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"
53 #include "brw_shader.h"
55 #include "brw_state.h"
57 #include "intel_batchbuffer.h"
58 #include "intel_buffer_objects.h"
59 #include "intel_buffers.h"
60 #include "intel_fbo.h"
61 #include "intel_mipmap_tree.h"
62 #include "intel_pixel.h"
63 #include "intel_image.h"
64 #include "intel_tex.h"
65 #include "intel_tex_obj.h"
67 #include "swrast_setup/swrast_setup.h"
69 #include "tnl/t_pipeline.h"
70 #include "util/ralloc.h"
72 /***************************************
73 * Mesa's Driver Functions
74 ***************************************/
77 brw_query_samples_for_format(struct gl_context
*ctx
, GLenum target
,
78 GLenum internalFormat
, int samples
[16])
80 struct brw_context
*brw
= brw_context(ctx
);
102 assert(brw
->gen
< 6);
108 const char *const brw_vendor_string
= "Intel Open Source Technology Center";
111 brw_get_renderer_string(unsigned deviceID
)
114 static char buffer
[128];
118 #define CHIPSET(id, symbol, str) case id: chipset = str; break;
119 #include "pci_ids/i965_pci_ids.h"
121 chipset
= "Unknown Intel Chipset";
125 (void) driGetRendererString(buffer
, chipset
, 0);
129 static const GLubyte
*
130 intel_get_string(struct gl_context
* ctx
, GLenum name
)
132 const struct brw_context
*const brw
= brw_context(ctx
);
136 return (GLubyte
*) brw_vendor_string
;
140 (GLubyte
*) brw_get_renderer_string(brw
->intelScreen
->deviceID
);
148 intel_viewport(struct gl_context
*ctx
)
150 struct brw_context
*brw
= brw_context(ctx
);
151 __DRIcontext
*driContext
= brw
->driContext
;
153 if (_mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
154 dri2InvalidateDrawable(driContext
->driDrawablePriv
);
155 dri2InvalidateDrawable(driContext
->driReadablePriv
);
160 intel_update_state(struct gl_context
* ctx
, GLuint new_state
)
162 struct brw_context
*brw
= brw_context(ctx
);
163 struct intel_texture_object
*tex_obj
;
164 struct intel_renderbuffer
*depth_irb
;
166 if (ctx
->swrast_context
)
167 _swrast_InvalidateState(ctx
, new_state
);
168 _vbo_InvalidateState(ctx
, new_state
);
170 brw
->NewGLState
|= new_state
;
172 _mesa_unlock_context_textures(ctx
);
174 /* Resolve the depth buffer's HiZ buffer. */
175 depth_irb
= intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
);
177 intel_renderbuffer_resolve_hiz(brw
, depth_irb
);
179 /* Resolve depth buffer and render cache of each enabled texture. */
180 int maxEnabledUnit
= ctx
->Texture
._MaxEnabledTexImageUnit
;
181 for (int i
= 0; i
<= maxEnabledUnit
; i
++) {
182 if (!ctx
->Texture
.Unit
[i
]._Current
)
184 tex_obj
= intel_texture_object(ctx
->Texture
.Unit
[i
]._Current
);
185 if (!tex_obj
|| !tex_obj
->mt
)
187 intel_miptree_all_slices_resolve_depth(brw
, tex_obj
->mt
);
188 intel_miptree_resolve_color(brw
, tex_obj
->mt
);
189 brw_render_cache_set_check_flush(brw
, tex_obj
->mt
->bo
);
192 _mesa_lock_context_textures(ctx
);
195 #define flushFront(screen) ((screen)->image.loader ? (screen)->image.loader->flushFrontBuffer : (screen)->dri2.loader->flushFrontBuffer)
198 intel_flush_front(struct gl_context
*ctx
)
200 struct brw_context
*brw
= brw_context(ctx
);
201 __DRIcontext
*driContext
= brw
->driContext
;
202 __DRIdrawable
*driDrawable
= driContext
->driDrawablePriv
;
203 __DRIscreen
*const screen
= brw
->intelScreen
->driScrnPriv
;
205 if (brw
->front_buffer_dirty
&& _mesa_is_winsys_fbo(ctx
->DrawBuffer
)) {
206 if (flushFront(screen
) && driDrawable
&&
207 driDrawable
->loaderPrivate
) {
209 /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
211 * This potentially resolves both front and back buffer. It
212 * is unnecessary to resolve the back, but harms nothing except
213 * performance. And no one cares about front-buffer render
216 intel_resolve_for_dri2_flush(brw
, driDrawable
);
217 intel_batchbuffer_flush(brw
);
219 flushFront(screen
)(driDrawable
, driDrawable
->loaderPrivate
);
221 /* We set the dirty bit in intel_prepare_render() if we're
222 * front buffer rendering once we get there.
224 brw
->front_buffer_dirty
= false;
230 intel_glFlush(struct gl_context
*ctx
)
232 struct brw_context
*brw
= brw_context(ctx
);
234 intel_batchbuffer_flush(brw
);
235 intel_flush_front(ctx
);
237 brw
->need_flush_throttle
= true;
241 intel_finish(struct gl_context
* ctx
)
243 struct brw_context
*brw
= brw_context(ctx
);
247 if (brw
->batch
.last_bo
)
248 drm_intel_bo_wait_rendering(brw
->batch
.last_bo
);
252 brw_init_driver_functions(struct brw_context
*brw
,
253 struct dd_function_table
*functions
)
255 _mesa_init_driver_functions(functions
);
257 /* GLX uses DRI2 invalidate events to handle window resizing.
258 * Unfortunately, EGL does not - libEGL is written in XCB (not Xlib),
259 * which doesn't provide a mechanism for snooping the event queues.
261 * So EGL still relies on viewport hacks to handle window resizing.
262 * This should go away with DRI3000.
