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/imports.h"
38 #include "main/macros.h"
39 #include "main/points.h"
40 #include "main/version.h"
41 #include "main/vtxfmt.h"
43 #include "vbo/vbo_context.h"
45 #include "drivers/common/driverfuncs.h"
46 #include "drivers/common/meta.h"
49 #include "brw_context.h"
50 #include "brw_defines.h"
52 #include "brw_state.h"
54 #include "intel_batchbuffer.h"
55 #include "intel_buffer_objects.h"
56 #include "intel_buffers.h"
57 #include "intel_fbo.h"
58 #include "intel_mipmap_tree.h"
59 #include "intel_pixel.h"
60 #include "intel_regions.h"
61 #include "intel_tex.h"
62 #include "intel_tex_obj.h"
64 #include "swrast_setup/swrast_setup.h"
66 #include "tnl/t_pipeline.h"
67 #include "glsl/ralloc.h"
69 /***************************************
70 * Mesa's Driver Functions
71 ***************************************/
74 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
75 GLenum internalFormat
, int samples
[16])
77 struct brw_context
*brw
= brw_context(ctx
);
97 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
100 brw_get_renderer_string(unsigned deviceID
)
103 static char buffer
[128];
107 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
108 #include "pci_ids/i965_pci_ids.h"
110 chipset
= "Unknown Intel Chipset";
114 (void) driGetRendererString(buffer
, chipset
, 0);
118 static const GLubyte
*
119 intelGetString(struct gl_context
* ctx
, GLenum name
)
121 const struct brw_context
*const brw
= brw_context(ctx
);
125 return (GLubyte
*) brw_vendor_string
;
129 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
137 intel_viewport(struct gl_context
*ctx
)
139 struct brw_context
*brw
= brw_context(ctx
);
140 __DRIcontext
*driContext
= brw
->driContext
;
142 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
143 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
144 dri2InvalidateDrawable(driContext
->driReadablePriv
);
149 intelInvalidateState(struct gl_context
* ctx
, GLuint new_state
)
151 struct brw_context
*brw
= brw_context(ctx
);
153 if (ctx
->swrast_context
)
154 _swrast_InvalidateState(ctx
, new_state
);
155 _vbo_InvalidateState(ctx
, new_state
);
157 brw
->NewGLState
|= new_state
;
160 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
163 intel_flush_front(struct gl_context
*ctx
)
165 struct brw_context
*brw
= brw_context(ctx
);
166 __DRIcontext
*driContext
= brw
->driContext
;
167 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
168 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
170 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
171 if (flushFront(screen
) && driDrawable
&&
172 driDrawable
->loaderPrivate
) {
174 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
176 * This potentially resolves both front and back buffer. It
177 * is unnecessary to resolve the back, but harms nothing except
178 * performance. And no one cares about front-buffer render
181 intel_resolve_for_dri2_flush(brw
, driDrawable
);
182 intel_batchbuffer_flush(brw
);
184 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
186 /* We set the dirty bit in intel_prepare_render() if we're
187 * front buffer rendering once we get there.
189 brw
->front_buffer_dirty
= false;
195 intel_glFlush(struct gl_context
*ctx
)
197 struct brw_context
*brw
= brw_context(ctx
);
199 intel_batchbuffer_flush(brw
);
200 intel_flush_front(ctx
);
201 if (brw
->is_front_buffer_rendering
)
202 brw
->need_throttle
= true;
206 intelFinish(struct gl_context
* ctx
)
208 struct brw_context
*brw
= brw_context(ctx
);
212 if (brw
->batch
.last_bo
)
213 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
217 brw_init_driver_functions(struct brw_context
*brw
,
218 struct dd_function_table
*functions
)
220 _mesa_init_driver_functions(functions
);
222 /* GLX uses DRI2 invalidate events to handle window resizing.
223 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
224 * which doesn't provide a mechanism for snooping the event queues.
226 * So EGL still relies on viewport hacks to handle window resizing.
227 * This should go away with DRI3000.
