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.
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:
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.
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.
27 **********************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
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 #include "main/framebuffer.h"
46 #include "vbo/vbo_context.h"
48 #include "drivers/common/driverfuncs.h"
49 #include "drivers/common/meta.h"
52 #include "brw_context.h"
53 #include "brw_defines.h"
54 #include "brw_compiler.h"
56 #include "brw_state.h"
58 #include "intel_batchbuffer.h"
59 #include "intel_buffer_objects.h"
60 #include "intel_buffers.h"
61 #include "intel_fbo.h"
62 #include "intel_mipmap_tree.h"
63 #include "intel_pixel.h"
64 #include "intel_image.h"
65 #include "intel_tex.h"
66 #include "intel_tex_obj.h"
68 #include "swrast_setup/swrast_setup.h"
70 #include "tnl/t_pipeline.h"
71 #include "util/ralloc.h"
72 #include "util/debug.h"
74 /***************************************
75 * Mesa's Driver Functions
76 ***************************************/
79 brw_query_samples_for_format(struct gl_context *ctx, GLenum target,
80 GLenum internalFormat, int samples[16])
82 struct brw_context *brw = brw_context(ctx);
110 assert(brw->gen < 6);
116 const char *const brw_vendor_string = "Intel Open Source Technology Center";
119 brw_get_renderer_string(unsigned deviceID)
122 static char buffer[128];
126 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
127 #include "pci_ids/i965_pci_ids.h"
129 chipset = "Unknown Intel Chipset";
133 (void) driGetRendererString(buffer, chipset, 0);
137 static const GLubyte *
138 intel_get_string(struct gl_context * ctx, GLenum name)
140 const struct brw_context *const brw = brw_context(ctx);
144 return (GLubyte *) brw_vendor_string;
148 (GLubyte *) brw_get_renderer_string(brw->intelScreen->deviceID);
156 intel_viewport(struct gl_context *ctx)
158 struct brw_context *brw = brw_context(ctx);
159 __DRIcontext *driContext = brw->driContext;
161 if (_mesa_is_winsys_fbo(ctx->DrawBuffer)) {
162 dri2InvalidateDrawable(driContext->driDrawablePriv);
163 dri2InvalidateDrawable(driContext->driReadablePriv);
168 intel_update_state(struct gl_context * ctx, GLuint new_state)
170 struct brw_context *brw = brw_context(ctx);
171 struct intel_texture_object *tex_obj;
172 struct intel_renderbuffer *depth_irb;
174 if (ctx->swrast_context)
175 _swrast_InvalidateState(ctx, new_state);
176 _vbo_InvalidateState(ctx, new_state);
178 brw->NewGLState |= new_state;
180 _mesa_unlock_context_textures(ctx);
182 /* Resolve the depth buffer's HiZ buffer. */
183 depth_irb = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
185 intel_renderbuffer_resolve_hiz(brw, depth_irb);
187 /* Resolve depth buffer and render cache of each enabled texture. */
188 int maxEnabledUnit = ctx->Texture._MaxEnabledTexImageUnit;
189 for (int i = 0; i <= maxEnabledUnit; i++) {
190 if (!ctx->Texture.Unit[i]._Current)
192 tex_obj = intel_texture_object(ctx->Texture.Unit[i]._Current);
193 if (!tex_obj || !tex_obj->mt)
195 intel_miptree_all_slices_resolve_depth(brw, tex_obj->mt);
196 intel_miptree_resolve_color(brw, tex_obj->mt);
197 brw_render_cache_set_check_flush(brw, tex_obj->mt->bo);
200 /* Resolve color for each active shader image. */
201 for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
202 const struct gl_shader *shader = ctx->_Shader->CurrentProgram[i] ?
203 ctx->_Shader->CurrentProgram[i]->_LinkedShaders[i] : NULL;
205 if (unlikely(shader && shader->NumImages)) {
206 for (unsigned j = 0; j < shader->NumImages; j++) {
207 struct gl_image_unit *u = &ctx->ImageUnits[shader->ImageUnits[j]];
208 tex_obj = intel_texture_object(u->TexObj);
210 if (tex_obj && tex_obj->mt) {
211 intel_miptree_resolve_color(brw, tex_obj->mt);
212 brw_render_cache_set_check_flush(brw, tex_obj->mt->bo);
218 /* If FRAMEBUFFER_SRGB is used on Gen9+ then we need to resolve any of the
219 * single-sampled color renderbuffers because the CCS buffer isn't
220 * supported for SRGB formats. This only matters if FRAMEBUFFER_SRGB is
221 * enabled because otherwise the surface state will be programmed with the
222 * linear equivalent format anyway.
224 if (brw->gen >= 9 && ctx->Color.sRGBEnabled) {
225 struct gl_framebuffer *fb = ctx->DrawBuffer;
226 for (int i = 0; i < fb->_NumColorDrawBuffers; i++) {
227 struct gl_renderbuffer *rb = fb->_ColorDrawBuffers[i];
232 struct intel_renderbuffer *irb = intel_renderbuffer(rb);
233 struct intel_mipmap_tree *mt = irb->mt;
236 mt->num_samples > 1 ||
237 _mesa_get_srgb_format_linear(mt->format) == mt->format)
240 intel_miptree_resolve_color(brw, mt);
241 brw_render_cache_set_check_flush(brw, mt->bo);
245 _mesa_lock_context_textures(ctx);
248 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
251 intel_flush_front(struct gl_context *ctx)
253 struct brw_context *brw = brw_context(ctx);
254 __DRIcontext *driContext = brw->driContext;
255 __DRIdrawable *driDrawable = driContext->driDrawablePriv;
256 __DRIscreen *const screen = brw->intelScreen->driScrnPriv;
258 if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) {
259 if (flushFront(screen) && driDrawable &&
260 driDrawable->loaderPrivate) {
262 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
264 * This potentially resolves both front and back buffer. It
265 * is unnecessary to resolve the back, but harms nothing except
266 * performance. And no one cares about front-buffer render
269 intel_resolve_for_dri2_flush(brw, driDrawable);
270 intel_batchbuffer_flush(brw);
272 flushFront(screen)(driDrawable, driDrawable->loaderPrivate);
274 /* We set the dirty bit in intel_prepare_render() if we're
275 * front buffer rendering once we get there.