264 if (!brw
->driContext
->driScreenPriv
->dri2
.useInvalidate
)
265 functions
->Viewport
= intel_viewport
;
267 functions
->Flush
= intel_glFlush
;
268 functions
->Finish
= intel_finish
;
269 functions
->GetString
= intel_get_string
;
270 functions
->UpdateState
= intel_update_state
;
272 intelInitTextureFuncs(functions
);
273 intelInitTextureImageFuncs(functions
);
274 intelInitTextureSubImageFuncs(functions
);
275 intelInitTextureCopyImageFuncs(functions
);
276 intelInitCopyImageFuncs(functions
);
277 intelInitClearFuncs(functions
);
278 intelInitBufferFuncs(functions
);
279 intelInitPixelFuncs(functions
);
280 intelInitBufferObjectFuncs(functions
);
281 intel_init_syncobj_functions(functions
);
282 brw_init_object_purgeable_functions(functions
);
284 brwInitFragProgFuncs( functions
);
285 brw_init_common_queryobj_functions(functions
);
287 gen6_init_queryobj_functions(functions
);
289 gen4_init_queryobj_functions(functions
);
290 brw_init_compute_functions(functions
);
292 brw_init_conditional_render_functions(functions
);
294 functions
->QuerySamplesForFormat
= brw_query_samples_for_format
;
296 functions
->NewTransformFeedback
= brw_new_transform_feedback
;
297 functions
->DeleteTransformFeedback
= brw_delete_transform_feedback
;
298 functions
->GetTransformFeedbackVertexCount
=
299 brw_get_transform_feedback_vertex_count
;
301 functions
->BeginTransformFeedback
= gen7_begin_transform_feedback
;
302 functions
->EndTransformFeedback
= gen7_end_transform_feedback
;
303 functions
->PauseTransformFeedback
= gen7_pause_transform_feedback
;
304 functions
->ResumeTransformFeedback
= gen7_resume_transform_feedback
;
306 functions
->BeginTransformFeedback
= brw_begin_transform_feedback
;
307 functions
->EndTransformFeedback
= brw_end_transform_feedback
;
311 functions
->GetSamplePosition
= gen6_get_sample_position
;
315 brw_initialize_context_constants(struct brw_context
*brw
)
317 struct gl_context
*ctx
= &brw
->ctx
;
319 unsigned max_samplers
=
320 brw
->gen
>= 8 || brw
->is_haswell
? BRW_MAX_TEX_UNIT
: 16;
322 ctx
->Const
.QueryCounterBits
.Timestamp
= 36;
324 ctx
->Const
.StripTextureBorder
= true;
326 ctx
->Const
.MaxUniformBlockSize
= 65536;
327 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
328 struct gl_program_constants
*prog
= &ctx
->Const
.Program
[i
];
329 prog
->MaxUniformBlocks
= 12;
330 prog
->MaxCombinedUniformComponents
=
331 prog
->MaxUniformComponents
+
332 ctx
->Const
.MaxUniformBlockSize
/ 4 * prog
->MaxUniformBlocks
;
335 ctx
->Const
.MaxDualSourceDrawBuffers
= 1;
336 ctx
->Const
.MaxDrawBuffers
= BRW_MAX_DRAW_BUFFERS
;
337 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
= max_samplers
;
338 ctx
->Const
.MaxTextureCoordUnits
= 8; /* Mesa limit */
339 ctx
->Const
.MaxTextureUnits
=
340 MIN2(ctx
->Const
.MaxTextureCoordUnits
,
341 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
);
342 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
= max_samplers
;
344 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= max_samplers
;
346 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
= 0;
347 if (_mesa_extension_override_enables
.ARB_compute_shader
) {
348 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= BRW_MAX_TEX_UNIT
;
349 ctx
->Const
.MaxUniformBufferBindings
+= 12;
351 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
= 0;
353 ctx
->Const
.MaxCombinedTextureImageUnits
=
354 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTextureImageUnits
+
355 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
+
356 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxTextureImageUnits
+
357 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
359 ctx
->Const
.MaxTextureLevels
= 14; /* 8192 */
360 if (ctx
->Const
.MaxTextureLevels
> MAX_TEXTURE_LEVELS
)
361 ctx
->Const
.MaxTextureLevels
= MAX_TEXTURE_LEVELS
;
362 ctx
->Const
.Max3DTextureLevels
= 12; /* 2048 */
363 ctx
->Const
.MaxCubeTextureLevels
= 14; /* 8192 */
364 ctx
->Const
.MaxTextureMbytes
= 1536;
367 ctx
->Const
.MaxArrayTextureLayers
= 2048;
369 ctx
->Const
.MaxArrayTextureLayers
= 512;
371 ctx
->Const
.MaxTextureRectSize
= 1 << 12;
373 ctx
->Const
.MaxTextureMaxAnisotropy
= 16.0;
375 ctx
->Const
.MaxRenderbufferSize
= 8192;
377 /* Hardware only supports a limited number of transform feedback buffers.
378 * So we need to override the Mesa default (which is based only on software
381 ctx
->Const
.MaxTransformFeedbackBuffers
= BRW_MAX_SOL_BUFFERS
;
383 /* On Gen6, in the worst case, we use up one binding table entry per
384 * transform feedback component (see comments above the definition of
385 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
386 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
387 * BRW_MAX_SOL_BINDINGS.
389 * In "separate components" mode, we need to divide this value by
390 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
391 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
393 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
= BRW_MAX_SOL_BINDINGS
;
394 ctx
->Const
.MaxTransformFeedbackSeparateComponents
=
395 BRW_MAX_SOL_BINDINGS
/ BRW_MAX_SOL_BUFFERS
;
397 ctx
->Const
.AlwaysUseGetTransformFeedbackVertexCount
= true;
400 const int *msaa_modes
= intel_supported_msaa_modes(brw
->intelScreen
);
401 const int clamp_max_samples
=
402 driQueryOptioni(&brw
->optionCache
, "clamp_max_samples");
404 if (clamp_max_samples
< 0) {
405 max_samples
= msaa_modes
[0];
407 /* Select the largest supported MSAA mode that does not exceed
411 for (int i
= 0; msaa_modes
[i
] != 0; ++i
) {
412 if (msaa_modes
[i
] <= clamp_max_samples
) {
413 max_samples
= msaa_modes
[i
];
419 ctx
->Const
.MaxSamples
= max_samples
;
420 ctx
->Const
.MaxColorTextureSamples
= max_samples
;
421 ctx
->Const
.MaxDepthTextureSamples
= max_samples
;
422 ctx
->Const
.MaxIntegerSamples
= max_samples
;
424 /* gen6_set_sample_maps() sets SampleMap{2,4,8}x variables which are used
425 * to map indices of rectangular grid to sample numbers within a pixel.
426 * These variables are used by GL_EXT_framebuffer_multisample_blit_scaled
427 * extension implementation. For more details see the comment above
428 * gen6_set_sample_maps() definition.
