tu: Switch to the bindless descriptor model
[mesa.git] / src / freedreno / vulkan / tu_device.c
1 /*
2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
4 *
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the next
16 * paragraph) shall be included in all copies or substantial portions of the
17 * Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
25 * DEALINGS IN THE SOFTWARE.
26 */
27
28 #include "tu_private.h"
29
30 #include <fcntl.h>
31 #include <libsync.h>
32 #include <stdbool.h>
33 #include <string.h>
34 #include <sys/mman.h>
35 #include <sys/sysinfo.h>
36 #include <unistd.h>
37 #include <xf86drm.h>
38
39 #include "compiler/glsl_types.h"
40 #include "util/debug.h"
41 #include "util/disk_cache.h"
42 #include "vk_format.h"
43 #include "vk_util.h"
44
45 #include "drm-uapi/msm_drm.h"
46
47 static int
48 tu_device_get_cache_uuid(uint16_t family, void *uuid)
49 {
50 uint32_t mesa_timestamp;
51 uint16_t f = family;
52 memset(uuid, 0, VK_UUID_SIZE);
53 if (!disk_cache_get_function_timestamp(tu_device_get_cache_uuid,
54 &mesa_timestamp))
55 return -1;
56
57 memcpy(uuid, &mesa_timestamp, 4);
58 memcpy((char *) uuid + 4, &f, 2);
59 snprintf((char *) uuid + 6, VK_UUID_SIZE - 10, "tu");
60 return 0;
61 }
62
63 static void
64 tu_get_driver_uuid(void *uuid)
65 {
66 memset(uuid, 0, VK_UUID_SIZE);
67 snprintf(uuid, VK_UUID_SIZE, "freedreno");
68 }
69
70 static void
71 tu_get_device_uuid(void *uuid)
72 {
73 memset(uuid, 0, VK_UUID_SIZE);
74 }
75
76 static VkResult
77 tu_bo_init(struct tu_device *dev,
78 struct tu_bo *bo,
79 uint32_t gem_handle,
80 uint64_t size)
81 {
82 uint64_t iova = tu_gem_info_iova(dev, gem_handle);
83 if (!iova)
84 return VK_ERROR_OUT_OF_DEVICE_MEMORY;
85
86 *bo = (struct tu_bo) {
87 .gem_handle = gem_handle,
88 .size = size,
89 .iova = iova,
90 };
91
92 return VK_SUCCESS;
93 }
94
95 VkResult
96 tu_bo_init_new(struct tu_device *dev, struct tu_bo *bo, uint64_t size)
97 {
98 /* TODO: Choose better flags. As of 2018-11-12, freedreno/drm/msm_bo.c
99 * always sets `flags = MSM_BO_WC`, and we copy that behavior here.
100 */
101 uint32_t gem_handle = tu_gem_new(dev, size, MSM_BO_WC);
102 if (!gem_handle)
103 return vk_error(dev->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
104
105 VkResult result = tu_bo_init(dev, bo, gem_handle, size);
106 if (result != VK_SUCCESS) {
107 tu_gem_close(dev, gem_handle);
108 return vk_error(dev->instance, result);
109 }
110
111 return VK_SUCCESS;
112 }
113
114 VkResult
115 tu_bo_init_dmabuf(struct tu_device *dev,
116 struct tu_bo *bo,
117 uint64_t size,
118 int fd)
119 {
120 uint32_t gem_handle = tu_gem_import_dmabuf(dev, fd, size);
121 if (!gem_handle)
122 return vk_error(dev->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
123
124 VkResult result = tu_bo_init(dev, bo, gem_handle, size);
125 if (result != VK_SUCCESS) {
126 tu_gem_close(dev, gem_handle);
127 return vk_error(dev->instance, result);
128 }
129
130 return VK_SUCCESS;
131 }
132
133 int
134 tu_bo_export_dmabuf(struct tu_device *dev, struct tu_bo *bo)
135 {
136 return tu_gem_export_dmabuf(dev, bo->gem_handle);
137 }
138
139 VkResult
140 tu_bo_map(struct tu_device *dev, struct tu_bo *bo)
141 {
142 if (bo->map)
143 return VK_SUCCESS;
144
145 uint64_t offset = tu_gem_info_offset(dev, bo->gem_handle);
146 if (!offset)
147 return vk_error(dev->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
148
149 /* TODO: Should we use the wrapper os_mmap() like Freedreno does? */
150 void *map = mmap(0, bo->size, PROT_READ | PROT_WRITE, MAP_SHARED,
151 dev->physical_device->local_fd, offset);
152 if (map == MAP_FAILED)
153 return vk_error(dev->instance, VK_ERROR_MEMORY_MAP_FAILED);
154
155 bo->map = map;
156 return VK_SUCCESS;
157 }
158
159 void
160 tu_bo_finish(struct tu_device *dev, struct tu_bo *bo)
161 {
162 assert(bo->gem_handle);
163
164 if (bo->map)
165 munmap(bo->map, bo->size);
166
167 tu_gem_close(dev, bo->gem_handle);
168 }
169
170 static VkResult
171 tu_physical_device_init(struct tu_physical_device *device,
172 struct tu_instance *instance,
173 drmDevicePtr drm_device)
174 {
175 const char *path = drm_device->nodes[DRM_NODE_RENDER];
176 VkResult result = VK_SUCCESS;
177 drmVersionPtr version;
178 int fd;
179 int master_fd = -1;
180
181 fd = open(path, O_RDWR | O_CLOEXEC);
182 if (fd < 0) {
183 return vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
184 "failed to open device %s", path);
185 }
186
187 /* Version 1.3 added MSM_INFO_IOVA. */
188 const int min_version_major = 1;
189 const int min_version_minor = 3;
190
191 version = drmGetVersion(fd);
192 if (!version) {
193 close(fd);
194 return vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
195 "failed to query kernel driver version for device %s",
196 path);
197 }
198
199 if (strcmp(version->name, "msm")) {
200 drmFreeVersion(version);
201 close(fd);
202 return vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
203 "device %s does not use the msm kernel driver", path);
204 }
205
206 if (version->version_major != min_version_major ||
207 version->version_minor < min_version_minor) {
208 result = vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
209 "kernel driver for device %s has version %d.%d, "
210 "but Vulkan requires version >= %d.%d",
211 path, version->version_major, version->version_minor,
212 min_version_major, min_version_minor);
213 drmFreeVersion(version);
214 close(fd);
215 return result;
216 }
217
218 drmFreeVersion(version);
219
220 if (instance->debug_flags & TU_DEBUG_STARTUP)
221 tu_logi("Found compatible device '%s'.", path);
222
223 device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
224 device->instance = instance;
225 assert(strlen(path) < ARRAY_SIZE(device->path));
226 strncpy(device->path, path, ARRAY_SIZE(device->path));
227
228 if (instance->enabled_extensions.KHR_display) {
229 master_fd =
230 open(drm_device->nodes[DRM_NODE_PRIMARY], O_RDWR | O_CLOEXEC);
231 if (master_fd >= 0) {
232 /* TODO: free master_fd is accel is not working? */
233 }
234 }
235
236 device->master_fd = master_fd;
237 device->local_fd = fd;
238
239 if (tu_drm_get_gpu_id(device, &device->gpu_id)) {
240 if (instance->debug_flags & TU_DEBUG_STARTUP)
241 tu_logi("Could not query the GPU ID");
242 result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
243 "could not get GPU ID");
244 goto fail;
245 }
246
247 if (tu_drm_get_gmem_size(device, &device->gmem_size)) {
248 if (instance->debug_flags & TU_DEBUG_STARTUP)
249 tu_logi("Could not query the GMEM size");
250 result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
251 "could not get GMEM size");
252 goto fail;
253 }
254
255 if (tu_drm_get_gmem_base(device, &device->gmem_base)) {
256 if (instance->debug_flags & TU_DEBUG_STARTUP)
257 tu_logi("Could not query the GMEM size");
258 result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
259 "could not get GMEM size");
260 goto fail;
261 }
262
263 memset(device->name, 0, sizeof(device->name));
264 sprintf(device->name, "FD%d", device->gpu_id);
265
266 switch (device->gpu_id) {
267 case 618:
268 device->magic.RB_UNKNOWN_8E04_blit = 0x00100000;
269 device->ccu_offset_gmem = 0x7c000; /* 0x7e000 in some cases? */
270 device->ccu_offset_bypass = 0x10000;
271 device->magic.PC_UNKNOWN_9805 = 0x0;
272 device->magic.SP_UNKNOWN_A0F8 = 0x0;
273 break;
274 case 630:
275 case 640:
276 device->magic.RB_UNKNOWN_8E04_blit = 0x01000000;
277 device->ccu_offset_gmem = 0xf8000;
278 device->ccu_offset_bypass = 0x20000;
279 device->magic.PC_UNKNOWN_9805 = 0x1;
280 device->magic.SP_UNKNOWN_A0F8 = 0x1;
281 break;
282 default:
283 result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
284 "device %s is unsupported", device->name);
285 goto fail;
286 }
287 if (tu_device_get_cache_uuid(device->gpu_id, device->cache_uuid)) {
288 result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
289 "cannot generate UUID");
290 goto fail;
291 }
292
293 /* The gpu id is already embedded in the uuid so we just pass "tu"
294 * when creating the cache.
