2 Copyright 2003 VMware, Inc.
3 Copyright (C) Intel Corp. 2006. All Rights Reserved.
4 Intel funded Tungsten Graphics to
5 develop this 3D driver.
7 Permission is hereby granted, free of charge, to any person obtaining
8 a copy of this software and associated documentation files (the
9 "Software"), to deal in the Software without restriction, including
10 without limitation the rights to use, copy, modify, merge, publish,
11 distribute, sublicense, and/or sell copies of the Software, and to
12 permit persons to whom the Software is furnished to do so, subject to
13 the following conditions:
15 The above copyright notice and this permission notice (including the
16 next paragraph) shall be included in all copies or substantial
17 portions of the Software.
19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
20 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
22 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
23 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
24 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
25 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **********************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
34 #include "main/api_exec.h"
35 #include "main/context.h"
36 #include "main/fbobject.h"
37 #include "main/extensions.h"
38 #include "main/imports.h"
39 #include "main/macros.h"
40 #include "main/points.h"
41 #include "main/version.h"
42 #include "main/vtxfmt.h"
43 #include "main/texobj.h"
45 #include "vbo/vbo_context.h"
47 #include "drivers/common/driverfuncs.h"
48 #include "drivers/common/meta.h"
51 #include "brw_context.h"
52 #include "brw_defines.h"
54 #include "brw_state.h"
56 #include "intel_batchbuffer.h"
57 #include "intel_buffer_objects.h"
58 #include "intel_buffers.h"
59 #include "intel_fbo.h"
60 #include "intel_mipmap_tree.h"
61 #include "intel_pixel.h"
62 #include "intel_image.h"
63 #include "intel_tex.h"
64 #include "intel_tex_obj.h"
66 #include "swrast_setup/swrast_setup.h"
68 #include "tnl/t_pipeline.h"
69 #include "util/ralloc.h"
71 #include "glsl/nir/nir.h"
73 /***************************************
74 * Mesa's Driver Functions
75 ***************************************/
78 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
79 GLenum internalFormat
, int samples
[16])
81 struct brw_context
*brw
= brw_context(ctx
);
103 assert(brw
->gen
< 6);
109 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
112 brw_get_renderer_string(unsigned deviceID
)
115 static char buffer
[128];
119 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
120 #include "pci_ids/i965_pci_ids.h"
122 chipset
= "Unknown Intel Chipset";
126 (void) driGetRendererString(buffer
, chipset
, 0);
130 static const GLubyte
*
131 intel_get_string(struct gl_context
* ctx
, GLenum name
)
133 const struct brw_context
*const brw
= brw_context(ctx
);
137 return (GLubyte
*) brw_vendor_string
;
141 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
149 intel_viewport(struct gl_context
*ctx
)
151 struct brw_context
*brw
= brw_context(ctx
);
152 __DRIcontext
*driContext
= brw
->driContext
;
154 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
155 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
156 dri2InvalidateDrawable(driContext
->driReadablePriv
);
161 intel_update_state(struct gl_context
* ctx
, GLuint new_state
)
163 struct brw_context
*brw
= brw_context(ctx
);
164 struct intel_texture_object
*tex_obj
;
165 struct intel_renderbuffer
*depth_irb
;
167 if (ctx
->swrast_context
)
168 _swrast_InvalidateState(ctx
, new_state
);
169 _vbo_InvalidateState(ctx
, new_state
);
171 brw
->NewGLState
|= new_state
;
173 _mesa_unlock_context_textures(ctx
);
175 /* Resolve the depth buffer's HiZ buffer. */
176 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
178 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
180 /* Resolve depth buffer and render cache of each enabled texture. */
181 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
182 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
183 if (!ctx
->Texture
.Unit
[i
]._Current
)
185 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
186 if (!tex_obj
|| !tex_obj
->mt
)
188 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
189 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
190 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
193 _mesa_lock_context_textures(ctx
);
196 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
199 intel_flush_front(struct gl_context
*ctx
)
201 struct brw_context
*brw
= brw_context(ctx
);
202 __DRIcontext
*driContext
= brw
->driContext
;
203 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
204 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
206 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
207 if (flushFront(screen
) && driDrawable
&&
208 driDrawable
->loaderPrivate
) {
210 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
212 * This potentially resolves both front and back buffer. It
213 * is unnecessary to resolve the back, but harms nothing except
214 * performance. And no one cares about front-buffer render
217 intel_resolve_for_dri2_flush(brw
, driDrawable
);
218 intel_batchbuffer_flush(brw
);
220 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
222 /* We set the dirty bit in intel_prepare_render() if we're
223 * front buffer rendering once we get there.
