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"
49 #include "vbo/vbo_context.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"
78 /***************************************
79 * Mesa's Driver Functions
80 ***************************************/
82 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
85 get_bsw_model(const struct intel_screen
*screen
)
87 switch (screen
->eu_total
) {
98 brw_get_renderer_string(const struct intel_screen
*screen
)
101 static char buffer
[128];
104 switch (screen
->deviceID
) {
106 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
107 #include "pci_ids/i965_pci_ids.h"
109 chipset
= "Unknown Intel Chipset";
113 /* Braswell branding is funny, so we have to fix it up here */
114 if (screen
->deviceID
== 0x22B1) {
115 bsw
= strdup(chipset
);
116 char *needle
= strstr(bsw
, "XXX");
118 memcpy(needle
, get_bsw_model(screen
), 3);
123 (void) driGetRendererString(buffer
, chipset
, 0);
128 static const GLubyte
*
129 intel_get_string(struct gl_context
* ctx
, GLenum name
)
131 const struct brw_context
*const brw
= brw_context(ctx
);
135 return (GLubyte
*) brw_vendor_string
;
139 (GLubyte
*) brw_get_renderer_string(brw
->screen
);
147 intel_viewport(struct gl_context
*ctx
)
149 struct brw_context
*brw
= brw_context(ctx
);
150 __DRIcontext
*driContext
= brw
->driContext
;
152 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
153 if (driContext
->driDrawablePriv
)
154 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
155 if (driContext
->driReadablePriv
)
156 dri2InvalidateDrawable(driContext
->driReadablePriv
);
161 intel_update_framebuffer(struct gl_context
*ctx
,
162 struct gl_framebuffer
*fb
)
164 struct brw_context
*brw
= brw_context(ctx
);
166 /* Quantize the derived default number of samples
168 fb
->DefaultGeometry
._NumSamples
=
169 intel_quantize_num_samples(brw
->screen
,
170 fb
->DefaultGeometry
.NumSamples
);
174 intel_update_state(struct gl_context
* ctx
)
176 GLuint new_state
= ctx
->NewState
;
177 struct brw_context
*brw
= brw_context(ctx
);
179 if (ctx
->swrast_context
)
180 _swrast_InvalidateState(ctx
, new_state
);
182 brw
->NewGLState
|= new_state
;
184 if (new_state
& (_NEW_SCISSOR
| _NEW_BUFFERS
| _NEW_VIEWPORT
))
185 _mesa_update_draw_buffer_bounds(ctx
, ctx
->DrawBuffer
);
187 if (new_state
& (_NEW_STENCIL
| _NEW_BUFFERS
)) {
188 brw
->stencil_enabled
= _mesa_stencil_is_enabled(ctx
);
189 brw
->stencil_two_sided
= _mesa_stencil_is_two_sided(ctx
);
190 brw
->stencil_write_enabled
=
191 _mesa_stencil_is_write_enabled(ctx
, brw
->stencil_two_sided
);
194 if (new_state
& _NEW_POLYGON
)
195 brw
->polygon_front_bit
= _mesa_polygon_get_front_bit(ctx
);
197 if (new_state
& _NEW_BUFFERS
) {
198 intel_update_framebuffer(ctx
, ctx
->DrawBuffer
);
199 if (ctx
->DrawBuffer
!= ctx
->ReadBuffer
)
200 intel_update_framebuffer(ctx
, ctx
->ReadBuffer
);
204 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
207 intel_flush_front(struct gl_context
*ctx
)
209 struct brw_context
*brw
= brw_context(ctx
);
210 __DRIcontext
*driContext
= brw
->driContext
;
211 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
212 __DRIscreen
*const dri_screen
= brw
->screen
->driScrnPriv
;
214 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
215 if (flushFront(dri_screen
) && driDrawable
&&
216 driDrawable
->loaderPrivate
) {
218 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
220 * This potentially resolves both front and back buffer. It
221 * is unnecessary to resolve the back, but harms nothing except
222 * performance. And no one cares about front-buffer render
225 intel_resolve_for_dri2_flush(brw
, driDrawable
);
226 intel_batchbuffer_flush(brw
);
228 flushFront(dri_screen
)(driDrawable
, driDrawable
->loaderPrivate
);
230 /* We set the dirty bit in intel_prepare_render() if we're
231 * front buffer rendering once we get there.
233 brw
->front_buffer_dirty
= false;
239 intel_glFlush(struct gl_context
*ctx
)
241 struct brw_context
*brw
= brw_context(ctx
);
243 intel_batchbuffer_flush(brw
);
244 intel_flush_front(ctx
);
246 brw
->need_flush_throttle
= true;
250 intel_finish(struct gl_context
* ctx
)
252 struct brw_context
*brw
= brw_context(ctx
);
256 if (brw
->batch
.last_bo
)
257 brw_bo_wait_rendering(brw
->batch
.last_bo
);
261 brw_init_driver_functions(struct brw_context
*brw
,
262 struct dd_function_table
*functions
)
264 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
266 _mesa_init_driver_functions(functions
);
268 /* GLX uses DRI2 invalidate events to handle window resizing.
269 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
270 * which doesn't provide a mechanism for snooping the event queues.
272 * So EGL still relies on viewport hacks to handle window resizing.
273 * This should go away with DRI3000.
