 const char *region_name; 1164 1165 /* Set this up front, so that in case our buffers get invalidated 1166 * while we're getting new buffers, we don't clobber the stamp and 1167 * thus ignore the invalidate. */ 1168 drawable->lastStamp = drawable->dri2.stamp; 1169 1170 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1171 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1172 1173 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1174 1175 if (buffers == NULL) 1176 return; 1177 1178 for (i = 0; i < count; i++) { 1179 switch (buffers[i].attachment) { 1180 case __DRI_BUFFER_FRONT_LEFT: 1181 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1182 region_name = "dri2 front buffer"; 1183 break; 1184 1185 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1186 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1187 region_name = "dri2 fake front buffer"; 1188 break; 1189 1190 case __DRI_BUFFER_BACK_LEFT: 1191 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1192 region_name = "dri2 back buffer"; 1193 break; 1194 1195 case __DRI_BUFFER_DEPTH: 1196 case __DRI_BUFFER_HIZ: 1197 case __DRI_BUFFER_DEPTH_STENCIL: 1198 case __DRI_BUFFER_STENCIL: 1199 case __DRI_BUFFER_ACCUM: 1200 default: 1201 fprintf(stderr, 1202 "unhandled buffer attach event, attachment type %d\n", 1203 buffers[i].attachment); 1204 return; 1205 } 1206 1207 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1208 } 1209 1210} 1211 1212void 1213intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1214{ 1215 struct brw_context *brw = context->driverPrivate; 1216 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1217 1218 /* Set this up front, so that in case our buffers get invalidated 1219 * while we're getting new buffers, we don't clobber the stamp and 1220 * thus ignore the invalidate. */ 1221 drawable->lastStamp = drawable->dri2.stamp; 1222 1223 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1224 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1225 1226 if (screen->image.loader) 1227 intel_update_image_buffers(brw, drawable); 1228 else 1229 intel_update_dri2_buffers(brw, drawable); 1230 1231 driUpdateFramebufferSize(&brw->ctx, drawable); 1232} 1233 1234/** 1235 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1236 * state is required. 1237 */ 1238void 1239intel_prepare_render(struct brw_context *brw) 1240{ 1241 struct gl_context *ctx = &brw->ctx; 1242 __DRIcontext *driContext = brw->driContext; 1243 __DRIdrawable *drawable; 1244 1245 drawable = driContext->driDrawablePriv; 1246 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1247 if (drawable->lastStamp != drawable->dri2.stamp) 1248 intel_update_renderbuffers(driContext, drawable); 1249 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1250 } 1251 1252 drawable = driContext->driReadablePriv; 1253 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1254 if (drawable->lastStamp != drawable->dri2.stamp) 1255 intel_update_renderbuffers(driContext, drawable); 1256 driContext->dri2.read_stamp = drawable->dri2.stamp; 1257 } 1258 1259 /* If we're currently rendering to the front buffer, the rendering 1260 * that will happen next will probably dirty the front buffer. So 1261 * mark it as dirty here. 1262 */ 1263 if (brw_is_front_buffer_drawing(ctx->DrawBuffer)) 1264 brw->front_buffer_dirty = true; 1265} 1266 1267/** 1268 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1269 * 1270 * To determine which DRI buffers to request, examine the renderbuffers 1271 * attached to the drawable's framebuffer. Then request the buffers with 1272 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1273 * 1274 * This is called from intel_update_renderbuffers(). 1275 * 1276 * \param drawable Drawable whose buffers are queried. 1277 * \param buffers [out] List of buffers returned by DRI2 query. 1278 * \param buffer_count [out] Number of buffers returned. 