295 */
296 char buf[VK_UUID_SIZE * 2 + 1];
297 disk_cache_format_hex_id(buf, device->cache_uuid, VK_UUID_SIZE * 2);
298 device->disk_cache = disk_cache_create(device->name, buf, 0);
299
300 fprintf(stderr, "WARNING: tu is not a conformant vulkan implementation, "
301 "testing use only.\n");
302
303 tu_get_driver_uuid(&device->device_uuid);
304 tu_get_device_uuid(&device->device_uuid);
305
306 tu_fill_device_extension_table(device, &device->supported_extensions);
307
308 if (result != VK_SUCCESS) {
309 vk_error(instance, result);
310 goto fail;
311 }
312
313 result = tu_wsi_init(device);
314 if (result != VK_SUCCESS) {
315 vk_error(instance, result);
316 goto fail;
317 }
318
319 return VK_SUCCESS;
320
321 fail:
322 close(fd);
323 if (master_fd != -1)
324 close(master_fd);
325 return result;
326 }
327
328 static void
329 tu_physical_device_finish(struct tu_physical_device *device)
330 {
331 tu_wsi_finish(device);
332
333 disk_cache_destroy(device->disk_cache);
334 close(device->local_fd);
335 if (device->master_fd != -1)
336 close(device->master_fd);
337 }
338
339 static VKAPI_ATTR void *
340 default_alloc_func(void *pUserData,
341 size_t size,
342 size_t align,
343 VkSystemAllocationScope allocationScope)
344 {
345 return malloc(size);
346 }
347
348 static VKAPI_ATTR void *
349 default_realloc_func(void *pUserData,
350 void *pOriginal,
351 size_t size,
352 size_t align,
353 VkSystemAllocationScope allocationScope)
354 {
355 return realloc(pOriginal, size);
356 }
357
358 static VKAPI_ATTR void
359 default_free_func(void *pUserData, void *pMemory)
360 {
361 free(pMemory);
362 }
363
364 static const VkAllocationCallbacks default_alloc = {
365 .pUserData = NULL,
366 .pfnAllocation = default_alloc_func,
367 .pfnReallocation = default_realloc_func,
368 .pfnFree = default_free_func,
369 };
370
371 static const struct debug_control tu_debug_options[] = {
372 { "startup", TU_DEBUG_STARTUP },
373 { "nir", TU_DEBUG_NIR },
374 { "ir3", TU_DEBUG_IR3 },
375 { "nobin", TU_DEBUG_NOBIN },
376 { "sysmem", TU_DEBUG_SYSMEM },
377 { "forcebin", TU_DEBUG_FORCEBIN },
378 { NULL, 0 }
379 };
380
381 const char *
382 tu_get_debug_option_name(int id)
383 {
384 assert(id < ARRAY_SIZE(tu_debug_options) - 1);
385 return tu_debug_options[id].string;
386 }
387
388 static int
389 tu_get_instance_extension_index(const char *name)
390 {
391 for (unsigned i = 0; i < TU_INSTANCE_EXTENSION_COUNT; ++i) {
392 if (strcmp(name, tu_instance_extensions[i].extensionName) == 0)
393 return i;
394 }
395 return -1;
396 }
397
398 VkResult
399 tu_CreateInstance(const VkInstanceCreateInfo *pCreateInfo,
400 const VkAllocationCallbacks *pAllocator,
401 VkInstance *pInstance)
402 {
403 struct tu_instance *instance;
404 VkResult result;
405
406 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);
407
408 uint32_t client_version;
409 if (pCreateInfo->pApplicationInfo &&
410 pCreateInfo->pApplicationInfo->apiVersion != 0) {
411 client_version = pCreateInfo->pApplicationInfo->apiVersion;
412 } else {
413 tu_EnumerateInstanceVersion(&client_version);
414 }
415
416 instance = vk_zalloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
417 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
418 if (!instance)
419 return vk_error(NULL, VK_ERROR_OUT_OF_HOST_MEMORY);
420
421 instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
422
423 if (pAllocator)
424 instance->alloc = *pAllocator;
425 else
426 instance->alloc = default_alloc;
427
428 instance->api_version = client_version;
429 instance->physical_device_count = -1;
430
431 instance->debug_flags =
432 parse_debug_string(getenv("TU_DEBUG"), tu_debug_options);
433
434 if (instance->debug_flags & TU_DEBUG_STARTUP)
435 tu_logi("Created an instance");
436
437 for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
438 const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i];
439 int index = tu_get_instance_extension_index(ext_name);
440
441 if (index < 0 || !tu_supported_instance_extensions.extensions[index]) {
442 vk_free2(&default_alloc, pAllocator, instance);
443 return vk_error(instance, VK_ERROR_EXTENSION_NOT_PRESENT);
444 }
445
446 instance->enabled_extensions.extensions[index] = true;
447 }
448
449 result = vk_debug_report_instance_init(&instance->debug_report_callbacks);
450 if (result != VK_SUCCESS) {
451 vk_free2(&default_alloc, pAllocator, instance);
452 return vk_error(instance, result);
453 }
454
455 glsl_type_singleton_init_or_ref();
456
457 VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
458
459 *pInstance = tu_instance_to_handle(instance);
460
461 return VK_SUCCESS;
462 }
463
464 void
465 tu_DestroyInstance(VkInstance _instance,
466 const VkAllocationCallbacks *pAllocator)
467 {
468 TU_FROM_HANDLE(tu_instance, instance, _instance);
469
470 if (!instance)
471 return;
472
473 for (int i = 0; i < instance->physical_device_count; ++i) {
474 tu_physical_device_finish(instance->physical_devices + i);
475 }
476
477 VG(VALGRIND_DESTROY_MEMPOOL(instance));
478
479 glsl_type_singleton_decref();
480
481 vk_debug_report_instance_destroy(&instance->debug_report_callbacks);
482
483 vk_free(&instance->alloc, instance);
484 }
485
486 static VkResult
487 tu_enumerate_devices(struct tu_instance *instance)
488 {
489 /* TODO: Check for more devices ? */
490 drmDevicePtr devices[8];
491 VkResult result = VK_ERROR_INCOMPATIBLE_DRIVER;
492 int max_devices;
493
494 instance->physical_device_count = 0;
495
496 max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
497
498 if (instance->debug_flags & TU_DEBUG_STARTUP)
499 tu_logi("Found %d drm nodes", max_devices);
500
501 if (max_devices < 1)
502 return vk_error(instance, VK_ERROR_INCOMPATIBLE_DRIVER);
503
504 for (unsigned i = 0; i < (unsigned) max_devices; i++) {
505 if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER &&
506 devices[i]->bustype == DRM_BUS_PLATFORM) {
507
508 result = tu_physical_device_init(
509 instance->physical_devices + instance->physical_device_count,
510 instance, devices[i]);
511 if (result == VK_SUCCESS)
512 ++instance->physical_device_count;
513 else if (result != VK_ERROR_INCOMPATIBLE_DRIVER)
514 break;
515 }
516 }
517 drmFreeDevices(devices, max_devices);
518
519 return result;
520 }
521
522 VkResult
523 tu_EnumeratePhysicalDevices(VkInstance _instance,
524 uint32_t *pPhysicalDeviceCount,
525 VkPhysicalDevice *pPhysicalDevices)
526 {
527 TU_FROM_HANDLE(tu_instance, instance, _instance);
528 VK_OUTARRAY_MAKE(out, pPhysicalDevices, pPhysicalDeviceCount);
529
530 VkResult result;
531
532 if (instance->physical_device_count < 0) {
533 result = tu_enumerate_devices(instance);
534 if (result != VK_SUCCESS && result != VK_ERROR_INCOMPATIBLE_DRIVER)
535 return result;
536 }
537
538 for (uint32_t i = 0; i < instance->physical_device_count; ++i) {
539 vk_outarray_append(&out, p)
540 {
541 *p = tu_physical_device_to_handle(instance->physical_devices + i);
542 }
543 }
544
545 return vk_outarray_status(&out);
546 }
547
548 VkResult
549 tu_EnumeratePhysicalDeviceGroups(
550 VkInstance _instance,
551 uint32_t *pPhysicalDeviceGroupCount,
552 VkPhysicalDeviceGroupProperties *pPhysicalDeviceGroupProperties)
553 {
554 TU_FROM_HANDLE(tu_instance, instance, _instance);
555 VK_OUTARRAY_MAKE(out, pPhysicalDeviceGroupProperties,
556 pPhysicalDeviceGroupCount);
557 VkResult result;
558
559 if (instance->physical_device_count < 0) {
560 result = tu_enumerate_devices(instance);
561 if (result != VK_SUCCESS && result != VK_ERROR_INCOMPATIBLE_DRIVER)
562 return result;
563 }
564
565 for (uint32_t i = 0; i < instance->physical_device_count; ++i) {
566 vk_outarray_append(&out, p)
567 {
568 p->physicalDeviceCount = 1;
569 p->physicalDevices[0] =
570 tu_physical_device_to_handle(instance->physical_devices + i);
571 p->subsetAllocation = false;
572 }
573 }
574
575 return vk_outarray_status(&out);
576 }
577
578 void
579 tu_GetPhysicalDeviceFeatures(VkPhysicalDevice physicalDevice,
580 VkPhysicalDeviceFeatures *pFeatures)
581 {
582 memset(pFeatures, 0, sizeof(*pFeatures));
583
584 *pFeatures = (VkPhysicalDeviceFeatures) {
585 .robustBufferAccess = false,
586 .fullDrawIndexUint32 = true,
587 .imageCubeArray = false,
588 .independentBlend = true,
589 .geometryShader = true,
590 .tessellationShader = false,
591 .sampleRateShading = true,
592 .dualSrcBlend = true,
593 .logicOp = true,
594 .multiDrawIndirect = false,
595 .drawIndirectFirstInstance = false,
596 .depthClamp = true,
597 .depthBiasClamp = false,
598 .fillModeNonSolid = false,
599 .depthBounds = false,
600 .wideLines = false,
601 .largePoints = false,
602 .alphaToOne = false,
603 .multiViewport = false,
604 .samplerAnisotropy = true,
605 .textureCompressionETC2 = true,
606 .textureCompressionASTC_LDR = true,
607 .textureCompressionBC = true,
608 .occlusionQueryPrecise = true,
609 .pipelineStatisticsQuery = false,
610 .vertexPipelineStoresAndAtomics = false,
611 .fragmentStoresAndAtomics = false,
612 .shaderTessellationAndGeometryPointSize = false,
613 .shaderImageGatherExtended = false,
614 .shaderStorageImageExtendedFormats = false,
615 .shaderStorageImageMultisample = false,
616 .shaderUniformBufferArrayDynamicIndexing = false,
617 .shaderSampledImageArrayDynamicIndexing = false,
618 .shaderStorageBufferArrayDynamicIndexing = false,
619 .shaderStorageImageArrayDynamicIndexing = false,
620 .shaderStorageImageReadWithoutFormat = false,
621 .shaderStorageImageWriteWithoutFormat = false,
622 .shaderClipDistance = false,
623 .shaderCullDistance = false,
624 .shaderFloat64 = false,
625 .shaderInt64 = false,
626 .shaderInt16 = false,
627 .sparseBinding = false,
628 .variableMultisampleRate = false,
629 .inheritedQueries = false,
630 };
631 }
632
633 void
634 tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
635 VkPhysicalDeviceFeatures2 *pFeatures)
636 {
637 vk_foreach_struct(ext, pFeatures->pNext)
638 {
639 switch (ext->sType) {
640 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: {
641 VkPhysicalDeviceVariablePointersFeatures *features = (void *) ext;
642 features->variablePointersStorageBuffer = false;
643 features->variablePointers = false;
644 break;
645 }
646 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
647 VkPhysicalDeviceMultiviewFeatures *features =
648 (VkPhysicalDeviceMultiviewFeatures *) ext;
649 features->multiview = false;
