2 * Mesa 3-D graphics library
4 * Copyright (C) 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 "brw_defines.h"
29 #include "intel_reg.h"
32 #include "ilo_format.h"
33 #include "ilo_resource.h"
34 #include "ilo_shader.h"
35 #include "ilo_gpe_gen7.h"
38 gen7_emit_GPGPU_WALKER(const struct ilo_dev_info
*dev
,
41 assert(!"GPGPU_WALKER unsupported");
45 gen7_emit_3DSTATE_CLEAR_PARAMS(const struct ilo_dev_info
*dev
,
49 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x04);
50 const uint8_t cmd_len
= 3;
52 ILO_GPE_VALID_GEN(dev
, 7, 7);
54 ilo_cp_begin(cp
, cmd_len
);
55 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
56 ilo_cp_write(cp
, clear_val
);
62 gen7_emit_3DSTATE_DEPTH_BUFFER(const struct ilo_dev_info
*dev
,
63 const struct pipe_surface
*surface
,
64 const struct pipe_depth_stencil_alpha_state
*dsa
,
68 ilo_gpe_gen6_emit_3DSTATE_DEPTH_BUFFER(dev
, surface
, dsa
, hiz
, cp
);
72 gen7_emit_3dstate_pointer(const struct ilo_dev_info
*dev
,
73 int subop
, uint32_t pointer
,
76 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, subop
);
77 const uint8_t cmd_len
= 2;
79 ILO_GPE_VALID_GEN(dev
, 7, 7);
81 ilo_cp_begin(cp
, cmd_len
);
82 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
83 ilo_cp_write(cp
, pointer
);
88 gen7_emit_3DSTATE_CC_STATE_POINTERS(const struct ilo_dev_info
*dev
,
89 uint32_t color_calc_state
,
92 gen7_emit_3dstate_pointer(dev
, 0x0e, color_calc_state
, cp
);
96 gen7_emit_3DSTATE_GS(const struct ilo_dev_info
*dev
,
97 const struct ilo_shader
*gs
,
101 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x11);
102 const uint8_t cmd_len
= 7;
103 uint32_t dw2
, dw4
, dw5
;
106 ILO_GPE_VALID_GEN(dev
, 7, 7);
110 max_threads
= (dev
->gt
== 2) ? 128 : 36;
118 ilo_cp_begin(cp
, cmd_len
);
119 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
124 ilo_cp_write(cp
, GEN6_GS_STATISTICS_ENABLE
);
130 dw2
= ((num_samplers
+ 3) / 4) << GEN6_GS_SAMPLER_COUNT_SHIFT
;
132 dw4
= ((gs
->in
.count
+ 1) / 2) << GEN6_GS_URB_READ_LENGTH_SHIFT
|
133 GEN7_GS_INCLUDE_VERTEX_HANDLES
|
134 0 << GEN6_GS_URB_ENTRY_READ_OFFSET_SHIFT
|
135 gs
->in
.start_grf
<< GEN6_GS_DISPATCH_START_GRF_SHIFT
;
137 dw5
= (max_threads
- 1) << GEN6_GS_MAX_THREADS_SHIFT
|
138 GEN6_GS_STATISTICS_ENABLE
|
141 ilo_cp_begin(cp
, cmd_len
);
142 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
143 ilo_cp_write(cp
, gs
->cache_offset
);
144 ilo_cp_write(cp
, dw2
);
145 ilo_cp_write(cp
, 0); /* scratch */
146 ilo_cp_write(cp
, dw4
);
147 ilo_cp_write(cp
, dw5
);
153 gen7_emit_3DSTATE_SF(const struct ilo_dev_info
*dev
,
154 const struct pipe_rasterizer_state
*rasterizer
,
155 const struct pipe_surface
*zs_surf
,
158 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x13);
159 const uint8_t cmd_len
= 7;
162 ILO_GPE_VALID_GEN(dev
, 7, 7);
164 ilo_gpe_gen6_fill_3dstate_sf_raster(dev
, rasterizer
,
165 1, (zs_surf
) ? zs_surf
->format
: PIPE_FORMAT_NONE
, true,
168 ilo_cp_begin(cp
, cmd_len
);
169 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
170 ilo_cp_write_multi(cp
, dw
, 6);
175 gen7_emit_3DSTATE_WM(const struct ilo_dev_info
*dev
,
176 const struct ilo_shader
*fs
,
177 const struct pipe_rasterizer_state
*rasterizer
,
181 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x14);
182 const uint8_t cmd_len
= 3;
183 const int num_samples
= 1;
186 ILO_GPE_VALID_GEN(dev
, 7, 7);
188 dw1
= GEN7_WM_STATISTICS_ENABLE
|
189 GEN7_WM_LINE_AA_WIDTH_2_0
;
192 dw1
|= GEN7_WM_DEPTH_CLEAR
;
193 dw1
|= GEN7_WM_DEPTH_RESOLVE
;
194 dw1
|= GEN7_WM_HIERARCHICAL_DEPTH_RESOLVE
;
201 * a) fs writes colors and color is not masked, or
202 * b) fs writes depth, or
205 dw1
|= GEN7_WM_DISPATCH_ENABLE
;
208 * From the Ivy Bridge PRM, volume 2 part 1, page 278:
210 * "This bit (Pixel Shader Kill Pixel), if ENABLED, indicates that
211 * the PS kernel or color calculator has the ability to kill
212 * (discard) pixels or samples, other than due to depth or stencil
213 * testing. This bit is required to be ENABLED in the following
216 * - The API pixel shader program contains "killpix" or "discard"
217 * instructions, or other code in the pixel shader kernel that
218 * can cause the final pixel mask to differ from the pixel mask
219 * received on dispatch.
221 * - A sampler with chroma key enabled with kill pixel mode is used
222 * by the pixel shader.
224 * - Any render target has Alpha Test Enable or AlphaToCoverage
227 * - The pixel shader kernel generates and outputs oMask.
229 * Note: As ClipDistance clipping is fully supported in hardware
230 * and therefore not via PS instructions, there should be no need
231 * to ENABLE this bit due to ClipDistance clipping."
233 if (fs
->has_kill
|| cc_may_kill
)
234 dw1
|= GEN7_WM_KILL_ENABLE
;
237 dw1
|= GEN7_WM_PSCDEPTH_ON
;
239 dw1
|= GEN7_WM_USES_SOURCE_DEPTH
| GEN7_WM_USES_SOURCE_W
;
241 dw1
|= fs
->in
.barycentric_interpolation_mode
<<
242 GEN7_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT
;
244 else if (cc_may_kill
) {
245 dw1
|= GEN7_WM_DISPATCH_ENABLE
|
249 dw1
|= GEN7_WM_POSITION_ZW_PIXEL
;
251 /* same value as in 3DSTATE_SF */
252 if (rasterizer
->line_smooth
)
253 dw1
|= GEN7_WM_LINE_END_CAP_AA_WIDTH_1_0
;
255 if (rasterizer
->poly_stipple_enable
)
256 dw1
|= GEN7_WM_POLYGON_STIPPLE_ENABLE
;
257 if (rasterizer
->line_stipple_enable
)
258 dw1
|= GEN7_WM_LINE_STIPPLE_ENABLE
;
260 if (rasterizer
->bottom_edge_rule
)
261 dw1
|= GEN7_WM_POINT_RASTRULE_UPPER_RIGHT
;
263 if (num_samples
> 1) {
264 if (rasterizer
->multisample
)
265 dw1
|= GEN7_WM_MSRAST_ON_PATTERN
;
267 dw1
|= GEN7_WM_MSRAST_OFF_PIXEL
;
269 dw2
= GEN7_WM_MSDISPMODE_PERPIXEL
;
272 dw1
|= GEN7_WM_MSRAST_OFF_PIXEL
;
274 dw2
= GEN7_WM_MSDISPMODE_PERSAMPLE
;
277 ilo_cp_begin(cp
, cmd_len
);
278 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
279 ilo_cp_write(cp
, dw1
);
280 ilo_cp_write(cp
, dw2
);
285 gen7_emit_3dstate_constant(const struct ilo_dev_info
*dev
,
287 const uint32_t *bufs
, const int *sizes
,
291 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, subop
);
292 const uint8_t cmd_len
= 7;
294 int total_read_length
, i
;
296 ILO_GPE_VALID_GEN(dev
, 7, 7);
298 /* VS, HS, DS, GS, and PS variants */
299 assert(subop
>= 0x15 && subop
<= 0x1a && subop
!= 0x18);
301 assert(num_bufs
<= 4);
306 total_read_length
= 0;
307 for (i
= 0; i
< 4; i
++) {
311 * From the Ivy Bridge PRM, volume 2 part 1, page 112:
313 * "Constant buffers must be enabled in order from Constant Buffer 0
314 * to Constant Buffer 3 within this command. For example, it is
315 * not allowed to enable Constant Buffer 1 by programming a
316 * non-zero value in the VS Constant Buffer 1 Read Length without a
317 * non-zero value in VS Constant Buffer 0 Read Length."
