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 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
282 ctx
->Const
.StripTextureBorder
= true;
284 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
285 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
286 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
287 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
288 ctx
->Const
.MaxTextureUnits
=
289 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
290 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
291 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
293 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
295 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
296 ctx
->Const
.MaxCombinedTextureImageUnits
=
297 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
298 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
299 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
;
301 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
302 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
303 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
304 ctx
->Const
.Max3DTextureLevels
= 9;
305 ctx
->Const
.MaxCubeTextureLevels
= 12;
308 ctx
->Const
.MaxArrayTextureLayers
= 2048;
310 ctx
->Const
.MaxArrayTextureLayers
= 512;
312 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
314 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
316 ctx
->Const
.MaxRenderbufferSize
= 8192;
318 /* Hardware only supports a limited number of transform feedback buffers.
319 * So we need to override the Mesa default (which is based only on software
322 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
324 /* On Gen6, in the worst case, we use up one binding table entry per
325 * transform feedback component (see comments above the definition of
326 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
327 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
328 * BRW_MAX_SOL_BINDINGS.
330 * In "separate components" mode, we need to divide this value by
331 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
332 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
334 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
335 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
336 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
338 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
341 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
342 const int clamp_max_samples
=
343 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
345 if (clamp_max_samples
< 0) {
346 max_samples
= msaa_modes
[0];
348 /* Select the largest supported MSAA mode that does not exceed
352 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
353 if (msaa_modes
[i
] <= clamp_max_samples
) {
354 max_samples
= msaa_modes
[i
];
360 ctx
->Const
.MaxSamples
= max_samples
;
361 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
362 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
363 ctx
->Const
.MaxIntegerSamples
= max_samples
;
366 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
368 ctx
->Const
.MinLineWidth
= 1.0;
369 ctx
->Const
.MinLineWidthAA
= 1.0;
370 ctx
->Const
.MaxLineWidth
= 5.0;
371 ctx
->Const
.MaxLineWidthAA
= 5.0;
372 ctx
->Const
.LineWidthGranularity
= 0.5;
374 ctx
->Const
.MinPointSize
= 1.0;
375 ctx
->Const
.MinPointSizeAA
= 1.0;
376 ctx
->Const
.MaxPointSize
= 255.0;
377 ctx
->Const
.MaxPointSizeAA
= 255.0;
378 ctx
->Const
.PointSizeGranularity
= 1.0;
380 if (brw
->gen
>= 5 || brw
->is_g4x
)
381 ctx
->Const
.MaxClipPlanes
= 8;
383 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
384 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
385 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
386 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
387 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
388 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
389 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
390 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
391 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
392 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
393 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
394 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
395 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
396 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
398 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
399 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
400 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
401 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
402 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
403 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
404 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
405 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
406 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
407 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
408 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
410 /* Fragment shaders use real, 32-bit twos-complement integers for all
413 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
414 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
415 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
416 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
417 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
420 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
421 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
422 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
423 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
424 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
425 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
426 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
429 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
430 * but we're not sure how it's actually done for vertex order,
431 * that affect provoking vertex decision. Always use last vertex
432 * convention for quad primitive which works as expected for now.
435 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
437 ctx
->Const
.NativeIntegers
= true;
438 ctx
->Const
.UniformBooleanTrue
= 1;
440 /* From the gen4 PRM, volume 4 page 127:
442 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
443 * the base address of the first element of the surface, computed in
444 * software by adding the surface base address to the byte offset of
445 * the element in the buffer."
447 * However, unaligned accesses are slower, so enforce buffer alignment.
449 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
450 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
453 ctx
->Const
.MaxVarying
= 32;
454 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
455 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
456 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
457 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
460 /* We want the GLSL compiler to emit code that uses condition codes */
461 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
462 ctx
->ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
463 ctx
->ShaderCompilerOptions
[i
].EmitCondCodes
= true;
464 ctx
->ShaderCompilerOptions
[i
].EmitNoNoise
= true;
465 ctx
->ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
466 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
467 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectOutput
= true;
469 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectUniform
=
470 (i
== MESA_SHADER_FRAGMENT
);
471 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
472 (i
== MESA_SHADER_FRAGMENT
);
473 ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
= true;
476 ctx
->ShaderCompilerOptions
[MESA_SHADER_VERTEX
].PreferDP4
= true;
480 * Process driconf (drirc) options, setting appropriate context flags.