650 features->multiviewGeometryShader = false;
651 features->multiviewTessellationShader = false;
652 break;
653 }
654 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
655 VkPhysicalDeviceShaderDrawParametersFeatures *features =
656 (VkPhysicalDeviceShaderDrawParametersFeatures *) ext;
657 features->shaderDrawParameters = false;
658 break;
659 }
660 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
661 VkPhysicalDeviceProtectedMemoryFeatures *features =
662 (VkPhysicalDeviceProtectedMemoryFeatures *) ext;
663 features->protectedMemory = false;
664 break;
665 }
666 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
667 VkPhysicalDevice16BitStorageFeatures *features =
668 (VkPhysicalDevice16BitStorageFeatures *) ext;
669 features->storageBuffer16BitAccess = false;
670 features->uniformAndStorageBuffer16BitAccess = false;
671 features->storagePushConstant16 = false;
672 features->storageInputOutput16 = false;
673 break;
674 }
675 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: {
676 VkPhysicalDeviceSamplerYcbcrConversionFeatures *features =
677 (VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext;
678 features->samplerYcbcrConversion = false;
679 break;
680 }
681 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: {
682 VkPhysicalDeviceDescriptorIndexingFeaturesEXT *features =
683 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT *) ext;
684 features->shaderInputAttachmentArrayDynamicIndexing = false;
685 features->shaderUniformTexelBufferArrayDynamicIndexing = false;
686 features->shaderStorageTexelBufferArrayDynamicIndexing = false;
687 features->shaderUniformBufferArrayNonUniformIndexing = false;
688 features->shaderSampledImageArrayNonUniformIndexing = false;
689 features->shaderStorageBufferArrayNonUniformIndexing = false;
690 features->shaderStorageImageArrayNonUniformIndexing = false;
691 features->shaderInputAttachmentArrayNonUniformIndexing = false;
692 features->shaderUniformTexelBufferArrayNonUniformIndexing = false;
693 features->shaderStorageTexelBufferArrayNonUniformIndexing = false;
694 features->descriptorBindingUniformBufferUpdateAfterBind = false;
695 features->descriptorBindingSampledImageUpdateAfterBind = false;
696 features->descriptorBindingStorageImageUpdateAfterBind = false;
697 features->descriptorBindingStorageBufferUpdateAfterBind = false;
698 features->descriptorBindingUniformTexelBufferUpdateAfterBind = false;
699 features->descriptorBindingStorageTexelBufferUpdateAfterBind = false;
700 features->descriptorBindingUpdateUnusedWhilePending = false;
701 features->descriptorBindingPartiallyBound = false;
702 features->descriptorBindingVariableDescriptorCount = false;
703 features->runtimeDescriptorArray = false;
704 break;
705 }
706 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
707 VkPhysicalDeviceConditionalRenderingFeaturesEXT *features =
708 (VkPhysicalDeviceConditionalRenderingFeaturesEXT *) ext;
709 features->conditionalRendering = false;
710 features->inheritedConditionalRendering = false;
711 break;
712 }
713 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT: {
714 VkPhysicalDeviceTransformFeedbackFeaturesEXT *features =
715 (VkPhysicalDeviceTransformFeedbackFeaturesEXT *) ext;
716 features->transformFeedback = true;
717 features->geometryStreams = false;
718 break;
719 }
720 default:
721 break;
722 }
723 }
724 return tu_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
725 }
726
727 void
728 tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice,
729 VkPhysicalDeviceProperties *pProperties)
730 {
731 TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
732 VkSampleCountFlags sample_counts =
733 VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT;
734
735 /* I have no idea what the maximum size is, but the hardware supports very
736 * large numbers of descriptors (at least 2^16). This limit is based on
737 * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
738 * we don't have to think about what to do if that overflows, but really
739 * nothing is likely to get close to this.
740 */
741 const size_t max_descriptor_set_size = (1 << 28) / A6XX_TEX_CONST_DWORDS;
742
743 VkPhysicalDeviceLimits limits = {
744 .maxImageDimension1D = (1 << 14),
745 .maxImageDimension2D = (1 << 14),
746 .maxImageDimension3D = (1 << 11),
747 .maxImageDimensionCube = (1 << 14),
748 .maxImageArrayLayers = (1 << 11),
749 .maxTexelBufferElements = 128 * 1024 * 1024,
750 .maxUniformBufferRange = MAX_UNIFORM_BUFFER_RANGE,
751 .maxStorageBufferRange = MAX_STORAGE_BUFFER_RANGE,
752 .maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE,
753 .maxMemoryAllocationCount = UINT32_MAX,
754 .maxSamplerAllocationCount = 64 * 1024,
755 .bufferImageGranularity = 64, /* A cache line */
756 .sparseAddressSpaceSize = 0xffffffffu, /* buffer max size */
757 .maxBoundDescriptorSets = MAX_SETS,
758 .maxPerStageDescriptorSamplers = max_descriptor_set_size,
759 .maxPerStageDescriptorUniformBuffers = max_descriptor_set_size,
760 .maxPerStageDescriptorStorageBuffers = max_descriptor_set_size,
761 .maxPerStageDescriptorSampledImages = max_descriptor_set_size,
762 .maxPerStageDescriptorStorageImages = max_descriptor_set_size,
763 .maxPerStageDescriptorInputAttachments = MAX_RTS,
764 .maxPerStageResources = max_descriptor_set_size,
765 .maxDescriptorSetSamplers = max_descriptor_set_size,
766 .maxDescriptorSetUniformBuffers = max_descriptor_set_size,
767 .maxDescriptorSetUniformBuffersDynamic = MAX_DYNAMIC_UNIFORM_BUFFERS,
768 .maxDescriptorSetStorageBuffers = max_descriptor_set_size,
769 .maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_STORAGE_BUFFERS,
770 .maxDescriptorSetSampledImages = max_descriptor_set_size,
771 .maxDescriptorSetStorageImages = max_descriptor_set_size,
772 .maxDescriptorSetInputAttachments = MAX_RTS,
773 .maxVertexInputAttributes = 32,
774 .maxVertexInputBindings = 32,
775 .maxVertexInputAttributeOffset = 4095,
776 .maxVertexInputBindingStride = 2048,
777 .maxVertexOutputComponents = 128,
778 .maxTessellationGenerationLevel = 64,
779 .maxTessellationPatchSize = 32,
780 .maxTessellationControlPerVertexInputComponents = 128,
781 .maxTessellationControlPerVertexOutputComponents = 128,
782 .maxTessellationControlPerPatchOutputComponents = 120,
783 .maxTessellationControlTotalOutputComponents = 4096,
784 .maxTessellationEvaluationInputComponents = 128,
785 .maxTessellationEvaluationOutputComponents = 128,
786 .maxGeometryShaderInvocations = 32,
787 .maxGeometryInputComponents = 64,
788 .maxGeometryOutputComponents = 128,
789 .maxGeometryOutputVertices = 256,
790 .maxGeometryTotalOutputComponents = 1024,
791 .maxFragmentInputComponents = 128,
792 .maxFragmentOutputAttachments = 8,
793 .maxFragmentDualSrcAttachments = 1,
794 .maxFragmentCombinedOutputResources = 8,
795 .maxComputeSharedMemorySize = 32768,
796 .maxComputeWorkGroupCount = { 65535, 65535, 65535 },
797 .maxComputeWorkGroupInvocations = 2048,
798 .maxComputeWorkGroupSize = { 2048, 2048, 2048 },
799 .subPixelPrecisionBits = 8,
800 .subTexelPrecisionBits = 4 /* FIXME */,
801 .mipmapPrecisionBits = 4 /* FIXME */,
802 .maxDrawIndexedIndexValue = UINT32_MAX,
803 .maxDrawIndirectCount = UINT32_MAX,
804 .maxSamplerLodBias = 16,
805 .maxSamplerAnisotropy = 16,
806 .maxViewports = MAX_VIEWPORTS,
807 .maxViewportDimensions = { (1 << 14), (1 << 14) },
808 .viewportBoundsRange = { INT16_MIN, INT16_MAX },
809 .viewportSubPixelBits = 8,
810 .minMemoryMapAlignment = 4096, /* A page */
811 .minTexelBufferOffsetAlignment = 64,
812 .minUniformBufferOffsetAlignment = 64,
813 .minStorageBufferOffsetAlignment = 64,
814 .minTexelOffset = -32,
815 .maxTexelOffset = 31,
816 .minTexelGatherOffset = -32,
817 .maxTexelGatherOffset = 31,
818 .minInterpolationOffset = -2,
819 .maxInterpolationOffset = 2,
820 .subPixelInterpolationOffsetBits = 8,
821 .maxFramebufferWidth = (1 << 14),
822 .maxFramebufferHeight = (1 << 14),
823 .maxFramebufferLayers = (1 << 10),
824 .framebufferColorSampleCounts = sample_counts,
825 .framebufferDepthSampleCounts = sample_counts,
826 .framebufferStencilSampleCounts = sample_counts,
827 .framebufferNoAttachmentsSampleCounts = sample_counts,
828 .maxColorAttachments = MAX_RTS,
829 .sampledImageColorSampleCounts = sample_counts,
830 .sampledImageIntegerSampleCounts = VK_SAMPLE_COUNT_1_BIT,
831 .sampledImageDepthSampleCounts = sample_counts,
832 .sampledImageStencilSampleCounts = sample_counts,
833 .storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
834 .maxSampleMaskWords = 1,
835 .timestampComputeAndGraphics = true,
836 .timestampPeriod = 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
837 .maxClipDistances = 8,
838 .maxCullDistances = 8,
839 .maxCombinedClipAndCullDistances = 8,
840 .discreteQueuePriorities = 1,
841 .pointSizeRange = { 0.125, 255.875 },
842 .lineWidthRange = { 0.0, 7.9921875 },
843 .pointSizeGranularity = (1.0 / 8.0),
844 .lineWidthGranularity = (1.0 / 128.0),
845 .strictLines = false, /* FINISHME */
846 .standardSampleLocations = true,
847 .optimalBufferCopyOffsetAlignment = 128,
848 .optimalBufferCopyRowPitchAlignment = 128,
849 .nonCoherentAtomSize = 64,
850 };
851
852 *pProperties = (VkPhysicalDeviceProperties) {
853 .apiVersion = tu_physical_device_api_version(pdevice),
854 .driverVersion = vk_get_driver_version(),
855 .vendorID = 0, /* TODO */
856 .deviceID = 0,
857 .deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
858 .limits = limits,
859 .sparseProperties = { 0 },
860 };
861
862 strcpy(pProperties->deviceName, pdevice->name);
863 memcpy(pProperties->pipelineCacheUUID, pdevice->cache_uuid, VK_UUID_SIZE);
864 }
865
866 void
867 tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice,
868 VkPhysicalDeviceProperties2 *pProperties)
869 {
870 TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
871 tu_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
872
873 vk_foreach_struct(ext, pProperties->pNext)
874 {
875 switch (ext->sType) {
876 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
877 VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
878 (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
879 properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
880 break;
881 }
882 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: {
883 VkPhysicalDeviceIDProperties *properties =
