projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
vk: Add func anv_clear_mask()
[mesa.git] / src / vulkan / device.c
1 /*
2 * Copyright © 2015 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 #include <assert.h>
25 #include <stdbool.h>
26 #include <string.h>
27 #include <unistd.h>
28 #include <fcntl.h>
29
30 #include "private.h"
31
32 static int
33 anv_env_get_int(const char *name)
34 {
35 const char *val = getenv(name);
36
37 if (!val)
38 return 0;
39
40 return strtol(val, NULL, 0);
41 }
42
43 static VkResult
44 fill_physical_device(struct anv_physical_device *device,
45 struct anv_instance *instance,
46 const char *path)
47 {
48 int fd;
49
50 fd = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC);
51 if (fd < 0)
52 return vk_error(VK_ERROR_UNAVAILABLE);
53
54 device->instance = instance;
55 device->path = path;
56
57 device->chipset_id = anv_env_get_int("INTEL_DEVID_OVERRIDE");
58 device->no_hw = false;
59 if (device->chipset_id) {
60 /* INTEL_DEVID_OVERRIDE implies INTEL_NO_HW. */
61 device->no_hw = true;
62 } else {
63 device->chipset_id = anv_gem_get_param(fd, I915_PARAM_CHIPSET_ID);
64 }
65 if (!device->chipset_id)
66 goto fail;
67
68 device->name = brw_get_device_name(device->chipset_id);
69 device->info = brw_get_device_info(device->chipset_id, -1);
70 if (!device->info)
71 goto fail;
72
73 if (!anv_gem_get_param(fd, I915_PARAM_HAS_WAIT_TIMEOUT))
74 goto fail;
75
76 if (!anv_gem_get_param(fd, I915_PARAM_HAS_EXECBUF2))
77 goto fail;
78
79 if (!anv_gem_get_param(fd, I915_PARAM_HAS_LLC))
80 goto fail;
81
82 if (!anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_CONSTANTS))
83 goto fail;
84
85 close(fd);
86
87 return VK_SUCCESS;
88
89 fail:
90 close(fd);
91
92 return vk_error(VK_ERROR_UNAVAILABLE);
93 }
94
95 static void *default_alloc(
96 void* pUserData,
97 size_t size,
98 size_t alignment,
99 VkSystemAllocType allocType)
100 {
101 return malloc(size);
102 }
103
104 static void default_free(
105 void* pUserData,
106 void* pMem)
107 {
108 free(pMem);
109 }
110
111 static const VkAllocCallbacks default_alloc_callbacks = {
112 .pUserData = NULL,
113 .pfnAlloc = default_alloc,
114 .pfnFree = default_free
115 };
116
117 VkResult anv_CreateInstance(
118 const VkInstanceCreateInfo* pCreateInfo,
119 VkInstance* pInstance)
120 {
121 struct anv_instance *instance;
122 const VkAllocCallbacks *alloc_callbacks = &default_alloc_callbacks;
123 void *user_data = NULL;
124 VkResult result;
125
126 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);
127
128 if (pCreateInfo->pAllocCb) {
129 alloc_callbacks = pCreateInfo->pAllocCb;
130 user_data = pCreateInfo->pAllocCb->pUserData;
131 }
132 instance = alloc_callbacks->pfnAlloc(user_data, sizeof(*instance), 8,
133 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
134 if (!instance)
135 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
136
137 instance->pAllocUserData = alloc_callbacks->pUserData;
138 instance->pfnAlloc = alloc_callbacks->pfnAlloc;
139 instance->pfnFree = alloc_callbacks->pfnFree;
140 instance->apiVersion = pCreateInfo->pAppInfo->apiVersion;
141
142 instance->physicalDeviceCount = 0;
143 result = fill_physical_device(&instance->physicalDevice,
144 instance, "/dev/dri/renderD128");
145
146 if (result != VK_SUCCESS)
147 return result;
148
149 instance->physicalDeviceCount++;
150 *pInstance = (VkInstance) instance;
151
152 return VK_SUCCESS;
153 }
154
155 VkResult anv_DestroyInstance(
156 VkInstance _instance)
157 {
158 struct anv_instance *instance = (struct anv_instance *) _instance;
159
160 instance->pfnFree(instance->pAllocUserData, instance);
161
162 return VK_SUCCESS;
163 }
164
165 VkResult anv_EnumeratePhysicalDevices(
166 VkInstance _instance,
167 uint32_t* pPhysicalDeviceCount,
168 VkPhysicalDevice* pPhysicalDevices)
169 {
170 struct anv_instance *instance = (struct anv_instance *) _instance;
171
172 if (*pPhysicalDeviceCount >= 1)
173 pPhysicalDevices[0] = (VkPhysicalDevice) &instance->physicalDevice;
174 *pPhysicalDeviceCount = instance->physicalDeviceCount;
175
176 return VK_SUCCESS;
177 }
178
179 VkResult anv_GetPhysicalDeviceInfo(
180 VkPhysicalDevice physicalDevice,
181 VkPhysicalDeviceInfoType infoType,
182 size_t* pDataSize,
183 void* pData)
184 {
185 struct anv_physical_device *device = (struct anv_physical_device *) physicalDevice;
186 VkPhysicalDeviceProperties *properties;
187 VkPhysicalDevicePerformance *performance;
188 VkPhysicalDeviceQueueProperties *queue_properties;
189 VkPhysicalDeviceMemoryProperties *memory_properties;
190 VkDisplayPropertiesWSI *display_properties;
191 uint64_t ns_per_tick = 80;
192
193 switch ((uint32_t) infoType) {
194 case VK_PHYSICAL_DEVICE_INFO_TYPE_PROPERTIES:
195 properties = pData;
196
197 *pDataSize = sizeof(*properties);
198 if (pData == NULL)
199 return VK_SUCCESS;
200
201 properties->apiVersion = 1;
202 properties->driverVersion = 1;
203 properties->vendorId = 0x8086;
204 properties->deviceId = device->chipset_id;
205 properties->deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU;
206 strcpy(properties->deviceName, device->name);
207 properties->maxInlineMemoryUpdateSize = 0;
208 properties->maxBoundDescriptorSets = MAX_SETS;
209 properties->maxThreadGroupSize = 512;
210 properties->timestampFrequency = 1000 * 1000 * 1000 / ns_per_tick;
211 properties->multiColorAttachmentClears = true;
212 properties->maxDescriptorSets = 8;
213 properties->maxViewports = 16;
214 properties->maxColorAttachments = 8;
215 return VK_SUCCESS;
216
217 case VK_PHYSICAL_DEVICE_INFO_TYPE_PERFORMANCE:
218 performance = pData;
219
220 *pDataSize = sizeof(*performance);
221 if (pData == NULL)
222 return VK_SUCCESS;
223
224 performance->maxDeviceClock = 1.0;
225 performance->aluPerClock = 1.0;
226 performance->texPerClock = 1.0;
227 performance->primsPerClock = 1.0;
228 performance->pixelsPerClock = 1.0;
229 return VK_SUCCESS;
230
231 case VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PROPERTIES:
232 queue_properties = pData;
233
234 *pDataSize = sizeof(*queue_properties);
235 if (pData == NULL)
236 return VK_SUCCESS;
237
238 queue_properties->queueFlags = 0;
239 queue_properties->queueCount = 1;
240 queue_properties->supportsTimestamps = true;
241 return VK_SUCCESS;
242
243 case VK_PHYSICAL_DEVICE_INFO_TYPE_MEMORY_PROPERTIES:
244 memory_properties = pData;
245
246 *pDataSize = sizeof(*memory_properties);
247 if (pData == NULL)
248 return VK_SUCCESS;
249
250 memory_properties->supportsMigration = false;
251 memory_properties->supportsPinning = false;
252 return VK_SUCCESS;
253
254 case VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI:
255 anv_finishme("VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI");
256
257 *pDataSize = sizeof(*display_properties);
258 if (pData == NULL)
259 return VK_SUCCESS;
260
261 display_properties = pData;
262 display_properties->display = 0;
263 display_properties->physicalResolution = (VkExtent2D) { 0, 0 };
264 return VK_SUCCESS;
265
266 case VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI:
267 anv_finishme("VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI");
268 return VK_SUCCESS;
269
270
271 default:
272 return VK_UNSUPPORTED;
273 }
274
275 }
276
277 void * vkGetProcAddr(
278 VkPhysicalDevice physicalDevice,
279 const char* pName)
280 {
281 return anv_lookup_entrypoint(pName);
282 }
283
284 static void
285 parse_debug_flags(struct anv_device *device)
286 {
287 const char *debug, *p, *end;
288
289 debug = getenv("INTEL_DEBUG");
290 device->dump_aub = false;
291 if (debug) {
292 for (p = debug; *p; p = end + 1) {
293 end = strchrnul(p, ',');
294 if (end - p == 3 && memcmp(p, "aub", 3) == 0)
295 device->dump_aub = true;
296 if (end - p == 5 && memcmp(p, "no_hw", 5) == 0)
297 device->no_hw = true;
298 if (*end == '\0')
299 break;
300 }
301 }
302 }
303
304 static VkResult
305 anv_queue_init(struct anv_device *device, struct anv_queue *queue)
306 {
307 queue->device = device;
308 queue->pool = &device->surface_state_pool;
309
310 queue->completed_serial = anv_state_pool_alloc(queue->pool, 4, 4);
311 if (queue->completed_serial.map == NULL)
312 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
313
314 *(uint32_t *)queue->completed_serial.map = 0;
315 queue->next_serial = 1;
316
317 return VK_SUCCESS;
318 }
319
320 static void
321 anv_queue_finish(struct anv_queue *queue)
322 {
323 #ifdef HAVE_VALGRIND
324 /* This gets torn down with the device so we only need to do this if
325 * valgrind is present.
326 */
327 anv_state_pool_free(queue->pool, queue->completed_serial);
328 #endif
329 }
330
331 static void
332 anv_device_init_border_colors(struct anv_device *device)
333 {
334 float float_border_colors[][4] = {
335 [VK_BORDER_COLOR_OPAQUE_WHITE] = { 1.0, 1.0, 1.0, 1.0 },
336 [VK_BORDER_COLOR_TRANSPARENT_BLACK] = { 0.0, 0.0, 0.0, 0.0 },
337 [VK_BORDER_COLOR_OPAQUE_BLACK] = { 0.0, 0.0, 0.0, 1.0 }
338 };
339
340 uint32_t uint32_border_colors[][4] = {
341 [VK_BORDER_COLOR_OPAQUE_WHITE] = { 1, 1, 1, 1 },
342 [VK_BORDER_COLOR_TRANSPARENT_BLACK] = { 0, 0, 0, 0 },
343 [VK_BORDER_COLOR_OPAQUE_BLACK] = { 0, 0, 0, 1 }
344 };
345
346 device->float_border_colors =
347 anv_state_pool_alloc(&device->dynamic_state_pool,
348 sizeof(float_border_colors), 32);
349 memcpy(device->float_border_colors.map,
350 float_border_colors, sizeof(float_border_colors));
351
352 device->uint32_border_colors =
353 anv_state_pool_alloc(&device->dynamic_state_pool,
354 sizeof(uint32_border_colors), 32);
355 memcpy(device->uint32_border_colors.map,
356 uint32_border_colors, sizeof(uint32_border_colors));
357
358 }
359
360 static const uint32_t BATCH_SIZE = 8192;
361
362 VkResult anv_CreateDevice(
363 VkPhysicalDevice _physicalDevice,
364 const VkDeviceCreateInfo* pCreateInfo,
365 VkDevice* pDevice)
366 {
367 struct anv_physical_device *physicalDevice =
368 (struct anv_physical_device *) _physicalDevice;
369 struct anv_instance *instance = physicalDevice->instance;
370 struct anv_device *device;
371
372 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO);
373
374 device = instance->pfnAlloc(instance->pAllocUserData,
375 sizeof(*device), 8,
376 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
377 if (!device)
378 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
379
380 device->no_hw = physicalDevice->no_hw;
381 parse_debug_flags(device);
382
383 device->instance = physicalDevice->instance;
384 device->fd = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC);
385 if (device->fd == -1)
386 goto fail_device;
387
388 device->context_id = anv_gem_create_context(device);
389 if (device->context_id == -1)
390 goto fail_fd;
391
392 anv_bo_pool_init(&device->batch_bo_pool, device, BATCH_SIZE);
393
394 anv_block_pool_init(&device->dynamic_state_block_pool, device, 2048);
395
396 anv_state_pool_init(&device->dynamic_state_pool,
397 &device->dynamic_state_block_pool);
398
399 anv_block_pool_init(&device->instruction_block_pool, device, 2048);
400 anv_block_pool_init(&device->surface_state_block_pool, device, 2048);
401
402 anv_state_pool_init(&device->surface_state_pool,
403 &device->surface_state_block_pool);
404
405 anv_block_pool_init(&device->scratch_block_pool, device, 0x10000);
406
407 device->info = *physicalDevice->info;
408
409 device->compiler = anv_compiler_create(device);
410 device->aub_writer = NULL;
411
412 pthread_mutex_init(&device->mutex, NULL);
413
414 anv_queue_init(device, &device->queue);
415
416 anv_device_init_meta(device);
417
418 anv_device_init_border_colors(device);
419
420 *pDevice = (VkDevice) device;
421
422 return VK_SUCCESS;
423
424 fail_fd:
425 close(device->fd);
426 fail_device:
427 anv_device_free(device, device);
428
429 return vk_error(VK_ERROR_UNAVAILABLE);
430 }
431
432 VkResult anv_DestroyDevice(
433 VkDevice _device)
434 {
435 struct anv_device *device = (struct anv_device *) _device;
436
437 anv_compiler_destroy(device->compiler);
438
439 anv_queue_finish(&device->queue);
440
441 anv_device_finish_meta(device);
442
443 #ifdef HAVE_VALGRIND
444 /* We only need to free these to prevent valgrind errors. The backing
445 * BO will go away in a couple of lines so we don't actually leak.