430 gen6_set_sample_maps(ctx
);
433 ctx
->Const
.MaxProgramTextureGatherComponents
= 4;
434 else if (brw
->gen
== 6)
435 ctx
->Const
.MaxProgramTextureGatherComponents
= 1;
437 ctx
->Const
.MinLineWidth
= 1.0;
438 ctx
->Const
.MinLineWidthAA
= 1.0;
440 ctx
->Const
.MaxLineWidth
= 7.375;
441 ctx
->Const
.MaxLineWidthAA
= 7.375;
442 ctx
->Const
.LineWidthGranularity
= 0.125;
444 ctx
->Const
.MaxLineWidth
= 7.0;
445 ctx
->Const
.MaxLineWidthAA
= 7.0;
446 ctx
->Const
.LineWidthGranularity
= 0.5;
449 /* For non-antialiased lines, we have to round the line width to the
450 * nearest whole number. Make sure that we don't advertise a line
451 * width that, when rounded, will be beyond the actual hardware
454 assert(roundf(ctx
->Const
.MaxLineWidth
) <= ctx
->Const
.MaxLineWidth
);
456 ctx
->Const
.MinPointSize
= 1.0;
457 ctx
->Const
.MinPointSizeAA
= 1.0;
458 ctx
->Const
.MaxPointSize
= 255.0;
459 ctx
->Const
.MaxPointSizeAA
= 255.0;
460 ctx
->Const
.PointSizeGranularity
= 1.0;
462 if (brw
->gen
>= 5 || brw
->is_g4x
)
463 ctx
->Const
.MaxClipPlanes
= 8;
465 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeInstructions
= 16 * 1024;
466 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAluInstructions
= 0;
467 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexInstructions
= 0;
468 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTexIndirections
= 0;
469 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAluInstructions
= 0;
470 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexInstructions
= 0;
471 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTexIndirections
= 0;
472 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAttribs
= 16;
473 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeTemps
= 256;
474 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeAddressRegs
= 1;
475 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
= 1024;
476 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
=
477 MIN2(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxNativeParameters
,
478 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxEnvParams
);
480 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeInstructions
= 1024;
481 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAluInstructions
= 1024;
482 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexInstructions
= 1024;
483 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTexIndirections
= 1024;
484 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAttribs
= 12;
485 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeTemps
= 256;
486 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeAddressRegs
= 0;
487 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
= 1024;
488 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
=
489 MIN2(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxNativeParameters
,
490 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxEnvParams
);
492 /* Fragment shaders use real, 32-bit twos-complement integers for all
495 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMin
= 31;
496 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.RangeMax
= 30;
497 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
.Precision
= 0;
498 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
499 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].LowInt
;
501 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMin
= 31;
502 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.RangeMax
= 30;
503 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
.Precision
= 0;
504 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].HighInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
505 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MediumInt
= ctx
->Const
.Program
[MESA_SHADER_VERTEX
].LowInt
;
508 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
509 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
510 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
511 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicCounters
= MAX_ATOMIC_COUNTERS
;
512 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAtomicBuffers
= BRW_MAX_ABO
;
513 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAtomicBuffers
= BRW_MAX_ABO
;
514 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxAtomicBuffers
= BRW_MAX_ABO
;
515 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxAtomicBuffers
= BRW_MAX_ABO
;
516 ctx
->Const
.MaxCombinedAtomicBuffers
= 3 * BRW_MAX_ABO
;
518 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxImageUniforms
=
520 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxImageUniforms
=
521 (brw
->intelScreen
->compiler
->scalar_vs
? BRW_MAX_IMAGES
: 0);
522 ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxImageUniforms
=
524 ctx
->Const
.MaxImageUnits
= MAX_IMAGE_UNITS
;
525 ctx
->Const
.MaxCombinedShaderOutputResources
=
526 MAX_IMAGE_UNITS
+ BRW_MAX_DRAW_BUFFERS
;
527 ctx
->Const
.MaxImageSamples
= 0;
528 ctx
->Const
.MaxCombinedImageUniforms
= 3 * BRW_MAX_IMAGES
;
531 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
532 * but we're not sure how it's actually done for vertex order,
533 * that affect provoking vertex decision. Always use last vertex
534 * convention for quad primitive which works as expected for now.
537 ctx
->Const
.QuadsFollowProvokingVertexConvention
= false;
539 ctx
->Const
.NativeIntegers
= true;
540 ctx
->Const
.VertexID_is_zero_based
= true;
542 /* Regarding the CMP instruction, the Ivybridge PRM says:
544 * "For each enabled channel 0b or 1b is assigned to the appropriate flag
545 * bit and 0/all zeros or all ones (e.g, byte 0xFF, word 0xFFFF, DWord
546 * 0xFFFFFFFF) is assigned to dst."
548 * but PRMs for earlier generations say
550 * "In dword format, one GRF may store up to 8 results. When the register
551 * is used later as a vector of Booleans, as only LSB at each channel
552 * contains meaning [sic] data, software should make sure all higher bits
553 * are masked out (e.g. by 'and-ing' an [sic] 0x01 constant)."
555 * We select the representation of a true boolean uniform to be ~0, and fix
556 * the results of Gen <= 5 CMP instruction's with -(result & 1).
558 ctx
->Const
.UniformBooleanTrue
= ~0;
560 /* From the gen4 PRM, volume 4 page 127:
562 * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
563 * the base address of the first element of the surface, computed in
564 * software by adding the surface base address to the byte offset of
565 * the element in the buffer."
567 * However, unaligned accesses are slower, so enforce buffer alignment.
569 ctx
->Const
.UniformBufferOffsetAlignment
= 16;
570 ctx
->Const
.TextureBufferOffsetAlignment
= 16;
571 ctx
->Const
.MaxTextureBufferSize
= 128 * 1024 * 1024;
574 ctx
->Const
.MaxVarying
= 32;
575 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxOutputComponents
= 128;
576 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxInputComponents
= 64;
577 ctx
->Const
.Program
[MESA_SHADER_GEOMETRY
].MaxOutputComponents
= 128;
578 ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxInputComponents
= 128;
581 /* We want the GLSL compiler to emit code that uses condition codes */
582 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
583 ctx
->Const
.ShaderCompilerOptions
[i
] =
584 brw
->intelScreen
->compiler
->glsl_compiler_options
[i
];
587 /* ARB_viewport_array */
588 if (brw
->gen
>= 6 && ctx
->API
== API_OPENGL_CORE
) {
589 ctx
->Const
.MaxViewports
= GEN6_NUM_VIEWPORTS
;
590 ctx
->Const
.ViewportSubpixelBits
= 0;
592 /* Cast to float before negating because MaxViewportWidth is unsigned.
594 ctx
->Const
.ViewportBounds
.Min
= -(float)ctx
->Const
.MaxViewportWidth
;
595 ctx
->Const
.ViewportBounds
.Max
= ctx
->Const
.MaxViewportWidth
;
598 /* ARB_gpu_shader5 */
600 ctx
->Const
.MaxVertexStreams
= MIN2(4, MAX_VERTEX_STREAMS
);
602 /* ARB_framebuffer_no_attachments */
603 ctx
->Const
.MaxFramebufferWidth
= ctx
->Const
.MaxViewportWidth
;
604 ctx
->Const
.MaxFramebufferHeight
= ctx
->Const
.MaxViewportHeight
;
605 ctx
->Const
.MaxFramebufferLayers
= ctx
->Const
.MaxArrayTextureLayers
;
606 ctx
->Const
.MaxFramebufferSamples
= max_samples
;
610 brw_adjust_cs_context_constants(struct brw_context
*brw
)
612 struct gl_context
*ctx
= &brw
->ctx
;
614 /* For ES, we set these constants based on SIMD8.
