brw_context.c revision 1bc3b62d4aad22b94b8031c29c654a8f90ccc24d
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 brw_init_compute_functions(functions); 292 if (brw->gen >= 7) 293 brw_init_conditional_render_functions(functions); 294 295 functions->QuerySamplesForFormat = brw_query_samples_for_format; 296 297 functions->NewTransformFeedback = brw_new_transform_feedback; 298 functions->DeleteTransformFeedback = brw_delete_transform_feedback; 299 functions->GetTransformFeedbackVertexCount = 300 brw_get_transform_feedback_vertex_count; 301 if (brw->gen >= 7) { 302 functions->BeginTransformFeedback = gen7_begin_transform_feedback; 303 functions->EndTransformFeedback = gen7_end_transform_feedback; 304 functions->PauseTransformFeedback = gen7_pause_transform_feedback; 305 functions->ResumeTransformFeedback = gen7_resume_transform_feedback; 306 } else { 307 functions->BeginTransformFeedback = brw_begin_transform_feedback; 308 functions->EndTransformFeedback = brw_end_transform_feedback; 309 } 310 311 if (brw->gen >= 6) 312 functions->GetSamplePosition = gen6_get_sample_position; 313} 314 315static void 316brw_initialize_context_constants(struct brw_context *brw) 317{ 318 struct gl_context *ctx = &brw->ctx; 319 320 unsigned max_samplers = 321 brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16; 322 323 ctx->Const.QueryCounterBits.Timestamp = 36; 324 325 ctx->Const.StripTextureBorder = true; 326 327 ctx->Const.MaxDualSourceDrawBuffers = 1; 328 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS; 329 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits = max_samplers; 330 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */ 331 ctx->Const.MaxTextureUnits = 332 MIN2(ctx->Const.MaxTextureCoordUnits, 333 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits); 334 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits = max_samplers; 335 if (brw->gen >= 6) 336 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = max_samplers; 337 else 338 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits = 0; 339 if (_mesa_extension_override_enables.ARB_compute_shader) { 340 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits = BRW_MAX_TEX_UNIT; 341 ctx->Const.MaxUniformBufferBindings += 12; 342 } else { 343 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits = 0; 344 } 345 ctx->Const.MaxCombinedTextureImageUnits = 346 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits + 347 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits + 348 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits + 349 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits; 350 351 ctx->Const.MaxTextureLevels = 14; /* 8192 */ 352 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS) 353 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS; 354 ctx->Const.Max3DTextureLevels = 12; /* 2048 */ 355 ctx->Const.MaxCubeTextureLevels = 14; /* 8192 */ 356 ctx->Const.MaxTextureMbytes = 1536; 357 358 if (brw->gen >= 7) 359 ctx->Const.MaxArrayTextureLayers = 2048; 360 else 361 ctx->Const.MaxArrayTextureLayers = 512; 362 363 ctx->Const.MaxTextureRectSize = 1 << 12; 364 365 ctx->Const.MaxTextureMaxAnisotropy = 16.0; 366 367 ctx->Const.MaxRenderbufferSize = 8192; 368 369 /* Hardware only supports a limited number of transform feedback buffers. 370 * So we need to override the Mesa default (which is based only on software 371 * limits). 372 */ 373 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS; 374 375 /* On Gen6, in the worst case, we use up one binding table entry per 376 * transform feedback component (see comments above the definition of 377 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value 378 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to 379 * BRW_MAX_SOL_BINDINGS. 380 * 381 * In "separate components" mode, we need to divide this value by 382 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries 383 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS. 384 */ 385 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS; 386 ctx->Const.MaxTransformFeedbackSeparateComponents = 387 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS; 388 389 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true; 390 391 int max_samples; 392 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen); 393 const int clamp_max_samples = 394 driQueryOptioni(&brw->optionCache, "clamp_max_samples"); 395 396 if (clamp_max_samples < 0) { 397 max_samples = msaa_modes[0]; 398 } else { 399 /* Select the largest supported MSAA mode that does not exceed 400 * clamp_max_samples. 401 */ 402 max_samples = 0; 403 for (int i = 0; msaa_modes[i] != 0; ++i) { 404 if (msaa_modes[i] <= clamp_max_samples) { 405 max_samples = msaa_modes[i]; 406 break; 407 } 408 } 409 } 410 411 ctx->Const.MaxSamples = max_samples; 412 ctx->Const.MaxColorTextureSamples = max_samples; 413 ctx->Const.MaxDepthTextureSamples = max_samples; 414 ctx->Const.MaxIntegerSamples = max_samples; 415 416 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used 417 * to map indices of rectangular grid to sample numbers within a pixel. 418 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled 419 * extension implementation. For more details see the comment above 420 * gen6_set_sample_maps() definition. 