225 brw
->front_buffer_dirty
= false;
231 intel_glFlush(struct gl_context
*ctx
)
233 struct brw_context
*brw
= brw_context(ctx
);
235 intel_batchbuffer_flush(brw
);
236 intel_flush_front(ctx
);
238 brw
->need_flush_throttle
= true;
242 intel_finish(struct gl_context
* ctx
)
244 struct brw_context
*brw
= brw_context(ctx
);
248 if (brw
->batch
.last_bo
)
249 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
253 brw_init_driver_functions(struct brw_context
*brw
,
254 struct dd_function_table
*functions
)
256 _mesa_init_driver_functions(functions
);
258 /* GLX uses DRI2 invalidate events to handle window resizing.
259 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
260 * which doesn't provide a mechanism for snooping the event queues.
262 * So EGL still relies on viewport hacks to handle window resizing.
263 * This should go away with DRI3000.
265 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
266 functions
->Viewport
= intel_viewport
;
268 functions
->Flush
= intel_glFlush
;
269 functions
->Finish
= intel_finish
;
270 functions
->GetString
= intel_get_string
;
271 functions
->UpdateState
= intel_update_state
;
273 intelInitTextureFuncs(functions
);
274 intelInitTextureImageFuncs(functions
);
275 intelInitTextureSubImageFuncs(functions
);
276 intelInitTextureCopyImageFuncs(functions
);
277 intelInitCopyImageFuncs(functions
);
278 intelInitClearFuncs(functions
);
279 intelInitBufferFuncs(functions
);
280 intelInitPixelFuncs(functions
);
281 intelInitBufferObjectFuncs(functions
);
282 intel_init_syncobj_functions(functions
);
283 brw_init_object_purgeable_functions(functions
);
285 brwInitFragProgFuncs( functions
);
286 brw_init_common_queryobj_functions(functions
);
288 gen6_init_queryobj_functions(functions
);
290 gen4_init_queryobj_functions(functions
);
291 brw_init_compute_functions(functions
);
293 brw_init_conditional_render_functions(functions
);
295 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
297 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
298 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
299 functions
->GetTransformFeedbackVertexCount
=
300 brw_get_transform_feedback_vertex_count
;
302 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
303 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
304 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
305 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
307 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
308 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
312 functions
->GetSamplePosition
= gen6_get_sample_position
;
316 brw_initialize_context_constants(struct brw_context
*brw
)
318 struct gl_context
*ctx
= &brw
->ctx
;
320 unsigned max_samplers
=
321 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
323 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
325 ctx
->Const
.StripTextureBorder
= true;
327 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
328 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
329 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
330 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
331 ctx
->Const
.MaxTextureUnits
=
332 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
333 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
334 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
336 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
338 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
339 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
340 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
341 ctx
->Const
.MaxUniformBufferBindings
+= 12;
343 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
345 ctx
->Const
.MaxCombinedTextureImageUnits
=
346 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
347 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
348 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
349 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
351 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
352 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
353 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
354 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
355 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
356 ctx
->Const
.MaxTextureMbytes
= 1536;
359 ctx
->Const
.MaxArrayTextureLayers
= 2048;
361 ctx
->Const
.MaxArrayTextureLayers
= 512;
363 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
365 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
367 ctx
->Const
.MaxRenderbufferSize
= 8192;
369 /* Hardware only supports a limited number of transform feedback buffers.
370 * So we need to override the Mesa default (which is based only on software
373 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
375 /* On Gen6, in the worst case, we use up one binding table entry per
376 * transform feedback component (see comments above the definition of
377 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
378 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
379 * BRW_MAX_SOL_BINDINGS.
381 * In "separate components" mode, we need to divide this value by
382 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
383 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
385 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
386 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
387 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
389 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
392 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
393 const int clamp_max_samples
=
394 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
396 if (clamp_max_samples
< 0) {
397 max_samples
= msaa_modes
[0];
399 /* Select the largest supported MSAA mode that does not exceed
403 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
404 if (msaa_modes
[i
] <= clamp_max_samples
) {
405 max_samples
= msaa_modes
[i
];
411 ctx
->Const
.MaxSamples
= max_samples
;
412 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
413 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
414 ctx
->Const
.MaxIntegerSamples
= max_samples
;
416 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
417 * to map indices of rectangular grid to sample numbers within a pixel.
418 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
419 * extension implementation. For more details see the comment above
420 * gen6_set_sample_maps() definition.