275 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
276 functions
->Viewport
= intel_viewport
;
278 functions
->Flush
= intel_glFlush
;
279 functions
->Finish
= intel_finish
;
280 functions
->GetString
= intel_get_string
;
281 functions
->UpdateState
= intel_update_state
;
283 intelInitTextureFuncs(functions
);
284 intelInitTextureImageFuncs(functions
);
285 intelInitTextureSubImageFuncs(functions
);
286 intelInitTextureCopyImageFuncs(functions
);
287 intelInitCopyImageFuncs(functions
);
288 intelInitClearFuncs(functions
);
289 intelInitBufferFuncs(functions
);
290 intelInitPixelFuncs(functions
);
291 intelInitBufferObjectFuncs(functions
);
292 brw_init_syncobj_functions(functions
);
293 brw_init_object_purgeable_functions(functions
);
295 brwInitFragProgFuncs( functions
);
296 brw_init_common_queryobj_functions(functions
);
297 if (devinfo
->gen
>= 8 || devinfo
->is_haswell
)
298 hsw_init_queryobj_functions(functions
);
299 else if (devinfo
->gen
>= 6)
300 gen6_init_queryobj_functions(functions
);
302 gen4_init_queryobj_functions(functions
);
303 brw_init_compute_functions(functions
);
304 brw_init_conditional_render_functions(functions
);
306 functions
->QueryInternalFormat
= brw_query_internal_format
;
308 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
309 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
310 if (can_do_mi_math_and_lrr(brw
->screen
)) {
311 functions
->BeginTransformFeedback
= hsw_begin_transform_feedback
;
312 functions
->EndTransformFeedback
= hsw_end_transform_feedback
;
313 functions
->PauseTransformFeedback
= hsw_pause_transform_feedback
;
314 functions
->ResumeTransformFeedback
= hsw_resume_transform_feedback
;
315 } else if (devinfo
->gen
>= 7) {
316 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
317 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
318 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
319 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
320 functions
->GetTransformFeedbackVertexCount
=
321 brw_get_transform_feedback_vertex_count
;
323 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
324 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
325 functions
->PauseTransformFeedback
= brw_pause_transform_feedback
;
326 functions
->ResumeTransformFeedback
= brw_resume_transform_feedback
;
327 functions
->GetTransformFeedbackVertexCount
=
328 brw_get_transform_feedback_vertex_count
;
331 if (devinfo
->gen
>= 6)
332 functions
->GetSamplePosition
= gen6_get_sample_position
;
336 brw_initialize_context_constants(struct brw_context
*brw
)
338 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
339 struct gl_context
*ctx
= &brw
->ctx
;
340 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
342 const bool stage_exists
[MESA_SHADER_STAGES
] = {
343 [MESA_SHADER_VERTEX
] = true,
344 [MESA_SHADER_TESS_CTRL
] = devinfo
->gen
>= 7,
345 [MESA_SHADER_TESS_EVAL
] = devinfo
->gen
>= 7,
346 [MESA_SHADER_GEOMETRY
] = devinfo
->gen
>= 6,
347 [MESA_SHADER_FRAGMENT
] = true,
348 [MESA_SHADER_COMPUTE
] =
349 ((ctx
->API
== API_OPENGL_COMPAT
|| ctx
->API
== API_OPENGL_CORE
) &&
350 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 1024) ||
351 (ctx
->API
== API_OPENGLES2
&&
352 ctx
->Const
.MaxComputeWorkGroupSize
[0] >= 128) ||
353 _mesa_extension_override_enables
.ARB_compute_shader
,
356 unsigned num_stages
= 0;
357 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
362 unsigned max_samplers
=
363 devinfo
->gen
>= 8 || devinfo
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
365 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
366 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
367 ctx
->Const
.MaxCombinedShaderOutputResources
=
368 MAX_IMAGE_UNITS
+ BRW_MAX_DRAW_BUFFERS
;
370 /* The timestamp register we can read for glGetTimestamp() is
371 * sometimes only 32 bits, before scaling to nanoseconds (depending
374 * Once scaled to nanoseconds the timestamp would roll over at a
375 * non-power-of-two, so an application couldn't use
376 * GL_QUERY_COUNTER_BITS to handle rollover correctly. Instead, we
377 * report 36 bits and truncate at that (rolling over 5 times as
378 * often as the HW counter), and when the 32-bit counter rolls
379 * over, it happens to also be at a rollover in the reported value
380 * from near (1<<36) to 0.
382 * The low 32 bits rolls over in ~343 seconds. Our 36-bit result
383 * rolls over every ~69 seconds.
385 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
387 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
388 ctx
->Const
.MaxImageUnits
= MAX_IMAGE_UNITS
;
389 if (devinfo
->gen
>= 7) {
390 ctx
->Const
.MaxRenderbufferSize
= 16384;
391 ctx
->Const
.MaxTextureLevels
= MIN2(15 /* 16384 */, MAX_TEXTURE_LEVELS
);
392 ctx
->Const
.MaxCubeTextureLevels
= 15; /* 16384 */
394 ctx
->Const
.MaxRenderbufferSize
= 8192;
395 ctx
->Const
.MaxTextureLevels
= MIN2(14 /* 8192 */, MAX_TEXTURE_LEVELS
);
396 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
398 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
399 ctx
->Const
.MaxArrayTextureLayers
= devinfo
->gen
>= 7 ? 2048 : 512;
400 ctx
->Const
.MaxTextureMbytes
= 1536;
401 ctx
->Const
.MaxTextureRectSize
= devinfo
->gen
>= 7 ? 16384 : 8192;
402 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
403 ctx
->Const
.MaxTextureLodBias
= 15.0;
404 ctx
->Const
.StripTextureBorder
= true;
405 if (devinfo
->gen
>= 7) {
406 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
407 ctx
->Const
.MinProgramTextureGatherOffset
= -32;
408 ctx
->Const
.MaxProgramTextureGatherOffset
= 31;
409 } else if (devinfo
->gen
== 6) {
410 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
411 ctx
->Const
.MinProgramTextureGatherOffset
= -8;
412 ctx
->Const
.MaxProgramTextureGatherOffset
= 7;
415 ctx
->Const
.MaxUniformBlockSize
= 65536;
417 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
418 struct gl_program_constants
*prog
= &ctx
->Const
.Program
[i
];
420 if (!stage_exists
[i
])
423 prog
->MaxTextureImageUnits
= max_samplers
;
425 prog
->MaxUniformBlocks
= BRW_MAX_UBO
;
426 prog
->MaxCombinedUniformComponents
=
427 prog
->MaxUniformComponents
+
428 ctx
->Const
.MaxUniformBlockSize
/ 4 * prog
->MaxUniformBlocks
;
430 prog
->MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
431 prog
->MaxAtomicBuffers
= BRW_MAX_ABO
;
432 prog
->MaxImageUniforms
= compiler
->scalar_stage
[i
] ? BRW_MAX_IMAGES
: 0;
433 prog
->MaxShaderStorageBlocks
= BRW_MAX_SSBO
;
436 ctx
->Const
.MaxTextureUnits
=
437 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
438 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
440 ctx
->Const
.MaxUniformBufferBindings
= num_stages
* BRW_MAX_UBO
;
441 ctx
->Const
.MaxCombinedUniformBlocks
= num_stages
* BRW_MAX_UBO
;
442 ctx
->Const
.MaxCombinedAtomicBuffers
= num_stages
* BRW_MAX_ABO
;
443 ctx
->Const
.MaxCombinedShaderStorageBlocks
= num_stages
* BRW_MAX_SSBO
;
444 ctx
->Const
.MaxShaderStorageBufferBindings
= num_stages
* BRW_MAX_SSBO
;
445 ctx
->Const
.MaxCombinedTextureImageUnits
= num_stages
* max_samplers
;
446 ctx
->Const
.MaxCombinedImageUniforms
= num_stages
* BRW_MAX_IMAGES
;
449 /* Hardware only supports a limited number of transform feedback buffers.
