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_image.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
);
103 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
106 brw_get_renderer_string(unsigned deviceID
)
109 static char buffer
[128];
113 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
114 #include "pci_ids/i965_pci_ids.h"
116 chipset
= "Unknown Intel Chipset";
120 (void) driGetRendererString(buffer
, chipset
, 0);
124 static const GLubyte
*
125 intelGetString(struct gl_context
* ctx
, GLenum name
)
127 const struct brw_context
*const brw
= brw_context(ctx
);
131 return (GLubyte
*) brw_vendor_string
;
135 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
143 intel_viewport(struct gl_context
*ctx
)
145 struct brw_context
*brw
= brw_context(ctx
);
146 __DRIcontext
*driContext
= brw
->driContext
;
148 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
149 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
150 dri2InvalidateDrawable(driContext
->driReadablePriv
);
155 intelInvalidateState(struct gl_context
* ctx
, GLuint new_state
)
157 struct brw_context
*brw
= brw_context(ctx
);
159 if (ctx
->swrast_context
)
160 _swrast_InvalidateState(ctx
, new_state
);
161 _vbo_InvalidateState(ctx
, new_state
);
163 brw
->NewGLState
|= new_state
;
166 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
169 intel_flush_front(struct gl_context
*ctx
)
171 struct brw_context
*brw
= brw_context(ctx
);
172 __DRIcontext
*driContext
= brw
->driContext
;
173 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
174 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
176 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
177 if (flushFront(screen
) && driDrawable
&&
178 driDrawable
->loaderPrivate
) {
180 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
182 * This potentially resolves both front and back buffer. It
183 * is unnecessary to resolve the back, but harms nothing except
184 * performance. And no one cares about front-buffer render
187 intel_resolve_for_dri2_flush(brw
, driDrawable
);
188 intel_batchbuffer_flush(brw
);
190 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
192 /* We set the dirty bit in intel_prepare_render() if we're
193 * front buffer rendering once we get there.
195 brw
->front_buffer_dirty
= false;
201 intel_glFlush(struct gl_context
*ctx
)
203 struct brw_context
*brw
= brw_context(ctx
);
205 intel_batchbuffer_flush(brw
);
206 intel_flush_front(ctx
);
207 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
208 brw
->need_throttle
= true;
212 intelFinish(struct gl_context
* ctx
)
214 struct brw_context
*brw
= brw_context(ctx
);
218 if (brw
->batch
.last_bo
)
219 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
223 brw_init_driver_functions(struct brw_context
*brw
,
224 struct dd_function_table
*functions
)
226 _mesa_init_driver_functions(functions
);
228 /* GLX uses DRI2 invalidate events to handle window resizing.
229 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
230 * which doesn't provide a mechanism for snooping the event queues.
232 * So EGL still relies on viewport hacks to handle window resizing.
233 * This should go away with DRI3000.
235 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
236 functions
->Viewport
= intel_viewport
;
238 functions
->Flush
= intel_glFlush
;
239 functions
->Finish
= intelFinish
;
240 functions
->GetString
= intelGetString
;
241 functions
->UpdateState
= intelInvalidateState
;
243 intelInitTextureFuncs(functions
);
244 intelInitTextureImageFuncs(functions
);
245 intelInitTextureSubImageFuncs(functions
);
246 intelInitTextureCopyImageFuncs(functions
);
247 intelInitClearFuncs(functions
);
248 intelInitBufferFuncs(functions
);
249 intelInitPixelFuncs(functions
);
250 intelInitBufferObjectFuncs(functions
);
251 intel_init_syncobj_functions(functions
);
252 brw_init_object_purgeable_functions(functions
);
254 brwInitFragProgFuncs( functions
);
255 brw_init_common_queryobj_functions(functions
);
257 gen6_init_queryobj_functions(functions
);
259 gen4_init_queryobj_functions(functions
);
261 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
263 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
264 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
265 functions
->GetTransformFeedbackVertexCount
=
266 brw_get_transform_feedback_vertex_count
;
268 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
269 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
270 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
271 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
273 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
274 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
278 functions
->GetSamplePosition
= gen6_get_sample_position
;
282 brw_initialize_context_constants(struct brw_context
*brw
)
284 struct gl_context
*ctx
= &brw
->ctx
;
286 unsigned max_samplers
=
287 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
289 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
291 ctx
->Const
.StripTextureBorder
= true;
293 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
294 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
295 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
296 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
297 ctx
->Const
.MaxTextureUnits
=
298 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
299 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
300 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
302 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
304 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
305 if (getenv("INTEL_COMPUTE_SHADER")) {
306 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
307 ctx
->Const
.MaxUniformBufferBindings
+= 12;
309 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
311 ctx
->Const
.MaxCombinedTextureImageUnits
=
312 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
313 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
314 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
315 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
317 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
318 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
319 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
320 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
321 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
322 ctx
->Const
.MaxTextureMbytes
= 1536;
325 ctx
->Const
.MaxArrayTextureLayers
= 2048;
327 ctx
->Const
.MaxArrayTextureLayers
= 512;
329 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
331 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
333 ctx
->Const
.MaxRenderbufferSize
= 8192;
335 /* Hardware only supports a limited number of transform feedback buffers.
