2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
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:
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
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.
28 #include "tu_private.h"
35 #include <sys/sysinfo.h>
39 #include "compiler/glsl_types.h"
40 #include "util/debug.h"
41 #include "util/disk_cache.h"
42 #include "util/u_atomic.h"
43 #include "vk_format.h"
46 #include "drm-uapi/msm_drm.h"
48 /* for fd_get_driver/device_uuid() */
49 #include "freedreno/common/freedreno_uuid.h"
52 tu_semaphore_remove_temp(struct tu_device
*device
,
53 struct tu_semaphore
*sem
);
56 tu_device_get_cache_uuid(uint16_t family
, void *uuid
)
58 uint32_t mesa_timestamp
;
60 memset(uuid
, 0, VK_UUID_SIZE
);
61 if (!disk_cache_get_function_timestamp(tu_device_get_cache_uuid
,
65 memcpy(uuid
, &mesa_timestamp
, 4);
66 memcpy((char *) uuid
+ 4, &f
, 2);
67 snprintf((char *) uuid
+ 6, VK_UUID_SIZE
- 10, "tu");
72 tu_bo_init(struct tu_device
*dev
,
77 uint64_t iova
= tu_gem_info_iova(dev
, gem_handle
);
79 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
81 *bo
= (struct tu_bo
) {
82 .gem_handle
= gem_handle
,
91 tu_bo_init_new(struct tu_device
*dev
, struct tu_bo
*bo
, uint64_t size
)
93 /* TODO: Choose better flags. As of 2018-11-12, freedreno/drm/msm_bo.c
94 * always sets `flags = MSM_BO_WC`, and we copy that behavior here.
96 uint32_t gem_handle
= tu_gem_new(dev
, size
, MSM_BO_WC
);
98 return vk_error(dev
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
100 VkResult result
= tu_bo_init(dev
, bo
, gem_handle
, size
);
101 if (result
!= VK_SUCCESS
) {
102 tu_gem_close(dev
, gem_handle
);
103 return vk_error(dev
->instance
, result
);
110 tu_bo_init_dmabuf(struct tu_device
*dev
,
115 uint32_t gem_handle
= tu_gem_import_dmabuf(dev
, fd
, size
);
117 return vk_error(dev
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE
);
119 VkResult result
= tu_bo_init(dev
, bo
, gem_handle
, size
);
120 if (result
!= VK_SUCCESS
) {
121 tu_gem_close(dev
, gem_handle
);
122 return vk_error(dev
->instance
, result
);
129 tu_bo_export_dmabuf(struct tu_device
*dev
, struct tu_bo
*bo
)
131 return tu_gem_export_dmabuf(dev
, bo
->gem_handle
);
135 tu_bo_map(struct tu_device
*dev
, struct tu_bo
*bo
)
140 uint64_t offset
= tu_gem_info_offset(dev
, bo
->gem_handle
);
142 return vk_error(dev
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
144 /* TODO: Should we use the wrapper os_mmap() like Freedreno does? */
145 void *map
= mmap(0, bo
->size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
146 dev
->physical_device
->local_fd
, offset
);
147 if (map
== MAP_FAILED
)
148 return vk_error(dev
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
155 tu_bo_finish(struct tu_device
*dev
, struct tu_bo
*bo
)
157 assert(bo
->gem_handle
);
160 munmap(bo
->map
, bo
->size
);
162 tu_gem_close(dev
, bo
->gem_handle
);
166 tu_physical_device_init(struct tu_physical_device
*device
,
167 struct tu_instance
*instance
,
168 drmDevicePtr drm_device
)
170 const char *path
= drm_device
->nodes
[DRM_NODE_RENDER
];
171 VkResult result
= VK_SUCCESS
;
172 drmVersionPtr version
;
176 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
178 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
179 "failed to open device %s", path
);
182 /* Version 1.3 added MSM_INFO_IOVA. */
183 const int min_version_major
= 1;
184 const int min_version_minor
= 3;
186 version
= drmGetVersion(fd
);
189 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
190 "failed to query kernel driver version for device %s",
194 if (strcmp(version
->name
, "msm")) {
195 drmFreeVersion(version
);
197 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
198 "device %s does not use the msm kernel driver", path
);
201 if (version
->version_major
!= min_version_major
||
202 version
->version_minor
< min_version_minor
) {
203 result
= vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
204 "kernel driver for device %s has version %d.%d, "
205 "but Vulkan requires version >= %d.%d",
206 path
, version
->version_major
, version
->version_minor
,
207 min_version_major
, min_version_minor
);
208 drmFreeVersion(version
);
213 device
->msm_major_version
= version
->version_major
;
214 device
->msm_minor_version
= version
->version_minor
;
216 drmFreeVersion(version
);
218 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
219 tu_logi("Found compatible device '%s'.", path
);
221 vk_object_base_init(NULL
, &device
->base
, VK_OBJECT_TYPE_PHYSICAL_DEVICE
);
222 device
->instance
= instance
;
223 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
224 strncpy(device
->path
, path
, ARRAY_SIZE(device
->path
));
226 if (instance
->enabled_extensions
.KHR_display
) {
228 open(drm_device
->nodes
[DRM_NODE_PRIMARY
], O_RDWR
| O_CLOEXEC
);
229 if (master_fd
>= 0) {
230 /* TODO: free master_fd is accel is not working? */
234 device
->master_fd
= master_fd
;
235 device
->local_fd
= fd
;
237 if (tu_drm_get_gpu_id(device
