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 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].NirOptions
= &nir_options
;
596 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_FRAGMENT
].NirOptions
= &nir_options
;
597 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_COMPUTE
].NirOptions
= &nir_options
;
599 /* ARB_viewport_array */
600 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
601 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
602 ctx
->Const
.ViewportSubpixelBits
= 0;
604 /* Cast to float before negating because MaxViewportWidth is unsigned.
606 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
607 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
610 /* ARB_gpu_shader5 */
612 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
616 brw_adjust_cs_context_constants(struct brw_context
*brw
)
618 struct gl_context
*ctx
= &brw
->ctx
;
620 /* For ES, we set these constants based on SIMD8.
622 * TODO: Once we can always generate SIMD16, we should update this.
624 * For GL, we assume we can generate a SIMD16 program, but this currently
625 * is not always true. This allows us to run more test cases, and will be
626 * required based on desktop GL compute shader requirements.
628 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
630 const uint32_t max_invocations
= simd_size
* brw
->max_cs_threads
;
631 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
632 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
633 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
634 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
638 * Process driconf (drirc) options, setting appropriate context flags.
640 * intelInitExtensions still pokes at optionCache directly, in order to
641 * avoid advertising various extensions. No flags are set, so it makes
642 * sense to continue doing that there.
645 brw_process_driconf_options(struct brw_context
*brw
)
647 struct gl_context
*ctx
= &brw
->ctx
;
649 driOptionCache
*options
= &brw
->optionCache
;
650 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
651 brw
->driContext
->driScreenPriv
->myNum
, "i965");
653 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
654 switch (bo_reuse_mode
) {
655 case DRI_CONF_BO_REUSE_DISABLED
:
657 case DRI_CONF_BO_REUSE_ALL
:
658 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
662 if (!driQueryOptionb(options
, "hiz")) {
663 brw
->has_hiz
= false;
664 /* On gen6, you can only do separate stencil with HIZ. */
666 brw
->has_separate_stencil
= false;
669 if (driQueryOptionb(options
, "always_flush_batch")) {
670 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
671 brw
->always_flush_batch
= true;
674 if (driQueryOptionb(options
, "always_flush_cache")) {
675 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
676 brw
->always_flush_cache
= true;
679 if (driQueryOptionb(options
, "disable_throttling")) {
680 fprintf(stderr
, "disabling flush throttling\n");
681 brw
->disable_throttling
= true;
684 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
686 ctx
->Const
.ForceGLSLExtensionsWarn
=
687 driQueryOptionb(options
, "force_glsl_extensions_warn");
689 ctx
->Const
.DisableGLSLLineContinuations
=
690 driQueryOptionb(options
, "disable_glsl_line_continuations");
692 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
693 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
697 brwCreateContext(gl_api api
,
698 const struct gl_config
*mesaVis
,
699 __DRIcontext
*driContextPriv
,
700 unsigned major_version
,
701 unsigned minor_version
,
704 unsigned *dri_ctx_error
,
705 void *sharedContextPrivate
)
707 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
708 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
709 struct intel_screen
*screen
= sPriv
->driverPrivate
;
710 const struct brw_device_info
*devinfo
= screen
->devinfo
;
711 struct dd_function_table functions
;
713 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
714 * provides us with context reset notifications.