446 */
447 anv_state_pool_free(&device->dynamic_state_pool,
448 device->float_border_colors);
449 anv_state_pool_free(&device->dynamic_state_pool,
450 device->uint32_border_colors);
451 #endif
452
453 anv_bo_pool_finish(&device->batch_bo_pool);
454 anv_block_pool_finish(&device->dynamic_state_block_pool);
455 anv_block_pool_finish(&device->instruction_block_pool);
456 anv_block_pool_finish(&device->surface_state_block_pool);
457
458 close(device->fd);
459
460 if (device->aub_writer)
461 anv_aub_writer_destroy(device->aub_writer);
462
463 anv_device_free(device, device);
464
465 return VK_SUCCESS;
466 }
467
468 VkResult anv_GetGlobalExtensionInfo(
469 VkExtensionInfoType infoType,
470 uint32_t extensionIndex,
471 size_t* pDataSize,
472 void* pData)
473 {
474 static const VkExtensionProperties extensions[] = {
475 {
476 .extName = "VK_WSI_LunarG",
477 .version = 3
478 }
479 };
480 uint32_t count = ARRAY_SIZE(extensions);
481
482 switch (infoType) {
483 case VK_EXTENSION_INFO_TYPE_COUNT:
484 memcpy(pData, &count, sizeof(count));
485 *pDataSize = sizeof(count);
486 return VK_SUCCESS;
487
488 case VK_EXTENSION_INFO_TYPE_PROPERTIES:
489 if (extensionIndex >= count)
490 return vk_error(VK_ERROR_INVALID_EXTENSION);
491
492 memcpy(pData, &extensions[extensionIndex], sizeof(extensions[0]));
493 *pDataSize = sizeof(extensions[0]);
494 return VK_SUCCESS;
495
496 default:
497 return VK_UNSUPPORTED;
498 }
499 }
500
501 VkResult anv_GetPhysicalDeviceExtensionInfo(
502 VkPhysicalDevice physicalDevice,
503 VkExtensionInfoType infoType,
504 uint32_t extensionIndex,
505 size_t* pDataSize,
506 void* pData)
507 {
508 uint32_t *count;
509
510 switch (infoType) {
511 case VK_EXTENSION_INFO_TYPE_COUNT:
512 *pDataSize = 4;
513 if (pData == NULL)
514 return VK_SUCCESS;
515
516 count = pData;
517 *count = 0;
518 return VK_SUCCESS;
519
520 case VK_EXTENSION_INFO_TYPE_PROPERTIES:
521 return vk_error(VK_ERROR_INVALID_EXTENSION);
522
523 default:
524 return VK_UNSUPPORTED;
525 }
526 }
527
528 VkResult anv_EnumerateLayers(
529 VkPhysicalDevice physicalDevice,
530 size_t maxStringSize,
531 size_t* pLayerCount,
532 char* const* pOutLayers,
533 void* pReserved)
534 {
535 *pLayerCount = 0;
536
537 return VK_SUCCESS;
538 }
539
540 VkResult anv_GetDeviceQueue(
541 VkDevice _device,
542 uint32_t queueNodeIndex,
543 uint32_t queueIndex,
544 VkQueue* pQueue)
545 {
546 struct anv_device *device = (struct anv_device *) _device;
547
548 assert(queueIndex == 0);
549
550 *pQueue = (VkQueue) &device->queue;
551
552 return VK_SUCCESS;
553 }
554
555 VkResult
556 anv_reloc_list_init(struct anv_reloc_list *list, struct anv_device *device)
557 {
558 list->num_relocs = 0;
559 list->array_length = 256;
560 list->relocs =
561 anv_device_alloc(device, list->array_length * sizeof(*list->relocs), 8,
562 VK_SYSTEM_ALLOC_TYPE_INTERNAL);
563
564 if (list->relocs == NULL)
565 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
566
567 list->reloc_bos =
568 anv_device_alloc(device, list->array_length * sizeof(*list->reloc_bos), 8,
569 VK_SYSTEM_ALLOC_TYPE_INTERNAL);
570
571 if (list->relocs == NULL) {
572 anv_device_free(device, list->relocs);
573 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
574 }
575
576 return VK_SUCCESS;
577 }
578
579 void
580 anv_reloc_list_finish(struct anv_reloc_list *list, struct anv_device *device)
581 {
582 anv_device_free(device, list->relocs);
583 anv_device_free(device, list->reloc_bos);
584 }
585
586 static VkResult
587 anv_reloc_list_grow(struct anv_reloc_list *list, struct anv_device *device,
588 size_t num_additional_relocs)
589 {
590 if (list->num_relocs + num_additional_relocs <= list->array_length)
591 return VK_SUCCESS;
592
593 size_t new_length = list->array_length * 2;
594 while (new_length < list->num_relocs + num_additional_relocs)
595 new_length *= 2;
596
597 struct drm_i915_gem_relocation_entry *new_relocs =
598 anv_device_alloc(device, new_length * sizeof(*list->relocs), 8,
599 VK_SYSTEM_ALLOC_TYPE_INTERNAL);
600 if (new_relocs == NULL)
601 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
602
603 struct anv_bo **new_reloc_bos =
604 anv_device_alloc(device, new_length * sizeof(*list->reloc_bos), 8,
605 VK_SYSTEM_ALLOC_TYPE_INTERNAL);
606 if (new_relocs == NULL) {
607 anv_device_free(device, new_relocs);
608 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
609 }
610
611 memcpy(new_relocs, list->relocs, list->num_relocs * sizeof(*list->relocs));
612 memcpy(new_reloc_bos, list->reloc_bos,
613 list->num_relocs * sizeof(*list->reloc_bos));
614
615 anv_device_free(device, list->relocs);
616 anv_device_free(device, list->reloc_bos);
617
618 list->relocs = new_relocs;
619 list->reloc_bos = new_reloc_bos;
620
621 return VK_SUCCESS;
622 }
623
624 static VkResult
625 anv_batch_bo_create(struct anv_device *device, struct anv_batch_bo **bbo_out)
626 {
627 VkResult result;
628
629 struct anv_batch_bo *bbo =
630 anv_device_alloc(device, sizeof(*bbo), 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL);
631 if (bbo == NULL)
632 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
633
634 bbo->num_relocs = 0;
635 bbo->prev_batch_bo = NULL;
636
637 result = anv_bo_pool_alloc(&device->batch_bo_pool, &bbo->bo);
638 if (result != VK_SUCCESS) {
639 anv_device_free(device, bbo);
640 return result;
641 }
642
643 *bbo_out = bbo;
644
645 return VK_SUCCESS;
646 }
647
648 static void
649 anv_batch_bo_start(struct anv_batch_bo *bbo, struct anv_batch *batch,
650 size_t batch_padding)
651 {
652 batch->next = batch->start = bbo->bo.map;
653 batch->end = bbo->bo.map + bbo->bo.size - batch_padding;
654 bbo->first_reloc = batch->relocs.num_relocs;
655 }
656
657 static void
658 anv_batch_bo_finish(struct anv_batch_bo *bbo, struct anv_batch *batch)
659 {
660 assert(batch->start == bbo->bo.map);
661 bbo->length = batch->next - batch->start;
662 VG(VALGRIND_CHECK_MEM_IS_DEFINED(batch->start, bbo->length));
663 bbo->num_relocs = batch->relocs.num_relocs - bbo->first_reloc;
664 }
665
666 static void
667 anv_batch_bo_destroy(struct anv_batch_bo *bbo, struct anv_device *device)
668 {
669 anv_bo_pool_free(&device->batch_bo_pool, &bbo->bo);
670 anv_device_free(device, bbo);
671 }
672
673 void *
674 anv_batch_emit_dwords(struct anv_batch *batch, int num_dwords)
675 {
676 if (batch->next + num_dwords * 4 > batch->end)
677 batch->extend_cb(batch, batch->user_data);
678
679 void *p = batch->next;
680
681 batch->next += num_dwords * 4;
682 assert(batch->next <= batch->end);
683
684 return p;
685 }
686
687 static void
688 anv_reloc_list_append(struct anv_reloc_list *list, struct anv_device *device,
689 struct anv_reloc_list *other, uint32_t offset)
690 {
691 anv_reloc_list_grow(list, device, other->num_relocs);
692 /* TODO: Handle failure */
693
694 memcpy(&list->relocs[list->num_relocs], &other->relocs[0],
695 other->num_relocs * sizeof(other->relocs[0]));
696 memcpy(&list->reloc_bos[list->num_relocs], &other->reloc_bos[0],
697 other->num_relocs * sizeof(other->reloc_bos[0]));
698
699 for (uint32_t i = 0; i < other->num_relocs; i++)
700 list->relocs[i + list->num_relocs].offset += offset;
701
702 list->num_relocs += other->num_relocs;
703 }
704
705 static uint64_t
706 anv_reloc_list_add(struct anv_reloc_list *list, struct anv_device *device,
707 uint32_t offset, struct anv_bo *target_bo, uint32_t delta)
708 {
709 struct drm_i915_gem_relocation_entry *entry;
710 int index;
711
712 anv_reloc_list_grow(list, device, 1);
713 /* TODO: Handle failure */
714
715 /* XXX: Can we use I915_EXEC_HANDLE_LUT? */
716 index = list->num_relocs++;
717 list->reloc_bos[index] = target_bo;
718 entry = &list->relocs[index];
719 entry->target_handle = target_bo->gem_handle;
720 entry->delta = delta;
721 entry->offset = offset;
722 entry->presumed_offset = target_bo->offset;
723 entry->read_domains = 0;
724 entry->write_domain = 0;
725
726 return target_bo->offset + delta;
727 }
728
729 void
730 anv_batch_emit_batch(struct anv_batch *batch, struct anv_batch *other)
731 {
732 uint32_t size, offset;
733
734 size = other->next - other->start;
735 assert(size % 4 == 0);
736
737 if (batch->next + size > batch->end)
738 batch->extend_cb(batch, batch->user_data);
739
740 assert(batch->next + size <= batch->end);
741
742 memcpy(batch->next, other->start, size);
743
744 offset = batch->next - batch->start;
745 anv_reloc_list_append(&batch->relocs, batch->device,
746 &other->relocs, offset);
747
748 batch->next += size;
749 }
750
751 uint64_t
752 anv_batch_emit_reloc(struct anv_batch *batch,
753 void *location, struct anv_bo *bo, uint32_t delta)
754 {
755 return anv_reloc_list_add(&batch->relocs, batch->device,
756 location - batch->start, bo, delta);
757 }
758
759 VkResult anv_QueueSubmit(
760 VkQueue _queue,
761 uint32_t cmdBufferCount,
762 const VkCmdBuffer* pCmdBuffers,
763 VkFence _fence)
764 {
765 struct anv_queue *queue = (struct anv_queue *) _queue;
766 struct anv_device *device = queue->device;
767 struct anv_fence *fence = (struct anv_fence *) _fence;
768 int ret;
769
770 for (uint32_t i = 0; i < cmdBufferCount; i++) {
771 struct anv_cmd_buffer *cmd_buffer =
772 (struct anv_cmd_buffer *) pCmdBuffers[i];
773
774 if (device->dump_aub)
775 anv_cmd_buffer_dump(cmd_buffer);
776
777 if (!device->no_hw) {
778 ret = anv_gem_execbuffer(device, &cmd_buffer->execbuf);
779 if (ret != 0)
780 return vk_error(VK_ERROR_UNKNOWN);
781
782 if (fence) {
783 ret = anv_gem_execbuffer(device, &fence->execbuf);
784 if (ret != 0)
785 return vk_error(VK_ERROR_UNKNOWN);
786 }
787
788 for (uint32_t i = 0; i < cmd_buffer->bo_count; i++)
789 cmd_buffer->exec2_bos[i]->offset = cmd_buffer->exec2_objects[i].offset;
790 } else {
791 *(uint32_t *)queue->completed_serial.map = cmd_buffer->serial;
792 }
793 }
794
795 return VK_SUCCESS;
796 }
797
798 VkResult anv_QueueWaitIdle(
799 VkQueue _queue)
800 {
801 struct anv_queue *queue = (struct anv_queue *) _queue;
802
803 return vkDeviceWaitIdle((VkDevice) queue->device);
804 }
805
806 VkResult anv_DeviceWaitIdle(
807 VkDevice _device)
808 {
809 struct anv_device *device = (struct anv_device *) _device;
810 struct anv_state state;
811 struct anv_batch batch;
812 struct drm_i915_gem_execbuffer2 execbuf;
813 struct drm_i915_gem_exec_object2 exec2_objects[1];
814 struct anv_bo *bo = NULL;
815 VkResult result;
816 int64_t timeout;
817 int ret;
818
819 state = anv_state_pool_alloc(&device->dynamic_state_pool, 32, 32);
820 bo = &device->dynamic_state_pool.block_pool->bo;
821 batch.start = batch.next = state.map;
822 batch.end = state.map + 32;
823 anv_batch_emit(&batch, GEN8_MI_BATCH_BUFFER_END);
824 anv_batch_emit(&batch, GEN8_MI_NOOP);
825
826 exec2_objects[0].handle = bo->gem_handle;
827 exec2_objects[0].relocation_count = 0;
828 exec2_objects[0].relocs_ptr = 0;
829 exec2_objects[0].alignment = 0;
830 exec2_objects[0].offset = bo->offset;
831 exec2_objects[0].flags = 0;
832 exec2_objects[0].rsvd1 = 0;
833 exec2_objects[0].rsvd2 = 0;
834
835 execbuf.buffers_ptr = (uintptr_t) exec2_objects;
836 execbuf.buffer_count = 1;
837 execbuf.batch_start_offset = state.offset;
838 execbuf.batch_len = batch.next - state.map;
839 execbuf.cliprects_ptr = 0;
840 execbuf.num_cliprects = 0;
841 execbuf.DR1 = 0;
842 execbuf.DR4 = 0;
843
844 execbuf.flags =
845 I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
846 execbuf.rsvd1 = device->context_id;
847 execbuf.rsvd2 = 0;
848
849 if (!device->no_hw) {
850 ret = anv_gem_execbuffer(device, &execbuf);
851 if (ret != 0) {
852 result = vk_error(VK_ERROR_UNKNOWN);
853 goto fail;
854 }
855
856 timeout = INT64_MAX;
857 ret = anv_gem_wait(device, bo->gem_handle, &timeout);
858 if (ret != 0) {
859 result = vk_error(VK_ERROR_UNKNOWN);
860 goto fail;
861 }
862 }
863
864 anv_state_pool_free(&device->dynamic_state_pool, state);
865
866 return VK_SUCCESS;
867
868 fail:
869 anv_state_pool_free(&device->dynamic_state_pool, state);
870
871 return result;
872 }
873
874 void *
875 anv_device_alloc(struct anv_device * device,
876 size_t size,
877 size_t alignment,
878 VkSystemAllocType allocType)
879 {
880 return device->instance->pfnAlloc(device->instance->pAllocUserData,
881 size,
882 alignment,
883 allocType);
884 }
885
886 void
887 anv_device_free(struct anv_device * device,
888 void * mem)
889 {
890 return device->instance->pfnFree(device->instance->pAllocUserData,
891 mem);
892 }
893
894 VkResult
895 anv_bo_init_new(struct anv_bo *bo, struct anv_device *device, uint64_t size)
896 {
897 bo->gem_handle = anv_gem_create(device, size);
898 if (!bo->gem_handle)
899 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
900
901 bo->map = NULL;
902 bo->index = 0;
903 bo->offset = 0;
904 bo->size = size;
905
906 return VK_SUCCESS;
907 }
908
909 VkResult anv_AllocMemory(
910 VkDevice _device,
911 const VkMemoryAllocInfo* pAllocInfo,
912 VkDeviceMemory* pMem)
913 {
914 struct anv_device *device = (struct anv_device *) _device;
915 struct anv_device_memory *mem;
916 VkResult result;
917
918 assert(pAllocInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOC_INFO);
919
920 mem = anv_device_alloc(device, sizeof(*mem), 8,
921 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
922 if (mem == NULL)
923 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
924
925 result = anv_bo_init_new(&mem->bo, device, pAllocInfo->allocationSize);
926 if (result != VK_SUCCESS)
927 goto fail;
928
929 *pMem = (VkDeviceMemory) mem;
930
931 return VK_SUCCESS;
932
933 fail:
934 anv_device_free(device, mem);
935
936 return result;
937 }
938
939 VkResult anv_FreeMemory(
940 VkDevice _device,
941 VkDeviceMemory _mem)
942 {
943 struct anv_device *device = (struct anv_device *) _device;
944 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
945
946 if (mem->bo.map)
947 anv_gem_munmap(mem->bo.map, mem->bo.size);
948
949 if (mem->bo.gem_handle != 0)
950 anv_gem_close(device, mem->bo.gem_handle);
951
952 anv_device_free(device, mem);
953
954 return VK_SUCCESS;
955 }
956
957 VkResult anv_MapMemory(
958 VkDevice _device,
959 VkDeviceMemory _mem,
960 VkDeviceSize offset,
961 VkDeviceSize size,
962 VkMemoryMapFlags flags,
963 void** ppData)
964 {
965 struct anv_device *device = (struct anv_device *) _device;
966 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
967
968 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
969 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
970 * at a time is valid. We could just mmap up front and return an offset
971 * pointer here, but that may exhaust virtual memory on 32 bit
972 * userspace. */
973
974 mem->map = anv_gem_mmap(device, mem->bo.gem_handle, offset, size);
975 mem->map_size = size;
976
977 *ppData = mem->map;
978
979 return VK_SUCCESS;
980 }
981
982 VkResult anv_UnmapMemory(
983 VkDevice _device,
984 VkDeviceMemory _mem)
985 {
986 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
987
988 anv_gem_munmap(mem->map, mem->map_size);
989
990 return VK_SUCCESS;
991 }
992
993 VkResult anv_FlushMappedMemory(
994 VkDevice device,
995 VkDeviceMemory mem,
996 VkDeviceSize offset,
997 VkDeviceSize size)
998 {
999 /* clflush here for !llc platforms */
1000
1001 return VK_SUCCESS;
1002 }
1003
1004 VkResult anv_PinSystemMemory(
1005 VkDevice device,
1006 const void* pSysMem,
1007 size_t memSize,
1008 VkDeviceMemory* pMem)
1009 {
1010 return VK_SUCCESS;
1011 }
1012
1013 VkResult anv_DestroyObject(
1014 VkDevice _device,
1015 VkObjectType objType,
1016 VkObject _object)
1017 {
1018 struct anv_device *device = (struct anv_device *) _device;
1019 struct anv_object *object = (struct anv_object *) _object;
1020
1021 switch (objType) {
1022 case VK_OBJECT_TYPE_INSTANCE:
1023 return anv_DestroyInstance((VkInstance) _object);
1024
1025 case VK_OBJECT_TYPE_PHYSICAL_DEVICE:
1026 /* We don't want to actually destroy physical devices */
1027 return VK_SUCCESS;
1028
1029 case VK_OBJECT_TYPE_DEVICE:
1030 assert(_device == (VkDevice) _object);
1031 return anv_DestroyDevice((VkDevice) _object);
1032
1033 case VK_OBJECT_TYPE_QUEUE:
1034 /* TODO */
1035 return VK_SUCCESS;
1036
1037 case VK_OBJECT_TYPE_DEVICE_MEMORY:
1038 return anv_FreeMemory(_device, (VkDeviceMemory) _object);
1039
1040 case VK_OBJECT_TYPE_DESCRIPTOR_POOL:
1041 /* These are just dummys anyway, so we don't need to destroy them */
1042 return VK_SUCCESS;
1043
1044 case VK_OBJECT_TYPE_BUFFER:
1045 case VK_OBJECT_TYPE_IMAGE:
1046 case VK_OBJECT_TYPE_DEPTH_STENCIL_VIEW:
1047 case VK_OBJECT_TYPE_SHADER:
1048 case VK_OBJECT_TYPE_PIPELINE_LAYOUT:
1049 case VK_OBJECT_TYPE_SAMPLER:
1050 case VK_OBJECT_TYPE_DESCRIPTOR_SET:
1051 case VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT:
1052 case VK_OBJECT_TYPE_DYNAMIC_RS_STATE:
1053 case VK_OBJECT_TYPE_DYNAMIC_CB_STATE:
1054 case VK_OBJECT_TYPE_DYNAMIC_DS_STATE:
1055 case VK_OBJECT_TYPE_RENDER_PASS:
1056 /* These are trivially destroyable */
1057 anv_device_free(device, (void *) _object);
1058 return VK_SUCCESS;
1059
1060 case VK_OBJECT_TYPE_COMMAND_BUFFER:
1061 case VK_OBJECT_TYPE_PIPELINE:
1062 case VK_OBJECT_TYPE_DYNAMIC_VP_STATE:
1063 case VK_OBJECT_TYPE_FENCE:
1064 case VK_OBJECT_TYPE_QUERY_POOL:
1065 case VK_OBJECT_TYPE_FRAMEBUFFER:
1066 case VK_OBJECT_TYPE_BUFFER_VIEW:
1067 case VK_OBJECT_TYPE_IMAGE_VIEW:
1068 case VK_OBJECT_TYPE_COLOR_ATTACHMENT_VIEW:
1069 (object->destructor)(device, object, objType);
1070 return VK_SUCCESS;
1071
1072 case VK_OBJECT_TYPE_SEMAPHORE:
1073 case VK_OBJECT_TYPE_EVENT:
1074 stub_return(VK_UNSUPPORTED);
1075
1076 default:
1077 unreachable("Invalid object type");
1078 }
1079 }
1080
1081 static void
1082 fill_memory_requirements(
1083 VkObjectType objType,