616 * TODO: Once we can always generate SIMD16, we should update this.
618 * For GL, we assume we can generate a SIMD16 program, but this currently
619 * is not always true. This allows us to run more test cases, and will be
620 * required based on desktop GL compute shader requirements.
622 const int simd_size
= ctx
->API
== API_OPENGL_CORE
? 16 : 8;
624 const uint32_t max_invocations
= simd_size
* brw
->max_cs_threads
;
625 ctx
->Const
.MaxComputeWorkGroupSize
[0] = max_invocations
;
626 ctx
->Const
.MaxComputeWorkGroupSize
[1] = max_invocations
;
627 ctx
->Const
.MaxComputeWorkGroupSize
[2] = max_invocations
;
628 ctx
->Const
.MaxComputeWorkGroupInvocations
= max_invocations
;
632 * Process driconf (drirc) options, setting appropriate context flags.
634 * intelInitExtensions still pokes at optionCache directly, in order to
635 * avoid advertising various extensions. No flags are set, so it makes
636 * sense to continue doing that there.
639 brw_process_driconf_options(struct brw_context
*brw
)
641 struct gl_context
*ctx
= &brw
->ctx
;
643 driOptionCache
*options
= &brw
->optionCache
;
644 driParseConfigFiles(options
, &brw
->intelScreen
->optionCache
,
645 brw
->driContext
->driScreenPriv
->myNum
, "i965");
647 int bo_reuse_mode
= driQueryOptioni(options
, "bo_reuse");
648 switch (bo_reuse_mode
) {
649 case DRI_CONF_BO_REUSE_DISABLED
:
651 case DRI_CONF_BO_REUSE_ALL
:
652 intel_bufmgr_gem_enable_reuse(brw
->bufmgr
);
656 if (!driQueryOptionb(options
, "hiz")) {
657 brw
->has_hiz
= false;
658 /* On gen6, you can only do separate stencil with HIZ. */
660 brw
->has_separate_stencil
= false;
663 if (driQueryOptionb(options
, "always_flush_batch")) {
664 fprintf(stderr
, "flushing batchbuffer before/after each draw call\n");
665 brw
->always_flush_batch
= true;
668 if (driQueryOptionb(options
, "always_flush_cache")) {
669 fprintf(stderr
, "flushing GPU caches before/after each draw call\n");
670 brw
->always_flush_cache
= true;
673 if (driQueryOptionb(options
, "disable_throttling")) {
674 fprintf(stderr
, "disabling flush throttling\n");
675 brw
->disable_throttling
= true;
678 brw
->precompile
= driQueryOptionb(&brw
->optionCache
, "shader_precompile");
680 ctx
->Const
.ForceGLSLExtensionsWarn
=
681 driQueryOptionb(options
, "force_glsl_extensions_warn");
683 ctx
->Const
.DisableGLSLLineContinuations
=
684 driQueryOptionb(options
, "disable_glsl_line_continuations");
686 ctx
->Const
.AllowGLSLExtensionDirectiveMidShader
=
687 driQueryOptionb(options
, "allow_glsl_extension_directive_midshader");
691 brwCreateContext(gl_api api
,
692 const struct gl_config
*mesaVis
,
693 __DRIcontext
*driContextPriv
,
694 unsigned major_version
,
695 unsigned minor_version
,
698 unsigned *dri_ctx_error
,
699 void *sharedContextPrivate
)
701 __DRIscreen
*sPriv
= driContextPriv
->driScreenPriv
;
702 struct gl_context
*shareCtx
= (struct gl_context
*) sharedContextPrivate
;
703 struct intel_screen
*screen
= sPriv
->driverPrivate
;
704 const struct brw_device_info
*devinfo
= screen
->devinfo
;
705 struct dd_function_table functions
;
707 /* Only allow the __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag if the kernel
708 * provides us with context reset notifications.
710 uint32_t allowed_flags
= __DRI_CTX_FLAG_DEBUG
711 | __DRI_CTX_FLAG_FORWARD_COMPATIBLE
;
713 if (screen
->has_context_reset_notification
)
714 allowed_flags
|= __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
;
716 if (flags
& ~allowed_flags
) {
717 *dri_ctx_error
= __DRI_CTX_ERROR_UNKNOWN_FLAG
;
721 struct brw_context
*brw
= rzalloc(NULL
, struct brw_context
);
723 fprintf(stderr
, "%s: failed to alloc context\n", __func__
);
724 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
728 driContextPriv
->driverPrivate
= brw
;
729 brw
->driContext
= driContextPriv
;
730 brw
->intelScreen
= screen
;
731 brw
->bufmgr
= screen
->bufmgr
;
733 brw
->gen
= devinfo
->gen
;
734 brw
->gt
= devinfo
->gt
;
735 brw
->is_g4x
= devinfo
->is_g4x
;
736 brw
->is_baytrail
= devinfo
->is_baytrail
;
737 brw
->is_haswell
= devinfo
->is_haswell
;
738 brw
->is_cherryview
= devinfo
->is_cherryview
;
739 brw
->is_broxton
= devinfo
->is_broxton
;
740 brw
->has_llc
= devinfo
->has_llc
;
741 brw
->has_hiz
= devinfo
->has_hiz_and_separate_stencil
;
742 brw
->has_separate_stencil
= devinfo
->has_hiz_and_separate_stencil
;
743 brw
->has_pln
= devinfo
->has_pln
;
744 brw
->has_compr4
= devinfo
->has_compr4
;
745 brw
->has_surface_tile_offset
= devinfo
->has_surface_tile_offset
;
746 brw
->has_negative_rhw_bug
= devinfo
->has_negative_rhw_bug
;
747 brw
->needs_unlit_centroid_workaround
=
748 devinfo
->needs_unlit_centroid_workaround
;
750 brw
->must_use_separate_stencil
= screen
->hw_must_use_separate_stencil
;
751 brw
->has_swizzling
= screen
->hw_has_swizzling
;
753 brw
->vs
.base
.stage
= MESA_SHADER_VERTEX
;
754 brw
->gs
.base
.stage
= MESA_SHADER_GEOMETRY
;
755 brw
->wm
.base
.stage
= MESA_SHADER_FRAGMENT
;
757 gen8_init_vtable_surface_functions(brw
);
758 brw
->vtbl
.emit_depth_stencil_hiz
= gen8_emit_depth_stencil_hiz
;
759 } else if (brw
->gen
>= 7) {
760 gen7_init_vtable_surface_functions(brw
);
761 brw
->vtbl
.emit_depth_stencil_hiz
= gen7_emit_depth_stencil_hiz
;
762 } else if (brw
->gen
>= 6) {
763 gen6_init_vtable_surface_functions(brw
);
764 brw
->vtbl
.emit_depth_stencil_hiz
= gen6_emit_depth_stencil_hiz
;
766 gen4_init_vtable_surface_functions(brw
);
767 brw
->vtbl
.emit_depth_stencil_hiz
= brw_emit_depth_stencil_hiz
;
770 brw_init_driver_functions(brw
, &functions
);
773 functions
.GetGraphicsResetStatus
= brw_get_graphics_reset_status
;
775 struct gl_context
*ctx
= &brw
->ctx
;
777 if (!_mesa_initialize_context(ctx
, api
, mesaVis
, shareCtx
, &functions
)) {
778 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
779 fprintf(stderr
, "%s: failed to init mesa context\n", __func__
);
780 intelDestroyContext(driContextPriv
);
784 driContextSetFlags(ctx
, flags
);
786 /* Initialize the software rasterizer and helper modules.