716 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
717 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
719 if (screen
->has_context_reset_notification
)
720 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
722 if (flags
& ~allowed_flags
) {
723 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
727 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
729 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
730 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
734 driContextPriv
->driverPrivate
= brw
;
735 brw
->driContext
= driContextPriv
;
736 brw
->intelScreen
= screen
;
737 brw
->bufmgr
= screen
->bufmgr
;
739 brw
->gen
= devinfo
->gen
;
740 brw
->gt
= devinfo
->gt
;
741 brw
->is_g4x
= devinfo
->is_g4x
;
742 brw
->is_baytrail
= devinfo
->is_baytrail
;
743 brw
->is_haswell
= devinfo
->is_haswell
;
744 brw
->is_cherryview
= devinfo
->is_cherryview
;
745 brw
->has_llc
= devinfo
->has_llc
;
746 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
747 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
748 brw
->has_pln
= devinfo
->has_pln
;
749 brw
->has_compr4
= devinfo
->has_compr4
;
750 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
751 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
752 brw
->needs_unlit_centroid_workaround
=
753 devinfo
->needs_unlit_centroid_workaround
;
755 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
756 brw
->has_swizzling
= screen
->hw_has_swizzling
;
758 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
759 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
760 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
762 gen8_init_vtable_surface_functions(brw
);
763 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
764 } else if (brw
->gen
>= 7) {
765 gen7_init_vtable_surface_functions(brw
);
766 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
767 } else if (brw
->gen
>= 6) {
768 gen6_init_vtable_surface_functions(brw
);
769 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
771 gen4_init_vtable_surface_functions(brw
);
772 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
775 brw_init_driver_functions(brw
, &functions
);
778 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
780 struct gl_context
*ctx
= &brw
->ctx
;
782 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
783 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
784 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
785 intelDestroyContext(driContextPriv
);
789 driContextSetFlags(ctx
, flags
);
791 /* Initialize the software rasterizer and helper modules.
793 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
794 * software fallbacks (which we have to support on legacy GL to do weird
795 * glDrawPixels(), glBitmap(), and other functions).
797 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
798 _swrast_CreateContext(ctx
);
801 _vbo_CreateContext(ctx
);
802 if (ctx
->swrast_context
) {
803 _tnl_CreateContext(ctx
);
804 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
805 _swsetup_CreateContext(ctx
);
807 /* Configure swrast to match hardware characteristics: */
808 _swrast_allow_pixel_fog(ctx
, false);
809 _swrast_allow_vertex_fog(ctx
, true);
812 _mesa_meta_init(ctx
);
814 brw_process_driconf_options(brw
);
815 brw_process_intel_debug_variable(brw
);
817 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
818 brw
->scalar_vs
= true;
820 brw_initialize_context_constants(brw
);
822 ctx
->Const
.ResetStrategy
= notify_reset
823 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
825 /* Reinitialize the context point state. It depends on ctx->Const values. */
826 _mesa_init_point(ctx
);
830 intel_batchbuffer_init(brw
);
833 /* Create a new hardware context. Using a hardware context means that
834 * our GPU state will be saved/restored on context switch, allowing us
835 * to assume that the GPU is in the same state we left it in.
837 * This is required for transform feedback buffer offsets, query objects,
838 * and also allows us to reduce how much state we have to emit.
840 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
843 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
844 intelDestroyContext(driContextPriv
);
851 intelInitExtensions(ctx
);
853 brw_init_surface_formats(brw
);
855 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
856 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
857 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
858 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
859 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
860 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
861 brw
->urb
.size
= devinfo
->urb
.size
;
862 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
863 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
864 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
865 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
866 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
868 brw_adjust_cs_context_constants(brw
);
870 /* Estimate the size of the mappable aperture into the GTT. There's an
871 * ioctl to get the whole GTT size, but not one to get the mappable subset.
872 * It turns out it's basically always 256MB, though some ancient hardware
875 uint32_t gtt_size
= 256 * 1024 * 1024;
877 /* We don't want to map two objects such that a memcpy between them would
878 * just fault one mapping in and then the other over and over forever. So
879 * we would need to divide the GTT size by 2. Additionally, some GTT is
880 * taken up by things like the framebuffer and the ringbuffer and such, so
881 * be more conservative.
