intel: Split gen_device_info out into libintel_dev
[mesa.git] / src / intel / blorp / blorp_genX_exec.h
1 /*
2 * Copyright © 2016 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #ifndef BLORP_GENX_EXEC_H
25 #define BLORP_GENX_EXEC_H
26
27 #include "blorp_priv.h"
28 #include "dev/gen_device_info.h"
29 #include "common/gen_sample_positions.h"
30 #include "genxml/gen_macros.h"
31
32 /**
33 * This file provides the blorp pipeline setup and execution functionality.
34 * It defines the following function:
35 *
36 * static void
37 * blorp_exec(struct blorp_context *blorp, void *batch_data,
38 * const struct blorp_params *params);
39 *
40 * It is the job of whoever includes this header to wrap this in something
41 * to get an externally visible symbol.
42 *
43 * In order for the blorp_exec function to work, the driver must provide
44 * implementations of the following static helper functions.
45 */
46
47 static void *
48 blorp_emit_dwords(struct blorp_batch *batch, unsigned n);
49
50 static uint64_t
51 blorp_emit_reloc(struct blorp_batch *batch,
52 void *location, struct blorp_address address, uint32_t delta);
53
54 static void *
55 blorp_alloc_dynamic_state(struct blorp_batch *batch,
56 uint32_t size,
57 uint32_t alignment,
58 uint32_t *offset);
59 static void *
60 blorp_alloc_vertex_buffer(struct blorp_batch *batch, uint32_t size,
61 struct blorp_address *addr);
62
63 #if GEN_GEN >= 8
64 static struct blorp_address
65 blorp_get_workaround_page(struct blorp_batch *batch);
66 #endif
67
68 static void
69 blorp_alloc_binding_table(struct blorp_batch *batch, unsigned num_entries,
70 unsigned state_size, unsigned state_alignment,
71 uint32_t *bt_offset, uint32_t *surface_offsets,
72 void **surface_maps);
73
74 static void
75 blorp_flush_range(struct blorp_batch *batch, void *start, size_t size);
76
77 static void
78 blorp_surface_reloc(struct blorp_batch *batch, uint32_t ss_offset,
79 struct blorp_address address, uint32_t delta);
80
81 #if GEN_GEN >= 7
82 static struct blorp_address
83 blorp_get_surface_base_address(struct blorp_batch *batch);
84 #endif
85
86 static void
87 blorp_emit_urb_config(struct blorp_batch *batch,
88 unsigned vs_entry_size, unsigned sf_entry_size);
89
90 static void
91 blorp_emit_pipeline(struct blorp_batch *batch,
92 const struct blorp_params *params);
93
94 /***** BEGIN blorp_exec implementation ******/
95
96 static uint64_t
97 _blorp_combine_address(struct blorp_batch *batch, void *location,
98 struct blorp_address address, uint32_t delta)
99 {
100 if (address.buffer == NULL) {
101 return address.offset + delta;
102 } else {
103 return blorp_emit_reloc(batch, location, address, delta);
104 }
105 }
106
107 #define __gen_address_type struct blorp_address
108 #define __gen_user_data struct blorp_batch
109 #define __gen_combine_address _blorp_combine_address
110
111 #include "genxml/genX_pack.h"
112
113 #define _blorp_cmd_length(cmd) cmd ## _length
114 #define _blorp_cmd_length_bias(cmd) cmd ## _length_bias
115 #define _blorp_cmd_header(cmd) cmd ## _header
116 #define _blorp_cmd_pack(cmd) cmd ## _pack
117
118 #define blorp_emit(batch, cmd, name) \
119 for (struct cmd name = { _blorp_cmd_header(cmd) }, \
120 *_dst = blorp_emit_dwords(batch, _blorp_cmd_length(cmd)); \
121 __builtin_expect(_dst != NULL, 1); \
122 _blorp_cmd_pack(cmd)(batch, (void *)_dst, &name), \
123 _dst = NULL)
124
125 #define blorp_emitn(batch, cmd, n) ({ \
126 uint32_t *_dw = blorp_emit_dwords(batch, n); \
127 if (_dw) { \
128 struct cmd template = { \
129 _blorp_cmd_header(cmd), \
130 .DWordLength = n - _blorp_cmd_length_bias(cmd), \
131 }; \
132 _blorp_cmd_pack(cmd)(batch, _dw, &template); \
133 } \
134 _dw ? _dw + 1 : NULL; /* Array starts at dw[1] */ \
135 })
136
137 #define STRUCT_ZERO(S) ({ struct S t; memset(&t, 0, sizeof(t)); t; })
138
139 #define blorp_emit_dynamic(batch, state, name, align, offset) \
140 for (struct state name = STRUCT_ZERO(state), \
141 *_dst = blorp_alloc_dynamic_state(batch, \
142 _blorp_cmd_length(state) * 4, \
143 align, offset); \
144 __builtin_expect(_dst != NULL, 1); \
145 _blorp_cmd_pack(state)(batch, (void *)_dst, &name), \
146 blorp_flush_range(batch, _dst, _blorp_cmd_length(state) * 4), \
147 _dst = NULL)
148
149 /* 3DSTATE_URB
150 * 3DSTATE_URB_VS
151 * 3DSTATE_URB_HS
152 * 3DSTATE_URB_DS
153 * 3DSTATE_URB_GS
154 *
155 * Assign the entire URB to the VS. Even though the VS disabled, URB space
156 * is still needed because the clipper loads the VUE's from the URB. From
157 * the Sandybridge PRM, Volume 2, Part 1, Section 3DSTATE,
158 * Dword 1.15:0 "VS Number of URB Entries":
159 * This field is always used (even if VS Function Enable is DISABLED).
160 *
161 * The warning below appears in the PRM (Section 3DSTATE_URB), but we can
162 * safely ignore it because this batch contains only one draw call.
163 * Because of URB corruption caused by allocating a previous GS unit
164 * URB entry to the VS unit, software is required to send a “GS NULL
165 * Fence” (Send URB fence with VS URB size == 1 and GS URB size == 0)
166 * plus a dummy DRAW call before any case where VS will be taking over
167 * GS URB space.
168 *
169 * If the 3DSTATE_URB_VS is emitted, than the others must be also.
170 * From the Ivybridge PRM, Volume 2 Part 1, section 1.7.1 3DSTATE_URB_VS:
171 *
172 * 3DSTATE_URB_HS, 3DSTATE_URB_DS, and 3DSTATE_URB_GS must also be
173 * programmed in order for the programming of this state to be
174 * valid.
175 */
176 static void
177 emit_urb_config(struct blorp_batch *batch,
178 const struct blorp_params *params)
179 {
180 /* Once vertex fetcher has written full VUE entries with complete
181 * header the space requirement is as follows per vertex (in bytes):
182 *
183 * Header Position Program constants
184 * +--------+------------+-------------------+
185 * | 16 | 16 | n x 16 |
186 * +--------+------------+-------------------+
187 *
188 * where 'n' stands for number of varying inputs expressed as vec4s.
189 */
190 const unsigned num_varyings =
191 params->wm_prog_data ? params->wm_prog_data->num_varying_inputs : 0;
192 const unsigned total_needed = 16 + 16 + num_varyings * 16;
193
194 /* The URB size is expressed in units of 64 bytes (512 bits) */
195 const unsigned vs_entry_size = DIV_ROUND_UP(total_needed, 64);
196
197 const unsigned sf_entry_size =
198 params->sf_prog_data ? params->sf_prog_data->urb_entry_size : 0;
199
200 blorp_emit_urb_config(batch, vs_entry_size, sf_entry_size);
201 }
202
203 static void
204 blorp_emit_vertex_data(struct blorp_batch *batch,
205 const struct blorp_params *params,
206 struct blorp_address *addr,
207 uint32_t *size)
208 {
209 const float vertices[] = {
210 /* v0 */ (float)params->x1, (float)params->y1, params->z,
211 /* v1 */ (float)params->x0, (float)params->y1, params->z,
212 /* v2 */ (float)params->x0, (float)params->y0, params->z,
213 };
214
215 void *data = blorp_alloc_vertex_buffer(batch, sizeof(vertices), addr);
216 memcpy(data, vertices, sizeof(vertices));
217 *size = sizeof(vertices);
218 blorp_flush_range(batch, data, *size);
219 }
220
221 static void
222 blorp_emit_input_varying_data(struct blorp_batch *batch,
223 const struct blorp_params *params,
224 struct blorp_address *addr,
225 uint32_t *size)
226 {
227 const unsigned vec4_size_in_bytes = 4 * sizeof(float);
228 const unsigned max_num_varyings =
229 DIV_ROUND_UP(sizeof(params->wm_inputs), vec4_size_in_bytes);
230 const unsigned num_varyings =
231 params->wm_prog_data ? params->wm_prog_data->num_varying_inputs : 0;
232
233 *size = 16 + num_varyings * vec4_size_in_bytes;
234
235 const uint32_t *const inputs_src = (const uint32_t *)&params->wm_inputs;
236 void *data = blorp_alloc_vertex_buffer(batch, *size, addr);
237 uint32_t *inputs = data;
238
239 /* Copy in the VS inputs */
240 assert(sizeof(params->vs_inputs) == 16);
241 memcpy(inputs, &params->vs_inputs, sizeof(params->vs_inputs));
242 inputs += 4;
243
244 if (params->wm_prog_data) {
245 /* Walk over the attribute slots, determine if the attribute is used by
246 * the program and when necessary copy the values from the input storage
247 * to the vertex data buffer.
