2 Copyright 2003 VMware, Inc.
3 Copyright (C) Intel Corp. 2006. All Rights Reserved.
4 Intel funded Tungsten Graphics to
5 develop this 3D driver.
7 Permission is hereby granted, free of charge, to any person obtaining
8 a copy of this software and associated documentation files (the
9 "Software"), to deal in the Software without restriction, including
10 without limitation the rights to use, copy, modify, merge, publish,
11 distribute, sublicense, and/or sell copies of the Software, and to
12 permit persons to whom the Software is furnished to do so, subject to
13 the following conditions:
15 The above copyright notice and this permission notice (including the
16 next paragraph) shall be included in all copies or substantial
17 portions of the Software.
19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **********************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
34 #include "main/api_exec.h"
35 #include "main/context.h"
36 #include "main/fbobject.h"
37 #include "main/extensions.h"
38 #include "main/imports.h"
39 #include "main/macros.h"
40 #include "main/points.h"
41 #include "main/version.h"
42 #include "main/vtxfmt.h"
43 #include "main/texobj.h"
44 #include "main/framebuffer.h"
46 #include "vbo/vbo_context.h"
48 #include "drivers/common/driverfuncs.h"
49 #include "drivers/common/meta.h"
52 #include "brw_context.h"
53 #include "brw_defines.h"
54 #include "brw_compiler.h"
56 #include "brw_state.h"
58 #include "intel_batchbuffer.h"
59 #include "intel_buffer_objects.h"
60 #include "intel_buffers.h"
61 #include "intel_fbo.h"
62 #include "intel_mipmap_tree.h"
63 #include "intel_pixel.h"
64 #include "intel_image.h"
65 #include "intel_tex.h"
66 #include "intel_tex_obj.h"
68 #include "swrast_setup/swrast_setup.h"
70 #include "tnl/t_pipeline.h"
71 #include "util/ralloc.h"
72 #include "util/debug.h"
74 /***************************************
75 * Mesa's Driver Functions
76 ***************************************/
79 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
80 GLenum internalFormat
, int samples
[16])
82 struct brw_context
*brw
= brw_context(ctx
);
110 assert(brw
->gen
< 6);
116 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
119 brw_get_renderer_string(unsigned deviceID
)
122 static char buffer
[128];
126 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
127 #include "pci_ids/i965_pci_ids.h"
129 chipset
= "Unknown Intel Chipset";
133 (void) driGetRendererString(buffer
, chipset
, 0);
137 static const GLubyte
*
138 intel_get_string(struct gl_context
* ctx
, GLenum name
)
140 const struct brw_context
*const brw
= brw_context(ctx
);
144 return (GLubyte
*) brw_vendor_string
;
148 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
156 intel_viewport(struct gl_context
*ctx
)
158 struct brw_context
*brw
= brw_context(ctx
);
159 __DRIcontext
*driContext
= brw
->driContext
;
161 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
162 if (driContext
->driDrawablePriv
)
163 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
164 if (driContext
->driReadablePriv
)
165 dri2InvalidateDrawable(driContext
->driReadablePriv
);
170 intel_update_framebuffer(struct gl_context
*ctx
,
171 struct gl_framebuffer
*fb
)
173 struct brw_context
*brw
= brw_context(ctx
);
175 /* Quantize the derived default number of samples
177 fb
->DefaultGeometry
._NumSamples
=
178 intel_quantize_num_samples(brw
->intelScreen
,
179 fb
->DefaultGeometry
.NumSamples
);
183 intel_update_state(struct gl_context
* ctx
, GLuint new_state
)
185 struct brw_context
*brw
= brw_context(ctx
);
186 struct intel_texture_object
*tex_obj
;
187 struct intel_renderbuffer
*depth_irb
;
189 if (ctx
->swrast_context
)
190 _swrast_InvalidateState(ctx
, new_state
);
191 _vbo_InvalidateState(ctx
, new_state
);
193 brw
->NewGLState
|= new_state
;
195 _mesa_unlock_context_textures(ctx
);
197 /* Resolve the depth buffer's HiZ buffer. */
198 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
200 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
202 /* Resolve depth buffer and render cache of each enabled texture. */
203 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
204 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
205 if (!ctx
->Texture
.Unit
[i
]._Current
)
207 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
208 if (!tex_obj
|| !tex_obj
->mt
)
210 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
211 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
212 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
215 /* Resolve color for each active shader image. */
216 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
217 const struct gl_shader
*shader
= ctx
->_Shader
->CurrentProgram
[i
] ?
218 ctx
->_Shader
->CurrentProgram
[i
]->_LinkedShaders
[i
] : NULL
;
220 if (unlikely(shader
&& shader
->NumImages
)) {
221 for (unsigned j
= 0; j
< shader
->NumImages
; j
++) {
222 struct gl_image_unit
*u
= &ctx
->ImageUnits
[shader
->ImageUnits
[j
]];
223 tex_obj
= intel_texture_object(u
->TexObj
);
225 if (tex_obj
&& tex_obj
->mt
) {
226 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
227 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
233 /* If FRAMEBUFFER_SRGB is used on Gen9+ then we need to resolve any of the
234 * single-sampled color renderbuffers because the CCS buffer isn't
235 * supported for SRGB formats. This only matters if FRAMEBUFFER_SRGB is
236 * enabled because otherwise the surface state will be programmed with the
237 * linear equivalent format anyway.
