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
);
237 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
238 brw
->need_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,
558 /* We want the GLSL compiler to emit code that uses condition codes */
559 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
560 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
561 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
562 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
563 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
564 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
565 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
566 (i
== MESA_SHADER_FRAGMENT
);
567 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
568 (i
== MESA_SHADER_FRAGMENT
);
569 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
570 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
571 ctx
->Const
.ShaderCompilerOptions
[i
].NirOptions
= &nir_options
;
574 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
575 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
577 if (brw
->scalar_vs
) {
578 /* If we're using the scalar backend for vertex shaders, we need to
579 * configure these accordingly.
581 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectOutput
= true;
582 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectTemp
= true;
583 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= false;
586 /* ARB_viewport_array */
587 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
588 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
589 ctx
->Const
.ViewportSubpixelBits
= 0;
591 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
593 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
594 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
597 /* ARB_gpu_shader5 */
599 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
603 * Process driconf (drirc) options, setting appropriate context flags.
605 * intelInitExtensions still pokes at optionCache directly, in order to
606 * avoid advertising various extensions. No flags are set, so it makes
607 * sense to continue doing that there.
610 brw_process_driconf_options(struct brw_context
*brw
)
612 struct gl_context
*ctx
= &brw
->ctx
;
614 driOptionCache
*options
= &brw
->optionCache
;
615 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
616 brw
->driContext
->driScreenPriv
->myNum
, "i965");
618 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
619 switch (bo_reuse_mode
) {
620 case DRI_CONF_BO_REUSE_DISABLED
:
622 case DRI_CONF_BO_REUSE_ALL
:
623 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
627 if (!driQueryOptionb(options
, "hiz")) {
628 brw
->has_hiz
= false;
629 /* On gen6, you can only do separate stencil with HIZ. */
631 brw
->has_separate_stencil
= false;
634 if (driQueryOptionb(options
, "always_flush_batch")) {
635 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
636 brw
->always_flush_batch
= true;
639 if (driQueryOptionb(options
, "always_flush_cache")) {
640 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
641 brw
->always_flush_cache
= true;
644 if (driQueryOptionb(options
, "disable_throttling")) {
645 fprintf(stderr
, "disabling flush throttling\n");
646 brw
->disable_throttling
= true;
649 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
651 ctx
->Const
.ForceGLSLExtensionsWarn
=
652 driQueryOptionb(options
, "force_glsl_extensions_warn");
654 ctx
->Const
.DisableGLSLLineContinuations
=
655 driQueryOptionb(options
, "disable_glsl_line_continuations");
657 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
658 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
662 brwCreateContext(gl_api api
,
663 const struct gl_config
*mesaVis
,
664 __DRIcontext
*driContextPriv
,
665 unsigned major_version
,
666 unsigned minor_version
,
669 unsigned *dri_ctx_error
,
670 void *sharedContextPrivate
)
672 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
673 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
674 struct intel_screen
*screen
= sPriv
->driverPrivate
;
675 const struct brw_device_info
*devinfo
= screen
->devinfo
;
676 struct dd_function_table functions
;
678 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
679 * provides us with context reset notifications.