229 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
230 functions
->Viewport
= intel_viewport
;
232 functions
->Flush
= intel_glFlush
;
233 functions
->Finish
= intelFinish
;
234 functions
->GetString
= intelGetString
;
235 functions
->UpdateState
= intelInvalidateState
;
237 intelInitTextureFuncs(functions
);
238 intelInitTextureImageFuncs(functions
);
239 intelInitTextureSubImageFuncs(functions
);
240 intelInitTextureCopyImageFuncs(functions
);
241 intelInitClearFuncs(functions
);
242 intelInitBufferFuncs(functions
);
243 intelInitPixelFuncs(functions
);
244 intelInitBufferObjectFuncs(functions
);
245 intel_init_syncobj_functions(functions
);
246 brw_init_object_purgeable_functions(functions
);
248 brwInitFragProgFuncs( functions
);
249 brw_init_common_queryobj_functions(functions
);
251 gen6_init_queryobj_functions(functions
);
253 gen4_init_queryobj_functions(functions
);
255 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
257 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
258 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
259 functions
->GetTransformFeedbackVertexCount
=
260 brw_get_transform_feedback_vertex_count
;
262 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
263 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
264 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
265 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
267 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
268 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
272 functions
->GetSamplePosition
= gen6_get_sample_position
;
276 brw_initialize_context_constants(struct brw_context
*brw
)
278 struct gl_context
*ctx
= &brw
->ctx
;
280 unsigned max_samplers
=
281 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
283 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
285 ctx
->Const
.StripTextureBorder
= true;
287 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
288 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
289 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
290 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
291 ctx
->Const
.MaxTextureUnits
=
292 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
293 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
294 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
296 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
298 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
299 ctx
->Const
.MaxCombinedTextureImageUnits
=
300 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
301 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
302 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
;
304 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
305 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
306 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
307 ctx
->Const
.Max3DTextureLevels
= 9;
308 ctx
->Const
.MaxCubeTextureLevels
= 12;
311 ctx
->Const
.MaxArrayTextureLayers
= 2048;
313 ctx
->Const
.MaxArrayTextureLayers
= 512;
315 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
317 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
319 ctx
->Const
.MaxRenderbufferSize
= 8192;
321 /* Hardware only supports a limited number of transform feedback buffers.
322 * So we need to override the Mesa default (which is based only on software
325 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
327 /* On Gen6, in the worst case, we use up one binding table entry per
328 * transform feedback component (see comments above the definition of
329 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
330 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
331 * BRW_MAX_SOL_BINDINGS.
333 * In "separate components" mode, we need to divide this value by
334 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
335 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
337 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
338 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
339 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
341 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
344 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
345 const int clamp_max_samples
=
346 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
348 if (clamp_max_samples
< 0) {
349 max_samples
= msaa_modes
[0];
351 /* Select the largest supported MSAA mode that does not exceed
355 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
356 if (msaa_modes
[i
] <= clamp_max_samples
) {
357 max_samples
= msaa_modes
[i
];
363 ctx
->Const
.MaxSamples
= max_samples
;
364 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
365 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
366 ctx
->Const
.MaxIntegerSamples
= max_samples
;
369 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
371 ctx
->Const
.MinLineWidth
= 1.0;
372 ctx
->Const
.MinLineWidthAA
= 1.0;
373 ctx
->Const
.MaxLineWidth
= 5.0;
374 ctx
->Const
.MaxLineWidthAA
= 5.0;
375 ctx
->Const
.LineWidthGranularity
= 0.5;
377 ctx
->Const
.MinPointSize
= 1.0;
378 ctx
->Const
.MinPointSizeAA
= 1.0;
379 ctx
->Const
.MaxPointSize
= 255.0;
380 ctx
->Const
.MaxPointSizeAA
= 255.0;
381 ctx
->Const
.PointSizeGranularity
= 1.0;
383 if (brw
->gen
>= 5 || brw
->is_g4x
)
384 ctx
->Const
.MaxClipPlanes
= 8;
386 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
387 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
388 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
389 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
390 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
391 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
392 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
393 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
394 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
395 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
396 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
397 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
398 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
399 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
401 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
402 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
403 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
404 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
405 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
406 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
407 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
408 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
409 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
410 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
411 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
413 /* Fragment shaders use real, 32-bit twos-complement integers for all
416 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
417 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
418 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
419 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
420 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
423 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
424 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
425 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
426 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
427 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
428 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
429 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
432 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
433 * but we're not sure how it's actually done for vertex order,
434 * that affect provoking vertex decision. Always use last vertex
435 * convention for quad primitive which works as expected for now.
