2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
33 #include "main/context.h"
34 #include "main/blend.h"
35 #include "main/mtypes.h"
36 #include "main/samplerobj.h"
37 #include "program/prog_parameter.h"
39 #include "intel_mipmap_tree.h"
40 #include "intel_batchbuffer.h"
41 #include "intel_tex.h"
42 #include "intel_fbo.h"
43 #include "intel_buffer_objects.h"
45 #include "brw_context.h"
46 #include "brw_state.h"
47 #include "brw_defines.h"
51 translate_tex_target(GLenum target
)
55 case GL_TEXTURE_1D_ARRAY_EXT
:
56 return BRW_SURFACE_1D
;
58 case GL_TEXTURE_RECTANGLE_NV
:
59 return BRW_SURFACE_2D
;
62 case GL_TEXTURE_2D_ARRAY_EXT
:
63 case GL_TEXTURE_EXTERNAL_OES
:
64 case GL_TEXTURE_2D_MULTISAMPLE
:
65 case GL_TEXTURE_2D_MULTISAMPLE_ARRAY
:
66 return BRW_SURFACE_2D
;
69 return BRW_SURFACE_3D
;
71 case GL_TEXTURE_CUBE_MAP
:
72 case GL_TEXTURE_CUBE_MAP_ARRAY
:
73 return BRW_SURFACE_CUBE
;
82 brw_get_surface_tiling_bits(uint32_t tiling
)
86 return BRW_SURFACE_TILED
;
88 return BRW_SURFACE_TILED
| BRW_SURFACE_TILED_Y
;
96 brw_get_surface_num_multisamples(unsigned num_samples
)
99 return BRW_SURFACE_MULTISAMPLECOUNT_4
;
101 return BRW_SURFACE_MULTISAMPLECOUNT_1
;
106 * Compute the combination of DEPTH_TEXTURE_MODE and EXT_texture_swizzle
110 brw_get_texture_swizzle(const struct gl_context
*ctx
,
111 const struct gl_texture_object
*t
)
113 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
115 int swizzles
[SWIZZLE_NIL
+ 1] = {
125 if (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
126 img
->_BaseFormat
== GL_DEPTH_STENCIL
) {
127 GLenum depth_mode
= t
->DepthMode
;
129 /* In ES 3.0, DEPTH_TEXTURE_MODE is expected to be GL_RED for textures
130 * with depth component data specified with a sized internal format.
131 * Otherwise, it's left at the old default, GL_LUMINANCE.
133 if (_mesa_is_gles3(ctx
) &&
134 img
->InternalFormat
!= GL_DEPTH_COMPONENT
&&
135 img
->InternalFormat
!= GL_DEPTH_STENCIL
) {
139 switch (depth_mode
) {
141 swizzles
[0] = SWIZZLE_ZERO
;
142 swizzles
[1] = SWIZZLE_ZERO
;
143 swizzles
[2] = SWIZZLE_ZERO
;
144 swizzles
[3] = SWIZZLE_X
;
147 swizzles
[0] = SWIZZLE_X
;
148 swizzles
[1] = SWIZZLE_X
;
149 swizzles
[2] = SWIZZLE_X
;
150 swizzles
[3] = SWIZZLE_ONE
;
153 swizzles
[0] = SWIZZLE_X
;
154 swizzles
[1] = SWIZZLE_X
;
155 swizzles
[2] = SWIZZLE_X
;
156 swizzles
[3] = SWIZZLE_X
;
159 swizzles
[0] = SWIZZLE_X
;
160 swizzles
[1] = SWIZZLE_ZERO
;
161 swizzles
[2] = SWIZZLE_ZERO
;
162 swizzles
[3] = SWIZZLE_ONE
;
167 /* If the texture's format is alpha-only, force R, G, and B to
168 * 0.0. Similarly, if the texture's format has no alpha channel,
169 * force the alpha value read to 1.0. This allows for the
170 * implementation to use an RGBA texture for any of these formats
171 * without leaking any unexpected values.