1084 VkObject object,
1085 VkMemoryRequirements * memory_requirements)
1086 {
1087 struct anv_buffer *buffer;
1088 struct anv_image *image;
1089
1090 memory_requirements->memPropsAllowed =
1091 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
1092 /* VK_MEMORY_PROPERTY_HOST_NON_COHERENT_BIT | */
1093 /* VK_MEMORY_PROPERTY_HOST_UNCACHED_BIT | */
1094 VK_MEMORY_PROPERTY_HOST_WRITE_COMBINED_BIT;
1095
1096 memory_requirements->memPropsRequired = 0;
1097
1098 switch (objType) {
1099 case VK_OBJECT_TYPE_BUFFER:
1100 buffer = (struct anv_buffer *) object;
1101 memory_requirements->size = buffer->size;
1102 memory_requirements->alignment = 16;
1103 break;
1104 case VK_OBJECT_TYPE_IMAGE:
1105 image = (struct anv_image *) object;
1106 memory_requirements->size = image->size;
1107 memory_requirements->alignment = image->alignment;
1108 break;
1109 default:
1110 memory_requirements->size = 0;
1111 break;
1112 }
1113 }
1114
1115 static uint32_t
1116 get_allocation_count(VkObjectType objType)
1117 {
1118 switch (objType) {
1119 case VK_OBJECT_TYPE_BUFFER:
1120 case VK_OBJECT_TYPE_IMAGE:
1121 return 1;
1122 default:
1123 return 0;
1124 }
1125 }
1126
1127 VkResult anv_GetObjectInfo(
1128 VkDevice _device,
1129 VkObjectType objType,
1130 VkObject object,
1131 VkObjectInfoType infoType,
1132 size_t* pDataSize,
1133 void* pData)
1134 {
1135 VkMemoryRequirements memory_requirements;
1136 uint32_t *count;
1137
1138 switch (infoType) {
1139 case VK_OBJECT_INFO_TYPE_MEMORY_REQUIREMENTS:
1140 *pDataSize = sizeof(memory_requirements);
1141 if (pData == NULL)
1142 return VK_SUCCESS;
1143
1144 fill_memory_requirements(objType, object, pData);
1145 return VK_SUCCESS;
1146
1147 case VK_OBJECT_INFO_TYPE_MEMORY_ALLOCATION_COUNT:
1148 *pDataSize = sizeof(count);
1149 if (pData == NULL)
1150 return VK_SUCCESS;
1151
1152 count = pData;
1153 *count = get_allocation_count(objType);
1154 return VK_SUCCESS;
1155
1156 default:
1157 return vk_error(VK_UNSUPPORTED);
1158 }
1159
1160 }
1161
1162 VkResult anv_QueueBindObjectMemory(
1163 VkQueue queue,
1164 VkObjectType objType,
1165 VkObject object,
1166 uint32_t allocationIdx,
1167 VkDeviceMemory _mem,
1168 VkDeviceSize memOffset)
1169 {
1170 struct anv_buffer *buffer;
1171 struct anv_image *image;
1172 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
1173
1174 switch (objType) {
1175 case VK_OBJECT_TYPE_BUFFER:
1176 buffer = (struct anv_buffer *) object;
1177 buffer->bo = &mem->bo;
1178 buffer->offset = memOffset;
1179 break;
1180 case VK_OBJECT_TYPE_IMAGE:
1181 image = (struct anv_image *) object;
1182 image->bo = &mem->bo;
1183 image->offset = memOffset;
1184 break;
1185 default:
1186 break;
1187 }
1188
1189 return VK_SUCCESS;
1190 }
1191
1192 VkResult anv_QueueBindObjectMemoryRange(
1193 VkQueue queue,
1194 VkObjectType objType,
1195 VkObject object,
1196 uint32_t allocationIdx,
1197 VkDeviceSize rangeOffset,
1198 VkDeviceSize rangeSize,
1199 VkDeviceMemory mem,
1200 VkDeviceSize memOffset)
1201 {
1202 stub_return(VK_UNSUPPORTED);
1203 }
1204
1205 VkResult anv_QueueBindImageMemoryRange(
1206 VkQueue queue,
1207 VkImage image,
1208 uint32_t allocationIdx,
1209 const VkImageMemoryBindInfo* pBindInfo,
1210 VkDeviceMemory mem,
1211 VkDeviceSize memOffset)
1212 {
1213 stub_return(VK_UNSUPPORTED);
1214 }
1215
1216 static void
1217 anv_fence_destroy(struct anv_device *device,
1218 struct anv_object *object,
1219 VkObjectType obj_type)
1220 {
1221 struct anv_fence *fence = (struct anv_fence *) object;
1222
1223 assert(obj_type == VK_OBJECT_TYPE_FENCE);
1224
1225 anv_gem_munmap(fence->bo.map, fence->bo.size);
1226 anv_gem_close(device, fence->bo.gem_handle);
1227 anv_device_free(device, fence);
1228 }
1229
1230 VkResult anv_CreateFence(
1231 VkDevice _device,
1232 const VkFenceCreateInfo* pCreateInfo,
1233 VkFence* pFence)
1234 {
1235 struct anv_device *device = (struct anv_device *) _device;
1236 struct anv_fence *fence;
1237 struct anv_batch batch;
1238 VkResult result;
1239
1240 const uint32_t fence_size = 128;
1241
1242 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FENCE_CREATE_INFO);
1243
1244 fence = anv_device_alloc(device, sizeof(*fence), 8,
1245 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1246 if (fence == NULL)
1247 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1248
1249 result = anv_bo_init_new(&fence->bo, device, fence_size);
1250 if (result != VK_SUCCESS)
1251 goto fail;
1252
1253 fence->base.destructor = anv_fence_destroy;
1254
1255 fence->bo.map =
1256 anv_gem_mmap(device, fence->bo.gem_handle, 0, fence->bo.size);
1257 batch.next = batch.start = fence->bo.map;
1258 batch.end = fence->bo.map + fence->bo.size;
1259 anv_batch_emit(&batch, GEN8_MI_BATCH_BUFFER_END);
1260 anv_batch_emit(&batch, GEN8_MI_NOOP);
1261
1262 fence->exec2_objects[0].handle = fence->bo.gem_handle;
1263 fence->exec2_objects[0].relocation_count = 0;
1264 fence->exec2_objects[0].relocs_ptr = 0;
1265 fence->exec2_objects[0].alignment = 0;
1266 fence->exec2_objects[0].offset = fence->bo.offset;
1267 fence->exec2_objects[0].flags = 0;
1268 fence->exec2_objects[0].rsvd1 = 0;
1269 fence->exec2_objects[0].rsvd2 = 0;
1270
1271 fence->execbuf.buffers_ptr = (uintptr_t) fence->exec2_objects;
1272 fence->execbuf.buffer_count = 1;
1273 fence->execbuf.batch_start_offset = 0;
1274 fence->execbuf.batch_len = batch.next - fence->bo.map;
1275 fence->execbuf.cliprects_ptr = 0;
1276 fence->execbuf.num_cliprects = 0;
1277 fence->execbuf.DR1 = 0;
1278 fence->execbuf.DR4 = 0;
1279
1280 fence->execbuf.flags =
1281 I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
1282 fence->execbuf.rsvd1 = device->context_id;
1283 fence->execbuf.rsvd2 = 0;
1284
1285 *pFence = (VkFence) fence;
1286
1287 return VK_SUCCESS;
1288
1289 fail:
1290 anv_device_free(device, fence);
1291
1292 return result;
1293 }
1294
1295 VkResult anv_ResetFences(
1296 VkDevice _device,
1297 uint32_t fenceCount,
1298 VkFence* pFences)
1299 {
1300 struct anv_fence **fences = (struct anv_fence **) pFences;
1301
1302 for (uint32_t i = 0; i < fenceCount; i++)
1303 fences[i]->ready = false;
1304
1305 return VK_SUCCESS;
1306 }
1307
1308 VkResult anv_GetFenceStatus(
1309 VkDevice _device,
1310 VkFence _fence)
1311 {
1312 struct anv_device *device = (struct anv_device *) _device;
1313 struct anv_fence *fence = (struct anv_fence *) _fence;
1314 int64_t t = 0;
1315 int ret;
1316
1317 if (fence->ready)
1318 return VK_SUCCESS;
1319
1320 ret = anv_gem_wait(device, fence->bo.gem_handle, &t);
1321 if (ret == 0) {
1322 fence->ready = true;
1323 return VK_SUCCESS;
1324 }
1325
1326 return VK_NOT_READY;
1327 }
1328
1329 VkResult anv_WaitForFences(
1330 VkDevice _device,
1331 uint32_t fenceCount,
1332 const VkFence* pFences,
1333 bool32_t waitAll,
1334 uint64_t timeout)
1335 {
1336 struct anv_device *device = (struct anv_device *) _device;
1337 struct anv_fence **fences = (struct anv_fence **) pFences;
1338 int64_t t = timeout;
1339 int ret;
1340
1341 /* FIXME: handle !waitAll */
1342
1343 for (uint32_t i = 0; i < fenceCount; i++) {
1344 ret = anv_gem_wait(device, fences[i]->bo.gem_handle, &t);
1345 if (ret == -1 && errno == ETIME)
1346 return VK_TIMEOUT;
1347 else if (ret == -1)
1348 return vk_error(VK_ERROR_UNKNOWN);
1349 }
1350
1351 return VK_SUCCESS;
1352 }
1353
1354 // Queue semaphore functions
1355
1356 VkResult anv_CreateSemaphore(
1357 VkDevice device,
1358 const VkSemaphoreCreateInfo* pCreateInfo,
1359 VkSemaphore* pSemaphore)
1360 {
1361 stub_return(VK_UNSUPPORTED);
1362 }
1363
1364 VkResult anv_QueueSignalSemaphore(
1365 VkQueue queue,
1366 VkSemaphore semaphore)
1367 {
1368 stub_return(VK_UNSUPPORTED);
1369 }
1370
1371 VkResult anv_QueueWaitSemaphore(
1372 VkQueue queue,
1373 VkSemaphore semaphore)
1374 {
1375 stub_return(VK_UNSUPPORTED);
1376 }
1377
1378 // Event functions
1379
1380 VkResult anv_CreateEvent(
1381 VkDevice device,
1382 const VkEventCreateInfo* pCreateInfo,
1383 VkEvent* pEvent)
1384 {
1385 stub_return(VK_UNSUPPORTED);
1386 }
1387
1388 VkResult anv_GetEventStatus(
1389 VkDevice device,
1390 VkEvent event)
1391 {
1392 stub_return(VK_UNSUPPORTED);
1393 }
1394
1395 VkResult anv_SetEvent(
1396 VkDevice device,
1397 VkEvent event)
1398 {
1399 stub_return(VK_UNSUPPORTED);
1400 }
1401
1402 VkResult anv_ResetEvent(
1403 VkDevice device,
1404 VkEvent event)
1405 {
1406 stub_return(VK_UNSUPPORTED);
1407 }
1408
1409 // Buffer functions
1410
1411 VkResult anv_CreateBuffer(
1412 VkDevice _device,
1413 const VkBufferCreateInfo* pCreateInfo,
1414 VkBuffer* pBuffer)
1415 {
1416 struct anv_device *device = (struct anv_device *) _device;
1417 struct anv_buffer *buffer;
1418
1419 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO);
1420
1421 buffer = anv_device_alloc(device, sizeof(*buffer), 8,
1422 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1423 if (buffer == NULL)
1424 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1425
1426 buffer->size = pCreateInfo->size;
1427 buffer->bo = NULL;
1428 buffer->offset = 0;
1429
1430 *pBuffer = (VkBuffer) buffer;
1431
1432 return VK_SUCCESS;
1433 }
1434
1435 // Buffer view functions
1436
1437 static void
1438 fill_buffer_surface_state(void *state, VkFormat format,
1439 uint32_t offset, uint32_t range)
1440 {
1441 const struct anv_format *info;
1442
1443 info = anv_format_for_vk_format(format);
1444 /* This assumes RGBA float format. */
1445 uint32_t stride = 4;
1446 uint32_t num_elements = range / stride;
1447
1448 struct GEN8_RENDER_SURFACE_STATE surface_state = {
1449 .SurfaceType = SURFTYPE_BUFFER,
1450 .SurfaceArray = false,
1451 .SurfaceFormat = info->surface_format,
1452 .SurfaceVerticalAlignment = VALIGN4,
1453 .SurfaceHorizontalAlignment = HALIGN4,
1454 .TileMode = LINEAR,
1455 .VerticalLineStride = 0,
1456 .VerticalLineStrideOffset = 0,
1457 .SamplerL2BypassModeDisable = true,
1458 .RenderCacheReadWriteMode = WriteOnlyCache,
1459 .MemoryObjectControlState = GEN8_MOCS,
1460 .BaseMipLevel = 0.0,
1461 .SurfaceQPitch = 0,
1462 .Height = (num_elements >> 7) & 0x3fff,
1463 .Width = num_elements & 0x7f,
1464 .Depth = (num_elements >> 21) & 0x3f,
1465 .SurfacePitch = stride - 1,
1466 .MinimumArrayElement = 0,
1467 .NumberofMultisamples = MULTISAMPLECOUNT_1,
1468 .XOffset = 0,
1469 .YOffset = 0,
1470 .SurfaceMinLOD = 0,
1471 .MIPCountLOD = 0,
1472 .AuxiliarySurfaceMode = AUX_NONE,
1473 .RedClearColor = 0,
1474 .GreenClearColor = 0,
1475 .BlueClearColor = 0,
1476 .AlphaClearColor = 0,
1477 .ShaderChannelSelectRed = SCS_RED,
1478 .ShaderChannelSelectGreen = SCS_GREEN,
1479 .ShaderChannelSelectBlue = SCS_BLUE,
1480 .ShaderChannelSelectAlpha = SCS_ALPHA,
1481 .ResourceMinLOD = 0.0,
1482 /* FIXME: We assume that the image must be bound at this time. */
1483 .SurfaceBaseAddress = { NULL, offset },
1484 };
1485
1486 GEN8_RENDER_SURFACE_STATE_pack(NULL, state, &surface_state);
1487 }
1488
1489 VkResult anv_CreateBufferView(
1490 VkDevice _device,
1491 const VkBufferViewCreateInfo* pCreateInfo,
1492 VkBufferView* pView)
1493 {
1494 struct anv_device *device = (struct anv_device *) _device;
1495 struct anv_buffer *buffer = (struct anv_buffer *) pCreateInfo->buffer;
1496 struct anv_surface_view *view;
1497
1498 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO);
1499
1500 view = anv_device_alloc(device, sizeof(*view), 8,
1501 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1502 if (view == NULL)
1503 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1504
1505 view->base.destructor = anv_surface_view_destroy;
1506
1507 view->bo = buffer->bo;
1508 view->offset = buffer->offset + pCreateInfo->offset;
1509 view->surface_state =
1510 anv_state_pool_alloc(&device->surface_state_pool, 64, 64);
1511 view->format = pCreateInfo->format;
1512 view->range = pCreateInfo->range;
1513
1514 fill_buffer_surface_state(view->surface_state.map,
1515 pCreateInfo->format, view->offset, pCreateInfo->range);
1516
1517 *pView = (VkBufferView) view;
1518
1519 return VK_SUCCESS;
1520 }
1521
1522 // Sampler functions
1523
1524 VkResult anv_CreateSampler(
1525 VkDevice _device,
1526 const VkSamplerCreateInfo* pCreateInfo,
1527 VkSampler* pSampler)
1528 {
1529 struct anv_device *device = (struct anv_device *) _device;
1530 struct anv_sampler *sampler;
1531 uint32_t mag_filter, min_filter, max_anisotropy;
1532
1533 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
1534
1535 sampler = anv_device_alloc(device, sizeof(*sampler), 8,
1536 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1537 if (!sampler)
1538 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1539
1540 static const uint32_t vk_to_gen_tex_filter[] = {
1541 [VK_TEX_FILTER_NEAREST] = MAPFILTER_NEAREST,
1542 [VK_TEX_FILTER_LINEAR] = MAPFILTER_LINEAR
1543 };
1544
1545 static const uint32_t vk_to_gen_mipmap_mode[] = {
1546 [VK_TEX_MIPMAP_MODE_BASE] = MIPFILTER_NONE,
1547 [VK_TEX_MIPMAP_MODE_NEAREST] = MIPFILTER_NEAREST,
1548 [VK_TEX_MIPMAP_MODE_LINEAR] = MIPFILTER_LINEAR
1549 };
1550
1551 static const uint32_t vk_to_gen_tex_address[] = {
1552 [VK_TEX_ADDRESS_WRAP] = TCM_WRAP,
1553 [VK_TEX_ADDRESS_MIRROR] = TCM_MIRROR,
1554 [VK_TEX_ADDRESS_CLAMP] = TCM_CLAMP,
1555 [VK_TEX_ADDRESS_MIRROR_ONCE] = TCM_MIRROR_ONCE,
1556 [VK_TEX_ADDRESS_CLAMP_BORDER] = TCM_CLAMP_BORDER,
1557 };
1558
1559 static const uint32_t vk_to_gen_compare_op[] = {
1560 [VK_COMPARE_OP_NEVER] = PREFILTEROPNEVER,
1561 [VK_COMPARE_OP_LESS] = PREFILTEROPLESS,
1562 [VK_COMPARE_OP_EQUAL] = PREFILTEROPEQUAL,
1563 [VK_COMPARE_OP_LESS_EQUAL] = PREFILTEROPLEQUAL,
1564 [VK_COMPARE_OP_GREATER] = PREFILTEROPGREATER,
1565 [VK_COMPARE_OP_NOT_EQUAL] = PREFILTEROPNOTEQUAL,
1566 [VK_COMPARE_OP_GREATER_EQUAL] = PREFILTEROPGEQUAL,
1567 [VK_COMPARE_OP_ALWAYS] = PREFILTEROPALWAYS,
1568 };
1569
1570 if (pCreateInfo->maxAnisotropy > 1) {
1571 mag_filter = MAPFILTER_ANISOTROPIC;
1572 min_filter = MAPFILTER_ANISOTROPIC;
1573 max_anisotropy = (pCreateInfo->maxAnisotropy - 2) / 2;
1574 } else {
1575 mag_filter = vk_to_gen_tex_filter[pCreateInfo->magFilter];
1576 min_filter = vk_to_gen_tex_filter[pCreateInfo->minFilter];
1577 max_anisotropy = RATIO21;
1578 }
1579
1580 struct GEN8_SAMPLER_STATE sampler_state = {
1581 .SamplerDisable = false,
1582 .TextureBorderColorMode = DX10OGL,
1583 .LODPreClampMode = 0,
1584 .BaseMipLevel = 0.0,
1585 .MipModeFilter = vk_to_gen_mipmap_mode[pCreateInfo->mipMode],
1586 .MagModeFilter = mag_filter,
1587 .MinModeFilter = min_filter,
1588 .TextureLODBias = pCreateInfo->mipLodBias * 256,
1589 .AnisotropicAlgorithm = EWAApproximation,
1590 .MinLOD = pCreateInfo->minLod,
1591 .MaxLOD = pCreateInfo->maxLod,
1592 .ChromaKeyEnable = 0,
1593 .ChromaKeyIndex = 0,
1594 .ChromaKeyMode = 0,
1595 .ShadowFunction = vk_to_gen_compare_op[pCreateInfo->compareOp],
1596 .CubeSurfaceControlMode = 0,
1597
1598 .IndirectStatePointer =
1599 device->float_border_colors.offset +
1600 pCreateInfo->borderColor * sizeof(float) * 4,
1601
1602 .LODClampMagnificationMode = MIPNONE,
1603 .MaximumAnisotropy = max_anisotropy,
1604 .RAddressMinFilterRoundingEnable = 0,
1605 .RAddressMagFilterRoundingEnable = 0,
1606 .VAddressMinFilterRoundingEnable = 0,
1607 .VAddressMagFilterRoundingEnable = 0,
1608 .UAddressMinFilterRoundingEnable = 0,
1609 .UAddressMagFilterRoundingEnable = 0,
1610 .TrilinearFilterQuality = 0,
1611 .NonnormalizedCoordinateEnable = 0,
1612 .TCXAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressU],
1613 .TCYAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressV],
1614 .TCZAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressW],
1615 };
1616
1617 GEN8_SAMPLER_STATE_pack(NULL, sampler->state, &sampler_state);
1618
1619 *pSampler = (VkSampler) sampler;
1620
1621 return VK_SUCCESS;
1622 }
1623
1624 // Descriptor set functions
1625
1626 VkResult anv_CreateDescriptorSetLayout(
1627 VkDevice _device,
1628 const VkDescriptorSetLayoutCreateInfo* pCreateInfo,
1629 VkDescriptorSetLayout* pSetLayout)
1630 {
1631 struct anv_device *device = (struct anv_device *) _device;
1632 struct anv_descriptor_set_layout *set_layout;
1633
1634 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO);
1635
1636 uint32_t sampler_count[VK_SHADER_STAGE_NUM] = { 0, };
1637 uint32_t surface_count[VK_SHADER_STAGE_NUM] = { 0, };
1638 uint32_t num_dynamic_buffers = 0;
1639 uint32_t count = 0;
1640 uint32_t stages = 0;
1641 uint32_t s;
1642
1643 for (uint32_t i = 0; i < pCreateInfo->count; i++) {
1644 switch (pCreateInfo->pBinding[i].descriptorType) {
1645 case VK_DESCRIPTOR_TYPE_SAMPLER:
1646 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1647 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1648 sampler_count[s] += pCreateInfo->pBinding[i].arraySize;
1649 break;
1650 default:
1651 break;
1652 }
1653
1654 switch (pCreateInfo->pBinding[i].descriptorType) {
1655 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1656 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
1657 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
1658 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
1659 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
1660 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
1661 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
1662 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1663 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1664 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1665 surface_count[s] += pCreateInfo->pBinding[i].arraySize;
1666 break;
1667 default:
1668 break;
1669 }
1670
1671 switch (pCreateInfo->pBinding[i].descriptorType) {
