2 * Mesa 3-D graphics library
4 * Copyright (C) 2012-2013 LunarG, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
25 * Chia-I Wu <olv@lunarg.com>
28 #include "util/u_dual_blend.h"
29 #include "util/u_half.h"
30 #include "brw_defines.h"
31 #include "intel_reg.h"
33 #include "ilo_context.h"
34 #include "ilo_format.h"
35 #include "ilo_resource.h"
36 #include "ilo_shader.h"
37 #include "ilo_state.h"
38 #include "ilo_gpe_gen6.h"
41 * Translate a pipe logicop to the matching hardware logicop.
44 gen6_translate_pipe_logicop(unsigned logicop
)
47 case PIPE_LOGICOP_CLEAR
: return BRW_LOGICOPFUNCTION_CLEAR
;
48 case PIPE_LOGICOP_NOR
: return BRW_LOGICOPFUNCTION_NOR
;
49 case PIPE_LOGICOP_AND_INVERTED
: return BRW_LOGICOPFUNCTION_AND_INVERTED
;
50 case PIPE_LOGICOP_COPY_INVERTED
: return BRW_LOGICOPFUNCTION_COPY_INVERTED
;
51 case PIPE_LOGICOP_AND_REVERSE
: return BRW_LOGICOPFUNCTION_AND_REVERSE
;
52 case PIPE_LOGICOP_INVERT
: return BRW_LOGICOPFUNCTION_INVERT
;
53 case PIPE_LOGICOP_XOR
: return BRW_LOGICOPFUNCTION_XOR
;
54 case PIPE_LOGICOP_NAND
: return BRW_LOGICOPFUNCTION_NAND
;
55 case PIPE_LOGICOP_AND
: return BRW_LOGICOPFUNCTION_AND
;
56 case PIPE_LOGICOP_EQUIV
: return BRW_LOGICOPFUNCTION_EQUIV
;
57 case PIPE_LOGICOP_NOOP
: return BRW_LOGICOPFUNCTION_NOOP
;
58 case PIPE_LOGICOP_OR_INVERTED
: return BRW_LOGICOPFUNCTION_OR_INVERTED
;
59 case PIPE_LOGICOP_COPY
: return BRW_LOGICOPFUNCTION_COPY
;
60 case PIPE_LOGICOP_OR_REVERSE
: return BRW_LOGICOPFUNCTION_OR_REVERSE
;
61 case PIPE_LOGICOP_OR
: return BRW_LOGICOPFUNCTION_OR
;
62 case PIPE_LOGICOP_SET
: return BRW_LOGICOPFUNCTION_SET
;
64 assert(!"unknown logicop function");
65 return BRW_LOGICOPFUNCTION_CLEAR
;
70 * Translate a pipe blend function to the matching hardware blend function.
73 gen6_translate_pipe_blend(unsigned blend
)
76 case PIPE_BLEND_ADD
: return BRW_BLENDFUNCTION_ADD
;
77 case PIPE_BLEND_SUBTRACT
: return BRW_BLENDFUNCTION_SUBTRACT
;
78 case PIPE_BLEND_REVERSE_SUBTRACT
: return BRW_BLENDFUNCTION_REVERSE_SUBTRACT
;
79 case PIPE_BLEND_MIN
: return BRW_BLENDFUNCTION_MIN
;
80 case PIPE_BLEND_MAX
: return BRW_BLENDFUNCTION_MAX
;
82 assert(!"unknown blend function");
83 return BRW_BLENDFUNCTION_ADD
;
88 * Translate a pipe blend factor to the matching hardware blend factor.
91 gen6_translate_pipe_blendfactor(unsigned blendfactor
)
93 switch (blendfactor
) {
94 case PIPE_BLENDFACTOR_ONE
: return BRW_BLENDFACTOR_ONE
;
95 case PIPE_BLENDFACTOR_SRC_COLOR
: return BRW_BLENDFACTOR_SRC_COLOR
;
96 case PIPE_BLENDFACTOR_SRC_ALPHA
: return BRW_BLENDFACTOR_SRC_ALPHA
;
97 case PIPE_BLENDFACTOR_DST_ALPHA
: return BRW_BLENDFACTOR_DST_ALPHA
;
98 case PIPE_BLENDFACTOR_DST_COLOR
: return BRW_BLENDFACTOR_DST_COLOR
;
99 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
: return BRW_BLENDFACTOR_SRC_ALPHA_SATURATE
;
100 case PIPE_BLENDFACTOR_CONST_COLOR
: return BRW_BLENDFACTOR_CONST_COLOR
;
101 case PIPE_BLENDFACTOR_CONST_ALPHA
: return BRW_BLENDFACTOR_CONST_ALPHA
;
102 case PIPE_BLENDFACTOR_SRC1_COLOR
: return BRW_BLENDFACTOR_SRC1_COLOR
;
103 case PIPE_BLENDFACTOR_SRC1_ALPHA
: return BRW_BLENDFACTOR_SRC1_ALPHA
;
104 case PIPE_BLENDFACTOR_ZERO
: return BRW_BLENDFACTOR_ZERO
;
105 case PIPE_BLENDFACTOR_INV_SRC_COLOR
: return BRW_BLENDFACTOR_INV_SRC_COLOR
;
106 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
: return BRW_BLENDFACTOR_INV_SRC_ALPHA
;
107 case PIPE_BLENDFACTOR_INV_DST_ALPHA
: return BRW_BLENDFACTOR_INV_DST_ALPHA
;
108 case PIPE_BLENDFACTOR_INV_DST_COLOR
: return BRW_BLENDFACTOR_INV_DST_COLOR
;
109 case PIPE_BLENDFACTOR_INV_CONST_COLOR
: return BRW_BLENDFACTOR_INV_CONST_COLOR
;
110 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
: return BRW_BLENDFACTOR_INV_CONST_ALPHA
;
111 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
: return BRW_BLENDFACTOR_INV_SRC1_COLOR
;
112 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
: return BRW_BLENDFACTOR_INV_SRC1_ALPHA
;
114 assert(!"unknown blend factor");
115 return BRW_BLENDFACTOR_ONE
;
120 * Translate a pipe stencil op to the matching hardware stencil op.
123 gen6_translate_pipe_stencil_op(unsigned stencil_op
)
125 switch (stencil_op
) {
126 case PIPE_STENCIL_OP_KEEP
: return BRW_STENCILOP_KEEP
;
127 case PIPE_STENCIL_OP_ZERO
: return BRW_STENCILOP_ZERO
;
128 case PIPE_STENCIL_OP_REPLACE
: return BRW_STENCILOP_REPLACE
;
129 case PIPE_STENCIL_OP_INCR
: return BRW_STENCILOP_INCRSAT
;
130 case PIPE_STENCIL_OP_DECR
: return BRW_STENCILOP_DECRSAT
;
131 case PIPE_STENCIL_OP_INCR_WRAP
: return BRW_STENCILOP_INCR
;
132 case PIPE_STENCIL_OP_DECR_WRAP
: return BRW_STENCILOP_DECR
;
133 case PIPE_STENCIL_OP_INVERT
: return BRW_STENCILOP_INVERT
;
135 assert(!"unknown stencil op");
136 return BRW_STENCILOP_KEEP
;
141 * Translate a pipe texture mipfilter to the matching hardware mipfilter.
144 gen6_translate_tex_mipfilter(unsigned filter
)
147 case PIPE_TEX_MIPFILTER_NEAREST
: return BRW_MIPFILTER_NEAREST
;
148 case PIPE_TEX_MIPFILTER_LINEAR
: return BRW_MIPFILTER_LINEAR
;
149 case PIPE_TEX_MIPFILTER_NONE
: return BRW_MIPFILTER_NONE
;
151 assert(!"unknown mipfilter");
152 return BRW_MIPFILTER_NONE
;
157 * Translate a pipe texture filter to the matching hardware mapfilter.
160 gen6_translate_tex_filter(unsigned filter
)
163 case PIPE_TEX_FILTER_NEAREST
: return BRW_MAPFILTER_NEAREST
;
164 case PIPE_TEX_FILTER_LINEAR
: return BRW_MAPFILTER_LINEAR
;
166 assert(!"unknown sampler filter");
167 return BRW_MAPFILTER_NEAREST
;
172 * Translate a pipe texture coordinate wrapping mode to the matching hardware
176 gen6_translate_tex_wrap(unsigned wrap
, bool clamp_to_edge
)
178 /* clamp to edge or border? */
179 if (wrap
== PIPE_TEX_WRAP_CLAMP
) {
180 wrap
= (clamp_to_edge
) ?
181 PIPE_TEX_WRAP_CLAMP_TO_EDGE
: PIPE_TEX_WRAP_CLAMP_TO_BORDER
;
185 case PIPE_TEX_WRAP_REPEAT
: return BRW_TEXCOORDMODE_WRAP
;
186 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
: return BRW_TEXCOORDMODE_CLAMP
;
187 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
: return BRW_TEXCOORDMODE_CLAMP_BORDER
;
188 case PIPE_TEX_WRAP_MIRROR_REPEAT
: return BRW_TEXCOORDMODE_MIRROR
;
189 case PIPE_TEX_WRAP_CLAMP
:
190 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
191 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
192 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
194 assert(!"unknown sampler wrap mode");
195 return BRW_TEXCOORDMODE_WRAP
;
200 * Translate a pipe shadow compare function to the matching hardware shadow
204 gen6_translate_shadow_func(unsigned func
)
207 * For PIPE_FUNC_x, the reference value is on the left-hand side of the
208 * comparison, and 1.0 is returned when the comparison is true.
210 * For BRW_PREFILTER_x, the reference value is on the right-hand side of
211 * the comparison, and 0.0 is returned when the comparison is true.
214 case PIPE_FUNC_NEVER
: return BRW_PREFILTER_ALWAYS
;
215 case PIPE_FUNC_LESS
: return BRW_PREFILTER_LEQUAL
;
216 case PIPE_FUNC_EQUAL
: return BRW_PREFILTER_NOTEQUAL
;
217 case PIPE_FUNC_LEQUAL
: return BRW_PREFILTER_LESS
;
218 case PIPE_FUNC_GREATER
: return BRW_PREFILTER_GEQUAL
;
219 case PIPE_FUNC_NOTEQUAL
: return BRW_PREFILTER_EQUAL
;
220 case PIPE_FUNC_GEQUAL
: return BRW_PREFILTER_GREATER
;
221 case PIPE_FUNC_ALWAYS
: return BRW_PREFILTER_NEVER
;
223 assert(!"unknown shadow compare function");
224 return BRW_PREFILTER_NEVER
;
229 * Translate a pipe DSA test function to the matching hardware compare
233 gen6_translate_dsa_func(unsigned func
)
236 case PIPE_FUNC_NEVER
: return BRW_COMPAREFUNCTION_NEVER
;
237 case PIPE_FUNC_LESS
: return BRW_COMPAREFUNCTION_LESS
;
238 case PIPE_FUNC_EQUAL
: return BRW_COMPAREFUNCTION_EQUAL
;
239 case PIPE_FUNC_LEQUAL
: return BRW_COMPAREFUNCTION_LEQUAL
;
240 case PIPE_FUNC_GREATER
: return BRW_COMPAREFUNCTION_GREATER
;
241 case PIPE_FUNC_NOTEQUAL
: return BRW_COMPAREFUNCTION_NOTEQUAL
;
242 case PIPE_FUNC_GEQUAL
: return BRW_COMPAREFUNCTION_GEQUAL
;
243 case PIPE_FUNC_ALWAYS
: return BRW_COMPAREFUNCTION_ALWAYS
;
245 assert(!"unknown depth/stencil/alpha test function");
246 return BRW_COMPAREFUNCTION_NEVER
;
251 ve_init_cso(const struct ilo_dev_info
*dev
,
252 const struct pipe_vertex_element
*state
,
254 struct ilo_ve_cso
*cso
)
257 BRW_VE1_COMPONENT_STORE_SRC
,
258 BRW_VE1_COMPONENT_STORE_SRC
,
259 BRW_VE1_COMPONENT_STORE_SRC
,
260 BRW_VE1_COMPONENT_STORE_SRC
,
264 ILO_GPE_VALID_GEN(dev
, 6, 7);
266 switch (util_format_get_nr_components(state
->src_format
)) {
267 case 1: comp
[1] = BRW_VE1_COMPONENT_STORE_0
;
268 case 2: comp
[2] = BRW_VE1_COMPONENT_STORE_0
;
269 case 3: comp
[3] = (util_format_is_pure_integer(state
->src_format
)) ?