239 if (brw
->gen
>= 9 && ctx
->Color
.sRGBEnabled
) {
240 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
241 for (int i
= 0; i
< fb
->_NumColorDrawBuffers
; i
++) {
242 struct gl_renderbuffer
*rb
= fb
->_ColorDrawBuffers
[i
];
247 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
248 struct intel_mipmap_tree
*mt
= irb
->mt
;
251 mt
->num_samples
> 1 ||
252 _mesa_get_srgb_format_linear(mt
->format
) == mt
->format
)
255 intel_miptree_resolve_color(brw
, mt
);
256 brw_render_cache_set_check_flush(brw
, mt
->bo
);
260 _mesa_lock_context_textures(ctx
);
262 if (new_state
& _NEW_BUFFERS
) {
263 intel_update_framebuffer(ctx
, ctx
->DrawBuffer
);
264 if (ctx
->DrawBuffer
!= ctx
->ReadBuffer
)
265 intel_update_framebuffer(ctx
, ctx
->ReadBuffer
);
269 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
272 intel_flush_front(struct gl_context
*ctx
)
274 struct brw_context
*brw
= brw_context(ctx
);
275 __DRIcontext
*driContext
= brw
->driContext
;
276 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
277 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
279 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
280 if (flushFront(screen
) && driDrawable
&&
281 driDrawable
->loaderPrivate
) {
283 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
285 * This potentially resolves both front and back buffer. It
286 * is unnecessary to resolve the back, but harms nothing except
287 * performance. And no one cares about front-buffer render
290 intel_resolve_for_dri2_flush(brw
, driDrawable
);
291 intel_batchbuffer_flush(brw
);
293 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
295 /* We set the dirty bit in intel_prepare_render() if we're
296 * front buffer rendering once we get there.
298 brw
->front_buffer_dirty
= false;
304 intel_glFlush(struct gl_context
*ctx
)
306 struct brw_context
*brw
= brw_context(ctx
);
308 intel_batchbuffer_flush(brw
);
309 intel_flush_front(ctx
);
311 brw
->need_flush_throttle
= true;
315 intel_finish(struct gl_context
* ctx
)
317 struct brw_context
*brw
= brw_context(ctx
);
321 if (brw
->batch
.last_bo
)
322 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
326 brw_init_driver_functions(struct brw_context
*brw
,
327 struct dd_function_table
*functions
)
329 _mesa_init_driver_functions(functions
);
331 /* GLX uses DRI2 invalidate events to handle window resizing.
332 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
333 * which doesn't provide a mechanism for snooping the event queues.
335 * So EGL still relies on viewport hacks to handle window resizing.
336 * This should go away with DRI3000.
338 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
339 functions
->Viewport
= intel_viewport
;
341 functions
->Flush
= intel_glFlush
;
342 functions
->Finish
= intel_finish
;
343 functions
->GetString
= intel_get_string
;
344 functions
->UpdateState
= intel_update_state
;
346 intelInitTextureFuncs(functions
);
347 intelInitTextureImageFuncs(functions
);
348 intelInitTextureSubImageFuncs(functions
);
349 intelInitTextureCopyImageFuncs(functions
);
350 intelInitCopyImageFuncs(functions
);
351 intelInitClearFuncs(functions
);
352 intelInitBufferFuncs(functions
);
353 intelInitPixelFuncs(functions
);
354 intelInitBufferObjectFuncs(functions
);
355 intel_init_syncobj_functions(functions
);
356 brw_init_object_purgeable_functions(functions
);
358 brwInitFragProgFuncs( functions
);
359 brw_init_common_queryobj_functions(functions
);
361 gen6_init_queryobj_functions(functions
);
363 gen4_init_queryobj_functions(functions
);
364 brw_init_compute_functions(functions
);
366 brw_init_conditional_render_functions(functions
);
368 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
370 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
371 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
372 functions
->GetTransformFeedbackVertexCount
=
373 brw_get_transform_feedback_vertex_count
;
375 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
376 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
377 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
378 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
380 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
381 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
385 functions
->GetSamplePosition
= gen6_get_sample_position
;
389 brw_initialize_context_constants(struct brw_context
*brw
)
391 struct gl_context
*ctx
= &brw
->ctx
;
392 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
394 const bool stage_exists
[MESA_SHADER_STAGES
] = {
395 [MESA_SHADER_VERTEX
] = true,
396 [MESA_SHADER_TESS_CTRL
] = brw
->gen
>= 7,
397 [MESA_SHADER_TESS_EVAL
] = brw
->gen
>= 7,
398 [MESA_SHADER_GEOMETRY
] = brw
->gen
>= 6,
399 [MESA_SHADER_FRAGMENT
] = true,
400 [MESA_SHADER_COMPUTE
] =
401 (ctx
->API
== API_OPENGL_CORE
&&
402 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 1024) ||
403 (ctx
->API
== API_OPENGLES2
&&
404 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 128) ||
405 _mesa_extension_override_enables
.ARB_compute_shader
,
408 unsigned num_stages
= 0;
409 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
414 unsigned max_samplers
=
415 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
417 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
418 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
419 ctx
->Const
.MaxCombinedShaderOutputResources
=
420 MAX_IMAGE_UNITS
+ BRW_MAX_DRAW_BUFFERS
;
422 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
424 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
425 ctx
->Const
.MaxImageUnits
= MAX_IMAGE_UNITS
;
426 ctx
->Const
.MaxRenderbufferSize
= 8192;
427 ctx
->Const
.MaxTextureLevels
= MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS
);
428 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
429 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
430 ctx
->Const
.MaxArrayTextureLayers
= brw
->gen
>= 7 ? 2048 : 512;
431 ctx
->Const
.MaxTextureMbytes
= 1536;
432 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
433 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
434 ctx
->Const
.StripTextureBorder
= true;
436 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
437 else if (brw
->gen
== 6)
438 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
440 ctx
->Const
.MaxUniformBlockSize
= 65536;
442 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
443 struct gl_program_constants
*prog
= &ctx
->Const