277 brw->front_buffer_dirty = false;
283 intel_glFlush(struct gl_context *ctx)
285 struct brw_context *brw = brw_context(ctx);
287 intel_batchbuffer_flush(brw);
288 intel_flush_front(ctx);
290 brw->need_flush_throttle = true;
294 intel_finish(struct gl_context * ctx)
296 struct brw_context *brw = brw_context(ctx);
300 if (brw->batch.last_bo)
301 drm_intel_bo_wait_rendering(brw->batch.last_bo);
305 brw_init_driver_functions(struct brw_context *brw,
306 struct dd_function_table *functions)
308 _mesa_init_driver_functions(functions);
310 /* GLX uses DRI2 invalidate events to handle window resizing.
311 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
312 * which doesn't provide a mechanism for snooping the event queues.
314 * So EGL still relies on viewport hacks to handle window resizing.
315 * This should go away with DRI3000.
317 if (!brw->driContext->driScreenPriv->dri2.useInvalidate)
318 functions->Viewport = intel_viewport;
320 functions->Flush = intel_glFlush;
321 functions->Finish = intel_finish;
322 functions->GetString = intel_get_string;
323 functions->UpdateState = intel_update_state;
325 intelInitTextureFuncs(functions);
326 intelInitTextureImageFuncs(functions);
327 intelInitTextureSubImageFuncs(functions);
328 intelInitTextureCopyImageFuncs(functions);
329 intelInitCopyImageFuncs(functions);
330 intelInitClearFuncs(functions);
331 intelInitBufferFuncs(functions);
332 intelInitPixelFuncs(functions);
333 intelInitBufferObjectFuncs(functions);
334 intel_init_syncobj_functions(functions);
335 brw_init_object_purgeable_functions(functions);
337 brwInitFragProgFuncs( functions );
338 brw_init_common_queryobj_functions(functions);
340 gen6_init_queryobj_functions(functions);
342 gen4_init_queryobj_functions(functions);
343 brw_init_compute_functions(functions);
345 brw_init_conditional_render_functions(functions);
347 functions->QuerySamplesForFormat = brw_query_samples_for_format;
349 functions->NewTransformFeedback = brw_new_transform_feedback;
350 functions->DeleteTransformFeedback = brw_delete_transform_feedback;
351 functions->GetTransformFeedbackVertexCount =
352 brw_get_transform_feedback_vertex_count;
354 functions->BeginTransformFeedback = gen7_begin_transform_feedback;
355 functions->EndTransformFeedback = gen7_end_transform_feedback;
356 functions->PauseTransformFeedback = gen7_pause_transform_feedback;
357 functions->ResumeTransformFeedback = gen7_resume_transform_feedback;
359 functions->BeginTransformFeedback = brw_begin_transform_feedback;
360 functions->EndTransformFeedback = brw_end_transform_feedback;
364 functions->GetSamplePosition = gen6_get_sample_position;
368 brw_initialize_context_constants(struct brw_context *brw)
370 struct gl_context *ctx = &brw->ctx;
371 const struct brw_compiler *compiler = brw->intelScreen->compiler;
373 const bool stage_exists[MESA_SHADER_STAGES] = {
374 [MESA_SHADER_VERTEX] = true,
375 [MESA_SHADER_TESS_CTRL] = brw->gen >= 8,
376 [MESA_SHADER_TESS_EVAL] = brw->gen >= 8,
377 [MESA_SHADER_GEOMETRY] = brw->gen >= 6,
378 [MESA_SHADER_FRAGMENT] = true,
379 [MESA_SHADER_COMPUTE] =
380 (ctx->Const.MaxComputeWorkGroupSize[0] >= 1024) ||
381 _mesa_extension_override_enables.ARB_compute_shader,
384 unsigned num_stages = 0;
385 for (int i = 0; i < MESA_SHADER_STAGES; i++) {
390 unsigned max_samplers =
391 brw->gen >= 8 || brw->is_haswell ? BRW_MAX_TEX_UNIT : 16;
393 ctx->Const.MaxDualSourceDrawBuffers = 1;
394 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS;
395 ctx->Const.MaxCombinedShaderOutputResources =
396 MAX_IMAGE_UNITS + BRW_MAX_DRAW_BUFFERS;
398 ctx->Const.QueryCounterBits.Timestamp = 36;
400 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */
401 ctx->Const.MaxImageUnits = MAX_IMAGE_UNITS;
402 ctx->Const.MaxRenderbufferSize = 8192;
403 ctx->Const.MaxTextureLevels = MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS);
404 ctx->Const.Max3DTextureLevels = 12; /* 2048 */
405 ctx->Const.MaxCubeTextureLevels = 14; /* 8192 */
406 ctx->Const.MaxArrayTextureLayers = brw->gen >= 7 ? 2048 : 512;
407 ctx->Const.MaxTextureMbytes = 1536;
408 ctx->Const.MaxTextureRectSize = 1 << 12;
409 ctx->Const.MaxTextureMaxAnisotropy = 16.0;
410 ctx->Const.StripTextureBorder = true;
412 ctx->Const.MaxProgramTextureGatherComponents = 4;
413 else if (brw->gen == 6)
414 ctx->Const.MaxProgramTextureGatherComponents = 1;
416 ctx->Const.MaxUniformBlockSize = 65536;
418 for (int i = 0; i < MESA_SHADER_STAGES; i++) {
419 struct gl_program_constants *prog = &ctx->Const.Program[i];
421 if (!stage_exists[i])
424 prog->MaxTextureImageUnits = max_samplers;
426 prog->MaxUniformBlocks = BRW_MAX_UBO;
427 prog->MaxCombinedUniformComponents =
428 prog->MaxUniformComponents +
429 ctx->Const.MaxUniformBlockSize / 4 * prog->MaxUniformBlocks;
431 prog->MaxAtomicCounters = MAX_ATOMIC_COUNTERS;
432 prog->MaxAtomicBuffers = BRW_MAX_ABO;
433 prog->MaxImageUniforms = compiler->scalar_stage[i] ? BRW_MAX_IMAGES : 0;
434 prog->MaxShaderStorageBlocks = BRW_MAX_SSBO;
437 ctx->Const.MaxTextureUnits =
438 MIN2(ctx->Const.MaxTextureCoordUnits,
439 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits);
441 ctx->Const.MaxUniformBufferBindings = num_stages * BRW_MAX_UBO;
442 ctx->Const.MaxCombinedUniformBlocks = num_stages * BRW_MAX_UBO;
443 ctx->Const.MaxCombinedAtomicBuffers = num_stages * BRW_MAX_ABO;
444 ctx->Const.MaxCombinedShaderStorageBlocks = num_stages * BRW_MAX_SSBO;
445 ctx->Const.MaxShaderStorageBufferBindings = num_stages * BRW_MAX_SSBO;
446 ctx->Const.MaxCombinedTextureImageUnits = num_stages * max_samplers;
447 ctx->Const.MaxCombinedImageUniforms = num_stages * BRW_MAX_IMAGES;
450 /* Hardware only supports a limited number of transform feedback buffers.
