2 Copyright 2003 VMware, Inc.
3 Copyright (C) Intel Corp. 2006. All Rights Reserved.
4 Intel funded Tungsten Graphics to
5 develop this 3D driver.
7 Permission is hereby granted, free of charge, to any person obtaining
8 a copy of this software and associated documentation files (the
9 "Software"), to deal in the Software without restriction, including
10 without limitation the rights to use, copy, modify, merge, publish,
11 distribute, sublicense, and/or sell copies of the Software, and to
12 permit persons to whom the Software is furnished to do so, subject to
13 the following conditions:
15 The above copyright notice and this permission notice (including the
16 next paragraph) shall be included in all copies or substantial
17 portions of the Software.
19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **********************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
34 #include "main/api_exec.h"
35 #include "main/context.h"
36 #include "main/fbobject.h"
37 #include "main/extensions.h"
38 #include "main/imports.h"
39 #include "main/macros.h"
40 #include "main/points.h"
41 #include "main/version.h"
42 #include "main/vtxfmt.h"
43 #include "main/texobj.h"
45 #include "vbo/vbo_context.h"
47 #include "drivers/common/driverfuncs.h"
48 #include "drivers/common/meta.h"
51 #include "brw_context.h"
52 #include "brw_defines.h"
54 #include "brw_state.h"
56 #include "intel_batchbuffer.h"
57 #include "intel_buffer_objects.h"
58 #include "intel_buffers.h"
59 #include "intel_fbo.h"
60 #include "intel_mipmap_tree.h"
61 #include "intel_pixel.h"
62 #include "intel_image.h"
63 #include "intel_tex.h"
64 #include "intel_tex_obj.h"
66 #include "swrast_setup/swrast_setup.h"
68 #include "tnl/t_pipeline.h"
69 #include "util/ralloc.h"
71 /***************************************
72 * Mesa's Driver Functions
73 ***************************************/
76 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
77 GLenum internalFormat
, int samples
[16])
79 struct brw_context
*brw
= brw_context(ctx
);
101 assert(brw
->gen
< 6);
107 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
110 brw_get_renderer_string(unsigned deviceID
)
113 static char buffer
[128];
117 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
118 #include "pci_ids/i965_pci_ids.h"
120 chipset
= "Unknown Intel Chipset";
124 (void) driGetRendererString(buffer
, chipset
, 0);
128 static const GLubyte
*
129 intel_get_string(struct gl_context
* ctx
, GLenum name
)
131 const struct brw_context
*const brw
= brw_context(ctx
);
135 return (GLubyte
*) brw_vendor_string
;
139 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
147 intel_viewport(struct gl_context
*ctx
)
149 struct brw_context
*brw
= brw_context(ctx
);
150 __DRIcontext
*driContext
= brw
->driContext
;
152 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
153 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
154 dri2InvalidateDrawable(driContext
->driReadablePriv
);
159 intel_update_state(struct gl_context
* ctx
, GLuint new_state
)
161 struct brw_context
*brw
= brw_context(ctx
);
162 struct intel_texture_object
*tex_obj
;
163 struct intel_renderbuffer
*depth_irb
;
165 if (ctx
->swrast_context
)
166 _swrast_InvalidateState(ctx
, new_state
);
167 _vbo_InvalidateState(ctx
, new_state
);
169 brw
->NewGLState
|= new_state
;
171 _mesa_unlock_context_textures(ctx
);
173 /* Resolve the depth buffer's HiZ buffer. */
174 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
176 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
178 /* Resolve depth buffer and render cache of each enabled texture. */
179 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
180 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
181 if (!ctx
->Texture
.Unit
[i
]._Current
)
183 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
184 if (!tex_obj
|| !tex_obj
->mt
)
186 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
187 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
188 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
191 _mesa_lock_context_textures(ctx
);
194 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
197 intel_flush_front(struct gl_context
*ctx
)
199 struct brw_context
*brw
= brw_context(ctx
);
200 __DRIcontext
*driContext
= brw
->driContext
;
201 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
202 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
204 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
205 if (flushFront(screen
) && driDrawable
&&
206 driDrawable
->loaderPrivate
) {
208 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
210 * This potentially resolves both front and back buffer. It
211 * is unnecessary to resolve the back, but harms nothing except
212 * performance. And no one cares about front-buffer render
215 intel_resolve_for_dri2_flush(brw
, driDrawable
);
216 intel_batchbuffer_flush(brw
);
218 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
220 /* We set the dirty bit in intel_prepare_render() if we're
221 * front buffer rendering once we get there.
