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
;
409 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
410 * to map indices of rectangular grid to sample numbers within a pixel.
411 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
412 * extension implementation. For more details see the comment above
413 * gen6_set_sample_maps() definition.
415 gen6_set_sample_maps(ctx
);
418 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
419 else if (brw
->gen
== 6)
420 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
422 ctx
->Const
.MinLineWidth
= 1.0;
423 ctx
->Const
.MinLineWidthAA
= 1.0;
425 ctx
->Const
.MaxLineWidth
= 7.875;
426 ctx
->Const
.MaxLineWidthAA
= 7.875;
427 ctx
->Const
.LineWidthGranularity
= 0.125;
429 ctx
->Const
.MaxLineWidth
= 7.0;
430 ctx
->Const
.MaxLineWidthAA
= 7.0;
431 ctx
->Const
.LineWidthGranularity
= 0.5;
434 ctx
->Const
.MinPointSize
= 1.0;
435 ctx
->Const
.MinPointSizeAA
= 1.0;
436 ctx
->Const
.MaxPointSize
= 255.0;
437 ctx
->Const
.MaxPointSizeAA
= 255.0;
438 ctx
->Const
.PointSizeGranularity
= 1.0;
440 if (brw
->gen
>= 5 || brw
->is_g4x
)
441 ctx
->Const
.MaxClipPlanes
= 8;
443 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
444 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
445 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
446 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
447 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
448 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
449 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
450 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
451 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
452 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
453 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
454 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
455 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
456 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
458 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
459 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
460 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
461 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
462 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
463 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
464 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
465 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
466 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
467 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
468 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
470 /* Fragment shaders use real, 32-bit twos-complement integers for all
473 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
474 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
475 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
476 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
477 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
480 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
481 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
482 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
483 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
484 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
485 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
486 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
487 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
488 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
491 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
492 * but we're not sure how it's actually done for vertex order,
493 * that affect provoking vertex decision. Always use last vertex
494 * convention for quad primitive which works as expected for now.
497 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
499 ctx
->Const
.NativeIntegers
= true;
500 ctx
->Const
.VertexID_is_zero_based
= true;
502 /* Regarding the CMP instruction, the Ivybridge PRM says:
504 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
505 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
506 * 0xFFFFFFFF) is assigned to dst."
508 * but PRMs for earlier generations say
510 * "In dword format, one GRF may store up to 8 results. When the register
511 * is used later as a vector of Booleans, as only LSB at each channel
512 * contains meaning [sic] data, software should make sure all higher bits
513 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
515 * We select the representation of a true boolean uniform to match what the
516 * CMP instruction returns.
518 * The Sandybridge BSpec's description of the CMP instruction matches that
519 * of the Ivybridge PRM. (The description in the Sandybridge PRM is seems
520 * to have not been updated from Ironlake). Its CMP instruction behaves like
521 * Ivybridge and newer.
524 ctx
->Const
.UniformBooleanTrue
= ~0;
526 ctx
->Const
.UniformBooleanTrue
= 1;
528 /* From the gen4 PRM, volume 4 page 127:
530 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
531 * the base address of the first element of the surface, computed in
532 * software by adding the surface base address to the byte offset of
533 * the element in the buffer."
535 * However, unaligned accesses are slower, so enforce buffer alignment.
537 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
538 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
541 ctx
->Const
.MaxVarying
= 32;
542 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
543 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
544 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
545 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
548 /* We want the GLSL compiler to emit code that uses condition codes */
549 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
550 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
551 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
552 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
553 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
554 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
555 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
556 (i
== MESA_SHADER_FRAGMENT
);
557 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
558 (i
== MESA_SHADER_FRAGMENT
);
559 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
560 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
563 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
564 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
566 /* ARB_viewport_array */
567 if (brw
->gen
>= 7 && ctx
->API
== API_OPENGL_CORE
) {
568 ctx
->Const
.MaxViewports
= GEN7_NUM_VIEWPORTS
;
569 ctx
->Const
.ViewportSubpixelBits
= 0;
571 /* Cast to float before negating becuase MaxViewportWidth is unsigned.