788 * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
789 * software fallbacks (which we have to support on legacy GL to do weird
790 * glDrawPixels(), glBitmap(), and other functions).
792 if (api
!= API_OPENGL_CORE
&& api
!= API_OPENGLES2
) {
793 _swrast_CreateContext(ctx
);
796 _vbo_CreateContext(ctx
);
797 if (ctx
->swrast_context
) {
798 _tnl_CreateContext(ctx
);
799 TNL_CONTEXT(ctx
)->Driver
.RunPipeline
= _tnl_run_pipeline
;
800 _swsetup_CreateContext(ctx
);
802 /* Configure swrast to match hardware characteristics: */
803 _swrast_allow_pixel_fog(ctx
, false);
804 _swrast_allow_vertex_fog(ctx
, true);
807 _mesa_meta_init(ctx
);
809 brw_process_driconf_options(brw
);
811 if (INTEL_DEBUG
& DEBUG_PERF
)
812 brw
->perf_debug
= true;
814 brw_initialize_context_constants(brw
);
816 ctx
->Const
.ResetStrategy
= notify_reset
817 ? GL_LOSE_CONTEXT_ON_RESET_ARB
: GL_NO_RESET_NOTIFICATION_ARB
;
819 /* Reinitialize the context point state. It depends on ctx->Const values. */
820 _mesa_init_point(ctx
);
824 intel_batchbuffer_init(brw
);
827 /* Create a new hardware context. Using a hardware context means that
828 * our GPU state will be saved/restored on context switch, allowing us
829 * to assume that the GPU is in the same state we left it in.
831 * This is required for transform feedback buffer offsets, query objects,
832 * and also allows us to reduce how much state we have to emit.
834 brw
->hw_ctx
= drm_intel_gem_context_create(brw
->bufmgr
);
837 fprintf(stderr
, "Gen6+ requires Kernel 3.6 or later.\n");
838 intelDestroyContext(driContextPriv
);
843 if (brw_init_pipe_control(brw
, devinfo
)) {
844 *dri_ctx_error
= __DRI_CTX_ERROR_NO_MEMORY
;
845 intelDestroyContext(driContextPriv
);
851 intelInitExtensions(ctx
);
853 brw_init_surface_formats(brw
);
855 brw
->max_vs_threads
= devinfo
->max_vs_threads
;
856 brw
->max_hs_threads
= devinfo
->max_hs_threads
;
857 brw
->max_ds_threads
= devinfo
->max_ds_threads
;
858 brw
->max_gs_threads
= devinfo
->max_gs_threads
;
859 brw
->max_wm_threads
= devinfo
->max_wm_threads
;
860 brw
->max_cs_threads
= devinfo
->max_cs_threads
;
861 brw
->urb
.size
= devinfo
->urb
.size
;
862 brw
->urb
.min_vs_entries
= devinfo
->urb
.min_vs_entries
;
863 brw
->urb
.max_vs_entries
= devinfo
->urb
.max_vs_entries
;
864 brw
->urb
.max_hs_entries
= devinfo
->urb
.max_hs_entries
;
865 brw
->urb
.max_ds_entries
= devinfo
->urb
.max_ds_entries
;
866 brw
->urb
.max_gs_entries
= devinfo
->urb
.max_gs_entries
;
868 brw_adjust_cs_context_constants(brw
);
870 /* Estimate the size of the mappable aperture into the GTT. There's an
871 * ioctl to get the whole GTT size, but not one to get the mappable subset.
872 * It turns out it's basically always 256MB, though some ancient hardware
875 uint32_t gtt_size
= 256 * 1024 * 1024;
877 /* We don't want to map two objects such that a memcpy between them would
878 * just fault one mapping in and then the other over and over forever. So
879 * we would need to divide the GTT size by 2. Additionally, some GTT is
880 * taken up by things like the framebuffer and the ringbuffer and such, so
881 * be more conservative.