451 * So we need to override the Mesa default (which is based only on software
454 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS;
456 /* On Gen6, in the worst case, we use up one binding table entry per
457 * transform feedback component (see comments above the definition of
458 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
459 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
460 * BRW_MAX_SOL_BINDINGS.
462 * In "separate components" mode, we need to divide this value by
463 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
464 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
466 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS;
467 ctx->Const.MaxTransformFeedbackSeparateComponents =
468 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS;
470 ctx->Const.AlwaysUseGetTransformFeedbackVertexCount = true;
473 const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen);
474 const int clamp_max_samples =
475 driQueryOptioni(&brw->optionCache, "clamp_max_samples");
477 if (clamp_max_samples < 0) {
478 max_samples = msaa_modes[0];
480 /* Select the largest supported MSAA mode that does not exceed
484 for (int i = 0; msaa_modes[i] != 0; ++i) {
485 if (msaa_modes[i] <= clamp_max_samples) {
486 max_samples = msaa_modes[i];
492 ctx->Const.MaxSamples = max_samples;
493 ctx->Const.MaxColorTextureSamples = max_samples;
494 ctx->Const.MaxDepthTextureSamples = max_samples;
495 ctx->Const.MaxIntegerSamples = max_samples;
496 ctx->Const.MaxImageSamples = 0;
498 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
499 * to map indices of rectangular grid to sample numbers within a pixel.
500 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
501 * extension implementation. For more details see the comment above
502 * gen6_set_sample_maps() definition.
504 gen6_set_sample_maps(ctx);
506 ctx->Const.MinLineWidth = 1.0;
507 ctx->Const.MinLineWidthAA = 1.0;
509 ctx->Const.MaxLineWidth = 7.375;
510 ctx->Const.MaxLineWidthAA = 7.375;
511 ctx->Const.LineWidthGranularity = 0.125;
513 ctx->Const.MaxLineWidth = 7.0;
514 ctx->Const.MaxLineWidthAA = 7.0;
515 ctx->Const.LineWidthGranularity = 0.5;
518 /* For non-antialiased lines, we have to round the line width to the
519 * nearest whole number. Make sure that we don't advertise a line
520 * width that, when rounded, will be beyond the actual hardware
523 assert(roundf(ctx->Const.MaxLineWidth) <= ctx->Const.MaxLineWidth);
525 ctx->Const.MinPointSize = 1.0;
526 ctx->Const.MinPointSizeAA = 1.0;
527 ctx->Const.MaxPointSize = 255.0;
528 ctx->Const.MaxPointSizeAA = 255.0;
529 ctx->Const.PointSizeGranularity = 1.0;
531 if (brw->gen >= 5 || brw->is_g4x)
532 ctx->Const.MaxClipPlanes = 8;
534 ctx->Const.LowerTessLevel = true;
536 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeInstructions = 16 * 1024;
537 ctx->Const.Program[MESA_SHADER_VERTEX].MaxAluInstructions = 0;
538 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexInstructions = 0;
539 ctx->Const.Program[MESA_SHADER_VERTEX].MaxTexIndirections = 0;
540 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAluInstructions = 0;
541 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexInstructions = 0;
542 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTexIndirections = 0;
543 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAttribs = 16;
544 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeTemps = 256;
545 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeAddressRegs = 1;
546 ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters = 1024;
547 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams =
548 MIN2(ctx->Const.Program[MESA_SHADER_VERTEX].MaxNativeParameters,
549 ctx->Const.Program[MESA_SHADER_VERTEX].MaxEnvParams);
551 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeInstructions = 1024;
552 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAluInstructions = 1024;
553 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexInstructions = 1024;
554 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTexIndirections = 1024;
555 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAttribs = 12;
556 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeTemps = 256;
557 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeAddressRegs = 0;
558 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters = 1024;
559 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams =
560 MIN2(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxNativeParameters,
561 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxEnvParams);
563 /* Fragment shaders use real, 32-bit twos-complement integers for all
566 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMin = 31;
567 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.RangeMax = 30;
568 ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt.Precision = 0;
569 ctx->Const.Program[MESA_SHADER_FRAGMENT].HighInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt;
570 ctx->Const.Program[MESA_SHADER_FRAGMENT].MediumInt = ctx->Const.Program[MESA_SHADER_FRAGMENT].LowInt;
572 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMin = 31;
573 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.RangeMax = 30;
574 ctx->Const.Program[MESA_SHADER_VERTEX].LowInt.Precision = 0;
575 ctx->Const.Program[MESA_SHADER_VERTEX].HighInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt;
576 ctx->Const.Program[MESA_SHADER_VERTEX].MediumInt = ctx->Const.Program[MESA_SHADER_VERTEX].LowInt;
578 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
579 * but we're not sure how it's actually done for vertex order,
580 * that affect provoking vertex decision. Always use last vertex
581 * convention for quad primitive which works as expected for now.
584 ctx->Const.QuadsFollowProvokingVertexConvention = false;
586 ctx->Const.NativeIntegers = true;
587 ctx->Const.VertexID_is_zero_based = true;
589 /* Regarding the CMP instruction, the Ivybridge PRM says:
591 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
592 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
593 * 0xFFFFFFFF) is assigned to dst."
595 * but PRMs for earlier generations say
597 * "In dword format, one GRF may store up to 8 results. When the register
598 * is used later as a vector of Booleans, as only LSB at each channel
599 * contains meaning [sic] data, software should make sure all higher bits
600 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
602 * We select the representation of a true boolean uniform to be ~0, and fix
603 * the results of Gen <= 5 CMP instruction's with -(result & 1).
605 ctx->Const.UniformBooleanTrue = ~0;
607 /* From the gen4 PRM, volume 4 page 127:
609 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
610 * the base address of the first element of the surface, computed in
611 * software by adding the surface base address to the byte offset of
612 * the element in the buffer."
614 * However, unaligned accesses are slower, so enforce buffer alignment.
616 ctx->Const.UniformBufferOffsetAlignment = 16;
618 /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so
619 * that we can safely have the CPU and GPU writing the same SSBO on
620 * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never
621 * writes, so there's no problem. For an SSBO, the GPU and the CPU can
622 * be updating disjoint regions of the buffer simultaneously and that will
623 * break if the regions overlap the same cacheline.