336 * So we need to override the Mesa default (which is based only on software
339 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
341 /* On Gen6, in the worst case, we use up one binding table entry per
342 * transform feedback component (see comments above the definition of
343 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
344 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
345 * BRW_MAX_SOL_BINDINGS.
347 * In "separate components" mode, we need to divide this value by
348 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
349 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
351 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
352 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
353 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
355 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
358 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
359 const int clamp_max_samples
=
360 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
362 if (clamp_max_samples
< 0) {
363 max_samples
= msaa_modes
[0];
365 /* Select the largest supported MSAA mode that does not exceed
369 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
370 if (msaa_modes
[i
] <= clamp_max_samples
) {
371 max_samples
= msaa_modes
[i
];
377 ctx
->Const
.MaxSamples
= max_samples
;
378 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
379 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
380 ctx
->Const
.MaxIntegerSamples
= max_samples
;
383 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
384 else if (brw
->gen
== 6)
385 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
387 ctx
->Const
.MinLineWidth
= 1.0;
388 ctx
->Const
.MinLineWidthAA
= 1.0;
389 ctx
->Const
.MaxLineWidth
= 5.0;
390 ctx
->Const
.MaxLineWidthAA
= 5.0;
391 ctx
->Const
.LineWidthGranularity
= 0.5;
393 ctx
->Const
.MinPointSize
= 1.0;
394 ctx
->Const
.MinPointSizeAA
= 1.0;
395 ctx
->Const
.MaxPointSize
= 255.0;
396 ctx
->Const
.MaxPointSizeAA
= 255.0;
397 ctx
->Const
.PointSizeGranularity
= 1.0;
399 if (brw
->gen
>= 5 || brw
->is_g4x
)
400 ctx
->Const
.MaxClipPlanes
= 8;
402 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
403 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
404 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
405 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
406 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
407 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
408 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
409 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
410 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
411 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
412 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
413 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
414 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
415 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
417 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
418 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
419 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
420 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
421 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
422 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
423 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
424 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
425 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
426 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
427 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
429 /* Fragment shaders use real, 32-bit twos-complement integers for all
432 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
433 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
434 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
435 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
436 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
439 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
440 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
441 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
442 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
443 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
444 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
445 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
446 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
447 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
450 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
451 * but we're not sure how it's actually done for vertex order,
452 * that affect provoking vertex decision. Always use last vertex
453 * convention for quad primitive which works as expected for now.
456 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
458 ctx
->Const
.NativeIntegers
= true;
459 ctx
->Const
.UniformBooleanTrue
= 1;
461 /* From the gen4 PRM, volume 4 page 127:
463 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
464 * the base address of the first element of the surface, computed in
465 * software by adding the surface base address to the byte offset of
466 * the element in the buffer."
468 * However, unaligned accesses are slower, so enforce buffer alignment.