573 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
574 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
577 /* ARB_gpu_shader5 */
579 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
583 * Process driconf (drirc) options, setting appropriate context flags.
585 * intelInitExtensions still pokes at optionCache directly, in order to
586 * avoid advertising various extensions. No flags are set, so it makes
587 * sense to continue doing that there.
590 brw_process_driconf_options(struct brw_context
*brw
)
592 struct gl_context
*ctx
= &brw
->ctx
;
594 driOptionCache
*options
= &brw
->optionCache
;
595 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
596 brw
->driContext
->driScreenPriv
->myNum
, "i965");
598 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
599 switch (bo_reuse_mode
) {
600 case DRI_CONF_BO_REUSE_DISABLED
:
602 case DRI_CONF_BO_REUSE_ALL
:
603 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
607 if (!driQueryOptionb(options
, "hiz")) {
608 brw
->has_hiz
= false;
609 /* On gen6, you can only do separate stencil with HIZ. */
611 brw
->has_separate_stencil
= false;
614 if (driQueryOptionb(options
, "always_flush_batch")) {
615 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
616 brw
->always_flush_batch
= true;
619 if (driQueryOptionb(options
, "always_flush_cache")) {
620 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
621 brw
->always_flush_cache
= true;
624 if (driQueryOptionb(options
, "disable_throttling")) {
625 fprintf(stderr
, "disabling flush throttling\n");
626 brw
->disable_throttling
= true;
629 brw
->disable_derivative_optimization
=
630 driQueryOptionb(&brw
->optionCache
, "disable_derivative_optimization");
632 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
634 ctx
->Const
.ForceGLSLExtensionsWarn
=
635 driQueryOptionb(options
, "force_glsl_extensions_warn");
637 ctx
->Const
.DisableGLSLLineContinuations
=
638 driQueryOptionb(options
, "disable_glsl_line_continuations");
640 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
641 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
645 brwCreateContext(gl_api api
,
646 const struct gl_config
*mesaVis
,
647 __DRIcontext
*driContextPriv
,
648 unsigned major_version
,
649 unsigned minor_version
,
652 unsigned *dri_ctx_error
,
653 void *sharedContextPrivate
)
655 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
656 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
657 struct intel_screen
*screen
= sPriv
->driverPrivate
;
658 const struct brw_device_info
*devinfo
= screen
->devinfo
;
659 struct dd_function_table functions
;
661 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
662 * provides us with context reset notifications.
664 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
665 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
667 if (screen
->has_context_reset_notification
)
668 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
670 if (flags
& ~allowed_flags
) {
671 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
675 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
677 fprintf(stderr
, "%s: failed to alloc context\n", __FUNCTION__
);
678 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
682 driContextPriv
->driverPrivate
= brw
;
683 brw
->driContext
= driContextPriv
;
684 brw
->intelScreen
= screen
;
685 brw
->bufmgr
= screen
->bufmgr
;
687 brw
->gen
= devinfo
->gen
;
688 brw
->gt
= devinfo
->gt
;
689 brw
->is_g4x
= devinfo
->is_g4x
;
690 brw
->is_baytrail
= devinfo
->is_baytrail
;
691 brw
->is_haswell
= devinfo
->is_haswell
;
692 brw
->is_cherryview
= devinfo
->is_cherryview
;
693 brw
->has_llc
= devinfo
->has_llc
;
694 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
695 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
696 brw
->has_pln
= devinfo
->has_pln
;
697 brw
->has_compr4
= devinfo
->has_compr4
;
698 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
699 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
700 brw
->needs_unlit_centroid_workaround
=
701 devinfo
->needs_unlit_centroid_workaround
;
703 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
704 brw
->has_swizzling
= screen
->hw_has_swizzling
;
706 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
707 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
708 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
710 gen8_init_vtable_surface_functions(brw
);
711 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
712 } else if (brw
->gen
>= 7) {
713 gen7_init_vtable_surface_functions(brw
);
714 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
715 } else if (brw
->gen
>= 6) {
716 gen6_init_vtable_surface_functions(brw
);
717 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
719 gen4_init_vtable_surface_functions(brw
);
720 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
723 brw_init_driver_functions(brw
, &functions
);
726 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
728 struct gl_context
*ctx
= &brw
->ctx
;
730 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
731 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
732 fprintf(stderr
, "%s: failed to init mesa context\n", __FUNCTION__
);
733 intelDestroyContext(driContextPriv
);
737 driContextSetFlags(ctx
, flags
);
739 /* Initialize the software rasterizer and helper modules.
