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 "util/u_resource.h"
29 #include "brw_defines.h"
30 #include "intel_reg.h"
33 #include "ilo_format.h"
34 #include "ilo_resource.h"
35 #include "ilo_shader.h"
36 #include "ilo_gpe_gen7.h"
39 gen7_emit_GPGPU_WALKER(const struct ilo_dev_info *dev,
42 assert(!"GPGPU_WALKER unsupported");
46 gen7_emit_3DSTATE_CLEAR_PARAMS(const struct ilo_dev_info *dev,
50 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x04);
51 const uint8_t cmd_len = 3;
53 ILO_GPE_VALID_GEN(dev, 7, 7);
55 ilo_cp_begin(cp, cmd_len);
56 ilo_cp_write(cp, cmd | (cmd_len - 2));
57 ilo_cp_write(cp, clear_val);
63 gen7_emit_3dstate_pointer(const struct ilo_dev_info *dev,
64 int subop, uint32_t pointer,
67 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, subop);
68 const uint8_t cmd_len = 2;
70 ILO_GPE_VALID_GEN(dev, 7, 7);
72 ilo_cp_begin(cp, cmd_len);
73 ilo_cp_write(cp, cmd | (cmd_len - 2));
74 ilo_cp_write(cp, pointer);
79 gen7_emit_3DSTATE_CC_STATE_POINTERS(const struct ilo_dev_info *dev,
80 uint32_t color_calc_state,
83 gen7_emit_3dstate_pointer(dev, 0x0e, color_calc_state, cp);
87 gen7_emit_3DSTATE_GS(const struct ilo_dev_info *dev,
88 const struct ilo_shader *gs,
92 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x11);
93 const uint8_t cmd_len = 7;
94 uint32_t dw2, dw4, dw5;
97 ILO_GPE_VALID_GEN(dev, 7, 7);
101 max_threads = (dev->gt == 2) ? 128 : 36;
109 ilo_cp_begin(cp, cmd_len);
110 ilo_cp_write(cp, cmd | (cmd_len - 2));
115 ilo_cp_write(cp, GEN6_GS_STATISTICS_ENABLE);
121 dw2 = ((num_samplers + 3) / 4) << GEN6_GS_SAMPLER_COUNT_SHIFT;
123 dw4 = ((gs->in.count + 1) / 2) << GEN6_GS_URB_READ_LENGTH_SHIFT |
124 GEN7_GS_INCLUDE_VERTEX_HANDLES |
125 0 << GEN6_GS_URB_ENTRY_READ_OFFSET_SHIFT |
126 gs->in.start_grf << GEN6_GS_DISPATCH_START_GRF_SHIFT;
128 dw5 = (max_threads - 1) << GEN6_GS_MAX_THREADS_SHIFT |
129 GEN6_GS_STATISTICS_ENABLE |
132 ilo_cp_begin(cp, cmd_len);
133 ilo_cp_write(cp, cmd | (cmd_len - 2));
134 ilo_cp_write(cp, gs->cache_offset);
135 ilo_cp_write(cp, dw2);
136 ilo_cp_write(cp, 0); /* scratch */
137 ilo_cp_write(cp, dw4);
138 ilo_cp_write(cp, dw5);
144 gen7_emit_3DSTATE_SF(const struct ilo_dev_info *dev,
145 const struct ilo_rasterizer_state *rasterizer,
146 const struct pipe_surface *zs_surf,
149 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x13);
150 const uint8_t cmd_len = 7;
151 const int num_samples = 1;
154 ILO_GPE_VALID_GEN(dev, 7, 7);
156 ilo_gpe_gen6_fill_3dstate_sf_raster(dev,
157 &rasterizer->sf, num_samples,
158 (zs_surf) ? zs_surf->format : PIPE_FORMAT_NONE,
159 payload, Elements(payload));
161 ilo_cp_begin(cp, cmd_len);
162 ilo_cp_write(cp, cmd | (cmd_len - 2));
163 ilo_cp_write_multi(cp, payload, 6);
168 gen7_emit_3DSTATE_WM(const struct ilo_dev_info *dev,
169 const struct ilo_shader *fs,
170 const struct pipe_rasterizer_state *rasterizer,
174 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x14);
175 const uint8_t cmd_len = 3;
176 const int num_samples = 1;
179 ILO_GPE_VALID_GEN(dev, 7, 7);
181 dw1 = GEN7_WM_STATISTICS_ENABLE |
182 GEN7_WM_LINE_AA_WIDTH_2_0;
185 dw1 |= GEN7_WM_DEPTH_CLEAR;
186 dw1 |= GEN7_WM_DEPTH_RESOLVE;
187 dw1 |= GEN7_WM_HIERARCHICAL_DEPTH_RESOLVE;
194 * a) fs writes colors and color is not masked, or
195 * b) fs writes depth, or
198 dw1 |= GEN7_WM_DISPATCH_ENABLE;
201 * From the Ivy Bridge PRM, volume 2 part 1, page 278:
203 * "This bit (Pixel Shader Kill Pixel), if ENABLED, indicates that
204 * the PS kernel or color calculator has the ability to kill
205 * (discard) pixels or samples, other than due to depth or stencil
206 * testing. This bit is required to be ENABLED in the following
209 * - The API pixel shader program contains "killpix" or "discard"
210 * instructions, or other code in the pixel shader kernel that
211 * can cause the final pixel mask to differ from the pixel mask
212 * received on dispatch.
214 * - A sampler with chroma key enabled with kill pixel mode is used
215 * by the pixel shader.
217 * - Any render target has Alpha Test Enable or AlphaToCoverage
220 * - The pixel shader kernel generates and outputs oMask.
222 * Note: As ClipDistance clipping is fully supported in hardware
223 * and therefore not via PS instructions, there should be no need
224 * to ENABLE this bit due to ClipDistance clipping."
226 if (fs->has_kill || cc_may_kill)
227 dw1 |= GEN7_WM_KILL_ENABLE;
230 dw1 |= GEN7_WM_PSCDEPTH_ON;
232 dw1 |= GEN7_WM_USES_SOURCE_DEPTH | GEN7_WM_USES_SOURCE_W;
234 dw1 |= fs->in.barycentric_interpolation_mode <<
235 GEN7_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT;
237 else if (cc_may_kill) {
238 dw1 |= GEN7_WM_DISPATCH_ENABLE |
242 dw1 |= GEN7_WM_POSITION_ZW_PIXEL;
244 /* same value as in 3DSTATE_SF */
245 if (rasterizer->line_smooth)
246 dw1 |= GEN7_WM_LINE_END_CAP_AA_WIDTH_1_0;
248 if (rasterizer->poly_stipple_enable)
249 dw1 |= GEN7_WM_POLYGON_STIPPLE_ENABLE;
250 if (rasterizer->line_stipple_enable)
251 dw1 |= GEN7_WM_LINE_STIPPLE_ENABLE;
253 if (rasterizer->bottom_edge_rule)
254 dw1 |= GEN7_WM_POINT_RASTRULE_UPPER_RIGHT;
256 if (num_samples > 1) {
257 if (rasterizer->multisample)
258 dw1 |= GEN7_WM_MSRAST_ON_PATTERN;
260 dw1 |= GEN7_WM_MSRAST_OFF_PIXEL;
262 dw2 = GEN7_WM_MSDISPMODE_PERPIXEL;
265 dw1 |= GEN7_WM_MSRAST_OFF_PIXEL;
267 dw2 = GEN7_WM_MSDISPMODE_PERSAMPLE;
270 ilo_cp_begin(cp, cmd_len);
271 ilo_cp_write(cp, cmd | (cmd_len - 2));
272 ilo_cp_write(cp, dw1);
273 ilo_cp_write(cp, dw2);
278 gen7_emit_3dstate_constant(const struct ilo_dev_info *dev,
280 const uint32_t *bufs, const int *sizes,
284 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, subop);
285 const uint8_t cmd_len = 7;
287 int total_read_length, i;
289 ILO_GPE_VALID_GEN(dev, 7, 7);
291 /* VS, HS, DS, GS, and PS variants */
292 assert(subop >= 0x15 && subop <= 0x1a && subop != 0x18);
294 assert(num_bufs <= 4);
299 total_read_length = 0;
300 for (i = 0; i < 4; i++) {
304 * From the Ivy Bridge PRM, volume 2 part 1, page 112:
306 * "Constant buffers must be enabled in order from Constant Buffer 0
307 * to Constant Buffer 3 within this command. For example, it is
308 * not allowed to enable Constant Buffer 1 by programming a
309 * non-zero value in the VS Constant Buffer 1 Read Length without a
310 * non-zero value in VS Constant Buffer 0 Read Length."