.Program
[i
];
445 if (!stage_exists
[i
])
448 prog
->MaxTextureImageUnits
= max_samplers
;
450 prog
->MaxUniformBlocks
= BRW_MAX_UBO
;
451 prog
->MaxCombinedUniformComponents
=
452 prog
->MaxUniformComponents
+
453 ctx
->Const
.MaxUniformBlockSize
/ 4 * prog
->MaxUniformBlocks
;
455 prog
->MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
456 prog
->MaxAtomicBuffers
= BRW_MAX_ABO
;
457 prog
->MaxImageUniforms
= compiler
->scalar_stage
[i
] ? BRW_MAX_IMAGES
: 0;
458 prog
->MaxShaderStorageBlocks
= BRW_MAX_SSBO
;
461 ctx
->Const
.MaxTextureUnits
=
462 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
463 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
465 ctx
->Const
.MaxUniformBufferBindings
= num_stages
* BRW_MAX_UBO
;
466 ctx
->Const
.MaxCombinedUniformBlocks
= num_stages
* BRW_MAX_UBO
;
467 ctx
->Const
.MaxCombinedAtomicBuffers
= num_stages
* BRW_MAX_ABO
;
468 ctx
->Const
.MaxCombinedShaderStorageBlocks
= num_stages
* BRW_MAX_SSBO
;
469 ctx
->Const
.MaxShaderStorageBufferBindings
= num_stages
* BRW_MAX_SSBO
;
470 ctx
->Const
.MaxCombinedTextureImageUnits
= num_stages
* max_samplers
;
471 ctx
->Const
.MaxCombinedImageUniforms
= num_stages
* BRW_MAX_IMAGES
;
474 /* Hardware only supports a limited number of transform feedback buffers.
475 * So we need to override the Mesa default (which is based only on software
478 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
480 /* On Gen6, in the worst case, we use up one binding table entry per
481 * transform feedback component (see comments above the definition of
482 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
483 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
484 * BRW_MAX_SOL_BINDINGS.
486 * In "separate components" mode, we need to divide this value by
487 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
488 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
490 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
491 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
492 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
494 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
497 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
498 const int clamp_max_samples
=
499 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
501 if (clamp_max_samples
< 0) {
502 max_samples
= msaa_modes
[0];
504 /* Select the largest supported MSAA mode that does not exceed
508 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
509 if (msaa_modes
[i
] <= clamp_max_samples
) {
510 max_samples
= msaa_modes
[i
];
516 ctx
->Const
.MaxSamples
= max_samples
;
517 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
518 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
519 ctx
->Const
.MaxIntegerSamples
= max_samples
;
520 ctx
->Const
.MaxImageSamples
= 0;
522 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
523 * to map indices of rectangular grid to sample numbers within a pixel.
524 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
525 * extension implementation. For more details see the comment above
526 * gen6_set_sample_maps() definition.
528 gen6_set_sample_maps(ctx
);
530 ctx
->Const
.MinLineWidth
= 1.0;
531 ctx
->Const
.MinLineWidthAA
= 1.0;
533 ctx
->Const
.MaxLineWidth
= 7.375;
534 ctx
->Const
.MaxLineWidthAA
= 7.375;
535 ctx
->Const
.LineWidthGranularity
= 0.125;
537 ctx
->Const
.MaxLineWidth
= 7.0;
538 ctx
->Const
.MaxLineWidthAA
= 7.0;
539 ctx
->Const
.LineWidthGranularity
= 0.5;
542 /* For non-antialiased lines, we have to round the line width to the
543 * nearest whole number. Make sure that we don't advertise a line
544 * width that, when rounded, will be beyond the actual hardware
547 assert(roundf(ctx
->Const
.MaxLineWidth
) <= ctx
->Const
.MaxLineWidth
);
549 ctx
->Const
.MinPointSize
= 1.0;
550 ctx
->Const
.MinPointSizeAA
= 1.0;
551 ctx
->Const
.MaxPointSize
= 255.0;
552 ctx
->Const
.MaxPointSizeAA
= 255.0;
553 ctx
->Const
.PointSizeGranularity
= 1.0;
555 if (brw
->gen
>= 5 || brw
->is_g4x
)
556 ctx
->Const
.MaxClipPlanes
= 8;
558 ctx
->Const
.LowerTessLevel
= true;
560 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
561 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
562 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
563 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
564 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
565 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
566 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
567 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
568 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
569 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
570 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
571 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
572 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
573 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
575 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
576 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
577 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
578 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
579 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
580 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
581 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
582 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
583 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
584 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
585 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
587 /* Fragment shaders use real, 32-bit twos-complement integers for all
590 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
591 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
592 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
593 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
594 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
596 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
597 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
598 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
599 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
600 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
602 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
603 * but we're not sure how it's actually done for vertex order,
604 * that affect provoking vertex decision. Always use last vertex
605 * convention for quad primitive which works as expected for now.