741 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
742 * software fallbacks (which we have to support on legacy GL to do weird
743 * glDrawPixels(), glBitmap(), and other functions).
745 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
746 _swrast_CreateContext(ctx
);
749 _vbo_CreateContext(ctx
);
750 if (ctx
->swrast_context
) {
751 _tnl_CreateContext(ctx
);
752 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
753 _swsetup_CreateContext(ctx
);
755 /* Configure swrast to match hardware characteristics: */
756 _swrast_allow_pixel_fog(ctx
, false);
757 _swrast_allow_vertex_fog(ctx
, true);
760 _mesa_meta_init(ctx
);
762 brw_process_driconf_options(brw
);
763 brw_process_intel_debug_variable(brw
);
764 brw_initialize_context_constants(brw
);
766 ctx
->Const
.ResetStrategy
= notify_reset
767 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
769 /* Reinitialize the context point state. It depends on ctx->Const values. */
770 _mesa_init_point(ctx
);
774 intel_batchbuffer_init(brw
);
777 /* Create a new hardware context. Using a hardware context means that
778 * our GPU state will be saved/restored on context switch, allowing us
779 * to assume that the GPU is in the same state we left it in.
781 * This is required for transform feedback buffer offsets, query objects,
782 * and also allows us to reduce how much state we have to emit.
784 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
787 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
788 intelDestroyContext(driContextPriv
);
795 intelInitExtensions(ctx
);
797 brw_init_surface_formats(brw
);
799 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
800 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
801 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
802 brw
->urb
.size
= devinfo
->urb
.size
;
803 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
804 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
805 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
807 /* Estimate the size of the mappable aperture into the GTT. There's an
808 * ioctl to get the whole GTT size, but not one to get the mappable subset.
809 * It turns out it's basically always 256MB, though some ancient hardware
812 uint32_t gtt_size
= 256 * 1024 * 1024;
814 /* We don't want to map two objects such that a memcpy between them would
815 * just fault one mapping in and then the other over and over forever. So
816 * we would need to divide the GTT size by 2. Additionally, some GTT is
817 * taken up by things like the framebuffer and the ringbuffer and such, so
818 * be more conservative.
820 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
823 brw
->urb
.gs_present
= false;
825 brw
->prim_restart
.in_progress
= false;
826 brw
->prim_restart
.enable_cut_index
= false;
827 brw
->gs
.enabled
= false;
828 brw
->sf
.viewport_transform_enable
= true;
830 ctx
->VertexProgram
._MaintainTnlProgram
= true;
831 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
833 brw_draw_init( brw
);
835 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
836 /* Turn on some extra GL_ARB_debug_output generation. */
837 brw
->perf_debug
= true;
840 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
841 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
843 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
844 brw_init_shader_time(brw
);
846 _mesa_compute_version(ctx
);
848 _mesa_initialize_dispatch_tables(ctx
);
849 _mesa_initialize_vbo_vtxfmt(ctx
);
851 if (ctx
->Extensions
.AMD_performance_monitor
) {
852 brw_init_performance_monitors(brw
);
855 vbo_use_buffer_objects(ctx
);
856 vbo_always_unmap_buffers(ctx
);
862 intelDestroyContext(__DRIcontext
* driContextPriv
)
864 struct brw_context
*brw
=
865 (struct brw_context
*) driContextPriv
->driverPrivate
;
866 struct gl_context
*ctx
= &brw
->ctx
;
868 assert(brw
); /* should never be null */
872 /* Dump a final BMP in case the application doesn't call SwapBuffers */
873 if (INTEL_DEBUG
& DEBUG_AUB
) {
874 intel_batchbuffer_flush(brw
);
875 aub_dump_bmp(&brw
->ctx
);
878 _mesa_meta_free(&brw
->ctx
);
879 brw_meta_fast_clear_free(brw
);
881 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
882 /* Force a report. */
883 brw
->shader_time
.report_time