422 gen6_set_sample_maps(ctx
);
425 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
426 else if (brw
->gen
== 6)
427 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
429 ctx
->Const
.MinLineWidth
= 1.0;
430 ctx
->Const
.MinLineWidthAA
= 1.0;
431 if (brw
->gen
>= 9 || brw
->is_cherryview
) {
432 ctx
->Const
.MaxLineWidth
= 40.0;
433 ctx
->Const
.MaxLineWidthAA
= 40.0;
434 ctx
->Const
.LineWidthGranularity
= 0.125;
435 } else if (brw
->gen
>= 6) {
436 ctx
->Const
.MaxLineWidth
= 7.375;
437 ctx
->Const
.MaxLineWidthAA
= 7.375;
438 ctx
->Const
.LineWidthGranularity
= 0.125;
440 ctx
->Const
.MaxLineWidth
= 7.0;
441 ctx
->Const
.MaxLineWidthAA
= 7.0;
442 ctx
->Const
.LineWidthGranularity
= 0.5;
445 /* For non-antialiased lines, we have to round the line width to the
446 * nearest whole number. Make sure that we don't advertise a line
447 * width that, when rounded, will be beyond the actual hardware
450 assert(roundf(ctx
->Const
.MaxLineWidth
) <= ctx
->Const
.MaxLineWidth
);
452 ctx
->Const
.MinPointSize
= 1.0;
453 ctx
->Const
.MinPointSizeAA
= 1.0;
454 ctx
->Const
.MaxPointSize
= 255.0;
455 ctx
->Const
.MaxPointSizeAA
= 255.0;
456 ctx
->Const
.PointSizeGranularity
= 1.0;
458 if (brw
->gen
>= 5 || brw
->is_g4x
)
459 ctx
->Const
.MaxClipPlanes
= 8;
461 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
462 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
463 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
464 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
465 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
466 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
467 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
468 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
469 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
470 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
471 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
472 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
473 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
474 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
476 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
477 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
478 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
479 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
480 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
481 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
482 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
483 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
484 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
485 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
486 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
488 /* Fragment shaders use real, 32-bit twos-complement integers for all
491 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
492 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
493 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
494 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
495 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
497 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
498 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
499 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
500 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
501 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
504 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
505 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
506 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
507 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
508 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
509 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
510 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
511 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
512 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
515 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
516 * but we're not sure how it's actually done for vertex order,
517 * that affect provoking vertex decision. Always use last vertex
518 * convention for quad primitive which works as expected for now.
521 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
523 ctx
->Const
.NativeIntegers
= true;
524 ctx
->Const
.VertexID_is_zero_based
= true;
526 /* Regarding the CMP instruction, the Ivybridge PRM says:
528 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
529 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
530 * 0xFFFFFFFF) is assigned to dst."
532 * but PRMs for earlier generations say
534 * "In dword format, one GRF may store up to 8 results. When the register
535 * is used later as a vector of Booleans, as only LSB at each channel
536 * contains meaning [sic] data, software should make sure all higher bits
537 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
539 * We select the representation of a true boolean uniform to be ~0, and fix
540 * the results of Gen <= 5 CMP instruction's with -(result & 1).
542 ctx
->Const
.UniformBooleanTrue
= ~0;
544 /* From the gen4 PRM, volume 4 page 127:
546 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
547 * the base address of the first element of the surface, computed in
548 * software by adding the surface base address to the byte offset of
549 * the element in the buffer."
551 * However, unaligned accesses are slower, so enforce buffer alignment.
553 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
554 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
555 ctx
->Const
.MaxTextureBufferSize
= 128 * 1024 * 1024;
558 ctx
->Const
.MaxVarying
= 32;
559 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
560 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
561 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
562 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
565 static const nir_shader_compiler_options nir_options
= {
566 .native_integers
= true,
567 /* In order to help allow for better CSE at the NIR level we tell NIR
568 * to split all ffma instructions during opt_algebraic and we then
569 * re-combine them as a later step.
575 /* We want the GLSL compiler to emit code that uses condition codes */
576 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
577 ctx
->Const
.ShaderCompilerOptions
[i
].MaxIfDepth
= brw
->gen
< 6 ? 16 : UINT_MAX
;
578 ctx
->Const
.ShaderCompilerOptions
[i
].EmitCondCodes
= true;
579 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoNoise
= true;
580 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoMainReturn
= true;
581 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectInput
= true;
582 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectOutput
=
583 (i
== MESA_SHADER_FRAGMENT
);
584 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectTemp
=
585 (i
== MESA_SHADER_FRAGMENT
);
586 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectUniform
= false;
587 ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
= true;
590 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= true;
591 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_GEOMETRY
].OptimizeForAOS
= true;
593 if (brw
->scalar_vs
) {
594 /* If we're using the scalar backend for vertex shaders, we need to
595 * configure these accordingly.
