2 Copyright 2003 VMware, Inc.
3 Copyright (C) Intel Corp. 2006. All Rights Reserved.
4 Intel funded Tungsten Graphics to
5 develop this 3D driver.
7 Permission is hereby granted, free of charge, to any person obtaining
8 a copy of this software and associated documentation files (the
9 "Software"), to deal in the Software without restriction, including
10 without limitation the rights to use, copy, modify, merge, publish,
11 distribute, sublicense, and/or sell copies of the Software, and to
12 permit persons to whom the Software is furnished to do so, subject to
13 the following conditions:
15 The above copyright notice and this permission notice (including the
16 next paragraph) shall be included in all copies or substantial
17 portions of the Software.
19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **********************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
34 #include "main/api_exec.h"
35 #include "main/context.h"
36 #include "main/fbobject.h"
37 #include "main/extensions.h"
38 #include "main/imports.h"
39 #include "main/macros.h"
40 #include "main/points.h"
41 #include "main/version.h"
42 #include "main/vtxfmt.h"
43 #include "main/texobj.h"
45 #include "vbo/vbo_context.h"
47 #include "drivers/common/driverfuncs.h"
48 #include "drivers/common/meta.h"
51 #include "brw_context.h"
52 #include "brw_defines.h"
54 #include "brw_state.h"
56 #include "intel_batchbuffer.h"
57 #include "intel_buffer_objects.h"
58 #include "intel_buffers.h"
59 #include "intel_fbo.h"
60 #include "intel_mipmap_tree.h"
61 #include "intel_pixel.h"
62 #include "intel_image.h"
63 #include "intel_tex.h"
64 #include "intel_tex_obj.h"
66 #include "swrast_setup/swrast_setup.h"
68 #include "tnl/t_pipeline.h"
69 #include "util/ralloc.h"
71 #include "glsl/nir/nir.h"
73 /***************************************
74 * Mesa's Driver Functions
75 ***************************************/
78 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
79 GLenum internalFormat
, int samples
[16])
81 struct brw_context
*brw
= brw_context(ctx
);
103 assert(brw
->gen
< 6);
109 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
112 brw_get_renderer_string(unsigned deviceID
)
115 static char buffer
[128];
119 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
120 #include "pci_ids/i965_pci_ids.h"
122 chipset
= "Unknown Intel Chipset";
126 (void) driGetRendererString(buffer
, chipset
, 0);
130 static const GLubyte
*
131 intel_get_string(struct gl_context
* ctx
, GLenum name
)
133 const struct brw_context
*const brw
= brw_context(ctx
);
137 return (GLubyte
*) brw_vendor_string
;
141 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
149 intel_viewport(struct gl_context
*ctx
)
151 struct brw_context
*brw
= brw_context(ctx
);
152 __DRIcontext
*driContext
= brw
->driContext
;
154 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
155 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
156 dri2InvalidateDrawable(driContext
->driReadablePriv
);
161 intel_update_state(struct gl_context
* ctx
, GLuint new_state
)
163 struct brw_context
*brw
= brw_context(ctx
);
164 struct intel_texture_object
*tex_obj
;
165 struct intel_renderbuffer
*depth_irb
;
167 if (ctx
->swrast_context
)
168 _swrast_InvalidateState(ctx
, new_state
);
169 _vbo_InvalidateState(ctx
, new_state
);
171 brw
->NewGLState
|= new_state
;
173 _mesa_unlock_context_textures(ctx
);
175 /* Resolve the depth buffer's HiZ buffer. */
176 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
178 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
180 /* Resolve depth buffer and render cache of each enabled texture. */
181 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
182 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
183 if (!ctx
->Texture
.Unit
[i
]._Current
)
185 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
186 if (!tex_obj
|| !tex_obj
->mt
)
188 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
189 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
190 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
193 _mesa_lock_context_textures(ctx
);
196 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
199 intel_flush_front(struct gl_context
*ctx
)
201 struct brw_context
*brw
= brw_context(ctx
);
202 __DRIcontext
*driContext
= brw
->driContext
;
203 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
204 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
206 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
207 if (flushFront(screen
) && driDrawable
&&
208 driDrawable
->loaderPrivate
) {
210 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
212 * This potentially resolves both front and back buffer. It
213 * is unnecessary to resolve the back, but harms nothing except
214 * performance. And no one cares about front-buffer render
217 intel_resolve_for_dri2_flush(brw
, driDrawable
);
218 intel_batchbuffer_flush(brw
);
220 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
222 /* We set the dirty bit in intel_prepare_render() if we're
223 * front buffer rendering once we get there.