, &device
->gpu_id
)) {
238 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
239 tu_logi("Could not query the GPU ID");
240 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
241 "could not get GPU ID");
245 if (tu_drm_get_gmem_size(device
, &device
->gmem_size
)) {
246 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
247 tu_logi("Could not query the GMEM size");
248 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
249 "could not get GMEM size");
253 if (tu_drm_get_gmem_base(device
, &device
->gmem_base
)) {
254 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
255 tu_logi("Could not query the GMEM size");
256 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
257 "could not get GMEM size");
261 memset(device
->name
, 0, sizeof(device
->name
));
262 sprintf(device
->name
, "FD%d", device
->gpu_id
);
264 device
->limited_z24s8
= (device
->gpu_id
== 630);
266 switch (device
->gpu_id
) {
268 device
->ccu_offset_gmem
= 0x7c000; /* 0x7e000 in some cases? */
269 device
->ccu_offset_bypass
= 0x10000;
270 device
->tile_align_w
= 32;
271 device
->magic
.PC_UNKNOWN_9805
= 0x0;
272 device
->magic
.SP_UNKNOWN_A0F8
= 0x0;
276 device
->ccu_offset_gmem
= 0xf8000;
277 device
->ccu_offset_bypass
= 0x20000;
278 device
->tile_align_w
= 32;
279 device
->magic
.PC_UNKNOWN_9805
= 0x1;
280 device
->magic
.SP_UNKNOWN_A0F8
= 0x1;
283 device
->ccu_offset_gmem
= 0x114000;
284 device
->ccu_offset_bypass
= 0x30000;
285 device
->tile_align_w
= 96;
286 device
->magic
.PC_UNKNOWN_9805
= 0x2;
287 device
->magic
.SP_UNKNOWN_A0F8
= 0x2;
290 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
291 "device %s is unsupported", device
->name
);
294 if (tu_device_get_cache_uuid(device
->gpu_id
, device
->cache_uuid
)) {
295 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
296 "cannot generate UUID");
300 /* The gpu id is already embedded in the uuid so we just pass "tu"
301 * when creating the cache.
303 char buf
[VK_UUID_SIZE
* 2 + 1];
304 disk_cache_format_hex_id(buf
, device
->cache_uuid
, VK_UUID_SIZE
* 2);
305 device
->disk_cache
= disk_cache_create(device
->name
, buf
, 0);
307 fprintf(stderr
, "WARNING: tu is not a conformant vulkan implementation, "
308 "testing use only.\n");
310 fd_get_driver_uuid(device
->driver_uuid
);
311 fd_get_device_uuid(device
->device_uuid
, device
->gpu_id
);
313 tu_physical_device_get_supported_extensions(device
, &device
->supported_extensions
);
315 if (result
!= VK_SUCCESS
) {
316 vk_error(instance
, result
);
320 result
= tu_wsi_init(device
);
321 if (result
!= VK_SUCCESS
) {
322 vk_error(instance
, result
);
336 tu_physical_device_finish(struct tu_physical_device
*device
)
338 tu_wsi_finish(device
);
340 disk_cache_destroy(device
->disk_cache
);
341 close(device
->local_fd
);
342 if (device
->master_fd
!= -1)
343 close(device
->master_fd
);
345 vk_object_base_finish(&device
->base
);
348 static VKAPI_ATTR
void *
349 default_alloc_func(void *pUserData
,
352 VkSystemAllocationScope allocationScope
)
357 static VKAPI_ATTR
void *
358 default_realloc_func(void *pUserData
,
362 VkSystemAllocationScope allocationScope
)
364 return realloc(pOriginal
, size
);
367 static VKAPI_ATTR
void
368 default_free_func(void *pUserData
, void *pMemory
)
373 static const VkAllocationCallbacks default_alloc
= {
375 .pfnAllocation
= default_alloc_func
,
376 .pfnReallocation
= default_realloc_func
,
377 .pfnFree
= default_free_func
,
380 static const struct debug_control tu_debug_options
[] = {
381 { "startup", TU_DEBUG_STARTUP
},
382 { "nir", TU_DEBUG_NIR
},
383 { "ir3", TU_DEBUG_IR3
},
384 { "nobin", TU_DEBUG_NOBIN
},
385 { "sysmem", TU_DEBUG_SYSMEM
},
386 { "forcebin", TU_DEBUG_FORCEBIN
},
387 { "noubwc", TU_DEBUG_NOUBWC
},
392 tu_get_debug_option_name(int id
)
394 assert(id
< ARRAY_SIZE(tu_debug_options
) - 1);
395 return tu_debug_options
[id
].string
;
399 tu_get_instance_extension_index(const char *name
)
401 for (unsigned i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; ++i
) {
402 if (strcmp(name
, tu_instance_extensions
[i
].extensionName
) == 0)
409 tu_CreateInstance(const VkInstanceCreateInfo
*pCreateInfo
,
410 const VkAllocationCallbacks
*pAllocator
,
411 VkInstance
*pInstance
)
413 struct tu_instance
*instance
;
416 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
418 uint32_t client_version
;
419 if (pCreateInfo
->pApplicationInfo
&&
420 pCreateInfo
->pApplicationInfo
->apiVersion
!= 0) {
421 client_version
= pCreateInfo
->pApplicationInfo
->apiVersion
;
423 tu_EnumerateInstanceVersion(&client_version
);
426 instance
= vk_zalloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
427 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
430 return vk_error(NULL
, VK_ERROR_OUT_OF_HOST_MEMORY
);
432 vk_object_base_init(NULL
, &instance
->base
, VK_OBJECT_TYPE_INSTANCE
);
435 instance
->alloc
= *pAllocator
;
437 instance
->alloc
= default_alloc
;
439 instance
->api_version
= client_version
;
440 instance
->physical_device_count
= -1;
442 instance
->debug_flags
=