884 (VkPhysicalDeviceIDProperties *) ext;
885 memcpy(properties->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE);
886 memcpy(properties->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE);
887 properties->deviceLUIDValid = false;
888 break;
889 }
890 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES: {
891 VkPhysicalDeviceMultiviewProperties *properties =
892 (VkPhysicalDeviceMultiviewProperties *) ext;
893 properties->maxMultiviewViewCount = MAX_VIEWS;
894 properties->maxMultiviewInstanceIndex = INT_MAX;
895 break;
896 }
897 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES: {
898 VkPhysicalDevicePointClippingProperties *properties =
899 (VkPhysicalDevicePointClippingProperties *) ext;
900 properties->pointClippingBehavior =
901 VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES;
902 break;
903 }
904 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: {
905 VkPhysicalDeviceMaintenance3Properties *properties =
906 (VkPhysicalDeviceMaintenance3Properties *) ext;
907 /* Make sure everything is addressable by a signed 32-bit int, and
908 * our largest descriptors are 96 bytes. */
909 properties->maxPerSetDescriptors = (1ull << 31) / 96;
910 /* Our buffer size fields allow only this much */
911 properties->maxMemoryAllocationSize = 0xFFFFFFFFull;
912 break;
913 }
914 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT: {
915 VkPhysicalDeviceTransformFeedbackPropertiesEXT *properties =
916 (VkPhysicalDeviceTransformFeedbackPropertiesEXT *)ext;
917
918 properties->maxTransformFeedbackStreams = IR3_MAX_SO_STREAMS;
919 properties->maxTransformFeedbackBuffers = IR3_MAX_SO_BUFFERS;
920 properties->maxTransformFeedbackBufferSize = UINT32_MAX;
921 properties->maxTransformFeedbackStreamDataSize = 512;
922 properties->maxTransformFeedbackBufferDataSize = 512;
923 properties->maxTransformFeedbackBufferDataStride = 512;
924 /* TODO: enable xfb query */
925 properties->transformFeedbackQueries = false;
926 properties->transformFeedbackStreamsLinesTriangles = false;
927 properties->transformFeedbackRasterizationStreamSelect = false;
928 properties->transformFeedbackDraw = true;
929 break;
930 }
931 default:
932 break;
933 }
934 }
935 }
936
937 static const VkQueueFamilyProperties tu_queue_family_properties = {
938 .queueFlags =
939 VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_COMPUTE_BIT | VK_QUEUE_TRANSFER_BIT,
940 .queueCount = 1,
941 .timestampValidBits = 48,
942 .minImageTransferGranularity = { 1, 1, 1 },
943 };
944
945 void
946 tu_GetPhysicalDeviceQueueFamilyProperties(
947 VkPhysicalDevice physicalDevice,
948 uint32_t *pQueueFamilyPropertyCount,
949 VkQueueFamilyProperties *pQueueFamilyProperties)
950 {
951 VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pQueueFamilyPropertyCount);
952
953 vk_outarray_append(&out, p) { *p = tu_queue_family_properties; }
954 }
955
956 void
957 tu_GetPhysicalDeviceQueueFamilyProperties2(
958 VkPhysicalDevice physicalDevice,
959 uint32_t *pQueueFamilyPropertyCount,
960 VkQueueFamilyProperties2 *pQueueFamilyProperties)
961 {
962 VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pQueueFamilyPropertyCount);
963
964 vk_outarray_append(&out, p)
965 {
966 p->queueFamilyProperties = tu_queue_family_properties;
967 }
968 }
969
970 static uint64_t
971 tu_get_system_heap_size()
972 {
973 struct sysinfo info;
974 sysinfo(&info);
975
976 uint64_t total_ram = (uint64_t) info.totalram * (uint64_t) info.mem_unit;
977
978 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
979 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
980 */
981 uint64_t available_ram;
982 if (total_ram <= 4ull * 1024ull * 1024ull * 1024ull)
983 available_ram = total_ram / 2;
984 else
985 available_ram = total_ram * 3 / 4;
986
987 return available_ram;
988 }
989
990 void
991 tu_GetPhysicalDeviceMemoryProperties(
992 VkPhysicalDevice physicalDevice,
993 VkPhysicalDeviceMemoryProperties *pMemoryProperties)
994 {
995 pMemoryProperties->memoryHeapCount = 1;
996 pMemoryProperties->memoryHeaps[0].size = tu_get_system_heap_size();
997 pMemoryProperties->memoryHeaps[0].flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT;
998
999 pMemoryProperties->memoryTypeCount = 1;
1000 pMemoryProperties->memoryTypes[0].propertyFlags =
1001 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
1002 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
1003 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
1004 pMemoryProperties->memoryTypes[0].heapIndex = 0;
1005 }
1006
1007 void
1008 tu_GetPhysicalDeviceMemoryProperties2(
1009 VkPhysicalDevice physicalDevice,
1010 VkPhysicalDeviceMemoryProperties2 *pMemoryProperties)
1011 {
1012 return tu_GetPhysicalDeviceMemoryProperties(
1013 physicalDevice, &pMemoryProperties->memoryProperties);
1014 }
1015
1016 static VkResult
1017 tu_queue_init(struct tu_device *device,
1018 struct tu_queue *queue,
1019 uint32_t queue_family_index,
1020 int idx,
1021 VkDeviceQueueCreateFlags flags)
1022 {
1023 queue->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
1024 queue->device = device;
1025 queue->queue_family_index = queue_family_index;
1026 queue->queue_idx = idx;
1027 queue->flags = flags;
1028
1029 int ret = tu_drm_submitqueue_new(device, 0, &queue->msm_queue_id);
1030 if (ret)
1031 return VK_ERROR_INITIALIZATION_FAILED;
1032
1033 tu_fence_init(&queue->submit_fence, false);
1034
1035 return VK_SUCCESS;
1036 }
1037
1038 static void
1039 tu_queue_finish(struct tu_queue *queue)
1040 {
1041 tu_fence_finish(&queue->submit_fence);
1042 tu_drm_submitqueue_close(queue->device, queue->msm_queue_id);
1043 }
1044
1045 static int
1046 tu_get_device_extension_index(const char *name)
1047 {
1048 for (unsigned i = 0; i < TU_DEVICE_EXTENSION_COUNT; ++i) {
1049 if (strcmp(name, tu_device_extensions[i].extensionName) == 0)
1050 return i;
1051 }
1052 return -1;
1053 }
1054
1055 struct PACKED bcolor_entry {
1056 uint32_t fp32[4];
1057 uint16_t ui16[4];
1058 int16_t si16[4];
1059 uint16_t fp16[4];
1060 uint16_t rgb565;
1061 uint16_t rgb5a1;
1062 uint16_t rgba4;
1063 uint8_t __pad0[2];
1064 uint8_t ui8[4];
1065 int8_t si8[4];
1066 uint32_t rgb10a2;
1067 uint32_t z24; /* also s8? */
1068 uint16_t srgb[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
1069 uint8_t __pad1[56];
1070 } border_color[] = {
1071 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = {},
1072 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = {},
1073 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = {
1074 .fp32[3] = 0x3f800000,
1075 .ui16[3] = 0xffff,
1076 .si16[3] = 0x7fff,
1077 .fp16[3] = 0x3c00,
1078 .rgb5a1 = 0x8000,
1079 .rgba4 = 0xf000,
1080 .ui8[3] = 0xff,
1081 .si8[3] = 0x7f,
1082 .rgb10a2 = 0xc0000000,
1083 .srgb[3] = 0x3c00,
1084 },
1085 [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = {
1086 .fp32[3] = 1,
1087 .fp16[3] = 1,
1088 },
1089 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = {
1090 .fp32[0 ... 3] = 0x3f800000,
1091 .ui16[0 ... 3] = 0xffff,
1092 .si16[0 ... 3] = 0x7fff,
1093 .fp16[0 ... 3] = 0x3c00,
1094 .rgb565 = 0xffff,
1095 .rgb5a1 = 0xffff,
1096 .rgba4 = 0xffff,
1097 .ui8[0 ... 3] = 0xff,
1098 .si8[0 ... 3] = 0x7f,
1099 .rgb10a2 = 0xffffffff,
1100 .z24 = 0xffffff,
1101 .srgb[0 ... 3] = 0x3c00,
1102 },
1103 [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = {
1104 .fp32[0 ... 3] = 1,
1105 .fp16[0 ... 3] = 1,
1106 },
1107 };
1108
1109
1110 VkResult
1111 tu_CreateDevice(VkPhysicalDevice physicalDevice,
1112 const VkDeviceCreateInfo *pCreateInfo,
1113 const VkAllocationCallbacks *pAllocator,
1114 VkDevice *pDevice)
1115 {
1116 TU_FROM_HANDLE(tu_physical_device, physical_device, physicalDevice);
1117 VkResult result;
1118 struct tu_device *device;
1119
1120 /* Check enabled features */
1121 if (pCreateInfo->pEnabledFeatures) {
1122 VkPhysicalDeviceFeatures supported_features;
1123 tu_GetPhysicalDeviceFeatures(physicalDevice, &supported_features);
1124 VkBool32 *supported_feature = (VkBool32 *) &supported_features;
1125 VkBool32 *enabled_feature = (VkBool32 *) pCreateInfo->pEnabledFeatures;
1126 unsigned num_features =
1127 sizeof(VkPhysicalDeviceFeatures) / sizeof(VkBool32);
1128 for (uint32_t i = 0; i < num_features; i++) {
1129 if (enabled_feature[i] && !supported_feature[i])
1130 return vk_error(physical_device->instance,
1131 VK_ERROR_FEATURE_NOT_PRESENT);
1132 }
1133 }
1134
1135 device = vk_zalloc2(&physical_device->instance->alloc, pAllocator,
1136 sizeof(*device), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
1137 if (!device)
1138 return vk_error(physical_device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1139
1140 device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
1141 device->instance = physical_device->instance;
1142 device->physical_device = physical_device;
1143
1144 if (pAllocator)
1145 device->alloc = *pAllocator;
1146 else
1147 device->alloc = physical_device->instance->alloc;
1148
1149 for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
1150 const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i];
1151 int index = tu_get_device_extension_index(ext_name);
1152 if (index < 0 ||
1153 !physical_device->supported_extensions.extensions[index]) {
1154 vk_free(&device->alloc, device);
1155 return vk_error(physical_device->instance,
1156 VK_ERROR_EXTENSION_NOT_PRESENT);
1157 }
1158
1159 device->enabled_extensions.extensions[index] = true;
1160 }
1161
1162 for (unsigned i = 0; i < pCreateInfo->queueCreateInfoCount; i++) {
1163 const VkDeviceQueueCreateInfo *queue_create =
1164 &pCreateInfo->pQueueCreateInfos[i];
1165 uint32_t qfi = queue_create->queueFamilyIndex;
1166 device->queues[qfi] = vk_alloc(
1167 &device->alloc, queue_create->queueCount * sizeof(struct tu_queue),
1168 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
1169 if (!device->queues[qfi]) {
1170 result = VK_ERROR_OUT_OF_HOST_MEMORY;
1171 goto fail_queues;
1172 }
1173
1174 memset(device->queues[qfi], 0,
1175 queue_create->queueCount * sizeof(struct tu_queue));
1176
1177 device->queue_count[qfi] = queue_create->queueCount;
1178
1179 for (unsigned q = 0; q < queue_create->queueCount; q++) {
1180 result = tu_queue_init(device, &device->queues[qfi][q], qfi, q,
1181 queue_create->flags);
1182 if (result != VK_SUCCESS)
1183 goto fail_queues;
1184 }
1185 }