438 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
440 ctx
->Const
.NativeIntegers
= true;
441 ctx
->Const
.UniformBooleanTrue
= 1;
443 /* From the gen4 PRM, volume 4 page 127:
445 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
446 * the base address of the first element of the surface, computed in
447 * software by adding the surface base address to the byte offset of
448 * the element in the buffer."
450 * However, unaligned accesses are slower, so enforce buffer alignment.
452 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
453 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
456 ctx
->Const
.MaxVarying
= 32;
457 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
458 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
459 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
460 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
463 /* We want the GLSL compiler to emit code that uses condition codes */
464 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
465 ctx
->ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
466 ctx
->ShaderCompilerOptions
[i
].EmitCondCodes
= true;
467 ctx
->ShaderCompilerOptions
[i
].EmitNoNoise
= true;
468 ctx
->ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
469 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
470 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectOutput
= true;
472 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectUniform
=
473 (i
== MESA_SHADER_FRAGMENT
);
474 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
475 (i
== MESA_SHADER_FRAGMENT
);
476 ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
= true;
479 ctx
->ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
480 ctx
->ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
482 /* ARB_viewport_array */
483 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
484 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
485 ctx
->Const
.ViewportSubpixelBits
= 0;
487 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
489 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
490 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
495 * Process driconf (drirc) options, setting appropriate context flags.
497 * intelInitExtensions still pokes at optionCache directly, in order to
498 * avoid advertising various extensions. No flags are set, so it makes
499 * sense to continue doing that there.
502 brw_process_driconf_options(struct brw_context
*brw
)
504 struct gl_context
*ctx
= &brw
->ctx
;
506 driOptionCache
*options
= &brw
->optionCache
;
507 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
508 brw
->driContext
->driScreenPriv
->myNum
, "i965");
510 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
511 switch (bo_reuse_mode
) {
512 case DRI_CONF_BO_REUSE_DISABLED
:
514 case DRI_CONF_BO_REUSE_ALL
:
515 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
519 if (!driQueryOptionb(options
, "hiz")) {
520 brw
->has_hiz
= false;
521 /* On gen6, you can only do separate stencil with HIZ. */
523 brw
->has_separate_stencil
= false;
526 if (driQueryOptionb(options
, "always_flush_batch")) {
527 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
528 brw
->always_flush_batch
= true;
531 if (driQueryOptionb(options
, "always_flush_cache")) {
532 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
533 brw
->always_flush_cache
= true;
536 if (driQueryOptionb(options
, "disable_throttling")) {
537 fprintf(stderr
, "disabling flush throttling\n");
538 brw
->disable_throttling
= true;
541 brw
->disable_derivative_optimization
=
542 driQueryOptionb(&brw
->optionCache
, "disable_derivative_optimization");
544 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
546 ctx
->Const
.ForceGLSLExtensionsWarn
=
547 driQueryOptionb(options
, "force_glsl_extensions_warn");
549 ctx
->Const
.DisableGLSLLineContinuations
=
550 driQueryOptionb(options
, "disable_glsl_line_continuations");
554 brwCreateContext(gl_api api
,
555 const struct gl_config
*mesaVis
,
556 __DRIcontext
*driContextPriv
,
557 unsigned major_version
,
558 unsigned minor_version
,
561 unsigned *dri_ctx_error
,
562 void *sharedContextPrivate
)
564 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
565 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
566 struct intel_screen
*screen
= sPriv
->driverPrivate
;
567 const struct brw_device_info
*devinfo
= screen
->devinfo
;
568 struct dd_function_table functions
;
569 struct gl_config visual
;
571 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
572 * provides us with context reset notifications.