597 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectOutput
= true;
598 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].EmitNoIndirectTemp
= true;
599 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
= false;
601 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].NirOptions
= &nir_options
;
604 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_FRAGMENT
].NirOptions
= &nir_options
;
605 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_COMPUTE
].NirOptions
= &nir_options
;
607 /* ARB_viewport_array */
608 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
609 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
610 ctx
->Const
.ViewportSubpixelBits
= 0;
612 /* Cast to float before negating because MaxViewportWidth is unsigned.
614 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
615 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
618 /* ARB_gpu_shader5 */
620 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
624 brw_adjust_cs_context_constants(struct brw_context
*brw
)
626 struct gl_context
*ctx
= &brw
->ctx
;
628 /* For ES, we set these constants based on SIMD8.
630 * TODO: Once we can always generate SIMD16, we should update this.
632 * For GL, we assume we can generate a SIMD16 program, but this currently
633 * is not always true. This allows us to run more test cases, and will be
634 * required based on desktop GL compute shader requirements.
636 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
638 const uint32_t max_invocations
= simd_size
* brw
->max_cs_threads
;
639 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
640 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
641 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
642 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
646 * Process driconf (drirc) options, setting appropriate context flags.
648 * intelInitExtensions still pokes at optionCache directly, in order to
649 * avoid advertising various extensions. No flags are set, so it makes
650 * sense to continue doing that there.
653 brw_process_driconf_options(struct brw_context
*brw
)
655 struct gl_context
*ctx
= &brw
->ctx
;
657 driOptionCache
*options
= &brw
->optionCache
;
658 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
659 brw
->driContext
->driScreenPriv
->myNum
, "i965");
661 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
662 switch (bo_reuse_mode
) {
663 case DRI_CONF_BO_REUSE_DISABLED
:
665 case DRI_CONF_BO_REUSE_ALL
:
666 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
670 if (!driQueryOptionb(options
, "hiz")) {
671 brw
->has_hiz
= false;
672 /* On gen6, you can only do separate stencil with HIZ. */
674 brw
->has_separate_stencil
= false;
677 if (driQueryOptionb(options
, "always_flush_batch")) {
678 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
679 brw
->always_flush_batch
= true;
682 if (driQueryOptionb(options
, "always_flush_cache")) {
683 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
684 brw
->always_flush_cache
= true;
687 if (driQueryOptionb(options
, "disable_throttling")) {
688 fprintf(stderr
, "disabling flush throttling\n");
689 brw
->disable_throttling
= true;
692 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
694 ctx
->Const
.ForceGLSLExtensionsWarn
=
695 driQueryOptionb(options
, "force_glsl_extensions_warn");
697 ctx
->Const
.DisableGLSLLineContinuations
=
698 driQueryOptionb(options
, "disable_glsl_line_continuations");
700 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
701 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
705 brwCreateContext(gl_api api
,
706 const struct gl_config
*mesaVis
,
707 __DRIcontext
*driContextPriv
,
708 unsigned major_version
,
709 unsigned minor_version
,
712 unsigned *dri_ctx_error
,
713 void *sharedContextPrivate
)
715 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
716 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
717 struct intel_screen
*screen
= sPriv
->driverPrivate
;
718 const struct brw_device_info
*devinfo
= screen
->devinfo
;
719 struct dd_function_table functions
;
721 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
722 * provides us with context reset notifications.
724 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
725 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
727 if (screen
->has_context_reset_notification
)
728 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
730 if (flags
& ~allowed_flags
) {
731 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
735 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
737 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
738 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
742 driContextPriv
->driverPrivate
= brw
;
743 brw
->driContext
= driContextPriv
;
744 brw
->intelScreen
= screen
;
745 brw
->bufmgr
= screen
->bufmgr
;
747 brw
->gen
= devinfo
->gen
;
748 brw
->gt
= devinfo
->gt
;
749 brw
->is_g4x
= devinfo
->is_g4x
;
750 brw
->is_baytrail
= devinfo
->is_baytrail
;
751 brw
->is_haswell
= devinfo
->is_haswell
;
752 brw
->is_cherryview
= devinfo
->is_cherryview
;
753 brw
->has_llc
= devinfo
->has_llc
;
754 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
755 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
756 brw
->has_pln
= devinfo
->has_pln
;
757 brw
->has_compr4
= devinfo
->has_compr4
;
758 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
759 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
760 brw
->needs_unlit_centroid_workaround
=
761 devinfo
->needs_unlit_centroid_workaround
;
763 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
764 brw
->has_swizzling
= screen
->hw_has_swizzling
;
766 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
767 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
768 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
770 gen8_init_vtable_surface_functions(brw
);
771 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
772 } else if (brw
->gen
>= 7) {
773 gen7_init_vtable_surface_functions(brw
);
774 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
775 } else if (brw
->gen
>= 6) {
776 gen6_init_vtable_surface_functions(brw
);
777 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
779 gen4_init_vtable_surface_functions(brw
);
780 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
783 brw_init_driver_functions(brw
, &functions
);
786 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
788 struct gl_context
*ctx
= &brw
->ctx
;
790 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
791 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
792 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
793 intelDestroyContext(driContextPriv
);
797 driContextSetFlags(ctx
, flags
);
799 /* Initialize the software rasterizer and helper modules.