225 brw
->front_buffer_dirty
= false;
231 intel_glFlush(struct gl_context
*ctx
)
233 struct brw_context
*brw
= brw_context(ctx
);
235 intel_batchbuffer_flush(brw
);
236 intel_flush_front(ctx
);
238 brw
->need_flush_throttle
= true;
242 intel_finish(struct gl_context
* ctx
)
244 struct brw_context
*brw
= brw_context(ctx
);
248 if (brw
->batch
.last_bo
)
249 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
253 brw_init_driver_functions(struct brw_context
*brw
,
254 struct dd_function_table
*functions
)
256 _mesa_init_driver_functions(functions
);
258 /* GLX uses DRI2 invalidate events to handle window resizing.
259 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
260 * which doesn't provide a mechanism for snooping the event queues.
262 * So EGL still relies on viewport hacks to handle window resizing.
263 * This should go away with DRI3000.
265 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
266 functions
->Viewport
= intel_viewport
;
268 functions
->Flush
= intel_glFlush
;
269 functions
->Finish
= intel_finish
;
270 functions
->GetString
= intel_get_string
;
271 functions
->UpdateState
= intel_update_state
;
273 intelInitTextureFuncs(functions
);
274 intelInitTextureImageFuncs(functions
);
275 intelInitTextureSubImageFuncs(functions
);
276 intelInitTextureCopyImageFuncs(functions
);
277 intelInitCopyImageFuncs(functions
);
278 intelInitClearFuncs(functions
);
279 intelInitBufferFuncs(functions
);
280 intelInitPixelFuncs(functions
);
281 intelInitBufferObjectFuncs(functions
);
282 intel_init_syncobj_functions(functions
);
283 brw_init_object_purgeable_functions(functions
);
285 brwInitFragProgFuncs( functions
);
286 brw_init_common_queryobj_functions(functions
);
288 gen6_init_queryobj_functions(functions
);
290 gen4_init_queryobj_functions(functions
);
292 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
294 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
295 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
296 functions
->GetTransformFeedbackVertexCount
=
297 brw_get_transform_feedback_vertex_count
;
299 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
300 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
301 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
302 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
304 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
305 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
309 functions
->GetSamplePosition
= gen6_get_sample_position
;
313 brw_initialize_context_constants(struct brw_context
*brw
)
315 struct gl_context
*ctx
= &brw
->ctx
;
317 unsigned max_samplers
=
318 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
320 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
322 ctx
->Const
.StripTextureBorder
= true;
324 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
325 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
326 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
327 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
328 ctx
->Const
.MaxTextureUnits
=
329 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
330 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
331 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
333 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
335 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
336 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
337 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
338 ctx
->Const
.MaxUniformBufferBindings
+= 12;
340 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
342 ctx
->Const
.MaxCombinedTextureImageUnits
=
343 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
344 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
345 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
346 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
348 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
349 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
350 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
351 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
352 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
353 ctx
->Const
.MaxTextureMbytes
= 1536;
356 ctx
->Const
.MaxArrayTextureLayers
= 2048;
358 ctx
->Const
.MaxArrayTextureLayers
= 512;
360 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
362 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
364 ctx
->Const
.MaxRenderbufferSize
= 8192;
366 /* Hardware only supports a limited number of transform feedback buffers.
367 * So we need to override the Mesa default (which is based only on software
370 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
372 /* On Gen6, in the worst case, we use up one binding table entry per
373 * transform feedback component (see comments above the definition of
374 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
375 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
376 * BRW_MAX_SOL_BINDINGS.
378 * In "separate components" mode, we need to divide this value by
379 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
380 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
382 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
383 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
384 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
386 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
389 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
390 const int clamp_max_samples
=
391 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
393 if (clamp_max_samples
< 0) {
394 max_samples
= msaa_modes
[0];
396 /* Select the largest supported MSAA mode that does not exceed
400 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
401 if (msaa_modes
[i
] <= clamp_max_samples
) {
402 max_samples
= msaa_modes
[i
];
408 ctx
->Const
.MaxSamples
= max_samples
;
409 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
410 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
411 ctx
->Const
.MaxIntegerSamples
= max_samples
;
413 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
414 * to map indices of rectangular grid to sample numbers within a pixel.
415 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
416 * extension implementation. For more details see the comment above
417 * gen6_set_sample_maps() definition.
