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 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
295 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
296 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
297 functions
->GetTransformFeedbackVertexCount
=
298 brw_get_transform_feedback_vertex_count
;
300 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
301 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
302 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
303 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
305 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
306 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
310 functions
->GetSamplePosition
= gen6_get_sample_position
;
314 brw_initialize_context_constants(struct brw_context
*brw
)
316 struct gl_context
*ctx
= &brw
->ctx
;
318 unsigned max_samplers
=
319 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
321 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
323 ctx
->Const
.StripTextureBorder
= true;
325 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
326 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
327 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
328 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
329 ctx
->Const
.MaxTextureUnits
=
330 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
331 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
332 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
334 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
336 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
337 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
338 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
339 ctx
->Const
.MaxUniformBufferBindings
+= 12;
341 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
343 ctx
->Const
.MaxCombinedTextureImageUnits
=
344 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
345 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
346 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
347 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
349 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
350 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
351 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
352 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
353 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
354 ctx
->Const
.MaxTextureMbytes
= 1536;
357 ctx
->Const
.MaxArrayTextureLayers
= 2048;
359 ctx
->Const
.MaxArrayTextureLayers
= 512;
361 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
363 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
365 ctx
->Const
.MaxRenderbufferSize
= 8192;
367 /* Hardware only supports a limited number of transform feedback buffers.
368 * So we need to override the Mesa default (which is based only on software
371 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
373 /* On Gen6, in the worst case, we use up one binding table entry per
374 * transform feedback component (see comments above the definition of
375 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
376 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
377 * BRW_MAX_SOL_BINDINGS.
379 * In "separate components" mode, we need to divide this value by
380 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
381 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
383 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
384 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
385 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
387 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
390 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
391 const int clamp_max_samples
=
392 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
394 if (clamp_max_samples
< 0) {
395 max_samples
= msaa_modes
[0];
397 /* Select the largest supported MSAA mode that does not exceed
401 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
402 if (msaa_modes
[i
] <= clamp_max_samples
) {
403 max_samples
= msaa_modes
[i
];
409 ctx
->Const
.MaxSamples
= max_samples
;
410 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
411 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
412 ctx
->Const
.MaxIntegerSamples
= max_samples
;
414 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
415 * to map indices of rectangular grid to sample numbers within a pixel.
416 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
417 * extension implementation. For more details see the comment above
418 * gen6_set_sample_maps() definition.
420 gen6_set_sample_maps(ctx
);
423 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
424 else if (brw
->gen
== 6)
425 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
427 ctx
->Const
.MinLineWidth
= 1.0;
428 ctx
->Const
.MinLineWidthAA
= 1.0;
429 if (brw
->gen
>= 9 || brw
->is_cherryview
) {
430 ctx
->Const
.MaxLineWidth
= 40.0;
431 ctx
->Const
.MaxLineWidthAA
= 40.0;
432 ctx
->Const
.LineWidthGranularity
= 0.125;
433 } else if (brw
->gen
>= 6) {
434 ctx
->Const
.MaxLineWidth
= 7.375;
435 ctx
->Const
.MaxLineWidthAA
= 7.375;
436 ctx
->Const
.LineWidthGranularity
= 0.125;
438 ctx
->Const
.MaxLineWidth
= 7.0;
439 ctx
->Const
.MaxLineWidthAA
= 7.0;
440 ctx
->Const
.LineWidthGranularity
= 0.5;
443 ctx
->Const
.MinPointSize
= 1.0;
444 ctx
->Const
.MinPointSizeAA
= 1.0;
445 ctx
->Const
.MaxPointSize
= 255.0;
446 ctx
->Const
.MaxPointSizeAA
= 255.0;
447 ctx
->Const
.PointSizeGranularity
= 1.0;
449 if (brw
->gen
>= 5 || brw
->is_g4x
)
450 ctx
->Const
.MaxClipPlanes
= 8;
452 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
453 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
454 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
455 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
456 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
457 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
458 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
459 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
460 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
461 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
462 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
463 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
464 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
465 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
467 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
468 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
469 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
470 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
471 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
472 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
473 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
474 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
475 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
476 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
477 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
479 /* Fragment shaders use real, 32-bit twos-complement integers for all
482 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
483 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
484 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
485 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
486 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
488 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
489 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
490 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
491 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
492 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
495 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
496 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
497 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
498 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
499 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
500 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
501 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
502 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
503 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
506 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
507 * but we're not sure how it's actually done for vertex order,
508 * that affect provoking vertex decision. Always use last vertex
509 * convention for quad primitive which works as expected for now.