883 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
886 brw
->urb
.gs_present
= false;
888 brw
->prim_restart
.in_progress
= false;
889 brw
->prim_restart
.enable_cut_index
= false;
890 brw
->gs
.enabled
= false;
891 brw
->sf
.viewport_transform_enable
= true;
893 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
895 ctx
->VertexProgram
._MaintainTnlProgram
= true;
896 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
898 brw_draw_init( brw
);
900 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
901 /* Turn on some extra GL_ARB_debug_output generation. */
902 brw
->perf_debug
= true;
905 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
906 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
908 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
909 brw_init_shader_time(brw
);
911 _mesa_compute_version(ctx
);
913 _mesa_initialize_dispatch_tables(ctx
);
914 _mesa_initialize_vbo_vtxfmt(ctx
);
916 if (ctx
->Extensions
.AMD_performance_monitor
) {
917 brw_init_performance_monitors(brw
);
920 vbo_use_buffer_objects(ctx
);
921 vbo_always_unmap_buffers(ctx
);
927 intelDestroyContext(__DRIcontext
* driContextPriv
)
929 struct brw_context
*brw
=
930 (struct brw_context
*) driContextPriv
->driverPrivate
;
931 struct gl_context
*ctx
= &brw
->ctx
;
933 /* Dump a final BMP in case the application doesn't call SwapBuffers */
934 if (INTEL_DEBUG
& DEBUG_AUB
) {
935 intel_batchbuffer_flush(brw
);
936 aub_dump_bmp(&brw
->ctx
);
939 _mesa_meta_free(&brw
->ctx
);
940 brw_meta_fast_clear_free(brw
);
942 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
943 /* Force a report. */
944 brw
->shader_time
.report_time
= 0;
946 brw_collect_and_report_shader_time(brw
);
947 brw_destroy_shader_time(brw
);
950 brw_destroy_state(brw
);
951 brw_draw_destroy(brw
);
953 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
954 if (brw
->vs
.base
.scratch_bo
)
955 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
956 if (brw
->gs
.base
.scratch_bo
)
957 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
958 if (brw
->wm
.base
.scratch_bo
)
959 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
961 drm_intel_gem_context_destroy(brw
->hw_ctx
);
963 if (ctx
->swrast_context
) {
964 _swsetup_DestroyContext(&brw
->ctx
);
965 _tnl_DestroyContext(&brw
->ctx
);
967 _vbo_DestroyContext(&brw
->ctx
);
969 if (ctx
->swrast_context
)
970 _swrast_DestroyContext(&brw
->ctx
);
972 intel_batchbuffer_free(brw
);
974 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
975 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
976 brw
->throttle_batch
[1] = NULL
;
977 brw
->throttle_batch
[0] = NULL
;
979 driDestroyOptionCache(&brw
->optionCache
);
981 /* free the Mesa context */
982 _mesa_free_context_data(&brw
->ctx
);
985 driContextPriv
->driverPrivate
= NULL
;
989 intelUnbindContext(__DRIcontext
* driContextPriv
)
991 /* Unset current context and dispath table */
992 _mesa_make_current(NULL
, NULL
, NULL
);
998 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
999 * on window system framebuffers.
1001 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1002 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1003 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1004 * for a visual where you're guaranteed to be capable, but it turns out that
1005 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1006 * incapable ones, because there's no difference between the two in resources
1007 * used. Applications thus get built that accidentally rely on the default
1008 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1011 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1012 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1013 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1014 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1015 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1016 * and get no sRGB encode (assuming that both kinds of visual are available).
1017 * Thus our choice to support sRGB by default on our visuals for desktop would
1018 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1020 * Unfortunately, renderbuffer setup happens before a context is created. So
1021 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1022 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1023 * yet), we go turn that back off before anyone finds out.
1026 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1027 struct gl_framebuffer
*fb
)
1029 struct gl_context
*ctx
= &brw
->ctx
;
1031 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1034 /* Some day when we support the sRGB capable bit on visuals available for
1035 * GLES, we'll need to respect that and not disable things here.
