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 /***************************************
72 * Mesa's Driver Functions
73 ***************************************/
76 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
77 GLenum internalFormat
, int samples
[16])
79 struct brw_context
*brw
= brw_context(ctx
);
105 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
108 brw_get_renderer_string(unsigned deviceID
)
111 static char buffer
[128];
115 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
116 #include "pci_ids/i965_pci_ids.h"
118 chipset
= "Unknown Intel Chipset";
122 (void) driGetRendererString(buffer
, chipset
, 0);
126 static const GLubyte
*
127 intel_get_string(struct gl_context
* ctx
, GLenum name
)
129 const struct brw_context
*const brw
= brw_context(ctx
);
133 return (GLubyte
*) brw_vendor_string
;
137 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
145 intel_viewport(struct gl_context
*ctx
)
147 struct brw_context
*brw
= brw_context(ctx
);
148 __DRIcontext
*driContext
= brw
->driContext
;
150 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
151 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
152 dri2InvalidateDrawable(driContext
->driReadablePriv
);
157 intel_update_state(struct gl_context
* ctx
, GLuint new_state
)
159 struct brw_context
*brw
= brw_context(ctx
);
160 struct intel_texture_object
*tex_obj
;
161 struct intel_renderbuffer
*depth_irb
;
163 if (ctx
->swrast_context
)
164 _swrast_InvalidateState(ctx
, new_state
);
165 _vbo_InvalidateState(ctx
, new_state
);
167 brw
->NewGLState
|= new_state
;
169 _mesa_unlock_context_textures(ctx
);
171 /* Resolve the depth buffer's HiZ buffer. */
172 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
174 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
176 /* Resolve depth buffer and render cache of each enabled texture. */
177 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
178 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
179 if (!ctx
->Texture
.Unit
[i
]._Current
)
181 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
182 if (!tex_obj
|| !tex_obj
->mt
)
184 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
185 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
186 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
189 _mesa_lock_context_textures(ctx
);
192 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
195 intel_flush_front(struct gl_context
*ctx
)
197 struct brw_context
*brw
= brw_context(ctx
);
198 __DRIcontext
*driContext
= brw
->driContext
;
199 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
200 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
202 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
203 if (flushFront(screen
) && driDrawable
&&
204 driDrawable
->loaderPrivate
) {
206 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
208 * This potentially resolves both front and back buffer. It
209 * is unnecessary to resolve the back, but harms nothing except
210 * performance. And no one cares about front-buffer render
213 intel_resolve_for_dri2_flush(brw
, driDrawable
);
214 intel_batchbuffer_flush(brw
);
216 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
218 /* We set the dirty bit in intel_prepare_render() if we're
219 * front buffer rendering once we get there.
221 brw
->front_buffer_dirty
= false;
227 intel_glFlush(struct gl_context
*ctx
)
229 struct brw_context
*brw
= brw_context(ctx
);
231 intel_batchbuffer_flush(brw
);
232 intel_flush_front(ctx
);
233 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
234 brw
->need_throttle
= true;
238 intel_finish(struct gl_context
* ctx
)
240 struct brw_context
*brw
= brw_context(ctx
);
244 if (brw
->batch
.last_bo
)
245 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
249 brw_init_driver_functions(struct brw_context
*brw
,
250 struct dd_function_table
*functions
)
252 _mesa_init_driver_functions(functions
);
254 /* GLX uses DRI2 invalidate events to handle window resizing.
255 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
256 * which doesn't provide a mechanism for snooping the event queues.
258 * So EGL still relies on viewport hacks to handle window resizing.
259 * This should go away with DRI3000.