450 * So we need to override the Mesa default (which is based only on software
453 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
455 /* On Gen6, in the worst case, we use up one binding table entry per
456 * transform feedback component (see comments above the definition of
457 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
458 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
459 * BRW_MAX_SOL_BINDINGS.
461 * In "separate components" mode, we need to divide this value by
462 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
463 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
465 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
466 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
467 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
469 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
=
470 !can_do_mi_math_and_lrr(brw
->screen
);
473 const int *msaa_modes
= intel_supported_msaa_modes(brw
->screen
);
474 const int clamp_max_samples
=
475 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
477 if (clamp_max_samples
< 0) {
478 max_samples
= msaa_modes
[0];
480 /* Select the largest supported MSAA mode that does not exceed
484 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
485 if (msaa_modes
[i
] <= clamp_max_samples
) {
486 max_samples
= msaa_modes
[i
];
492 ctx
->Const
.MaxSamples
= max_samples
;
493 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
494 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
495 ctx
->Const
.MaxIntegerSamples
= max_samples
;
496 ctx
->Const
.MaxImageSamples
= 0;
498 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
499 * to map indices of rectangular grid to sample numbers within a pixel.
500 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
501 * extension implementation. For more details see the comment above
502 * gen6_set_sample_maps() definition.
504 gen6_set_sample_maps(ctx
);
506 ctx
->Const
.MinLineWidth
= 1.0;
507 ctx
->Const
.MinLineWidthAA
= 1.0;
508 if (devinfo
->gen
>= 6) {
509 ctx
->Const
.MaxLineWidth
= 7.375;
510 ctx
->Const
.MaxLineWidthAA
= 7.375;
511 ctx
->Const
.LineWidthGranularity
= 0.125;
513 ctx
->Const
.MaxLineWidth
= 7.0;
514 ctx
->Const
.MaxLineWidthAA
= 7.0;
515 ctx
->Const
.LineWidthGranularity
= 0.5;
518 /* For non-antialiased lines, we have to round the line width to the
519 * nearest whole number. Make sure that we don't advertise a line
520 * width that, when rounded, will be beyond the actual hardware
523 assert(roundf(ctx
->Const
.MaxLineWidth
) <= ctx
->Const
.MaxLineWidth
);
525 ctx
->Const
.MinPointSize
= 1.0;
526 ctx
->Const
.MinPointSizeAA
= 1.0;
527 ctx
->Const
.MaxPointSize
= 255.0;
528 ctx
->Const
.MaxPointSizeAA
= 255.0;
529 ctx
->Const
.PointSizeGranularity
= 1.0;
531 if (devinfo
->gen
>= 5 || devinfo
->is_g4x
)
532 ctx
->Const
.MaxClipPlanes
= 8;
534 ctx
->Const
.GLSLTessLevelsAsInputs
= true;
535 ctx
->Const
.LowerTCSPatchVerticesIn
= devinfo
->gen
>= 8;
536 ctx
->Const
.LowerTESPatchVerticesIn
= true;
537 ctx
->Const
.PrimitiveRestartForPatches
= true;
539 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
540 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
541 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
542 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
543 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
544 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
545 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
546 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
547 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
548 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
549 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
550 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
551 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
552 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
554 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
555 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
556 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
557 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
558 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
559 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
560 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
561 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
562 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
563 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
564 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
566 /* Fragment shaders use real, 32-bit twos-complement integers for all
569 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
570 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
571 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
572 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
573 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
575 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
576 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
577 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
578 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
579 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
581 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
582 * but we're not sure how it's actually done for vertex order,
583 * that affect provoking vertex decision. Always use last vertex
584 * convention for quad primitive which works as expected for now.
586 if (devinfo
->gen
>= 6)
587 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
589 ctx
->Const
.NativeIntegers
= true;
590 ctx
->Const
.VertexID_is_zero_based
= true;
592 /* Regarding the CMP instruction, the Ivybridge PRM says:
594 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
595 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
596 * 0xFFFFFFFF) is assigned to dst."
598 * but PRMs for earlier generations say
600 * "In dword format, one GRF may store up to 8 results. When the register
601 * is used later as a vector of Booleans, as only LSB at each channel
602 * contains meaning [sic] data, software should make sure all higher bits
603 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
605 * We select the representation of a true boolean uniform to be ~0, and fix
606 * the results of Gen <= 5 CMP instruction's with -(result & 1).
608 ctx
->Const
.UniformBooleanTrue
= ~0;
610 /* From the gen4 PRM, volume 4 page 127:
612 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
613 * the base address of the first element of the surface, computed in
614 * software by adding the surface base address to the byte offset of
615 * the element in the buffer."
617 * However, unaligned accesses are slower, so enforce buffer alignment.
619 * In order to push UBO data, 3DSTATE_CONSTANT_XS imposes an additional
620 * restriction: the start of the buffer needs to be 32B aligned.