482 * intelInitExtensions still pokes at optionCache directly, in order to
483 * avoid advertising various extensions. No flags are set, so it makes
484 * sense to continue doing that there.
487 brw_process_driconf_options(struct brw_context
*brw
)
489 struct gl_context
*ctx
= &brw
->ctx
;
491 driOptionCache
*options
= &brw
->optionCache
;
492 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
493 brw
->driContext
->driScreenPriv
->myNum
, "i965");
495 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
496 switch (bo_reuse_mode
) {
497 case DRI_CONF_BO_REUSE_DISABLED
:
499 case DRI_CONF_BO_REUSE_ALL
:
500 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
504 if (!driQueryOptionb(options
, "hiz")) {
505 brw
->has_hiz
= false;
506 /* On gen6, you can only do separate stencil with HIZ. */
508 brw
->has_separate_stencil
= false;
511 if (driQueryOptionb(options
, "always_flush_batch")) {
512 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
513 brw
->always_flush_batch
= true;
516 if (driQueryOptionb(options
, "always_flush_cache")) {
517 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
518 brw
->always_flush_cache
= true;
521 if (driQueryOptionb(options
, "disable_throttling")) {
522 fprintf(stderr
, "disabling flush throttling\n");
523 brw
->disable_throttling
= true;
526 brw
->disable_derivative_optimization
=
527 driQueryOptionb(&brw
->optionCache
, "disable_derivative_optimization");
529 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
531 ctx
->Const
.ForceGLSLExtensionsWarn
=
532 driQueryOptionb(options
, "force_glsl_extensions_warn");
534 ctx
->Const
.DisableGLSLLineContinuations
=
535 driQueryOptionb(options
, "disable_glsl_line_continuations");
539 brwCreateContext(gl_api api
,
540 const struct gl_config
*mesaVis
,
541 __DRIcontext
*driContextPriv
,
542 unsigned major_version
,
543 unsigned minor_version
,
546 unsigned *dri_ctx_error
,
547 void *sharedContextPrivate
)
549 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
550 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
551 struct intel_screen
*screen
= sPriv
->driverPrivate
;
552 const struct brw_device_info
*devinfo
= screen
->devinfo
;
553 struct dd_function_table functions
;
554 struct gl_config visual
;
556 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
557 * provides us with context reset notifications.
559 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
560 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
562 if (screen
->has_context_reset_notification
)
563 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
565 if (flags
& ~allowed_flags
) {
566 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
570 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
572 printf("%s: failed to alloc context\n", __FUNCTION__
);
573 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
577 driContextPriv
->driverPrivate
= brw
;
578 brw
->driContext
= driContextPriv
;
579 brw
->intelScreen
= screen
;
580 brw
->bufmgr
= screen
->bufmgr
;
582 brw
->gen
= devinfo
->gen
;
583 brw
->gt
= devinfo
->gt
;
584 brw
->is_g4x
= devinfo
->is_g4x
;
585 brw
->is_baytrail
= devinfo
->is_baytrail
;
586 brw
->is_haswell
= devinfo
->is_haswell
;
587 brw
->has_llc
= devinfo
->has_llc
;
588 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
&& brw
->gen
< 8;
589 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
590 brw
->has_pln
= devinfo
->has_pln
;
591 brw
->has_compr4
= devinfo
->has_compr4
;
592 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
593 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
594 brw
->needs_unlit_centroid_workaround
=
595 devinfo
->needs_unlit_centroid_workaround
;
597 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
598 brw
->has_swizzling
= screen
->hw_has_swizzling
;
601 gen7_init_vtable_surface_functions(brw
);
602 gen7_init_vtable_sampler_functions(brw
);
603 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
605 gen4_init_vtable_surface_functions(brw
);
606 gen4_init_vtable_sampler_functions(brw
);
607 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
610 brw_init_driver_functions(brw
, &functions
);
613 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
615 struct gl_context
*ctx
= &brw
->ctx
;
617 if (mesaVis
== NULL
) {
618 memset(&visual
, 0, sizeof visual
);
622 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
623 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
624 printf("%s: failed to init mesa context\n", __FUNCTION__
);
625 intelDestroyContext(driContextPriv
);
629 driContextSetFlags(ctx
, flags
);
631 /* Initialize the software rasterizer and helper modules.