223 brw
->front_buffer_dirty
= false;
229 intel_glFlush(struct gl_context
*ctx
)
231 struct brw_context
*brw
= brw_context(ctx
);
233 intel_batchbuffer_flush(brw
);
234 intel_flush_front(ctx
);
235 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
236 brw
->need_throttle
= true;
240 intel_finish(struct gl_context
* ctx
)
242 struct brw_context
*brw
= brw_context(ctx
);
246 if (brw
->batch
.last_bo
)
247 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
251 brw_init_driver_functions(struct brw_context
*brw
,
252 struct dd_function_table
*functions
)
254 _mesa_init_driver_functions(functions
);
256 /* GLX uses DRI2 invalidate events to handle window resizing.
257 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
258 * which doesn't provide a mechanism for snooping the event queues.
260 * So EGL still relies on viewport hacks to handle window resizing.
261 * This should go away with DRI3000.
263 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
264 functions
->Viewport
= intel_viewport
;
266 functions
->Flush
= intel_glFlush
;
267 functions
->Finish
= intel_finish
;
268 functions
->GetString
= intel_get_string
;
269 functions
->UpdateState
= intel_update_state
;
271 intelInitTextureFuncs(functions
);
272 intelInitTextureImageFuncs(functions
);
273 intelInitTextureSubImageFuncs(functions
);
274 intelInitTextureCopyImageFuncs(functions
);
275 intelInitCopyImageFuncs(functions
);
276 intelInitClearFuncs(functions
);
277 intelInitBufferFuncs(functions
);
278 intelInitPixelFuncs(functions
);
279 intelInitBufferObjectFuncs(functions
);
280 intel_init_syncobj_functions(functions
);
281 brw_init_object_purgeable_functions(functions
);
283 brwInitFragProgFuncs( functions
);
284 brw_init_common_queryobj_functions(functions
);
286 gen6_init_queryobj_functions(functions
);
288 gen4_init_queryobj_functions(functions
);
290 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
292 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
293 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
294 functions
->GetTransformFeedbackVertexCount
=
295 brw_get_transform_feedback_vertex_count
;
297 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
298 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
299 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
300 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
302 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
303 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
307 functions
->GetSamplePosition
= gen6_get_sample_position
;
311 brw_initialize_context_constants(struct brw_context
*brw
)
313 struct gl_context
*ctx
= &brw
->ctx
;
315 unsigned max_samplers
=
316 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
318 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
320 ctx
->Const
.StripTextureBorder
= true;
322 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
323 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
324 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
325 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
326 ctx
->Const
.MaxTextureUnits
=
327 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
328 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
329 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
331 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
333 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
334 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
335 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
336 ctx
->Const
.MaxUniformBufferBindings
+= 12;
338 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
340 ctx
->Const
.MaxCombinedTextureImageUnits
=
341 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
342 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
343 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
344 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
346 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
347 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
348 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
349 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
350 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
351 ctx
->Const
.MaxTextureMbytes
= 1536;
354 ctx
->Const
.MaxArrayTextureLayers
= 2048;
356 ctx
->Const
.MaxArrayTextureLayers
= 512;
358 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
360 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
362 ctx
->Const
.MaxRenderbufferSize
= 8192;
364 /* Hardware only supports a limited number of transform feedback buffers.
365 * So we need to override the Mesa default (which is based only on software
368 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
370 /* On Gen6, in the worst case, we use up one binding table entry per
371 * transform feedback component (see comments above the definition of
372 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
373 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
374 * BRW_MAX_SOL_BINDINGS.
376 * In "separate components" mode, we need to divide this value by
377 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
378 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
380 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
381 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
382 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
384 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
387 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
388 const int clamp_max_samples
=
389 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
391 if (clamp_max_samples
< 0) {
392 max_samples
= msaa_modes
[0];
394 /* Select the largest supported MSAA mode that does not exceed
398 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
399 if (msaa_modes
[i
] <= clamp_max_samples
) {
400 max_samples
= msaa_modes
[i
];
406 ctx
->Const
.MaxSamples
= max_samples
;
407 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
408 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
409 ctx
->Const
.MaxIntegerSamples
= max_samples
;
411 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
412 * to map indices of rectangular grid to sample numbers within a pixel.
413 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
414 * extension implementation. For more details see the comment above
415 * gen6_set_sample_maps() definition.
