vk: Emit color calc state
[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->maxAtomicCounters = 0;
241 queue_properties->supportsTimestamps = true;
242 queue_properties->maxMemReferences = 256;
243 return VK_SUCCESS;
244
245 case VK_PHYSICAL_DEVICE_INFO_TYPE_MEMORY_PROPERTIES:
246 memory_properties = pData;
247
248 *pDataSize = sizeof(*memory_properties);
249 if (pData == NULL)
250 return VK_SUCCESS;
251
252 memory_properties->supportsMigration = false;
253 memory_properties->supportsPinning = false;
254 return VK_SUCCESS;
255
256 case VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI:
257 anv_finishme("VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI");
258
259 *pDataSize = sizeof(*display_properties);
260 if (pData == NULL)
261 return VK_SUCCESS;
262
263 display_properties = pData;
264 display_properties->display = 0;
265 display_properties->physicalResolution = (VkExtent2D) { 0, 0 };
266 return VK_SUCCESS;
267
268 case VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI:
269 anv_finishme("VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI");
270 return VK_SUCCESS;
271
272
273 default:
274 return VK_UNSUPPORTED;
275 }
276
277 }
278
279 void * vkGetProcAddr(
280 VkPhysicalDevice physicalDevice,
281 const char* pName)
282 {
283 return anv_lookup_entrypoint(pName);
284 }
285
286 static void
287 parse_debug_flags(struct anv_device *device)
288 {
289 const char *debug, *p, *end;
290
291 debug = getenv("INTEL_DEBUG");
292 device->dump_aub = false;
293 if (debug) {
294 for (p = debug; *p; p = end + 1) {
295 end = strchrnul(p, ',');
296 if (end - p == 3 && memcmp(p, "aub", 3) == 0)
297 device->dump_aub = true;
298 if (end - p == 5 && memcmp(p, "no_hw", 5) == 0)
299 device->no_hw = true;
300 if (*end == '\0')
301 break;
302 }
303 }
304 }
305
306 VkResult anv_CreateDevice(
307 VkPhysicalDevice _physicalDevice,
308 const VkDeviceCreateInfo* pCreateInfo,
309 VkDevice* pDevice)
310 {
311 struct anv_physical_device *physicalDevice =
312 (struct anv_physical_device *) _physicalDevice;
313 struct anv_instance *instance = physicalDevice->instance;
314 struct anv_device *device;
315
316 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO);
317
318 device = instance->pfnAlloc(instance->pAllocUserData,
319 sizeof(*device), 8,
320 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
321 if (!device)
322 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
323
324 device->no_hw = physicalDevice->no_hw;
325 parse_debug_flags(device);
326
327 device->instance = physicalDevice->instance;
328 device->fd = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC);
329 if (device->fd == -1)
330 goto fail_device;
331
332 device->context_id = anv_gem_create_context(device);
333 if (device->context_id == -1)
334 goto fail_fd;
335
336 anv_block_pool_init(&device->dynamic_state_block_pool, device, 2048);
337
338 anv_state_pool_init(&device->dynamic_state_pool,
339 &device->dynamic_state_block_pool);
340
341 anv_block_pool_init(&device->instruction_block_pool, device, 2048);
342 anv_block_pool_init(&device->surface_state_block_pool, device, 2048);
343
344
345 /* Binding table pointers are only 16 bits so we have to make sure that
346 * they get allocated at the beginning of the surface state BO. To
347 * handle this, we create a separate block pool that works out of the
348 * first 64 KB of the surface state BO.
349 */
350 anv_block_pool_init_slave(&device->binding_table_block_pool,
351 &device->surface_state_block_pool, 32);
352
353 anv_state_pool_init(&device->surface_state_pool,
354 &device->surface_state_block_pool);
355
356 device->compiler = anv_compiler_create(device->fd);
357 device->aub_writer = NULL;
358
359 device->info = *physicalDevice->info;
360
361 pthread_mutex_init(&device->mutex, NULL);
362
363 anv_device_init_meta(device);
364
365 *pDevice = (VkDevice) device;
366
367 return VK_SUCCESS;
368
369 fail_fd:
370 close(device->fd);
371 fail_device:
372 anv_device_free(device, device);
373
374 return vk_error(VK_ERROR_UNAVAILABLE);
375 }
376
377 VkResult anv_DestroyDevice(
378 VkDevice _device)
379 {
380 struct anv_device *device = (struct anv_device *) _device;
381
382 anv_compiler_destroy(device->compiler);
383
384 anv_block_pool_finish(&device->dynamic_state_block_pool);
385 anv_block_pool_finish(&device->instruction_block_pool);
386 anv_block_pool_finish(&device->surface_state_block_pool);
387
388 close(device->fd);
389
390 if (device->aub_writer)
391 anv_aub_writer_destroy(device->aub_writer);
392
393 anv_device_free(device, device);
394
395 return VK_SUCCESS;
396 }
397
398 VkResult anv_GetGlobalExtensionInfo(
399 VkExtensionInfoType infoType,
400 uint32_t extensionIndex,
401 size_t* pDataSize,
402 void* pData)
403 {
404 static const VkExtensionProperties extensions[] = {
405 {
406 .extName = "VK_WSI_LunarG",
407 .version = 3
408 }
409 };
410 uint32_t count = ARRAY_SIZE(extensions);
411
412 switch (infoType) {
413 case VK_EXTENSION_INFO_TYPE_COUNT:
414 memcpy(pData, &count, sizeof(count));
415 *pDataSize = sizeof(count);
416 return VK_SUCCESS;
417
418 case VK_EXTENSION_INFO_TYPE_PROPERTIES:
419 if (extensionIndex >= count)
420 return vk_error(VK_ERROR_INVALID_EXTENSION);
421
422 memcpy(pData, &extensions[extensionIndex], sizeof(extensions[0]));
423 *pDataSize = sizeof(extensions[0]);
424 return VK_SUCCESS;
425
426 default:
427 return VK_UNSUPPORTED;
428 }
429 }
430
431 VkResult anv_GetPhysicalDeviceExtensionInfo(
432 VkPhysicalDevice physicalDevice,
433 VkExtensionInfoType infoType,
434 uint32_t extensionIndex,
435 size_t* pDataSize,
436 void* pData)
437 {
438 uint32_t *count;
439
440 switch (infoType) {
441 case VK_EXTENSION_INFO_TYPE_COUNT:
442 *pDataSize = 4;
443 if (pData == NULL)
444 return VK_SUCCESS;
445
446 count = pData;
447 *count = 0;
448 return VK_SUCCESS;
449
450 case VK_EXTENSION_INFO_TYPE_PROPERTIES:
451 return vk_error(VK_ERROR_INVALID_EXTENSION);
452
453 default:
454 return VK_UNSUPPORTED;
455 }
456 }
457
458 VkResult anv_EnumerateLayers(
459 VkPhysicalDevice physicalDevice,
460 size_t maxStringSize,
461 size_t* pLayerCount,
462 char* const* pOutLayers,
463 void* pReserved)
464 {
465 *pLayerCount = 0;
466
467 return VK_SUCCESS;
468 }
469
470 VkResult anv_GetDeviceQueue(
471 VkDevice _device,
472 uint32_t queueNodeIndex,
473 uint32_t queueIndex,
474 VkQueue* pQueue)
475 {
476 struct anv_device *device = (struct anv_device *) _device;
477 struct anv_queue *queue;
478
479 /* FIXME: Should allocate these at device create time. */
480
481 queue = anv_device_alloc(device, sizeof(*queue), 8,
482 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
483 if (queue == NULL)
484 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
485
486 queue->device = device;
487 queue->pool = &device->surface_state_pool;
488
489 queue->completed_serial = anv_state_pool_alloc(queue->pool, 4, 4);
490 *(uint32_t *)queue->completed_serial.map = 0;
491 queue->next_serial = 1;
492
493 *pQueue = (VkQueue) queue;
494
495 return VK_SUCCESS;
496 }
497
498 static const uint32_t BATCH_SIZE = 8192;
499
500 VkResult
501 anv_batch_init(struct anv_batch *batch, struct anv_device *device)
502 {
503 VkResult result;
504
505 result = anv_bo_init_new(&batch->bo, device, BATCH_SIZE);
506 if (result != VK_SUCCESS)
507 return result;
508
509 batch->bo.map =
510 anv_gem_mmap(device, batch->bo.gem_handle, 0, BATCH_SIZE);
511 if (batch->bo.map == NULL) {
512 result = vk_error(VK_ERROR_MEMORY_MAP_FAILED);
513 goto fail_bo;
514 }
515
516 batch->cmd_relocs.num_relocs = 0;
517 batch->next = batch->bo.map;
518
519 return VK_SUCCESS;
520
521 fail_bo:
522 anv_gem_close(device, batch->bo.gem_handle);
523
524 return result;
525
526 }
527
528 void
529 anv_batch_finish(struct anv_batch *batch, struct anv_device *device)
530 {
531 anv_gem_munmap(batch->bo.map, BATCH_SIZE);
532 anv_gem_close(device, batch->bo.gem_handle);
533 }
534
535 void
536 anv_batch_reset(struct anv_batch *batch)
537 {
538 batch->next = batch->bo.map;
539 batch->cmd_relocs.num_relocs = 0;
540 }
541
542 void *
543 anv_batch_emit_dwords(struct anv_batch *batch, int num_dwords)
544 {
545 void *p = batch->next;
546
547 batch->next += num_dwords * 4;
548
549 return p;
550 }
551
552 static void
553 anv_reloc_list_append(struct anv_reloc_list *list,
554 struct anv_reloc_list *other, uint32_t offset)
555 {
556 uint32_t i, count;
557
558 count = list->num_relocs;
559 memcpy(&list->relocs[count], &other->relocs[0],
560 other->num_relocs * sizeof(other->relocs[0]));
561 memcpy(&list->reloc_bos[count], &other->reloc_bos[0],
562 other->num_relocs * sizeof(other->reloc_bos[0]));
563 for (i = 0; i < other->num_relocs; i++)
564 list->relocs[i + count].offset += offset;
565
566 count += other->num_relocs;
567 }
568
569 static uint64_t
570 anv_reloc_list_add(struct anv_reloc_list *list,
571 uint32_t offset,
572 struct anv_bo *target_bo, uint32_t delta)
573 {
574 struct drm_i915_gem_relocation_entry *entry;
575 int index;
576
577 assert(list->num_relocs < ANV_BATCH_MAX_RELOCS);
578
579 /* XXX: Can we use I915_EXEC_HANDLE_LUT? */
580 index = list->num_relocs++;
581 list->reloc_bos[index] = target_bo;
582 entry = &list->relocs[index];
583 entry->target_handle = target_bo->gem_handle;
584 entry->delta = delta;
585 entry->offset = offset;
586 entry->presumed_offset = target_bo->offset;
587 entry->read_domains = 0;
588 entry->write_domain = 0;
589
590 return target_bo->offset + delta;
591 }
592
593 void
594 anv_batch_emit_batch(struct anv_batch *batch, struct anv_batch *other)
595 {
596 uint32_t size, offset;
597
598 size = other->next - other->bo.map;
599 memcpy(batch->next, other->bo.map, size);
600
601 offset = batch->next - batch->bo.map;
602 anv_reloc_list_append(&batch->cmd_relocs, &other->cmd_relocs, offset);
603
604 batch->next += size;
605 }
606
607 uint64_t
608 anv_batch_emit_reloc(struct anv_batch *batch,
609 void *location, struct anv_bo *bo, uint32_t delta)
610 {
611 return anv_reloc_list_add(&batch->cmd_relocs,
612 location - batch->bo.map, bo, delta);
613 }
614
615 VkResult anv_QueueSubmit(
616 VkQueue _queue,
617 uint32_t cmdBufferCount,
618 const VkCmdBuffer* pCmdBuffers,
619 VkFence _fence)
620 {
621 struct anv_queue *queue = (struct anv_queue *) _queue;
622 struct anv_device *device = queue->device;
623 struct anv_fence *fence = (struct anv_fence *) _fence;
624 int ret;
625
626 for (uint32_t i = 0; i < cmdBufferCount; i++) {
627 struct anv_cmd_buffer *cmd_buffer =
628 (struct anv_cmd_buffer *) pCmdBuffers[i];
629
630 if (device->dump_aub)
631 anv_cmd_buffer_dump(cmd_buffer);
632
633 if (!device->no_hw) {
634 ret = anv_gem_execbuffer(device, &cmd_buffer->execbuf);
635 if (ret != 0)
636 return vk_error(VK_ERROR_UNKNOWN);
637
638 if (fence) {
639 ret = anv_gem_execbuffer(device, &fence->execbuf);
640 if (ret != 0)
641 return vk_error(VK_ERROR_UNKNOWN);
642 }
643
644 for (uint32_t i = 0; i < cmd_buffer->bo_count; i++)
645 cmd_buffer->exec2_bos[i]->offset = cmd_buffer->exec2_objects[i].offset;
646 } else {
647 *(uint32_t *)queue->completed_serial.map = cmd_buffer->serial;
648 }
649 }
650
651 return VK_SUCCESS;
652 }
653
654 VkResult anv_QueueAddMemReferences(
655 VkQueue queue,
656 uint32_t count,
657 const VkDeviceMemory* pMems)
658 {
659 return VK_SUCCESS;
660 }
661
662 VkResult anv_QueueRemoveMemReferences(
663 VkQueue queue,
664 uint32_t count,
665 const VkDeviceMemory* pMems)
666 {
667 return VK_SUCCESS;
668 }
669
670 VkResult anv_QueueWaitIdle(
671 VkQueue _queue)
672 {
673 struct anv_queue *queue = (struct anv_queue *) _queue;
674
675 return vkDeviceWaitIdle((VkDevice) queue->device);
676 }
677
678 VkResult anv_DeviceWaitIdle(
679 VkDevice _device)
680 {
681 struct anv_device *device = (struct anv_device *) _device;
682 struct anv_state state;
683 struct anv_batch batch;
684 struct drm_i915_gem_execbuffer2 execbuf;
685 struct drm_i915_gem_exec_object2 exec2_objects[1];
686 struct anv_bo *bo = NULL;
687 VkResult result;
688 int64_t timeout;
689 int ret;
690
691 state = anv_state_pool_alloc(&device->dynamic_state_pool, 32, 32);
692 bo = &device->dynamic_state_pool.block_pool->bo;
693 batch.next = state.map;
694 anv_batch_emit(&batch, GEN8_MI_BATCH_BUFFER_END);
695 anv_batch_emit(&batch, GEN8_MI_NOOP);
696
697 exec2_objects[0].handle = bo->gem_handle;
698 exec2_objects[0].relocation_count = 0;
699 exec2_objects[0].relocs_ptr = 0;
700 exec2_objects[0].alignment = 0;
701 exec2_objects[0].offset = bo->offset;
702 exec2_objects[0].flags = 0;
703 exec2_objects[0].rsvd1 = 0;
704 exec2_objects[0].rsvd2 = 0;
705
706 execbuf.buffers_ptr = (uintptr_t) exec2_objects;
707 execbuf.buffer_count = 1;
708 execbuf.batch_start_offset = state.offset;
709 execbuf.batch_len = batch.next - state.map;
710 execbuf.cliprects_ptr = 0;