261 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
262 functions
->Viewport
= intel_viewport
;
264 functions
->Flush
= intel_glFlush
;
265 functions
->Finish
= intel_finish
;
266 functions
->GetString
= intel_get_string
;
267 functions
->UpdateState
= intel_update_state
;
269 intelInitTextureFuncs(functions
);
270 intelInitTextureImageFuncs(functions
);
271 intelInitTextureSubImageFuncs(functions
);
272 intelInitTextureCopyImageFuncs(functions
);
273 intelInitCopyImageFuncs(functions
);
274 intelInitClearFuncs(functions
);
275 intelInitBufferFuncs(functions
);
276 intelInitPixelFuncs(functions
);
277 intelInitBufferObjectFuncs(functions
);
278 intel_init_syncobj_functions(functions
);
279 brw_init_object_purgeable_functions(functions
);
281 brwInitFragProgFuncs( functions
);
282 brw_init_common_queryobj_functions(functions
);
284 gen6_init_queryobj_functions(functions
);
286 gen4_init_queryobj_functions(functions
);
288 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
290 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
291 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
292 functions
->GetTransformFeedbackVertexCount
=
293 brw_get_transform_feedback_vertex_count
;
295 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
296 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
297 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
298 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
300 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
301 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
305 functions
->GetSamplePosition
= gen6_get_sample_position
;
309 brw_initialize_context_constants(struct brw_context
*brw
)
311 struct gl_context
*ctx
= &brw
->ctx
;
313 unsigned max_samplers
=
314 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
316 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
318 ctx
->Const
.StripTextureBorder
= true;
320 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
321 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
322 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
323 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
324 ctx
->Const
.MaxTextureUnits
=
325 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
326 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
327 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
329 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
331 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
332 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
333 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
334 ctx
->Const
.MaxUniformBufferBindings
+= 12;
336 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
338 ctx
->Const
.MaxCombinedTextureImageUnits
=
339 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
340 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
341 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
342 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
344 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
345 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
346 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
347 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
348 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
349 ctx
->Const
.MaxTextureMbytes
= 1536;
352 ctx
->Const
.MaxArrayTextureLayers
= 2048;
354 ctx
->Const
.MaxArrayTextureLayers
= 512;
356 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
358 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
360 ctx
->Const
.MaxRenderbufferSize
= 8192;
362 /* Hardware only supports a limited number of transform feedback buffers.
363 * So we need to override the Mesa default (which is based only on software
366 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
368 /* On Gen6, in the worst case, we use up one binding table entry per
369 * transform feedback component (see comments above the definition of
370 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
371 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
372 * BRW_MAX_SOL_BINDINGS.
374 * In "separate components" mode, we need to divide this value by
375 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
376 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
378 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
379 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
380 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
382 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
385 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
386 const int clamp_max_samples
=
387 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
389 if (clamp_max_samples
< 0) {
390 max_samples
= msaa_modes
[0];
392 /* Select the largest supported MSAA mode that does not exceed
396 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
397 if (msaa_modes
[i
] <= clamp_max_samples
) {
398 max_samples
= msaa_modes
[i
];
404 ctx
->Const
.MaxSamples
= max_samples
;
405 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
406 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
407 ctx
->Const
.MaxIntegerSamples
= max_samples
;
410 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
411 else if (brw
->gen
== 6)
412 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
414 ctx
->Const
.MinLineWidth
= 1.0;
415 ctx
->Const
.MinLineWidthAA
= 1.0;
416 ctx
->Const
.MaxLineWidth
= 5.0;
417 ctx
->Const
.MaxLineWidthAA
= 5.0;
418 ctx
->Const
.LineWidthGranularity
= 0.5;
420 ctx
->Const
.MinPointSize
= 1.0;
421 ctx
->Const
.MinPointSizeAA
= 1.0;
422 ctx
->Const
.MaxPointSize
= 255.0;
423 ctx
->Const
.MaxPointSizeAA
= 255.0;
424 ctx
->Const
.PointSizeGranularity
= 1.0;
426 if (brw
->gen
>= 5 || brw
->is_g4x
)
427 ctx
->Const
.MaxClipPlanes
= 8;
429 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
430 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
431 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
432 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
433 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
434 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
435 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
436 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
437 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
438 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
439 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
440 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
441 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
442 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
444 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
445 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
446 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
447 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
448 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
449 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
450 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
451 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
452 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
453 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
454 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
456 /* Fragment shaders use real, 32-bit twos-complement integers for all
459 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
460 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
461 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
462 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
463 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
466 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
467 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
468 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
469 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
470 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
471 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
472 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
473 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
474 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
477 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
478 * but we're not sure how it's actually done for vertex order,
479 * that affect provoking vertex decision. Always use last vertex
480 * convention for quad primitive which works as expected for now.