248 */
249 for (unsigned i = 0; i < max_num_varyings; i++) {
250 const gl_varying_slot attr = VARYING_SLOT_VAR0 + i;
251
252 const int input_index = params->wm_prog_data->urb_setup[attr];
253 if (input_index < 0)
254 continue;
255
256 memcpy(inputs, inputs_src + i * 4, vec4_size_in_bytes);
257
258 inputs += 4;
259 }
260 }
261
262 blorp_flush_range(batch, data, *size);
263 }
264
265 static void
266 blorp_fill_vertex_buffer_state(struct blorp_batch *batch,
267 struct GENX(VERTEX_BUFFER_STATE) *vb,
268 unsigned idx,
269 struct blorp_address addr, uint32_t size,
270 uint32_t stride)
271 {
272 vb[idx].VertexBufferIndex = idx;
273 vb[idx].BufferStartingAddress = addr;
274 vb[idx].BufferPitch = stride;
275
276 #if GEN_GEN >= 6
277 vb[idx].VertexBufferMOCS = addr.mocs;
278 #endif
279
280 #if GEN_GEN >= 7
281 vb[idx].AddressModifyEnable = true;
282 #endif
283
284 #if GEN_GEN >= 8
285 vb[idx].BufferSize = size;
286 #elif GEN_GEN >= 5
287 vb[idx].BufferAccessType = stride > 0 ? VERTEXDATA : INSTANCEDATA;
288 vb[idx].EndAddress = vb[idx].BufferStartingAddress;
289 vb[idx].EndAddress.offset += size - 1;
290 #elif GEN_GEN == 4
291 vb[idx].BufferAccessType = stride > 0 ? VERTEXDATA : INSTANCEDATA;
292 vb[idx].MaxIndex = stride > 0 ? size / stride : 0;
293 #endif
294 }
295
296 static void
297 blorp_emit_vertex_buffers(struct blorp_batch *batch,
298 const struct blorp_params *params)
299 {
300 struct GENX(VERTEX_BUFFER_STATE) vb[3];
301 memset(vb, 0, sizeof(vb));
302
303 struct blorp_address addr;
304 uint32_t size;
305 blorp_emit_vertex_data(batch, params, &addr, &size);
306 blorp_fill_vertex_buffer_state(batch, vb, 0, addr, size, 3 * sizeof(float));
307
308 blorp_emit_input_varying_data(batch, params, &addr, &size);
309 blorp_fill_vertex_buffer_state(batch, vb, 1, addr, size, 0);
310
311 uint32_t num_vbs = 2;
312 if (params->dst_clear_color_as_input) {
313 blorp_fill_vertex_buffer_state(batch, vb, num_vbs++,
314 params->dst.clear_color_addr,
315 batch->blorp->isl_dev->ss.clear_value_size,
316 0);
317 }
318
319 const unsigned num_dwords = 1 + num_vbs * GENX(VERTEX_BUFFER_STATE_length);
320 uint32_t *dw = blorp_emitn(batch, GENX(3DSTATE_VERTEX_BUFFERS), num_dwords);
321 if (!dw)
322 return;
323
324 for (unsigned i = 0; i < num_vbs; i++) {
325 GENX(VERTEX_BUFFER_STATE_pack)(batch, dw, &vb[i]);
326 dw += GENX(VERTEX_BUFFER_STATE_length);
327 }
328 }
329
330 static void
331 blorp_emit_vertex_elements(struct blorp_batch *batch,
332 const struct blorp_params *params)
333 {
334 const unsigned num_varyings =
335 params->wm_prog_data ? params->wm_prog_data->num_varying_inputs : 0;
336 bool need_ndc = batch->blorp->compiler->devinfo->gen <= 5;
337 const unsigned num_elements = 2 + need_ndc + num_varyings;
338
339 struct GENX(VERTEX_ELEMENT_STATE) ve[num_elements];
340 memset(ve, 0, num_elements * sizeof(*ve));
341
342 /* Setup VBO for the rectangle primitive..
343 *
344 * A rectangle primitive (3DPRIM_RECTLIST) consists of only three
345 * vertices. The vertices reside in screen space with DirectX
346 * coordinates (that is, (0, 0) is the upper left corner).
347 *
348 * v2 ------ implied
349 * | |
350 * | |
351 * v1 ----- v0
352 *
353 * Since the VS is disabled, the clipper loads each VUE directly from
354 * the URB. This is controlled by the 3DSTATE_VERTEX_BUFFERS and
355 * 3DSTATE_VERTEX_ELEMENTS packets below. The VUE contents are as follows:
356 * dw0: Reserved, MBZ.
357 * dw1: Render Target Array Index. Below vertex fetcher gets programmed
358 * to assign this with primitive instance identifier which will be
359 * used for layered clears. All other renders have only one instance
360 * and therefore the value will be effectively zero.
361 * dw2: Viewport Index. The HiZ op disables viewport mapping and
362 * scissoring, so set the dword to 0.
363 * dw3: Point Width: The HiZ op does not emit the POINTLIST primitive,
364 * so set the dword to 0.
365 * dw4: Vertex Position X.
366 * dw5: Vertex Position Y.
367 * dw6: Vertex Position Z.
368 * dw7: Vertex Position W.
369 *
370 * dw8: Flat vertex input 0
371 * dw9: Flat vertex input 1
372 * ...
373 * dwn: Flat vertex input n - 8
374 *
375 * For details, see the Sandybridge PRM, Volume 2, Part 1, Section 1.5.1
376 * "Vertex URB Entry (VUE) Formats".
377 *
378 * Only vertex position X and Y are going to be variable, Z is fixed to
379 * zero and W to one. Header words dw0,2,3 are zero. There is no need to
380 * include the fixed values in the vertex buffer. Vertex fetcher can be
381 * instructed to fill vertex elements with constant values of one and zero
382 * instead of reading them from the buffer.
383 * Flat inputs are program constants that are not interpolated. Moreover
384 * their values will be the same between vertices.
385 *
386 * See the vertex element setup below.
387 */
388 unsigned slot = 0;
389
390 ve[slot] = (struct GENX(VERTEX_ELEMENT_STATE)) {
391 .VertexBufferIndex = 1,
392 .Valid = true,
393 .SourceElementFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R32G32B32A32_FLOAT,
394 .SourceElementOffset = 0,
395 .Component0Control = VFCOMP_STORE_SRC,
396
397 /* From Gen8 onwards hardware is no more instructed to overwrite
398 * components using an element specifier. Instead one has separate
399 * 3DSTATE_VF_SGVS (System Generated Value Setup) state packet for it.
400 */
401 #if GEN_GEN >= 8
402 .Component1Control = VFCOMP_STORE_0,
403 #elif GEN_GEN >= 5
404 .Component1Control = VFCOMP_STORE_IID,
405 #else
406 .Component1Control = VFCOMP_STORE_0,
407 #endif
408 .Component2Control = VFCOMP_STORE_0,
409 .Component3Control = VFCOMP_STORE_0,
410 #if GEN_GEN <= 5
411 .DestinationElementOffset = slot * 4,
412 #endif
413 };
414 slot++;
415
416 #if GEN_GEN <= 5
417 /* On Iron Lake and earlier, a native device coordinates version of the
418 * position goes right after the normal VUE header and before position.
419 * Since w == 1 for all of our coordinates, this is just a copy of the
420 * position.
