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 "compiler/nir/nir.h"
35 #include "main/api_exec.h"
36 #include "main/context.h"
37 #include "main/fbobject.h"
38 #include "main/extensions.h"
39 #include "main/imports.h"
40 #include "main/macros.h"
41 #include "main/points.h"
42 #include "main/version.h"
43 #include "main/vtxfmt.h"
44 #include "main/texobj.h"
45 #include "main/framebuffer.h"
46 #include "main/stencil.h"
47 #include "main/state.h"
51 #include "drivers/common/driverfuncs.h"
52 #include "drivers/common/meta.h"
55 #include "brw_context.h"
56 #include "brw_defines.h"
57 #include "brw_blorp.h"
59 #include "brw_state.h"
61 #include "intel_batchbuffer.h"
62 #include "intel_buffer_objects.h"
63 #include "intel_buffers.h"
64 #include "intel_fbo.h"
65 #include "intel_mipmap_tree.h"
66 #include "intel_pixel.h"
67 #include "intel_image.h"
68 #include "intel_tex.h"
69 #include "intel_tex_obj.h"
71 #include "swrast_setup/swrast_setup.h"
73 #include "tnl/t_pipeline.h"
74 #include "util/ralloc.h"
75 #include "util/debug.h"
76 #include "util/disk_cache.h"
79 #include "common/gen_defines.h"
81 #include "compiler/spirv/nir_spirv.h"
82 /***************************************
83 * Mesa's Driver Functions
84 ***************************************/
86 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
89 get_bsw_model(const struct intel_screen
*screen
)
91 switch (screen
->eu_total
) {
102 brw_get_renderer_string(const struct intel_screen
*screen
)
105 static char buffer
[128];
108 switch (screen
->deviceID
) {
110 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
111 #include "pci_ids/i965_pci_ids.h"
113 chipset
= "Unknown Intel Chipset";
117 /* Braswell branding is funny, so we have to fix it up here */
118 if (screen
->deviceID
== 0x22B1) {
119 bsw
= strdup(chipset
);
120 char *needle
= strstr(bsw
, "XXX");
122 memcpy(needle
, get_bsw_model(screen
), 3);
127 (void) driGetRendererString(buffer
, chipset
, 0);
132 static const GLubyte
*
133 intel_get_string(struct gl_context
* ctx
, GLenum name
)
135 const struct brw_context
*const brw
= brw_context(ctx
);
139 return (GLubyte
*) brw_vendor_string
;
143 (GLubyte
*) brw_get_renderer_string(brw
->screen
);
151 intel_viewport(struct gl_context
*ctx
)
153 struct brw_context
*brw
= brw_context(ctx
);
154 __DRIcontext
*driContext
= brw
->driContext
;
156 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
157 if (driContext
->driDrawablePriv
)
158 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
159 if (driContext
->driReadablePriv
)
160 dri2InvalidateDrawable(driContext
->driReadablePriv
);
165 intel_update_framebuffer(struct gl_context
*ctx
,
166 struct gl_framebuffer
*fb
)
168 struct brw_context
*brw
= brw_context(ctx
);
170 /* Quantize the derived default number of samples
172 fb
->DefaultGeometry
._NumSamples
=
173 intel_quantize_num_samples(brw
->screen
,
174 fb
->DefaultGeometry
.NumSamples
);
178 intel_update_state(struct gl_context
* ctx
)
180 GLuint new_state
= ctx
->NewState
;
181 struct brw_context
*brw
= brw_context(ctx
);
183 if (ctx
->swrast_context
)
184 _swrast_InvalidateState(ctx
, new_state
);
186 brw
->NewGLState
|= new_state
;
188 if (new_state
& (_NEW_SCISSOR
| _NEW_BUFFERS
| _NEW_VIEWPORT
))
189 _mesa_update_draw_buffer_bounds(ctx
, ctx
->DrawBuffer
);
191 if (new_state
& (_NEW_STENCIL
| _NEW_BUFFERS
)) {
192 brw
->stencil_enabled
= _mesa_stencil_is_enabled(ctx
);
193 brw
->stencil_two_sided
= _mesa_stencil_is_two_sided(ctx
);
194 brw
->stencil_write_enabled
=
195 _mesa_stencil_is_write_enabled(ctx
, brw
->stencil_two_sided
);
198 if (new_state
& _NEW_POLYGON
)
199 brw
->polygon_front_bit
= _mesa_polygon_get_front_bit(ctx
);
201 if (new_state
& _NEW_BUFFERS
) {
202 intel_update_framebuffer(ctx
, ctx
->DrawBuffer
);
203 if (ctx
->DrawBuffer
!= ctx
->ReadBuffer
)
204 intel_update_framebuffer(ctx
, ctx
->ReadBuffer
);
208 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
211 intel_flush_front(struct gl_context
*ctx
)
213 struct brw_context
*brw
= brw_context(ctx
);
214 __DRIcontext
*driContext
= brw
->driContext
;
215 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
216 __DRIscreen
*const dri_screen
= brw
->screen
->driScrnPriv
;
218 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
219 if (flushFront(dri_screen
) && driDrawable
&&
220 driDrawable
->loaderPrivate
) {
222 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
224 * This potentially resolves both front and back buffer. It
225 * is unnecessary to resolve the back, but harms nothing except
226 * performance. And no one cares about front-buffer render
229 intel_resolve_for_dri2_flush(brw
, driDrawable
);
230 intel_batchbuffer_flush(brw
);
232 flushFront(dri_screen
)(driDrawable
, driDrawable
->loaderPrivate
);
234 /* We set the dirty bit in intel_prepare_render() if we're
235 * front buffer rendering once we get there.
237 brw
->front_buffer_dirty
= false;
243 intel_glFlush(struct gl_context
*ctx
)
245 struct brw_context
*brw
= brw_context(ctx
);
247 intel_batchbuffer_flush(brw
);
248 intel_flush_front(ctx
);
250 brw
->need_flush_throttle
= true;
254 intel_finish(struct gl_context
* ctx
)
256 struct brw_context
*brw
= brw_context(ctx
);
260 if (brw
->batch
.last_bo
)
261 brw_bo_wait_rendering(brw
->batch
.last_bo
);
265 brw_init_driver_functions(struct brw_context
*brw
,
266 struct dd_function_table
*functions
)
268 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
270 _mesa_init_driver_functions(functions
);
272 /* GLX uses DRI2 invalidate events to handle window resizing.
273 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
274 * which doesn't provide a mechanism for snooping the event queues.
276 * So EGL still relies on viewport hacks to handle window resizing.
277 * This should go away with DRI3000.