1672 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1673 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1674 num_dynamic_buffers += pCreateInfo->pBinding[i].arraySize;
1675 break;
1676 default:
1677 break;
1678 }
1679
1680 stages |= pCreateInfo->pBinding[i].stageFlags;
1681 count += pCreateInfo->pBinding[i].arraySize;
1682 }
1683
1684 uint32_t sampler_total = 0;
1685 uint32_t surface_total = 0;
1686 for (uint32_t s = 0; s < VK_SHADER_STAGE_NUM; s++) {
1687 sampler_total += sampler_count[s];
1688 surface_total += surface_count[s];
1689 }
1690
1691 size_t size = sizeof(*set_layout) +
1692 (sampler_total + surface_total) * sizeof(set_layout->entries[0]);
1693 set_layout = anv_device_alloc(device, size, 8,
1694 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1695 if (!set_layout)
1696 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1697
1698 set_layout->num_dynamic_buffers = num_dynamic_buffers;
1699 set_layout->count = count;
1700 set_layout->shader_stages = stages;
1701
1702 struct anv_descriptor_slot *p = set_layout->entries;
1703 struct anv_descriptor_slot *sampler[VK_SHADER_STAGE_NUM];
1704 struct anv_descriptor_slot *surface[VK_SHADER_STAGE_NUM];
1705 for (uint32_t s = 0; s < VK_SHADER_STAGE_NUM; s++) {
1706 set_layout->stage[s].surface_count = surface_count[s];
1707 set_layout->stage[s].surface_start = surface[s] = p;
1708 p += surface_count[s];
1709 set_layout->stage[s].sampler_count = sampler_count[s];
1710 set_layout->stage[s].sampler_start = sampler[s] = p;
1711 p += sampler_count[s];
1712 }
1713
1714 uint32_t descriptor = 0;
1715 int8_t dynamic_slot = 0;
1716 bool is_dynamic;
1717 for (uint32_t i = 0; i < pCreateInfo->count; i++) {
1718 switch (pCreateInfo->pBinding[i].descriptorType) {
1719 case VK_DESCRIPTOR_TYPE_SAMPLER:
1720 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1721 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1722 for (uint32_t j = 0; j < pCreateInfo->pBinding[i].arraySize; j++) {
1723 sampler[s]->index = descriptor + j;
1724 sampler[s]->dynamic_slot = -1;
1725 sampler[s]++;
1726 }
1727 break;
1728 default:
1729 break;
1730 }
1731
1732 switch (pCreateInfo->pBinding[i].descriptorType) {
1733 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1734 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1735 is_dynamic = true;
1736 break;
1737 default:
1738 is_dynamic = false;
1739 break;
1740 }
1741
1742 switch (pCreateInfo->pBinding[i].descriptorType) {
1743 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1744 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
1745 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
1746 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
1747 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
1748 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
1749 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
1750 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1751 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1752 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1753 for (uint32_t j = 0; j < pCreateInfo->pBinding[i].arraySize; j++) {
1754 surface[s]->index = descriptor + j;
1755 if (is_dynamic)
1756 surface[s]->dynamic_slot = dynamic_slot + j;
1757 else
1758 surface[s]->dynamic_slot = -1;
1759 surface[s]++;
1760 }
1761 break;
1762 default:
1763 break;
1764 }
1765
1766 if (is_dynamic)
1767 dynamic_slot += pCreateInfo->pBinding[i].arraySize;
1768
1769 descriptor += pCreateInfo->pBinding[i].arraySize;
1770 }
1771
1772 *pSetLayout = (VkDescriptorSetLayout) set_layout;
1773
1774 return VK_SUCCESS;
1775 }
1776
1777 VkResult anv_CreateDescriptorPool(
1778 VkDevice device,
1779 VkDescriptorPoolUsage poolUsage,
1780 uint32_t maxSets,
1781 const VkDescriptorPoolCreateInfo* pCreateInfo,
1782 VkDescriptorPool* pDescriptorPool)
1783 {
1784 *pDescriptorPool = 1;
1785
1786 return VK_SUCCESS;
1787 }
1788
1789 VkResult anv_ResetDescriptorPool(
1790 VkDevice device,
1791 VkDescriptorPool descriptorPool)
1792 {
1793 return VK_SUCCESS;
1794 }
1795
1796 VkResult anv_AllocDescriptorSets(
1797 VkDevice _device,
1798 VkDescriptorPool descriptorPool,
1799 VkDescriptorSetUsage setUsage,
1800 uint32_t count,
1801 const VkDescriptorSetLayout* pSetLayouts,
1802 VkDescriptorSet* pDescriptorSets,
1803 uint32_t* pCount)
1804 {
1805 struct anv_device *device = (struct anv_device *) _device;
1806 const struct anv_descriptor_set_layout *layout;
1807 struct anv_descriptor_set *set;
1808 size_t size;
1809
1810 for (uint32_t i = 0; i < count; i++) {
1811 layout = (struct anv_descriptor_set_layout *) pSetLayouts[i];
1812 size = sizeof(*set) + layout->count * sizeof(set->descriptors[0]);
1813 set = anv_device_alloc(device, size, 8,
1814 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1815 if (!set) {
1816 *pCount = i;
1817 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1818 }
1819
1820 /* Descriptor sets may not be 100% filled out so we need to memset to
1821 * ensure that we can properly detect and handle holes.
1822 */
1823 memset(set, 0, size);
1824
1825 pDescriptorSets[i] = (VkDescriptorSet) set;
1826 }
1827
1828 *pCount = count;
1829
1830 return VK_SUCCESS;
1831 }
1832
1833 void anv_UpdateDescriptors(
1834 VkDevice _device,
1835 VkDescriptorSet descriptorSet,
1836 uint32_t updateCount,
1837 const void** ppUpdateArray)
1838 {
1839 struct anv_descriptor_set *set = (struct anv_descriptor_set *) descriptorSet;
1840 VkUpdateSamplers *update_samplers;
1841 VkUpdateSamplerTextures *update_sampler_textures;
1842 VkUpdateImages *update_images;
1843 VkUpdateBuffers *update_buffers;
1844 VkUpdateAsCopy *update_as_copy;
1845
1846 for (uint32_t i = 0; i < updateCount; i++) {
1847 const struct anv_common *common = ppUpdateArray[i];
1848
1849 switch (common->sType) {
1850 case VK_STRUCTURE_TYPE_UPDATE_SAMPLERS:
1851 update_samplers = (VkUpdateSamplers *) common;
1852
1853 for (uint32_t j = 0; j < update_samplers->count; j++) {
1854 set->descriptors[update_samplers->binding + j].sampler =
1855 (struct anv_sampler *) update_samplers->pSamplers[j];
1856 }
1857 break;
1858
1859 case VK_STRUCTURE_TYPE_UPDATE_SAMPLER_TEXTURES:
1860 /* FIXME: Shouldn't this be *_UPDATE_SAMPLER_IMAGES? */
1861 update_sampler_textures = (VkUpdateSamplerTextures *) common;
1862
1863 for (uint32_t j = 0; j < update_sampler_textures->count; j++) {
1864 set->descriptors[update_sampler_textures->binding + j].view =
1865 (struct anv_surface_view *)
1866 update_sampler_textures->pSamplerImageViews[j].pImageView->view;
1867 set->descriptors[update_sampler_textures->binding + j].sampler =
1868 (struct anv_sampler *)
1869 update_sampler_textures->pSamplerImageViews[j].sampler;
1870 }
1871 break;
1872
1873 case VK_STRUCTURE_TYPE_UPDATE_IMAGES:
1874 update_images = (VkUpdateImages *) common;
1875
1876 for (uint32_t j = 0; j < update_images->count; j++) {
1877 set->descriptors[update_images->binding + j].view =
1878 (struct anv_surface_view *) update_images->pImageViews[j].view;
1879 }
1880 break;
1881
1882 case VK_STRUCTURE_TYPE_UPDATE_BUFFERS:
1883 update_buffers = (VkUpdateBuffers *) common;
1884
1885 for (uint32_t j = 0; j < update_buffers->count; j++) {
1886 set->descriptors[update_buffers->binding + j].view =
1887 (struct anv_surface_view *) update_buffers->pBufferViews[j].view;
1888 }
1889 /* FIXME: descriptor arrays? */
1890 break;
1891
1892 case VK_STRUCTURE_TYPE_UPDATE_AS_COPY:
1893 update_as_copy = (VkUpdateAsCopy *) common;
1894 (void) update_as_copy;
1895 break;
1896
1897 default:
1898 break;
1899 }
1900 }
1901 }
1902
1903 // State object functions
1904
1905 static inline int64_t
1906 clamp_int64(int64_t x, int64_t min, int64_t max)
1907 {
1908 if (x < min)
1909 return min;
1910 else if (x < max)
1911 return x;
1912 else
1913 return max;
1914 }
1915
1916 static void
1917 anv_dynamic_vp_state_destroy(struct anv_device *device,
1918 struct anv_object *object,
1919 VkObjectType obj_type)
1920 {
1921 struct anv_dynamic_vp_state *state = (void *)object;
1922
1923 assert(obj_type == VK_OBJECT_TYPE_DYNAMIC_VP_STATE);
1924
1925 anv_state_pool_free(&device->dynamic_state_pool, state->sf_clip_vp);
1926 anv_state_pool_free(&device->dynamic_state_pool, state->cc_vp);
1927 anv_state_pool_free(&device->dynamic_state_pool, state->scissor);
1928
1929 anv_device_free(device, state);
1930 }
1931
1932 VkResult anv_CreateDynamicViewportState(
1933 VkDevice _device,
1934 const VkDynamicVpStateCreateInfo* pCreateInfo,
1935 VkDynamicVpState* pState)
1936 {
1937 struct anv_device *device = (struct anv_device *) _device;
1938 struct anv_dynamic_vp_state *state;
1939
1940 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO);
1941
1942 state = anv_device_alloc(device, sizeof(*state), 8,
1943 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1944 if (state == NULL)
1945 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1946
1947 state->base.destructor = anv_dynamic_vp_state_destroy;
1948
1949 unsigned count = pCreateInfo->viewportAndScissorCount;
1950 state->sf_clip_vp = anv_state_pool_alloc(&device->dynamic_state_pool,
1951 count * 64, 64);
1952 state->cc_vp = anv_state_pool_alloc(&device->dynamic_state_pool,
1953 count * 8, 32);
1954 state->scissor = anv_state_pool_alloc(&device->dynamic_state_pool,
1955 count * 32, 32);
1956
1957 for (uint32_t i = 0; i < pCreateInfo->viewportAndScissorCount; i++) {
1958 const VkViewport *vp = &pCreateInfo->pViewports[i];
1959 const VkRect2D *s = &pCreateInfo->pScissors[i];
1960
1961 struct GEN8_SF_CLIP_VIEWPORT sf_clip_viewport = {
1962 .ViewportMatrixElementm00 = vp->width / 2,
1963 .ViewportMatrixElementm11 = vp->height / 2,
1964 .ViewportMatrixElementm22 = (vp->maxDepth - vp->minDepth) / 2,
1965 .ViewportMatrixElementm30 = vp->originX + vp->width / 2,
1966 .ViewportMatrixElementm31 = vp->originY + vp->height / 2,
1967 .ViewportMatrixElementm32 = (vp->maxDepth + vp->minDepth) / 2,
1968 .XMinClipGuardband = -1.0f,
1969 .XMaxClipGuardband = 1.0f,
1970 .YMinClipGuardband = -1.0f,
1971 .YMaxClipGuardband = 1.0f,
1972 .XMinViewPort = vp->originX,
1973 .XMaxViewPort = vp->originX + vp->width - 1,
1974 .YMinViewPort = vp->originY,
1975 .YMaxViewPort = vp->originY + vp->height - 1,
1976 };
1977
1978 struct GEN8_CC_VIEWPORT cc_viewport = {
1979 .MinimumDepth = vp->minDepth,
1980 .MaximumDepth = vp->maxDepth
1981 };
1982
1983 /* Since xmax and ymax are inclusive, we have to have xmax < xmin or
1984 * ymax < ymin for empty clips. In case clip x, y, width height are all
1985 * 0, the clamps below produce 0 for xmin, ymin, xmax, ymax, which isn't
1986 * what we want. Just special case empty clips and produce a canonical
1987 * empty clip. */
1988 static const struct GEN8_SCISSOR_RECT empty_scissor = {
1989 .ScissorRectangleYMin = 1,
1990 .ScissorRectangleXMin = 1,
1991 .ScissorRectangleYMax = 0,
1992 .ScissorRectangleXMax = 0
1993 };
1994
1995 const int max = 0xffff;
1996 struct GEN8_SCISSOR_RECT scissor = {
1997 /* Do this math using int64_t so overflow gets clamped correctly. */
1998 .ScissorRectangleYMin = clamp_int64(s->offset.y, 0, max),
1999 .ScissorRectangleXMin = clamp_int64(s->offset.x, 0, max),
2000 .ScissorRectangleYMax = clamp_int64((uint64_t) s->offset.y + s->extent.height - 1, 0, max),
2001 .ScissorRectangleXMax = clamp_int64((uint64_t) s->offset.x + s->extent.width - 1, 0, max)
2002 };
2003
2004 GEN8_SF_CLIP_VIEWPORT_pack(NULL, state->sf_clip_vp.map + i * 64, &sf_clip_viewport);
2005 GEN8_CC_VIEWPORT_pack(NULL, state->cc_vp.map + i * 32, &cc_viewport);
2006
2007 if (s->extent.width <= 0 || s->extent.height <= 0) {
2008 GEN8_SCISSOR_RECT_pack(NULL, state->scissor.map + i * 32, &empty_scissor);
2009 } else {
2010 GEN8_SCISSOR_RECT_pack(NULL, state->scissor.map + i * 32, &scissor);
2011 }
2012 }
2013
2014 *pState = (VkDynamicVpState) state;
2015
2016 return VK_SUCCESS;
2017 }
2018
2019 VkResult anv_CreateDynamicRasterState(
2020 VkDevice _device,
2021 const VkDynamicRsStateCreateInfo* pCreateInfo,
2022 VkDynamicRsState* pState)
2023 {
2024 struct anv_device *device = (struct anv_device *) _device;
2025 struct anv_dynamic_rs_state *state;
2026
2027 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_RS_STATE_CREATE_INFO);
2028
2029 state = anv_device_alloc(device, sizeof(*state), 8,
2030 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2031 if (state == NULL)
2032 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2033
2034 struct GEN8_3DSTATE_SF sf = {
2035 GEN8_3DSTATE_SF_header,
2036 .LineWidth = pCreateInfo->lineWidth,
2037 .PointWidth = 1.0,
2038 };
2039
2040 GEN8_3DSTATE_SF_pack(NULL, state->state_sf, &sf);
2041
2042 bool enable_bias = pCreateInfo->depthBias != 0.0f ||
2043 pCreateInfo->slopeScaledDepthBias != 0.0f;
2044 struct GEN8_3DSTATE_RASTER raster = {
2045 .GlobalDepthOffsetEnableSolid = enable_bias,
2046 .GlobalDepthOffsetEnableWireframe = enable_bias,
2047 .GlobalDepthOffsetEnablePoint = enable_bias,
2048 .GlobalDepthOffsetConstant = pCreateInfo->depthBias,
2049 .GlobalDepthOffsetScale = pCreateInfo->slopeScaledDepthBias,
2050 .GlobalDepthOffsetClamp = pCreateInfo->depthBiasClamp
2051 };
2052
2053 GEN8_3DSTATE_RASTER_pack(NULL, state->state_raster, &raster);
2054
2055 *pState = (VkDynamicRsState) state;
2056
2057 return VK_SUCCESS;
2058 }
2059
2060 VkResult anv_CreateDynamicColorBlendState(
2061 VkDevice _device,
2062 const VkDynamicCbStateCreateInfo* pCreateInfo,
2063 VkDynamicCbState* pState)
2064 {
2065 struct anv_device *device = (struct anv_device *) _device;
2066 struct anv_dynamic_cb_state *state;
2067
2068 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_CB_STATE_CREATE_INFO);
2069
2070 state = anv_device_alloc(device, sizeof(*state), 8,
2071 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2072 if (state == NULL)
2073 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2074
2075 struct GEN8_COLOR_CALC_STATE color_calc_state = {
2076 .BlendConstantColorRed = pCreateInfo->blendConst[0],
2077 .BlendConstantColorGreen = pCreateInfo->blendConst[1],
2078 .BlendConstantColorBlue = pCreateInfo->blendConst[2],
2079 .BlendConstantColorAlpha = pCreateInfo->blendConst[3]
2080 };
2081
2082 GEN8_COLOR_CALC_STATE_pack(NULL, state->state_color_calc, &color_calc_state);
2083
2084 *pState = (VkDynamicCbState) state;
2085
2086 return VK_SUCCESS;
2087 }
2088
2089 VkResult anv_CreateDynamicDepthStencilState(
2090 VkDevice _device,
2091 const VkDynamicDsStateCreateInfo* pCreateInfo,
2092 VkDynamicDsState* pState)
2093 {
2094 struct anv_device *device = (struct anv_device *) _device;
2095 struct anv_dynamic_ds_state *state;
2096
2097 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_DS_STATE_CREATE_INFO);
2098
2099 state = anv_device_alloc(device, sizeof(*state), 8,
2100 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2101 if (state == NULL)
2102 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2103
2104 struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
2105 GEN8_3DSTATE_WM_DEPTH_STENCIL_header,
2106
2107 /* Is this what we need to do? */
2108 .StencilBufferWriteEnable = pCreateInfo->stencilWriteMask != 0,
2109
2110 .StencilTestMask = pCreateInfo->stencilReadMask & 0xff,
2111 .StencilWriteMask = pCreateInfo->stencilWriteMask & 0xff,
2112
2113 .BackfaceStencilTestMask = pCreateInfo->stencilReadMask & 0xff,
2114 .BackfaceStencilWriteMask = pCreateInfo->stencilWriteMask & 0xff,
2115 };
2116
2117 GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, state->state_wm_depth_stencil,
2118 &wm_depth_stencil);
2119
2120 struct GEN8_COLOR_CALC_STATE color_calc_state = {
2121 .StencilReferenceValue = pCreateInfo->stencilFrontRef,
2122 .BackFaceStencilReferenceValue = pCreateInfo->stencilBackRef
2123 };
2124
2125 GEN8_COLOR_CALC_STATE_pack(NULL, state->state_color_calc, &color_calc_state);
2126
2127 *pState = (VkDynamicDsState) state;
2128
2129 return VK_SUCCESS;
2130 }
2131
2132 // Command buffer functions
2133
2134 static void
2135 anv_cmd_buffer_destroy(struct anv_device *device,
2136 struct anv_object *object,
2137 VkObjectType obj_type)
2138 {
2139 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) object;
2140
2141 assert(obj_type == VK_OBJECT_TYPE_COMMAND_BUFFER);
2142
2143 /* Destroy all of the batch buffers */
2144 struct anv_batch_bo *bbo = cmd_buffer->last_batch_bo;
2145 while (bbo) {
2146 struct anv_batch_bo *prev = bbo->prev_batch_bo;
2147 anv_batch_bo_destroy(bbo, device);
2148 bbo = prev;
2149 }
2150 anv_reloc_list_finish(&cmd_buffer->batch.relocs, device);
2151
2152 /* Destroy all of the surface state buffers */
2153 bbo = cmd_buffer->surface_batch_bo;
2154 while (bbo) {
2155 struct anv_batch_bo *prev = bbo->prev_batch_bo;
2156 anv_batch_bo_destroy(bbo, device);
2157 bbo = prev;
2158 }
2159 anv_reloc_list_finish(&cmd_buffer->surface_relocs, device);
2160
2161 anv_state_stream_finish(&cmd_buffer->surface_state_stream);
2162 anv_state_stream_finish(&cmd_buffer->dynamic_state_stream);
2163 anv_device_free(device, cmd_buffer->exec2_objects);
2164 anv_device_free(device, cmd_buffer->exec2_bos);
2165 anv_device_free(device, cmd_buffer);
2166 }
2167
2168 static VkResult
2169 anv_cmd_buffer_chain_batch(struct anv_batch *batch, void *_data)
2170 {
2171 struct anv_cmd_buffer *cmd_buffer = _data;
2172
2173 struct anv_batch_bo *new_bbo, *old_bbo = cmd_buffer->last_batch_bo;
2174
2175 VkResult result = anv_batch_bo_create(cmd_buffer->device, &new_bbo);
2176 if (result != VK_SUCCESS)
2177 return result;
2178
2179 /* We set the end of the batch a little short so we would be sure we
2180 * have room for the chaining command. Since we're about to emit the
2181 * chaining command, let's set it back where it should go.
