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_state(struct gl_context
* ctx
, GLuint new_state
)
172 struct brw_context
*brw
= brw_context(ctx
);
173 struct intel_texture_object
*tex_obj
;
174 struct intel_renderbuffer
*depth_irb
;
176 if (ctx
->swrast_context
)
177 _swrast_InvalidateState(ctx
, new_state
);
178 _vbo_InvalidateState(ctx
, new_state
);
180 brw
->NewGLState
|= new_state
;
182 _mesa_unlock_context_textures(ctx
);
184 /* Resolve the depth buffer's HiZ buffer. */
185 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
187 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
189 /* Resolve depth buffer and render cache of each enabled texture. */
190 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
191 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
192 if (!ctx
->Texture
.Unit
[i
]._Current
)
194 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
195 if (!tex_obj
|| !tex_obj
->mt
)
197 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
198 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
199 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
202 /* Resolve color for each active shader image. */
203 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
204 const struct gl_shader
*shader
= ctx
->_Shader
->CurrentProgram
[i
] ?
205 ctx
->_Shader
->CurrentProgram
[i
]->_LinkedShaders
[i
] : NULL
;
207 if (unlikely(shader
&& shader
->NumImages
)) {
208 for (unsigned j
= 0; j
< shader
->NumImages
; j
++) {
209 struct gl_image_unit
*u
= &ctx
->ImageUnits
[shader
->ImageUnits
[j
]];
210 tex_obj
= intel_texture_object(u
->TexObj
);
212 if (tex_obj
&& tex_obj
->mt
) {
213 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
214 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
220 /* If FRAMEBUFFER_SRGB is used on Gen9+ then we need to resolve any of the
221 * single-sampled color renderbuffers because the CCS buffer isn't
222 * supported for SRGB formats. This only matters if FRAMEBUFFER_SRGB is
223 * enabled because otherwise the surface state will be programmed with the
224 * linear equivalent format anyway.
226 if (brw
->gen
>= 9 && ctx
->Color
.sRGBEnabled
) {
227 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
228 for (int i
= 0; i
< fb
->_NumColorDrawBuffers
; i
++) {
229 struct gl_renderbuffer
*rb
= fb
->_ColorDrawBuffers
[i
];
234 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
235 struct intel_mipmap_tree
*mt
= irb
->mt
;
238 mt
->num_samples
> 1 ||
239 _mesa_get_srgb_format_linear(mt
->format
) == mt
->format
)
242 intel_miptree_resolve_color(brw
, mt
);
243 brw_render_cache_set_check_flush(brw
, mt
->bo
);
247 _mesa_lock_context_textures(ctx
);
250 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
253 intel_flush_front(struct gl_context
*ctx
)
255 struct brw_context
*brw
= brw_context(ctx
);
256 __DRIcontext
*driContext
= brw
->driContext
;
257 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
258 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
260 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
261 if (flushFront(screen
) && driDrawable
&&
262 driDrawable
->loaderPrivate
) {
264 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
266 * This potentially resolves both front and back buffer. It
267 * is unnecessary to resolve the back, but harms nothing except
268 * performance. And no one cares about front-buffer render
271 intel_resolve_for_dri2_flush(brw
, driDrawable
);
272 intel_batchbuffer_flush(brw
);
274 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
276 /* We set the dirty bit in intel_prepare_render() if we're
277 * front buffer rendering once we get there.
279 brw
->front_buffer_dirty
= false;
285 intel_glFlush(struct gl_context
*ctx
)
287 struct brw_context
*brw
= brw_context(ctx
);
289 intel_batchbuffer_flush(brw
);
290 intel_flush_front(ctx
);
292 brw
->need_flush_throttle
= true;
296 intel_finish(struct gl_context
* ctx
)
298 struct brw_context
*brw
= brw_context(ctx
);
302 if (brw
->batch
.last_bo
)
303 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
307 brw_init_driver_functions(struct brw_context
*brw
,
308 struct dd_function_table
*functions
)
310 _mesa_init_driver_functions(functions
);
312 /* GLX uses DRI2 invalidate events to handle window resizing.
313 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
314 * which doesn't provide a mechanism for snooping the event queues.
316 * So EGL still relies on viewport hacks to handle window resizing.
317 * This should go away with DRI3000.