270 BRW_VE1_COMPONENT_STORE_1_INT
:
271 BRW_VE1_COMPONENT_STORE_1_FLT
;
274 format
= ilo_translate_vertex_format(state
->src_format
);
276 STATIC_ASSERT(Elements(cso
->payload
) >= 2);
278 vb_index
<< GEN6_VE0_INDEX_SHIFT
|
280 format
<< BRW_VE0_FORMAT_SHIFT
|
281 state
->src_offset
<< BRW_VE0_SRC_OFFSET_SHIFT
;
284 comp
[0] << BRW_VE1_COMPONENT_0_SHIFT
|
285 comp
[1] << BRW_VE1_COMPONENT_1_SHIFT
|
286 comp
[2] << BRW_VE1_COMPONENT_2_SHIFT
|
287 comp
[3] << BRW_VE1_COMPONENT_3_SHIFT
;
291 ilo_gpe_init_ve(const struct ilo_dev_info
*dev
,
293 const struct pipe_vertex_element
*states
,
294 struct ilo_ve_state
*ve
)
298 ILO_GPE_VALID_GEN(dev
, 6, 7);
300 ve
->count
= num_states
;
303 for (i
= 0; i
< num_states
; i
++) {
304 const unsigned pipe_idx
= states
[i
].vertex_buffer_index
;
305 const unsigned instance_divisor
= states
[i
].instance_divisor
;
309 * map the pipe vb to the hardware vb, which has a fixed instance
312 for (hw_idx
= 0; hw_idx
< ve
->vb_count
; hw_idx
++) {
313 if (ve
->vb_mapping
[hw_idx
] == pipe_idx
&&
314 ve
->instance_divisors
[hw_idx
] == instance_divisor
)
318 /* create one if there is no matching hardware vb */
319 if (hw_idx
>= ve
->vb_count
) {
320 hw_idx
= ve
->vb_count
++;
322 ve
->vb_mapping
[hw_idx
] = pipe_idx
;
323 ve
->instance_divisors
[hw_idx
] = instance_divisor
;
326 ve_init_cso(dev
, &states
[i
], hw_idx
, &ve
->cso
[i
]);
331 ilo_gpe_init_vs_cso(const struct ilo_dev_info
*dev
,
332 const struct ilo_shader_state
*vs
,
333 struct ilo_shader_cso
*cso
)
335 int start_grf
, vue_read_len
, max_threads
;
336 uint32_t dw2
, dw4
, dw5
;
338 ILO_GPE_VALID_GEN(dev
, 6, 7);
340 start_grf
= ilo_shader_get_kernel_param(vs
, ILO_KERNEL_URB_DATA_START_REG
);
341 vue_read_len
= ilo_shader_get_kernel_param(vs
, ILO_KERNEL_INPUT_COUNT
);
344 * From the Sandy Bridge PRM, volume 2 part 1, page 135:
346 * "(Vertex URB Entry Read Length) Specifies the number of pairs of
347 * 128-bit vertex elements to be passed into the payload for each
350 * "It is UNDEFINED to set this field to 0 indicating no Vertex URB
351 * data to be read and passed to the thread."
353 vue_read_len
= (vue_read_len
+ 1) / 2;
360 * From the Sandy Bridge PRM, volume 1 part 1, page 22:
362 * "Device # of EUs #Threads/EU
366 max_threads
= (dev
->gt
== 2) ? 60 : 24;
370 * From the Ivy Bridge PRM, volume 1 part 1, page 18:
372 * "Device # of EUs #Threads/EU
373 * Ivy Bridge (GT2) 16 8
374 * Ivy Bridge (GT1) 6 6"
376 max_threads
= (dev
->gt
== 2) ? 128 : 36;
379 /* see brwCreateContext() */
380 max_threads
= (dev
->gt
== 2) ? 280 : 70;
387 dw2
= (true) ? 0 : GEN6_VS_FLOATING_POINT_MODE_ALT
;
389 dw4
= start_grf
<< GEN6_VS_DISPATCH_START_GRF_SHIFT
|
390 vue_read_len
<< GEN6_VS_URB_READ_LENGTH_SHIFT
|
391 0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT
;
393 dw5
= GEN6_VS_STATISTICS_ENABLE
|
396 if (dev
->gen
>= ILO_GEN(7.5))
397 dw5
|= (max_threads
- 1) << HSW_VS_MAX_THREADS_SHIFT
;
399 dw5
|= (max_threads
- 1) << GEN6_VS_MAX_THREADS_SHIFT
;
401 STATIC_ASSERT(Elements(cso
->payload
) >= 3);
402 cso
->payload
[0] = dw2
;
403 cso
->payload
[1] = dw4
;
404 cso
->payload
[2] = dw5
;
408 ilo_gpe_init_gs_cso_gen6(const struct ilo_dev_info
*dev
,
409 const struct ilo_shader_state
*gs
,
410 struct ilo_shader_cso
*cso
)
412 int start_grf
, vue_read_len
, max_threads
;
413 uint32_t dw2
, dw4
, dw5
, dw6
;
415 ILO_GPE_VALID_GEN(dev
, 6, 6);
417 if (ilo_shader_get_type(gs
) == PIPE_SHADER_GEOMETRY
) {
418 start_grf
= ilo_shader_get_kernel_param(gs
,
419 ILO_KERNEL_URB_DATA_START_REG
);
421 vue_read_len
= ilo_shader_get_kernel_param(gs
, ILO_KERNEL_INPUT_COUNT
);
424 start_grf
= ilo_shader_get_kernel_param(gs
,
425 ILO_KERNEL_VS_GEN6_SO_START_REG
);
427 vue_read_len
= ilo_shader_get_kernel_param(gs
, ILO_KERNEL_OUTPUT_COUNT
);
431 * From the Sandy Bridge PRM, volume 2 part 1, page 153:
433 * "Specifies the amount of URB data read and passed in the thread
434 * payload for each Vertex URB entry, in 256-bit register increments.
436 * It is UNDEFINED to set this field (Vertex URB Entry Read Length) to
437 * 0 indicating no Vertex URB data to be read and passed to the
440 vue_read_len
= (vue_read_len
+ 1) / 2;
445 * From the Sandy Bridge PRM, volume 2 part 1, page 154:
447 * "Maximum Number of Threads valid range is [0,27] when Rendering
448 * Enabled bit is set."
450 * From the Sandy Bridge PRM, volume 2 part 1, page 173:
452 * "Programming Note: If the GS stage is enabled, software must always
453 * allocate at least one GS URB Entry. This is true even if the GS
454 * thread never needs to output vertices to the pipeline, e.g., when
455 * only performing stream output. This is an artifact of the need to
456 * pass the GS thread an initial destination URB handle."
458 * As such, we always enable rendering, and limit the number of threads.
461 /* maximum is 60, but limited to 28 */
465 /* maximum is 24, but limited to 21 (see brwCreateContext()) */
469 dw2
= GEN6_GS_SPF_MODE
;
471 dw4
= vue_read_len
<< GEN6_GS_URB_READ_LENGTH_SHIFT
|
472 0 << GEN6_GS_URB_ENTRY_READ_OFFSET_SHIFT
|
473 start_grf
<< GEN6_GS_DISPATCH_START_GRF_SHIFT
;
475 dw5
= (max_threads
- 1) << GEN6_GS_MAX_THREADS_SHIFT
|
476 GEN6_GS_STATISTICS_ENABLE
|
477 GEN6_GS_SO_STATISTICS_ENABLE
|
478 GEN6_GS_RENDERING_ENABLE
;
481 * we cannot make use of GEN6_GS_REORDER because it will reorder
482 * triangle strips according to D3D rules (triangle 2N+1 uses vertices
483 * (2N+1, 2N+3, 2N+2)), instead of GL rules (triangle 2N+1 uses vertices
484 * (2N+2, 2N+1, 2N+3)).
486 dw6
= GEN6_GS_ENABLE
;
488 if (ilo_shader_get_kernel_param(gs
, ILO_KERNEL_GS_DISCARD_ADJACENCY
))
489 dw6
|= GEN6_GS_DISCARD_ADJACENCY
;
491 if (ilo_shader_get_kernel_param(gs
, ILO_KERNEL_VS_GEN6_SO
)) {
492 const uint32_t svbi_post_inc
=
493 ilo_shader_get_kernel_param(gs
, ILO_KERNEL_GS_GEN6_SVBI_POST_INC
);
495 dw6
|= GEN6_GS_SVBI_PAYLOAD_ENABLE
;
497 dw6
|= GEN6_GS_SVBI_POSTINCREMENT_ENABLE
|
498 svbi_post_inc
<< GEN6_GS_SVBI_POSTINCREMENT_VALUE_SHIFT
;
502 STATIC_ASSERT(Elements(cso
->payload
) >= 4);
503 cso
->payload
[0] = dw2
;
504 cso
->payload
[1] = dw4
;
505 cso
->payload
[2] = dw5
;
506 cso
->payload
[3] = dw6
;
510 ilo_gpe_init_rasterizer_clip(const struct ilo_dev_info
*dev
,
511 const struct pipe_rasterizer_state
*state
,
512 struct ilo_rasterizer_clip
*clip
)
514 uint32_t dw1
, dw2
, dw3
;
516 ILO_GPE_VALID_GEN(dev
, 6, 7);
518 dw1
= GEN6_CLIP_STATISTICS_ENABLE
;
520 if (dev
->gen
>= ILO_GEN(7)) {
522 * From the Ivy Bridge PRM, volume 2 part 1, page 219:
524 * "Workaround : Due to Hardware issue "EarlyCull" needs to be
525 * enabled only for the cases where the incoming primitive topology
526 * into the clipper guaranteed to be Trilist."
528 * What does this mean?
531 GEN7_CLIP_EARLY_CULL
;
533 if (state
->front_ccw
)
534 dw1
|= GEN7_CLIP_WINDING_CCW
;
536 switch (state
->cull_face
) {
538 dw1
|= GEN7_CLIP_CULLMODE_NONE
;
540 case PIPE_FACE_FRONT
:
541 dw1
|= GEN7_CLIP_CULLMODE_FRONT
;
544 dw1
|= GEN7_CLIP_CULLMODE_BACK
;
546 case PIPE_FACE_FRONT_AND_BACK
:
547 dw1
|= GEN7_CLIP_CULLMODE_BOTH
;
552 dw2
= GEN6_CLIP_ENABLE
|
554 state
->clip_plane_enable
<< GEN6_USER_CLIP_CLIP_DISTANCES_SHIFT
|
555 GEN6_CLIP_MODE_NORMAL
;
557 if (state
->clip_halfz
)
558 dw2
|= GEN6_CLIP_API_D3D
;
560 dw2
|= GEN6_CLIP_API_OGL
;
562 if (state
->depth_clip
)
563 dw2
|= GEN6_CLIP_Z_TEST
;
565 if (state
->flatshade_first
) {
566 dw2
|= 0 << GEN6_CLIP_TRI_PROVOKE_SHIFT
|
567 0 << GEN6_CLIP_LINE_PROVOKE_SHIFT
|
568 1 << GEN6_CLIP_TRIFAN_PROVOKE_SHIFT
;
571 dw2
|= 2 << GEN6_CLIP_TRI_PROVOKE_SHIFT
|
572 1 << GEN6_CLIP_LINE_PROVOKE_SHIFT
|
573 2 << GEN6_CLIP_TRIFAN_PROVOKE_SHIFT
;
576 dw3
= 0x1 << GEN6_CLIP_MIN_POINT_WIDTH_SHIFT
|
577 0x7ff << GEN6_CLIP_MAX_POINT_WIDTH_SHIFT
;
579 clip
->payload
[0] = dw1
;
580 clip
->payload
[1] = dw2
;
581 clip
->payload
[2] = dw3
;
583 clip
->can_enable_guardband
= true;
586 * There are several reasons that guard band test should be disabled
588 * - GL wide points (to avoid partially visibie object)
589 * - GL wide or AA lines (to avoid partially visibie object)
591 if (state
->point_size_per_vertex
|| state
->point_size
> 1.0f
)
592 clip
->can_enable_guardband
= false;
593 if (state
->line_smooth
|| state
->line_width
> 1.0f
)
594 clip
->can_enable_guardband
= false;
598 ilo_gpe_init_rasterizer_sf(const struct ilo_dev_info
*dev
,
599 const struct pipe_rasterizer_state
*state
,
600 struct ilo_rasterizer_sf
*sf
)
602 float offset_const
, offset_scale
, offset_clamp
;
603 int line_width
, point_width
;
604 uint32_t dw1
, dw2
, dw3
;
606 ILO_GPE_VALID_GEN(dev
, 6, 7);
609 * Scale the constant term. The minimum representable value used by the HW
610 * is not large enouch to be the minimum resolvable difference.