625 ctx->Const.ShaderStorageBufferOffsetAlignment = 64;
626 ctx->Const.TextureBufferOffsetAlignment = 16;
627 ctx->Const.MaxTextureBufferSize = 128 * 1024 * 1024;
630 ctx->Const.MaxVarying = 32;
631 ctx->Const.Program[MESA_SHADER_VERTEX].MaxOutputComponents = 128;
632 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxInputComponents = 64;
633 ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxOutputComponents = 128;
634 ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxInputComponents = 128;
635 ctx->Const.Program[MESA_SHADER_TESS_CTRL].MaxInputComponents = 128;
636 ctx->Const.Program[MESA_SHADER_TESS_CTRL].MaxOutputComponents = 128;
637 ctx->Const.Program[MESA_SHADER_TESS_EVAL].MaxInputComponents = 128;
638 ctx->Const.Program[MESA_SHADER_TESS_EVAL].MaxOutputComponents = 128;
641 /* We want the GLSL compiler to emit code that uses condition codes */
642 for (int i = 0; i < MESA_SHADER_STAGES; i++) {
643 ctx->Const.ShaderCompilerOptions[i] =
644 brw->intelScreen->compiler->glsl_compiler_options[i];
647 /* ARB_viewport_array */
648 if (brw->gen >= 6 && ctx->API == API_OPENGL_CORE) {
649 ctx->Const.MaxViewports = GEN6_NUM_VIEWPORTS;
650 ctx->Const.ViewportSubpixelBits = 0;
652 /* Cast to float before negating because MaxViewportWidth is unsigned.
654 ctx->Const.ViewportBounds.Min = -(float)ctx->Const.MaxViewportWidth;
655 ctx->Const.ViewportBounds.Max = ctx->Const.MaxViewportWidth;
658 /* ARB_gpu_shader5 */
660 ctx->Const.MaxVertexStreams = MIN2(4, MAX_VERTEX_STREAMS);
662 /* ARB_framebuffer_no_attachments */
663 ctx->Const.MaxFramebufferWidth = ctx->Const.MaxViewportWidth;
664 ctx->Const.MaxFramebufferHeight = ctx->Const.MaxViewportHeight;
665 ctx->Const.MaxFramebufferLayers = ctx->Const.MaxArrayTextureLayers;
666 ctx->Const.MaxFramebufferSamples = max_samples;
670 brw_initialize_cs_context_constants(struct brw_context *brw, unsigned max_threads)
672 struct gl_context *ctx = &brw->ctx;
674 /* For ES, we set these constants based on SIMD8.
676 * TODO: Once we can always generate SIMD16, we should update this.
678 * For GL, we assume we can generate a SIMD16 program, but this currently
679 * is not always true. This allows us to run more test cases, and will be
680 * required based on desktop GL compute shader requirements.
682 const int simd_size = ctx->API == API_OPENGL_CORE ? 16 : 8;
684 const uint32_t max_invocations = simd_size * max_threads;
685 ctx->Const.MaxComputeWorkGroupSize[0] = max_invocations;
686 ctx->Const.MaxComputeWorkGroupSize[1] = max_invocations;
687 ctx->Const.MaxComputeWorkGroupSize[2] = max_invocations;
688 ctx->Const.MaxComputeWorkGroupInvocations = max_invocations;
692 * Process driconf (drirc) options, setting appropriate context flags.
694 * intelInitExtensions still pokes at optionCache directly, in order to
695 * avoid advertising various extensions. No flags are set, so it makes
696 * sense to continue doing that there.
699 brw_process_driconf_options(struct brw_context *brw)
701 struct gl_context *ctx = &brw->ctx;
703 driOptionCache *options = &brw->optionCache;
704 driParseConfigFiles(options, &brw->intelScreen->optionCache,
705 brw->driContext->driScreenPriv->myNum, "i965");
707 int bo_reuse_mode = driQueryOptioni(options, "bo_reuse");
708 switch (bo_reuse_mode) {
709 case DRI_CONF_BO_REUSE_DISABLED:
711 case DRI_CONF_BO_REUSE_ALL:
712 intel_bufmgr_gem_enable_reuse(brw->bufmgr);
716 if (!driQueryOptionb(options, "hiz")) {
717 brw->has_hiz = false;
718 /* On gen6, you can only do separate stencil with HIZ. */
720 brw->has_separate_stencil = false;
723 if (driQueryOptionb(options, "always_flush_batch")) {
724 fprintf(stderr, "flushing batchbuffer before/after each draw call\n");
725 brw->always_flush_batch = true;
728 if (driQueryOptionb(options, "always_flush_cache")) {
729 fprintf(stderr, "flushing GPU caches before/after each draw call\n");
730 brw->always_flush_cache = true;
733 if (driQueryOptionb(options, "disable_throttling")) {
734 fprintf(stderr, "disabling flush throttling\n");
735 brw->disable_throttling = true;
738 brw->precompile = driQueryOptionb(&brw->optionCache, "shader_precompile");
740 ctx->Const.ForceGLSLExtensionsWarn =
741 driQueryOptionb(options, "force_glsl_extensions_warn");
743 ctx->Const.DisableGLSLLineContinuations =
744 driQueryOptionb(options, "disable_glsl_line_continuations");
746 ctx->Const.AllowGLSLExtensionDirectiveMidShader =
747 driQueryOptionb(options, "allow_glsl_extension_directive_midshader");
751 brwCreateContext(gl_api api,
752 const struct gl_config *mesaVis,
753 __DRIcontext *driContextPriv,
754 unsigned major_version,
755 unsigned minor_version,
758 unsigned *dri_ctx_error,
759 void *sharedContextPrivate)
761 __DRIscreen *sPriv = driContextPriv->driScreenPriv;
762 struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate;
763 struct intel_screen *screen = sPriv->driverPrivate;
764 const struct brw_device_info *devinfo = screen->devinfo;
765 struct dd_function_table functions;
767 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
768 * provides us with context reset notifications.
