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
);
291 brw_init_compute_functions(functions
);
293 brw_init_conditional_render_functions(functions
);
295 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
297 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
298 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
299 functions
->GetTransformFeedbackVertexCount
=
300 brw_get_transform_feedback_vertex_count
;
302 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
303 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
304 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
305 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
307 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
308 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
312 functions
->GetSamplePosition
= gen6_get_sample_position
;
316 brw_initialize_context_constants(struct brw_context
*brw
)
318 struct gl_context
*ctx
= &brw
->ctx
;
320 unsigned max_samplers
=
321 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
323 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
325 ctx
->Const
.StripTextureBorder
= true;
327 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
328 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
329 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
330 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
331 ctx
->Const
.MaxTextureUnits
=
332 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
333 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
334 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
336 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
338 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
339 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
340 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
341 ctx
->Const
.MaxUniformBufferBindings
+= 12;
343 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
345 ctx
->Const
.MaxCombinedTextureImageUnits
=
346 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
347 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
348 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
349 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
351 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
352 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
353 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
354 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
355 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
356 ctx
->Const
.MaxTextureMbytes
= 1536;
359 ctx
->Const
.MaxArrayTextureLayers
= 2048;
361 ctx
->Const
.MaxArrayTextureLayers
= 512;
363 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
365 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
367 ctx
->Const
.MaxRenderbufferSize
= 8192;
369 /* Hardware only supports a limited number of transform feedback buffers.
370 * So we need to override the Mesa default (which is based only on software
373 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
375 /* On Gen6, in the worst case, we use up one binding table entry per
376 * transform feedback component (see comments above the definition of
377 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
378 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
379 * BRW_MAX_SOL_BINDINGS.
381 * In "separate components" mode, we need to divide this value by
382 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
383 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
385 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
386 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
387 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
389 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
392 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
393 const int clamp_max_samples
=
394 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
396 if (clamp_max_samples
< 0) {
397 max_samples
= msaa_modes
[0];
399 /* Select the largest supported MSAA mode that does not exceed
403 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
404 if (msaa_modes
[i
] <= clamp_max_samples
) {
405 max_samples
= msaa_modes
[i
];
411 ctx
->Const
.MaxSamples
= max_samples
;
412 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
413 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
414 ctx
->Const
.MaxIntegerSamples
= max_samples
;
416 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
417 * to map indices of rectangular grid to sample numbers within a pixel.
418 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
419 * extension implementation. For more details see the comment above
420 * gen6_set_sample_maps() definition.
422 gen6_set_sample_maps(ctx
);
425 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
426 else if (brw
->gen
== 6)
427 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
429 ctx
->Const
.MinLineWidth
= 1.0;
430 ctx
->Const
.MinLineWidthAA
= 1.0;
431 if (brw
->gen
>= 9 || brw
->is_cherryview
) {
432 ctx
->Const
.MaxLineWidth
= 40.0;
433 ctx
->Const
.MaxLineWidthAA
= 40.0;
434 ctx
->Const
.LineWidthGranularity
= 0.125;
435 } else if (brw
->gen
>= 6) {
436 ctx
->Const
.MaxLineWidth
= 7.375;
437 ctx
->Const
.MaxLineWidthAA
= 7.375;
438 ctx
->Const
.LineWidthGranularity
= 0.125;
440 ctx
->Const
.MaxLineWidth
= 7.0;
441 ctx
->Const
.MaxLineWidthAA
= 7.0;
442 ctx
->Const
.LineWidthGranularity
= 0.5;
445 ctx
->Const
.MinPointSize
= 1.0;
446 ctx
->Const
.MinPointSizeAA
= 1.0;
447 ctx
->Const
.MaxPointSize
= 255.0;
448 ctx
->Const
.MaxPointSizeAA
= 255.0;
449 ctx
->Const
.PointSizeGranularity
= 1.0;
451 if (brw
->gen
>= 5 || brw
->is_g4x
)
452 ctx
->Const
.MaxClipPlanes
= 8;
454 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
455 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
456 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
457 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
458 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
459 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
460 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
461 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
462 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
463 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
464 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
465 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
466 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
467 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
469 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
470 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
471 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
472 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
473 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
474 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
475 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
476 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
477 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
478 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
479 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
481 /* Fragment shaders use real, 32-bit twos-complement integers for all
484 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
485 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
486 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
487 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
488 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
490 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
491 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
492 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
493 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
494 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
497 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
498 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
499 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
500 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
501 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
502 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
503 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
504 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
505 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
508 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
509 * but we're not sure how it's actually done for vertex order,
510 * that affect provoking vertex decision. Always use last vertex
511 * convention for quad primitive which works as expected for now.