443 parse_debug_string(getenv("TU_DEBUG"), tu_debug_options
);
445 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
446 tu_logi("Created an instance");
448 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
449 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
450 int index
= tu_get_instance_extension_index(ext_name
);
452 if (index
< 0 || !tu_instance_extensions_supported
.extensions
[index
]) {
453 vk_object_base_finish(&instance
->base
);
454 vk_free2(&default_alloc
, pAllocator
, instance
);
455 return vk_error(instance
, VK_ERROR_EXTENSION_NOT_PRESENT
);
458 instance
->enabled_extensions
.extensions
[index
] = true;
461 result
= vk_debug_report_instance_init(&instance
->debug_report_callbacks
);
462 if (result
!= VK_SUCCESS
) {
463 vk_object_base_finish(&instance
->base
);
464 vk_free2(&default_alloc
, pAllocator
, instance
);
465 return vk_error(instance
, result
);
468 glsl_type_singleton_init_or_ref();
470 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
472 *pInstance
= tu_instance_to_handle(instance
);
478 tu_DestroyInstance(VkInstance _instance
,
479 const VkAllocationCallbacks
*pAllocator
)
481 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
486 for (int i
= 0; i
< instance
->physical_device_count
; ++i
) {
487 tu_physical_device_finish(instance
->physical_devices
+ i
);
490 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
492 glsl_type_singleton_decref();
494 vk_debug_report_instance_destroy(&instance
->debug_report_callbacks
);
496 vk_object_base_finish(&instance
->base
);
497 vk_free(&instance
->alloc
, instance
);
501 tu_enumerate_devices(struct tu_instance
*instance
)
503 /* TODO: Check for more devices ? */
504 drmDevicePtr devices
[8];
505 VkResult result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
508 instance
->physical_device_count
= 0;
510 max_devices
= drmGetDevices2(0, devices
, ARRAY_SIZE(devices
));
512 if (instance
->debug_flags
& TU_DEBUG_STARTUP
) {
514 tu_logi("drmGetDevices2 returned error: %s\n", strerror(max_devices
));
516 tu_logi("Found %d drm nodes", max_devices
);
520 return vk_error(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
);
522 for (unsigned i
= 0; i
< (unsigned) max_devices
; i
++) {
523 if (devices
[i
]->available_nodes
& 1 << DRM_NODE_RENDER
&&
524 devices
[i
]->bustype
== DRM_BUS_PLATFORM
) {
526 result
= tu_physical_device_init(
527 instance
->physical_devices
+ instance
->physical_device_count
,
528 instance
, devices
[i
]);
529 if (result
== VK_SUCCESS
)
530 ++instance
->physical_device_count
;
531 else if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
535 drmFreeDevices(devices
, max_devices
);
541 tu_EnumeratePhysicalDevices(VkInstance _instance
,
542 uint32_t *pPhysicalDeviceCount
,
543 VkPhysicalDevice
*pPhysicalDevices
)
545 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
546 VK_OUTARRAY_MAKE(out
, pPhysicalDevices
, pPhysicalDeviceCount
);
550 if (instance
->physical_device_count
< 0) {
551 result
= tu_enumerate_devices(instance
);
552 if (result
!= VK_SUCCESS
&& result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
556 for (uint32_t i
= 0; i
< instance
->physical_device_count
; ++i
) {
557 vk_outarray_append(&out
, p
)
559 *p
= tu_physical_device_to_handle(instance
->physical_devices
+ i
);
563 return vk_outarray_status(&out
);
567 tu_EnumeratePhysicalDeviceGroups(
568 VkInstance _instance
,
569 uint32_t *pPhysicalDeviceGroupCount
,
570 VkPhysicalDeviceGroupProperties
*pPhysicalDeviceGroupProperties
)
572 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
573 VK_OUTARRAY_MAKE(out
, pPhysicalDeviceGroupProperties
,
574 pPhysicalDeviceGroupCount
);
577 if (instance
->physical_device_count
< 0) {
578 result
= tu_enumerate_devices(instance
);
579 if (result
!= VK_SUCCESS
&& result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
583 for (uint32_t i
= 0; i
< instance
->physical_device_count
; ++i
) {
584 vk_outarray_append(&out
, p
)
586 p
->physicalDeviceCount
= 1;
587 p
->physicalDevices
[0] =
588 tu_physical_device_to_handle(instance
->physical_devices
+ i
);
589 p
->subsetAllocation
= false;
593 return vk_outarray_status(&out
);
597 tu_GetPhysicalDeviceFeatures(VkPhysicalDevice physicalDevice
,
598 VkPhysicalDeviceFeatures
*pFeatures
)
600 memset(pFeatures
, 0, sizeof(*pFeatures
));
602 *pFeatures
= (VkPhysicalDeviceFeatures
) {
603 .robustBufferAccess
= true,
604 .fullDrawIndexUint32
= true,
605 .imageCubeArray
= true,
606 .independentBlend
= true,
607 .geometryShader
= true,
608 .tessellationShader
= true,
609 .sampleRateShading
= true,
610 .dualSrcBlend
= true,
612 .multiDrawIndirect
= true,
613 .drawIndirectFirstInstance
= true,
615 .depthBiasClamp
= true,
616 .fillModeNonSolid
= true,
621 .multiViewport
= false,
622 .samplerAnisotropy
= true,
623 .textureCompressionETC2
= true,
624 .textureCompressionASTC_LDR
= true,
625 .textureCompressionBC
= true,
626 .occlusionQueryPrecise
= true,
627 .pipelineStatisticsQuery
= false,
628 .vertexPipelineStoresAndAtomics