312 if (i >= num_bufs || !sizes[i]) {
314 assert(i >= num_bufs || !sizes[i]);
320 /* read lengths are in 256-bit units */
321 read_len = (sizes[i] + 31) / 32;
322 /* the lower 5 bits are used for memory object control state */
323 assert(bufs[i] % 32 == 0);
325 dw[i / 2] |= read_len << ((i % 2) ? 16 : 0);
328 total_read_length += read_len;
332 * From the Ivy Bridge PRM, volume 2 part 1, page 113:
334 * "The sum of all four read length fields must be less than or equal
337 assert(total_read_length <= 64);
339 ilo_cp_begin(cp, cmd_len);
340 ilo_cp_write(cp, cmd | (cmd_len - 2));
341 ilo_cp_write_multi(cp, dw, 6);
346 gen7_emit_3DSTATE_CONSTANT_VS(const struct ilo_dev_info *dev,
347 const uint32_t *bufs, const int *sizes,
351 gen7_emit_3dstate_constant(dev, 0x15, bufs, sizes, num_bufs, cp);
355 gen7_emit_3DSTATE_CONSTANT_GS(const struct ilo_dev_info *dev,
356 const uint32_t *bufs, const int *sizes,
360 gen7_emit_3dstate_constant(dev, 0x16, bufs, sizes, num_bufs, cp);
364 gen7_emit_3DSTATE_CONSTANT_PS(const struct ilo_dev_info *dev,
365 const uint32_t *bufs, const int *sizes,
369 gen7_emit_3dstate_constant(dev, 0x17, bufs, sizes, num_bufs, cp);
373 gen7_emit_3DSTATE_SAMPLE_MASK(const struct ilo_dev_info *dev,
374 unsigned sample_mask,
378 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x18);
379 const uint8_t cmd_len = 2;
380 const unsigned valid_mask = ((1 << num_samples) - 1) | 0x1;
382 ILO_GPE_VALID_GEN(dev, 7, 7);
385 * From the Ivy Bridge PRM, volume 2 part 1, page 294:
387 * "If Number of Multisamples is NUMSAMPLES_1, bits 7:1 of this field
388 * (Sample Mask) must be zero.
390 * If Number of Multisamples is NUMSAMPLES_4, bits 7:4 of this field
393 sample_mask &= valid_mask;
395 ilo_cp_begin(cp, cmd_len);
396 ilo_cp_write(cp, cmd | (cmd_len - 2));
397 ilo_cp_write(cp, sample_mask);
402 gen7_emit_3DSTATE_CONSTANT_HS(const struct ilo_dev_info *dev,
403 const uint32_t *bufs, const int *sizes,
407 gen7_emit_3dstate_constant(dev, 0x19, bufs, sizes, num_bufs, cp);
411 gen7_emit_3DSTATE_CONSTANT_DS(const struct ilo_dev_info *dev,
412 const uint32_t *bufs, const int *sizes,
416 gen7_emit_3dstate_constant(dev, 0x1a, bufs, sizes, num_bufs, cp);
420 gen7_emit_3DSTATE_HS(const struct ilo_dev_info *dev,
421 const struct ilo_shader *hs,
422 int max_threads, int num_samplers,
425 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x1b);
426 const uint8_t cmd_len = 7;
427 uint32_t dw1, dw2, dw5;
429 ILO_GPE_VALID_GEN(dev, 7, 7);
432 ilo_cp_begin(cp, cmd_len);
433 ilo_cp_write(cp, cmd | (cmd_len - 2));
445 dw1 = (num_samplers + 3) / 4 << 27 |
451 dw2 = 1 << 31 | /* HS Enable */
452 1 << 29 | /* HS Statistics Enable */
453 0; /* Instance Count */
455 dw5 = hs->in.start_grf << 19 |
459 ilo_cp_begin(cp, cmd_len);
460 ilo_cp_write(cp, cmd | (cmd_len - 2));
461 ilo_cp_write(cp, dw1);
462 ilo_cp_write(cp, dw2);
463 ilo_cp_write(cp, hs->cache_offset);
465 ilo_cp_write(cp, dw5);
471 gen7_emit_3DSTATE_TE(const struct ilo_dev_info *dev,
474 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x1c);
475 const uint8_t cmd_len = 4;
477 ILO_GPE_VALID_GEN(dev, 7, 7);
479 ilo_cp_begin(cp, cmd_len);
480 ilo_cp_write(cp, cmd | (cmd_len - 2));
488 gen7_emit_3DSTATE_DS(const struct ilo_dev_info *dev,
489 const struct ilo_shader *ds,
490 int max_threads, int num_samplers,
493 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x1d);
494 const uint8_t cmd_len = 6;
495 uint32_t dw2, dw4, dw5;
497 ILO_GPE_VALID_GEN(dev, 7, 7);
500 ilo_cp_begin(cp, cmd_len);
501 ilo_cp_write(cp, cmd | (cmd_len - 2));
512 dw2 = (num_samplers + 3) / 4 << 27 |
518 dw4 = ds->in.start_grf << 20 |
522 dw5 = (max_threads - 1) << 25 |
526 ilo_cp_begin(cp, cmd_len);
527 ilo_cp_write(cp, cmd | (cmd_len - 2));
528 ilo_cp_write(cp, ds->cache_offset);
529 ilo_cp_write(cp, dw2);
531 ilo_cp_write(cp, dw4);
532 ilo_cp_write(cp, dw5);
537 gen7_emit_3DSTATE_STREAMOUT(const struct ilo_dev_info *dev,
538 unsigned buffer_mask,
539 int vertex_attrib_count,
540 bool rasterizer_discard,
543 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x1e);
544 const uint8_t cmd_len = 3;
545 const bool enable = (buffer_mask != 0);
549 ILO_GPE_VALID_GEN(dev, 7, 7);
552 dw1 = 0 << SO_RENDER_STREAM_SELECT_SHIFT;
553 if (rasterizer_discard)
554 dw1 |= SO_RENDERING_DISABLE;
558 ilo_cp_begin(cp, cmd_len);
559 ilo_cp_write(cp, cmd | (cmd_len - 2));
560 ilo_cp_write(cp, dw1);
561 ilo_cp_write(cp, dw2);
566 read_len = (vertex_attrib_count + 1) / 2;
570 dw1 = SO_FUNCTION_ENABLE |
571 0 << SO_RENDER_STREAM_SELECT_SHIFT |
572 SO_STATISTICS_ENABLE |
575 if (rasterizer_discard)
576 dw1 |= SO_RENDERING_DISABLE;
580 dw1 |= SO_REORDER_TRAILING;
582 dw2 = 0 << SO_STREAM_3_VERTEX_READ_OFFSET_SHIFT |
583 0 << SO_STREAM_3_VERTEX_READ_LENGTH_SHIFT |
584 0 << SO_STREAM_2_VERTEX_READ_OFFSET_SHIFT |
585 0 << SO_STREAM_2_VERTEX_READ_LENGTH_SHIFT |
586 0 << SO_STREAM_1_VERTEX_READ_OFFSET_SHIFT |
587 0 << SO_STREAM_1_VERTEX_READ_LENGTH_SHIFT |
588 0 << SO_STREAM_0_VERTEX_READ_OFFSET_SHIFT |
589 (read_len - 1) << SO_STREAM_0_VERTEX_READ_LENGTH_SHIFT;
591 ilo_cp_begin(cp, cmd_len);
592 ilo_cp_write(cp, cmd | (cmd_len - 2));
593 ilo_cp_write(cp, dw1);
594 ilo_cp_write(cp, dw2);
599 gen7_emit_3DSTATE_SBE(const struct ilo_dev_info *dev,
600 const struct pipe_rasterizer_state *rasterizer,