801 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
802 * software fallbacks (which we have to support on legacy GL to do weird
803 * glDrawPixels(), glBitmap(), and other functions).
805 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
806 _swrast_CreateContext(ctx
);
809 _vbo_CreateContext(ctx
);
810 if (ctx
->swrast_context
) {
811 _tnl_CreateContext(ctx
);
812 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
813 _swsetup_CreateContext(ctx
);
815 /* Configure swrast to match hardware characteristics: */
816 _swrast_allow_pixel_fog(ctx
, false);
817 _swrast_allow_vertex_fog(ctx
, true);
820 _mesa_meta_init(ctx
);
822 brw_process_driconf_options(brw
);
823 brw_process_intel_debug_variable(brw
);
825 if (brw
->gen
>= 8 && !(INTEL_DEBUG
& DEBUG_VEC4VS
))
826 brw
->scalar_vs
= true;
828 brw_initialize_context_constants(brw
);
830 ctx
->Const
.ResetStrategy
= notify_reset
831 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
833 /* Reinitialize the context point state. It depends on ctx->Const values. */
834 _mesa_init_point(ctx
);
838 intel_batchbuffer_init(brw
);
841 /* Create a new hardware context. Using a hardware context means that
842 * our GPU state will be saved/restored on context switch, allowing us
843 * to assume that the GPU is in the same state we left it in.
845 * This is required for transform feedback buffer offsets, query objects,
846 * and also allows us to reduce how much state we have to emit.
848 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
851 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
852 intelDestroyContext(driContextPriv
);
859 intelInitExtensions(ctx
);
861 brw_init_surface_formats(brw
);
863 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
864 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
865 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
866 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
867 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
868 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
869 brw
->urb
.size
= devinfo
->urb
.size
;
870 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
871 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
872 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
873 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
874 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
876 brw_adjust_cs_context_constants(brw
);
878 /* Estimate the size of the mappable aperture into the GTT. There's an
879 * ioctl to get the whole GTT size, but not one to get the mappable subset.
880 * It turns out it's basically always 256MB, though some ancient hardware
883 uint32_t gtt_size
= 256 * 1024 * 1024;
885 /* We don't want to map two objects such that a memcpy between them would
886 * just fault one mapping in and then the other over and over forever. So
887 * we would need to divide the GTT size by 2. Additionally, some GTT is
888 * taken up by things like the framebuffer and the ringbuffer and such, so
889 * be more conservative.
891 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
894 brw
->urb
.gs_present
= false;
896 brw
->prim_restart
.in_progress
= false;
897 brw
->prim_restart
.enable_cut_index
= false;
898 brw
->gs
.enabled
= false;
899 brw
->sf
.viewport_transform_enable
= true;
901 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
903 ctx
->VertexProgram
._MaintainTnlProgram
= true;
904 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
906 brw_draw_init( brw
);
908 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
909 /* Turn on some extra GL_ARB_debug_output generation. */
910 brw
->perf_debug
= true;
913 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
914 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
916 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
917 brw_init_shader_time(brw
);
919 _mesa_compute_version(ctx
);
921 _mesa_initialize_dispatch_tables(ctx
);
922 _mesa_initialize_vbo_vtxfmt(ctx
);
924 if (ctx
->Extensions
.AMD_performance_monitor
) {
925 brw_init_performance_monitors(brw
);
928 vbo_use_buffer_objects(ctx
);
929 vbo_always_unmap_buffers(ctx
);
935 intelDestroyContext(__DRIcontext
* driContextPriv
)
937 struct brw_context
*brw
=
938 (struct brw_context
*) driContextPriv
->driverPrivate
;
939 struct gl_context
*ctx
= &brw
->ctx
;
941 /* Dump a final BMP in case the application doesn't call SwapBuffers */
942 if (INTEL_DEBUG
& DEBUG_AUB
) {
943 intel_batchbuffer_flush(brw
);
944 aub_dump_bmp(&brw
->ctx
);
947 _mesa_meta_free(&brw
->ctx
);
948 brw_meta_fast_clear_free(brw
);
950 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
951 /* Force a report. */
952 brw
->shader_time
.report_time
= 0;
954 brw_collect_and_report_shader_time(brw
);
955 brw_destroy_shader_time(brw
);
958 brw_destroy_state(brw
);
959 brw_draw_destroy(brw
);
961 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
962 if (brw
->vs
.base
.scratch_bo
)
963 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
964 if (brw
->gs
.base
.scratch_bo
)
965 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
966 if (brw
->wm
.base
.scratch_bo
)
967 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
969 drm_intel_gem_context_destroy(brw
->hw_ctx
);
971 if (ctx
->swrast_context
) {
972 _swsetup_DestroyContext(&brw
->ctx
);
973 _tnl_DestroyContext(&brw
->ctx
);
975 _vbo_DestroyContext(&brw
->ctx
);
977 if (ctx
->swrast_context
)
978 _swrast_DestroyContext(&brw
->ctx
);
980 intel_batchbuffer_free(brw
);
982 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
983 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
984 brw
->throttle_batch
[1] = NULL
;
985 brw
->throttle_batch
[0] = NULL
;
987 driDestroyOptionCache(&brw
->optionCache
);
989 /* free the Mesa context */
990 _mesa_free_context_data(&brw
->ctx
);
993 driContextPriv
->driverPrivate
= NULL
;
997 intelUnbindContext(__DRIcontext
* driContextPriv
)
999 /* Unset current context and dispath table */
1000 _mesa_make_current(NULL
, NULL
, NULL
);
1006 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1007 * on window system framebuffers.