419 gen6_set_sample_maps(ctx
);
422 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
423 else if (brw
->gen
== 6)
424 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
426 ctx
->Const
.MinLineWidth
= 1.0;
427 ctx
->Const
.MinLineWidthAA
= 1.0;
428 if (brw
->gen
>= 9 || brw
->is_cherryview
) {
429 ctx
->Const
.MaxLineWidth
= 40.0;
430 ctx
->Const
.MaxLineWidthAA
= 40.0;
431 ctx
->Const
.LineWidthGranularity
= 0.125;
432 } else if (brw
->gen
>= 6) {
433 ctx
->Const
.MaxLineWidth
= 7.375;
434 ctx
->Const
.MaxLineWidthAA
= 7.375;
435 ctx
->Const
.LineWidthGranularity
= 0.125;
437 ctx
->Const
.MaxLineWidth
= 7.0;
438 ctx
->Const
.MaxLineWidthAA
= 7.0;
439 ctx
->Const
.LineWidthGranularity
= 0.5;
442 ctx
->Const
.MinPointSize
= 1.0;
443 ctx
->Const
.MinPointSizeAA
= 1.0;
444 ctx
->Const
.MaxPointSize
= 255.0;
445 ctx
->Const
.MaxPointSizeAA
= 255.0;
446 ctx
->Const
.PointSizeGranularity
= 1.0;
448 if (brw
->gen
>= 5 || brw
->is_g4x
)
449 ctx
->Const
.MaxClipPlanes
= 8;
451 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
452 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
453 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
454 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
455 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
456 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
457 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
458 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
459 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
460 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
461 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
462 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
463 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
464 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
466 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
467 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
468 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
469 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
470 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
471 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
472 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
473 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
474 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
475 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
476 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
478 /* Fragment shaders use real, 32-bit twos-complement integers for all
481 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
482 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
483 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
484 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
485 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
487 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
488 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
489 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
490 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
491 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
494 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
495 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
496 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
497 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
498 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
499 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
500 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
501 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
502 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
505 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
506 * but we're not sure how it's actually done for vertex order,
507 * that affect provoking vertex decision. Always use last vertex
508 * convention for quad primitive which works as expected for now.
511 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
513 ctx
->Const
.NativeIntegers
= true;
514 ctx
->Const
.VertexID_is_zero_based
= true;
516 /* Regarding the CMP instruction, the Ivybridge PRM says:
518 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
519 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
520 * 0xFFFFFFFF) is assigned to dst."
522 * but PRMs for earlier generations say
524 * "In dword format, one GRF may store up to 8 results. When the register
525 * is used later as a vector of Booleans, as only LSB at each channel
526 * contains meaning [sic] data, software should make sure all higher bits
527 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
529 * We select the representation of a true boolean uniform to be ~0, and fix
530 * the results of Gen <= 5 CMP instruction's with -(result & 1).
532 ctx
->Const
.UniformBooleanTrue
= ~0;
534 /* From the gen4 PRM, volume 4 page 127:
536 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
537 * the base address of the first element of the surface, computed in
538 * software by adding the surface base address to the byte offset of
539 * the element in the buffer."
541 * However, unaligned accesses are slower, so enforce buffer alignment.
543 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
544 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
547 ctx
->Const
.MaxVarying
= 32;
548 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
549 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
550 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
551 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
554 static const nir_shader_compiler_options nir_options
= {
555 .native_integers
= true,
556 /* In order to help allow for better CSE at the NIR level we tell NIR
557 * to split all ffma instructions during opt_algebraic and we then
558 * re-combine them as a later step.
563 /* We want the GLSL compiler to emit code that uses condition codes */
564 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
565 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
566 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
567 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
568 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
569 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
570 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
571 (i
== MESA_SHADER_FRAGMENT
);
572 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
573 (i
== MESA_SHADER_FRAGMENT
);
574 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
575 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
576 ctx
->Const
.ShaderCompilerOptions
[i
].NirOptions
= &nir_options
;
579 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
580 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
582 if (brw
->scalar_vs
) {
583 /* If we're using the scalar backend for vertex shaders, we need to
584 * configure these accordingly.
586 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectOutput
= true;
587 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectTemp
= true;
588 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= false;
591 /* ARB_viewport_array */
592 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
593 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
594 ctx
->Const
.ViewportSubpixelBits
= 0;
596 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
598 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
599 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
602 /* ARB_gpu_shader5 */
604 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
608 * Process driconf (drirc) options, setting appropriate context flags.
610 * intelInitExtensions still pokes at optionCache directly, in order to
611 * avoid advertising various extensions. No flags are set, so it makes
612 * sense to continue doing that there.