319 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
320 functions
->Viewport
= intel_viewport
;
322 functions
->Flush
= intel_glFlush
;
323 functions
->Finish
= intel_finish
;
324 functions
->GetString
= intel_get_string
;
325 functions
->UpdateState
= intel_update_state
;
327 intelInitTextureFuncs(functions
);
328 intelInitTextureImageFuncs(functions
);
329 intelInitTextureSubImageFuncs(functions
);
330 intelInitTextureCopyImageFuncs(functions
);
331 intelInitCopyImageFuncs(functions
);
332 intelInitClearFuncs(functions
);
333 intelInitBufferFuncs(functions
);
334 intelInitPixelFuncs(functions
);
335 intelInitBufferObjectFuncs(functions
);
336 intel_init_syncobj_functions(functions
);
337 brw_init_object_purgeable_functions(functions
);
339 brwInitFragProgFuncs( functions
);
340 brw_init_common_queryobj_functions(functions
);
342 gen6_init_queryobj_functions(functions
);
344 gen4_init_queryobj_functions(functions
);
345 brw_init_compute_functions(functions
);
347 brw_init_conditional_render_functions(functions
);
349 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
351 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
352 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
353 functions
->GetTransformFeedbackVertexCount
=
354 brw_get_transform_feedback_vertex_count
;
356 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
357 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
358 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
359 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
361 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
362 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
366 functions
->GetSamplePosition
= gen6_get_sample_position
;
370 brw_initialize_context_constants(struct brw_context
*brw
)
372 struct gl_context
*ctx
= &brw
->ctx
;
373 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
375 const bool stage_exists
[MESA_SHADER_STAGES
] = {
376 [MESA_SHADER_VERTEX
] = true,
377 [MESA_SHADER_TESS_CTRL
] = brw
->gen
>= 7,
378 [MESA_SHADER_TESS_EVAL
] = brw
->gen
>= 7,
379 [MESA_SHADER_GEOMETRY
] = brw
->gen
>= 6,
380 [MESA_SHADER_FRAGMENT
] = true,
381 [MESA_SHADER_COMPUTE
] =
382 (ctx
->API
== API_OPENGL_CORE
&&
383 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 1024) ||
384 (ctx
->API
== API_OPENGLES2
&&
385 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 128) ||
386 _mesa_extension_override_enables
.ARB_compute_shader
,
389 unsigned num_stages
= 0;
390 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
395 unsigned max_samplers
=
396 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
398 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
399 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
400 ctx
->Const
.MaxCombinedShaderOutputResources
=
401 MAX_IMAGE_UNITS
+ BRW_MAX_DRAW_BUFFERS
;
403 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
405 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
406 ctx
->Const
.MaxImageUnits
= MAX_IMAGE_UNITS
;
407 ctx
->Const
.MaxRenderbufferSize
= 8192;
408 ctx
->Const
.MaxTextureLevels
= MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS
);
409 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
410 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
411 ctx
->Const
.MaxArrayTextureLayers
= brw
->gen
>= 7 ? 2048 : 512;
412 ctx
->Const
.MaxTextureMbytes
= 1536;
413 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
414 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
415 ctx
->Const
.StripTextureBorder
= true;
417 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
418 else if (brw
->gen
== 6)
419 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
421 ctx
->Const
.MaxUniformBlockSize
= 65536;
423 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
424 struct gl_program_constants
*prog
= &ctx
->Const
.Program
[i
];
426 if (!stage_exists
[i
])
429 prog
->MaxTextureImageUnits
= max_samplers
;
431 prog
->MaxUniformBlocks
= BRW_MAX_UBO
;
432 prog
->MaxCombinedUniformComponents
=
433 prog
->MaxUniformComponents
+
434 ctx
->Const
.MaxUniformBlockSize
/ 4 * prog
->MaxUniformBlocks
;
436 prog
->MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
437 prog
->MaxAtomicBuffers
= BRW_MAX_ABO
;
438 prog
->MaxImageUniforms
= compiler
->scalar_stage
[i
] ? BRW_MAX_IMAGES
: 0;
439 prog
->MaxShaderStorageBlocks
= BRW_MAX_SSBO
;
442 ctx
->Const
.MaxTextureUnits
=
443 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
444 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
446 ctx
->Const
.MaxUniformBufferBindings
= num_stages
* BRW_MAX_UBO
;
447 ctx
->Const
.MaxCombinedUniformBlocks
= num_stages
* BRW_MAX_UBO
;
448 ctx
->Const
.MaxCombinedAtomicBuffers
= num_stages
* BRW_MAX_ABO
;
449 ctx
->Const
.MaxCombinedShaderStorageBlocks
= num_stages
* BRW_MAX_SSBO
;
450 ctx
->Const
.MaxShaderStorageBufferBindings
= num_stages
* BRW_MAX_SSBO
;
451 ctx
->Const
.MaxCombinedTextureImageUnits
= num_stages
* max_samplers
;
452 ctx
->Const
.MaxCombinedImageUniforms
= num_stages
* BRW_MAX_IMAGES
;
455 /* Hardware only supports a limited number of transform feedback buffers.
456 * So we need to override the Mesa default (which is based only on software
459 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
461 /* On Gen6, in the worst case, we use up one binding table entry per
462 * transform feedback component (see comments above the definition of
463 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
464 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
465 * BRW_MAX_SOL_BINDINGS.