633 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
634 * software fallbacks (which we have to support on legacy GL to do weird
635 * glDrawPixels(), glBitmap(), and other functions).
637 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
638 _swrast_CreateContext(ctx
);
641 _vbo_CreateContext(ctx
);
642 if (ctx
->swrast_context
) {
643 _tnl_CreateContext(ctx
);
644 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
645 _swsetup_CreateContext(ctx
);
647 /* Configure swrast to match hardware characteristics: */
648 _swrast_allow_pixel_fog(ctx
, false);
649 _swrast_allow_vertex_fog(ctx
, true);
652 _mesa_meta_init(ctx
);
654 brw_process_driconf_options(brw
);
655 brw_process_intel_debug_variable(brw
);
656 brw_initialize_context_constants(brw
);
658 ctx
->Const
.ResetStrategy
= notify_reset
659 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
661 /* Reinitialize the context point state. It depends on ctx->Const values. */
662 _mesa_init_point(ctx
);
664 intel_batchbuffer_init(brw
);
668 intelInitExtensions(ctx
);
673 /* Create a new hardware context. Using a hardware context means that
674 * our GPU state will be saved/restored on context switch, allowing us
675 * to assume that the GPU is in the same state we left it in.
677 * This is required for transform feedback buffer offsets, query objects,
678 * and also allows us to reduce how much state we have to emit.
680 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
683 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
684 intelDestroyContext(driContextPriv
);
689 brw_init_surface_formats(brw
);
691 if (brw
->is_g4x
|| brw
->gen
>= 5) {
692 brw
->CMD_VF_STATISTICS
= GM45_3DSTATE_VF_STATISTICS
;
693 brw
->CMD_PIPELINE_SELECT
= CMD_PIPELINE_SELECT_GM45
;
695 brw
->CMD_VF_STATISTICS
= GEN4_3DSTATE_VF_STATISTICS
;
696 brw
->CMD_PIPELINE_SELECT
= CMD_PIPELINE_SELECT_965
;
699 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
700 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
701 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
702 brw
->urb
.size
= devinfo
->urb
.size
;
703 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
704 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
705 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
707 /* Estimate the size of the mappable aperture into the GTT. There's an
708 * ioctl to get the whole GTT size, but not one to get the mappable subset.
709 * It turns out it's basically always 256MB, though some ancient hardware
712 uint32_t gtt_size
= 256 * 1024 * 1024;
714 /* We don't want to map two objects such that a memcpy between them would
715 * just fault one mapping in and then the other over and over forever. So
716 * we would need to divide the GTT size by 2. Additionally, some GTT is
717 * taken up by things like the framebuffer and the ringbuffer and such, so
718 * be more conservative.