2182 */
2183 batch->end += GEN8_MI_BATCH_BUFFER_START_length * 4;
2184 assert(batch->end == old_bbo->bo.map + old_bbo->bo.size);
2185
2186 anv_batch_emit(batch, GEN8_MI_BATCH_BUFFER_START,
2187 GEN8_MI_BATCH_BUFFER_START_header,
2188 ._2ndLevelBatchBuffer = _1stlevelbatch,
2189 .AddressSpaceIndicator = ASI_PPGTT,
2190 .BatchBufferStartAddress = { &new_bbo->bo, 0 },
2191 );
2192
2193 /* Pad out to a 2-dword aligned boundary with zeros */
2194 if ((uintptr_t)batch->next % 8 != 0) {
2195 *(uint32_t *)batch->next = 0;
2196 batch->next += 4;
2197 }
2198
2199 anv_batch_bo_finish(cmd_buffer->last_batch_bo, batch);
2200
2201 new_bbo->prev_batch_bo = old_bbo;
2202 cmd_buffer->last_batch_bo = new_bbo;
2203
2204 anv_batch_bo_start(new_bbo, batch, GEN8_MI_BATCH_BUFFER_START_length * 4);
2205
2206 return VK_SUCCESS;
2207 }
2208
2209 VkResult anv_CreateCommandBuffer(
2210 VkDevice _device,
2211 const VkCmdBufferCreateInfo* pCreateInfo,
2212 VkCmdBuffer* pCmdBuffer)
2213 {
2214 struct anv_device *device = (struct anv_device *) _device;
2215 struct anv_cmd_buffer *cmd_buffer;
2216 VkResult result;
2217
2218 cmd_buffer = anv_device_alloc(device, sizeof(*cmd_buffer), 8,
2219 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2220 if (cmd_buffer == NULL)
2221 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2222
2223 cmd_buffer->base.destructor = anv_cmd_buffer_destroy;
2224
2225 cmd_buffer->device = device;
2226 cmd_buffer->rs_state = NULL;
2227 cmd_buffer->vp_state = NULL;
2228 cmd_buffer->cb_state = NULL;
2229 cmd_buffer->ds_state = NULL;
2230 memset(&cmd_buffer->state_vf, 0, sizeof(cmd_buffer->state_vf));
2231 memset(&cmd_buffer->descriptors, 0, sizeof(cmd_buffer->descriptors));
2232
2233 result = anv_batch_bo_create(device, &cmd_buffer->last_batch_bo);
2234 if (result != VK_SUCCESS)
2235 goto fail;
2236
2237 result = anv_reloc_list_init(&cmd_buffer->batch.relocs, device);
2238 if (result != VK_SUCCESS)
2239 goto fail_batch_bo;
2240
2241 cmd_buffer->batch.device = device;
2242 cmd_buffer->batch.extend_cb = anv_cmd_buffer_chain_batch;
2243 cmd_buffer->batch.user_data = cmd_buffer;
2244
2245 anv_batch_bo_start(cmd_buffer->last_batch_bo, &cmd_buffer->batch,
2246 GEN8_MI_BATCH_BUFFER_START_length * 4);
2247
2248 result = anv_batch_bo_create(device, &cmd_buffer->surface_batch_bo);
2249 if (result != VK_SUCCESS)
2250 goto fail_batch_relocs;
2251 cmd_buffer->surface_batch_bo->first_reloc = 0;
2252
2253 result = anv_reloc_list_init(&cmd_buffer->surface_relocs, device);
2254 if (result != VK_SUCCESS)
2255 goto fail_ss_batch_bo;
2256
2257 /* Start surface_next at 1 so surface offset 0 is invalid. */
2258 cmd_buffer->surface_next = 1;
2259
2260 cmd_buffer->exec2_objects = NULL;
2261 cmd_buffer->exec2_bos = NULL;
2262 cmd_buffer->exec2_array_length = 0;
2263
2264 anv_state_stream_init(&cmd_buffer->surface_state_stream,
2265 &device->surface_state_block_pool);
2266 anv_state_stream_init(&cmd_buffer->dynamic_state_stream,
2267 &device->dynamic_state_block_pool);
2268
2269 cmd_buffer->dirty = 0;
2270 cmd_buffer->vb_dirty = 0;
2271 cmd_buffer->descriptors_dirty = 0;
2272 cmd_buffer->pipeline = NULL;
2273 cmd_buffer->vp_state = NULL;
2274 cmd_buffer->rs_state = NULL;
2275 cmd_buffer->ds_state = NULL;
2276
2277 *pCmdBuffer = (VkCmdBuffer) cmd_buffer;
2278
2279 return VK_SUCCESS;
2280
2281 fail_ss_batch_bo:
2282 anv_batch_bo_destroy(cmd_buffer->surface_batch_bo, device);
2283 fail_batch_relocs:
2284 anv_reloc_list_finish(&cmd_buffer->batch.relocs, device);
2285 fail_batch_bo:
2286 anv_batch_bo_destroy(cmd_buffer->last_batch_bo, device);
2287 fail:
2288 anv_device_free(device, cmd_buffer);
2289
2290 return result;
2291 }
2292
2293 static void
2294 anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer *cmd_buffer)
2295 {
2296 struct anv_device *device = cmd_buffer->device;
2297 struct anv_bo *scratch_bo = NULL;
2298
2299 cmd_buffer->scratch_size = device->scratch_block_pool.size;
2300 if (cmd_buffer->scratch_size > 0)
2301 scratch_bo = &device->scratch_block_pool.bo;
2302
2303 anv_batch_emit(&cmd_buffer->batch, GEN8_STATE_BASE_ADDRESS,
2304 .GeneralStateBaseAddress = { scratch_bo, 0 },
2305 .GeneralStateMemoryObjectControlState = GEN8_MOCS,
2306 .GeneralStateBaseAddressModifyEnable = true,
2307 .GeneralStateBufferSize = 0xfffff,
2308 .GeneralStateBufferSizeModifyEnable = true,
2309
2310 .SurfaceStateBaseAddress = { &cmd_buffer->surface_batch_bo->bo, 0 },
2311 .SurfaceStateMemoryObjectControlState = GEN8_MOCS,
2312 .SurfaceStateBaseAddressModifyEnable = true,
2313
2314 .DynamicStateBaseAddress = { &device->dynamic_state_block_pool.bo, 0 },
2315 .DynamicStateMemoryObjectControlState = GEN8_MOCS,
2316 .DynamicStateBaseAddressModifyEnable = true,
2317 .DynamicStateBufferSize = 0xfffff,
2318 .DynamicStateBufferSizeModifyEnable = true,
2319
2320 .IndirectObjectBaseAddress = { NULL, 0 },
2321 .IndirectObjectMemoryObjectControlState = GEN8_MOCS,
2322 .IndirectObjectBaseAddressModifyEnable = true,
2323 .IndirectObjectBufferSize = 0xfffff,
2324 .IndirectObjectBufferSizeModifyEnable = true,
2325
2326 .InstructionBaseAddress = { &device->instruction_block_pool.bo, 0 },
2327 .InstructionMemoryObjectControlState = GEN8_MOCS,
2328 .InstructionBaseAddressModifyEnable = true,
2329 .InstructionBufferSize = 0xfffff,
2330 .InstructionBuffersizeModifyEnable = true);
2331 }
2332
2333 VkResult anv_BeginCommandBuffer(
2334 VkCmdBuffer cmdBuffer,
2335 const VkCmdBufferBeginInfo* pBeginInfo)
2336 {
2337 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2338
2339 anv_cmd_buffer_emit_state_base_address(cmd_buffer);
2340 cmd_buffer->current_pipeline = UINT32_MAX;
2341
2342 return VK_SUCCESS;
2343 }
2344
2345 static VkResult
2346 anv_cmd_buffer_add_bo(struct anv_cmd_buffer *cmd_buffer,
2347 struct anv_bo *bo,
2348 struct drm_i915_gem_relocation_entry *relocs,
2349 size_t num_relocs)
2350 {
2351 struct drm_i915_gem_exec_object2 *obj;
2352
2353 if (bo->index < cmd_buffer->bo_count &&
2354 cmd_buffer->exec2_bos[bo->index] == bo)
2355 return VK_SUCCESS;
2356
2357 if (cmd_buffer->bo_count >= cmd_buffer->exec2_array_length) {
2358 uint32_t new_len = cmd_buffer->exec2_objects ?
2359 cmd_buffer->exec2_array_length * 2 : 64;
2360
2361 struct drm_i915_gem_exec_object2 *new_objects =
2362 anv_device_alloc(cmd_buffer->device, new_len * sizeof(*new_objects),
2363 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL);
2364 if (new_objects == NULL)
2365 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2366
2367 struct anv_bo **new_bos =
2368 anv_device_alloc(cmd_buffer->device, new_len * sizeof(*new_bos),
2369 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL);
2370 if (new_objects == NULL) {
2371 anv_device_free(cmd_buffer->device, new_objects);
2372 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2373 }
2374
2375 if (cmd_buffer->exec2_objects) {
2376 memcpy(new_objects, cmd_buffer->exec2_objects,
2377 cmd_buffer->bo_count * sizeof(*new_objects));
2378 memcpy(new_bos, cmd_buffer->exec2_bos,
2379 cmd_buffer->bo_count * sizeof(*new_bos));
2380 }
2381
2382 cmd_buffer->exec2_objects = new_objects;
2383 cmd_buffer->exec2_bos = new_bos;
2384 cmd_buffer->exec2_array_length = new_len;
2385 }
2386
2387 assert(cmd_buffer->bo_count < cmd_buffer->exec2_array_length);
2388
2389 bo->index = cmd_buffer->bo_count++;
2390 obj = &cmd_buffer->exec2_objects[bo->index];
2391 cmd_buffer->exec2_bos[bo->index] = bo;
2392
2393 obj->handle = bo->gem_handle;
2394 obj->relocation_count = 0;
2395 obj->relocs_ptr = 0;
2396 obj->alignment = 0;
2397 obj->offset = bo->offset;
2398 obj->flags = 0;
2399 obj->rsvd1 = 0;
2400 obj->rsvd2 = 0;
2401
2402 if (relocs) {
2403 obj->relocation_count = num_relocs;
2404 obj->relocs_ptr = (uintptr_t) relocs;
2405 }
2406
2407 return VK_SUCCESS;
2408 }
2409
2410 static void
2411 anv_cmd_buffer_add_validate_bos(struct anv_cmd_buffer *cmd_buffer,
2412 struct anv_reloc_list *list)
2413 {
2414 for (size_t i = 0; i < list->num_relocs; i++)
2415 anv_cmd_buffer_add_bo(cmd_buffer, list->reloc_bos[i], NULL, 0);
2416 }
2417
2418 static void
2419 anv_cmd_buffer_process_relocs(struct anv_cmd_buffer *cmd_buffer,
2420 struct anv_reloc_list *list)
2421 {
2422 struct anv_bo *bo;
2423
2424 /* If the kernel supports I915_EXEC_NO_RELOC, it will compare offset in
2425 * struct drm_i915_gem_exec_object2 against the bos current offset and if
2426 * all bos haven't moved it will skip relocation processing alltogether.
2427 * If I915_EXEC_NO_RELOC is not supported, the kernel ignores the incoming
2428 * value of offset so we can set it either way. For that to work we need
2429 * to make sure all relocs use the same presumed offset.