601 const struct ilo_shader *fs,
602 const struct ilo_shader *last_sh,
605 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x1f);
606 const uint8_t cmd_len = 14;
609 ILO_GPE_VALID_GEN(dev, 7, 7);
611 ilo_gpe_gen6_fill_3dstate_sf_sbe(dev, rasterizer,
612 fs, last_sh, dw, Elements(dw));
614 ilo_cp_begin(cp, cmd_len);
615 ilo_cp_write(cp, cmd | (cmd_len - 2));
616 ilo_cp_write_multi(cp, dw, 13);
621 gen7_emit_3DSTATE_PS(const struct ilo_dev_info *dev,
622 const struct ilo_shader *fs,
623 int num_samplers, bool dual_blend,
626 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, 0x20);
627 const uint8_t cmd_len = 8;
628 uint32_t dw2, dw4, dw5;
631 ILO_GPE_VALID_GEN(dev, 7, 7);
633 /* see brwCreateContext() */
634 max_threads = (dev->gt == 2) ? 172 : 48;
637 ilo_cp_begin(cp, cmd_len);
638 ilo_cp_write(cp, cmd | (cmd_len - 2));
642 /* GPU hangs if none of the dispatch enable bits is set */
643 ilo_cp_write(cp, (max_threads - 1) << IVB_PS_MAX_THREADS_SHIFT |
644 GEN7_PS_8_DISPATCH_ENABLE);
653 dw2 = (num_samplers + 3) / 4 << GEN7_PS_SAMPLER_COUNT_SHIFT |
654 0 << GEN7_PS_BINDING_TABLE_ENTRY_COUNT_SHIFT;
656 dw2 |= GEN7_PS_FLOATING_POINT_MODE_ALT;
658 dw4 = (max_threads - 1) << IVB_PS_MAX_THREADS_SHIFT |
659 GEN7_PS_POSOFFSET_NONE;
662 dw4 |= GEN7_PS_PUSH_CONSTANT_ENABLE;
664 dw4 |= GEN7_PS_ATTRIBUTE_ENABLE;
666 dw4 |= GEN7_PS_DUAL_SOURCE_BLEND_ENABLE;
669 dw4 |= GEN7_PS_16_DISPATCH_ENABLE;
671 dw4 |= GEN7_PS_8_DISPATCH_ENABLE;
673 dw5 = fs->in.start_grf << GEN7_PS_DISPATCH_START_GRF_SHIFT_0 |
674 0 << GEN7_PS_DISPATCH_START_GRF_SHIFT_1 |
675 0 << GEN7_PS_DISPATCH_START_GRF_SHIFT_2;
677 ilo_cp_begin(cp, cmd_len);
678 ilo_cp_write(cp, cmd | (cmd_len - 2));
679 ilo_cp_write(cp, fs->cache_offset);
680 ilo_cp_write(cp, dw2);
681 ilo_cp_write(cp, 0); /* scratch */
682 ilo_cp_write(cp, dw4);
683 ilo_cp_write(cp, dw5);
684 ilo_cp_write(cp, 0); /* kernel 1 */
685 ilo_cp_write(cp, 0); /* kernel 2 */
690 gen7_emit_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP(const struct ilo_dev_info *dev,
691 uint32_t sf_clip_viewport,
694 gen7_emit_3dstate_pointer(dev, 0x21, sf_clip_viewport, cp);
698 gen7_emit_3DSTATE_VIEWPORT_STATE_POINTERS_CC(const struct ilo_dev_info *dev,
699 uint32_t cc_viewport,
702 gen7_emit_3dstate_pointer(dev, 0x23, cc_viewport, cp);
706 gen7_emit_3DSTATE_BLEND_STATE_POINTERS(const struct ilo_dev_info *dev,
707 uint32_t blend_state,
710 gen7_emit_3dstate_pointer(dev, 0x24, blend_state, cp);
714 gen7_emit_3DSTATE_DEPTH_STENCIL_STATE_POINTERS(const struct ilo_dev_info *dev,
715 uint32_t depth_stencil_state,
718 gen7_emit_3dstate_pointer(dev, 0x25, depth_stencil_state, cp);
722 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_VS(const struct ilo_dev_info *dev,
723 uint32_t binding_table,
726 gen7_emit_3dstate_pointer(dev, 0x26, binding_table, cp);
730 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_HS(const struct ilo_dev_info *dev,
731 uint32_t binding_table,
734 gen7_emit_3dstate_pointer(dev, 0x27, binding_table, cp);
738 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_DS(const struct ilo_dev_info *dev,
739 uint32_t binding_table,
742 gen7_emit_3dstate_pointer(dev, 0x28, binding_table, cp);
746 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_GS(const struct ilo_dev_info *dev,
747 uint32_t binding_table,
750 gen7_emit_3dstate_pointer(dev, 0x29, binding_table, cp);
754 gen7_emit_3DSTATE_BINDING_TABLE_POINTERS_PS(const struct ilo_dev_info *dev,
755 uint32_t binding_table,
758 gen7_emit_3dstate_pointer(dev, 0x2a, binding_table, cp);
762 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_VS(const struct ilo_dev_info *dev,
763 uint32_t sampler_state,
766 gen7_emit_3dstate_pointer(dev, 0x2b, sampler_state, cp);
770 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_HS(const struct ilo_dev_info *dev,
771 uint32_t sampler_state,
774 gen7_emit_3dstate_pointer(dev, 0x2c, sampler_state, cp);
778 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_DS(const struct ilo_dev_info *dev,
779 uint32_t sampler_state,
782 gen7_emit_3dstate_pointer(dev, 0x2d, sampler_state, cp);
786 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_GS(const struct ilo_dev_info *dev,
787 uint32_t sampler_state,
790 gen7_emit_3dstate_pointer(dev, 0x2e, sampler_state, cp);
794 gen7_emit_3DSTATE_SAMPLER_STATE_POINTERS_PS(const struct ilo_dev_info *dev,
795 uint32_t sampler_state,
798 gen7_emit_3dstate_pointer(dev, 0x2f, sampler_state, cp);
802 gen7_emit_3dstate_urb(const struct ilo_dev_info *dev,
803 int subop, int offset, int size,
807 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x0, subop);
808 const uint8_t cmd_len = 2;
809 const int row_size = 64; /* 512 bits */
810 int alloc_size, num_entries, min_entries, max_entries;
812 ILO_GPE_VALID_GEN(dev, 7, 7);
814 /* VS, HS, DS, and GS variants */
815 assert(subop >= 0x30 && subop <= 0x33);
817 /* in multiples of 8KB */
818 assert(offset % 8192 == 0);
821 /* in multiple of 512-bit rows */
822 alloc_size = (entry_size + row_size - 1) / row_size;
827 * From the Ivy Bridge PRM, volume 2 part 1, page 34:
829 * "VS URB Entry Allocation Size equal to 4(5 512-bit URB rows) may
830 * cause performance to decrease due to banking in the URB. Element
831 * sizes of 16 to 20 should be programmed with six 512-bit URB rows."