417 gen6_set_sample_maps(ctx
);
420 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
421 else if (brw
->gen
== 6)
422 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
424 ctx
->Const
.MinLineWidth
= 1.0;
425 ctx
->Const
.MinLineWidthAA
= 1.0;
426 if (brw
->gen
>= 9 || brw
->is_cherryview
) {
427 ctx
->Const
.MaxLineWidth
= 40.0;
428 ctx
->Const
.MaxLineWidthAA
= 40.0;
429 ctx
->Const
.LineWidthGranularity
= 0.125;
430 } else if (brw
->gen
>= 6) {
431 ctx
->Const
.MaxLineWidth
= 7.375;
432 ctx
->Const
.MaxLineWidthAA
= 7.375;
433 ctx
->Const
.LineWidthGranularity
= 0.125;
435 ctx
->Const
.MaxLineWidth
= 7.0;
436 ctx
->Const
.MaxLineWidthAA
= 7.0;
437 ctx
->Const
.LineWidthGranularity
= 0.5;
440 ctx
->Const
.MinPointSize
= 1.0;
441 ctx
->Const
.MinPointSizeAA
= 1.0;
442 ctx
->Const
.MaxPointSize
= 255.0;
443 ctx
->Const
.MaxPointSizeAA
= 255.0;
444 ctx
->Const
.PointSizeGranularity
= 1.0;
446 if (brw
->gen
>= 5 || brw
->is_g4x
)
447 ctx
->Const
.MaxClipPlanes
= 8;
449 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
450 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
451 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
452 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
453 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
454 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
455 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
456 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
457 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
458 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
459 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
460 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
461 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
462 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
464 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
465 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
466 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
467 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
468 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
469 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
470 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
471 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
472 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
473 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
474 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
476 /* Fragment shaders use real, 32-bit twos-complement integers for all
479 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
480 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
481 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
482 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
483 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
485 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
486 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
487 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
488 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
489 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
492 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
493 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
494 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
495 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
496 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
497 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
498 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
499 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
500 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
503 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
504 * but we're not sure how it's actually done for vertex order,
505 * that affect provoking vertex decision. Always use last vertex
506 * convention for quad primitive which works as expected for now.
509 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
511 ctx
->Const
.NativeIntegers
= true;
512 ctx
->Const
.VertexID_is_zero_based
= true;
514 /* Regarding the CMP instruction, the Ivybridge PRM says:
516 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
517 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
518 * 0xFFFFFFFF) is assigned to dst."
520 * but PRMs for earlier generations say
522 * "In dword format, one GRF may store up to 8 results. When the register
523 * is used later as a vector of Booleans, as only LSB at each channel
524 * contains meaning [sic] data, software should make sure all higher bits
525 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
527 * We select the representation of a true boolean uniform to be ~0, and fix
528 * the results of Gen <= 5 CMP instruction's with -(result & 1).
530 ctx
->Const
.UniformBooleanTrue
= ~0;
532 /* From the gen4 PRM, volume 4 page 127:
534 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
535 * the base address of the first element of the surface, computed in
536 * software by adding the surface base address to the byte offset of
537 * the element in the buffer."
539 * However, unaligned accesses are slower, so enforce buffer alignment.
541 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
542 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
545 ctx
->Const
.MaxVarying
= 32;
546 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
547 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
548 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
549 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
552 /* We want the GLSL compiler to emit code that uses condition codes */
553 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
554 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
555 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
556 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
557 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
558 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
559 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
560 (i
== MESA_SHADER_FRAGMENT
);
561 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
562 (i
== MESA_SHADER_FRAGMENT
);
563 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
564 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
567 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
568 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
570 if (brw
->scalar_vs
) {
571 /* If we're using the scalar backend for vertex shaders, we need to
572 * configure these accordingly.
574 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectOutput
= true;
575 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectTemp
= true;
576 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= false;
579 /* ARB_viewport_array */
580 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
581 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
582 ctx
->Const
.ViewportSubpixelBits
= 0;
584 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
586 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
587 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
590 /* ARB_gpu_shader5 */
592 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
596 * Process driconf (drirc) options, setting appropriate context flags.
598 * intelInitExtensions still pokes at optionCache directly, in order to
599 * avoid advertising various extensions. No flags are set, so it makes
600 * sense to continue doing that there.