421 */
422 ve[slot] = (struct GENX(VERTEX_ELEMENT_STATE)) {
423 .VertexBufferIndex = 0,
424 .Valid = true,
425 .SourceElementFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R32G32B32_FLOAT,
426 .SourceElementOffset = 0,
427 .Component0Control = VFCOMP_STORE_SRC,
428 .Component1Control = VFCOMP_STORE_SRC,
429 .Component2Control = VFCOMP_STORE_SRC,
430 .Component3Control = VFCOMP_STORE_1_FP,
431 .DestinationElementOffset = slot * 4,
432 };
433 slot++;
434 #endif
435
436 ve[slot] = (struct GENX(VERTEX_ELEMENT_STATE)) {
437 .VertexBufferIndex = 0,
438 .Valid = true,
439 .SourceElementFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R32G32B32_FLOAT,
440 .SourceElementOffset = 0,
441 .Component0Control = VFCOMP_STORE_SRC,
442 .Component1Control = VFCOMP_STORE_SRC,
443 .Component2Control = VFCOMP_STORE_SRC,
444 .Component3Control = VFCOMP_STORE_1_FP,
445 #if GEN_GEN <= 5
446 .DestinationElementOffset = slot * 4,
447 #endif
448 };
449 slot++;
450
451 if (params->dst_clear_color_as_input) {
452 /* If the caller wants the destination indirect clear color, redirect
453 * to vertex buffer 2 where we stored it earlier. The only users of
454 * an indirect clear color source have that as their only vertex
455 * attribute.
456 */
457 assert(num_varyings == 1);
458 ve[slot] = (struct GENX(VERTEX_ELEMENT_STATE)) {
459 .VertexBufferIndex = 2,
460 .Valid = true,
461 .SourceElementOffset = 0,
462 .Component0Control = VFCOMP_STORE_SRC,
463 #if GEN_GEN >= 9
464 .SourceElementFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R32G32B32A32_FLOAT,
465 .Component1Control = VFCOMP_STORE_SRC,
466 .Component2Control = VFCOMP_STORE_SRC,
467 .Component3Control = VFCOMP_STORE_SRC,
468 #else
469 /* Clear colors on gen7-8 are for bits out of one dword */
470 .SourceElementFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R32_FLOAT,
471 .Component1Control = VFCOMP_STORE_0,
472 .Component2Control = VFCOMP_STORE_0,
473 .Component3Control = VFCOMP_STORE_0,
474 #endif
475 };
476 slot++;
477 } else {
478 for (unsigned i = 0; i < num_varyings; ++i) {
479 ve[slot] = (struct GENX(VERTEX_ELEMENT_STATE)) {
480 .VertexBufferIndex = 1,
481 .Valid = true,
482 .SourceElementFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R32G32B32A32_FLOAT,
483 .SourceElementOffset = 16 + i * 4 * sizeof(float),
484 .Component0Control = VFCOMP_STORE_SRC,
485 .Component1Control = VFCOMP_STORE_SRC,
486 .Component2Control = VFCOMP_STORE_SRC,
487 .Component3Control = VFCOMP_STORE_SRC,
488 #if GEN_GEN <= 5
489 .DestinationElementOffset = slot * 4,
490 #endif
491 };
492 slot++;
493 }
494 }
495
496 const unsigned num_dwords =
497 1 + GENX(VERTEX_ELEMENT_STATE_length) * num_elements;
498 uint32_t *dw = blorp_emitn(batch, GENX(3DSTATE_VERTEX_ELEMENTS), num_dwords);
499 if (!dw)
500 return;
501
502 for (unsigned i = 0; i < num_elements; i++) {
503 GENX(VERTEX_ELEMENT_STATE_pack)(batch, dw, &ve[i]);
504 dw += GENX(VERTEX_ELEMENT_STATE_length);
505 }
506
507 #if GEN_GEN >= 8
508 /* Overwrite Render Target Array Index (2nd dword) in the VUE header with
509 * primitive instance identifier. This is used for layered clears.
510 */
511 blorp_emit(batch, GENX(3DSTATE_VF_SGVS), sgvs) {
512 sgvs.InstanceIDEnable = true;
513 sgvs.InstanceIDComponentNumber = COMP_1;
514 sgvs.InstanceIDElementOffset = 0;
515 }
516
517 for (unsigned i = 0; i < num_elements; i++) {
518 blorp_emit(batch, GENX(3DSTATE_VF_INSTANCING), vf) {
519 vf.VertexElementIndex = i;
520 vf.InstancingEnable = false;
521 }
522 }
523
524 blorp_emit(batch, GENX(3DSTATE_VF_TOPOLOGY), topo) {
525 topo.PrimitiveTopologyType = _3DPRIM_RECTLIST;
526 }
527 #endif
528 }
529
530 /* 3DSTATE_VIEWPORT_STATE_POINTERS */
531 static uint32_t
532 blorp_emit_cc_viewport(struct blorp_batch *batch)
533 {
534 uint32_t cc_vp_offset;
535 blorp_emit_dynamic(batch, GENX(CC_VIEWPORT), vp, 32, &cc_vp_offset) {
536 vp.MinimumDepth = 0.0;
537 vp.MaximumDepth = 1.0;
538 }
539
540 #if GEN_GEN >= 7
541 blorp_emit(batch, GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), vsp) {
542 vsp.CCViewportPointer = cc_vp_offset;
543 }
544 #elif GEN_GEN == 6
545 blorp_emit(batch, GENX(3DSTATE_VIEWPORT_STATE_POINTERS), vsp) {
546 vsp.CCViewportStateChange = true;
547 vsp.PointertoCC_VIEWPORT = cc_vp_offset;
548 }
549 #endif
550
551 return cc_vp_offset;
552 }
553
554 static uint32_t
555 blorp_emit_sampler_state(struct blorp_batch *batch)
556 {
557 uint32_t offset;
558 blorp_emit_dynamic(batch, GENX(SAMPLER_STATE), sampler, 32, &offset) {
559 sampler.MipModeFilter = MIPFILTER_NONE;
560 sampler.MagModeFilter = MAPFILTER_LINEAR;
561 sampler.MinModeFilter = MAPFILTER_LINEAR;
562 sampler.MinLOD = 0;
563 sampler.MaxLOD = 0;
564 sampler.TCXAddressControlMode = TCM_CLAMP;
565 sampler.TCYAddressControlMode = TCM_CLAMP;
566 sampler.TCZAddressControlMode = TCM_CLAMP;
567 sampler.MaximumAnisotropy = RATIO21;
568 sampler.RAddressMinFilterRoundingEnable = true;
569 sampler.RAddressMagFilterRoundingEnable = true;
570 sampler.VAddressMinFilterRoundingEnable = true;
571 sampler.VAddressMagFilterRoundingEnable = true;
572 sampler.UAddressMinFilterRoundingEnable = true;
573 sampler.UAddressMagFilterRoundingEnable = true;
574 #if GEN_GEN > 6
575 sampler.NonnormalizedCoordinateEnable = true;
576 #endif
577 }
578
579 #if GEN_GEN >= 7
580 blorp_emit(batch, GENX(3DSTATE_SAMPLER_STATE_POINTERS_PS), ssp) {
581 ssp.PointertoPSSamplerState = offset;
582 }
583 #elif GEN_GEN == 6
584 blorp_emit(batch, GENX(3DSTATE_SAMPLER_STATE_POINTERS), ssp) {
585 ssp.VSSamplerStateChange = true;
586 ssp.GSSamplerStateChange = true;
587 ssp.PSSamplerStateChange = true;
588 ssp.PointertoPSSamplerState = offset;
589 }
590 #endif
591
592 return offset;
593 }
594
595 /* What follows is the code for setting up a "pipeline" on Sandy Bridge and
596 * later hardware. This file will be included by i965 for gen4-5 as well, so
597 * this code is guarded by GEN_GEN >= 6.
598 */
599 #if GEN_GEN >= 6
600
601 static void
602 blorp_emit_vs_config(struct blorp_batch *batch,
603 const struct blorp_params *params)
604 {
605 struct brw_vs_prog_data *vs_prog_data = params->vs_prog_data;
606 assert(!vs_prog_data || GEN_GEN < 11 ||
607 vs_prog_data->base.dispatch_mode == DISPATCH_MODE_SIMD8);
608
609 blorp_emit(batch, GENX(3DSTATE_VS), vs) {
610 if (vs_prog_data) {
611 vs.Enable = true;
612
613 vs.KernelStartPointer = params->vs_prog_kernel;
614
615 vs.DispatchGRFStartRegisterForURBData =
616 vs_prog_data->base.base.dispatch_grf_start_reg;
617 vs.VertexURBEntryReadLength =
618 vs_prog_data->base.urb_read_length;
619 vs.VertexURBEntryReadOffset = 0;
620
621 vs.MaximumNumberofThreads =
622 batch->blorp->isl_dev->info->max_vs_threads - 1;
623
624 #if GEN_GEN >= 8
625 vs.SIMD8DispatchEnable =
626 vs_prog_data->base.dispatch_mode == DISPATCH_MODE_SIMD8;
627 #endif
628 }
629 }
630 }
631
632 static void
633 blorp_emit_sf_config(struct blorp_batch *batch,
634 const struct blorp_params *params)
635 {
636 const struct brw_wm_prog_data *prog_data = params->wm_prog_data;
637
638 /* 3DSTATE_SF
639 *
640 * Disable ViewportTransformEnable (dw2.1)
641 *
642 * From the SandyBridge PRM, Volume 2, Part 1, Section 1.3, "3D
643 * Primitives Overview":
644 * RECTLIST: Viewport Mapping must be DISABLED (as is typical with the
645 * use of screen- space coordinates).