833 if (subop == 0x30 && alloc_size == 5)
836 /* in multiples of 8 */
837 num_entries = (size / row_size / alloc_size) & ~7;
840 case 0x30: /* 3DSTATE_URB_VS */
842 max_entries = (dev->gt == 2) ? 704 : 512;
844 assert(num_entries >= min_entries);
845 if (num_entries > max_entries)
846 num_entries = max_entries;
848 case 0x31: /* 3DSTATE_URB_HS */
849 max_entries = (dev->gt == 2) ? 64 : 32;
850 if (num_entries > max_entries)
851 num_entries = max_entries;
853 case 0x32: /* 3DSTATE_URB_DS */
855 assert(num_entries >= 138);
857 case 0x33: /* 3DSTATE_URB_GS */
858 max_entries = (dev->gt == 2) ? 320 : 192;
859 if (num_entries > max_entries)
860 num_entries = max_entries;
866 ilo_cp_begin(cp, cmd_len);
867 ilo_cp_write(cp, cmd | (cmd_len - 2));
868 ilo_cp_write(cp, offset << GEN7_URB_STARTING_ADDRESS_SHIFT |
869 (alloc_size - 1) << GEN7_URB_ENTRY_SIZE_SHIFT |
875 gen7_emit_3DSTATE_URB_VS(const struct ilo_dev_info *dev,
876 int offset, int size, int entry_size,
879 gen7_emit_3dstate_urb(dev, 0x30, offset, size, entry_size, cp);
883 gen7_emit_3DSTATE_URB_HS(const struct ilo_dev_info *dev,
884 int offset, int size, int entry_size,
887 gen7_emit_3dstate_urb(dev, 0x31, offset, size, entry_size, cp);
891 gen7_emit_3DSTATE_URB_DS(const struct ilo_dev_info *dev,
892 int offset, int size, int entry_size,
895 gen7_emit_3dstate_urb(dev, 0x32, offset, size, entry_size, cp);
899 gen7_emit_3DSTATE_URB_GS(const struct ilo_dev_info *dev,
900 int offset, int size, int entry_size,
903 gen7_emit_3dstate_urb(dev, 0x33, offset, size, entry_size, cp);
907 gen7_emit_3dstate_push_constant_alloc(const struct ilo_dev_info *dev,
908 int subop, int offset, int size,
911 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x1, subop);
912 const uint8_t cmd_len = 2;
915 ILO_GPE_VALID_GEN(dev, 7, 7);
917 /* VS, HS, DS, GS, and PS variants */
918 assert(subop >= 0x12 && subop <= 0x16);
921 * From the Ivy Bridge PRM, volume 2 part 1, page 68:
923 * "(A table that says the maximum size of each constant buffer is
926 * From the Ivy Bridge PRM, volume 2 part 1, page 115:
928 * "The sum of the Constant Buffer Offset and the Constant Buffer Size
929 * may not exceed the maximum value of the Constant Buffer Size."
931 * Thus, the valid range of buffer end is [0KB, 16KB].
933 end = (offset + size) / 1024;
935 assert(!"invalid constant buffer end");
939 /* the valid range of buffer offset is [0KB, 15KB] */
940 offset = (offset + 1023) / 1024;
942 assert(!"invalid constant buffer offset");
951 /* the valid range of buffer size is [0KB, 15KB] */
954 assert(!"invalid constant buffer size");
958 ilo_cp_begin(cp, cmd_len);
959 ilo_cp_write(cp, cmd | (cmd_len - 2));
960 ilo_cp_write(cp, offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT |
966 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_VS(const struct ilo_dev_info *dev,
967 int offset, int size,
970 gen7_emit_3dstate_push_constant_alloc(dev, 0x12, offset, size, cp);
974 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_HS(const struct ilo_dev_info *dev,
975 int offset, int size,
978 gen7_emit_3dstate_push_constant_alloc(dev, 0x13, offset, size, cp);
982 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_DS(const struct ilo_dev_info *dev,
983 int offset, int size,
986 gen7_emit_3dstate_push_constant_alloc(dev, 0x14, offset, size, cp);
990 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_GS(const struct ilo_dev_info *dev,
991 int offset, int size,
994 gen7_emit_3dstate_push_constant_alloc(dev, 0x15, offset, size, cp);
998 gen7_emit_3DSTATE_PUSH_CONSTANT_ALLOC_PS(const struct ilo_dev_info *dev,
999 int offset, int size,
1002 gen7_emit_3dstate_push_constant_alloc(dev, 0x16, offset, size, cp);
1006 gen7_emit_3DSTATE_SO_DECL_LIST(const struct ilo_dev_info *dev,
1007 const struct pipe_stream_output_info *so_info,
1008 const struct ilo_shader *sh,
1011 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x1, 0x17);
1013 int buffer_selects, num_entries, i;
1014 uint16_t so_decls[128];
1016 ILO_GPE_VALID_GEN(dev, 7, 7);
1022 int buffer_offsets[PIPE_MAX_SO_BUFFERS];
1024 memset(buffer_offsets, 0, sizeof(buffer_offsets));
1026 for (i = 0; i < so_info->num_outputs; i++) {
1027 unsigned decl, buf, attr, mask;
1029 buf = so_info->output[i].output_buffer;
1031 /* pad with holes */
1032 assert(buffer_offsets[buf] <= so_info->output[i].dst_offset);
1033 while (buffer_offsets[buf] < so_info->output[i].dst_offset) {
1036 num_dwords = so_info->output[i].dst_offset - buffer_offsets[buf];
1040 decl = buf << SO_DECL_OUTPUT_BUFFER_SLOT_SHIFT |
1042 ((1 << num_dwords) - 1) << SO_DECL_COMPONENT_MASK_SHIFT;
1044 so_decls[num_entries++] = decl;
1045 buffer_offsets[buf] += num_dwords;
1048 /* figure out which attribute is sourced */
1049 for (attr = 0; attr < sh->out.count; attr++) {
1050 const int idx = sh->out.register_indices[attr];
1051 if (idx == so_info->output[i].register_index)
1055 decl = buf << SO_DECL_OUTPUT_BUFFER_SLOT_SHIFT;
1057 if (attr < sh->out.count) {
1058 mask = ((1 << so_info->output[i].num_components) - 1) <<
1059 so_info->output[i].start_component;
1061 /* PSIZE is at W channel */
1062 if (sh->out.semantic_names[attr] == TGSI_SEMANTIC_PSIZE) {
1063 assert(mask == 0x1);
1064 mask = (mask << 3) & 0xf;
1067 decl |= attr << SO_DECL_REGISTER_INDEX_SHIFT |
1068 mask << SO_DECL_COMPONENT_MASK_SHIFT;
1071 assert(!"stream output an undefined register");
1072 mask = (1 << so_info->output[i].num_components) - 1;
1073 decl |= SO_DECL_HOLE_FLAG |
1074 mask << SO_DECL_COMPONENT_MASK_SHIFT;
1077 so_decls[num_entries++] = decl;
1078 buffer_selects |= 1 << buf;
1079 buffer_offsets[buf] += so_info->output[i].num_components;
1084 * From the Ivy Bridge PRM, volume 2 part 1, page 201:
1086 * "Errata: All 128 decls for all four streams must be included
1087 * whenever this command is issued. The "Num Entries [n]" fields still
1088 * contain the actual numbers of valid decls."
1090 * Also note that "DWord Length" has 9 bits for this command, and the type
1091 * of cmd_len is thus uint16_t.