770 uint32_t allowed_flags = __DRI_CTX_FLAG_DEBUG
771 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE;
773 if (screen->has_context_reset_notification)
774 allowed_flags |= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS;
776 if (flags & ~allowed_flags) {
777 *dri_ctx_error = __DRI_CTX_ERROR_UNKNOWN_FLAG;
781 struct brw_context *brw = rzalloc(NULL, struct brw_context);
783 fprintf(stderr, "%s: failed to alloc context\n", __func__);
784 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
788 driContextPriv->driverPrivate = brw;
789 brw->driContext = driContextPriv;
790 brw->intelScreen = screen;
791 brw->bufmgr = screen->bufmgr;
793 brw->gen = devinfo->gen;
794 brw->gt = devinfo->gt;
795 brw->is_g4x = devinfo->is_g4x;
796 brw->is_baytrail = devinfo->is_baytrail;
797 brw->is_haswell = devinfo->is_haswell;
798 brw->is_cherryview = devinfo->is_cherryview;
799 brw->is_broxton = devinfo->is_broxton;
800 brw->has_llc = devinfo->has_llc;
801 brw->has_hiz = devinfo->has_hiz_and_separate_stencil;
802 brw->has_separate_stencil = devinfo->has_hiz_and_separate_stencil;
803 brw->has_pln = devinfo->has_pln;
804 brw->has_compr4 = devinfo->has_compr4;
805 brw->has_surface_tile_offset = devinfo->has_surface_tile_offset;
806 brw->has_negative_rhw_bug = devinfo->has_negative_rhw_bug;
807 brw->needs_unlit_centroid_workaround =
808 devinfo->needs_unlit_centroid_workaround;
810 brw->must_use_separate_stencil = screen->hw_must_use_separate_stencil;
811 brw->has_swizzling = screen->hw_has_swizzling;
813 brw->vs.base.stage = MESA_SHADER_VERTEX;
814 brw->tcs.base.stage = MESA_SHADER_TESS_CTRL;
815 brw->tes.base.stage = MESA_SHADER_TESS_EVAL;
816 brw->gs.base.stage = MESA_SHADER_GEOMETRY;
817 brw->wm.base.stage = MESA_SHADER_FRAGMENT;
819 gen8_init_vtable_surface_functions(brw);
820 brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz;
821 } else if (brw->gen >= 7) {
822 gen7_init_vtable_surface_functions(brw);
823 brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz;
824 } else if (brw->gen >= 6) {
825 gen6_init_vtable_surface_functions(brw);
826 brw->vtbl.emit_depth_stencil_hiz = gen6_emit_depth_stencil_hiz;
828 gen4_init_vtable_surface_functions(brw);
829 brw->vtbl.emit_depth_stencil_hiz = brw_emit_depth_stencil_hiz;
832 brw_init_driver_functions(brw, &functions);
835 functions.GetGraphicsResetStatus = brw_get_graphics_reset_status;
837 struct gl_context *ctx = &brw->ctx;
839 if (!_mesa_initialize_context(ctx, api, mesaVis, shareCtx, &functions)) {
840 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
841 fprintf(stderr, "%s: failed to init mesa context\n", __func__);
842 intelDestroyContext(driContextPriv);
846 driContextSetFlags(ctx, flags);
848 /* Initialize the software rasterizer and helper modules.
850 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
851 * software fallbacks (which we have to support on legacy GL to do weird
852 * glDrawPixels(), glBitmap(), and other functions).
854 if (api != API_OPENGL_CORE && api != API_OPENGLES2) {
855 _swrast_CreateContext(ctx);
858 _vbo_CreateContext(ctx);
859 if (ctx->swrast_context) {
860 _tnl_CreateContext(ctx);
861 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
862 _swsetup_CreateContext(ctx);
864 /* Configure swrast to match hardware characteristics: */
865 _swrast_allow_pixel_fog(ctx, false);
866 _swrast_allow_vertex_fog(ctx, true);
869 _mesa_meta_init(ctx);
871 brw_process_driconf_options(brw);
873 if (INTEL_DEBUG & DEBUG_PERF)
874 brw->perf_debug = true;
876 brw_initialize_cs_context_constants(brw, devinfo->max_cs_threads);
877 brw_initialize_context_constants(brw);
879 ctx->Const.ResetStrategy = notify_reset
880 ? GL_LOSE_CONTEXT_ON_RESET_ARB : GL_NO_RESET_NOTIFICATION_ARB;
882 /* Reinitialize the context point state. It depends on ctx->Const values. */
883 _mesa_init_point(ctx);
887 intel_batchbuffer_init(brw);
890 /* Create a new hardware context. Using a hardware context means that
891 * our GPU state will be saved/restored on context switch, allowing us
892 * to assume that the GPU is in the same state we left it in.
894 * This is required for transform feedback buffer offsets, query objects,
895 * and also allows us to reduce how much state we have to emit.
897 brw->hw_ctx = drm_intel_gem_context_create(brw->bufmgr);
900 fprintf(stderr, "Gen6+ requires Kernel 3.6 or later.\n");
901 intelDestroyContext(driContextPriv);
906 if (brw_init_pipe_control(brw, devinfo)) {
907 *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
908 intelDestroyContext(driContextPriv);
914 intelInitExtensions(ctx);
916 brw_init_surface_formats(brw);
918 brw->max_vs_threads = devinfo->max_vs_threads;
919 brw->max_hs_threads = devinfo->max_hs_threads;
920 brw->max_ds_threads = devinfo->max_ds_threads;
921 brw->max_gs_threads = devinfo->max_gs_threads;
922 brw->max_wm_threads = devinfo->max_wm_threads;
923 brw->max_cs_threads = devinfo->max_cs_threads;
924 brw->urb.size = devinfo->urb.size;
925 brw->urb.min_vs_entries = devinfo->urb.min_vs_entries;
926 brw->urb.max_vs_entries = devinfo->urb.max_vs_entries;
927 brw->urb.max_hs_entries = devinfo->urb.max_hs_entries;
928 brw->urb.max_ds_entries = devinfo->urb.max_ds_entries;
929 brw->urb.max_gs_entries = devinfo->urb.max_gs_entries;
931 /* Estimate the size of the mappable aperture into the GTT. There's an
932 * ioctl to get the whole GTT size, but not one to get the mappable subset.
933 * It turns out it's basically always 256MB, though some ancient hardware
936 uint32_t gtt_size = 256 * 1024 * 1024;
938 /* We don't want to map two objects such that a memcpy between them would
939 * just fault one mapping in and then the other over and over forever. So
940 * we would need to divide the GTT size by 2. Additionally, some GTT is
941 * taken up by things like the framebuffer and the ringbuffer and such, so
942 * be more conservative.