622 ctx
->Const
.UniformBufferOffsetAlignment
= 32;
624 /* ShaderStorageBufferOffsetAlignment should be a cacheline (64 bytes) so
625 * that we can safely have the CPU and GPU writing the same SSBO on
626 * non-cachecoherent systems (our Atom CPUs). With UBOs, the GPU never
627 * writes, so there's no problem. For an SSBO, the GPU and the CPU can
628 * be updating disjoint regions of the buffer simultaneously and that will
629 * break if the regions overlap the same cacheline.
631 ctx
->Const
.ShaderStorageBufferOffsetAlignment
= 64;
632 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
633 ctx
->Const
.MaxTextureBufferSize
= 128 * 1024 * 1024;
635 if (devinfo
->gen
>= 6) {
636 ctx
->Const
.MaxVarying
= 32;
637 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
638 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
639 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
640 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
641 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxInputComponents
= 128;
642 ctx
->Const
.Program
[MESA_SHADER_TESS_CTRL
].MaxOutputComponents
= 128;
643 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxInputComponents
= 128;
644 ctx
->Const
.Program
[MESA_SHADER_TESS_EVAL
].MaxOutputComponents
= 128;
647 /* We want the GLSL compiler to emit code that uses condition codes */
648 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
649 ctx
->Const
.ShaderCompilerOptions
[i
] =
650 brw
->screen
->compiler
->glsl_compiler_options
[i
];
653 if (devinfo
->gen
>= 7) {
654 ctx
->Const
.MaxViewportWidth
= 32768;
655 ctx
->Const
.MaxViewportHeight
= 32768;
658 /* ARB_viewport_array, OES_viewport_array */
659 if (devinfo
->gen
>= 6) {
660 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
661 ctx
->Const
.ViewportSubpixelBits
= 0;
663 /* Cast to float before negating because MaxViewportWidth is unsigned.
665 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
666 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
669 /* ARB_gpu_shader5 */
670 if (devinfo
->gen
>= 7)
671 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
673 /* ARB_framebuffer_no_attachments */
674 ctx
->Const
.MaxFramebufferWidth
= 16384;
675 ctx
->Const
.MaxFramebufferHeight
= 16384;
676 ctx
->Const
.MaxFramebufferLayers
= ctx
->Const
.MaxArrayTextureLayers
;
677 ctx
->Const
.MaxFramebufferSamples
= max_samples
;
679 /* OES_primitive_bounding_box */
680 ctx
->Const
.NoPrimitiveBoundingBoxOutput
= true;
682 /* TODO: We should be able to use STD430 packing by default on all hardware
683 * but some piglit tests [1] currently fail on SNB when this is enabled.
684 * The problem is the messages we're using for doing uniform pulls
685 * in the vec4 back-end on SNB is the OWORD block load instruction, which
686 * takes its offset in units of OWORDS (16 bytes). On IVB+, we use the
687 * sampler which doesn't have these restrictions.
689 * In the scalar back-end, we use the sampler for dynamic uniform loads and
690 * pull an entire cache line at a time for constant offset loads both of
691 * which support almost any alignment.
693 * [1] glsl-1.40/uniform_buffer/vs-float-array-variable-index.shader_test
695 if (devinfo
->gen
>= 7)
696 ctx
->Const
.UseSTD430AsDefaultPacking
= true;
700 brw_initialize_cs_context_constants(struct brw_context
*brw
)
702 struct gl_context
*ctx
= &brw
->ctx
;
703 const struct intel_screen
*screen
= brw
->screen
;
704 struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
706 /* FINISHME: Do this for all platforms that the kernel supports */
707 if (devinfo
->is_cherryview
&&
708 screen
->subslice_total
> 0 && screen
->eu_total
> 0) {
709 /* Logical CS threads = EUs per subslice * 7 threads per EU */
710 uint32_t max_cs_threads
= screen
->eu_total
/ screen
->subslice_total
* 7;
712 /* Fuse configurations may give more threads than expected, never less. */
713 if (max_cs_threads
> devinfo
->max_cs_threads
)
714 devinfo
->max_cs_threads
= max_cs_threads
;
717 /* Maximum number of scalar compute shader invocations that can be run in
718 * parallel in the same subslice assuming SIMD32 dispatch.
720 * We don't advertise more than 64 threads, because we are limited to 64 by
721 * our usage of thread_width_max in the gpgpu walker command. This only
722 * currently impacts Haswell, which otherwise might be able to advertise 70
723 * threads. With SIMD32 and 64 threads, Haswell still provides twice the
724 * required the number of invocation needed for ARB_compute_shader.
726 const unsigned max_threads
= MIN2(64, devinfo
->max_cs_threads
);
727 const uint32_t max_invocations
= 32 * max_threads
;
728 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
729 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
730 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
731 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
732 ctx
->Const
.MaxComputeSharedMemorySize
= 64 * 1024;
736 * Process driconf (drirc) options, setting appropriate context flags.
738 * intelInitExtensions still pokes at optionCache directly, in order to
739 * avoid advertising various extensions. No flags are set, so it makes
740 * sense to continue doing that there.