1279 * 1280 * \see intel_update_renderbuffers() 1281 * \see DRI2GetBuffers() 1282 * \see DRI2GetBuffersWithFormat() 1283 */ 1284static void 1285intel_query_dri2_buffers(struct brw_context *brw, 1286 __DRIdrawable *drawable, 1287 __DRIbuffer **buffers, 1288 int *buffer_count) 1289{ 1290 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1291 struct gl_framebuffer *fb = drawable->driverPrivate; 1292 int i = 0; 1293 unsigned attachments[8]; 1294 1295 struct intel_renderbuffer *front_rb; 1296 struct intel_renderbuffer *back_rb; 1297 1298 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1299 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1300 1301 memset(attachments, 0, sizeof(attachments)); 1302 if ((brw_is_front_buffer_drawing(fb) || 1303 brw_is_front_buffer_reading(fb) || 1304 !back_rb) && front_rb) { 1305 /* If a fake front buffer is in use, then querying for 1306 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1307 * the real front buffer to the fake front buffer. So before doing the 1308 * query, we need to make sure all the pending drawing has landed in the 1309 * real front buffer. 1310 */ 1311 intel_batchbuffer_flush(brw); 1312 intel_flush_front(&brw->ctx); 1313 1314 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1315 attachments[i++] = intel_bits_per_pixel(front_rb); 1316 } else if (front_rb && brw->front_buffer_dirty) { 1317 /* We have pending front buffer rendering, but we aren't querying for a 1318 * front buffer. If the front buffer we have is a fake front buffer, 1319 * the X server is going to throw it away when it processes the query. 1320 * So before doing the query, make sure all the pending drawing has 1321 * landed in the real front buffer. 1322 */ 1323 intel_batchbuffer_flush(brw); 1324 intel_flush_front(&brw->ctx); 1325 } 1326 1327 if (back_rb) { 1328 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1329 attachments[i++] = intel_bits_per_pixel(back_rb); 1330 } 1331 1332 assert(i <= ARRAY_SIZE(attachments)); 1333 1334 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1335 &drawable->w, 1336 &drawable->h, 1337 attachments, i / 2, 1338 buffer_count, 1339 drawable->loaderPrivate); 1340} 1341 1342/** 1343 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1344 * 1345 * This is called from intel_update_renderbuffers(). 1346 * 1347 * \par Note: 1348 * DRI buffers whose attachment point is DRI2BufferStencil or 1349 * DRI2BufferDepthStencil are handled as special cases. 1350 * 1351 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1352 * that is passed to drm_intel_bo_gem_create_from_name(). 1353 * 1354 * \see intel_update_renderbuffers() 1355 */ 1356static void 1357intel_process_dri2_buffer(struct brw_context *brw, 1358 __DRIdrawable *drawable, 1359 __DRIbuffer *buffer, 1360 struct intel_renderbuffer *rb, 1361 const char *buffer_name) 1362{ 1363 struct gl_framebuffer *fb = drawable->driverPrivate; 1364 drm_intel_bo *bo; 1365 1366 if (!rb) 1367 return; 1368 1369 unsigned num_samples = rb->Base.Base.NumSamples; 1370 1371 /* We try to avoid closing and reopening the same BO name, because the first 1372 * use of a mapping of the buffer involves a bunch of page faulting which is 1373 * moderately expensive. 1374 */ 1375 struct intel_mipmap_tree *last_mt; 1376 if (num_samples == 0) 1377 last_mt = rb->mt; 1378 else 1379 last_mt = rb->singlesample_mt; 1380 1381 uint32_t old_name = 0; 1382 if (last_mt) { 1383 /* The bo already has a name because the miptree was created by a 1384 * previous call to intel_process_dri2_buffer(). If a bo already has a 1385 * name, then drm_intel_bo_flink() is a low-cost getter. It does not 1386 * create a new name. 1387 */ 1388 drm_intel_bo_flink(last_mt->bo, &old_name); 1389 } 1390 1391 if (old_name == buffer->name) 1392 return; 1393 1394 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1395 fprintf(stderr, 1396 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1397 buffer->name, buffer->attachment, 1398 buffer->cpp, buffer->pitch); 1399 } 1400 1401 intel_miptree_release(&rb->mt); 1402 bo = drm_intel_bo_gem_create_from_name(brw->bufmgr, buffer_name, 1403 buffer->name); 1404 if (!bo) { 1405 fprintf(stderr, 1406 "Failed to open BO for returned DRI2 buffer " 1407 "(%dx%d, %s, named %d).