467 * In "separate components" mode, we need to divide this value by
468 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
469 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
471 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
472 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
473 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
475 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
478 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
479 const int clamp_max_samples
=
480 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
482 if (clamp_max_samples
< 0) {
483 max_samples
= msaa_modes
[0];
485 /* Select the largest supported MSAA mode that does not exceed
489 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
490 if (msaa_modes
[i
] <= clamp_max_samples
) {
491 max_samples
= msaa_modes
[i
];
497 ctx
->Const
.MaxSamples
= max_samples
;
498 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
499 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
500 ctx
->Const
.MaxIntegerSamples
= max_samples
;
501 ctx
->Const
.MaxImageSamples
= 0;
503 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
504 * to map indices of rectangular grid to sample numbers within a pixel.
505 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
506 * extension implementation. For more details see the comment above
507 * gen6_set_sample_maps() definition.
509 gen6_set_sample_maps(ctx
);
511 ctx
->Const
.MinLineWidth
= 1.0;
512 ctx
->Const
.MinLineWidthAA
= 1.0;
514 ctx
->Const
.MaxLineWidth
= 7.375;
515 ctx
->Const
.MaxLineWidthAA
= 7.375;
516 ctx
->Const
.LineWidthGranularity
= 0.125;
518 ctx
->Const
.MaxLineWidth
= 7.0;
519 ctx
->Const
.MaxLineWidthAA
= 7.0;
520 ctx
->Const
.LineWidthGranularity
= 0.5;
523 /* For non-antialiased lines, we have to round the line width to the
524 * nearest whole number. Make sure that we don't advertise a line
525 * width that, when rounded, will be beyond the actual hardware
528 assert(roundf(ctx
->Const
.MaxLineWidth
) <= ctx
->Const
.MaxLineWidth
);
530 ctx
->Const
.MinPointSize
= 1.0;
531 ctx
->Const
.MinPointSizeAA
= 1.0;
532 ctx
->Const
.MaxPointSize
= 255.0;
533 ctx
->Const
.MaxPointSizeAA
= 255.0;
534 ctx
->Const
.PointSizeGranularity
= 1.0;
536 if (brw
->gen
>= 5 || brw
->is_g4x
)
537 ctx
->Const
.MaxClipPlanes
= 8;
539 ctx
->Const
.LowerTessLevel
= true;
541 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
542 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
543 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
544 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
545 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
546 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
547 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
548 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
549 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
550 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
551 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
552 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
553 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
554 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
556 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
557 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
558 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
559 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
560 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
561 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
562 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
563 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
564 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
565 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
566 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
568 /* Fragment shaders use real, 32-bit twos-complement integers for all
571 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
572 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
573 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
574 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
575 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
577 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
578 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
579 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
580 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
581 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
583 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
584 * but we're not sure how it's actually done for vertex order,
585 * that affect provoking vertex decision. Always use last vertex
586 * convention for quad primitive which works as expected for now.
589 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
591 ctx
->Const
.NativeIntegers
= true;
592 ctx
->Const
.VertexID_is_zero_based
= true;
594 /* Regarding the CMP instruction, the Ivybridge PRM says:
596 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
597 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
598 * 0xFFFFFFFF) is assigned to dst."
600 * but PRMs for earlier generations say
602 * "In dword format, one GRF may store up to 8 results. When the register
603 * is used later as a vector of Booleans, as only LSB at each channel
604 * contains meaning [sic] data, software should make sure all higher bits
605 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
607 * We select the representation of a true boolean uniform to be ~0, and fix
608 * the results of Gen <= 5 CMP instruction's with -(result & 1).
610 ctx
->Const
.UniformBooleanTrue
= ~0;
612 /* From the gen4 PRM, volume 4 page 127:
614 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
615 * the base address of the first element of the surface, computed in
616 * software by adding the surface base address to the byte offset of
617 * the element in the buffer."
619 * However, unaligned accesses are slower, so enforce buffer alignment.
621 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
623 /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so
624 * that we can safely have the CPU and GPU writing the same SSBO on
625 * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never
626 * writes, so there's no problem. For an SSBO, the GPU and the CPU can
627 * be updating disjoint regions of the buffer simultaneously and that will
628 * break if the regions overlap the same cacheline.
630 ctx
->Const
.ShaderStorageBufferOffsetAlignment
= 64;
631 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
632 ctx
->Const
.MaxTextureBufferSize
= 128 * 1024 * 1024;
635 ctx
->Const
.MaxVarying
= 32;
636 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
637 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
638 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
639 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
640 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxInputComponents
= 128;
641 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxOutputComponents
= 128;
642 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxInputComponents
= 128;
643 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxOutputComponents
= 128;
646 /* We want the GLSL compiler to emit code that uses condition codes */
647 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
648 ctx
->Const
.ShaderCompilerOptions
[i
] =
649 brw
->intelScreen
->compiler
->glsl_compiler_options
[i
];
652 /* ARB_viewport_array */
653 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
654 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
655 ctx
->Const
.ViewportSubpixelBits
= 0;
657 /* Cast to float before negating because MaxViewportWidth is unsigned.