574 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
575 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
577 if (screen
->has_context_reset_notification
)
578 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
580 if (flags
& ~allowed_flags
) {
581 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
585 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
587 printf("%s: failed to alloc context\n", __FUNCTION__
);
588 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
592 driContextPriv
->driverPrivate
= brw
;
593 brw
->driContext
= driContextPriv
;
594 brw
->intelScreen
= screen
;
595 brw
->bufmgr
= screen
->bufmgr
;
597 brw
->gen
= devinfo
->gen
;
598 brw
->gt
= devinfo
->gt
;
599 brw
->is_g4x
= devinfo
->is_g4x
;
600 brw
->is_baytrail
= devinfo
->is_baytrail
;
601 brw
->is_haswell
= devinfo
->is_haswell
;
602 brw
->has_llc
= devinfo
->has_llc
;
603 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
&& brw
->gen
< 8;
604 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
605 brw
->has_pln
= devinfo
->has_pln
;
606 brw
->has_compr4
= devinfo
->has_compr4
;
607 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
608 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
609 brw
->needs_unlit_centroid_workaround
=
610 devinfo
->needs_unlit_centroid_workaround
;
612 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
613 brw
->has_swizzling
= screen
->hw_has_swizzling
;
616 gen8_init_vtable_surface_functions(brw
);
617 gen7_init_vtable_sampler_functions(brw
);
618 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
619 } else if (brw
->gen
>= 7) {
620 gen7_init_vtable_surface_functions(brw
);
621 gen7_init_vtable_sampler_functions(brw
);
622 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
624 gen4_init_vtable_surface_functions(brw
);
625 gen4_init_vtable_sampler_functions(brw
);
626 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
629 brw_init_driver_functions(brw
, &functions
);
632 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
634 struct gl_context
*ctx
= &brw
->ctx
;
636 if (mesaVis
== NULL
) {
637 memset(&visual
, 0, sizeof visual
);
641 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
642 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
643 printf("%s: failed to init mesa context\n", __FUNCTION__
);
644 intelDestroyContext(driContextPriv
);
648 driContextSetFlags(ctx
, flags
);
650 /* Initialize the software rasterizer and helper modules.
652 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
653 * software fallbacks (which we have to support on legacy GL to do weird
654 * glDrawPixels(), glBitmap(), and other functions).
656 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
657 _swrast_CreateContext(ctx
);
660 _vbo_CreateContext(ctx
);
661 if (ctx
->swrast_context
) {
662 _tnl_CreateContext(ctx
);
663 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
664 _swsetup_CreateContext(ctx
);
666 /* Configure swrast to match hardware characteristics: */
667 _swrast_allow_pixel_fog(ctx
, false);
668 _swrast_allow_vertex_fog(ctx
, true);
671 _mesa_meta_init(ctx
);
673 brw_process_driconf_options(brw
);
674 brw_process_intel_debug_variable(brw
);
675 brw_initialize_context_constants(brw
);
677 ctx
->Const
.ResetStrategy
= notify_reset
678 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
680 /* Reinitialize the context point state. It depends on ctx->Const values. */
681 _mesa_init_point(ctx
);
683 intel_batchbuffer_init(brw
);
687 intelInitExtensions(ctx
);
692 /* Create a new hardware context. Using a hardware context means that
693 * our GPU state will be saved/restored on context switch, allowing us
694 * to assume that the GPU is in the same state we left it in.
696 * This is required for transform feedback buffer offsets, query objects,
697 * and also allows us to reduce how much state we have to emit.
699 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
702 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
703 intelDestroyContext(driContextPriv
);
708 brw_init_surface_formats(brw
);
710 if (brw
->is_g4x
|| brw
->gen
>= 5) {
711 brw
->CMD_VF_STATISTICS
= GM45_3DSTATE_VF_STATISTICS
;
712 brw
->CMD_PIPELINE_SELECT
= CMD_PIPELINE_SELECT_GM45
;
714 brw
->CMD_VF_STATISTICS
= GEN4_3DSTATE_VF_STATISTICS
;
715 brw
->CMD_PIPELINE_SELECT
= CMD_PIPELINE_SELECT_965
;
718 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
719 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
720 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
721 brw
->urb
.size
= devinfo
->urb
.size
;
722 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
723 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
724 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
726 /* Estimate the size of the mappable aperture into the GTT. There's an
727 * ioctl to get the whole GTT size, but not one to get the mappable subset.
728 * It turns out it's basically always 256MB, though some ancient hardware
731 uint32_t gtt_size
= 256 * 1024 * 1024;
733 /* We don't want to map two objects such that a memcpy between them would
734 * just fault one mapping in and then the other over and over forever. So
735 * we would need to divide the GTT size by 2. Additionally, some GTT is
736 * taken up by things like the framebuffer and the ringbuffer and such, so
737 * be more conservative.