512 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
514 ctx
->Const
.NativeIntegers
= true;
515 ctx
->Const
.VertexID_is_zero_based
= true;
517 /* Regarding the CMP instruction, the Ivybridge PRM says:
519 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
520 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
521 * 0xFFFFFFFF) is assigned to dst."
523 * but PRMs for earlier generations say
525 * "In dword format, one GRF may store up to 8 results. When the register
526 * is used later as a vector of Booleans, as only LSB at each channel
527 * contains meaning [sic] data, software should make sure all higher bits
528 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
530 * We select the representation of a true boolean uniform to be ~0, and fix
531 * the results of Gen <= 5 CMP instruction's with -(result & 1).
533 ctx
->Const
.UniformBooleanTrue
= ~0;
535 /* From the gen4 PRM, volume 4 page 127:
537 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
538 * the base address of the first element of the surface, computed in
539 * software by adding the surface base address to the byte offset of
540 * the element in the buffer."
542 * However, unaligned accesses are slower, so enforce buffer alignment.
544 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
545 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
548 ctx
->Const
.MaxVarying
= 32;
549 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
550 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
551 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
552 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
555 static const nir_shader_compiler_options nir_options
= {
556 .native_integers
= true,
557 /* In order to help allow for better CSE at the NIR level we tell NIR
558 * to split all ffma instructions during opt_algebraic and we then
559 * re-combine them as a later step.
565 /* We want the GLSL compiler to emit code that uses condition codes */
566 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
567 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
568 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
569 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
570 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
571 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
572 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
573 (i
== MESA_SHADER_FRAGMENT
);
574 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
575 (i
== MESA_SHADER_FRAGMENT
);
576 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
577 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
580 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
581 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
583 if (brw
->scalar_vs
) {
584 /* If we're using the scalar backend for vertex shaders, we need to
585 * configure these accordingly.
587 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectOutput
= true;
588 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectTemp
= true;
589 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= false;
591 if (brw_env_var_as_boolean("INTEL_USE_NIR", false))
592 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].NirOptions
= &nir_options
;
595 if (brw_env_var_as_boolean("INTEL_USE_NIR", true))
596 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_FRAGMENT
].NirOptions
= &nir_options
;
598 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_COMPUTE
].NirOptions
= &nir_options
;
600 /* ARB_viewport_array */
601 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
602 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
603 ctx
->Const
.ViewportSubpixelBits
= 0;
605 /* Cast to float before negating because MaxViewportWidth is unsigned.
607 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
608 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
611 /* ARB_gpu_shader5 */
613 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
617 brw_adjust_cs_context_constants(struct brw_context
*brw
)
619 struct gl_context
*ctx
= &brw
->ctx
;
621 /* For ES, we set these constants based on SIMD8.
623 * TODO: Once we can always generate SIMD16, we should update this.
625 * For GL, we assume we can generate a SIMD16 program, but this currently
626 * is not always true. This allows us to run more test cases, and will be
627 * required based on desktop GL compute shader requirements.
629 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
631 const uint32_t max_invocations
= simd_size
* brw
->max_cs_threads
;
632 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
633 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
634 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
635 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
639 * Process driconf (drirc) options, setting appropriate context flags.
641 * intelInitExtensions still pokes at optionCache directly, in order to
642 * avoid advertising various extensions. No flags are set, so it makes
643 * sense to continue doing that there.