608 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
610 ctx
->Const
.NativeIntegers
= true;
611 ctx
->Const
.VertexID_is_zero_based
= true;
613 /* Regarding the CMP instruction, the Ivybridge PRM says:
615 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
616 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
617 * 0xFFFFFFFF) is assigned to dst."
619 * but PRMs for earlier generations say
621 * "In dword format, one GRF may store up to 8 results. When the register
622 * is used later as a vector of Booleans, as only LSB at each channel
623 * contains meaning [sic] data, software should make sure all higher bits
624 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
626 * We select the representation of a true boolean uniform to be ~0, and fix
627 * the results of Gen <= 5 CMP instruction's with -(result & 1).
629 ctx
->Const
.UniformBooleanTrue
= ~0;
631 /* From the gen4 PRM, volume 4 page 127:
633 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
634 * the base address of the first element of the surface, computed in
635 * software by adding the surface base address to the byte offset of
636 * the element in the buffer."
638 * However, unaligned accesses are slower, so enforce buffer alignment.
640 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
642 /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so
643 * that we can safely have the CPU and GPU writing the same SSBO on
644 * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never
645 * writes, so there's no problem. For an SSBO, the GPU and the CPU can
646 * be updating disjoint regions of the buffer simultaneously and that will
647 * break if the regions overlap the same cacheline.
649 ctx
->Const
.ShaderStorageBufferOffsetAlignment
= 64;
650 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
651 ctx
->Const
.MaxTextureBufferSize
= 128 * 1024 * 1024;
654 ctx
->Const
.MaxVarying
= 32;
655 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
656 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
657 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
658 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
659 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxInputComponents
= 128;
660 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxOutputComponents
= 128;
661 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxInputComponents
= 128;
662 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxOutputComponents
= 128;
665 /* We want the GLSL compiler to emit code that uses condition codes */
666 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
667 ctx
->Const
.ShaderCompilerOptions
[i
] =
668 brw
->intelScreen
->compiler
->glsl_compiler_options
[i
];
671 /* ARB_viewport_array */
672 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
673 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
674 ctx
->Const
.ViewportSubpixelBits
= 0;
676 /* Cast to float before negating because MaxViewportWidth is unsigned.
678 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
679 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
682 /* ARB_gpu_shader5 */
684 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
686 /* ARB_framebuffer_no_attachments */
687 ctx
->Const
.MaxFramebufferWidth
= ctx
->Const
.MaxViewportWidth
;
688 ctx
->Const
.MaxFramebufferHeight
= ctx
->Const
.MaxViewportHeight
;
689 ctx
->Const
.MaxFramebufferLayers
= ctx
->Const
.MaxArrayTextureLayers
;
690 ctx
->Const
.MaxFramebufferSamples
= max_samples
;
694 brw_initialize_cs_context_constants(struct brw_context
*brw
, unsigned max_threads
)
696 struct gl_context
*ctx
= &brw
->ctx
;
698 /* For ES, we set these constants based on SIMD8.
700 * TODO: Once we can always generate SIMD16, we should update this.
702 * For GL, we assume we can generate a SIMD16 program, but this currently
703 * is not always true. This allows us to run more test cases, and will be
704 * required based on desktop GL compute shader requirements.
706 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
708 const uint32_t max_invocations
= simd_size
* max_threads
;
709 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
710 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
711 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
712 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
716 * Process driconf (drirc) options, setting appropriate context flags.
718 * intelInitExtensions still pokes at optionCache directly, in order to
719 * avoid advertising various extensions. No flags are set, so it makes
720 * sense to continue doing that there.
723 brw_process_driconf_options(struct brw_context
*brw
)
725 struct gl_context
*ctx
= &brw
->ctx
;
727 driOptionCache
*options
= &brw
->optionCache
;
728 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
729 brw
->driContext
->driScreenPriv
->myNum
, "i965");
731 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
732 switch (bo_reuse_mode
) {
733 case DRI_CONF_BO_REUSE_DISABLED
:
735 case DRI_CONF_BO_REUSE_ALL
:
736 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
740 if (!driQueryOptionb(options
, "hiz")) {
741 brw
->has_hiz
= false;
742 /* On gen6, you can only do separate stencil with HIZ. */
744 brw
->has_separate_stencil
= false;
747 if (driQueryOptionb(options
, "always_flush_batch")) {
748 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
749 brw
->always_flush_batch
= true;
752 if (driQueryOptionb(options
, "always_flush_cache")) {
753 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
754 brw
->always_flush_cache
= true;
757 if (driQueryOptionb(options
, "disable_throttling")) {
758 fprintf(stderr
, "disabling flush throttling\n");
759 brw
->disable_throttling
= true;
762 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
764 ctx
->Const
.ForceGLSLExtensionsWarn
=
765 driQueryOptionb(options
, "force_glsl_extensions_warn");
767 ctx
->Const
.DisableGLSLLineContinuations
=
768 driQueryOptionb(options
, "disable_glsl_line_continuations");
770 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
771 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
773 brw
->dual_color_blend_by_location
=
774 driQueryOptionb(options
, "dual_color_blend_by_location");
778 brwCreateContext(gl_api api
,
779 const struct gl_config
*mesaVis
,
780 __DRIcontext
*driContextPriv
,
781 unsigned major_version
,
782 unsigned minor_version
,
785 unsigned *dri_ctx_error
,
786 void *sharedContextPrivate
)
788 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
789 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
790 struct intel_screen
*screen
= sPriv
->driverPrivate
;
791 const struct brw_device_info
*devinfo
= screen
->devinfo
;
792 struct dd_function_table functions
;
794 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
795 * provides us with context reset notifications.