603 brw_process_driconf_options(struct brw_context
*brw
)
605 struct gl_context
*ctx
= &brw
->ctx
;
607 driOptionCache
*options
= &brw
->optionCache
;
608 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
609 brw
->driContext
->driScreenPriv
->myNum
, "i965");
611 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
612 switch (bo_reuse_mode
) {
613 case DRI_CONF_BO_REUSE_DISABLED
:
615 case DRI_CONF_BO_REUSE_ALL
:
616 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
620 if (!driQueryOptionb(options
, "hiz")) {
621 brw
->has_hiz
= false;
622 /* On gen6, you can only do separate stencil with HIZ. */
624 brw
->has_separate_stencil
= false;
627 if (driQueryOptionb(options
, "always_flush_batch")) {
628 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
629 brw
->always_flush_batch
= true;
632 if (driQueryOptionb(options
, "always_flush_cache")) {
633 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
634 brw
->always_flush_cache
= true;
637 if (driQueryOptionb(options
, "disable_throttling")) {
638 fprintf(stderr
, "disabling flush throttling\n");
639 brw
->disable_throttling
= true;
642 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
644 ctx
->Const
.ForceGLSLExtensionsWarn
=
645 driQueryOptionb(options
, "force_glsl_extensions_warn");
647 ctx
->Const
.DisableGLSLLineContinuations
=
648 driQueryOptionb(options
, "disable_glsl_line_continuations");
650 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
651 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
655 brwCreateContext(gl_api api
,
656 const struct gl_config
*mesaVis
,
657 __DRIcontext
*driContextPriv
,
658 unsigned major_version
,
659 unsigned minor_version
,
662 unsigned *dri_ctx_error
,
663 void *sharedContextPrivate
)
665 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
666 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
667 struct intel_screen
*screen
= sPriv
->driverPrivate
;
668 const struct brw_device_info
*devinfo
= screen
->devinfo
;
669 struct dd_function_table functions
;
671 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
672 * provides us with context reset notifications.
674 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
675 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
677 if (screen
->has_context_reset_notification
)
678 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
680 if (flags
& ~allowed_flags
) {
681 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
685 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
687 fprintf(stderr
, "%s: failed to alloc context\n", __FUNCTION__
);
688 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
692 driContextPriv
->driverPrivate
= brw
;
693 brw
->driContext
= driContextPriv
;
694 brw
->intelScreen
= screen
;
695 brw
->bufmgr
= screen
->bufmgr
;
697 brw
->gen
= devinfo
->gen
;
698 brw
->gt
= devinfo
->gt
;
699 brw
->is_g4x
= devinfo
->is_g4x
;
700 brw
->is_baytrail
= devinfo
->is_baytrail
;
701 brw
->is_haswell
= devinfo
->is_haswell
;
702 brw
->is_cherryview
= devinfo
->is_cherryview
;
703 brw
->has_llc
= devinfo
->has_llc
;
704 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
705 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
706 brw
->has_pln
= devinfo
->has_pln
;
707 brw
->has_compr4
= devinfo
->has_compr4
;
708 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
709 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
710 brw
->needs_unlit_centroid_workaround
=
711 devinfo
->needs_unlit_centroid_workaround
;
713 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
714 brw
->has_swizzling
= screen
->hw_has_swizzling
;
716 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
717 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
718 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
720 gen8_init_vtable_surface_functions(brw
);
721 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
722 } else if (brw
->gen
>= 7) {
723 gen7_init_vtable_surface_functions(brw
);
724 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
725 } else if (brw
->gen
>= 6) {
726 gen6_init_vtable_surface_functions(brw
);
727 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
729 gen4_init_vtable_surface_functions(brw
);
730 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
733 brw_init_driver_functions(brw
, &functions
);
736 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
738 struct gl_context
*ctx
= &brw
->ctx
;
740 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
741 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
742 fprintf(stderr
, "%s: failed to init mesa context\n", __FUNCTION__
);
743 intelDestroyContext(driContextPriv
);
747 driContextSetFlags(ctx
, flags
);
749 /* Initialize the software rasterizer and helper modules.
751 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
752 * software fallbacks (which we have to support on legacy GL to do weird
753 * glDrawPixels(), glBitmap(), and other functions).
755 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
756 _swrast_CreateContext(ctx
);
759 _vbo_CreateContext(ctx
);
760 if (ctx
->swrast_context
) {
761 _tnl_CreateContext(ctx
);
762 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
763 _swsetup_CreateContext(ctx
);
765 /* Configure swrast to match hardware characteristics: */
766 _swrast_allow_pixel_fog(ctx
, false);
767 _swrast_allow_vertex_fog(ctx
, true);
770 _mesa_meta_init(ctx
);
772 brw_process_driconf_options(brw
);
773 brw_process_intel_debug_variable(brw
);
775 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
776 brw
->scalar_vs
= true;
778 brw_initialize_context_constants(brw
);
780 ctx
->Const
.ResetStrategy
= notify_reset
781 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
783 /* Reinitialize the context point state. It depends on ctx->Const values. */
784 _mesa_init_point(ctx
);
788 intel_batchbuffer_init(brw
);
791 /* Create a new hardware context. Using a hardware context means that
792 * our GPU state will be saved/restored on context switch, allowing us
793 * to assume that the GPU is in the same state we left it in.