= true,
629 .fragmentStoresAndAtomics
= true,
630 .shaderTessellationAndGeometryPointSize
= false,
631 .shaderImageGatherExtended
= false,
632 .shaderStorageImageExtendedFormats
= false,
633 .shaderStorageImageMultisample
= false,
634 .shaderUniformBufferArrayDynamicIndexing
= true,
635 .shaderSampledImageArrayDynamicIndexing
= true,
636 .shaderStorageBufferArrayDynamicIndexing
= true,
637 .shaderStorageImageArrayDynamicIndexing
= true,
638 .shaderStorageImageReadWithoutFormat
= false,
639 .shaderStorageImageWriteWithoutFormat
= false,
640 .shaderClipDistance
= false,
641 .shaderCullDistance
= false,
642 .shaderFloat64
= false,
643 .shaderInt64
= false,
644 .shaderInt16
= false,
645 .sparseBinding
= false,
646 .variableMultisampleRate
= false,
647 .inheritedQueries
= false,
652 tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice
,
653 VkPhysicalDeviceFeatures2
*pFeatures
)
655 vk_foreach_struct(ext
, pFeatures
->pNext
)
657 switch (ext
->sType
) {
658 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES
: {
659 VkPhysicalDeviceVulkan11Features
*features
= (void *) ext
;
660 features
->storageBuffer16BitAccess
= false;
661 features
->uniformAndStorageBuffer16BitAccess
= false;
662 features
->storagePushConstant16
= false;
663 features
->storageInputOutput16
= false;
664 features
->multiview
= false;
665 features
->multiviewGeometryShader
= false;
666 features
->multiviewTessellationShader
= false;
667 features
->variablePointersStorageBuffer
= true;
668 features
->variablePointers
= true;
669 features
->protectedMemory
= false;
670 features
->samplerYcbcrConversion
= true;
671 features
->shaderDrawParameters
= true;
674 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES
: {
675 VkPhysicalDeviceVulkan12Features
*features
= (void *) ext
;
676 features
->samplerMirrorClampToEdge
= true;
677 features
->drawIndirectCount
= true;
678 features
->storageBuffer8BitAccess
= false;
679 features
->uniformAndStorageBuffer8BitAccess
= false;
680 features
->storagePushConstant8
= false;
681 features
->shaderBufferInt64Atomics
= false;
682 features
->shaderSharedInt64Atomics
= false;
683 features
->shaderFloat16
= false;
684 features
->shaderInt8
= false;
686 features
->descriptorIndexing
= false;
687 features
->shaderInputAttachmentArrayDynamicIndexing
= false;
688 features
->shaderUniformTexelBufferArrayDynamicIndexing
= false;
689 features
->shaderStorageTexelBufferArrayDynamicIndexing
= false;
690 features
->shaderUniformBufferArrayNonUniformIndexing
= false;
691 features
->shaderSampledImageArrayNonUniformIndexing
= false;
692 features
->shaderStorageBufferArrayNonUniformIndexing
= false;
693 features
->shaderStorageImageArrayNonUniformIndexing
= false;
694 features
->shaderInputAttachmentArrayNonUniformIndexing
= false;
695 features
->shaderUniformTexelBufferArrayNonUniformIndexing
= false;
696 features
->shaderStorageTexelBufferArrayNonUniformIndexing
= false;
697 features
->descriptorBindingUniformBufferUpdateAfterBind
= false;
698 features
->descriptorBindingSampledImageUpdateAfterBind
= false;
699 features
->descriptorBindingStorageImageUpdateAfterBind
= false;
700 features
->descriptorBindingStorageBufferUpdateAfterBind
= false;
701 features
->descriptorBindingUniformTexelBufferUpdateAfterBind
= false;
702 features
->descriptorBindingStorageTexelBufferUpdateAfterBind
= false;
703 features
->descriptorBindingUpdateUnusedWhilePending
= false;
704 features
->descriptorBindingPartiallyBound
= false;
705 features
->descriptorBindingVariableDescriptorCount
= false;
706 features
->runtimeDescriptorArray
= false;
708 features
->samplerFilterMinmax
= true;
709 features
->scalarBlockLayout
= false;
710 features
->imagelessFramebuffer
= false;
711 features
->uniformBufferStandardLayout
= false;
712 features
->shaderSubgroupExtendedTypes
= false;
713 features
->separateDepthStencilLayouts
= false;
714 features
->hostQueryReset
= false;
715 features
->timelineSemaphore
= false;
716 features
->bufferDeviceAddress
= false;
717 features
->bufferDeviceAddressCaptureReplay
= false;
718 features
->bufferDeviceAddressMultiDevice
= false;
719 features
->vulkanMemoryModel
= false;
720 features
->vulkanMemoryModelDeviceScope
= false;
721 features
->vulkanMemoryModelAvailabilityVisibilityChains
= false;
722 features
->shaderOutputViewportIndex
= false;
723 features
->shaderOutputLayer
= false;
724 features
->subgroupBroadcastDynamicId
= false;
727 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES
: {
728 VkPhysicalDeviceVariablePointersFeatures
*features
= (void *) ext
;
729 features
->variablePointersStorageBuffer
= true;
730 features
->variablePointers
= true;
733 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES
: {
734 VkPhysicalDeviceMultiviewFeatures
*features
=
735 (VkPhysicalDeviceMultiviewFeatures
*) ext
;
736 features
->multiview
= false;
737 features
->multiviewGeometryShader
= false;
738 features
->multiviewTessellationShader
= false;
741 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES
: {
742 VkPhysicalDeviceShaderDrawParametersFeatures
*features
=
743 (VkPhysicalDeviceShaderDrawParametersFeatures
*) ext
;
744 features
->shaderDrawParameters
= true;
747 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES
: {
748 VkPhysicalDeviceProtectedMemoryFeatures
*features
=
749 (VkPhysicalDeviceProtectedMemoryFeatures
*) ext
;
750 features
->protectedMemory
= false;
753 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES
: {
754 VkPhysicalDevice16BitStorageFeatures
*features
=
755 (VkPhysicalDevice16BitStorageFeatures
*) ext
;
756 features
->storageBuffer16BitAccess
= false;
757 features
->uniformAndStorageBuffer16BitAccess
= false;
758 features
->storagePushConstant16
= false;
759 features
->storageInputOutput16
= false;
762 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES
: {
763 VkPhysicalDeviceSamplerYcbcrConversionFeatures
*features
=
764 (VkPhysicalDeviceSamplerYcbcrConversionFeatures
*) ext
;
765 features
->samplerYcbcrConversion
= true;
768 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT
: {
769 VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*features
=
770 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*) ext
;
771 features
->shaderInputAttachmentArrayDynamicIndexing
= false;
772 features
->shaderUniformTexelBufferArrayDynamicIndexing
= false;
773 features
->shaderStorageTexelBufferArrayDynamicIndexing
= false;
774 features
->shaderUniformBufferArrayNonUniformIndexing
= false;
775 features
->shaderSampledImageArrayNonUniformIndexing
= false;
776 features
->shaderStorageBufferArrayNonUniformIndexing
= false;
777 features
->shaderStorageImageArrayNonUniformIndexing
= false;
778 features
->shaderInputAttachmentArrayNonUniformIndexing
= false;
779 features
->shaderUniformTexelBufferArrayNonUniformIndexing
= false;
780 features
->shaderStorageTexelBufferArrayNonUniformIndexing
= false;
781 features
->descriptorBindingUniformBufferUpdateAfterBind
= false;
782 features
->descriptorBindingSampledImageUpdateAfterBind
= false;
783 features
->descriptorBindingStorageImageUpdateAfterBind
= false;
784 features
->descriptorBindingStorageBufferUpdateAfterBind
= false;
785 features
->descriptorBindingUniformTexelBufferUpdateAfterBind
= false;
786 features
->descriptorBindingStorageTexelBufferUpdateAfterBind
= false;
787 features
->descriptorBindingUpdateUnusedWhilePending
= false;
788 features
->descriptorBindingPartiallyBound
= false;
789 features
->descriptorBindingVariableDescriptorCount
= false;
790 features
->runtimeDescriptorArray
= false;
793 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT
: {
794 VkPhysicalDeviceConditionalRenderingFeaturesEXT
*features
=
795 (VkPhysicalDeviceConditionalRenderingFeaturesEXT
*) ext
;
796 features
->conditionalRendering
= true;
797 features
->inheritedConditionalRendering
= true;
800 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT
: {
801 VkPhysicalDeviceTransformFeedbackFeaturesEXT
*features
=
802 (VkPhysicalDeviceTransformFeedbackFeaturesEXT
*) ext
;
803 features
->transformFeedback
= true;
804 features
->geometryStreams
= false;
807 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT
: {
808 VkPhysicalDeviceIndexTypeUint8FeaturesEXT
*features
=
809 (VkPhysicalDeviceIndexTypeUint8FeaturesEXT
*)ext
;
810 features
->indexTypeUint8
= true;
813 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT
: {
814 VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*features
=
815 (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*)ext
;
816 features
->vertexAttributeInstanceRateDivisor
= true;
817 features
->vertexAttributeInstanceRateZeroDivisor
= true;
820 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRIVATE_DATA_FEATURES_EXT
: {
821 VkPhysicalDevicePrivateDataFeaturesEXT
*features
=
822 (VkPhysicalDevicePrivateDataFeaturesEXT
*)ext
;
823 features
->privateData
= true;
826 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT
: {
827 VkPhysicalDeviceDepthClipEnableFeaturesEXT
*features
=
828 (VkPhysicalDeviceDepthClipEnableFeaturesEXT
*)ext
;
829 features
->depthClipEnable
= true;
832 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_4444_FORMATS_FEATURES_EXT
: {
833 VkPhysicalDevice4444FormatsFeaturesEXT
*features
= (void *)ext
;
834 features
->formatA4R4G4B4
= true;
835 features
->formatA4B4G4R4
= true;
842 return tu_GetPhysicalDeviceFeatures(physicalDevice
, &pFeatures
->features
);
846 tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice
,
847 VkPhysicalDeviceProperties
*pProperties
)
849 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
850 VkSampleCountFlags sample_counts
=
851 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
853 /* I have no idea what the maximum size is, but the hardware supports very
854 * large numbers of descriptors (at least 2^16). This limit is based on
855 * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
856 * we don't have to think about what to do if that overflows, but really
857 * nothing is likely to get close to this.