421 */ 422 gen6_set_sample_maps(ctx); 423 424 if (brw->gen >= 7) 425 ctx->Const.MaxProgramTextureGatherComponents = 4; 426 else if (brw->gen == 6) 427 ctx->Const.MaxProgramTextureGatherComponents = 1; 428 429 ctx->Const.MinLineWidth = 1.0; 430 ctx->Const.MinLineWidthAA = 1.0; 431 if (brw->gen >= 6) { 432 ctx->Const.MaxLineWidth = 7.375; 433 ctx->Const.MaxLineWidthAA = 7.375; 434 ctx->Const.LineWidthGranularity = 0.125; 435 } else { 436 ctx->Const.MaxLineWidth = 7.0; 437 ctx->Const.MaxLineWidthAA = 7.0; 438 ctx->Const.LineWidthGranularity = 0.5; 439 } 440 441 /* For non-antialiased lines, we have to round the line width to the 442 * nearest whole number. Make sure that we don't advertise a line 443 * width that, when rounded, will be beyond the actual hardware 444 * maximum. 445 */ 446 assert(roundf(ctx->Const.MaxLineWidth) <= ctx->Const.MaxLineWidth); 447 448 ctx->Const.MinPointSize = 1.0; 449 ctx->Const.MinPointSizeAA = 1.0; 450 ctx->Const.MaxPointSize = 255.0; 451 ctx->Const.MaxPointSizeAA = 255.0; 452 ctx->Const.PointSizeGranularity = 1.0; 453 454 if (brw->gen >= 5 || brw->is_g4x) 455 ctx->Const.MaxClipPlanes = 8; 456 457 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024; 458 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0; 459 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0; 460 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0; 461 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0; 462 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0; 463 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0; 464 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16; 465 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256; 466 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1; 467 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024; 468 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams = 469 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters, 470 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams); 471 472 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024; 473 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024; 474 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024; 475 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024; 476 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12; 477 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256; 478 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0; 479 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024; 480 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams = 481 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters, 482 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams); 483 484 /* Fragment shaders use real, 32-bit twos-complement integers for all 485 * integer types. 486 */ 487 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31; 488 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30; 489 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0; 490 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 491 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt; 492 493 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMin = 31; 494 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMax = 30; 495 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.Precision = 0; 496 ctx->Const.Program[MESA_SHADER_VERTEX].HighInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt; 497 ctx->Const.Program[MESA_SHADER_VERTEX].MediumInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt; 498 499 if (brw->gen >= 7) { 500 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 501 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 502 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 503 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxAtomicCounters = MAX_ATOMIC_COUNTERS; 504 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAtomicBuffers = BRW_MAX_ABO; 505 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAtomicBuffers = BRW_MAX_ABO; 506 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxAtomicBuffers = BRW_MAX_ABO; 507 ctx->Const.Program[MESA_SHADER_COMPUTE].MaxAtomicBuffers = BRW_MAX_ABO; 508 ctx->Const.MaxCombinedAtomicBuffers = 3 * BRW_MAX_ABO; 509 } 510 511 /* Gen6 converts quads to polygon in beginning of 3D pipeline, 512 * but we're not sure how it's actually done for vertex order, 513 * that affect provoking vertex decision. Always use last vertex 514 * convention for quad primitive which works as expected for now. 515 */ 516 if (brw->gen >= 6) 517 ctx->Const.QuadsFollowProvokingVertexConvention = false; 518 519 ctx->Const.NativeIntegers = true; 520 ctx->Const.VertexID_is_zero_based = true; 521 522 /* Regarding the CMP instruction, the Ivybridge PRM says: 523 * 524 * "For each enabled channel 0b or 1b is assigned to the appropriate flag 525 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord 526 * 0xFFFFFFFF) is assigned to dst." 527 * 528 * but PRMs for earlier generations say 529 * 530 * "In dword format, one GRF may store up to 8 results. When the register 531 * is used later as a vector of Booleans, as only LSB at each channel 532 * contains meaning [sic] data, software should make sure all higher bits 533 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)." 534 * 535 * We select the representation of a true boolean uniform to be ~0, and fix 536 * the results of Gen <= 5 CMP instruction's with -(result & 1). 537 */ 538 ctx->Const.UniformBooleanTrue = ~0; 539 540 /* From the gen4 PRM, volume 4 page 127: 541 * 542 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies 543 * the base address of the first element of the surface, computed in 544 * software by adding the surface base address to the byte offset of 545 * the element in the buffer." 546 * 547 * However, unaligned accesses are slower, so enforce buffer alignment. 