646 brw_process_driconf_options(struct brw_context
*brw
)
648 struct gl_context
*ctx
= &brw
->ctx
;
650 driOptionCache
*options
= &brw
->optionCache
;
651 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
652 brw
->driContext
->driScreenPriv
->myNum
, "i965");
654 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
655 switch (bo_reuse_mode
) {
656 case DRI_CONF_BO_REUSE_DISABLED
:
658 case DRI_CONF_BO_REUSE_ALL
:
659 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
663 if (!driQueryOptionb(options
, "hiz")) {
664 brw
->has_hiz
= false;
665 /* On gen6, you can only do separate stencil with HIZ. */
667 brw
->has_separate_stencil
= false;
670 if (driQueryOptionb(options
, "always_flush_batch")) {
671 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
672 brw
->always_flush_batch
= true;
675 if (driQueryOptionb(options
, "always_flush_cache")) {
676 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
677 brw
->always_flush_cache
= true;
680 if (driQueryOptionb(options
, "disable_throttling")) {
681 fprintf(stderr
, "disabling flush throttling\n");
682 brw
->disable_throttling
= true;
685 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
687 ctx
->Const
.ForceGLSLExtensionsWarn
=
688 driQueryOptionb(options
, "force_glsl_extensions_warn");
690 ctx
->Const
.DisableGLSLLineContinuations
=
691 driQueryOptionb(options
, "disable_glsl_line_continuations");
693 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
694 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
698 brwCreateContext(gl_api api
,
699 const struct gl_config
*mesaVis
,
700 __DRIcontext
*driContextPriv
,
701 unsigned major_version
,
702 unsigned minor_version
,
705 unsigned *dri_ctx_error
,
706 void *sharedContextPrivate
)
708 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
709 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
710 struct intel_screen
*screen
= sPriv
->driverPrivate
;
711 const struct brw_device_info
*devinfo
= screen
->devinfo
;
712 struct dd_function_table functions
;
714 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
715 * provides us with context reset notifications.
717 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
718 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
720 if (screen
->has_context_reset_notification
)
721 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
723 if (flags
& ~allowed_flags
) {
724 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
728 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
730 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
731 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
735 driContextPriv
->driverPrivate
= brw
;
736 brw
->driContext
= driContextPriv
;
737 brw
->intelScreen
= screen
;
738 brw
->bufmgr
= screen
->bufmgr
;
740 brw
->gen
= devinfo
->gen
;
741 brw
->gt
= devinfo
->gt
;
742 brw
->is_g4x
= devinfo
->is_g4x
;
743 brw
->is_baytrail
= devinfo
->is_baytrail
;
744 brw
->is_haswell
= devinfo
->is_haswell
;
745 brw
->is_cherryview
= devinfo
->is_cherryview
;
746 brw
->has_llc
= devinfo
->has_llc
;
747 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
748 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
749 brw
->has_pln
= devinfo
->has_pln
;
750 brw
->has_compr4
= devinfo
->has_compr4
;
751 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
752 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
753 brw
->needs_unlit_centroid_workaround
=
754 devinfo
->needs_unlit_centroid_workaround
;
756 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
757 brw
->has_swizzling
= screen
->hw_has_swizzling
;
759 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
760 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
761 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
763 gen8_init_vtable_surface_functions(brw
);
764 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
765 } else if (brw
->gen
>= 7) {
766 gen7_init_vtable_surface_functions(brw
);
767 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
768 } else if (brw
->gen
>= 6) {
769 gen6_init_vtable_surface_functions(brw
);
770 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
772 gen4_init_vtable_surface_functions(brw
);
773 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
776 brw_init_driver_functions(brw
, &functions
);
779 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
781 struct gl_context
*ctx
= &brw
->ctx
;
783 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
784 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
785 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
786 intelDestroyContext(driContextPriv
);
790 driContextSetFlags(ctx
, flags
);
792 /* Initialize the software rasterizer and helper modules.
794 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
795 * software fallbacks (which we have to support on legacy GL to do weird
796 * glDrawPixels(), glBitmap(), and other functions).
798 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
799 _swrast_CreateContext(ctx
);
802 _vbo_CreateContext(ctx
);
803 if (ctx
->swrast_context
) {
804 _tnl_CreateContext(ctx
);
805 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
806 _swsetup_CreateContext(ctx
);
808 /* Configure swrast to match hardware characteristics: */
809 _swrast_allow_pixel_fog(ctx
, false);
810 _swrast_allow_vertex_fog(ctx
, true);
813 _mesa_meta_init(ctx
);
815 brw_process_driconf_options(brw
);
816 brw_process_intel_debug_variable(brw
);
818 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
819 brw
->scalar_vs
= true;
821 brw_initialize_context_constants(brw
);
823 ctx
->Const
.ResetStrategy
= notify_reset
824 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
826 /* Reinitialize the context point state. It depends on ctx->Const values. */
827 _mesa_init_point(ctx
);
831 intel_batchbuffer_init(brw
);
834 /* Create a new hardware context. Using a hardware context means that
835 * our GPU state will be saved/restored on context switch, allowing us
836 * to assume that the GPU is in the same state we left it in.