279 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
280 functions
->Viewport
= intel_viewport
;
282 functions
->Flush
= intel_glFlush
;
283 functions
->Finish
= intel_finish
;
284 functions
->GetString
= intel_get_string
;
285 functions
->UpdateState
= intel_update_state
;
287 brw_init_draw_functions(functions
);
288 intelInitTextureFuncs(functions
);
289 intelInitTextureImageFuncs(functions
);
290 intelInitTextureCopyImageFuncs(functions
);
291 intelInitCopyImageFuncs(functions
);
292 intelInitClearFuncs(functions
);
293 intelInitBufferFuncs(functions
);
294 intelInitPixelFuncs(functions
);
295 intelInitBufferObjectFuncs(functions
);
296 brw_init_syncobj_functions(functions
);
297 brw_init_object_purgeable_functions(functions
);
299 brwInitFragProgFuncs( functions
);
300 brw_init_common_queryobj_functions(functions
);
301 if (devinfo
->gen
>= 8 || devinfo
->is_haswell
)
302 hsw_init_queryobj_functions(functions
);
303 else if (devinfo
->gen
>= 6)
304 gen6_init_queryobj_functions(functions
);
306 gen4_init_queryobj_functions(functions
);
307 brw_init_compute_functions(functions
);
308 brw_init_conditional_render_functions(functions
);
310 functions
->GenerateMipmap
= brw_generate_mipmap
;
312 functions
->QueryInternalFormat
= brw_query_internal_format
;
314 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
315 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
316 if (can_do_mi_math_and_lrr(brw
->screen
)) {
317 functions
->BeginTransformFeedback
= hsw_begin_transform_feedback
;
318 functions
->EndTransformFeedback
= hsw_end_transform_feedback
;
319 functions
->PauseTransformFeedback
= hsw_pause_transform_feedback
;
320 functions
->ResumeTransformFeedback
= hsw_resume_transform_feedback
;
321 } else if (devinfo
->gen
>= 7) {
322 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
323 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
324 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
325 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
326 functions
->GetTransformFeedbackVertexCount
=
327 brw_get_transform_feedback_vertex_count
;
329 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
330 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
331 functions
->PauseTransformFeedback
= brw_pause_transform_feedback
;
332 functions
->ResumeTransformFeedback
= brw_resume_transform_feedback
;
333 functions
->GetTransformFeedbackVertexCount
=
334 brw_get_transform_feedback_vertex_count
;
337 if (devinfo
->gen
>= 6)
338 functions
->GetSamplePosition
= gen6_get_sample_position
;
340 /* GL_ARB_get_program_binary */
341 brw_program_binary_init(brw
->screen
->deviceID
);
342 functions
->GetProgramBinaryDriverSHA1
= brw_get_program_binary_driver_sha1
;
343 functions
->ProgramBinarySerializeDriverBlob
= brw_serialize_program_binary
;
344 functions
->ProgramBinaryDeserializeDriverBlob
=
345 brw_deserialize_program_binary
;
347 if (brw
->screen
->disk_cache
) {
348 functions
->ShaderCacheSerializeDriverBlob
= brw_program_serialize_nir
;
353 brw_initialize_spirv_supported_capabilities(struct brw_context
*brw
)
355 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
356 struct gl_context
*ctx
= &brw
->ctx
;
358 /* The following SPIR-V capabilities are only supported on gen7+. In theory
359 * you should enable the extension only on gen7+, but just in case let's
362 assert(devinfo
->gen
>= 7);
364 ctx
->Const
.SpirVCapabilities
.float64
= devinfo
->gen
>= 8;
365 ctx
->Const
.SpirVCapabilities
.int64
= devinfo
->gen
>= 8;
366 ctx
->Const
.SpirVCapabilities
.tessellation
= true;
367 ctx
->Const
.SpirVCapabilities
.draw_parameters
= true;
368 ctx
->Const
.SpirVCapabilities
.image_write_without_format
= true;
369 ctx
->Const
.SpirVCapabilities
.variable_pointers
= true;
370 ctx
->Const
.SpirVCapabilities
.atomic_storage
= devinfo
->gen
>= 7;
371 ctx
->Const
.SpirVCapabilities
.transform_feedback
= devinfo
->gen
>= 7;
372 ctx
->Const
.SpirVCapabilities
.geometry_streams
= devinfo
->gen
>= 7;
376 brw_initialize_context_constants(struct brw_context
*brw
)
378 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
379 struct gl_context
*ctx
= &brw
->ctx
;
380 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
382 const bool stage_exists
[MESA_SHADER_STAGES
] = {
383 [MESA_SHADER_VERTEX
] = true,
384 [MESA_SHADER_TESS_CTRL
] = devinfo
->gen
>= 7,
385 [MESA_SHADER_TESS_EVAL
] = devinfo
->gen
>= 7,
386 [MESA_SHADER_GEOMETRY
] = devinfo
->gen
>= 6,
387 [MESA_SHADER_FRAGMENT
] = true,
388 [MESA_SHADER_COMPUTE
] =
389 (_mesa_is_desktop_gl(ctx
) &&
390 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 1024) ||
391 (ctx
->API
== API_OPENGLES2
&&
392 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 128),
395 unsigned num_stages
= 0;
396 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
401 unsigned max_samplers
=
402 devinfo
->gen
>= 8 || devinfo
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
404 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
405 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
406 ctx
->Const
.MaxCombinedShaderOutputResources
=
407 MAX_IMAGE_UNITS
+ BRW_MAX_DRAW_BUFFERS
;
409 /* The timestamp register we can read for glGetTimestamp() is
410 * sometimes only 32 bits, before scaling to nanoseconds (depending
413 * Once scaled to nanoseconds the timestamp would roll over at a
414 * non-power-of-two, so an application couldn't use
415 * GL_QUERY_COUNTER_BITS to handle rollover correctly. Instead, we
416 * report 36 bits and truncate at that (rolling over 5 times as
417 * often as the HW counter), and when the 32-bit counter rolls
418 * over, it happens to also be at a rollover in the reported value
419 * from near (1<<36) to 0.
421 * The low 32 bits rolls over in ~343 seconds. Our 36-bit result
422 * rolls over every ~69 seconds.