483 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
485 ctx
->Const
.NativeIntegers
= true;
487 /* Regarding the CMP instruction, the Ivybridge PRM says:
489 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
490 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
491 * 0xFFFFFFFF) is assigned to dst."
493 * but PRMs for earlier generations say
495 * "In dword format, one GRF may store up to 8 results. When the register
496 * is used later as a vector of Booleans, as only LSB at each channel
497 * contains meaning [sic] data, software should make sure all higher bits
498 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
500 * We select the representation of a true boolean uniform to match what the
501 * CMP instruction returns.
503 * The Sandybridge BSpec's description of the CMP instruction matches that
504 * of the Ivybridge PRM. (The description in the Sandybridge PRM is seems
505 * to have not been updated from Ironlake). Its CMP instruction behaves like
506 * Ivybridge and newer.
509 ctx
->Const
.UniformBooleanTrue
= ~0;
511 ctx
->Const
.UniformBooleanTrue
= 1;
513 /* From the gen4 PRM, volume 4 page 127:
515 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
516 * the base address of the first element of the surface, computed in
517 * software by adding the surface base address to the byte offset of
518 * the element in the buffer."
520 * However, unaligned accesses are slower, so enforce buffer alignment.
522 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
523 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
526 ctx
->Const
.MaxVarying
= 32;
527 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
528 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
529 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
530 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
533 /* We want the GLSL compiler to emit code that uses condition codes */
534 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
535 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
536 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
537 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
538 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
539 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
540 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
541 (i
== MESA_SHADER_FRAGMENT
);
542 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
543 (i
== MESA_SHADER_FRAGMENT
);
544 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
545 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
548 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
549 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
551 /* ARB_viewport_array */
552 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
553 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
554 ctx
->Const
.ViewportSubpixelBits
= 0;
556 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
558 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
559 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
562 /* ARB_gpu_shader5 */
564 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
568 * Process driconf (drirc) options, setting appropriate context flags.
570 * intelInitExtensions still pokes at optionCache directly, in order to
571 * avoid advertising various extensions. No flags are set, so it makes
572 * sense to continue doing that there.
575 brw_process_driconf_options(struct brw_context
*brw
)
577 struct gl_context
*ctx
= &brw
->ctx
;
579 driOptionCache
*options
= &brw
->optionCache
;
580 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
581 brw
->driContext
->driScreenPriv
->myNum
, "i965");
583 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
584 switch (bo_reuse_mode
) {
585 case DRI_CONF_BO_REUSE_DISABLED
:
587 case DRI_CONF_BO_REUSE_ALL
:
588 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
592 if (!driQueryOptionb(options
, "hiz")) {
593 brw
->has_hiz
= false;
594 /* On gen6, you can only do separate stencil with HIZ. */
596 brw
->has_separate_stencil
= false;
599 if (driQueryOptionb(options
, "always_flush_batch")) {
600 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
601 brw
->always_flush_batch
= true;
604 if (driQueryOptionb(options
, "always_flush_cache")) {
605 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
606 brw
->always_flush_cache
= true;
609 if (driQueryOptionb(options
, "disable_throttling")) {
610 fprintf(stderr
, "disabling flush throttling\n");
611 brw
->disable_throttling
= true;
614 brw
->disable_derivative_optimization
=
615 driQueryOptionb(&brw
->optionCache
, "disable_derivative_optimization");
617 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
619 ctx
->Const
.ForceGLSLExtensionsWarn
=
620 driQueryOptionb(options
, "force_glsl_extensions_warn");
622 ctx
->Const
.DisableGLSLLineContinuations
=
623 driQueryOptionb(options
, "disable_glsl_line_continuations");
625 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
626 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
630 brwCreateContext(gl_api api
,
631 const struct gl_config
*mesaVis
,
632 __DRIcontext
*driContextPriv
,
633 unsigned major_version
,
634 unsigned minor_version
,
637 unsigned *dri_ctx_error
,
638 void *sharedContextPrivate
)
640 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
641 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
642 struct intel_screen
*screen
= sPriv
->driverPrivate
;
643 const struct brw_device_info
*devinfo
= screen
->devinfo
;
644 struct dd_function_table functions
;
646 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
647 * provides us with context reset notifications.