1186
1187 device->compiler = ir3_compiler_create(NULL, physical_device->gpu_id);
1188 if (!device->compiler)
1189 goto fail_queues;
1190
1191 #define VSC_DATA_SIZE(pitch) ((pitch) * 32 + 0x100) /* extra size to store VSC_SIZE */
1192 #define VSC_DATA2_SIZE(pitch) ((pitch) * 32)
1193
1194 device->vsc_data_pitch = 0x440 * 4;
1195 device->vsc_data2_pitch = 0x1040 * 4;
1196
1197 result = tu_bo_init_new(device, &device->vsc_data, VSC_DATA_SIZE(device->vsc_data_pitch));
1198 if (result != VK_SUCCESS)
1199 goto fail_vsc_data;
1200
1201 result = tu_bo_init_new(device, &device->vsc_data2, VSC_DATA2_SIZE(device->vsc_data2_pitch));
1202 if (result != VK_SUCCESS)
1203 goto fail_vsc_data2;
1204
1205 STATIC_ASSERT(sizeof(struct bcolor_entry) == 128);
1206 result = tu_bo_init_new(device, &device->border_color, sizeof(border_color));
1207 if (result != VK_SUCCESS)
1208 goto fail_border_color;
1209
1210 result = tu_bo_map(device, &device->border_color);
1211 if (result != VK_SUCCESS)
1212 goto fail_border_color_map;
1213
1214 memcpy(device->border_color.map, border_color, sizeof(border_color));
1215
1216 VkPipelineCacheCreateInfo ci;
1217 ci.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO;
1218 ci.pNext = NULL;
1219 ci.flags = 0;
1220 ci.pInitialData = NULL;
1221 ci.initialDataSize = 0;
1222 VkPipelineCache pc;
1223 result =
1224 tu_CreatePipelineCache(tu_device_to_handle(device), &ci, NULL, &pc);
1225 if (result != VK_SUCCESS)
1226 goto fail_pipeline_cache;
1227
1228 device->mem_cache = tu_pipeline_cache_from_handle(pc);
1229
1230 *pDevice = tu_device_to_handle(device);
1231 return VK_SUCCESS;
1232
1233 fail_pipeline_cache:
1234 fail_border_color_map:
1235 tu_bo_finish(device, &device->border_color);
1236
1237 fail_border_color:
1238 tu_bo_finish(device, &device->vsc_data2);
1239
1240 fail_vsc_data2:
1241 tu_bo_finish(device, &device->vsc_data);
1242
1243 fail_vsc_data:
1244 ralloc_free(device->compiler);
1245
1246 fail_queues:
1247 for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) {
1248 for (unsigned q = 0; q < device->queue_count[i]; q++)
1249 tu_queue_finish(&device->queues[i][q]);
1250 if (device->queue_count[i])
1251 vk_free(&device->alloc, device->queues[i]);
1252 }
1253
1254 vk_free(&device->alloc, device);
1255 return result;
1256 }
1257
1258 void
1259 tu_DestroyDevice(VkDevice _device, const VkAllocationCallbacks *pAllocator)
1260 {
1261 TU_FROM_HANDLE(tu_device, device, _device);
1262
1263 if (!device)
1264 return;
1265
1266 tu_bo_finish(device, &device->vsc_data);
1267 tu_bo_finish(device, &device->vsc_data2);
1268
1269 for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) {
1270 for (unsigned q = 0; q < device->queue_count[i]; q++)
1271 tu_queue_finish(&device->queues[i][q]);
1272 if (device->queue_count[i])
1273 vk_free(&device->alloc, device->queues[i]);
1274 }
1275
1276 /* the compiler does not use pAllocator */
1277 ralloc_free(device->compiler);
1278
1279 VkPipelineCache pc = tu_pipeline_cache_to_handle(device->mem_cache);
1280 tu_DestroyPipelineCache(tu_device_to_handle(device), pc, NULL);
1281
1282 vk_free(&device->alloc, device);
1283 }
1284
1285 VkResult
1286 tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount,
1287 VkLayerProperties *pProperties)
1288 {
1289 *pPropertyCount = 0;
1290 return VK_SUCCESS;
1291 }
1292
1293 VkResult
1294 tu_EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice,
1295 uint32_t *pPropertyCount,
1296 VkLayerProperties *pProperties)
1297 {
1298 *pPropertyCount = 0;
1299 return VK_SUCCESS;
1300 }
1301
1302 void
1303 tu_GetDeviceQueue2(VkDevice _device,
1304 const VkDeviceQueueInfo2 *pQueueInfo,
1305 VkQueue *pQueue)
1306 {
1307 TU_FROM_HANDLE(tu_device, device, _device);
1308 struct tu_queue *queue;
1309
1310 queue =
1311 &device->queues[pQueueInfo->queueFamilyIndex][pQueueInfo->queueIndex];
1312 if (pQueueInfo->flags != queue->flags) {
1313 /* From the Vulkan 1.1.70 spec:
1314 *
1315 * "The queue returned by vkGetDeviceQueue2 must have the same
1316 * flags value from this structure as that used at device
1317 * creation time in a VkDeviceQueueCreateInfo instance. If no
1318 * matching flags were specified at device creation time then
1319 * pQueue will return VK_NULL_HANDLE."
1320 */
1321 *pQueue = VK_NULL_HANDLE;
1322 return;
1323 }
1324
1325 *pQueue = tu_queue_to_handle(queue);
1326 }
1327
1328 void
1329 tu_GetDeviceQueue(VkDevice _device,
1330 uint32_t queueFamilyIndex,
1331 uint32_t queueIndex,
1332 VkQueue *pQueue)
1333 {
1334 const VkDeviceQueueInfo2 info =
1335 (VkDeviceQueueInfo2) { .sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2,
1336 .queueFamilyIndex = queueFamilyIndex,
1337 .queueIndex = queueIndex };
1338
1339 tu_GetDeviceQueue2(_device, &info, pQueue);
1340 }
1341
1342 VkResult
1343 tu_QueueSubmit(VkQueue _queue,
1344 uint32_t submitCount,
1345 const VkSubmitInfo *pSubmits,
1346 VkFence _fence)
1347 {
1348 TU_FROM_HANDLE(tu_queue, queue, _queue);
1349
1350 for (uint32_t i = 0; i < submitCount; ++i) {
1351 const VkSubmitInfo *submit = pSubmits + i;
1352 const bool last_submit = (i == submitCount - 1);
1353 struct tu_bo_list bo_list;
1354 tu_bo_list_init(&bo_list);
1355
1356 uint32_t entry_count = 0;
1357 for (uint32_t j = 0; j < submit->commandBufferCount; ++j) {
1358 TU_FROM_HANDLE(tu_cmd_buffer, cmdbuf, submit->pCommandBuffers[j]);
1359 entry_count += cmdbuf->cs.entry_count;
1360 }
1361
1362 struct drm_msm_gem_submit_cmd cmds[entry_count];
1363 uint32_t entry_idx = 0;
1364 for (uint32_t j = 0; j < submit->commandBufferCount; ++j) {
1365 TU_FROM_HANDLE(tu_cmd_buffer, cmdbuf, submit->pCommandBuffers[j]);
1366 struct tu_cs *cs = &cmdbuf->cs;
1367 for (unsigned i = 0; i < cs->entry_count; ++i, ++entry_idx) {
1368 cmds[entry_idx].type = MSM_SUBMIT_CMD_BUF;
1369 cmds[entry_idx].submit_idx =
1370 tu_bo_list_add(&bo_list, cs->entries[i].bo,
1371 MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_DUMP);
1372 cmds[entry_idx].submit_offset = cs->entries[i].offset;
1373 cmds[entry_idx].size = cs->entries[i].size;
1374 cmds[entry_idx].pad = 0;
1375 cmds[entry_idx].nr_relocs = 0;
1376 cmds[entry_idx].relocs = 0;
1377 }
1378
1379 tu_bo_list_merge(&bo_list, &cmdbuf->bo_list);
1380 }
1381
1382 uint32_t flags = MSM_PIPE_3D0;
1383 if (last_submit) {
1384 flags |= MSM_SUBMIT_FENCE_FD_OUT;
1385 }
1386
1387 struct drm_msm_gem_submit req = {
1388 .flags = flags,
1389 .queueid = queue->msm_queue_id,
1390 .bos = (uint64_t)(uintptr_t) bo_list.bo_infos,
1391 .nr_bos = bo_list.count,
1392 .cmds = (uint64_t)(uintptr_t)cmds,
1393 .nr_cmds = entry_count,
1394 };
1395
1396 int ret = drmCommandWriteRead(queue->device->physical_device->local_fd,
1397 DRM_MSM_GEM_SUBMIT,
1398 &req, sizeof(req));
1399 if (ret) {
1400 fprintf(stderr, "submit failed: %s\n", strerror(errno));
1401 abort();
1402 }
1403
1404 tu_bo_list_destroy(&bo_list);
1405
1406 if (last_submit) {
1407 /* no need to merge fences as queue execution is serialized */
1408 tu_fence_update_fd(&queue->submit_fence, req.fence_fd);
1409 }
1410 }
1411
1412 if (_fence != VK_NULL_HANDLE) {
1413 TU_FROM_HANDLE(tu_fence, fence, _fence);
1414 tu_fence_copy(fence, &queue->submit_fence);
1415 }
1416
1417 return VK_SUCCESS;
1418 }
1419
1420 VkResult
1421 tu_QueueWaitIdle(VkQueue _queue)
1422 {
1423 TU_FROM_HANDLE(tu_queue, queue, _queue);
1424
1425 tu_fence_wait_idle(&queue->submit_fence);
1426
1427 return VK_SUCCESS;
1428 }
1429
1430 VkResult
1431 tu_DeviceWaitIdle(VkDevice _device)
1432 {
1433 TU_FROM_HANDLE(tu_device, device, _device);
1434
1435 for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) {
1436 for (unsigned q = 0; q < device->queue_count[i]; q++) {
1437 tu_QueueWaitIdle(tu_queue_to_handle(&device->queues[i][q]));
1438 }
1439 }
1440 return VK_SUCCESS;
1441 }
1442
1443 VkResult
1444 tu_ImportSemaphoreFdKHR(VkDevice _device,
1445 const VkImportSemaphoreFdInfoKHR *pImportSemaphoreFdInfo)
1446 {
1447 tu_stub();
1448
1449 return VK_SUCCESS;
1450 }
1451
1452 VkResult
1453 tu_GetSemaphoreFdKHR(VkDevice _device,
1454 const VkSemaphoreGetFdInfoKHR *pGetFdInfo,
1455 int *pFd)
1456 {
1457 tu_stub();
1458
1459 return VK_SUCCESS;
1460 }
1461
1462 VkResult
1463 tu_ImportFenceFdKHR(VkDevice _device,
1464 const VkImportFenceFdInfoKHR *pImportFenceFdInfo)
1465 {
1466 tu_stub();
1467
1468 return VK_SUCCESS;
1469 }
1470
1471 VkResult
1472 tu_GetFenceFdKHR(VkDevice _device,
1473 const VkFenceGetFdInfoKHR *pGetFdInfo,
1474 int *pFd)
1475 {
1476 tu_stub();
1477
1478 return VK_SUCCESS;
1479 }
1480
1481 VkResult
1482 tu_EnumerateInstanceExtensionProperties(const char *pLayerName,
1483 uint32_t *pPropertyCount,
1484 VkExtensionProperties *pProperties)
1485 {
1486 VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
1487
1488 /* We spport no lyaers */
1489 if (pLayerName)
1490 return vk_error(NULL, VK_ERROR_LAYER_NOT_PRESENT);
1491
1492 for (int i = 0; i < TU_INSTANCE_EXTENSION_COUNT; i++) {
1493 if (tu_supported_instance_extensions.extensions[i]) {
1494 vk_outarray_append(&out, prop) { *prop = tu_instance_extensions[i]; }
1495 }
1496 }
1497
1498 return vk_outarray_status(&out);
1499 }
1500
1501 VkResult
1502 tu_EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice,
1503 const char *pLayerName,
1504 uint32_t *pPropertyCount,
1505 VkExtensionProperties *pProperties)
1506 {
1507 /* We spport no lyaers */
1508 TU_FROM_HANDLE(tu_physical_device, device, physicalDevice);
1509 VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
1510
1511 /* We spport no lyaers */
1512 if (pLayerName)
1513 return vk_error(NULL, VK_ERROR_LAYER_NOT_PRESENT);
1514
1515 for (int i = 0; i < TU_DEVICE_EXTENSION_COUNT; i++) {
1516 if (device->supported_extensions.extensions[i]) {
1517 vk_outarray_append(&out, prop) { *prop = tu_device_extensions[i]; }
1518 }
1519 }
1520
1521 return vk_outarray_status(&out);
1522 }
1523
1524 PFN_vkVoidFunction
1525 tu_GetInstanceProcAddr(VkInstance _instance, const char *pName)
1526 {
1527 TU_FROM_HANDLE(tu_instance, instance, _instance);
1528
1529 return tu_lookup_entrypoint_checked(
1530 pName, instance ? instance->api_version : 0,
1531 instance ? &instance->enabled_extensions : NULL, NULL);
1532 }
1533
1534 /* The loader wants us to expose a second GetInstanceProcAddr function
1535 * to work around certain LD_PRELOAD issues seen in apps.