470 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
471 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
474 ctx
->Const
.MaxVarying
= 32;
475 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
476 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
477 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
478 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
481 /* We want the GLSL compiler to emit code that uses condition codes */
482 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
483 ctx
->ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
484 ctx
->ShaderCompilerOptions
[i
].EmitCondCodes
= true;
485 ctx
->ShaderCompilerOptions
[i
].EmitNoNoise
= true;
486 ctx
->ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
487 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
488 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
489 (i
== MESA_SHADER_FRAGMENT
);
490 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
491 (i
== MESA_SHADER_FRAGMENT
);
492 ctx
->ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
493 ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
= true;
496 ctx
->ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
497 ctx
->ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
499 /* ARB_viewport_array */
500 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
501 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
502 ctx
->Const
.ViewportSubpixelBits
= 0;
504 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
506 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
507 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
512 * Process driconf (drirc) options, setting appropriate context flags.
514 * intelInitExtensions still pokes at optionCache directly, in order to
515 * avoid advertising various extensions. No flags are set, so it makes
516 * sense to continue doing that there.
519 brw_process_driconf_options(struct brw_context
*brw
)
521 struct gl_context
*ctx
= &brw
->ctx
;
523 driOptionCache
*options
= &brw
->optionCache
;
524 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
525 brw
->driContext
->driScreenPriv
->myNum
, "i965");
527 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
528 switch (bo_reuse_mode
) {
529 case DRI_CONF_BO_REUSE_DISABLED
:
531 case DRI_CONF_BO_REUSE_ALL
:
532 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
536 if (!driQueryOptionb(options
, "hiz")) {
537 brw
->has_hiz
= false;
538 /* On gen6, you can only do separate stencil with HIZ. */
540 brw
->has_separate_stencil
= false;
543 if (driQueryOptionb(options
, "always_flush_batch")) {
544 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
545 brw
->always_flush_batch
= true;
548 if (driQueryOptionb(options
, "always_flush_cache")) {
549 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
550 brw
->always_flush_cache
= true;
553 if (driQueryOptionb(options
, "disable_throttling")) {
554 fprintf(stderr
, "disabling flush throttling\n");
555 brw
->disable_throttling
= true;
558 brw
->disable_derivative_optimization
=
559 driQueryOptionb(&brw
->optionCache
, "disable_derivative_optimization");
561 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
563 ctx
->Const
.ForceGLSLExtensionsWarn
=
564 driQueryOptionb(options
, "force_glsl_extensions_warn");
566 ctx
->Const
.DisableGLSLLineContinuations
=
567 driQueryOptionb(options
, "disable_glsl_line_continuations");
571 brwCreateContext(gl_api api
,
572 const struct gl_config
*mesaVis
,
573 __DRIcontext
*driContextPriv
,
574 unsigned major_version
,
575 unsigned minor_version
,
578 unsigned *dri_ctx_error
,
579 void *sharedContextPrivate
)
581 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
582 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
583 struct intel_screen
*screen
= sPriv
->driverPrivate
;
584 const struct brw_device_info
*devinfo
= screen
->devinfo
;
585 struct dd_function_table functions
;
587 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
588 * provides us with context reset notifications.