548 */ 549 ctx->Const.UniformBufferOffsetAlignment = 16; 550 ctx->Const.TextureBufferOffsetAlignment = 16; 551 ctx->Const.MaxTextureBufferSize = 128 * 1024 * 1024; 552 553 if (brw->gen >= 6) { 554 ctx->Const.MaxVarying = 32; 555 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128; 556 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64; 557 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128; 558 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128; 559 } 560 561 static const nir_shader_compiler_options nir_options = { 562 .native_integers = true, 563 /* In order to help allow for better CSE at the NIR level we tell NIR 564 * to split all ffma instructions during opt_algebraic and we then 565 * re-combine them as a later step. 566 */ 567 .lower_ffma = true, 568 .lower_sub = true, 569 }; 570 571 /* We want the GLSL compiler to emit code that uses condition codes */ 572 for (int i = 0; i < MESA_SHADER_STAGES; i++) { 573 ctx->Const.ShaderCompilerOptions[i].MaxIfDepth = brw->gen < 6 ? 16 : UINT_MAX; 574 ctx->Const.ShaderCompilerOptions[i].EmitCondCodes = true; 575 ctx->Const.ShaderCompilerOptions[i].EmitNoNoise = true; 576 ctx->Const.ShaderCompilerOptions[i].EmitNoMainReturn = true; 577 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectInput = true; 578 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectOutput = 579 (i == MESA_SHADER_FRAGMENT); 580 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectTemp = 581 (i == MESA_SHADER_FRAGMENT); 582 ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectUniform = false; 583 ctx->Const.ShaderCompilerOptions[i].LowerClipDistance = true; 584 } 585 586 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = true; 587 ctx->Const.ShaderCompilerOptions[MESA_SHADER_GEOMETRY].OptimizeForAOS = true; 588 589 if (brw->scalar_vs) { 590 /* If we're using the scalar backend for vertex shaders, we need to 591 * configure these accordingly. 592 */ 593 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].EmitNoIndirectOutput = true; 594 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].EmitNoIndirectTemp = true; 595 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = false; 596 597 ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].NirOptions = &nir_options; 598 } 599 600 ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT].NirOptions = &nir_options; 601 ctx->Const.ShaderCompilerOptions[MESA_SHADER_COMPUTE].NirOptions = &nir_options; 602 603 /* ARB_viewport_array */ 604 if (brw->gen >= 6 && ctx->API == API_OPENGL_CORE) { 605 ctx->Const.MaxViewports = GEN6_NUM_VIEWPORTS; 606 ctx->Const.ViewportSubpixelBits = 0; 607 608 /* Cast to float before negating because MaxViewportWidth is unsigned. 609 */ 610 ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth; 611 ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth; 612 } 613 614 /* ARB_gpu_shader5 */ 615 if (brw->gen >= 7) 616 ctx->Const.MaxVertexStreams = MIN2(4, MAX_VERTEX_STREAMS); 617 618 /* ARB_framebuffer_no_attachments */ 619 ctx->Const.MaxFramebufferWidth = ctx->Const.MaxViewportWidth; 620 ctx->Const.MaxFramebufferHeight = ctx->Const.MaxViewportHeight; 621 ctx->Const.MaxFramebufferLayers = ctx->Const.MaxArrayTextureLayers; 622 ctx->Const.MaxFramebufferSamples = max_samples; 623} 624 625static void 626brw_adjust_cs_context_constants(struct brw_context *brw) 627{ 628 struct gl_context *ctx = &brw->ctx; 629 630 /* For ES, we set these constants based on SIMD8. 631 * 632 * TODO: Once we can always generate SIMD16, we should update this. 633 * 634 * For GL, we assume we can generate a SIMD16 program, but this currently 635 * is not always true. This allows us to run more test cases, and will be 636 * required based on desktop GL compute shader requirements. 637 */ 638 const int simd_size = ctx->API == API_OPENGL_CORE ? 16 : 8; 639 640 const uint32_t max_invocations = simd_size * brw->max_cs_threads; 641 ctx->Const.MaxComputeWorkGroupSize[0] = max_invocations; 642 ctx->Const.MaxComputeWorkGroupSize[1] = max_invocations; 643 ctx->Const.MaxComputeWorkGroupSize[2] = max_invocations; 644 ctx->Const.MaxComputeWorkGroupInvocations = max_invocations; 645} 646 647/** 648 * Process driconf (drirc) options, setting appropriate context flags. 649 * 650 * intelInitExtensions still pokes at optionCache directly, in order to 651 * avoid advertising various extensions. No flags are set, so it makes 652 * sense to continue doing that there. 653 */ 654static void 655brw_process_driconf_options(struct brw_context *brw) 656{ 657 struct gl_context *ctx = &brw->ctx; 658 659 driOptionCache *options = &brw->optionCache; 660 driParseConfigFiles(options, &brw->intelScreen->optionCache, 661 brw->driContext->driScreenPriv->myNum, "i965"); 662 663 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse"); 664 switch (bo_reuse_mode) { 665 case DRI_CONF_BO_REUSE_DISABLED: 666 break; 667 case DRI_CONF_BO_REUSE_ALL: 668 intel_bufmgr_gem_enable_reuse(brw->bufmgr); 669 break; 670 } 671 672 if (!driQueryOptionb(options, "hiz")) { 673 brw->has_hiz = false; 674 /* On gen6, you can only do separate stencil with HIZ. */ 675 if (brw->gen == 6) 676 brw->has_separate_stencil = false; 677 } 678 679 if (driQueryOptionb(options, "always_flush_batch")) { 680 fprintf(stderr, "flushing batchbuffer before/after each draw call\n"); 681 brw->always_flush_batch = true; 682 } 683 684 if (driQueryOptionb(options, "always_flush_cache")) { 685 fprintf(stderr, "flushing GPU caches before/after each draw call\n"); 686 brw->always_flush_cache = true; 687 } 688 689 if (driQueryOptionb(options, "disable_throttling")) { 690 fprintf(stderr, "disabling flush throttling\n"); 691 brw->disable_throttling = true; 692 } 693 694 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile"); 695 696 ctx->Const.ForceGLSLExtensionsWarn = 697 driQueryOptionb(options, "force_glsl_extensions_warn"); 698 699 ctx->Const.DisableGLSLLineContinuations = 700 driQueryOptionb(options, "disable_glsl_line_continuations"); 701 702 ctx->Const.AllowGLSLExtensionDirectiveMidShader = 703 driQueryOptionb(options, "allow_glsl_extension_directive_midshader"); 704} 705 706GLboolean 707brwCreateContext(gl_api api, 708 const struct gl_config *mesaVis, 709 __DRIcontext *driContextPriv, 710 unsigned major_version, 711 unsigned minor_version, 712 uint32_t flags, 713 bool notify_reset, 714 unsigned *dri_ctx_error, 715 void *sharedContextPrivate) 716{ 717 __DRIscreen *sPriv = driContextPriv->driScreenPriv; 718 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate; 719 struct intel_screen *screen = sPriv->driverPrivate; 720 const struct brw_device_info *devinfo = screen->devinfo; 721 struct dd_function_table functions; 722 723 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel 724 * provides us with context reset notifications. 725 */ 726 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG 727 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE; 728 729 if (screen->has_context_reset_notification) 730 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS; 731 732 if (flags & ~allowed_flags) { 733 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG; 734 return false; 735 } 736 737 struct brw_context *brw = rzalloc(NULL, struct brw_context); 738 if (!brw) { 739 fprintf(stderr, "%s: failed to alloc context\n", __func__); 740 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 741 return false; 742 } 743 744 driContextPriv->driverPrivate = brw; 745 brw->driContext = driContextPriv; 746 brw->intelScreen = screen; 747 brw->bufmgr = screen->bufmgr; 748 749 brw->gen = devinfo->gen; 750 brw->gt = devinfo->gt; 751 brw->is_g4x = devinfo->is_g4x; 752 brw->is_baytrail = devinfo->is_baytrail; 753 brw->is_haswell = devinfo->is_haswell; 754 brw->is_cherryview = devinfo->is_cherryview; 755 brw->has_llc = devinfo->has_llc; 756 brw->has_hiz = devinfo->has_hiz_and_separate_stencil; 757 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil; 758 brw->has_pln = devinfo->has_pln; 759 brw->has_compr4 = devinfo->has_compr4; 760 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset; 761 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug; 762 brw->needs_unlit_centroid_workaround = 763 devinfo->needs_unlit_centroid_workaround; 764 765 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil; 766 brw->has_swizzling = screen->hw_has_swizzling; 767 768 brw->vs.base.stage = MESA_SHADER_VERTEX; 769 brw->gs.base.stage = MESA_SHADER_GEOMETRY; 770 brw->wm.base.stage = MESA_SHADER_FRAGMENT; 771 if (brw->gen >= 8) { 772 gen8_init_vtable_surface_functions(brw); 773 brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz; 774 } else if (brw->gen >= 7) { 775 gen7_init_vtable_surface_functions(brw); 776 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; 777 } else if (brw->gen >= 6) { 778 gen6_init_vtable_surface_functions(brw); 779 brw->vtbl.emit_depth_stencil_hiz = gen6_emit_depth_stencil_hiz; 780 } else { 781 gen4_init_vtable_surface_functions(brw); 782 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz; 783 } 784 785 brw_init_driver_functions(brw, &functions); 786 787 if (notify_reset) 788 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status; 789 790 struct gl_context *ctx = &brw->ctx; 791 792 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) { 793 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY; 794 fprintf(stderr, "%s: failed to init mesa context\n", __func__); 795 intelDestroyContext(driContextPriv); 796 return false; 797 } 798 799 driContextSetFlags(ctx, flags); 800 801 /* Initialize the software rasterizer and helper modules. 802 * 803 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for 804 * software fallbacks (which we have to support on legacy GL to do weird 805 * glDrawPixels(), glBitmap(), and other functions). 806 */ 807 if (api != API_OPENGL_CORE && api != API_OPENGLES2) { 808 _swrast_CreateContext(ctx); 809 } 810 811 _vbo_CreateContext(ctx); 812 if (ctx->swrast_context) { 813 _tnl_CreateContext(ctx); 814 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; 815 _swsetup_CreateContext(ctx); 816 817 /* Configure swrast to match hardware characteristics: */ 818 _swrast_allow_pixel_fog(ctx, false); 819 _swrast_allow_vertex_fog(ctx, true); 820 } 821 822 _mesa_meta_init(ctx); 823 824 brw_process_driconf_options(brw); 825 826 if (INTEL_DEBUG & DEBUG_PERF) 827 brw->perf_debug = true; 828 829 if (brw->gen >= 8 && !(INTEL_DEBUG & DEBUG_VEC4VS)) 830 brw->scalar_vs = true; 831 832 brw_initialize_context_constants(brw); 833 834 ctx->Const.ResetStrategy = notify_reset 835 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB; 836 837 /* Reinitialize the context point state. It depends on ctx->Const values. */ 838 _mesa_init_point(ctx); 839 840 intel_fbo_init(brw); 841 842 intel_batchbuffer_init(brw); 843 844 if (brw->gen >= 6) { 845 /* Create a new hardware context. Using a hardware context means that 846 * our GPU state will be saved/restored on context switch, allowing us 847 * to assume that the GPU is in the same state we left it in. 848 * 849 * This is required for transform feedback buffer offsets, query objects, 850 * and also allows us to reduce how much state we have to emit. 851 */ 852 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr); 853 854 if (!brw->hw_ctx) { 855 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n"); 856 intelDestroyContext(driContextPriv); 857 return false; 858 } 859 } 860 861 brw_init_state(brw); 862 863 