319 if (i
>= num_bufs
|| !sizes
[i
]) {
321 assert(i
>= num_bufs
|| !sizes
[i
]);
327 /* read lengths are in 256-bit units */
328 read_len
= (sizes
[i
] + 31) / 32;
329 /* the lower 5 bits are used for memory object control state */
330 assert(bufs
[i
] % 32 == 0);
332 dw
[i
/ 2] |= read_len
<< ((i
% 2) ? 16 : 0);
335 total_read_length
+= read_len
;
339 * From the Ivy Bridge PRM, volume 2 part 1, page 113:
341 * "The sum of all four read length fields must be less than or equal
344 assert(total_read_length
<= 64);
346 ilo_cp_begin(cp
, cmd_len
);
347 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
348 ilo_cp_write_multi(cp
, dw
, 6);
353 gen7_emit_3DSTATE_CONSTANT_VS(const struct ilo_dev_info
*dev
,
354 const uint32_t *bufs
, const int *sizes
,
358 gen7_emit_3dstate_constant(dev
, 0x15, bufs
, sizes
, num_bufs
, cp
);
362 gen7_emit_3DSTATE_CONSTANT_GS(const struct ilo_dev_info
*dev
,
363 const uint32_t *bufs
, const int *sizes
,
367 gen7_emit_3dstate_constant(dev
, 0x16, bufs
, sizes
, num_bufs
, cp
);
371 gen7_emit_3DSTATE_CONSTANT_PS(const struct ilo_dev_info
*dev
,
372 const uint32_t *bufs
, const int *sizes
,
376 gen7_emit_3dstate_constant(dev
, 0x17, bufs
, sizes
, num_bufs
, cp
);
380 gen7_emit_3DSTATE_SAMPLE_MASK(const struct ilo_dev_info
*dev
,
381 unsigned sample_mask
,
385 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x18);
386 const uint8_t cmd_len
= 2;
387 const unsigned valid_mask
= ((1 << num_samples
) - 1) | 0x1;
389 ILO_GPE_VALID_GEN(dev
, 7, 7);
392 * From the Ivy Bridge PRM, volume 2 part 1, page 294:
394 * "If Number of Multisamples is NUMSAMPLES_1, bits 7:1 of this field
395 * (Sample Mask) must be zero.
397 * If Number of Multisamples is NUMSAMPLES_4, bits 7:4 of this field
400 sample_mask
&= valid_mask
;
402 ilo_cp_begin(cp
, cmd_len
);
403 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
404 ilo_cp_write(cp
, sample_mask
);
409 gen7_emit_3DSTATE_CONSTANT_HS(const struct ilo_dev_info
*dev
,
410 const uint32_t *bufs
, const int *sizes
,
414 gen7_emit_3dstate_constant(dev
, 0x19, bufs
, sizes
, num_bufs
, cp
);
418 gen7_emit_3DSTATE_CONSTANT_DS(const struct ilo_dev_info
*dev
,
419 const uint32_t *bufs
, const int *sizes
,
423 gen7_emit_3dstate_constant(dev
, 0x1a, bufs
, sizes
, num_bufs
, cp
);
427 gen7_emit_3DSTATE_HS(const struct ilo_dev_info
*dev
,
428 const struct ilo_shader
*hs
,
429 int max_threads
, int num_samplers
,
432 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x1b);
433 const uint8_t cmd_len
= 7;
434 uint32_t dw1
, dw2
, dw5
;
436 ILO_GPE_VALID_GEN(dev
, 7, 7);
439 ilo_cp_begin(cp
, cmd_len
);
440 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
452 dw1
= (num_samplers
+ 3) / 4 << 27 |
458 dw2
= 1 << 31 | /* HS Enable */
459 1 << 29 | /* HS Statistics Enable */
460 0; /* Instance Count */
462 dw5
= hs
->in
.start_grf
<< 19 |
466 ilo_cp_begin(cp
, cmd_len
);
467 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
468 ilo_cp_write(cp
, dw1
);
469 ilo_cp_write(cp
, dw2
);
470 ilo_cp_write(cp
, hs
->cache_offset
);
472 ilo_cp_write(cp
, dw5
);
478 gen7_emit_3DSTATE_TE(const struct ilo_dev_info
*dev
,
481 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x1c);
482 const uint8_t cmd_len
= 4;
484 ILO_GPE_VALID_GEN(dev
, 7, 7);
486 ilo_cp_begin(cp
, cmd_len
);
487 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
495 gen7_emit_3DSTATE_DS(const struct ilo_dev_info
*dev
,
496 const struct ilo_shader
*ds
,
497 int max_threads
, int num_samplers
,
500 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x1d);
501 const uint8_t cmd_len
= 6;
502 uint32_t dw2
, dw4
, dw5
;
504 ILO_GPE_VALID_GEN(dev
, 7, 7);
507 ilo_cp_begin(cp
, cmd_len
);
508 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
519 dw2
= (num_samplers
+ 3) / 4 << 27 |
525 dw4
= ds
->in
.start_grf
<< 20 |
529 dw5
= (max_threads
- 1) << 25 |
533 ilo_cp_begin(cp
, cmd_len
);
534 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
535 ilo_cp_write(cp
, ds
->cache_offset
);
536 ilo_cp_write(cp
, dw2
);
538 ilo_cp_write(cp
, dw4
);
539 ilo_cp_write(cp
, dw5
);
544 gen7_emit_3DSTATE_STREAMOUT(const struct ilo_dev_info
*dev
,
545 unsigned buffer_mask
,
546 int vertex_attrib_count
,
547 bool rasterizer_discard
,
550 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x1e);
551 const uint8_t cmd_len
= 3;
552 const bool enable
= (buffer_mask
!= 0);
556 ILO_GPE_VALID_GEN(dev
, 7, 7);
559 dw1
= 0 << SO_RENDER_STREAM_SELECT_SHIFT
;
560 if (rasterizer_discard
)
561 dw1
|= SO_RENDERING_DISABLE
;
565 ilo_cp_begin(cp
, cmd_len
);
566 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
567 ilo_cp_write(cp
, dw1
);
568 ilo_cp_write(cp
, dw2
);
573 read_len
= (vertex_attrib_count
+ 1) / 2;
577 dw1
= SO_FUNCTION_ENABLE
|
578 0 << SO_RENDER_STREAM_SELECT_SHIFT
|
579 SO_STATISTICS_ENABLE
|
582 if (rasterizer_discard
)
583 dw1
|= SO_RENDERING_DISABLE
;
587 dw1
|= SO_REORDER_TRAILING
;
589 dw2
= 0 << SO_STREAM_3_VERTEX_READ_OFFSET_SHIFT
|
590 0 << SO_STREAM_3_VERTEX_READ_LENGTH_SHIFT
|
591 0 << SO_STREAM_2_VERTEX_READ_OFFSET_SHIFT
|
592 0 << SO_STREAM_2_VERTEX_READ_LENGTH_SHIFT
|
593 0 << SO_STREAM_1_VERTEX_READ_OFFSET_SHIFT
|
594 0 << SO_STREAM_1_VERTEX_READ_LENGTH_SHIFT
|
595 0 << SO_STREAM_0_VERTEX_READ_OFFSET_SHIFT
|
596 (read_len
- 1) << SO_STREAM_0_VERTEX_READ_LENGTH_SHIFT
;
598 ilo_cp_begin(cp
, cmd_len
);
599 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
600 ilo_cp_write(cp
, dw1
);
601 ilo_cp_write(cp
, dw2
);
606 gen7_emit_3DSTATE_SBE(const struct ilo_dev_info
*dev
,
607 const struct pipe_rasterizer_state
*rasterizer
,
608 const struct ilo_shader
*fs
,
609 const struct ilo_shader
*last_sh
,
612 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x1f);
613 const uint8_t cmd_len
= 14;
616 ILO_GPE_VALID_GEN(dev
, 7, 7);
618 ilo_gpe_gen6_fill_3dstate_sf_sbe(dev
, rasterizer
,
619 fs
, last_sh
, dw
, Elements(dw
));
621 ilo_cp_begin(cp
, cmd_len
);
622 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
623 ilo_cp_write_multi(cp
, dw
, 13);
628 gen7_emit_3DSTATE_PS(const struct ilo_dev_info
*dev
,
629 const struct ilo_shader
*fs
,
630 int num_samplers
, bool dual_blend
,
633 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, 0x20);
634 const uint8_t cmd_len
= 8;
635 uint32_t dw2
, dw4
, dw5
;
638 ILO_GPE_VALID_GEN(dev
, 7, 7);
640 /* see brwCreateContext() */
641 max_threads
= (dev
->gt
== 2) ? 172 : 48;
644 ilo_cp_begin(cp
, cmd_len
);
645 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
649 /* GPU hangs if none of the dispatch enable bits is set */
650 ilo_cp_write(cp
, (max_threads
- 1) << IVB_PS_MAX_THREADS_SHIFT
|
651 GEN7_PS_8_DISPATCH_ENABLE
);
660 dw2
= (num_samplers
+ 3) / 4 << GEN7_PS_SAMPLER_COUNT_SHIFT
|