739 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
742 brw
->urb
.gen6_gs_previously_active
= false;
744 brw
->prim_restart
.in_progress
= false;
745 brw
->prim_restart
.enable_cut_index
= false;
746 brw
->gs
.enabled
= false;
749 brw
->curbe
.last_buf
= calloc(1, 4096);
750 brw
->curbe
.next_buf
= calloc(1, 4096);
753 ctx
->VertexProgram
._MaintainTnlProgram
= true;
754 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
756 brw_draw_init( brw
);
758 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
759 /* Turn on some extra GL_ARB_debug_output generation. */
760 brw
->perf_debug
= true;
763 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
764 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
766 brw_fs_alloc_reg_sets(brw
);
767 brw_vec4_alloc_reg_set(brw
);
769 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
770 brw_init_shader_time(brw
);
772 _mesa_compute_version(ctx
);
774 _mesa_initialize_dispatch_tables(ctx
);
775 _mesa_initialize_vbo_vtxfmt(ctx
);
777 if (ctx
->Extensions
.AMD_performance_monitor
) {
778 brw_init_performance_monitors(brw
);
785 intelDestroyContext(__DRIcontext
* driContextPriv
)
787 struct brw_context
*brw
=
788 (struct brw_context
*) driContextPriv
->driverPrivate
;
789 struct gl_context
*ctx
= &brw
->ctx
;
791 assert(brw
); /* should never be null */
795 /* Dump a final BMP in case the application doesn't call SwapBuffers */
796 if (INTEL_DEBUG
& DEBUG_AUB
) {
797 intel_batchbuffer_flush(brw
);
798 aub_dump_bmp(&brw
->ctx
);
801 _mesa_meta_free(&brw
->ctx
);
803 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
804 /* Force a report. */
805 brw
->shader_time
.report_time
= 0;
807 brw_collect_and_report_shader_time(brw
);
808 brw_destroy_shader_time(brw
);
811 brw_destroy_state(brw
);
812 brw_draw_destroy(brw
);
814 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
815 drm_intel_bo_unreference(brw
->vs
.base
.const_bo
);
816 drm_intel_bo_unreference(brw
->wm
.base
.const_bo
);
818 free(brw
->curbe
.last_buf
);
819 free(brw
->curbe
.next_buf
);
821 drm_intel_gem_context_destroy(brw
->hw_ctx
);
823 if (ctx
->swrast_context
) {
824 _swsetup_DestroyContext(&brw
->ctx
);
825 _tnl_DestroyContext(&brw
->ctx
);
827 _vbo_DestroyContext(&brw
->ctx
);
829 if (ctx
->swrast_context
)
830 _swrast_DestroyContext(&brw
->ctx
);
832 intel_batchbuffer_free(brw
);
834 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
835 brw
->first_post_swapbuffers_batch
= NULL
;
837 driDestroyOptionCache(&brw
->optionCache
);
839 /* free the Mesa context */
840 _mesa_free_context_data(&brw
->ctx
);
843 driContextPriv
->driverPrivate
= NULL
;
847 intelUnbindContext(__DRIcontext
* driContextPriv
)
849 /* Unset current context and dispath table */
850 _mesa_make_current(NULL
, NULL
, NULL
);
856 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
857 * on window system framebuffers.
859 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
860 * your renderbuffer can do sRGB encode, and you can flip a switch that does
861 * sRGB encode if the renderbuffer can handle it. You can ask specifically
862 * for a visual where you're guaranteed to be capable, but it turns out that
863 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
864 * incapable ones, becuase there's no difference between the two in resources
865 * used. Applications thus get built that accidentally rely on the default
866 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
869 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
870 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
871 * So they removed the enable knob and made it "if the renderbuffer is sRGB
872 * capable, do sRGB encode". Then, for your window system renderbuffers, you
873 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
874 * and get no sRGB encode (assuming that both kinds of visual are available).
875 * Thus our choice to support sRGB by default on our visuals for desktop would
876 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
878 * Unfortunately, renderbuffer setup happens before a context is created. So
879 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
880 * context (without an sRGB visual, though we don't have sRGB visuals exposed
881 * yet), we go turn that back off before anyone finds out.