646 *
647 * A solid rectangle must be rendered, so set FrontFaceFillMode (dw2.4:3)
648 * and BackFaceFillMode (dw2.5:6) to SOLID(0).
649 *
650 * From the Sandy Bridge PRM, Volume 2, Part 1, Section
651 * 6.4.1.1 3DSTATE_SF, Field FrontFaceFillMode:
652 * SOLID: Any triangle or rectangle object found to be front-facing
653 * is rendered as a solid object. This setting is required when
654 * (rendering rectangle (RECTLIST) objects.
655 */
656
657 #if GEN_GEN >= 8
658
659 blorp_emit(batch, GENX(3DSTATE_SF), sf);
660
661 blorp_emit(batch, GENX(3DSTATE_RASTER), raster) {
662 raster.CullMode = CULLMODE_NONE;
663 }
664
665 blorp_emit(batch, GENX(3DSTATE_SBE), sbe) {
666 sbe.VertexURBEntryReadOffset = 1;
667 if (prog_data) {
668 sbe.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
669 sbe.VertexURBEntryReadLength = brw_blorp_get_urb_length(prog_data);
670 sbe.ConstantInterpolationEnable = prog_data->flat_inputs;
671 } else {
672 sbe.NumberofSFOutputAttributes = 0;
673 sbe.VertexURBEntryReadLength = 1;
674 }
675 sbe.ForceVertexURBEntryReadLength = true;
676 sbe.ForceVertexURBEntryReadOffset = true;
677
678 #if GEN_GEN >= 9
679 for (unsigned i = 0; i < 32; i++)
680 sbe.AttributeActiveComponentFormat[i] = ACF_XYZW;
681 #endif
682 }
683
684 #elif GEN_GEN >= 7
685
686 blorp_emit(batch, GENX(3DSTATE_SF), sf) {
687 sf.FrontFaceFillMode = FILL_MODE_SOLID;
688 sf.BackFaceFillMode = FILL_MODE_SOLID;
689
690 sf.MultisampleRasterizationMode = params->num_samples > 1 ?
691 MSRASTMODE_ON_PATTERN : MSRASTMODE_OFF_PIXEL;
692
693 #if GEN_GEN == 7
694 sf.DepthBufferSurfaceFormat = params->depth_format;
695 #endif
696 }
697
698 blorp_emit(batch, GENX(3DSTATE_SBE), sbe) {
699 sbe.VertexURBEntryReadOffset = 1;
700 if (prog_data) {
701 sbe.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
702 sbe.VertexURBEntryReadLength = brw_blorp_get_urb_length(prog_data);
703 sbe.ConstantInterpolationEnable = prog_data->flat_inputs;
704 } else {
705 sbe.NumberofSFOutputAttributes = 0;
706 sbe.VertexURBEntryReadLength = 1;
707 }
708 }
709
710 #else /* GEN_GEN <= 6 */
711
712 blorp_emit(batch, GENX(3DSTATE_SF), sf) {
713 sf.FrontFaceFillMode = FILL_MODE_SOLID;
714 sf.BackFaceFillMode = FILL_MODE_SOLID;
715
716 sf.MultisampleRasterizationMode = params->num_samples > 1 ?
717 MSRASTMODE_ON_PATTERN : MSRASTMODE_OFF_PIXEL;
718
719 sf.VertexURBEntryReadOffset = 1;
720 if (prog_data) {
721 sf.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
722 sf.VertexURBEntryReadLength = brw_blorp_get_urb_length(prog_data);
723 sf.ConstantInterpolationEnable = prog_data->flat_inputs;
724 } else {
725 sf.NumberofSFOutputAttributes = 0;
726 sf.VertexURBEntryReadLength = 1;
727 }
728 }
729
730 #endif /* GEN_GEN */
731 }
732
733 static void
734 blorp_emit_ps_config(struct blorp_batch *batch,
735 const struct blorp_params *params)
736 {
737 const struct brw_wm_prog_data *prog_data = params->wm_prog_data;
738
739 /* Even when thread dispatch is disabled, max threads (dw5.25:31) must be
740 * nonzero to prevent the GPU from hanging. While the documentation doesn't
741 * mention this explicitly, it notes that the valid range for the field is
742 * [1,39] = [2,40] threads, which excludes zero.
743 *
744 * To be safe (and to minimize extraneous code) we go ahead and fully
745 * configure the WM state whether or not there is a WM program.
746 */
747
748 #if GEN_GEN >= 8
749
750 blorp_emit(batch, GENX(3DSTATE_WM), wm);
751
752 blorp_emit(batch, GENX(3DSTATE_PS), ps) {
753 if (params->src.enabled) {
754 ps.SamplerCount = 1; /* Up to 4 samplers */
755 ps.BindingTableEntryCount = 2;
756 } else {
757 ps.BindingTableEntryCount = 1;
758 }
759
760 if (prog_data) {
761 ps.DispatchGRFStartRegisterForConstantSetupData0 =
762 prog_data->base.dispatch_grf_start_reg;
763 ps.DispatchGRFStartRegisterForConstantSetupData2 =
764 prog_data->dispatch_grf_start_reg_2;
765
766 ps._8PixelDispatchEnable = prog_data->dispatch_8;
767 ps._16PixelDispatchEnable = prog_data->dispatch_16;
768
769 ps.KernelStartPointer0 = params->wm_prog_kernel;
770 ps.KernelStartPointer2 =
771 params->wm_prog_kernel + prog_data->prog_offset_2;
772 }
773
774 /* 3DSTATE_PS expects the number of threads per PSD, which is always 64
775 * for pre Gen11 and 128 for gen11+; On gen11+ If a programmed value is
776 * k, it implies 2(k+1) threads. It implicitly scales for different GT
777 * levels (which have some # of PSDs).
778 *
779 * In Gen8 the format is U8-2 whereas in Gen9+ it is U9-1.
780 */
781 if (GEN_GEN >= 9)
782 ps.MaximumNumberofThreadsPerPSD = 64 - 1;
783 else
784 ps.MaximumNumberofThreadsPerPSD = 64 - 2;
785
786 switch (params->fast_clear_op) {
787 case ISL_AUX_OP_NONE:
788 break;
789 #if GEN_GEN >= 9
790 case ISL_AUX_OP_PARTIAL_RESOLVE:
791 ps.RenderTargetResolveType = RESOLVE_PARTIAL;
792 break;
793 case ISL_AUX_OP_FULL_RESOLVE:
794 ps.RenderTargetResolveType = RESOLVE_FULL;
795 break;
796 #else
797 case ISL_AUX_OP_FULL_RESOLVE:
798 ps.RenderTargetResolveEnable = true;
799 break;
800 #endif
801 case ISL_AUX_OP_FAST_CLEAR:
802 ps.RenderTargetFastClearEnable = true;
803 break;
804 default:
805 unreachable("Invalid fast clear op");
806 }
807 }
808
809 blorp_emit(batch, GENX(3DSTATE_PS_EXTRA), psx) {
810 if (prog_data) {
811 psx.PixelShaderValid = true;
812 psx.AttributeEnable = prog_data->num_varying_inputs > 0;
813 psx.PixelShaderIsPerSample = prog_data->persample_dispatch;
814 }
815
816 if (params->src.enabled)
817 psx.PixelShaderKillsPixel = true;
818 }
819
820 #elif GEN_GEN >= 7
821
822 blorp_emit(batch, GENX(3DSTATE_WM), wm) {
823 switch (params->hiz_op) {
824 case ISL_AUX_OP_FAST_CLEAR:
825 wm.DepthBufferClear = true;
826 break;
827 case ISL_AUX_OP_FULL_RESOLVE:
828 wm.DepthBufferResolveEnable = true;
829 break;
830 case ISL_AUX_OP_AMBIGUATE:
831 wm.HierarchicalDepthBufferResolveEnable = true;
832 break;
833 case ISL_AUX_OP_NONE:
834 break;
835 default:
836 unreachable("not reached");
837 }
838
839 if (prog_data)
840 wm.ThreadDispatchEnable = true;
841
842 if (params->src.enabled)
843 wm.PixelShaderKillsPixel = true;
844
845 if (params->num_samples > 1) {
846 wm.MultisampleRasterizationMode = MSRASTMODE_ON_PATTERN;
847 wm.MultisampleDispatchMode =
848 (prog_data && prog_data->persample_dispatch) ?