681 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
682 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
684 if (screen
->has_context_reset_notification
)
685 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
687 if (flags
& ~allowed_flags
) {
688 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
692 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
694 fprintf(stderr
, "%s: failed to alloc context\n", __FUNCTION__
);
695 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
699 driContextPriv
->driverPrivate
= brw
;
700 brw
->driContext
= driContextPriv
;
701 brw
->intelScreen
= screen
;
702 brw
->bufmgr
= screen
->bufmgr
;
704 brw
->gen
= devinfo
->gen
;
705 brw
->gt
= devinfo
->gt
;
706 brw
->is_g4x
= devinfo
->is_g4x
;
707 brw
->is_baytrail
= devinfo
->is_baytrail
;
708 brw
->is_haswell
= devinfo
->is_haswell
;
709 brw
->is_cherryview
= devinfo
->is_cherryview
;
710 brw
->has_llc
= devinfo
->has_llc
;
711 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
712 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
713 brw
->has_pln
= devinfo
->has_pln
;
714 brw
->has_compr4
= devinfo
->has_compr4
;
715 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
716 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
717 brw
->needs_unlit_centroid_workaround
=
718 devinfo
->needs_unlit_centroid_workaround
;
720 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
721 brw
->has_swizzling
= screen
->hw_has_swizzling
;
723 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
724 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
725 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
727 gen8_init_vtable_surface_functions(brw
);
728 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
729 } else if (brw
->gen
>= 7) {
730 gen7_init_vtable_surface_functions(brw
);
731 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
732 } else if (brw
->gen
>= 6) {
733 gen6_init_vtable_surface_functions(brw
);
734 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
736 gen4_init_vtable_surface_functions(brw
);
737 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
740 brw_init_driver_functions(brw
, &functions
);
743 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
745 struct gl_context
*ctx
= &brw
->ctx
;
747 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
748 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
749 fprintf(stderr
, "%s: failed to init mesa context\n", __FUNCTION__
);
750 intelDestroyContext(driContextPriv
);
754 driContextSetFlags(ctx
, flags
);
756 /* Initialize the software rasterizer and helper modules.
758 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
759 * software fallbacks (which we have to support on legacy GL to do weird
760 * glDrawPixels(), glBitmap(), and other functions).
762 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
763 _swrast_CreateContext(ctx
);
766 _vbo_CreateContext(ctx
);
767 if (ctx
->swrast_context
) {
768 _tnl_CreateContext(ctx
);
769 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
770 _swsetup_CreateContext(ctx
);
772 /* Configure swrast to match hardware characteristics: */
773 _swrast_allow_pixel_fog(ctx
, false);
774 _swrast_allow_vertex_fog(ctx
, true);
777 _mesa_meta_init(ctx
);
779 brw_process_driconf_options(brw
);
780 brw_process_intel_debug_variable(brw
);
782 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
783 brw
->scalar_vs
= true;
785 brw_initialize_context_constants(brw
);
787 ctx
->Const
.ResetStrategy
= notify_reset
788 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
790 /* Reinitialize the context point state. It depends on ctx->Const values. */
791 _mesa_init_point(ctx
);
795 intel_batchbuffer_init(brw
);
798 /* Create a new hardware context. Using a hardware context means that
799 * our GPU state will be saved/restored on context switch, allowing us
800 * to assume that the GPU is in the same state we left it in.
802 * This is required for transform feedback buffer offsets, query objects,
803 * and also allows us to reduce how much state we have to emit.
805 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
808 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
809 intelDestroyContext(driContextPriv
);
816 intelInitExtensions(ctx
);
818 brw_init_surface_formats(brw
);
820 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
821 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
822 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
823 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
824 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
825 brw
->urb
.size
= devinfo
->urb
.size
;
826 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
827 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
828 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
829 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
830 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
832 /* Estimate the size of the mappable aperture into the GTT. There's an
833 * ioctl to get the whole GTT size, but not one to get the mappable subset.
834 * It turns out it's basically always 256MB, though some ancient hardware
837 uint32_t gtt_size
= 256 * 1024 * 1024;
839 /* We don't want to map two objects such that a memcpy between them would
840 * just fault one mapping in and then the other over and over forever. So
841 * we would need to divide the GTT size by 2. Additionally, some GTT is
842 * taken up by things like the framebuffer and the ringbuffer and such, so
843 * be more conservative.