797 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
798 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
800 if (screen
->has_context_reset_notification
)
801 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
803 if (flags
& ~allowed_flags
) {
804 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
808 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
810 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
811 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
815 driContextPriv
->driverPrivate
= brw
;
816 brw
->driContext
= driContextPriv
;
817 brw
->intelScreen
= screen
;
818 brw
->bufmgr
= screen
->bufmgr
;
820 brw
->gen
= devinfo
->gen
;
821 brw
->gt
= devinfo
->gt
;
822 brw
->is_g4x
= devinfo
->is_g4x
;
823 brw
->is_baytrail
= devinfo
->is_baytrail
;
824 brw
->is_haswell
= devinfo
->is_haswell
;
825 brw
->is_cherryview
= devinfo
->is_cherryview
;
826 brw
->is_broxton
= devinfo
->is_broxton
;
827 brw
->has_llc
= devinfo
->has_llc
;
828 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
829 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
830 brw
->has_pln
= devinfo
->has_pln
;
831 brw
->has_compr4
= devinfo
->has_compr4
;
832 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
833 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
834 brw
->needs_unlit_centroid_workaround
=
835 devinfo
->needs_unlit_centroid_workaround
;
837 brw
->must_use_separate_stencil
= devinfo
->must_use_separate_stencil
;
838 brw
->has_swizzling
= screen
->hw_has_swizzling
;
840 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
841 brw
->tcs
.base
.stage
= MESA_SHADER_TESS_CTRL
;
842 brw
->tes
.base
.stage
= MESA_SHADER_TESS_EVAL
;
843 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
844 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
846 gen8_init_vtable_surface_functions(brw
);
847 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
848 } else if (brw
->gen
>= 7) {
849 gen7_init_vtable_surface_functions(brw
);
850 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
851 } else if (brw
->gen
>= 6) {
852 gen6_init_vtable_surface_functions(brw
);
853 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
855 gen4_init_vtable_surface_functions(brw
);
856 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
859 brw_init_driver_functions(brw
, &functions
);
862 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
864 struct gl_context
*ctx
= &brw
->ctx
;
866 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
867 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
868 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
869 intelDestroyContext(driContextPriv
);
873 driContextSetFlags(ctx
, flags
);
875 /* Initialize the software rasterizer and helper modules.
877 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
878 * software fallbacks (which we have to support on legacy GL to do weird
879 * glDrawPixels(), glBitmap(), and other functions).
881 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
882 _swrast_CreateContext(ctx
);
885 _vbo_CreateContext(ctx
);
886 if (ctx
->swrast_context
) {
887 _tnl_CreateContext(ctx
);
888 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
889 _swsetup_CreateContext(ctx
);
891 /* Configure swrast to match hardware characteristics: */
892 _swrast_allow_pixel_fog(ctx
, false);
893 _swrast_allow_vertex_fog(ctx
, true);
896 _mesa_meta_init(ctx
);
898 brw_process_driconf_options(brw
);
900 if (INTEL_DEBUG
& DEBUG_PERF
)
901 brw
->perf_debug
= true;
903 brw_initialize_cs_context_constants(brw
, devinfo
->max_cs_threads
);
904 brw_initialize_context_constants(brw
);
906 ctx
->Const
.ResetStrategy
= notify_reset
907 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
909 /* Reinitialize the context point state. It depends on ctx->Const values. */
910 _mesa_init_point(ctx
);
914 intel_batchbuffer_init(brw
);
917 /* Create a new hardware context. Using a hardware context means that
918 * our GPU state will be saved/restored on context switch, allowing us
919 * to assume that the GPU is in the same state we left it in.
921 * This is required for transform feedback buffer offsets, query objects,
922 * and also allows us to reduce how much state we have to emit.
924 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
927 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
928 intelDestroyContext(driContextPriv
);
933 if (brw_init_pipe_control(brw
, devinfo
)) {
934 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
935 intelDestroyContext(driContextPriv
);
941 intelInitExtensions(ctx
);
943 brw_init_surface_formats(brw
);
945 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
946 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
947 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
948 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
949 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
950 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
951 brw
->urb
.size
= devinfo
->urb
.size
;
952 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
953 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
954 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
955 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
956 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
958 /* Estimate the size of the mappable aperture into the GTT. There's an
959 * ioctl to get the whole GTT size, but not one to get the mappable subset.
960 * It turns out it's basically always 256MB, though some ancient hardware
963 uint32_t gtt_size
= 256 * 1024 * 1024;
965 /* We don't want to map two objects such that a memcpy between them would
966 * just fault one mapping in and then the other over and over forever. So
967 * we would need to divide the GTT size by 2. Additionally, some GTT is
968 * taken up by things like the framebuffer and the ringbuffer and such, so
969 * be more conservative.