1009 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1010 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1011 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1012 * for a visual where you're guaranteed to be capable, but it turns out that
1013 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1014 * incapable ones, because there's no difference between the two in resources
1015 * used. Applications thus get built that accidentally rely on the default
1016 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1019 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1020 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1021 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1022 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1023 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1024 * and get no sRGB encode (assuming that both kinds of visual are available).
1025 * Thus our choice to support sRGB by default on our visuals for desktop would
1026 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1028 * Unfortunately, renderbuffer setup happens before a context is created. So
1029 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1030 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1031 * yet), we go turn that back off before anyone finds out.
1034 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1035 struct gl_framebuffer
*fb
)
1037 struct gl_context
*ctx
= &brw
->ctx
;
1039 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1042 /* Some day when we support the sRGB capable bit on visuals available for
1043 * GLES, we'll need to respect that and not disable things here.
1045 fb
->Visual
.sRGBCapable
= false;
1046 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1047 if (fb
->Attachment
[i
].Renderbuffer
&&
1048 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1049 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1055 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1056 __DRIdrawable
* driDrawPriv
,
1057 __DRIdrawable
* driReadPriv
)
1059 struct brw_context
*brw
;
1060 GET_CURRENT_CONTEXT(curCtx
);
1063 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1067 /* According to the glXMakeCurrent() man page: "Pending commands to
1068 * the previous context, if any, are flushed before it is released."
1069 * But only flush if we're actually changing contexts.
1071 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1072 _mesa_flush(curCtx
);
1075 if (driContextPriv
) {
1076 struct gl_context
*ctx
= &brw
->ctx
;
1077 struct gl_framebuffer
*fb
, *readFb
;
1079 if (driDrawPriv
== NULL
) {
1080 fb
= _mesa_get_incomplete_framebuffer();
1082 fb
= driDrawPriv
->driverPrivate
;
1083 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1086 if (driReadPriv
== NULL
) {
1087 readFb
= _mesa_get_incomplete_framebuffer();
1089 readFb
= driReadPriv
->driverPrivate
;
1090 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1093 /* The sRGB workaround changes the renderbuffer's format. We must change
1094 * the format before the renderbuffer's miptree get's allocated, otherwise
1095 * the formats of the renderbuffer and its miptree will differ.
1097 intel_gles3_srgb_workaround(brw
, fb
);
1098 intel_gles3_srgb_workaround(brw
, readFb
);
1100 /* If the context viewport hasn't been initialized, force a call out to
1101 * the loader to get buffers so we have a drawable size for the initial
1103 if (!brw
->ctx
.ViewportInitialized
)
1104 intel_prepare_render(brw
);
1106 _mesa_make_current(ctx
, fb
, readFb
);
1108 _mesa_make_current(NULL
, NULL
, NULL
);
1115 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1116 __DRIdrawable
*drawable
)
1119 /* MSAA and fast color clear are not supported, so don't waste time
1120 * checking whether a resolve is needed.
1125 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1126 struct intel_renderbuffer
*rb
;
1128 /* Usually, only the back buffer will need to be downsampled. However,
1129 * the front buffer will also need it if the user has rendered into it.