590 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
591 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
593 if (screen
->has_context_reset_notification
)
594 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
596 if (flags
& ~allowed_flags
) {
597 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
601 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
603 fprintf(stderr
, "%s: failed to alloc context\n", __FUNCTION__
);
604 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
608 driContextPriv
->driverPrivate
= brw
;
609 brw
->driContext
= driContextPriv
;
610 brw
->intelScreen
= screen
;
611 brw
->bufmgr
= screen
->bufmgr
;
613 brw
->gen
= devinfo
->gen
;
614 brw
->gt
= devinfo
->gt
;
615 brw
->is_g4x
= devinfo
->is_g4x
;
616 brw
->is_baytrail
= devinfo
->is_baytrail
;
617 brw
->is_haswell
= devinfo
->is_haswell
;
618 brw
->has_llc
= devinfo
->has_llc
;
619 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
620 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
621 brw
->has_pln
= devinfo
->has_pln
;
622 brw
->has_compr4
= devinfo
->has_compr4
;
623 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
624 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
625 brw
->needs_unlit_centroid_workaround
=
626 devinfo
->needs_unlit_centroid_workaround
;
628 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
629 brw
->has_swizzling
= screen
->hw_has_swizzling
;
631 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
632 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
633 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
635 gen8_init_vtable_surface_functions(brw
);
636 gen7_init_vtable_sampler_functions(brw
);
637 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
638 } else if (brw
->gen
>= 7) {
639 gen7_init_vtable_surface_functions(brw
);
640 gen7_init_vtable_sampler_functions(brw
);
641 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
643 gen4_init_vtable_surface_functions(brw
);
644 gen4_init_vtable_sampler_functions(brw
);
645 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
648 brw_init_driver_functions(brw
, &functions
);
651 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
653 struct gl_context
*ctx
= &brw
->ctx
;
655 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
656 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
657 fprintf(stderr
, "%s: failed to init mesa context\n", __FUNCTION__
);
658 intelDestroyContext(driContextPriv
);
662 driContextSetFlags(ctx
, flags
);
664 /* Initialize the software rasterizer and helper modules.
666 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
667 * software fallbacks (which we have to support on legacy GL to do weird
668 * glDrawPixels(), glBitmap(), and other functions).
670 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
671 _swrast_CreateContext(ctx
);
674 _vbo_CreateContext(ctx
);
675 if (ctx
->swrast_context
) {
676 _tnl_CreateContext(ctx
);
677 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
678 _swsetup_CreateContext(ctx
);
680 /* Configure swrast to match hardware characteristics: */
681 _swrast_allow_pixel_fog(ctx
, false);
682 _swrast_allow_vertex_fog(ctx
, true);
685 _mesa_meta_init(ctx
);
687 brw_process_driconf_options(brw
);
688 brw_process_intel_debug_variable(brw
);
689 brw_initialize_context_constants(brw
);
691 ctx
->Const
.ResetStrategy
= notify_reset
692 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
694 /* Reinitialize the context point state. It depends on ctx->Const values. */
695 _mesa_init_point(ctx
);
699 intel_batchbuffer_init(brw
);
702 /* Create a new hardware context. Using a hardware context means that
703 * our GPU state will be saved/restored on context switch, allowing us
704 * to assume that the GPU is in the same state we left it in.
706 * This is required for transform feedback buffer offsets, query objects,
707 * and also allows us to reduce how much state we have to emit.
709 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
712 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
713 intelDestroyContext(driContextPriv
);
720 intelInitExtensions(ctx
);
722 brw_init_surface_formats(brw
);
724 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
725 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
726 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
727 brw
->urb
.size
= devinfo
->urb
.size
;
728 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
729 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
730 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
732 /* Estimate the size of the mappable aperture into the GTT. There's an
733 * ioctl to get the whole GTT size, but not one to get the mappable subset.
734 * It turns out it's basically always 256MB, though some ancient hardware
737 uint32_t gtt_size
= 256 * 1024 * 1024;
739 /* We don't want to map two objects such that a memcpy between them would
740 * just fault one mapping in and then the other over and over forever. So
741 * we would need to divide the GTT size by 2. Additionally, some GTT is
742 * taken up by things like the framebuffer and the ringbuffer and such, so
743 * be more conservative.