743 brw_process_driconf_options(struct brw_context
*brw
)
745 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
746 struct gl_context
*ctx
= &brw
->ctx
;
748 driOptionCache
*options
= &brw
->optionCache
;
749 driParseConfigFiles(options
, &brw
->screen
->optionCache
,
750 brw
->driContext
->driScreenPriv
->myNum
, "i965");
752 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
753 switch (bo_reuse_mode
) {
754 case DRI_CONF_BO_REUSE_DISABLED
:
756 case DRI_CONF_BO_REUSE_ALL
:
757 brw_bufmgr_enable_reuse(brw
->bufmgr
);
761 if (INTEL_DEBUG
& DEBUG_NO_HIZ
) {
762 brw
->has_hiz
= false;
763 /* On gen6, you can only do separate stencil with HIZ. */
764 if (devinfo
->gen
== 6)
765 brw
->has_separate_stencil
= false;
768 if (driQueryOptionb(options
, "mesa_no_error"))
769 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_NO_ERROR_BIT_KHR
;
771 if (driQueryOptionb(options
, "always_flush_batch")) {
772 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
773 brw
->always_flush_batch
= true;
776 if (driQueryOptionb(options
, "always_flush_cache")) {
777 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
778 brw
->always_flush_cache
= true;
781 if (driQueryOptionb(options
, "disable_throttling")) {
782 fprintf(stderr
, "disabling flush throttling\n");
783 brw
->disable_throttling
= true;
786 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
788 if (driQueryOptionb(&brw
->optionCache
, "precise_trig"))
789 brw
->screen
->compiler
->precise_trig
= true;
791 ctx
->Const
.ForceGLSLExtensionsWarn
=
792 driQueryOptionb(options
, "force_glsl_extensions_warn");
794 ctx
->Const
.ForceGLSLVersion
=
795 driQueryOptioni(options
, "force_glsl_version");
797 ctx
->Const
.DisableGLSLLineContinuations
=
798 driQueryOptionb(options
, "disable_glsl_line_continuations");
800 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
801 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
803 ctx
->Const
.AllowGLSLBuiltinVariableRedeclaration
=
804 driQueryOptionb(options
, "allow_glsl_builtin_variable_redeclaration");
806 ctx
->Const
.AllowHigherCompatVersion
=
807 driQueryOptionb(options
, "allow_higher_compat_version");
809 ctx
->Const
.ForceGLSLAbsSqrt
=
810 driQueryOptionb(options
, "force_glsl_abs_sqrt");
812 ctx
->Const
.GLSLZeroInit
= driQueryOptionb(options
, "glsl_zero_init");
814 brw
->dual_color_blend_by_location
=
815 driQueryOptionb(options
, "dual_color_blend_by_location");
819 brwCreateContext(gl_api api
,
820 const struct gl_config
*mesaVis
,
821 __DRIcontext
*driContextPriv
,
822 unsigned major_version
,
823 unsigned minor_version
,
826 unsigned *dri_ctx_error
,
827 void *sharedContextPrivate
)
829 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
830 struct intel_screen
*screen
= driContextPriv
->driScreenPriv
->driverPrivate
;
831 const struct gen_device_info
*devinfo
= &screen
->devinfo
;
832 struct dd_function_table functions
;
834 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
835 * provides us with context reset notifications.
837 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
|
838 __DRI_CTX_FLAG_FORWARD_COMPATIBLE
|
839 __DRI_CTX_FLAG_NO_ERROR
;
841 if (screen
->has_context_reset_notification
)
842 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
844 if (flags
& ~allowed_flags
) {
845 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
849 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
851 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
852 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
856 driContextPriv
->driverPrivate
= brw
;
857 brw
->driContext
= driContextPriv
;
858 brw
->screen
= screen
;
859 brw
->bufmgr
= screen
->bufmgr
;
861 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
862 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
863 brw
->has_pln
= devinfo
->has_pln
;
864 brw
->has_compr4
= devinfo
->has_compr4
;
865 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
866 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
867 brw
->needs_unlit_centroid_workaround
=
868 devinfo
->needs_unlit_centroid_workaround
;
870 brw
->must_use_separate_stencil
= devinfo
->must_use_separate_stencil
;
871 brw
->has_swizzling
= screen
->hw_has_swizzling
;
873 brw
->isl_dev
= screen
->isl_dev
;
875 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
876 brw
->tcs
.base
.stage
= MESA_SHADER_TESS_CTRL
;
877 brw
->tes
.base
.stage
= MESA_SHADER_TESS_EVAL
;
878 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
879 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
880 if (devinfo
->gen
>= 8) {
881 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
882 } else if (devinfo
->gen
>= 7) {
883 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
884 } else if (devinfo
->gen
>= 6) {
885 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
887 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
890 brw_init_driver_functions(brw
, &functions
);
893 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
895 struct gl_context
*ctx
= &brw
->ctx
;
897 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
898 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
899 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
900 intelDestroyContext(driContextPriv
);
904 driContextSetFlags(ctx
, flags
);
906 /* Initialize the software rasterizer and helper modules.
908 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
909 * software fallbacks (which we have to support on legacy GL to do weird
910 * glDrawPixels(), glBitmap(), and other functions).
912 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
913 _swrast_CreateContext(ctx
);
916 _vbo_CreateContext(ctx
);
917 if (ctx
->swrast_context
) {
918 _tnl_CreateContext(ctx
);
919 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
920 _swsetup_CreateContext(ctx
);
922 /* Configure swrast to match hardware characteristics: */
923 _swrast_allow_pixel_fog(ctx
, false);
924 _swrast_allow_vertex_fog(ctx
, true);
927 _mesa_meta_init(ctx
);
929 brw_process_driconf_options(brw
);
931 if (INTEL_DEBUG
& DEBUG_PERF
)
932 brw
->perf_debug
= true;
934 brw_initialize_cs_context_constants(brw
);
935 brw_initialize_context_constants(brw
);
937 ctx
->Const
.ResetStrategy
= notify_reset
938 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
940 /* Reinitialize the context point state. It depends on ctx->Const values. */
941 _mesa_init_point(ctx
);
945 intel_batchbuffer_init(screen
, &brw
->batch
);
947 if (devinfo
->gen
>= 6) {
948 /* Create a new hardware context. Using a hardware context means that
949 * our GPU state will be saved/restored on context switch, allowing us
950 * to assume that the GPU is in the same state we left it in.
952 * This is required for transform feedback buffer offsets, query objects,
953 * and also allows us to reduce how much state we have to emit.