661 0 << GEN7_PS_BINDING_TABLE_ENTRY_COUNT_SHIFT
;
663 dw2
|= GEN7_PS_FLOATING_POINT_MODE_ALT
;
665 dw4
= (max_threads
- 1) << IVB_PS_MAX_THREADS_SHIFT
|
666 GEN7_PS_POSOFFSET_NONE
;
669 dw4
|= GEN7_PS_PUSH_CONSTANT_ENABLE
;
671 dw4
|= GEN7_PS_ATTRIBUTE_ENABLE
;
673 dw4
|= GEN7_PS_DUAL_SOURCE_BLEND_ENABLE
;
676 dw4
|= GEN7_PS_16_DISPATCH_ENABLE
;
678 dw4
|= GEN7_PS_8_DISPATCH_ENABLE
;
680 dw5
= fs
->in
.start_grf
<< GEN7_PS_DISPATCH_START_GRF_SHIFT_0
|
681 0 << GEN7_PS_DISPATCH_START_GRF_SHIFT_1
|
682 0 << GEN7_PS_DISPATCH_START_GRF_SHIFT_2
;
684 ilo_cp_begin(cp
, cmd_len
);
685 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
686 ilo_cp_write(cp
, fs
->cache_offset
);
687 ilo_cp_write(cp
, dw2
);
688 ilo_cp_write(cp
, 0); /* scratch */
689 ilo_cp_write(cp
, dw4
);
690 ilo_cp_write(cp
, dw5
);
691 ilo_cp_write(cp
, 0); /* kernel 1 */
692 ilo_cp_write(cp
, 0); /* kernel 2 */
697 gen7_emit_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP(const struct ilo_dev_info
*dev
,
698 uint32_t sf_clip_viewport
,
701 gen7_emit_3dstate_pointer(dev
, 0x21, sf_clip_viewport
, cp
);
705 gen7_emit_3DSTATE_VIEWPORT_STATE_POINTERS_CC(const struct ilo_dev_info
*dev
,
706 uint32_t cc_viewport
,
709 gen7_emit_3dstate_pointer(dev
, 0x23, cc_viewport
, cp
);
713 gen7_emit_3DSTATE_BLEND_STATE_POINTERS(const struct ilo_dev_info
*dev
,
714 uint32_t blend_state
,
717 gen7_emit_3dstate_pointer(dev
, 0x24, blend_state
, cp
);
721 gen7_emit_3DSTATE_DEPTH_STENCIL_STATE_POINTERS(const struct ilo_dev_info
*dev
,
722 uint32_t depth_stencil_state
,
725 gen7_emit_3dstate_pointer(dev
, 0x25, depth_stencil_state
, cp
);
729 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_VS(const struct ilo_dev_info
*dev
,
730 uint32_t binding_table
,
733 gen7_emit_3dstate_pointer(dev
, 0x26, binding_table
, cp
);
737 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_HS(const struct ilo_dev_info
*dev
,
738 uint32_t binding_table
,
741 gen7_emit_3dstate_pointer(dev
, 0x27, binding_table
, cp
);
745 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_DS(const struct ilo_dev_info
*dev
,
746 uint32_t binding_table
,
749 gen7_emit_3dstate_pointer(dev
, 0x28, binding_table
, cp
);
753 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_GS(const struct ilo_dev_info
*dev
,
754 uint32_t binding_table
,
757 gen7_emit_3dstate_pointer(dev
, 0x29, binding_table
, cp
);
761 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_PS(const struct ilo_dev_info
*dev
,
762 uint32_t binding_table
,
765 gen7_emit_3dstate_pointer(dev
, 0x2a, binding_table
, cp
);
769 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_VS(const struct ilo_dev_info
*dev
,
770 uint32_t sampler_state
,
773 gen7_emit_3dstate_pointer(dev
, 0x2b, sampler_state
, cp
);
777 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_HS(const struct ilo_dev_info
*dev
,
778 uint32_t sampler_state
,
781 gen7_emit_3dstate_pointer(dev
, 0x2c, sampler_state
, cp
);
785 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_DS(const struct ilo_dev_info
*dev
,
786 uint32_t sampler_state
,
789 gen7_emit_3dstate_pointer(dev
, 0x2d, sampler_state
, cp
);
793 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_GS(const struct ilo_dev_info
*dev
,
794 uint32_t sampler_state
,
797 gen7_emit_3dstate_pointer(dev
, 0x2e, sampler_state
, cp
);
801 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_PS(const struct ilo_dev_info
*dev
,
802 uint32_t sampler_state
,
805 gen7_emit_3dstate_pointer(dev
, 0x2f, sampler_state
, cp
);
809 gen7_emit_3dstate_urb(const struct ilo_dev_info
*dev
,
810 int subop
, int offset
, int size
,
814 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x0, subop
);
815 const uint8_t cmd_len
= 2;
816 const int row_size
= 64; /* 512 bits */
817 int alloc_size
, num_entries
, min_entries
, max_entries
;
819 ILO_GPE_VALID_GEN(dev
, 7, 7);
821 /* VS, HS, DS, and GS variants */
822 assert(subop
>= 0x30 && subop
<= 0x33);
824 /* in multiples of 8KB */
825 assert(offset
% 8192 == 0);
828 /* in multiple of 512-bit rows */
829 alloc_size
= (entry_size
+ row_size
- 1) / row_size
;
834 * From the Ivy Bridge PRM, volume 2 part 1, page 34:
836 * "VS URB Entry Allocation Size equal to 4(5 512-bit URB rows) may
837 * cause performance to decrease due to banking in the URB. Element
838 * sizes of 16 to 20 should be programmed with six 512-bit URB rows."
840 if (subop
== 0x30 && alloc_size
== 5)
843 /* in multiples of 8 */
844 num_entries
= (size
/ row_size
/ alloc_size
) & ~7;
847 case 0x30: /* 3DSTATE_URB_VS */
849 max_entries
= (dev
->gt
== 2) ? 704 : 512;
851 assert(num_entries
>= min_entries
);
852 if (num_entries
> max_entries
)
853 num_entries
= max_entries
;
855 case 0x31: /* 3DSTATE_URB_HS */
856 max_entries
= (dev
->gt
== 2) ? 64 : 32;
857 if (num_entries
> max_entries
)
858 num_entries
= max_entries
;
860 case 0x32: /* 3DSTATE_URB_DS */
862 assert(num_entries
>= 138);
864 case 0x33: /* 3DSTATE_URB_GS */
865 max_entries
= (dev
->gt
== 2) ? 320 : 192;
866 if (num_entries
> max_entries
)
867 num_entries
= max_entries
;
873 ilo_cp_begin(cp
, cmd_len
);
874 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
875 ilo_cp_write(cp
, offset
<< GEN7_URB_STARTING_ADDRESS_SHIFT
|
876 (alloc_size
- 1) << GEN7_URB_ENTRY_SIZE_SHIFT
|
882 gen7_emit_3DSTATE_URB_VS(const struct ilo_dev_info
*dev
,
883 int offset
, int size
, int entry_size
,
886 gen7_emit_3dstate_urb(dev
, 0x30, offset
, size
, entry_size
, cp
);
890 gen7_emit_3DSTATE_URB_HS(const struct ilo_dev_info
*dev
,
891 int offset
, int size
, int entry_size
,
894 gen7_emit_3dstate_urb(dev
, 0x31, offset
, size
, entry_size
, cp
);
898 gen7_emit_3DSTATE_URB_DS(const struct ilo_dev_info
*dev
,
899 int offset
, int size
, int entry_size
,
902 gen7_emit_3dstate_urb(dev
, 0x32, offset
, size
, entry_size
, cp
);
906 gen7_emit_3DSTATE_URB_GS(const struct ilo_dev_info
*dev
,
907 int offset
, int size
, int entry_size
,
910 gen7_emit_3dstate_urb(dev
, 0x33, offset
, size
, entry_size
, cp
);
914 gen7_emit_3dstate_push_constant_alloc(const struct ilo_dev_info
*dev
,
915 int subop
, int offset
, int size
,
918 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x1, subop
);
919 const uint8_t cmd_len
= 2;
922 ILO_GPE_VALID_GEN(dev
, 7, 7);
924 /* VS, HS, DS, GS, and PS variants */
925 assert(subop
>= 0x12 && subop
<= 0x16);
928 * From the Ivy Bridge PRM, volume 2 part 1, page 68:
930 * "(A table that says the maximum size of each constant buffer is
933 * From the Ivy Bridge PRM, volume 2 part 1, page 115:
935 * "The sum of the Constant Buffer Offset and the Constant Buffer Size
936 * may not exceed the maximum value of the Constant Buffer Size."
938 * Thus, the valid range of buffer end is [0KB, 16KB].