849 MSDISPMODE_PERSAMPLE : MSDISPMODE_PERPIXEL;
850 } else {
851 wm.MultisampleRasterizationMode = MSRASTMODE_OFF_PIXEL;
852 wm.MultisampleDispatchMode = MSDISPMODE_PERSAMPLE;
853 }
854 }
855
856 blorp_emit(batch, GENX(3DSTATE_PS), ps) {
857 ps.MaximumNumberofThreads =
858 batch->blorp->isl_dev->info->max_wm_threads - 1;
859
860 #if GEN_IS_HASWELL
861 ps.SampleMask = 1;
862 #endif
863
864 if (prog_data) {
865 ps.DispatchGRFStartRegisterForConstantSetupData0 =
866 prog_data->base.dispatch_grf_start_reg;
867 ps.DispatchGRFStartRegisterForConstantSetupData2 =
868 prog_data->dispatch_grf_start_reg_2;
869
870 ps.KernelStartPointer0 = params->wm_prog_kernel;
871 ps.KernelStartPointer2 =
872 params->wm_prog_kernel + prog_data->prog_offset_2;
873
874 ps._8PixelDispatchEnable = prog_data->dispatch_8;
875 ps._16PixelDispatchEnable = prog_data->dispatch_16;
876
877 ps.AttributeEnable = prog_data->num_varying_inputs > 0;
878 } else {
879 /* Gen7 hardware gets angry if we don't enable at least one dispatch
880 * mode, so just enable 16-pixel dispatch if we don't have a program.
881 */
882 ps._16PixelDispatchEnable = true;
883 }
884
885 if (params->src.enabled)
886 ps.SamplerCount = 1; /* Up to 4 samplers */
887
888 switch (params->fast_clear_op) {
889 case ISL_AUX_OP_NONE:
890 break;
891 case ISL_AUX_OP_FULL_RESOLVE:
892 ps.RenderTargetResolveEnable = true;
893 break;
894 case ISL_AUX_OP_FAST_CLEAR:
895 ps.RenderTargetFastClearEnable = true;
896 break;
897 default:
898 unreachable("Invalid fast clear op");
899 }
900 }
901
902 #else /* GEN_GEN <= 6 */
903
904 blorp_emit(batch, GENX(3DSTATE_WM), wm) {
905 wm.MaximumNumberofThreads =
906 batch->blorp->isl_dev->info->max_wm_threads - 1;
907
908 switch (params->hiz_op) {
909 case ISL_AUX_OP_FAST_CLEAR:
910 wm.DepthBufferClear = true;
911 break;
912 case ISL_AUX_OP_FULL_RESOLVE:
913 wm.DepthBufferResolveEnable = true;
914 break;
915 case ISL_AUX_OP_AMBIGUATE:
916 wm.HierarchicalDepthBufferResolveEnable = true;
917 break;
918 case ISL_AUX_OP_NONE:
919 break;
920 default:
921 unreachable("not reached");
922 }
923
924 if (prog_data) {
925 wm.ThreadDispatchEnable = true;
926
927 wm.DispatchGRFStartRegisterForConstantSetupData0 =
928 prog_data->base.dispatch_grf_start_reg;
929 wm.DispatchGRFStartRegisterForConstantSetupData2 =
930 prog_data->dispatch_grf_start_reg_2;
931
932 wm.KernelStartPointer0 = params->wm_prog_kernel;
933 wm.KernelStartPointer2 =
934 params->wm_prog_kernel + prog_data->prog_offset_2;
935
936 wm._8PixelDispatchEnable = prog_data->dispatch_8;
937 wm._16PixelDispatchEnable = prog_data->dispatch_16;
938
939 wm.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
940 }
941
942 if (params->src.enabled) {
943 wm.SamplerCount = 1; /* Up to 4 samplers */
944 wm.PixelShaderKillsPixel = true; /* TODO: temporarily smash on */
945 }
946
947 if (params->num_samples > 1) {
948 wm.MultisampleRasterizationMode = MSRASTMODE_ON_PATTERN;
949 wm.MultisampleDispatchMode =
950 (prog_data && prog_data->persample_dispatch) ?
951 MSDISPMODE_PERSAMPLE : MSDISPMODE_PERPIXEL;
952 } else {
953 wm.MultisampleRasterizationMode = MSRASTMODE_OFF_PIXEL;
954 wm.MultisampleDispatchMode = MSDISPMODE_PERSAMPLE;
955 }
956 }
957
958 #endif /* GEN_GEN */
959 }
960
961 static uint32_t
962 blorp_emit_blend_state(struct blorp_batch *batch,
963 const struct blorp_params *params)
964 {
965 struct GENX(BLEND_STATE) blend;
966 memset(&blend, 0, sizeof(blend));
967
968 uint32_t offset;
969 int size = GENX(BLEND_STATE_length) * 4;
970 size += GENX(BLEND_STATE_ENTRY_length) * 4 * params->num_draw_buffers;
971 uint32_t *state = blorp_alloc_dynamic_state(batch, size, 64, &offset);
972 uint32_t *pos = state;
973
974 GENX(BLEND_STATE_pack)(NULL, pos, &blend);
975 pos += GENX(BLEND_STATE_length);
976
977 for (unsigned i = 0; i < params->num_draw_buffers; ++i) {
978 struct GENX(BLEND_STATE_ENTRY) entry = {
979 .PreBlendColorClampEnable = true,
980 .PostBlendColorClampEnable = true,
981 .ColorClampRange = COLORCLAMP_RTFORMAT,
982
983 .WriteDisableRed = params->color_write_disable[0],
984 .WriteDisableGreen = params->color_write_disable[1],
985 .WriteDisableBlue = params->color_write_disable[2],
986 .WriteDisableAlpha = params->color_write_disable[3],
987 };
988 GENX(BLEND_STATE_ENTRY_pack)(NULL, pos, &entry);
989 pos += GENX(BLEND_STATE_ENTRY_length);
990 }
991
992 blorp_flush_range(batch, state, size);
993
994 #if GEN_GEN >= 7
995 blorp_emit(batch, GENX(3DSTATE_BLEND_STATE_POINTERS), sp) {
996 sp.BlendStatePointer = offset;
997 #if GEN_GEN >= 8
998 sp.BlendStatePointerValid = true;
999 #endif
1000 }
1001 #endif
1002
1003 #if GEN_GEN >= 8
1004 blorp_emit(batch, GENX(3DSTATE_PS_BLEND), ps_blend) {
1005 ps_blend.HasWriteableRT = true;
1006 }
1007 #endif
1008
1009 return offset;
1010 }
1011
1012 static uint32_t
1013 blorp_emit_color_calc_state(struct blorp_batch *batch,
1014 MAYBE_UNUSED const struct blorp_params *params)
1015 {
1016 uint32_t offset;
1017 blorp_emit_dynamic(batch, GENX(COLOR_CALC_STATE), cc, 64, &offset) {
1018 #if GEN_GEN <= 8
1019 cc.StencilReferenceValue = params->stencil_ref;
1020 #endif
1021 }
1022
1023 #if GEN_GEN >= 7
1024 blorp_emit(batch, GENX(3DSTATE_CC_STATE_POINTERS), sp) {
1025 sp.ColorCalcStatePointer = offset;
1026 #if GEN_GEN >= 8
1027 sp.ColorCalcStatePointerValid = true;
1028 #endif
1029 }
1030 #endif
1031
1032 return offset;
1033 }
1034
1035 static uint32_t
1036 blorp_emit_depth_stencil_state(struct blorp_batch *batch,
1037 const struct blorp_params *params)
1038 {
1039 #if GEN_GEN >= 8
1040 struct GENX(3DSTATE_WM_DEPTH_STENCIL) ds = {
1041 GENX(3DSTATE_WM_DEPTH_STENCIL_header),
1042 };
1043 #else
1044 struct GENX(DEPTH_STENCIL_STATE) ds = { 0 };
1045 #endif
1046
1047 if (params->depth.enabled) {
1048 ds.DepthBufferWriteEnable = true;
1049
1050 switch (params->hiz_op) {
1051 case ISL_AUX_OP_NONE:
1052 ds.DepthTestEnable = true;
1053 ds.DepthTestFunction = COMPAREFUNCTION_ALWAYS;
1054 break;
1055
1056 /* See the following sections of the Sandy Bridge PRM, Volume 2, Part1:
1057 * - 7.5.3.1 Depth Buffer Clear
1058 * - 7.5.3.2 Depth Buffer Resolve
1059 * - 7.5.3.3 Hierarchical Depth Buffer Resolve
1060 */
1061 case ISL_AUX_OP_FULL_RESOLVE:
1062 ds.DepthTestEnable = true;
1063 ds.DepthTestFunction = COMPAREFUNCTION_NEVER;
1064 break;
1065
1066 case ISL_AUX_OP_FAST_CLEAR:
1067 case ISL_AUX_OP_AMBIGUATE:
1068 ds.DepthTestEnable = false;
1069 break;
1070 case ISL_AUX_OP_PARTIAL_RESOLVE:
1071 unreachable("Invalid HIZ op");
1072 }
1073 }
1074
1075 if (params->stencil.enabled) {
1076 ds.StencilBufferWriteEnable = true;
1077 ds.StencilTestEnable = true;
1078 ds.DoubleSidedStencilEnable = false;
1079
1080 ds.StencilTestFunction = COMPAREFUNCTION_ALWAYS;
1081 ds.StencilPassDepthPassOp = STENCILOP_REPLACE;
1082
1083 ds.StencilWriteMask = params->stencil_mask;
1084 #if GEN_GEN >= 9
1085 ds.StencilReferenceValue = params->stencil_ref;
1086 #endif
1087 }
1088
1089 #if GEN_GEN >= 8
1090 uint32_t offset = 0;
1091 uint32_t *dw = blorp_emit_dwords(batch,
1092 GENX(3DSTATE_WM_DEPTH_STENCIL_length));
1093 if (!dw)
1094 return 0;
1095
1096 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, dw, &ds);
1097 #else
1098 uint32_t offset;
1099 void *state = blorp_alloc_dynamic_state(batch,
1100 GENX(DEPTH_STENCIL_STATE_length) * 4,
1101 64, &offset);
1102 GENX(DEPTH_STENCIL_STATE_pack)(NULL, state, &ds);
1103 blorp_flush_range(batch, state, GENX(DEPTH_STENCIL_STATE_length) * 4);
1104 #endif
1105
1106 #if GEN_GEN == 7
1107 blorp_emit(batch, GENX(3DSTATE_DEPTH_STENCIL_STATE_POINTERS), sp) {
1108 sp.PointertoDEPTH_STENCIL_STATE = offset;
1109 }
1110 #endif
1111
1112 return offset;
1113 }
1114
1115 static void
1116 blorp_emit_3dstate_multisample(struct blorp_batch *batch,
1117 const struct blorp_params *params)
1118 {
1119 blorp_emit(batch, GENX(3DSTATE_MULTISAMPLE), ms) {
1120 ms.NumberofMultisamples = __builtin_ffs(params->num_samples) - 1;
1121
1122 #if GEN_GEN >= 8
1123 /* The PRM says that this bit is valid only for DX9:
1124 *
1125 * SW can choose to set this bit only for DX9 API. DX10/OGL API's
1126 * should not have any effect by setting or not setting this bit.