= 0;
885 brw_collect_and_report_shader_time(brw
);
886 brw_destroy_shader_time(brw
);
889 brw_destroy_state(brw
);
890 brw_draw_destroy(brw
);
892 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
894 drm_intel_gem_context_destroy(brw
->hw_ctx
);
896 if (ctx
->swrast_context
) {
897 _swsetup_DestroyContext(&brw
->ctx
);
898 _tnl_DestroyContext(&brw
->ctx
);
900 _vbo_DestroyContext(&brw
->ctx
);
902 if (ctx
->swrast_context
)
903 _swrast_DestroyContext(&brw
->ctx
);
905 intel_batchbuffer_free(brw
);
907 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
908 brw
->first_post_swapbuffers_batch
= NULL
;
910 driDestroyOptionCache(&brw
->optionCache
);
912 /* free the Mesa context */
913 _mesa_free_context_data(&brw
->ctx
);
916 driContextPriv
->driverPrivate
= NULL
;
920 intelUnbindContext(__DRIcontext
* driContextPriv
)
922 /* Unset current context and dispath table */
923 _mesa_make_current(NULL
, NULL
, NULL
);
929 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
930 * on window system framebuffers.
932 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
933 * your renderbuffer can do sRGB encode, and you can flip a switch that does
934 * sRGB encode if the renderbuffer can handle it. You can ask specifically
935 * for a visual where you're guaranteed to be capable, but it turns out that
936 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
937 * incapable ones, becuase there's no difference between the two in resources
938 * used. Applications thus get built that accidentally rely on the default
939 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
942 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
943 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
944 * So they removed the enable knob and made it "if the renderbuffer is sRGB
945 * capable, do sRGB encode". Then, for your window system renderbuffers, you
946 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
947 * and get no sRGB encode (assuming that both kinds of visual are available).
948 * Thus our choice to support sRGB by default on our visuals for desktop would
949 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
951 * Unfortunately, renderbuffer setup happens before a context is created. So
952 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
953 * context (without an sRGB visual, though we don't have sRGB visuals exposed
954 * yet), we go turn that back off before anyone finds out.
957 intel_gles3_srgb_workaround(struct brw_context
*brw
,
958 struct gl_framebuffer
*fb
)
960 struct gl_context
*ctx
= &brw
->ctx
;
962 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
965 /* Some day when we support the sRGB capable bit on visuals available for
966 * GLES, we'll need to respect that and not disable things here.
968 fb
->Visual
.sRGBCapable
= false;
969 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
970 if (fb
->Attachment
[i
].Renderbuffer
&&
971 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
972 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
978 intelMakeCurrent(__DRIcontext
* driContextPriv
,
979 __DRIdrawable
* driDrawPriv
,
980 __DRIdrawable
* driReadPriv
)
982 struct brw_context
*brw
;
983 GET_CURRENT_CONTEXT(curCtx
);
986 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
990 /* According to the glXMakeCurrent() man page: "Pending commands to
991 * the previous context, if any, are flushed before it is released."
992 * But only flush if we're actually changing contexts.
994 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
998 if (driContextPriv
) {
999 struct gl_context
*ctx
= &brw
->ctx
;
1000 struct gl_framebuffer
*fb
, *readFb
;
1002 if (driDrawPriv
== NULL
) {
1003 fb
= _mesa_get_incomplete_framebuffer();
1005 fb
= driDrawPriv
->driverPrivate
;
1006 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1009 if (driReadPriv
== NULL
) {
1010 readFb
= _mesa_get_incomplete_framebuffer();
1012 readFb
= driReadPriv
->driverPrivate
;
1013 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1016 /* The sRGB workaround changes the renderbuffer's format. We must change
1017 * the format before the renderbuffer's miptree get's allocated, otherwise
1018 * the formats of the renderbuffer and its miptree will differ.