intelInitExtensions(ctx); 864 865 brw_init_surface_formats(brw); 866 867 brw->max_vs_threads = devinfo->max_vs_threads; 868 brw->max_hs_threads = devinfo->max_hs_threads; 869 brw->max_ds_threads = devinfo->max_ds_threads; 870 brw->max_gs_threads = devinfo->max_gs_threads; 871 brw->max_wm_threads = devinfo->max_wm_threads; 872 brw->max_cs_threads = devinfo->max_cs_threads; 873 brw->urb.size = devinfo->urb.size; 874 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries; 875 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries; 876 brw->urb.max_hs_entries = devinfo->urb.max_hs_entries; 877 brw->urb.max_ds_entries = devinfo->urb.max_ds_entries; 878 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries; 879 880 brw_adjust_cs_context_constants(brw); 881 882 /* Estimate the size of the mappable aperture into the GTT. There's an 883 * ioctl to get the whole GTT size, but not one to get the mappable subset. 884 * It turns out it's basically always 256MB, though some ancient hardware 885 * was smaller. 886 */ 887 uint32_t gtt_size = 256 * 1024 * 1024; 888 889 /* We don't want to map two objects such that a memcpy between them would 890 * just fault one mapping in and then the other over and over forever. So 891 * we would need to divide the GTT size by 2. Additionally, some GTT is 892 * taken up by things like the framebuffer and the ringbuffer and such, so 893 * be more conservative. 894 */ 895 brw->max_gtt_map_object_size = gtt_size / 4; 896 897 if (brw->gen == 6) 898 brw->urb.gs_present = false; 899 900 brw->prim_restart.in_progress = false; 901 brw->prim_restart.enable_cut_index = false; 902 brw->gs.enabled = false; 903 brw->sf.viewport_transform_enable = true; 904 905 brw->predicate.state = BRW_PREDICATE_STATE_RENDER; 906 907 ctx->VertexProgram._MaintainTnlProgram = true; 908 ctx->FragmentProgram._MaintainTexEnvProgram = true; 909 910 brw_draw_init( brw ); 911 912 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) { 913 /* Turn on some extra GL_ARB_debug_output generation. */ 914 brw->perf_debug = true; 915 } 916 917 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0) 918 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB; 919 920 if (INTEL_DEBUG & DEBUG_SHADER_TIME) 921 brw_init_shader_time(brw); 922 923 _mesa_compute_version(ctx); 924 925 _mesa_initialize_dispatch_tables(ctx); 926 _mesa_initialize_vbo_vtxfmt(ctx); 927 928 if (ctx->Extensions.AMD_performance_monitor) { 929 brw_init_performance_monitors(brw); 930 } 931 932 vbo_use_buffer_objects(ctx); 933 vbo_always_unmap_buffers(ctx); 934 935 return true; 936} 937 938void 939intelDestroyContext(__DRIcontext * driContextPriv) 940{ 941 struct brw_context *brw = 942 (struct brw_context *) driContextPriv->driverPrivate; 943 struct gl_context *ctx = &brw->ctx; 944 945 /* Dump a final BMP in case the application doesn't call SwapBuffers */ 946 if (INTEL_DEBUG & DEBUG_AUB) { 947 intel_batchbuffer_flush(brw); 948 aub_dump_bmp(&brw->ctx); 949 } 950 951 _mesa_meta_free(&brw->ctx); 952 brw_meta_fast_clear_free(brw); 953 954 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 955 /* Force a report. */ 956 brw->shader_time.report_time = 0; 957 958 brw_collect_and_report_shader_time(brw); 959 brw_destroy_shader_time(brw); 960 } 961 962 brw_destroy_state(brw); 963 brw_draw_destroy(brw); 964 965 drm_intel_bo_unreference(brw->curbe.curbe_bo); 966 if (brw->vs.base.scratch_bo) 967 drm_intel_bo_unreference(brw->vs.base.scratch_bo); 968 if (brw->gs.base.scratch_bo) 969 drm_intel_bo_unreference(brw->gs.base.scratch_bo); 970 if (brw->wm.base.scratch_bo) 971 drm_intel_bo_unreference(brw->wm.base.scratch_bo); 972 973 drm_intel_gem_context_destroy(brw->hw_ctx); 974 975 if (ctx->swrast_context) { 976 _swsetup_DestroyContext(&brw->ctx); 977 _tnl_DestroyContext(&brw->ctx); 978 } 979 _vbo_DestroyContext(&brw->ctx); 980 981 if (ctx->swrast_context) 982 _swrast_DestroyContext(&brw->ctx); 983 984 intel_batchbuffer_free(brw); 985 986 drm_intel_bo_unreference(brw->throttle_batch[1]); 987 drm_intel_bo_unreference(brw->throttle_batch[0]); 988 brw->throttle_batch[1] = NULL; 989 brw->throttle_batch[0] = NULL; 990 991 driDestroyOptionCache(&brw->optionCache); 992 993 /* free the Mesa context */ 994 _mesa_free_context_data(&brw->ctx); 995 996 ralloc_free(brw); 997 driContextPriv->driverPrivate = NULL; 998} 999 1000GLboolean 1001intelUnbindContext(__DRIcontext * driContextPriv) 1002{ 1003 /* Unset current context and dispath table */ 1004 _mesa_make_current(NULL, NULL, NULL); 1005 1006 return true; 1007} 1008 1009/** 1010 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior 1011 * on window system framebuffers. 1012 * 1013 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if 1014 * your renderbuffer can do sRGB encode, and you can flip a switch that does 1015 * sRGB encode if the renderbuffer can handle it. You can ask specifically 1016 * for a visual where you're guaranteed to be capable, but it turns out that 1017 * everyone just makes all their ARGB8888 visuals capable and doesn't offer 1018 * incapable ones, because there's no difference between the two in resources 1019 * used. Applications thus get built that accidentally rely on the default 1020 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds 1021 * great... 1022 * 1023 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode 1024 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent. 1025 * So they removed the enable knob and made it "if the renderbuffer is sRGB 1026 * capable, do sRGB encode". Then, for your window system renderbuffers, you 1027 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals 1028 * and get no sRGB encode (assuming that both kinds of visual are available). 