1131 static const gl_buffer_index buffers
[2] = {
1136 for (int i
= 0; i
< 2; ++i
) {
1137 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1138 if (rb
== NULL
|| rb
->mt
== NULL
)
1140 if (rb
->mt
->num_samples
<= 1)
1141 intel_miptree_resolve_color(brw
, rb
->mt
);
1143 intel_renderbuffer_downsample(brw
, rb
);
1148 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1150 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1154 intel_query_dri2_buffers(struct brw_context
*brw
,
1155 __DRIdrawable
*drawable
,
1156 __DRIbuffer
**buffers
,
1160 intel_process_dri2_buffer(struct brw_context
*brw
,
1161 __DRIdrawable
*drawable
,
1162 __DRIbuffer
*buffer
,
1163 struct intel_renderbuffer
*rb
,
1164 const char *buffer_name
);
1167 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1170 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1172 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1173 struct intel_renderbuffer
*rb
;
1174 __DRIbuffer
*buffers
= NULL
;
1176 const char *region_name
;
1178 /* Set this up front, so that in case our buffers get invalidated
1179 * while we're getting new buffers, we don't clobber the stamp and
1180 * thus ignore the invalidate. */
1181 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1183 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1184 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1186 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1188 if (buffers
== NULL
)
1191 for (i
= 0; i
< count
; i
++) {
1192 switch (buffers
[i
].attachment
) {
1193 case __DRI_BUFFER_FRONT_LEFT
:
1194 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1195 region_name
= "dri2 front buffer";
1198 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1199 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1200 region_name
= "dri2 fake front buffer";
1203 case __DRI_BUFFER_BACK_LEFT
:
1204 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1205 region_name
= "dri2 back buffer";
1208 case __DRI_BUFFER_DEPTH
:
1209 case __DRI_BUFFER_HIZ
:
1210 case __DRI_BUFFER_DEPTH_STENCIL
:
1211 case __DRI_BUFFER_STENCIL
:
1212 case __DRI_BUFFER_ACCUM
:
1215 "unhandled buffer attach event, attachment type %d\n",
1216 buffers
[i
].attachment
);
1220 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1226 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1228 struct brw_context
*brw
= context
->driverPrivate
;
1229 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1231 /* Set this up front, so that in case our buffers get invalidated
1232 * while we're getting new buffers, we don't clobber the stamp and
1233 * thus ignore the invalidate. */
1234 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1236 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1237 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1239 if (screen
->image
.loader
)
1240 intel_update_image_buffers(brw
, drawable
);
1242 intel_update_dri2_buffers(brw
, drawable
);
1244 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1248 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1249 * state is required.
1252 intel_prepare_render(struct brw_context
*brw
)
1254 struct gl_context
*ctx
= &brw
->ctx
;
1255 __DRIcontext
*driContext
= brw
->driContext
;
1256 __DRIdrawable
*drawable
;
1258 drawable
= driContext
->driDrawablePriv
;
1259 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1260 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1261 intel_update_renderbuffers(driContext
, drawable
);
1262 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1265 drawable
= driContext
->driReadablePriv
;
1266 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1267 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1268 intel_update_renderbuffers(driContext
, drawable
);
1269 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1272 /* If we're currently rendering to the front buffer, the rendering
1273 * that will happen next will probably dirty the front buffer. So
1274 * mark it as dirty here.
1276 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1277 brw
->front_buffer_dirty
= true;
1281 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1283 * To determine which DRI buffers to request, examine the renderbuffers
1284 * attached to the drawable's framebuffer. Then request the buffers with
1285 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1287 * This is called from intel_update_renderbuffers().
1289 * \param drawable Drawable whose buffers are queried.
1290 * \param buffers [out] List of buffers returned by DRI2 query.
1291 * \param buffer_count [out] Number of buffers returned.
1293 * \see intel_update_renderbuffers()
1294 * \see DRI2GetBuffers()
1295 * \see DRI2GetBuffersWithFormat()
1298 intel_query_dri2_buffers(struct brw_context
*brw
,
1299 __DRIdrawable
*drawable
,
1300 __DRIbuffer
**buffers
,
1303 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1304 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1306 unsigned attachments
[8];
1308 struct intel_renderbuffer
*front_rb
;
1309 struct intel_renderbuffer
*back_rb
;
1311 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1312 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1314 memset(attachments
, 0, sizeof(attachments
));
1315 if ((brw_is_front_buffer_drawing(fb
) ||
1316 brw_is_front_buffer_reading(fb
) ||
1317 !back_rb
) && front_rb
) {
1318 /* If a fake front buffer is in use, then querying for
1319 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1320 * the real front buffer to the fake front buffer. So before doing the
1321 * query, we need to make sure all the pending drawing has landed in the
1322 * real front buffer.
1324 intel_batchbuffer_flush(brw
);
1325 intel_flush_front(&brw
->ctx
);
1327 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1328 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1329 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1330 /* We have pending front buffer rendering, but we aren't querying for a
1331 * front buffer. If the front buffer we have is a fake front buffer,
1332 * the X server is going to throw it away when it processes the query.