1127 */
1128 ms.PixelPositionOffsetEnable = false;
1129 #elif GEN_GEN >= 7
1130
1131 switch (params->num_samples) {
1132 case 1:
1133 GEN_SAMPLE_POS_1X(ms.Sample);
1134 break;
1135 case 2:
1136 GEN_SAMPLE_POS_2X(ms.Sample);
1137 break;
1138 case 4:
1139 GEN_SAMPLE_POS_4X(ms.Sample);
1140 break;
1141 case 8:
1142 GEN_SAMPLE_POS_8X(ms.Sample);
1143 break;
1144 default:
1145 break;
1146 }
1147 #else
1148 GEN_SAMPLE_POS_4X(ms.Sample);
1149 #endif
1150 ms.PixelLocation = CENTER;
1151 }
1152 }
1153
1154 static void
1155 blorp_emit_pipeline(struct blorp_batch *batch,
1156 const struct blorp_params *params)
1157 {
1158 uint32_t blend_state_offset = 0;
1159 uint32_t color_calc_state_offset;
1160 uint32_t depth_stencil_state_offset;
1161
1162 emit_urb_config(batch, params);
1163
1164 if (params->wm_prog_data) {
1165 blend_state_offset = blorp_emit_blend_state(batch, params);
1166 }
1167 color_calc_state_offset = blorp_emit_color_calc_state(batch, params);
1168 depth_stencil_state_offset = blorp_emit_depth_stencil_state(batch, params);
1169
1170 #if GEN_GEN == 6
1171 /* 3DSTATE_CC_STATE_POINTERS
1172 *
1173 * The pointer offsets are relative to
1174 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
1175 *
1176 * The HiZ op doesn't use BLEND_STATE or COLOR_CALC_STATE.
1177 *
1178 * The dynamic state emit helpers emit their own STATE_POINTERS packets on
1179 * gen7+. However, on gen6 and earlier, they're all lumpped together in
1180 * one CC_STATE_POINTERS packet so we have to emit that here.
1181 */
1182 blorp_emit(batch, GENX(3DSTATE_CC_STATE_POINTERS), cc) {
1183 cc.BLEND_STATEChange = true;
1184 cc.ColorCalcStatePointerValid = true;
1185 cc.DEPTH_STENCIL_STATEChange = true;
1186 cc.PointertoBLEND_STATE = blend_state_offset;
1187 cc.ColorCalcStatePointer = color_calc_state_offset;
1188 cc.PointertoDEPTH_STENCIL_STATE = depth_stencil_state_offset;
1189 }
1190 #else
1191 (void)blend_state_offset;
1192 (void)color_calc_state_offset;
1193 (void)depth_stencil_state_offset;
1194 #endif
1195
1196 blorp_emit(batch, GENX(3DSTATE_CONSTANT_VS), vs);
1197 #if GEN_GEN >= 7
1198 blorp_emit(batch, GENX(3DSTATE_CONSTANT_HS), hs);
1199 blorp_emit(batch, GENX(3DSTATE_CONSTANT_DS), DS);
1200 #endif
1201 blorp_emit(batch, GENX(3DSTATE_CONSTANT_GS), gs);
1202 blorp_emit(batch, GENX(3DSTATE_CONSTANT_PS), ps);
1203
1204 if (params->src.enabled)
1205 blorp_emit_sampler_state(batch);
1206
1207 blorp_emit_3dstate_multisample(batch, params);
1208
1209 blorp_emit(batch, GENX(3DSTATE_SAMPLE_MASK), mask) {
1210 mask.SampleMask = (1 << params->num_samples) - 1;
1211 }
1212
1213 /* From the BSpec, 3D Pipeline > Geometry > Vertex Shader > State,
1214 * 3DSTATE_VS, Dword 5.0 "VS Function Enable":
1215 *
1216 * [DevSNB] A pipeline flush must be programmed prior to a
1217 * 3DSTATE_VS command that causes the VS Function Enable to
1218 * toggle. Pipeline flush can be executed by sending a PIPE_CONTROL
1219 * command with CS stall bit set and a post sync operation.
1220 *
1221 * We've already done one at the start of the BLORP operation.
1222 */
1223 blorp_emit_vs_config(batch, params);
1224 #if GEN_GEN >= 7
1225 blorp_emit(batch, GENX(3DSTATE_HS), hs);
1226 blorp_emit(batch, GENX(3DSTATE_TE), te);
1227 blorp_emit(batch, GENX(3DSTATE_DS), DS);
1228 blorp_emit(batch, GENX(3DSTATE_STREAMOUT), so);
1229 #endif
1230 blorp_emit(batch, GENX(3DSTATE_GS), gs);
1231
1232 blorp_emit(batch, GENX(3DSTATE_CLIP), clip) {
1233 clip.PerspectiveDivideDisable = true;
1234 }
1235
1236 blorp_emit_sf_config(batch, params);
1237 blorp_emit_ps_config(batch, params);
1238
1239 blorp_emit_cc_viewport(batch);
1240 }
1241
1242 /******** This is the end of the pipeline setup code ********/
1243
1244 #endif /* GEN_GEN >= 6 */
1245
1246 #if GEN_GEN >= 7 && GEN_GEN <= 10
1247 static void
1248 blorp_emit_memcpy(struct blorp_batch *batch,
1249 struct blorp_address dst,
1250 struct blorp_address src,
1251 uint32_t size)
1252 {
1253 assert(size % 4 == 0);
1254
1255 for (unsigned dw = 0; dw < size; dw += 4) {
1256 #if GEN_GEN >= 8
1257 blorp_emit(batch, GENX(MI_COPY_MEM_MEM), cp) {
1258 cp.DestinationMemoryAddress = dst;
1259 cp.SourceMemoryAddress = src;
1260 }
1261 #else
1262 /* IVB does not have a general purpose register for command streamer
1263 * commands. Therefore, we use an alternate temporary register.