720 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
723 brw
->urb
.gen6_gs_previously_active
= false;
725 brw
->prim_restart
.in_progress
= false;
726 brw
->prim_restart
.enable_cut_index
= false;
727 brw
->gs
.enabled
= false;
730 brw
->curbe
.last_buf
= calloc(1, 4096);
731 brw
->curbe
.next_buf
= calloc(1, 4096);
734 ctx
->VertexProgram
._MaintainTnlProgram
= true;
735 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
737 brw_draw_init( brw
);
739 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
740 /* Turn on some extra GL_ARB_debug_output generation. */
741 brw
->perf_debug
= true;
744 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
745 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
747 brw_fs_alloc_reg_sets(brw
);
748 brw_vec4_alloc_reg_set(brw
);
750 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
751 brw_init_shader_time(brw
);
753 _mesa_compute_version(ctx
);
755 _mesa_initialize_dispatch_tables(ctx
);
756 _mesa_initialize_vbo_vtxfmt(ctx
);
758 if (ctx
->Extensions
.AMD_performance_monitor
) {
759 brw_init_performance_monitors(brw
);
766 intelDestroyContext(__DRIcontext
* driContextPriv
)
768 struct brw_context
*brw
=
769 (struct brw_context
*) driContextPriv
->driverPrivate
;
770 struct gl_context
*ctx
= &brw
->ctx
;
772 assert(brw
); /* should never be null */
776 /* Dump a final BMP in case the application doesn't call SwapBuffers */
777 if (INTEL_DEBUG
& DEBUG_AUB
) {
778 intel_batchbuffer_flush(brw
);
779 aub_dump_bmp(&brw
->ctx
);
782 _mesa_meta_free(&brw
->ctx
);
784 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
785 /* Force a report. */
786 brw
->shader_time
.report_time
= 0;
788 brw_collect_and_report_shader_time(brw
);
789 brw_destroy_shader_time(brw
);
792 brw_destroy_state(brw
);
793 brw_draw_destroy(brw
);
795 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
796 drm_intel_bo_unreference(brw
->vs
.base
.const_bo
);
797 drm_intel_bo_unreference(brw
->wm
.base
.const_bo
);
799 free(brw
->curbe
.last_buf
);
800 free(brw
->curbe
.next_buf
);
802 drm_intel_gem_context_destroy(brw
->hw_ctx
);
804 if (ctx
->swrast_context
) {
805 _swsetup_DestroyContext(&brw
->ctx
);
806 _tnl_DestroyContext(&brw
->ctx
);
808 _vbo_DestroyContext(&brw
->ctx
);
810 if (ctx
->swrast_context
)
811 _swrast_DestroyContext(&brw
->ctx
);
813 intel_batchbuffer_free(brw
);
815 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
816 brw
->first_post_swapbuffers_batch
= NULL
;
818 driDestroyOptionCache(&brw
->optionCache
);
820 /* free the Mesa context */
821 _mesa_free_context_data(&brw
->ctx
);
824 driContextPriv
->driverPrivate
= NULL
;
828 intelUnbindContext(__DRIcontext
* driContextPriv
)
830 /* Unset current context and dispath table */
831 _mesa_make_current(NULL
, NULL
, NULL
);
837 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
838 * on window system framebuffers.
840 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
841 * your renderbuffer can do sRGB encode, and you can flip a switch that does
842 * sRGB encode if the renderbuffer can handle it. You can ask specifically
843 * for a visual where you're guaranteed to be capable, but it turns out that
844 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
845 * incapable ones, becuase there's no difference between the two in resources
846 * used. Applications thus get built that accidentally rely on the default
847 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
850 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
851 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
852 * So they removed the enable knob and made it "if the renderbuffer is sRGB
853 * capable, do sRGB encode". Then, for your window system renderbuffers, you
854 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
855 * and get no sRGB encode (assuming that both kinds of visual are available).
856 * Thus our choice to support sRGB by default on our visuals for desktop would
857 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
859 * Unfortunately, renderbuffer setup happens before a context is created. So
860 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
861 * context (without an sRGB visual, though we don't have sRGB visuals exposed
862 * yet), we go turn that back off before anyone finds out.
865 intel_gles3_srgb_workaround(struct brw_context
*brw
,
866 struct gl_framebuffer
*fb
)
868 struct gl_context
*ctx
= &brw
->ctx
;
870 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
873 /* Some day when we support the sRGB capable bit on visuals available for
874 * GLES, we'll need to respect that and not disable things here.