1037 fb
->Visual
.sRGBCapable
= false;
1038 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1039 if (fb
->Attachment
[i
].Renderbuffer
&&
1040 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1041 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1047 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1048 __DRIdrawable
* driDrawPriv
,
1049 __DRIdrawable
* driReadPriv
)
1051 struct brw_context
*brw
;
1052 GET_CURRENT_CONTEXT(curCtx
);
1055 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1059 /* According to the glXMakeCurrent() man page: "Pending commands to
1060 * the previous context, if any, are flushed before it is released."
1061 * But only flush if we're actually changing contexts.
1063 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1064 _mesa_flush(curCtx
);
1067 if (driContextPriv
) {
1068 struct gl_context
*ctx
= &brw
->ctx
;
1069 struct gl_framebuffer
*fb
, *readFb
;
1071 if (driDrawPriv
== NULL
) {
1072 fb
= _mesa_get_incomplete_framebuffer();
1074 fb
= driDrawPriv
->driverPrivate
;
1075 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1078 if (driReadPriv
== NULL
) {
1079 readFb
= _mesa_get_incomplete_framebuffer();
1081 readFb
= driReadPriv
->driverPrivate
;
1082 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1085 /* The sRGB workaround changes the renderbuffer's format. We must change
1086 * the format before the renderbuffer's miptree get's allocated, otherwise
1087 * the formats of the renderbuffer and its miptree will differ.
1089 intel_gles3_srgb_workaround(brw
, fb
);
1090 intel_gles3_srgb_workaround(brw
, readFb
);
1092 /* If the context viewport hasn't been initialized, force a call out to
1093 * the loader to get buffers so we have a drawable size for the initial
1095 if (!brw
->ctx
.ViewportInitialized
)
1096 intel_prepare_render(brw
);
1098 _mesa_make_current(ctx
, fb
, readFb
);
1100 _mesa_make_current(NULL
, NULL
, NULL
);
1107 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1108 __DRIdrawable
*drawable
)
1111 /* MSAA and fast color clear are not supported, so don't waste time
1112 * checking whether a resolve is needed.
1117 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1118 struct intel_renderbuffer
*rb
;
1120 /* Usually, only the back buffer will need to be downsampled. However,
1121 * the front buffer will also need it if the user has rendered into it.
1123 static const gl_buffer_index buffers
[2] = {
1128 for (int i
= 0; i
< 2; ++i
) {
1129 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1130 if (rb
== NULL
|| rb
->mt
== NULL
)
1132 if (rb
->mt
->num_samples
<= 1)
1133 intel_miptree_resolve_color(brw
, rb
->mt
);
1135 intel_renderbuffer_downsample(brw
, rb
);
1140 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1142 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1146 intel_query_dri2_buffers(struct brw_context
*brw
,
1147 __DRIdrawable
*drawable
,
1148 __DRIbuffer
**buffers
,
1152 intel_process_dri2_buffer(struct brw_context
*brw
,
1153 __DRIdrawable
*drawable
,
1154 __DRIbuffer
*buffer
,
1155 struct intel_renderbuffer
*rb
,
1156 const char *buffer_name
);
1159 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1162 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1164 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1165 struct intel_renderbuffer
*rb
;
1166 __DRIbuffer
*buffers
= NULL
;
1168 const char *region_name
;
1170 /* Set this up front, so that in case our buffers get invalidated
1171 * while we're getting new buffers, we don't clobber the stamp and
1172 * thus ignore the invalidate. */
1173 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1175 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1176 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1178 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1180 if (buffers
== NULL
)
1183 for (i
= 0; i
< count
; i
++) {
1184 switch (buffers
[i
].attachment
) {
1185 case __DRI_BUFFER_FRONT_LEFT
:
1186 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1187 region_name
= "dri2 front buffer";
1190 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1191 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1192 region_name
= "dri2 fake front buffer";
1195 case __DRI_BUFFER_BACK_LEFT
:
1196 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1197 region_name
= "dri2 back buffer";
1200 case __DRI_BUFFER_DEPTH
:
1201 case __DRI_BUFFER_HIZ
:
1202 case __DRI_BUFFER_DEPTH_STENCIL
:
1203 case __DRI_BUFFER_STENCIL
:
1204 case __DRI_BUFFER_ACCUM
:
1207 "unhandled buffer attach event, attachment type %d\n",
1208 buffers
[i
].attachment
);
1212 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1218 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1220 struct brw_context
*brw
= context
->driverPrivate
;
1221 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1223 /* Set this up front, so that in case our buffers get invalidated
1224 * while we're getting new buffers, we don't clobber the stamp and
1225 * thus ignore the invalidate. */
1226 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1228 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1229 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1231 if (screen
->image
.loader
)
1232 intel_update_image_buffers(brw
, drawable
);
1234 intel_update_dri2_buffers(brw
, drawable
);
1236 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1240 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1241 * state is required.