1029 * Thus our choice to support sRGB by default on our visuals for desktop would 1030 * result in broken rendering of GLES apps that aren't expecting sRGB encode. 1031 * 1032 * Unfortunately, renderbuffer setup happens before a context is created. So 1033 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3 1034 * context (without an sRGB visual, though we don't have sRGB visuals exposed 1035 * yet), we go turn that back off before anyone finds out. 1036 */ 1037static void 1038intel_gles3_srgb_workaround(struct brw_context *brw, 1039 struct gl_framebuffer *fb) 1040{ 1041 struct gl_context *ctx = &brw->ctx; 1042 1043 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable) 1044 return; 1045 1046 /* Some day when we support the sRGB capable bit on visuals available for 1047 * GLES, we'll need to respect that and not disable things here. 1048 */ 1049 fb->Visual.sRGBCapable = false; 1050 for (int i = 0; i < BUFFER_COUNT; i++) { 1051 if (fb->Attachment[i].Renderbuffer && 1052 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) { 1053 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM; 1054 } 1055 } 1056} 1057 1058GLboolean 1059intelMakeCurrent(__DRIcontext * driContextPriv, 1060 __DRIdrawable * driDrawPriv, 1061 __DRIdrawable * driReadPriv) 1062{ 1063 struct brw_context *brw; 1064 GET_CURRENT_CONTEXT(curCtx); 1065 1066 if (driContextPriv) 1067 brw = (struct brw_context *) driContextPriv->driverPrivate; 1068 else 1069 brw = NULL; 1070 1071 /* According to the glXMakeCurrent() man page: "Pending commands to 1072 * the previous context, if any, are flushed before it is released." 1073 * But only flush if we're actually changing contexts. 1074 */ 1075 if (brw_context(curCtx) && brw_context(curCtx) != brw) { 1076 _mesa_flush(curCtx); 1077 } 1078 1079 if (driContextPriv) { 1080 struct gl_context *ctx = &brw->ctx; 1081 struct gl_framebuffer *fb, *readFb; 1082 1083 if (driDrawPriv == NULL) { 1084 fb = _mesa_get_incomplete_framebuffer(); 1085 } else { 1086 fb = driDrawPriv->driverPrivate; 1087 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1; 1088 } 1089 1090 if (driReadPriv == NULL) { 1091 readFb = _mesa_get_incomplete_framebuffer(); 1092 } else { 1093 readFb = driReadPriv->driverPrivate; 1094 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1; 1095 } 1096 1097 /* The sRGB workaround changes the renderbuffer's format. We must change 1098 * the format before the renderbuffer's miptree get's allocated, otherwise 1099 * the formats of the renderbuffer and its miptree will differ. 1100 */ 1101 intel_gles3_srgb_workaround(brw, fb); 1102 intel_gles3_srgb_workaround(brw, readFb); 1103 1104 /* If the context viewport hasn't been initialized, force a call out to 1105 * the loader to get buffers so we have a drawable size for the initial 1106 * viewport. */ 1107 if (!brw->ctx.ViewportInitialized) 1108 intel_prepare_render(brw); 1109 1110 _mesa_make_current(ctx, fb, readFb); 1111 } else { 1112 _mesa_make_current(NULL, NULL, NULL); 1113 } 1114 1115 return true; 1116} 1117 1118void 1119intel_resolve_for_dri2_flush(struct brw_context *brw, 1120 __DRIdrawable *drawable) 1121{ 1122 if (brw->gen < 6) { 1123 /* MSAA and fast color clear are not supported, so don't waste time 1124 * checking whether a resolve is needed. 1125 */ 1126 return; 1127 } 1128 1129 struct gl_framebuffer *fb = drawable->driverPrivate; 1130 struct intel_renderbuffer *rb; 1131 1132 /* Usually, only the back buffer will need to be downsampled. However, 1133 * the front buffer will also need it if the user has rendered into it. 1134 */ 1135 static const gl_buffer_index buffers[2] = { 1136 BUFFER_BACK_LEFT, 1137 BUFFER_FRONT_LEFT, 1138 }; 1139 1140 for (int i = 0; i < 2; ++i) { 1141 rb = intel_get_renderbuffer(fb, buffers[i]); 1142 if (rb == NULL || rb->mt == NULL) 1143 continue; 1144 if (rb->mt->num_samples <= 1) 1145 intel_miptree_resolve_color(brw, rb->mt); 1146 else 1147 intel_renderbuffer_downsample(brw, rb); 1148 } 1149} 1150 1151static unsigned 1152intel_bits_per_pixel(const struct intel_renderbuffer *rb) 1153{ 1154 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8; 1155} 1156 1157static void 1158intel_query_dri2_buffers(struct brw_context *brw, 1159 __DRIdrawable *drawable, 1160 __DRIbuffer **buffers, 1161 int *count); 1162 1163static void 1164intel_process_dri2_buffer(struct brw_context *brw, 1165 __DRIdrawable *drawable, 1166 __DRIbuffer *buffer, 1167 struct intel_renderbuffer *rb, 1168 const char *buffer_name); 1169 1170static void 1171intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable); 1172 1173static void 1174intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1175{ 1176 struct gl_framebuffer *fb = drawable->driverPrivate; 1177 struct intel_renderbuffer *rb; 1178 __DRIbuffer *buffers = NULL; 1179 int i, count; 1180 const char *region_name; 1181 1182 /* Set this up front, so that in case our buffers get invalidated 1183 * while we're getting new buffers, we don't clobber the stamp and 1184 * thus ignore the invalidate. */ 1185 drawable->lastStamp = drawable->dri2.stamp; 1186 1187 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1188 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1189 1190 intel_query_dri2_buffers(brw, drawable, &buffers, &count); 1191 1192 if (buffers == NULL) 1193 return; 1194 1195 for (i = 0; i < count; i++) { 1196 switch (buffers[i].attachment) { 1197 case __DRI_BUFFER_FRONT_LEFT: 1198 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1199 region_name = "dri2 front buffer"; 1200 break; 1201 1202 case __DRI_BUFFER_FAKE_FRONT_LEFT: 