173 switch (img
->_BaseFormat
) {
175 swizzles
[0] = SWIZZLE_ZERO
;
176 swizzles
[1] = SWIZZLE_ZERO
;
177 swizzles
[2] = SWIZZLE_ZERO
;
182 if (_mesa_get_format_bits(img
->TexFormat
, GL_ALPHA_BITS
) > 0)
183 swizzles
[3] = SWIZZLE_ONE
;
187 return MAKE_SWIZZLE4(swizzles
[GET_SWZ(t
->_Swizzle
, 0)],
188 swizzles
[GET_SWZ(t
->_Swizzle
, 1)],
189 swizzles
[GET_SWZ(t
->_Swizzle
, 2)],
190 swizzles
[GET_SWZ(t
->_Swizzle
, 3)]);
194 gen4_emit_buffer_surface_state(struct brw_context
*brw
,
195 uint32_t *out_offset
,
197 unsigned buffer_offset
,
198 unsigned surface_format
,
199 unsigned buffer_size
,
202 uint32_t *surf
= brw_state_batch(brw
, AUB_TRACE_SURFACE_STATE
,
203 6 * 4, 32, out_offset
);
204 memset(surf
, 0, 6 * 4);
206 surf
[0] = BRW_SURFACE_BUFFER
<< BRW_SURFACE_TYPE_SHIFT
|
207 surface_format
<< BRW_SURFACE_FORMAT_SHIFT
|
208 (brw
->gen
>= 6 ? BRW_SURFACE_RC_READ_WRITE
: 0);
209 surf
[1] = (bo
? bo
->offset
: 0) + buffer_offset
; /* reloc */
210 surf
[2] = (buffer_size
& 0x7f) << BRW_SURFACE_WIDTH_SHIFT
|
211 ((buffer_size
>> 7) & 0x1fff) << BRW_SURFACE_HEIGHT_SHIFT
;
212 surf
[3] = ((buffer_size
>> 20) & 0x7f) << BRW_SURFACE_DEPTH_SHIFT
|
213 (pitch
- 1) << BRW_SURFACE_PITCH_SHIFT
;
215 /* Emit relocation to surface contents. The 965 PRM, Volume 4, section
216 * 5.1.2 "Data Cache" says: "the data cache does not exist as a separate
217 * physical cache. It is mapped in hardware to the sampler cache."
220 drm_intel_bo_emit_reloc(brw
->batch
.bo
, *out_offset
+ 4,
222 I915_GEM_DOMAIN_SAMPLER
, 0);
227 brw_update_buffer_texture_surface(struct gl_context
*ctx
,
229 uint32_t *surf_offset
)
231 struct brw_context
*brw
= brw_context(ctx
);
232 struct gl_texture_object
*tObj
= ctx
->Texture
.Unit
[unit
]._Current
;
233 struct intel_buffer_object
*intel_obj
=
234 intel_buffer_object(tObj
->BufferObject
);
235 uint32_t size
= tObj
->BufferSize
;
236 drm_intel_bo
*bo
= NULL
;
237 gl_format format
= tObj
->_BufferObjectFormat
;
238 uint32_t brw_format
= brw_format_for_mesa_format(format
);
239 int texel_size
= _mesa_get_format_bytes(format
);
242 size
= MIN2(size
, intel_obj
->Base
.Size
);
243 bo
= intel_bufferobj_buffer(brw
, intel_obj
, tObj
->BufferOffset
, size
);
246 if (brw_format
== 0 && format
!= MESA_FORMAT_RGBA_FLOAT32
) {
247 _mesa_problem(NULL
, "bad format %s for texture buffer\n",
248 _mesa_get_format_name(format
));
251 gen4_emit_buffer_surface_state(brw
, surf_offset
, bo
,
259 brw_update_texture_surface(struct gl_context
*ctx
,
261 uint32_t *surf_offset
,
264 struct brw_context
*brw
= brw_context(ctx
);
265 struct gl_texture_object
*tObj
= ctx
->Texture
.Unit
[unit
]._Current
;
266 struct intel_texture_object
*intelObj
= intel_texture_object(tObj
);
267 struct intel_mipmap_tree
*mt
= intelObj
->mt
;
268 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit
);
271 /* BRW_NEW_UNIFORM_BUFFER */
272 if (tObj
->Target
== GL_TEXTURE_BUFFER
) {
273 brw_update_buffer_texture_surface(ctx
, unit
, surf_offset
);
277 surf
= brw_state_batch(brw
, AUB_TRACE_SURFACE_STATE
,
278 6 * 4, 32, surf_offset
);