514 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
516 ctx
->Const
.NativeIntegers
= true;
517 ctx
->Const
.VertexID_is_zero_based
= true;
519 /* Regarding the CMP instruction, the Ivybridge PRM says:
521 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
522 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
523 * 0xFFFFFFFF) is assigned to dst."
525 * but PRMs for earlier generations say
527 * "In dword format, one GRF may store up to 8 results. When the register
528 * is used later as a vector of Booleans, as only LSB at each channel
529 * contains meaning [sic] data, software should make sure all higher bits
530 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
532 * We select the representation of a true boolean uniform to be ~0, and fix
533 * the results of Gen <= 5 CMP instruction's with -(result & 1).
535 ctx
->Const
.UniformBooleanTrue
= ~0;
537 /* From the gen4 PRM, volume 4 page 127:
539 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
540 * the base address of the first element of the surface, computed in
541 * software by adding the surface base address to the byte offset of
542 * the element in the buffer."
544 * However, unaligned accesses are slower, so enforce buffer alignment.
546 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
547 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
550 ctx
->Const
.MaxVarying
= 32;
551 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
552 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
553 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
554 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
557 static const nir_shader_compiler_options nir_options
= {
558 .native_integers
= true,
559 /* In order to help allow for better CSE at the NIR level we tell NIR
560 * to split all ffma instructions during opt_algebraic and we then
561 * re-combine them as a later step.
567 /* We want the GLSL compiler to emit code that uses condition codes */
568 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
569 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
570 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
571 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
572 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
573 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
574 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
575 (i
== MESA_SHADER_FRAGMENT
);
576 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
577 (i
== MESA_SHADER_FRAGMENT
);
578 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
579 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
582 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
583 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
585 if (brw
->scalar_vs
) {
586 /* If we're using the scalar backend for vertex shaders, we need to
587 * configure these accordingly.
589 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectOutput
= true;
590 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectTemp
= true;
591 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= false;
593 if (brw_env_var_as_boolean("INTEL_USE_NIR", false))
594 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].NirOptions
= &nir_options
;
597 if (brw_env_var_as_boolean("INTEL_USE_NIR", true))
598 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_FRAGMENT
].NirOptions
= &nir_options
;
600 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_COMPUTE
].NirOptions
= &nir_options
;
602 /* ARB_viewport_array */
603 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
604 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
605 ctx
->Const
.ViewportSubpixelBits
= 0;
607 /* Cast to float before negating because MaxViewportWidth is unsigned.
609 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
610 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
613 /* ARB_gpu_shader5 */
615 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
619 brw_adjust_cs_context_constants(struct brw_context
*brw
)
621 struct gl_context
*ctx
= &brw
->ctx
;
623 /* For ES, we set these constants based on SIMD8.
625 * TODO: Once we can always generate SIMD16, we should update this.
627 * For GL, we assume we can generate a SIMD16 program, but this currently
628 * is not always true. This allows us to run more test cases, and will be
629 * required based on desktop GL compute shader requirements.
631 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
633 const uint32_t max_invocations
= simd_size
* brw
->max_cs_threads
;
634 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
635 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
636 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
637 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
641 * Process driconf (drirc) options, setting appropriate context flags.
643 * intelInitExtensions still pokes at optionCache directly, in order to
644 * avoid advertising various extensions. No flags are set, so it makes
645 * sense to continue doing that there.