1536 */
1537 PUBLIC
1538 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1539 vk_icdGetInstanceProcAddr(VkInstance instance, const char *pName);
1540
1541 PUBLIC
1542 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1543 vk_icdGetInstanceProcAddr(VkInstance instance, const char *pName)
1544 {
1545 return tu_GetInstanceProcAddr(instance, pName);
1546 }
1547
1548 PFN_vkVoidFunction
1549 tu_GetDeviceProcAddr(VkDevice _device, const char *pName)
1550 {
1551 TU_FROM_HANDLE(tu_device, device, _device);
1552
1553 return tu_lookup_entrypoint_checked(pName, device->instance->api_version,
1554 &device->instance->enabled_extensions,
1555 &device->enabled_extensions);
1556 }
1557
1558 static VkResult
1559 tu_alloc_memory(struct tu_device *device,
1560 const VkMemoryAllocateInfo *pAllocateInfo,
1561 const VkAllocationCallbacks *pAllocator,
1562 VkDeviceMemory *pMem)
1563 {
1564 struct tu_device_memory *mem;
1565 VkResult result;
1566
1567 assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
1568
1569 if (pAllocateInfo->allocationSize == 0) {
1570 /* Apparently, this is allowed */
1571 *pMem = VK_NULL_HANDLE;
1572 return VK_SUCCESS;
1573 }
1574
1575 mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
1576 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1577 if (mem == NULL)
1578 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1579
1580 const VkImportMemoryFdInfoKHR *fd_info =
1581 vk_find_struct_const(pAllocateInfo->pNext, IMPORT_MEMORY_FD_INFO_KHR);
1582 if (fd_info && !fd_info->handleType)
1583 fd_info = NULL;
1584
1585 if (fd_info) {
1586 assert(fd_info->handleType ==
1587 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT ||
1588 fd_info->handleType ==
1589 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
1590
1591 /*
1592 * TODO Importing the same fd twice gives us the same handle without
1593 * reference counting. We need to maintain a per-instance handle-to-bo
1594 * table and add reference count to tu_bo.
1595 */
1596 result = tu_bo_init_dmabuf(device, &mem->bo,
1597 pAllocateInfo->allocationSize, fd_info->fd);
1598 if (result == VK_SUCCESS) {
1599 /* take ownership and close the fd */
1600 close(fd_info->fd);
1601 }
1602 } else {
1603 result =
1604 tu_bo_init_new(device, &mem->bo, pAllocateInfo->allocationSize);
1605 }
1606
1607 if (result != VK_SUCCESS) {
1608 vk_free2(&device->alloc, pAllocator, mem);
1609 return result;
1610 }
1611
1612 mem->size = pAllocateInfo->allocationSize;
1613 mem->type_index = pAllocateInfo->memoryTypeIndex;
1614
1615 mem->map = NULL;
1616 mem->user_ptr = NULL;
1617
1618 *pMem = tu_device_memory_to_handle(mem);
1619
1620 return VK_SUCCESS;
1621 }
1622
1623 VkResult
1624 tu_AllocateMemory(VkDevice _device,
1625 const VkMemoryAllocateInfo *pAllocateInfo,
1626 const VkAllocationCallbacks *pAllocator,
1627 VkDeviceMemory *pMem)
1628 {
1629 TU_FROM_HANDLE(tu_device, device, _device);
1630 return tu_alloc_memory(device, pAllocateInfo, pAllocator, pMem);
1631 }
1632
1633 void
1634 tu_FreeMemory(VkDevice _device,
1635 VkDeviceMemory _mem,
1636 const VkAllocationCallbacks *pAllocator)
1637 {
1638 TU_FROM_HANDLE(tu_device, device, _device);
1639 TU_FROM_HANDLE(tu_device_memory, mem, _mem);
1640
1641 if (mem == NULL)
1642 return;
1643
1644 tu_bo_finish(device, &mem->bo);
1645 vk_free2(&device->alloc, pAllocator, mem);
1646 }
1647
1648 VkResult
1649 tu_MapMemory(VkDevice _device,
1650 VkDeviceMemory _memory,
1651 VkDeviceSize offset,
1652 VkDeviceSize size,
1653 VkMemoryMapFlags flags,
1654 void **ppData)
1655 {
1656 TU_FROM_HANDLE(tu_device, device, _device);
1657 TU_FROM_HANDLE(tu_device_memory, mem, _memory);
1658 VkResult result;
1659
1660 if (mem == NULL) {
1661 *ppData = NULL;
1662 return VK_SUCCESS;
1663 }
1664
1665 if (mem->user_ptr) {
1666 *ppData = mem->user_ptr;
1667 } else if (!mem->map) {
1668 result = tu_bo_map(device, &mem->bo);
1669 if (result != VK_SUCCESS)
1670 return result;
1671 *ppData = mem->map = mem->bo.map;
1672 } else
1673 *ppData = mem->map;
1674
1675 if (*ppData) {
1676 *ppData += offset;
1677 return VK_SUCCESS;
1678 }
1679
1680 return vk_error(device->instance, VK_ERROR_MEMORY_MAP_FAILED);
1681 }
1682
1683 void
1684 tu_UnmapMemory(VkDevice _device, VkDeviceMemory _memory)
1685 {
1686 /* I do not see any unmapping done by the freedreno Gallium driver. */
1687 }
1688
1689 VkResult
1690 tu_FlushMappedMemoryRanges(VkDevice _device,
1691 uint32_t memoryRangeCount,
1692 const VkMappedMemoryRange *pMemoryRanges)
1693 {
1694 return VK_SUCCESS;
1695 }
1696
1697 VkResult
1698 tu_InvalidateMappedMemoryRanges(VkDevice _device,
1699 uint32_t memoryRangeCount,
1700 const VkMappedMemoryRange *pMemoryRanges)
1701 {
1702 return VK_SUCCESS;
1703 }
1704
1705 void
1706 tu_GetBufferMemoryRequirements(VkDevice _device,
1707 VkBuffer _buffer,
1708 VkMemoryRequirements *pMemoryRequirements)
1709 {
1710 TU_FROM_HANDLE(tu_buffer, buffer, _buffer);
1711
1712 pMemoryRequirements->memoryTypeBits = 1;
1713 pMemoryRequirements->alignment = 64;
1714 pMemoryRequirements->size =
1715 align64(buffer->size, pMemoryRequirements->alignment);
1716 }
1717
1718 void
1719 tu_GetBufferMemoryRequirements2(
1720 VkDevice device,
1721 const VkBufferMemoryRequirementsInfo2 *pInfo,
1722 VkMemoryRequirements2 *pMemoryRequirements)
1723 {
1724 tu_GetBufferMemoryRequirements(device, pInfo->buffer,
1725 &pMemoryRequirements->memoryRequirements);
1726 }
1727
1728 void
1729 tu_GetImageMemoryRequirements(VkDevice _device,
1730 VkImage _image,
1731 VkMemoryRequirements *pMemoryRequirements)
1732 {
1733 TU_FROM_HANDLE(tu_image, image, _image);
1734
1735 pMemoryRequirements->memoryTypeBits = 1;
1736 pMemoryRequirements->size = image->layout.size;
1737 pMemoryRequirements->alignment = image->layout.base_align;
1738 }
1739
1740 void
1741 tu_GetImageMemoryRequirements2(VkDevice device,
1742 const VkImageMemoryRequirementsInfo2 *pInfo,
1743 VkMemoryRequirements2 *pMemoryRequirements)
1744 {
1745 tu_GetImageMemoryRequirements(device, pInfo->image,
1746 &pMemoryRequirements->memoryRequirements);
1747 }
1748
1749 void
1750 tu_GetImageSparseMemoryRequirements(
1751 VkDevice device,
1752 VkImage image,
1753 uint32_t *pSparseMemoryRequirementCount,
1754 VkSparseImageMemoryRequirements *pSparseMemoryRequirements)
1755 {
1756 tu_stub();
1757 }
1758
1759 void
1760 tu_GetImageSparseMemoryRequirements2(
1761 VkDevice device,
1762 const VkImageSparseMemoryRequirementsInfo2 *pInfo,
1763 uint32_t *pSparseMemoryRequirementCount,
1764 VkSparseImageMemoryRequirements2 *pSparseMemoryRequirements)
1765 {
1766 tu_stub();
1767 }
1768
1769 void
1770 tu_GetDeviceMemoryCommitment(VkDevice device,
1771 VkDeviceMemory memory,
1772 VkDeviceSize *pCommittedMemoryInBytes)
1773 {
1774 *pCommittedMemoryInBytes = 0;
1775 }
1776
1777 VkResult
1778 tu_BindBufferMemory2(VkDevice device,
1779 uint32_t bindInfoCount,
1780 const VkBindBufferMemoryInfo *pBindInfos)
1781 {
1782 for (uint32_t i = 0; i < bindInfoCount; ++i) {
1783 TU_FROM_HANDLE(tu_device_memory, mem, pBindInfos[i].memory);
1784 TU_FROM_HANDLE(tu_buffer, buffer, pBindInfos[i].buffer);
1785
1786 if (mem) {
1787 buffer->bo = &mem->bo;
1788 buffer->bo_offset = pBindInfos[i].memoryOffset;
1789 } else {
1790 buffer->bo = NULL;
1791 }
1792 }
1793 return VK_SUCCESS;
1794 }
1795
1796 VkResult
1797 tu_BindBufferMemory(VkDevice device,
1798 VkBuffer buffer,
1799 VkDeviceMemory memory,
1800 VkDeviceSize memoryOffset)
1801 {
1802 const VkBindBufferMemoryInfo info = {
1803 .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO,
1804 .buffer = buffer,
1805 .memory = memory,
1806 .memoryOffset = memoryOffset
1807 };
1808
1809 return tu_BindBufferMemory2(device, 1, &info);
1810 }
1811
1812 VkResult
1813 tu_BindImageMemory2(VkDevice device,
1814 uint32_t bindInfoCount,
1815 const VkBindImageMemoryInfo *pBindInfos)
1816 {
1817 for (uint32_t i = 0; i < bindInfoCount; ++i) {
1818 TU_FROM_HANDLE(tu_image, image, pBindInfos[i].image);
1819 TU_FROM_HANDLE(tu_device_memory, mem, pBindInfos[i].memory);
1820
1821 if (mem) {
1822 image->bo = &mem->bo;
1823 image->bo_offset = pBindInfos[i].memoryOffset;
1824 } else {
1825 image->bo = NULL;