280 (void) for_gather
; /* no w/a to apply for this gen */
282 surf
[0] = (translate_tex_target(tObj
->Target
) << BRW_SURFACE_TYPE_SHIFT
|
283 BRW_SURFACE_MIPMAPLAYOUT_BELOW
<< BRW_SURFACE_MIPLAYOUT_SHIFT
|
284 BRW_SURFACE_CUBEFACE_ENABLES
|
285 (translate_tex_format(brw
,
288 sampler
->sRGBDecode
) <<
289 BRW_SURFACE_FORMAT_SHIFT
));
291 surf
[1] = intelObj
->mt
->region
->bo
->offset
+ intelObj
->mt
->offset
; /* reloc */
293 surf
[2] = ((intelObj
->_MaxLevel
- tObj
->BaseLevel
) << BRW_SURFACE_LOD_SHIFT
|
294 (mt
->logical_width0
- 1) << BRW_SURFACE_WIDTH_SHIFT
|
295 (mt
->logical_height0
- 1) << BRW_SURFACE_HEIGHT_SHIFT
);
297 surf
[3] = (brw_get_surface_tiling_bits(intelObj
->mt
->region
->tiling
) |
298 (mt
->logical_depth0
- 1) << BRW_SURFACE_DEPTH_SHIFT
|
299 (intelObj
->mt
->region
->pitch
- 1) <<
300 BRW_SURFACE_PITCH_SHIFT
);
302 surf
[4] = (brw_get_surface_num_multisamples(intelObj
->mt
->num_samples
) |
303 SET_FIELD(tObj
->BaseLevel
- mt
->first_level
, BRW_SURFACE_MIN_LOD
));
305 surf
[5] = mt
->align_h
== 4 ? BRW_SURFACE_VERTICAL_ALIGN_ENABLE
: 0;
307 /* Emit relocation to surface contents */
308 drm_intel_bo_emit_reloc(brw
->batch
.bo
,
310 intelObj
->mt
->region
->bo
,
311 surf
[1] - intelObj
->mt
->region
->bo
->offset
,
312 I915_GEM_DOMAIN_SAMPLER
, 0);
316 * Create the constant buffer surface. Vertex/fragment shader constants will be
317 * read from this buffer with Data Port Read instructions/messages.
320 brw_create_constant_surface(struct brw_context
*brw
,
324 uint32_t *out_offset
,
327 uint32_t stride
= dword_pitch
? 4 : 16;
328 uint32_t elements
= ALIGN(size
, stride
) / stride
;
330 gen4_emit_buffer_surface_state(brw
, out_offset
, bo
, offset
,
331 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
,
336 * Set up a binding table entry for use by stream output logic (transform
339 * buffer_size_minus_1 must me less than BRW_MAX_NUM_BUFFER_ENTRIES.
342 brw_update_sol_surface(struct brw_context
*brw
,
343 struct gl_buffer_object
*buffer_obj
,
344 uint32_t *out_offset
, unsigned num_vector_components
,
345 unsigned stride_dwords
, unsigned offset_dwords
)
347 struct intel_buffer_object
*intel_bo
= intel_buffer_object(buffer_obj
);
348 uint32_t offset_bytes
= 4 * offset_dwords
;
349 drm_intel_bo
*bo
= intel_bufferobj_buffer(brw
, intel_bo
,
351 buffer_obj
->Size
- offset_bytes
);
352 uint32_t *surf
= brw_state_batch(brw
, AUB_TRACE_SURFACE_STATE
, 6 * 4, 32,
354 uint32_t pitch_minus_1
= 4*stride_dwords
- 1;
355 size_t size_dwords
= buffer_obj
->Size
/ 4;
356 uint32_t buffer_size_minus_1
, width
, height
, depth
, surface_format
;
358 /* FIXME: can we rely on core Mesa to ensure that the buffer isn't
359 * too big to map using a single binding table entry?
361 assert((size_dwords
- offset_dwords
) / stride_dwords
362 <= BRW_MAX_NUM_BUFFER_ENTRIES
);
364 if (size_dwords
> offset_dwords
+ num_vector_components
) {
365 /* There is room for at least 1 transform feedback output in the buffer.
366 * Compute the number of additional transform feedback outputs the
367 * buffer has room for.