648 brw_process_driconf_options(struct brw_context
*brw
)
650 struct gl_context
*ctx
= &brw
->ctx
;
652 driOptionCache
*options
= &brw
->optionCache
;
653 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
654 brw
->driContext
->driScreenPriv
->myNum
, "i965");
656 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
657 switch (bo_reuse_mode
) {
658 case DRI_CONF_BO_REUSE_DISABLED
:
660 case DRI_CONF_BO_REUSE_ALL
:
661 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
665 if (!driQueryOptionb(options
, "hiz")) {
666 brw
->has_hiz
= false;
667 /* On gen6, you can only do separate stencil with HIZ. */
669 brw
->has_separate_stencil
= false;
672 if (driQueryOptionb(options
, "always_flush_batch")) {
673 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
674 brw
->always_flush_batch
= true;
677 if (driQueryOptionb(options
, "always_flush_cache")) {
678 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
679 brw
->always_flush_cache
= true;
682 if (driQueryOptionb(options
, "disable_throttling")) {
683 fprintf(stderr
, "disabling flush throttling\n");
684 brw
->disable_throttling
= true;
687 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
689 ctx
->Const
.ForceGLSLExtensionsWarn
=
690 driQueryOptionb(options
, "force_glsl_extensions_warn");
692 ctx
->Const
.DisableGLSLLineContinuations
=
693 driQueryOptionb(options
, "disable_glsl_line_continuations");
695 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
696 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
700 brwCreateContext(gl_api api
,
701 const struct gl_config
*mesaVis
,
702 __DRIcontext
*driContextPriv
,
703 unsigned major_version
,
704 unsigned minor_version
,
707 unsigned *dri_ctx_error
,
708 void *sharedContextPrivate
)
710 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
711 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
712 struct intel_screen
*screen
= sPriv
->driverPrivate
;
713 const struct brw_device_info
*devinfo
= screen
->devinfo
;
714 struct dd_function_table functions
;
716 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
717 * provides us with context reset notifications.
719 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
720 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
722 if (screen
->has_context_reset_notification
)
723 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
725 if (flags
& ~allowed_flags
) {
726 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
730 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
732 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
733 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
737 driContextPriv
->driverPrivate
= brw
;
738 brw
->driContext
= driContextPriv
;
739 brw
->intelScreen
= screen
;
740 brw
->bufmgr
= screen
->bufmgr
;
742 brw
->gen
= devinfo
->gen
;
743 brw
->gt
= devinfo
->gt
;
744 brw
->is_g4x
= devinfo
->is_g4x
;
745 brw
->is_baytrail
= devinfo
->is_baytrail
;
746 brw
->is_haswell
= devinfo
->is_haswell
;
747 brw
->is_cherryview
= devinfo
->is_cherryview
;
748 brw
->has_llc
= devinfo
->has_llc
;
749 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
750 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
751 brw
->has_pln
= devinfo
->has_pln
;
752 brw
->has_compr4
= devinfo
->has_compr4
;
753 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
754 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
755 brw
->needs_unlit_centroid_workaround
=
756 devinfo
->needs_unlit_centroid_workaround
;
758 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
759 brw
->has_swizzling
= screen
->hw_has_swizzling
;
761 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
762 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
763 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
765 gen8_init_vtable_surface_functions(brw
);
766 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
767 } else if (brw
->gen
>= 7) {
768 gen7_init_vtable_surface_functions(brw
);
769 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
770 } else if (brw
->gen
>= 6) {
771 gen6_init_vtable_surface_functions(brw
);
772 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
774 gen4_init_vtable_surface_functions(brw
);
775 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
778 brw_init_driver_functions(brw
, &functions
);
781 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
783 struct gl_context
*ctx
= &brw
->ctx
;
785 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
786 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
787 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
788 intelDestroyContext(driContextPriv
);
792 driContextSetFlags(ctx
, flags
);
794 /* Initialize the software rasterizer and helper modules.
796 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
797 * software fallbacks (which we have to support on legacy GL to do weird
798 * glDrawPixels(), glBitmap(), and other functions).
800 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
801 _swrast_CreateContext(ctx
);
804 _vbo_CreateContext(ctx
);
805 if (ctx
->swrast_context
) {
806 _tnl_CreateContext(ctx
);
807 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
808 _swsetup_CreateContext(ctx
);
810 /* Configure swrast to match hardware characteristics: */
811 _swrast_allow_pixel_fog(ctx
, false);
812 _swrast_allow_vertex_fog(ctx
, true);
815 _mesa_meta_init(ctx
);
817 brw_process_driconf_options(brw
);
818 brw_process_intel_debug_variable(brw
);
820 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
821 brw
->scalar_vs
= true;
823 brw_initialize_context_constants(brw
);
825 ctx
->Const
.ResetStrategy
= notify_reset
826 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
828 /* Reinitialize the context point state. It depends on ctx->Const values. */
829 _mesa_init_point(ctx
);
833 intel_batchbuffer_init(brw
);
836 /* Create a new hardware context. Using a hardware context means that
837 * our GPU state will be saved/restored on context switch, allowing us
838 * to assume that the GPU is in the same state we left it in.