2430 */
2431
2432 for (size_t i = 0; i < list->num_relocs; i++) {
2433 bo = list->reloc_bos[i];
2434 if (bo->offset != list->relocs[i].presumed_offset)
2435 cmd_buffer->need_reloc = true;
2436
2437 list->relocs[i].target_handle = bo->index;
2438 }
2439 }
2440
2441 VkResult anv_EndCommandBuffer(
2442 VkCmdBuffer cmdBuffer)
2443 {
2444 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2445 struct anv_device *device = cmd_buffer->device;
2446 struct anv_batch *batch = &cmd_buffer->batch;
2447
2448 anv_batch_emit(batch, GEN8_MI_BATCH_BUFFER_END);
2449
2450 /* Round batch up to an even number of dwords. */
2451 if ((batch->next - batch->start) & 4)
2452 anv_batch_emit(batch, GEN8_MI_NOOP);
2453
2454 anv_batch_bo_finish(cmd_buffer->last_batch_bo, &cmd_buffer->batch);
2455 cmd_buffer->surface_batch_bo->num_relocs =
2456 cmd_buffer->surface_relocs.num_relocs - cmd_buffer->surface_batch_bo->first_reloc;
2457 cmd_buffer->surface_batch_bo->length = cmd_buffer->surface_next;
2458
2459 cmd_buffer->bo_count = 0;
2460 cmd_buffer->need_reloc = false;
2461
2462 /* Lock for access to bo->index. */
2463 pthread_mutex_lock(&device->mutex);
2464
2465 /* Add surface state bos first so we can add them with their relocs. */
2466 for (struct anv_batch_bo *bbo = cmd_buffer->surface_batch_bo;
2467 bbo != NULL; bbo = bbo->prev_batch_bo) {
2468 anv_cmd_buffer_add_bo(cmd_buffer, &bbo->bo,
2469 &cmd_buffer->surface_relocs.relocs[bbo->first_reloc],
2470 bbo->num_relocs);
2471 }
2472
2473 /* Add all of the BOs referenced by surface state */
2474 anv_cmd_buffer_add_validate_bos(cmd_buffer, &cmd_buffer->surface_relocs);
2475
2476 /* Add all but the first batch BO */
2477 struct anv_batch_bo *batch_bo = cmd_buffer->last_batch_bo;
2478 while (batch_bo->prev_batch_bo) {
2479 anv_cmd_buffer_add_bo(cmd_buffer, &batch_bo->bo,
2480 &batch->relocs.relocs[batch_bo->first_reloc],
2481 batch_bo->num_relocs);
2482 batch_bo = batch_bo->prev_batch_bo;
2483 }
2484
2485 /* Add everything referenced by the batches */
2486 anv_cmd_buffer_add_validate_bos(cmd_buffer, &batch->relocs);
2487
2488 /* Add the first batch bo last */
2489 assert(batch_bo->prev_batch_bo == NULL && batch_bo->first_reloc == 0);
2490 anv_cmd_buffer_add_bo(cmd_buffer, &batch_bo->bo,
2491 &batch->relocs.relocs[batch_bo->first_reloc],
2492 batch_bo->num_relocs);
2493 assert(batch_bo->bo.index == cmd_buffer->bo_count - 1);
2494
2495 anv_cmd_buffer_process_relocs(cmd_buffer, &cmd_buffer->surface_relocs);
2496 anv_cmd_buffer_process_relocs(cmd_buffer, &batch->relocs);
2497
2498 cmd_buffer->execbuf.buffers_ptr = (uintptr_t) cmd_buffer->exec2_objects;
2499 cmd_buffer->execbuf.buffer_count = cmd_buffer->bo_count;
2500 cmd_buffer->execbuf.batch_start_offset = 0;
2501 cmd_buffer->execbuf.batch_len = batch->next - batch->start;
2502 cmd_buffer->execbuf.cliprects_ptr = 0;
2503 cmd_buffer->execbuf.num_cliprects = 0;
2504 cmd_buffer->execbuf.DR1 = 0;
2505 cmd_buffer->execbuf.DR4 = 0;
2506
2507 cmd_buffer->execbuf.flags = I915_EXEC_HANDLE_LUT;
2508 if (!cmd_buffer->need_reloc)
2509 cmd_buffer->execbuf.flags |= I915_EXEC_NO_RELOC;
2510 cmd_buffer->execbuf.flags |= I915_EXEC_RENDER;
2511 cmd_buffer->execbuf.rsvd1 = device->context_id;
2512 cmd_buffer->execbuf.rsvd2 = 0;
2513
2514 pthread_mutex_unlock(&device->mutex);
2515
2516 return VK_SUCCESS;
2517 }
2518
2519 VkResult anv_ResetCommandBuffer(
2520 VkCmdBuffer cmdBuffer)
2521 {
2522 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2523
2524 /* Delete all but the first batch bo */
2525 while (cmd_buffer->last_batch_bo->prev_batch_bo) {
2526 struct anv_batch_bo *prev = cmd_buffer->last_batch_bo->prev_batch_bo;
2527 anv_batch_bo_destroy(cmd_buffer->last_batch_bo, cmd_buffer->device);
2528 cmd_buffer->last_batch_bo = prev;
2529 }
2530 assert(cmd_buffer->last_batch_bo->prev_batch_bo == NULL);
2531
2532 cmd_buffer->batch.relocs.num_relocs = 0;
2533 anv_batch_bo_start(cmd_buffer->last_batch_bo, &cmd_buffer->batch,
2534 GEN8_MI_BATCH_BUFFER_START_length * 4);
2535
2536 /* Delete all but the first batch bo */
2537 while (cmd_buffer->surface_batch_bo->prev_batch_bo) {
2538 struct anv_batch_bo *prev = cmd_buffer->surface_batch_bo->prev_batch_bo;
2539 anv_batch_bo_destroy(cmd_buffer->surface_batch_bo, cmd_buffer->device);
2540 cmd_buffer->surface_batch_bo = prev;
2541 }
2542 assert(cmd_buffer->surface_batch_bo->prev_batch_bo == NULL);
2543
2544 cmd_buffer->surface_next = 1;
2545 cmd_buffer->surface_relocs.num_relocs = 0;
2546
2547 cmd_buffer->rs_state = NULL;
2548 cmd_buffer->vp_state = NULL;
2549 cmd_buffer->cb_state = NULL;
2550 cmd_buffer->ds_state = NULL;
2551
2552 return VK_SUCCESS;
2553 }
2554
2555 // Command buffer building functions
2556
2557 void anv_CmdBindPipeline(
2558 VkCmdBuffer cmdBuffer,
2559 VkPipelineBindPoint pipelineBindPoint,
2560 VkPipeline _pipeline)
2561 {
2562 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2563 struct anv_pipeline *pipeline = (struct anv_pipeline *) _pipeline;
2564
2565 switch (pipelineBindPoint) {
2566 case VK_PIPELINE_BIND_POINT_COMPUTE:
2567 cmd_buffer->compute_pipeline = pipeline;
2568 cmd_buffer->compute_dirty |= ANV_CMD_BUFFER_PIPELINE_DIRTY;
2569 break;
2570
2571 case VK_PIPELINE_BIND_POINT_GRAPHICS:
2572 cmd_buffer->pipeline = pipeline;
2573 cmd_buffer->vb_dirty |= pipeline->vb_used;
2574 cmd_buffer->dirty |= ANV_CMD_BUFFER_PIPELINE_DIRTY;
2575 break;
2576
2577 default:
2578 assert(!"invalid bind point");
2579 break;
2580 }
2581 }
2582
2583 void anv_CmdBindDynamicStateObject(
2584 VkCmdBuffer cmdBuffer,
2585 VkStateBindPoint stateBindPoint,
2586 VkDynamicStateObject dynamicState)
2587 {
2588 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2589
2590 switch (stateBindPoint) {
2591 case VK_STATE_BIND_POINT_VIEWPORT:
2592 cmd_buffer->vp_state = (struct anv_dynamic_vp_state *) dynamicState;
2593 cmd_buffer->dirty |= ANV_CMD_BUFFER_VP_DIRTY;
2594 break;
2595 case VK_STATE_BIND_POINT_RASTER:
2596 cmd_buffer->rs_state = (struct anv_dynamic_rs_state *) dynamicState;
2597 cmd_buffer->dirty |= ANV_CMD_BUFFER_RS_DIRTY;
2598 break;
2599 case VK_STATE_BIND_POINT_COLOR_BLEND:
2600 cmd_buffer->cb_state = (struct anv_dynamic_cb_state *) dynamicState;
2601 cmd_buffer->dirty |= ANV_CMD_BUFFER_CB_DIRTY;
2602 break;
2603 case VK_STATE_BIND_POINT_DEPTH_STENCIL:
2604 cmd_buffer->ds_state = (struct anv_dynamic_ds_state *) dynamicState;
2605 cmd_buffer->dirty |= ANV_CMD_BUFFER_DS_DIRTY;
2606 break;
2607 default:
2608 break;
2609 };
2610 }
2611
2612 static struct anv_state
2613 anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer *cmd_buffer,
2614 uint32_t size, uint32_t alignment)
2615 {
2616 struct anv_state state;
2617
2618 state.offset = align_u32(cmd_buffer->surface_next, alignment);
2619 if (state.offset + size > cmd_buffer->surface_batch_bo->bo.size)
2620 return (struct anv_state) { 0 };
2621
2622 state.map = cmd_buffer->surface_batch_bo->bo.map + state.offset;
2623 state.alloc_size = size;
2624 cmd_buffer->surface_next = state.offset + size;
2625
2626 assert(state.offset + size <= cmd_buffer->surface_batch_bo->bo.size);
2627
2628 return state;
2629 }
2630
2631 static VkResult
2632 anv_cmd_buffer_new_surface_state_bo(struct anv_cmd_buffer *cmd_buffer)
2633 {
2634 struct anv_batch_bo *new_bbo, *old_bbo = cmd_buffer->surface_batch_bo;
2635
2636 /* Finish off the old buffer */
2637 old_bbo->num_relocs =
2638 cmd_buffer->surface_relocs.num_relocs - old_bbo->first_reloc;
2639 old_bbo->length = cmd_buffer->surface_next;
2640
2641 VkResult result = anv_batch_bo_create(cmd_buffer->device, &new_bbo);
2642 if (result != VK_SUCCESS)
2643 return result;
2644
2645 new_bbo->first_reloc = cmd_buffer->surface_relocs.num_relocs;
2646 cmd_buffer->surface_next = 1;
2647
2648 new_bbo->prev_batch_bo = old_bbo;
2649 cmd_buffer->surface_batch_bo = new_bbo;
2650
2651 /* Re-emit state base addresses so we get the new surface state base
2652 * address before we start emitting binding tables etc.
2653 */
2654 anv_cmd_buffer_emit_state_base_address(cmd_buffer);
2655
2656 /* It seems like just changing the state base addresses isn't enough.
2657 * Invalidating the cache seems to be enough to cause things to
2658 * propagate. However, I'm not 100% sure what we're supposed to do.
2659 */
2660 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
2661 .TextureCacheInvalidationEnable = true);
2662
2663 return VK_SUCCESS;
2664 }
2665
2666 void anv_CmdBindDescriptorSets(
2667 VkCmdBuffer cmdBuffer,
2668 VkPipelineBindPoint pipelineBindPoint,
2669 uint32_t firstSet,
2670 uint32_t setCount,
2671 const VkDescriptorSet* pDescriptorSets,
2672 uint32_t dynamicOffsetCount,
2673 const uint32_t* pDynamicOffsets)
2674 {
2675 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2676 struct anv_pipeline_layout *layout;
2677 struct anv_descriptor_set *set;
2678 struct anv_descriptor_set_layout *set_layout;
2679
2680 assert(firstSet + setCount < MAX_SETS);
2681
2682 if (pipelineBindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS)
2683 layout = cmd_buffer->pipeline->layout;
2684 else
2685 layout = cmd_buffer->compute_pipeline->layout;
2686
2687 uint32_t dynamic_slot = 0;
2688 for (uint32_t i = 0; i < setCount; i++) {
2689 set = (struct anv_descriptor_set *) pDescriptorSets[i];
2690 set_layout = layout->set[firstSet + i].layout;
2691
2692 cmd_buffer->descriptors[firstSet + i].set = set;
2693
2694 assert(set_layout->num_dynamic_buffers <
2695 ARRAY_SIZE(cmd_buffer->descriptors[0].dynamic_offsets));
2696 memcpy(cmd_buffer->descriptors[firstSet + i].dynamic_offsets,
2697 pDynamicOffsets + dynamic_slot,
2698 set_layout->num_dynamic_buffers * sizeof(*pDynamicOffsets));
2699
2700 cmd_buffer->descriptors_dirty |= set_layout->shader_stages;
2701
2702 dynamic_slot += set_layout->num_dynamic_buffers;
2703 }
2704 }
2705
2706 void anv_CmdBindIndexBuffer(
2707 VkCmdBuffer cmdBuffer,
2708 VkBuffer _buffer,
2709 VkDeviceSize offset,
2710 VkIndexType indexType)
2711 {
2712 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2713 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
2714
2715 static const uint32_t vk_to_gen_index_type[] = {
2716 [VK_INDEX_TYPE_UINT16] = INDEX_WORD,
2717 [VK_INDEX_TYPE_UINT32] = INDEX_DWORD,
2718 };
2719
2720 struct GEN8_3DSTATE_VF vf = {
2721 GEN8_3DSTATE_VF_header,
2722 .CutIndex = (indexType == VK_INDEX_TYPE_UINT16) ? UINT16_MAX : UINT32_MAX,
2723 };
2724 GEN8_3DSTATE_VF_pack(NULL, cmd_buffer->state_vf, &vf);
2725
2726 cmd_buffer->dirty |= ANV_CMD_BUFFER_INDEX_BUFFER_DIRTY;
2727
2728 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_INDEX_BUFFER,
2729 .IndexFormat = vk_to_gen_index_type[indexType],
2730 .MemoryObjectControlState = GEN8_MOCS,
2731 .BufferStartingAddress = { buffer->bo, buffer->offset + offset },
2732 .BufferSize = buffer->size - offset);
2733 }
2734
2735 void anv_CmdBindVertexBuffers(
2736 VkCmdBuffer cmdBuffer,
2737 uint32_t startBinding,
2738 uint32_t bindingCount,
2739 const VkBuffer* pBuffers,
2740 const VkDeviceSize* pOffsets)
2741 {
2742 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2743 struct anv_vertex_binding *vb = cmd_buffer->vertex_bindings;
2744
2745 /* We have to defer setting up vertex buffer since we need the buffer
2746 * stride from the pipeline. */
2747
2748 assert(startBinding + bindingCount < MAX_VBS);
2749 for (uint32_t i = 0; i < bindingCount; i++) {
2750 vb[startBinding + i].buffer = (struct anv_buffer *) pBuffers[i];
2751 vb[startBinding + i].offset = pOffsets[i];
2752 cmd_buffer->vb_dirty |= 1 << (startBinding + i);
2753 }
2754 }
2755
2756 static VkResult
2757 cmd_buffer_emit_binding_table(struct anv_cmd_buffer *cmd_buffer,
2758 unsigned stage, struct anv_state *bt_state)
2759 {
2760 struct anv_pipeline_layout *layout;
2761 uint32_t color_attachments, bias, size;
2762
2763 if (stage == VK_SHADER_STAGE_COMPUTE)
2764 layout = cmd_buffer->compute_pipeline->layout;
2765 else
2766 layout = cmd_buffer->pipeline->layout;
2767
2768 if (stage == VK_SHADER_STAGE_FRAGMENT) {
2769 bias = MAX_RTS;
2770 color_attachments = cmd_buffer->framebuffer->color_attachment_count;
2771 } else {
2772 bias = 0;
2773 color_attachments = 0;
2774 }
2775
2776 /* This is a little awkward: layout can be NULL but we still have to
2777 * allocate and set a binding table for the PS stage for render
2778 * targets. */
2779 uint32_t surface_count = layout ? layout->stage[stage].surface_count : 0;
2780
2781 if (color_attachments + surface_count == 0)
2782 return VK_SUCCESS;
2783
2784 size = (bias + surface_count) * sizeof(uint32_t);
2785 *bt_state = anv_cmd_buffer_alloc_surface_state(cmd_buffer, size, 32);
2786 uint32_t *bt_map = bt_state->map;
2787
2788 if (bt_state->map == NULL)
2789 return VK_ERROR_OUT_OF_DEVICE_MEMORY;
2790
2791 for (uint32_t ca = 0; ca < color_attachments; ca++) {
2792 const struct anv_surface_view *view =
2793 cmd_buffer->framebuffer->color_attachments[ca];
2794
2795 struct anv_state state =
2796 anv_cmd_buffer_alloc_surface_state(cmd_buffer, 64, 64);
2797
2798 if (state.map == NULL)
2799 return VK_ERROR_OUT_OF_DEVICE_MEMORY;
2800
2801 memcpy(state.map, view->surface_state.map, 64);
2802
2803 /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
2804 *(uint64_t *)(state.map + 8 * 4) =
2805 anv_reloc_list_add(&cmd_buffer->surface_relocs,
2806 cmd_buffer->device,
2807 state.offset + 8 * 4,
2808 view->bo, view->offset);
2809
2810 bt_map[ca] = state.offset;
2811 }
2812
2813 if (layout == NULL)
2814 return VK_SUCCESS;
2815
2816 for (uint32_t set = 0; set < layout->num_sets; set++) {
2817 struct anv_descriptor_set_binding *d = &cmd_buffer->descriptors[set];
2818 struct anv_descriptor_set_layout *set_layout = layout->set[set].layout;
2819 struct anv_descriptor_slot *surface_slots =
2820 set_layout->stage[stage].surface_start;
2821
2822 uint32_t start = bias + layout->set[set].surface_start[stage];
2823
2824 for (uint32_t b = 0; b < set_layout->stage[stage].surface_count; b++) {
2825 struct anv_surface_view *view =
2826 d->set->descriptors[surface_slots[b].index].view;
2827
2828 if (!view)
2829 continue;
2830
2831 struct anv_state state =
2832 anv_cmd_buffer_alloc_surface_state(cmd_buffer, 64, 64);
2833
2834 if (state.map == NULL)
2835 return VK_ERROR_OUT_OF_DEVICE_MEMORY;
2836
2837 uint32_t offset;
2838 if (surface_slots[b].dynamic_slot >= 0) {
2839 uint32_t dynamic_offset =
2840 d->dynamic_offsets[surface_slots[b].dynamic_slot];
2841
2842 offset = view->offset + dynamic_offset;
2843 fill_buffer_surface_state(state.map, view->format, offset,
2844 view->range - dynamic_offset);
2845 } else {
2846 offset = view->offset;
2847 memcpy(state.map, view->surface_state.map, 64);
2848 }
2849
2850 /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
2851 *(uint64_t *)(state.map + 8 * 4) =
2852 anv_reloc_list_add(&cmd_buffer->surface_relocs,
2853 cmd_buffer->device,
2854 state.offset + 8 * 4,
2855 view->bo, offset);
2856
2857 bt_map[start + b] = state.offset;
2858 }
2859 }
2860
2861 return VK_SUCCESS;
2862 }
2863
2864 static VkResult
2865 cmd_buffer_emit_samplers(struct anv_cmd_buffer *cmd_buffer,
2866 unsigned stage, struct anv_state *state)
2867 {
2868 struct anv_pipeline_layout *layout;
2869 uint32_t sampler_count;
2870
2871 if (stage == VK_SHADER_STAGE_COMPUTE)
2872 layout = cmd_buffer->compute_pipeline->layout;
2873 else
2874 layout = cmd_buffer->pipeline->layout;
2875
2876 sampler_count = layout ? layout->stage[stage].sampler_count : 0;
2877 if (sampler_count == 0)
2878 return VK_SUCCESS;
2879
2880 uint32_t size = sampler_count * 16;
2881 *state = anv_state_stream_alloc(&cmd_buffer->dynamic_state_stream, size, 32);
2882
2883 if (state->map == NULL)
2884 return VK_ERROR_OUT_OF_DEVICE_MEMORY;
2885
2886 for (uint32_t set = 0; set < layout->num_sets; set++) {
2887 struct anv_descriptor_set_binding *d = &cmd_buffer->descriptors[set];
2888 struct anv_descriptor_set_layout *set_layout = layout->set[set].layout;
2889 struct anv_descriptor_slot *sampler_slots =
2890 set_layout->stage[stage].sampler_start;
2891
2892 uint32_t start = layout->set[set].sampler_start[stage];
2893
2894 for (uint32_t b = 0; b < set_layout->stage[stage].sampler_count; b++) {
2895 struct anv_sampler *sampler =
2896 d->set->descriptors[sampler_slots[b].index].sampler;
2897
2898 if (!sampler)
2899 continue;
2900
2901 memcpy(state->map + (start + b) * 16,
2902 sampler->state, sizeof(sampler->state));
2903 }
2904 }
2905
2906 return VK_SUCCESS;
2907 }
2908
2909 static VkResult
2910 flush_descriptor_set(struct anv_cmd_buffer *cmd_buffer, uint32_t stage)
2911 {
2912 struct anv_state surfaces = { 0, }, samplers = { 0, };
2913 VkResult result;
2914
2915 result = cmd_buffer_emit_samplers(cmd_buffer, stage, &samplers);
2916 if (result != VK_SUCCESS)
2917 return result;
2918 result = cmd_buffer_emit_binding_table(cmd_buffer, stage, &surfaces);
2919 if (result != VK_SUCCESS)
2920 return result;
2921
2922 static const uint32_t sampler_state_opcodes[] = {
2923 [VK_SHADER_STAGE_VERTEX] = 43,
2924 [VK_SHADER_STAGE_TESS_CONTROL] = 44, /* HS */
2925 [VK_SHADER_STAGE_TESS_EVALUATION] = 45, /* DS */
2926 [VK_SHADER_STAGE_GEOMETRY] = 46,
2927 [VK_SHADER_STAGE_FRAGMENT] = 47,
2928 [VK_SHADER_STAGE_COMPUTE] = 0,
2929 };
2930
2931 static const uint32_t binding_table_opcodes[] = {
2932 [VK_SHADER_STAGE_VERTEX] = 38,
2933 [VK_SHADER_STAGE_TESS_CONTROL] = 39,
2934 [VK_SHADER_STAGE_TESS_EVALUATION] = 40,
2935 [VK_SHADER_STAGE_GEOMETRY] = 41,
2936 [VK_SHADER_STAGE_FRAGMENT] = 42,