369 buffer_size_minus_1
=
370 (size_dwords
- offset_dwords
- num_vector_components
) / stride_dwords
;
372 /* There isn't even room for a single transform feedback output in the
373 * buffer. We can't configure the binding table entry to prevent output
374 * entirely; we'll have to rely on the geometry shader to detect
375 * overflow. But to minimize the damage in case of a bug, set up the
376 * binding table entry to just allow a single output.
378 buffer_size_minus_1
= 0;
380 width
= buffer_size_minus_1
& 0x7f;
381 height
= (buffer_size_minus_1
& 0xfff80) >> 7;
382 depth
= (buffer_size_minus_1
& 0x7f00000) >> 20;
384 switch (num_vector_components
) {
386 surface_format
= BRW_SURFACEFORMAT_R32_FLOAT
;
389 surface_format
= BRW_SURFACEFORMAT_R32G32_FLOAT
;
392 surface_format
= BRW_SURFACEFORMAT_R32G32B32_FLOAT
;
395 surface_format
= BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
;
398 assert(!"Invalid vector size for transform feedback output");
399 surface_format
= BRW_SURFACEFORMAT_R32_FLOAT
;
403 surf
[0] = BRW_SURFACE_BUFFER
<< BRW_SURFACE_TYPE_SHIFT
|
404 BRW_SURFACE_MIPMAPLAYOUT_BELOW
<< BRW_SURFACE_MIPLAYOUT_SHIFT
|
405 surface_format
<< BRW_SURFACE_FORMAT_SHIFT
|
406 BRW_SURFACE_RC_READ_WRITE
;
407 surf
[1] = bo
->offset
+ offset_bytes
; /* reloc */
408 surf
[2] = (width
<< BRW_SURFACE_WIDTH_SHIFT
|
409 height
<< BRW_SURFACE_HEIGHT_SHIFT
);
410 surf
[3] = (depth
<< BRW_SURFACE_DEPTH_SHIFT
|
411 pitch_minus_1
<< BRW_SURFACE_PITCH_SHIFT
);
415 /* Emit relocation to surface contents. */
416 drm_intel_bo_emit_reloc(brw
->batch
.bo
,
419 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
);
422 /* Creates a new WM constant buffer reflecting the current fragment program's
423 * constants, if needed by the fragment program.
425 * Otherwise, constants go through the CURBEs using the brw_constant_buffer
429 brw_upload_wm_pull_constants(struct brw_context
*brw
)
431 struct gl_context
*ctx
= &brw
->ctx
;
432 /* BRW_NEW_FRAGMENT_PROGRAM */
433 struct brw_fragment_program
*fp
=
434 (struct brw_fragment_program
*) brw
->fragment_program
;
435 struct gl_program_parameter_list
*params
= fp
->program
.Base
.Parameters
;
436 const int size
= brw
->wm
.prog_data
->nr_pull_params
* sizeof(float);
437 const int surf_index
=
438 brw
->wm
.prog_data
->base
.binding_table
.pull_constants_start
;
442 _mesa_load_state_parameters(ctx
, params
);
444 /* CACHE_NEW_WM_PROG */
445 if (brw
->wm
.prog_data
->nr_pull_params
== 0) {
446 if (brw
->wm
.base
.const_bo
) {
447 drm_intel_bo_unreference(brw
->wm
.base
.const_bo
);
448 brw
->wm
.base
.const_bo
= NULL
;
449 brw
->wm
.base
.surf_offset
[surf_index
] = 0;
450 brw
->state
.dirty
.brw
|= BRW_NEW_SURFACES
;
455 drm_intel_bo_unreference(brw
->wm
.base
.const_bo
);
456 brw
->wm
.base
.const_bo
= drm_intel_bo_alloc(brw
->bufmgr
, "WM const bo",
459 /* _NEW_PROGRAM_CONSTANTS */
460 drm_intel_gem_bo_map_gtt(brw
->wm
.base
.const_bo
);
461 constants
= brw
->wm
.base
.const_bo
->virtual;
462 for (i
= 0; i
< brw
->wm
.prog_data
->nr_pull_params
; i
++) {
463 constants
[i
] = *brw
->wm
.prog_data
->pull_param
[i
];
465 drm_intel_gem_bo_unmap_gtt(brw
->wm
.base
.const_bo
);
467 brw
->vtbl
.create_constant_surface(brw
, brw
->wm
.base
.const_bo
, 0, size
,
468 &brw
->wm
.base
.surf_offset
[surf_index
],
471 brw
->state
.dirty
.brw
|= BRW_NEW_SURFACES
;
474 const struct brw_tracked_state brw_wm_pull_constants
= {
476 .mesa
= (_NEW_PROGRAM_CONSTANTS
),
477 .brw
= (BRW_NEW_BATCH
| BRW_NEW_FRAGMENT_PROGRAM
),
478 .cache
= CACHE_NEW_WM_PROG
,
480 .emit