1093 cmd_len = 2 * 128 + 3;
1095 ilo_cp_begin(cp, cmd_len);
1096 ilo_cp_write(cp, cmd | (cmd_len - 2));
1097 ilo_cp_write(cp, 0 << SO_STREAM_TO_BUFFER_SELECTS_3_SHIFT |
1098 0 << SO_STREAM_TO_BUFFER_SELECTS_2_SHIFT |
1099 0 << SO_STREAM_TO_BUFFER_SELECTS_1_SHIFT |
1100 buffer_selects << SO_STREAM_TO_BUFFER_SELECTS_0_SHIFT);
1101 ilo_cp_write(cp, 0 << SO_NUM_ENTRIES_3_SHIFT |
1102 0 << SO_NUM_ENTRIES_2_SHIFT |
1103 0 << SO_NUM_ENTRIES_1_SHIFT |
1104 num_entries << SO_NUM_ENTRIES_0_SHIFT);
1106 for (i = 0; i < num_entries; i++) {
1107 ilo_cp_write(cp, so_decls[i]);
1108 ilo_cp_write(cp, 0);
1110 for (; i < 128; i++) {
1111 ilo_cp_write(cp, 0);
1112 ilo_cp_write(cp, 0);
1119 gen7_emit_3DSTATE_SO_BUFFER(const struct ilo_dev_info *dev,
1120 int index, int base, int stride,
1121 const struct pipe_stream_output_target *so_target,
1124 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x1, 0x18);
1125 const uint8_t cmd_len = 4;
1126 struct ilo_buffer *buf;
1129 ILO_GPE_VALID_GEN(dev, 7, 7);
1131 if (!so_target || !so_target->buffer) {
1132 ilo_cp_begin(cp, cmd_len);
1133 ilo_cp_write(cp, cmd | (cmd_len - 2));
1134 ilo_cp_write(cp, index << SO_BUFFER_INDEX_SHIFT);
1135 ilo_cp_write(cp, 0);
1136 ilo_cp_write(cp, 0);
1141 buf = ilo_buffer(so_target->buffer);
1144 assert(stride % 4 == 0 && base % 4 == 0);
1145 assert(so_target->buffer_offset % 4 == 0);
1148 base = (base + so_target->buffer_offset) & ~3;
1149 end = (base + so_target->buffer_size) & ~3;
1151 ilo_cp_begin(cp, cmd_len);
1152 ilo_cp_write(cp, cmd | (cmd_len - 2));
1153 ilo_cp_write(cp, index << SO_BUFFER_INDEX_SHIFT |
1155 ilo_cp_write_bo(cp, base, buf->bo, INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER);
1156 ilo_cp_write_bo(cp, end, buf->bo, INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER);
1161 gen7_emit_3DPRIMITIVE(const struct ilo_dev_info *dev,
1162 const struct pipe_draw_info *info,
1166 const uint32_t cmd = ILO_GPE_CMD(0x3, 0x3, 0x00);
1167 const uint8_t cmd_len = 7;
1168 const int prim = (rectlist) ?
1169 _3DPRIM_RECTLIST : ilo_gpe_gen6_translate_pipe_prim(info->mode);
1170 const int vb_access = (info->indexed) ?
1171 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM :
1172 GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
1174 ILO_GPE_VALID_GEN(dev, 7, 7);
1176 ilo_cp_begin(cp, cmd_len);
1177 ilo_cp_write(cp, cmd | (cmd_len - 2));
1178 ilo_cp_write(cp, vb_access | prim);
1179 ilo_cp_write(cp, info->count);
1180 ilo_cp_write(cp, info->start);
1181 ilo_cp_write(cp, info->instance_count);
1182 ilo_cp_write(cp, info->start_instance);
1183 ilo_cp_write(cp, info->index_bias);
1188 gen7_emit_SF_CLIP_VIEWPORT(const struct ilo_dev_info *dev,
1189 const struct ilo_viewport_cso *viewports,
1190 unsigned num_viewports,
1193 const int state_align = 64 / 4;
1194 const int state_len = 16 * num_viewports;
1195 uint32_t state_offset, *dw;
1198 ILO_GPE_VALID_GEN(dev, 7, 7);
1201 * From the Ivy Bridge PRM, volume 2 part 1, page 270:
1203 * "The viewport-specific state used by both the SF and CL units
1204 * (SF_CLIP_VIEWPORT) is stored as an array of up to 16 elements, each
1205 * of which contains the DWords described below. The start of each
1206 * element is spaced 16 DWords apart. The location of first element of
1207 * the array, as specified by both Pointer to SF_VIEWPORT and Pointer
1208 * to CLIP_VIEWPORT, is aligned to a 64-byte boundary."
1210 assert(num_viewports && num_viewports <= 16);
1212 dw = ilo_cp_steal_ptr(cp, "SF_CLIP_VIEWPORT",
1213 state_len, state_align, &state_offset);
1215 for (i = 0; i < num_viewports; i++) {
1216 const struct ilo_viewport_cso *vp = &viewports[i];
1218 dw[0] = fui(vp->m00);
1219 dw[1] = fui(vp->m11);
1220 dw[2] = fui(vp->m22);
1221 dw[3] = fui(vp->m30);
1222 dw[4] = fui(vp->m31);
1223 dw[5] = fui(vp->m32);
1226 dw[8] = fui(vp->min_gbx);
1227 dw[9] = fui(vp->max_gbx);
1228 dw[10] = fui(vp->min_gby);
1229 dw[11] = fui(vp->max_gby);
1238 return state_offset;
1242 ilo_gpe_init_view_surface_null_gen7(const struct ilo_dev_info *dev,
1243 unsigned width, unsigned height,
1244 unsigned depth, unsigned level,
1245 struct ilo_view_surface *surf)
1249 ILO_GPE_VALID_GEN(dev, 7, 7);
1252 * From the Ivy Bridge PRM, volume 4 part 1, page 62:
1254 * "A null surface is used in instances where an actual surface is not
1255 * bound. When a write message is generated to a null surface, no
1256 * actual surface is written to. When a read message (including any
1257 * sampling engine message) is generated to a null surface, the result
1258 * is all zeros. Note that a null surface type is allowed to be used
1259 * with all messages, even if it is not specificially indicated as
1260 * supported. All of the remaining fields in surface state are ignored
1261 * for null surfaces, with the following exceptions:
1263 * * Width, Height, Depth, LOD, and Render Target View Extent fields
1264 * must match the depth buffer's corresponding state for all render
1265 * target surfaces, including null.
1266 * * All sampling engine and data port messages support null surfaces
1267 * with the above behavior, even if not mentioned as specifically
1268 * supported, except for the following:
1269 * * Data Port Media Block Read/Write messages.
1270 * * The Surface Type of a surface used as a render target (accessed
1271 * via the Data Port's Render Target Write message) must be the same
1272 * as the Surface Type of all other render targets and of the depth
1273 * buffer (defined in 3DSTATE_DEPTH_BUFFER), unless either the depth
1274 * buffer or render targets are SURFTYPE_NULL."
1276 * From the Ivy Bridge PRM, volume 4 part 1, page 65:
1278 * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be
1282 STATIC_ASSERT(Elements(surf->payload) >= 8);
1285 dw[0] = BRW_SURFACE_NULL << BRW_SURFACE_TYPE_SHIFT |
1286 BRW_SURFACEFORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT |
1287 BRW_SURFACE_TILED << 13;
1291 dw[2] = SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT) |
1292 SET_FIELD(width - 1, GEN7_SURFACE_WIDTH);
1294 dw[3] = SET_FIELD(depth - 1, BRW_SURFACE_DEPTH);
1306 ilo_gpe_init_view_surface_for_buffer_gen7(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 struct ilo_view_surface *surf)
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;
1322 ILO_GPE_VALID_GEN(dev, 7, 7);
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 STATIC_ASSERT(Elements(surf->payload) >= 8);
1400 dw[0] = surface_type << BRW_SURFACE_TYPE_SHIFT |
1401 surface_format << BRW_SURFACE_FORMAT_SHIFT;
1402 if (render_cache_rw)
1403 dw[0] |= BRW_SURFACE_RC_READ_WRITE;
1407 dw[2] = SET_FIELD(height, GEN7_SURFACE_HEIGHT) |
1408 SET_FIELD(width, GEN7_SURFACE_WIDTH);
1410 dw[3] = SET_FIELD(depth, BRW_SURFACE_DEPTH) |
1419 /* do not increment reference count */
1424 ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
1425 const struct ilo_texture *tex,
1426 enum pipe_format format,
1427 unsigned first_level,
1428 unsigned num_levels,
1429 unsigned first_layer,
1430 unsigned num_layers,
1431 bool is_rt, bool render_cache_rw,
1432 struct ilo_view_surface *surf)
1434 int surface_type, surface_format;
1435 int width, height, depth, pitch, lod;
1436 unsigned layer_offset, x_offset, y_offset;
1439 ILO_GPE_VALID_GEN(dev, 7, 7);
1441 surface_type = ilo_gpe_gen6_translate_texture(tex->base.target);
1442 assert(surface_type != BRW_SURFACE_BUFFER);
1444 if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT && tex->separate_s8)
1445 format = PIPE_FORMAT_Z32_FLOAT;
1448 surface_format = ilo_translate_render_format(format);
1450 surface_format = ilo_translate_texture_format(format);
1451 assert(surface_format >= 0);
1453 width = tex->base.width0;
1454 height = tex->base.height0;
1455 depth = (tex->base.target == PIPE_TEXTURE_3D) ?