615 brw_process_driconf_options(struct brw_context
*brw
)
617 struct gl_context
*ctx
= &brw
->ctx
;
619 driOptionCache
*options
= &brw
->optionCache
;
620 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
621 brw
->driContext
->driScreenPriv
->myNum
, "i965");
623 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
624 switch (bo_reuse_mode
) {
625 case DRI_CONF_BO_REUSE_DISABLED
:
627 case DRI_CONF_BO_REUSE_ALL
:
628 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
632 if (!driQueryOptionb(options
, "hiz")) {
633 brw
->has_hiz
= false;
634 /* On gen6, you can only do separate stencil with HIZ. */
636 brw
->has_separate_stencil
= false;
639 if (driQueryOptionb(options
, "always_flush_batch")) {
640 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
641 brw
->always_flush_batch
= true;
644 if (driQueryOptionb(options
, "always_flush_cache")) {
645 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
646 brw
->always_flush_cache
= true;
649 if (driQueryOptionb(options
, "disable_throttling")) {
650 fprintf(stderr
, "disabling flush throttling\n");
651 brw
->disable_throttling
= true;
654 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
656 ctx
->Const
.ForceGLSLExtensionsWarn
=
657 driQueryOptionb(options
, "force_glsl_extensions_warn");
659 ctx
->Const
.DisableGLSLLineContinuations
=
660 driQueryOptionb(options
, "disable_glsl_line_continuations");
662 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
663 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
666 /* drop when libdrm 2.4.61 is released */
667 #ifndef I915_PARAM_REVISION
668 #define I915_PARAM_REVISION 32
672 brw_get_revision(int fd
)
674 struct drm_i915_getparam gp
;
678 memset(&gp
, 0, sizeof(gp
));
679 gp
.param
= I915_PARAM_REVISION
;
680 gp
.value
= &revision
;
682 ret
= drmCommandWriteRead(fd
, DRM_I915_GETPARAM
, &gp
, sizeof(gp
));
690 brwCreateContext(gl_api api
,
691 const struct gl_config
*mesaVis
,
692 __DRIcontext
*driContextPriv
,
693 unsigned major_version
,
694 unsigned minor_version
,
697 unsigned *dri_ctx_error
,
698 void *sharedContextPrivate
)
700 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
701 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
702 struct intel_screen
*screen
= sPriv
->driverPrivate
;
703 const struct brw_device_info
*devinfo
= screen
->devinfo
;
704 struct dd_function_table functions
;
706 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
707 * provides us with context reset notifications.
709 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
710 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
712 if (screen
->has_context_reset_notification
)
713 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
715 if (flags
& ~allowed_flags
) {
716 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
720 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
722 fprintf(stderr
, "%s: failed to alloc context\n", __FUNCTION__
);
723 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
727 driContextPriv
->driverPrivate
= brw
;
728 brw
->driContext
= driContextPriv
;
729 brw
->intelScreen
= screen
;
730 brw
->bufmgr
= screen
->bufmgr
;
732 brw
->gen
= devinfo
->gen
;
733 brw
->gt
= devinfo
->gt
;
734 brw
->is_g4x
= devinfo
->is_g4x
;
735 brw
->is_baytrail
= devinfo
->is_baytrail
;
736 brw
->is_haswell
= devinfo
->is_haswell
;
737 brw
->is_cherryview
= devinfo
->is_cherryview
;
738 brw
->has_llc
= devinfo
->has_llc
;
739 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
740 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
741 brw
->has_pln
= devinfo
->has_pln
;
742 brw
->has_compr4
= devinfo
->has_compr4
;
743 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
744 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
745 brw
->needs_unlit_centroid_workaround
=
746 devinfo
->needs_unlit_centroid_workaround
;
747 brw
->revision
= brw_get_revision(sPriv
->fd
);
749 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
750 brw
->has_swizzling
= screen
->hw_has_swizzling
;
752 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
753 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
754 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
756 gen8_init_vtable_surface_functions(brw
);
757 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
758 } else if (brw
->gen
>= 7) {
759 gen7_init_vtable_surface_functions(brw
);
760 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
761 } else if (brw
->gen
>= 6) {
762 gen6_init_vtable_surface_functions(brw
);
763 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
765 gen4_init_vtable_surface_functions(brw
);
766 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
769 brw_init_driver_functions(brw
, &functions
);
772 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
774 struct gl_context
*ctx
= &brw
->ctx
;
776 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
777 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
778 fprintf(stderr
, "%s: failed to init mesa context\n", __FUNCTION__
);
779 intelDestroyContext(driContextPriv
);
783 driContextSetFlags(ctx
, flags
);
785 /* Initialize the software rasterizer and helper modules.
787 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
788 * software fallbacks (which we have to support on legacy GL to do weird
789 * glDrawPixels(), glBitmap(), and other functions).