884 intel_gles3_srgb_workaround(struct brw_context
*brw
,
885 struct gl_framebuffer
*fb
)
887 struct gl_context
*ctx
= &brw
->ctx
;
889 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
892 /* Some day when we support the sRGB capable bit on visuals available for
893 * GLES, we'll need to respect that and not disable things here.
895 fb
->Visual
.sRGBCapable
= false;
896 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
897 if (fb
->Attachment
[i
].Renderbuffer
&&
898 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
899 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
905 intelMakeCurrent(__DRIcontext
* driContextPriv
,
906 __DRIdrawable
* driDrawPriv
,
907 __DRIdrawable
* driReadPriv
)
909 struct brw_context
*brw
;
910 GET_CURRENT_CONTEXT(curCtx
);
913 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
917 /* According to the glXMakeCurrent() man page: "Pending commands to
918 * the previous context, if any, are flushed before it is released."
919 * But only flush if we're actually changing contexts.
921 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
925 if (driContextPriv
) {
926 struct gl_context
*ctx
= &brw
->ctx
;
927 struct gl_framebuffer
*fb
, *readFb
;
929 if (driDrawPriv
== NULL
&& driReadPriv
== NULL
) {
930 fb
= _mesa_get_incomplete_framebuffer();
931 readFb
= _mesa_get_incomplete_framebuffer();
933 fb
= driDrawPriv
->driverPrivate
;
934 readFb
= driReadPriv
->driverPrivate
;
935 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
936 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
939 /* The sRGB workaround changes the renderbuffer's format. We must change
940 * the format before the renderbuffer's miptree get's allocated, otherwise
941 * the formats of the renderbuffer and its miptree will differ.
943 intel_gles3_srgb_workaround(brw
, fb
);
944 intel_gles3_srgb_workaround(brw
, readFb
);
946 /* If the context viewport hasn't been initialized, force a call out to
947 * the loader to get buffers so we have a drawable size for the initial
949 if (!brw
->ctx
.ViewportInitialized
)
950 intel_prepare_render(brw
);
952 _mesa_make_current(ctx
, fb
, readFb
);
954 _mesa_make_current(NULL
, NULL
, NULL
);
961 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
962 __DRIdrawable
*drawable
)
965 /* MSAA and fast color clear are not supported, so don't waste time
966 * checking whether a resolve is needed.
971 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
972 struct intel_renderbuffer
*rb
;
974 /* Usually, only the back buffer will need to be downsampled. However,
975 * the front buffer will also need it if the user has rendered into it.
977 static const gl_buffer_index buffers
[2] = {
982 for (int i
= 0; i
< 2; ++i
) {
983 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
984 if (rb
== NULL
|| rb
->mt
== NULL
)
986 if (rb
->mt
->num_samples
<= 1)
987 intel_miptree_resolve_color(brw
, rb
->mt
);
989 intel_miptree_downsample(brw
, rb
->mt
);
994 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
996 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1000 intel_query_dri2_buffers(struct brw_context
*brw
,
1001 __DRIdrawable
*drawable
,
1002 __DRIbuffer
**buffers
,
1006 intel_process_dri2_buffer(struct brw_context
*brw
,
1007 __DRIdrawable
*drawable
,
1008 __DRIbuffer
*buffer
,
1009 struct intel_renderbuffer
*rb
,
1010 const char *buffer_name
);
1013 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1016 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1018 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1019 struct intel_renderbuffer
*rb
;
1020 __DRIbuffer
*buffers
= NULL
;
1022 const char *region_name
;
1024 /* Set this up front, so that in case our buffers get invalidated
1025 * while we're getting new buffers, we don't clobber the stamp and
1026 * thus ignore the invalidate. */
1027 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1029 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1030 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1032 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1034 if (buffers
== NULL
)
1037 for (i
= 0; i
< count
; i
++) {
1038 switch (buffers
[i
].attachment
) {
1039 case __DRI_BUFFER_FRONT_LEFT
:
1040 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1041 region_name
= "dri2 front buffer";
1044 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1045 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1046 region_name
= "dri2 fake front buffer";
1049 case __DRI_BUFFER_BACK_LEFT
:
1050 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1051 region_name
= "dri2 back buffer";
1054 case __DRI_BUFFER_DEPTH
:
1055 case __DRI_BUFFER_HIZ
:
1056 case __DRI_BUFFER_DEPTH_STENCIL
:
1057 case __DRI_BUFFER_STENCIL
:
1058 case __DRI_BUFFER_ACCUM
:
1061 "unhandled buffer attach event, attachment type %d\n",
1062 buffers
[i
].attachment
);
1066 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1072 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1074 struct brw_context
*brw
= context
->driverPrivate
;
1075 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1077 /* Set this up front, so that in case our buffers get invalidated
1078 * while we're getting new buffers, we don't clobber the stamp and
1079 * thus ignore the invalidate. */
1080 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1082 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1083 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1085 if (screen
->image
.loader
)
1086 intel_update_image_buffers(brw
, drawable
);
1088 intel_update_dri2_buffers(brw
, drawable
);
1090 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1094 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1095 * state is required.