= brw_upload_wm_pull_constants
,
484 brw_update_null_renderbuffer_surface(struct brw_context
*brw
, unsigned int unit
)
486 /* From the Sandy bridge PRM, Vol4 Part1 p71 (Surface Type: Programming
489 * A null surface will be used in instances where an actual surface is
490 * not bound. When a write message is generated to a null surface, no
491 * actual surface is written to. When a read message (including any
492 * sampling engine message) is generated to a null surface, the result
493 * is all zeros. Note that a null surface type is allowed to be used
494 * with all messages, even if it is not specificially indicated as
495 * supported. All of the remaining fields in surface state are ignored
496 * for null surfaces, with the following exceptions:
498 * - [DevSNB+]: Width, Height, Depth, and LOD fields must match the
499 * depth buffer’s corresponding state for all render target surfaces,
502 * - Surface Format must be R8G8B8A8_UNORM.
504 struct gl_context
*ctx
= &brw
->ctx
;
506 unsigned surface_type
= BRW_SURFACE_NULL
;
507 drm_intel_bo
*bo
= NULL
;
508 unsigned pitch_minus_1
= 0;
509 uint32_t multisampling_state
= 0;
510 uint32_t surf_index
=
511 brw
->wm
.prog_data
->binding_table
.render_target_start
+ unit
;
514 const struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
516 surf
= brw_state_batch(brw
, AUB_TRACE_SURFACE_STATE
, 6 * 4, 32,
517 &brw
->wm
.base
.surf_offset
[surf_index
]);
519 if (fb
->Visual
.samples
> 1) {
520 /* On Gen6, null render targets seem to cause GPU hangs when
521 * multisampling. So work around this problem by rendering into dummy
524 * To decrease the amount of memory needed by the workaround buffer, we
525 * set its pitch to 128 bytes (the width of a Y tile). This means that
526 * the amount of memory needed for the workaround buffer is
527 * (width_in_tiles + height_in_tiles - 1) tiles.
529 * Note that since the workaround buffer will be interpreted by the
530 * hardware as an interleaved multisampled buffer, we need to compute
531 * width_in_tiles and height_in_tiles by dividing the width and height
532 * by 16 rather than the normal Y-tile size of 32.
534 unsigned width_in_tiles
= ALIGN(fb
->Width
, 16) / 16;
535 unsigned height_in_tiles
= ALIGN(fb
->Height
, 16) / 16;
536 unsigned size_needed
= (width_in_tiles
+ height_in_tiles
- 1) * 4096;
537 brw_get_scratch_bo(brw
, &brw
->wm
.multisampled_null_render_target_bo
,
539 bo
= brw
->wm
.multisampled_null_render_target_bo
;
540 surface_type
= BRW_SURFACE_2D
;
542 multisampling_state
=
543 brw_get_surface_num_multisamples(fb
->Visual
.samples
);
546 surf
[0] = (surface_type
<< BRW_SURFACE_TYPE_SHIFT
|
547 BRW_SURFACEFORMAT_B8G8R8A8_UNORM
<< BRW_SURFACE_FORMAT_SHIFT
);
549 surf
[0] |= (1 << BRW_SURFACE_WRITEDISABLE_R_SHIFT
|
550 1 << BRW_SURFACE_WRITEDISABLE_G_SHIFT
|
551 1 << BRW_SURFACE_WRITEDISABLE_B_SHIFT
|
552 1 << BRW_SURFACE_WRITEDISABLE_A_SHIFT
);
554 surf
[1] = bo
? bo
->offset
: 0;
555 surf
[2] = ((fb
->Width
- 1) << BRW_SURFACE_WIDTH_SHIFT
|
556 (fb
->Height
- 1) << BRW_SURFACE_HEIGHT_SHIFT
);
558 /* From Sandy bridge PRM, Vol4 Part1 p82 (Tiled Surface: Programming
561 * If Surface Type is SURFTYPE_NULL, this field must be TRUE
563 surf
[3] = (BRW_SURFACE_TILED
| BRW_SURFACE_TILED_Y
|
564 pitch_minus_1
<< BRW_SURFACE_PITCH_SHIFT
);
565 surf
[4] = multisampling_state
;
569 drm_intel_bo_emit_reloc(brw
->batch
.bo
,
570 brw
->wm
.base
.surf_offset
[surf_index
] + 4,
572 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
);
577 * Sets up a surface state structure to point at the given region.