940 end
= (offset
+ size
) / 1024;
942 assert(!"invalid constant buffer end");
946 /* the valid range of buffer offset is [0KB, 15KB] */
947 offset
= (offset
+ 1023) / 1024;
949 assert(!"invalid constant buffer offset");
958 /* the valid range of buffer size is [0KB, 15KB] */
961 assert(!"invalid constant buffer size");
965 ilo_cp_begin(cp
, cmd_len
);
966 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
967 ilo_cp_write(cp
, offset
<< GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT
|
973 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_VS(const struct ilo_dev_info
*dev
,
974 int offset
, int size
,
977 gen7_emit_3dstate_push_constant_alloc(dev
, 0x12, offset
, size
, cp
);
981 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_HS(const struct ilo_dev_info
*dev
,
982 int offset
, int size
,
985 gen7_emit_3dstate_push_constant_alloc(dev
, 0x13, offset
, size
, cp
);
989 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_DS(const struct ilo_dev_info
*dev
,
990 int offset
, int size
,
993 gen7_emit_3dstate_push_constant_alloc(dev
, 0x14, offset
, size
, cp
);
997 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_GS(const struct ilo_dev_info
*dev
,
998 int offset
, int size
,
1001 gen7_emit_3dstate_push_constant_alloc(dev
, 0x15, offset
, size
, cp
);
1005 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_PS(const struct ilo_dev_info
*dev
,
1006 int offset
, int size
,
1009 gen7_emit_3dstate_push_constant_alloc(dev
, 0x16, offset
, size
, cp
);
1013 gen7_emit_3DSTATE_SO_DECL_LIST(const struct ilo_dev_info
*dev
,
1014 const struct pipe_stream_output_info
*so_info
,
1015 const struct ilo_shader
*sh
,
1018 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x1, 0x17);
1020 int buffer_selects
, num_entries
, i
;
1021 uint16_t so_decls
[128];
1023 ILO_GPE_VALID_GEN(dev
, 7, 7);
1029 int buffer_offsets
[PIPE_MAX_SO_BUFFERS
];
1031 memset(buffer_offsets
, 0, sizeof(buffer_offsets
));
1033 for (i
= 0; i
< so_info
->num_outputs
; i
++) {
1034 unsigned decl
, buf
, attr
, mask
;
1036 buf
= so_info
->output
[i
].output_buffer
;
1038 /* pad with holes */
1039 assert(buffer_offsets
[buf
] <= so_info
->output
[i
].dst_offset
);
1040 while (buffer_offsets
[buf
] < so_info
->output
[i
].dst_offset
) {
1043 num_dwords
= so_info
->output
[i
].dst_offset
- buffer_offsets
[buf
];
1047 decl
= buf
<< SO_DECL_OUTPUT_BUFFER_SLOT_SHIFT
|
1049 ((1 << num_dwords
) - 1) << SO_DECL_COMPONENT_MASK_SHIFT
;
1051 so_decls
[num_entries
++] = decl
;
1052 buffer_offsets
[buf
] += num_dwords
;
1055 /* figure out which attribute is sourced */
1056 for (attr
= 0; attr
< sh
->out
.count
; attr
++) {
1057 const int idx
= sh
->out
.register_indices
[attr
];
1058 if (idx
== so_info
->output
[i
].register_index
)
1062 decl
= buf
<< SO_DECL_OUTPUT_BUFFER_SLOT_SHIFT
;
1064 if (attr
< sh
->out
.count
) {
1065 mask
= ((1 << so_info
->output
[i
].num_components
) - 1) <<
1066 so_info
->output
[i
].start_component
;
1068 /* PSIZE is at W channel */
1069 if (sh
->out
.semantic_names
[attr
] == TGSI_SEMANTIC_PSIZE
) {
1070 assert(mask
== 0x1);
1071 mask
= (mask
<< 3) & 0xf;
1074 decl
|= attr
<< SO_DECL_REGISTER_INDEX_SHIFT
|
1075 mask
<< SO_DECL_COMPONENT_MASK_SHIFT
;
1078 assert(!"stream output an undefined register");
1079 mask
= (1 << so_info
->output
[i
].num_components
) - 1;
1080 decl
|= SO_DECL_HOLE_FLAG
|
1081 mask
<< SO_DECL_COMPONENT_MASK_SHIFT
;
1084 so_decls
[num_entries
++] = decl
;
1085 buffer_selects
|= 1 << buf
;
1086 buffer_offsets
[buf
] += so_info
->output
[i
].num_components
;
1091 * From the Ivy Bridge PRM, volume 2 part 1, page 201:
1093 * "Errata: All 128 decls for all four streams must be included
1094 * whenever this command is issued. The "Num Entries [n]" fields still
1095 * contain the actual numbers of valid decls."
1097 * Also note that "DWord Length" has 9 bits for this command, and the type
1098 * of cmd_len is thus uint16_t.
1100 cmd_len
= 2 * 128 + 3;
1102 ilo_cp_begin(cp
, cmd_len
);
1103 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
1104 ilo_cp_write(cp
, 0 << SO_STREAM_TO_BUFFER_SELECTS_3_SHIFT
|
1105 0 << SO_STREAM_TO_BUFFER_SELECTS_2_SHIFT
|
1106 0 << SO_STREAM_TO_BUFFER_SELECTS_1_SHIFT
|
1107 buffer_selects
<< SO_STREAM_TO_BUFFER_SELECTS_0_SHIFT
);
1108 ilo_cp_write(cp
, 0 << SO_NUM_ENTRIES_3_SHIFT
|
1109 0 << SO_NUM_ENTRIES_2_SHIFT
|
1110 0 << SO_NUM_ENTRIES_1_SHIFT
|
1111 num_entries
<< SO_NUM_ENTRIES_0_SHIFT
);
1113 for (i
= 0; i
< num_entries
; i
++) {
1114 ilo_cp_write(cp
, so_decls
[i
]);
1115 ilo_cp_write(cp
, 0);
1117 for (; i
< 128; i
++) {
1118 ilo_cp_write(cp
, 0);
1119 ilo_cp_write(cp
, 0);
1126 gen7_emit_3DSTATE_SO_BUFFER(const struct ilo_dev_info
*dev
,
1127 int index
, int base
, int stride
,
1128 const struct pipe_stream_output_target
*so_target
,
1131 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x1, 0x18);
1132 const uint8_t cmd_len
= 4;
1133 struct ilo_buffer
*buf
;
1136 ILO_GPE_VALID_GEN(dev
, 7, 7);
1138 if (!so_target
|| !so_target
->buffer
) {
1139 ilo_cp_begin(cp
, cmd_len
);
1140 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
1141 ilo_cp_write(cp
, index
<< SO_BUFFER_INDEX_SHIFT
);
1142 ilo_cp_write(cp
, 0);
1143 ilo_cp_write(cp
, 0);
1148 buf
= ilo_buffer(so_target
->buffer
);
1151 assert(stride
% 4 == 0 && base
% 4 == 0);
1152 assert(so_target
->buffer_offset
% 4 == 0);
1155 base
= (base
+ so_target
->buffer_offset
) & ~3;
1156 end
= (base
+ so_target
->buffer_size
) & ~3;
1158 ilo_cp_begin(cp
, cmd_len
);
1159 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
1160 ilo_cp_write(cp
, index
<< SO_BUFFER_INDEX_SHIFT
|
1162 ilo_cp_write_bo(cp
, base
, buf
->bo
, INTEL_DOMAIN_RENDER
, INTEL_DOMAIN_RENDER
);
1163 ilo_cp_write_bo(cp
, end
, buf
->bo
, INTEL_DOMAIN_RENDER
, INTEL_DOMAIN_RENDER
);
1168 gen7_emit_3DPRIMITIVE(const struct ilo_dev_info
*dev
,
1169 const struct pipe_draw_info
*info
,
1173 const uint32_t cmd
= ILO_GPE_CMD(0x3, 0x3, 0x00);
1174 const uint8_t cmd_len
= 7;
1175 const int prim
= (rectlist
) ?
1176 _3DPRIM_RECTLIST
: ilo_gpe_gen6_translate_pipe_prim(info
->mode
);
1177 const int vb_access
= (info
->indexed
) ?
1178 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
:
1179 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
;
1181 ILO_GPE_VALID_GEN(dev
, 7, 7);
1183 ilo_cp_begin(cp
, cmd_len
);
1184 ilo_cp_write(cp
, cmd
| (cmd_len
- 2));
1185 ilo_cp_write(cp
, vb_access
| prim
);
1186 ilo_cp_write(cp
, info
->count
);
1187 ilo_cp_write(cp
, info
->start
);
1188 ilo_cp_write(cp
, info
->instance_count
);
1189 ilo_cp_write(cp
, info
->start_instance
);
1190 ilo_cp_write(cp
, info
->index_bias
);
1195 gen7_emit_SF_CLIP_VIEWPORT(const struct ilo_dev_info
*dev
,
1196 const struct ilo_viewport_cso
*viewports
,
1197 unsigned num_viewports
,
1200 const int state_align
= 64 / 4;
1201 const int state_len
= 16 * num_viewports
;
1202 uint32_t state_offset
, *dw
;
1205 ILO_GPE_VALID_GEN(dev
, 7, 7);
1208 * From the Ivy Bridge PRM, volume 2 part 1, page 270:
1210 * "The viewport-specific state used by both the SF and CL units
1211 * (SF_CLIP_VIEWPORT) is stored as an array of up to 16 elements, each
1212 * of which contains the DWords described below. The start of each
1213 * element is spaced 16 DWords apart. The location of first element of
1214 * the array, as specified by both Pointer to SF_VIEWPORT and Pointer
1215 * to CLIP_VIEWPORT, is aligned to a 64-byte boundary."