745 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
748 brw
->urb
.gen6_gs_previously_active
= false;
750 brw
->prim_restart
.in_progress
= false;
751 brw
->prim_restart
.enable_cut_index
= false;
752 brw
->gs
.enabled
= false;
755 brw
->curbe
.last_buf
= calloc(1, 4096);
756 brw
->curbe
.next_buf
= calloc(1, 4096);
759 ctx
->VertexProgram
._MaintainTnlProgram
= true;
760 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
762 brw_draw_init( brw
);
764 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
765 /* Turn on some extra GL_ARB_debug_output generation. */
766 brw
->perf_debug
= true;
769 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
770 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
772 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
773 brw_init_shader_time(brw
);
775 _mesa_compute_version(ctx
);
777 _mesa_initialize_dispatch_tables(ctx
);
778 _mesa_initialize_vbo_vtxfmt(ctx
);
780 if (ctx
->Extensions
.AMD_performance_monitor
) {
781 brw_init_performance_monitors(brw
);
788 intelDestroyContext(__DRIcontext
* driContextPriv
)
790 struct brw_context
*brw
=
791 (struct brw_context
*) driContextPriv
->driverPrivate
;
792 struct gl_context
*ctx
= &brw
->ctx
;
794 assert(brw
); /* should never be null */
798 /* Dump a final BMP in case the application doesn't call SwapBuffers */
799 if (INTEL_DEBUG
& DEBUG_AUB
) {
800 intel_batchbuffer_flush(brw
);
801 aub_dump_bmp(&brw
->ctx
);
804 _mesa_meta_free(&brw
->ctx
);
806 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
807 /* Force a report. */
808 brw
->shader_time
.report_time
= 0;
810 brw_collect_and_report_shader_time(brw
);
811 brw_destroy_shader_time(brw
);
814 brw_destroy_state(brw
);
815 brw_draw_destroy(brw
);
817 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
819 free(brw
->curbe
.last_buf
);
820 free(brw
->curbe
.next_buf
);
822 drm_intel_gem_context_destroy(brw
->hw_ctx
);
824 if (ctx
->swrast_context
) {
825 _swsetup_DestroyContext(&brw
->ctx
);
826 _tnl_DestroyContext(&brw
->ctx
);
828 _vbo_DestroyContext(&brw
->ctx
);
830 if (ctx
->swrast_context
)
831 _swrast_DestroyContext(&brw
->ctx
);
833 intel_batchbuffer_free(brw
);
835 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
836 brw
->first_post_swapbuffers_batch
= NULL
;
838 driDestroyOptionCache(&brw
->optionCache
);
840 /* free the Mesa context */
841 _mesa_free_context_data(&brw
->ctx
);
844 driContextPriv
->driverPrivate
= NULL
;
848 intelUnbindContext(__DRIcontext
* driContextPriv
)
850 /* Unset current context and dispath table */
851 _mesa_make_current(NULL
, NULL
, NULL
);
857 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
858 * on window system framebuffers.
860 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
861 * your renderbuffer can do sRGB encode, and you can flip a switch that does
862 * sRGB encode if the renderbuffer can handle it. You can ask specifically
863 * for a visual where you're guaranteed to be capable, but it turns out that
864 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
865 * incapable ones, becuase there's no difference between the two in resources
866 * used. Applications thus get built that accidentally rely on the default
867 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
870 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
871 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
872 * So they removed the enable knob and made it "if the renderbuffer is sRGB
873 * capable, do sRGB encode". Then, for your window system renderbuffers, you
874 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
875 * and get no sRGB encode (assuming that both kinds of visual are available).
876 * Thus our choice to support sRGB by default on our visuals for desktop would
877 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
879 * Unfortunately, renderbuffer setup happens before a context is created. So
880 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
881 * context (without an sRGB visual, though we don't have sRGB visuals exposed
882 * yet), we go turn that back off before anyone finds out.
885 intel_gles3_srgb_workaround(struct brw_context
*brw
,
886 struct gl_framebuffer
*fb
)
888 struct gl_context
*ctx
= &brw
->ctx
;
890 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
893 /* Some day when we support the sRGB capable bit on visuals available for
894 * GLES, we'll need to respect that and not disable things here.
896 fb
->Visual
.sRGBCapable
= false;
897 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
898 if (fb
->Attachment
[i
].Renderbuffer
&&
899 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
900 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
906 intelMakeCurrent(__DRIcontext
* driContextPriv
,
907 __DRIdrawable
* driDrawPriv
,
908 __DRIdrawable
* driReadPriv
)
910 struct brw_context
*brw
;
911 GET_CURRENT_CONTEXT(curCtx
);
914 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
918 /* According to the glXMakeCurrent() man page: "Pending commands to
919 * the previous context, if any, are flushed before it is released."
920 * But only flush if we're actually changing contexts.