612 offset_const
= state
->offset_units
* 2.0f
;
614 offset_scale
= state
->offset_scale
;
615 offset_clamp
= state
->offset_clamp
;
618 * From the Sandy Bridge PRM, volume 2 part 1, page 248:
620 * "This bit (Statistics Enable) should be set whenever clipping is
621 * enabled and the Statistics Enable bit is set in CLIP_STATE. It
622 * should be cleared if clipping is disabled or Statistics Enable in
623 * CLIP_STATE is clear."
625 dw1
= GEN6_SF_STATISTICS_ENABLE
|
626 GEN6_SF_VIEWPORT_TRANSFORM_ENABLE
;
628 /* XXX GEN6 path seems to work fine for GEN7 */
629 if (false && dev
->gen
>= ILO_GEN(7)) {
631 * From the Ivy Bridge PRM, volume 2 part 1, page 258:
633 * "This bit (Legacy Global Depth Bias Enable, Global Depth Offset
634 * Enable Solid , Global Depth Offset Enable Wireframe, and Global
635 * Depth Offset Enable Point) should be set whenever non zero depth
636 * bias (Slope, Bias) values are used. Setting this bit may have
637 * some degradation of performance for some workloads."
639 if (state
->offset_tri
|| state
->offset_line
|| state
->offset_point
) {
640 /* XXX need to scale offset_const according to the depth format */
641 dw1
|= GEN6_SF_LEGACY_GLOBAL_DEPTH_BIAS
;
643 dw1
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID
|
644 GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME
|
645 GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT
;
654 if (state
->offset_tri
)
655 dw1
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID
;
656 if (state
->offset_line
)
657 dw1
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME
;
658 if (state
->offset_point
)
659 dw1
|= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT
;
662 switch (state
->fill_front
) {
663 case PIPE_POLYGON_MODE_FILL
:
664 dw1
|= GEN6_SF_FRONT_SOLID
;
666 case PIPE_POLYGON_MODE_LINE
:
667 dw1
|= GEN6_SF_FRONT_WIREFRAME
;
669 case PIPE_POLYGON_MODE_POINT
:
670 dw1
|= GEN6_SF_FRONT_POINT
;
674 switch (state
->fill_back
) {
675 case PIPE_POLYGON_MODE_FILL
:
676 dw1
|= GEN6_SF_BACK_SOLID
;
678 case PIPE_POLYGON_MODE_LINE
:
679 dw1
|= GEN6_SF_BACK_WIREFRAME
;
681 case PIPE_POLYGON_MODE_POINT
:
682 dw1
|= GEN6_SF_BACK_POINT
;
686 if (state
->front_ccw
)
687 dw1
|= GEN6_SF_WINDING_CCW
;
691 if (state
->line_smooth
) {
693 * From the Sandy Bridge PRM, volume 2 part 1, page 251:
695 * "This field (Anti-aliasing Enable) must be disabled if any of the
696 * render targets have integer (UINT or SINT) surface format."
698 * From the Sandy Bridge PRM, volume 2 part 1, page 317:
700 * "This field (Hierarchical Depth Buffer Enable) must be disabled
701 * if Anti-aliasing Enable in 3DSTATE_SF is enabled.
703 * TODO We do not check those yet.
705 dw2
|= GEN6_SF_LINE_AA_ENABLE
|
706 GEN6_SF_LINE_END_CAP_WIDTH_1_0
;
709 switch (state
->cull_face
) {
711 dw2
|= GEN6_SF_CULL_NONE
;
713 case PIPE_FACE_FRONT
:
714 dw2
|= GEN6_SF_CULL_FRONT
;
717 dw2
|= GEN6_SF_CULL_BACK
;
719 case PIPE_FACE_FRONT_AND_BACK
:
720 dw2
|= GEN6_SF_CULL_BOTH
;
725 * Smooth lines should intersect ceil(line_width) or (ceil(line_width) + 1)
726 * pixels in the minor direction. We have to make the lines slightly
727 * thicker, 0.5 pixel on both sides, so that they intersect that many
728 * pixels are considered into the lines.
730 * Line width is in U3.7.
732 line_width
= (int) ((state
->line_width
+
733 (float) state
->line_smooth
) * 128.0f
+ 0.5f
);
734 line_width
= CLAMP(line_width
, 0, 1023);
736 if (line_width
== 128 && !state
->line_smooth
) {
741 dw2
|= line_width
<< GEN6_SF_LINE_WIDTH_SHIFT
;
744 dw2
|= GEN6_SF_SCISSOR_ENABLE
;
746 dw3
= GEN6_SF_LINE_AA_MODE_TRUE
|
747 GEN6_SF_VERTEX_SUBPIXEL_8BITS
;
749 if (state
->line_last_pixel
)
752 if (state
->flatshade_first
) {
753 dw3
|= 0 << GEN6_SF_TRI_PROVOKE_SHIFT
|
754 0 << GEN6_SF_LINE_PROVOKE_SHIFT
|
755 1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
;
758 dw3
|= 2 << GEN6_SF_TRI_PROVOKE_SHIFT
|
759 1 << GEN6_SF_LINE_PROVOKE_SHIFT
|
760 2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT
;
763 if (!state
->point_size_per_vertex
)
764 dw3
|= GEN6_SF_USE_STATE_POINT_WIDTH
;
767 point_width
= (int) (state
->point_size
* 8.0f
+ 0.5f
);
768 point_width
= CLAMP(point_width
, 1, 2047);
772 STATIC_ASSERT(Elements(sf
->payload
) >= 6);
773 sf
->payload
[0] = dw1
;
774 sf
->payload
[1] = dw2
;
775 sf
->payload
[2] = dw3
;
776 sf
->payload
[3] = fui(offset_const
);
777 sf
->payload
[4] = fui(offset_scale
);
778 sf
->payload
[5] = fui(offset_clamp
);
780 if (state
->multisample
) {
781 sf
->dw_msaa
= GEN6_SF_MSRAST_ON_PATTERN
;
784 * From the Sandy Bridge PRM, volume 2 part 1, page 251:
786 * "Software must not program a value of 0.0 when running in
787 * MSRASTMODE_ON_xxx modes - zero-width lines are not available
788 * when multisampling rasterization is enabled."
791 line_width
= 128; /* 1.0f */
793 sf
->dw_msaa
|= line_width
<< GEN6_SF_LINE_WIDTH_SHIFT
;
802 ilo_gpe_init_rasterizer_wm_gen6(const struct ilo_dev_info
*dev
,
803 const struct pipe_rasterizer_state
*state
,
804 struct ilo_rasterizer_wm
*wm
)
808 ILO_GPE_VALID_GEN(dev
, 6, 6);
810 /* only the FF unit states are set, as in GEN7 */
812 dw5
= GEN6_WM_LINE_AA_WIDTH_2_0
;
814 /* same value as in 3DSTATE_SF */
815 if (state
->line_smooth
)
816 dw5
|= GEN6_WM_LINE_END_CAP_AA_WIDTH_1_0
;
818 if (state
->poly_stipple_enable
)
819 dw5
|= GEN6_WM_POLYGON_STIPPLE_ENABLE
;
820 if (state
->line_stipple_enable
)
821 dw5
|= GEN6_WM_LINE_STIPPLE_ENABLE
;
823 dw6
= GEN6_WM_POSITION_ZW_PIXEL
|
824 GEN6_WM_MSRAST_OFF_PIXEL
|
825 GEN6_WM_MSDISPMODE_PERSAMPLE
;
827 if (state
->bottom_edge_rule
)
828 dw6
|= GEN6_WM_POINT_RASTRULE_UPPER_RIGHT
;
831 * assertion that makes sure
833 * dw6 |= wm->dw_msaa_rast | wm->dw_msaa_disp;
837 STATIC_ASSERT(GEN6_WM_MSRAST_OFF_PIXEL
== 0 &&
838 GEN6_WM_MSDISPMODE_PERSAMPLE
== 0);
841 (state
->multisample
) ? GEN6_WM_MSRAST_ON_PATTERN
: 0;
842 wm
->dw_msaa_disp
= GEN6_WM_MSDISPMODE_PERPIXEL
;
844 STATIC_ASSERT(Elements(wm
->payload
) >= 2);
845 wm
->payload
[0] = dw5
;
846 wm
->payload
[1] = dw6
;
850 ilo_gpe_init_fs_cso_gen6(const struct ilo_dev_info
*dev
,
851 const struct ilo_shader_state
*fs
,
852 struct ilo_shader_cso
*cso
)
854 int start_grf
, input_count
, interps
, max_threads
;
855 uint32_t dw2
, dw4
, dw5
, dw6
;
857 ILO_GPE_VALID_GEN(dev
, 6, 6);
859 start_grf
= ilo_shader_get_kernel_param(fs
, ILO_KERNEL_URB_DATA_START_REG
);
860 input_count
= ilo_shader_get_kernel_param(fs
, ILO_KERNEL_INPUT_COUNT
);
861 interps
= ilo_shader_get_kernel_param(fs
,
862 ILO_KERNEL_FS_BARYCENTRIC_INTERPOLATIONS
);
864 /* see brwCreateContext() */
865 max_threads
= (dev
->gt
== 2) ? 80 : 40;
867 dw2
= (true) ? 0 : GEN6_WM_FLOATING_POINT_MODE_ALT
;
869 dw4
= start_grf
<< GEN6_WM_DISPATCH_START_GRF_SHIFT_0
|
870 0 << GEN6_WM_DISPATCH_START_GRF_SHIFT_1
|
871 0 << GEN6_WM_DISPATCH_START_GRF_SHIFT_2
;
873 dw5
= (max_threads
- 1) << GEN6_WM_MAX_THREADS_SHIFT
;
876 * From the Sandy Bridge PRM, volume 2 part 1, page 275:
878 * "This bit (Pixel Shader Kill Pixel), if ENABLED, indicates that the
879 * PS kernel or color calculator has the ability to kill (discard)
880 * pixels or samples, other than due to depth or stencil testing.
881 * This bit is required to be ENABLED in the following situations:
883 * The API pixel shader program contains "killpix" or "discard"
884 * instructions, or other code in the pixel shader kernel that can
885 * cause the final pixel mask to differ from the pixel mask received
888 * A sampler with chroma key enabled with kill pixel mode is used by
891 * Any render target has Alpha Test Enable or AlphaToCoverage Enable
894 * The pixel shader kernel generates and outputs oMask.
896 * Note: As ClipDistance clipping is fully supported in hardware and
897 * therefore not via PS instructions, there should be no need to
898 * ENABLE this bit due to ClipDistance clipping."
900 if (ilo_shader_get_kernel_param(fs
, ILO_KERNEL_FS_USE_KILL
))
901 dw5
|= GEN6_WM_KILL_ENABLE
;
904 * From the Sandy Bridge PRM, volume 2 part 1, page 275:
906 * "If a NULL Depth Buffer is selected, the Pixel Shader Computed Depth
907 * field must be set to disabled."
909 * TODO This is not checked yet.
911 if (ilo_shader_get_kernel_param(fs
, ILO_KERNEL_FS_OUTPUT_Z
))
912 dw5
|= GEN6_WM_COMPUTED_DEPTH
;
914 if (ilo_shader_get_kernel_param(fs
, ILO_KERNEL_FS_INPUT_Z
))
915 dw5
|= GEN6_WM_USES_SOURCE_DEPTH
;
917 if (ilo_shader_get_kernel_param(fs
, ILO_KERNEL_FS_INPUT_W
))
918 dw5
|= GEN6_WM_USES_SOURCE_W
;
921 * TODO set this bit only when
923 * a) fs writes colors and color is not masked, or
924 * b) fs writes depth, or
928 dw5
|= GEN6_WM_DISPATCH_ENABLE
;
930 assert(!ilo_shader_get_kernel_param(fs
, ILO_KERNEL_FS_DISPATCH_16_OFFSET
));
931 dw5
|= GEN6_WM_8_DISPATCH_ENABLE
;
933 dw6
= input_count
<< GEN6_WM_NUM_SF_OUTPUTS_SHIFT
|
934 GEN6_WM_POSOFFSET_NONE
|
935 interps
<< GEN6_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT
;
937 STATIC_ASSERT(Elements(cso
->payload
) >= 4);
938 cso
->payload
[0] = dw2
;
939 cso
->payload
[1] = dw4
;
940 cso
->payload
[2] = dw5
;
941 cso
->payload
[3] = dw6
;
944 struct ilo_zs_surface_info
{
951 enum intel_tiling_mode tiling
;
955 unsigned width
, height
, depth
;
956 unsigned lod
, first_layer
, num_layers
;
957 uint32_t x_offset
, y_offset
;
961 zs_init_info_null(const struct ilo_dev_info
*dev
,
962 struct ilo_zs_surface_info
*info
)
964 ILO_GPE_VALID_GEN(dev
, 6, 7);
966 memset(info
, 0, sizeof(*info
));
968 info
->surface_type
= BRW_SURFACE_NULL
;
969 info
->format
= BRW_DEPTHFORMAT_D32_FLOAT
;
973 info
->num_layers
= 1;
977 zs_init_info(const struct ilo_dev_info
*dev
,
978 const struct ilo_texture
*tex
,
979 enum pipe_format format
,
981 unsigned first_layer
, unsigned num_layers
,
982 struct ilo_zs_surface_info
*info
)
984 const bool rebase_layer
= true;
985 struct intel_bo
* const hiz_bo
= NULL
;
986 bool separate_stencil
;
987 uint32_t x_offset
[3], y_offset
[3];
989 ILO_GPE_VALID_GEN(dev
, 6, 7);
991 memset(info
, 0, sizeof(*info
));
993 info
->surface_type
= ilo_gpe_gen6_translate_texture(tex
->base
.target
);
995 if (info
->surface_type
== BRW_SURFACE_CUBE
) {
997 * From the Sandy Bridge PRM, volume 2 part 1, page 325-326:
999 * "For Other Surfaces (Cube Surfaces):
1000 * This field (Minimum Array Element) is ignored."