845 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
848 brw
->urb
.gs_present
= false;
850 brw
->prim_restart
.in_progress
= false;
851 brw
->prim_restart
.enable_cut_index
= false;
852 brw
->gs
.enabled
= false;
853 brw
->sf
.viewport_transform_enable
= true;
855 ctx
->VertexProgram
._MaintainTnlProgram
= true;
856 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
858 brw_draw_init( brw
);
860 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
861 /* Turn on some extra GL_ARB_debug_output generation. */
862 brw
->perf_debug
= true;
865 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
866 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
868 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
869 brw_init_shader_time(brw
);
871 _mesa_compute_version(ctx
);
873 _mesa_initialize_dispatch_tables(ctx
);
874 _mesa_initialize_vbo_vtxfmt(ctx
);
876 if (ctx
->Extensions
.AMD_performance_monitor
) {
877 brw_init_performance_monitors(brw
);
880 vbo_use_buffer_objects(ctx
);
881 vbo_always_unmap_buffers(ctx
);
887 intelDestroyContext(__DRIcontext
* driContextPriv
)
889 struct brw_context
*brw
=
890 (struct brw_context
*) driContextPriv
->driverPrivate
;
891 struct gl_context
*ctx
= &brw
->ctx
;
893 assert(brw
); /* should never be null */
897 /* Dump a final BMP in case the application doesn't call SwapBuffers */
898 if (INTEL_DEBUG
& DEBUG_AUB
) {
899 intel_batchbuffer_flush(brw
);
900 aub_dump_bmp(&brw
->ctx
);
903 _mesa_meta_free(&brw
->ctx
);
904 brw_meta_fast_clear_free(brw
);
906 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
907 /* Force a report. */
908 brw
->shader_time
.report_time
= 0;
910 brw_collect_and_report_shader_time(brw
);
911 brw_destroy_shader_time(brw
);
914 brw_destroy_state(brw
);
915 brw_draw_destroy(brw
);
917 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
918 if (brw
->vs
.base
.scratch_bo
)
919 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
920 if (brw
->gs
.base
.scratch_bo
)
921 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
922 if (brw
->wm
.base
.scratch_bo
)
923 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
925 drm_intel_gem_context_destroy(brw
->hw_ctx
);
927 if (ctx
->swrast_context
) {
928 _swsetup_DestroyContext(&brw
->ctx
);
929 _tnl_DestroyContext(&brw
->ctx
);
931 _vbo_DestroyContext(&brw
->ctx
);
933 if (ctx
->swrast_context
)
934 _swrast_DestroyContext(&brw
->ctx
);
936 intel_batchbuffer_free(brw
);
938 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
939 brw
->first_post_swapbuffers_batch
= NULL
;
941 driDestroyOptionCache(&brw
->optionCache
);
943 /* free the Mesa context */
944 _mesa_free_context_data(&brw
->ctx
);
947 driContextPriv
->driverPrivate
= NULL
;
951 intelUnbindContext(__DRIcontext
* driContextPriv
)
953 /* Unset current context and dispath table */
954 _mesa_make_current(NULL
, NULL
, NULL
);
960 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
961 * on window system framebuffers.
963 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
964 * your renderbuffer can do sRGB encode, and you can flip a switch that does
965 * sRGB encode if the renderbuffer can handle it. You can ask specifically
966 * for a visual where you're guaranteed to be capable, but it turns out that
967 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
968 * incapable ones, becuase there's no difference between the two in resources
969 * used. Applications thus get built that accidentally rely on the default
970 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
973 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
974 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
975 * So they removed the enable knob and made it "if the renderbuffer is sRGB
976 * capable, do sRGB encode". Then, for your window system renderbuffers, you
977 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
978 * and get no sRGB encode (assuming that both kinds of visual are available).
979 * Thus our choice to support sRGB by default on our visuals for desktop would
980 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
982 * Unfortunately, renderbuffer setup happens before a context is created. So
983 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
984 * context (without an sRGB visual, though we don't have sRGB visuals exposed
985 * yet), we go turn that back off before anyone finds out.
988 intel_gles3_srgb_workaround(struct brw_context
*brw
,
989 struct gl_framebuffer
*fb
)
991 struct gl_context
*ctx
= &brw
->ctx
;
993 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
996 /* Some day when we support the sRGB capable bit on visuals available for
997 * GLES, we'll need to respect that and not disable things here.
999 fb
->Visual
.sRGBCapable
= false;
1000 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1001 if (fb
->Attachment
[i
].Renderbuffer
&&
1002 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1003 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1009 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1010 __DRIdrawable
* driDrawPriv
,
1011 __DRIdrawable
* driReadPriv
)
1013 struct brw_context
*brw
;
1014 GET_CURRENT_CONTEXT(curCtx
);
1017 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1021 /* According to the glXMakeCurrent() man page: "Pending commands to
1022 * the previous context, if any, are flushed before it is released."