955 brw
->hw_ctx
= brw_create_hw_context(brw
->bufmgr
);
958 fprintf(stderr
, "Failed to create hardware context.\n");
959 intelDestroyContext(driContextPriv
);
964 if (brw_init_pipe_control(brw
, devinfo
)) {
965 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
966 intelDestroyContext(driContextPriv
);
972 intelInitExtensions(ctx
);
974 brw_init_surface_formats(brw
);
978 brw
->urb
.size
= devinfo
->urb
.size
;
980 if (devinfo
->gen
== 6)
981 brw
->urb
.gs_present
= false;
983 brw
->prim_restart
.in_progress
= false;
984 brw
->prim_restart
.enable_cut_index
= false;
985 brw
->gs
.enabled
= false;
986 brw
->clip
.viewport_count
= 1;
988 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
990 brw
->max_gtt_map_object_size
= screen
->max_gtt_map_object_size
;
992 ctx
->VertexProgram
._MaintainTnlProgram
= true;
993 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
995 brw_draw_init( brw
);
997 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
998 /* Turn on some extra GL_ARB_debug_output generation. */
999 brw
->perf_debug
= true;
1002 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0) {
1003 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
1004 ctx
->Const
.RobustAccess
= GL_TRUE
;
1007 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
1008 brw_init_shader_time(brw
);
1010 _mesa_compute_version(ctx
);
1012 _mesa_initialize_dispatch_tables(ctx
);
1013 _mesa_initialize_vbo_vtxfmt(ctx
);
1015 if (ctx
->Extensions
.INTEL_performance_query
)
1016 brw_init_performance_queries(brw
);
1018 vbo_use_buffer_objects(ctx
);
1019 vbo_always_unmap_buffers(ctx
);
1025 intelDestroyContext(__DRIcontext
* driContextPriv
)
1027 struct brw_context
*brw
=
1028 (struct brw_context
*) driContextPriv
->driverPrivate
;
1029 struct gl_context
*ctx
= &brw
->ctx
;
1030 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
1032 _mesa_meta_free(&brw
->ctx
);
1034 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
1035 /* Force a report. */
1036 brw
->shader_time
.report_time
= 0;
1038 brw_collect_and_report_shader_time(brw
);
1039 brw_destroy_shader_time(brw
);
1042 if (devinfo
->gen
>= 6)
1043 blorp_finish(&brw
->blorp
);
1045 brw_destroy_state(brw
);
1046 brw_draw_destroy(brw
);
1048 brw_bo_unreference(brw
->curbe
.curbe_bo
);
1049 if (brw
->vs
.base
.scratch_bo
)
1050 brw_bo_unreference(brw
->vs
.base
.scratch_bo
);
1051 if (brw
->tcs
.base
.scratch_bo
)
1052 brw_bo_unreference(brw
->tcs
.base
.scratch_bo
);
1053 if (brw
->tes
.base
.scratch_bo
)
1054 brw_bo_unreference(brw
->tes
.base
.scratch_bo
);
1055 if (brw
->gs
.base
.scratch_bo
)
1056 brw_bo_unreference(brw
->gs
.base
.scratch_bo
);
1057 if (brw
->wm
.base
.scratch_bo
)
1058 brw_bo_unreference(brw
->wm
.base
.scratch_bo
);
1060 brw_destroy_hw_context(brw
->bufmgr
, brw
->hw_ctx
);
1062 if (ctx
->swrast_context
) {
1063 _swsetup_DestroyContext(&brw
->ctx
);
1064 _tnl_DestroyContext(&brw
->ctx
);
1066 _vbo_DestroyContext(&brw
->ctx
);
1068 if (ctx
->swrast_context
)
1069 _swrast_DestroyContext(&brw
->ctx
);
1071 brw_fini_pipe_control(brw
);
1072 intel_batchbuffer_free(&brw
->batch
);
1074 brw_bo_unreference(brw
->throttle_batch
[1]);
1075 brw_bo_unreference(brw
->throttle_batch
[0]);
1076 brw
->throttle_batch
[1] = NULL
;
1077 brw
->throttle_batch
[0] = NULL
;
1079 driDestroyOptionCache(&brw
->optionCache
);
1081 /* free the Mesa context */
1082 _mesa_free_context_data(&brw
->ctx
);
1085 driContextPriv
->driverPrivate
= NULL
;
1089 intelUnbindContext(__DRIcontext
* driContextPriv
)
1091 /* Unset current context and dispath table */
1092 _mesa_make_current(NULL
, NULL
, NULL
);
1098 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1099 * on window system framebuffers.
1101 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1102 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1103 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1104 * for a visual where you're guaranteed to be capable, but it turns out that
1105 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1106 * incapable ones, because there's no difference between the two in resources
1107 * used. Applications thus get built that accidentally rely on the default
1108 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1111 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1112 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1113 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1114 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1115 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1116 * and get no sRGB encode (assuming that both kinds of visual are available).
1117 * Thus our choice to support sRGB by default on our visuals for desktop would
1118 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1120 * Unfortunately, renderbuffer setup happens before a context is created. So
1121 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1122 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1123 * yet), we go turn that back off before anyone finds out.
1126 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1127 struct gl_framebuffer
*fb
)
1129 struct gl_context
*ctx
= &brw
->ctx
;
1131 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1134 /* Some day when we support the sRGB capable bit on visuals available for
1135 * GLES, we'll need to respect that and not disable things here.
1137 fb
->Visual
.sRGBCapable
= false;
1138 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1139 struct gl_renderbuffer
*rb
= fb
->Attachment
[i
].Renderbuffer
;
1141 rb
->Format
= _mesa_get_srgb_format_linear(rb
->Format
);
1146 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1147 __DRIdrawable
* driDrawPriv
,
1148 __DRIdrawable
* driReadPriv
)
1150 struct brw_context
*brw
;
1151 GET_CURRENT_CONTEXT(curCtx
);
1154 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1158 /* According to the glXMakeCurrent() man page: "Pending commands to
1159 * the previous context, if any, are flushed before it is released."
1160 * But only flush if we're actually changing contexts.
1162 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1163 _mesa_flush(curCtx
);
1166 if (driContextPriv
) {
1167 struct gl_context
*ctx
= &brw
->ctx
;
1168 struct gl_framebuffer
*fb
, *readFb
;
1170 if (driDrawPriv
== NULL
) {
1171 fb
= _mesa_get_incomplete_framebuffer();
1173 fb
= driDrawPriv
->driverPrivate
;
1174 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1177 if (driReadPriv
== NULL
) {
1178 readFb
= _mesa_get_incomplete_framebuffer();
1180 readFb
= driReadPriv
->driverPrivate
;
1181 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1184 /* The sRGB workaround changes the renderbuffer's format. We must change
1185 * the format before the renderbuffer's miptree get's allocated, otherwise
1186 * the formats of the renderbuffer and its miptree will differ.