1826 image->bo_offset = 0;
1827 }
1828 }
1829
1830 return VK_SUCCESS;
1831 }
1832
1833 VkResult
1834 tu_BindImageMemory(VkDevice device,
1835 VkImage image,
1836 VkDeviceMemory memory,
1837 VkDeviceSize memoryOffset)
1838 {
1839 const VkBindImageMemoryInfo info = {
1840 .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO,
1841 .image = image,
1842 .memory = memory,
1843 .memoryOffset = memoryOffset
1844 };
1845
1846 return tu_BindImageMemory2(device, 1, &info);
1847 }
1848
1849 VkResult
1850 tu_QueueBindSparse(VkQueue _queue,
1851 uint32_t bindInfoCount,
1852 const VkBindSparseInfo *pBindInfo,
1853 VkFence _fence)
1854 {
1855 return VK_SUCCESS;
1856 }
1857
1858 // Queue semaphore functions
1859
1860 VkResult
1861 tu_CreateSemaphore(VkDevice _device,
1862 const VkSemaphoreCreateInfo *pCreateInfo,
1863 const VkAllocationCallbacks *pAllocator,
1864 VkSemaphore *pSemaphore)
1865 {
1866 TU_FROM_HANDLE(tu_device, device, _device);
1867
1868 struct tu_semaphore *sem =
1869 vk_alloc2(&device->alloc, pAllocator, sizeof(*sem), 8,
1870 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1871 if (!sem)
1872 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1873
1874 *pSemaphore = tu_semaphore_to_handle(sem);
1875 return VK_SUCCESS;
1876 }
1877
1878 void
1879 tu_DestroySemaphore(VkDevice _device,
1880 VkSemaphore _semaphore,
1881 const VkAllocationCallbacks *pAllocator)
1882 {
1883 TU_FROM_HANDLE(tu_device, device, _device);
1884 TU_FROM_HANDLE(tu_semaphore, sem, _semaphore);
1885 if (!_semaphore)
1886 return;
1887
1888 vk_free2(&device->alloc, pAllocator, sem);
1889 }
1890
1891 VkResult
1892 tu_CreateEvent(VkDevice _device,
1893 const VkEventCreateInfo *pCreateInfo,
1894 const VkAllocationCallbacks *pAllocator,
1895 VkEvent *pEvent)
1896 {
1897 TU_FROM_HANDLE(tu_device, device, _device);
1898 struct tu_event *event =
1899 vk_alloc2(&device->alloc, pAllocator, sizeof(*event), 8,
1900 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1901
1902 if (!event)
1903 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1904
1905 VkResult result = tu_bo_init_new(device, &event->bo, 0x1000);
1906 if (result != VK_SUCCESS)
1907 goto fail_alloc;
1908
1909 result = tu_bo_map(device, &event->bo);
1910 if (result != VK_SUCCESS)
1911 goto fail_map;
1912
1913 *pEvent = tu_event_to_handle(event);
1914
1915 return VK_SUCCESS;
1916
1917 fail_map:
1918 tu_bo_finish(device, &event->bo);
1919 fail_alloc:
1920 vk_free2(&device->alloc, pAllocator, event);
1921 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1922 }
1923
1924 void
1925 tu_DestroyEvent(VkDevice _device,
1926 VkEvent _event,
1927 const VkAllocationCallbacks *pAllocator)
1928 {
1929 TU_FROM_HANDLE(tu_device, device, _device);
1930 TU_FROM_HANDLE(tu_event, event, _event);
1931
1932 if (!event)
1933 return;
1934
1935 tu_bo_finish(device, &event->bo);
1936 vk_free2(&device->alloc, pAllocator, event);
1937 }
1938
1939 VkResult
1940 tu_GetEventStatus(VkDevice _device, VkEvent _event)
1941 {
1942 TU_FROM_HANDLE(tu_event, event, _event);
1943
1944 if (*(uint64_t*) event->bo.map == 1)
1945 return VK_EVENT_SET;
1946 return VK_EVENT_RESET;
1947 }
1948
1949 VkResult
1950 tu_SetEvent(VkDevice _device, VkEvent _event)
1951 {
1952 TU_FROM_HANDLE(tu_event, event, _event);
1953 *(uint64_t*) event->bo.map = 1;
1954
1955 return VK_SUCCESS;
1956 }
1957
1958 VkResult
1959 tu_ResetEvent(VkDevice _device, VkEvent _event)
1960 {
1961 TU_FROM_HANDLE(tu_event, event, _event);
1962 *(uint64_t*) event->bo.map = 0;
1963
1964 return VK_SUCCESS;
1965 }
1966
1967 VkResult
1968 tu_CreateBuffer(VkDevice _device,
1969 const VkBufferCreateInfo *pCreateInfo,
1970 const VkAllocationCallbacks *pAllocator,
1971 VkBuffer *pBuffer)
1972 {
1973 TU_FROM_HANDLE(tu_device, device, _device);
1974 struct tu_buffer *buffer;
1975
1976 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO);
1977
1978 buffer = vk_alloc2(&device->alloc, pAllocator, sizeof(*buffer), 8,
1979 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1980 if (buffer == NULL)
1981 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1982
1983 buffer->size = pCreateInfo->size;
1984 buffer->usage = pCreateInfo->usage;
1985 buffer->flags = pCreateInfo->flags;
1986
1987 *pBuffer = tu_buffer_to_handle(buffer);
1988
1989 return VK_SUCCESS;
1990 }
1991
1992 void
1993 tu_DestroyBuffer(VkDevice _device,
1994 VkBuffer _buffer,
1995 const VkAllocationCallbacks *pAllocator)
1996 {
1997 TU_FROM_HANDLE(tu_device, device, _device);
1998 TU_FROM_HANDLE(tu_buffer, buffer, _buffer);
1999
2000 if (!buffer)
2001 return;
2002
2003 vk_free2(&device->alloc, pAllocator, buffer);
2004 }
2005
2006 static uint32_t
2007 tu_surface_max_layer_count(struct tu_image_view *iview)
2008 {
2009 return iview->type == VK_IMAGE_VIEW_TYPE_3D
2010 ? iview->extent.depth
2011 : (iview->base_layer + iview->layer_count);
2012 }
2013
2014 VkResult
2015 tu_CreateFramebuffer(VkDevice _device,
2016 const VkFramebufferCreateInfo *pCreateInfo,
2017 const VkAllocationCallbacks *pAllocator,
2018 VkFramebuffer *pFramebuffer)
2019 {
2020 TU_FROM_HANDLE(tu_device, device, _device);
2021 struct tu_framebuffer *framebuffer;
2022
2023 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
2024
2025 size_t size = sizeof(*framebuffer) + sizeof(struct tu_attachment_info) *
2026 pCreateInfo->attachmentCount;
2027 framebuffer = vk_alloc2(&device->alloc, pAllocator, size, 8,
2028 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
2029 if (framebuffer == NULL)
2030 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
2031
2032 framebuffer->attachment_count = pCreateInfo->attachmentCount;
2033 framebuffer->width = pCreateInfo->width;
2034 framebuffer->height = pCreateInfo->height;
2035 framebuffer->layers = pCreateInfo->layers;
2036 for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
2037 VkImageView _iview = pCreateInfo->pAttachments[i];
2038 struct tu_image_view *iview = tu_image_view_from_handle(_iview);
2039 framebuffer->attachments[i].attachment = iview;
2040
2041 framebuffer->width = MIN2(framebuffer->width, iview->extent.width);
2042 framebuffer->height = MIN2(framebuffer->height, iview->extent.height);
2043 framebuffer->layers =
2044 MIN2(framebuffer->layers, tu_surface_max_layer_count(iview));
2045 }
2046
2047 *pFramebuffer = tu_framebuffer_to_handle(framebuffer);
2048 return VK_SUCCESS;
2049 }
2050
2051 void
2052 tu_DestroyFramebuffer(VkDevice _device,
2053 VkFramebuffer _fb,
2054 const VkAllocationCallbacks *pAllocator)
2055 {
2056 TU_FROM_HANDLE(tu_device, device, _device);
2057 TU_FROM_HANDLE(tu_framebuffer, fb, _fb);
2058
2059 if (!fb)
2060 return;
2061 vk_free2(&device->alloc, pAllocator, fb);
2062 }
2063
2064 static enum a6xx_tex_clamp
2065 tu6_tex_wrap(VkSamplerAddressMode address_mode)
2066 {
2067 switch (address_mode) {
2068 case VK_SAMPLER_ADDRESS_MODE_REPEAT:
2069 return A6XX_TEX_REPEAT;
2070 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT:
2071 return A6XX_TEX_MIRROR_REPEAT;
2072 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE:
2073 return A6XX_TEX_CLAMP_TO_EDGE;
2074 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER:
2075 return A6XX_TEX_CLAMP_TO_BORDER;
2076 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE:
2077 /* only works for PoT.. need to emulate otherwise! */
2078 return A6XX_TEX_MIRROR_CLAMP;
2079 default:
2080 unreachable("illegal tex wrap mode");
2081 break;
2082 }
2083 }
2084
2085 static enum a6xx_tex_filter
2086 tu6_tex_filter(VkFilter filter, unsigned aniso)
2087 {
2088 switch (filter) {
2089 case VK_FILTER_NEAREST:
2090 return A6XX_TEX_NEAREST;
2091 case VK_FILTER_LINEAR:
2092 return aniso ? A6XX_TEX_ANISO : A6XX_TEX_LINEAR;
2093 case VK_FILTER_CUBIC_IMG:
2094 default:
2095 unreachable("illegal texture filter");
2096 break;
2097 }
2098 }
2099
2100 static inline enum adreno_compare_func
2101 tu6_compare_func(VkCompareOp op)
2102 {
2103 return (enum adreno_compare_func) op;
2104 }
2105
2106 static void
2107 tu_init_sampler(struct tu_device *device,
2108 struct tu_sampler *sampler,
2109 const VkSamplerCreateInfo *pCreateInfo)
2110 {
2111 unsigned aniso = pCreateInfo->anisotropyEnable ?