659 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
660 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
663 /* ARB_gpu_shader5 */
665 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
667 /* ARB_framebuffer_no_attachments */
668 ctx
->Const
.MaxFramebufferWidth
= ctx
->Const
.MaxViewportWidth
;
669 ctx
->Const
.MaxFramebufferHeight
= ctx
->Const
.MaxViewportHeight
;
670 ctx
->Const
.MaxFramebufferLayers
= ctx
->Const
.MaxArrayTextureLayers
;
671 ctx
->Const
.MaxFramebufferSamples
= max_samples
;
675 brw_initialize_cs_context_constants(struct brw_context
*brw
, unsigned max_threads
)
677 struct gl_context
*ctx
= &brw
->ctx
;
679 /* For ES, we set these constants based on SIMD8.
681 * TODO: Once we can always generate SIMD16, we should update this.
683 * For GL, we assume we can generate a SIMD16 program, but this currently
684 * is not always true. This allows us to run more test cases, and will be
685 * required based on desktop GL compute shader requirements.
687 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
689 const uint32_t max_invocations
= simd_size
* max_threads
;
690 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
691 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
692 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
693 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
697 * Process driconf (drirc) options, setting appropriate context flags.
699 * intelInitExtensions still pokes at optionCache directly, in order to
700 * avoid advertising various extensions. No flags are set, so it makes
701 * sense to continue doing that there.
704 brw_process_driconf_options(struct brw_context
*brw
)
706 struct gl_context
*ctx
= &brw
->ctx
;
708 driOptionCache
*options
= &brw
->optionCache
;
709 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
710 brw
->driContext
->driScreenPriv
->myNum
, "i965");
712 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
713 switch (bo_reuse_mode
) {
714 case DRI_CONF_BO_REUSE_DISABLED
:
716 case DRI_CONF_BO_REUSE_ALL
:
717 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
721 if (!driQueryOptionb(options
, "hiz")) {
722 brw
->has_hiz
= false;
723 /* On gen6, you can only do separate stencil with HIZ. */
725 brw
->has_separate_stencil
= false;
728 if (driQueryOptionb(options
, "always_flush_batch")) {
729 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
730 brw
->always_flush_batch
= true;
733 if (driQueryOptionb(options
, "always_flush_cache")) {
734 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
735 brw
->always_flush_cache
= true;
738 if (driQueryOptionb(options
, "disable_throttling")) {
739 fprintf(stderr
, "disabling flush throttling\n");
740 brw
->disable_throttling
= true;
743 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
745 ctx
->Const
.ForceGLSLExtensionsWarn
=
746 driQueryOptionb(options
, "force_glsl_extensions_warn");
748 ctx
->Const
.DisableGLSLLineContinuations
=
749 driQueryOptionb(options
, "disable_glsl_line_continuations");
751 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
752 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
756 brwCreateContext(gl_api api
,
757 const struct gl_config
*mesaVis
,
758 __DRIcontext
*driContextPriv
,
759 unsigned major_version
,
760 unsigned minor_version
,
763 unsigned *dri_ctx_error
,
764 void *sharedContextPrivate
)
766 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
767 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
768 struct intel_screen
*screen
= sPriv
->driverPrivate
;
769 const struct brw_device_info
*devinfo
= screen
->devinfo
;
770 struct dd_function_table functions
;
772 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
773 * provides us with context reset notifications.
775 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
776 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
778 if (screen
->has_context_reset_notification
)
779 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
781 if (flags
& ~allowed_flags
) {
782 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
786 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
788 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
789 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
793 driContextPriv
->driverPrivate
= brw
;
794 brw
->driContext
= driContextPriv
;
795 brw
->intelScreen
= screen
;
796 brw
->bufmgr
= screen
->bufmgr
;
798 brw
->gen
= devinfo
->gen
;
799 brw
->gt
= devinfo
->gt
;
800 brw
->is_g4x
= devinfo
->is_g4x
;
801 brw
->is_baytrail
= devinfo
->is_baytrail
;
802 brw
->is_haswell
= devinfo
->is_haswell
;
803 brw
->is_cherryview
= devinfo
->is_cherryview
;
804 brw
->is_broxton
= devinfo
->is_broxton
;
805 brw
->has_llc
= devinfo
->has_llc
;
806 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
807 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
808 brw
->has_pln
= devinfo
->has_pln
;
809 brw
->has_compr4
= devinfo
->has_compr4
;
810 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
811 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
812 brw
->needs_unlit_centroid_workaround
=
813 devinfo
->needs_unlit_centroid_workaround
;
815 brw
->must_use_separate_stencil
= devinfo
->must_use_separate_stencil
;
816 brw
->has_swizzling
= screen
->hw_has_swizzling
;
818 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
819 brw
->tcs
.base
.stage
= MESA_SHADER_TESS_CTRL
;
820 brw
->tes
.base
.stage
= MESA_SHADER_TESS_EVAL
;
821 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
822 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
824 gen8_init_vtable_surface_functions(brw
);
825 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
826 } else if (brw
->gen
>= 7) {
827 gen7_init_vtable_surface_functions(brw
);
828 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
829 } else if (brw
->gen
>= 6) {
830 gen6_init_vtable_surface_functions(brw
);
831 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
833 gen4_init_vtable_surface_functions(brw
);
834 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
837 brw_init_driver_functions(brw
, &functions
);
840 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
842 struct gl_context
*ctx
= &brw
->ctx
;
844 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
845 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
846 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
847 intelDestroyContext(driContextPriv
);
851 driContextSetFlags(ctx
, flags
);
853 /* Initialize the software rasterizer and helper modules.