1333 * So before doing the query, make sure all the pending drawing has
1334 * landed in the real front buffer.
1336 intel_batchbuffer_flush(brw
);
1337 intel_flush_front(&brw
->ctx
);
1341 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1342 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1345 assert(i
<= ARRAY_SIZE(attachments
));
1347 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1352 drawable
->loaderPrivate
);
1356 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1358 * This is called from intel_update_renderbuffers().
1361 * DRI buffers whose attachment point is DRI2BufferStencil or
1362 * DRI2BufferDepthStencil are handled as special cases.
1364 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1365 * that is passed to drm_intel_bo_gem_create_from_name().
1367 * \see intel_update_renderbuffers()
1370 intel_process_dri2_buffer(struct brw_context
*brw
,
1371 __DRIdrawable
*drawable
,
1372 __DRIbuffer
*buffer
,
1373 struct intel_renderbuffer
*rb
,
1374 const char *buffer_name
)
1376 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1382 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1384 /* We try to avoid closing and reopening the same BO name, because the first
1385 * use of a mapping of the buffer involves a bunch of page faulting which is
1386 * moderately expensive.
1388 struct intel_mipmap_tree
*last_mt
;
1389 if (num_samples
== 0)
1392 last_mt
= rb
->singlesample_mt
;
1394 uint32_t old_name
= 0;
1396 /* The bo already has a name because the miptree was created by a
1397 * previous call to intel_process_dri2_buffer(). If a bo already has a
1398 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1399 * create a new name.
1401 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1404 if (old_name
== buffer
->name
)
1407 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1409 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1410 buffer
->name
, buffer
->attachment
,
1411 buffer
->cpp
, buffer
->pitch
);
1414 intel_miptree_release(&rb
->mt
);
1415 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1419 "Failed to open BO for returned DRI2 buffer "
1420 "(%dx%d, %s, named %d).\n"
1421 "This is likely a bug in the X Server that will lead to a "
1423 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1427 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1428 drawable
->w
, drawable
->h
,
1431 if (brw_is_front_buffer_drawing(fb
) &&
1432 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1433 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1434 rb
->Base
.Base
.NumSamples
> 1) {
1435 intel_renderbuffer_upsample(brw
, rb
);
1440 drm_intel_bo_unreference(bo
);
1444 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1446 * To determine which DRI buffers to request, examine the renderbuffers
1447 * attached to the drawable's framebuffer. Then request the buffers from
1450 * This is called from intel_update_renderbuffers().
1452 * \param drawable Drawable whose buffers are queried.
1453 * \param buffers [out] List of buffers returned by DRI2 query.
1454 * \param buffer_count [out] Number of buffers returned.
1456 * \see intel_update_renderbuffers()
1460 intel_update_image_buffer(struct brw_context
*intel
,
1461 __DRIdrawable
*drawable
,
1462 struct intel_renderbuffer
*rb
,
1464 enum __DRIimageBufferMask buffer_type
)
1466 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1468 if (!rb
|| !buffer
->bo
)
1471 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1473 /* Check and see if we're already bound to the right
1476 struct intel_mipmap_tree
*last_mt
;
1477 if (num_samples
== 0)
1480 last_mt
= rb
->singlesample_mt
;
1482 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1485 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1486 buffer
->width
, buffer
->height
,
1489 if (brw_is_front_buffer_drawing(fb
) &&
1490 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1491 rb
->Base
.Base
.NumSamples
> 1) {
1492 intel_renderbuffer_upsample(intel
, rb
);
1497 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1499 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1500 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1501 struct intel_renderbuffer
*front_rb
;
1502 struct intel_renderbuffer
*back_rb
;
1503 struct __DRIimageList images
;
1504 unsigned int format
;
1505 uint32_t buffer_mask
= 0;
1507 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1508 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1511 format
= intel_rb_format(back_rb
);
1513 format
= intel_rb_format(front_rb
);
1517 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1518 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1519 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1523 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1525 (*screen
->image
.loader
->getBuffers
) (drawable
,
1526 driGLFormatToImageFormat(format
),
1527 &drawable
->dri2
.stamp
,
1528 drawable
->loaderPrivate
,
1532 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1533 drawable
->w
= images
.front
->width
;
1534 drawable
->h
= images
.front
->height
;
1535 intel_update_image_buffer(brw
,
1539 __DRI_IMAGE_BUFFER_FRONT
);
1541 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1542 drawable
->w
= images
.back
->width
;
1543 drawable
->h
= images
.back
->height
;
1544 intel_update_image_buffer(brw
,
1548 __DRI_IMAGE_BUFFER_BACK
);