1217 assert(num_viewports
&& num_viewports
<= 16);
1219 dw
= ilo_cp_steal_ptr(cp
, "SF_CLIP_VIEWPORT",
1220 state_len
, state_align
, &state_offset
);
1222 for (i
= 0; i
< num_viewports
; i
++) {
1223 const struct ilo_viewport_cso
*vp
= &viewports
[i
];
1225 dw
[0] = fui(vp
->m00
);
1226 dw
[1] = fui(vp
->m11
);
1227 dw
[2] = fui(vp
->m22
);
1228 dw
[3] = fui(vp
->m30
);
1229 dw
[4] = fui(vp
->m31
);
1230 dw
[5] = fui(vp
->m32
);
1233 dw
[8] = fui(vp
->min_gbx
);
1234 dw
[9] = fui(vp
->max_gbx
);
1235 dw
[10] = fui(vp
->min_gby
);
1236 dw
[11] = fui(vp
->max_gby
);
1245 return state_offset
;
1249 gen7_fill_null_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1250 unsigned width
, unsigned height
,
1251 unsigned depth
, unsigned lod
,
1252 uint32_t *dw
, int num_dwords
)
1254 ILO_GPE_VALID_GEN(dev
, 7, 7);
1255 assert(num_dwords
== 8);
1258 * From the Ivy Bridge PRM, volume 4 part 1, page 62:
1260 * "A null surface is used in instances where an actual surface is not
1261 * bound. When a write message is generated to a null surface, no
1262 * actual surface is written to. When a read message (including any
1263 * sampling engine message) is generated to a null surface, the result
1264 * is all zeros. Note that a null surface type is allowed to be used
1265 * with all messages, even if it is not specificially indicated as
1266 * supported. All of the remaining fields in surface state are ignored
1267 * for null surfaces, with the following exceptions:
1269 * * Width, Height, Depth, LOD, and Render Target View Extent fields
1270 * must match the depth buffer's corresponding state for all render
1271 * target surfaces, including null.
1272 * * All sampling engine and data port messages support null surfaces
1273 * with the above behavior, even if not mentioned as specifically
1274 * supported, except for the following:
1275 * * Data Port Media Block Read/Write messages.
1276 * * The Surface Type of a surface used as a render target (accessed
1277 * via the Data Port's Render Target Write message) must be the same
1278 * as the Surface Type of all other render targets and of the depth
1279 * buffer (defined in 3DSTATE_DEPTH_BUFFER), unless either the depth
1280 * buffer or render targets are SURFTYPE_NULL."
1282 * From the Ivy Bridge PRM, volume 4 part 1, page 65:
1284 * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be
1288 dw
[0] = BRW_SURFACE_NULL
<< BRW_SURFACE_TYPE_SHIFT
|
1289 BRW_SURFACEFORMAT_B8G8R8A8_UNORM
<< BRW_SURFACE_FORMAT_SHIFT
|
1290 BRW_SURFACE_TILED
<< 13;
1294 dw
[2] = SET_FIELD(height
- 1, GEN7_SURFACE_HEIGHT
) |
1295 SET_FIELD(width
- 1, GEN7_SURFACE_WIDTH
);
1297 dw
[3] = SET_FIELD(depth
- 1, BRW_SURFACE_DEPTH
);
1306 gen7_fill_buffer_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1307 const struct ilo_buffer
*buf
,
1308 unsigned offset
, unsigned size
,
1309 unsigned struct_size
,
1310 enum pipe_format elem_format
,
1311 bool is_rt
, bool render_cache_rw
,
1312 uint32_t *dw
, int num_dwords
)
1314 const bool typed
= (elem_format
!= PIPE_FORMAT_NONE
);
1315 const bool structured
= (!typed
&& struct_size
> 1);
1316 const int elem_size
= (typed
) ?
1317 util_format_get_blocksize(elem_format
) : 1;
1318 int width
, height
, depth
, pitch
;
1319 int surface_type
, surface_format
, num_entries
;
1321 ILO_GPE_VALID_GEN(dev
, 7, 7);
1322 assert(num_dwords
== 8);
1324 surface_type
= (structured
) ? 5 : BRW_SURFACE_BUFFER
;
1326 surface_format
= (typed
) ?
1327 ilo_translate_color_format(elem_format
) : BRW_SURFACEFORMAT_RAW
;
1329 num_entries
= size
/ struct_size
;
1330 /* see if there is enough space to fit another element */
1331 if (size
% struct_size
>= elem_size
&& !structured
)
1335 * From the Ivy Bridge PRM, volume 4 part 1, page 67:
1337 * "For SURFTYPE_BUFFER render targets, this field (Surface Base
1338 * Address) specifies the base address of first element of the
1339 * surface. The surface is interpreted as a simple array of that
1340 * single element type. The address must be naturally-aligned to the
1341 * element size (e.g., a buffer containing R32G32B32A32_FLOAT elements
1342 * must be 16-byte aligned)
1344 * For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies
1345 * the base address of the first element of the surface, computed in
1346 * software by adding the surface base address to the byte offset of
1347 * the element in the buffer."
1350 assert(offset
% elem_size
== 0);
1353 * From the Ivy Bridge PRM, volume 4 part 1, page 68:
1355 * "For typed buffer and structured buffer surfaces, the number of
1356 * entries in the buffer ranges from 1 to 2^27. For raw buffer
1357 * surfaces, the number of entries in the buffer is the number of
1358 * bytes which can range from 1 to 2^30."
1360 assert(num_entries
>= 1 &&
1361 num_entries
<= 1 << ((typed
|| structured
) ? 27 : 30));
1364 * From the Ivy Bridge PRM, volume 4 part 1, page 69:
1366 * "For SURFTYPE_BUFFER: The low two bits of this field (Width) must be
1367 * 11 if the Surface Format is RAW (the size of the buffer must be a
1368 * multiple of 4 bytes)."
1370 * From the Ivy Bridge PRM, volume 4 part 1, page 70:
1372 * "For surfaces of type SURFTYPE_BUFFER and SURFTYPE_STRBUF, this
1373 * field (Surface Pitch) indicates the size of the structure."
1375 * "For linear surfaces with Surface Type of SURFTYPE_STRBUF, the pitch
1376 * must be a multiple of 4 bytes."
1379 assert(struct_size
% 4 == 0);
1381 assert(num_entries
% 4 == 0);
1383 pitch
= struct_size
;
1388 width
= (num_entries
& 0x0000007f);
1390 height
= (num_entries
& 0x001fff80) >> 7;
1392 depth
= (num_entries
& 0x7fe00000) >> 21;
1393 /* limit to [26:21] */
1394 if (typed
|| structured
)
1397 dw
[0] = surface_type
<< BRW_SURFACE_TYPE_SHIFT
|
1398 surface_format
<< BRW_SURFACE_FORMAT_SHIFT
;
1399 if (render_cache_rw
)
1400 dw
[0] |= BRW_SURFACE_RC_READ_WRITE
;
1404 dw
[2] = SET_FIELD(height
, GEN7_SURFACE_HEIGHT
) |
1405 SET_FIELD(width
, GEN7_SURFACE_WIDTH
);
1407 dw
[3] = SET_FIELD(depth
, BRW_SURFACE_DEPTH
) |
1417 gen7_fill_normal_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1418 struct ilo_texture
*tex
,
1419 enum pipe_format format
,
1420 unsigned first_level
, unsigned num_levels
,
1421 unsigned first_layer
, unsigned num_layers
,
1422 bool is_rt
, bool render_cache_rw
,
1423 uint32_t *dw
, int num_dwords
)
1425 int surface_type
, surface_format
;
1426 int width
, height
, depth
, pitch
, lod
;
1427 unsigned layer_offset
, x_offset
, y_offset
;
1429 ILO_GPE_VALID_GEN(dev
, 7, 7);
1430 assert(num_dwords
== 8);
1432 surface_type
= ilo_gpe_gen6_translate_texture(tex
->base
.target
);
1433 assert(surface_type
!= BRW_SURFACE_BUFFER
);
1435 if (format
== PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
&& tex
->separate_s8
)
1436 format
= PIPE_FORMAT_Z32_FLOAT
;
1439 surface_format
= ilo_translate_render_format(format
);
1441 surface_format
= ilo_translate_texture_format(format
);
1442 assert(surface_format
>= 0);
1444 width
= tex
->base
.width0
;
1445 height
= tex
->base
.height0
;
1446 depth
= (tex
->base
.target
== PIPE_TEXTURE_3D
) ?
1447 tex
->base
.depth0
: num_layers
;
1448 pitch
= tex
->bo_stride
;
1450 if (surface_type
== BRW_SURFACE_CUBE
) {
1452 * From the Ivy Bridge PRM, volume 4 part 1, page 70:
1454 * "For SURFTYPE_CUBE:For Sampling Engine Surfaces, the range of
1455 * this field is [0,340], indicating the number of cube array
1456 * elements (equal to the number of underlying 2D array elements
1457 * divided by 6). For other surfaces, this field must be zero."
1459 * When is_rt is true, we treat the texture as a 2D one to avoid the
1463 surface_type
= BRW_SURFACE_2D
;
1466 assert(num_layers
% 6 == 0);
1467 depth
= num_layers
/ 6;
1471 /* sanity check the size */
1472 assert(width
>= 1 && height
>= 1 && depth
>= 1 && pitch
>= 1);
1473 assert(first_layer
< 2048 && num_layers
<= 2048);
1474 switch (surface_type
) {
1475 case BRW_SURFACE_1D
:
1476 assert(width
<= 16384 && height
== 1 && depth
<= 2048);
1478 case BRW_SURFACE_2D
:
1479 assert(width
<= 16384 && height
<= 16384 && depth
<= 2048);
1481 case BRW_SURFACE_3D
:
1482 assert(width
<= 2048 && height
<= 2048 && depth
<= 2048);
1484 assert(first_layer
== 0);
1486 case BRW_SURFACE_CUBE
:
1487 assert(width
<= 16384 && height
<= 16384 && depth
<= 86);
1488 assert(width
== height
);
1490 assert(first_layer
== 0);
1493 assert(!"unexpected surface type");
1499 * Compute the offset to the layer manually.