711 execbuf.num_cliprects = 0;
712 execbuf.DR1 = 0;
713 execbuf.DR4 = 0;
714
715 execbuf.flags =
716 I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
717 execbuf.rsvd1 = device->context_id;
718 execbuf.rsvd2 = 0;
719
720 if (!device->no_hw) {
721 ret = anv_gem_execbuffer(device, &execbuf);
722 if (ret != 0) {
723 result = vk_error(VK_ERROR_UNKNOWN);
724 goto fail;
725 }
726
727 timeout = INT64_MAX;
728 ret = anv_gem_wait(device, bo->gem_handle, &timeout);
729 if (ret != 0) {
730 result = vk_error(VK_ERROR_UNKNOWN);
731 goto fail;
732 }
733 }
734
735 anv_state_pool_free(&device->dynamic_state_pool, state);
736
737 return VK_SUCCESS;
738
739 fail:
740 anv_state_pool_free(&device->dynamic_state_pool, state);
741
742 return result;
743 }
744
745 void *
746 anv_device_alloc(struct anv_device * device,
747 size_t size,
748 size_t alignment,
749 VkSystemAllocType allocType)
750 {
751 return device->instance->pfnAlloc(device->instance->pAllocUserData,
752 size,
753 alignment,
754 allocType);
755 }
756
757 void
758 anv_device_free(struct anv_device * device,
759 void * mem)
760 {
761 return device->instance->pfnFree(device->instance->pAllocUserData,
762 mem);
763 }
764
765 VkResult
766 anv_bo_init_new(struct anv_bo *bo, struct anv_device *device, uint64_t size)
767 {
768 bo->gem_handle = anv_gem_create(device, size);
769 if (!bo->gem_handle)
770 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
771
772 bo->map = NULL;
773 bo->index = 0;
774 bo->offset = 0;
775 bo->size = size;
776
777 return VK_SUCCESS;
778 }
779
780 VkResult anv_AllocMemory(
781 VkDevice _device,
782 const VkMemoryAllocInfo* pAllocInfo,
783 VkDeviceMemory* pMem)
784 {
785 struct anv_device *device = (struct anv_device *) _device;
786 struct anv_device_memory *mem;
787 VkResult result;
788
789 assert(pAllocInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOC_INFO);
790
791 mem = anv_device_alloc(device, sizeof(*mem), 8,
792 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
793 if (mem == NULL)
794 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
795
796 result = anv_bo_init_new(&mem->bo, device, pAllocInfo->allocationSize);
797 if (result != VK_SUCCESS)
798 goto fail;
799
800 *pMem = (VkDeviceMemory) mem;
801
802 return VK_SUCCESS;
803
804 fail:
805 anv_device_free(device, mem);
806
807 return result;
808 }
809
810 VkResult anv_FreeMemory(
811 VkDevice _device,
812 VkDeviceMemory _mem)
813 {
814 struct anv_device *device = (struct anv_device *) _device;
815 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
816
817 if (mem->bo.map)
818 anv_gem_munmap(mem->bo.map, mem->bo.size);
819
820 if (mem->bo.gem_handle != 0)
821 anv_gem_close(device, mem->bo.gem_handle);
822
823 anv_device_free(device, mem);
824
825 return VK_SUCCESS;
826 }
827
828 VkResult anv_SetMemoryPriority(
829 VkDevice device,
830 VkDeviceMemory mem,
831 VkMemoryPriority priority)
832 {
833 return VK_SUCCESS;
834 }
835
836 VkResult anv_MapMemory(
837 VkDevice _device,
838 VkDeviceMemory _mem,
839 VkDeviceSize offset,
840 VkDeviceSize size,
841 VkMemoryMapFlags flags,
842 void** ppData)
843 {
844 struct anv_device *device = (struct anv_device *) _device;
845 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
846
847 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
848 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
849 * at a time is valid. We could just mmap up front and return an offset
850 * pointer here, but that may exhaust virtual memory on 32 bit
851 * userspace. */
852
853 mem->map = anv_gem_mmap(device, mem->bo.gem_handle, offset, size);
854 mem->map_size = size;
855
856 *ppData = mem->map;
857
858 return VK_SUCCESS;
859 }
860
861 VkResult anv_UnmapMemory(
862 VkDevice _device,
863 VkDeviceMemory _mem)
864 {
865 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
866
867 anv_gem_munmap(mem->map, mem->map_size);
868
869 return VK_SUCCESS;
870 }
871
872 VkResult anv_FlushMappedMemory(
873 VkDevice device,
874 VkDeviceMemory mem,
875 VkDeviceSize offset,
876 VkDeviceSize size)
877 {
878 /* clflush here for !llc platforms */
879
880 return VK_SUCCESS;
881 }
882
883 VkResult anv_PinSystemMemory(
884 VkDevice device,
885 const void* pSysMem,
886 size_t memSize,
887 VkDeviceMemory* pMem)
888 {
889 return VK_SUCCESS;
890 }
891
892 VkResult anv_GetMultiDeviceCompatibility(
893 VkPhysicalDevice physicalDevice0,
894 VkPhysicalDevice physicalDevice1,
895 VkPhysicalDeviceCompatibilityInfo* pInfo)
896 {
897 return VK_UNSUPPORTED;
898 }
899
900 VkResult anv_OpenSharedMemory(
901 VkDevice device,
902 const VkMemoryOpenInfo* pOpenInfo,
903 VkDeviceMemory* pMem)
904 {
905 return VK_UNSUPPORTED;
906 }
907
908 VkResult anv_OpenSharedSemaphore(
909 VkDevice device,
910 const VkSemaphoreOpenInfo* pOpenInfo,
911 VkSemaphore* pSemaphore)
912 {
913 return VK_UNSUPPORTED;
914 }
915
916 VkResult anv_OpenPeerMemory(
917 VkDevice device,
918 const VkPeerMemoryOpenInfo* pOpenInfo,
919 VkDeviceMemory* pMem)
920 {
921 return VK_UNSUPPORTED;
922 }
923
924 VkResult anv_OpenPeerImage(
925 VkDevice device,
926 const VkPeerImageOpenInfo* pOpenInfo,
927 VkImage* pImage,
928 VkDeviceMemory* pMem)
929 {
930 return VK_UNSUPPORTED;
931 }
932
933 VkResult anv_DestroyObject(
934 VkDevice _device,
935 VkObjectType objType,
936 VkObject _object)
937 {
938 struct anv_device *device = (struct anv_device *) _device;
939 struct anv_object *object = (struct anv_object *) _object;
940
941 switch (objType) {
942 case VK_OBJECT_TYPE_INSTANCE:
943 return anv_DestroyInstance((VkInstance) _object);
944
945 case VK_OBJECT_TYPE_PHYSICAL_DEVICE:
946 /* We don't want to actually destroy physical devices */
947 return VK_SUCCESS;
948
949 case VK_OBJECT_TYPE_DEVICE:
950 assert(_device == (VkDevice) _object);
951 return anv_DestroyDevice((VkDevice) _object);
952
953 case VK_OBJECT_TYPE_QUEUE:
954 /* TODO */
955 return VK_SUCCESS;
956
957 case VK_OBJECT_TYPE_DEVICE_MEMORY:
958 return anv_FreeMemory(_device, (VkDeviceMemory) _object);
959
960 case VK_OBJECT_TYPE_DESCRIPTOR_POOL:
961 /* These are just dummys anyway, so we don't need to destroy them */
962 return VK_SUCCESS;
963
964 case VK_OBJECT_TYPE_BUFFER:
965 case VK_OBJECT_TYPE_BUFFER_VIEW:
966 case VK_OBJECT_TYPE_IMAGE:
967 case VK_OBJECT_TYPE_IMAGE_VIEW:
968 case VK_OBJECT_TYPE_COLOR_ATTACHMENT_VIEW:
969 case VK_OBJECT_TYPE_DEPTH_STENCIL_VIEW:
970 case VK_OBJECT_TYPE_SHADER:
971 case VK_OBJECT_TYPE_PIPELINE_LAYOUT:
972 case VK_OBJECT_TYPE_SAMPLER:
973 case VK_OBJECT_TYPE_DESCRIPTOR_SET:
974 case VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT:
975 case VK_OBJECT_TYPE_DYNAMIC_RS_STATE:
976 case VK_OBJECT_TYPE_DYNAMIC_CB_STATE:
977 case VK_OBJECT_TYPE_DYNAMIC_DS_STATE:
978 case VK_OBJECT_TYPE_RENDER_PASS:
979 /* These are trivially destroyable */
980 anv_device_free(device, (void *) _object);
981 return VK_SUCCESS;
982
983 case VK_OBJECT_TYPE_COMMAND_BUFFER:
984 case VK_OBJECT_TYPE_PIPELINE:
985 case VK_OBJECT_TYPE_DYNAMIC_VP_STATE:
986 case VK_OBJECT_TYPE_FENCE:
987 case VK_OBJECT_TYPE_QUERY_POOL:
988 case VK_OBJECT_TYPE_FRAMEBUFFER:
989 (object->destructor)(device, object, objType);
990 return VK_SUCCESS;
991
992 case VK_OBJECT_TYPE_SEMAPHORE:
993 case VK_OBJECT_TYPE_EVENT:
994 stub_return(VK_UNSUPPORTED);
995
996 default:
997 unreachable("Invalid object type");
998 }
999 }
1000
1001 static void
1002 fill_memory_requirements(
1003 VkObjectType objType,
1004 VkObject object,
1005 VkMemoryRequirements * memory_requirements)
1006 {
1007 struct anv_buffer *buffer;
1008 struct anv_image *image;
1009
1010 memory_requirements->memPropsAllowed =
1011 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
1012 VK_MEMORY_PROPERTY_HOST_DEVICE_COHERENT_BIT |
1013 /* VK_MEMORY_PROPERTY_HOST_UNCACHED_BIT | */
1014 VK_MEMORY_PROPERTY_HOST_WRITE_COMBINED_BIT |
1015 VK_MEMORY_PROPERTY_PREFER_HOST_LOCAL |
1016 VK_MEMORY_PROPERTY_SHAREABLE_BIT;
1017
1018 memory_requirements->memPropsRequired = 0;
1019
1020 switch (objType) {
1021 case VK_OBJECT_TYPE_BUFFER:
1022 buffer = (struct anv_buffer *) object;
1023 memory_requirements->size = buffer->size;
1024 memory_requirements->alignment = 16;
1025 break;
1026 case VK_OBJECT_TYPE_IMAGE:
1027 image = (struct anv_image *) object;
1028 memory_requirements->size = image->size;
1029 memory_requirements->alignment = image->alignment;
1030 break;
1031 default:
1032 memory_requirements->size = 0;
1033 break;
1034 }
1035 }
1036
1037 static uint32_t
1038 get_allocation_count(VkObjectType objType)
1039 {
1040 switch (objType) {
1041 case VK_OBJECT_TYPE_BUFFER:
1042 case VK_OBJECT_TYPE_IMAGE:
1043 return 1;
1044 default:
1045 return 0;
1046 }
1047 }
1048
1049 VkResult anv_GetObjectInfo(
1050 VkDevice _device,
1051 VkObjectType objType,
1052 VkObject object,
1053 VkObjectInfoType infoType,
1054 size_t* pDataSize,
1055 void* pData)
1056 {
1057 VkMemoryRequirements memory_requirements;
1058 uint32_t *count;
1059
1060 switch (infoType) {
1061 case VK_OBJECT_INFO_TYPE_MEMORY_REQUIREMENTS:
1062 *pDataSize = sizeof(memory_requirements);
1063 if (pData == NULL)
1064 return VK_SUCCESS;
1065
1066 fill_memory_requirements(objType, object, pData);
1067 return VK_SUCCESS;
1068
1069 case VK_OBJECT_INFO_TYPE_MEMORY_ALLOCATION_COUNT:
1070 *pDataSize = sizeof(count);
1071 if (pData == NULL)
1072 return VK_SUCCESS;
1073
1074 count = pData;
1075 *count = get_allocation_count(objType);
1076 return VK_SUCCESS;
1077
1078 default:
1079 return VK_UNSUPPORTED;
1080 }
1081
1082 }
1083
1084 VkResult anv_QueueBindObjectMemory(
1085 VkQueue queue,
1086 VkObjectType objType,
1087 VkObject object,
1088 uint32_t allocationIdx,
1089 VkDeviceMemory _mem,
1090 VkDeviceSize memOffset)
1091 {
1092 struct anv_buffer *buffer;
1093 struct anv_image *image;
1094 struct anv_device_memory *mem = (struct anv_device_memory *) _mem;
1095
1096 switch (objType) {
1097 case VK_OBJECT_TYPE_BUFFER:
1098 buffer = (struct anv_buffer *) object;
1099 buffer->bo = &mem->bo;
1100 buffer->offset = memOffset;
1101 break;
1102 case VK_OBJECT_TYPE_IMAGE:
1103 image = (struct anv_image *) object;
1104 image->bo = &mem->bo;
1105 image->offset = memOffset;
1106 break;
1107 default:
1108 break;
1109 }
1110
1111 return VK_SUCCESS;
1112 }
1113
1114 VkResult anv_QueueBindObjectMemoryRange(
1115 VkQueue queue,
1116 VkObjectType objType,
1117 VkObject object,
1118 uint32_t allocationIdx,
1119 VkDeviceSize rangeOffset,
1120 VkDeviceSize rangeSize,
1121 VkDeviceMemory mem,
1122 VkDeviceSize memOffset)
1123 {
1124 stub_return(VK_UNSUPPORTED);
1125 }
1126
1127 VkResult anv_QueueBindImageMemoryRange(
1128 VkQueue queue,
1129 VkImage image,
1130 uint32_t allocationIdx,
1131 const VkImageMemoryBindInfo* pBindInfo,
1132 VkDeviceMemory mem,
1133 VkDeviceSize memOffset)
1134 {
1135 stub_return(VK_UNSUPPORTED);
1136 }
1137
1138 static void
1139 anv_fence_destroy(struct anv_device *device,
1140 struct anv_object *object,
1141 VkObjectType obj_type)
1142 {
1143 struct anv_fence *fence = (struct anv_fence *) object;
1144
1145 assert(obj_type == VK_OBJECT_TYPE_FENCE);
1146
1147 anv_gem_munmap(fence->bo.map, fence->bo.size);
1148 anv_gem_close(device, fence->bo.gem_handle);
1149 anv_device_free(device, fence);
1150 }
1151
1152 VkResult anv_CreateFence(
1153 VkDevice _device,
1154 const VkFenceCreateInfo* pCreateInfo,
1155 VkFence* pFence)
1156 {
1157 struct anv_device *device = (struct anv_device *) _device;
1158 struct anv_fence *fence;
1159 struct anv_batch batch;
1160 VkResult result;
1161
1162 const uint32_t fence_size = 128;
1163
1164 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FENCE_CREATE_INFO);
1165
1166 fence = anv_device_alloc(device, sizeof(*fence), 8,
1167 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1168 if (fence == NULL)
1169 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1170
1171 result = anv_bo_init_new(&fence->bo, device, fence_size);
1172 if (result != VK_SUCCESS)
1173 goto fail;
1174
1175 fence->base.destructor = anv_fence_destroy;
1176
1177 fence->bo.map =
1178 anv_gem_mmap(device, fence->bo.gem_handle, 0, fence->bo.size);
1179 batch.next = fence->bo.map;
1180 anv_batch_emit(&batch, GEN8_MI_BATCH_BUFFER_END);
1181 anv_batch_emit(&batch, GEN8_MI_NOOP);
1182
1183 fence->exec2_objects[0].handle = fence->bo.gem_handle;
1184 fence->exec2_objects[0].relocation_count = 0;
1185 fence->exec2_objects[0].relocs_ptr = 0;
1186 fence->exec2_objects[0].alignment = 0;
1187 fence->exec2_objects[0].offset = fence->bo.offset;
1188 fence->exec2_objects[0].flags = 0;
1189 fence->exec2_objects[0].rsvd1 = 0;
1190 fence->exec2_objects[0].rsvd2 = 0;
1191
1192 fence->execbuf.buffers_ptr = (uintptr_t) fence->exec2_objects;
1193 fence->execbuf.buffer_count = 1;
1194 fence->execbuf.batch_start_offset = 0;