922 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
926 if (driContextPriv
) {
927 struct gl_context
*ctx
= &brw
->ctx
;
928 struct gl_framebuffer
*fb
, *readFb
;
930 if (driDrawPriv
== NULL
&& driReadPriv
== NULL
) {
931 fb
= _mesa_get_incomplete_framebuffer();
932 readFb
= _mesa_get_incomplete_framebuffer();
934 fb
= driDrawPriv
->driverPrivate
;
935 readFb
= driReadPriv
->driverPrivate
;
936 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
937 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
940 /* The sRGB workaround changes the renderbuffer's format. We must change
941 * the format before the renderbuffer's miptree get's allocated, otherwise
942 * the formats of the renderbuffer and its miptree will differ.
944 intel_gles3_srgb_workaround(brw
, fb
);
945 intel_gles3_srgb_workaround(brw
, readFb
);
947 /* If the context viewport hasn't been initialized, force a call out to
948 * the loader to get buffers so we have a drawable size for the initial
950 if (!brw
->ctx
.ViewportInitialized
)
951 intel_prepare_render(brw
);
953 _mesa_make_current(ctx
, fb
, readFb
);
955 _mesa_make_current(NULL
, NULL
, NULL
);
962 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
963 __DRIdrawable
*drawable
)
966 /* MSAA and fast color clear are not supported, so don't waste time
967 * checking whether a resolve is needed.
972 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
973 struct intel_renderbuffer
*rb
;
975 /* Usually, only the back buffer will need to be downsampled. However,
976 * the front buffer will also need it if the user has rendered into it.
978 static const gl_buffer_index buffers
[2] = {
983 for (int i
= 0; i
< 2; ++i
) {
984 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
985 if (rb
== NULL
|| rb
->mt
== NULL
)
987 if (rb
->mt
->num_samples
<= 1)
988 intel_miptree_resolve_color(brw
, rb
->mt
);
990 intel_renderbuffer_downsample(brw
, rb
);
995 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
997 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1001 intel_query_dri2_buffers(struct brw_context
*brw
,
1002 __DRIdrawable
*drawable
,
1003 __DRIbuffer
**buffers
,
1007 intel_process_dri2_buffer(struct brw_context
*brw
,
1008 __DRIdrawable
*drawable
,
1009 __DRIbuffer
*buffer
,
1010 struct intel_renderbuffer
*rb
,
1011 const char *buffer_name
);
1014 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1017 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1019 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1020 struct intel_renderbuffer
*rb
;
1021 __DRIbuffer
*buffers
= NULL
;
1023 const char *region_name
;
1025 /* Set this up front, so that in case our buffers get invalidated
1026 * while we're getting new buffers, we don't clobber the stamp and
1027 * thus ignore the invalidate. */
1028 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1030 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1031 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1033 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1035 if (buffers
== NULL
)
1038 for (i
= 0; i
< count
; i
++) {
1039 switch (buffers
[i
].attachment
) {
1040 case __DRI_BUFFER_FRONT_LEFT
:
1041 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1042 region_name
= "dri2 front buffer";
1045 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1046 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1047 region_name
= "dri2 fake front buffer";
1050 case __DRI_BUFFER_BACK_LEFT
:
1051 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1052 region_name
= "dri2 back buffer";
1055 case __DRI_BUFFER_DEPTH
:
1056 case __DRI_BUFFER_HIZ
:
1057 case __DRI_BUFFER_DEPTH_STENCIL
:
1058 case __DRI_BUFFER_STENCIL
:
1059 case __DRI_BUFFER_ACCUM
:
1062 "unhandled buffer attach event, attachment type %d\n",
1063 buffers
[i
].attachment
);
1067 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1073 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1075 struct brw_context
*brw
= context
->driverPrivate
;
1076 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1078 /* Set this up front, so that in case our buffers get invalidated
1079 * while we're getting new buffers, we don't clobber the stamp and
1080 * thus ignore the invalidate. */
1081 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1083 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1084 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1086 if (screen
->image
.loader
)
1087 intel_update_image_buffers(brw
, drawable
);
1089 intel_update_dri2_buffers(brw
, drawable
);
1091 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1095 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1096 * state is required.