1264 */
1265 #define BLORP_TEMP_REG 0x2440 /* GEN7_3DPRIM_BASE_VERTEX */
1266 blorp_emit(batch, GENX(MI_LOAD_REGISTER_MEM), load) {
1267 load.RegisterAddress = BLORP_TEMP_REG;
1268 load.MemoryAddress = src;
1269 }
1270 blorp_emit(batch, GENX(MI_STORE_REGISTER_MEM), store) {
1271 store.RegisterAddress = BLORP_TEMP_REG;
1272 store.MemoryAddress = dst;
1273 }
1274 #undef BLORP_TEMP_REG
1275 #endif
1276 dst.offset += 4;
1277 src.offset += 4;
1278 }
1279 }
1280 #endif
1281
1282 static void
1283 blorp_emit_surface_state(struct blorp_batch *batch,
1284 const struct brw_blorp_surface_info *surface,
1285 void *state, uint32_t state_offset,
1286 const bool color_write_disables[4],
1287 bool is_render_target)
1288 {
1289 const struct isl_device *isl_dev = batch->blorp->isl_dev;
1290 struct isl_surf surf = surface->surf;
1291
1292 if (surf.dim == ISL_SURF_DIM_1D &&
1293 surf.dim_layout == ISL_DIM_LAYOUT_GEN4_2D) {
1294 assert(surf.logical_level0_px.height == 1);
1295 surf.dim = ISL_SURF_DIM_2D;
1296 }
1297
1298 /* Blorp doesn't support HiZ in any of the blit or slow-clear paths */
1299 enum isl_aux_usage aux_usage = surface->aux_usage;
1300 if (aux_usage == ISL_AUX_USAGE_HIZ)
1301 aux_usage = ISL_AUX_USAGE_NONE;
1302
1303 isl_channel_mask_t write_disable_mask = 0;
1304 if (is_render_target && GEN_GEN <= 5) {
1305 if (color_write_disables[0])
1306 write_disable_mask |= ISL_CHANNEL_RED_BIT;
1307 if (color_write_disables[1])
1308 write_disable_mask |= ISL_CHANNEL_GREEN_BIT;
1309 if (color_write_disables[2])
1310 write_disable_mask |= ISL_CHANNEL_BLUE_BIT;
1311 if (color_write_disables[3])
1312 write_disable_mask |= ISL_CHANNEL_ALPHA_BIT;
1313 }
1314
1315 isl_surf_fill_state(batch->blorp->isl_dev, state,
1316 .surf = &surf, .view = &surface->view,
1317 .aux_surf = &surface->aux_surf, .aux_usage = aux_usage,
1318 .mocs = surface->addr.mocs,
1319 .clear_color = surface->clear_color,
1320 .write_disables = write_disable_mask);
1321
1322 blorp_surface_reloc(batch, state_offset + isl_dev->ss.addr_offset,
1323 surface->addr, 0);
1324
1325 if (aux_usage != ISL_AUX_USAGE_NONE) {
1326 /* On gen7 and prior, the bottom 12 bits of the MCS base address are
1327 * used to store other information. This should be ok, however, because
1328 * surface buffer addresses are always 4K page alinged.
1329 */
1330 assert((surface->aux_addr.offset & 0xfff) == 0);
1331 uint32_t *aux_addr = state + isl_dev->ss.aux_addr_offset;
1332 blorp_surface_reloc(batch, state_offset + isl_dev->ss.aux_addr_offset,
1333 surface->aux_addr, *aux_addr);
1334 }
1335
1336 blorp_flush_range(batch, state, GENX(RENDER_SURFACE_STATE_length) * 4);
1337
1338 if (surface->clear_color_addr.buffer) {
1339 #if GEN_GEN > 10
1340 unreachable("Implement indirect clear support on gen11+");
1341 #elif GEN_GEN >= 7 && GEN_GEN <= 10
1342 struct blorp_address dst_addr = blorp_get_surface_base_address(batch);
1343 dst_addr.offset += state_offset + isl_dev->ss.clear_value_offset;
1344 blorp_emit_memcpy(batch, dst_addr, surface->clear_color_addr,
1345 isl_dev->ss.clear_value_size);
1346 #else
1347 unreachable("Fast clears are only supported on gen7+");
1348 #endif
1349 }
1350 }
1351
1352 static void
1353 blorp_emit_null_surface_state(struct blorp_batch *batch,
1354 const struct brw_blorp_surface_info *surface,
1355 uint32_t *state)
1356 {
1357 struct GENX(RENDER_SURFACE_STATE) ss = {
1358 .SurfaceType = SURFTYPE_NULL,
1359 .SurfaceFormat = (enum GENX(SURFACE_FORMAT)) ISL_FORMAT_R8G8B8A8_UNORM,
1360 .Width = surface->surf.logical_level0_px.width - 1,
1361 .Height = surface->surf.logical_level0_px.height - 1,
1362 .MIPCountLOD = surface->view.base_level,
1363 .MinimumArrayElement = surface->view.base_array_layer,
1364 .Depth = surface->view.array_len - 1,
1365 .RenderTargetViewExtent = surface->view.array_len - 1,
1366 #if GEN_GEN >= 6
1367 .NumberofMultisamples = ffs(surface->surf.samples) - 1,
1368 #endif
1369
1370 #if GEN_GEN >= 7
1371 .SurfaceArray = surface->surf.dim != ISL_SURF_DIM_3D,
1372 #endif
1373
1374 #if GEN_GEN >= 8
1375 .TileMode = YMAJOR,
1376 #else
1377 .TiledSurface = true,
1378 #endif
1379 };
1380
1381 GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &ss);
1382
1383 blorp_flush_range(batch, state, GENX(RENDER_SURFACE_STATE_length) * 4);
1384 }
1385
1386 static void
1387 blorp_emit_surface_states(struct blorp_batch *batch,
1388 const struct blorp_params *params)
1389 {
1390 const struct isl_device *isl_dev = batch->blorp->isl_dev;
1391 uint32_t bind_offset, surface_offsets[2];
1392 void *surface_maps[2];
1393
1394 MAYBE_UNUSED bool has_indirect_clear_color = false;
1395 if (params->use_pre_baked_binding_table) {
1396 bind_offset = params->pre_baked_binding_table_offset;
1397 } else {
1398 unsigned num_surfaces = 1 + params->src.enabled;
1399 blorp_alloc_binding_table(batch, num_surfaces,
1400 isl_dev->ss.size, isl_dev->ss.align,
1401 &bind_offset, surface_offsets, surface_maps);
1402
1403 if (params->dst.enabled) {
1404 blorp_emit_surface_state(batch, &params->dst,
1405 surface_maps[BLORP_RENDERBUFFER_BT_INDEX],
1406 surface_offsets[BLORP_RENDERBUFFER_BT_INDEX],
1407 params->color_write_disable, true);
1408 if (params->dst.clear_color_addr.buffer != NULL)
1409 has_indirect_clear_color = true;
1410 } else {
1411 assert(params->depth.enabled || params->stencil.enabled);
1412 const struct brw_blorp_surface_info *surface =
1413 params->depth.enabled ? &params->depth : &params->stencil;
1414 blorp_emit_null_surface_state(batch, surface,
1415 surface_maps[BLORP_RENDERBUFFER_BT_INDEX]);
1416 }
1417
1418 if (params->src.enabled) {
1419 blorp_emit_surface_state(batch, &params->src,
1420 surface_maps[BLORP_TEXTURE_BT_INDEX],
1421 surface_offsets[BLORP_TEXTURE_BT_INDEX],
1422 NULL, false);
1423 if (params->src.clear_color_addr.buffer != NULL)
1424 has_indirect_clear_color = true;
1425 }
1426 }
1427
1428 #if GEN_GEN >= 7
1429 if (has_indirect_clear_color) {
1430 /* Updating a surface state object may require that the state cache be
1431 * invalidated. From the SKL PRM, Shared Functions -> State -> State
1432 * Caching:
1433 *
1434 * Whenever the RENDER_SURFACE_STATE object in memory pointed to by
1435 * the Binding Table Pointer (BTP) and Binding Table Index (BTI) is
1436 * modified [...], the L1 state cache must be invalidated to ensure
1437 * the new surface or sampler state is fetched from system memory.