883 brw
->max_gtt_map_object_size
= gtt_size
/ 4;
886 brw
->urb
.gs_present
= false;
888 brw
->prim_restart
.in_progress
= false;
889 brw
->prim_restart
.enable_cut_index
= false;
890 brw
->gs
.enabled
= false;
891 brw
->sf
.viewport_transform_enable
= true;
893 brw
->predicate
.state
= BRW_PREDICATE_STATE_RENDER
;
895 brw
->use_resource_streamer
= screen
->has_resource_streamer
&&
896 (brw_env_var_as_boolean("INTEL_USE_HW_BT", false) ||
897 brw_env_var_as_boolean("INTEL_USE_GATHER", false));
899 ctx
->VertexProgram
._MaintainTnlProgram
= true;
900 ctx
->FragmentProgram
._MaintainTexEnvProgram
= true;
902 brw_draw_init( brw
);
904 if ((flags
& __DRI_CTX_FLAG_DEBUG
) != 0) {
905 /* Turn on some extra GL_ARB_debug_output generation. */
906 brw
->perf_debug
= true;
909 if ((flags
& __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS
) != 0)
910 ctx
->Const
.ContextFlags
|= GL_CONTEXT_FLAG_ROBUST_ACCESS_BIT_ARB
;
912 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
913 brw_init_shader_time(brw
);
915 _mesa_compute_version(ctx
);
917 _mesa_initialize_dispatch_tables(ctx
);
918 _mesa_initialize_vbo_vtxfmt(ctx
);
920 if (ctx
->Extensions
.AMD_performance_monitor
) {
921 brw_init_performance_monitors(brw
);
924 vbo_use_buffer_objects(ctx
);
925 vbo_always_unmap_buffers(ctx
);
931 intelDestroyContext(__DRIcontext
* driContextPriv
)
933 struct brw_context
*brw
=
934 (struct brw_context
*) driContextPriv
->driverPrivate
;
935 struct gl_context
*ctx
= &brw
->ctx
;
937 /* Dump a final BMP in case the application doesn't call SwapBuffers */
938 if (INTEL_DEBUG
& DEBUG_AUB
) {
939 intel_batchbuffer_flush(brw
);
940 aub_dump_bmp(&brw
->ctx
);
943 _mesa_meta_free(&brw
->ctx
);
944 brw_meta_fast_clear_free(brw
);
946 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
947 /* Force a report. */
948 brw
->shader_time
.report_time
= 0;
950 brw_collect_and_report_shader_time(brw
);
951 brw_destroy_shader_time(brw
);
954 brw_destroy_state(brw
);
955 brw_draw_destroy(brw
);
957 drm_intel_bo_unreference(brw
->curbe
.curbe_bo
);
958 if (brw
->vs
.base
.scratch_bo
)
959 drm_intel_bo_unreference(brw
->vs
.base
.scratch_bo
);
960 if (brw
->gs
.base
.scratch_bo
)
961 drm_intel_bo_unreference(brw
->gs
.base
.scratch_bo
);
962 if (brw
->wm
.base
.scratch_bo
)
963 drm_intel_bo_unreference(brw
->wm
.base
.scratch_bo
);
965 gen7_reset_hw_bt_pool_offsets(brw
);
966 drm_intel_bo_unreference(brw
->hw_bt_pool
.bo
);
967 brw
->hw_bt_pool
.bo
= NULL
;
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 brw_fini_pipe_control(brw
);
981 intel_batchbuffer_free(brw
);
983 drm_intel_bo_unreference(brw
->throttle_batch
[1]);
984 drm_intel_bo_unreference(brw
->throttle_batch
[0]);
985 brw
->throttle_batch
[1] = NULL
;
986 brw
->throttle_batch
[0] = NULL
;
988 driDestroyOptionCache(&brw
->optionCache
);
990 /* free the Mesa context */
991 _mesa_free_context_data(&brw
->ctx
);
994 driContextPriv
->driverPrivate
= NULL
;
998 intelUnbindContext(__DRIcontext
* driContextPriv
)
1000 /* Unset current context and dispath table */
1001 _mesa_make_current(NULL
, NULL
, NULL
);
1007 * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
1008 * on window system framebuffers.
1010 * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
1011 * your renderbuffer can do sRGB encode, and you can flip a switch that does
1012 * sRGB encode if the renderbuffer can handle it. You can ask specifically
1013 * for a visual where you're guaranteed to be capable, but it turns out that
1014 * everyone just makes all their ARGB8888 visuals capable and doesn't offer
1015 * incapable ones, because there's no difference between the two in resources
1016 * used. Applications thus get built that accidentally rely on the default
1017 * visual choice being sRGB, so we make ours sRGB capable. Everything sounds
1020 * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
1021 * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
1022 * So they removed the enable knob and made it "if the renderbuffer is sRGB
1023 * capable, do sRGB encode". Then, for your window system renderbuffers, you
1024 * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
1025 * and get no sRGB encode (assuming that both kinds of visual are available).
1026 * Thus our choice to support sRGB by default on our visuals for desktop would
1027 * result in broken rendering of GLES apps that aren't expecting sRGB encode.
1029 * Unfortunately, renderbuffer setup happens before a context is created. So
1030 * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
1031 * context (without an sRGB visual, though we don't have sRGB visuals exposed
1032 * yet), we go turn that back off before anyone finds out.
1035 intel_gles3_srgb_workaround(struct brw_context
*brw
,
1036 struct gl_framebuffer
*fb
)
1038 struct gl_context
*ctx
= &brw
->ctx
;
1040 if (_mesa_is_desktop_gl(ctx
) || !fb
->Visual
.sRGBCapable
)
1043 /* Some day when we support the sRGB capable bit on visuals available for
1044 * GLES, we'll need to respect that and not disable things here.
1046 fb
->Visual
.sRGBCapable
= false;
1047 for (int i
= 0; i
< BUFFER_COUNT
; i
++) {
1048 if (fb
->Attachment
[i
].Renderbuffer
&&
1049 fb
->Attachment
[i
].Renderbuffer
->Format
== MESA_FORMAT_B8G8R8A8_SRGB
) {
1050 fb
->Attachment
[i
].Renderbuffer
->Format
= MESA_FORMAT_B8G8R8A8_UNORM
;
1056 intelMakeCurrent(__DRIcontext
* driContextPriv
,
1057 __DRIdrawable
* driDrawPriv
,
1058 __DRIdrawable
* driReadPriv
)
1060 struct brw_context
*brw
;
1061 GET_CURRENT_CONTEXT(curCtx
);
1064 brw
= (struct brw_context
*) driContextPriv
->driverPrivate
;
1068 /* According to the glXMakeCurrent() man page: "Pending commands to
1069 * the previous context, if any, are flushed before it is released."
1070 * But only flush if we're actually changing contexts.
1072 if (brw_context(curCtx
) && brw_context(curCtx
) != brw
) {
1073 _mesa_flush(curCtx
);
1076 if (driContextPriv
) {
1077 struct gl_context
*ctx
= &brw
->ctx
;
1078 struct gl_framebuffer
*fb
, *readFb
;
1080 if (driDrawPriv
== NULL
) {
1081 fb
= _mesa_get_incomplete_framebuffer();
1083 fb
= driDrawPriv
->driverPrivate
;
1084 driContextPriv
->dri2
.draw_stamp
= driDrawPriv
->dri2
.stamp
- 1;
1087 if (driReadPriv
== NULL
) {
1088 readFb
= _mesa_get_incomplete_framebuffer();
1090 readFb
= driReadPriv
->driverPrivate
;
1091 driContextPriv
->dri2
.read_stamp
= driReadPriv
->dri2
.stamp
- 1;
1094 /* The sRGB workaround changes the renderbuffer's format. We must change
1095 * the format before the renderbuffer's miptree get's allocated, otherwise
1096 * the formats of the renderbuffer and its miptree will differ.
1098 intel_gles3_srgb_workaround(brw
, fb
);
1099 intel_gles3_srgb_workaround(brw
, readFb
);
1101 /* If the context viewport hasn't been initialized, force a call out to
1102 * the loader to get buffers so we have a drawable size for the initial
1104 if (!brw
->ctx
.ViewportInitialized
)
1105 intel_prepare_render(brw
);
1107 _mesa_make_current(ctx
, fb
, readFb
);
1109 _mesa_make_current(NULL
, NULL
, NULL
);
1116 intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1117 __DRIdrawable
*drawable
)
1120 /* MSAA and fast color clear are not supported, so don't waste time
1121 * checking whether a resolve is needed.
1126 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1127 struct intel_renderbuffer
*rb
;
1129 /* Usually, only the back buffer will need to be downsampled. However,
1130 * the front buffer will also need it if the user has rendered into it.