971 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
974 brw
->urb
.gs_present
= false;
976 brw
->prim_restart
.in_progress
= false;
977 brw
->prim_restart
.enable_cut_index
= false;
978 brw
->gs
.enabled
= false;
979 brw
->sf
.viewport_transform_enable
= true;
981 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
983 brw
->use_resource_streamer
= screen
->has_resource_streamer
&&
984 (env_var_as_boolean("INTEL_USE_HW_BT", false) ||
985 env_var_as_boolean("INTEL_USE_GATHER", false));
987 ctx
->VertexProgram
._MaintainTnlProgram
= true;
988 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
990 brw_draw_init( brw
);
992 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
993 /* Turn on some extra GL_ARB_debug_output generation. */
994 brw
->perf_debug
= true;
997 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
998 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
1000 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
1001 brw_init_shader_time(brw
);
1003 _mesa_compute_version(ctx
);
1005 _mesa_initialize_dispatch_tables(ctx
);
1006 _mesa_initialize_vbo_vtxfmt(ctx
);
1008 if (ctx
->Extensions
.AMD_performance_monitor
) {
1009 brw_init_performance_monitors(brw
);
1012 vbo_use_buffer_objects(ctx
);
1013 vbo_always_unmap_buffers(ctx
);
1019 intelDestroyContext(__DRIcontext
* driContextPriv
)
1021 struct brw_context
*brw
=
1022 (struct brw_context
*) driContextPriv
->driverPrivate
;
1023 struct gl_context
*ctx
= &brw
->ctx
;
1025 /* Dump a final BMP in case the application doesn't call SwapBuffers */
1026 if (INTEL_DEBUG
& DEBUG_AUB
) {
1027 intel_batchbuffer_flush(brw
);
1028 aub_dump_bmp(&brw
->ctx
);
1031 _mesa_meta_free(&brw
->ctx
);
1032 brw_meta_fast_clear_free(brw
);
1034 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
1035 /* Force a report. */
1036 brw
->shader_time
.report_time
= 0;
1038 brw_collect_and_report_shader_time(brw
);
1039 brw_destroy_shader_time(brw
);
1042 brw_destroy_state(brw
);
1043 brw_draw_destroy(brw
);
1045 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
1046 if (brw
->vs
.base
.scratch_bo
)
1047 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
1048 if (brw
->gs
.base
.scratch_bo
)
1049 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
1050 if (brw
->wm
.base
.scratch_bo
)
1051 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
1053 gen7_reset_hw_bt_pool_offsets(brw
);
1054 drm_intel_bo_unreference(brw
->hw_bt_pool
.bo
);
1055 brw
->hw_bt_pool
.bo
= NULL
;
1057 drm_intel_gem_context_destroy(brw
->hw_ctx
);
1059 if (ctx
->swrast_context
) {
1060 _swsetup_DestroyContext(&brw
->ctx
);
1061 _tnl_DestroyContext(&brw
->ctx
);
1063 _vbo_DestroyContext(&brw
->ctx
);
1065 if (ctx
->swrast_context
)
1066 _swrast_DestroyContext(&brw
->ctx
);
1068 brw_fini_pipe_control(brw
);
1069 intel_batchbuffer_free(brw
);
1071 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
1072 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
1073 brw
->throttle_batch
[1] = NULL
;
1074 brw
->throttle_batch
[0] = NULL
;
1076 driDestroyOptionCache(&brw
->optionCache
);
1078 /* free the Mesa context */
1079 _mesa_free_context_data(&brw
->ctx
);
1082 driContextPriv
->driverPrivate
= NULL
;
1086 intelUnbindContext(__DRIcontext
* driContextPriv
)
1088 /* Unset current context and dispath table */
1089 _mesa_make_current(NULL
, NULL
, NULL
);
1095 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1096 * on window system framebuffers.
1098 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1099 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1100 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1101 * for a visual where you're guaranteed to be capable, but it turns out that
1102 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1103 * incapable ones, because there's no difference between the two in resources
1104 * used. Applications thus get built that accidentally rely on the default
1105 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1108 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1109 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1110 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1111 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1112 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1113 * and get no sRGB encode (assuming that both kinds of visual are available).
1114 * Thus our choice to support sRGB by default on our visuals for desktop would
1115 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1117 * Unfortunately, renderbuffer setup happens before a context is created. So
1118 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1119 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1120 * yet), we go turn that back off before anyone finds out.
1123 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1124 struct gl_framebuffer
*fb
)
1126 struct gl_context
*ctx
= &brw
->ctx
;
1128 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1131 /* Some day when we support the sRGB capable bit on visuals available for
1132 * GLES, we'll need to respect that and not disable things here.
1134 fb
->Visual
.sRGBCapable
= false;
1135 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1136 if (fb
->Attachment
[i
].Renderbuffer
&&
1137 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1138 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1144 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1145 __DRIdrawable
* driDrawPriv
,
1146 __DRIdrawable
* driReadPriv
)
1148 struct brw_context
*brw
;
1149 GET_CURRENT_CONTEXT(curCtx
);
1152 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1156 /* According to the glXMakeCurrent() man page: "Pending commands to
1157 * the previous context, if any, are flushed before it is released."
1158 * But only flush if we're actually changing contexts.
1160 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1161 _mesa_flush(curCtx
);
1164 if (driContextPriv
) {
1165 struct gl_context
*ctx
= &brw
->ctx
;
1166 struct gl_framebuffer
*fb
, *readFb
;
1168 if (driDrawPriv
== NULL
) {
1169 fb
= _mesa_get_incomplete_framebuffer();
1171 fb
= driDrawPriv
->driverPrivate
;
1172 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1175 if (driReadPriv
== NULL
) {
1176 readFb
= _mesa_get_incomplete_framebuffer();
1178 readFb
= driReadPriv
->driverPrivate
;
1179 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1182 /* The sRGB workaround changes the renderbuffer's format. We must change
1183 * the format before the renderbuffer's miptree get's allocated, otherwise
1184 * the formats of the renderbuffer and its miptree will differ.