424 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
426 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
427 ctx
->Const
.MaxImageUnits
= MAX_IMAGE_UNITS
;
428 if (devinfo
->gen
>= 7) {
429 ctx
->Const
.MaxRenderbufferSize
= 16384;
430 ctx
->Const
.MaxTextureLevels
= MIN2(15 /* 16384 */, MAX_TEXTURE_LEVELS
);
431 ctx
->Const
.MaxCubeTextureLevels
= 15; /* 16384 */
433 ctx
->Const
.MaxRenderbufferSize
= 8192;
434 ctx
->Const
.MaxTextureLevels
= MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS
);
435 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
437 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
438 ctx
->Const
.MaxArrayTextureLayers
= devinfo
->gen
>= 7 ? 2048 : 512;
439 ctx
->Const
.MaxTextureMbytes
= 1536;
440 ctx
->Const
.MaxTextureRectSize
= devinfo
->gen
>= 7 ? 16384 : 8192;
441 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
442 ctx
->Const
.MaxTextureLodBias
= 15.0;
443 ctx
->Const
.StripTextureBorder
= true;
444 if (devinfo
->gen
>= 7) {
445 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
446 ctx
->Const
.MinProgramTextureGatherOffset
= -32;
447 ctx
->Const
.MaxProgramTextureGatherOffset
= 31;
448 } else if (devinfo
->gen
== 6) {
449 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
450 ctx
->Const
.MinProgramTextureGatherOffset
= -8;
451 ctx
->Const
.MaxProgramTextureGatherOffset
= 7;
454 ctx
->Const
.MaxUniformBlockSize
= 65536;
456 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
457 struct gl_program_constants
*prog
= &ctx
->Const
.Program
[i
];
459 if (!stage_exists
[i
])
462 prog
->MaxTextureImageUnits
= max_samplers
;
464 prog
->MaxUniformBlocks
= BRW_MAX_UBO
;
465 prog
->MaxCombinedUniformComponents
=
466 prog
->MaxUniformComponents
+
467 ctx
->Const
.MaxUniformBlockSize
/ 4 * prog
->MaxUniformBlocks
;
469 prog
->MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
470 prog
->MaxAtomicBuffers
= BRW_MAX_ABO
;
471 prog
->MaxImageUniforms
= compiler
->scalar_stage
[i
] ? BRW_MAX_IMAGES
: 0;
472 prog
->MaxShaderStorageBlocks
= BRW_MAX_SSBO
;
475 ctx
->Const
.MaxTextureUnits
=
476 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
477 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
479 ctx
->Const
.MaxUniformBufferBindings
= num_stages
* BRW_MAX_UBO
;
480 ctx
->Const
.MaxCombinedUniformBlocks
= num_stages
* BRW_MAX_UBO
;
481 ctx
->Const
.MaxCombinedAtomicBuffers
= num_stages
* BRW_MAX_ABO
;
482 ctx
->Const
.MaxCombinedShaderStorageBlocks
= num_stages
* BRW_MAX_SSBO
;
483 ctx
->Const
.MaxShaderStorageBufferBindings
= num_stages
* BRW_MAX_SSBO
;
484 ctx
->Const
.MaxCombinedTextureImageUnits
= num_stages
* max_samplers
;
485 ctx
->Const
.MaxCombinedImageUniforms
= num_stages
* BRW_MAX_IMAGES
;
488 /* Hardware only supports a limited number of transform feedback buffers.
489 * So we need to override the Mesa default (which is based only on software
492 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
494 /* On Gen6, in the worst case, we use up one binding table entry per
495 * transform feedback component (see comments above the definition of
496 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
497 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
498 * BRW_MAX_SOL_BINDINGS.
500 * In "separate components" mode, we need to divide this value by
501 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
502 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
504 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
505 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
506 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
508 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
=
509 !can_do_mi_math_and_lrr(brw
->screen
);
512 const int *msaa_modes
= intel_supported_msaa_modes(brw
->screen
);
513 const int clamp_max_samples
=
514 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
516 if (clamp_max_samples
< 0) {
517 max_samples
= msaa_modes
[0];
519 /* Select the largest supported MSAA mode that does not exceed
523 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
524 if (msaa_modes
[i
] <= clamp_max_samples
) {
525 max_samples
= msaa_modes
[i
];
531 ctx
->Const
.MaxSamples
= max_samples
;
532 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
533 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
534 ctx
->Const
.MaxIntegerSamples
= max_samples
;
535 ctx
->Const
.MaxImageSamples
= 0;
537 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
538 * to map indices of rectangular grid to sample numbers within a pixel.
539 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
540 * extension implementation. For more details see the comment above
541 * gen6_set_sample_maps() definition.
543 gen6_set_sample_maps(ctx
);
545 ctx
->Const
.MinLineWidth
= 1.0;
546 ctx
->Const
.MinLineWidthAA
= 1.0;
547 if (devinfo
->gen
>= 6) {
548 ctx
->Const
.MaxLineWidth
= 7.375;
549 ctx
->Const
.MaxLineWidthAA
= 7.375;
550 ctx
->Const
.LineWidthGranularity
= 0.125;
552 ctx
->Const
.MaxLineWidth
= 7.0;
553 ctx
->Const
.MaxLineWidthAA
= 7.0;
554 ctx
->Const
.LineWidthGranularity
= 0.5;
557 /* For non-antialiased lines, we have to round the line width to the
558 * nearest whole number. Make sure that we don't advertise a line
559 * width that, when rounded, will be beyond the actual hardware
562 assert(roundf(ctx
->Const
.MaxLineWidth
) <= ctx
->Const
.MaxLineWidth
);
564 ctx
->Const
.MinPointSize
= 1.0;
565 ctx
->Const
.MinPointSizeAA
= 1.0;
566 ctx
->Const
.MaxPointSize
= 255.0;
567 ctx
->Const
.MaxPointSizeAA
= 255.0;
568 ctx
->Const
.PointSizeGranularity
= 1.0;
570 if (devinfo
->gen
>= 5 || devinfo
->is_g4x
)
571 ctx
->Const
.MaxClipPlanes
= 8;
573 ctx
->Const
.GLSLTessLevelsAsInputs
= true;
574 ctx
->Const
.PrimitiveRestartForPatches
= true;
576 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
577 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
578 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
579 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
580 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
581 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
582 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
583 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
584 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
585 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
586 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
587 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
588 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
589 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
591 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
592 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
593 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
594 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
595 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
596 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
597 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
598 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
599 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
600 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
601 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
603 /* Fragment shaders use real, 32-bit twos-complement integers for all
606 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
607 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
608 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
609 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
610 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
612 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
613 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
614 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
615 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
616 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
618 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
619 * but we're not sure how it's actually done for vertex order,
620 * that affect provoking vertex decision. Always use last vertex
621 * convention for quad primitive which works as expected for now.
623 if (devinfo
->gen
>= 6)
624 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
626 ctx
->Const
.NativeIntegers
= true;
628 /* Regarding the CMP instruction, the Ivybridge PRM says:
630 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
631 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
632 * 0xFFFFFFFF) is assigned to dst."
634 * but PRMs for earlier generations say
636 * "In dword format, one GRF may store up to 8 results. When the register
637 * is used later as a vector of Booleans, as only LSB at each channel
638 * contains meaning [sic] data, software should make sure all higher bits
639 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
641 * We select the representation of a true boolean uniform to be ~0, and fix
642 * the results of Gen <= 5 CMP instruction's with -(result & 1).
644 ctx
->Const
.UniformBooleanTrue
= ~0;
646 /* From the gen4 PRM, volume 4 page 127:
648 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
649 * the base address of the first element of the surface, computed in
650 * software by adding the surface base address to the byte offset of
651 * the element in the buffer."
653 * However, unaligned accesses are slower, so enforce buffer alignment.