1002 * "For Other Surfaces (Cube Surfaces):
1003 * This field (Render Target View Extent) is ignored."
1005 * As such, we cannot set first_layer and num_layers on cube surfaces.
1006 * To work around that, treat it as a 2D surface.
1008 info
->surface_type
= BRW_SURFACE_2D
;
1011 if (dev
->gen
>= ILO_GEN(7)) {
1012 separate_stencil
= true;
1016 * From the Sandy Bridge PRM, volume 2 part 1, page 317:
1018 * "This field (Separate Stencil Buffer Enable) must be set to the
1019 * same value (enabled or disabled) as Hierarchical Depth Buffer
1022 separate_stencil
= (hiz_bo
!= NULL
);
1026 * From the Sandy Bridge PRM, volume 2 part 1, page 317:
1028 * "If this field (Hierarchical Depth Buffer Enable) is enabled, the
1029 * Surface Format of the depth buffer cannot be
1030 * D32_FLOAT_S8X24_UINT or D24_UNORM_S8_UINT. Use of stencil
1031 * requires the separate stencil buffer."
1033 * From the Ironlake PRM, volume 2 part 1, page 330:
1035 * "If this field (Separate Stencil Buffer Enable) is disabled, the
1036 * Surface Format of the depth buffer cannot be D24_UNORM_X8_UINT."
1038 * There is no similar restriction for GEN6. But when D24_UNORM_X8_UINT
1039 * is indeed used, the depth values output by the fragment shaders will
1040 * be different when read back.
1042 * As for GEN7+, separate_stencil is always true.
1045 case PIPE_FORMAT_Z16_UNORM
:
1046 info
->format
= BRW_DEPTHFORMAT_D16_UNORM
;
1048 case PIPE_FORMAT_Z32_FLOAT
:
1049 info
->format
= BRW_DEPTHFORMAT_D32_FLOAT
;
1051 case PIPE_FORMAT_Z24X8_UNORM
:
1052 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
1053 info
->format
= (separate_stencil
) ?
1054 BRW_DEPTHFORMAT_D24_UNORM_X8_UINT
:
1055 BRW_DEPTHFORMAT_D24_UNORM_S8_UINT
;
1057 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1058 info
->format
= (separate_stencil
) ?
1059 BRW_DEPTHFORMAT_D32_FLOAT
:
1060 BRW_DEPTHFORMAT_D32_FLOAT_S8X24_UINT
;
1062 case PIPE_FORMAT_S8_UINT
:
1063 if (separate_stencil
) {
1064 info
->format
= BRW_DEPTHFORMAT_D32_FLOAT
;
1069 assert(!"unsupported depth/stencil format");
1070 zs_init_info_null(dev
, info
);
1075 if (format
!= PIPE_FORMAT_S8_UINT
) {
1076 info
->zs
.bo
= tex
->bo
;
1077 info
->zs
.stride
= tex
->bo_stride
;
1078 info
->zs
.tiling
= tex
->tiling
;
1081 info
->zs
.offset
= ilo_texture_get_slice_offset(tex
,
1082 level
, first_layer
, &x_offset
[0], &y_offset
[0]);
1086 if (tex
->separate_s8
|| format
== PIPE_FORMAT_S8_UINT
) {
1087 const struct ilo_texture
*s8_tex
=
1088 (tex
->separate_s8
) ? tex
->separate_s8
: tex
;
1090 info
->stencil
.bo
= s8_tex
->bo
;
1093 * From the Sandy Bridge PRM, volume 2 part 1, page 329:
1095 * "The pitch must be set to 2x the value computed based on width,
1096 * as the stencil buffer is stored with two rows interleaved."
1098 * According to the classic driver, we need to do the same for GEN7+
1099 * even though the Ivy Bridge PRM does not say anything about it.
1101 info
->stencil
.stride
= s8_tex
->bo_stride
* 2;
1103 info
->stencil
.tiling
= s8_tex
->tiling
;
1106 info
->stencil
.offset
= ilo_texture_get_slice_offset(s8_tex
,
1107 level
, first_layer
, &x_offset
[1], &y_offset
[1]);
1112 info
->hiz
.bo
= hiz_bo
;
1113 info
->hiz
.stride
= 0;
1114 info
->hiz
.tiling
= 0;
1115 info
->hiz
.offset
= 0;
1120 info
->width
= tex
->base
.width0
;
1121 info
->height
= tex
->base
.height0
;
1122 info
->depth
= (tex
->base
.target
== PIPE_TEXTURE_3D
) ?
1123 tex
->base
.depth0
: num_layers
;
1126 info
->first_layer
= first_layer
;
1127 info
->num_layers
= num_layers
;
1130 /* the size of the layer */
1131 info
->width
= u_minify(info
->width
, level
);
1132 info
->height
= u_minify(info
->height
, level
);
1133 if (info
->surface_type
== BRW_SURFACE_3D
)
1134 info
->depth
= u_minify(info
->depth
, level
);
1138 /* no layered rendering */
1139 assert(num_layers
== 1);
1142 info
->first_layer
= 0;
1143 info
->num_layers
= 1;
1145 /* all three share the same X/Y offsets */
1147 if (info
->stencil
.bo
) {
1148 assert(x_offset
[0] == x_offset
[1]);
1149 assert(y_offset
[0] == y_offset
[1]);
1152 info
->x_offset
= x_offset
[0];
1153 info
->y_offset
= y_offset
[0];
1156 assert(info
->stencil
.bo
);
1158 info
->x_offset
= x_offset
[1];
1159 info
->y_offset
= y_offset
[1];
1163 assert(info
->x_offset
== x_offset
[2]);
1164 assert(info
->y_offset
== y_offset
[2]);
1168 * From the Sandy Bridge PRM, volume 2 part 1, page 326:
1170 * "The 3 LSBs of both offsets (Depth Coordinate Offset Y and Depth
1171 * Coordinate Offset X) must be zero to ensure correct alignment"
1173 * XXX Skip the check for gen6, which seems to be fine. We need to make
1174 * sure that does not happen eventually.
1176 if (dev
->gen
>= ILO_GEN(7)) {
1177 assert((info
->x_offset
& 7) == 0 && (info
->y_offset
& 7) == 0);
1178 info
->x_offset
&= ~7;
1179 info
->y_offset
&= ~7;
1182 info
->width
+= info
->x_offset
;
1183 info
->height
+= info
->y_offset
;
1185 /* we have to treat them as 2D surfaces */
1186 if (info
->surface_type
== BRW_SURFACE_CUBE
) {
1187 assert(tex
->base
.width0
== tex
->base
.height0
);
1188 /* we will set slice_offset to point to the single face */
1189 info
->surface_type
= BRW_SURFACE_2D
;
1191 else if (info
->surface_type
== BRW_SURFACE_1D
&& info
->height
> 1) {
1192 assert(tex
->base
.height0
== 1);
1193 info
->surface_type
= BRW_SURFACE_2D
;
1199 ilo_gpe_init_zs_surface(const struct ilo_dev_info
*dev
,
1200 const struct ilo_texture
*tex
,
1201 enum pipe_format format
,
1203 unsigned first_layer
, unsigned num_layers
,
1204 struct ilo_zs_surface
*zs
)
1206 const int max_2d_size
= (dev
->gen
>= ILO_GEN(7)) ? 16384 : 8192;
1207 const int max_array_size
= (dev
->gen
>= ILO_GEN(7)) ? 2048 : 512;
1208 struct ilo_zs_surface_info info
;
1209 uint32_t dw1
, dw2
, dw3
, dw4
, dw5
, dw6
;
1211 ILO_GPE_VALID_GEN(dev
, 6, 7);
1214 zs_init_info(dev
, tex
, format
, level
, first_layer
, num_layers
, &info
);
1216 zs_init_info_null(dev
, &info
);
1218 switch (info
.surface_type
) {
1219 case BRW_SURFACE_NULL
:
1221 case BRW_SURFACE_1D
:
1222 assert(info
.width
<= max_2d_size
&& info
.height
== 1 &&
1223 info
.depth
<= max_array_size
);
1224 assert(info
.first_layer
< max_array_size
- 1 &&
1225 info
.num_layers
<= max_array_size
);
1227 case BRW_SURFACE_2D
:
1228 assert(info
.width
<= max_2d_size
&& info
.height
<= max_2d_size
&&
1229 info
.depth
<= max_array_size
);
1230 assert(info
.first_layer
< max_array_size
- 1 &&
1231 info
.num_layers
<= max_array_size
);
1233 case BRW_SURFACE_3D
:
1234 assert(info
.width
<= 2048 && info
.height
<= 2048 && info
.depth
<= 2048);
1235 assert(info
.first_layer
< 2048 && info
.num_layers
<= max_array_size
);
1236 assert(info
.x_offset
== 0 && info
.y_offset
== 0);
1238 case BRW_SURFACE_CUBE
:
1239 assert(info
.width
<= max_2d_size
&& info
.height
<= max_2d_size
&&
1241 assert(info
.first_layer
== 0 && info
.num_layers
== 1);
1242 assert(info
.width
== info
.height
);
1243 assert(info
.x_offset
== 0 && info
.y_offset
== 0);
1246 assert(!"unexpected depth surface type");
1250 dw1
= info
.surface_type
<< 29 |
1254 /* required for GEN6+ */
1255 assert(info
.zs
.tiling
== INTEL_TILING_Y
);
1256 assert(info
.zs
.stride
> 0 && info
.zs
.stride
< 128 * 1024 &&
1257 info
.zs
.stride
% 128 == 0);
1258 assert(info
.width
<= info
.zs
.stride
);
1260 dw1
|= (info
.zs
.stride
- 1);
1261 dw2
= info
.zs
.offset
;
1267 if (dev
->gen
>= ILO_GEN(7)) {
1271 if (info
.stencil
.bo
)
1277 dw3
= (info
.height
- 1) << 18 |
1278 (info
.width
- 1) << 4 |
1281 dw4
= (info
.depth
- 1) << 21 |
1282 info
.first_layer
<< 10;
1284 dw5
= info
.y_offset
<< 16 | info
.x_offset
;
1286 dw6
= (info
.num_layers
- 1) << 21;
1289 /* always Y-tiled */
1298 dw3
= (info
.height
- 1) << 19 |
1299 (info
.width
- 1) << 6 |
1301 BRW_SURFACE_MIPMAPLAYOUT_BELOW
<< 1;
1303 dw4
= (info
.depth
- 1) << 21 |
1304 info
.first_layer
<< 10 |
1305 (info
.num_layers
- 1) << 1;
1307 dw5
= info
.y_offset
<< 16 | info
.x_offset
;
1312 STATIC_ASSERT(Elements(zs
->payload
) >= 10);
1314 zs
->payload
[0] = dw1
;
1315 zs
->payload
[1] = dw2
;
1316 zs
->payload
[2] = dw3
;
1317 zs
->payload
[3] = dw4
;
1318 zs
->payload
[4] = dw5
;
1319 zs
->payload
[5] = dw6
;
1321 /* do not increment reference count */
1322 zs
->bo
= info
.zs
.bo
;
1324 /* separate stencil */
1325 if (info
.stencil
.bo
) {
1326 assert(info
.stencil
.stride
> 0 && info
.stencil
.stride
< 128 * 1024 &&
1327 info
.stencil
.stride
% 128 == 0);
1329 zs
->payload
[6] = info
.stencil
.stride
- 1;
1330 zs
->payload
[7] = info
.stencil
.offset
;
1332 /* do not increment reference count */
1333 zs
->separate_s8_bo
= info
.stencil
.bo
;
1338 zs
->separate_s8_bo
= NULL
;
1343 zs
->payload
[8] = info
.hiz
.stride
- 1;
1344 zs
->payload
[9] = info
.hiz
.offset
;
1346 /* do not increment reference count */
1347 zs
->hiz_bo
= info
.hiz
.bo
;
1357 viewport_get_guardband(const struct ilo_dev_info
*dev
,
1358 int center_x
, int center_y
,
1359 int *min_gbx
, int *max_gbx
,
1360 int *min_gby
, int *max_gby
)
1363 * From the Sandy Bridge PRM, volume 2 part 1, page 234:
1365 * "Per-Device Guardband Extents
1367 * - Supported X,Y ScreenSpace "Guardband" Extent: [-16K,16K-1]
1368 * - Maximum Post-Clamp Delta (X or Y): 16K"
1370 * "In addition, in order to be correctly rendered, objects must have a
1371 * screenspace bounding box not exceeding 8K in the X or Y direction.