840 * This is required for transform feedback buffer offsets, query objects,
841 * and also allows us to reduce how much state we have to emit.
843 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
846 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
847 intelDestroyContext(driContextPriv
);
854 intelInitExtensions(ctx
);
856 brw_init_surface_formats(brw
);
858 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
859 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
860 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
861 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
862 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
863 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
864 brw
->urb
.size
= devinfo
->urb
.size
;
865 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
866 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
867 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
868 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
869 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
871 brw_adjust_cs_context_constants(brw
);
873 /* Estimate the size of the mappable aperture into the GTT. There's an
874 * ioctl to get the whole GTT size, but not one to get the mappable subset.
875 * It turns out it's basically always 256MB, though some ancient hardware
878 uint32_t gtt_size
= 256 * 1024 * 1024;
880 /* We don't want to map two objects such that a memcpy between them would
881 * just fault one mapping in and then the other over and over forever. So
882 * we would need to divide the GTT size by 2. Additionally, some GTT is
883 * taken up by things like the framebuffer and the ringbuffer and such, so
884 * be more conservative.
886 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
889 brw
->urb
.gs_present
= false;
891 brw
->prim_restart
.in_progress
= false;
892 brw
->prim_restart
.enable_cut_index
= false;
893 brw
->gs
.enabled
= false;
894 brw
->sf
.viewport_transform_enable
= true;
896 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
898 ctx
->VertexProgram
._MaintainTnlProgram
= true;
899 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
901 brw_draw_init( brw
);
903 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
904 /* Turn on some extra GL_ARB_debug_output generation. */
905 brw
->perf_debug
= true;
908 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
909 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
911 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
912 brw_init_shader_time(brw
);
914 _mesa_compute_version(ctx
);
916 _mesa_initialize_dispatch_tables(ctx
);
917 _mesa_initialize_vbo_vtxfmt(ctx
);
919 if (ctx
->Extensions
.AMD_performance_monitor
) {
920 brw_init_performance_monitors(brw
);
923 vbo_use_buffer_objects(ctx
);
924 vbo_always_unmap_buffers(ctx
);
930 intelDestroyContext(__DRIcontext
* driContextPriv
)
932 struct brw_context
*brw
=
933 (struct brw_context
*) driContextPriv
->driverPrivate
;
934 struct gl_context
*ctx
= &brw
->ctx
;
936 /* Dump a final BMP in case the application doesn't call SwapBuffers */
937 if (INTEL_DEBUG
& DEBUG_AUB
) {
938 intel_batchbuffer_flush(brw
);
939 aub_dump_bmp(&brw
->ctx
);
942 _mesa_meta_free(&brw
->ctx
);
943 brw_meta_fast_clear_free(brw
);
945 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
946 /* Force a report. */
947 brw
->shader_time
.report_time
= 0;
949 brw_collect_and_report_shader_time(brw
);
950 brw_destroy_shader_time(brw
);
953 brw_destroy_state(brw
);
954 brw_draw_destroy(brw
);
956 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
957 if (brw
->vs
.base
.scratch_bo
)
958 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
959 if (brw
->gs
.base
.scratch_bo
)
960 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
961 if (brw
->wm
.base
.scratch_bo
)
962 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
964 drm_intel_gem_context_destroy(brw
->hw_ctx
);
966 if (ctx
->swrast_context
) {
967 _swsetup_DestroyContext(&brw
->ctx
);
968 _tnl_DestroyContext(&brw
->ctx
);
970 _vbo_DestroyContext(&brw
->ctx
);
972 if (ctx
->swrast_context
)
973 _swrast_DestroyContext(&brw
->ctx
);
975 intel_batchbuffer_free(brw
);
977 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
978 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
979 brw
->throttle_batch
[1] = NULL
;
980 brw
->throttle_batch
[0] = NULL
;
982 driDestroyOptionCache(&brw
->optionCache
);
984 /* free the Mesa context */
985 _mesa_free_context_data(&brw
->ctx
);
988 driContextPriv
->driverPrivate
= NULL
;
992 intelUnbindContext(__DRIcontext
* driContextPriv
)
994 /* Unset current context and dispath table */
995 _mesa_make_current(NULL
, NULL
, NULL
);
1001 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1002 * on window system framebuffers.