859 const size_t max_descriptor_set_size
= (1 << 28) / A6XX_TEX_CONST_DWORDS
;
861 VkPhysicalDeviceLimits limits
= {
862 .maxImageDimension1D
= (1 << 14),
863 .maxImageDimension2D
= (1 << 14),
864 .maxImageDimension3D
= (1 << 11),
865 .maxImageDimensionCube
= (1 << 14),
866 .maxImageArrayLayers
= (1 << 11),
867 .maxTexelBufferElements
= 128 * 1024 * 1024,
868 .maxUniformBufferRange
= MAX_UNIFORM_BUFFER_RANGE
,
869 .maxStorageBufferRange
= MAX_STORAGE_BUFFER_RANGE
,
870 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
871 .maxMemoryAllocationCount
= UINT32_MAX
,
872 .maxSamplerAllocationCount
= 64 * 1024,
873 .bufferImageGranularity
= 64, /* A cache line */
874 .sparseAddressSpaceSize
= 0xffffffffu
, /* buffer max size */
875 .maxBoundDescriptorSets
= MAX_SETS
,
876 .maxPerStageDescriptorSamplers
= max_descriptor_set_size
,
877 .maxPerStageDescriptorUniformBuffers
= max_descriptor_set_size
,
878 .maxPerStageDescriptorStorageBuffers
= max_descriptor_set_size
,
879 .maxPerStageDescriptorSampledImages
= max_descriptor_set_size
,
880 .maxPerStageDescriptorStorageImages
= max_descriptor_set_size
,
881 .maxPerStageDescriptorInputAttachments
= MAX_RTS
,
882 .maxPerStageResources
= max_descriptor_set_size
,
883 .maxDescriptorSetSamplers
= max_descriptor_set_size
,
884 .maxDescriptorSetUniformBuffers
= max_descriptor_set_size
,
885 .maxDescriptorSetUniformBuffersDynamic
= MAX_DYNAMIC_UNIFORM_BUFFERS
,
886 .maxDescriptorSetStorageBuffers
= max_descriptor_set_size
,
887 .maxDescriptorSetStorageBuffersDynamic
= MAX_DYNAMIC_STORAGE_BUFFERS
,
888 .maxDescriptorSetSampledImages
= max_descriptor_set_size
,
889 .maxDescriptorSetStorageImages
= max_descriptor_set_size
,
890 .maxDescriptorSetInputAttachments
= MAX_RTS
,
891 .maxVertexInputAttributes
= 32,
892 .maxVertexInputBindings
= 32,
893 .maxVertexInputAttributeOffset
= 4095,
894 .maxVertexInputBindingStride
= 2048,
895 .maxVertexOutputComponents
= 128,
896 .maxTessellationGenerationLevel
= 64,
897 .maxTessellationPatchSize
= 32,
898 .maxTessellationControlPerVertexInputComponents
= 128,
899 .maxTessellationControlPerVertexOutputComponents
= 128,
900 .maxTessellationControlPerPatchOutputComponents
= 120,
901 .maxTessellationControlTotalOutputComponents
= 4096,
902 .maxTessellationEvaluationInputComponents
= 128,
903 .maxTessellationEvaluationOutputComponents
= 128,
904 .maxGeometryShaderInvocations
= 32,
905 .maxGeometryInputComponents
= 64,
906 .maxGeometryOutputComponents
= 128,
907 .maxGeometryOutputVertices
= 256,
908 .maxGeometryTotalOutputComponents
= 1024,
909 .maxFragmentInputComponents
= 124,
910 .maxFragmentOutputAttachments
= 8,
911 .maxFragmentDualSrcAttachments
= 1,
912 .maxFragmentCombinedOutputResources
= 8,
913 .maxComputeSharedMemorySize
= 32768,
914 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
915 .maxComputeWorkGroupInvocations
= 2048,
916 .maxComputeWorkGroupSize
= { 2048, 2048, 2048 },
917 .subPixelPrecisionBits
= 8,
918 .subTexelPrecisionBits
= 8,
919 .mipmapPrecisionBits
= 8,
920 .maxDrawIndexedIndexValue
= UINT32_MAX
,
921 .maxDrawIndirectCount
= UINT32_MAX
,
922 .maxSamplerLodBias
= 4095.0 / 256.0, /* [-16, 15.99609375] */
923 .maxSamplerAnisotropy
= 16,
924 .maxViewports
= MAX_VIEWPORTS
,
925 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
926 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
927 .viewportSubPixelBits
= 8,
928 .minMemoryMapAlignment
= 4096, /* A page */
929 .minTexelBufferOffsetAlignment
= 64,
930 .minUniformBufferOffsetAlignment
= 64,
931 .minStorageBufferOffsetAlignment
= 64,
932 .minTexelOffset
= -16,
933 .maxTexelOffset
= 15,
934 .minTexelGatherOffset
= -32,
935 .maxTexelGatherOffset
= 31,
936 .minInterpolationOffset
= -0.5,
937 .maxInterpolationOffset
= 0.4375,
938 .subPixelInterpolationOffsetBits
= 4,
939 .maxFramebufferWidth
= (1 << 14),
940 .maxFramebufferHeight
= (1 << 14),
941 .maxFramebufferLayers
= (1 << 10),
942 .framebufferColorSampleCounts
= sample_counts
,
943 .framebufferDepthSampleCounts
= sample_counts
,
944 .framebufferStencilSampleCounts
= sample_counts
,
945 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
946 .maxColorAttachments
= MAX_RTS
,
947 .sampledImageColorSampleCounts
= sample_counts
,
948 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
949 .sampledImageDepthSampleCounts