795 * This is required for transform feedback buffer offsets, query objects,
796 * and also allows us to reduce how much state we have to emit.
798 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
801 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
802 intelDestroyContext(driContextPriv
);
809 intelInitExtensions(ctx
);
811 brw_init_surface_formats(brw
);
813 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
814 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
815 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
816 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
817 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
818 brw
->urb
.size
= devinfo
->urb
.size
;
819 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
820 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
821 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
822 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
823 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
825 /* Estimate the size of the mappable aperture into the GTT. There's an
826 * ioctl to get the whole GTT size, but not one to get the mappable subset.
827 * It turns out it's basically always 256MB, though some ancient hardware
830 uint32_t gtt_size
= 256 * 1024 * 1024;
832 /* We don't want to map two objects such that a memcpy between them would
833 * just fault one mapping in and then the other over and over forever. So
834 * we would need to divide the GTT size by 2. Additionally, some GTT is
835 * taken up by things like the framebuffer and the ringbuffer and such, so
836 * be more conservative.
838 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
841 brw
->urb
.gs_present
= false;
843 brw
->prim_restart
.in_progress
= false;
844 brw
->prim_restart
.enable_cut_index
= false;
845 brw
->gs
.enabled
= false;
846 brw
->sf
.viewport_transform_enable
= true;
848 ctx
->VertexProgram
._MaintainTnlProgram
= true;
849 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
851 brw_draw_init( brw
);
853 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
854 /* Turn on some extra GL_ARB_debug_output generation. */
855 brw
->perf_debug
= true;
858 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
859 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
861 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
862 brw_init_shader_time(brw
);
864 _mesa_compute_version(ctx
);
866 _mesa_initialize_dispatch_tables(ctx
);
867 _mesa_initialize_vbo_vtxfmt(ctx
);
869 if (ctx
->Extensions
.AMD_performance_monitor
) {
870 brw_init_performance_monitors(brw
);
873 vbo_use_buffer_objects(ctx
);
874 vbo_always_unmap_buffers(ctx
);
880 intelDestroyContext(__DRIcontext
* driContextPriv
)
882 struct brw_context
*brw
=
883 (struct brw_context
*) driContextPriv
->driverPrivate
;
884 struct gl_context
*ctx
= &brw
->ctx
;
886 assert(brw
); /* should never be null */
890 /* Dump a final BMP in case the application doesn't call SwapBuffers */
891 if (INTEL_DEBUG
& DEBUG_AUB
) {
892 intel_batchbuffer_flush(brw
);
893 aub_dump_bmp(&brw
->ctx
);
896 _mesa_meta_free(&brw
->ctx
);
897 brw_meta_fast_clear_free(brw
);
899 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
900 /* Force a report. */
901 brw
->shader_time
.report_time
= 0;
903 brw_collect_and_report_shader_time(brw
);
904 brw_destroy_shader_time(brw
);
907 brw_destroy_state(brw
);
908 brw_draw_destroy(brw
);
910 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
911 if (brw
->vs
.base
.scratch_bo
)
912 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
913 if (brw
->gs
.base
.scratch_bo
)
914 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
915 if (brw
->wm
.base
.scratch_bo
)
916 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
918 drm_intel_gem_context_destroy(brw
->hw_ctx
);
920 if (ctx
->swrast_context
) {
921 _swsetup_DestroyContext(&brw
->ctx
);
922 _tnl_DestroyContext(&brw
->ctx
);
924 _vbo_DestroyContext(&brw
->ctx
);
926 if (ctx
->swrast_context
)
927 _swrast_DestroyContext(&brw
->ctx
);
929 intel_batchbuffer_free(brw
);
931 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
932 brw
->first_post_swapbuffers_batch
= NULL
;
934 driDestroyOptionCache(&brw
->optionCache
);
936 /* free the Mesa context */
937 _mesa_free_context_data(&brw
->ctx
);
940 driContextPriv
->driverPrivate
= NULL
;
944 intelUnbindContext(__DRIcontext
* driContextPriv
)
946 /* Unset current context and dispath table */
947 _mesa_make_current(NULL
, NULL
, NULL
);
953 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
954 * on window system framebuffers.
956 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
957 * your renderbuffer can do sRGB encode, and you can flip a switch that does
958 * sRGB encode if the renderbuffer can handle it. You can ask specifically
959 * for a visual where you're guaranteed to be capable, but it turns out that
960 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
961 * incapable ones, becuase there's no difference between the two in resources
962 * used. Applications thus get built that accidentally rely on the default
963 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
966 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
967 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
968 * So they removed the enable knob and made it "if the renderbuffer is sRGB
969 * capable, do sRGB encode". Then, for your window system renderbuffers, you
970 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
971 * and get no sRGB encode (assuming that both kinds of visual are available).