791 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
792 _swrast_CreateContext(ctx
);
795 _vbo_CreateContext(ctx
);
796 if (ctx
->swrast_context
) {
797 _tnl_CreateContext(ctx
);
798 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
799 _swsetup_CreateContext(ctx
);
801 /* Configure swrast to match hardware characteristics: */
802 _swrast_allow_pixel_fog(ctx
, false);
803 _swrast_allow_vertex_fog(ctx
, true);
806 _mesa_meta_init(ctx
);
808 brw_process_driconf_options(brw
);
809 brw_process_intel_debug_variable(brw
);
811 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
812 brw
->scalar_vs
= true;
814 brw_initialize_context_constants(brw
);
816 ctx
->Const
.ResetStrategy
= notify_reset
817 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
819 /* Reinitialize the context point state. It depends on ctx->Const values. */
820 _mesa_init_point(ctx
);
824 intel_batchbuffer_init(brw
);
827 /* Create a new hardware context. Using a hardware context means that
828 * our GPU state will be saved/restored on context switch, allowing us
829 * to assume that the GPU is in the same state we left it in.
831 * This is required for transform feedback buffer offsets, query objects,
832 * and also allows us to reduce how much state we have to emit.
834 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
837 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
838 intelDestroyContext(driContextPriv
);
845 intelInitExtensions(ctx
);
847 brw_init_surface_formats(brw
);
849 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
850 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
851 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
852 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
853 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
854 brw
->urb
.size
= devinfo
->urb
.size
;
855 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
856 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
857 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
858 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
859 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
861 /* Estimate the size of the mappable aperture into the GTT. There's an
862 * ioctl to get the whole GTT size, but not one to get the mappable subset.
863 * It turns out it's basically always 256MB, though some ancient hardware
866 uint32_t gtt_size
= 256 * 1024 * 1024;
868 /* We don't want to map two objects such that a memcpy between them would
869 * just fault one mapping in and then the other over and over forever. So
870 * we would need to divide the GTT size by 2. Additionally, some GTT is
871 * taken up by things like the framebuffer and the ringbuffer and such, so
872 * be more conservative.
874 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
877 brw
->urb
.gs_present
= false;
879 brw
->prim_restart
.in_progress
= false;
880 brw
->prim_restart
.enable_cut_index
= false;
881 brw
->gs
.enabled
= false;
882 brw
->sf
.viewport_transform_enable
= true;
884 ctx
->VertexProgram
._MaintainTnlProgram
= true;
885 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
887 brw_draw_init( brw
);
889 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
890 /* Turn on some extra GL_ARB_debug_output generation. */
891 brw
->perf_debug
= true;
894 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
895 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
897 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
898 brw_init_shader_time(brw
);
900 _mesa_compute_version(ctx
);
902 _mesa_initialize_dispatch_tables(ctx
);
903 _mesa_initialize_vbo_vtxfmt(ctx
);
905 if (ctx
->Extensions
.AMD_performance_monitor
) {
906 brw_init_performance_monitors(brw
);
909 vbo_use_buffer_objects(ctx
);
910 vbo_always_unmap_buffers(ctx
);
916 intelDestroyContext(__DRIcontext
* driContextPriv
)
918 struct brw_context
*brw
=
919 (struct brw_context
*) driContextPriv
->driverPrivate
;
920 struct gl_context
*ctx
= &brw
->ctx
;
922 assert(brw
); /* should never be null */
926 /* Dump a final BMP in case the application doesn't call SwapBuffers */
927 if (INTEL_DEBUG
& DEBUG_AUB
) {
928 intel_batchbuffer_flush(brw
);
929 aub_dump_bmp(&brw
->ctx
);
932 _mesa_meta_free(&brw
->ctx
);
933 brw_meta_fast_clear_free(brw
);
935 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
936 /* Force a report. */
937 brw
->shader_time
.report_time
= 0;
939 brw_collect_and_report_shader_time(brw
);
940 brw_destroy_shader_time(brw
);
943 brw_destroy_state(brw
);
944 brw_draw_destroy(brw
);
946 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
947 if (brw
->vs
.base
.scratch_bo
)
948 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
949 if (brw
->gs
.base
.scratch_bo
)
950 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
951 if (brw
->wm
.base
.scratch_bo
)
952 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
954 drm_intel_gem_context_destroy(brw
->hw_ctx
);
956 if (ctx
->swrast_context
) {
957 _swsetup_DestroyContext(&brw
->ctx
);
958 _tnl_DestroyContext(&brw
->ctx
);
960 _vbo_DestroyContext(&brw
->ctx
);
962 if (ctx
->swrast_context
)
963 _swrast_DestroyContext(&brw
->ctx
);
965 intel_batchbuffer_free(brw
);
967 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
968 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
969 brw
->throttle_batch
[1] = NULL
;
970 brw
->throttle_batch
[0] = NULL
;
972 driDestroyOptionCache(&brw
->optionCache
);
974 /* free the Mesa context */
975 _mesa_free_context_data(&brw
->ctx
);
978 driContextPriv
->driverPrivate
= NULL
;
982 intelUnbindContext(__DRIcontext
* driContextPriv
)
984 /* Unset current context and dispath table */
985 _mesa_make_current(NULL
, NULL
, NULL
);
991 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
992 * on window system framebuffers.