= sample_counts
,
950 .sampledImageStencilSampleCounts
= sample_counts
,
951 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
952 .maxSampleMaskWords
= 1,
953 .timestampComputeAndGraphics
= true,
954 .timestampPeriod
= 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
955 .maxClipDistances
= 8,
956 .maxCullDistances
= 8,
957 .maxCombinedClipAndCullDistances
= 8,
958 .discreteQueuePriorities
= 1,
959 .pointSizeRange
= { 1, 4092 },
960 .lineWidthRange
= { 0.0, 7.9921875 },
961 .pointSizeGranularity
= 0.0625,
962 .lineWidthGranularity
= (1.0 / 128.0),
963 .strictLines
= false, /* FINISHME */
964 .standardSampleLocations
= true,
965 .optimalBufferCopyOffsetAlignment
= 128,
966 .optimalBufferCopyRowPitchAlignment
= 128,
967 .nonCoherentAtomSize
= 64,
970 *pProperties
= (VkPhysicalDeviceProperties
) {
971 .apiVersion
= tu_physical_device_api_version(pdevice
),
972 .driverVersion
= vk_get_driver_version(),
973 .vendorID
= 0, /* TODO */
975 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
977 .sparseProperties
= { 0 },
980 strcpy(pProperties
->deviceName
, pdevice
->name
);
981 memcpy(pProperties
->pipelineCacheUUID
, pdevice
->cache_uuid
, VK_UUID_SIZE
);
985 tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice
,
986 VkPhysicalDeviceProperties2
*pProperties
)
988 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
989 tu_GetPhysicalDeviceProperties(physicalDevice
, &pProperties
->properties
);
991 vk_foreach_struct(ext
, pProperties
->pNext
)
993 switch (ext
->sType
) {
994 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR
: {
995 VkPhysicalDevicePushDescriptorPropertiesKHR
*properties
=
996 (VkPhysicalDevicePushDescriptorPropertiesKHR
*) ext
;
997 properties
->maxPushDescriptors
= MAX_PUSH_DESCRIPTORS
;
1000 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES
: {
1001 VkPhysicalDeviceIDProperties
*properties
=
1002 (VkPhysicalDeviceIDProperties
*) ext
;
1003 memcpy(properties
->driverUUID
, pdevice
->driver_uuid
, VK_UUID_SIZE
);
1004 memcpy(properties
->deviceUUID
, pdevice
->device_uuid
, VK_UUID_SIZE
);
1005 properties
->deviceLUIDValid
= false;
1008 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES
: {
1009 VkPhysicalDeviceMultiviewProperties
*properties
=
1010 (VkPhysicalDeviceMultiviewProperties
*) ext
;
1011 properties
->maxMultiviewViewCount
= MAX_VIEWS
;
1012 properties
->maxMultiviewInstanceIndex
= INT_MAX
;
1015 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES
: {
1016 VkPhysicalDevicePointClippingProperties
*properties
=
1017 (VkPhysicalDevicePointClippingProperties
*) ext
;
1018 properties
->pointClippingBehavior
=
1019 VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES
;
1022 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES
: {
1023 VkPhysicalDeviceMaintenance3Properties
*properties
=
1024 (VkPhysicalDeviceMaintenance3Properties
*) ext
;
1025 /* Make sure everything is addressable by a signed 32-bit int, and
1026 * our largest descriptors are 96 bytes. */
1027 properties
->maxPerSetDescriptors
= (1ull << 31) / 96;
1028 /* Our buffer size fields allow only this much */
1029 properties
->maxMemoryAllocationSize
= 0xFFFFFFFFull
;
1032 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT
: {
1033 VkPhysicalDeviceTransformFeedbackPropertiesEXT
*properties
=
1034 (VkPhysicalDeviceTransformFeedbackPropertiesEXT
*)ext
;
1036 properties
->maxTransformFeedbackStreams
= IR3_MAX_SO_STREAMS
;
1037 properties
->maxTransformFeedbackBuffers
= IR3_MAX_SO_BUFFERS
;
1038 properties
->maxTransformFeedbackBufferSize
= UINT32_MAX
;
1039 properties
->maxTransformFeedbackStreamDataSize
= 512;
1040 properties
->maxTransformFeedbackBufferDataSize
= 512;
1041 properties
->maxTransformFeedbackBufferDataStride
= 512;
1042 properties
->transformFeedbackQueries
= true;
1043 properties
->transformFeedbackStreamsLinesTriangles
= false;
1044 properties
->transformFeedbackRasterizationStreamSelect
= false;
1045 properties
->transformFeedbackDraw
= true;
1048 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT
: {
1049 VkPhysicalDeviceSampleLocationsPropertiesEXT
*properties
=
1050 (VkPhysicalDeviceSampleLocationsPropertiesEXT
*)ext
;
1051 properties
->sampleLocationSampleCounts
= 0;
1052 if (pdevice
->supported_extensions
.EXT_sample_locations
) {
1053 properties