838 * This is required for transform feedback buffer offsets, query objects,
839 * and also allows us to reduce how much state we have to emit.
841 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
844 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
845 intelDestroyContext(driContextPriv
);
852 intelInitExtensions(ctx
);
854 brw_init_surface_formats(brw
);
856 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
857 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
858 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
859 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
860 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
861 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
862 brw
->urb
.size
= devinfo
->urb
.size
;
863 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
864 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
865 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
866 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
867 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
869 brw_adjust_cs_context_constants(brw
);
871 /* Estimate the size of the mappable aperture into the GTT. There's an
872 * ioctl to get the whole GTT size, but not one to get the mappable subset.
873 * It turns out it's basically always 256MB, though some ancient hardware
876 uint32_t gtt_size
= 256 * 1024 * 1024;
878 /* We don't want to map two objects such that a memcpy between them would
879 * just fault one mapping in and then the other over and over forever. So
880 * we would need to divide the GTT size by 2. Additionally, some GTT is
881 * taken up by things like the framebuffer and the ringbuffer and such, so
882 * be more conservative.
884 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
887 brw
->urb
.gs_present
= false;
889 brw
->prim_restart
.in_progress
= false;
890 brw
->prim_restart
.enable_cut_index
= false;
891 brw
->gs
.enabled
= false;
892 brw
->sf
.viewport_transform_enable
= true;
894 ctx
->VertexProgram
._MaintainTnlProgram
= true;
895 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
897 brw_draw_init( brw
);
899 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
900 /* Turn on some extra GL_ARB_debug_output generation. */
901 brw
->perf_debug
= true;
904 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
905 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
907 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
908 brw_init_shader_time(brw
);
910 _mesa_compute_version(ctx
);
912 _mesa_initialize_dispatch_tables(ctx
);
913 _mesa_initialize_vbo_vtxfmt(ctx
);
915 if (ctx
->Extensions
.AMD_performance_monitor
) {
916 brw_init_performance_monitors(brw
);
919 vbo_use_buffer_objects(ctx
);
920 vbo_always_unmap_buffers(ctx
);
926 intelDestroyContext(__DRIcontext
* driContextPriv
)
928 struct brw_context
*brw
=
929 (struct brw_context
*) driContextPriv
->driverPrivate
;
930 struct gl_context
*ctx
= &brw
->ctx
;
932 /* Dump a final BMP in case the application doesn't call SwapBuffers */
933 if (INTEL_DEBUG
& DEBUG_AUB
) {
934 intel_batchbuffer_flush(brw
);
935 aub_dump_bmp(&brw
->ctx
);
938 _mesa_meta_free(&brw
->ctx
);
939 brw_meta_fast_clear_free(brw
);
941 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
942 /* Force a report. */
943 brw
->shader_time
.report_time
= 0;
945 brw_collect_and_report_shader_time(brw
);
946 brw_destroy_shader_time(brw
);
949 brw_destroy_state(brw
);
950 brw_draw_destroy(brw
);
952 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
953 if (brw
->vs
.base
.scratch_bo
)
954 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
955 if (brw
->gs
.base
.scratch_bo
)
956 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
957 if (brw
->wm
.base
.scratch_bo
)
958 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
960 drm_intel_gem_context_destroy(brw
->hw_ctx
);
962 if (ctx
->swrast_context
) {
963 _swsetup_DestroyContext(&brw
->ctx
);
964 _tnl_DestroyContext(&brw
->ctx
);
966 _vbo_DestroyContext(&brw
->ctx
);
968 if (ctx
->swrast_context
)
969 _swrast_DestroyContext(&brw
->ctx
);
971 intel_batchbuffer_free(brw
);
973 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
974 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
975 brw
->throttle_batch
[1] = NULL
;
976 brw
->throttle_batch
[0] = NULL
;
978 driDestroyOptionCache(&brw
->optionCache
);
980 /* free the Mesa context */
981 _mesa_free_context_data(&brw
->ctx
);
984 driContextPriv
->driverPrivate
= NULL
;
988 intelUnbindContext(__DRIcontext
* driContextPriv
)
990 /* Unset current context and dispath table */
991 _mesa_make_current(NULL
, NULL
, NULL
);
997 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
998 * on window system framebuffers.