1023 * But only flush if we're actually changing contexts.
1025 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1026 _mesa_flush(curCtx
);
1029 if (driContextPriv
) {
1030 struct gl_context
*ctx
= &brw
->ctx
;
1031 struct gl_framebuffer
*fb
, *readFb
;
1033 if (driDrawPriv
== NULL
) {
1034 fb
= _mesa_get_incomplete_framebuffer();
1036 fb
= driDrawPriv
->driverPrivate
;
1037 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1040 if (driReadPriv
== NULL
) {
1041 readFb
= _mesa_get_incomplete_framebuffer();
1043 readFb
= driReadPriv
->driverPrivate
;
1044 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1047 /* The sRGB workaround changes the renderbuffer's format. We must change
1048 * the format before the renderbuffer's miptree get's allocated, otherwise
1049 * the formats of the renderbuffer and its miptree will differ.
1051 intel_gles3_srgb_workaround(brw
, fb
);
1052 intel_gles3_srgb_workaround(brw
, readFb
);
1054 /* If the context viewport hasn't been initialized, force a call out to
1055 * the loader to get buffers so we have a drawable size for the initial
1057 if (!brw
->ctx
.ViewportInitialized
)
1058 intel_prepare_render(brw
);
1060 _mesa_make_current(ctx
, fb
, readFb
);
1062 _mesa_make_current(NULL
, NULL
, NULL
);
1069 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1070 __DRIdrawable
*drawable
)
1073 /* MSAA and fast color clear are not supported, so don't waste time
1074 * checking whether a resolve is needed.
1079 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1080 struct intel_renderbuffer
*rb
;
1082 /* Usually, only the back buffer will need to be downsampled. However,
1083 * the front buffer will also need it if the user has rendered into it.
1085 static const gl_buffer_index buffers
[2] = {
1090 for (int i
= 0; i
< 2; ++i
) {
1091 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1092 if (rb
== NULL
|| rb
->mt
== NULL
)
1094 if (rb
->mt
->num_samples
<= 1)
1095 intel_miptree_resolve_color(brw
, rb
->mt
);
1097 intel_renderbuffer_downsample(brw
, rb
);
1102 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1104 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1108 intel_query_dri2_buffers(struct brw_context
*brw
,
1109 __DRIdrawable
*drawable
,
1110 __DRIbuffer
**buffers
,
1114 intel_process_dri2_buffer(struct brw_context
*brw
,
1115 __DRIdrawable
*drawable
,
1116 __DRIbuffer
*buffer
,
1117 struct intel_renderbuffer
*rb
,
1118 const char *buffer_name
);
1121 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1124 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1126 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1127 struct intel_renderbuffer
*rb
;
1128 __DRIbuffer
*buffers
= NULL
;
1130 const char *region_name
;
1132 /* Set this up front, so that in case our buffers get invalidated
1133 * while we're getting new buffers, we don't clobber the stamp and
1134 * thus ignore the invalidate. */
1135 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1137 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1138 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1140 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1142 if (buffers
== NULL
)
1145 for (i
= 0; i
< count
; i
++) {
1146 switch (buffers
[i
].attachment
) {
1147 case __DRI_BUFFER_FRONT_LEFT
:
1148 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1149 region_name
= "dri2 front buffer";
1152 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1153 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1154 region_name
= "dri2 fake front buffer";
1157 case __DRI_BUFFER_BACK_LEFT
:
1158 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1159 region_name
= "dri2 back buffer";
1162 case __DRI_BUFFER_DEPTH
:
1163 case __DRI_BUFFER_HIZ
:
1164 case __DRI_BUFFER_DEPTH_STENCIL
:
1165 case __DRI_BUFFER_STENCIL
:
1166 case __DRI_BUFFER_ACCUM
:
1169 "unhandled buffer attach event, attachment type %d\n",
1170 buffers
[i
].attachment
);
1174 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1180 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1182 struct brw_context
*brw
= context
->driverPrivate
;
1183 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1185 /* Set this up front, so that in case our buffers get invalidated
1186 * while we're getting new buffers, we don't clobber the stamp and
1187 * thus ignore the invalidate. */
1188 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1190 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1191 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1193 if (screen
->image
.loader
)
1194 intel_update_image_buffers(brw
, drawable
);
1196 intel_update_dri2_buffers(brw
, drawable
);
1198 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1202 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1203 * state is required.