1203 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1204 region_name = "dri2 fake front buffer"; 1205 break; 1206 1207 case __DRI_BUFFER_BACK_LEFT: 1208 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1209 region_name = "dri2 back buffer"; 1210 break; 1211 1212 case __DRI_BUFFER_DEPTH: 1213 case __DRI_BUFFER_HIZ: 1214 case __DRI_BUFFER_DEPTH_STENCIL: 1215 case __DRI_BUFFER_STENCIL: 1216 case __DRI_BUFFER_ACCUM: 1217 default: 1218 fprintf(stderr, 1219 "unhandled buffer attach event, attachment type %d\n", 1220 buffers[i].attachment); 1221 return; 1222 } 1223 1224 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name); 1225 } 1226 1227} 1228 1229void 1230intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) 1231{ 1232 struct brw_context *brw = context->driverPrivate; 1233 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1234 1235 /* Set this up front, so that in case our buffers get invalidated 1236 * while we're getting new buffers, we don't clobber the stamp and 1237 * thus ignore the invalidate. */ 1238 drawable->lastStamp = drawable->dri2.stamp; 1239 1240 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) 1241 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); 1242 1243 if (screen->image.loader) 1244 intel_update_image_buffers(brw, drawable); 1245 else 1246 intel_update_dri2_buffers(brw, drawable); 1247 1248 driUpdateFramebufferSize(&brw->ctx, drawable); 1249} 1250 1251/** 1252 * intel_prepare_render should be called anywhere that curent read/drawbuffer 1253 * state is required. 1254 */ 1255void 1256intel_prepare_render(struct brw_context *brw) 1257{ 1258 struct gl_context *ctx = &brw->ctx; 1259 __DRIcontext *driContext = brw->driContext; 1260 __DRIdrawable *drawable; 1261 1262 drawable = driContext->driDrawablePriv; 1263 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) { 1264 if (drawable->lastStamp != drawable->dri2.stamp) 1265 intel_update_renderbuffers(driContext, drawable); 1266 driContext->dri2.draw_stamp = drawable->dri2.stamp; 1267 } 1268 1269 drawable = driContext->driReadablePriv; 1270 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) { 1271 if (drawable->lastStamp != drawable->dri2.stamp) 1272 intel_update_renderbuffers(driContext, drawable); 1273 driContext->dri2.read_stamp = drawable->dri2.stamp; 1274 } 1275 1276 /* If we're currently rendering to the front buffer, the rendering 1277 * that will happen next will probably dirty the front buffer. So 1278 * mark it as dirty here. 1279 */ 1280 if (brw_is_front_buffer_drawing(ctx->DrawBuffer)) 1281 brw->front_buffer_dirty = true; 1282} 1283 1284/** 1285 * \brief Query DRI2 to obtain a DRIdrawable's buffers. 1286 * 1287 * To determine which DRI buffers to request, examine the renderbuffers 1288 * attached to the drawable's framebuffer. Then request the buffers with 1289 * DRI2GetBuffers() or DRI2GetBuffersWithFormat(). 1290 * 1291 * This is called from intel_update_renderbuffers(). 1292 * 1293 * \param drawable Drawable whose buffers are queried. 1294 * \param buffers [out] List of buffers returned by DRI2 query. 1295 * \param buffer_count [out] Number of buffers returned. 1296 * 1297 * \see intel_update_renderbuffers() 1298 * \see DRI2GetBuffers() 1299 * \see DRI2GetBuffersWithFormat() 1300 */ 1301static void 1302intel_query_dri2_buffers(struct brw_context *brw, 1303 __DRIdrawable *drawable, 1304 __DRIbuffer **buffers, 1305 int *buffer_count) 1306{ 1307 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1308 struct gl_framebuffer *fb = drawable->driverPrivate; 1309 int i = 0; 1310 unsigned attachments[8]; 1311 1312 struct intel_renderbuffer *front_rb; 1313 struct intel_renderbuffer *back_rb; 1314 1315 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1316 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1317 1318 memset(attachments, 0, sizeof(attachments)); 1319 if ((brw_is_front_buffer_drawing(fb) || 1320 brw_is_front_buffer_reading(fb) || 1321 !back_rb) && front_rb) { 1322 /* If a fake front buffer is in use, then querying for 1323 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from 1324 * the real front buffer to the fake front buffer. So before doing the 1325 * query, we need to make sure all the pending drawing has landed in the 1326 * real front buffer. 1327 */ 1328 intel_batchbuffer_flush(brw); 1329 intel_flush_front(&brw->ctx); 1330 1331 attachments[i++] = __DRI_BUFFER_FRONT_LEFT; 1332 attachments[i++] = intel_bits_per_pixel(front_rb); 1333 } else if (front_rb && brw->front_buffer_dirty) { 1334 /* We have pending front buffer rendering, but we aren't querying for a 1335 * front buffer. If the front buffer we have is a fake front buffer, 1336 * the X server is going to throw it away when it processes the query. 1337 * So before doing the query, make sure all the pending drawing has 1338 * landed in the real front buffer. 1339 */ 1340 intel_batchbuffer_flush(brw); 1341 intel_flush_front(&brw->ctx); 1342 } 1343 1344 if (back_rb) { 1345 attachments[i++] = __DRI_BUFFER_BACK_LEFT; 1346 attachments[i++] = intel_bits_per_pixel(back_rb); 1347 } 1348 1349 assert(i <= ARRAY_SIZE(attachments)); 1350 1351 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable, 1352 &drawable->w, 1353 &drawable->h, 1354 attachments, i / 2, 1355 buffer_count, 1356 drawable->loaderPrivate); 1357} 1358 1359/** 1360 * \brief Assign a DRI buffer's DRM region to a renderbuffer. 1361 * 1362 * This is called from intel_update_renderbuffers(). 1363 * 1364 * \par Note: 1365 * DRI buffers whose attachment point is DRI2BufferStencil or 1366 * DRI2BufferDepthStencil are handled as special cases. 