1195 fence->execbuf.batch_len = batch.next - fence->bo.map;
1196 fence->execbuf.cliprects_ptr = 0;
1197 fence->execbuf.num_cliprects = 0;
1198 fence->execbuf.DR1 = 0;
1199 fence->execbuf.DR4 = 0;
1200
1201 fence->execbuf.flags =
1202 I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
1203 fence->execbuf.rsvd1 = device->context_id;
1204 fence->execbuf.rsvd2 = 0;
1205
1206 *pFence = (VkQueryPool) fence;
1207
1208 return VK_SUCCESS;
1209
1210 fail:
1211 anv_device_free(device, fence);
1212
1213 return result;
1214 }
1215
1216 VkResult anv_ResetFences(
1217 VkDevice _device,
1218 uint32_t fenceCount,
1219 VkFence* pFences)
1220 {
1221 struct anv_fence **fences = (struct anv_fence **) pFences;
1222
1223 for (uint32_t i; i < fenceCount; i++)
1224 fences[i]->ready = false;
1225
1226 return VK_SUCCESS;
1227 }
1228
1229 VkResult anv_GetFenceStatus(
1230 VkDevice _device,
1231 VkFence _fence)
1232 {
1233 struct anv_device *device = (struct anv_device *) _device;
1234 struct anv_fence *fence = (struct anv_fence *) _fence;
1235 int64_t t = 0;
1236 int ret;
1237
1238 if (fence->ready)
1239 return VK_SUCCESS;
1240
1241 ret = anv_gem_wait(device, fence->bo.gem_handle, &t);
1242 if (ret == 0) {
1243 fence->ready = true;
1244 return VK_SUCCESS;
1245 }
1246
1247 return VK_NOT_READY;
1248 }
1249
1250 VkResult anv_WaitForFences(
1251 VkDevice _device,
1252 uint32_t fenceCount,
1253 const VkFence* pFences,
1254 bool32_t waitAll,
1255 uint64_t timeout)
1256 {
1257 struct anv_device *device = (struct anv_device *) _device;
1258 struct anv_fence **fences = (struct anv_fence **) pFences;
1259 int64_t t = timeout;
1260 int ret;
1261
1262 /* FIXME: handle !waitAll */
1263
1264 for (uint32_t i = 0; i < fenceCount; i++) {
1265 ret = anv_gem_wait(device, fences[i]->bo.gem_handle, &t);
1266 if (ret == -1 && errno == ETIME)
1267 return VK_TIMEOUT;
1268 else if (ret == -1)
1269 return vk_error(VK_ERROR_UNKNOWN);
1270 }
1271
1272 return VK_SUCCESS;
1273 }
1274
1275 // Queue semaphore functions
1276
1277 VkResult anv_CreateSemaphore(
1278 VkDevice device,
1279 const VkSemaphoreCreateInfo* pCreateInfo,
1280 VkSemaphore* pSemaphore)
1281 {
1282 stub_return(VK_UNSUPPORTED);
1283 }
1284
1285 VkResult anv_QueueSignalSemaphore(
1286 VkQueue queue,
1287 VkSemaphore semaphore)
1288 {
1289 stub_return(VK_UNSUPPORTED);
1290 }
1291
1292 VkResult anv_QueueWaitSemaphore(
1293 VkQueue queue,
1294 VkSemaphore semaphore)
1295 {
1296 stub_return(VK_UNSUPPORTED);
1297 }
1298
1299 // Event functions
1300
1301 VkResult anv_CreateEvent(
1302 VkDevice device,
1303 const VkEventCreateInfo* pCreateInfo,
1304 VkEvent* pEvent)
1305 {
1306 stub_return(VK_UNSUPPORTED);
1307 }
1308
1309 VkResult anv_GetEventStatus(
1310 VkDevice device,
1311 VkEvent event)
1312 {
1313 stub_return(VK_UNSUPPORTED);
1314 }
1315
1316 VkResult anv_SetEvent(
1317 VkDevice device,
1318 VkEvent event)
1319 {
1320 stub_return(VK_UNSUPPORTED);
1321 }
1322
1323 VkResult anv_ResetEvent(
1324 VkDevice device,
1325 VkEvent event)
1326 {
1327 stub_return(VK_UNSUPPORTED);
1328 }
1329
1330 // Query functions
1331
1332 static void
1333 anv_query_pool_destroy(struct anv_device *device,
1334 struct anv_object *object,
1335 VkObjectType obj_type)
1336 {
1337 struct anv_query_pool *pool = (struct anv_query_pool *) object;
1338
1339 assert(obj_type == VK_OBJECT_TYPE_QUERY_POOL);
1340
1341 anv_gem_munmap(pool->bo.map, pool->bo.size);
1342 anv_gem_close(device, pool->bo.gem_handle);
1343 anv_device_free(device, pool);
1344 }
1345
1346 VkResult anv_CreateQueryPool(
1347 VkDevice _device,
1348 const VkQueryPoolCreateInfo* pCreateInfo,
1349 VkQueryPool* pQueryPool)
1350 {
1351 struct anv_device *device = (struct anv_device *) _device;
1352 struct anv_query_pool *pool;
1353 VkResult result;
1354 size_t size;
1355
1356 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO);
1357
1358 switch (pCreateInfo->queryType) {
1359 case VK_QUERY_TYPE_OCCLUSION:
1360 break;
1361 case VK_QUERY_TYPE_PIPELINE_STATISTICS:
1362 return VK_UNSUPPORTED;
1363 default:
1364 unreachable("");
1365 }
1366
1367 pool = anv_device_alloc(device, sizeof(*pool), 8,
1368 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1369 if (pool == NULL)
1370 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1371
1372 pool->base.destructor = anv_query_pool_destroy;
1373
1374 pool->type = pCreateInfo->queryType;
1375 size = pCreateInfo->slots * sizeof(struct anv_query_pool_slot);
1376 result = anv_bo_init_new(&pool->bo, device, size);
1377 if (result != VK_SUCCESS)
1378 goto fail;
1379
1380 pool->bo.map = anv_gem_mmap(device, pool->bo.gem_handle, 0, size);
1381
1382 *pQueryPool = (VkQueryPool) pool;
1383
1384 return VK_SUCCESS;
1385
1386 fail:
1387 anv_device_free(device, pool);
1388
1389 return result;
1390 }
1391
1392 VkResult anv_GetQueryPoolResults(
1393 VkDevice _device,
1394 VkQueryPool queryPool,
1395 uint32_t startQuery,
1396 uint32_t queryCount,
1397 size_t* pDataSize,
1398 void* pData,
1399 VkQueryResultFlags flags)
1400 {
1401 struct anv_device *device = (struct anv_device *) _device;
1402 struct anv_query_pool *pool = (struct anv_query_pool *) queryPool;
1403 struct anv_query_pool_slot *slot = pool->bo.map;
1404 int64_t timeout = INT64_MAX;
1405 uint32_t *dst32 = pData;
1406 uint64_t *dst64 = pData;
1407 uint64_t result;
1408 int ret;
1409
1410 if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
1411 /* Where is the availabilty info supposed to go? */
1412 anv_finishme("VK_QUERY_RESULT_WITH_AVAILABILITY_BIT");
1413 return VK_UNSUPPORTED;
1414 }
1415
1416 assert(pool->type == VK_QUERY_TYPE_OCCLUSION);
1417
1418 if (flags & VK_QUERY_RESULT_64_BIT)
1419 *pDataSize = queryCount * sizeof(uint64_t);
1420 else
1421 *pDataSize = queryCount * sizeof(uint32_t);
1422
1423 if (pData == NULL)
1424 return VK_SUCCESS;
1425
1426 if (flags & VK_QUERY_RESULT_WAIT_BIT) {
1427 ret = anv_gem_wait(device, pool->bo.gem_handle, &timeout);
1428 if (ret == -1)
1429 return vk_error(VK_ERROR_UNKNOWN);
1430 }
1431
1432 for (uint32_t i = 0; i < queryCount; i++) {
1433 result = slot[startQuery + i].end - slot[startQuery + i].begin;
1434 if (flags & VK_QUERY_RESULT_64_BIT) {
1435 *dst64++ = result;
1436 } else {
1437 if (result > UINT32_MAX)
1438 result = UINT32_MAX;
1439 *dst32++ = result;
1440 }
1441 }
1442
1443 return VK_SUCCESS;
1444 }
1445
1446 // Buffer functions
1447
1448 VkResult anv_CreateBuffer(
1449 VkDevice _device,
1450 const VkBufferCreateInfo* pCreateInfo,
1451 VkBuffer* pBuffer)
1452 {
1453 struct anv_device *device = (struct anv_device *) _device;
1454 struct anv_buffer *buffer;
1455
1456 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO);
1457
1458 buffer = anv_device_alloc(device, sizeof(*buffer), 8,
1459 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1460 if (buffer == NULL)
1461 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1462
1463 buffer->size = pCreateInfo->size;
1464 buffer->bo = NULL;
1465 buffer->offset = 0;
1466
1467 *pBuffer = (VkBuffer) buffer;
1468
1469 return VK_SUCCESS;
1470 }
1471
1472 // Buffer view functions
1473
1474 VkResult anv_CreateBufferView(
1475 VkDevice _device,
1476 const VkBufferViewCreateInfo* pCreateInfo,
1477 VkBufferView* pView)
1478 {
1479 struct anv_device *device = (struct anv_device *) _device;
1480 struct anv_buffer *buffer = (struct anv_buffer *) pCreateInfo->buffer;
1481 struct anv_surface_view *view;
1482 const struct anv_format *format;
1483
1484 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO);
1485
1486 view = anv_device_alloc(device, sizeof(*view), 8,
1487 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1488 if (view == NULL)
1489 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1490
1491 view->bo = buffer->bo;
1492 view->offset = buffer->offset + pCreateInfo->offset;
1493 view->surface_state =
1494 anv_state_pool_alloc(&device->surface_state_pool, 64, 64);
1495 view->format = pCreateInfo->format;
1496
1497 format = anv_format_for_vk_format(pCreateInfo->format);
1498 /* This assumes RGBA float format. */
1499 uint32_t stride = 4;
1500 uint32_t num_elements = pCreateInfo->range / stride;
1501 struct GEN8_RENDER_SURFACE_STATE surface_state = {
1502 .SurfaceType = SURFTYPE_BUFFER,
1503 .SurfaceArray = false,
1504 .SurfaceFormat = format->format,
1505 .SurfaceVerticalAlignment = VALIGN4,
1506 .SurfaceHorizontalAlignment = HALIGN4,
1507 .TileMode = LINEAR,
1508 .VerticalLineStride = 0,
1509 .VerticalLineStrideOffset = 0,
1510 .SamplerL2BypassModeDisable = true,
1511 .RenderCacheReadWriteMode = WriteOnlyCache,
1512 .MemoryObjectControlState = GEN8_MOCS,
1513 .BaseMipLevel = 0,
1514 .SurfaceQPitch = 0,
1515 .Height = (num_elements >> 7) & 0x3fff,
1516 .Width = num_elements & 0x7f,
1517 .Depth = (num_elements >> 21) & 0x3f,
1518 .SurfacePitch = stride - 1,
1519 .MinimumArrayElement = 0,
1520 .NumberofMultisamples = MULTISAMPLECOUNT_1,
1521 .XOffset = 0,
1522 .YOffset = 0,
1523 .SurfaceMinLOD = 0,
1524 .MIPCountLOD = 0,
1525 .AuxiliarySurfaceMode = AUX_NONE,
1526 .RedClearColor = 0,
1527 .GreenClearColor = 0,
1528 .BlueClearColor = 0,
1529 .AlphaClearColor = 0,
1530 .ShaderChannelSelectRed = SCS_RED,
1531 .ShaderChannelSelectGreen = SCS_GREEN,
1532 .ShaderChannelSelectBlue = SCS_BLUE,
1533 .ShaderChannelSelectAlpha = SCS_ALPHA,
1534 .ResourceMinLOD = 0,
1535 /* FIXME: We assume that the image must be bound at this time. */
1536 .SurfaceBaseAddress = { NULL, view->offset },
1537 };
1538
1539 GEN8_RENDER_SURFACE_STATE_pack(NULL, view->surface_state.map, &surface_state);
1540
1541 *pView = (VkImageView) view;
1542
1543 return VK_SUCCESS;
1544 }
1545
1546 // Sampler functions
1547
1548 VkResult anv_CreateSampler(
1549 VkDevice _device,
1550 const VkSamplerCreateInfo* pCreateInfo,
1551 VkSampler* pSampler)
1552 {
1553 struct anv_device *device = (struct anv_device *) _device;
1554 struct anv_sampler *sampler;
1555
1556 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
1557
1558 sampler = anv_device_alloc(device, sizeof(*sampler), 8,
1559 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1560 if (!sampler)
1561 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1562
1563 static const uint32_t vk_to_gen_tex_filter[] = {
1564 [VK_TEX_FILTER_NEAREST] = MAPFILTER_NEAREST,
1565 [VK_TEX_FILTER_LINEAR] = MAPFILTER_LINEAR
1566 };
1567
1568 static const uint32_t vk_to_gen_mipmap_mode[] = {
1569 [VK_TEX_MIPMAP_MODE_BASE] = MIPFILTER_NONE,
1570 [VK_TEX_MIPMAP_MODE_NEAREST] = MIPFILTER_NEAREST,
1571 [VK_TEX_MIPMAP_MODE_LINEAR] = MIPFILTER_LINEAR
1572 };
1573
1574 static const uint32_t vk_to_gen_tex_address[] = {
1575 [VK_TEX_ADDRESS_WRAP] = TCM_WRAP,
1576 [VK_TEX_ADDRESS_MIRROR] = TCM_MIRROR,
1577 [VK_TEX_ADDRESS_CLAMP] = TCM_CLAMP,
1578 [VK_TEX_ADDRESS_MIRROR_ONCE] = TCM_MIRROR_ONCE,
1579 [VK_TEX_ADDRESS_CLAMP_BORDER] = TCM_CLAMP_BORDER,
1580 };
1581
1582 static const uint32_t vk_to_gen_compare_op[] = {
1583 [VK_COMPARE_OP_NEVER] = PREFILTEROPNEVER,
1584 [VK_COMPARE_OP_LESS] = PREFILTEROPLESS,
1585 [VK_COMPARE_OP_EQUAL] = PREFILTEROPEQUAL,
1586 [VK_COMPARE_OP_LESS_EQUAL] = PREFILTEROPLEQUAL,
1587 [VK_COMPARE_OP_GREATER] = PREFILTEROPGREATER,
1588 [VK_COMPARE_OP_NOT_EQUAL] = PREFILTEROPNOTEQUAL,
1589 [VK_COMPARE_OP_GREATER_EQUAL] = PREFILTEROPGEQUAL,
1590 [VK_COMPARE_OP_ALWAYS] = PREFILTEROPALWAYS,
1591 };
1592
1593 if (pCreateInfo->maxAnisotropy > 0)
1594 anv_finishme("missing support for anisotropic filtering");
1595
1596 struct GEN8_SAMPLER_STATE sampler_state = {
1597 .SamplerDisable = false,
1598 .TextureBorderColorMode = DX10OGL,
1599 .LODPreClampMode = 0,
1600 .BaseMipLevel = 0,
1601 .MipModeFilter = vk_to_gen_mipmap_mode[pCreateInfo->mipMode],
1602 .MagModeFilter = vk_to_gen_tex_filter[pCreateInfo->magFilter],
1603 .MinModeFilter = vk_to_gen_tex_filter[pCreateInfo->minFilter],
1604 .TextureLODBias = pCreateInfo->mipLodBias * 256,
1605 .AnisotropicAlgorithm = EWAApproximation,
1606 .MinLOD = pCreateInfo->minLod * 256,
1607 .MaxLOD = pCreateInfo->maxLod * 256,
1608 .ChromaKeyEnable = 0,
1609 .ChromaKeyIndex = 0,
1610 .ChromaKeyMode = 0,
1611 .ShadowFunction = vk_to_gen_compare_op[pCreateInfo->compareOp],
1612 .CubeSurfaceControlMode = 0,
1613 .IndirectStatePointer = 0,
1614 .LODClampMagnificationMode = MIPNONE,
1615 .MaximumAnisotropy = 0,
1616 .RAddressMinFilterRoundingEnable = 0,
1617 .RAddressMagFilterRoundingEnable = 0,
1618 .VAddressMinFilterRoundingEnable = 0,
1619 .VAddressMagFilterRoundingEnable = 0,
1620 .UAddressMinFilterRoundingEnable = 0,
1621 .UAddressMagFilterRoundingEnable = 0,
1622 .TrilinearFilterQuality = 0,
1623 .NonnormalizedCoordinateEnable = 0,
1624 .TCXAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressU],
1625 .TCYAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressV],