1099 intel_prepare_render(struct brw_context
*brw
)
1101 struct gl_context
*ctx
= &brw
->ctx
;
1102 __DRIcontext
*driContext
= brw
->driContext
;
1103 __DRIdrawable
*drawable
;
1105 drawable
= driContext
->driDrawablePriv
;
1106 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1107 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1108 intel_update_renderbuffers(driContext
, drawable
);
1109 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1112 drawable
= driContext
->driReadablePriv
;
1113 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1114 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1115 intel_update_renderbuffers(driContext
, drawable
);
1116 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1119 /* If we're currently rendering to the front buffer, the rendering
1120 * that will happen next will probably dirty the front buffer. So
1121 * mark it as dirty here.
1123 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1124 brw
->front_buffer_dirty
= true;
1126 /* Wait for the swapbuffers before the one we just emitted, so we
1127 * don't get too many swaps outstanding for apps that are GPU-heavy
1128 * but not CPU-heavy.
1130 * We're using intelDRI2Flush (called from the loader before
1131 * swapbuffer) and glFlush (for front buffer rendering) as the
1132 * indicator that a frame is done and then throttle when we get
1133 * here as we prepare to render the next frame. At this point for
1134 * round trips for swap/copy and getting new buffers are done and
1135 * we'll spend less time waiting on the GPU.
1137 * Unfortunately, we don't have a handle to the batch containing
1138 * the swap, and getting our hands on that doesn't seem worth it,
1139 * so we just us the first batch we emitted after the last swap.
1141 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1142 if (!brw
->disable_throttling
)
1143 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1144 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1145 brw
->first_post_swapbuffers_batch
= NULL
;
1146 brw
->need_throttle
= false;
1151 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1153 * To determine which DRI buffers to request, examine the renderbuffers
1154 * attached to the drawable's framebuffer. Then request the buffers with
1155 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1157 * This is called from intel_update_renderbuffers().
1159 * \param drawable Drawable whose buffers are queried.
1160 * \param buffers [out] List of buffers returned by DRI2 query.
1161 * \param buffer_count [out] Number of buffers returned.
1163 * \see intel_update_renderbuffers()
1164 * \see DRI2GetBuffers()
1165 * \see DRI2GetBuffersWithFormat()
1168 intel_query_dri2_buffers(struct brw_context
*brw
,
1169 __DRIdrawable
*drawable
,
1170 __DRIbuffer
**buffers
,
1173 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1174 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1176 unsigned attachments
[8];
1178 struct intel_renderbuffer
*front_rb
;
1179 struct intel_renderbuffer
*back_rb
;
1181 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1182 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1184 memset(attachments
, 0, sizeof(attachments
));
1185 if ((brw_is_front_buffer_drawing(fb
) ||
1186 brw_is_front_buffer_reading(fb
) ||
1187 !back_rb
) && front_rb
) {
1188 /* If a fake front buffer is in use, then querying for
1189 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1190 * the real front buffer to the fake front buffer. So before doing the
1191 * query, we need to make sure all the pending drawing has landed in the
1192 * real front buffer.
1194 intel_batchbuffer_flush(brw
);
1195 intel_flush_front(&brw
->ctx
);
1197 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1198 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1199 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1200 /* We have pending front buffer rendering, but we aren't querying for a
1201 * front buffer. If the front buffer we have is a fake front buffer,
1202 * the X server is going to throw it away when it processes the query.
1203 * So before doing the query, make sure all the pending drawing has
1204 * landed in the real front buffer.
1206 intel_batchbuffer_flush(brw
);
1207 intel_flush_front(&brw
->ctx
);
1211 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1212 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1215 assert(i
<= ARRAY_SIZE(attachments
));
1217 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1222 drawable
->loaderPrivate
);
1226 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1228 * This is called from intel_update_renderbuffers().
1231 * DRI buffers whose attachment point is DRI2BufferStencil or
1232 * DRI2BufferDepthStencil are handled as special cases.