1000 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1001 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1002 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1003 * for a visual where you're guaranteed to be capable, but it turns out that
1004 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1005 * incapable ones, because there's no difference between the two in resources
1006 * used. Applications thus get built that accidentally rely on the default
1007 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1010 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1011 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1012 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1013 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1014 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1015 * and get no sRGB encode (assuming that both kinds of visual are available).
1016 * Thus our choice to support sRGB by default on our visuals for desktop would
1017 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1019 * Unfortunately, renderbuffer setup happens before a context is created. So
1020 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1021 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1022 * yet), we go turn that back off before anyone finds out.
1025 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1026 struct gl_framebuffer
*fb
)
1028 struct gl_context
*ctx
= &brw
->ctx
;
1030 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1033 /* Some day when we support the sRGB capable bit on visuals available for
1034 * GLES, we'll need to respect that and not disable things here.
1036 fb
->Visual
.sRGBCapable
= false;
1037 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1038 if (fb
->Attachment
[i
].Renderbuffer
&&
1039 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1040 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1046 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1047 __DRIdrawable
* driDrawPriv
,
1048 __DRIdrawable
* driReadPriv
)
1050 struct brw_context
*brw
;
1051 GET_CURRENT_CONTEXT(curCtx
);
1054 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1058 /* According to the glXMakeCurrent() man page: "Pending commands to
1059 * the previous context, if any, are flushed before it is released."
1060 * But only flush if we're actually changing contexts.
1062 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1063 _mesa_flush(curCtx
);
1066 if (driContextPriv
) {
1067 struct gl_context
*ctx
= &brw
->ctx
;
1068 struct gl_framebuffer
*fb
, *readFb
;
1070 if (driDrawPriv
== NULL
) {
1071 fb
= _mesa_get_incomplete_framebuffer();
1073 fb
= driDrawPriv
->driverPrivate
;
1074 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1077 if (driReadPriv
== NULL
) {
1078 readFb
= _mesa_get_incomplete_framebuffer();
1080 readFb
= driReadPriv
->driverPrivate
;
1081 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1084 /* The sRGB workaround changes the renderbuffer's format. We must change
1085 * the format before the renderbuffer's miptree get's allocated, otherwise
1086 * the formats of the renderbuffer and its miptree will differ.
1088 intel_gles3_srgb_workaround(brw
, fb
);
1089 intel_gles3_srgb_workaround(brw
, readFb
);
1091 /* If the context viewport hasn't been initialized, force a call out to
1092 * the loader to get buffers so we have a drawable size for the initial
1094 if (!brw
->ctx
.ViewportInitialized
)
1095 intel_prepare_render(brw
);
1097 _mesa_make_current(ctx
, fb
, readFb
);
1099 _mesa_make_current(NULL
, NULL
, NULL
);
1106 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1107 __DRIdrawable
*drawable
)
1110 /* MSAA and fast color clear are not supported, so don't waste time
1111 * checking whether a resolve is needed.
1116 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1117 struct intel_renderbuffer
*rb
;
1119 /* Usually, only the back buffer will need to be downsampled. However,
1120 * the front buffer will also need it if the user has rendered into it.