1626 .TCZAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressW],
1627 };
1628
1629 GEN8_SAMPLER_STATE_pack(NULL, sampler->state, &sampler_state);
1630
1631 *pSampler = (VkSampler) sampler;
1632
1633 return VK_SUCCESS;
1634 }
1635
1636 // Descriptor set functions
1637
1638 VkResult anv_CreateDescriptorSetLayout(
1639 VkDevice _device,
1640 const VkDescriptorSetLayoutCreateInfo* pCreateInfo,
1641 VkDescriptorSetLayout* pSetLayout)
1642 {
1643 struct anv_device *device = (struct anv_device *) _device;
1644 struct anv_descriptor_set_layout *set_layout;
1645
1646 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO);
1647
1648 uint32_t sampler_count[VK_NUM_SHADER_STAGE] = { 0, };
1649 uint32_t surface_count[VK_NUM_SHADER_STAGE] = { 0, };
1650 uint32_t num_dynamic_buffers = 0;
1651 uint32_t count = 0;
1652 uint32_t s;
1653
1654 for (uint32_t i = 0; i < pCreateInfo->count; i++) {
1655 switch (pCreateInfo->pBinding[i].descriptorType) {
1656 case VK_DESCRIPTOR_TYPE_SAMPLER:
1657 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1658 sampler_count[s] += pCreateInfo->pBinding[i].count;
1659 break;
1660
1661 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1662 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1663 sampler_count[s] += pCreateInfo->pBinding[i].count;
1664
1665 /* fall through */
1666
1667 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
1668 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
1669 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
1670 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
1671 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
1672 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
1673 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1674 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1675 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1676 surface_count[s] += pCreateInfo->pBinding[i].count;
1677 break;
1678 default:
1679 break;
1680 }
1681
1682 count += pCreateInfo->pBinding[i].count;
1683 }
1684
1685 for (uint32_t i = 0; i < pCreateInfo->count; i++) {
1686 switch (pCreateInfo->pBinding[i].descriptorType) {
1687 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1688 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1689 num_dynamic_buffers++;
1690 break;
1691 default:
1692 break;
1693 }
1694 }
1695
1696 uint32_t sampler_total = 0;
1697 uint32_t surface_total = 0;
1698 for (uint32_t s = 0; s < VK_NUM_SHADER_STAGE; s++) {
1699 sampler_total += sampler_count[s];
1700 surface_total += surface_count[s];
1701 }
1702
1703 size_t size = sizeof(*set_layout) +
1704 (sampler_total + surface_total) * sizeof(uint32_t);
1705 set_layout = anv_device_alloc(device, size, 8,
1706 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1707 if (!set_layout)
1708 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1709
1710 set_layout->num_dynamic_buffers = num_dynamic_buffers;
1711 set_layout->count = count;
1712
1713 uint32_t *p = set_layout->entries;
1714 uint32_t *sampler[VK_NUM_SHADER_STAGE];
1715 uint32_t *surface[VK_NUM_SHADER_STAGE];
1716 for (uint32_t s = 0; s < VK_NUM_SHADER_STAGE; s++) {
1717 set_layout->stage[s].surface_count = surface_count[s];
1718 set_layout->stage[s].surface_start = surface[s] = p;
1719 p += surface_count[s];
1720 set_layout->stage[s].sampler_count = sampler_count[s];
1721 set_layout->stage[s].sampler_start = sampler[s] = p;
1722 p += sampler_count[s];
1723 }
1724
1725 uint32_t descriptor = 0;
1726 for (uint32_t i = 0; i < pCreateInfo->count; i++) {
1727 switch (pCreateInfo->pBinding[i].descriptorType) {
1728 case VK_DESCRIPTOR_TYPE_SAMPLER:
1729 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1730 for (uint32_t j = 0; j < pCreateInfo->pBinding[i].count; j++)
1731 *(sampler[s])++ = descriptor + j;
1732 break;
1733
1734 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1735 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1736 for (uint32_t j = 0; j < pCreateInfo->pBinding[i].count; j++)
1737 *(sampler[s])++ = descriptor + j;
1738
1739 /* fallthrough */
1740
1741 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
1742 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
1743 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
1744 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
1745 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
1746 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
1747 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1748 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1749 for_each_bit(s, pCreateInfo->pBinding[i].stageFlags)
1750 for (uint32_t j = 0; j < pCreateInfo->pBinding[i].count; j++) {
1751 *(surface[s])++ = descriptor + j;
1752 }
1753 break;
1754 default:
1755 unreachable("");
1756 }
1757 descriptor += pCreateInfo->pBinding[i].count;
1758 }
1759
1760 *pSetLayout = (VkDescriptorSetLayout) set_layout;
1761
1762 return VK_SUCCESS;
1763 }
1764
1765 VkResult anv_BeginDescriptorPoolUpdate(
1766 VkDevice device,
1767 VkDescriptorUpdateMode updateMode)
1768 {
1769 return VK_SUCCESS;
1770 }
1771
1772 VkResult anv_EndDescriptorPoolUpdate(
1773 VkDevice device,
1774 VkCmdBuffer cmd)
1775 {
1776 return VK_SUCCESS;
1777 }
1778
1779 VkResult anv_CreateDescriptorPool(
1780 VkDevice device,
1781 VkDescriptorPoolUsage poolUsage,
1782 uint32_t maxSets,
1783 const VkDescriptorPoolCreateInfo* pCreateInfo,
1784 VkDescriptorPool* pDescriptorPool)
1785 {
1786 *pDescriptorPool = 1;
1787
1788 return VK_SUCCESS;
1789 }
1790
1791 VkResult anv_ResetDescriptorPool(
1792 VkDevice device,
1793 VkDescriptorPool descriptorPool)
1794 {
1795 return VK_SUCCESS;
1796 }
1797
1798 VkResult anv_AllocDescriptorSets(
1799 VkDevice _device,
1800 VkDescriptorPool descriptorPool,
1801 VkDescriptorSetUsage setUsage,
1802 uint32_t count,
1803 const VkDescriptorSetLayout* pSetLayouts,
1804 VkDescriptorSet* pDescriptorSets,
1805 uint32_t* pCount)
1806 {
1807 struct anv_device *device = (struct anv_device *) _device;
1808 const struct anv_descriptor_set_layout *layout;
1809 struct anv_descriptor_set *set;
1810 size_t size;
1811
1812 for (uint32_t i = 0; i < count; i++) {
1813 layout = (struct anv_descriptor_set_layout *) pSetLayouts[i];
1814 size = sizeof(*set) + layout->count * sizeof(set->descriptors[0]);
1815 set = anv_device_alloc(device, size, 8,
1816 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1817 if (!set) {
1818 *pCount = i;
1819 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1820 }
1821
1822 /* Descriptor sets may not be 100% filled out so we need to memset to
1823 * ensure that we can properly detect and handle holes.
1824 */
1825 memset(set, 0, size);
1826
1827 pDescriptorSets[i] = (VkDescriptorSet) set;
1828 }
1829
1830 *pCount = count;
1831
1832 return VK_SUCCESS;
1833 }
1834
1835 void anv_ClearDescriptorSets(
1836 VkDevice device,
1837 VkDescriptorPool descriptorPool,
1838 uint32_t count,
1839 const VkDescriptorSet* pDescriptorSets)
1840 {
1841 }
1842
1843 void anv_UpdateDescriptors(
1844 VkDevice _device,
1845 VkDescriptorSet descriptorSet,
1846 uint32_t updateCount,
1847 const void** ppUpdateArray)
1848 {
1849 struct anv_descriptor_set *set = (struct anv_descriptor_set *) descriptorSet;
1850 VkUpdateSamplers *update_samplers;
1851 VkUpdateSamplerTextures *update_sampler_textures;
1852 VkUpdateImages *update_images;
1853 VkUpdateBuffers *update_buffers;
1854 VkUpdateAsCopy *update_as_copy;
1855
1856 for (uint32_t i = 0; i < updateCount; i++) {
1857 const struct anv_common *common = ppUpdateArray[i];
1858
1859 switch (common->sType) {
1860 case VK_STRUCTURE_TYPE_UPDATE_SAMPLERS:
1861 update_samplers = (VkUpdateSamplers *) common;
1862
1863 for (uint32_t j = 0; j < update_samplers->count; j++) {
1864 set->descriptors[update_samplers->binding + j].sampler =
1865 (struct anv_sampler *) update_samplers->pSamplers[j];
1866 }
1867 break;
1868
1869 case VK_STRUCTURE_TYPE_UPDATE_SAMPLER_TEXTURES:
1870 /* FIXME: Shouldn't this be *_UPDATE_SAMPLER_IMAGES? */
1871 update_sampler_textures = (VkUpdateSamplerTextures *) common;
1872
1873 for (uint32_t j = 0; j < update_sampler_textures->count; j++) {
1874 set->descriptors[update_sampler_textures->binding + j].view =
1875 (struct anv_surface_view *)
1876 update_sampler_textures->pSamplerImageViews[j].pImageView->view;
1877 set->descriptors[update_sampler_textures->binding + j].sampler =
1878 (struct anv_sampler *)
1879 update_sampler_textures->pSamplerImageViews[j].sampler;
1880 }
1881 break;
1882
1883 case VK_STRUCTURE_TYPE_UPDATE_IMAGES:
1884 update_images = (VkUpdateImages *) common;
1885
1886 for (uint32_t j = 0; j < update_images->count; j++) {
1887 set->descriptors[update_images->binding + j].view =
1888 (struct anv_surface_view *) update_images->pImageViews[j].view;
1889 }
1890 break;
1891
1892 case VK_STRUCTURE_TYPE_UPDATE_BUFFERS:
1893 update_buffers = (VkUpdateBuffers *) common;
1894
1895 for (uint32_t j = 0; j < update_buffers->count; j++) {
1896 set->descriptors[update_buffers->binding + j].view =
1897 (struct anv_surface_view *) update_buffers->pBufferViews[j].view;
1898 }
1899 /* FIXME: descriptor arrays? */
1900 break;
1901
1902 case VK_STRUCTURE_TYPE_UPDATE_AS_COPY:
1903 update_as_copy = (VkUpdateAsCopy *) common;
1904 (void) update_as_copy;
1905 break;
1906
1907 default:
1908 break;
1909 }
1910 }
1911 }
1912
1913 // State object functions
1914
1915 static inline int64_t
1916 clamp_int64(int64_t x, int64_t min, int64_t max)
1917 {
1918 if (x < min)
1919 return min;
1920 else if (x < max)
1921 return x;
1922 else
1923 return max;
1924 }
1925
1926 static void
1927 anv_dynamic_vp_state_destroy(struct anv_device *device,
1928 struct anv_object *object,
1929 VkObjectType obj_type)
1930 {
1931 struct anv_dynamic_vp_state *state = (void *)object;
1932
1933 assert(obj_type == VK_OBJECT_TYPE_DYNAMIC_VP_STATE);
1934
1935 anv_state_pool_free(&device->dynamic_state_pool, state->sf_clip_vp);
1936 anv_state_pool_free(&device->dynamic_state_pool, state->cc_vp);
1937 anv_state_pool_free(&device->dynamic_state_pool, state->scissor);
1938
1939 anv_device_free(device, state);
1940 }
1941
1942 VkResult anv_CreateDynamicViewportState(
1943 VkDevice _device,
1944 const VkDynamicVpStateCreateInfo* pCreateInfo,
1945 VkDynamicVpState* pState)
1946 {
1947 struct anv_device *device = (struct anv_device *) _device;
1948 struct anv_dynamic_vp_state *state;
1949
1950 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO);
1951
1952 state = anv_device_alloc(device, sizeof(*state), 8,
1953 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
1954 if (state == NULL)
1955 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1956
1957 state->base.destructor = anv_dynamic_vp_state_destroy;
1958
1959 unsigned count = pCreateInfo->viewportAndScissorCount;
1960 state->sf_clip_vp = anv_state_pool_alloc(&device->dynamic_state_pool,
1961 count * 64, 64);
1962 state->cc_vp = anv_state_pool_alloc(&device->dynamic_state_pool,
1963 count * 8, 32);
1964 state->scissor = anv_state_pool_alloc(&device->dynamic_state_pool,
1965 count * 32, 32);
1966
1967 for (uint32_t i = 0; i < pCreateInfo->viewportAndScissorCount; i++) {
1968 const VkViewport *vp = &pCreateInfo->pViewports[i];
1969 const VkRect *s = &pCreateInfo->pScissors[i];
1970
1971 struct GEN8_SF_CLIP_VIEWPORT sf_clip_viewport = {
1972 .ViewportMatrixElementm00 = vp->width / 2,
1973 .ViewportMatrixElementm11 = vp->height / 2,
1974 .ViewportMatrixElementm22 = (vp->maxDepth - vp->minDepth) / 2,
1975 .ViewportMatrixElementm30 = vp->originX + vp->width / 2,
1976 .ViewportMatrixElementm31 = vp->originY + vp->height / 2,
1977 .ViewportMatrixElementm32 = (vp->maxDepth + vp->minDepth) / 2,
1978 .XMinClipGuardband = -1.0f,
1979 .XMaxClipGuardband = 1.0f,
1980 .YMinClipGuardband = -1.0f,
1981 .YMaxClipGuardband = 1.0f,
1982 .XMinViewPort = vp->originX,
1983 .XMaxViewPort = vp->originX + vp->width - 1,
1984 .YMinViewPort = vp->originY,
1985 .YMaxViewPort = vp->originY + vp->height - 1,
1986 };
1987
1988 struct GEN8_CC_VIEWPORT cc_viewport = {
1989 .MinimumDepth = vp->minDepth,
1990 .MaximumDepth = vp->maxDepth
1991 };
1992
1993 /* Since xmax and ymax are inclusive, we have to have xmax < xmin or
1994 * ymax < ymin for empty clips. In case clip x, y, width height are all
1995 * 0, the clamps below produce 0 for xmin, ymin, xmax, ymax, which isn't
1996 * what we want. Just special case empty clips and produce a canonical
1997 * empty clip. */
1998 static const struct GEN8_SCISSOR_RECT empty_scissor = {
1999 .ScissorRectangleYMin = 1,
2000 .ScissorRectangleXMin = 1,
2001 .ScissorRectangleYMax = 0,
2002 .ScissorRectangleXMax = 0
2003 };
2004
2005 const int max = 0xffff;
2006 struct GEN8_SCISSOR_RECT scissor = {
2007 /* Do this math using int64_t so overflow gets clamped correctly. */