972 * Thus our choice to support sRGB by default on our visuals for desktop would
973 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
975 * Unfortunately, renderbuffer setup happens before a context is created. So
976 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
977 * context (without an sRGB visual, though we don't have sRGB visuals exposed
978 * yet), we go turn that back off before anyone finds out.
981 intel_gles3_srgb_workaround(struct brw_context
*brw
,
982 struct gl_framebuffer
*fb
)
984 struct gl_context
*ctx
= &brw
->ctx
;
986 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
989 /* Some day when we support the sRGB capable bit on visuals available for
990 * GLES, we'll need to respect that and not disable things here.
992 fb
->Visual
.sRGBCapable
= false;
993 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
994 if (fb
->Attachment
[i
].Renderbuffer
&&
995 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
996 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1002 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1003 __DRIdrawable
* driDrawPriv
,
1004 __DRIdrawable
* driReadPriv
)
1006 struct brw_context
*brw
;
1007 GET_CURRENT_CONTEXT(curCtx
);
1010 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1014 /* According to the glXMakeCurrent() man page: "Pending commands to
1015 * the previous context, if any, are flushed before it is released."
1016 * But only flush if we're actually changing contexts.
1018 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1019 _mesa_flush(curCtx
);
1022 if (driContextPriv
) {
1023 struct gl_context
*ctx
= &brw
->ctx
;
1024 struct gl_framebuffer
*fb
, *readFb
;
1026 if (driDrawPriv
== NULL
) {
1027 fb
= _mesa_get_incomplete_framebuffer();
1029 fb
= driDrawPriv
->driverPrivate
;
1030 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1033 if (driReadPriv
== NULL
) {
1034 readFb
= _mesa_get_incomplete_framebuffer();
1036 readFb
= driReadPriv
->driverPrivate
;
1037 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1040 /* The sRGB workaround changes the renderbuffer's format. We must change
1041 * the format before the renderbuffer's miptree get's allocated, otherwise
1042 * the formats of the renderbuffer and its miptree will differ.
1044 intel_gles3_srgb_workaround(brw
, fb
);
1045 intel_gles3_srgb_workaround(brw
, readFb
);
1047 /* If the context viewport hasn't been initialized, force a call out to
1048 * the loader to get buffers so we have a drawable size for the initial
1050 if (!brw
->ctx
.ViewportInitialized
)
1051 intel_prepare_render(brw
);
1053 _mesa_make_current(ctx
, fb
, readFb
);
1055 _mesa_make_current(NULL
, NULL
, NULL
);
1062 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1063 __DRIdrawable
*drawable
)
1066 /* MSAA and fast color clear are not supported, so don't waste time
1067 * checking whether a resolve is needed.
1072 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1073 struct intel_renderbuffer
*rb
;
1075 /* Usually, only the back buffer will need to be downsampled. However,
1076 * the front buffer will also need it if the user has rendered into it.
1078 static const gl_buffer_index buffers
[2] = {
1083 for (int i
= 0; i
< 2; ++i
) {
1084 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1085 if (rb
== NULL
|| rb
->mt
== NULL
)
1087 if (rb
->mt
->num_samples
<= 1)
1088 intel_miptree_resolve_color(brw
, rb
->mt
);
1090 intel_renderbuffer_downsample(brw
, rb
);
1095 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1097 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1101 intel_query_dri2_buffers(struct brw_context
*brw
,
1102 __DRIdrawable
*drawable
,
1103 __DRIbuffer
**buffers
,
1107 intel_process_dri2_buffer(struct brw_context
*brw
,
1108 __DRIdrawable
*drawable
,
1109 __DRIbuffer
*buffer
,
1110 struct intel_renderbuffer
*rb
,
1111 const char *buffer_name
);
1114 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1117 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1119 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1120 struct intel_renderbuffer
*rb
;
1121 __DRIbuffer
*buffers
= NULL
;
1123 const char *region_name
;
1125 /* Set this up front, so that in case our buffers get invalidated
1126 * while we're getting new buffers, we don't clobber the stamp and
1127 * thus ignore the invalidate. */
1128 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1130 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1131 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1133 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1135 if (buffers
== NULL
)
1138 for (i
= 0; i
< count
; i
++) {
1139 switch (buffers
[i
].attachment
) {
1140 case __DRI_BUFFER_FRONT_LEFT
:
1141 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1142 region_name
= "dri2 front buffer";
1145 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1146 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1147 region_name
= "dri2 fake front buffer";
1150 case __DRI_BUFFER_BACK_LEFT
:
1151 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1152 region_name
= "dri2 back buffer";
1155 case __DRI_BUFFER_DEPTH
:
1156 case __DRI_BUFFER_HIZ
:
1157 case __DRI_BUFFER_DEPTH_STENCIL
:
1158 case __DRI_BUFFER_STENCIL
:
1159 case __DRI_BUFFER_ACCUM
:
1162 "unhandled buffer attach event, attachment type %d\n",
1163 buffers
[i
].attachment
);
1167 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1173 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1175 struct brw_context
*brw
= context
->driverPrivate
;
1176 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1178 /* Set this up front, so that in case our buffers get invalidated
1179 * while we're getting new buffers, we don't clobber the stamp and
1180 * thus ignore the invalidate. */
1181 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1183 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1184 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1186 if (screen
->image
.loader
)
1187 intel_update_image_buffers(brw
, drawable
);
1189 intel_update_dri2_buffers(brw
, drawable
);
1191 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1195 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1196 * state is required.