1132 static const gl_buffer_index buffers
[2] = {
1137 for (int i
= 0; i
< 2; ++i
) {
1138 rb
= intel_get_renderbuffer(fb
, buffers
[i
]);
1139 if (rb
== NULL
|| rb
->mt
== NULL
)
1141 if (rb
->mt
->num_samples
<= 1)
1142 intel_miptree_resolve_color(brw
, rb
->mt
);
1144 intel_renderbuffer_downsample(brw
, rb
);
1149 intel_bits_per_pixel(const struct intel_renderbuffer
*rb
)
1151 return _mesa_get_format_bytes(intel_rb_format(rb
)) * 8;
1155 intel_query_dri2_buffers(struct brw_context
*brw
,
1156 __DRIdrawable
*drawable
,
1157 __DRIbuffer
**buffers
,
1161 intel_process_dri2_buffer(struct brw_context
*brw
,
1162 __DRIdrawable
*drawable
,
1163 __DRIbuffer
*buffer
,
1164 struct intel_renderbuffer
*rb
,
1165 const char *buffer_name
);
1168 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
);
1171 intel_update_dri2_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1173 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1174 struct intel_renderbuffer
*rb
;
1175 __DRIbuffer
*buffers
= NULL
;
1177 const char *region_name
;
1179 /* Set this up front, so that in case our buffers get invalidated
1180 * while we're getting new buffers, we don't clobber the stamp and
1181 * thus ignore the invalidate. */
1182 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1184 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1185 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1187 intel_query_dri2_buffers(brw
, drawable
, &buffers
, &count
);
1189 if (buffers
== NULL
)
1192 for (i
= 0; i
< count
; i
++) {
1193 switch (buffers
[i
].attachment
) {
1194 case __DRI_BUFFER_FRONT_LEFT
:
1195 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1196 region_name
= "dri2 front buffer";
1199 case __DRI_BUFFER_FAKE_FRONT_LEFT
:
1200 rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1201 region_name
= "dri2 fake front buffer";
1204 case __DRI_BUFFER_BACK_LEFT
:
1205 rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1206 region_name
= "dri2 back buffer";
1209 case __DRI_BUFFER_DEPTH
:
1210 case __DRI_BUFFER_HIZ
:
1211 case __DRI_BUFFER_DEPTH_STENCIL
:
1212 case __DRI_BUFFER_STENCIL
:
1213 case __DRI_BUFFER_ACCUM
:
1216 "unhandled buffer attach event, attachment type %d\n",
1217 buffers
[i
].attachment
);
1221 intel_process_dri2_buffer(brw
, drawable
, &buffers
[i
], rb
, region_name
);
1227 intel_update_renderbuffers(__DRIcontext
*context
, __DRIdrawable
*drawable
)
1229 struct brw_context
*brw
= context
->driverPrivate
;
1230 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1232 /* Set this up front, so that in case our buffers get invalidated
1233 * while we're getting new buffers, we don't clobber the stamp and
1234 * thus ignore the invalidate. */
1235 drawable
->lastStamp
= drawable
->dri2
.stamp
;
1237 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
))
1238 fprintf(stderr
, "enter %s, drawable %p\n", __func__
, drawable
);
1240 if (screen
->image
.loader
)
1241 intel_update_image_buffers(brw
, drawable
);
1243 intel_update_dri2_buffers(brw
, drawable
);
1245 driUpdateFramebufferSize(&brw
->ctx
, drawable
);
1249 * intel_prepare_render should be called anywhere that curent read/drawbuffer
1250 * state is required.
1253 intel_prepare_render(struct brw_context
*brw
)
1255 struct gl_context
*ctx
= &brw
->ctx
;
1256 __DRIcontext
*driContext
= brw
->driContext
;
1257 __DRIdrawable
*drawable
;
1259 drawable
= driContext
->driDrawablePriv
;
1260 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.draw_stamp
) {
1261 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1262 intel_update_renderbuffers(driContext
, drawable
);
1263 driContext
->dri2
.draw_stamp
= drawable
->dri2
.stamp
;
1266 drawable
= driContext
->driReadablePriv
;
1267 if (drawable
&& drawable
->dri2
.stamp
!= driContext
->dri2
.read_stamp
) {
1268 if (drawable
->lastStamp
!= drawable
->dri2
.stamp
)
1269 intel_update_renderbuffers(driContext
, drawable
);
1270 driContext
->dri2
.read_stamp
= drawable
->dri2
.stamp
;
1273 /* If we're currently rendering to the front buffer, the rendering
1274 * that will happen next will probably dirty the front buffer. So
1275 * mark it as dirty here.
1277 if (brw_is_front_buffer_drawing(ctx
->DrawBuffer
))
1278 brw
->front_buffer_dirty
= true;
1282 * \brief Query DRI2 to obtain a DRIdrawable's buffers.
1284 * To determine which DRI buffers to request, examine the renderbuffers
1285 * attached to the drawable's framebuffer. Then request the buffers with
1286 * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
1288 * This is called from intel_update_renderbuffers().
1290 * \param drawable Drawable whose buffers are queried.
1291 * \param buffers [out] List of buffers returned by DRI2 query.
1292 * \param buffer_count [out] Number of buffers returned.
1294 * \see intel_update_renderbuffers()
1295 * \see DRI2GetBuffers()
1296 * \see DRI2GetBuffersWithFormat()
1299 intel_query_dri2_buffers(struct brw_context
*brw
,
1300 __DRIdrawable
*drawable
,
1301 __DRIbuffer
**buffers
,
1304 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1305 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1307 unsigned attachments
[8];
1309 struct intel_renderbuffer
*front_rb
;
1310 struct intel_renderbuffer
*back_rb
;
1312 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1313 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1315 memset(attachments
, 0, sizeof(attachments
));
1316 if ((brw_is_front_buffer_drawing(fb
) ||
1317 brw_is_front_buffer_reading(fb
) ||
1318 !back_rb
) && front_rb
) {
1319 /* If a fake front buffer is in use, then querying for
1320 * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
1321 * the real front buffer to the fake front buffer. So before doing the
1322 * query, we need to make sure all the pending drawing has landed in the
1323 * real front buffer.
1325 intel_batchbuffer_flush(brw
);
1326 intel_flush_front(&brw
->ctx
);
1328 attachments
[i
++] = __DRI_BUFFER_FRONT_LEFT
;
1329 attachments
[i
++] = intel_bits_per_pixel(front_rb
);
1330 } else if (front_rb
&& brw
->front_buffer_dirty
) {
1331 /* We have pending front buffer rendering, but we aren't querying for a
1332 * front buffer. If the front buffer we have is a fake front buffer,
1333 * the X server is going to throw it away when it processes the query.
1334 * So before doing the query, make sure all the pending drawing has
1335 * landed in the real front buffer.