1456 tex->base.depth0 : num_layers;
1457 pitch = tex->bo_stride;
1459 if (surface_type == BRW_SURFACE_CUBE) {
1461 * From the Ivy Bridge PRM, volume 4 part 1, page 70:
1463 * "For SURFTYPE_CUBE:For Sampling Engine Surfaces, the range of
1464 * this field is [0,340], indicating the number of cube array
1465 * elements (equal to the number of underlying 2D array elements
1466 * divided by 6). For other surfaces, this field must be zero."
1468 * When is_rt is true, we treat the texture as a 2D one to avoid the
1472 surface_type = BRW_SURFACE_2D;
1475 assert(num_layers % 6 == 0);
1476 depth = num_layers / 6;
1480 /* sanity check the size */
1481 assert(width >= 1 && height >= 1 && depth >= 1 && pitch >= 1);
1482 assert(first_layer < 2048 && num_layers <= 2048);
1483 switch (surface_type) {
1484 case BRW_SURFACE_1D:
1485 assert(width <= 16384 && height == 1 && depth <= 2048);
1487 case BRW_SURFACE_2D:
1488 assert(width <= 16384 && height <= 16384 && depth <= 2048);
1490 case BRW_SURFACE_3D:
1491 assert(width <= 2048 && height <= 2048 && depth <= 2048);
1493 assert(first_layer == 0);
1495 case BRW_SURFACE_CUBE:
1496 assert(width <= 16384 && height <= 16384 && depth <= 86);
1497 assert(width == height);
1499 assert(first_layer == 0);
1502 assert(!"unexpected surface type");
1508 * Compute the offset to the layer manually.
1510 * For rendering, the hardware requires LOD to be the same for all
1511 * render targets and the depth buffer. We need to compute the offset
1512 * to the layer manually and always set LOD to 0.
1515 /* we lose the capability for layered rendering */
1516 assert(num_layers == 1);
1518 layer_offset = ilo_texture_get_slice_offset(tex,
1519 first_level, first_layer, &x_offset, &y_offset);
1521 assert(x_offset % 4 == 0);
1522 assert(y_offset % 2 == 0);
1526 /* derive the size for the LOD */
1527 width = u_minify(width, first_level);
1528 height = u_minify(height, first_level);
1529 if (surface_type == BRW_SURFACE_3D)
1530 depth = u_minify(depth, first_level);
1544 assert(num_levels == 1);
1552 lod = num_levels - 1;
1556 * From the Ivy Bridge PRM, volume 4 part 1, page 68:
1558 * "The Base Address for linear render target surfaces and surfaces
1559 * accessed with the typed surface read/write data port messages must
1560 * be element-size aligned, for non-YUV surface formats, or a multiple
1561 * of 2 element-sizes for YUV surface formats. Other linear surfaces
1562 * have no alignment requirements (byte alignment is sufficient)."
1564 * From the Ivy Bridge PRM, volume 4 part 1, page 70:
1566 * "For linear render target surfaces and surfaces accessed with the
1567 * typed data port messages, the pitch must be a multiple of the
1568 * element size for non-YUV surface formats. Pitch must be a multiple
1569 * of 2 * element size for YUV surface formats. For linear surfaces
1570 * with Surface Type of SURFTYPE_STRBUF, the pitch must be a multiple
1571 * of 4 bytes.For other linear surfaces, the pitch can be any multiple
1574 * From the Ivy Bridge PRM, volume 4 part 1, page 74:
1576 * "For linear surfaces, this field (X Offset) must be zero."
1578 if (tex->tiling == INTEL_TILING_NONE) {
1580 const int elem_size = util_format_get_blocksize(format);
1581 assert(layer_offset % elem_size == 0);
1582 assert(pitch % elem_size == 0);
1588 STATIC_ASSERT(Elements(surf->payload) >= 8);
1591 dw[0] = surface_type << BRW_SURFACE_TYPE_SHIFT |
1592 surface_format << BRW_SURFACE_FORMAT_SHIFT |
1593 ilo_gpe_gen6_translate_winsys_tiling(tex->tiling) << 13;
1596 * From the Ivy Bridge PRM, volume 4 part 1, page 63:
1598 * "If this field (Surface Array) is enabled, the Surface Type must be
1599 * SURFTYPE_1D, SURFTYPE_2D, or SURFTYPE_CUBE. If this field is
1600 * disabled and Surface Type is SURFTYPE_1D, SURFTYPE_2D, or
1601 * SURFTYPE_CUBE, the Depth field must be set to zero."
1603 * For non-3D sampler surfaces, resinfo (the sampler message) always
1604 * returns zero for the number of layers when this field is not set.
1606 if (surface_type != BRW_SURFACE_3D) {
1607 if (util_resource_is_array_texture(&tex->base))
1608 dw[0] |= GEN7_SURFACE_IS_ARRAY;
1614 dw[0] |= GEN7_SURFACE_VALIGN_4;
1617 dw[0] |= GEN7_SURFACE_HALIGN_8;
1619 if (tex->array_spacing_full)
1620 dw[0] |= GEN7_SURFACE_ARYSPC_FULL;
1622 dw[0] |= GEN7_SURFACE_ARYSPC_LOD0;
1624 if (render_cache_rw)
1625 dw[0] |= BRW_SURFACE_RC_READ_WRITE;
1627 if (surface_type == BRW_SURFACE_CUBE && !is_rt)
1628 dw[0] |= BRW_SURFACE_CUBEFACE_ENABLES;
1630 dw[1] = layer_offset;
1632 dw[2] = SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT) |
1633 SET_FIELD(width - 1, GEN7_SURFACE_WIDTH);
1635 dw[3] = SET_FIELD(depth - 1, BRW_SURFACE_DEPTH) |
1638 dw[4] = first_layer << 18 |
1639 (num_layers - 1) << 7;
1642 * MSFMT_MSS means the samples are not interleaved and MSFMT_DEPTH_STENCIL
1643 * means the samples are interleaved. The layouts are the same when the
1644 * number of samples is 1.