994 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
995 * your renderbuffer can do sRGB encode, and you can flip a switch that does
996 * sRGB encode if the renderbuffer can handle it. You can ask specifically
997 * for a visual where you're guaranteed to be capable, but it turns out that
998 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
999 * incapable ones, becuase there's no difference between the two in resources
1000 * used. Applications thus get built that accidentally rely on the default
1001 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1004 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1005 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1006 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1007 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1008 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1009 * and get no sRGB encode (assuming that both kinds of visual are available).
1010 * Thus our choice to support sRGB by default on our visuals for desktop would
1011 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1013 * Unfortunately, renderbuffer setup happens before a context is created. So
1014 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1015 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1016 * yet), we go turn that back off before anyone finds out.
1019 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1020 struct gl_framebuffer
*fb
)
1022 struct gl_context
*ctx
= &brw
->ctx
;
1024 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1027 /* Some day when we support the sRGB capable bit on visuals available for
1028 * GLES, we'll need to respect that and not disable things here.
1030 fb
->Visual
.sRGBCapable
= false;
1031 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1032 if (fb
->Attachment
[i
].Renderbuffer
&&
1033 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1034 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1040 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1041 __DRIdrawable
* driDrawPriv
,
1042 __DRIdrawable
* driReadPriv
)
1044 struct brw_context
*brw
;
1045 GET_CURRENT_CONTEXT(curCtx
);
1048 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1052 /* According to the glXMakeCurrent() man page: "Pending commands to
1053 * the previous context, if any, are flushed before it is released."
1054 * But only flush if we're actually changing contexts.
1056 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1057 _mesa_flush(curCtx
);
1060 if (driContextPriv
) {
1061 struct gl_context
*ctx
= &brw
->ctx
;
1062 struct gl_framebuffer
*fb
, *readFb
;
1064 if (driDrawPriv
== NULL
) {
1065 fb
= _mesa_get_incomplete_framebuffer();
1067 fb
= driDrawPriv
->driverPrivate
;
1068 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1071 if (driReadPriv
== NULL
) {
1072 readFb
= _mesa_get_incomplete_framebuffer();
1074 readFb
= driReadPriv
->driverPrivate
;
1075 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1078 /* The sRGB workaround changes the renderbuffer's format. We must change
1079 * the format before the renderbuffer's miptree get's allocated, otherwise
1080 * the formats of the renderbuffer and its miptree will differ.
1082 intel_gles3_srgb_workaround(brw
, fb
);
1083 intel_gles3_srgb_workaround(brw
, readFb
);
1085 /* If the context viewport hasn't been initialized, force a call out to
1086 * the loader to get buffers so we have a drawable size for the initial
1088 if (!brw
->ctx
.ViewportInitialized
)
1089 intel_prepare_render(brw
);
1091 _mesa_make_current(ctx
, fb
, readFb
);
1093 _mesa_make_current(NULL
, NULL
, NULL
);
1100 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1101 __DRIdrawable
*drawable
)
1104 /* MSAA and fast color clear are not supported, so don't waste time
1105 * checking whether a resolve is needed.
1110 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1111 struct intel_renderbuffer
*rb
;
1113 /* Usually, only the back buffer will need to be downsampled. However,
1114 * the front buffer will also need it if the user has rendered into it.