944 brw->max_gtt_map_object_size = gtt_size / 4;
947 brw->urb.gs_present = false;
949 brw->prim_restart.in_progress = false;
950 brw->prim_restart.enable_cut_index = false;
951 brw->gs.enabled = false;
952 brw->sf.viewport_transform_enable = true;
954 brw->predicate.state = BRW_PREDICATE_STATE_RENDER;
956 brw->use_resource_streamer = screen->has_resource_streamer &&
957 (env_var_as_boolean("INTEL_USE_HW_BT", false) ||
958 env_var_as_boolean("INTEL_USE_GATHER", false));
960 ctx->VertexProgram._MaintainTnlProgram = true;
961 ctx->FragmentProgram._MaintainTexEnvProgram = true;
963 brw_draw_init( brw );
965 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0) {
966 /* Turn on some extra GL_ARB_debug_output generation. */
967 brw->perf_debug = true;
970 if ((flags & __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS) != 0)
971 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB;
973 if (INTEL_DEBUG & DEBUG_SHADER_TIME)
974 brw_init_shader_time(brw);
976 _mesa_compute_version(ctx);
978 _mesa_initialize_dispatch_tables(ctx);
979 _mesa_initialize_vbo_vtxfmt(ctx);
981 if (ctx->Extensions.AMD_performance_monitor) {
982 brw_init_performance_monitors(brw);
985 vbo_use_buffer_objects(ctx);
986 vbo_always_unmap_buffers(ctx);
992 intelDestroyContext(__DRIcontext * driContextPriv)
994 struct brw_context *brw =
995 (struct brw_context *) driContextPriv->driverPrivate;
996 struct gl_context *ctx = &brw->ctx;
998 /* Dump a final BMP in case the application doesn't call SwapBuffers */
999 if (INTEL_DEBUG & DEBUG_AUB) {
1000 intel_batchbuffer_flush(brw);
1001 aub_dump_bmp(&brw->ctx);
1004 _mesa_meta_free(&brw->ctx);
1005 brw_meta_fast_clear_free(brw);
1007 if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
1008 /* Force a report. */
1009 brw->shader_time.report_time = 0;
1011 brw_collect_and_report_shader_time(brw);
1012 brw_destroy_shader_time(brw);
1015 brw_destroy_state(brw);
1016 brw_draw_destroy(brw);
1018 drm_intel_bo_unreference(brw->curbe.curbe_bo);
1019 if (brw->vs.base.scratch_bo)
1020 drm_intel_bo_unreference(brw->vs.base.scratch_bo);
1021 if (brw->gs.base.scratch_bo)
1022 drm_intel_bo_unreference(brw->gs.base.scratch_bo);
1023 if (brw->wm.base.scratch_bo)
1024 drm_intel_bo_unreference(brw->wm.base.scratch_bo);
1026 gen7_reset_hw_bt_pool_offsets(brw);
1027 drm_intel_bo_unreference(brw->hw_bt_pool.bo);
1028 brw->hw_bt_pool.bo = NULL;
1030 drm_intel_gem_context_destroy(brw->hw_ctx);
1032 if (ctx->swrast_context) {
1033 _swsetup_DestroyContext(&brw->ctx);
1034 _tnl_DestroyContext(&brw->ctx);
1036 _vbo_DestroyContext(&brw->ctx);
1038 if (ctx->swrast_context)
1039 _swrast_DestroyContext(&brw->ctx);
1041 brw_fini_pipe_control(brw);
1042 intel_batchbuffer_free(brw);
1044 drm_intel_bo_unreference(brw->throttle_batch[1]);
1045 drm_intel_bo_unreference(brw->throttle_batch[0]);
1046 brw->throttle_batch[1] = NULL;
1047 brw->throttle_batch[0] = NULL;
1049 driDestroyOptionCache(&brw->optionCache);
1051 /* free the Mesa context */
1052 _mesa_free_context_data(&brw->ctx);
1055 driContextPriv->driverPrivate = NULL;
1059 intelUnbindContext(__DRIcontext * driContextPriv)
1061 /* Unset current context and dispath table */
1062 _mesa_make_current(NULL, NULL, NULL);
1068 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1069 * on window system framebuffers.
1071 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1072 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1073 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1074 * for a visual where you're guaranteed to be capable, but it turns out that
1075 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1076 * incapable ones, because there's no difference between the two in resources
1077 * used. Applications thus get built that accidentally rely on the default
1078 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1081 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1082 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1083 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1084 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1085 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1086 * and get no sRGB encode (assuming that both kinds of visual are available).
1087 * Thus our choice to support sRGB by default on our visuals for desktop would
1088 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1090 * Unfortunately, renderbuffer setup happens before a context is created. So
1091 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1092 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1093 * yet), we go turn that back off before anyone finds out.
1096 intel_gles3_srgb_workaround(struct brw_context *brw,
1097 struct gl_framebuffer *fb)
1099 struct gl_context *ctx = &brw->ctx;
1101 if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable)
1104 /* Some day when we support the sRGB capable bit on visuals available for
1105 * GLES, we'll need to respect that and not disable things here.
1107 fb->Visual.sRGBCapable = false;
1108 for (int i = 0; i < BUFFER_COUNT; i++) {
1109 if (fb->Attachment[i].Renderbuffer &&
1110 fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_B8G8R8A8_SRGB) {
1111 fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_B8G8R8A8_UNORM;
1117 intelMakeCurrent(__DRIcontext * driContextPriv,
1118 __DRIdrawable * driDrawPriv,
1119 __DRIdrawable * driReadPriv)
1121 struct brw_context *brw;
1122 GET_CURRENT_CONTEXT(curCtx);
1125 brw = (struct brw_context *) driContextPriv->driverPrivate;
1129 /* According to the glXMakeCurrent() man page: "Pending commands to
1130 * the previous context, if any, are flushed before it is released."
1131 * But only flush if we're actually changing contexts.
1133 if (brw_context(curCtx) && brw_context(curCtx) != brw) {
1134 _mesa_flush(curCtx);
1137 if (driContextPriv) {
1138 struct gl_context *ctx = &brw->ctx;
1139 struct gl_framebuffer *fb, *readFb;
1141 if (driDrawPriv == NULL) {
1142 fb = _mesa_get_incomplete_framebuffer();
1144 fb = driDrawPriv->driverPrivate;
1145 driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1;
1148 if (driReadPriv == NULL) {
1149 readFb = _mesa_get_incomplete_framebuffer();
1151 readFb = driReadPriv->driverPrivate;
1152 driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1;
1155 /* The sRGB workaround changes the renderbuffer's format. We must change
1156 * the format before the renderbuffer's miptree get's allocated, otherwise
1157 * the formats of the renderbuffer and its miptree will differ.