1646 if (tex->interleaved && tex->base.nr_samples > 1) {
1648 dw[4] |= GEN7_SURFACE_MSFMT_DEPTH_STENCIL;
1651 dw[4] |= GEN7_SURFACE_MSFMT_MSS;
1654 if (tex->base.nr_samples > 4)
1655 dw[4] |= GEN7_SURFACE_MULTISAMPLECOUNT_8;
1656 else if (tex->base.nr_samples > 2)
1657 dw[4] |= GEN7_SURFACE_MULTISAMPLECOUNT_4;
1659 dw[4] |= GEN7_SURFACE_MULTISAMPLECOUNT_1;
1661 dw[5] = x_offset << BRW_SURFACE_X_OFFSET_SHIFT |
1662 y_offset << BRW_SURFACE_Y_OFFSET_SHIFT |
1663 SET_FIELD(first_level, GEN7_SURFACE_MIN_LOD) |
1669 /* do not increment reference count */
1674 gen7_estimate_command_size(const struct ilo_dev_info *dev,
1675 enum ilo_gpe_gen7_command cmd,
1678 static const struct {
1681 } gen7_command_size_table[ILO_GPE_GEN7_COMMAND_COUNT] = {
1682 [ILO_GPE_GEN7_STATE_BASE_ADDRESS] = { 0, 10 },
1683 [ILO_GPE_GEN7_STATE_SIP] = { 0, 2 },
1684 [ILO_GPE_GEN7_3DSTATE_VF_STATISTICS] = { 0, 1 },
1685 [ILO_GPE_GEN7_PIPELINE_SELECT] = { 0, 1 },
1686 [ILO_GPE_GEN7_MEDIA_VFE_STATE] = { 0, 8 },
1687 [ILO_GPE_GEN7_MEDIA_CURBE_LOAD] = { 0, 4 },
1688 [ILO_GPE_GEN7_MEDIA_INTERFACE_DESCRIPTOR_LOAD] = { 0, 4 },
1689 [ILO_GPE_GEN7_MEDIA_STATE_FLUSH] = { 0, 2 },
1690 [ILO_GPE_GEN7_GPGPU_WALKER] = { 0, 11 },
1691 [ILO_GPE_GEN7_3DSTATE_CLEAR_PARAMS] = { 0, 3 },
1692 [ILO_GPE_GEN7_3DSTATE_DEPTH_BUFFER] = { 0, 7 },
1693 [ILO_GPE_GEN7_3DSTATE_STENCIL_BUFFER] = { 0, 3 },
1694 [ILO_GPE_GEN7_3DSTATE_HIER_DEPTH_BUFFER] = { 0, 3 },
1695 [ILO_GPE_GEN7_3DSTATE_VERTEX_BUFFERS] = { 1, 4 },
1696 [ILO_GPE_GEN7_3DSTATE_VERTEX_ELEMENTS] = { 1, 2 },
1697 [ILO_GPE_GEN7_3DSTATE_INDEX_BUFFER] = { 0, 3 },
1698 [ILO_GPE_GEN7_3DSTATE_CC_STATE_POINTERS] = { 0, 2 },
1699 [ILO_GPE_GEN7_3DSTATE_SCISSOR_STATE_POINTERS] = { 0, 2 },
1700 [ILO_GPE_GEN7_3DSTATE_VS] = { 0, 6 },
1701 [ILO_GPE_GEN7_3DSTATE_GS] = { 0, 7 },
1702 [ILO_GPE_GEN7_3DSTATE_CLIP] = { 0, 4 },
1703 [ILO_GPE_GEN7_3DSTATE_SF] = { 0, 7 },
1704 [ILO_GPE_GEN7_3DSTATE_WM] = { 0, 3 },
1705 [ILO_GPE_GEN7_3DSTATE_CONSTANT_VS] = { 0, 7 },
1706 [ILO_GPE_GEN7_3DSTATE_CONSTANT_GS] = { 0, 7 },
1707 [ILO_GPE_GEN7_3DSTATE_CONSTANT_PS] = { 0, 7 },
1708 [ILO_GPE_GEN7_3DSTATE_SAMPLE_MASK] = { 0, 2 },
1709 [ILO_GPE_GEN7_3DSTATE_CONSTANT_HS] = { 0, 7 },
1710 [ILO_GPE_GEN7_3DSTATE_CONSTANT_DS] = { 0, 7 },
1711 [ILO_GPE_GEN7_3DSTATE_HS] = { 0, 7 },
1712 [ILO_GPE_GEN7_3DSTATE_TE] = { 0, 4 },
1713 [ILO_GPE_GEN7_3DSTATE_DS] = { 0, 6 },
1714 [ILO_GPE_GEN7_3DSTATE_STREAMOUT] = { 0, 3 },
1715 [ILO_GPE_GEN7_3DSTATE_SBE] = { 0, 14 },
1716 [ILO_GPE_GEN7_3DSTATE_PS] = { 0, 8 },
1717 [ILO_GPE_GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP] = { 0, 2 },
1718 [ILO_GPE_GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC] = { 0, 2 },
1719 [ILO_GPE_GEN7_3DSTATE_BLEND_STATE_POINTERS] = { 0, 2 },
1720 [ILO_GPE_GEN7_3DSTATE_DEPTH_STENCIL_STATE_POINTERS] = { 0, 2 },
1721 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_VS] = { 0, 2 },
1722 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_HS] = { 0, 2 },
1723 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_DS] = { 0, 2 },
1724 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_GS] = { 0, 2 },
1725 [ILO_GPE_GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS] = { 0, 2 },
1726 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_VS] = { 0, 2 },
1727 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_HS] = { 0, 2 },
1728 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_DS] = { 0, 2 },
1729 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_GS] = { 0, 2 },
1730 [ILO_GPE_GEN7_3DSTATE_SAMPLER_STATE_POINTERS_PS] = { 0, 2 },
1731 [ILO_GPE_GEN7_3DSTATE_URB_VS] = { 0, 2 },
1732 [ILO_GPE_GEN7_3DSTATE_URB_HS] = { 0, 2 },
1733 [ILO_GPE_GEN7_3DSTATE_URB_DS] = { 0, 2 },
1734 [ILO_GPE_GEN7_3DSTATE_URB_GS] = { 0, 2 },
1735 [ILO_GPE_GEN7_3DSTATE_DRAWING_RECTANGLE] = { 0, 4 },
1736 [ILO_GPE_GEN7_3DSTATE_POLY_STIPPLE_OFFSET] = { 0, 2 },
1737 [ILO_GPE_GEN7_3DSTATE_POLY_STIPPLE_PATTERN] = { 0, 33, },
1738 [ILO_GPE_GEN7_3DSTATE_LINE_STIPPLE] = { 0, 3 },
1739 [ILO_GPE_GEN7_3DSTATE_AA_LINE_PARAMETERS] = { 0, 3 },
1740 [ILO_GPE_GEN7_3DSTATE_MULTISAMPLE] = { 0, 4 },
1741 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_VS] = { 0, 2 },
1742 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_HS] = { 0, 2 },
1743 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_DS] = { 0, 2 },
1744 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_GS] = { 0, 2 },
1745 [ILO_GPE_GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_PS] = { 0, 2 },
1746 [ILO_GPE_GEN7_3DSTATE_SO_DECL_LIST] = { 3, 2 },
1747 [ILO_GPE_GEN7_3DSTATE_SO_BUFFER] = { 0, 4 },
1748 [ILO_GPE_GEN7_PIPE_CONTROL] = { 0, 5 },
1749 [ILO_GPE_GEN7_3DPRIMITIVE] = { 0, 7 },
1751 const int header = gen7_command_size_table[cmd].header;
1752 const int body = gen7_command_size_table[cmd].body;
1753 const int count = arg;
1755 ILO_GPE_VALID_GEN(dev, 7, 7);
1756 assert(cmd < ILO_GPE_GEN7_COMMAND_COUNT);
1758 return (likely(count)) ? header + body * count : 0;
1762 gen7_estimate_state_size(const struct ilo_dev_info *dev,
1763 enum ilo_gpe_gen7_state state,
1766 static const struct {
1770 } gen7_state_size_table[ILO_GPE_GEN7_STATE_COUNT] = {
1771 [ILO_GPE_GEN7_INTERFACE_DESCRIPTOR_DATA] = { 8, 8, true },
1772 [ILO_GPE_GEN7_SF_CLIP_VIEWPORT] = { 16, 16, true },
1773 [ILO_GPE_GEN7_CC_VIEWPORT] = { 8, 2, true },