2008 .ScissorRectangleYMin = clamp_int64(s->offset.y, 0, max),
2009 .ScissorRectangleXMin = clamp_int64(s->offset.x, 0, max),
2010 .ScissorRectangleYMax = clamp_int64((uint64_t) s->offset.y + s->extent.height - 1, 0, max),
2011 .ScissorRectangleXMax = clamp_int64((uint64_t) s->offset.x + s->extent.width - 1, 0, max)
2012 };
2013
2014 GEN8_SF_CLIP_VIEWPORT_pack(NULL, state->sf_clip_vp.map + i * 64, &sf_clip_viewport);
2015 GEN8_CC_VIEWPORT_pack(NULL, state->cc_vp.map + i * 32, &cc_viewport);
2016
2017 if (s->extent.width <= 0 || s->extent.height <= 0) {
2018 GEN8_SCISSOR_RECT_pack(NULL, state->scissor.map + i * 32, &empty_scissor);
2019 } else {
2020 GEN8_SCISSOR_RECT_pack(NULL, state->scissor.map + i * 32, &scissor);
2021 }
2022 }
2023
2024 *pState = (VkDynamicVpState) state;
2025
2026 return VK_SUCCESS;
2027 }
2028
2029 VkResult anv_CreateDynamicRasterState(
2030 VkDevice _device,
2031 const VkDynamicRsStateCreateInfo* pCreateInfo,
2032 VkDynamicRsState* pState)
2033 {
2034 struct anv_device *device = (struct anv_device *) _device;
2035 struct anv_dynamic_rs_state *state;
2036
2037 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_RS_STATE_CREATE_INFO);
2038
2039 state = anv_device_alloc(device, sizeof(*state), 8,
2040 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2041 if (state == NULL)
2042 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2043
2044 /* Missing these:
2045 * float pointFadeThreshold;
2046 * // optional (GL45) - Size of point fade threshold
2047 */
2048
2049 struct GEN8_3DSTATE_SF sf = {
2050 GEN8_3DSTATE_SF_header,
2051 .LineWidth = pCreateInfo->lineWidth,
2052 .PointWidth = pCreateInfo->pointSize,
2053 };
2054
2055 GEN8_3DSTATE_SF_pack(NULL, state->state_sf, &sf);
2056
2057 bool enable_bias = pCreateInfo->depthBias != 0.0f ||
2058 pCreateInfo->slopeScaledDepthBias != 0.0f;
2059 struct GEN8_3DSTATE_RASTER raster = {
2060 .GlobalDepthOffsetEnableSolid = enable_bias,
2061 .GlobalDepthOffsetEnableWireframe = enable_bias,
2062 .GlobalDepthOffsetEnablePoint = enable_bias,
2063 .GlobalDepthOffsetConstant = pCreateInfo->depthBias,
2064 .GlobalDepthOffsetScale = pCreateInfo->slopeScaledDepthBias,
2065 .GlobalDepthOffsetClamp = pCreateInfo->depthBiasClamp
2066 };
2067
2068 GEN8_3DSTATE_RASTER_pack(NULL, state->state_raster, &raster);
2069
2070 *pState = (VkDynamicRsState) state;
2071
2072 return VK_SUCCESS;
2073 }
2074
2075 VkResult anv_CreateDynamicColorBlendState(
2076 VkDevice _device,
2077 const VkDynamicCbStateCreateInfo* pCreateInfo,
2078 VkDynamicCbState* pState)
2079 {
2080 struct anv_device *device = (struct anv_device *) _device;
2081 struct anv_dynamic_cb_state *state;
2082
2083 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_CB_STATE_CREATE_INFO);
2084
2085 state = anv_device_alloc(device, sizeof(*state), 8,
2086 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2087 if (state == NULL)
2088 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2089
2090 struct GEN8_COLOR_CALC_STATE color_calc_state = {
2091 .BlendConstantColorRed = pCreateInfo->blendConst[0],
2092 .BlendConstantColorGreen = pCreateInfo->blendConst[1],
2093 .BlendConstantColorBlue = pCreateInfo->blendConst[2],
2094 .BlendConstantColorAlpha = pCreateInfo->blendConst[3]
2095 };
2096
2097 GEN8_COLOR_CALC_STATE_pack(NULL, state->state_color_calc, &color_calc_state);
2098
2099 *pState = (VkDynamicCbState) state;
2100
2101 return VK_SUCCESS;
2102 }
2103
2104 VkResult anv_CreateDynamicDepthStencilState(
2105 VkDevice _device,
2106 const VkDynamicDsStateCreateInfo* pCreateInfo,
2107 VkDynamicDsState* pState)
2108 {
2109 struct anv_device *device = (struct anv_device *) _device;
2110 struct anv_dynamic_ds_state *state;
2111
2112 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DYNAMIC_DS_STATE_CREATE_INFO);
2113
2114 state = anv_device_alloc(device, sizeof(*state), 8,
2115 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2116 if (state == NULL)
2117 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2118
2119 struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
2120 GEN8_3DSTATE_WM_DEPTH_STENCIL_header,
2121
2122 /* Is this what we need to do? */
2123 .StencilBufferWriteEnable = pCreateInfo->stencilWriteMask != 0,
2124
2125 .StencilTestMask = pCreateInfo->stencilReadMask,
2126 .StencilWriteMask = pCreateInfo->stencilWriteMask,
2127
2128 .BackfaceStencilTestMask = pCreateInfo->stencilReadMask,
2129 .BackfaceStencilWriteMask = pCreateInfo->stencilWriteMask,
2130 };
2131
2132 GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, state->state_wm_depth_stencil,
2133 &wm_depth_stencil);
2134
2135 struct GEN8_COLOR_CALC_STATE color_calc_state = {
2136 .StencilReferenceValue = pCreateInfo->stencilFrontRef,
2137 .BackFaceStencilReferenceValue = pCreateInfo->stencilBackRef
2138 };
2139
2140 GEN8_COLOR_CALC_STATE_pack(NULL, state->state_color_calc, &color_calc_state);
2141
2142 *pState = (VkDynamicDsState) state;
2143
2144 return VK_SUCCESS;
2145 }
2146
2147 // Command buffer functions
2148
2149 static void
2150 anv_cmd_buffer_destroy(struct anv_device *device,
2151 struct anv_object *object,
2152 VkObjectType obj_type)
2153 {
2154 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) object;
2155
2156 assert(obj_type == VK_OBJECT_TYPE_COMMAND_BUFFER);
2157
2158 anv_gem_munmap(cmd_buffer->surface_bo.map, BATCH_SIZE);
2159 anv_gem_close(device, cmd_buffer->surface_bo.gem_handle);
2160 anv_state_stream_finish(&cmd_buffer->surface_state_stream);
2161 anv_state_stream_finish(&cmd_buffer->dynamic_state_stream);
2162 anv_state_stream_finish(&cmd_buffer->binding_table_state_stream);
2163 anv_batch_finish(&cmd_buffer->batch, device);
2164 anv_device_free(device, cmd_buffer->exec2_objects);
2165 anv_device_free(device, cmd_buffer->exec2_bos);
2166 anv_device_free(device, cmd_buffer);
2167 }
2168
2169 VkResult anv_CreateCommandBuffer(
2170 VkDevice _device,
2171 const VkCmdBufferCreateInfo* pCreateInfo,
2172 VkCmdBuffer* pCmdBuffer)
2173 {
2174 struct anv_device *device = (struct anv_device *) _device;
2175 struct anv_cmd_buffer *cmd_buffer;
2176 VkResult result;
2177
2178 cmd_buffer = anv_device_alloc(device, sizeof(*cmd_buffer), 8,
2179 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2180 if (cmd_buffer == NULL)
2181 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2182
2183 cmd_buffer->base.destructor = anv_cmd_buffer_destroy;
2184
2185 cmd_buffer->device = device;
2186 cmd_buffer->rs_state = NULL;
2187 cmd_buffer->vp_state = NULL;
2188 memset(&cmd_buffer->default_bindings, 0, sizeof(cmd_buffer->default_bindings));
2189 cmd_buffer->bindings = &cmd_buffer->default_bindings;
2190
2191 result = anv_batch_init(&cmd_buffer->batch, device);
2192 if (result != VK_SUCCESS)
2193 goto fail;
2194
2195 result = anv_bo_init_new(&cmd_buffer->surface_bo, device, BATCH_SIZE);
2196 if (result != VK_SUCCESS)
2197 goto fail_batch;
2198
2199 cmd_buffer->surface_bo.map =
2200 anv_gem_mmap(device, cmd_buffer->surface_bo.gem_handle, 0, BATCH_SIZE);
2201 if (cmd_buffer->surface_bo.map == NULL) {
2202 result = vk_error(VK_ERROR_MEMORY_MAP_FAILED);
2203 goto fail_surface_bo;
2204 }
2205
2206 /* Start surface_next at 1 so surface offset 0 is invalid. */
2207 cmd_buffer->surface_next = 1;
2208 cmd_buffer->surface_relocs.num_relocs = 0;
2209
2210 cmd_buffer->exec2_objects =
2211 anv_device_alloc(device, 8192 * sizeof(cmd_buffer->exec2_objects[0]), 8,
2212 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2213 if (cmd_buffer->exec2_objects == NULL) {
2214 result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2215 goto fail_surface_map;
2216 }
2217
2218 cmd_buffer->exec2_bos =
2219 anv_device_alloc(device, 8192 * sizeof(cmd_buffer->exec2_bos[0]), 8,
2220 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
2221 if (cmd_buffer->exec2_bos == NULL) {
2222 result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
2223 goto fail_exec2_objects;
2224 }
2225
2226 anv_state_stream_init(&cmd_buffer->binding_table_state_stream,
2227 &device->binding_table_block_pool);
2228 anv_state_stream_init(&cmd_buffer->surface_state_stream,
2229 &device->surface_state_block_pool);
2230 anv_state_stream_init(&cmd_buffer->dynamic_state_stream,
2231 &device->dynamic_state_block_pool);
2232
2233 cmd_buffer->dirty = 0;
2234 cmd_buffer->vb_dirty = 0;
2235 cmd_buffer->pipeline = NULL;
2236
2237 *pCmdBuffer = (VkCmdBuffer) cmd_buffer;
2238
2239 return VK_SUCCESS;
2240
2241 fail_exec2_objects:
2242 anv_device_free(device, cmd_buffer->exec2_objects);
2243 fail_surface_map:
2244 anv_gem_munmap(cmd_buffer->surface_bo.map, BATCH_SIZE);
2245 fail_surface_bo:
2246 anv_gem_close(device, cmd_buffer->surface_bo.gem_handle);
2247 fail_batch:
2248 anv_batch_finish(&cmd_buffer->batch, device);
2249 fail:
2250 anv_device_free(device, cmd_buffer);
2251
2252 return result;
2253 }
2254
2255 VkResult anv_BeginCommandBuffer(
2256 VkCmdBuffer cmdBuffer,
2257 const VkCmdBufferBeginInfo* pBeginInfo)
2258 {
2259 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2260 struct anv_device *device = cmd_buffer->device;
2261
2262 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPELINE_SELECT,
2263 .PipelineSelection = _3D);
2264 anv_batch_emit(&cmd_buffer->batch, GEN8_STATE_SIP);
2265
2266 anv_batch_emit(&cmd_buffer->batch, GEN8_STATE_BASE_ADDRESS,
2267 .GeneralStateBaseAddress = { NULL, 0 },
2268 .GeneralStateMemoryObjectControlState = GEN8_MOCS,
2269 .GeneralStateBaseAddressModifyEnable = true,
2270 .GeneralStateBufferSize = 0xfffff,
2271 .GeneralStateBufferSizeModifyEnable = true,
2272
2273 .SurfaceStateBaseAddress = { &cmd_buffer->surface_bo, 0 },
2274 .SurfaceStateMemoryObjectControlState = GEN8_MOCS,
2275 .SurfaceStateBaseAddressModifyEnable = true,
2276
2277 .DynamicStateBaseAddress = { &device->dynamic_state_block_pool.bo, 0 },
2278 .DynamicStateMemoryObjectControlState = GEN8_MOCS,
2279 .DynamicStateBaseAddressModifyEnable = true,
2280 .DynamicStateBufferSize = 0xfffff,
2281 .DynamicStateBufferSizeModifyEnable = true,
2282
2283 .IndirectObjectBaseAddress = { NULL, 0 },
2284 .IndirectObjectMemoryObjectControlState = GEN8_MOCS,
2285 .IndirectObjectBaseAddressModifyEnable = true,
2286 .IndirectObjectBufferSize = 0xfffff,
2287 .IndirectObjectBufferSizeModifyEnable = true,
2288
2289 .InstructionBaseAddress = { &device->instruction_block_pool.bo, 0 },
2290 .InstructionMemoryObjectControlState = GEN8_MOCS,
2291 .InstructionBaseAddressModifyEnable = true,
2292 .InstructionBufferSize = 0xfffff,
2293 .InstructionBuffersizeModifyEnable = true);
2294
2295 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VF_STATISTICS,
2296 .StatisticsEnable = true);
2297 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_HS, .Enable = false);
2298 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_TE, .TEEnable = false);
2299 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
2300 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
2301
2302 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
2303 .ConstantBufferOffset = 0,
2304 .ConstantBufferSize = 4);
2305 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
2306 .ConstantBufferOffset = 4,
2307 .ConstantBufferSize = 4);
2308 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
2309 .ConstantBufferOffset = 8,
2310 .ConstantBufferSize = 4);
2311
2312 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_WM_CHROMAKEY,
2313 .ChromaKeyKillEnable = false);
2314 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_SBE_SWIZ);
2315 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
2316
2317 return VK_SUCCESS;
2318 }
2319
2320 static void
2321 anv_cmd_buffer_add_bo(struct anv_cmd_buffer *cmd_buffer,
2322 struct anv_bo *bo, struct anv_reloc_list *list)
2323 {
2324 struct drm_i915_gem_exec_object2 *obj;
2325
2326 bo->index = cmd_buffer->bo_count;
2327 obj = &cmd_buffer->exec2_objects[bo->index];
2328 cmd_buffer->exec2_bos[bo->index] = bo;
2329 cmd_buffer->bo_count++;
2330
2331 obj->handle = bo->gem_handle;
2332 obj->relocation_count = 0;
2333 obj->relocs_ptr = 0;
2334 obj->alignment = 0;
2335 obj->offset = bo->offset;
2336 obj->flags = 0;
2337 obj->rsvd1 = 0;
2338 obj->rsvd2 = 0;
2339
2340 if (list) {
2341 obj->relocation_count = list->num_relocs;
2342 obj->relocs_ptr = (uintptr_t) list->relocs;
2343 }
2344 }
2345
2346 static void
2347 anv_cmd_buffer_add_validate_bos(struct anv_cmd_buffer *cmd_buffer,
2348 struct anv_reloc_list *list)
2349 {
2350 struct anv_bo *bo, *batch_bo;
2351
2352 batch_bo = &cmd_buffer->batch.bo;
2353 for (size_t i = 0; i < list->num_relocs; i++) {
2354 bo = list->reloc_bos[i];
2355 /* Skip any relocations targeting the batch bo. We need to make sure
2356 * it's the last in the list so we'll add it manually later.