855 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
856 * software fallbacks (which we have to support on legacy GL to do weird
857 * glDrawPixels(), glBitmap(), and other functions).
859 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
860 _swrast_CreateContext(ctx
);
863 _vbo_CreateContext(ctx
);
864 if (ctx
->swrast_context
) {
865 _tnl_CreateContext(ctx
);
866 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
867 _swsetup_CreateContext(ctx
);
869 /* Configure swrast to match hardware characteristics: */
870 _swrast_allow_pixel_fog(ctx
, false);
871 _swrast_allow_vertex_fog(ctx
, true);
874 _mesa_meta_init(ctx
);
876 brw_process_driconf_options(brw
);
878 if (INTEL_DEBUG
& DEBUG_PERF
)
879 brw
->perf_debug
= true;
881 brw_initialize_cs_context_constants(brw
, devinfo
->max_cs_threads
);
882 brw_initialize_context_constants(brw
);
884 ctx
->Const
.ResetStrategy
= notify_reset
885 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
887 /* Reinitialize the context point state. It depends on ctx->Const values. */
888 _mesa_init_point(ctx
);
892 intel_batchbuffer_init(brw
);
895 /* Create a new hardware context. Using a hardware context means that
896 * our GPU state will be saved/restored on context switch, allowing us
897 * to assume that the GPU is in the same state we left it in.
899 * This is required for transform feedback buffer offsets, query objects,
900 * and also allows us to reduce how much state we have to emit.
902 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
905 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
906 intelDestroyContext(driContextPriv
);
911 if (brw_init_pipe_control(brw
, devinfo
)) {
912 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
913 intelDestroyContext(driContextPriv
);
919 intelInitExtensions(ctx
);
921 brw_init_surface_formats(brw
);
923 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
924 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
925 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
926 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
927 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
928 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
929 brw
->urb
.size
= devinfo
->urb
.size
;
930 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
931 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
932 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
933 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
934 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
936 /* Estimate the size of the mappable aperture into the GTT. There's an
937 * ioctl to get the whole GTT size, but not one to get the mappable subset.
938 * It turns out it's basically always 256MB, though some ancient hardware
941 uint32_t gtt_size
= 256 * 1024 * 1024;
943 /* We don't want to map two objects such that a memcpy between them would
944 * just fault one mapping in and then the other over and over forever. So
945 * we would need to divide the GTT size by 2. Additionally, some GTT is
946 * taken up by things like the framebuffer and the ringbuffer and such, so
947 * be more conservative.
949 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
952 brw
->urb
.gs_present
= false;
954 brw
->prim_restart
.in_progress
= false;
955 brw
->prim_restart
.enable_cut_index
= false;
956 brw
->gs
.enabled
= false;
957 brw
->sf
.viewport_transform_enable
= true;
959 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
961 brw
->use_resource_streamer
= screen
->has_resource_streamer
&&
962 (env_var_as_boolean("INTEL_USE_HW_BT", false) ||
963 env_var_as_boolean("INTEL_USE_GATHER", false));
965 ctx
->VertexProgram
._MaintainTnlProgram
= true;
966 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
968 brw_draw_init( brw
);
970 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
971 /* Turn on some extra GL_ARB_debug_output generation. */
972 brw
->perf_debug
= true;
975 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
976 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
978 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
979 brw_init_shader_time(brw
);
981 _mesa_compute_version(ctx
);
983 _mesa_initialize_dispatch_tables(ctx
);
984 _mesa_initialize_vbo_vtxfmt(ctx
);
986 if (ctx
->Extensions
.AMD_performance_monitor
) {
987 brw_init_performance_monitors(brw
);
990 vbo_use_buffer_objects(ctx
);
991 vbo_always_unmap_buffers(ctx
);
997 intelDestroyContext(__DRIcontext
* driContextPriv
)
999 struct brw_context
*brw
=
1000 (struct brw_context
*) driContextPriv
->driverPrivate
;
1001 struct gl_context
*ctx
= &brw
->ctx
;
1003 /* Dump a final BMP in case the application doesn't call SwapBuffers */
1004 if (INTEL_DEBUG
& DEBUG_AUB
) {
1005 intel_batchbuffer_flush(brw
);
1006 aub_dump_bmp(&brw
->ctx
);
1009 _mesa_meta_free(&brw
->ctx
);
1010 brw_meta_fast_clear_free(brw
);
1012 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
1013 /* Force a report. */
1014 brw
->shader_time
.report_time
= 0;
1016 brw_collect_and_report_shader_time(brw
);
1017 brw_destroy_shader_time(brw
);
1020 brw_destroy_state(brw
);
1021 brw_draw_destroy(brw
);
1023 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
1024 if (brw
->vs
.base
.scratch_bo
)
1025 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
1026 if (brw
->gs