1337 intel_batchbuffer_flush(brw
);
1338 intel_flush_front(&brw
->ctx
);
1342 attachments
[i
++] = __DRI_BUFFER_BACK_LEFT
;
1343 attachments
[i
++] = intel_bits_per_pixel(back_rb
);
1346 assert(i
<= ARRAY_SIZE(attachments
));
1348 *buffers
= screen
->dri2
.loader
->getBuffersWithFormat(drawable
,
1353 drawable
->loaderPrivate
);
1357 * \brief Assign a DRI buffer's DRM region to a renderbuffer.
1359 * This is called from intel_update_renderbuffers().
1362 * DRI buffers whose attachment point is DRI2BufferStencil or
1363 * DRI2BufferDepthStencil are handled as special cases.
1365 * \param buffer_name is a human readable name, such as "dri2 front buffer",
1366 * that is passed to drm_intel_bo_gem_create_from_name().
1368 * \see intel_update_renderbuffers()
1371 intel_process_dri2_buffer(struct brw_context
*brw
,
1372 __DRIdrawable
*drawable
,
1373 __DRIbuffer
*buffer
,
1374 struct intel_renderbuffer
*rb
,
1375 const char *buffer_name
)
1377 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1383 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1385 /* We try to avoid closing and reopening the same BO name, because the first
1386 * use of a mapping of the buffer involves a bunch of page faulting which is
1387 * moderately expensive.
1389 struct intel_mipmap_tree
*last_mt
;
1390 if (num_samples
== 0)
1393 last_mt
= rb
->singlesample_mt
;
1395 uint32_t old_name
= 0;
1397 /* The bo already has a name because the miptree was created by a
1398 * previous call to intel_process_dri2_buffer(). If a bo already has a
1399 * name, then drm_intel_bo_flink() is a low-cost getter. It does not
1400 * create a new name.
1402 drm_intel_bo_flink(last_mt
->bo
, &old_name
);
1405 if (old_name
== buffer
->name
)
1408 if (unlikely(INTEL_DEBUG
& DEBUG_DRI
)) {
1410 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
1411 buffer
->name
, buffer
->attachment
,
1412 buffer
->cpp
, buffer
->pitch
);
1415 intel_miptree_release(&rb
->mt
);
1416 bo
= drm_intel_bo_gem_create_from_name(brw
->bufmgr
, buffer_name
,
1420 "Failed to open BO for returned DRI2 buffer "
1421 "(%dx%d, %s, named %d).\n"
1422 "This is likely a bug in the X Server that will lead to a "
1424 drawable
->w
, drawable
->h
, buffer_name
, buffer
->name
);
1428 intel_update_winsys_renderbuffer_miptree(brw
, rb
, bo
,
1429 drawable
->w
, drawable
->h
,
1432 if (brw_is_front_buffer_drawing(fb
) &&
1433 (buffer
->attachment
== __DRI_BUFFER_FRONT_LEFT
||
1434 buffer
->attachment
== __DRI_BUFFER_FAKE_FRONT_LEFT
) &&
1435 rb
->Base
.Base
.NumSamples
> 1) {
1436 intel_renderbuffer_upsample(brw
, rb
);
1441 drm_intel_bo_unreference(bo
);
1445 * \brief Query DRI image loader to obtain a DRIdrawable's buffers.
1447 * To determine which DRI buffers to request, examine the renderbuffers
1448 * attached to the drawable's framebuffer. Then request the buffers from
1451 * This is called from intel_update_renderbuffers().
1453 * \param drawable Drawable whose buffers are queried.
1454 * \param buffers [out] List of buffers returned by DRI2 query.
1455 * \param buffer_count [out] Number of buffers returned.
1457 * \see intel_update_renderbuffers()
1461 intel_update_image_buffer(struct brw_context
*intel
,
1462 __DRIdrawable
*drawable
,
1463 struct intel_renderbuffer
*rb
,
1465 enum __DRIimageBufferMask buffer_type
)
1467 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1469 if (!rb
|| !buffer
->bo
)
1472 unsigned num_samples
= rb
->Base
.Base
.NumSamples
;
1474 /* Check and see if we're already bound to the right
1477 struct intel_mipmap_tree
*last_mt
;
1478 if (num_samples
== 0)
1481 last_mt
= rb
->singlesample_mt
;
1483 if (last_mt
&& last_mt
->bo
== buffer
->bo
)
1486 intel_update_winsys_renderbuffer_miptree(intel
, rb
, buffer
->bo
,
1487 buffer
->width
, buffer
->height
,
1490 if (brw_is_front_buffer_drawing(fb
) &&
1491 buffer_type
== __DRI_IMAGE_BUFFER_FRONT
&&
1492 rb
->Base
.Base
.NumSamples
> 1) {
1493 intel_renderbuffer_upsample(intel
, rb
);
1498 intel_update_image_buffers(struct brw_context
*brw
, __DRIdrawable
*drawable
)
1500 struct gl_framebuffer
*fb
= drawable
->driverPrivate
;
1501 __DRIscreen
*screen
= brw
->intelScreen
->driScrnPriv
;
1502 struct intel_renderbuffer
*front_rb
;
1503 struct intel_renderbuffer
*back_rb
;
1504 struct __DRIimageList images
;
1505 unsigned int format
;
1506 uint32_t buffer_mask
= 0;
1508 front_rb
= intel_get_renderbuffer(fb
, BUFFER_FRONT_LEFT
);
1509 back_rb
= intel_get_renderbuffer(fb
, BUFFER_BACK_LEFT
);
1512 format
= intel_rb_format(back_rb
);
1514 format
= intel_rb_format(front_rb
);
1518 if (front_rb
&& (brw_is_front_buffer_drawing(fb
) ||
1519 brw_is_front_buffer_reading(fb
) || !back_rb
)) {
1520 buffer_mask
|= __DRI_IMAGE_BUFFER_FRONT
;
1524 buffer_mask
|= __DRI_IMAGE_BUFFER_BACK
;
1526 (*screen
->image
.loader
->getBuffers
) (drawable
,
1527 driGLFormatToImageFormat(format
),
1528 &drawable
->dri2
.stamp
,
1529 drawable
->loaderPrivate
,
1533 if (images
.image_mask
& __DRI_IMAGE_BUFFER_FRONT
) {
1534 drawable
->w
= images
.front
->width
;
1535 drawable
->h
= images
.front
->height
;
1536 intel_update_image_buffer(brw
,
1540 __DRI_IMAGE_BUFFER_FRONT
);
1542 if (images
.image_mask
& __DRI_IMAGE_BUFFER_BACK
) {
1543 drawable
->w
= images
.back
->width
;
1544 drawable
->h
= images
.back
->height
;
1545 intel_update_image_buffer(brw
,
1549 __DRI_IMAGE_BUFFER_BACK
);