1004 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1005 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1006 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1007 * for a visual where you're guaranteed to be capable, but it turns out that
1008 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1009 * incapable ones, because there's no difference between the two in resources
1010 * used. Applications thus get built that accidentally rely on the default
1011 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1014 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1015 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1016 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1017 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1018 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1019 * and get no sRGB encode (assuming that both kinds of visual are available).
1020 * Thus our choice to support sRGB by default on our visuals for desktop would
1021 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1023 * Unfortunately, renderbuffer setup happens before a context is created. So
1024 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1025 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1026 * yet), we go turn that back off before anyone finds out.
1029 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1030 struct gl_framebuffer
*fb
)
1032 struct gl_context
*ctx
= &brw
->ctx
;
1034 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1037 /* Some day when we support the sRGB capable bit on visuals available for
1038 * GLES, we'll need to respect that and not disable things here.
1040 fb
->Visual
.sRGBCapable
= false;
1041 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1042 if (fb
->Attachment
[i
].Renderbuffer
&&
1043 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1044 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1050 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1051 __DRIdrawable
* driDrawPriv
,
1052 __DRIdrawable
* driReadPriv
)
1054 struct brw_context
*brw
;
1055 GET_CURRENT_CONTEXT(curCtx
);
1058 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1062 /* According to the glXMakeCurrent() man page: "Pending commands to
1063 * the previous context, if any, are flushed before it is released."
1064 * But only flush if we're actually changing contexts.
1066 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1067 _mesa_flush(curCtx
);
1070 if (driContextPriv
) {
1071 struct gl_context
*ctx
= &brw
->ctx
;
1072 struct gl_framebuffer
*fb
, *readFb
;
1074 if (driDrawPriv
== NULL
) {
1075 fb
= _mesa_get_incomplete_framebuffer();
1077 fb
= driDrawPriv
->driverPrivate
;
1078 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1081 if (driReadPriv
== NULL
) {
1082 readFb
= _mesa_get_incomplete_framebuffer();
1084 readFb
= driReadPriv
->driverPrivate
;
1085 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1088 /* The sRGB workaround changes the renderbuffer's format. We must change
1089 * the format before the renderbuffer's miptree get's allocated, otherwise
1090 * the formats of the renderbuffer and its miptree will differ.
1092 intel_gles3_srgb_workaround(brw
, fb
);
1093 intel_gles3_srgb_workaround(brw
, readFb
);
1095 /* If the context viewport hasn't been initialized, force a call out to
1096 * the loader to get buffers so we have a drawable size for the initial
1098 if (!brw
->ctx
.ViewportInitialized
)
1099 intel_prepare_render(brw
);
1101 _mesa_make_current(ctx
, fb
, readFb
);
1103 _mesa_make_current(NULL
, NULL
, NULL
);
1110 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1111 __DRIdrawable
*drawable
)
1114 /* MSAA and fast color clear are not supported, so don't waste time
1115 * checking whether a resolve is needed.
1120 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1121 struct intel_renderbuffer
*rb
;
1123 /* Usually, only the back buffer will need to be downsampled. However,
1124 * the front buffer will also need it if the user has rendered into it.