1501 * For rendering, the hardware requires LOD to be the same for all
1502 * render targets and the depth buffer. We need to compute the offset
1503 * to the layer manually and always set LOD to 0.
1506 /* we lose the capability for layered rendering */
1507 assert(num_layers
== 1);
1509 layer_offset
= ilo_texture_get_slice_offset(tex
,
1510 first_level
, first_layer
, &x_offset
, &y_offset
);
1512 assert(x_offset
% 4 == 0);
1513 assert(y_offset
% 2 == 0);
1517 /* derive the size for the LOD */
1518 width
= u_minify(width
, first_level
);
1519 height
= u_minify(height
, first_level
);
1520 if (surface_type
== BRW_SURFACE_3D
)
1521 depth
= u_minify(depth
, first_level
);
1535 assert(num_levels
== 1);
1543 lod
= num_levels
- 1;
1547 * From the Ivy Bridge PRM, volume 4 part 1, page 68:
1549 * "The Base Address for linear render target surfaces and surfaces
1550 * accessed with the typed surface read/write data port messages must
1551 * be element-size aligned, for non-YUV surface formats, or a multiple
1552 * of 2 element-sizes for YUV surface formats. Other linear surfaces
1553 * have no alignment requirements (byte alignment is sufficient)."
1555 * From the Ivy Bridge PRM, volume 4 part 1, page 70:
1557 * "For linear render target surfaces and surfaces accessed with the
1558 * typed data port messages, the pitch must be a multiple of the
1559 * element size for non-YUV surface formats. Pitch must be a multiple
1560 * of 2 * element size for YUV surface formats. For linear surfaces
1561 * with Surface Type of SURFTYPE_STRBUF, the pitch must be a multiple
1562 * of 4 bytes.For other linear surfaces, the pitch can be any multiple
1565 * From the Ivy Bridge PRM, volume 4 part 1, page 74:
1567 * "For linear surfaces, this field (X Offset) must be zero."
1569 if (tex
->tiling
== INTEL_TILING_NONE
) {
1571 const int elem_size
= util_format_get_blocksize(format
);
1572 assert(layer_offset
% elem_size
== 0);
1573 assert(pitch
% elem_size
== 0);
1579 dw
[0] = surface_type
<< BRW_SURFACE_TYPE_SHIFT
|
1580 surface_format
<< BRW_SURFACE_FORMAT_SHIFT
|
1581 ilo_gpe_gen6_translate_winsys_tiling(tex
->tiling
) << 13;
1583 if (surface_type
!= BRW_SURFACE_3D
&& depth
> 1)
1584 dw
[0] |= GEN7_SURFACE_IS_ARRAY
;
1587 dw
[0] |= GEN7_SURFACE_VALIGN_4
;
1590 dw
[0] |= GEN7_SURFACE_HALIGN_8
;
1592 if (tex
->array_spacing_full
)
1593 dw
[0] |= GEN7_SURFACE_ARYSPC_FULL
;
1595 dw
[0] |= GEN7_SURFACE_ARYSPC_LOD0
;
1597 if (render_cache_rw
)
1598 dw
[0] |= BRW_SURFACE_RC_READ_WRITE
;
1600 if (surface_type
== BRW_SURFACE_CUBE
&& !is_rt
)
1601 dw
[0] |= BRW_SURFACE_CUBEFACE_ENABLES
;
1603 dw
[1] = layer_offset
;
1605 dw
[2] = SET_FIELD(height
- 1, GEN7_SURFACE_HEIGHT
) |
1606 SET_FIELD(width
- 1, GEN7_SURFACE_WIDTH
);
1608 dw
[3] = SET_FIELD(depth
- 1, BRW_SURFACE_DEPTH
) |
1611 dw
[4] = first_layer
<< 18 |
1612 (num_layers
- 1) << 7;
1615 * MSFMT_MSS means the samples are not interleaved and MSFMT_DEPTH_STENCIL
1616 * means the samples are interleaved. The layouts are the same when the
1617 * number of samples is 1.
1619 if (tex
->interleaved
&& tex
->base
.nr_samples
> 1) {
1621 dw
[4] |= GEN7_SURFACE_MSFMT_DEPTH_STENCIL
;
1624 dw
[4] |= GEN7_SURFACE_MSFMT_MSS
;
1627 if (tex
->base
.nr_samples
> 4)
1628 dw
[4] |= GEN7_SURFACE_MULTISAMPLECOUNT_8
;
1629 else if (tex
->base
.nr_samples
> 2)
1630 dw
[4] |= GEN7_SURFACE_MULTISAMPLECOUNT_4
;
1632 dw
[4] |= GEN7_SURFACE_MULTISAMPLECOUNT_1
;
1634 dw
[5] = x_offset
<< BRW_SURFACE_X_OFFSET_SHIFT
|
1635 y_offset
<< BRW_SURFACE_Y_OFFSET_SHIFT
|
1636 SET_FIELD(first_level
, GEN7_SURFACE_MIN_LOD
) |
1644 gen7_emit_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1645 struct intel_bo
*bo
, bool for_render
,
1646 const uint32_t *dw
, int num_dwords
,
1649 const int state_align
= 32 / 4;
1650 const int state_len
= 8;
1651 uint32_t state_offset
;
1652 uint32_t read_domains
, write_domain
;
1654 ILO_GPE_VALID_GEN(dev
, 7, 7);
1655 assert(num_dwords
== state_len
);
1658 read_domains
= INTEL_DOMAIN_RENDER
;
1659 write_domain
= INTEL_DOMAIN_RENDER
;
1662 read_domains
= INTEL_DOMAIN_SAMPLER
;
1666 ilo_cp_steal(cp
, "SURFACE_STATE", state_len
, state_align
, &state_offset
);
1667 ilo_cp_write(cp
, dw
[0]);
1668 ilo_cp_write_bo(cp
, dw
[1], bo
, read_domains
, write_domain
);
1669 ilo_cp_write(cp
, dw
[2]);
1670 ilo_cp_write(cp
, dw
[3]);
1671 ilo_cp_write(cp
, dw
[4]);
1672 ilo_cp_write(cp
, dw
[5]);
1673 ilo_cp_write(cp
, dw
[6]);
1674 ilo_cp_write(cp
, dw
[7]);
1677 return state_offset
;
1681 gen7_emit_surf_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1682 const struct pipe_surface
*surface
,
1685 struct intel_bo
*bo
;
1688 ILO_GPE_VALID_GEN(dev
, 7, 7);
1690 if (surface
&& surface
->texture
) {
1691 struct ilo_texture
*tex
= ilo_texture(surface
->texture
);
1696 * classic i965 sets render_cache_rw for constant buffers and sol
1697 * surfaces but not render buffers. Why?