2357 */
2358 if (bo == batch_bo)
2359 continue;
2360 if (bo->index < cmd_buffer->bo_count && cmd_buffer->exec2_bos[bo->index] == bo)
2361 continue;
2362
2363 anv_cmd_buffer_add_bo(cmd_buffer, bo, NULL);
2364 }
2365 }
2366
2367 static void
2368 anv_cmd_buffer_process_relocs(struct anv_cmd_buffer *cmd_buffer,
2369 struct anv_reloc_list *list)
2370 {
2371 struct anv_bo *bo;
2372
2373 /* If the kernel supports I915_EXEC_NO_RELOC, it will compare offset in
2374 * struct drm_i915_gem_exec_object2 against the bos current offset and if
2375 * all bos haven't moved it will skip relocation processing alltogether.
2376 * If I915_EXEC_NO_RELOC is not supported, the kernel ignores the incoming
2377 * value of offset so we can set it either way. For that to work we need
2378 * to make sure all relocs use the same presumed offset.
2379 */
2380
2381 for (size_t i = 0; i < list->num_relocs; i++) {
2382 bo = list->reloc_bos[i];
2383 if (bo->offset != list->relocs[i].presumed_offset)
2384 cmd_buffer->need_reloc = true;
2385
2386 list->relocs[i].target_handle = bo->index;
2387 }
2388 }
2389
2390 VkResult anv_EndCommandBuffer(
2391 VkCmdBuffer cmdBuffer)
2392 {
2393 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2394 struct anv_device *device = cmd_buffer->device;
2395 struct anv_batch *batch = &cmd_buffer->batch;
2396
2397 anv_batch_emit(batch, GEN8_MI_BATCH_BUFFER_END);
2398
2399 /* Round batch up to an even number of dwords. */
2400 if ((batch->next - batch->bo.map) & 4)
2401 anv_batch_emit(batch, GEN8_MI_NOOP);
2402
2403 cmd_buffer->bo_count = 0;
2404 cmd_buffer->need_reloc = false;
2405
2406 /* Lock for access to bo->index. */
2407 pthread_mutex_lock(&device->mutex);
2408
2409 /* Add block pool bos first so we can add them with their relocs. */
2410 anv_cmd_buffer_add_bo(cmd_buffer, &cmd_buffer->surface_bo,
2411 &cmd_buffer->surface_relocs);
2412
2413 anv_cmd_buffer_add_validate_bos(cmd_buffer, &cmd_buffer->surface_relocs);
2414 anv_cmd_buffer_add_validate_bos(cmd_buffer, &batch->cmd_relocs);
2415 anv_cmd_buffer_add_bo(cmd_buffer, &batch->bo, &batch->cmd_relocs);
2416 anv_cmd_buffer_process_relocs(cmd_buffer, &cmd_buffer->surface_relocs);
2417 anv_cmd_buffer_process_relocs(cmd_buffer, &batch->cmd_relocs);
2418
2419 cmd_buffer->execbuf.buffers_ptr = (uintptr_t) cmd_buffer->exec2_objects;
2420 cmd_buffer->execbuf.buffer_count = cmd_buffer->bo_count;
2421 cmd_buffer->execbuf.batch_start_offset = 0;
2422 cmd_buffer->execbuf.batch_len = batch->next - batch->bo.map;
2423 cmd_buffer->execbuf.cliprects_ptr = 0;
2424 cmd_buffer->execbuf.num_cliprects = 0;
2425 cmd_buffer->execbuf.DR1 = 0;
2426 cmd_buffer->execbuf.DR4 = 0;
2427
2428 cmd_buffer->execbuf.flags = I915_EXEC_HANDLE_LUT;
2429 if (!cmd_buffer->need_reloc)
2430 cmd_buffer->execbuf.flags |= I915_EXEC_NO_RELOC;
2431 cmd_buffer->execbuf.flags |= I915_EXEC_RENDER;
2432 cmd_buffer->execbuf.rsvd1 = device->context_id;
2433 cmd_buffer->execbuf.rsvd2 = 0;
2434
2435 pthread_mutex_unlock(&device->mutex);
2436
2437 return VK_SUCCESS;
2438 }
2439
2440 VkResult anv_ResetCommandBuffer(
2441 VkCmdBuffer cmdBuffer)
2442 {
2443 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2444
2445 anv_batch_reset(&cmd_buffer->batch);
2446 cmd_buffer->surface_next = 0;
2447 cmd_buffer->surface_relocs.num_relocs = 0;
2448
2449 return VK_SUCCESS;
2450 }
2451
2452 // Command buffer building functions
2453
2454 void anv_CmdBindPipeline(
2455 VkCmdBuffer cmdBuffer,
2456 VkPipelineBindPoint pipelineBindPoint,
2457 VkPipeline _pipeline)
2458 {
2459 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2460 struct anv_pipeline *pipeline = (struct anv_pipeline *) _pipeline;
2461
2462 cmd_buffer->pipeline = pipeline;
2463 cmd_buffer->vb_dirty |= pipeline->vb_used;
2464 cmd_buffer->dirty |= ANV_CMD_BUFFER_PIPELINE_DIRTY;
2465 }
2466
2467 void anv_CmdBindDynamicStateObject(
2468 VkCmdBuffer cmdBuffer,
2469 VkStateBindPoint stateBindPoint,
2470 VkDynamicStateObject dynamicState)
2471 {
2472 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2473 struct anv_dynamic_vp_state *vp_state;
2474
2475 switch (stateBindPoint) {
2476 case VK_STATE_BIND_POINT_VIEWPORT:
2477 vp_state = (struct anv_dynamic_vp_state *) dynamicState;
2478 /* We emit state immediately, but set cmd_buffer->vp_state to indicate
2479 * that vp state has been set in this command buffer. */
2480 cmd_buffer->vp_state = vp_state;
2481 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_SCISSOR_STATE_POINTERS,
2482 .ScissorRectPointer = vp_state->scissor.offset);
2483 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_CC,
2484 .CCViewportPointer = vp_state->cc_vp.offset);
2485 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP,
2486 .SFClipViewportPointer = vp_state->sf_clip_vp.offset);
2487 break;
2488 case VK_STATE_BIND_POINT_RASTER:
2489 cmd_buffer->rs_state = (struct anv_dynamic_rs_state *) dynamicState;
2490 cmd_buffer->dirty |= ANV_CMD_BUFFER_RS_DIRTY;
2491 break;
2492 case VK_STATE_BIND_POINT_COLOR_BLEND:
2493 break;
2494 case VK_STATE_BIND_POINT_DEPTH_STENCIL:
2495 cmd_buffer->ds_state = (struct anv_dynamic_ds_state *) dynamicState;
2496 cmd_buffer->dirty |= ANV_CMD_BUFFER_DS_DIRTY;
2497 break;
2498 default:
2499 break;
2500 };
2501 }
2502
2503 static struct anv_state
2504 anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer *cmd_buffer,
2505 uint32_t size, uint32_t alignment)
2506 {
2507 struct anv_state state;
2508
2509 state.offset = ALIGN_U32(cmd_buffer->surface_next, alignment);
2510 state.map = cmd_buffer->surface_bo.map + state.offset;
2511 state.alloc_size = size;
2512 cmd_buffer->surface_next = state.offset + size;
2513
2514 assert(state.offset + size < cmd_buffer->surface_bo.size);
2515
2516 return state;
2517 }
2518
2519 void anv_CmdBindDescriptorSets(
2520 VkCmdBuffer cmdBuffer,
2521 VkPipelineBindPoint pipelineBindPoint,
2522 uint32_t firstSet,
2523 uint32_t setCount,
2524 const VkDescriptorSet* pDescriptorSets,
2525 uint32_t dynamicOffsetCount,
2526 const uint32_t* pDynamicOffsets)
2527 {
2528 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2529 struct anv_pipeline_layout *layout = cmd_buffer->pipeline->layout;
2530 struct anv_bindings *bindings = cmd_buffer->bindings;
2531
2532 uint32_t offset = 0;
2533 for (uint32_t i = 0; i < setCount; i++) {
2534 struct anv_descriptor_set *set =
2535 (struct anv_descriptor_set *) pDescriptorSets[i];
2536 struct anv_descriptor_set_layout *set_layout = layout->set[firstSet + i].layout;
2537
2538 for (uint32_t s = 0; s < VK_NUM_SHADER_STAGE; s++) {
2539 uint32_t *surface_to_desc = set_layout->stage[s].surface_start;
2540 uint32_t *sampler_to_desc = set_layout->stage[s].sampler_start;
2541 uint32_t bias = s == VK_SHADER_STAGE_FRAGMENT ? MAX_RTS : 0;
2542 uint32_t start;
2543
2544 start = bias + layout->set[firstSet + i].surface_start[s];
2545 for (uint32_t b = 0; b < set_layout->stage[s].surface_count; b++) {
2546 struct anv_surface_view *view = set->descriptors[surface_to_desc[b]].view;
2547 if (!view)
2548 continue;
2549
2550 struct anv_state state =
2551 anv_cmd_buffer_alloc_surface_state(cmd_buffer, 64, 64);
2552 memcpy(state.map, view->surface_state.map, 64);
2553
2554 /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
2555 *(uint64_t *)(state.map + 8 * 4) =
2556 anv_reloc_list_add(&cmd_buffer->surface_relocs,
2557 state.offset + 8 * 4,
2558 view->bo, view->offset);
2559
2560 bindings->descriptors[s].surfaces[start + b] = state.offset;
2561 }
2562
2563 start = layout->set[firstSet + i].sampler_start[s];
2564 for (uint32_t b = 0; b < set_layout->stage[s].sampler_count; b++) {
2565 struct anv_sampler *sampler = set->descriptors[sampler_to_desc[b]].sampler;
2566 if (!sampler)
2567 continue;
2568
2569 memcpy(&bindings->descriptors[s].samplers[start + b],
2570 sampler->state, sizeof(sampler->state));
2571 }
2572 }
2573
2574 offset += layout->set[firstSet + i].layout->num_dynamic_buffers;
2575 }
2576
2577 cmd_buffer->dirty |= ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY;
2578 }
2579
2580 void anv_CmdBindIndexBuffer(
2581 VkCmdBuffer cmdBuffer,
2582 VkBuffer _buffer,
2583 VkDeviceSize offset,
2584 VkIndexType indexType)
2585 {
2586 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2587 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
2588
2589 static const uint32_t vk_to_gen_index_type[] = {
2590 [VK_INDEX_TYPE_UINT8] = INDEX_BYTE,
2591 [VK_INDEX_TYPE_UINT16] = INDEX_WORD,
2592 [VK_INDEX_TYPE_UINT32] = INDEX_DWORD,
2593 };
2594
2595 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_INDEX_BUFFER,
2596 .IndexFormat = vk_to_gen_index_type[indexType],
2597 .MemoryObjectControlState = GEN8_MOCS,
2598 .BufferStartingAddress = { buffer->bo, buffer->offset + offset },
2599 .BufferSize = buffer->size - offset);
2600 }
2601
2602 void anv_CmdBindVertexBuffers(
2603 VkCmdBuffer cmdBuffer,
2604 uint32_t startBinding,
2605 uint32_t bindingCount,
2606 const VkBuffer* pBuffers,
2607 const VkDeviceSize* pOffsets)
2608 {
2609 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2610 struct anv_bindings *bindings = cmd_buffer->bindings;
2611
2612 /* We have to defer setting up vertex buffer since we need the buffer
2613 * stride from the pipeline. */
2614
2615 for (uint32_t i = 0; i < bindingCount; i++) {
2616 bindings->vb[startBinding + i].buffer = (struct anv_buffer *) pBuffers[i];
2617 bindings->vb[startBinding + i].offset = pOffsets[i];
2618 cmd_buffer->vb_dirty |= 1 << (startBinding + i);
2619 }
2620 }
2621
2622 static void
2623 flush_descriptor_sets(struct anv_cmd_buffer *cmd_buffer)
2624 {
2625 struct anv_pipeline_layout *layout = cmd_buffer->pipeline->layout;
2626 struct anv_bindings *bindings = cmd_buffer->bindings;
2627 uint32_t layers = cmd_buffer->framebuffer->layers;
2628
2629 for (uint32_t s = 0; s < VK_NUM_SHADER_STAGE; s++) {
2630 uint32_t bias;
2631
2632 if (s == VK_SHADER_STAGE_FRAGMENT) {
2633 bias = MAX_RTS;
2634 layers = cmd_buffer->framebuffer->layers;
2635 } else {
2636 bias = 0;
2637 layers = 0;
2638 }
2639
2640 /* This is a little awkward: layout can be NULL but we still have to
2641 * allocate and set a binding table for the PS stage for render
2642 * targets. */
2643 uint32_t surface_count = layout ? layout->stage[s].surface_count : 0;
2644
2645 if (layers + surface_count > 0) {
2646 struct anv_state state;
2647 uint32_t size;
2648
2649 size = (bias + surface_count) * sizeof(uint32_t);
2650 state = anv_cmd_buffer_alloc_surface_state(cmd_buffer, size, 32);
2651 memcpy(state.map, bindings->descriptors[s].surfaces, size);
2652
2653 static const uint32_t binding_table_opcodes[] = {
2654 [VK_SHADER_STAGE_VERTEX] = 38,
2655 [VK_SHADER_STAGE_TESS_CONTROL] = 39,
2656 [VK_SHADER_STAGE_TESS_EVALUATION] = 40,
2657 [VK_SHADER_STAGE_GEOMETRY] = 41,
2658 [VK_SHADER_STAGE_FRAGMENT] = 42,
2659 [VK_SHADER_STAGE_COMPUTE] = 0,
2660 };
2661
2662 anv_batch_emit(&cmd_buffer->batch,
2663 GEN8_3DSTATE_BINDING_TABLE_POINTERS_VS,
2664 ._3DCommandSubOpcode = binding_table_opcodes[s],
2665 .PointertoVSBindingTable = state.offset);
2666 }
2667
2668 if (layout && layout->stage[s].sampler_count > 0) {
2669 struct anv_state state;
2670 size_t size;
2671
2672 size = layout->stage[s].sampler_count * 16;
2673 state = anv_cmd_buffer_alloc_surface_state(cmd_buffer, size, 32);
2674 memcpy(state.map, bindings->descriptors[s].samplers, size);
2675
2676 static const uint32_t sampler_state_opcodes[] = {
2677 [VK_SHADER_STAGE_VERTEX] = 43,
2678 [VK_SHADER_STAGE_TESS_CONTROL] = 44, /* HS */
2679 [VK_SHADER_STAGE_TESS_EVALUATION] = 45, /* DS */
2680 [VK_SHADER_STAGE_GEOMETRY] = 46,
2681 [VK_SHADER_STAGE_FRAGMENT] = 47,
2682 [VK_SHADER_STAGE_COMPUTE] = 0,
2683 };
2684
2685 anv_batch_emit(&cmd_buffer->batch,
2686 GEN8_3DSTATE_SAMPLER_STATE_POINTERS_VS,
2687 ._3DCommandSubOpcode = sampler_state_opcodes[s],
2688 .PointertoVSSamplerState = state.offset);
2689 }
2690 }
2691 }
2692
2693 static struct anv_state
2694 anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer *cmd_buffer,
2695 uint32_t *a, uint32_t dwords, uint32_t alignment)
2696 {
2697 struct anv_device *device = cmd_buffer->device;
2698 struct anv_state state;
2699