1116 static const gl_buffer_index buffers
[2] = {
1121 for (int i
= 0; i
< 2; ++i
) {
1122 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1123 if (rb
== NULL
|| rb
->mt
== NULL
)
1125 if (rb
->mt
->num_samples
<= 1)
1126 intel_miptree_resolve_color(brw
, rb
->mt
);
1128 intel_renderbuffer_downsample(brw
, rb
);
1133 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1135 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1139 intel_query_dri2_buffers(struct brw_context
*brw
,
1140 __DRIdrawable
*drawable
,
1141 __DRIbuffer
**buffers
,
1145 intel_process_dri2_buffer(struct brw_context
*brw
,
1146 __DRIdrawable
*drawable
,
1147 __DRIbuffer
*buffer
,
1148 struct intel_renderbuffer
*rb
,
1149 const char *buffer_name
);
1152 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1155 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1157 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1158 struct intel_renderbuffer
*rb
;
1159 __DRIbuffer
*buffers
= NULL
;
1161 const char *region_name
;
1163 /* Set this up front, so that in case our buffers get invalidated
1164 * while we're getting new buffers, we don't clobber the stamp and
1165 * thus ignore the invalidate. */
1166 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1168 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1169 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1171 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1173 if (buffers
== NULL
)
1176 for (i
= 0; i
< count
; i
++) {
1177 switch (buffers
[i
].attachment
) {
1178 case __DRI_BUFFER_FRONT_LEFT
:
1179 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1180 region_name
= "dri2 front buffer";
1183 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1184 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1185 region_name
= "dri2 fake front buffer";
1188 case __DRI_BUFFER_BACK_LEFT
:
1189 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1190 region_name
= "dri2 back buffer";
1193 case __DRI_BUFFER_DEPTH
:
1194 case __DRI_BUFFER_HIZ
:
1195 case __DRI_BUFFER_DEPTH_STENCIL
:
1196 case __DRI_BUFFER_STENCIL
:
1197 case __DRI_BUFFER_ACCUM
:
1200 "unhandled buffer attach event, attachment type %d\n",
1201 buffers
[i
].attachment
);
1205 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1211 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1213 struct brw_context
*brw
= context
->driverPrivate
;
1214 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1216 /* Set this up front, so that in case our buffers get invalidated
1217 * while we're getting new buffers, we don't clobber the stamp and
1218 * thus ignore the invalidate. */
1219 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1221 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1222 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1224 if (screen
->image
.loader
)
1225 intel_update_image_buffers(brw
, drawable
);
1227 intel_update_dri2_buffers(brw
, drawable
);
1229 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1233 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1234 * state is required.
1237 intel_prepare_render(struct brw_context
*brw
)
1239 struct gl_context
*ctx
= &brw
->ctx
;
1240 __DRIcontext
*driContext
= brw
->driContext
;
1241 __DRIdrawable
*drawable
;
1243 drawable
= driContext
->driDrawablePriv
;
1244 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1245 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1246 intel_update_renderbuffers(driContext
, drawable
);
1247 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1250 drawable
= driContext
->driReadablePriv
;
1251 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1252 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1253 intel_update_renderbuffers(driContext
, drawable
);
1254 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1257 /* If we're currently rendering to the front buffer, the rendering
1258 * that will happen next will probably dirty the front buffer. So
1259 * mark it as dirty here.
1261 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1262 brw
->front_buffer_dirty
= true;
1266 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1268 * To determine which DRI buffers to request, examine the renderbuffers
1269 * attached to the drawable's framebuffer. Then request the buffers with
1270 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1272 * This is called from intel_update_renderbuffers().
1274 * \param drawable Drawable whose buffers are queried.
1275 * \param buffers [out] List of buffers returned by DRI2 query.
1276 * \param buffer_count [out] Number of buffers returned.
1278 * \see intel_update_renderbuffers()
1279 * \see DRI2GetBuffers()
1280 * \see DRI2GetBuffersWithFormat()
1283 intel_query_dri2_buffers(struct brw_context
*brw
,
1284 __DRIdrawable
*drawable
,
1285 __DRIbuffer
**buffers
,
1288 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1289 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1291 unsigned attachments
[8];
1293 struct intel_renderbuffer
*front_rb
;
1294 struct intel_renderbuffer
*back_rb
;
1296 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1297 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1299 memset(attachments
, 0, sizeof(attachments
));
1300 if ((brw_is_front_buffer_drawing(fb
) ||
1301 brw_is_front_buffer_reading(fb
) ||
1302 !back_rb
) && front_rb
) {
1303 /* If a fake front buffer is in use, then querying for
1304 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1305 * the real front buffer to the fake front buffer. So before doing the
1306 * query, we need to make sure all the pending drawing has landed in the
1307 * real front buffer.
1309 intel_batchbuffer_flush(brw
);
1310 intel_flush_front(&brw
->ctx
);
1312 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1313 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1314 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1315 /* We have pending front buffer rendering, but we aren't querying for a
1316 * front buffer. If the front buffer we have is a fake front buffer,
1317 * the X server is going to throw it away when it processes the query.