1367 * 1368 * \param buffer_name is a human readable name, such as "dri2 front buffer", 1369 * that is passed to drm_intel_bo_gem_create_from_name(). 1370 * 1371 * \see intel_update_renderbuffers() 1372 */ 1373static void 1374intel_process_dri2_buffer(struct brw_context *brw, 1375 __DRIdrawable *drawable, 1376 __DRIbuffer *buffer, 1377 struct intel_renderbuffer *rb, 1378 const char *buffer_name) 1379{ 1380 struct gl_framebuffer *fb = drawable->driverPrivate; 1381 drm_intel_bo *bo; 1382 1383 if (!rb) 1384 return; 1385 1386 unsigned num_samples = rb->Base.Base.NumSamples; 1387 1388 /* We try to avoid closing and reopening the same BO name, because the first 1389 * use of a mapping of the buffer involves a bunch of page faulting which is 1390 * moderately expensive. 1391 */ 1392 struct intel_mipmap_tree *last_mt; 1393 if (num_samples == 0) 1394 last_mt = rb->mt; 1395 else 1396 last_mt = rb->singlesample_mt; 1397 1398 uint32_t old_name = 0; 1399 if (last_mt) { 1400 /* The bo already has a name because the miptree was created by a 1401 * previous call to intel_process_dri2_buffer(). If a bo already has a 1402 * name, then drm_intel_bo_flink() is a low-cost getter. It does not 1403 * create a new name. 1404 */ 1405 drm_intel_bo_flink(last_mt->bo, &old_name); 1406 } 1407 1408 if (old_name == buffer->name) 1409 return; 1410 1411 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) { 1412 fprintf(stderr, 1413 "attaching buffer %d, at %d, cpp %d, pitch %d\n", 1414 buffer->name, buffer->attachment, 1415 buffer->cpp, buffer->pitch); 1416 } 1417 1418 intel_miptree_release(&rb->mt); 1419 bo = drm_intel_bo_gem_create_from_name(brw->bufmgr, buffer_name, 1420 buffer->name); 1421 if (!bo) { 1422 fprintf(stderr, 1423 "Failed to open BO for returned DRI2 buffer " 1424 "(%dx%d, %s, named %d).\n" 1425 "This is likely a bug in the X Server that will lead to a " 1426 "crash soon.\n", 1427 drawable->w, drawable->h, buffer_name, buffer->name); 1428 return; 1429 } 1430 1431 intel_update_winsys_renderbuffer_miptree(brw, rb, bo, 1432 drawable->w, drawable->h, 1433 buffer->pitch); 1434 1435 if (brw_is_front_buffer_drawing(fb) && 1436 (buffer->attachment == __DRI_BUFFER_FRONT_LEFT || 1437 buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) && 1438 rb->Base.Base.NumSamples > 1) { 1439 intel_renderbuffer_upsample(brw, rb); 1440 } 1441 1442 assert(rb->mt); 1443 1444 drm_intel_bo_unreference(bo); 1445} 1446 1447/** 1448 * \brief Query DRI image loader to obtain a DRIdrawable's buffers. 1449 * 1450 * To determine which DRI buffers to request, examine the renderbuffers 1451 * attached to the drawable's framebuffer. Then request the buffers from 1452 * the image loader 1453 * 1454 * This is called from intel_update_renderbuffers(). 1455 * 1456 * \param drawable Drawable whose buffers are queried. 1457 * \param buffers [out] List of buffers returned by DRI2 query. 1458 * \param buffer_count [out] Number of buffers returned. 1459 * 1460 * \see intel_update_renderbuffers() 1461 */ 1462 1463static void 1464intel_update_image_buffer(struct brw_context *intel, 1465 __DRIdrawable *drawable, 1466 struct intel_renderbuffer *rb, 1467 __DRIimage *buffer, 1468 enum __DRIimageBufferMask buffer_type) 1469{ 1470 struct gl_framebuffer *fb = drawable->driverPrivate; 1471 1472 if (!rb || !buffer->bo) 1473 return; 1474 1475 unsigned num_samples = rb->Base.Base.NumSamples; 1476 1477 /* Check and see if we're already bound to the right 1478 * buffer object 1479 */ 1480 struct intel_mipmap_tree *last_mt; 1481 if (num_samples == 0) 1482 last_mt = rb->mt; 1483 else 1484 last_mt = rb->singlesample_mt; 1485 1486 if (last_mt && last_mt->bo == buffer->bo) 1487 return; 1488 1489 intel_update_winsys_renderbuffer_miptree(intel, rb, buffer->bo, 1490 buffer->width, buffer->height, 1491 buffer->pitch); 1492 1493 if (brw_is_front_buffer_drawing(fb) && 1494 buffer_type == __DRI_IMAGE_BUFFER_FRONT && 1495 rb->Base.Base.NumSamples > 1) { 1496 intel_renderbuffer_upsample(intel, rb); 1497 } 1498} 1499 1500static void 1501intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable) 1502{ 1503 struct gl_framebuffer *fb = drawable->driverPrivate; 1504 __DRIscreen *screen = brw->intelScreen->driScrnPriv; 1505 struct intel_renderbuffer *front_rb; 1506 struct intel_renderbuffer *back_rb; 1507 struct __DRIimageList images; 1508 unsigned int format; 1509 uint32_t buffer_mask = 0; 1510 1511 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT); 1512 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT); 1513 1514 if (back_rb) 1515 format = intel_rb_format(back_rb); 1516 else if (front_rb) 1517 format = intel_rb_format(front_rb); 1518 else 1519 return; 1520 1521 if (front_rb && (brw_is_front_buffer_drawing(fb) || 1522 brw_is_front_buffer_reading(fb) || !back_rb)) { 1523 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT; 1524 } 1525 1526 if (back_rb) 1527 buffer_mask |= __DRI_IMAGE_BUFFER_BACK; 1528 1529 (*screen->image.loader->getBuffers) (drawable, 1530 driGLFormatToImageFormat(format), 1531 &drawable->dri2.stamp, 1532 drawable->loaderPrivate, 1533 buffer_mask, 1534 &images); 1535 1536 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) { 1537 drawable->w = images.front->width; 1538 drawable->h = images.front->height; 1539 intel_update_image_buffer(brw, 1540 drawable, 1541 front_rb, 1542 images.front, 1543 __DRI_IMAGE_BUFFER_FRONT); 1544 } 1545 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) { 1546 drawable->w = images.back->width; 1547 drawable->h = images.back->height; 1548 intel_update_image_buffer(brw, 1549 drawable, 1550 back_rb, 1551 images.back, 1552 __DRI_IMAGE_BUFFER_BACK); 1553 } 1554} 1555