.base
.scratch_bo
)
1027 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
1028 if (brw
->wm
.base
.scratch_bo
)
1029 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
1031 gen7_reset_hw_bt_pool_offsets(brw
);
1032 drm_intel_bo_unreference(brw
->hw_bt_pool
.bo
);
1033 brw
->hw_bt_pool
.bo
= NULL
;
1035 drm_intel_gem_context_destroy(brw
->hw_ctx
);
1037 if (ctx
->swrast_context
) {
1038 _swsetup_DestroyContext(&brw
->ctx
);
1039 _tnl_DestroyContext(&brw
->ctx
);
1041 _vbo_DestroyContext(&brw
->ctx
);
1043 if (ctx
->swrast_context
)
1044 _swrast_DestroyContext(&brw
->ctx
);
1046 brw_fini_pipe_control(brw
);
1047 intel_batchbuffer_free(brw
);
1049 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
1050 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
1051 brw
->throttle_batch
[1] = NULL
;
1052 brw
->throttle_batch
[0] = NULL
;
1054 driDestroyOptionCache(&brw
->optionCache
);
1056 /* free the Mesa context */
1057 _mesa_free_context_data(&brw
->ctx
);
1060 driContextPriv
->driverPrivate
= NULL
;
1064 intelUnbindContext(__DRIcontext
* driContextPriv
)
1066 /* Unset current context and dispath table */
1067 _mesa_make_current(NULL
, NULL
, NULL
);
1073 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1074 * on window system framebuffers.
1076 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1077 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1078 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1079 * for a visual where you're guaranteed to be capable, but it turns out that
1080 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1081 * incapable ones, because there's no difference between the two in resources
1082 * used. Applications thus get built that accidentally rely on the default
1083 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1086 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1087 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1088 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1089 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1090 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1091 * and get no sRGB encode (assuming that both kinds of visual are available).
1092 * Thus our choice to support sRGB by default on our visuals for desktop would
1093 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1095 * Unfortunately, renderbuffer setup happens before a context is created. So
1096 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1097 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1098 * yet), we go turn that back off before anyone finds out.
1101 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1102 struct gl_framebuffer
*fb
)
1104 struct gl_context
*ctx
= &brw
->ctx
;
1106 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1109 /* Some day when we support the sRGB capable bit on visuals available for
1110 * GLES, we'll need to respect that and not disable things here.
1112 fb
->Visual
.sRGBCapable
= false;
1113 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1114 if (fb
->Attachment
[i
].Renderbuffer
&&
1115 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1116 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1122 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1123 __DRIdrawable
* driDrawPriv
,
1124 __DRIdrawable
* driReadPriv
)
1126 struct brw_context
*brw
;
1127 GET_CURRENT_CONTEXT(curCtx
);
1130 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1134 /* According to the glXMakeCurrent() man page: "Pending commands to
1135 * the previous context, if any, are flushed before it is released."
1136 * But only flush if we're actually changing contexts.
1138 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1139 _mesa_flush(curCtx
);
1142 if (driContextPriv
) {
1143 struct gl_context
*ctx
= &brw
->ctx
;
1144 struct gl_framebuffer
*fb
, *readFb
;
1146 if (driDrawPriv
== NULL
) {
1147 fb
= _mesa_get_incomplete_framebuffer();
1149 fb
= driDrawPriv
->driverPrivate
;
1150 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1153 if (driReadPriv
== NULL
) {
1154 readFb
= _mesa_get_incomplete_framebuffer();
1156 readFb
= driReadPriv
->driverPrivate
;
1157 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1160 /* The sRGB workaround changes the renderbuffer's format. We must change
1161 * the format before the renderbuffer's miptree get's allocated, otherwise
1162 * the formats of the renderbuffer and its miptree will differ.
1164 intel_gles3_srgb_workaround(brw
, fb
);
1165 intel_gles3_srgb_workaround(brw
, readFb
);
1167 /* If the context viewport hasn't been initialized, force a call out to
1168 * the loader to get buffers so we have a drawable size for the initial
1170 if (!brw
->ctx
.ViewportInitialized
)
1171 intel_prepare_render(brw
);
1173 _mesa_make_current(ctx
, fb
, readFb
);
1175 _mesa_make_current(NULL
, NULL
, NULL
);
1182 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1183 __DRIdrawable
*drawable
)
1186 /* MSAA and fast color clear are not supported, so don't waste time
1187 * checking whether a resolve is needed.
1192 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1193 struct intel_renderbuffer
*rb
;
1195 /* Usually, only the back buffer will need to be downsampled. However,
1196 * the front buffer will also need it if the user has rendered into it.