1020 intel_gles3_srgb_workaround(brw
, fb
);
1021 intel_gles3_srgb_workaround(brw
, readFb
);
1023 /* If the context viewport hasn't been initialized, force a call out to
1024 * the loader to get buffers so we have a drawable size for the initial
1026 if (!brw
->ctx
.ViewportInitialized
)
1027 intel_prepare_render(brw
);
1029 _mesa_make_current(ctx
, fb
, readFb
);
1031 _mesa_make_current(NULL
, NULL
, NULL
);
1038 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1039 __DRIdrawable
*drawable
)
1042 /* MSAA and fast color clear are not supported, so don't waste time
1043 * checking whether a resolve is needed.
1048 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1049 struct intel_renderbuffer
*rb
;
1051 /* Usually, only the back buffer will need to be downsampled. However,
1052 * the front buffer will also need it if the user has rendered into it.
1054 static const gl_buffer_index buffers
[2] = {
1059 for (int i
= 0; i
< 2; ++i
) {
1060 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1061 if (rb
== NULL
|| rb
->mt
== NULL
)
1063 if (rb
->mt
->num_samples
<= 1)
1064 intel_miptree_resolve_color(brw
, rb
->mt
);
1066 intel_renderbuffer_downsample(brw
, rb
);
1071 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1073 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1077 intel_query_dri2_buffers(struct brw_context
*brw
,
1078 __DRIdrawable
*drawable
,
1079 __DRIbuffer
**buffers
,
1083 intel_process_dri2_buffer(struct brw_context
*brw
,
1084 __DRIdrawable
*drawable
,
1085 __DRIbuffer
*buffer
,
1086 struct intel_renderbuffer
*rb
,
1087 const char *buffer_name
);
1090 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1093 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1095 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1096 struct intel_renderbuffer
*rb
;
1097 __DRIbuffer
*buffers
= NULL
;
1099 const char *region_name
;
1101 /* Set this up front, so that in case our buffers get invalidated
1102 * while we're getting new buffers, we don't clobber the stamp and
1103 * thus ignore the invalidate. */
1104 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1106 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1107 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1109 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1111 if (buffers
== NULL
)
1114 for (i
= 0; i
< count
; i
++) {
1115 switch (buffers
[i
].attachment
) {
1116 case __DRI_BUFFER_FRONT_LEFT
:
1117 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1118 region_name
= "dri2 front buffer";
1121 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1122 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1123 region_name
= "dri2 fake front buffer";
1126 case __DRI_BUFFER_BACK_LEFT
:
1127 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1128 region_name
= "dri2 back buffer";
1131 case __DRI_BUFFER_DEPTH
:
1132 case __DRI_BUFFER_HIZ
:
1133 case __DRI_BUFFER_DEPTH_STENCIL
:
1134 case __DRI_BUFFER_STENCIL
:
1135 case __DRI_BUFFER_ACCUM
:
1138 "unhandled buffer attach event, attachment type %d\n",
1139 buffers
[i
].attachment
);
1143 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1149 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1151 struct brw_context
*brw
= context
->driverPrivate
;
1152 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1154 /* Set this up front, so that in case our buffers get invalidated
1155 * while we're getting new buffers, we don't clobber the stamp and
1156 * thus ignore the invalidate. */
1157 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1159 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1160 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1162 if (screen
->image
.loader
)
1163 intel_update_image_buffers(brw
, drawable
);
1165 intel_update_dri2_buffers(brw
, drawable
);
1167 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1171 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1172 * state is required.