578 * While it is only used for the front/back buffer currently, it should be
579 * usable for further buffers when doing ARB_draw_buffer support.
582 brw_update_renderbuffer_surface(struct brw_context
*brw
,
583 struct gl_renderbuffer
*rb
,
587 struct gl_context
*ctx
= &brw
->ctx
;
588 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
589 struct intel_mipmap_tree
*mt
= irb
->mt
;
590 struct intel_region
*region
;
592 uint32_t tile_x
, tile_y
;
595 gl_format rb_format
= _mesa_get_render_format(ctx
, intel_rb_format(irb
));
596 uint32_t surf_index
=
597 brw
->wm
.prog_data
->binding_table
.render_target_start
+ unit
;
601 if (rb
->TexImage
&& !brw
->has_surface_tile_offset
) {
602 intel_renderbuffer_get_tile_offsets(irb
, &tile_x
, &tile_y
);
604 if (tile_x
!= 0 || tile_y
!= 0) {
605 /* Original gen4 hardware couldn't draw to a non-tile-aligned
606 * destination in a miptree unless you actually setup your renderbuffer
607 * as a miptree and used the fragile lod/array_index/etc. controls to
608 * select the image. So, instead, we just make a new single-level
609 * miptree and render into that.
611 intel_renderbuffer_move_to_temp(brw
, irb
, false);
616 intel_miptree_used_for_rendering(irb
->mt
);
618 region
= irb
->mt
->region
;
620 surf
= brw_state_batch(brw
, AUB_TRACE_SURFACE_STATE
, 6 * 4, 32,
621 &brw
->wm
.base
.surf_offset
[surf_index
]);
623 format
= brw
->render_target_format
[rb_format
];
624 if (unlikely(!brw
->format_supported_as_render_target
[rb_format
])) {
625 _mesa_problem(ctx
, "%s: renderbuffer format %s unsupported\n",
626 __FUNCTION__
, _mesa_get_format_name(rb_format
));
629 surf
[0] = (BRW_SURFACE_2D
<< BRW_SURFACE_TYPE_SHIFT
|
630 format
<< BRW_SURFACE_FORMAT_SHIFT
);
633 surf
[1] = (intel_renderbuffer_get_tile_offsets(irb
, &tile_x
, &tile_y
) +
636 surf
[2] = ((rb
->Width
- 1) << BRW_SURFACE_WIDTH_SHIFT
|
637 (rb
->Height
- 1) << BRW_SURFACE_HEIGHT_SHIFT
);
639 surf
[3] = (brw_get_surface_tiling_bits(region
->tiling
) |
640 (region
->pitch
- 1) << BRW_SURFACE_PITCH_SHIFT
);
642 surf
[4] = brw_get_surface_num_multisamples(mt
->num_samples
);
644 assert(brw
->has_surface_tile_offset
|| (tile_x
== 0 && tile_y
== 0));
645 /* Note that the low bits of these fields are missing, so
646 * there's the possibility of getting in trouble.
648 assert(tile_x
% 4 == 0);
649 assert(tile_y
% 2 == 0);
650 surf
[5] = ((tile_x
/ 4) << BRW_SURFACE_X_OFFSET_SHIFT
|
651 (tile_y
/ 2) << BRW_SURFACE_Y_OFFSET_SHIFT
|
652 (mt
->align_h
== 4 ? BRW_SURFACE_VERTICAL_ALIGN_ENABLE
: 0));
656 if (!ctx
->Color
.ColorLogicOpEnabled
&&
657 (ctx
->Color
.BlendEnabled
& (1 << unit
)))
658 surf
[0] |= BRW_SURFACE_BLEND_ENABLED
;
660 if (!ctx
->Color
.ColorMask
[unit
][0])
661 surf
[0] |= 1 << BRW_SURFACE_WRITEDISABLE_R_SHIFT
;
662 if (!ctx
->Color
.ColorMask
[unit
][1])
663 surf
[0] |= 1 << BRW_SURFACE_WRITEDISABLE_G_SHIFT
;
664 if (!ctx
->Color
.ColorMask
[unit
][2])
665 surf
[0] |= 1 << BRW_SURFACE_WRITEDISABLE_B_SHIFT
;
667 /* As mentioned above, disable writes to the alpha component when the
668 * renderbuffer is XRGB.