2937 [VK_SHADER_STAGE_COMPUTE] = 0,
2938 };
2939
2940 if (samplers.alloc_size > 0) {
2941 anv_batch_emit(&cmd_buffer->batch,
2942 GEN8_3DSTATE_SAMPLER_STATE_POINTERS_VS,
2943 ._3DCommandSubOpcode = sampler_state_opcodes[stage],
2944 .PointertoVSSamplerState = samplers.offset);
2945 }
2946
2947 if (surfaces.alloc_size > 0) {
2948 anv_batch_emit(&cmd_buffer->batch,
2949 GEN8_3DSTATE_BINDING_TABLE_POINTERS_VS,
2950 ._3DCommandSubOpcode = binding_table_opcodes[stage],
2951 .PointertoVSBindingTable = surfaces.offset);
2952 }
2953
2954 return VK_SUCCESS;
2955 }
2956
2957 static void
2958 flush_descriptor_sets(struct anv_cmd_buffer *cmd_buffer)
2959 {
2960 uint32_t s, dirty = cmd_buffer->descriptors_dirty &
2961 cmd_buffer->pipeline->active_stages;
2962
2963 VkResult result;
2964 for_each_bit(s, dirty) {
2965 result = flush_descriptor_set(cmd_buffer, s);
2966 if (result != VK_SUCCESS)
2967 break;
2968 }
2969
2970 if (result != VK_SUCCESS) {
2971 assert(result == VK_ERROR_OUT_OF_DEVICE_MEMORY);
2972
2973 result = anv_cmd_buffer_new_surface_state_bo(cmd_buffer);
2974 assert(result == VK_SUCCESS);
2975
2976 /* Re-emit all active binding tables */
2977 for_each_bit(s, cmd_buffer->pipeline->active_stages) {
2978 result = flush_descriptor_set(cmd_buffer, s);
2979
2980 /* It had better succeed this time */
2981 assert(result == VK_SUCCESS);
2982 }
2983 }
2984
2985 cmd_buffer->descriptors_dirty &= ~cmd_buffer->pipeline->active_stages;
2986 }
2987
2988 static struct anv_state
2989 anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer *cmd_buffer,
2990 uint32_t *a, uint32_t dwords, uint32_t alignment)
2991 {
2992 struct anv_state state;
2993
2994 state = anv_state_stream_alloc(&cmd_buffer->dynamic_state_stream,
2995 dwords * 4, alignment);
2996 memcpy(state.map, a, dwords * 4);
2997
2998 VG(VALGRIND_CHECK_MEM_IS_DEFINED(state.map, dwords * 4));
2999
3000 return state;
3001 }
3002
3003 static struct anv_state
3004 anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer *cmd_buffer,
3005 uint32_t *a, uint32_t *b,
3006 uint32_t dwords, uint32_t alignment)
3007 {
3008 struct anv_state state;
3009 uint32_t *p;
3010
3011 state = anv_state_stream_alloc(&cmd_buffer->dynamic_state_stream,
3012 dwords * 4, alignment);
3013 p = state.map;
3014 for (uint32_t i = 0; i < dwords; i++)
3015 p[i] = a[i] | b[i];
3016
3017 VG(VALGRIND_CHECK_MEM_IS_DEFINED(p, dwords * 4));
3018
3019 return state;
3020 }
3021
3022 static VkResult
3023 flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
3024 {
3025 struct anv_device *device = cmd_buffer->device;
3026 struct anv_pipeline *pipeline = cmd_buffer->compute_pipeline;
3027 struct anv_state surfaces = { 0, }, samplers = { 0, };
3028 VkResult result;
3029
3030 result = cmd_buffer_emit_samplers(cmd_buffer,
3031 VK_SHADER_STAGE_COMPUTE, &samplers);
3032 if (result != VK_SUCCESS)
3033 return result;
3034 result = cmd_buffer_emit_binding_table(cmd_buffer,
3035 VK_SHADER_STAGE_COMPUTE, &surfaces);
3036 if (result != VK_SUCCESS)
3037 return result;
3038
3039 struct GEN8_INTERFACE_DESCRIPTOR_DATA desc = {
3040 .KernelStartPointer = pipeline->cs_simd,
3041 .KernelStartPointerHigh = 0,
3042 .BindingTablePointer = surfaces.offset,
3043 .BindingTableEntryCount = 0,
3044 .SamplerStatePointer = samplers.offset,
3045 .SamplerCount = 0,
3046 .NumberofThreadsinGPGPUThreadGroup = 0 /* FIXME: Really? */
3047 };
3048
3049 uint32_t size = GEN8_INTERFACE_DESCRIPTOR_DATA_length * sizeof(uint32_t);
3050 struct anv_state state =
3051 anv_state_pool_alloc(&device->dynamic_state_pool, size, 64);
3052
3053 GEN8_INTERFACE_DESCRIPTOR_DATA_pack(NULL, state.map, &desc);
3054
3055 anv_batch_emit(&cmd_buffer->batch, GEN8_MEDIA_INTERFACE_DESCRIPTOR_LOAD,
3056 .InterfaceDescriptorTotalLength = size,
3057 .InterfaceDescriptorDataStartAddress = state.offset);
3058
3059 return VK_SUCCESS;
3060 }
3061
3062 static void
3063 anv_cmd_buffer_flush_compute_state(struct anv_cmd_buffer *cmd_buffer)
3064 {
3065 struct anv_pipeline *pipeline = cmd_buffer->compute_pipeline;
3066 VkResult result;
3067
3068 assert(pipeline->active_stages == VK_SHADER_STAGE_COMPUTE_BIT);
3069
3070 if (cmd_buffer->current_pipeline != GPGPU) {
3071 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPELINE_SELECT,
3072 .PipelineSelection = GPGPU);
3073 cmd_buffer->current_pipeline = GPGPU;
3074 }
3075
3076 if (cmd_buffer->compute_dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY)
3077 anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
3078
3079 if ((cmd_buffer->descriptors_dirty & VK_SHADER_STAGE_COMPUTE_BIT) ||
3080 (cmd_buffer->compute_dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY)) {
3081 result = flush_compute_descriptor_set(cmd_buffer);
3082 if (result != VK_SUCCESS) {
3083 result = anv_cmd_buffer_new_surface_state_bo(cmd_buffer);
3084 assert(result == VK_SUCCESS);
3085 result = flush_compute_descriptor_set(cmd_buffer);
3086 assert(result == VK_SUCCESS);
3087 }
3088 cmd_buffer->descriptors_dirty &= ~VK_SHADER_STAGE_COMPUTE;
3089 }
3090
3091 cmd_buffer->compute_dirty = 0;
3092 }
3093
3094 static void
3095 anv_cmd_buffer_flush_state(struct anv_cmd_buffer *cmd_buffer)
3096 {
3097 struct anv_pipeline *pipeline = cmd_buffer->pipeline;
3098 uint32_t *p;
3099
3100 uint32_t vb_emit = cmd_buffer->vb_dirty & pipeline->vb_used;
3101
3102 assert((pipeline->active_stages & VK_SHADER_STAGE_COMPUTE_BIT) == 0);
3103
3104 if (cmd_buffer->current_pipeline != _3D) {
3105 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPELINE_SELECT,
3106 .PipelineSelection = _3D);
3107 cmd_buffer->current_pipeline = _3D;
3108 }
3109
3110 if (vb_emit) {
3111 const uint32_t num_buffers = __builtin_popcount(vb_emit);
3112 const uint32_t num_dwords = 1 + num_buffers * 4;
3113
3114 p = anv_batch_emitn(&cmd_buffer->batch, num_dwords,
3115 GEN8_3DSTATE_VERTEX_BUFFERS);
3116 uint32_t vb, i = 0;
3117 for_each_bit(vb, vb_emit) {
3118 struct anv_buffer *buffer = cmd_buffer->vertex_bindings[vb].buffer;
3119 uint32_t offset = cmd_buffer->vertex_bindings[vb].offset;
3120
3121 struct GEN8_VERTEX_BUFFER_STATE state = {
3122 .VertexBufferIndex = vb,
3123 .MemoryObjectControlState = GEN8_MOCS,
3124 .AddressModifyEnable = true,
3125 .BufferPitch = pipeline->binding_stride[vb],
3126 .BufferStartingAddress = { buffer->bo, buffer->offset + offset },
3127 .BufferSize = buffer->size - offset
3128 };
3129
3130 GEN8_VERTEX_BUFFER_STATE_pack(&cmd_buffer->batch, &p[1 + i * 4], &state);
3131 i++;
3132 }
3133 }
3134
3135 if (cmd_buffer->dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY) {
3136 /* If somebody compiled a pipeline after starting a command buffer the
3137 * scratch bo may have grown since we started this cmd buffer (and
3138 * emitted STATE_BASE_ADDRESS). If we're binding that pipeline now,
3139 * reemit STATE_BASE_ADDRESS so that we use the bigger scratch bo. */
3140 if (cmd_buffer->scratch_size < pipeline->total_scratch)
3141 anv_cmd_buffer_emit_state_base_address(cmd_buffer);
3142
3143 anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
3144 }
3145
3146 if (cmd_buffer->descriptors_dirty)
3147 flush_descriptor_sets(cmd_buffer);
3148
3149 if (cmd_buffer->dirty & ANV_CMD_BUFFER_VP_DIRTY) {
3150 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_SCISSOR_STATE_POINTERS,
3151 .ScissorRectPointer = cmd_buffer->vp_state->scissor.offset);
3152 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_CC,
3153 .CCViewportPointer = cmd_buffer->vp_state->cc_vp.offset);
3154 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP,
3155 .SFClipViewportPointer = cmd_buffer->vp_state->sf_clip_vp.offset);
3156 }
3157
3158 if (cmd_buffer->dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY | ANV_CMD_BUFFER_RS_DIRTY)) {
3159 anv_batch_emit_merge(&cmd_buffer->batch,
3160 cmd_buffer->rs_state->state_sf, pipeline->state_sf);
3161 anv_batch_emit_merge(&cmd_buffer->batch,
3162 cmd_buffer->rs_state->state_raster, pipeline->state_raster);
3163 }
3164
3165 if (cmd_buffer->ds_state &&
3166 (cmd_buffer->dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY | ANV_CMD_BUFFER_DS_DIRTY)))
3167 anv_batch_emit_merge(&cmd_buffer->batch,
3168 cmd_buffer->ds_state->state_wm_depth_stencil,
3169 pipeline->state_wm_depth_stencil);
3170
3171 if (cmd_buffer->dirty & (ANV_CMD_BUFFER_CB_DIRTY | ANV_CMD_BUFFER_DS_DIRTY)) {
3172 struct anv_state state;
3173 if (cmd_buffer->ds_state == NULL)
3174 state = anv_cmd_buffer_emit_dynamic(cmd_buffer,
3175 cmd_buffer->cb_state->state_color_calc,
3176 GEN8_COLOR_CALC_STATE_length, 64);
3177 else if (cmd_buffer->cb_state == NULL)
3178 state = anv_cmd_buffer_emit_dynamic(cmd_buffer,
3179 cmd_buffer->ds_state->state_color_calc,
3180 GEN8_COLOR_CALC_STATE_length, 64);
3181 else
3182 state = anv_cmd_buffer_merge_dynamic(cmd_buffer,
3183 cmd_buffer->ds_state->state_color_calc,
3184 cmd_buffer->cb_state->state_color_calc,
3185 GEN8_COLOR_CALC_STATE_length, 64);
3186
3187 anv_batch_emit(&cmd_buffer->batch,
3188 GEN8_3DSTATE_CC_STATE_POINTERS,
3189 .ColorCalcStatePointer = state.offset,
3190 .ColorCalcStatePointerValid = true);
3191 }
3192
3193 if (cmd_buffer->dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY | ANV_CMD_BUFFER_INDEX_BUFFER_DIRTY)) {
3194 anv_batch_emit_merge(&cmd_buffer->batch,
3195 cmd_buffer->state_vf, pipeline->state_vf);
3196 }
3197
3198 cmd_buffer->vb_dirty &= ~vb_emit;
3199 cmd_buffer->dirty = 0;
3200 }
3201
3202 void anv_CmdDraw(
3203 VkCmdBuffer cmdBuffer,
3204 uint32_t firstVertex,
3205 uint32_t vertexCount,
3206 uint32_t firstInstance,
3207 uint32_t instanceCount)
3208 {
3209 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3210
3211 anv_cmd_buffer_flush_state(cmd_buffer);
3212
3213 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
3214 .VertexAccessType = SEQUENTIAL,
3215 .VertexCountPerInstance = vertexCount,
3216 .StartVertexLocation = firstVertex,
3217 .InstanceCount = instanceCount,
3218 .StartInstanceLocation = firstInstance,
3219 .BaseVertexLocation = 0);
3220 }
3221
3222 void anv_CmdDrawIndexed(
3223 VkCmdBuffer cmdBuffer,
3224 uint32_t firstIndex,
3225 uint32_t indexCount,
3226 int32_t vertexOffset,
3227 uint32_t firstInstance,
3228 uint32_t instanceCount)
3229 {
3230 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3231
3232 anv_cmd_buffer_flush_state(cmd_buffer);
3233
3234 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
3235 .VertexAccessType = RANDOM,
3236 .VertexCountPerInstance = indexCount,
3237 .StartVertexLocation = firstIndex,
3238 .InstanceCount = instanceCount,
3239 .StartInstanceLocation = firstInstance,
3240 .BaseVertexLocation = vertexOffset);
3241 }
3242
3243 static void
3244 anv_batch_lrm(struct anv_batch *batch,
3245 uint32_t reg, struct anv_bo *bo, uint32_t offset)
3246 {
3247 anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
3248 .RegisterAddress = reg,
3249 .MemoryAddress = { bo, offset });
3250 }
3251
3252 static void
3253 anv_batch_lri(struct anv_batch *batch, uint32_t reg, uint32_t imm)
3254 {
3255 anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_IMM,
3256 .RegisterOffset = reg,
3257 .DataDWord = imm);
3258 }
3259
3260 /* Auto-Draw / Indirect Registers */
3261 #define GEN7_3DPRIM_END_OFFSET 0x2420
3262 #define GEN7_3DPRIM_START_VERTEX 0x2430
3263 #define GEN7_3DPRIM_VERTEX_COUNT 0x2434
3264 #define GEN7_3DPRIM_INSTANCE_COUNT 0x2438
3265 #define GEN7_3DPRIM_START_INSTANCE 0x243C
3266 #define GEN7_3DPRIM_BASE_VERTEX 0x2440
3267
3268 void anv_CmdDrawIndirect(
3269 VkCmdBuffer cmdBuffer,
3270 VkBuffer _buffer,
3271 VkDeviceSize offset,
3272 uint32_t count,
3273 uint32_t stride)
3274 {
3275 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3276 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
3277 struct anv_bo *bo = buffer->bo;
3278 uint32_t bo_offset = buffer->offset + offset;
3279
3280 anv_cmd_buffer_flush_state(cmd_buffer);
3281
3282 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);
3283 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
3284 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);
3285 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 12);
3286 anv_batch_lri(&cmd_buffer->batch, GEN7_3DPRIM_BASE_VERTEX, 0);
3287
3288 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
3289 .IndirectParameterEnable = true,
3290 .VertexAccessType = SEQUENTIAL);
3291 }
3292
3293 void anv_CmdDrawIndexedIndirect(
3294 VkCmdBuffer cmdBuffer,
3295 VkBuffer _buffer,
3296 VkDeviceSize offset,
3297 uint32_t count,
3298 uint32_t stride)
3299 {
3300 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3301 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
3302 struct anv_bo *bo = buffer->bo;
3303 uint32_t bo_offset = buffer->offset + offset;
3304
3305 anv_cmd_buffer_flush_state(cmd_buffer);
3306
3307 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);
3308 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
3309 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);
3310 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_BASE_VERTEX, bo, bo_offset + 12);
3311 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 16);
3312
3313 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
3314 .IndirectParameterEnable = true,
3315 .VertexAccessType = RANDOM);
3316 }
3317
3318 void anv_CmdDispatch(
3319 VkCmdBuffer cmdBuffer,
3320 uint32_t x,
3321 uint32_t y,
3322 uint32_t z)
3323 {
3324 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3325 struct anv_pipeline *pipeline = cmd_buffer->compute_pipeline;
3326 struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
3327
3328 anv_cmd_buffer_flush_compute_state(cmd_buffer);
3329
3330 anv_batch_emit(&cmd_buffer->batch, GEN8_GPGPU_WALKER,
3331 .SIMDSize = prog_data->simd_size / 16,
3332 .ThreadDepthCounterMaximum = 0,
3333 .ThreadHeightCounterMaximum = 0,
3334 .ThreadWidthCounterMaximum = pipeline->cs_thread_width_max,
3335 .ThreadGroupIDXDimension = x,
3336 .ThreadGroupIDYDimension = y,
3337 .ThreadGroupIDZDimension = z,
3338 .RightExecutionMask = pipeline->cs_right_mask,
3339 .BottomExecutionMask = 0xffffffff);
3340
3341 anv_batch_emit(&cmd_buffer->batch, GEN8_MEDIA_STATE_FLUSH);
3342 }
3343
3344 #define GPGPU_DISPATCHDIMX 0x2500
3345 #define GPGPU_DISPATCHDIMY 0x2504
3346 #define GPGPU_DISPATCHDIMZ 0x2508
3347
3348 void anv_CmdDispatchIndirect(
3349 VkCmdBuffer cmdBuffer,
3350 VkBuffer _buffer,
3351 VkDeviceSize offset)
3352 {
3353 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3354 struct anv_pipeline *pipeline = cmd_buffer->compute_pipeline;
3355 struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
3356 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
3357 struct anv_bo *bo = buffer->bo;
3358 uint32_t bo_offset = buffer->offset + offset;
3359
3360 anv_cmd_buffer_flush_compute_state(cmd_buffer);
3361
3362 anv_batch_lrm(&cmd_buffer->batch, GPGPU_DISPATCHDIMX, bo, bo_offset);
3363 anv_batch_lrm(&cmd_buffer->batch, GPGPU_DISPATCHDIMY, bo, bo_offset + 4);
3364 anv_batch_lrm(&cmd_buffer->batch, GPGPU_DISPATCHDIMZ, bo, bo_offset + 8);
3365
3366 anv_batch_emit(&cmd_buffer->batch, GEN8_GPGPU_WALKER,
3367 .IndirectParameterEnable = true,
3368 .SIMDSize = prog_data->simd_size / 16,
3369 .ThreadDepthCounterMaximum = 0,
3370 .ThreadHeightCounterMaximum = 0,
3371 .ThreadWidthCounterMaximum = pipeline->cs_thread_width_max,
3372 .RightExecutionMask = pipeline->cs_right_mask,
3373 .BottomExecutionMask = 0xffffffff);
3374
3375 anv_batch_emit(&cmd_buffer->batch, GEN8_MEDIA_STATE_FLUSH);
3376 }
3377
3378 void anv_CmdSetEvent(
3379 VkCmdBuffer cmdBuffer,
3380 VkEvent event,
3381 VkPipeEvent pipeEvent)
3382 {
3383 stub();
3384 }
3385
3386 void anv_CmdResetEvent(
3387 VkCmdBuffer cmdBuffer,
3388 VkEvent event,
3389 VkPipeEvent pipeEvent)
3390 {
3391 stub();
3392 }
3393
3394 void anv_CmdWaitEvents(
3395 VkCmdBuffer cmdBuffer,
3396 VkWaitEvent waitEvent,
3397 uint32_t eventCount,
3398 const VkEvent* pEvents,
3399 uint32_t memBarrierCount,
3400 const void** ppMemBarriers)
3401 {
3402 stub();
3403 }
3404
3405 void anv_CmdPipelineBarrier(
3406 VkCmdBuffer cmdBuffer,
3407 VkWaitEvent waitEvent,
3408 uint32_t pipeEventCount,
3409 const VkPipeEvent* pPipeEvents,
3410 uint32_t memBarrierCount,
3411 const void** ppMemBarriers)
3412 {
3413 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *)cmdBuffer;
3414 uint32_t b, *dw;
3415
3416 struct GEN8_PIPE_CONTROL cmd = {
3417 GEN8_PIPE_CONTROL_header,
3418 .PostSyncOperation = NoWrite,
3419 };
3420
3421 /* XXX: I think waitEvent is a no-op on our HW. We should verify that. */
3422
3423 for (uint32_t i = 0; i < pipeEventCount; i++) {
3424 switch (pPipeEvents[i]) {
3425 case VK_PIPE_EVENT_TOP_OF_PIPE:
3426 /* This is just what PIPE_CONTROL does */
3427 break;
3428 case VK_PIPE_EVENT_VERTEX_PROCESSING_COMPLETE:
3429 case VK_PIPE_EVENT_LOCAL_FRAGMENT_PROCESSING_COMPLETE:
3430 case VK_PIPE_EVENT_FRAGMENT_PROCESSING_COMPLETE:
3431 cmd.StallAtPixelScoreboard = true;
3432 break;
3433 case VK_PIPE_EVENT_GRAPHICS_PIPELINE_COMPLETE:
3434 case VK_PIPE_EVENT_COMPUTE_PIPELINE_COMPLETE:
3435 case VK_PIPE_EVENT_TRANSFER_COMPLETE:
3436 case VK_PIPE_EVENT_COMMANDS_COMPLETE:
3437 cmd.CommandStreamerStallEnable = true;
3438 break;
3439 default:
3440 unreachable("Invalid VkPipeEvent");
3441 }
3442 }
3443
3444 /* XXX: Right now, we're really dumb and just flush whatever categories
3445 * the app asks for. One of these days we may make this a bit better
3446 * but right now that's all the hardware allows for in most areas.