655 * In order to push UBO data, 3DSTATE_CONSTANT_XS imposes an additional
656 * restriction: the start of the buffer needs to be 32B aligned.
658 ctx
->Const
.UniformBufferOffsetAlignment
= 32;
660 /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so
661 * that we can safely have the CPU and GPU writing the same SSBO on
662 * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never
663 * writes, so there's no problem. For an SSBO, the GPU and the CPU can
664 * be updating disjoint regions of the buffer simultaneously and that will
665 * break if the regions overlap the same cacheline.
667 ctx
->Const
.ShaderStorageBufferOffsetAlignment
= 64;
668 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
669 ctx
->Const
.MaxTextureBufferSize
= 128 * 1024 * 1024;
671 if (devinfo
->gen
>= 6) {
672 ctx
->Const
.MaxVarying
= 32;
673 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
674 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
=
675 compiler
->scalar_stage
[MESA_SHADER_GEOMETRY
] ? 128 : 64;
676 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
677 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
678 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxInputComponents
= 128;
679 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxOutputComponents
= 128;
680 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxInputComponents
= 128;
681 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxOutputComponents
= 128;
684 /* We want the GLSL compiler to emit code that uses condition codes */
685 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
686 ctx
->Const
.ShaderCompilerOptions
[i
] =
687 brw
->screen
->compiler
->glsl_compiler_options
[i
];
690 if (devinfo
->gen
>= 7) {
691 ctx
->Const
.MaxViewportWidth
= 32768;
692 ctx
->Const
.MaxViewportHeight
= 32768;
695 /* ARB_viewport_array, OES_viewport_array */
696 if (devinfo
->gen
>= 6) {
697 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
698 ctx
->Const
.ViewportSubpixelBits
= 0;
700 /* Cast to float before negating because MaxViewportWidth is unsigned.
702 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
703 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
706 /* ARB_gpu_shader5 */
707 if (devinfo
->gen
>= 7)
708 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
710 /* ARB_framebuffer_no_attachments */
711 ctx
->Const
.MaxFramebufferWidth
= 16384;
712 ctx
->Const
.MaxFramebufferHeight
= 16384;
713 ctx
->Const
.MaxFramebufferLayers
= ctx
->Const
.MaxArrayTextureLayers
;
714 ctx
->Const
.MaxFramebufferSamples
= max_samples
;
716 /* OES_primitive_bounding_box */
717 ctx
->Const
.NoPrimitiveBoundingBoxOutput
= true;
719 /* TODO: We should be able to use STD430 packing by default on all hardware
720 * but some piglit tests [1] currently fail on SNB when this is enabled.
721 * The problem is the messages we're using for doing uniform pulls
722 * in the vec4 back-end on SNB is the OWORD block load instruction, which
723 * takes its offset in units of OWORDS (16 bytes). On IVB+, we use the
724 * sampler which doesn't have these restrictions.
726 * In the scalar back-end, we use the sampler for dynamic uniform loads and
727 * pull an entire cache line at a time for constant offset loads both of
728 * which support almost any alignment.
730 * [1] glsl-1.40/uniform_buffer/vs-float-array-variable-index.shader_test
732 if (devinfo
->gen
>= 7)
733 ctx
->Const
.UseSTD430AsDefaultPacking
= true;
735 if (!(ctx
->Const
.ContextFlags
& GL_CONTEXT_FLAG_DEBUG_BIT
))
736 ctx
->Const
.AllowMappedBuffersDuringExecution
= true;
738 /* GL_ARB_get_program_binary */
739 ctx
->Const
.NumProgramBinaryFormats
= 1;
743 brw_initialize_cs_context_constants(struct brw_context
*brw
)
745 struct gl_context
*ctx
= &brw
->ctx
;
746 const struct intel_screen
*screen
= brw
->screen
;
747 struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
749 /* FINISHME: Do this for all platforms that the kernel supports */
750 if (devinfo
->is_cherryview
&&
751 screen
->subslice_total
> 0 && screen
->eu_total
> 0) {
752 /* Logical CS threads = EUs per subslice * 7 threads per EU */
753 uint32_t max_cs_threads
= screen
->eu_total
/ screen
->subslice_total
* 7;
755 /* Fuse configurations may give more threads than expected, never less. */
756 if (max_cs_threads
> devinfo
->max_cs_threads
)
757 devinfo
->max_cs_threads
= max_cs_threads
;
760 /* Maximum number of scalar compute shader invocations that can be run in
761 * parallel in the same subslice assuming SIMD32 dispatch.
763 * We don't advertise more than 64 threads, because we are limited to 64 by
764 * our usage of thread_width_max in the gpgpu walker command. This only
765 * currently impacts Haswell, which otherwise might be able to advertise 70
766 * threads. With SIMD32 and 64 threads, Haswell still provides twice the
767 * required the number of invocation needed for ARB_compute_shader.
769 const unsigned max_threads
= MIN2(64, devinfo
->max_cs_threads
);
770 const uint32_t max_invocations
= 32 * max_threads
;
771 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
772 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
773 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
774 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
775 ctx
->Const
.MaxComputeSharedMemorySize
= 64 * 1024;
779 * Process driconf (drirc) options, setting appropriate context flags.
781 * intelInitExtensions still pokes at optionCache directly, in order to
782 * avoid advertising various extensions. No flags are set, so it makes
783 * sense to continue doing that there.