670 if (ctx
->DrawBuffer
->Visual
.alphaBits
== 0 ||
671 !ctx
->Color
.ColorMask
[unit
][3]) {
672 surf
[0] |= 1 << BRW_SURFACE_WRITEDISABLE_A_SHIFT
;
676 drm_intel_bo_emit_reloc(brw
->batch
.bo
,
677 brw
->wm
.base
.surf_offset
[surf_index
] + 4,
679 surf
[1] - region
->bo
->offset
,
680 I915_GEM_DOMAIN_RENDER
,
681 I915_GEM_DOMAIN_RENDER
);
685 * Construct SURFACE_STATE objects for renderbuffers/draw buffers.
688 brw_update_renderbuffer_surfaces(struct brw_context
*brw
)
690 struct gl_context
*ctx
= &brw
->ctx
;
693 /* _NEW_BUFFERS | _NEW_COLOR */
694 /* Update surfaces for drawing buffers */
695 if (ctx
->DrawBuffer
->_NumColorDrawBuffers
>= 1) {
696 for (i
= 0; i
< ctx
->DrawBuffer
->_NumColorDrawBuffers
; i
++) {
697 if (intel_renderbuffer(ctx
->DrawBuffer
->_ColorDrawBuffers
[i
])) {
698 brw
->vtbl
.update_renderbuffer_surface(brw
, ctx
->DrawBuffer
->_ColorDrawBuffers
[i
],
699 ctx
->DrawBuffer
->Layered
, i
);
701 brw
->vtbl
.update_null_renderbuffer_surface(brw
, i
);
705 brw
->vtbl
.update_null_renderbuffer_surface(brw
, 0);
707 brw
->state
.dirty
.brw
|= BRW_NEW_SURFACES
;
710 const struct brw_tracked_state brw_renderbuffer_surfaces
= {
712 .mesa
= (_NEW_COLOR
|
714 .brw
= BRW_NEW_BATCH
,
717 .emit
= brw_update_renderbuffer_surfaces
,
720 const struct brw_tracked_state gen6_renderbuffer_surfaces
= {
722 .mesa
= _NEW_BUFFERS
,
723 .brw
= BRW_NEW_BATCH
,
726 .emit
= brw_update_renderbuffer_surfaces
,
731 update_stage_texture_surfaces(struct brw_context
*brw
,
732 const struct gl_program
*prog
,
733 struct brw_stage_state
*stage_state
,
739 struct gl_context
*ctx
= &brw
->ctx
;
741 uint32_t *surf_offset
= stage_state
->surf_offset
;
743 surf_offset
+= stage_state
->prog_data
->binding_table
.gather_texture_start
;
745 surf_offset
+= stage_state
->prog_data
->binding_table
.texture_start
;
747 unsigned num_samplers
= _mesa_fls(prog
->SamplersUsed
);
748 for (unsigned s
= 0; s
< num_samplers
; s
++) {
751 if (prog
->SamplersUsed
& (1 << s
)) {
752 const unsigned unit
= prog
->SamplerUnits
[s
];
755 if (ctx
->Texture
.Unit
[unit
]._ReallyEnabled
) {
756 brw
->vtbl
.update_texture_surface(ctx
, unit
, surf_offset
+ s
, for_gather
);
764 * Construct SURFACE_STATE objects for enabled textures.