649 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
650 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
652 if (screen
->has_context_reset_notification
)
653 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
655 if (flags
& ~allowed_flags
) {
656 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
660 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
662 fprintf(stderr
, "%s: failed to alloc context\n", __FUNCTION__
);
663 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
667 driContextPriv
->driverPrivate
= brw
;
668 brw
->driContext
= driContextPriv
;
669 brw
->intelScreen
= screen
;
670 brw
->bufmgr
= screen
->bufmgr
;
672 brw
->gen
= devinfo
->gen
;
673 brw
->gt
= devinfo
->gt
;
674 brw
->is_g4x
= devinfo
->is_g4x
;
675 brw
->is_baytrail
= devinfo
->is_baytrail
;
676 brw
->is_haswell
= devinfo
->is_haswell
;
677 brw
->is_cherryview
= devinfo
->is_cherryview
;
678 brw
->has_llc
= devinfo
->has_llc
;
679 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
680 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
681 brw
->has_pln
= devinfo
->has_pln
;
682 brw
->has_compr4
= devinfo
->has_compr4
;
683 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
684 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
685 brw
->needs_unlit_centroid_workaround
=
686 devinfo
->needs_unlit_centroid_workaround
;
688 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
689 brw
->has_swizzling
= screen
->hw_has_swizzling
;
691 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
692 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
693 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
695 gen8_init_vtable_surface_functions(brw
);
696 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
697 } else if (brw
->gen
>= 7) {
698 gen7_init_vtable_surface_functions(brw
);
699 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
700 } else if (brw
->gen
>= 6) {
701 gen6_init_vtable_surface_functions(brw
);
702 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
704 gen4_init_vtable_surface_functions(brw
);
705 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
708 brw_init_driver_functions(brw
, &functions
);
711 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
713 struct gl_context
*ctx
= &brw
->ctx
;
715 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
716 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
717 fprintf(stderr
, "%s: failed to init mesa context\n", __FUNCTION__
);
718 intelDestroyContext(driContextPriv
);
722 driContextSetFlags(ctx
, flags
);
724 /* Initialize the software rasterizer and helper modules.
726 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
727 * software fallbacks (which we have to support on legacy GL to do weird
728 * glDrawPixels(), glBitmap(), and other functions).
730 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
731 _swrast_CreateContext(ctx
);
734 _vbo_CreateContext(ctx
);
735 if (ctx
->swrast_context
) {
736 _tnl_CreateContext(ctx
);
737 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
738 _swsetup_CreateContext(ctx
);
740 /* Configure swrast to match hardware characteristics: */
741 _swrast_allow_pixel_fog(ctx
, false);
742 _swrast_allow_vertex_fog(ctx
, true);
745 _mesa_meta_init(ctx
);
747 brw_process_driconf_options(brw
);
748 brw_process_intel_debug_variable(brw
);
749 brw_initialize_context_constants(brw
);
751 ctx
->Const
.ResetStrategy
= notify_reset
752 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
754 /* Reinitialize the context point state. It depends on ctx->Const values. */
755 _mesa_init_point(ctx
);
759 intel_batchbuffer_init(brw
);
762 /* Create a new hardware context. Using a hardware context means that
763 * our GPU state will be saved/restored on context switch, allowing us
764 * to assume that the GPU is in the same state we left it in.
766 * This is required for transform feedback buffer offsets, query objects,
767 * and also allows us to reduce how much state we have to emit.
769 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
772 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
773 intelDestroyContext(driContextPriv
);
780 intelInitExtensions(ctx
);
782 brw_init_surface_formats(brw
);
784 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
785 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
786 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
787 brw
->urb
.size
= devinfo
->urb
.size
;
788 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
789 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
790 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
792 /* Estimate the size of the mappable aperture into the GTT. There's an
793 * ioctl to get the whole GTT size, but not one to get the mappable subset.
794 * It turns out it's basically always 256MB, though some ancient hardware
797 uint32_t gtt_size
= 256 * 1024 * 1024;
799 /* We don't want to map two objects such that a memcpy between them would
800 * just fault one mapping in and then the other over and over forever. So
801 * we would need to divide the GTT size by 2. Additionally, some GTT is
802 * taken up by things like the framebuffer and the ringbuffer and such, so
803 * be more conservative.