786 brw_process_driconf_options(struct brw_context
*brw
)
788 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
789 struct gl_context
*ctx
= &brw
->ctx
;
791 driOptionCache
*options
= &brw
->optionCache
;
792 driParseConfigFiles(options
, &brw
->screen
->optionCache
,
793 brw
->driContext
->driScreenPriv
->myNum
,
796 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
797 switch (bo_reuse_mode
) {
798 case DRI_CONF_BO_REUSE_DISABLED
:
800 case DRI_CONF_BO_REUSE_ALL
:
801 brw_bufmgr_enable_reuse(brw
->bufmgr
);
805 if (INTEL_DEBUG
& DEBUG_NO_HIZ
) {
806 brw
->has_hiz
= false;
807 /* On gen6, you can only do separate stencil with HIZ. */
808 if (devinfo
->gen
== 6)
809 brw
->has_separate_stencil
= false;
812 if (driQueryOptionb(options
, "mesa_no_error"))
813 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_NO_ERROR_BIT_KHR
;
815 if (driQueryOptionb(options
, "always_flush_batch")) {
816 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
817 brw
->always_flush_batch
= true;
820 if (driQueryOptionb(options
, "always_flush_cache")) {
821 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
822 brw
->always_flush_cache
= true;
825 if (driQueryOptionb(options
, "disable_throttling")) {
826 fprintf(stderr
, "disabling flush throttling\n");
827 brw
->disable_throttling
= true;
830 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
832 if (driQueryOptionb(&brw
->optionCache
, "precise_trig"))
833 brw
->screen
->compiler
->precise_trig
= true;
835 ctx
->Const
.ForceGLSLExtensionsWarn
=
836 driQueryOptionb(options
, "force_glsl_extensions_warn");
838 ctx
->Const
.ForceGLSLVersion
=
839 driQueryOptioni(options
, "force_glsl_version");
841 ctx
->Const
.DisableGLSLLineContinuations
=
842 driQueryOptionb(options
, "disable_glsl_line_continuations");
844 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
845 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
847 ctx
->Const
.AllowGLSLBuiltinVariableRedeclaration
=
848 driQueryOptionb(options
, "allow_glsl_builtin_variable_redeclaration");
850 ctx
->Const
.AllowHigherCompatVersion
=
851 driQueryOptionb(options
, "allow_higher_compat_version");
853 ctx
->Const
.ForceGLSLAbsSqrt
=
854 driQueryOptionb(options
, "force_glsl_abs_sqrt");
856 ctx
->Const
.GLSLZeroInit
= driQueryOptionb(options
, "glsl_zero_init");
858 brw
->dual_color_blend_by_location
=
859 driQueryOptionb(options
, "dual_color_blend_by_location");
861 ctx
->Const
.AllowGLSLCrossStageInterpolationMismatch
=
862 driQueryOptionb(options
, "allow_glsl_cross_stage_interpolation_mismatch");
864 ctx
->Const
.dri_config_options_sha1
= ralloc_array(brw
, unsigned char, 20);
865 driComputeOptionsSha1(&brw
->screen
->optionCache
,
866 ctx
->Const
.dri_config_options_sha1
);
870 brwCreateContext(gl_api api
,
871 const struct gl_config
*mesaVis
,
872 __DRIcontext
*driContextPriv
,
873 const struct __DriverContextConfig
*ctx_config
,
874 unsigned *dri_ctx_error
,
875 void *sharedContextPrivate
)
877 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
878 struct intel_screen
*screen
= driContextPriv
->driScreenPriv
->driverPrivate
;
879 const struct gen_device_info
*devinfo
= &screen
->devinfo
;
880 struct dd_function_table functions
;
882 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
883 * provides us with context reset notifications.
885 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
|
886 __DRI_CTX_FLAG_FORWARD_COMPATIBLE
|
887 __DRI_CTX_FLAG_NO_ERROR
;
889 if (screen
->has_context_reset_notification
)
890 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
892 if (ctx_config
->flags
& ~allowed_flags
) {
893 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
897 if (ctx_config
->attribute_mask
&
898 ~(__DRIVER_CONTEXT_ATTRIB_RESET_STRATEGY
|
899 __DRIVER_CONTEXT_ATTRIB_PRIORITY
)) {
900 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_ATTRIBUTE
;
905 ((ctx_config
->attribute_mask
& __DRIVER_CONTEXT_ATTRIB_RESET_STRATEGY
) &&
906 ctx_config
->reset_strategy
!= __DRI_CTX_RESET_NO_NOTIFICATION
);
908 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
910 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
911 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
915 driContextPriv
->driverPrivate
= brw
;
916 brw
->driContext
= driContextPriv
;
917 brw
->screen
= screen
;
918 brw
->bufmgr
= screen
->bufmgr
;
920 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
921 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
923 brw
->has_swizzling
= screen
->hw_has_swizzling
;
925 brw
->isl_dev
= screen
->isl_dev
;
927 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
928 brw
->tcs
.base
.stage
= MESA_SHADER_TESS_CTRL
;
929 brw
->tes
.base
.stage
= MESA_SHADER_TESS_EVAL
;
930 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
931 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
932 brw
->cs
.base
.stage
= MESA_SHADER_COMPUTE
;
934 brw_init_driver_functions(brw
, &functions
);
937 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
939 struct gl_context
*ctx
= &brw
->ctx
;
941 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
942 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
943 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
944 intelDestroyContext(driContextPriv
);
948 driContextSetFlags(ctx
, ctx_config
->flags
);
950 /* Initialize the software rasterizer and helper modules.
952 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
953 * software fallbacks (which we have to support on legacy GL to do weird
954 * glDrawPixels(), glBitmap(), and other functions).
956 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
957 _swrast_CreateContext(ctx
);
960 _vbo_CreateContext(ctx
);
961 if (ctx
->swrast_context
) {
962 _tnl_CreateContext(ctx
);
963 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
964 _swsetup_CreateContext(ctx
);
966 /* Configure swrast to match hardware characteristics: */
967 _swrast_allow_pixel_fog(ctx
, false);
968 _swrast_allow_vertex_fog(ctx
, true);
971 _mesa_meta_init(ctx
);
973 brw_process_driconf_options(brw
);
975 if (INTEL_DEBUG
& DEBUG_PERF
)
976 brw
->perf_debug
= true;
978 brw_initialize_cs_context_constants(brw
);
979 brw_initialize_context_constants(brw
);
981 ctx
->Const
.ResetStrategy
= notify_reset
982 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
984 /* Reinitialize the context point state. It depends on ctx->Const values. */
985 _mesa_init_point(ctx
);
989 intel_batchbuffer_init(brw
);
991 /* Create a new hardware context. Using a hardware context means that
992 * our GPU state will be saved/restored on context switch, allowing us
993 * to assume that the GPU is in the same state we left it in.
995 * This is required for transform feedback buffer offsets, query objects,
996 * and also allows us to reduce how much state we have to emit.