1206 intel_prepare_render(struct brw_context
*brw
)
1208 struct gl_context
*ctx
= &brw
->ctx
;
1209 __DRIcontext
*driContext
= brw
->driContext
;
1210 __DRIdrawable
*drawable
;
1212 drawable
= driContext
->driDrawablePriv
;
1213 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1214 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1215 intel_update_renderbuffers(driContext
, drawable
);
1216 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1219 drawable
= driContext
->driReadablePriv
;
1220 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1221 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1222 intel_update_renderbuffers(driContext
, drawable
);
1223 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1226 /* If we're currently rendering to the front buffer, the rendering
1227 * that will happen next will probably dirty the front buffer. So
1228 * mark it as dirty here.
1230 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1231 brw
->front_buffer_dirty
= true;
1233 /* Wait for the swapbuffers before the one we just emitted, so we
1234 * don't get too many swaps outstanding for apps that are GPU-heavy
1235 * but not CPU-heavy.
1237 * We're using intelDRI2Flush (called from the loader before
1238 * swapbuffer) and glFlush (for front buffer rendering) as the
1239 * indicator that a frame is done and then throttle when we get
1240 * here as we prepare to render the next frame. At this point for
1241 * round trips for swap/copy and getting new buffers are done and
1242 * we'll spend less time waiting on the GPU.
1244 * Unfortunately, we don't have a handle to the batch containing
1245 * the swap, and getting our hands on that doesn't seem worth it,
1246 * so we just us the first batch we emitted after the last swap.
1248 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1249 if (!brw
->disable_throttling
)
1250 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1251 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1252 brw
->first_post_swapbuffers_batch
= NULL
;
1253 brw
->need_throttle
= false;
1258 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1260 * To determine which DRI buffers to request, examine the renderbuffers
1261 * attached to the drawable's framebuffer. Then request the buffers with
1262 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1264 * This is called from intel_update_renderbuffers().
1266 * \param drawable Drawable whose buffers are queried.
1267 * \param buffers [out] List of buffers returned by DRI2 query.
1268 * \param buffer_count [out] Number of buffers returned.
1270 * \see intel_update_renderbuffers()
1271 * \see DRI2GetBuffers()
1272 * \see DRI2GetBuffersWithFormat()
1275 intel_query_dri2_buffers(struct brw_context
*brw
,
1276 __DRIdrawable
*drawable
,
1277 __DRIbuffer
**buffers
,
1280 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1281 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1283 unsigned attachments
[8];
1285 struct intel_renderbuffer
*front_rb
;
1286 struct intel_renderbuffer
*back_rb
;
1288 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1289 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1291 memset(attachments
, 0, sizeof(attachments
));
1292 if ((brw_is_front_buffer_drawing(fb
) ||
1293 brw_is_front_buffer_reading(fb
) ||
1294 !back_rb
) && front_rb
) {
1295 /* If a fake front buffer is in use, then querying for
1296 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1297 * the real front buffer to the fake front buffer. So before doing the
1298 * query, we need to make sure all the pending drawing has landed in the
1299 * real front buffer.
1301 intel_batchbuffer_flush(brw
);
1302 intel_flush_front(&brw
->ctx
);
1304 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1305 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1306 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1307 /* We have pending front buffer rendering, but we aren't querying for a
1308 * front buffer. If the front buffer we have is a fake front buffer,
1309 * the X server is going to throw it away when it processes the query.
1310 * So before doing the query, make sure all the pending drawing has
1311 * landed in the real front buffer.
1313 intel_batchbuffer_flush(brw
);
1314 intel_flush_front(&brw
->ctx
);
1318 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1319 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1322 assert(i
<= ARRAY_SIZE(attachments
));
1324 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1329 drawable
->loaderPrivate
);
1333 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1335 * This is called from intel_update_renderbuffers().
1338 * DRI buffers whose attachment point is DRI2BufferStencil or
1339 * DRI2BufferDepthStencil are handled as special cases.