876 fb
->Visual
.sRGBCapable
= false;
877 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
878 if (fb
->Attachment
[i
].Renderbuffer
&&
879 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_SARGB8
) {
880 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_ARGB8888
;
886 intelMakeCurrent(__DRIcontext
* driContextPriv
,
887 __DRIdrawable
* driDrawPriv
,
888 __DRIdrawable
* driReadPriv
)
890 struct brw_context
*brw
;
891 GET_CURRENT_CONTEXT(curCtx
);
894 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
898 /* According to the glXMakeCurrent() man page: "Pending commands to
899 * the previous context, if any, are flushed before it is released."
900 * But only flush if we're actually changing contexts.
902 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
906 if (driContextPriv
) {
907 struct gl_context
*ctx
= &brw
->ctx
;
908 struct gl_framebuffer
*fb
, *readFb
;
910 if (driDrawPriv
== NULL
&& driReadPriv
== NULL
) {
911 fb
= _mesa_get_incomplete_framebuffer();
912 readFb
= _mesa_get_incomplete_framebuffer();
914 fb
= driDrawPriv
->driverPrivate
;
915 readFb
= driReadPriv
->driverPrivate
;
916 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
917 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
920 /* The sRGB workaround changes the renderbuffer's format. We must change
921 * the format before the renderbuffer's miptree get's allocated, otherwise
922 * the formats of the renderbuffer and its miptree will differ.
924 intel_gles3_srgb_workaround(brw
, fb
);
925 intel_gles3_srgb_workaround(brw
, readFb
);
927 intel_prepare_render(brw
);
928 _mesa_make_current(ctx
, fb
, readFb
);
930 _mesa_make_current(NULL
, NULL
, NULL
);
937 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
938 __DRIdrawable
*drawable
)
941 /* MSAA and fast color clear are not supported, so don't waste time
942 * checking whether a resolve is needed.
947 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
948 struct intel_renderbuffer
*rb
;
950 /* Usually, only the back buffer will need to be downsampled. However,
951 * the front buffer will also need it if the user has rendered into it.
953 static const gl_buffer_index buffers
[2] = {
958 for (int i
= 0; i
< 2; ++i
) {
959 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
960 if (rb
== NULL
|| rb
->mt
== NULL
)
962 if (rb
->mt
->num_samples
<= 1)
963 intel_miptree_resolve_color(brw
, rb
->mt
);
965 intel_miptree_downsample(brw
, rb
->mt
);
970 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
972 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
976 intel_query_dri2_buffers(struct brw_context
*brw
,
977 __DRIdrawable
*drawable
,
978 __DRIbuffer
**buffers
,
982 intel_process_dri2_buffer(struct brw_context
*brw
,
983 __DRIdrawable
*drawable
,
985 struct intel_renderbuffer
*rb
,
986 const char *buffer_name
);
989 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
992 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
994 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
995 struct intel_renderbuffer
*rb
;
996 __DRIbuffer
*buffers
= NULL
;
998 const char *region_name
;
1000 /* Set this up front, so that in case our buffers get invalidated
1001 * while we're getting new buffers, we don't clobber the stamp and
1002 * thus ignore the invalidate. */
1003 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1005 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1006 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1008 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1010 if (buffers
== NULL
)
1013 for (i
= 0; i
< count
; i
++) {
1014 switch (buffers
[i
].attachment
) {
1015 case __DRI_BUFFER_FRONT_LEFT
:
1016 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1017 region_name
= "dri2 front buffer";
1020 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1021 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1022 region_name
= "dri2 fake front buffer";
1025 case __DRI_BUFFER_BACK_LEFT
:
1026 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1027 region_name
= "dri2 back buffer";
1030 case __DRI_BUFFER_DEPTH
:
1031 case __DRI_BUFFER_HIZ
:
1032 case __DRI_BUFFER_DEPTH_STENCIL
:
1033 case __DRI_BUFFER_STENCIL
:
1034 case __DRI_BUFFER_ACCUM
:
1037 "unhandled buffer attach event, attachment type %d\n",
1038 buffers
[i
].attachment
);
1042 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1048 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1050 struct brw_context
*brw
= context
->driverPrivate
;
1051 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1053 /* Set this up front, so that in case our buffers get invalidated
1054 * while we're getting new buffers, we don't clobber the stamp and
1055 * thus ignore the invalidate. */
1056 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1058 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1059 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1061 if (screen
->image
.loader
)
1062 intel_update_image_buffers(brw
, drawable
);
1064 intel_update_dri2_buffers(brw
, drawable
);
1066 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1070 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1071 * state is required.