1186 intel_gles3_srgb_workaround(brw
, fb
);
1187 intel_gles3_srgb_workaround(brw
, readFb
);
1189 /* If the context viewport hasn't been initialized, force a call out to
1190 * the loader to get buffers so we have a drawable size for the initial
1192 if (!brw
->ctx
.ViewportInitialized
)
1193 intel_prepare_render(brw
);
1195 _mesa_make_current(ctx
, fb
, readFb
);
1197 _mesa_make_current(NULL
, NULL
, NULL
);
1204 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1205 __DRIdrawable
*drawable
)
1208 /* MSAA and fast color clear are not supported, so don't waste time
1209 * checking whether a resolve is needed.
1214 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1215 struct intel_renderbuffer
*rb
;
1217 /* Usually, only the back buffer will need to be downsampled. However,
1218 * the front buffer will also need it if the user has rendered into it.
1220 static const gl_buffer_index buffers
[2] = {
1225 for (int i
= 0; i
< 2; ++i
) {
1226 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1227 if (rb
== NULL
|| rb
->mt
== NULL
)
1229 if (rb
->mt
->num_samples
<= 1)
1230 intel_miptree_resolve_color(brw
, rb
->mt
);
1232 intel_renderbuffer_downsample(brw
, rb
);
1237 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1239 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1243 intel_query_dri2_buffers(struct brw_context
*brw
,
1244 __DRIdrawable
*drawable
,
1245 __DRIbuffer
**buffers
,
1249 intel_process_dri2_buffer(struct brw_context
*brw
,
1250 __DRIdrawable
*drawable
,
1251 __DRIbuffer
*buffer
,
1252 struct intel_renderbuffer
*rb
,
1253 const char *buffer_name
);
1256 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1259 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1261 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1262 struct intel_renderbuffer
*rb
;
1263 __DRIbuffer
*buffers
= NULL
;
1265 const char *region_name
;
1267 /* Set this up front, so that in case our buffers get invalidated
1268 * while we're getting new buffers, we don't clobber the stamp and
1269 * thus ignore the invalidate. */
1270 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1272 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1273 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1275 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1277 if (buffers
== NULL
)
1280 for (i
= 0; i
< count
; i
++) {
1281 switch (buffers
[i
].attachment
) {
1282 case __DRI_BUFFER_FRONT_LEFT
:
1283 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1284 region_name
= "dri2 front buffer";
1287 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1288 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1289 region_name
= "dri2 fake front buffer";
1292 case __DRI_BUFFER_BACK_LEFT
:
1293 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1294 region_name
= "dri2 back buffer";
1297 case __DRI_BUFFER_DEPTH
:
1298 case __DRI_BUFFER_HIZ
:
1299 case __DRI_BUFFER_DEPTH_STENCIL
:
1300 case __DRI_BUFFER_STENCIL
:
1301 case __DRI_BUFFER_ACCUM
:
1304 "unhandled buffer attach event, attachment type %d\n",
1305 buffers
[i
].attachment
);
1309 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1315 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1317 struct brw_context
*brw
= context
->driverPrivate
;
1318 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1320 /* Set this up front, so that in case our buffers get invalidated
1321 * while we're getting new buffers, we don't clobber the stamp and
1322 * thus ignore the invalidate. */
1323 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1325 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1326 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1328 if (screen
->image
.loader
)
1329 intel_update_image_buffers(brw
, drawable
);
1331 intel_update_dri2_buffers(brw
, drawable
);
1333 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1337 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1338 * state is required.
1341 intel_prepare_render(struct brw_context
*brw
)
1343 struct gl_context
*ctx
= &brw
->ctx
;
1344 __DRIcontext
*driContext
= brw
->driContext
;
1345 __DRIdrawable
*drawable
;
1347 drawable
= driContext
->driDrawablePriv
;
1348 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1349 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1350 intel_update_renderbuffers(driContext
, drawable
);
1351 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1354 drawable
= driContext
->driReadablePriv
;
1355 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1356 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1357 intel_update_renderbuffers(driContext
, drawable
);
1358 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1361 /* If we're currently rendering to the front buffer, the rendering
1362 * that will happen next will probably dirty the front buffer. So
1363 * mark it as dirty here.
1365 if (_mesa_is_front_buffer_drawing(ctx
->DrawBuffer
))
1366 brw
->front_buffer_dirty
= true;
1370 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1372 * To determine which DRI buffers to request, examine the renderbuffers
1373 * attached to the drawable's framebuffer. Then request the buffers with
1374 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1376 * This is called from intel_update_renderbuffers().
1378 * \param drawable Drawable whose buffers are queried.
1379 * \param buffers [out] List of buffers returned by DRI2 query.
1380 * \param buffer_count [out] Number of buffers returned.
1382 * \see intel_update_renderbuffers()
1383 * \see DRI2GetBuffers()
1384 * \see DRI2GetBuffersWithFormat()
1387 intel_query_dri2_buffers(struct brw_context
*brw
,
1388 __DRIdrawable
*drawable
,
1389 __DRIbuffer
**buffers
,
1392 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1393 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1395 unsigned attachments
[8];
1397 struct intel_renderbuffer
*front_rb
;
1398 struct intel_renderbuffer
*back_rb
;
1400 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1401 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1403 memset(attachments
, 0, sizeof(attachments
));
1404 if ((_mesa_is_front_buffer_drawing(fb
) ||
1405 _mesa_is_front_buffer_reading(fb
) ||
1406 !back_rb
) && front_rb
) {
1407 /* If a fake front buffer is in use, then querying for
1408 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1409 * the real front buffer to the fake front buffer. So before doing the
1410 * query, we need to make sure all the pending drawing has landed in the
1411 * real front buffer.