1188 intel_gles3_srgb_workaround(brw
, fb
);
1189 intel_gles3_srgb_workaround(brw
, readFb
);
1191 /* If the context viewport hasn't been initialized, force a call out to
1192 * the loader to get buffers so we have a drawable size for the initial
1194 if (!brw
->ctx
.ViewportInitialized
)
1195 intel_prepare_render(brw
);
1197 _mesa_make_current(ctx
, fb
, readFb
);
1199 _mesa_make_current(NULL
, NULL
, NULL
);
1206 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1207 __DRIdrawable
*drawable
)
1209 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
1211 if (devinfo
->gen
< 6) {
1212 /* MSAA and fast color clear are not supported, so don't waste time
1213 * checking whether a resolve is needed.
1218 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1219 struct intel_renderbuffer
*rb
;
1221 /* Usually, only the back buffer will need to be downsampled. However,
1222 * the front buffer will also need it if the user has rendered into it.
1224 static const gl_buffer_index buffers
[2] = {
1229 for (int i
= 0; i
< 2; ++i
) {
1230 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1231 if (rb
== NULL
|| rb
->mt
== NULL
)
1233 if (rb
->mt
->surf
.samples
== 1) {
1234 assert(rb
->mt_layer
== 0 && rb
->mt_level
== 0 &&
1235 rb
->layer_count
== 1);
1236 intel_miptree_prepare_external(brw
, rb
->mt
);
1238 intel_renderbuffer_downsample(brw
, rb
);
1244 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1246 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1250 intel_query_dri2_buffers(struct brw_context
*brw
,
1251 __DRIdrawable
*drawable
,
1252 __DRIbuffer
**buffers
,
1256 intel_process_dri2_buffer(struct brw_context
*brw
,
1257 __DRIdrawable
*drawable
,
1258 __DRIbuffer
*buffer
,
1259 struct intel_renderbuffer
*rb
,
1260 const char *buffer_name
);
1263 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1266 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1268 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1269 struct intel_renderbuffer
*rb
;
1270 __DRIbuffer
*buffers
= NULL
;
1272 const char *region_name
;
1274 /* Set this up front, so that in case our buffers get invalidated
1275 * while we're getting new buffers, we don't clobber the stamp and
1276 * thus ignore the invalidate. */
1277 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1279 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1280 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1282 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1284 if (buffers
== NULL
)
1287 for (int i
= 0; i
< count
; i
++) {
1288 switch (buffers
[i
].attachment
) {
1289 case __DRI_BUFFER_FRONT_LEFT
:
1290 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1291 region_name
= "dri2 front buffer";
1294 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1295 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1296 region_name
= "dri2 fake front buffer";
1299 case __DRI_BUFFER_BACK_LEFT
:
1300 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1301 region_name
= "dri2 back buffer";
1304 case __DRI_BUFFER_DEPTH
:
1305 case __DRI_BUFFER_HIZ
:
1306 case __DRI_BUFFER_DEPTH_STENCIL
:
1307 case __DRI_BUFFER_STENCIL
:
1308 case __DRI_BUFFER_ACCUM
:
1311 "unhandled buffer attach event, attachment type %d\n",
1312 buffers
[i
].attachment
);
1316 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1322 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1324 struct brw_context
*brw
= context
->driverPrivate
;
1325 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
1327 /* Set this up front, so that in case our buffers get invalidated
1328 * while we're getting new buffers, we don't clobber the stamp and
1329 * thus ignore the invalidate. */
1330 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1332 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1333 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1335 if (dri_screen
->image
.loader
)
1336 intel_update_image_buffers(brw
, drawable
);
1338 intel_update_dri2_buffers(brw
, drawable
);
1340 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1344 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1345 * state is required.
1348 intel_prepare_render(struct brw_context
*brw
)
1350 struct gl_context
*ctx
= &brw
->ctx
;
1351 __DRIcontext
*driContext
= brw
->driContext
;
1352 __DRIdrawable
*drawable
;
1354 drawable
= driContext
->driDrawablePriv
;
1355 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1356 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1357 intel_update_renderbuffers(driContext
, drawable
);
1358 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1361 drawable
= driContext
->driReadablePriv
;
1362 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1363 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1364 intel_update_renderbuffers(driContext
, drawable
);
1365 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1368 /* If we're currently rendering to the front buffer, the rendering
1369 * that will happen next will probably dirty the front buffer. So
1370 * mark it as dirty here.
1372 if (_mesa_is_front_buffer_drawing(ctx
->DrawBuffer
))
1373 brw
->front_buffer_dirty
= true;
1377 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1379 * To determine which DRI buffers to request, examine the renderbuffers
1380 * attached to the drawable's framebuffer. Then request the buffers with
1381 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1383 * This is called from intel_update_renderbuffers().
1385 * \param drawable Drawable whose buffers are queried.
1386 * \param buffers [out] List of buffers returned by DRI2 query.
1387 * \param buffer_count [out] Number of buffers returned.
1389 * \see intel_update_renderbuffers()
1390 * \see DRI2GetBuffers()
1391 * \see DRI2GetBuffersWithFormat()
1394 intel_query_dri2_buffers(struct brw_context
*brw
,
1395 __DRIdrawable
*drawable
,
1396 __DRIbuffer
**buffers
,
1399 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
1400 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1402 unsigned attachments
[8];
1404 struct intel_renderbuffer
*front_rb
;
1405 struct intel_renderbuffer
*back_rb
;
1407 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1408 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1410 memset(attachments
, 0, sizeof(attachments
));
1411 if ((_mesa_is_front_buffer_drawing(fb
) ||
1412 _mesa_is_front_buffer_reading(fb
) ||
1413 !back_rb
) && front_rb
) {
1414 /* If a fake front buffer is in use, then querying for
1415 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1416 * the real front buffer to the fake front buffer. So before doing the
1417 * query, we need to make sure all the pending drawing has landed in the
1418 * real front buffer.
1420 intel_batchbuffer_flush(brw
);
1421 intel_flush_front(&brw
->ctx
);
1423 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1424 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1425 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1426 /* We have pending front buffer rendering, but we aren't querying for a
1427 * front buffer. If the front buffer we have is a fake front buffer,
1428 * the X server is going to throw it away when it processes the query.