805 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
808 brw
->urb
.gen6_gs_previously_active
= false;
810 brw
->prim_restart
.in_progress
= false;
811 brw
->prim_restart
.enable_cut_index
= false;
812 brw
->gs
.enabled
= false;
813 brw
->sf
.viewport_transform_enable
= true;
815 ctx
->VertexProgram
._MaintainTnlProgram
= true;
816 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
818 brw_draw_init( brw
);
820 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
821 /* Turn on some extra GL_ARB_debug_output generation. */
822 brw
->perf_debug
= true;
825 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
826 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
828 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
829 brw_init_shader_time(brw
);
831 _mesa_compute_version(ctx
);
833 _mesa_initialize_dispatch_tables(ctx
);
834 _mesa_initialize_vbo_vtxfmt(ctx
);
836 if (ctx
->Extensions
.AMD_performance_monitor
) {
837 brw_init_performance_monitors(brw
);
840 vbo_use_buffer_objects(ctx
);
841 vbo_always_unmap_buffers(ctx
);
847 intelDestroyContext(__DRIcontext
* driContextPriv
)
849 struct brw_context
*brw
=
850 (struct brw_context
*) driContextPriv
->driverPrivate
;
851 struct gl_context
*ctx
= &brw
->ctx
;
853 assert(brw
); /* should never be null */
857 /* Dump a final BMP in case the application doesn't call SwapBuffers */
858 if (INTEL_DEBUG
& DEBUG_AUB
) {
859 intel_batchbuffer_flush(brw
);
860 aub_dump_bmp(&brw
->ctx
);
863 _mesa_meta_free(&brw
->ctx
);
864 brw_meta_fast_clear_free(brw
);
866 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
867 /* Force a report. */
868 brw
->shader_time
.report_time
= 0;
870 brw_collect_and_report_shader_time(brw
);
871 brw_destroy_shader_time(brw
);
874 brw_destroy_state(brw
);
875 brw_draw_destroy(brw
);
877 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
879 drm_intel_gem_context_destroy(brw
->hw_ctx
);
881 if (ctx
->swrast_context
) {
882 _swsetup_DestroyContext(&brw
->ctx
);
883 _tnl_DestroyContext(&brw
->ctx
);
885 _vbo_DestroyContext(&brw
->ctx
);
887 if (ctx
->swrast_context
)
888 _swrast_DestroyContext(&brw
->ctx
);
890 intel_batchbuffer_free(brw
);
892 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
893 brw
->first_post_swapbuffers_batch
= NULL
;
895 driDestroyOptionCache(&brw
->optionCache
);
897 /* free the Mesa context */
898 _mesa_free_context_data(&brw
->ctx
);
901 driContextPriv
->driverPrivate
= NULL
;
905 intelUnbindContext(__DRIcontext
* driContextPriv
)
907 /* Unset current context and dispath table */
908 _mesa_make_current(NULL
, NULL
, NULL
);
914 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
915 * on window system framebuffers.
917 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
918 * your renderbuffer can do sRGB encode, and you can flip a switch that does
919 * sRGB encode if the renderbuffer can handle it. You can ask specifically
920 * for a visual where you're guaranteed to be capable, but it turns out that
921 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
922 * incapable ones, becuase there's no difference between the two in resources
923 * used. Applications thus get built that accidentally rely on the default
924 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
927 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
928 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
929 * So they removed the enable knob and made it "if the renderbuffer is sRGB
930 * capable, do sRGB encode". Then, for your window system renderbuffers, you
931 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
932 * and get no sRGB encode (assuming that both kinds of visual are available).
933 * Thus our choice to support sRGB by default on our visuals for desktop would
934 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
936 * Unfortunately, renderbuffer setup happens before a context is created. So
937 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
938 * context (without an sRGB visual, though we don't have sRGB visuals exposed
939 * yet), we go turn that back off before anyone finds out.
942 intel_gles3_srgb_workaround(struct brw_context
*brw
,
943 struct gl_framebuffer
*fb
)
945 struct gl_context
*ctx
= &brw
->ctx
;
947 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
950 /* Some day when we support the sRGB capable bit on visuals available for
951 * GLES, we'll need to respect that and not disable things here.