3447 */
3448 VkMemoryOutputFlags out_flags = 0;
3449 VkMemoryInputFlags in_flags = 0;
3450
3451 for (uint32_t i = 0; i < memBarrierCount; i++) {
3452 const struct anv_common *common = ppMemBarriers[i];
3453 switch (common->sType) {
3454 case VK_STRUCTURE_TYPE_MEMORY_BARRIER: {
3455 const VkMemoryBarrier *barrier = (VkMemoryBarrier *)common;
3456 out_flags |= barrier->outputMask;
3457 in_flags |= barrier->inputMask;
3458 break;
3459 }
3460 case VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER: {
3461 const VkBufferMemoryBarrier *barrier = (VkBufferMemoryBarrier *)common;
3462 out_flags |= barrier->outputMask;
3463 in_flags |= barrier->inputMask;
3464 break;
3465 }
3466 case VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER: {
3467 const VkImageMemoryBarrier *barrier = (VkImageMemoryBarrier *)common;
3468 out_flags |= barrier->outputMask;
3469 in_flags |= barrier->inputMask;
3470 break;
3471 }
3472 default:
3473 unreachable("Invalid memory barrier type");
3474 }
3475 }
3476
3477 for_each_bit(b, out_flags) {
3478 switch ((VkMemoryOutputFlags)(1 << b)) {
3479 case VK_MEMORY_OUTPUT_HOST_WRITE_BIT:
3480 break; /* FIXME: Little-core systems */
3481 case VK_MEMORY_OUTPUT_SHADER_WRITE_BIT:
3482 cmd.DCFlushEnable = true;
3483 break;
3484 case VK_MEMORY_OUTPUT_COLOR_ATTACHMENT_BIT:
3485 cmd.RenderTargetCacheFlushEnable = true;
3486 break;
3487 case VK_MEMORY_OUTPUT_DEPTH_STENCIL_ATTACHMENT_BIT:
3488 cmd.DepthCacheFlushEnable = true;
3489 break;
3490 case VK_MEMORY_OUTPUT_TRANSFER_BIT:
3491 cmd.RenderTargetCacheFlushEnable = true;
3492 cmd.DepthCacheFlushEnable = true;
3493 break;
3494 default:
3495 unreachable("Invalid memory output flag");
3496 }
3497 }
3498
3499 for_each_bit(b, out_flags) {
3500 switch ((VkMemoryInputFlags)(1 << b)) {
3501 case VK_MEMORY_INPUT_HOST_READ_BIT:
3502 break; /* FIXME: Little-core systems */
3503 case VK_MEMORY_INPUT_INDIRECT_COMMAND_BIT:
3504 case VK_MEMORY_INPUT_INDEX_FETCH_BIT:
3505 case VK_MEMORY_INPUT_VERTEX_ATTRIBUTE_FETCH_BIT:
3506 cmd.VFCacheInvalidationEnable = true;
3507 break;
3508 case VK_MEMORY_INPUT_UNIFORM_READ_BIT:
3509 cmd.ConstantCacheInvalidationEnable = true;
3510 /* fallthrough */
3511 case VK_MEMORY_INPUT_SHADER_READ_BIT:
3512 cmd.DCFlushEnable = true;
3513 cmd.TextureCacheInvalidationEnable = true;
3514 break;
3515 case VK_MEMORY_INPUT_COLOR_ATTACHMENT_BIT:
3516 case VK_MEMORY_INPUT_DEPTH_STENCIL_ATTACHMENT_BIT:
3517 break; /* XXX: Hunh? */
3518 case VK_MEMORY_INPUT_TRANSFER_BIT:
3519 cmd.TextureCacheInvalidationEnable = true;
3520 break;
3521 }
3522 }
3523
3524 dw = anv_batch_emit_dwords(&cmd_buffer->batch, GEN8_PIPE_CONTROL_length);
3525 GEN8_PIPE_CONTROL_pack(&cmd_buffer->batch, dw, &cmd);
3526 }
3527
3528 static void
3529 anv_framebuffer_destroy(struct anv_device *device,
3530 struct anv_object *object,
3531 VkObjectType obj_type)
3532 {
3533 struct anv_framebuffer *fb = (struct anv_framebuffer *)object;
3534
3535 assert(obj_type == VK_OBJECT_TYPE_FRAMEBUFFER);
3536
3537 anv_DestroyObject((VkDevice) device,
3538 VK_OBJECT_TYPE_DYNAMIC_VP_STATE,
3539 fb->vp_state);
3540
3541 anv_device_free(device, fb);
3542 }
3543
3544 VkResult anv_CreateFramebuffer(
3545 VkDevice _device,
3546 const VkFramebufferCreateInfo* pCreateInfo,
3547 VkFramebuffer* pFramebuffer)
3548 {
3549 struct anv_device *device = (struct anv_device *) _device;
3550 struct anv_framebuffer *framebuffer;
3551
3552 static const struct anv_depth_stencil_view null_view =
3553 { .depth_format = D16_UNORM, .depth_stride = 0, .stencil_stride = 0 };
3554
3555 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
3556
3557 framebuffer = anv_device_alloc(device, sizeof(*framebuffer), 8,
3558 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
3559 if (framebuffer == NULL)
3560 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
3561
3562 framebuffer->base.destructor = anv_framebuffer_destroy;
3563
3564 framebuffer->color_attachment_count = pCreateInfo->colorAttachmentCount;
3565 for (uint32_t i = 0; i < pCreateInfo->colorAttachmentCount; i++) {
3566 framebuffer->color_attachments[i] =
3567 (struct anv_surface_view *) pCreateInfo->pColorAttachments[i].view;
3568 }
3569
3570 if (pCreateInfo->pDepthStencilAttachment) {
3571 framebuffer->depth_stencil =
3572 (struct anv_depth_stencil_view *) pCreateInfo->pDepthStencilAttachment->view;
3573 } else {
3574 framebuffer->depth_stencil = &null_view;
3575 }
3576
3577 framebuffer->sample_count = pCreateInfo->sampleCount;
3578 framebuffer->width = pCreateInfo->width;
3579 framebuffer->height = pCreateInfo->height;
3580 framebuffer->layers = pCreateInfo->layers;
3581
3582 anv_CreateDynamicViewportState((VkDevice) device,
3583 &(VkDynamicVpStateCreateInfo) {
3584 .sType = VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO,
3585 .viewportAndScissorCount = 1,
3586 .pViewports = (VkViewport[]) {
3587 {
3588 .originX = 0,
3589 .originY = 0,
3590 .width = pCreateInfo->width,
3591 .height = pCreateInfo->height,
3592 .minDepth = 0,
3593 .maxDepth = 1
3594 },
3595 },
3596 .pScissors = (VkRect2D[]) {
3597 { { 0, 0 },
3598 { pCreateInfo->width, pCreateInfo->height } },
3599 }
3600 },
3601 &framebuffer->vp_state);
3602
3603 *pFramebuffer = (VkFramebuffer) framebuffer;
3604
3605 return VK_SUCCESS;
3606 }
3607
3608 VkResult anv_CreateRenderPass(
3609 VkDevice _device,
3610 const VkRenderPassCreateInfo* pCreateInfo,
3611 VkRenderPass* pRenderPass)
3612 {
3613 struct anv_device *device = (struct anv_device *) _device;
3614 struct anv_render_pass *pass;
3615 size_t size;
3616
3617 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
3618
3619 size = sizeof(*pass) +
3620 pCreateInfo->layers * sizeof(struct anv_render_pass_layer);
3621 pass = anv_device_alloc(device, size, 8,
3622 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
3623 if (pass == NULL)
3624 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
3625
3626 pass->render_area = pCreateInfo->renderArea;
3627
3628 pass->num_layers = pCreateInfo->layers;
3629
3630 pass->num_clear_layers = 0;
3631 for (uint32_t i = 0; i < pCreateInfo->layers; i++) {
3632 pass->layers[i].color_load_op = pCreateInfo->pColorLoadOps[i];
3633 pass->layers[i].clear_color = pCreateInfo->pColorLoadClearValues[i];
3634 if (pass->layers[i].color_load_op == VK_ATTACHMENT_LOAD_OP_CLEAR)
3635 pass->num_clear_layers++;
3636 }
3637
3638 *pRenderPass = (VkRenderPass) pass;
3639
3640 return VK_SUCCESS;
3641 }
3642
3643 static void
3644 anv_cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer,
3645 struct anv_render_pass *pass)
3646 {
3647 const struct anv_depth_stencil_view *view =
3648 cmd_buffer->framebuffer->depth_stencil;
3649
3650 /* FIXME: Implement the PMA stall W/A */
3651 /* FIXME: Width and Height are wrong */
3652
3653 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DEPTH_BUFFER,
3654 .SurfaceType = SURFTYPE_2D,
3655 .DepthWriteEnable = view->depth_stride > 0,
3656 .StencilWriteEnable = view->stencil_stride > 0,
3657 .HierarchicalDepthBufferEnable = false,
3658 .SurfaceFormat = view->depth_format,
3659 .SurfacePitch = view->depth_stride > 0 ? view->depth_stride - 1 : 0,
3660 .SurfaceBaseAddress = { view->bo, view->depth_offset },
3661 .Height = pass->render_area.extent.height - 1,
3662 .Width = pass->render_area.extent.width - 1,
3663 .LOD = 0,
3664 .Depth = 1 - 1,
3665 .MinimumArrayElement = 0,
3666 .DepthBufferObjectControlState = GEN8_MOCS,
3667 .RenderTargetViewExtent = 1 - 1,
3668 .SurfaceQPitch = view->depth_qpitch >> 2);
3669
3670 /* Disable hierarchial depth buffers. */
3671 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_HIER_DEPTH_BUFFER);
3672
3673 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_STENCIL_BUFFER,
3674 .StencilBufferEnable = view->stencil_stride > 0,
3675 .StencilBufferObjectControlState = GEN8_MOCS,
3676 .SurfacePitch = view->stencil_stride > 0 ? view->stencil_stride - 1 : 0,
3677 .SurfaceBaseAddress = { view->bo, view->stencil_offset },
3678 .SurfaceQPitch = view->stencil_qpitch >> 2);
3679
3680 /* Clear the clear params. */
3681 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_CLEAR_PARAMS);
3682 }
3683
3684 void anv_CmdBeginRenderPass(
3685 VkCmdBuffer cmdBuffer,
3686 const VkRenderPassBegin* pRenderPassBegin)
3687 {
3688 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3689 struct anv_render_pass *pass = (struct anv_render_pass *) pRenderPassBegin->renderPass;
3690 struct anv_framebuffer *framebuffer =
3691 (struct anv_framebuffer *) pRenderPassBegin->framebuffer;
3692
3693 cmd_buffer->framebuffer = framebuffer;
3694
3695 cmd_buffer->descriptors_dirty |= VK_SHADER_STAGE_FRAGMENT_BIT;
3696
3697 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DRAWING_RECTANGLE,
3698 .ClippedDrawingRectangleYMin = pass->render_area.offset.y,
3699 .ClippedDrawingRectangleXMin = pass->render_area.offset.x,
3700 .ClippedDrawingRectangleYMax =
3701 pass->render_area.offset.y + pass->render_area.extent.height - 1,
3702 .ClippedDrawingRectangleXMax =
3703 pass->render_area.offset.x + pass->render_area.extent.width - 1,
3704 .DrawingRectangleOriginY = 0,
3705 .DrawingRectangleOriginX = 0);
3706
3707 anv_cmd_buffer_emit_depth_stencil(cmd_buffer, pass);
3708
3709 anv_cmd_buffer_clear(cmd_buffer, pass);
3710 }
3711
3712 void anv_CmdEndRenderPass(
3713 VkCmdBuffer cmdBuffer,
3714 VkRenderPass renderPass)
3715 {
3716 /* Emit a flushing pipe control at the end of a pass. This is kind of a
3717 * hack but it ensures that render targets always actually get written.
3718 * Eventually, we should do flushing based on image format transitions
3719 * or something of that nature.
3720 */
3721 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *)cmdBuffer;
3722 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
3723 .PostSyncOperation = NoWrite,
3724 .RenderTargetCacheFlushEnable = true,
3725 .InstructionCacheInvalidateEnable = true,
3726 .DepthCacheFlushEnable = true,
3727 .VFCacheInvalidationEnable = true,
3728 .TextureCacheInvalidationEnable = true,
3729 .CommandStreamerStallEnable = true);
3730 }
3731
3732 void vkCmdDbgMarkerBegin(
3733 VkCmdBuffer cmdBuffer,
3734 const char* pMarker)
3735 __attribute__ ((visibility ("default")));
3736
3737 void vkCmdDbgMarkerEnd(
3738 VkCmdBuffer cmdBuffer)
3739 __attribute__ ((visibility ("default")));
3740
3741 VkResult vkDbgSetObjectTag(
3742 VkDevice device,
3743 VkObject object,
3744 size_t tagSize,
3745 const void* pTag)
3746 __attribute__ ((visibility ("default")));
3747
3748
3749 void vkCmdDbgMarkerBegin(
3750 VkCmdBuffer cmdBuffer,
3751 const char* pMarker)
3752 {
3753 }
3754
3755 void vkCmdDbgMarkerEnd(
3756 VkCmdBuffer cmdBuffer)
3757 {
3758 }
3759
3760 VkResult vkDbgSetObjectTag(
3761 VkDevice device,
3762 VkObject object,
3763 size_t tagSize,
3764 const void* pTag)
3765 {
3766 return VK_SUCCESS;
3767 }