1074 intel_prepare_render(struct brw_context
*brw
)
1076 __DRIcontext
*driContext
= brw
->driContext
;
1077 __DRIdrawable
*drawable
;
1079 drawable
= driContext
->driDrawablePriv
;
1080 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1081 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1082 intel_update_renderbuffers(driContext
, drawable
);
1083 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1086 drawable
= driContext
->driReadablePriv
;
1087 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1088 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1089 intel_update_renderbuffers(driContext
, drawable
);
1090 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1093 /* If we're currently rendering to the front buffer, the rendering
1094 * that will happen next will probably dirty the front buffer. So
1095 * mark it as dirty here.
1097 if (brw
->is_front_buffer_rendering
)
1098 brw
->front_buffer_dirty
= true;
1100 /* Wait for the swapbuffers before the one we just emitted, so we
1101 * don't get too many swaps outstanding for apps that are GPU-heavy
1102 * but not CPU-heavy.
1104 * We're using intelDRI2Flush (called from the loader before
1105 * swapbuffer) and glFlush (for front buffer rendering) as the
1106 * indicator that a frame is done and then throttle when we get
1107 * here as we prepare to render the next frame. At this point for
1108 * round trips for swap/copy and getting new buffers are done and
1109 * we'll spend less time waiting on the GPU.
1111 * Unfortunately, we don't have a handle to the batch containing
1112 * the swap, and getting our hands on that doesn't seem worth it,
1113 * so we just us the first batch we emitted after the last swap.
1115 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1116 if (!brw
->disable_throttling
)
1117 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1118 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1119 brw
->first_post_swapbuffers_batch
= NULL
;
1120 brw
->need_throttle
= false;
1125 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1127 * To determine which DRI buffers to request, examine the renderbuffers
1128 * attached to the drawable's framebuffer. Then request the buffers with
1129 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1131 * This is called from intel_update_renderbuffers().
1133 * \param drawable Drawable whose buffers are queried.
1134 * \param buffers [out] List of buffers returned by DRI2 query.
1135 * \param buffer_count [out] Number of buffers returned.
1137 * \see intel_update_renderbuffers()
1138 * \see DRI2GetBuffers()
1139 * \see DRI2GetBuffersWithFormat()
1142 intel_query_dri2_buffers(struct brw_context
*brw
,
1143 __DRIdrawable
*drawable
,
1144 __DRIbuffer
**buffers
,
1147 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1148 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1150 unsigned attachments
[8];
1152 struct intel_renderbuffer
*front_rb
;
1153 struct intel_renderbuffer
*back_rb
;
1155 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1156 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1158 memset(attachments
, 0, sizeof(attachments
));
1159 if ((brw
->is_front_buffer_rendering
||
1160 brw
->is_front_buffer_reading
||
1161 !back_rb
) && front_rb
) {
1162 /* If a fake front buffer is in use, then querying for
1163 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1164 * the real front buffer to the fake front buffer. So before doing the
1165 * query, we need to make sure all the pending drawing has landed in the
1166 * real front buffer.
1168 intel_batchbuffer_flush(brw
);
1169 intel_flush_front(&brw
->ctx
);
1171 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1172 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1173 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1174 /* We have pending front buffer rendering, but we aren't querying for a
1175 * front buffer. If the front buffer we have is a fake front buffer,
1176 * the X server is going to throw it away when it processes the query.
1177 * So before doing the query, make sure all the pending drawing has
1178 * landed in the real front buffer.
1180 intel_batchbuffer_flush(brw
);
1181 intel_flush_front(&brw
->ctx
);
1185 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1186 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1189 assert(i
<= ARRAY_SIZE(attachments
));
1191 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1196 drawable
->loaderPrivate
);
1200 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1202 * This is called from intel_update_renderbuffers().