1234 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1235 * that is passed to drm_intel_bo_gem_create_from_name().
1237 * \see intel_update_renderbuffers()
1240 intel_process_dri2_buffer(struct brw_context
*brw
,
1241 __DRIdrawable
*drawable
,
1242 __DRIbuffer
*buffer
,
1243 struct intel_renderbuffer
*rb
,
1244 const char *buffer_name
)
1246 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1252 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1254 /* We try to avoid closing and reopening the same BO name, because the first
1255 * use of a mapping of the buffer involves a bunch of page faulting which is
1256 * moderately expensive.
1258 struct intel_mipmap_tree
*last_mt
;
1259 if (num_samples
== 0)
1262 last_mt
= rb
->singlesample_mt
;
1264 uint32_t old_name
= 0;
1266 /* The bo already has a name because the miptree was created by a
1267 * previous call to intel_process_dri2_buffer(). If a bo already has a
1268 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1269 * create a new name.
1271 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1274 if (old_name
== buffer
->name
)
1277 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1279 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1280 buffer
->name
, buffer
->attachment
,
1281 buffer
->cpp
, buffer
->pitch
);
1284 intel_miptree_release(&rb
->mt
);
1285 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1289 "Failed to open BO for returned DRI2 buffer "
1290 "(%dx%d, %s, named %d).\n"
1291 "This is likely a bug in the X Server that will lead to a "
1293 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1297 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1298 drawable
->w
, drawable
->h
,
1301 if (brw_is_front_buffer_drawing(fb
) &&
1302 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1303 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1304 rb
->Base
.Base
.NumSamples
> 1) {
1305 intel_renderbuffer_upsample(brw
, rb
);
1310 drm_intel_bo_unreference(bo
);
1314 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1316 * To determine which DRI buffers to request, examine the renderbuffers
1317 * attached to the drawable's framebuffer. Then request the buffers from
1320 * This is called from intel_update_renderbuffers().
1322 * \param drawable Drawable whose buffers are queried.
1323 * \param buffers [out] List of buffers returned by DRI2 query.
1324 * \param buffer_count [out] Number of buffers returned.
1326 * \see intel_update_renderbuffers()
1330 intel_update_image_buffer(struct brw_context
*intel
,
1331 __DRIdrawable
*drawable
,
1332 struct intel_renderbuffer
*rb
,
1334 enum __DRIimageBufferMask buffer_type
)
1336 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1338 if (!rb
|| !buffer
->bo
)
1341 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1343 /* Check and see if we're already bound to the right
1346 struct intel_mipmap_tree
*last_mt
;
1347 if (num_samples
== 0)
1350 last_mt
= rb
->singlesample_mt
;
1352 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1355 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1356 buffer
->width
, buffer
->height
,
1359 if (brw_is_front_buffer_drawing(fb
) &&
1360 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1361 rb
->Base
.Base
.NumSamples
> 1) {
1362 intel_renderbuffer_upsample(intel
, rb
);
1367 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1369 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1370 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1371 struct intel_renderbuffer
*front_rb
;
1372 struct intel_renderbuffer
*back_rb
;
1373 struct __DRIimageList images
;
1374 unsigned int format
;
1375 uint32_t buffer_mask
= 0;
1377 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1378 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1381 format
= intel_rb_format(back_rb
);
1383 format
= intel_rb_format(front_rb
);
1387 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1388 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1389 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1393 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1395 (*screen
->image
.loader
->getBuffers
) (drawable
,
1396 driGLFormatToImageFormat(format
),
1397 &drawable
->dri2
.stamp
,
1398 drawable
->loaderPrivate
,
1402 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1403 drawable
->w
= images
.front
->width
;
1404 drawable
->h
= images
.front
->height
;
1405 intel_update_image_buffer(brw
,
1409 __DRI_IMAGE_BUFFER_FRONT
);
1411 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1412 drawable
->w
= images
.back
->width
;
1413 drawable
->h
= images
.back
->height
;
1414 intel_update_image_buffer(brw
,
1418 __DRI_IMAGE_BUFFER_BACK
);