1438 */
1439 blorp_emit(batch, GENX(PIPE_CONTROL), pipe) {
1440 pipe.StateCacheInvalidationEnable = true;
1441 }
1442 }
1443 #endif
1444
1445 #if GEN_GEN >= 7
1446 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS_VS), bt);
1447 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS_HS), bt);
1448 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS_DS), bt);
1449 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS_GS), bt);
1450
1451 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS_PS), bt) {
1452 bt.PointertoPSBindingTable = bind_offset;
1453 }
1454 #elif GEN_GEN >= 6
1455 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS), bt) {
1456 bt.PSBindingTableChange = true;
1457 bt.PointertoPSBindingTable = bind_offset;
1458 }
1459 #else
1460 blorp_emit(batch, GENX(3DSTATE_BINDING_TABLE_POINTERS), bt) {
1461 bt.PointertoPSBindingTable = bind_offset;
1462 }
1463 #endif
1464 }
1465
1466 static void
1467 blorp_emit_depth_stencil_config(struct blorp_batch *batch,
1468 const struct blorp_params *params)
1469 {
1470 const struct isl_device *isl_dev = batch->blorp->isl_dev;
1471
1472 uint32_t *dw = blorp_emit_dwords(batch, isl_dev->ds.size / 4);
1473 if (dw == NULL)
1474 return;
1475
1476 struct isl_depth_stencil_hiz_emit_info info = { };
1477
1478 if (params->depth.enabled) {
1479 info.view = &params->depth.view;
1480 info.mocs = params->depth.addr.mocs;
1481 } else if (params->stencil.enabled) {
1482 info.view = &params->stencil.view;
1483 info.mocs = params->stencil.addr.mocs;
1484 }
1485
1486 if (params->depth.enabled) {
1487 info.depth_surf = &params->depth.surf;
1488
1489 info.depth_address =
1490 blorp_emit_reloc(batch, dw + isl_dev->ds.depth_offset / 4,
1491 params->depth.addr, 0);
1492
1493 info.hiz_usage = params->depth.aux_usage;
1494 if (info.hiz_usage == ISL_AUX_USAGE_HIZ) {
1495 info.hiz_surf = &params->depth.aux_surf;
1496
1497 struct blorp_address hiz_address = params->depth.aux_addr;
1498 #if GEN_GEN == 6
1499 /* Sandy bridge hardware does not technically support mipmapped HiZ.
1500 * However, we have a special layout that allows us to make it work
1501 * anyway by manually offsetting to the specified miplevel.
1502 */
1503 assert(info.hiz_surf->dim_layout == ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ);
1504 uint32_t offset_B;
1505 isl_surf_get_image_offset_B_tile_sa(info.hiz_surf,
1506 info.view->base_level, 0, 0,
1507 &offset_B, NULL, NULL);
1508 hiz_address.offset += offset_B;
1509 #endif
1510
1511 info.hiz_address =
1512 blorp_emit_reloc(batch, dw + isl_dev->ds.hiz_offset / 4,
1513 hiz_address, 0);
1514
1515 info.depth_clear_value = params->depth.clear_color.f32[0];
1516 }
1517 }
1518
1519 if (params->stencil.enabled) {
1520 info.stencil_surf = &params->stencil.surf;
1521
1522 struct blorp_address stencil_address = params->stencil.addr;
1523 #if GEN_GEN == 6
1524 /* Sandy bridge hardware does not technically support mipmapped stencil.
1525 * However, we have a special layout that allows us to make it work
1526 * anyway by manually offsetting to the specified miplevel.
1527 */
1528 assert(info.stencil_surf->dim_layout == ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ);
1529 uint32_t offset_B;
1530 isl_surf_get_image_offset_B_tile_sa(info.stencil_surf,
1531 info.view->base_level, 0, 0,
1532 &offset_B, NULL, NULL);
1533 stencil_address.offset += offset_B;
1534 #endif
1535
1536 info.stencil_address =
1537 blorp_emit_reloc(batch, dw + isl_dev->ds.stencil_offset / 4,
1538 stencil_address, 0);
1539 }
1540
1541 isl_emit_depth_stencil_hiz_s(isl_dev, dw, &info);
1542 }
1543
1544 #if GEN_GEN >= 8
1545 /* Emits the Optimized HiZ sequence specified in the BDW+ PRMs. The
1546 * depth/stencil buffer extents are ignored to handle APIs which perform
1547 * clearing operations without such information.
1548 * */
1549 static void
1550 blorp_emit_gen8_hiz_op(struct blorp_batch *batch,
1551 const struct blorp_params *params)
1552 {
1553 /* We should be performing an operation on a depth or stencil buffer.
1554 */
1555 assert(params->depth.enabled || params->stencil.enabled);
1556
1557 /* The stencil buffer should only be enabled if a fast clear operation is
1558 * requested.
1559 */
1560 if (params->stencil.enabled)
1561 assert(params->hiz_op == ISL_AUX_OP_FAST_CLEAR);
1562
1563 /* From the BDW PRM Volume 2, 3DSTATE_WM_HZ_OP:
1564 *
1565 * 3DSTATE_MULTISAMPLE packet must be used prior to this packet to change
1566 * the Number of Multisamples. This packet must not be used to change
1567 * Number of Multisamples in a rendering sequence.
1568 *
1569 * Since HIZ may be the first thing in a batch buffer, play safe and always
1570 * emit 3DSTATE_MULTISAMPLE.
1571 */
1572 blorp_emit_3dstate_multisample(batch, params);
1573
1574 /* If we can't alter the depth stencil config and multiple layers are
1575 * involved, the HiZ op will fail. This is because the op requires that a
1576 * new config is emitted for each additional layer.
1577 */
1578 if (batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL) {
1579 assert(params->num_layers <= 1);
1580 } else {
1581 blorp_emit_depth_stencil_config(batch, params);
1582 }
1583
1584 blorp_emit(batch, GENX(3DSTATE_WM_HZ_OP), hzp) {
1585 switch (params->hiz_op) {
1586 case ISL_AUX_OP_FAST_CLEAR:
1587 hzp.StencilBufferClearEnable = params->stencil.enabled;
1588 hzp.DepthBufferClearEnable = params->depth.enabled;
1589 hzp.StencilClearValue = params->stencil_ref;
1590 hzp.FullSurfaceDepthandStencilClear = params->full_surface_hiz_op;
1591 break;
1592 case ISL_AUX_OP_FULL_RESOLVE:
1593 assert(params->full_surface_hiz_op);
1594 hzp.DepthBufferResolveEnable = true;
1595 break;
1596 case ISL_AUX_OP_AMBIGUATE:
1597 assert(params->full_surface_hiz_op);
1598 hzp.HierarchicalDepthBufferResolveEnable = true;
1599 break;
1600 case ISL_AUX_OP_PARTIAL_RESOLVE:
1601 case ISL_AUX_OP_NONE:
1602 unreachable("Invalid HIZ op");
1603 }
1604
1605 hzp.NumberofMultisamples = ffs(params->num_samples) - 1;
1606 hzp.SampleMask = 0xFFFF;
1607
1608 /* Due to a hardware issue, this bit MBZ */
1609 assert(hzp.ScissorRectangleEnable == false);
1610
1611 /* Contrary to the HW docs both fields are inclusive */
1612 hzp.ClearRectangleXMin = params->x0;
1613 hzp.ClearRectangleYMin = params->y0;
1614
1615 /* Contrary to the HW docs both fields are exclusive */
1616 hzp.ClearRectangleXMax = params->x1;
1617 hzp.ClearRectangleYMax = params->y1;
1618 }
1619
1620 /* PIPE_CONTROL w/ all bits clear except for “Post-Sync Operation” must set
1621 * to “Write Immediate Data” enabled.
1622 */
1623 blorp_emit(batch, GENX(PIPE_CONTROL), pc) {
1624 pc.PostSyncOperation = WriteImmediateData;
1625 pc.Address = blorp_get_workaround_page(batch);
1626 }
1627
1628 blorp_emit(batch, GENX(3DSTATE_WM_HZ_OP), hzp);
1629 }
1630 #endif
1631
1632 /**
1633 * \brief Execute a blit or render pass operation.
1634 *
1635 * To execute the operation, this function manually constructs and emits a
1636 * batch to draw a rectangle primitive. The batchbuffer is flushed before
1637 * constructing and after emitting the batch.
1638 *
1639 * This function alters no GL state.
1640 */
1641 static void
1642 blorp_exec(struct blorp_batch *batch, const struct blorp_params *params)
1643 {
1644 #if GEN_GEN >= 8
1645 if (params->hiz_op != ISL_AUX_OP_NONE) {
1646 blorp_emit_gen8_hiz_op(batch, params);
1647 return;
1648 }
1649 #endif
1650
1651 blorp_emit_vertex_buffers(batch, params);
1652 blorp_emit_vertex_elements(batch, params);
1653
1654 blorp_emit_pipeline(batch, params);
1655
1656 blorp_emit_surface_states(batch, params);
1657
1658 if (!(batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL))
1659 blorp_emit_depth_stencil_config(batch, params);
1660
1661 blorp_emit(batch, GENX(3DPRIMITIVE), prim) {
1662 prim.VertexAccessType = SEQUENTIAL;
1663 prim.PrimitiveTopologyType = _3DPRIM_RECTLIST;
1664 #if GEN_GEN >= 7
1665 prim.PredicateEnable = batch->flags & BLORP_BATCH_PREDICATE_ENABLE;
1666 #endif
1667 prim.VertexCountPerInstance = 3;
1668 prim.InstanceCount = params->num_layers;
1669 }
1670 }
1671
1672 #endif /* BLORP_GENX_EXEC_H */