1205 * DRI buffers whose attachment point is DRI2BufferStencil or
1206 * DRI2BufferDepthStencil are handled as special cases.
1208 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1209 * that is passed to intel_region_alloc_for_handle().
1211 * \see intel_update_renderbuffers()
1212 * \see intel_region_alloc_for_handle()
1215 intel_process_dri2_buffer(struct brw_context
*brw
,
1216 __DRIdrawable
*drawable
,
1217 __DRIbuffer
*buffer
,
1218 struct intel_renderbuffer
*rb
,
1219 const char *buffer_name
)
1221 struct intel_region
*region
= NULL
;
1226 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1228 /* We try to avoid closing and reopening the same BO name, because the first
1229 * use of a mapping of the buffer involves a bunch of page faulting which is
1230 * moderately expensive.
1232 if (num_samples
== 0) {
1235 rb
->mt
->region
->name
== buffer
->name
)
1239 rb
->mt
->singlesample_mt
&&
1240 rb
->mt
->singlesample_mt
->region
&&
1241 rb
->mt
->singlesample_mt
->region
->name
== buffer
->name
)
1245 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1247 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1248 buffer
->name
, buffer
->attachment
,
1249 buffer
->cpp
, buffer
->pitch
);
1252 intel_miptree_release(&rb
->mt
);
1253 region
= intel_region_alloc_for_handle(brw
->intelScreen
,
1263 rb
->mt
= intel_miptree_create_for_dri2_buffer(brw
,
1265 intel_rb_format(rb
),
1268 intel_region_release(®ion
);
1272 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1274 * To determine which DRI buffers to request, examine the renderbuffers
1275 * attached to the drawable's framebuffer. Then request the buffers from
1278 * This is called from intel_update_renderbuffers().
1280 * \param drawable Drawable whose buffers are queried.
1281 * \param buffers [out] List of buffers returned by DRI2 query.
1282 * \param buffer_count [out] Number of buffers returned.
1284 * \see intel_update_renderbuffers()
1288 intel_update_image_buffer(struct brw_context
*intel
,
1289 __DRIdrawable
*drawable
,
1290 struct intel_renderbuffer
*rb
,
1292 enum __DRIimageBufferMask buffer_type
)
1294 struct intel_region
*region
= buffer
->region
;
1299 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1301 /* Check and see if we're already bound to the right
1304 if (num_samples
== 0) {
1307 rb
->mt
->region
->bo
== region
->bo
)
1311 rb
->mt
->singlesample_mt
&&
1312 rb
->mt
->singlesample_mt
->region
&&
1313 rb
->mt
->singlesample_mt
->region
->bo
== region
->bo
)
1317 intel_miptree_release(&rb
->mt
);
1318 rb
->mt
= intel_miptree_create_for_image_buffer(intel
,
1320 intel_rb_format(rb
),
1326 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1328 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1329 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1330 struct intel_renderbuffer
*front_rb
;
1331 struct intel_renderbuffer
*back_rb
;
1332 struct __DRIimageList images
;
1333 unsigned int format
;
1334 uint32_t buffer_mask
= 0;
1336 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1337 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1340 format
= intel_rb_format(back_rb
);
1342 format
= intel_rb_format(front_rb
);
1346 if ((brw
->is_front_buffer_rendering
|| brw
->is_front_buffer_reading
|| !back_rb
) && front_rb
)
1347 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1350 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1352 (*screen
->image
.loader
->getBuffers
) (drawable
,
1353 driGLFormatToImageFormat(format
),
1354 &drawable
->dri2
.stamp
,
1355 drawable
->loaderPrivate
,
1359 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1360 drawable
->w
= images
.front
->width
;
1361 drawable
->h
= images
.front
->height
;
1362 intel_update_image_buffer(brw
,
1366 __DRI_IMAGE_BUFFER_FRONT
);
1368 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1369 drawable
->w
= images
.back
->width
;
1370 drawable
->h
= images
.back
->height
;
1371 intel_update_image_buffer(brw
,
1375 __DRI_IMAGE_BUFFER_BACK
);