2700 state = anv_state_pool_alloc(&device->dynamic_state_pool, dwords * 4, alignment);
2701 memcpy(state.map, a, dwords * 4);
2702
2703 return state;
2704 }
2705
2706 static struct anv_state
2707 anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer *cmd_buffer,
2708 uint32_t *a, uint32_t *b, uint32_t dwords, uint32_t alignment)
2709 {
2710 struct anv_device *device = cmd_buffer->device;
2711 struct anv_state state;
2712 uint32_t *p;
2713
2714 state = anv_state_pool_alloc(&device->dynamic_state_pool, dwords * 4, alignment);
2715 p = state.map;
2716 for (uint32_t i = 0; i < dwords; i++)
2717 p[i] = a[i] | b[i];
2718
2719 return state;
2720 }
2721
2722 static void
2723 anv_cmd_buffer_flush_state(struct anv_cmd_buffer *cmd_buffer)
2724 {
2725 struct anv_pipeline *pipeline = cmd_buffer->pipeline;
2726 struct anv_bindings *bindings = cmd_buffer->bindings;
2727 uint32_t *p;
2728
2729 uint32_t vb_emit = cmd_buffer->vb_dirty & pipeline->vb_used;
2730 const uint32_t num_buffers = __builtin_popcount(vb_emit);
2731 const uint32_t num_dwords = 1 + num_buffers * 4;
2732
2733 if (vb_emit) {
2734 p = anv_batch_emitn(&cmd_buffer->batch, num_dwords,
2735 GEN8_3DSTATE_VERTEX_BUFFERS);
2736 uint32_t vb, i = 0;
2737 for_each_bit(vb, vb_emit) {
2738 struct anv_buffer *buffer = bindings->vb[vb].buffer;
2739 uint32_t offset = bindings->vb[vb].offset;
2740
2741 struct GEN8_VERTEX_BUFFER_STATE state = {
2742 .VertexBufferIndex = vb,
2743 .MemoryObjectControlState = GEN8_MOCS,
2744 .AddressModifyEnable = true,
2745 .BufferPitch = pipeline->binding_stride[vb],
2746 .BufferStartingAddress = { buffer->bo, buffer->offset + offset },
2747 .BufferSize = buffer->size - offset
2748 };
2749
2750 GEN8_VERTEX_BUFFER_STATE_pack(&cmd_buffer->batch, &p[1 + i * 4], &state);
2751 i++;
2752 }
2753 }
2754
2755 if (cmd_buffer->dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY)
2756 anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
2757
2758 if (cmd_buffer->dirty & ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY)
2759 flush_descriptor_sets(cmd_buffer);
2760
2761 if (cmd_buffer->dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY | ANV_CMD_BUFFER_RS_DIRTY)) {
2762 anv_batch_emit_merge(&cmd_buffer->batch,
2763 cmd_buffer->rs_state->state_sf, pipeline->state_sf);
2764 anv_batch_emit_merge(&cmd_buffer->batch,
2765 cmd_buffer->rs_state->state_raster, pipeline->state_raster);
2766 }
2767
2768 if (cmd_buffer->ds_state &&
2769 (cmd_buffer->dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY | ANV_CMD_BUFFER_DS_DIRTY)))
2770 anv_batch_emit_merge(&cmd_buffer->batch,
2771 cmd_buffer->ds_state->state_wm_depth_stencil,
2772 pipeline->state_wm_depth_stencil);
2773
2774 if (cmd_buffer->dirty & (ANV_CMD_BUFFER_CB_DIRTY | ANV_CMD_BUFFER_DS_DIRTY)) {
2775 struct anv_state state;
2776 if (cmd_buffer->ds_state)
2777 state = anv_cmd_buffer_merge_dynamic(cmd_buffer,
2778 cmd_buffer->ds_state->state_color_calc,
2779 cmd_buffer->cb_state->state_color_calc,
2780 GEN8_COLOR_CALC_STATE_length, 32);
2781 else
2782 state = anv_cmd_buffer_emit_dynamic(cmd_buffer,
2783 cmd_buffer->cb_state->state_color_calc,
2784 GEN8_COLOR_CALC_STATE_length, 32);
2785
2786 anv_batch_emit(&cmd_buffer->batch,
2787 GEN8_3DSTATE_CC_STATE_POINTERS,
2788 .ColorCalcStatePointer = state.offset,
2789 .ColorCalcStatePointerValid = true);
2790 }
2791
2792 cmd_buffer->vb_dirty &= ~vb_emit;
2793 cmd_buffer->dirty = 0;
2794 }
2795
2796 void anv_CmdDraw(
2797 VkCmdBuffer cmdBuffer,
2798 uint32_t firstVertex,
2799 uint32_t vertexCount,
2800 uint32_t firstInstance,
2801 uint32_t instanceCount)
2802 {
2803 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2804
2805 anv_cmd_buffer_flush_state(cmd_buffer);
2806
2807 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
2808 .VertexAccessType = SEQUENTIAL,
2809 .VertexCountPerInstance = vertexCount,
2810 .StartVertexLocation = firstVertex,
2811 .InstanceCount = instanceCount,
2812 .StartInstanceLocation = firstInstance,
2813 .BaseVertexLocation = 0);
2814 }
2815
2816 void anv_CmdDrawIndexed(
2817 VkCmdBuffer cmdBuffer,
2818 uint32_t firstIndex,
2819 uint32_t indexCount,
2820 int32_t vertexOffset,
2821 uint32_t firstInstance,
2822 uint32_t instanceCount)
2823 {
2824 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2825
2826 anv_cmd_buffer_flush_state(cmd_buffer);
2827
2828 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
2829 .VertexAccessType = RANDOM,
2830 .VertexCountPerInstance = indexCount,
2831 .StartVertexLocation = firstIndex,
2832 .InstanceCount = instanceCount,
2833 .StartInstanceLocation = firstInstance,
2834 .BaseVertexLocation = 0);
2835 }
2836
2837 static void
2838 anv_batch_lrm(struct anv_batch *batch,
2839 uint32_t reg, struct anv_bo *bo, uint32_t offset)
2840 {
2841 anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
2842 .RegisterAddress = reg,
2843 .MemoryAddress = { bo, offset });
2844 }
2845
2846 static void
2847 anv_batch_lri(struct anv_batch *batch, uint32_t reg, uint32_t imm)
2848 {
2849 anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_IMM,
2850 .RegisterOffset = reg,
2851 .DataDWord = imm);
2852 }
2853
2854 /* Auto-Draw / Indirect Registers */
2855 #define GEN7_3DPRIM_END_OFFSET 0x2420
2856 #define GEN7_3DPRIM_START_VERTEX 0x2430
2857 #define GEN7_3DPRIM_VERTEX_COUNT 0x2434
2858 #define GEN7_3DPRIM_INSTANCE_COUNT 0x2438
2859 #define GEN7_3DPRIM_START_INSTANCE 0x243C
2860 #define GEN7_3DPRIM_BASE_VERTEX 0x2440
2861
2862 void anv_CmdDrawIndirect(
2863 VkCmdBuffer cmdBuffer,
2864 VkBuffer _buffer,
2865 VkDeviceSize offset,
2866 uint32_t count,
2867 uint32_t stride)
2868 {
2869 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2870 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
2871 struct anv_bo *bo = buffer->bo;
2872 uint32_t bo_offset = buffer->offset + offset;
2873
2874 anv_cmd_buffer_flush_state(cmd_buffer);
2875
2876 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);
2877 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
2878 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);
2879 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 12);
2880 anv_batch_lri(&cmd_buffer->batch, GEN7_3DPRIM_BASE_VERTEX, 0);
2881
2882 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
2883 .IndirectParameterEnable = true,
2884 .VertexAccessType = SEQUENTIAL);
2885 }
2886
2887 void anv_CmdDrawIndexedIndirect(
2888 VkCmdBuffer cmdBuffer,
2889 VkBuffer _buffer,
2890 VkDeviceSize offset,
2891 uint32_t count,
2892 uint32_t stride)
2893 {
2894 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2895 struct anv_buffer *buffer = (struct anv_buffer *) _buffer;
2896 struct anv_bo *bo = buffer->bo;
2897 uint32_t bo_offset = buffer->offset + offset;
2898
2899 anv_cmd_buffer_flush_state(cmd_buffer);
2900
2901 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);
2902 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
2903 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);
2904 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_BASE_VERTEX, bo, bo_offset + 12);
2905 anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 16);
2906
2907 anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
2908 .IndirectParameterEnable = true,
2909 .VertexAccessType = RANDOM);
2910 }
2911
2912 void anv_CmdDispatch(
2913 VkCmdBuffer cmdBuffer,
2914 uint32_t x,
2915 uint32_t y,
2916 uint32_t z)
2917 {
2918 stub();
2919 }
2920
2921 void anv_CmdDispatchIndirect(
2922 VkCmdBuffer cmdBuffer,
2923 VkBuffer buffer,
2924 VkDeviceSize offset)
2925 {
2926 stub();
2927 }
2928
2929 void anv_CmdSetEvent(
2930 VkCmdBuffer cmdBuffer,
2931 VkEvent event,
2932 VkPipeEvent pipeEvent)
2933 {
2934 stub();
2935 }
2936
2937 void anv_CmdResetEvent(
2938 VkCmdBuffer cmdBuffer,
2939 VkEvent event,
2940 VkPipeEvent pipeEvent)
2941 {
2942 stub();
2943 }
2944
2945 void anv_CmdWaitEvents(
2946 VkCmdBuffer cmdBuffer,
2947 VkWaitEvent waitEvent,
2948 uint32_t eventCount,
2949 const VkEvent* pEvents,
2950 uint32_t memBarrierCount,
2951 const void** ppMemBarriers)
2952 {
2953 stub();
2954 }
2955
2956 void anv_CmdPipelineBarrier(
2957 VkCmdBuffer cmdBuffer,
2958 VkWaitEvent waitEvent,
2959 uint32_t pipeEventCount,
2960 const VkPipeEvent* pPipeEvents,
2961 uint32_t memBarrierCount,
2962 const void** ppMemBarriers)
2963 {
2964 stub();
2965 }
2966
2967 static void
2968 anv_batch_emit_ps_depth_count(struct anv_batch *batch,
2969 struct anv_bo *bo, uint32_t offset)
2970 {
2971 anv_batch_emit(batch, GEN8_PIPE_CONTROL,
2972 .DestinationAddressType = DAT_PPGTT,
2973 .PostSyncOperation = WritePSDepthCount,
2974 .Address = { bo, offset }); /* FIXME: This is only lower 32 bits */
2975 }
2976
2977 void anv_CmdBeginQuery(
2978 VkCmdBuffer cmdBuffer,
2979 VkQueryPool queryPool,
2980 uint32_t slot,
2981 VkQueryControlFlags flags)
2982 {
2983 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
2984 struct anv_query_pool *pool = (struct anv_query_pool *) queryPool;
2985
2986 switch (pool->type) {
2987 case VK_QUERY_TYPE_OCCLUSION:
2988 anv_batch_emit_ps_depth_count(&cmd_buffer->batch, &pool->bo,
2989 slot * sizeof(struct anv_query_pool_slot));
2990 break;
2991
2992 case VK_QUERY_TYPE_PIPELINE_STATISTICS:
2993 default:
2994 unreachable("");
2995 }
2996 }
2997
2998 void anv_CmdEndQuery(
2999 VkCmdBuffer cmdBuffer,
3000 VkQueryPool queryPool,
3001 uint32_t slot)
3002 {
3003 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3004 struct anv_query_pool *pool = (struct anv_query_pool *) queryPool;
3005
3006 switch (pool->type) {
3007 case VK_QUERY_TYPE_OCCLUSION:
3008 anv_batch_emit_ps_depth_count(&cmd_buffer->batch, &pool->bo,
3009 slot * sizeof(struct anv_query_pool_slot) + 8);
3010 break;
3011
3012 case VK_QUERY_TYPE_PIPELINE_STATISTICS:
3013 default:
3014 unreachable("");
3015 }
3016 }
3017
3018 void anv_CmdResetQueryPool(
3019 VkCmdBuffer cmdBuffer,
3020 VkQueryPool queryPool,
3021 uint32_t startQuery,
3022 uint32_t queryCount)
3023 {
3024 stub();
3025 }
3026
3027 #define TIMESTAMP 0x2358
3028
3029 void anv_CmdWriteTimestamp(
3030 VkCmdBuffer cmdBuffer,
3031 VkTimestampType timestampType,
3032 VkBuffer destBuffer,
3033 VkDeviceSize destOffset)
3034 {
3035 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3036 struct anv_buffer *buffer = (struct anv_buffer *) destBuffer;
3037 struct anv_bo *bo = buffer->bo;
3038
3039 switch (timestampType) {
3040 case VK_TIMESTAMP_TYPE_TOP:
3041 anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
3042 .RegisterAddress = TIMESTAMP,
3043 .MemoryAddress = { bo, buffer->offset + destOffset });
3044 anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
3045 .RegisterAddress = TIMESTAMP + 4,
3046 .MemoryAddress = { bo, buffer->offset + destOffset + 4 });
3047 break;
3048
3049 case VK_TIMESTAMP_TYPE_BOTTOM:
3050 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
3051 .DestinationAddressType = DAT_PPGTT,
3052 .PostSyncOperation = WriteTimestamp,
3053 .Address = /* FIXME: This is only lower 32 bits */
3054 { bo, buffer->offset + destOffset });
3055 break;
3056
3057 default:
3058 break;
3059 }
3060 }
3061
3062 #define alu_opcode(v) __gen_field((v), 20, 31)
3063 #define alu_operand1(v) __gen_field((v), 10, 19)
3064 #define alu_operand2(v) __gen_field((v), 0, 9)
3065 #define alu(opcode, operand1, operand2) \
3066 alu_opcode(opcode) | alu_operand1(operand1) | alu_operand2(operand2)
3067
3068 #define OPCODE_NOOP 0x000
3069 #define OPCODE_LOAD 0x080
3070 #define OPCODE_LOADINV 0x480
3071 #define OPCODE_LOAD0 0x081
3072 #define OPCODE_LOAD1 0x481
3073 #define OPCODE_ADD 0x100
3074 #define OPCODE_SUB 0x101
3075 #define OPCODE_AND 0x102
3076 #define OPCODE_OR 0x103
3077 #define OPCODE_XOR 0x104
3078 #define OPCODE_STORE 0x180
3079 #define OPCODE_STOREINV 0x580
3080
3081 #define OPERAND_R0 0x00
3082 #define OPERAND_R1 0x01
3083 #define OPERAND_R2 0x02
3084 #define OPERAND_R3 0x03
3085 #define OPERAND_R4 0x04
3086 #define OPERAND_SRCA 0x20
3087 #define OPERAND_SRCB 0x21
3088 #define OPERAND_ACCU 0x31
3089 #define OPERAND_ZF 0x32
3090 #define OPERAND_CF 0x33
3091
3092 #define CS_GPR(n) (0x2600 + (n) * 8)
3093
3094 static void
3095 emit_load_alu_reg_u64(struct anv_batch *batch, uint32_t reg,
3096 struct anv_bo *bo, uint32_t offset)