953 fb
->Visual
.sRGBCapable
= false;
954 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
955 if (fb
->Attachment
[i
].Renderbuffer
&&
956 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
957 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
963 intelMakeCurrent(__DRIcontext
* driContextPriv
,
964 __DRIdrawable
* driDrawPriv
,
965 __DRIdrawable
* driReadPriv
)
967 struct brw_context
*brw
;
968 GET_CURRENT_CONTEXT(curCtx
);
971 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
975 /* According to the glXMakeCurrent() man page: "Pending commands to
976 * the previous context, if any, are flushed before it is released."
977 * But only flush if we're actually changing contexts.
979 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
983 if (driContextPriv
) {
984 struct gl_context
*ctx
= &brw
->ctx
;
985 struct gl_framebuffer
*fb
, *readFb
;
987 if (driDrawPriv
== NULL
&& driReadPriv
== NULL
) {
988 fb
= _mesa_get_incomplete_framebuffer();
989 readFb
= _mesa_get_incomplete_framebuffer();
991 fb
= driDrawPriv
->driverPrivate
;
992 readFb
= driReadPriv
->driverPrivate
;
993 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
994 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
997 /* The sRGB workaround changes the renderbuffer's format. We must change
998 * the format before the renderbuffer's miptree get's allocated, otherwise
999 * the formats of the renderbuffer and its miptree will differ.
1001 intel_gles3_srgb_workaround(brw
, fb
);
1002 intel_gles3_srgb_workaround(brw
, readFb
);
1004 /* If the context viewport hasn't been initialized, force a call out to
1005 * the loader to get buffers so we have a drawable size for the initial
1007 if (!brw
->ctx
.ViewportInitialized
)
1008 intel_prepare_render(brw
);
1010 _mesa_make_current(ctx
, fb
, readFb
);
1012 _mesa_make_current(NULL
, NULL
, NULL
);
1019 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1020 __DRIdrawable
*drawable
)
1023 /* MSAA and fast color clear are not supported, so don't waste time
1024 * checking whether a resolve is needed.
1029 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1030 struct intel_renderbuffer
*rb
;
1032 /* Usually, only the back buffer will need to be downsampled. However,
1033 * the front buffer will also need it if the user has rendered into it.
1035 static const gl_buffer_index buffers
[2] = {
1040 for (int i
= 0; i
< 2; ++i
) {
1041 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1042 if (rb
== NULL
|| rb
->mt
== NULL
)
1044 if (rb
->mt
->num_samples
<= 1)
1045 intel_miptree_resolve_color(brw
, rb
->mt
);
1047 intel_renderbuffer_downsample(brw
, rb
);
1052 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1054 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1058 intel_query_dri2_buffers(struct brw_context
*brw
,
1059 __DRIdrawable
*drawable
,
1060 __DRIbuffer
**buffers
,
1064 intel_process_dri2_buffer(struct brw_context
*brw
,
1065 __DRIdrawable
*drawable
,
1066 __DRIbuffer
*buffer
,
1067 struct intel_renderbuffer
*rb
,
1068 const char *buffer_name
);
1071 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1074 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1076 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1077 struct intel_renderbuffer
*rb
;
1078 __DRIbuffer
*buffers
= NULL
;
1080 const char *region_name
;
1082 /* Set this up front, so that in case our buffers get invalidated
1083 * while we're getting new buffers, we don't clobber the stamp and
1084 * thus ignore the invalidate. */
1085 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1087 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1088 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1090 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1092 if (buffers
== NULL
)
1095 for (i
= 0; i
< count
; i
++) {
1096 switch (buffers
[i
].attachment
) {
1097 case __DRI_BUFFER_FRONT_LEFT
:
1098 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1099 region_name
= "dri2 front buffer";
1102 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1103 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1104 region_name
= "dri2 fake front buffer";
1107 case __DRI_BUFFER_BACK_LEFT
:
1108 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1109 region_name
= "dri2 back buffer";
1112 case __DRI_BUFFER_DEPTH
:
1113 case __DRI_BUFFER_HIZ
:
1114 case __DRI_BUFFER_DEPTH_STENCIL
:
1115 case __DRI_BUFFER_STENCIL
:
1116 case __DRI_BUFFER_ACCUM
:
1119 "unhandled buffer attach event, attachment type %d\n",
1120 buffers
[i
].attachment
);
1124 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1130 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1132 struct brw_context
*brw
= context
->driverPrivate
;
1133 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1135 /* Set this up front, so that in case our buffers get invalidated
1136 * while we're getting new buffers, we don't clobber the stamp and
1137 * thus ignore the invalidate. */
1138 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1140 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1141 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1143 if (screen
->image
.loader
)
1144 intel_update_image_buffers(brw
, drawable
);
1146 intel_update_dri2_buffers(brw
, drawable
);
1148 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1152 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1153 * state is required.