1175 intel_prepare_render(struct brw_context
*brw
)
1177 struct gl_context
*ctx
= &brw
->ctx
;
1178 __DRIcontext
*driContext
= brw
->driContext
;
1179 __DRIdrawable
*drawable
;
1181 drawable
= driContext
->driDrawablePriv
;
1182 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1183 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1184 intel_update_renderbuffers(driContext
, drawable
);
1185 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1188 drawable
= driContext
->driReadablePriv
;
1189 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1190 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1191 intel_update_renderbuffers(driContext
, drawable
);
1192 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1195 /* If we're currently rendering to the front buffer, the rendering
1196 * that will happen next will probably dirty the front buffer. So
1197 * mark it as dirty here.
1199 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1200 brw
->front_buffer_dirty
= true;
1202 /* Wait for the swapbuffers before the one we just emitted, so we
1203 * don't get too many swaps outstanding for apps that are GPU-heavy
1204 * but not CPU-heavy.
1206 * We're using intelDRI2Flush (called from the loader before
1207 * swapbuffer) and glFlush (for front buffer rendering) as the
1208 * indicator that a frame is done and then throttle when we get
1209 * here as we prepare to render the next frame. At this point for
1210 * round trips for swap/copy and getting new buffers are done and
1211 * we'll spend less time waiting on the GPU.
1213 * Unfortunately, we don't have a handle to the batch containing
1214 * the swap, and getting our hands on that doesn't seem worth it,
1215 * so we just us the first batch we emitted after the last swap.
1217 if (brw
->need_throttle
&& brw
->first_post_swapbuffers_batch
) {
1218 if (!brw
->disable_throttling
)
1219 drm_intel_bo_wait_rendering(brw
->first_post_swapbuffers_batch
);
1220 drm_intel_bo_unreference(brw
->first_post_swapbuffers_batch
);
1221 brw
->first_post_swapbuffers_batch
= NULL
;
1222 brw
->need_throttle
= false;
1227 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1229 * To determine which DRI buffers to request, examine the renderbuffers
1230 * attached to the drawable's framebuffer. Then request the buffers with
1231 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1233 * This is called from intel_update_renderbuffers().
1235 * \param drawable Drawable whose buffers are queried.
1236 * \param buffers [out] List of buffers returned by DRI2 query.
1237 * \param buffer_count [out] Number of buffers returned.
1239 * \see intel_update_renderbuffers()
1240 * \see DRI2GetBuffers()
1241 * \see DRI2GetBuffersWithFormat()
1244 intel_query_dri2_buffers(struct brw_context
*brw
,
1245 __DRIdrawable
*drawable
,
1246 __DRIbuffer
**buffers
,
1249 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1250 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1252 unsigned attachments
[8];
1254 struct intel_renderbuffer
*front_rb
;
1255 struct intel_renderbuffer
*back_rb
;
1257 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1258 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1260 memset(attachments
, 0, sizeof(attachments
));
1261 if ((brw_is_front_buffer_drawing(fb
) ||
1262 brw_is_front_buffer_reading(fb
) ||
1263 !back_rb
) && front_rb
) {
1264 /* If a fake front buffer is in use, then querying for
1265 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1266 * the real front buffer to the fake front buffer. So before doing the
1267 * query, we need to make sure all the pending drawing has landed in the
1268 * real front buffer.
1270 intel_batchbuffer_flush(brw
);
1271 intel_flush_front(&brw
->ctx
);
1273 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1274 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1275 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1276 /* We have pending front buffer rendering, but we aren't querying for a
1277 * front buffer. If the front buffer we have is a fake front buffer,
1278 * the X server is going to throw it away when it processes the query.
1279 * So before doing the query, make sure all the pending drawing has
1280 * landed in the real front buffer.
1282 intel_batchbuffer_flush(brw
);
1283 intel_flush_front(&brw
->ctx
);
1287 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1288 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1291 assert(i
<= ARRAY_SIZE(attachments
));
1293 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1298 drawable
->loaderPrivate
);
1302 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1304 * This is called from intel_update_renderbuffers().
1307 * DRI buffers whose attachment point is DRI2BufferStencil or
1308 * DRI2BufferDepthStencil are handled as special cases.
1310 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1311 * that is passed to drm_intel_bo_gem_create_from_name().