1413 intel_batchbuffer_flush(brw
);
1414 intel_flush_front(&brw
->ctx
);
1416 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1417 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1418 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1419 /* We have pending front buffer rendering, but we aren't querying for a
1420 * front buffer. If the front buffer we have is a fake front buffer,
1421 * the X server is going to throw it away when it processes the query.
1422 * So before doing the query, make sure all the pending drawing has
1423 * landed in the real front buffer.
1425 intel_batchbuffer_flush(brw
);
1426 intel_flush_front(&brw
->ctx
);
1430 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1431 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1434 assert(i
<= ARRAY_SIZE(attachments
));
1436 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1441 drawable
->loaderPrivate
);
1445 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1447 * This is called from intel_update_renderbuffers().
1450 * DRI buffers whose attachment point is DRI2BufferStencil or
1451 * DRI2BufferDepthStencil are handled as special cases.
1453 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1454 * that is passed to drm_intel_bo_gem_create_from_name().
1456 * \see intel_update_renderbuffers()
1459 intel_process_dri2_buffer(struct brw_context
*brw
,
1460 __DRIdrawable
*drawable
,
1461 __DRIbuffer
*buffer
,
1462 struct intel_renderbuffer
*rb
,
1463 const char *buffer_name
)
1465 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1471 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1473 /* We try to avoid closing and reopening the same BO name, because the first
1474 * use of a mapping of the buffer involves a bunch of page faulting which is
1475 * moderately expensive.
1477 struct intel_mipmap_tree
*last_mt
;
1478 if (num_samples
== 0)
1481 last_mt
= rb
->singlesample_mt
;
1483 uint32_t old_name
= 0;
1485 /* The bo already has a name because the miptree was created by a
1486 * previous call to intel_process_dri2_buffer(). If a bo already has a
1487 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1488 * create a new name.
1490 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1493 if (old_name
== buffer
->name
)
1496 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1498 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1499 buffer
->name
, buffer
->attachment
,
1500 buffer
->cpp
, buffer
->pitch
);
1503 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1507 "Failed to open BO for returned DRI2 buffer "
1508 "(%dx%d, %s, named %d).\n"
1509 "This is likely a bug in the X Server that will lead to a "
1511 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1515 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1516 drawable
->w
, drawable
->h
,
1519 if (_mesa_is_front_buffer_drawing(fb
) &&
1520 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1521 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1522 rb
->Base
.Base
.NumSamples
> 1) {
1523 intel_renderbuffer_upsample(brw
, rb
);
1528 drm_intel_bo_unreference(bo
);
1532 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1534 * To determine which DRI buffers to request, examine the renderbuffers
1535 * attached to the drawable's framebuffer. Then request the buffers from
1538 * This is called from intel_update_renderbuffers().
1540 * \param drawable Drawable whose buffers are queried.
1541 * \param buffers [out] List of buffers returned by DRI2 query.
1542 * \param buffer_count [out] Number of buffers returned.
1544 * \see intel_update_renderbuffers()
1548 intel_update_image_buffer(struct brw_context
*intel
,
1549 __DRIdrawable
*drawable
,
1550 struct intel_renderbuffer
*rb
,
1552 enum __DRIimageBufferMask buffer_type
)
1554 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1556 if (!rb
|| !buffer
->bo
)
1559 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1561 /* Check and see if we're already bound to the right
1564 struct intel_mipmap_tree
*last_mt
;
1565 if (num_samples
== 0)
1568 last_mt
= rb
->singlesample_mt
;
1570 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1573 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1574 buffer
->width
, buffer
->height
,
1577 if (_mesa_is_front_buffer_drawing(fb
) &&
1578 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1579 rb
->Base
.Base
.NumSamples
> 1) {
1580 intel_renderbuffer_upsample(intel
, rb
);
1585 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1587 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1588 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1589 struct intel_renderbuffer
*front_rb
;
1590 struct intel_renderbuffer
*back_rb
;
1591 struct __DRIimageList images
;
1592 unsigned int format
;
1593 uint32_t buffer_mask
= 0;
1595 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1596 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1599 format
= intel_rb_format(back_rb
);
1601 format
= intel_rb_format(front_rb
);
1605 if (front_rb
&& (_mesa_is_front_buffer_drawing(fb
) ||
1606 _mesa_is_front_buffer_reading(fb
) || !back_rb
)) {
1607 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1611 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1613 (*screen
->image
.loader
->getBuffers
) (drawable
,
1614 driGLFormatToImageFormat(format
),
1615 &drawable
->dri2
.stamp
,
1616 drawable
->loaderPrivate
,
1620 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1621 drawable
->w
= images
.front
->width
;
1622 drawable
->h
= images
.front
->height
;
1623 intel_update_image_buffer(brw
,
1627 __DRI_IMAGE_BUFFER_FRONT
);
1629 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1630 drawable
->w
= images
.back
->width
;
1631 drawable
->h
= images
.back
->height
;
1632 intel_update_image_buffer(brw
,
1636 __DRI_IMAGE_BUFFER_BACK
);