1341 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1342 * that is passed to drm_intel_bo_gem_create_from_name().
1344 * \see intel_update_renderbuffers()
1347 intel_process_dri2_buffer(struct brw_context
*brw
,
1348 __DRIdrawable
*drawable
,
1349 __DRIbuffer
*buffer
,
1350 struct intel_renderbuffer
*rb
,
1351 const char *buffer_name
)
1353 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1359 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1361 /* We try to avoid closing and reopening the same BO name, because the first
1362 * use of a mapping of the buffer involves a bunch of page faulting which is
1363 * moderately expensive.
1365 struct intel_mipmap_tree
*last_mt
;
1366 if (num_samples
== 0)
1369 last_mt
= rb
->singlesample_mt
;
1371 uint32_t old_name
= 0;
1373 /* The bo already has a name because the miptree was created by a
1374 * previous call to intel_process_dri2_buffer(). If a bo already has a
1375 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1376 * create a new name.
1378 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1381 if (old_name
== buffer
->name
)
1384 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1386 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1387 buffer
->name
, buffer
->attachment
,
1388 buffer
->cpp
, buffer
->pitch
);
1391 intel_miptree_release(&rb
->mt
);
1392 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1396 "Failed to open BO for returned DRI2 buffer "
1397 "(%dx%d, %s, named %d).\n"
1398 "This is likely a bug in the X Server that will lead to a "
1400 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1404 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1405 drawable
->w
, drawable
->h
,
1408 if (brw_is_front_buffer_drawing(fb
) &&
1409 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1410 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1411 rb
->Base
.Base
.NumSamples
> 1) {
1412 intel_renderbuffer_upsample(brw
, rb
);
1417 drm_intel_bo_unreference(bo
);
1421 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1423 * To determine which DRI buffers to request, examine the renderbuffers
1424 * attached to the drawable's framebuffer. Then request the buffers from
1427 * This is called from intel_update_renderbuffers().
1429 * \param drawable Drawable whose buffers are queried.
1430 * \param buffers [out] List of buffers returned by DRI2 query.
1431 * \param buffer_count [out] Number of buffers returned.
1433 * \see intel_update_renderbuffers()
1437 intel_update_image_buffer(struct brw_context
*intel
,
1438 __DRIdrawable
*drawable
,
1439 struct intel_renderbuffer
*rb
,
1441 enum __DRIimageBufferMask buffer_type
)
1443 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1445 if (!rb
|| !buffer
->bo
)
1448 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1450 /* Check and see if we're already bound to the right
1453 struct intel_mipmap_tree
*last_mt
;
1454 if (num_samples
== 0)
1457 last_mt
= rb
->singlesample_mt
;
1459 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1462 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1463 buffer
->width
, buffer
->height
,
1466 if (brw_is_front_buffer_drawing(fb
) &&
1467 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1468 rb
->Base
.Base
.NumSamples
> 1) {
1469 intel_renderbuffer_upsample(intel
, rb
);
1474 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1476 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1477 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1478 struct intel_renderbuffer
*front_rb
;
1479 struct intel_renderbuffer
*back_rb
;
1480 struct __DRIimageList images
;
1481 unsigned int format
;
1482 uint32_t buffer_mask
= 0;
1484 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1485 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1488 format
= intel_rb_format(back_rb
);
1490 format
= intel_rb_format(front_rb
);
1494 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1495 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1496 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1500 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1502 (*screen
->image
.loader
->getBuffers
) (drawable
,
1503 driGLFormatToImageFormat(format
),
1504 &drawable
->dri2
.stamp
,
1505 drawable
->loaderPrivate
,
1509 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1510 drawable
->w
= images
.front
->width
;
1511 drawable
->h
= images
.front
->height
;
1512 intel_update_image_buffer(brw
,
1516 __DRI_IMAGE_BUFFER_FRONT
);
1518 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1519 drawable
->w
= images
.back
->width
;
1520 drawable
->h
= images
.back
->height
;
1521 intel_update_image_buffer(brw
,
1525 __DRI_IMAGE_BUFFER_BACK
);