1122 static const gl_buffer_index buffers
[2] = {
1127 for (int i
= 0; i
< 2; ++i
) {
1128 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1129 if (rb
== NULL
|| rb
->mt
== NULL
)
1131 if (rb
->mt
->num_samples
<= 1)
1132 intel_miptree_resolve_color(brw
, rb
->mt
);
1134 intel_renderbuffer_downsample(brw
, rb
);
1139 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1141 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1145 intel_query_dri2_buffers(struct brw_context
*brw
,
1146 __DRIdrawable
*drawable
,
1147 __DRIbuffer
**buffers
,
1151 intel_process_dri2_buffer(struct brw_context
*brw
,
1152 __DRIdrawable
*drawable
,
1153 __DRIbuffer
*buffer
,
1154 struct intel_renderbuffer
*rb
,
1155 const char *buffer_name
);
1158 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1161 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1163 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1164 struct intel_renderbuffer
*rb
;
1165 __DRIbuffer
*buffers
= NULL
;
1167 const char *region_name
;
1169 /* Set this up front, so that in case our buffers get invalidated
1170 * while we're getting new buffers, we don't clobber the stamp and
1171 * thus ignore the invalidate. */
1172 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1174 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1175 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1177 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1179 if (buffers
== NULL
)
1182 for (i
= 0; i
< count
; i
++) {
1183 switch (buffers
[i
].attachment
) {
1184 case __DRI_BUFFER_FRONT_LEFT
:
1185 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1186 region_name
= "dri2 front buffer";
1189 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1190 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1191 region_name
= "dri2 fake front buffer";
1194 case __DRI_BUFFER_BACK_LEFT
:
1195 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1196 region_name
= "dri2 back buffer";
1199 case __DRI_BUFFER_DEPTH
:
1200 case __DRI_BUFFER_HIZ
:
1201 case __DRI_BUFFER_DEPTH_STENCIL
:
1202 case __DRI_BUFFER_STENCIL
:
1203 case __DRI_BUFFER_ACCUM
:
1206 "unhandled buffer attach event, attachment type %d\n",
1207 buffers
[i
].attachment
);
1211 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1217 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1219 struct brw_context
*brw
= context
->driverPrivate
;
1220 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1222 /* Set this up front, so that in case our buffers get invalidated
1223 * while we're getting new buffers, we don't clobber the stamp and
1224 * thus ignore the invalidate. */
1225 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1227 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1228 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1230 if (screen
->image
.loader
)
1231 intel_update_image_buffers(brw
, drawable
);
1233 intel_update_dri2_buffers(brw
, drawable
);
1235 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1239 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1240 * state is required.
1243 intel_prepare_render(struct brw_context
*brw
)
1245 struct gl_context
*ctx
= &brw
->ctx
;
1246 __DRIcontext
*driContext
= brw
->driContext
;
1247 __DRIdrawable
*drawable
;
1249 drawable
= driContext
->driDrawablePriv
;
1250 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1251 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1252 intel_update_renderbuffers(driContext
, drawable
);
1253 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1256 drawable
= driContext
->driReadablePriv
;
1257 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1258 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1259 intel_update_renderbuffers(driContext
, drawable
);
1260 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1263 /* If we're currently rendering to the front buffer, the rendering
1264 * that will happen next will probably dirty the front buffer. So
1265 * mark it as dirty here.
1267 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1268 brw
->front_buffer_dirty
= true;
1272 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1274 * To determine which DRI buffers to request, examine the renderbuffers
1275 * attached to the drawable's framebuffer. Then request the buffers with
1276 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1278 * This is called from intel_update_renderbuffers().
1280 * \param drawable Drawable whose buffers are queried.
1281 * \param buffers [out] List of buffers returned by DRI2 query.
1282 * \param buffer_count [out] Number of buffers returned.
1284 * \see intel_update_renderbuffers()
1285 * \see DRI2GetBuffers()
1286 * \see DRI2GetBuffersWithFormat()
1289 intel_query_dri2_buffers(struct brw_context
*brw
,
1290 __DRIdrawable
*drawable
,
1291 __DRIbuffer
**buffers
,
1294 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1295 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1297 unsigned attachments
[8];
1299 struct intel_renderbuffer
*front_rb
;
1300 struct intel_renderbuffer
*back_rb
;
1302 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1303 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1305 memset(attachments
, 0, sizeof(attachments
));
1306 if ((brw_is_front_buffer_drawing(fb
) ||
1307 brw_is_front_buffer_reading(fb
) ||
1308 !back_rb
) && front_rb
) {
1309 /* If a fake front buffer is in use, then querying for
1310 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1311 * the real front buffer to the fake front buffer. So before doing the
1312 * query, we need to make sure all the pending drawing has landed in the
1313 * real front buffer.
1315 intel_batchbuffer_flush(brw
);
1316 intel_flush_front(&brw
->ctx
);
1318 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1319 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1320 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1321 /* We have pending front buffer rendering, but we aren't querying for a
1322 * front buffer. If the front buffer we have is a fake front buffer,
1323 * the X server is going to throw it away when it processes the query.
1324 * So before doing the query, make sure all the pending drawing has
1325 * landed in the real front buffer.
1327 intel_batchbuffer_flush(brw
);
1328 intel_flush_front(&brw
->ctx
);
1332 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1333 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1336 assert(i
<= ARRAY_SIZE(attachments
));
1338 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1343 drawable
->loaderPrivate
);
1347 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1349 * This is called from intel_update_renderbuffers().