1244 intel_prepare_render(struct brw_context
*brw
)
1246 struct gl_context
*ctx
= &brw
->ctx
;
1247 __DRIcontext
*driContext
= brw
->driContext
;
1248 __DRIdrawable
*drawable
;
1250 drawable
= driContext
->driDrawablePriv
;
1251 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1252 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1253 intel_update_renderbuffers(driContext
, drawable
);
1254 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1257 drawable
= driContext
->driReadablePriv
;
1258 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1259 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1260 intel_update_renderbuffers(driContext
, drawable
);
1261 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1264 /* If we're currently rendering to the front buffer, the rendering
1265 * that will happen next will probably dirty the front buffer. So
1266 * mark it as dirty here.
1268 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1269 brw
->front_buffer_dirty
= true;
1273 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1275 * To determine which DRI buffers to request, examine the renderbuffers
1276 * attached to the drawable's framebuffer. Then request the buffers with
1277 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1279 * This is called from intel_update_renderbuffers().
1281 * \param drawable Drawable whose buffers are queried.
1282 * \param buffers [out] List of buffers returned by DRI2 query.
1283 * \param buffer_count [out] Number of buffers returned.
1285 * \see intel_update_renderbuffers()
1286 * \see DRI2GetBuffers()
1287 * \see DRI2GetBuffersWithFormat()
1290 intel_query_dri2_buffers(struct brw_context
*brw
,
1291 __DRIdrawable
*drawable
,
1292 __DRIbuffer
**buffers
,
1295 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1296 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1298 unsigned attachments
[8];
1300 struct intel_renderbuffer
*front_rb
;
1301 struct intel_renderbuffer
*back_rb
;
1303 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1304 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1306 memset(attachments
, 0, sizeof(attachments
));
1307 if ((brw_is_front_buffer_drawing(fb
) ||
1308 brw_is_front_buffer_reading(fb
) ||
1309 !back_rb
) && front_rb
) {
1310 /* If a fake front buffer is in use, then querying for
1311 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1312 * the real front buffer to the fake front buffer. So before doing the
1313 * query, we need to make sure all the pending drawing has landed in the
1314 * real front buffer.
1316 intel_batchbuffer_flush(brw
);
1317 intel_flush_front(&brw
->ctx
);
1319 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1320 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1321 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1322 /* We have pending front buffer rendering, but we aren't querying for a
1323 * front buffer. If the front buffer we have is a fake front buffer,
1324 * the X server is going to throw it away when it processes the query.
1325 * So before doing the query, make sure all the pending drawing has
1326 * landed in the real front buffer.
1328 intel_batchbuffer_flush(brw
);
1329 intel_flush_front(&brw
->ctx
);
1333 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1334 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1337 assert(i
<= ARRAY_SIZE(attachments
));
1339 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1344 drawable
->loaderPrivate
);
1348 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1350 * This is called from intel_update_renderbuffers().
1353 * DRI buffers whose attachment point is DRI2BufferStencil or
1354 * DRI2BufferDepthStencil are handled as special cases.
1356 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1357 * that is passed to drm_intel_bo_gem_create_from_name().