1699 gen7_fill_normal_SURFACE_STATE(dev
, tex
, surface
->format
,
1700 surface
->u
.tex
.level
, 1,
1701 surface
->u
.tex
.first_layer
,
1702 surface
->u
.tex
.last_layer
- surface
->u
.tex
.first_layer
+ 1,
1703 true, true, dw
, Elements(dw
));
1707 gen7_fill_null_SURFACE_STATE(dev
,
1708 surface
->width
, surface
->height
, 1, 0, dw
, Elements(dw
));
1711 return gen7_emit_SURFACE_STATE(dev
, bo
, true, dw
, Elements(dw
), cp
);
1715 gen7_emit_view_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1716 const struct pipe_sampler_view
*view
,
1719 struct intel_bo
*bo
;
1722 ILO_GPE_VALID_GEN(dev
, 7, 7);
1724 if (view
->texture
->target
== PIPE_BUFFER
) {
1725 const unsigned elem_size
= util_format_get_blocksize(view
->format
);
1726 const unsigned first_elem
= view
->u
.buf
.first_element
;
1727 const unsigned num_elems
= view
->u
.buf
.last_element
- first_elem
+ 1;
1728 struct ilo_buffer
*buf
= ilo_buffer(view
->texture
);
1730 gen7_fill_buffer_SURFACE_STATE(dev
, buf
,
1731 first_elem
* elem_size
, num_elems
* elem_size
,
1732 elem_size
, view
->format
, false, false, dw
, Elements(dw
));
1737 struct ilo_texture
*tex
= ilo_texture(view
->texture
);
1739 gen7_fill_normal_SURFACE_STATE(dev
, tex
, view
->format
,
1740 view
->u
.tex
.first_level
,
1741 view
->u
.tex
.last_level
- view
->u
.tex
.first_level
+ 1,
1742 view
->u
.tex
.first_layer
,
1743 view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1,
1744 false, false, dw
, Elements(dw
));
1749 return gen7_emit_SURFACE_STATE(dev
, bo
, false, dw
, Elements(dw
), cp
);
1753 gen7_emit_cbuf_SURFACE_STATE(const struct ilo_dev_info
*dev
,
1754 const struct pipe_constant_buffer
*cbuf
,
1757 const enum pipe_format elem_format
= PIPE_FORMAT_R32G32B32A32_FLOAT
;
1758 struct ilo_buffer
*buf
= ilo_buffer(cbuf
->buffer
);
1761 ILO_GPE_VALID_GEN(dev
, 7, 7);
1763 gen7_fill_buffer_SURFACE_STATE(dev
, buf
,
1764 cbuf
->buffer_offset
, cbuf
->buffer_size
,
1765 util_format_get_blocksize(elem_format
), elem_format
,
1766 false, false, dw
, Elements(dw
));
1768 return gen7_emit_SURFACE_STATE(dev
, buf
->bo
, false, dw
, Elements(dw
), cp
);
1772 gen7_estimate_command_size(const struct ilo_dev_info
*dev
,
1773 enum ilo_gpe_gen7_command cmd
,
1776 static const struct {
1779 } gen7_command_size_table
[ILO_GPE_GEN7_COMMAND_COUNT
] = {
1780 [ILO_GPE_GEN7_STATE_BASE_ADDRESS
] = { 0, 10 },
1781 [ILO_GPE_GEN7_STATE_SIP
] = { 0, 2 },
1782 [ILO_GPE_GEN7_3DSTATE_VF_STATISTICS
] = { 0, 1 },
1783 [ILO_GPE_GEN7_PIPELINE_SELECT
] = { 0, 1 },
1784 [ILO_GPE_GEN7_MEDIA_VFE_STATE
] = { 0, 8 },
1785 [ILO_GPE_GEN7_MEDIA_CURBE_LOAD
] = { 0, 4 },
1786 [ILO_GPE_GEN7_MEDIA_INTERFACE_DESCRIPTOR_LOAD
] = { 0, 4 },
1787 [ILO_GPE_GEN7_MEDIA_STATE_FLUSH
] = { 0, 2 },
1788 [ILO_GPE_GEN7_GPGPU_WALKER
] = { 0, 11 },
1789 [ILO_GPE_GEN7_3DSTATE_CLEAR_PARAMS
] = { 0, 3 },
1790 [ILO_GPE_GEN7_3DSTATE_DEPTH_BUFFER
] = { 0, 7 },
1791 [ILO_GPE_GEN7_3DSTATE_STENCIL_BUFFER
] = { 0, 3 },
1792 [ILO_GPE_GEN7_3DSTATE_HIER_DEPTH_BUFFER
] = { 0, 3 },
1793 [ILO_GPE_GEN7_3DSTATE_VERTEX_BUFFERS
] = { 1, 4 },
1794 [ILO_GPE_GEN7_3DSTATE_VERTEX_ELEMENTS
] = { 1, 2 },
1795 [ILO_GPE_GEN7_3DSTATE_INDEX_BUFFER
] = { 0, 3 },
1796 [ILO_GPE_GEN7_3DSTATE_CC_STATE_POINTERS
] = { 0, 2 },
1797 [ILO_GPE_GEN7_3DSTATE_SCISSOR_STATE_POINTERS
] = { 0, 2 },
1798 [ILO_GPE_GEN7_3DSTATE_VS
] = { 0, 6 },
1799 [ILO_GPE_GEN7_3DSTATE_GS
] = { 0, 7 },
1800 [ILO_GPE_GEN7_3DSTATE_CLIP
] = { 0, 4 },
1801 [ILO_GPE_GEN7_3DSTATE_SF
] = { 0, 7 },
1802 [ILO_GPE_GEN7_3DSTATE_WM
] = { 0, 3 },
1803 [ILO_GPE_GEN7_3DSTATE_CONSTANT_VS
] = { 0, 7 },
1804 [ILO_GPE_GEN7_3DSTATE_CONSTANT_GS
] = { 0, 7 },
1805 [ILO_GPE_GEN7_3DSTATE_CONSTANT_PS
] = { 0, 7 },
1806 [ILO_GPE_GEN7_3DSTATE_SAMPLE_MASK
] = { 0, 2 },
1807 [ILO_GPE_GEN7_3DSTATE_CONSTANT_HS
] = { 0, 7 },
1808 [ILO_GPE_GEN7_3DSTATE_CONSTANT_DS
] = { 0, 7 },
1809 [ILO_GPE_GEN7_3DSTATE_HS
] = { 0, 7 },
1810 [ILO_GPE_GEN7_3DSTATE_TE
] = { 0, 4 },
1811 [ILO_GPE_GEN7_3DSTATE_DS
] = { 0, 6 },
1812 [ILO_GPE_GEN7_3DSTATE_STREAMOUT
] = { 0, 3 },
1813 [ILO_GPE_GEN7_3DSTATE_SBE
] = { 0, 14 },
1814 [ILO_GPE_GEN7_3DSTATE_PS
] = { 0, 8 },
1815 [ILO_GPE_GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP
] = { 0, 2 },
1816 [ILO_GPE_GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC
] = { 0, 2 },
1817 [ILO_GPE_GEN7_3DSTATE_BLEND_STATE_POINTERS
] = { 0, 2 },
1818 [ILO_GPE_GEN7_3DSTATE_DEPTH_STENCIL_STATE_POINTERS
] = { 0, 2 },
1819 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_VS
] = { 0, 2 },
1820 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_HS
] = { 0, 2 },
1821 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_DS
] = { 0, 2 },
1822 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_GS
] = { 0, 2 },
1823 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS
] = { 0, 2 },
1824 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_VS
] = { 0, 2 },
1825 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_HS
] = { 0, 2 },
1826 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_DS
] = { 0, 2 },
1827 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_GS
] = { 0, 2 },
1828 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_PS
] = { 0, 2 },
1829 [ILO_GPE_GEN7_3DSTATE_URB_VS
] = { 0, 2 },
1830 [ILO_GPE_GEN7_3DSTATE_URB_HS
] = { 0, 2 },
1831 [ILO_GPE_GEN7_3DSTATE_URB_DS
] = { 0, 2 },
1832 [ILO_GPE_GEN7_3DSTATE_URB_GS
] = { 0, 2 },
1833 [ILO_GPE_GEN7_3DSTATE_DRAWING_RECTANGLE
] = { 0, 4 },
1834 [ILO_GPE_GEN7_3DSTATE_POLY_STIPPLE_OFFSET
] = { 0, 2 },
1835 [ILO_GPE_GEN7_3DSTATE_POLY_STIPPLE_PATTERN
] = { 0, 33, },
1836 [ILO_GPE_GEN7_3DSTATE_LINE_STIPPLE
] = { 0, 3 },
1837 [ILO_GPE_GEN7_3DSTATE_AA_LINE_PARAMETERS
] = { 0, 3 },
1838 [ILO_GPE_GEN7_3DSTATE_MULTISAMPLE
] = { 0, 4 },
1839 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_VS
] = { 0, 2 },
1840 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_HS
] = { 0, 2 },
1841 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_DS
] = { 0, 2 },
1842 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_GS
] = { 0, 2 },
1843 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_PS
] = { 0, 2 },
1844 [ILO_GPE_GEN7_3DSTATE_SO_DECL_LIST
] = { 3, 2 },
1845 [ILO_GPE_GEN7_3DSTATE_SO_BUFFER
] = { 0, 4 },
1846 [ILO_GPE_GEN7_PIPE_CONTROL
] = { 0, 5 },
1847 [ILO_GPE_GEN7_3DPRIMITIVE
] = { 0, 7 },