767 brw_update_texture_surfaces(struct brw_context
*brw
)
769 /* BRW_NEW_VERTEX_PROGRAM */
770 struct gl_program
*vs
= (struct gl_program
*) brw
->vertex_program
;
772 /* BRW_NEW_GEOMETRY_PROGRAM */
773 struct gl_program
*gs
= (struct gl_program
*) brw
->geometry_program
;
775 /* BRW_NEW_FRAGMENT_PROGRAM */
776 struct gl_program
*fs
= (struct gl_program
*) brw
->fragment_program
;
779 update_stage_texture_surfaces(brw
, vs
, &brw
->vs
.base
, false);
780 update_stage_texture_surfaces(brw
, gs
, &brw
->gs
.base
, false);
781 update_stage_texture_surfaces(brw
, fs
, &brw
->wm
.base
, false);
783 /* emit alternate set of surface state for gather. this
784 * allows the surface format to be overriden for only the
785 * gather4 messages. */
786 if (vs
&& vs
->UsesGather
)
787 update_stage_texture_surfaces(brw
, vs
, &brw
->vs
.base
, true);
788 if (gs
&& gs
->UsesGather
)
789 update_stage_texture_surfaces(brw
, gs
, &brw
->gs
.base
, true);
790 if (fs
&& fs
->UsesGather
)
791 update_stage_texture_surfaces(brw
, fs
, &brw
->wm
.base
, true);
793 brw
->state
.dirty
.brw
|= BRW_NEW_SURFACES
;
796 const struct brw_tracked_state brw_texture_surfaces
= {
798 .mesa
= _NEW_TEXTURE
,
799 .brw
= BRW_NEW_BATCH
|
800 BRW_NEW_UNIFORM_BUFFER
|
801 BRW_NEW_VERTEX_PROGRAM
|
802 BRW_NEW_GEOMETRY_PROGRAM
|
803 BRW_NEW_FRAGMENT_PROGRAM
,
806 .emit
= brw_update_texture_surfaces
,
810 brw_upload_ubo_surfaces(struct brw_context
*brw
,
811 struct gl_shader
*shader
,
812 struct brw_stage_state
*stage_state
,
813 struct brw_stage_prog_data
*prog_data
)
815 struct gl_context
*ctx
= &brw
->ctx
;
820 uint32_t *surf_offsets
=
821 &stage_state
->surf_offset
[prog_data
->binding_table
.ubo_start
];
823 for (int i
= 0; i
< shader
->NumUniformBlocks
; i
++) {
824 struct gl_uniform_buffer_binding
*binding
;
825 struct intel_buffer_object
*intel_bo
;
827 binding
= &ctx
->UniformBufferBindings
[shader
->UniformBlocks
[i
].Binding
];
828 intel_bo
= intel_buffer_object(binding
->BufferObject
);
830 intel_bufferobj_buffer(brw
, intel_bo
,
832 binding
->BufferObject
->Size
- binding
->Offset
);
834 /* Because behavior for referencing outside of the binding's size in the
835 * glBindBufferRange case is undefined, we can just bind the whole buffer
836 * glBindBufferBase wants and be a correct implementation.
838 brw
->vtbl
.create_constant_surface(brw
, bo
, binding
->Offset
,
839 bo
->size
- binding
->Offset
,
841 shader
->Type
== GL_FRAGMENT_SHADER
);
844 if (shader
->NumUniformBlocks
)
845 brw
->state
.dirty
.brw
|= BRW_NEW_SURFACES
;
849 brw_upload_wm_ubo_surfaces(struct brw_context
*brw
)
851 struct gl_context
*ctx
= &brw
->ctx
;
853 struct gl_shader_program
*prog
= ctx
->Shader
._CurrentFragmentProgram
;
858 /* CACHE_NEW_WM_PROG */
859 brw_upload_ubo_surfaces(brw
, prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
860 &brw
->wm
.base
, &brw
->wm
.prog_data
->base
);
863 const struct brw_tracked_state brw_wm_ubo_surfaces
= {
865 .mesa
= _NEW_PROGRAM
,
866 .brw
= BRW_NEW_BATCH
| BRW_NEW_UNIFORM_BUFFER
,
867 .cache
= CACHE_NEW_WM_PROG
,
869 .emit
= brw_upload_wm_ubo_surfaces
,
873 gen4_init_vtable_surface_functions(struct brw_context
*brw
)
875 brw
->vtbl
.update_texture_surface
= brw_update_texture_surface
;
876 brw
->vtbl
.update_renderbuffer_surface
= brw_update_renderbuffer_surface
;
877 brw
->vtbl
.update_null_renderbuffer_surface
=
878 brw_update_null_renderbuffer_surface
;
879 brw
->vtbl
.create_constant_surface
= brw_create_constant_surface
;