1372 * This additional restriction must also be comprehended by software,
1373 * i.e., enforced by use of clipping."
1375 * From the Ivy Bridge PRM, volume 2 part 1, page 248:
1377 * "Per-Device Guardband Extents
1379 * - Supported X,Y ScreenSpace "Guardband" Extent: [-32K,32K-1]
1380 * - Maximum Post-Clamp Delta (X or Y): N/A"
1382 * "In addition, in order to be correctly rendered, objects must have a
1383 * screenspace bounding box not exceeding 8K in the X or Y direction.
1384 * This additional restriction must also be comprehended by software,
1385 * i.e., enforced by use of clipping."
1387 * Combined, the bounding box of any object can not exceed 8K in both
1390 * Below we set the guardband as a squre of length 8K, centered at where
1391 * the viewport is. This makes sure all objects passing the GB test are
1392 * valid to the renderer, and those failing the XY clipping have a
1393 * better chance of passing the GB test.
1395 const int max_extent
= (dev
->gen
>= ILO_GEN(7)) ? 32768 : 16384;
1396 const int half_len
= 8192 / 2;
1398 /* make sure the guardband is within the valid range */
1399 if (center_x
- half_len
< -max_extent
)
1400 center_x
= -max_extent
+ half_len
;
1401 else if (center_x
+ half_len
> max_extent
- 1)
1402 center_x
= max_extent
- half_len
;
1404 if (center_y
- half_len
< -max_extent
)
1405 center_y
= -max_extent
+ half_len
;
1406 else if (center_y
+ half_len
> max_extent
- 1)
1407 center_y
= max_extent
- half_len
;
1409 *min_gbx
= (float) (center_x
- half_len
);
1410 *max_gbx
= (float) (center_x
+ half_len
);
1411 *min_gby
= (float) (center_y
- half_len
);
1412 *max_gby
= (float) (center_y
+ half_len
);
1416 ilo_gpe_set_viewport_cso(const struct ilo_dev_info
*dev
,
1417 const struct pipe_viewport_state
*state
,
1418 struct ilo_viewport_cso
*vp
)
1420 const float scale_x
= fabs(state
->scale
[0]);
1421 const float scale_y
= fabs(state
->scale
[1]);
1422 const float scale_z
= fabs(state
->scale
[2]);
1423 int min_gbx
, max_gbx
, min_gby
, max_gby
;
1425 ILO_GPE_VALID_GEN(dev
, 6, 7);
1427 viewport_get_guardband(dev
,
1428 (int) state
->translate
[0],
1429 (int) state
->translate
[1],
1430 &min_gbx
, &max_gbx
, &min_gby
, &max_gby
);
1433 vp
->m00
= state
->scale
[0];
1434 vp
->m11
= state
->scale
[1];
1435 vp
->m22
= state
->scale
[2];
1436 vp
->m30
= state
->translate
[0];
1437 vp
->m31
= state
->translate
[1];
1438 vp
->m32
= state
->translate
[2];
1440 /* guardband in NDC space */
1441 vp
->min_gbx
= ((float) min_gbx
- state
->translate
[0]) / scale_x
;
1442 vp
->max_gbx
= ((float) max_gbx
- state
->translate
[0]) / scale_x
;
1443 vp
->min_gby
= ((float) min_gby
- state
->translate
[1]) / scale_y
;
1444 vp
->max_gby
= ((float) max_gby
- state
->translate
[1]) / scale_y
;
1446 /* viewport in screen space */
1447 vp
->min_x
= scale_x
* -1.0f
+ state
->translate
[0];
1448 vp
->max_x
= scale_x
* 1.0f
+ state
->translate
[0];
1449 vp
->min_y
= scale_y
* -1.0f
+ state
->translate
[1];
1450 vp
->max_y
= scale_y
* 1.0f
+ state
->translate
[1];
1451 vp
->min_z
= scale_z
* -1.0f
+ state
->translate
[2];
1452 vp
->max_z
= scale_z
* 1.0f
+ state
->translate
[2];
1456 gen6_blend_factor_dst_alpha_forced_one(int factor
)
1459 case BRW_BLENDFACTOR_DST_ALPHA
:
1460 return BRW_BLENDFACTOR_ONE
;
1461 case BRW_BLENDFACTOR_INV_DST_ALPHA
:
1462 case BRW_BLENDFACTOR_SRC_ALPHA_SATURATE
:
1463 return BRW_BLENDFACTOR_ZERO
;
1470 blend_get_rt_blend_enable(const struct ilo_dev_info
*dev
,
1471 const struct pipe_rt_blend_state
*rt
,
1472 bool dst_alpha_forced_one
)
1474 int rgb_src
, rgb_dst
, a_src
, a_dst
;
1477 if (!rt
->blend_enable
)
1480 rgb_src
= gen6_translate_pipe_blendfactor(rt
->rgb_src_factor
);
1481 rgb_dst
= gen6_translate_pipe_blendfactor(rt
->rgb_dst_factor
);
1482 a_src
= gen6_translate_pipe_blendfactor(rt
->alpha_src_factor
);
1483 a_dst
= gen6_translate_pipe_blendfactor(rt
->alpha_dst_factor
);
1485 if (dst_alpha_forced_one
) {
1486 rgb_src
= gen6_blend_factor_dst_alpha_forced_one(rgb_src
);
1487 rgb_dst
= gen6_blend_factor_dst_alpha_forced_one(rgb_dst
);
1488 a_src
= gen6_blend_factor_dst_alpha_forced_one(a_src
);
1489 a_dst
= gen6_blend_factor_dst_alpha_forced_one(a_dst
);
1493 gen6_translate_pipe_blend(rt
->alpha_func
) << 26 |
1496 gen6_translate_pipe_blend(rt
->rgb_func
) << 11 |
1500 if (rt
->rgb_func
!= rt
->alpha_func
||
1501 rgb_src
!= a_src
|| rgb_dst
!= a_dst
)
1508 ilo_gpe_init_blend(const struct ilo_dev_info
*dev
,
1509 const struct pipe_blend_state
*state
,
1510 struct ilo_blend_state
*blend
)
1512 unsigned num_cso
, i
;
1514 ILO_GPE_VALID_GEN(dev
, 6, 7);
1516 if (state
->independent_blend_enable
) {
1517 num_cso
= Elements(blend
->cso
);
1520 memset(blend
->cso
, 0, sizeof(blend
->cso
));
1524 blend
->independent_blend_enable
= state
->independent_blend_enable
;
1525 blend
->alpha_to_coverage
= state
->alpha_to_coverage
;
1526 blend
->dual_blend
= false;
1528 for (i
= 0; i
< num_cso
; i
++) {
1529 const struct pipe_rt_blend_state
*rt
= &state
->rt
[i
];
1530 struct ilo_blend_cso
*cso
= &blend
->cso
[i
];
1533 cso
->payload
[0] = 0;
1534 cso
->payload
[1] = BRW_RENDERTARGET_CLAMPRANGE_FORMAT
<< 2 |
1537 if (!(rt
->colormask
& PIPE_MASK_A
))
1538 cso
->payload
[1] |= 1 << 27;
1539 if (!(rt
->colormask
& PIPE_MASK_R
))
1540 cso
->payload
[1] |= 1 << 26;
1541 if (!(rt
->colormask
& PIPE_MASK_G
))
1542 cso
->payload
[1] |= 1 << 25;
1543 if (!(rt
->colormask
& PIPE_MASK_B
))
1544 cso
->payload
[1] |= 1 << 24;
1547 cso
->payload
[1] |= 1 << 12;
1550 * From the Sandy Bridge PRM, volume 2 part 1, page 365:
1552 * "Color Buffer Blending and Logic Ops must not be enabled
1553 * simultaneously, or behavior is UNDEFINED."
1555 * Since state->logicop_enable takes precedence over rt->blend_enable,
1556 * no special care is needed.
1558 if (state
->logicop_enable
) {
1559 cso
->dw_logicop
= 1 << 22 |
1560 gen6_translate_pipe_logicop(state
->logicop_func
) << 18;
1563 cso
->dw_blend_dst_alpha_forced_one
= 0;
1568 cso
->dw_logicop
= 0;
1570 cso
->dw_blend
= blend_get_rt_blend_enable(dev
, rt
, false);
1571 cso
->dw_blend_dst_alpha_forced_one
=
1572 blend_get_rt_blend_enable(dev
, rt
, true);
1574 dual_blend
= (rt
->blend_enable
&&
1575 util_blend_state_is_dual(state
, i
));
1578 cso
->dw_alpha_mod
= 0;
1580 if (state
->alpha_to_coverage
) {
1581 cso
->dw_alpha_mod
|= 1 << 31;
1583 if (dev
->gen
>= ILO_GEN(7))
1584 cso
->dw_alpha_mod
|= 1 << 29;
1588 * From the Sandy Bridge PRM, volume 2 part 1, page 378:
1590 * "If Dual Source Blending is enabled, this bit (AlphaToOne Enable)
1591 * must be disabled."
1593 if (state
->alpha_to_one
&& !dual_blend
)
1594 cso
->dw_alpha_mod
|= 1 << 30;
1597 blend
->dual_blend
= true;
1602 ilo_gpe_init_dsa(const struct ilo_dev_info
*dev
,
1603 const struct pipe_depth_stencil_alpha_state
*state
,
1604 struct ilo_dsa_state
*dsa
)
1606 const struct pipe_depth_state
*depth
= &state
->depth
;
1607 const struct pipe_stencil_state
*stencil0
= &state
->stencil
[0];
1608 const struct pipe_stencil_state
*stencil1
= &state
->stencil
[1];
1609 const struct pipe_alpha_state
*alpha
= &state
->alpha
;
1612 ILO_GPE_VALID_GEN(dev
, 6, 7);
1614 STATIC_ASSERT(Elements(dsa
->payload
) >= 3);
1618 * From the Sandy Bridge PRM, volume 2 part 1, page 359:
1620 * "If the Depth Buffer is either undefined or does not have a surface
1621 * format of D32_FLOAT_S8X24_UINT or D24_UNORM_S8_UINT and separate
1622 * stencil buffer is disabled, Stencil Test Enable must be DISABLED"
1624 * From the Sandy Bridge PRM, volume 2 part 1, page 370:
1626 * "This field (Stencil Test Enable) cannot be enabled if
1627 * Surface Format in 3DSTATE_DEPTH_BUFFER is set to D16_UNORM."
1629 * TODO We do not check these yet.
1631 if (stencil0
->enabled
) {
1633 gen6_translate_dsa_func(stencil0
->func
) << 28 |
1634 gen6_translate_pipe_stencil_op(stencil0
->fail_op
) << 25 |
1635 gen6_translate_pipe_stencil_op(stencil0
->zfail_op
) << 22 |
1636 gen6_translate_pipe_stencil_op(stencil0
->zpass_op
) << 19;
1637 if (stencil0
->writemask
)
1640 dw
[1] = stencil0
->valuemask
<< 24 |
1641 stencil0
->writemask
<< 16;
1643 if (stencil1
->enabled
) {
1645 gen6_translate_dsa_func(stencil1
->func
) << 12 |
1646 gen6_translate_pipe_stencil_op(stencil1
->fail_op
) << 9 |
1647 gen6_translate_pipe_stencil_op(stencil1
->zfail_op
) << 6 |
1648 gen6_translate_pipe_stencil_op(stencil1
->zpass_op
) << 3;
1649 if (stencil1
->writemask
)
1652 dw
[1] |= stencil1
->valuemask
<< 8 |
1653 stencil1
->writemask
;
1662 * From the Sandy Bridge PRM, volume 2 part 1, page 360:
1664 * "Enabling the Depth Test function without defining a Depth Buffer is
1667 * From the Sandy Bridge PRM, volume 2 part 1, page 375:
1669 * "A Depth Buffer must be defined before enabling writes to it, or
1670 * operation is UNDEFINED."