1198 static const gl_buffer_index buffers
[2] = {
1203 for (int i
= 0; i
< 2; ++i
) {
1204 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1205 if (rb
== NULL
|| rb
->mt
== NULL
)
1207 if (rb
->mt
->num_samples
<= 1)
1208 intel_miptree_resolve_color(brw
, rb
->mt
);
1210 intel_renderbuffer_downsample(brw
, rb
);
1215 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1217 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1221 intel_query_dri2_buffers(struct brw_context
*brw
,
1222 __DRIdrawable
*drawable
,
1223 __DRIbuffer
**buffers
,
1227 intel_process_dri2_buffer(struct brw_context
*brw
,
1228 __DRIdrawable
*drawable
,
1229 __DRIbuffer
*buffer
,
1230 struct intel_renderbuffer
*rb
,
1231 const char *buffer_name
);
1234 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1237 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1239 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1240 struct intel_renderbuffer
*rb
;
1241 __DRIbuffer
*buffers
= NULL
;
1243 const char *region_name
;
1245 /* Set this up front, so that in case our buffers get invalidated
1246 * while we're getting new buffers, we don't clobber the stamp and
1247 * thus ignore the invalidate. */
1248 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1250 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1251 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1253 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1255 if (buffers
== NULL
)
1258 for (i
= 0; i
< count
; i
++) {
1259 switch (buffers
[i
].attachment
) {
1260 case __DRI_BUFFER_FRONT_LEFT
:
1261 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1262 region_name
= "dri2 front buffer";
1265 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1266 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1267 region_name
= "dri2 fake front buffer";
1270 case __DRI_BUFFER_BACK_LEFT
:
1271 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1272 region_name
= "dri2 back buffer";
1275 case __DRI_BUFFER_DEPTH
:
1276 case __DRI_BUFFER_HIZ
:
1277 case __DRI_BUFFER_DEPTH_STENCIL
:
1278 case __DRI_BUFFER_STENCIL
:
1279 case __DRI_BUFFER_ACCUM
:
1282 "unhandled buffer attach event, attachment type %d\n",
1283 buffers
[i
].attachment
);
1287 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1293 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1295 struct brw_context
*brw
= context
->driverPrivate
;
1296 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1298 /* Set this up front, so that in case our buffers get invalidated
1299 * while we're getting new buffers, we don't clobber the stamp and
1300 * thus ignore the invalidate. */
1301 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1303 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1304 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1306 if (screen
->image
.loader
)
1307 intel_update_image_buffers(brw
, drawable
);
1309 intel_update_dri2_buffers(brw
, drawable
);
1311 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1315 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1316 * state is required.
1319 intel_prepare_render(struct brw_context
*brw
)
1321 struct gl_context
*ctx
= &brw
->ctx
;
1322 __DRIcontext
*driContext
= brw
->driContext
;
1323 __DRIdrawable
*drawable
;
1325 drawable
= driContext
->driDrawablePriv
;
1326 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1327 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1328 intel_update_renderbuffers(driContext
, drawable
);
1329 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1332 drawable
= driContext
->driReadablePriv
;
1333 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1334 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1335 intel_update_renderbuffers(driContext
, drawable
);
1336 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1339 /* If we're currently rendering to the front buffer, the rendering
1340 * that will happen next will probably dirty the front buffer. So
1341 * mark it as dirty here.
1343 if (_mesa_is_front_buffer_drawing(ctx
->DrawBuffer
))
1344 brw
->front_buffer_dirty
= true;
1348 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1350 * To determine which DRI buffers to request, examine the renderbuffers
1351 * attached to the drawable's framebuffer. Then request the buffers with
1352 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1354 * This is called from intel_update_renderbuffers().
1356 * \param drawable Drawable whose buffers are queried.
1357 * \param buffers [out] List of buffers returned by DRI2 query.
1358 * \param buffer_count [out] Number of buffers returned.
1360 * \see intel_update_renderbuffers()
1361 * \see DRI2GetBuffers()
1362 * \see DRI2GetBuffersWithFormat()
1365 intel_query_dri2_buffers(struct brw_context
*brw
,
1366 __DRIdrawable
*drawable
,
1367 __DRIbuffer
**buffers
,
1370 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1371 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1373 unsigned attachments
[8];
1375 struct intel_renderbuffer
*front_rb
;
1376 struct intel_renderbuffer
*back_rb
;
1378 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1379 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1381 memset(attachments
, 0, sizeof(attachments
));
1382 if ((_mesa_is_front_buffer_drawing(fb
) ||
1383 _mesa_is_front_buffer_reading(fb
) ||
1384 !back_rb
) && front_rb
) {
1385 /* If a fake front buffer is in use, then querying for
1386 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1387 * the real front buffer to the fake front buffer. So before doing the
1388 * query, we need to make sure all the pending drawing has landed in the
1389 * real front buffer.