1159 intel_gles3_srgb_workaround(brw, fb);
1160 intel_gles3_srgb_workaround(brw, readFb);
1162 /* If the context viewport hasn't been initialized, force a call out to
1163 * the loader to get buffers so we have a drawable size for the initial
1165 if (!brw->ctx.ViewportInitialized)
1166 intel_prepare_render(brw);
1168 _mesa_make_current(ctx, fb, readFb);
1170 _mesa_make_current(NULL, NULL, NULL);
1177 intel_resolve_for_dri2_flush(struct brw_context *brw,
1178 __DRIdrawable *drawable)
1181 /* MSAA and fast color clear are not supported, so don't waste time
1182 * checking whether a resolve is needed.
1187 struct gl_framebuffer *fb = drawable->driverPrivate;
1188 struct intel_renderbuffer *rb;
1190 /* Usually, only the back buffer will need to be downsampled. However,
1191 * the front buffer will also need it if the user has rendered into it.
1193 static const gl_buffer_index buffers[2] = {
1198 for (int i = 0; i < 2; ++i) {
1199 rb = intel_get_renderbuffer(fb, buffers[i]);
1200 if (rb == NULL || rb->mt == NULL)
1202 if (rb->mt->num_samples <= 1)
1203 intel_miptree_resolve_color(brw, rb->mt);
1205 intel_renderbuffer_downsample(brw, rb);
1210 intel_bits_per_pixel(const struct intel_renderbuffer *rb)
1212 return _mesa_get_format_bytes(intel_rb_format(rb)) * 8;
1216 intel_query_dri2_buffers(struct brw_context *brw,
1217 __DRIdrawable *drawable,
1218 __DRIbuffer **buffers,
1222 intel_process_dri2_buffer(struct brw_context *brw,
1223 __DRIdrawable *drawable,
1224 __DRIbuffer *buffer,
1225 struct intel_renderbuffer *rb,
1226 const char *buffer_name);
1229 intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable);
1232 intel_update_dri2_buffers(struct brw_context *brw, __DRIdrawable *drawable)
1234 struct gl_framebuffer *fb = drawable->driverPrivate;
1235 struct intel_renderbuffer *rb;
1236 __DRIbuffer *buffers = NULL;
1238 const char *region_name;
1240 /* Set this up front, so that in case our buffers get invalidated
1241 * while we're getting new buffers, we don't clobber the stamp and
1242 * thus ignore the invalidate. */
1243 drawable->lastStamp = drawable->dri2.stamp;
1245 if (unlikely(INTEL_DEBUG & DEBUG_DRI))
1246 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
1248 intel_query_dri2_buffers(brw, drawable, &buffers, &count);
1250 if (buffers == NULL)
1253 for (i = 0; i < count; i++) {
1254 switch (buffers[i].attachment) {
1255 case __DRI_BUFFER_FRONT_LEFT:
1256 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1257 region_name = "dri2 front buffer";
1260 case __DRI_BUFFER_FAKE_FRONT_LEFT:
1261 rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1262 region_name = "dri2 fake front buffer";
1265 case __DRI_BUFFER_BACK_LEFT:
1266 rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
1267 region_name = "dri2 back buffer";
1270 case __DRI_BUFFER_DEPTH:
1271 case __DRI_BUFFER_HIZ:
1272 case __DRI_BUFFER_DEPTH_STENCIL:
1273 case __DRI_BUFFER_STENCIL:
1274 case __DRI_BUFFER_ACCUM:
1277 "unhandled buffer attach event, attachment type %d\n",
1278 buffers[i].attachment);
1282 intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name);
1288 intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable)
1290 struct brw_context *brw = context->driverPrivate;
1291 __DRIscreen *screen = brw->intelScreen->driScrnPriv;
1293 /* Set this up front, so that in case our buffers get invalidated
1294 * while we're getting new buffers, we don't clobber the stamp and
1295 * thus ignore the invalidate. */
1296 drawable->lastStamp = drawable->dri2.stamp;
1298 if (unlikely(INTEL_DEBUG & DEBUG_DRI))
1299 fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
1301 if (screen->image.loader)
1302 intel_update_image_buffers(brw, drawable);
1304 intel_update_dri2_buffers(brw, drawable);
1306 driUpdateFramebufferSize(&brw->ctx, drawable);
1310 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1311 * state is required.
1314 intel_prepare_render(struct brw_context *brw)
1316 struct gl_context *ctx = &brw->ctx;
1317 __DRIcontext *driContext = brw->driContext;
1318 __DRIdrawable *drawable;
1320 drawable = driContext->driDrawablePriv;
1321 if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) {
1322 if (drawable->lastStamp != drawable->dri2.stamp)
1323 intel_update_renderbuffers(driContext, drawable);
1324 driContext->dri2.draw_stamp = drawable->dri2.stamp;
1327 drawable = driContext->driReadablePriv;
1328 if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) {
1329 if (drawable->lastStamp != drawable->dri2.stamp)
1330 intel_update_renderbuffers(driContext, drawable);
1331 driContext->dri2.read_stamp = drawable->dri2.stamp;
1334 /* If we're currently rendering to the front buffer, the rendering
1335 * that will happen next will probably dirty the front buffer. So
1336 * mark it as dirty here.
1338 if (_mesa_is_front_buffer_drawing(ctx->DrawBuffer))
1339 brw->front_buffer_dirty = true;
1343 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1345 * To determine which DRI buffers to request, examine the renderbuffers
1346 * attached to the drawable's framebuffer. Then request the buffers with
1347 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1349 * This is called from intel_update_renderbuffers().
1351 * \param drawable Drawable whose buffers are queried.
1352 * \param buffers [out] List of buffers returned by DRI2 query.
1353 * \param buffer_count [out] Number of buffers returned.
1355 * \see intel_update_renderbuffers()
1356 * \see DRI2GetBuffers()
1357 * \see DRI2GetBuffersWithFormat()
1360 intel_query_dri2_buffers(struct brw_context *brw,
1361 __DRIdrawable *drawable,
1362 __DRIbuffer **buffers,
1365 __DRIscreen *screen = brw->intelScreen->driScrnPriv;
1366 struct gl_framebuffer *fb = drawable->driverPrivate;
1368 unsigned attachments[8];
1370 struct intel_renderbuffer *front_rb;
1371 struct intel_renderbuffer *back_rb;
1373 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1374 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
1376 memset(attachments, 0, sizeof(attachments));
1377 if ((_mesa_is_front_buffer_drawing(fb) ||
1378 _mesa_is_front_buffer_reading(fb) ||
1379 !back_rb) && front_rb) {
1380 /* If a fake front buffer is in use, then querying for
1381 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1382 * the real front buffer to the fake front buffer. So before doing the
1383 * query, we need to make sure all the pending drawing has landed in the
1384 * real front buffer.