1126 static const gl_buffer_index buffers
[2] = {
1131 for (int i
= 0; i
< 2; ++i
) {
1132 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1133 if (rb
== NULL
|| rb
->mt
== NULL
)
1135 if (rb
->mt
->num_samples
<= 1)
1136 intel_miptree_resolve_color(brw
, rb
->mt
);
1138 intel_renderbuffer_downsample(brw
, rb
);
1143 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1145 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1149 intel_query_dri2_buffers(struct brw_context
*brw
,
1150 __DRIdrawable
*drawable
,
1151 __DRIbuffer
**buffers
,
1155 intel_process_dri2_buffer(struct brw_context
*brw
,
1156 __DRIdrawable
*drawable
,
1157 __DRIbuffer
*buffer
,
1158 struct intel_renderbuffer
*rb
,
1159 const char *buffer_name
);
1162 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1165 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1167 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1168 struct intel_renderbuffer
*rb
;
1169 __DRIbuffer
*buffers
= NULL
;
1171 const char *region_name
;
1173 /* Set this up front, so that in case our buffers get invalidated
1174 * while we're getting new buffers, we don't clobber the stamp and
1175 * thus ignore the invalidate. */
1176 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1178 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1179 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1181 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1183 if (buffers
== NULL
)
1186 for (i
= 0; i
< count
; i
++) {
1187 switch (buffers
[i
].attachment
) {
1188 case __DRI_BUFFER_FRONT_LEFT
:
1189 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1190 region_name
= "dri2 front buffer";
1193 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1194 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1195 region_name
= "dri2 fake front buffer";
1198 case __DRI_BUFFER_BACK_LEFT
:
1199 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1200 region_name
= "dri2 back buffer";
1203 case __DRI_BUFFER_DEPTH
:
1204 case __DRI_BUFFER_HIZ
:
1205 case __DRI_BUFFER_DEPTH_STENCIL
:
1206 case __DRI_BUFFER_STENCIL
:
1207 case __DRI_BUFFER_ACCUM
:
1210 "unhandled buffer attach event, attachment type %d\n",
1211 buffers
[i
].attachment
);
1215 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1221 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1223 struct brw_context
*brw
= context
->driverPrivate
;
1224 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1226 /* Set this up front, so that in case our buffers get invalidated
1227 * while we're getting new buffers, we don't clobber the stamp and
1228 * thus ignore the invalidate. */
1229 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1231 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1232 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1234 if (screen
->image
.loader
)
1235 intel_update_image_buffers(brw
, drawable
);
1237 intel_update_dri2_buffers(brw
, drawable
);
1239 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1243 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1244 * state is required.
1247 intel_prepare_render(struct brw_context
*brw
)
1249 struct gl_context
*ctx
= &brw
->ctx
;
1250 __DRIcontext
*driContext
= brw
->driContext
;
1251 __DRIdrawable
*drawable
;
1253 drawable
= driContext
->driDrawablePriv
;
1254 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1255 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1256 intel_update_renderbuffers(driContext
, drawable
);
1257 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1260 drawable
= driContext
->driReadablePriv
;
1261 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1262 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1263 intel_update_renderbuffers(driContext
, drawable
);
1264 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1267 /* If we're currently rendering to the front buffer, the rendering
1268 * that will happen next will probably dirty the front buffer. So
1269 * mark it as dirty here.
1271 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1272 brw
->front_buffer_dirty
= true;
1276 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1278 * To determine which DRI buffers to request, examine the renderbuffers
1279 * attached to the drawable's framebuffer. Then request the buffers with
1280 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1282 * This is called from intel_update_renderbuffers().
1284 * \param drawable Drawable whose buffers are queried.
1285 * \param buffers [out] List of buffers returned by DRI2 query.
1286 * \param buffer_count [out] Number of buffers returned.
1288 * \see intel_update_renderbuffers()
1289 * \see DRI2GetBuffers()
1290 * \see DRI2GetBuffersWithFormat()
1293 intel_query_dri2_buffers(struct brw_context
*brw
,
1294 __DRIdrawable
*drawable
,
1295 __DRIbuffer
**buffers
,
1298 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1299 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1301 unsigned attachments
[8];
1303 struct intel_renderbuffer
*front_rb
;
1304 struct intel_renderbuffer
*back_rb
;
1306 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1307 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1309 memset(attachments
, 0, sizeof(attachments
));
1310 if ((brw_is_front_buffer_drawing(fb
) ||
1311 brw_is_front_buffer_reading(fb
) ||
1312 !back_rb
) && front_rb
) {
1313 /* If a fake front buffer is in use, then querying for
1314 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1315 * the real front buffer to the fake front buffer. So before doing the
1316 * query, we need to make sure all the pending drawing has landed in the
1317 * real front buffer.
1319 intel_batchbuffer_flush(brw
);
1320 intel_flush_front(&brw
->ctx
);
1322 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1323 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1324 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1325 /* We have pending front buffer rendering, but we aren't querying for a
1326 * front buffer. If the front buffer we have is a fake front buffer,
1327 * the X server is going to throw it away when it processes the query.
1328 * So before doing the query, make sure all the pending drawing has
1329 * landed in the real front buffer.
1331 intel_batchbuffer_flush(brw
);
1332 intel_flush_front(&brw
->ctx
);
1336 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1337 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1340 assert(i
<= ARRAY_SIZE(attachments
));
1342 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1347 drawable
->loaderPrivate
);
1351 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1353 * This is called from intel_update_renderbuffers().