2112 util_last_bit(MIN2((uint32_t)pCreateInfo->maxAnisotropy >> 1, 8)) : 0;
2113 bool miplinear = (pCreateInfo->mipmapMode == VK_SAMPLER_MIPMAP_MODE_LINEAR);
2114
2115 sampler->descriptor[0] =
2116 COND(miplinear, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR) |
2117 A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo->magFilter, aniso)) |
2118 A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo->minFilter, aniso)) |
2119 A6XX_TEX_SAMP_0_ANISO(aniso) |
2120 A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo->addressModeU)) |
2121 A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo->addressModeV)) |
2122 A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo->addressModeW)) |
2123 A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo->mipLodBias);
2124 sampler->descriptor[1] =
2125 /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
2126 COND(pCreateInfo->unnormalizedCoordinates, A6XX_TEX_SAMP_1_UNNORM_COORDS) |
2127 A6XX_TEX_SAMP_1_MIN_LOD(pCreateInfo->minLod) |
2128 A6XX_TEX_SAMP_1_MAX_LOD(pCreateInfo->maxLod) |
2129 COND(pCreateInfo->compareEnable,
2130 A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo->compareOp)));
2131 /* This is an offset into the border_color BO, which we fill with all the
2132 * possible Vulkan border colors in the correct order, so we can just use
2133 * the Vulkan enum with no translation necessary.
2134 */
2135 sampler->descriptor[2] =
2136 A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo->borderColor *
2137 sizeof(struct bcolor_entry));
2138 sampler->descriptor[3] = 0;
2139
2140 /* TODO:
2141 * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
2142 */
2143 }
2144
2145 VkResult
2146 tu_CreateSampler(VkDevice _device,
2147 const VkSamplerCreateInfo *pCreateInfo,
2148 const VkAllocationCallbacks *pAllocator,
2149 VkSampler *pSampler)
2150 {
2151 TU_FROM_HANDLE(tu_device, device, _device);
2152 struct tu_sampler *sampler;
2153
2154 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
2155
2156 sampler = vk_alloc2(&device->alloc, pAllocator, sizeof(*sampler), 8,
2157 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
2158 if (!sampler)
2159 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
2160
2161 tu_init_sampler(device, sampler, pCreateInfo);
2162 *pSampler = tu_sampler_to_handle(sampler);
2163
2164 return VK_SUCCESS;
2165 }
2166
2167 void
2168 tu_DestroySampler(VkDevice _device,
2169 VkSampler _sampler,
2170 const VkAllocationCallbacks *pAllocator)
2171 {
2172 TU_FROM_HANDLE(tu_device, device, _device);
2173 TU_FROM_HANDLE(tu_sampler, sampler, _sampler);
2174
2175 if (!sampler)
2176 return;
2177 vk_free2(&device->alloc, pAllocator, sampler);
2178 }
2179
2180 /* vk_icd.h does not declare this function, so we declare it here to
2181 * suppress Wmissing-prototypes.
2182 */
2183 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2184 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion);
2185
2186 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2187 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion)
2188 {
2189 /* For the full details on loader interface versioning, see
2190 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2191 * What follows is a condensed summary, to help you navigate the large and
2192 * confusing official doc.
2193 *
2194 * - Loader interface v0 is incompatible with later versions. We don't
2195 * support it.
2196 *
2197 * - In loader interface v1:
2198 * - The first ICD entrypoint called by the loader is
2199 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2200 * entrypoint.
2201 * - The ICD must statically expose no other Vulkan symbol unless it
2202 * is linked with -Bsymbolic.
2203 * - Each dispatchable Vulkan handle created by the ICD must be
2204 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2205 * ICD must initialize VK_LOADER_DATA.loadMagic to
2206 * ICD_LOADER_MAGIC.
2207 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2208 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2209 * such loader-managed surfaces.
2210 *
2211 * - Loader interface v2 differs from v1 in:
2212 * - The first ICD entrypoint called by the loader is
2213 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2214 * statically expose this entrypoint.
2215 *
2216 * - Loader interface v3 differs from v2 in:
2217 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2218 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2219 * because the loader no longer does so.
2220 */
2221 *pSupportedVersion = MIN2(*pSupportedVersion, 3u);
2222 return VK_SUCCESS;
2223 }
2224
2225 VkResult
2226 tu_GetMemoryFdKHR(VkDevice _device,
2227 const VkMemoryGetFdInfoKHR *pGetFdInfo,
2228 int *pFd)
2229 {
2230 TU_FROM_HANDLE(tu_device, device, _device);
2231 TU_FROM_HANDLE(tu_device_memory, memory, pGetFdInfo->memory);
2232
2233 assert(pGetFdInfo->sType == VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR);
2234
2235 /* At the moment, we support only the below handle types. */
2236 assert(pGetFdInfo->handleType ==
2237 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT ||
2238 pGetFdInfo->handleType ==
2239 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
2240
2241 int prime_fd = tu_bo_export_dmabuf(device, &memory->bo);
2242 if (prime_fd < 0)
2243 return vk_error(device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
2244
2245 *pFd = prime_fd;
2246 return VK_SUCCESS;
2247 }
2248
2249 VkResult
2250 tu_GetMemoryFdPropertiesKHR(VkDevice _device,
2251 VkExternalMemoryHandleTypeFlagBits handleType,
2252 int fd,
2253 VkMemoryFdPropertiesKHR *pMemoryFdProperties)
2254 {
2255 assert(handleType == VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
2256 pMemoryFdProperties->memoryTypeBits = 1;
2257 return VK_SUCCESS;
2258 }
2259
2260 void
2261 tu_GetPhysicalDeviceExternalSemaphoreProperties(
2262 VkPhysicalDevice physicalDevice,
2263 const VkPhysicalDeviceExternalSemaphoreInfo *pExternalSemaphoreInfo,
2264 VkExternalSemaphoreProperties *pExternalSemaphoreProperties)
2265 {
2266 pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0;
2267 pExternalSemaphoreProperties->compatibleHandleTypes = 0;
2268 pExternalSemaphoreProperties->externalSemaphoreFeatures = 0;
2269 }
2270
2271 void
2272 tu_GetPhysicalDeviceExternalFenceProperties(
2273 VkPhysicalDevice physicalDevice,
2274 const VkPhysicalDeviceExternalFenceInfo *pExternalFenceInfo,
2275 VkExternalFenceProperties *pExternalFenceProperties)
2276 {
2277 pExternalFenceProperties->exportFromImportedHandleTypes = 0;
2278 pExternalFenceProperties->compatibleHandleTypes = 0;
2279 pExternalFenceProperties->externalFenceFeatures = 0;
2280 }
2281
2282 VkResult
2283 tu_CreateDebugReportCallbackEXT(
2284 VkInstance _instance,
2285 const VkDebugReportCallbackCreateInfoEXT *pCreateInfo,
2286 const VkAllocationCallbacks *pAllocator,
2287 VkDebugReportCallbackEXT *pCallback)
2288 {
2289 TU_FROM_HANDLE(tu_instance, instance, _instance);
2290 return vk_create_debug_report_callback(&instance->debug_report_callbacks,
2291 pCreateInfo, pAllocator,
2292 &instance->alloc, pCallback);
2293 }
2294
2295 void
2296 tu_DestroyDebugReportCallbackEXT(VkInstance _instance,
2297 VkDebugReportCallbackEXT _callback,
2298 const VkAllocationCallbacks *pAllocator)
2299 {
2300 TU_FROM_HANDLE(tu_instance, instance, _instance);
2301 vk_destroy_debug_report_callback(&instance->debug_report_callbacks,
2302 _callback, pAllocator, &instance->alloc);
2303 }
2304
2305 void
2306 tu_DebugReportMessageEXT(VkInstance _instance,
2307 VkDebugReportFlagsEXT flags,
2308 VkDebugReportObjectTypeEXT objectType,
2309 uint64_t object,
2310 size_t location,
2311 int32_t messageCode,
2312 const char *pLayerPrefix,
2313 const char *pMessage)
2314 {
2315 TU_FROM_HANDLE(tu_instance, instance, _instance);
2316 vk_debug_report(&instance->debug_report_callbacks, flags, objectType,
2317 object, location, messageCode, pLayerPrefix, pMessage);
2318 }
2319
2320 void
2321 tu_GetDeviceGroupPeerMemoryFeatures(
2322 VkDevice device,
2323 uint32_t heapIndex,
2324 uint32_t localDeviceIndex,
2325 uint32_t remoteDeviceIndex,
2326 VkPeerMemoryFeatureFlags *pPeerMemoryFeatures)
2327 {
2328 assert(localDeviceIndex == remoteDeviceIndex);
2329
2330 *pPeerMemoryFeatures = VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT |
2331 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT |
2332 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
2333 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
2334 }