1774 [ILO_GPE_GEN7_COLOR_CALC_STATE] = { 16, 6, false },
1775 [ILO_GPE_GEN7_BLEND_STATE] = { 16, 2, true },
1776 [ILO_GPE_GEN7_DEPTH_STENCIL_STATE] = { 16, 3, false },
1777 [ILO_GPE_GEN7_SCISSOR_RECT] = { 8, 2, true },
1778 [ILO_GPE_GEN7_BINDING_TABLE_STATE] = { 8, 1, true },
1779 [ILO_GPE_GEN7_SURFACE_STATE] = { 8, 8, false },
1780 [ILO_GPE_GEN7_SAMPLER_STATE] = { 8, 4, true },
1781 [ILO_GPE_GEN7_SAMPLER_BORDER_COLOR_STATE] = { 8, 4, false },
1782 [ILO_GPE_GEN7_PUSH_CONSTANT_BUFFER] = { 8, 1, true },
1784 const int alignment = gen7_state_size_table[state].alignment;
1785 const int body = gen7_state_size_table[state].body;
1786 const bool is_array = gen7_state_size_table[state].is_array;
1787 const int count = arg;
1790 ILO_GPE_VALID_GEN(dev, 7, 7);
1791 assert(state < ILO_GPE_GEN7_STATE_COUNT);
1793 if (likely(count)) {
1795 estimate = (alignment - 1) + body * count;
1798 estimate = (alignment - 1) + body;
1799 /* all states are aligned */
1801 estimate += util_align_npot(body, alignment) * (count - 1);
1812 gen7_init(struct ilo_gpe_gen7 *gen7)
1814 const struct ilo_gpe_gen6 *gen6 = ilo_gpe_gen6_get();
1816 gen7->estimate_command_size = gen7_estimate_command_size;
1817 gen7->estimate_state_size = gen7_estimate_state_size;
1819 #define GEN7_USE(gen7, name, from) gen7->emit_ ## name = from->emit_ ## name
1820 #define GEN7_SET(gen7, name) gen7->emit_ ## name = gen7_emit_ ## name
1821 GEN7_USE(gen7, STATE_BASE_ADDRESS, gen6);
1822 GEN7_USE(gen7, STATE_SIP, gen6);
1823 GEN7_USE(gen7, 3DSTATE_VF_STATISTICS, gen6);
1824 GEN7_USE(gen7, PIPELINE_SELECT, gen6);
1825 GEN7_USE(gen7, MEDIA_VFE_STATE, gen6);
1826 GEN7_USE(gen7, MEDIA_CURBE_LOAD, gen6);
1827 GEN7_USE(gen7, MEDIA_INTERFACE_DESCRIPTOR_LOAD, gen6);
1828 GEN7_USE(gen7, MEDIA_STATE_FLUSH, gen6);
1829 GEN7_SET(gen7, GPGPU_WALKER);
1830 GEN7_SET(gen7, 3DSTATE_CLEAR_PARAMS);
1831 GEN7_USE(gen7, 3DSTATE_DEPTH_BUFFER, gen6);
1832 GEN7_USE(gen7, 3DSTATE_STENCIL_BUFFER, gen6);
1833 GEN7_USE(gen7, 3DSTATE_HIER_DEPTH_BUFFER, gen6);
1834 GEN7_USE(gen7, 3DSTATE_VERTEX_BUFFERS, gen6);
1835 GEN7_USE(gen7, 3DSTATE_VERTEX_ELEMENTS, gen6);
1836 GEN7_USE(gen7, 3DSTATE_INDEX_BUFFER, gen6);
1837 GEN7_SET(gen7, 3DSTATE_CC_STATE_POINTERS);
1838 GEN7_USE(gen7, 3DSTATE_SCISSOR_STATE_POINTERS, gen6);
1839 GEN7_USE(gen7, 3DSTATE_VS, gen6);
1840 GEN7_SET(gen7, 3DSTATE_GS);
1841 GEN7_USE(gen7, 3DSTATE_CLIP, gen6);
1842 GEN7_SET(gen7, 3DSTATE_SF);
1843 GEN7_SET(gen7, 3DSTATE_WM);
1844 GEN7_SET(gen7, 3DSTATE_CONSTANT_VS);
1845 GEN7_SET(gen7, 3DSTATE_CONSTANT_GS);
1846 GEN7_SET(gen7, 3DSTATE_CONSTANT_PS);
1847 GEN7_SET(gen7, 3DSTATE_SAMPLE_MASK);
1848 GEN7_SET(gen7, 3DSTATE_CONSTANT_HS);
1849 GEN7_SET(gen7, 3DSTATE_CONSTANT_DS);
1850 GEN7_SET(gen7, 3DSTATE_HS);
1851 GEN7_SET(gen7, 3DSTATE_TE);
1852 GEN7_SET(gen7, 3DSTATE_DS);
1853 GEN7_SET(gen7, 3DSTATE_STREAMOUT);
1854 GEN7_SET(gen7, 3DSTATE_SBE);
1855 GEN7_SET(gen7, 3DSTATE_PS);
1856 GEN7_SET(gen7, 3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP);
1857 GEN7_SET(gen7, 3DSTATE_VIEWPORT_STATE_POINTERS_CC);
1858 GEN7_SET(gen7, 3DSTATE_BLEND_STATE_POINTERS);
1859 GEN7_SET(gen7, 3DSTATE_DEPTH_STENCIL_STATE_POINTERS);
1860 GEN7_SET(gen7, 3DSTATE_BINDING_TABLE_POINTERS_VS);
1861 GEN7_SET(gen7, 3DSTATE_BINDING_TABLE_POINTERS_HS);
1862 GEN7_SET(gen7, 3DSTATE_BINDING_TABLE_POINTERS_DS);
1863 GEN7_SET(gen7, 3DSTATE_BINDING_TABLE_POINTERS_GS);
1864 GEN7_SET(gen7, 3DSTATE_BINDING_TABLE_POINTERS_PS);
1865 GEN7_SET(gen7, 3DSTATE_SAMPLER_STATE_POINTERS_VS);
1866 GEN7_SET(gen7, 3DSTATE_SAMPLER_STATE_POINTERS_HS);
1867 GEN7_SET(gen7, 3DSTATE_SAMPLER_STATE_POINTERS_DS);
1868 GEN7_SET(gen7, 3DSTATE_SAMPLER_STATE_POINTERS_GS);
1869 GEN7_SET(gen7, 3DSTATE_SAMPLER_STATE_POINTERS_PS);
1870 GEN7_SET(gen7, 3DSTATE_URB_VS);
1871 GEN7_SET(gen7, 3DSTATE_URB_HS);
1872 GEN7_SET(gen7, 3DSTATE_URB_DS);
1873 GEN7_SET(gen7, 3DSTATE_URB_GS);
1874 GEN7_USE(gen7, 3DSTATE_DRAWING_RECTANGLE, gen6);
1875 GEN7_USE(gen7, 3DSTATE_POLY_STIPPLE_OFFSET, gen6);
1876 GEN7_USE(gen7, 3DSTATE_POLY_STIPPLE_PATTERN, gen6);
1877 GEN7_USE(gen7, 3DSTATE_LINE_STIPPLE, gen6);
1878 GEN7_USE(gen7, 3DSTATE_AA_LINE_PARAMETERS, gen6);
1879 GEN7_USE(gen7, 3DSTATE_MULTISAMPLE, gen6);
1880 GEN7_SET(gen7, 3DSTATE_PUSH_CONSTANT_ALLOC_VS);
1881 GEN7_SET(gen7, 3DSTATE_PUSH_CONSTANT_ALLOC_HS);
1882 GEN7_SET(gen7, 3DSTATE_PUSH_CONSTANT_ALLOC_DS);
1883 GEN7_SET(gen7, 3DSTATE_PUSH_CONSTANT_ALLOC_GS);
1884 GEN7_SET(gen7, 3DSTATE_PUSH_CONSTANT_ALLOC_PS);
1885 GEN7_SET(gen7, 3DSTATE_SO_DECL_LIST);
1886 GEN7_SET(gen7, 3DSTATE_SO_BUFFER);
1887 GEN7_USE(gen7, PIPE_CONTROL, gen6);
1888 GEN7_SET(gen7, 3DPRIMITIVE);
1889 GEN7_USE(gen7, INTERFACE_DESCRIPTOR_DATA, gen6);
1890 GEN7_SET(gen7, SF_CLIP_VIEWPORT);
1891 GEN7_USE(gen7, CC_VIEWPORT, gen6);
1892 GEN7_USE(gen7, COLOR_CALC_STATE, gen6);
1893 GEN7_USE(gen7, BLEND_STATE, gen6);
1894 GEN7_USE(gen7, DEPTH_STENCIL_STATE, gen6);
1895 GEN7_USE(gen7, SCISSOR_RECT, gen6);
1896 GEN7_USE(gen7, BINDING_TABLE_STATE, gen6);
1897 GEN7_USE(gen7, SURFACE_STATE, gen6);
1898 GEN7_USE(gen7, SAMPLER_STATE, gen6);
1899 GEN7_USE(gen7, SAMPLER_BORDER_COLOR_STATE, gen6);
1900 GEN7_USE(gen7, push_constant_buffer, gen6);
1905 static struct ilo_gpe_gen7 gen7_gpe;
1907 const struct ilo_gpe_gen7 *
1908 ilo_gpe_gen7_get(void)
1910 if (!gen7_gpe.estimate_command_size)
1911 gen7_init(&gen7_gpe);