998 brw
->hw_ctx
= brw_create_hw_context(brw
->bufmgr
);
999 if (!brw
->hw_ctx
&& devinfo
->gen
>= 6) {
1000 fprintf(stderr
, "Failed to create hardware context.\n");
1001 intelDestroyContext(driContextPriv
);
1006 int hw_priority
= GEN_CONTEXT_MEDIUM_PRIORITY
;
1007 if (ctx_config
->attribute_mask
& __DRIVER_CONTEXT_ATTRIB_PRIORITY
) {
1008 switch (ctx_config
->priority
) {
1009 case __DRI_CTX_PRIORITY_LOW
:
1010 hw_priority
= GEN_CONTEXT_LOW_PRIORITY
;
1012 case __DRI_CTX_PRIORITY_HIGH
:
1013 hw_priority
= GEN_CONTEXT_HIGH_PRIORITY
;
1017 if (hw_priority
!= I915_CONTEXT_DEFAULT_PRIORITY
&&
1018 brw_hw_context_set_priority(brw
->bufmgr
, brw
->hw_ctx
, hw_priority
)) {
1020 "Failed to set priority [%d:%d] for hardware context.\n",
1021 ctx_config
->priority
, hw_priority
);
1022 intelDestroyContext(driContextPriv
);
1027 if (brw_init_pipe_control(brw
, devinfo
)) {
1028 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
1029 intelDestroyContext(driContextPriv
);
1033 if (devinfo
->gen
== 11) {
1035 "WARNING: i965 does not fully support Gen11 yet.\n"
1036 "Instability or lower performance might occur.\n");
1040 brw_upload_init(&brw
->upload
, brw
->bufmgr
, 65536);
1042 brw_init_state(brw
);
1044 intelInitExtensions(ctx
);
1046 brw_init_surface_formats(brw
);
1048 brw_blorp_init(brw
);
1050 brw
->urb
.size
= devinfo
->urb
.size
;
1052 if (devinfo
->gen
== 6)
1053 brw
->urb
.gs_present
= false;
1055 brw
->prim_restart
.in_progress
= false;
1056 brw
->prim_restart
.enable_cut_index
= false;
1057 brw
->gs
.enabled
= false;
1058 brw
->clip
.viewport_count
= 1;
1060 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
1062 brw
->max_gtt_map_object_size
= screen
->max_gtt_map_object_size
;
1064 ctx
->VertexProgram
._MaintainTnlProgram
= true;
1065 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
1067 brw_draw_init( brw
);
1069 if ((ctx_config
->flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
1070 /* Turn on some extra GL_ARB_debug_output generation. */
1071 brw
->perf_debug
= true;
1074 if ((ctx_config
->flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0) {
1075 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
1076 ctx
->Const
.RobustAccess
= GL_TRUE
;
1079 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
1080 brw_init_shader_time(brw
);
1082 _mesa_override_extensions(ctx
);
1083 _mesa_compute_version(ctx
);
1085 /* GL_ARB_gl_spirv */
1086 if (ctx
->Extensions
.ARB_gl_spirv
)
1087 brw_initialize_spirv_supported_capabilities(brw
);
1089 _mesa_initialize_dispatch_tables(ctx
);
1090 _mesa_initialize_vbo_vtxfmt(ctx
);
1092 if (ctx
->Extensions
.INTEL_performance_query
)
1093 brw_init_performance_queries(brw
);
1095 vbo_use_buffer_objects(ctx
);
1096 vbo_always_unmap_buffers(ctx
);
1098 brw
->ctx
.Cache
= brw
->screen
->disk_cache
;
1104 intelDestroyContext(__DRIcontext
* driContextPriv
)
1106 struct brw_context
*brw
=
1107 (struct brw_context
*) driContextPriv
->driverPrivate
;
1108 struct gl_context
*ctx
= &brw
->ctx
;
1110 _mesa_meta_free(&brw
->ctx
);
1112 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
1113 /* Force a report. */
1114 brw
->shader_time
.report_time
= 0;
1116 brw_collect_and_report_shader_time(brw
);
1117 brw_destroy_shader_time(brw
);
1120 blorp_finish(&brw
->blorp
);
1122 brw_destroy_state(brw
);
1123 brw_draw_destroy(brw
);
1125 brw_bo_unreference(brw
->curbe
.curbe_bo
);
1127 brw_bo_unreference(brw
->vs
.base
.scratch_bo
);
1128 brw_bo_unreference(brw
->tcs
.base
.scratch_bo
);
1129 brw_bo_unreference(brw
->tes
.base
.scratch_bo
);
1130 brw_bo_unreference(brw
->gs
.base
.scratch_bo
);
1131 brw_bo_unreference(brw
->wm
.base
.scratch_bo
);
1133 brw_bo_unreference(brw
->vs
.base
.push_const_bo
);
1134 brw_bo_unreference(brw
->tcs
.base
.push_const_bo
);
1135 brw_bo_unreference(brw
->tes
.base
.push_const_bo
);
1136 brw_bo_unreference(brw
->gs
.base
.push_const_bo
);
1137 brw_bo_unreference(brw
->wm
.base
.push_const_bo
);
1139 brw_destroy_hw_context(brw
->bufmgr
, brw
->hw_ctx
);
1141 if (ctx
->swrast_context
) {
1142 _swsetup_DestroyContext(&brw
->ctx
);
1143 _tnl_DestroyContext(&brw
->ctx
);
1145 _vbo_DestroyContext(&brw
->ctx
);
1147 if (ctx
->swrast_context
)
1148 _swrast_DestroyContext(&brw
->ctx
);
1150 brw_fini_pipe_control(brw
);
1151 intel_batchbuffer_free(&brw
->batch
);
1153 brw_bo_unreference(brw
->throttle_batch
[1]);
1154 brw_bo_unreference(brw
->throttle_batch
[0]);
1155 brw
->throttle_batch
[1] = NULL
;
1156 brw
->throttle_batch
[0] = NULL
;
1158 driDestroyOptionCache(&brw
->optionCache
);
1160 /* free the Mesa context */
1161 _mesa_free_context_data(&brw
->ctx
);
1164 driContextPriv
->driverPrivate
= NULL
;
1168 intelUnbindContext(__DRIcontext
* driContextPriv
)
1170 /* Unset current context and dispath table */
1171 _mesa_make_current(NULL
, NULL
, NULL
);
1177 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1178 * on window system framebuffers.
1180 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1181 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1182 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1183 * for a visual where you're guaranteed to be capable, but it turns out that
1184 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1185 * incapable ones, because there's no difference between the two in resources
1186 * used. Applications thus get built that accidentally rely on the default
1187 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1190 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1191 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1192 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1193 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1194 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1195 * and get no sRGB encode (assuming that both kinds of visual are available).
1196 * Thus our choice to support sRGB by default on our visuals for desktop would
1197 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1199 * Unfortunately, renderbuffer setup happens before a context is created. So
1200 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1201 * context (without an sRGB visual), we go turn that back off before anyone
1205 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1206 struct gl_framebuffer
*fb
)
1208 struct gl_context
*ctx
= &brw
->ctx
;
1210 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1213 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1214 struct gl_renderbuffer
*rb
= fb
->Attachment
[i
].Renderbuffer
;
1216 /* Check if sRGB was specifically asked for. */
1217 struct intel_renderbuffer
*irb
= intel_get_renderbuffer(fb
, i
);
1218 if (irb
&& irb
->need_srgb
)
1222 rb
->Format
= _mesa_get_srgb_format_linear(rb
->Format
);
1224 /* Disable sRGB from framebuffers that are not compatible. */
1225 fb
->Visual
.sRGBCapable
= false;
1229 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1230 __DRIdrawable
* driDrawPriv
,
1231 __DRIdrawable
* driReadPriv
)
1233 struct brw_context
*brw
;
1236 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1240 if (driContextPriv
) {
1241 struct gl_context
*ctx
= &brw
->ctx
;
1242 struct gl_framebuffer
*fb
, *readFb
;
1244 if (driDrawPriv
== NULL
) {
1245 fb
= _mesa_get_incomplete_framebuffer();
1247 fb
= driDrawPriv
->driverPrivate
;
1248 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1251 if (driReadPriv
== NULL
) {
1252 readFb
= _mesa_get_incomplete_framebuffer();
1254 readFb
= driReadPriv
->driverPrivate
;
1255 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1258 /* The sRGB workaround changes the renderbuffer's format. We must change
1259 * the format before the renderbuffer's miptree get's allocated, otherwise
1260 * the formats of the renderbuffer and its miptree will differ.