1199 intel_prepare_render(struct brw_context
*brw
)
1201 struct gl_context
*ctx
= &brw
->ctx
;
1202 __DRIcontext
*driContext
= brw
->driContext
;
1203 __DRIdrawable
*drawable
;
1205 drawable
= driContext
->driDrawablePriv
;
1206 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1207 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1208 intel_update_renderbuffers(driContext
, drawable
);
1209 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1212 drawable
= driContext
->driReadablePriv
;
1213 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1214 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1215 intel_update_renderbuffers(driContext
, drawable
);
1216 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1219 /* If we're currently rendering to the front buffer, the rendering
1220 * that will happen next will probably dirty the front buffer. So
1221 * mark it as dirty here.
1223 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1224 brw
->front_buffer_dirty
= true;
1226 /* Wait for the swapbuffers before the one we just emitted, so we
1227 * don't get too many swaps outstanding for apps that are GPU-heavy
1228 * but not CPU-heavy.
1230 * We're using intelDRI2Flush (called from the loader before
1231 * swapbuffer) and glFlush (for front buffer rendering) as the
1232 * indicator that a frame is done and then throttle when we get
1233 * here as we prepare to render the next frame. At this point for
1234 * round trips for swap/copy and getting new buffers are done and
1235 * we'll spend less time waiting on the GPU.
1237 * Unfortunately, we don't have a handle to the batch containing
1238 * the swap, and getting our hands on that doesn't seem worth it,
1239 * so we just us the first batch we emitted after the last swap.
1241 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1242 if (!brw
->disable_throttling
)
1243 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1244 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1245 brw
->first_post_swapbuffers_batch
= NULL
;
1246 brw
->need_throttle
= false;
1251 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1253 * To determine which DRI buffers to request, examine the renderbuffers
1254 * attached to the drawable's framebuffer. Then request the buffers with
1255 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1257 * This is called from intel_update_renderbuffers().
1259 * \param drawable Drawable whose buffers are queried.
1260 * \param buffers [out] List of buffers returned by DRI2 query.
1261 * \param buffer_count [out] Number of buffers returned.
1263 * \see intel_update_renderbuffers()
1264 * \see DRI2GetBuffers()
1265 * \see DRI2GetBuffersWithFormat()
1268 intel_query_dri2_buffers(struct brw_context
*brw
,
1269 __DRIdrawable
*drawable
,
1270 __DRIbuffer
**buffers
,
1273 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1274 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1276 unsigned attachments
[8];
1278 struct intel_renderbuffer
*front_rb
;
1279 struct intel_renderbuffer
*back_rb
;
1281 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1282 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1284 memset(attachments
, 0, sizeof(attachments
));
1285 if ((brw_is_front_buffer_drawing(fb
) ||
1286 brw_is_front_buffer_reading(fb
) ||
1287 !back_rb
) && front_rb
) {
1288 /* If a fake front buffer is in use, then querying for
1289 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1290 * the real front buffer to the fake front buffer. So before doing the
1291 * query, we need to make sure all the pending drawing has landed in the
1292 * real front buffer.
1294 intel_batchbuffer_flush(brw
);
1295 intel_flush_front(&brw
->ctx
);
1297 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1298 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1299 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1300 /* We have pending front buffer rendering, but we aren't querying for a
1301 * front buffer. If the front buffer we have is a fake front buffer,
1302 * the X server is going to throw it away when it processes the query.