1386 intel_batchbuffer_flush(brw);
1387 intel_flush_front(&brw->ctx);
1389 attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
1390 attachments[i++] = intel_bits_per_pixel(front_rb);
1391 } else if (front_rb && brw->front_buffer_dirty) {
1392 /* We have pending front buffer rendering, but we aren't querying for a
1393 * front buffer. If the front buffer we have is a fake front buffer,
1394 * the X server is going to throw it away when it processes the query.
1395 * So before doing the query, make sure all the pending drawing has
1396 * landed in the real front buffer.
1398 intel_batchbuffer_flush(brw);
1399 intel_flush_front(&brw->ctx);
1403 attachments[i++] = __DRI_BUFFER_BACK_LEFT;
1404 attachments[i++] = intel_bits_per_pixel(back_rb);
1407 assert(i <= ARRAY_SIZE(attachments));
1409 *buffers = screen->dri2.loader->getBuffersWithFormat(drawable,
1414 drawable->loaderPrivate);
1418 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1420 * This is called from intel_update_renderbuffers().
1423 * DRI buffers whose attachment point is DRI2BufferStencil or
1424 * DRI2BufferDepthStencil are handled as special cases.
1426 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1427 * that is passed to drm_intel_bo_gem_create_from_name().
1429 * \see intel_update_renderbuffers()
1432 intel_process_dri2_buffer(struct brw_context *brw,
1433 __DRIdrawable *drawable,
1434 __DRIbuffer *buffer,
1435 struct intel_renderbuffer *rb,
1436 const char *buffer_name)
1438 struct gl_framebuffer *fb = drawable->driverPrivate;
1444 unsigned num_samples = rb->Base.Base.NumSamples;
1446 /* We try to avoid closing and reopening the same BO name, because the first
1447 * use of a mapping of the buffer involves a bunch of page faulting which is
1448 * moderately expensive.
1450 struct intel_mipmap_tree *last_mt;
1451 if (num_samples == 0)
1454 last_mt = rb->singlesample_mt;
1456 uint32_t old_name = 0;
1458 /* The bo already has a name because the miptree was created by a
1459 * previous call to intel_process_dri2_buffer(). If a bo already has a
1460 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1461 * create a new name.
1463 drm_intel_bo_flink(last_mt->bo, &old_name);
1466 if (old_name == buffer->name)
1469 if (unlikely(INTEL_DEBUG & DEBUG_DRI)) {
1471 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1472 buffer->name, buffer->attachment,
1473 buffer->cpp, buffer->pitch);
1476 bo = drm_intel_bo_gem_create_from_name(brw->bufmgr, buffer_name,
1480 "Failed to open BO for returned DRI2 buffer "
1481 "(%dx%d, %s, named %d).\n"
1482 "This is likely a bug in the X Server that will lead to a "
1484 drawable->w, drawable->h, buffer_name, buffer->name);
1488 intel_update_winsys_renderbuffer_miptree(brw, rb, bo,
1489 drawable->w, drawable->h,
1492 if (_mesa_is_front_buffer_drawing(fb) &&
1493 (buffer->attachment == __DRI_BUFFER_FRONT_LEFT ||
1494 buffer->attachment == __DRI_BUFFER_FAKE_FRONT_LEFT) &&
1495 rb->Base.Base.NumSamples > 1) {
1496 intel_renderbuffer_upsample(brw, rb);
1501 drm_intel_bo_unreference(bo);
1505 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1507 * To determine which DRI buffers to request, examine the renderbuffers
1508 * attached to the drawable's framebuffer. Then request the buffers from
1511 * This is called from intel_update_renderbuffers().
1513 * \param drawable Drawable whose buffers are queried.
1514 * \param buffers [out] List of buffers returned by DRI2 query.
1515 * \param buffer_count [out] Number of buffers returned.
1517 * \see intel_update_renderbuffers()
1521 intel_update_image_buffer(struct brw_context *intel,
1522 __DRIdrawable *drawable,
1523 struct intel_renderbuffer *rb,
1525 enum __DRIimageBufferMask buffer_type)
1527 struct gl_framebuffer *fb = drawable->driverPrivate;
1529 if (!rb || !buffer->bo)
1532 unsigned num_samples = rb->Base.Base.NumSamples;
1534 /* Check and see if we're already bound to the right
1537 struct intel_mipmap_tree *last_mt;
1538 if (num_samples == 0)
1541 last_mt = rb->singlesample_mt;
1543 if (last_mt && last_mt->bo == buffer->bo)
1546 intel_update_winsys_renderbuffer_miptree(intel, rb, buffer->bo,
1547 buffer->width, buffer->height,
1550 if (_mesa_is_front_buffer_drawing(fb) &&
1551 buffer_type == __DRI_IMAGE_BUFFER_FRONT &&
1552 rb->Base.Base.NumSamples > 1) {
1553 intel_renderbuffer_upsample(intel, rb);
1558 intel_update_image_buffers(struct brw_context *brw, __DRIdrawable *drawable)
1560 struct gl_framebuffer *fb = drawable->driverPrivate;
1561 __DRIscreen *screen = brw->intelScreen->driScrnPriv;
1562 struct intel_renderbuffer *front_rb;
1563 struct intel_renderbuffer *back_rb;
1564 struct __DRIimageList images;
1565 unsigned int format;
1566 uint32_t buffer_mask = 0;
1568 front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
1569 back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
1572 format = intel_rb_format(back_rb);
1574 format = intel_rb_format(front_rb);
1578 if (front_rb && (_mesa_is_front_buffer_drawing(fb) ||
1579 _mesa_is_front_buffer_reading(fb) || !back_rb)) {
1580 buffer_mask |= __DRI_IMAGE_BUFFER_FRONT;
1584 buffer_mask |= __DRI_IMAGE_BUFFER_BACK;
1586 (*screen->image.loader->getBuffers) (drawable,
1587 driGLFormatToImageFormat(format),
1588 &drawable->dri2.stamp,
1589 drawable->loaderPrivate,
1593 if (images.image_mask & __DRI_IMAGE_BUFFER_FRONT) {
1594 drawable->w = images.front->width;
1595 drawable->h = images.front->height;
1596 intel_update_image_buffer(brw,
1600 __DRI_IMAGE_BUFFER_FRONT);
1602 if (images.image_mask & __DRI_IMAGE_BUFFER_BACK) {
1603 drawable->w = images.back->width;
1604 drawable->h = images.back->height;
1605 intel_update_image_buffer(brw,
1609 __DRI_IMAGE_BUFFER_BACK);