1262 intel_gles3_srgb_workaround(brw
, fb
);
1263 intel_gles3_srgb_workaround(brw
, readFb
);
1265 /* If the context viewport hasn't been initialized, force a call out to
1266 * the loader to get buffers so we have a drawable size for the initial
1268 if (!brw
->ctx
.ViewportInitialized
)
1269 intel_prepare_render(brw
);
1271 _mesa_make_current(ctx
, fb
, readFb
);
1273 _mesa_make_current(NULL
, NULL
, NULL
);
1280 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1281 __DRIdrawable
*drawable
)
1283 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
1285 if (devinfo
->gen
< 6) {
1286 /* MSAA and fast color clear are not supported, so don't waste time
1287 * checking whether a resolve is needed.
1292 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1293 struct intel_renderbuffer
*rb
;
1295 /* Usually, only the back buffer will need to be downsampled. However,
1296 * the front buffer will also need it if the user has rendered into it.
1298 static const gl_buffer_index buffers
[2] = {
1303 for (int i
= 0; i
< 2; ++i
) {
1304 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1305 if (rb
== NULL
|| rb
->mt
== NULL
)
1307 if (rb
->mt
->surf
.samples
== 1) {
1308 assert(rb
->mt_layer
== 0 && rb
->mt_level
== 0 &&
1309 rb
->layer_count
== 1);
1310 intel_miptree_prepare_external(brw
, rb
->mt
);
1312 intel_renderbuffer_downsample(brw
, rb
);
1314 /* Call prepare_external on the single-sample miptree to do any
1315 * needed resolves prior to handing it off to the window system.
1316 * This is needed in the case that rb->singlesample_mt is Y-tiled
1317 * with CCS_E enabled but without I915_FORMAT_MOD_Y_TILED_CCS_E. In
1318 * this case, the MSAA resolve above will write compressed data into
1319 * rb->singlesample_mt.
1321 * TODO: Some day, if we decide to care about the tiny performance
1322 * hit we're taking by doing the MSAA resolve and then a CCS resolve,
1323 * we could detect this case and just allocate the single-sampled
1324 * miptree without aux. However, that would be a lot of plumbing and
1325 * this is a rather exotic case so it's not really worth it.
1327 intel_miptree_prepare_external(brw
, rb
->singlesample_mt
);
1333 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1335 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1339 intel_query_dri2_buffers(struct brw_context
*brw
,
1340 __DRIdrawable
*drawable
,
1341 __DRIbuffer
**buffers
,
1345 intel_process_dri2_buffer(struct brw_context
*brw
,
1346 __DRIdrawable
*drawable
,
1347 __DRIbuffer
*buffer
,
1348 struct intel_renderbuffer
*rb
,
1349 const char *buffer_name
);
1352 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1355 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1357 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1358 struct intel_renderbuffer
*rb
;
1359 __DRIbuffer
*buffers
= NULL
;
1361 const char *region_name
;
1363 /* Set this up front, so that in case our buffers get invalidated
1364 * while we're getting new buffers, we don't clobber the stamp and
1365 * thus ignore the invalidate. */
1366 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1368 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1369 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1371 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1373 if (buffers
== NULL
)
1376 for (int i
= 0; i
< count
; i
++) {
1377 switch (buffers
[i
].attachment
) {
1378 case __DRI_BUFFER_FRONT_LEFT
:
1379 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1380 region_name
= "dri2 front buffer";
1383 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1384 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1385 region_name
= "dri2 fake front buffer";
1388 case __DRI_BUFFER_BACK_LEFT
:
1389 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1390 region_name
= "dri2 back buffer";
1393 case __DRI_BUFFER_DEPTH
:
1394 case __DRI_BUFFER_HIZ
:
1395 case __DRI_BUFFER_DEPTH_STENCIL
:
1396 case __DRI_BUFFER_STENCIL
:
1397 case __DRI_BUFFER_ACCUM
:
1400 "unhandled buffer attach event, attachment type %d\n",
1401 buffers
[i
].attachment
);
1405 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1411 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1413 struct brw_context
*brw
= context
->driverPrivate
;
1414 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
1416 /* Set this up front, so that in case our buffers get invalidated
1417 * while we're getting new buffers, we don't clobber the stamp and
1418 * thus ignore the invalidate. */
1419 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1421 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1422 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1424 if (dri_screen
->image
.loader
)
1425 intel_update_image_buffers(brw
, drawable
);
1427 intel_update_dri2_buffers(brw
, drawable
);
1429 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1433 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1434 * state is required.
1437 intel_prepare_render(struct brw_context
*brw
)
1439 struct gl_context
*ctx
= &brw
->ctx
;
1440 __DRIcontext
*driContext
= brw
->driContext
;
1441 __DRIdrawable
*drawable
;
1443 drawable
= driContext
->driDrawablePriv
;
1444 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1445 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1446 intel_update_renderbuffers(driContext
, drawable
);
1447 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1450 drawable
= driContext
->driReadablePriv
;
1451 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1452 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1453 intel_update_renderbuffers(driContext
, drawable
);
1454 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1457 /* If we're currently rendering to the front buffer, the rendering
1458 * that will happen next will probably dirty the front buffer. So
1459 * mark it as dirty here.
1461 if (_mesa_is_front_buffer_drawing(ctx
->DrawBuffer
))
1462 brw
->front_buffer_dirty
= true;
1466 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1468 * To determine which DRI buffers to request, examine the renderbuffers
1469 * attached to the drawable's framebuffer. Then request the buffers with
1470 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1472 * This is called from intel_update_renderbuffers().
1474 * \param drawable Drawable whose buffers are queried.
1475 * \param buffers [out] List of buffers returned by DRI2 query.
1476 * \param buffer_count [out] Number of buffers returned.
1478 * \see intel_update_renderbuffers()
1479 * \see DRI2GetBuffers()
1480 * \see DRI2GetBuffersWithFormat()
1483 intel_query_dri2_buffers(struct brw_context
*brw
,
1484 __DRIdrawable
*drawable
,
1485 __DRIbuffer
**buffers
,
1488 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
1489 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1491 unsigned attachments
[8];
1493 struct intel_renderbuffer
*front_rb
;
1494 struct intel_renderbuffer
*back_rb
;
1496 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1497 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1499 memset(attachments
, 0, sizeof(attachments
));
1500 if ((_mesa_is_front_buffer_drawing(fb
) ||
1501 _mesa_is_front_buffer_reading(fb
) ||
1502 !back_rb
) && front_rb
) {
1503 /* If a fake front buffer is in use, then querying for
1504 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1505 * the real front buffer to the fake front buffer. So before doing the
1506 * query, we need to make sure all the pending drawing has landed in the
1507 * real front buffer.