1672 * TODO We do not check these yet.
1674 dw
[2] = depth
->enabled
<< 31 |
1675 depth
->writemask
<< 26;
1677 dw
[2] |= gen6_translate_dsa_func(depth
->func
) << 27;
1679 dw
[2] |= BRW_COMPAREFUNCTION_ALWAYS
<< 27;
1681 /* dw_alpha will be ORed to BLEND_STATE */
1682 if (alpha
->enabled
) {
1683 dsa
->dw_alpha
= 1 << 16 |
1684 gen6_translate_dsa_func(alpha
->func
) << 13;
1690 dsa
->alpha_ref
= float_to_ubyte(alpha
->ref_value
);
1694 ilo_gpe_set_scissor(const struct ilo_dev_info
*dev
,
1695 unsigned start_slot
,
1696 unsigned num_states
,
1697 const struct pipe_scissor_state
*states
,
1698 struct ilo_scissor_state
*scissor
)
1702 ILO_GPE_VALID_GEN(dev
, 6, 7);
1704 for (i
= 0; i
< num_states
; i
++) {
1705 uint16_t min_x
, min_y
, max_x
, max_y
;
1707 /* both max and min are inclusive in SCISSOR_RECT */
1708 if (states
[i
].minx
< states
[i
].maxx
&&
1709 states
[i
].miny
< states
[i
].maxy
) {
1710 min_x
= states
[i
].minx
;
1711 min_y
= states
[i
].miny
;
1712 max_x
= states
[i
].maxx
- 1;
1713 max_y
= states
[i
].maxy
- 1;
1716 /* we have to make min greater than max */
1723 scissor
->payload
[(start_slot
+ i
) * 2 + 0] = min_y
<< 16 | min_x
;
1724 scissor
->payload
[(start_slot
+ i
) * 2 + 1] = max_y
<< 16 | max_x
;
1727 if (!start_slot
&& num_states
)
1728 scissor
->scissor0
= states
[0];
1732 ilo_gpe_set_scissor_null(const struct ilo_dev_info
*dev
,
1733 struct ilo_scissor_state
*scissor
)
1737 for (i
= 0; i
< Elements(scissor
->payload
); i
+= 2) {
1738 scissor
->payload
[i
+ 0] = 1 << 16 | 1;
1739 scissor
->payload
[i
+ 1] = 0;
1744 ilo_gpe_init_view_surface_null_gen6(const struct ilo_dev_info
*dev
,
1745 unsigned width
, unsigned height
,
1746 unsigned depth
, unsigned level
,
1747 struct ilo_view_surface
*surf
)
1751 ILO_GPE_VALID_GEN(dev
, 6, 6);
1754 * From the Sandy Bridge PRM, volume 4 part 1, page 71:
1756 * "A null surface will be used in instances where an actual surface is
1757 * not bound. When a write message is generated to a null surface, no
1758 * actual surface is written to. When a read message (including any
1759 * sampling engine message) is generated to a null surface, the result
1760 * is all zeros. Note that a null surface type is allowed to be used
1761 * with all messages, even if it is not specificially indicated as
1762 * supported. All of the remaining fields in surface state are ignored
1763 * for null surfaces, with the following exceptions:
1765 * * [DevSNB+]: Width, Height, Depth, and LOD fields must match the
1766 * depth buffer's corresponding state for all render target
1767 * surfaces, including null.
1768 * * Surface Format must be R8G8B8A8_UNORM."
1770 * From the Sandy Bridge PRM, volume 4 part 1, page 82:
1772 * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be
1776 STATIC_ASSERT(Elements(surf
->payload
) >= 6);
1779 dw
[0] = BRW_SURFACE_NULL
<< BRW_SURFACE_TYPE_SHIFT
|
1780 BRW_SURFACEFORMAT_B8G8R8A8_UNORM
<< BRW_SURFACE_FORMAT_SHIFT
;
1784 dw
[2] = (height
- 1) << BRW_SURFACE_HEIGHT_SHIFT
|
1785 (width
- 1) << BRW_SURFACE_WIDTH_SHIFT
|
1786 level
<< BRW_SURFACE_LOD_SHIFT
;
1788 dw
[3] = (depth
- 1) << BRW_SURFACE_DEPTH_SHIFT
|
1798 ilo_gpe_init_view_surface_for_buffer_gen6(const struct ilo_dev_info
*dev
,
1799 const struct ilo_buffer
*buf
,
1800 unsigned offset
, unsigned size
,
1801 unsigned struct_size
,
1802 enum pipe_format elem_format
,
1803 bool is_rt
, bool render_cache_rw
,
1804 struct ilo_view_surface
*surf
)
1806 const int elem_size
= util_format_get_blocksize(elem_format
);
1807 int width
, height
, depth
, pitch
;
1808 int surface_format
, num_entries
;
1811 ILO_GPE_VALID_GEN(dev
, 6, 6);
1814 * For SURFTYPE_BUFFER, a SURFACE_STATE specifies an element of a
1815 * structure in a buffer.
1818 surface_format
= ilo_translate_color_format(elem_format
);
1820 num_entries
= size
/ struct_size
;
1821 /* see if there is enough space to fit another element */
1822 if (size
% struct_size
>= elem_size
)
1826 * From the Sandy Bridge PRM, volume 4 part 1, page 76:
1828 * "For SURFTYPE_BUFFER render targets, this field (Surface Base
1829 * Address) specifies the base address of first element of the
1830 * surface. The surface is interpreted as a simple array of that
1831 * single element type. The address must be naturally-aligned to the
1832 * element size (e.g., a buffer containing R32G32B32A32_FLOAT elements
1833 * must be 16-byte aligned).
1835 * For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
1836 * the base address of the first element of the surface, computed in
1837 * software by adding the surface base address to the byte offset of
1838 * the element in the buffer."
1841 assert(offset
% elem_size
== 0);
1844 * From the Sandy Bridge PRM, volume 4 part 1, page 77:
1846 * "For buffer surfaces, the number of entries in the buffer ranges
1849 assert(num_entries
>= 1 && num_entries
<= 1 << 27);
1852 * From the Sandy Bridge PRM, volume 4 part 1, page 81:
1854 * "For surfaces of type SURFTYPE_BUFFER, this field (Surface Pitch)
1855 * indicates the size of the structure."
1857 pitch
= struct_size
;
1862 width
= (num_entries
& 0x0000007f);
1864 height
= (num_entries
& 0x000fff80) >> 7;
1866 depth
= (num_entries
& 0x07f00000) >> 20;
1868 STATIC_ASSERT(Elements(surf
->payload
) >= 6);
1871 dw
[0] = BRW_SURFACE_BUFFER
<< BRW_SURFACE_TYPE_SHIFT
|
1872 surface_format
<< BRW_SURFACE_FORMAT_SHIFT
;
1873 if (render_cache_rw
)
1874 dw
[0] |= BRW_SURFACE_RC_READ_WRITE
;
1878 dw
[2] = height
<< BRW_SURFACE_HEIGHT_SHIFT
|
1879 width
<< BRW_SURFACE_WIDTH_SHIFT
;
1881 dw
[3] = depth
<< BRW_SURFACE_DEPTH_SHIFT
|
1882 pitch
<< BRW_SURFACE_PITCH_SHIFT
;
1887 /* do not increment reference count */
1892 ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info
*dev
,
1893 const struct ilo_texture
*tex
,
1894 enum pipe_format format
,
1895 unsigned first_level
,
1896 unsigned num_levels
,
1897 unsigned first_layer
,
1898 unsigned num_layers
,
1899 bool is_rt
, bool render_cache_rw
,
1900 struct ilo_view_surface
*surf
)
1902 int surface_type
, surface_format
;
1903 int width
, height
, depth
, pitch
, lod
;
1904 unsigned layer_offset
, x_offset
, y_offset
;
1907 ILO_GPE_VALID_GEN(dev
, 6, 6);
1909 surface_type
= ilo_gpe_gen6_translate_texture(tex
->base
.target
);
1910 assert(surface_type
!= BRW_SURFACE_BUFFER
);
1912 if (format
== PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
&& tex
->separate_s8
)
1913 format
= PIPE_FORMAT_Z32_FLOAT
;
1916 surface_format
= ilo_translate_render_format(format
);
1918 surface_format
= ilo_translate_texture_format(format
);
1919 assert(surface_format
>= 0);
1921 width
= tex
->base
.width0
;
1922 height
= tex
->base
.height0
;
1923 depth
= (tex
->base
.target
== PIPE_TEXTURE_3D
) ?
1924 tex
->base
.depth0
: num_layers
;
1925 pitch
= tex
->bo_stride
;
1927 if (surface_type
== BRW_SURFACE_CUBE
) {
1929 * From the Sandy Bridge PRM, volume 4 part 1, page 81:
1931 * "For SURFTYPE_CUBE: [DevSNB+]: for Sampling Engine Surfaces, the
1932 * range of this field (Depth) is [0,84], indicating the number of
1933 * cube array elements (equal to the number of underlying 2D array
1934 * elements divided by 6). For other surfaces, this field must be
1937 * When is_rt is true, we treat the texture as a 2D one to avoid the
1941 surface_type
= BRW_SURFACE_2D
;
1944 assert(num_layers
% 6 == 0);
1945 depth
= num_layers
/ 6;
1949 /* sanity check the size */
1950 assert(width
>= 1 && height
>= 1 && depth
>= 1 && pitch
>= 1);
1951 switch (surface_type
) {
1952 case BRW_SURFACE_1D
:
1953 assert(width
<= 8192 && height
== 1 && depth
<= 512);
1954 assert(first_layer
< 512 && num_layers
<= 512);
1956 case BRW_SURFACE_2D
:
1957 assert(width
<= 8192 && height
<= 8192 && depth
<= 512);
1958 assert(first_layer
< 512 && num_layers
<= 512);
1960 case BRW_SURFACE_3D
:
1961 assert(width
<= 2048 && height
<= 2048 && depth
<= 2048);
1962 assert(first_layer
< 2048 && num_layers
<= 512);
1964 assert(first_layer
== 0);
1966 case BRW_SURFACE_CUBE
:
1967 assert(width
<= 8192 && height
<= 8192 && depth
<= 85);
1968 assert(width
== height
);
1969 assert(first_layer
< 512 && num_layers
<= 512);
1971 assert(first_layer
== 0);
1974 assert(!"unexpected surface type");
1978 /* non-full array spacing is supported only on GEN7+ */
1979 assert(tex
->array_spacing_full
);
1980 /* non-interleaved samples are supported only on GEN7+ */
1981 if (tex
->base
.nr_samples
> 1)
1982 assert(tex
->interleaved
);
1986 * Compute the offset to the layer manually.
1988 * For rendering, the hardware requires LOD to be the same for all
1989 * render targets and the depth buffer. We need to compute the offset
1990 * to the layer manually and always set LOD to 0.
1993 /* we lose the capability for layered rendering */
1994 assert(num_layers
== 1);
1996 layer_offset
= ilo_texture_get_slice_offset(tex
,
1997 first_level
, first_layer
, &x_offset
, &y_offset
);
1999 assert(x_offset
% 4 == 0);
2000 assert(y_offset
% 2 == 0);
2004 /* derive the size for the LOD */
2005 width
= u_minify(width
, first_level
);
2006 height
= u_minify(height
, first_level
);
2007 if (surface_type
== BRW_SURFACE_3D
)
2008 depth
= u_minify(depth
, first_level
);
2022 assert(num_levels
== 1);
2030 lod
= num_levels
- 1;
2034 * From the Sandy Bridge PRM, volume 4 part 1, page 76:
2036 * "Linear render target surface base addresses must be element-size
2037 * aligned, for non-YUV surface formats, or a multiple of 2
2038 * element-sizes for YUV surface formats. Other linear surfaces have
2039 * no alignment requirements (byte alignment is sufficient.)"
2041 * From the Sandy Bridge PRM, volume 4 part 1, page 81:
2043 * "For linear render target surfaces, the pitch must be a multiple
2044 * of the element size for non-YUV surface formats. Pitch must be a
2045 * multiple of 2 * element size for YUV surface formats."