1313 * \see intel_update_renderbuffers()
1316 intel_process_dri2_buffer(struct brw_context
*brw
,
1317 __DRIdrawable
*drawable
,
1318 __DRIbuffer
*buffer
,
1319 struct intel_renderbuffer
*rb
,
1320 const char *buffer_name
)
1322 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1328 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1330 /* We try to avoid closing and reopening the same BO name, because the first
1331 * use of a mapping of the buffer involves a bunch of page faulting which is
1332 * moderately expensive.
1334 struct intel_mipmap_tree
*last_mt
;
1335 if (num_samples
== 0)
1338 last_mt
= rb
->singlesample_mt
;
1340 uint32_t old_name
= 0;
1342 /* The bo already has a name because the miptree was created by a
1343 * previous call to intel_process_dri2_buffer(). If a bo already has a
1344 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1345 * create a new name.
1347 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1350 if (old_name
== buffer
->name
)
1353 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1355 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1356 buffer
->name
, buffer
->attachment
,
1357 buffer
->cpp
, buffer
->pitch
);
1360 intel_miptree_release(&rb
->mt
);
1361 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1365 "Failed to open BO for returned DRI2 buffer "
1366 "(%dx%d, %s, named %d).\n"
1367 "This is likely a bug in the X Server that will lead to a "
1369 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1373 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1374 drawable
->w
, drawable
->h
,
1377 if (brw_is_front_buffer_drawing(fb
) &&
1378 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1379 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1380 rb
->Base
.Base
.NumSamples
> 1) {
1381 intel_renderbuffer_upsample(brw
, rb
);
1386 drm_intel_bo_unreference(bo
);
1390 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1392 * To determine which DRI buffers to request, examine the renderbuffers
1393 * attached to the drawable's framebuffer. Then request the buffers from
1396 * This is called from intel_update_renderbuffers().
1398 * \param drawable Drawable whose buffers are queried.
1399 * \param buffers [out] List of buffers returned by DRI2 query.
1400 * \param buffer_count [out] Number of buffers returned.
1402 * \see intel_update_renderbuffers()
1406 intel_update_image_buffer(struct brw_context
*intel
,
1407 __DRIdrawable
*drawable
,
1408 struct intel_renderbuffer
*rb
,
1410 enum __DRIimageBufferMask buffer_type
)
1412 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1414 if (!rb
|| !buffer
->bo
)
1417 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1419 /* Check and see if we're already bound to the right
1422 struct intel_mipmap_tree
*last_mt
;
1423 if (num_samples
== 0)
1426 last_mt
= rb
->singlesample_mt
;
1428 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1431 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1432 buffer
->width
, buffer
->height
,
1435 if (brw_is_front_buffer_drawing(fb
) &&
1436 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1437 rb
->Base
.Base
.NumSamples
> 1) {
1438 intel_renderbuffer_upsample(intel
, rb
);
1443 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1445 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1446 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1447 struct intel_renderbuffer
*front_rb
;
1448 struct intel_renderbuffer
*back_rb
;
1449 struct __DRIimageList images
;
1450 unsigned int format
;
1451 uint32_t buffer_mask
= 0;
1453 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1454 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1457 format
= intel_rb_format(back_rb
);
1459 format
= intel_rb_format(front_rb
);
1463 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1464 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1465 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1469 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1471 (*screen
->image
.loader
->getBuffers
) (drawable
,
1472 driGLFormatToImageFormat(format
),
1473 &drawable
->dri2
.stamp
,
1474 drawable
->loaderPrivate
,
1478 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1479 drawable
->w
= images
.front
->width
;
1480 drawable
->h
= images
.front
->height
;
1481 intel_update_image_buffer(brw
,
1485 __DRI_IMAGE_BUFFER_FRONT
);
1487 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1488 drawable
->w
= images
.back
->width
;
1489 drawable
->h
= images
.back
->height
;
1490 intel_update_image_buffer(brw
,
1494 __DRI_IMAGE_BUFFER_BACK
);