1509 intel_batchbuffer_flush(brw
);
1510 intel_flush_front(&brw
->ctx
);
1512 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1513 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1514 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1515 /* We have pending front buffer rendering, but we aren't querying for a
1516 * front buffer. If the front buffer we have is a fake front buffer,
1517 * the X server is going to throw it away when it processes the query.
1518 * So before doing the query, make sure all the pending drawing has
1519 * landed in the real front buffer.
1521 intel_batchbuffer_flush(brw
);
1522 intel_flush_front(&brw
->ctx
);
1526 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1527 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1530 assert(i
<= ARRAY_SIZE(attachments
));
1533 dri_screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1538 drawable
->loaderPrivate
);
1542 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1544 * This is called from intel_update_renderbuffers().
1547 * DRI buffers whose attachment point is DRI2BufferStencil or
1548 * DRI2BufferDepthStencil are handled as special cases.
1550 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1551 * that is passed to brw_bo_gem_create_from_name().
1553 * \see intel_update_renderbuffers()
1556 intel_process_dri2_buffer(struct brw_context
*brw
,
1557 __DRIdrawable
*drawable
,
1558 __DRIbuffer
*buffer
,
1559 struct intel_renderbuffer
*rb
,
1560 const char *buffer_name
)
1562 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1568 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1570 /* We try to avoid closing and reopening the same BO name, because the first
1571 * use of a mapping of the buffer involves a bunch of page faulting which is
1572 * moderately expensive.
1574 struct intel_mipmap_tree
*last_mt
;
1575 if (num_samples
== 0)
1578 last_mt
= rb
->singlesample_mt
;
1580 uint32_t old_name
= 0;
1582 /* The bo already has a name because the miptree was created by a
1583 * previous call to intel_process_dri2_buffer(). If a bo already has a
1584 * name, then brw_bo_flink() is a low-cost getter. It does not
1585 * create a new name.
1587 brw_bo_flink(last_mt
->bo
, &old_name
);
1590 if (old_name
== buffer
->name
)
1593 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1595 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1596 buffer
->name
, buffer
->attachment
,
1597 buffer
->cpp
, buffer
->pitch
);
1600 bo
= brw_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1604 "Failed to open BO for returned DRI2 buffer "
1605 "(%dx%d, %s, named %d).\n"
1606 "This is likely a bug in the X Server that will lead to a "
1608 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1612 uint32_t tiling
, swizzle
;
1613 brw_bo_get_tiling(bo
, &tiling
, &swizzle
);
1615 struct intel_mipmap_tree
*mt
=
1616 intel_miptree_create_for_bo(brw
,
1618 intel_rb_format(rb
),
1624 isl_tiling_from_i915_tiling(tiling
),
1625 MIPTREE_CREATE_DEFAULT
);
1627 brw_bo_unreference(bo
);
1631 /* We got this BO from X11. We cana't assume that we have coherent texture
1632 * access because X may suddenly decide to use it for scan-out which would
1633 * destroy coherency.
1635 bo
->cache_coherent
= false;
1637 if (!intel_update_winsys_renderbuffer_miptree(brw
, rb
, mt
,
1638 drawable
->w
, drawable
->h
,
1640 brw_bo_unreference(bo
);
1641 intel_miptree_release(&mt
);
1645 if (_mesa_is_front_buffer_drawing(fb
) &&
1646 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1647 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1648 rb
->Base
.Base
.NumSamples
> 1) {
1649 intel_renderbuffer_upsample(brw
, rb
);
1654 brw_bo_unreference(bo
);
1658 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1660 * To determine which DRI buffers to request, examine the renderbuffers
1661 * attached to the drawable's framebuffer. Then request the buffers from
1664 * This is called from intel_update_renderbuffers().
1666 * \param drawable Drawable whose buffers are queried.
1667 * \param buffers [out] List of buffers returned by DRI2 query.
1668 * \param buffer_count [out] Number of buffers returned.
1670 * \see intel_update_renderbuffers()
1674 intel_update_image_buffer(struct brw_context
*intel
,
1675 __DRIdrawable
*drawable
,
1676 struct intel_renderbuffer
*rb
,
1678 enum __DRIimageBufferMask buffer_type
)
1680 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1682 if (!rb
|| !buffer
->bo
)
1685 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1687 /* Check and see if we're already bound to the right
1690 struct intel_mipmap_tree
*last_mt
;
1691 if (num_samples
== 0)
1694 last_mt
= rb
->singlesample_mt
;
1696 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1699 /* Only allow internal compression if samples == 0. For multisampled
1700 * window system buffers, the only thing the single-sampled buffer is used
1701 * for is as a resolve target. If we do any compression beyond what is
1702 * supported by the window system, we will just have to resolve so it's
1703 * probably better to just not bother.
1705 const bool allow_internal_aux
= (num_samples
== 0);
1707 struct intel_mipmap_tree
*mt
=
1708 intel_miptree_create_for_dri_image(intel
, buffer
, GL_TEXTURE_2D
,
1709 intel_rb_format(rb
),
1710 allow_internal_aux
);
1714 if (!intel_update_winsys_renderbuffer_miptree(intel
, rb
, mt
,
1715 buffer
->width
, buffer
->height
,
1717 intel_miptree_release(&mt
);
1721 if (_mesa_is_front_buffer_drawing(fb
) &&
1722 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1723 rb
->Base
.Base
.NumSamples
> 1) {
1724 intel_renderbuffer_upsample(intel
, rb
);
1729 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1731 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1732 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
1733 struct intel_renderbuffer
*front_rb
;
1734 struct intel_renderbuffer
*back_rb
;
1735 struct __DRIimageList images
;
1737 uint32_t buffer_mask
= 0;
1740 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1741 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1744 format
= intel_rb_format(back_rb
);
1746 format
= intel_rb_format(front_rb
);
1750 if (front_rb
&& (_mesa_is_front_buffer_drawing(fb
) ||
1751 _mesa_is_front_buffer_reading(fb
) || !back_rb
)) {
1752 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1756 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1758 ret
= dri_screen
->image
.loader
->getBuffers(drawable
,
1759 driGLFormatToImageFormat(format
),
1760 &drawable
->dri2
.stamp
,
1761 drawable
->loaderPrivate
,
1767 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1768 drawable
->w
= images
.front
->width
;
1769 drawable
->h
= images
.front
->height
;
1770 intel_update_image_buffer(brw
,
1774 __DRI_IMAGE_BUFFER_FRONT
);
1777 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1778 drawable
->w
= images
.back
->width
;
1779 drawable
->h
= images
.back
->height
;
1780 intel_update_image_buffer(brw
,
1784 __DRI_IMAGE_BUFFER_BACK
);