2047 * From the Sandy Bridge PRM, volume 4 part 1, page 86:
2049 * "For linear surfaces, this field (X Offset) must be zero"
2051 if (tex
->tiling
== INTEL_TILING_NONE
) {
2053 const int elem_size
= util_format_get_blocksize(format
);
2054 assert(layer_offset
% elem_size
== 0);
2055 assert(pitch
% elem_size
== 0);
2061 STATIC_ASSERT(Elements(surf
->payload
) >= 6);
2064 dw
[0] = surface_type
<< BRW_SURFACE_TYPE_SHIFT
|
2065 surface_format
<< BRW_SURFACE_FORMAT_SHIFT
|
2066 BRW_SURFACE_MIPMAPLAYOUT_BELOW
<< BRW_SURFACE_MIPLAYOUT_SHIFT
;
2068 if (surface_type
== BRW_SURFACE_CUBE
&& !is_rt
) {
2070 BRW_SURFACE_CUBEFACE_ENABLES
;
2073 if (render_cache_rw
)
2074 dw
[0] |= BRW_SURFACE_RC_READ_WRITE
;
2076 dw
[1] = layer_offset
;
2078 dw
[2] = (height
- 1) << BRW_SURFACE_HEIGHT_SHIFT
|
2079 (width
- 1) << BRW_SURFACE_WIDTH_SHIFT
|
2080 lod
<< BRW_SURFACE_LOD_SHIFT
;
2082 dw
[3] = (depth
- 1) << BRW_SURFACE_DEPTH_SHIFT
|
2083 (pitch
- 1) << BRW_SURFACE_PITCH_SHIFT
|
2084 ilo_gpe_gen6_translate_winsys_tiling(tex
->tiling
);
2086 dw
[4] = first_level
<< BRW_SURFACE_MIN_LOD_SHIFT
|
2088 (num_layers
- 1) << 8 |
2089 ((tex
->base
.nr_samples
> 1) ? BRW_SURFACE_MULTISAMPLECOUNT_4
:
2090 BRW_SURFACE_MULTISAMPLECOUNT_1
);
2092 dw
[5] = x_offset
<< BRW_SURFACE_X_OFFSET_SHIFT
|
2093 y_offset
<< BRW_SURFACE_Y_OFFSET_SHIFT
;
2095 dw
[5] |= BRW_SURFACE_VERTICAL_ALIGN_ENABLE
;
2097 /* do not increment reference count */
2102 sampler_init_border_color_gen6(const struct ilo_dev_info
*dev
,
2103 const union pipe_color_union
*color
,
2104 uint32_t *dw
, int num_dwords
)
2107 color
->f
[0], color
->f
[1], color
->f
[2], color
->f
[3],
2110 ILO_GPE_VALID_GEN(dev
, 6, 6);
2112 assert(num_dwords
>= 12);
2115 * This state is not documented in the Sandy Bridge PRM, but in the
2116 * Ironlake PRM. SNORM8 seems to be in DW11 instead of DW1.
2120 dw
[1] = fui(rgba
[0]);
2121 dw
[2] = fui(rgba
[1]);
2122 dw
[3] = fui(rgba
[2]);
2123 dw
[4] = fui(rgba
[3]);
2126 dw
[5] = util_float_to_half(rgba
[0]) |
2127 util_float_to_half(rgba
[1]) << 16;
2128 dw
[6] = util_float_to_half(rgba
[2]) |
2129 util_float_to_half(rgba
[3]) << 16;
2131 /* clamp to [-1.0f, 1.0f] */
2132 rgba
[0] = CLAMP(rgba
[0], -1.0f
, 1.0f
);
2133 rgba
[1] = CLAMP(rgba
[1], -1.0f
, 1.0f
);
2134 rgba
[2] = CLAMP(rgba
[2], -1.0f
, 1.0f
);
2135 rgba
[3] = CLAMP(rgba
[3], -1.0f
, 1.0f
);
2138 dw
[9] = (int16_t) util_iround(rgba
[0] * 32767.0f
) |
2139 (int16_t) util_iround(rgba
[1] * 32767.0f
) << 16;
2140 dw
[10] = (int16_t) util_iround(rgba
[2] * 32767.0f
) |
2141 (int16_t) util_iround(rgba
[3] * 32767.0f
) << 16;
2144 dw
[11] = (int8_t) util_iround(rgba
[0] * 127.0f
) |
2145 (int8_t) util_iround(rgba
[1] * 127.0f
) << 8 |
2146 (int8_t) util_iround(rgba
[2] * 127.0f
) << 16 |
2147 (int8_t) util_iround(rgba
[3] * 127.0f
) << 24;
2149 /* clamp to [0.0f, 1.0f] */
2150 rgba
[0] = CLAMP(rgba
[0], 0.0f
, 1.0f
);
2151 rgba
[1] = CLAMP(rgba
[1], 0.0f
, 1.0f
);
2152 rgba
[2] = CLAMP(rgba
[2], 0.0f
, 1.0f
);
2153 rgba
[3] = CLAMP(rgba
[3], 0.0f
, 1.0f
);
2156 dw
[0] = (uint8_t) util_iround(rgba
[0] * 255.0f
) |
2157 (uint8_t) util_iround(rgba
[1] * 255.0f
) << 8 |
2158 (uint8_t) util_iround(rgba
[2] * 255.0f
) << 16 |
2159 (uint8_t) util_iround(rgba
[3] * 255.0f
) << 24;
2162 dw
[7] = (uint16_t) util_iround(rgba
[0] * 65535.0f
) |
2163 (uint16_t) util_iround(rgba
[1] * 65535.0f
) << 16;
2164 dw
[8] = (uint16_t) util_iround(rgba
[2] * 65535.0f
) |
2165 (uint16_t) util_iround(rgba
[3] * 65535.0f
) << 16;
2169 ilo_gpe_init_sampler_cso(const struct ilo_dev_info
*dev
,
2170 const struct pipe_sampler_state
*state
,
2171 struct ilo_sampler_cso
*sampler
)
2173 int mip_filter
, min_filter
, mag_filter
, max_aniso
;
2174 int lod_bias
, max_lod
, min_lod
;
2175 int wrap_s
, wrap_t
, wrap_r
, wrap_cube
;
2176 bool clamp_is_to_edge
;
2177 uint32_t dw0
, dw1
, dw3
;
2179 ILO_GPE_VALID_GEN(dev
, 6, 7);
2181 memset(sampler
, 0, sizeof(*sampler
));
2183 mip_filter
= gen6_translate_tex_mipfilter(state
->min_mip_filter
);
2184 min_filter
= gen6_translate_tex_filter(state
->min_img_filter
);
2185 mag_filter
= gen6_translate_tex_filter(state
->mag_img_filter
);
2187 sampler
->anisotropic
= state
->max_anisotropy
;
2189 if (state
->max_anisotropy
>= 2 && state
->max_anisotropy
<= 16)
2190 max_aniso
= state
->max_anisotropy
/ 2 - 1;
2191 else if (state
->max_anisotropy
> 16)
2192 max_aniso
= BRW_ANISORATIO_16
;
2194 max_aniso
= BRW_ANISORATIO_2
;
2198 * Here is how the hardware calculate per-pixel LOD, from my reading of the
2201 * 1) LOD is set to log2(ratio of texels to pixels) if not specified in
2202 * other ways. The number of texels is measured using level
2204 * 2) Bias is added to LOD.
2205 * 3) LOD is clamped to [MinLod, MaxLod], and the clamped value is
2206 * compared with Base to determine whether magnification or
2207 * minification is needed. (if preclamp is disabled, LOD is compared
2208 * with Base before clamping)
2209 * 4) If magnification is needed, or no mipmapping is requested, LOD is
2210 * set to floor(MinLod).
2211 * 5) LOD is clamped to [0, MIPCnt], and SurfMinLod is added to LOD.
2213 * With Gallium interface, Base is always zero and
2214 * pipe_sampler_view::u.tex.first_level specifies SurfMinLod.
2216 if (dev
->gen
>= ILO_GEN(7)) {
2217 const float scale
= 256.0f
;
2219 /* [-16.0, 16.0) in S4.8 */
2221 (CLAMP(state
->lod_bias
, -16.0f
, 15.9f
) * scale
);
2224 /* [0.0, 14.0] in U4.8 */
2225 max_lod
= (int) (CLAMP(state
->max_lod
, 0.0f
, 14.0f
) * scale
);
2226 min_lod
= (int) (CLAMP(state
->min_lod
, 0.0f
, 14.0f
) * scale
);
2229 const float scale
= 64.0f
;
2231 /* [-16.0, 16.0) in S4.6 */
2233 (CLAMP(state
->lod_bias
, -16.0f
, 15.9f
) * scale
);
2236 /* [0.0, 13.0] in U4.6 */
2237 max_lod
= (int) (CLAMP(state
->max_lod
, 0.0f
, 13.0f
) * scale
);
2238 min_lod
= (int) (CLAMP(state
->min_lod
, 0.0f
, 13.0f
) * scale
);
2242 * We want LOD to be clamped to determine magnification/minification, and
2243 * get set to zero when it is magnification or when mipmapping is disabled.
2244 * The hardware would set LOD to floor(MinLod) and that is a problem when
2245 * MinLod is greater than or equal to 1.0f.
2247 * With Base being zero, it is always minification when MinLod is non-zero.
2248 * To achieve our goal, we just need to set MinLod to zero and set
2249 * MagFilter to MinFilter when mipmapping is disabled.
2251 if (state
->min_mip_filter
== PIPE_TEX_MIPFILTER_NONE
&& min_lod
) {
2253 mag_filter
= min_filter
;
2257 * For nearest filtering, PIPE_TEX_WRAP_CLAMP means
2258 * PIPE_TEX_WRAP_CLAMP_TO_EDGE; for linear filtering, PIPE_TEX_WRAP_CLAMP
2259 * means PIPE_TEX_WRAP_CLAMP_TO_BORDER while additionally clamping the
2260 * texture coordinates to [0.0, 1.0].
2262 * The clamping will be taken care of in the shaders. There are two
2263 * filters here, but let the minification one has a say.
2265 clamp_is_to_edge
= (state
->min_img_filter
== PIPE_TEX_FILTER_NEAREST
);
2266 if (!clamp_is_to_edge
) {
2267 sampler
->saturate_s
= (state
->wrap_s
== PIPE_TEX_WRAP_CLAMP
);
2268 sampler
->saturate_t
= (state
->wrap_t
== PIPE_TEX_WRAP_CLAMP
);
2269 sampler
->saturate_r
= (state
->wrap_r
== PIPE_TEX_WRAP_CLAMP
);
2272 /* determine wrap s/t/r */
2273 wrap_s
= gen6_translate_tex_wrap(state
->wrap_s
, clamp_is_to_edge
);
2274 wrap_t
= gen6_translate_tex_wrap(state
->wrap_t
, clamp_is_to_edge
);
2275 wrap_r
= gen6_translate_tex_wrap(state
->wrap_r
, clamp_is_to_edge
);
2278 * From the Sandy Bridge PRM, volume 4 part 1, page 107:
2280 * "When using cube map texture coordinates, only TEXCOORDMODE_CLAMP
2281 * and TEXCOORDMODE_CUBE settings are valid, and each TC component
2282 * must have the same Address Control mode."
2284 * From the Ivy Bridge PRM, volume 4 part 1, page 96:
2286 * "This field (Cube Surface Control Mode) must be set to
2287 * CUBECTRLMODE_PROGRAMMED"
2289 * Therefore, we cannot use "Cube Surface Control Mode" for semless cube
2292 if (state
->seamless_cube_map
&&
2293 (state
->min_img_filter
!= PIPE_TEX_FILTER_NEAREST
||
2294 state
->mag_img_filter
!= PIPE_TEX_FILTER_NEAREST
)) {
2295 wrap_cube
= BRW_TEXCOORDMODE_CUBE
;
2298 wrap_cube
= BRW_TEXCOORDMODE_CLAMP
;
2301 if (!state
->normalized_coords
) {
2303 * From the Ivy Bridge PRM, volume 4 part 1, page 98:
2305 * "The following state must be set as indicated if this field
2306 * (Non-normalized Coordinate Enable) is enabled:
2308 * - TCX/Y/Z Address Control Mode must be TEXCOORDMODE_CLAMP,
2309 * TEXCOORDMODE_HALF_BORDER, or TEXCOORDMODE_CLAMP_BORDER.
2310 * - Surface Type must be SURFTYPE_2D or SURFTYPE_3D.
2311 * - Mag Mode Filter must be MAPFILTER_NEAREST or
2313 * - Min Mode Filter must be MAPFILTER_NEAREST or
2315 * - Mip Mode Filter must be MIPFILTER_NONE.
2316 * - Min LOD must be 0.
2317 * - Max LOD must be 0.
2318 * - MIP Count must be 0.
2319 * - Surface Min LOD must be 0.
2320 * - Texture LOD Bias must be 0."