1359 * \see intel_update_renderbuffers()
1362 intel_process_dri2_buffer(struct brw_context
*brw
,
1363 __DRIdrawable
*drawable
,
1364 __DRIbuffer
*buffer
,
1365 struct intel_renderbuffer
*rb
,
1366 const char *buffer_name
)
1368 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1374 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1376 /* We try to avoid closing and reopening the same BO name, because the first
1377 * use of a mapping of the buffer involves a bunch of page faulting which is
1378 * moderately expensive.
1380 struct intel_mipmap_tree
*last_mt
;
1381 if (num_samples
== 0)
1384 last_mt
= rb
->singlesample_mt
;
1386 uint32_t old_name
= 0;
1388 /* The bo already has a name because the miptree was created by a
1389 * previous call to intel_process_dri2_buffer(). If a bo already has a
1390 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1391 * create a new name.
1393 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1396 if (old_name
== buffer
->name
)
1399 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1401 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1402 buffer
->name
, buffer
->attachment
,
1403 buffer
->cpp
, buffer
->pitch
);
1406 intel_miptree_release(&rb
->mt
);
1407 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1411 "Failed to open BO for returned DRI2 buffer "
1412 "(%dx%d, %s, named %d).\n"
1413 "This is likely a bug in the X Server that will lead to a "
1415 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1419 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1420 drawable
->w
, drawable
->h
,
1423 if (brw_is_front_buffer_drawing(fb
) &&
1424 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1425 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1426 rb
->Base
.Base
.NumSamples
> 1) {
1427 intel_renderbuffer_upsample(brw
, rb
);
1432 drm_intel_bo_unreference(bo
);
1436 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1438 * To determine which DRI buffers to request, examine the renderbuffers
1439 * attached to the drawable's framebuffer. Then request the buffers from
1442 * This is called from intel_update_renderbuffers().
1444 * \param drawable Drawable whose buffers are queried.
1445 * \param buffers [out] List of buffers returned by DRI2 query.
1446 * \param buffer_count [out] Number of buffers returned.
1448 * \see intel_update_renderbuffers()
1452 intel_update_image_buffer(struct brw_context
*intel
,
1453 __DRIdrawable
*drawable
,
1454 struct intel_renderbuffer
*rb
,
1456 enum __DRIimageBufferMask buffer_type
)
1458 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1460 if (!rb
|| !buffer
->bo
)
1463 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1465 /* Check and see if we're already bound to the right
1468 struct intel_mipmap_tree
*last_mt
;
1469 if (num_samples
== 0)
1472 last_mt
= rb
->singlesample_mt
;
1474 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1477 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1478 buffer
->width
, buffer
->height
,
1481 if (brw_is_front_buffer_drawing(fb
) &&
1482 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1483 rb
->Base
.Base
.NumSamples
> 1) {
1484 intel_renderbuffer_upsample(intel
, rb
);
1489 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1491 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1492 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1493 struct intel_renderbuffer
*front_rb
;
1494 struct intel_renderbuffer
*back_rb
;
1495 struct __DRIimageList images
;
1496 unsigned int format
;
1497 uint32_t buffer_mask
= 0;
1499 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1500 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1503 format
= intel_rb_format(back_rb
);
1505 format
= intel_rb_format(front_rb
);
1509 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1510 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1511 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1515 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1517 (*screen
->image
.loader
->getBuffers
) (drawable
,
1518 driGLFormatToImageFormat(format
),
1519 &drawable
->dri2
.stamp
,
1520 drawable
->loaderPrivate
,
1524 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1525 drawable
->w
= images
.front
->width
;
1526 drawable
->h
= images
.front
->height
;
1527 intel_update_image_buffer(brw
,
1531 __DRI_IMAGE_BUFFER_FRONT
);
1533 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1534 drawable
->w
= images
.back
->width
;
1535 drawable
->h
= images
.back
->height
;
1536 intel_update_image_buffer(brw
,
1540 __DRI_IMAGE_BUFFER_BACK
);