1098 intel_prepare_render(struct brw_context
*brw
)
1100 __DRIcontext
*driContext
= brw
->driContext
;
1101 __DRIdrawable
*drawable
;
1103 drawable
= driContext
->driDrawablePriv
;
1104 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1105 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1106 intel_update_renderbuffers(driContext
, drawable
);
1107 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1110 drawable
= driContext
->driReadablePriv
;
1111 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1112 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1113 intel_update_renderbuffers(driContext
, drawable
);
1114 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1117 /* If we're currently rendering to the front buffer, the rendering
1118 * that will happen next will probably dirty the front buffer. So
1119 * mark it as dirty here.
1121 if (brw
->is_front_buffer_rendering
)
1122 brw
->front_buffer_dirty
= true;
1124 /* Wait for the swapbuffers before the one we just emitted, so we
1125 * don't get too many swaps outstanding for apps that are GPU-heavy
1126 * but not CPU-heavy.
1128 * We're using intelDRI2Flush (called from the loader before
1129 * swapbuffer) and glFlush (for front buffer rendering) as the
1130 * indicator that a frame is done and then throttle when we get
1131 * here as we prepare to render the next frame. At this point for
1132 * round trips for swap/copy and getting new buffers are done and
1133 * we'll spend less time waiting on the GPU.
1135 * Unfortunately, we don't have a handle to the batch containing
1136 * the swap, and getting our hands on that doesn't seem worth it,
1137 * so we just us the first batch we emitted after the last swap.
1139 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1140 if (!brw
->disable_throttling
)
1141 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1142 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1143 brw
->first_post_swapbuffers_batch
= NULL
;
1144 brw
->need_throttle
= false;
1149 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1151 * To determine which DRI buffers to request, examine the renderbuffers
1152 * attached to the drawable's framebuffer. Then request the buffers with
1153 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1155 * This is called from intel_update_renderbuffers().
1157 * \param drawable Drawable whose buffers are queried.
1158 * \param buffers [out] List of buffers returned by DRI2 query.
1159 * \param buffer_count [out] Number of buffers returned.
1161 * \see intel_update_renderbuffers()
1162 * \see DRI2GetBuffers()
1163 * \see DRI2GetBuffersWithFormat()
1166 intel_query_dri2_buffers(struct brw_context
*brw
,
1167 __DRIdrawable
*drawable
,
1168 __DRIbuffer
**buffers
,
1171 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1172 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1174 unsigned attachments
[8];
1176 struct intel_renderbuffer
*front_rb
;
1177 struct intel_renderbuffer
*back_rb
;
1179 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1180 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1182 memset(attachments
, 0, sizeof(attachments
));
1183 if ((brw
->is_front_buffer_rendering
||
1184 brw
->is_front_buffer_reading
||
1185 !back_rb
) && front_rb
) {
1186 /* If a fake front buffer is in use, then querying for
1187 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1188 * the real front buffer to the fake front buffer. So before doing the
1189 * query, we need to make sure all the pending drawing has landed in the
1190 * real front buffer.
1192 intel_batchbuffer_flush(brw
);
1193 intel_flush_front(&brw
->ctx
);
1195 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1196 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1197 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1198 /* We have pending front buffer rendering, but we aren't querying for a
1199 * front buffer. If the front buffer we have is a fake front buffer,
1200 * the X server is going to throw it away when it processes the query.
1201 * So before doing the query, make sure all the pending drawing has
1202 * landed in the real front buffer.