\n" 1408 "This is likely a bug in the X Server that will lead to a " 1409 "crash soon.\n", 1410 drawable->w, drawable->h, buffer_name, buffer->name); 1411 return; 1412 } 1413 1414 intel_update_winsys_renderbuffer_miptree(brw, rb, bo, 1415 drawable->w, drawable->h, 1416 buffer->pitch); 1417 1418 if (brw_is_front_buffer_drawing(fb) && 1419 (buffer->attachment == __DRI_BUFFER_FRONT_LEFT || 1420 buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) && 1421 rb->Base.Base.NumSamples > 1) { 1422 intel_renderbuffer_upsample(brw, rb); 1423 } 1424 1425 assert(rb->mt); 1426 1427 drm_intel_bo_unreference(bo); 1428} 1429 1430/** 1431 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1432 * 1433 * To determine which DRI buffers to request, examine the renderbuffers 1434 * attached to the drawable's framebuffer. Then request the buffers from 1435 * the image loader 1436 * 1437 * This is called from intel_update_renderbuffers(). 1438 * 1439 * \param drawable Drawable whose buffers are queried. 1440 * \param buffers [out] List of buffers returned by DRI2 query. 1441 * \param buffer_count [out] Number of buffers returned. 1442 * 1443 * \see intel_update_renderbuffers() 1444 */ 1445 1446static void 1447intel_update_image_buffer(struct brw_context *intel, 1448 __DRIdrawable *drawable, 1449 struct intel_renderbuffer *rb, 1450 __DRIimage *buffer, 1451 enum __DRIimageBufferMask buffer_type) 1452{ 1453 struct gl_framebuffer *fb = drawable->driverPrivate; 1454 1455 if (!rb || !buffer->bo) 1456 return; 1457 1458 unsigned num_samples = rb->Base.Base.NumSamples; 1459 1460 /* Check and see if we're already bound to the right 1461 * buffer object 1462 */ 1463 struct intel_mipmap_tree *last_mt; 1464 if (num_samples == 0) 1465 last_mt = rb->mt; 1466 else 1467 last_mt = rb->singlesample_mt; 1468 1469 if (last_mt && last_mt->bo == buffer->bo) 1470 return; 1471 1472 intel_update_winsys_renderbuffer_miptree(intel, rb, buffer->bo, 1473 buffer->width, buffer->height, 1474 buffer->pitch); 1475 1476 if (brw_is_front_buffer_drawing(fb) && 1477 buffer_type == __DRI_IMAGE_BUFFER_FRONT && 1478 rb->Base.Base.NumSamples > 1) { 1479 intel_renderbuffer_upsample(intel, rb); 1480 } 1481} 1482 1483static void 1484intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1485{ 1486 struct gl_framebuffer *fb = drawable->driverPrivate; 1487 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1488 struct intel_renderbuffer *front_rb; 1489 struct intel_renderbuffer *back_rb; 1490 struct __DRIimageList images; 1491 unsigned int format; 1492 uint32_t buffer_mask = 0; 1493 1494 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1495 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1496 1497 if (back_rb) 1498 format = intel_rb_format(back_rb); 1499 else if (front_rb) 1500 format = intel_rb_format(front_rb); 1501 else 1502 return; 1503 1504 if (front_rb && (brw_is_front_buffer_drawing(fb) || 1505 brw_is_front_buffer_reading(fb) || !back_rb)) { 1506 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1507 } 1508 1509 if (back_rb) 1510 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1511 1512 (*screen->image.loader->getBuffers) (drawable, 1513 driGLFormatToImageFormat(format), 1514 &drawable->dri2.stamp, 1515 drawable->loaderPrivate, 1516 buffer_mask, 1517 &images); 1518 1519 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1520 drawable->w = images.front->width; 1521 drawable->h = images.front->height; 1522 intel_update_image_buffer(brw, 1523 drawable, 1524 front_rb, 1525 images.front, 1526 __DRI_IMAGE_BUFFER_FRONT); 1527 } 1528 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1529 drawable->w = images.back->width; 1530 drawable->h = images.back->height; 1531 intel_update_image_buffer(brw, 1532 drawable, 1533 back_rb, 1534 images.back, 1535 __DRI_IMAGE_BUFFER_BACK); 1536 } 1537} 1538