1429 * So before doing the query, make sure all the pending drawing has
1430 * landed in the real front buffer.
1432 intel_batchbuffer_flush(brw
);
1433 intel_flush_front(&brw
->ctx
);
1437 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1438 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1441 assert(i
<= ARRAY_SIZE(attachments
));
1444 dri_screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1449 drawable
->loaderPrivate
);
1453 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1455 * This is called from intel_update_renderbuffers().
1458 * DRI buffers whose attachment point is DRI2BufferStencil or
1459 * DRI2BufferDepthStencil are handled as special cases.
1461 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1462 * that is passed to brw_bo_gem_create_from_name().
1464 * \see intel_update_renderbuffers()
1467 intel_process_dri2_buffer(struct brw_context
*brw
,
1468 __DRIdrawable
*drawable
,
1469 __DRIbuffer
*buffer
,
1470 struct intel_renderbuffer
*rb
,
1471 const char *buffer_name
)
1473 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1479 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1481 /* We try to avoid closing and reopening the same BO name, because the first
1482 * use of a mapping of the buffer involves a bunch of page faulting which is
1483 * moderately expensive.
1485 struct intel_mipmap_tree
*last_mt
;
1486 if (num_samples
== 0)
1489 last_mt
= rb
->singlesample_mt
;
1491 uint32_t old_name
= 0;
1493 /* The bo already has a name because the miptree was created by a
1494 * previous call to intel_process_dri2_buffer(). If a bo already has a
1495 * name, then brw_bo_flink() is a low-cost getter. It does not
1496 * create a new name.
1498 brw_bo_flink(last_mt
->bo
, &old_name
);
1501 if (old_name
== buffer
->name
)
1504 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1506 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1507 buffer
->name
, buffer
->attachment
,
1508 buffer
->cpp
, buffer
->pitch
);
1511 bo
= brw_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1515 "Failed to open BO for returned DRI2 buffer "
1516 "(%dx%d, %s, named %d).\n"
1517 "This is likely a bug in the X Server that will lead to a "
1519 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1523 struct intel_mipmap_tree
*mt
=
1524 intel_miptree_create_for_bo(brw
,
1526 intel_rb_format(rb
),
1532 MIPTREE_CREATE_DEFAULT
);
1534 brw_bo_unreference(bo
);
1538 /* We got this BO from X11. We cana't assume that we have coherent texture
1539 * access because X may suddenly decide to use it for scan-out which would
1540 * destroy coherency.
1542 bo
->cache_coherent
= false;
1544 if (!intel_update_winsys_renderbuffer_miptree(brw
, rb
, mt
,
1545 drawable
->w
, drawable
->h
,
1547 brw_bo_unreference(bo
);
1548 intel_miptree_release(&mt
);
1552 if (_mesa_is_front_buffer_drawing(fb
) &&
1553 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1554 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1555 rb
->Base
.Base
.NumSamples
> 1) {
1556 intel_renderbuffer_upsample(brw
, rb
);
1561 brw_bo_unreference(bo
);
1565 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1567 * To determine which DRI buffers to request, examine the renderbuffers
1568 * attached to the drawable's framebuffer. Then request the buffers from
1571 * This is called from intel_update_renderbuffers().
1573 * \param drawable Drawable whose buffers are queried.
1574 * \param buffers [out] List of buffers returned by DRI2 query.
1575 * \param buffer_count [out] Number of buffers returned.
1577 * \see intel_update_renderbuffers()
1581 intel_update_image_buffer(struct brw_context
*intel
,
1582 __DRIdrawable
*drawable
,
1583 struct intel_renderbuffer
*rb
,
1585 enum __DRIimageBufferMask buffer_type
)
1587 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1589 if (!rb
|| !buffer
->bo
)
1592 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1594 /* Check and see if we're already bound to the right
1597 struct intel_mipmap_tree
*last_mt
;
1598 if (num_samples
== 0)
1601 last_mt
= rb
->singlesample_mt
;
1603 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1606 enum isl_colorspace colorspace
;
1607 switch (_mesa_get_format_color_encoding(intel_rb_format(rb
))) {
1609 colorspace
= ISL_COLORSPACE_SRGB
;
1612 colorspace
= ISL_COLORSPACE_LINEAR
;
1615 unreachable("Invalid color encoding");
1618 struct intel_mipmap_tree
*mt
=
1619 intel_miptree_create_for_dri_image(intel
, buffer
, GL_TEXTURE_2D
,
1624 if (!intel_update_winsys_renderbuffer_miptree(intel
, rb
, mt
,
1625 buffer
->width
, buffer
->height
,
1627 intel_miptree_release(&mt
);
1631 if (_mesa_is_front_buffer_drawing(fb
) &&
1632 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1633 rb
->Base
.Base
.NumSamples
> 1) {
1634 intel_renderbuffer_upsample(intel
, rb
);
1639 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1641 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1642 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
1643 struct intel_renderbuffer
*front_rb
;
1644 struct intel_renderbuffer
*back_rb
;
1645 struct __DRIimageList images
;
1647 uint32_t buffer_mask
= 0;
1650 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1651 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1654 format
= intel_rb_format(back_rb
);
1656 format
= intel_rb_format(front_rb
);
1660 if (front_rb
&& (_mesa_is_front_buffer_drawing(fb
) ||
1661 _mesa_is_front_buffer_reading(fb
) || !back_rb
)) {
1662 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1666 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1668 ret
= dri_screen
->image
.loader
->getBuffers(drawable
,
1669 driGLFormatToImageFormat(format
),
1670 &drawable
->dri2
.stamp
,
1671 drawable
->loaderPrivate
,
1677 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1678 drawable
->w
= images
.front
->width
;
1679 drawable
->h
= images
.front
->height
;
1680 intel_update_image_buffer(brw
,
1684 __DRI_IMAGE_BUFFER_FRONT
);
1687 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1688 drawable
->w
= images
.back
->width
;
1689 drawable
->h
= images
.back
->height
;
1690 intel_update_image_buffer(brw
,
1694 __DRI_IMAGE_BUFFER_BACK
);