1391 intel_batchbuffer_flush(brw
);
1392 intel_flush_front(&brw
->ctx
);
1394 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1395 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1396 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1397 /* We have pending front buffer rendering, but we aren't querying for a
1398 * front buffer. If the front buffer we have is a fake front buffer,
1399 * the X server is going to throw it away when it processes the query.
1400 * So before doing the query, make sure all the pending drawing has
1401 * landed in the real front buffer.
1403 intel_batchbuffer_flush(brw
);
1404 intel_flush_front(&brw
->ctx
);
1408 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1409 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1412 assert(i
<= ARRAY_SIZE(attachments
));
1414 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1419 drawable
->loaderPrivate
);
1423 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1425 * This is called from intel_update_renderbuffers().
1428 * DRI buffers whose attachment point is DRI2BufferStencil or
1429 * DRI2BufferDepthStencil are handled as special cases.
1431 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1432 * that is passed to drm_intel_bo_gem_create_from_name().
1434 * \see intel_update_renderbuffers()
1437 intel_process_dri2_buffer(struct brw_context
*brw
,
1438 __DRIdrawable
*drawable
,
1439 __DRIbuffer
*buffer
,
1440 struct intel_renderbuffer
*rb
,
1441 const char *buffer_name
)
1443 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1449 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1451 /* We try to avoid closing and reopening the same BO name, because the first
1452 * use of a mapping of the buffer involves a bunch of page faulting which is
1453 * moderately expensive.
1455 struct intel_mipmap_tree
*last_mt
;
1456 if (num_samples
== 0)
1459 last_mt
= rb
->singlesample_mt
;
1461 uint32_t old_name
= 0;
1463 /* The bo already has a name because the miptree was created by a
1464 * previous call to intel_process_dri2_buffer(). If a bo already has a
1465 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1466 * create a new name.
1468 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1471 if (old_name
== buffer
->name
)
1474 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1476 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1477 buffer
->name
, buffer
->attachment
,
1478 buffer
->cpp
, buffer
->pitch
);
1481 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1485 "Failed to open BO for returned DRI2 buffer "
1486 "(%dx%d, %s, named %d).\n"
1487 "This is likely a bug in the X Server that will lead to a "
1489 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1493 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1494 drawable
->w
, drawable
->h
,
1497 if (_mesa_is_front_buffer_drawing(fb
) &&
1498 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1499 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1500 rb
->Base
.Base
.NumSamples
> 1) {
1501 intel_renderbuffer_upsample(brw
, rb
);
1506 drm_intel_bo_unreference(bo
);
1510 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1512 * To determine which DRI buffers to request, examine the renderbuffers
1513 * attached to the drawable's framebuffer. Then request the buffers from
1516 * This is called from intel_update_renderbuffers().
1518 * \param drawable Drawable whose buffers are queried.
1519 * \param buffers [out] List of buffers returned by DRI2 query.
1520 * \param buffer_count [out] Number of buffers returned.
1522 * \see intel_update_renderbuffers()
1526 intel_update_image_buffer(struct brw_context
*intel
,
1527 __DRIdrawable
*drawable
,
1528 struct intel_renderbuffer
*rb
,
1530 enum __DRIimageBufferMask buffer_type
)
1532 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1534 if (!rb
|| !buffer
->bo
)
1537 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1539 /* Check and see if we're already bound to the right
1542 struct intel_mipmap_tree
*last_mt
;
1543 if (num_samples
== 0)
1546 last_mt
= rb
->singlesample_mt
;
1548 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1551 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1552 buffer
->width
, buffer
->height
,
1555 if (_mesa_is_front_buffer_drawing(fb
) &&
1556 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1557 rb
->Base
.Base
.NumSamples
> 1) {
1558 intel_renderbuffer_upsample(intel
, rb
);
1563 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1565 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1566 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1567 struct intel_renderbuffer
*front_rb
;
1568 struct intel_renderbuffer
*back_rb
;
1569 struct __DRIimageList images
;
1570 unsigned int format
;
1571 uint32_t buffer_mask
= 0;
1573 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1574 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1577 format
= intel_rb_format(back_rb
);
1579 format
= intel_rb_format(front_rb
);
1583 if (front_rb
&& (_mesa_is_front_buffer_drawing(fb
) ||
1584 _mesa_is_front_buffer_reading(fb
) || !back_rb
)) {
1585 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1589 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1591 (*screen
->image
.loader
->getBuffers
) (drawable
,
1592 driGLFormatToImageFormat(format
),
1593 &drawable
->dri2
.stamp
,
1594 drawable
->loaderPrivate
,
1598 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1599 drawable
->w
= images
.front
->width
;
1600 drawable
->h
= images
.front
->height
;
1601 intel_update_image_buffer(brw
,
1605 __DRI_IMAGE_BUFFER_FRONT
);
1607 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1608 drawable
->w
= images
.back
->width
;
1609 drawable
->h
= images
.back
->height
;
1610 intel_update_image_buffer(brw
,
1614 __DRI_IMAGE_BUFFER_BACK
);