2322 assert(wrap_s
== BRW_TEXCOORDMODE_CLAMP
||
2323 wrap_s
== BRW_TEXCOORDMODE_CLAMP_BORDER
);
2324 assert(wrap_t
== BRW_TEXCOORDMODE_CLAMP
||
2325 wrap_t
== BRW_TEXCOORDMODE_CLAMP_BORDER
);
2326 assert(wrap_r
== BRW_TEXCOORDMODE_CLAMP
||
2327 wrap_r
== BRW_TEXCOORDMODE_CLAMP_BORDER
);
2329 assert(mag_filter
== BRW_MAPFILTER_NEAREST
||
2330 mag_filter
== BRW_MAPFILTER_LINEAR
);
2331 assert(min_filter
== BRW_MAPFILTER_NEAREST
||
2332 min_filter
== BRW_MAPFILTER_LINEAR
);
2334 /* work around a bug in util_blitter */
2335 mip_filter
= BRW_MIPFILTER_NONE
;
2337 assert(mip_filter
== BRW_MIPFILTER_NONE
);
2340 if (dev
->gen
>= ILO_GEN(7)) {
2345 sampler
->dw_filter
= mag_filter
<< 17 |
2348 sampler
->dw_filter_aniso
= BRW_MAPFILTER_ANISOTROPIC
<< 17 |
2349 BRW_MAPFILTER_ANISOTROPIC
<< 14 |
2352 dw1
= min_lod
<< 20 |
2355 if (state
->compare_mode
!= PIPE_TEX_COMPARE_NONE
)
2356 dw1
|= gen6_translate_shadow_func(state
->compare_func
) << 1;
2358 dw3
= max_aniso
<< 19;
2360 /* round the coordinates for linear filtering */
2361 if (min_filter
!= BRW_MAPFILTER_NEAREST
) {
2362 dw3
|= (BRW_ADDRESS_ROUNDING_ENABLE_U_MIN
|
2363 BRW_ADDRESS_ROUNDING_ENABLE_V_MIN
|
2364 BRW_ADDRESS_ROUNDING_ENABLE_R_MIN
) << 13;
2366 if (mag_filter
!= BRW_MAPFILTER_NEAREST
) {
2367 dw3
|= (BRW_ADDRESS_ROUNDING_ENABLE_U_MAG
|
2368 BRW_ADDRESS_ROUNDING_ENABLE_V_MAG
|
2369 BRW_ADDRESS_ROUNDING_ENABLE_R_MAG
) << 13;
2372 if (!state
->normalized_coords
)
2375 sampler
->dw_wrap
= wrap_s
<< 6 |
2380 * As noted in the classic i965 driver, the HW may still reference
2381 * wrap_t and wrap_r for 1D textures. We need to set them to a safe
2384 sampler
->dw_wrap_1d
= wrap_s
<< 6 |
2385 BRW_TEXCOORDMODE_WRAP
<< 3 |
2386 BRW_TEXCOORDMODE_WRAP
;
2388 sampler
->dw_wrap_cube
= wrap_cube
<< 6 |
2392 STATIC_ASSERT(Elements(sampler
->payload
) >= 7);
2394 sampler
->payload
[0] = dw0
;
2395 sampler
->payload
[1] = dw1
;
2396 sampler
->payload
[2] = dw3
;
2398 memcpy(&sampler
->payload
[3],
2399 state
->border_color
.ui
, sizeof(state
->border_color
.ui
));
2406 if (state
->compare_mode
!= PIPE_TEX_COMPARE_NONE
)
2407 dw0
|= gen6_translate_shadow_func(state
->compare_func
);
2409 sampler
->dw_filter
= (min_filter
!= mag_filter
) << 27 |
2413 sampler
->dw_filter_aniso
= BRW_MAPFILTER_ANISOTROPIC
<< 17 |
2414 BRW_MAPFILTER_ANISOTROPIC
<< 14;
2416 dw1
= min_lod
<< 22 |
2419 sampler
->dw_wrap
= wrap_s
<< 6 |
2423 sampler
->dw_wrap_1d
= wrap_s
<< 6 |
2424 BRW_TEXCOORDMODE_WRAP
<< 3 |
2425 BRW_TEXCOORDMODE_WRAP
;
2427 sampler
->dw_wrap_cube
= wrap_cube
<< 6 |
2431 dw3
= max_aniso
<< 19;
2433 /* round the coordinates for linear filtering */
2434 if (min_filter
!= BRW_MAPFILTER_NEAREST
) {
2435 dw3
|= (BRW_ADDRESS_ROUNDING_ENABLE_U_MIN
|
2436 BRW_ADDRESS_ROUNDING_ENABLE_V_MIN
|
2437 BRW_ADDRESS_ROUNDING_ENABLE_R_MIN
) << 13;
2439 if (mag_filter
!= BRW_MAPFILTER_NEAREST
) {
2440 dw3
|= (BRW_ADDRESS_ROUNDING_ENABLE_U_MAG
|
2441 BRW_ADDRESS_ROUNDING_ENABLE_V_MAG
|
2442 BRW_ADDRESS_ROUNDING_ENABLE_R_MAG
) << 13;
2445 if (!state
->normalized_coords
)
2448 STATIC_ASSERT(Elements(sampler
->payload
) >= 15);
2450 sampler
->payload
[0] = dw0
;
2451 sampler
->payload
[1] = dw1
;
2452 sampler
->payload
[2] = dw3
;
2454 sampler_init_border_color_gen6(dev
,
2455 &state
->border_color
, &sampler
->payload
[3], 12);
2460 ilo_gpe_gen6_estimate_command_size(const struct ilo_dev_info
*dev
,
2461 enum ilo_gpe_gen6_command cmd
,
2464 static const struct {
2467 } gen6_command_size_table
[ILO_GPE_GEN6_COMMAND_COUNT
] = {
2468 [ILO_GPE_GEN6_STATE_BASE_ADDRESS
] = { 0, 10 },
2469 [ILO_GPE_GEN6_STATE_SIP
] = { 0, 2 },
2470 [ILO_GPE_GEN6_3DSTATE_VF_STATISTICS
] = { 0, 1 },
2471 [ILO_GPE_GEN6_PIPELINE_SELECT
] = { 0, 1 },
2472 [ILO_GPE_GEN6_MEDIA_VFE_STATE
] = { 0, 8 },
2473 [ILO_GPE_GEN6_MEDIA_CURBE_LOAD
] = { 0, 4 },
2474 [ILO_GPE_GEN6_MEDIA_INTERFACE_DESCRIPTOR_LOAD
] = { 0, 4 },
2475 [ILO_GPE_GEN6_MEDIA_GATEWAY_STATE
] = { 0, 2 },
2476 [ILO_GPE_GEN6_MEDIA_STATE_FLUSH
] = { 0, 2 },
2477 [ILO_GPE_GEN6_MEDIA_OBJECT_WALKER
] = { 17, 1 },
2478 [ILO_GPE_GEN6_3DSTATE_BINDING_TABLE_POINTERS
] = { 0, 4 },
2479 [ILO_GPE_GEN6_3DSTATE_SAMPLER_STATE_POINTERS
] = { 0, 4 },
2480 [ILO_GPE_GEN6_3DSTATE_URB
] = { 0, 3 },
2481 [ILO_GPE_GEN6_3DSTATE_VERTEX_BUFFERS
] = { 1, 4 },
2482 [ILO_GPE_GEN6_3DSTATE_VERTEX_ELEMENTS
] = { 1, 2 },
2483 [ILO_GPE_GEN6_3DSTATE_INDEX_BUFFER
] = { 0, 3 },
2484 [ILO_GPE_GEN6_3DSTATE_VIEWPORT_STATE_POINTERS
] = { 0, 4 },
2485 [ILO_GPE_GEN6_3DSTATE_CC_STATE_POINTERS
] = { 0, 4 },
2486 [ILO_GPE_GEN6_3DSTATE_SCISSOR_STATE_POINTERS
] = { 0, 2 },
2487 [ILO_GPE_GEN6_3DSTATE_VS
] = { 0, 6 },
2488 [ILO_GPE_GEN6_3DSTATE_GS
] = { 0, 7 },
2489 [ILO_GPE_GEN6_3DSTATE_CLIP
] = { 0, 4 },
2490 [ILO_GPE_GEN6_3DSTATE_SF
] = { 0, 20 },
2491 [ILO_GPE_GEN6_3DSTATE_WM
] = { 0, 9 },
2492 [ILO_GPE_GEN6_3DSTATE_CONSTANT_VS
] = { 0, 5 },
2493 [ILO_GPE_GEN6_3DSTATE_CONSTANT_GS
] = { 0, 5 },
2494 [ILO_GPE_GEN6_3DSTATE_CONSTANT_PS
] = { 0, 5 },
2495 [ILO_GPE_GEN6_3DSTATE_SAMPLE_MASK
] = { 0, 2 },
2496 [ILO_GPE_GEN6_3DSTATE_DRAWING_RECTANGLE
] = { 0, 4 },
2497 [ILO_GPE_GEN6_3DSTATE_DEPTH_BUFFER
] = { 0, 7 },
2498 [ILO_GPE_GEN6_3DSTATE_POLY_STIPPLE_OFFSET
] = { 0, 2 },
2499 [ILO_GPE_GEN6_3DSTATE_POLY_STIPPLE_PATTERN
] = { 0, 33 },
2500 [ILO_GPE_GEN6_3DSTATE_LINE_STIPPLE
] = { 0, 3 },
2501 [ILO_GPE_GEN6_3DSTATE_AA_LINE_PARAMETERS
] = { 0, 3 },
2502 [ILO_GPE_GEN6_3DSTATE_GS_SVB_INDEX
] = { 0, 4 },
2503 [ILO_GPE_GEN6_3DSTATE_MULTISAMPLE
] = { 0, 3 },
2504 [ILO_GPE_GEN6_3DSTATE_STENCIL_BUFFER
] = { 0, 3 },
2505 [ILO_GPE_GEN6_3DSTATE_HIER_DEPTH_BUFFER
] = { 0, 3 },
2506 [ILO_GPE_GEN6_3DSTATE_CLEAR_PARAMS
] = { 0, 2 },
2507 [ILO_GPE_GEN6_PIPE_CONTROL
] = { 0, 5 },
2508 [ILO_GPE_GEN6_3DPRIMITIVE
] = { 0, 6 },
2510 const int header
= gen6_command_size_table
[cmd
].header
;
2511 const int body
= gen6_command_size_table
[arg
].body
;
2512 const int count
= arg
;
2514 ILO_GPE_VALID_GEN(dev
, 6, 6);
2515 assert(cmd
< ILO_GPE_GEN6_COMMAND_COUNT
);
2517 return (likely(count
)) ? header
+ body
* count
: 0;
2521 ilo_gpe_gen6_estimate_state_size(const struct ilo_dev_info
*dev
,
2522 enum ilo_gpe_gen6_state state
,
2525 static const struct {
2529 } gen6_state_size_table
[ILO_GPE_GEN6_STATE_COUNT
] = {
2530 [ILO_GPE_GEN6_INTERFACE_DESCRIPTOR_DATA
] = { 8, 8, true },
2531 [ILO_GPE_GEN6_SF_VIEWPORT
] = { 8, 8, true },
2532 [ILO_GPE_GEN6_CLIP_VIEWPORT
] = { 8, 4, true },
2533 [ILO_GPE_GEN6_CC_VIEWPORT
] = { 8, 2, true },
2534 [ILO_GPE_GEN6_COLOR_CALC_STATE
] = { 16, 6, false },
2535 [ILO_GPE_GEN6_BLEND_STATE
] = { 16, 2, true },
2536 [ILO_GPE_GEN6_DEPTH_STENCIL_STATE
] = { 16, 3, false },
2537 [ILO_GPE_GEN6_SCISSOR_RECT
] = { 8, 2, true },
2538 [ILO_GPE_GEN6_BINDING_TABLE_STATE
] = { 8, 1, true },
2539 [ILO_GPE_GEN6_SURFACE_STATE
] = { 8, 6, false },
2540 [ILO_GPE_GEN6_SAMPLER_STATE
] = { 8, 4, true },
2541 [ILO_GPE_GEN6_SAMPLER_BORDER_COLOR_STATE
] = { 8, 12, false },
2542 [ILO_GPE_GEN6_PUSH_CONSTANT_BUFFER
] = { 8, 1, true },
2544 const int alignment
= gen6_state_size_table
[state
].alignment
;
2545 const int body
= gen6_state_size_table
[state
].body
;
2546 const bool is_array
= gen6_state_size_table
[state
].is_array
;
2547 const int count
= arg
;
2550 ILO_GPE_VALID_GEN(dev
, 6, 6);
2551 assert(state
< ILO_GPE_GEN6_STATE_COUNT
);
2553 if (likely(count
)) {
2555 estimate
= (alignment
- 1) + body
* count
;
2558 estimate
= (alignment
- 1) + body
;
2559 /* all states are aligned */
2561 estimate
+= util_align_npot(body
, alignment
) * (count
- 1);