1849 const int header
= gen7_command_size_table
[cmd
].header
;
1850 const int body
= gen7_command_size_table
[cmd
].body
;
1851 const int count
= arg
;
1853 ILO_GPE_VALID_GEN(dev
, 7, 7);
1854 assert(cmd
< ILO_GPE_GEN7_COMMAND_COUNT
);
1856 return (likely(count
)) ? header
+ body
* count
: 0;
1860 gen7_estimate_state_size(const struct ilo_dev_info
*dev
,
1861 enum ilo_gpe_gen7_state state
,
1864 static const struct {
1868 } gen7_state_size_table
[ILO_GPE_GEN7_STATE_COUNT
] = {
1869 [ILO_GPE_GEN7_INTERFACE_DESCRIPTOR_DATA
] = { 8, 8, true },
1870 [ILO_GPE_GEN7_SF_CLIP_VIEWPORT
] = { 16, 16, true },
1871 [ILO_GPE_GEN7_CC_VIEWPORT
] = { 8, 2, true },
1872 [ILO_GPE_GEN7_COLOR_CALC_STATE
] = { 16, 6, false },
1873 [ILO_GPE_GEN7_BLEND_STATE
] = { 16, 2, true },
1874 [ILO_GPE_GEN7_DEPTH_STENCIL_STATE
] = { 16, 3, false },
1875 [ILO_GPE_GEN7_SCISSOR_RECT
] = { 8, 2, true },
1876 [ILO_GPE_GEN7_BINDING_TABLE_STATE
] = { 8, 1, true },
1877 [ILO_GPE_GEN7_SURFACE_STATE
] = { 8, 8, false },
1878 [ILO_GPE_GEN7_SAMPLER_STATE
] = { 8, 4, true },
1879 [ILO_GPE_GEN7_SAMPLER_BORDER_COLOR_STATE
] = { 8, 4, false },
1880 [ILO_GPE_GEN7_PUSH_CONSTANT_BUFFER
] = { 8, 1, true },
1882 const int alignment
= gen7_state_size_table
[state
].alignment
;
1883 const int body
= gen7_state_size_table
[state
].body
;
1884 const bool is_array
= gen7_state_size_table
[state
].is_array
;
1885 const int count
= arg
;
1888 ILO_GPE_VALID_GEN(dev
, 7, 7);
1889 assert(state
< ILO_GPE_GEN7_STATE_COUNT
);
1891 if (likely(count
)) {
1893 estimate
= (alignment
- 1) + body
* count
;
1896 estimate
= (alignment
- 1) + body
;
1897 /* all states are aligned */
1899 estimate
+= util_align_npot(body
, alignment
) * (count
- 1);
1910 gen7_init(struct ilo_gpe_gen7
*gen7
)
1912 const struct ilo_gpe_gen6
*gen6
= ilo_gpe_gen6_get();
1914 gen7
->estimate_command_size
= gen7_estimate_command_size
;
1915 gen7
->estimate_state_size
= gen7_estimate_state_size
;
1917 #define GEN7_USE(gen7, name, from) gen7->emit_ ## name = from->emit_ ## name
1918 #define GEN7_SET(gen7, name) gen7->emit_ ## name = gen7_emit_ ## name
1919 GEN7_USE(gen7
, STATE_BASE_ADDRESS
, gen6
);
1920 GEN7_USE(gen7
, STATE_SIP
, gen6
);
1921 GEN7_USE(gen7
, 3DSTATE_VF_STATISTICS
, gen6
);
1922 GEN7_USE(gen7
, PIPELINE_SELECT
, gen6
);
1923 GEN7_USE(gen7
, MEDIA_VFE_STATE
, gen6
);
1924 GEN7_USE(gen7
, MEDIA_CURBE_LOAD
, gen6
);
1925 GEN7_USE(gen7
, MEDIA_INTERFACE_DESCRIPTOR_LOAD
, gen6
);
1926 GEN7_USE(gen7
, MEDIA_STATE_FLUSH
, gen6
);
1927 GEN7_SET(gen7
, GPGPU_WALKER
);
1928 GEN7_SET(gen7
, 3DSTATE_CLEAR_PARAMS
);
1929 GEN7_SET(gen7
, 3DSTATE_DEPTH_BUFFER
);
1930 GEN7_USE(gen7
, 3DSTATE_STENCIL_BUFFER
, gen6
);
1931 GEN7_USE(gen7
, 3DSTATE_HIER_DEPTH_BUFFER
, gen6
);
1932 GEN7_USE(gen7
, 3DSTATE_VERTEX_BUFFERS
, gen6
);
1933 GEN7_USE(gen7
, 3DSTATE_VERTEX_ELEMENTS
, gen6
);
1934 GEN7_USE(gen7
, 3DSTATE_INDEX_BUFFER
, gen6
);
1935 GEN7_SET(gen7
, 3DSTATE_CC_STATE_POINTERS
);
1936 GEN7_USE(gen7
, 3DSTATE_SCISSOR_STATE_POINTERS
, gen6
);
1937 GEN7_USE(gen7
, 3DSTATE_VS
, gen6
);
1938 GEN7_SET(gen7
, 3DSTATE_GS
);
1939 GEN7_USE(gen7
, 3DSTATE_CLIP
, gen6
);
1940 GEN7_SET(gen7
, 3DSTATE_SF
);
1941 GEN7_SET(gen7
, 3DSTATE_WM
);
1942 GEN7_SET(gen7
, 3DSTATE_CONSTANT_VS
);
1943 GEN7_SET(gen7
, 3DSTATE_CONSTANT_GS
);
1944 GEN7_SET(gen7
, 3DSTATE_CONSTANT_PS
);
1945 GEN7_SET(gen7
, 3DSTATE_SAMPLE_MASK
);
1946 GEN7_SET(gen7
, 3DSTATE_CONSTANT_HS
);
1947 GEN7_SET(gen7
, 3DSTATE_CONSTANT_DS
);
1948 GEN7_SET(gen7
, 3DSTATE_HS
);
1949 GEN7_SET(gen7
, 3DSTATE_TE
);
1950 GEN7_SET(gen7
, 3DSTATE_DS
);
1951 GEN7_SET(gen7
, 3DSTATE_STREAMOUT
);
1952 GEN7_SET(gen7
, 3DSTATE_SBE
);
1953 GEN7_SET(gen7
, 3DSTATE_PS
);
1954 GEN7_SET(gen7
, 3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP
);
1955 GEN7_SET(gen7
, 3DSTATE_VIEWPORT_STATE_POINTERS_CC
);
1956 GEN7_SET(gen7
, 3DSTATE_BLEND_STATE_POINTERS
);
1957 GEN7_SET(gen7
, 3DSTATE_DEPTH_STENCIL_STATE_POINTERS
);
1958 GEN7_SET(gen7
, 3DSTATE_BINDING_TABLE_POINTERS_VS
);
1959 GEN7_SET(gen7
, 3DSTATE_BINDING_TABLE_POINTERS_HS
);
1960 GEN7_SET(gen7
, 3DSTATE_BINDING_TABLE_POINTERS_DS
);
1961 GEN7_SET(gen7
, 3DSTATE_BINDING_TABLE_POINTERS_GS
);
1962 GEN7_SET(gen7
, 3DSTATE_BINDING_TABLE_POINTERS_PS
);
1963 GEN7_SET(gen7
, 3DSTATE_SAMPLER_STATE_POINTERS_VS
);
1964 GEN7_SET(gen7
, 3DSTATE_SAMPLER_STATE_POINTERS_HS
);
1965 GEN7_SET(gen7
, 3DSTATE_SAMPLER_STATE_POINTERS_DS
);
1966 GEN7_SET(gen7
, 3DSTATE_SAMPLER_STATE_POINTERS_GS
);
1967 GEN7_SET(gen7
, 3DSTATE_SAMPLER_STATE_POINTERS_PS
);
1968 GEN7_SET(gen7
, 3DSTATE_URB_VS
);
1969 GEN7_SET(gen7
, 3DSTATE_URB_HS
);
1970 GEN7_SET(gen7
, 3DSTATE_URB_DS
);
1971 GEN7_SET(gen7
, 3DSTATE_URB_GS
);
1972 GEN7_USE(gen7
, 3DSTATE_DRAWING_RECTANGLE
, gen6
);
1973 GEN7_USE(gen7
, 3DSTATE_POLY_STIPPLE_OFFSET
, gen6
);
1974 GEN7_USE(gen7
, 3DSTATE_POLY_STIPPLE_PATTERN
, gen6
);
1975 GEN7_USE(gen7
, 3DSTATE_LINE_STIPPLE
, gen6
);
1976 GEN7_USE(gen7
, 3DSTATE_AA_LINE_PARAMETERS
, gen6
);
1977 GEN7_USE(gen7
, 3DSTATE_MULTISAMPLE
, gen6
);
1978 GEN7_SET(gen7
, 3DSTATE_PUSH_CONSTANT_ALLOC_VS
);
1979 GEN7_SET(gen7
, 3DSTATE_PUSH_CONSTANT_ALLOC_HS
);
1980 GEN7_SET(gen7
, 3DSTATE_PUSH_CONSTANT_ALLOC_DS
);
1981 GEN7_SET(gen7
, 3DSTATE_PUSH_CONSTANT_ALLOC_GS
);
1982 GEN7_SET(gen7
, 3DSTATE_PUSH_CONSTANT_ALLOC_PS
);
1983 GEN7_SET(gen7
, 3DSTATE_SO_DECL_LIST
);
1984 GEN7_SET(gen7
, 3DSTATE_SO_BUFFER
);
1985 GEN7_USE(gen7
, PIPE_CONTROL
, gen6
);
1986 GEN7_SET(gen7
, 3DPRIMITIVE
);
1987 GEN7_USE(gen7
, INTERFACE_DESCRIPTOR_DATA
, gen6
);
1988 GEN7_SET(gen7
, SF_CLIP_VIEWPORT
);
1989 GEN7_USE(gen7
, CC_VIEWPORT
, gen6
);
1990 GEN7_USE(gen7
, COLOR_CALC_STATE
, gen6
);
1991 GEN7_USE(gen7
, BLEND_STATE
, gen6
);
1992 GEN7_USE(gen7
, DEPTH_STENCIL_STATE
, gen6
);
1993 GEN7_USE(gen7
, SCISSOR_RECT
, gen6
);
1994 GEN7_USE(gen7
, BINDING_TABLE_STATE
, gen6
);
1995 GEN7_SET(gen7
, surf_SURFACE_STATE
);
1996 GEN7_SET(gen7
, view_SURFACE_STATE
);
1997 GEN7_SET(gen7
, cbuf_SURFACE_STATE
);
1998 GEN7_USE(gen7
, SAMPLER_STATE
, gen6
);
1999 GEN7_USE(gen7
, SAMPLER_BORDER_COLOR_STATE
, gen6
);
2000 GEN7_USE(gen7
, push_constant_buffer
, gen6
);
2005 static struct ilo_gpe_gen7 gen7_gpe
;
2007 const struct ilo_gpe_gen7
*
2008 ilo_gpe_gen7_get(void)
2010 if (!gen7_gpe
.estimate_command_size
)
2011 gen7_init(&gen7_gpe
);