3097 {
3098 anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
3099 .RegisterAddress = reg,
3100 .MemoryAddress = { bo, offset });
3101 anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
3102 .RegisterAddress = reg + 4,
3103 .MemoryAddress = { bo, offset + 4 });
3104 }
3105
3106 void anv_CmdCopyQueryPoolResults(
3107 VkCmdBuffer cmdBuffer,
3108 VkQueryPool queryPool,
3109 uint32_t startQuery,
3110 uint32_t queryCount,
3111 VkBuffer destBuffer,
3112 VkDeviceSize destOffset,
3113 VkDeviceSize destStride,
3114 VkQueryResultFlags flags)
3115 {
3116 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3117 struct anv_query_pool *pool = (struct anv_query_pool *) queryPool;
3118 struct anv_buffer *buffer = (struct anv_buffer *) destBuffer;
3119 uint32_t slot_offset, dst_offset;
3120
3121 if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
3122 /* Where is the availabilty info supposed to go? */
3123 anv_finishme("VK_QUERY_RESULT_WITH_AVAILABILITY_BIT");
3124 return;
3125 }
3126
3127 assert(pool->type == VK_QUERY_TYPE_OCCLUSION);
3128
3129 /* FIXME: If we're not waiting, should we just do this on the CPU? */
3130 if (flags & VK_QUERY_RESULT_WAIT_BIT)
3131 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
3132 .CommandStreamerStallEnable = true);
3133
3134 dst_offset = buffer->offset + destOffset;
3135 for (uint32_t i = 0; i < queryCount; i++) {
3136
3137 slot_offset = (startQuery + i) * sizeof(struct anv_query_pool_slot);
3138
3139 emit_load_alu_reg_u64(&cmd_buffer->batch, CS_GPR(0), &pool->bo, slot_offset);
3140 emit_load_alu_reg_u64(&cmd_buffer->batch, CS_GPR(1), &pool->bo, slot_offset + 8);
3141
3142 /* FIXME: We need to clamp the result for 32 bit. */
3143
3144 uint32_t *dw = anv_batch_emitn(&cmd_buffer->batch, 5, GEN8_MI_MATH);
3145 dw[1] = alu(OPCODE_LOAD, OPERAND_SRCA, OPERAND_R1);
3146 dw[2] = alu(OPCODE_LOAD, OPERAND_SRCB, OPERAND_R0);
3147 dw[3] = alu(OPCODE_SUB, 0, 0);
3148 dw[4] = alu(OPCODE_STORE, OPERAND_R2, OPERAND_ACCU);
3149
3150 anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
3151 .RegisterAddress = CS_GPR(2),
3152 /* FIXME: This is only lower 32 bits */
3153 .MemoryAddress = { buffer->bo, dst_offset });
3154
3155 if (flags & VK_QUERY_RESULT_64_BIT)
3156 anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
3157 .RegisterAddress = CS_GPR(2) + 4,
3158 /* FIXME: This is only lower 32 bits */
3159 .MemoryAddress = { buffer->bo, dst_offset + 4 });
3160
3161 dst_offset += destStride;
3162 }
3163 }
3164
3165 void anv_CmdInitAtomicCounters(
3166 VkCmdBuffer cmdBuffer,
3167 VkPipelineBindPoint pipelineBindPoint,
3168 uint32_t startCounter,
3169 uint32_t counterCount,
3170 const uint32_t* pData)
3171 {
3172 stub();
3173 }
3174
3175 void anv_CmdLoadAtomicCounters(
3176 VkCmdBuffer cmdBuffer,
3177 VkPipelineBindPoint pipelineBindPoint,
3178 uint32_t startCounter,
3179 uint32_t counterCount,
3180 VkBuffer srcBuffer,
3181 VkDeviceSize srcOffset)
3182 {
3183 stub();
3184 }
3185
3186 void anv_CmdSaveAtomicCounters(
3187 VkCmdBuffer cmdBuffer,
3188 VkPipelineBindPoint pipelineBindPoint,
3189 uint32_t startCounter,
3190 uint32_t counterCount,
3191 VkBuffer destBuffer,
3192 VkDeviceSize destOffset)
3193 {
3194 stub();
3195 }
3196
3197 static void
3198 anv_framebuffer_destroy(struct anv_device *device,
3199 struct anv_object *object,
3200 VkObjectType obj_type)
3201 {
3202 struct anv_framebuffer *fb = (struct anv_framebuffer *)object;
3203
3204 assert(obj_type == VK_OBJECT_TYPE_FRAMEBUFFER);
3205
3206 anv_DestroyObject((VkDevice) device,
3207 VK_OBJECT_TYPE_DYNAMIC_VP_STATE,
3208 fb->vp_state);
3209
3210 anv_device_free(device, fb);
3211 }
3212
3213 VkResult anv_CreateFramebuffer(
3214 VkDevice _device,
3215 const VkFramebufferCreateInfo* pCreateInfo,
3216 VkFramebuffer* pFramebuffer)
3217 {
3218 struct anv_device *device = (struct anv_device *) _device;
3219 struct anv_framebuffer *framebuffer;
3220
3221 static const struct anv_depth_stencil_view null_view =
3222 { .depth_format = D16_UNORM, .depth_stride = 0, .stencil_stride = 0 };
3223
3224 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
3225
3226 framebuffer = anv_device_alloc(device, sizeof(*framebuffer), 8,
3227 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
3228 if (framebuffer == NULL)
3229 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
3230
3231 framebuffer->base.destructor = anv_framebuffer_destroy;
3232
3233 framebuffer->color_attachment_count = pCreateInfo->colorAttachmentCount;
3234 for (uint32_t i = 0; i < pCreateInfo->colorAttachmentCount; i++) {
3235 framebuffer->color_attachments[i] =
3236 (struct anv_surface_view *) pCreateInfo->pColorAttachments[i].view;
3237 }
3238
3239 if (pCreateInfo->pDepthStencilAttachment) {
3240 framebuffer->depth_stencil =
3241 (struct anv_depth_stencil_view *) pCreateInfo->pDepthStencilAttachment->view;
3242 } else {
3243 framebuffer->depth_stencil = &null_view;
3244 }
3245
3246 framebuffer->sample_count = pCreateInfo->sampleCount;
3247 framebuffer->width = pCreateInfo->width;
3248 framebuffer->height = pCreateInfo->height;
3249 framebuffer->layers = pCreateInfo->layers;
3250
3251 vkCreateDynamicViewportState((VkDevice) device,
3252 &(VkDynamicVpStateCreateInfo) {
3253 .sType = VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO,
3254 .viewportAndScissorCount = 1,
3255 .pViewports = (VkViewport[]) {
3256 {
3257 .originX = 0,
3258 .originY = 0,
3259 .width = pCreateInfo->width,
3260 .height = pCreateInfo->height,
3261 .minDepth = 0,
3262 .maxDepth = 1
3263 },
3264 },
3265 .pScissors = (VkRect[]) {
3266 { { 0, 0 },
3267 { pCreateInfo->width, pCreateInfo->height } },
3268 }
3269 },
3270 &framebuffer->vp_state);
3271
3272 *pFramebuffer = (VkFramebuffer) framebuffer;
3273
3274 return VK_SUCCESS;
3275 }
3276
3277 VkResult anv_CreateRenderPass(
3278 VkDevice _device,
3279 const VkRenderPassCreateInfo* pCreateInfo,
3280 VkRenderPass* pRenderPass)
3281 {
3282 struct anv_device *device = (struct anv_device *) _device;
3283 struct anv_render_pass *pass;
3284 size_t size;
3285
3286 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
3287
3288 size = sizeof(*pass) +
3289 pCreateInfo->layers * sizeof(struct anv_render_pass_layer);
3290 pass = anv_device_alloc(device, size, 8,
3291 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
3292 if (pass == NULL)
3293 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
3294
3295 pass->render_area = pCreateInfo->renderArea;
3296
3297 pass->num_layers = pCreateInfo->layers;
3298
3299 pass->num_clear_layers = 0;
3300 for (uint32_t i = 0; i < pCreateInfo->layers; i++) {
3301 pass->layers[i].color_load_op = pCreateInfo->pColorLoadOps[i];
3302 pass->layers[i].clear_color = pCreateInfo->pColorLoadClearValues[i];
3303 if (pass->layers[i].color_load_op == VK_ATTACHMENT_LOAD_OP_CLEAR)
3304 pass->num_clear_layers++;
3305 }
3306
3307 *pRenderPass = (VkRenderPass) pass;
3308
3309 return VK_SUCCESS;
3310 }
3311
3312 void
3313 anv_cmd_buffer_fill_render_targets(struct anv_cmd_buffer *cmd_buffer)
3314 {
3315 struct anv_framebuffer *framebuffer = cmd_buffer->framebuffer;
3316 struct anv_bindings *bindings = cmd_buffer->bindings;
3317
3318 for (uint32_t i = 0; i < framebuffer->color_attachment_count; i++) {
3319 const struct anv_surface_view *view = framebuffer->color_attachments[i];
3320
3321 struct anv_state state =
3322 anv_cmd_buffer_alloc_surface_state(cmd_buffer, 64, 64);
3323 memcpy(state.map, view->surface_state.map, 64);
3324
3325 /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
3326 *(uint64_t *)(state.map + 8 * 4) =
3327 anv_reloc_list_add(&cmd_buffer->surface_relocs,
3328 state.offset + 8 * 4,
3329 view->bo, view->offset);
3330
3331 bindings->descriptors[VK_SHADER_STAGE_FRAGMENT].surfaces[i] = state.offset;
3332 }
3333 cmd_buffer->dirty |= ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY;
3334 }
3335
3336 static void
3337 anv_cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer,
3338 struct anv_render_pass *pass)
3339 {
3340 const struct anv_depth_stencil_view *view =
3341 cmd_buffer->framebuffer->depth_stencil;
3342
3343 /* FIXME: Implement the PMA stall W/A */
3344
3345 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DEPTH_BUFFER,
3346 .SurfaceType = SURFTYPE_2D,
3347 .DepthWriteEnable = view->depth_stride > 0,
3348 .StencilWriteEnable = view->stencil_stride > 0,
3349 .HierarchicalDepthBufferEnable = false,
3350 .SurfaceFormat = view->depth_format,
3351 .SurfacePitch = view->depth_stride > 0 ? view->depth_stride - 1 : 0,
3352 .SurfaceBaseAddress = { view->bo, view->depth_offset },
3353 .Height = pass->render_area.extent.height - 1,
3354 .Width = pass->render_area.extent.width - 1,
3355 .LOD = 0,
3356 .Depth = 1 - 1,
3357 .MinimumArrayElement = 0,
3358 .DepthBufferObjectControlState = GEN8_MOCS,
3359 .RenderTargetViewExtent = 1 - 1,
3360 .SurfaceQPitch = 0);
3361
3362 /* Disable hierarchial depth buffers. */
3363 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_HIER_DEPTH_BUFFER);
3364
3365 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_STENCIL_BUFFER,
3366 .StencilBufferEnable = view->stencil_stride > 0,
3367 .StencilBufferObjectControlState = GEN8_MOCS,
3368 .SurfacePitch = view->stencil_stride > 0 ? view->stencil_stride - 1 : 0,
3369 .SurfaceBaseAddress = { view->bo, view->stencil_offset },
3370 .SurfaceQPitch = 0);
3371
3372 /* Clear the clear params. */
3373 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_CLEAR_PARAMS);
3374 }
3375
3376 void anv_CmdBeginRenderPass(
3377 VkCmdBuffer cmdBuffer,
3378 const VkRenderPassBegin* pRenderPassBegin)
3379 {
3380 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
3381 struct anv_render_pass *pass = (struct anv_render_pass *) pRenderPassBegin->renderPass;
3382 struct anv_framebuffer *framebuffer =
3383 (struct anv_framebuffer *) pRenderPassBegin->framebuffer;
3384
3385 cmd_buffer->framebuffer = framebuffer;
3386
3387 anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DRAWING_RECTANGLE,
3388 .ClippedDrawingRectangleYMin = pass->render_area.offset.y,
3389 .ClippedDrawingRectangleXMin = pass->render_area.offset.x,
3390 .ClippedDrawingRectangleYMax =
3391 pass->render_area.offset.y + pass->render_area.extent.height - 1,
3392 .ClippedDrawingRectangleXMax =
3393 pass->render_area.offset.x + pass->render_area.extent.width - 1,
3394 .DrawingRectangleOriginY = 0,
3395 .DrawingRectangleOriginX = 0);
3396
3397 anv_cmd_buffer_fill_render_targets(cmd_buffer);
3398
3399 anv_cmd_buffer_emit_depth_stencil(cmd_buffer, pass);
3400
3401 anv_cmd_buffer_clear(cmd_buffer, pass);
3402 }
3403
3404 void anv_CmdEndRenderPass(
3405 VkCmdBuffer cmdBuffer,
3406 VkRenderPass renderPass)
3407 {
3408 /* Emit a flushing pipe control at the end of a pass. This is kind of a
3409 * hack but it ensures that render targets always actually get written.
3410 * Eventually, we should do flushing based on image format transitions
3411 * or something of that nature.
3412 */
3413 struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *)cmdBuffer;
3414 anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
3415 .PostSyncOperation = NoWrite,
3416 .RenderTargetCacheFlushEnable = true,
3417 .InstructionCacheInvalidateEnable = true,
3418 .DepthCacheFlushEnable = true,
3419 .VFCacheInvalidationEnable = true,
3420 .TextureCacheInvalidationEnable = true,
3421 .CommandStreamerStallEnable = true);
3422 }
3423
3424 void vkCmdDbgMarkerBegin(
3425 VkCmdBuffer cmdBuffer,
3426 const char* pMarker)
3427 __attribute__ ((visibility ("default")));
3428
3429 void vkCmdDbgMarkerEnd(
3430 VkCmdBuffer cmdBuffer)
3431 __attribute__ ((visibility ("default")));
3432
3433 VkResult vkDbgSetObjectTag(
3434 VkDevice device,
3435 VkObject object,
3436 size_t tagSize,
3437 const void* pTag)
3438 __attribute__ ((visibility ("default")));
3439
3440
3441 void vkCmdDbgMarkerBegin(
3442 VkCmdBuffer cmdBuffer,
3443 const char* pMarker)
3444 {
3445 }
3446
3447 void vkCmdDbgMarkerEnd(
3448 VkCmdBuffer cmdBuffer)
3449 {
3450 }
3451
3452 VkResult vkDbgSetObjectTag(
3453 VkDevice device,
3454 VkObject object,
3455 size_t tagSize,
3456 const void* pTag)
3457 {
3458 return VK_SUCCESS;
3459 }