1303 * So before doing the query, make sure all the pending drawing has
1304 * landed in the real front buffer.
1306 intel_batchbuffer_flush(brw
);
1307 intel_flush_front(&brw
->ctx
);
1311 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1312 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1315 assert(i
<= ARRAY_SIZE(attachments
));
1317 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1322 drawable
->loaderPrivate
);
1326 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1328 * This is called from intel_update_renderbuffers().
1331 * DRI buffers whose attachment point is DRI2BufferStencil or
1332 * DRI2BufferDepthStencil are handled as special cases.
1334 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1335 * that is passed to drm_intel_bo_gem_create_from_name().
1337 * \see intel_update_renderbuffers()
1340 intel_process_dri2_buffer(struct brw_context
*brw
,
1341 __DRIdrawable
*drawable
,
1342 __DRIbuffer
*buffer
,
1343 struct intel_renderbuffer
*rb
,
1344 const char *buffer_name
)
1346 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1352 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1354 /* We try to avoid closing and reopening the same BO name, because the first
1355 * use of a mapping of the buffer involves a bunch of page faulting which is
1356 * moderately expensive.
1358 struct intel_mipmap_tree
*last_mt
;
1359 if (num_samples
== 0)
1362 last_mt
= rb
->singlesample_mt
;
1364 uint32_t old_name
= 0;
1366 /* The bo already has a name because the miptree was created by a
1367 * previous call to intel_process_dri2_buffer(). If a bo already has a
1368 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1369 * create a new name.
1371 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1374 if (old_name
== buffer
->name
)
1377 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1379 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1380 buffer
->name
, buffer
->attachment
,
1381 buffer
->cpp
, buffer
->pitch
);
1384 intel_miptree_release(&rb
->mt
);
1385 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1389 "Failed to open BO for returned DRI2 buffer "
1390 "(%dx%d, %s, named %d).\n"
1391 "This is likely a bug in the X Server that will lead to a "
1393 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1397 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1398 drawable
->w
, drawable
->h
,
1401 if (brw_is_front_buffer_drawing(fb
) &&
1402 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1403 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1404 rb
->Base
.Base
.NumSamples
> 1) {
1405 intel_renderbuffer_upsample(brw
, rb
);
1410 drm_intel_bo_unreference(bo
);
1414 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1416 * To determine which DRI buffers to request, examine the renderbuffers
1417 * attached to the drawable's framebuffer. Then request the buffers from
1420 * This is called from intel_update_renderbuffers().
1422 * \param drawable Drawable whose buffers are queried.
1423 * \param buffers [out] List of buffers returned by DRI2 query.
1424 * \param buffer_count [out] Number of buffers returned.
1426 * \see intel_update_renderbuffers()
1430 intel_update_image_buffer(struct brw_context
*intel
,
1431 __DRIdrawable
*drawable
,
1432 struct intel_renderbuffer
*rb
,
1434 enum __DRIimageBufferMask buffer_type
)
1436 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1438 if (!rb
|| !buffer
->bo
)
1441 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1443 /* Check and see if we're already bound to the right
1446 struct intel_mipmap_tree
*last_mt
;
1447 if (num_samples
== 0)
1450 last_mt
= rb
->singlesample_mt
;
1452 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1455 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1456 buffer
->width
, buffer
->height
,
1459 if (brw_is_front_buffer_drawing(fb
) &&
1460 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1461 rb
->Base
.Base
.NumSamples
> 1) {
1462 intel_renderbuffer_upsample(intel
, rb
);
1467 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1469 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1470 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1471 struct intel_renderbuffer
*front_rb
;
1472 struct intel_renderbuffer
*back_rb
;
1473 struct __DRIimageList images
;
1474 unsigned int format
;
1475 uint32_t buffer_mask
= 0;
1477 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1478 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1481 format
= intel_rb_format(back_rb
);
1483 format
= intel_rb_format(front_rb
);
1487 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1488 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1489 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1493 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1495 (*screen
->image
.loader
->getBuffers
) (drawable
,
1496 driGLFormatToImageFormat(format
),
1497 &drawable
->dri2
.stamp
,
1498 drawable
->loaderPrivate
,
1502 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1503 drawable
->w
= images
.front
->width
;
1504 drawable
->h
= images
.front
->height
;
1505 intel_update_image_buffer(brw
,
1509 __DRI_IMAGE_BUFFER_FRONT
);
1511 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1512 drawable
->w
= images
.back
->width
;
1513 drawable
->h
= images
.back
->height
;
1514 intel_update_image_buffer(brw
,
1518 __DRI_IMAGE_BUFFER_BACK
);