1156 intel_prepare_render(struct brw_context
*brw
)
1158 struct gl_context
*ctx
= &brw
->ctx
;
1159 __DRIcontext
*driContext
= brw
->driContext
;
1160 __DRIdrawable
*drawable
;
1162 drawable
= driContext
->driDrawablePriv
;
1163 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1164 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1165 intel_update_renderbuffers(driContext
, drawable
);
1166 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1169 drawable
= driContext
->driReadablePriv
;
1170 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1171 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1172 intel_update_renderbuffers(driContext
, drawable
);
1173 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1176 /* If we're currently rendering to the front buffer, the rendering
1177 * that will happen next will probably dirty the front buffer. So
1178 * mark it as dirty here.
1180 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1181 brw
->front_buffer_dirty
= true;
1183 /* Wait for the swapbuffers before the one we just emitted, so we
1184 * don't get too many swaps outstanding for apps that are GPU-heavy
1185 * but not CPU-heavy.
1187 * We're using intelDRI2Flush (called from the loader before
1188 * swapbuffer) and glFlush (for front buffer rendering) as the
1189 * indicator that a frame is done and then throttle when we get
1190 * here as we prepare to render the next frame. At this point for
1191 * round trips for swap/copy and getting new buffers are done and
1192 * we'll spend less time waiting on the GPU.
1194 * Unfortunately, we don't have a handle to the batch containing
1195 * the swap, and getting our hands on that doesn't seem worth it,
1196 * so we just us the first batch we emitted after the last swap.
1198 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1199 if (!brw
->disable_throttling
)
1200 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1201 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1202 brw
->first_post_swapbuffers_batch
= NULL
;
1203 brw
->need_throttle
= false;
1208 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1210 * To determine which DRI buffers to request, examine the renderbuffers
1211 * attached to the drawable's framebuffer. Then request the buffers with
1212 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1214 * This is called from intel_update_renderbuffers().
1216 * \param drawable Drawable whose buffers are queried.
1217 * \param buffers [out] List of buffers returned by DRI2 query.
1218 * \param buffer_count [out] Number of buffers returned.
1220 * \see intel_update_renderbuffers()
1221 * \see DRI2GetBuffers()
1222 * \see DRI2GetBuffersWithFormat()
1225 intel_query_dri2_buffers(struct brw_context
*brw
,
1226 __DRIdrawable
*drawable
,
1227 __DRIbuffer
**buffers
,
1230 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1231 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1233 unsigned attachments
[8];
1235 struct intel_renderbuffer
*front_rb
;
1236 struct intel_renderbuffer
*back_rb
;
1238 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1239 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1241 memset(attachments
, 0, sizeof(attachments
));
1242 if ((brw_is_front_buffer_drawing(fb
) ||
1243 brw_is_front_buffer_reading(fb
) ||
1244 !back_rb
) && front_rb
) {
1245 /* If a fake front buffer is in use, then querying for
1246 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1247 * the real front buffer to the fake front buffer. So before doing the
1248 * query, we need to make sure all the pending drawing has landed in the
1249 * real front buffer.
1251 intel_batchbuffer_flush(brw
);
1252 intel_flush_front(&brw
->ctx
);
1254 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1255 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1256 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1257 /* We have pending front buffer rendering, but we aren't querying for a
1258 * front buffer. If the front buffer we have is a fake front buffer,
1259 * the X server is going to throw it away when it processes the query.
1260 * So before doing the query, make sure all the pending drawing has
1261 * landed in the real front buffer.
1263 intel_batchbuffer_flush(brw
);
1264 intel_flush_front(&brw
->ctx
);
1268 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1269 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1272 assert(i
<= ARRAY_SIZE(attachments
));
1274 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1279 drawable
->loaderPrivate
);
1283 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1285 * This is called from intel_update_renderbuffers().