1356 * DRI buffers whose attachment point is DRI2BufferStencil or
1357 * DRI2BufferDepthStencil are handled as special cases.
1359 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1360 * that is passed to drm_intel_bo_gem_create_from_name().
1362 * \see intel_update_renderbuffers()
1365 intel_process_dri2_buffer(struct brw_context
*brw
,
1366 __DRIdrawable
*drawable
,
1367 __DRIbuffer
*buffer
,
1368 struct intel_renderbuffer
*rb
,
1369 const char *buffer_name
)
1371 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1377 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1379 /* We try to avoid closing and reopening the same BO name, because the first
1380 * use of a mapping of the buffer involves a bunch of page faulting which is
1381 * moderately expensive.
1383 struct intel_mipmap_tree
*last_mt
;
1384 if (num_samples
== 0)
1387 last_mt
= rb
->singlesample_mt
;
1389 uint32_t old_name
= 0;
1391 /* The bo already has a name because the miptree was created by a
1392 * previous call to intel_process_dri2_buffer(). If a bo already has a
1393 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1394 * create a new name.
1396 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1399 if (old_name
== buffer
->name
)
1402 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1404 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1405 buffer
->name
, buffer
->attachment
,
1406 buffer
->cpp
, buffer
->pitch
);
1409 intel_miptree_release(&rb
->mt
);
1410 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1414 "Failed to open BO for returned DRI2 buffer "
1415 "(%dx%d, %s, named %d).\n"
1416 "This is likely a bug in the X Server that will lead to a "
1418 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1422 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1423 drawable
->w
, drawable
->h
,
1426 if (brw_is_front_buffer_drawing(fb
) &&
1427 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1428 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1429 rb
->Base
.Base
.NumSamples
> 1) {
1430 intel_renderbuffer_upsample(brw
, rb
);
1435 drm_intel_bo_unreference(bo
);
1439 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1441 * To determine which DRI buffers to request, examine the renderbuffers
1442 * attached to the drawable's framebuffer. Then request the buffers from
1445 * This is called from intel_update_renderbuffers().
1447 * \param drawable Drawable whose buffers are queried.
1448 * \param buffers [out] List of buffers returned by DRI2 query.
1449 * \param buffer_count [out] Number of buffers returned.
1451 * \see intel_update_renderbuffers()
1455 intel_update_image_buffer(struct brw_context
*intel
,
1456 __DRIdrawable
*drawable
,
1457 struct intel_renderbuffer
*rb
,
1459 enum __DRIimageBufferMask buffer_type
)
1461 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1463 if (!rb
|| !buffer
->bo
)
1466 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1468 /* Check and see if we're already bound to the right
1471 struct intel_mipmap_tree
*last_mt
;
1472 if (num_samples
== 0)
1475 last_mt
= rb
->singlesample_mt
;
1477 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1480 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1481 buffer
->width
, buffer
->height
,
1484 if (brw_is_front_buffer_drawing(fb
) &&
1485 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1486 rb
->Base
.Base
.NumSamples
> 1) {
1487 intel_renderbuffer_upsample(intel
, rb
);
1492 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1494 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1495 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1496 struct intel_renderbuffer
*front_rb
;
1497 struct intel_renderbuffer
*back_rb
;
1498 struct __DRIimageList images
;
1499 unsigned int format
;
1500 uint32_t buffer_mask
= 0;
1502 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1503 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1506 format
= intel_rb_format(back_rb
);
1508 format
= intel_rb_format(front_rb
);
1512 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1513 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1514 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1518 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1520 (*screen
->image
.loader
->getBuffers
) (drawable
,
1521 driGLFormatToImageFormat(format
),
1522 &drawable
->dri2
.stamp
,
1523 drawable
->loaderPrivate
,
1527 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1528 drawable
->w
= images
.front
->width
;
1529 drawable
->h
= images
.front
->height
;
1530 intel_update_image_buffer(brw
,
1534 __DRI_IMAGE_BUFFER_FRONT
);
1536 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1537 drawable
->w
= images
.back
->width
;
1538 drawable
->h
= images
.back
->height
;
1539 intel_update_image_buffer(brw
,
1543 __DRI_IMAGE_BUFFER_BACK
);