1318 * So before doing the query, make sure all the pending drawing has
1319 * landed in the real front buffer.
1321 intel_batchbuffer_flush(brw
);
1322 intel_flush_front(&brw
->ctx
);
1326 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1327 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1330 assert(i
<= ARRAY_SIZE(attachments
));
1332 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1337 drawable
->loaderPrivate
);
1341 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1343 * This is called from intel_update_renderbuffers().
1346 * DRI buffers whose attachment point is DRI2BufferStencil or
1347 * DRI2BufferDepthStencil are handled as special cases.
1349 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1350 * that is passed to drm_intel_bo_gem_create_from_name().
1352 * \see intel_update_renderbuffers()
1355 intel_process_dri2_buffer(struct brw_context
*brw
,
1356 __DRIdrawable
*drawable
,
1357 __DRIbuffer
*buffer
,
1358 struct intel_renderbuffer
*rb
,
1359 const char *buffer_name
)
1361 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1367 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1369 /* We try to avoid closing and reopening the same BO name, because the first
1370 * use of a mapping of the buffer involves a bunch of page faulting which is
1371 * moderately expensive.
1373 struct intel_mipmap_tree
*last_mt
;
1374 if (num_samples
== 0)
1377 last_mt
= rb
->singlesample_mt
;
1379 uint32_t old_name
= 0;
1381 /* The bo already has a name because the miptree was created by a
1382 * previous call to intel_process_dri2_buffer(). If a bo already has a
1383 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1384 * create a new name.
1386 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1389 if (old_name
== buffer
->name
)
1392 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1394 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1395 buffer
->name
, buffer
->attachment
,
1396 buffer
->cpp
, buffer
->pitch
);
1399 intel_miptree_release(&rb
->mt
);
1400 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1404 "Failed to open BO for returned DRI2 buffer "
1405 "(%dx%d, %s, named %d).\n"
1406 "This is likely a bug in the X Server that will lead to a "
1408 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1412 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1413 drawable
->w
, drawable
->h
,
1416 if (brw_is_front_buffer_drawing(fb
) &&
1417 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1418 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1419 rb
->Base
.Base
.NumSamples
> 1) {
1420 intel_renderbuffer_upsample(brw
, rb
);
1425 drm_intel_bo_unreference(bo
);
1429 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1431 * To determine which DRI buffers to request, examine the renderbuffers
1432 * attached to the drawable's framebuffer. Then request the buffers from
1435 * This is called from intel_update_renderbuffers().
1437 * \param drawable Drawable whose buffers are queried.
1438 * \param buffers [out] List of buffers returned by DRI2 query.
1439 * \param buffer_count [out] Number of buffers returned.
1441 * \see intel_update_renderbuffers()
1445 intel_update_image_buffer(struct brw_context
*intel
,
1446 __DRIdrawable
*drawable
,
1447 struct intel_renderbuffer
*rb
,
1449 enum __DRIimageBufferMask buffer_type
)
1451 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1453 if (!rb
|| !buffer
->bo
)
1456 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1458 /* Check and see if we're already bound to the right
1461 struct intel_mipmap_tree
*last_mt
;
1462 if (num_samples
== 0)
1465 last_mt
= rb
->singlesample_mt
;
1467 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1470 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1471 buffer
->width
, buffer
->height
,
1474 if (brw_is_front_buffer_drawing(fb
) &&
1475 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1476 rb
->Base
.Base
.NumSamples
> 1) {
1477 intel_renderbuffer_upsample(intel
, rb
);
1482 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1484 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1485 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1486 struct intel_renderbuffer
*front_rb
;
1487 struct intel_renderbuffer
*back_rb
;
1488 struct __DRIimageList images
;
1489 unsigned int format
;
1490 uint32_t buffer_mask
= 0;
1492 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1493 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1496 format
= intel_rb_format(back_rb
);
1498 format
= intel_rb_format(front_rb
);
1502 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1503 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1504 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1508 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1510 (*screen
->image
.loader
->getBuffers
) (drawable
,
1511 driGLFormatToImageFormat(format
),
1512 &drawable
->dri2
.stamp
,
1513 drawable
->loaderPrivate
,
1517 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1518 drawable
->w
= images
.front
->width
;
1519 drawable
->h
= images
.front
->height
;
1520 intel_update_image_buffer(brw
,
1524 __DRI_IMAGE_BUFFER_FRONT
);
1526 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1527 drawable
->w
= images
.back
->width
;
1528 drawable
->h
= images
.back
->height
;
1529 intel_update_image_buffer(brw
,
1533 __DRI_IMAGE_BUFFER_BACK
);