1288 * DRI buffers whose attachment point is DRI2BufferStencil or
1289 * DRI2BufferDepthStencil are handled as special cases.
1291 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1292 * that is passed to drm_intel_bo_gem_create_from_name().
1294 * \see intel_update_renderbuffers()
1297 intel_process_dri2_buffer(struct brw_context
*brw
,
1298 __DRIdrawable
*drawable
,
1299 __DRIbuffer
*buffer
,
1300 struct intel_renderbuffer
*rb
,
1301 const char *buffer_name
)
1303 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1309 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1311 /* We try to avoid closing and reopening the same BO name, because the first
1312 * use of a mapping of the buffer involves a bunch of page faulting which is
1313 * moderately expensive.
1315 struct intel_mipmap_tree
*last_mt
;
1316 if (num_samples
== 0)
1319 last_mt
= rb
->singlesample_mt
;
1321 uint32_t old_name
= 0;
1323 /* The bo already has a name because the miptree was created by a
1324 * previous call to intel_process_dri2_buffer(). If a bo already has a
1325 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1326 * create a new name.
1328 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1331 if (old_name
== buffer
->name
)
1334 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1336 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1337 buffer
->name
, buffer
->attachment
,
1338 buffer
->cpp
, buffer
->pitch
);
1341 intel_miptree_release(&rb
->mt
);
1342 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1346 "Failed to open BO for returned DRI2 buffer "
1347 "(%dx%d, %s, named %d).\n"
1348 "This is likely a bug in the X Server that will lead to a "
1350 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1354 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1355 drawable
->w
, drawable
->h
,
1358 if (brw_is_front_buffer_drawing(fb
) &&
1359 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1360 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1361 rb
->Base
.Base
.NumSamples
> 1) {
1362 intel_renderbuffer_upsample(brw
, rb
);
1367 drm_intel_bo_unreference(bo
);
1371 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1373 * To determine which DRI buffers to request, examine the renderbuffers
1374 * attached to the drawable's framebuffer. Then request the buffers from
1377 * This is called from intel_update_renderbuffers().
1379 * \param drawable Drawable whose buffers are queried.
1380 * \param buffers [out] List of buffers returned by DRI2 query.
1381 * \param buffer_count [out] Number of buffers returned.
1383 * \see intel_update_renderbuffers()
1387 intel_update_image_buffer(struct brw_context
*intel
,
1388 __DRIdrawable
*drawable
,
1389 struct intel_renderbuffer
*rb
,
1391 enum __DRIimageBufferMask buffer_type
)
1393 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1395 if (!rb
|| !buffer
->bo
)
1398 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1400 /* Check and see if we're already bound to the right
1403 struct intel_mipmap_tree
*last_mt
;
1404 if (num_samples
== 0)
1407 last_mt
= rb
->singlesample_mt
;
1409 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1412 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1413 buffer
->width
, buffer
->height
,
1416 if (brw_is_front_buffer_drawing(fb
) &&
1417 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1418 rb
->Base
.Base
.NumSamples
> 1) {
1419 intel_renderbuffer_upsample(intel
, rb
);
1424 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1426 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1427 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1428 struct intel_renderbuffer
*front_rb
;
1429 struct intel_renderbuffer
*back_rb
;
1430 struct __DRIimageList images
;
1431 unsigned int format
;
1432 uint32_t buffer_mask
= 0;
1434 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1435 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1438 format
= intel_rb_format(back_rb
);
1440 format
= intel_rb_format(front_rb
);
1444 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1445 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1446 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1450 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1452 (*screen
->image
.loader
->getBuffers
) (drawable
,
1453 driGLFormatToImageFormat(format
),
1454 &drawable
->dri2
.stamp
,
1455 drawable
->loaderPrivate
,
1459 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1460 drawable
->w
= images
.front
->width
;
1461 drawable
->h
= images
.front
->height
;
1462 intel_update_image_buffer(brw
,
1466 __DRI_IMAGE_BUFFER_FRONT
);
1468 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1469 drawable
->w
= images
.back
->width
;
1470 drawable
->h
= images
.back
->height
;
1471 intel_update_image_buffer(brw
,
1475 __DRI_IMAGE_BUFFER_BACK
);