1204 intel_batchbuffer_flush(brw
);
1205 intel_flush_front(&brw
->ctx
);
1209 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1210 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1213 assert(i
<= ARRAY_SIZE(attachments
));
1215 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1220 drawable
->loaderPrivate
);
1224 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1226 * This is called from intel_update_renderbuffers().
1229 * DRI buffers whose attachment point is DRI2BufferStencil or
1230 * DRI2BufferDepthStencil are handled as special cases.
1232 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1233 * that is passed to intel_region_alloc_for_handle().
1235 * \see intel_update_renderbuffers()
1236 * \see intel_region_alloc_for_handle()
1239 intel_process_dri2_buffer(struct brw_context
*brw
,
1240 __DRIdrawable
*drawable
,
1241 __DRIbuffer
*buffer
,
1242 struct intel_renderbuffer
*rb
,
1243 const char *buffer_name
)
1245 struct intel_region
*region
= NULL
;
1250 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1252 /* We try to avoid closing and reopening the same BO name, because the first
1253 * use of a mapping of the buffer involves a bunch of page faulting which is
1254 * moderately expensive.
1256 if (num_samples
== 0) {
1259 rb
->mt
->region
->name
== buffer
->name
)
1263 rb
->mt
->singlesample_mt
&&
1264 rb
->mt
->singlesample_mt
->region
&&
1265 rb
->mt
->singlesample_mt
->region
->name
== buffer
->name
)
1269 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1271 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1272 buffer
->name
, buffer
->attachment
,
1273 buffer
->cpp
, buffer
->pitch
);
1276 intel_miptree_release(&rb
->mt
);
1277 region
= intel_region_alloc_for_handle(brw
->intelScreen
,
1287 rb
->mt
= intel_miptree_create_for_dri2_buffer(brw
,
1289 intel_rb_format(rb
),
1292 intel_region_release(®ion
);
1296 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1298 * To determine which DRI buffers to request, examine the renderbuffers
1299 * attached to the drawable's framebuffer. Then request the buffers from
1302 * This is called from intel_update_renderbuffers().
1304 * \param drawable Drawable whose buffers are queried.
1305 * \param buffers [out] List of buffers returned by DRI2 query.
1306 * \param buffer_count [out] Number of buffers returned.
1308 * \see intel_update_renderbuffers()
1312 intel_update_image_buffer(struct brw_context
*intel
,
1313 __DRIdrawable
*drawable
,
1314 struct intel_renderbuffer
*rb
,
1316 enum __DRIimageBufferMask buffer_type
)
1318 struct intel_region
*region
= buffer
->region
;
1323 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1325 /* Check and see if we're already bound to the right
1328 if (num_samples
== 0) {
1331 rb
->mt
->region
->bo
== region
->bo
)
1335 rb
->mt
->singlesample_mt
&&
1336 rb
->mt
->singlesample_mt
->region
&&
1337 rb
->mt
->singlesample_mt
->region
->bo
== region
->bo
)
1341 intel_miptree_release(&rb
->mt
);
1342 rb
->mt
= intel_miptree_create_for_image_buffer(intel
,
1344 intel_rb_format(rb
),
1350 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1352 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1353 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1354 struct intel_renderbuffer
*front_rb
;
1355 struct intel_renderbuffer
*back_rb
;
1356 struct __DRIimageList images
;
1357 unsigned int format
;
1358 uint32_t buffer_mask
= 0;
1360 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1361 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1364 format
= intel_rb_format(back_rb
);
1366 format
= intel_rb_format(front_rb
);
1370 if ((brw
->is_front_buffer_rendering
|| brw
->is_front_buffer_reading
|| !back_rb
) && front_rb
)
1371 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1374 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1376 (*screen
->image
.loader
->getBuffers
) (drawable
,
1377 driGLFormatToImageFormat(format
),
1378 &drawable
->dri2
.stamp
,
1379 drawable
->loaderPrivate
,
1383 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1384 drawable
->w
= images
.front
->width
;
1385 drawable
->h
= images
.front
->height
;
1386 intel_update_image_buffer(brw
,
1390 __DRI_IMAGE_BUFFER_FRONT
);
1392 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1393 drawable
->w
= images
.back
->width
;
1394 drawable
->h
= images
.back
->height
;
1395 intel_update_image_buffer(brw
,
1399 __DRI_IMAGE_BUFFER_BACK
);