1352 * DRI buffers whose attachment point is DRI2BufferStencil or
1353 * DRI2BufferDepthStencil are handled as special cases.
1355 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1356 * that is passed to drm_intel_bo_gem_create_from_name().
1358 * \see intel_update_renderbuffers()
1361 intel_process_dri2_buffer(struct brw_context
*brw
,
1362 __DRIdrawable
*drawable
,
1363 __DRIbuffer
*buffer
,
1364 struct intel_renderbuffer
*rb
,
1365 const char *buffer_name
)
1367 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1373 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1375 /* We try to avoid closing and reopening the same BO name, because the first
1376 * use of a mapping of the buffer involves a bunch of page faulting which is
1377 * moderately expensive.
1379 struct intel_mipmap_tree
*last_mt
;
1380 if (num_samples
== 0)
1383 last_mt
= rb
->singlesample_mt
;
1385 uint32_t old_name
= 0;
1387 /* The bo already has a name because the miptree was created by a
1388 * previous call to intel_process_dri2_buffer(). If a bo already has a
1389 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1390 * create a new name.
1392 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1395 if (old_name
== buffer
->name
)
1398 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1400 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1401 buffer
->name
, buffer
->attachment
,
1402 buffer
->cpp
, buffer
->pitch
);
1405 intel_miptree_release(&rb
->mt
);
1406 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1410 "Failed to open BO for returned DRI2 buffer "
1411 "(%dx%d, %s, named %d).\n"
1412 "This is likely a bug in the X Server that will lead to a "
1414 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1418 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1419 drawable
->w
, drawable
->h
,
1422 if (brw_is_front_buffer_drawing(fb
) &&
1423 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1424 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1425 rb
->Base
.Base
.NumSamples
> 1) {
1426 intel_renderbuffer_upsample(brw
, rb
);
1431 drm_intel_bo_unreference(bo
);
1435 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1437 * To determine which DRI buffers to request, examine the renderbuffers
1438 * attached to the drawable's framebuffer. Then request the buffers from
1441 * This is called from intel_update_renderbuffers().
1443 * \param drawable Drawable whose buffers are queried.
1444 * \param buffers [out] List of buffers returned by DRI2 query.
1445 * \param buffer_count [out] Number of buffers returned.
1447 * \see intel_update_renderbuffers()
1451 intel_update_image_buffer(struct brw_context
*intel
,
1452 __DRIdrawable
*drawable
,
1453 struct intel_renderbuffer
*rb
,
1455 enum __DRIimageBufferMask buffer_type
)
1457 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1459 if (!rb
|| !buffer
->bo
)
1462 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1464 /* Check and see if we're already bound to the right
1467 struct intel_mipmap_tree
*last_mt
;
1468 if (num_samples
== 0)
1471 last_mt
= rb
->singlesample_mt
;
1473 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1476 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1477 buffer
->width
, buffer
->height
,
1480 if (brw_is_front_buffer_drawing(fb
) &&
1481 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1482 rb
->Base
.Base
.NumSamples
> 1) {
1483 intel_renderbuffer_upsample(intel
, rb
);
1488 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1490 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1491 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1492 struct intel_renderbuffer
*front_rb
;
1493 struct intel_renderbuffer
*back_rb
;
1494 struct __DRIimageList images
;
1495 unsigned int format
;
1496 uint32_t buffer_mask
= 0;
1498 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1499 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1502 format
= intel_rb_format(back_rb
);
1504 format
= intel_rb_format(front_rb
);
1508 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1509 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1510 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1514 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1516 (*screen
->image
.loader
->getBuffers
) (drawable
,
1517 driGLFormatToImageFormat(format
),
1518 &drawable
->dri2
.stamp
,
1519 drawable
->loaderPrivate
,
1523 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1524 drawable
->w
= images
.front
->width
;
1525 drawable
->h
= images
.front
->height
;
1526 intel_update_image_buffer(brw
,
1530 __DRI_IMAGE_BUFFER_FRONT
);
1532 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1533 drawable
->w
= images
.back
->width
;
1534 drawable
->h
= images
.back
->height
;
1535 intel_update_image_buffer(brw
,
1539 __DRI_IMAGE_BUFFER_BACK
);