->sampleLocationSampleCounts
=
1054 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
1056 properties
->maxSampleLocationGridSize
= (VkExtent2D
) { 1 , 1 };
1057 properties
->sampleLocationCoordinateRange
[0] = 0.0f
;
1058 properties
->sampleLocationCoordinateRange
[1] = 0.9375f
;
1059 properties
->sampleLocationSubPixelBits
= 4;
1060 properties
->variableSampleLocations
= true;
1063 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES
: {
1064 VkPhysicalDeviceSamplerFilterMinmaxProperties
*properties
=
1065 (VkPhysicalDeviceSamplerFilterMinmaxProperties
*)ext
;
1066 properties
->filterMinmaxImageComponentMapping
= true;
1067 properties
->filterMinmaxSingleComponentFormats
= true;
1070 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES
: {
1071 VkPhysicalDeviceSubgroupProperties
*properties
=
1072 (VkPhysicalDeviceSubgroupProperties
*)ext
;
1073 properties
->subgroupSize
= 64;
1074 properties
->supportedStages
= VK_SHADER_STAGE_COMPUTE_BIT
;
1075 properties
->supportedOperations
= VK_SUBGROUP_FEATURE_BASIC_BIT
|
1076 VK_SUBGROUP_FEATURE_VOTE_BIT
;
1077 properties
->quadOperationsInAllStages
= false;
1080 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT
: {
1081 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*props
=
1082 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*)ext
;
1083 props
->maxVertexAttribDivisor
= UINT32_MAX
;
1092 static const VkQueueFamilyProperties tu_queue_family_properties
= {
1094 VK_QUEUE_GRAPHICS_BIT
| VK_QUEUE_COMPUTE_BIT
| VK_QUEUE_TRANSFER_BIT
,
1096 .timestampValidBits
= 48,
1097 .minImageTransferGranularity
= { 1, 1, 1 },
1101 tu_GetPhysicalDeviceQueueFamilyProperties(
1102 VkPhysicalDevice physicalDevice
,
1103 uint32_t *pQueueFamilyPropertyCount
,
1104 VkQueueFamilyProperties
*pQueueFamilyProperties
)
1106 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1108 vk_outarray_append(&out
, p
) { *p
= tu_queue_family_properties
; }
1112 tu_GetPhysicalDeviceQueueFamilyProperties2(
1113 VkPhysicalDevice physicalDevice
,
1114 uint32_t *pQueueFamilyPropertyCount
,
1115 VkQueueFamilyProperties2
*pQueueFamilyProperties
)
1117 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1119 vk_outarray_append(&out
, p
)
1121 p
->queueFamilyProperties
= tu_queue_family_properties
;
1126 tu_get_system_heap_size()
1128 struct sysinfo info
;
1131 uint64_t total_ram
= (uint64_t) info
.totalram
* (uint64_t) info
.mem_unit
;
1133 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
1134 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
1136 uint64_t available_ram
;
1137 if (total_ram
<= 4ull * 1024ull * 1024ull * 1024ull)
1138 available_ram
= total_ram
/ 2;
1140 available_ram
= total_ram
* 3 / 4;
1142 return available_ram
;
1146 tu_GetPhysicalDeviceMemoryProperties(
1147 VkPhysicalDevice physicalDevice
,
1148 VkPhysicalDeviceMemoryProperties
*pMemoryProperties
)
1150 pMemoryProperties
->memoryHeapCount
= 1;
1151 pMemoryProperties
->memoryHeaps
[0].size
= tu_get_system_heap_size();
1152 pMemoryProperties
->memoryHeaps
[0].flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
;
1154 pMemoryProperties
->memoryTypeCount
= 1;
1155 pMemoryProperties
->memoryTypes
[0].propertyFlags
=
1156 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
1157 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
1158 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
;
1159 pMemoryProperties
->memoryTypes
[0].heapIndex
= 0;
1163 tu_GetPhysicalDeviceMemoryProperties2(
1164 VkPhysicalDevice physicalDevice
,
1165 VkPhysicalDeviceMemoryProperties2
*pMemoryProperties
)
1167 return tu_GetPhysicalDeviceMemoryProperties(
1168 physicalDevice
, &pMemoryProperties
->memoryProperties
);
1172 tu_queue_init(struct tu_device
*device
,
1173 struct tu_queue
*queue
,
1174 uint32_t queue_family_index
,
1176 VkDeviceQueueCreateFlags flags
)
1178 vk_object_base_init(&device
->vk
, &queue
->base
, VK_OBJECT_TYPE_QUEUE
);
1180 queue
->device
= device
;
1181 queue
->queue_family_index
= queue_family_index
;
1182 queue
->queue_idx
= idx
;
1183 queue
->flags
= flags
;
1185 int ret
= tu_drm_submitqueue_new(device
, 0, &queue
->msm_queue_id
);
1187 return VK_ERROR_INITIALIZATION_FAILED
;
1189 tu_fence_init(&queue
->submit_fence
, false);
1195 tu_queue_finish(struct tu_queue
*queue
)
1197 tu_fence_finish(&queue
->submit_fence
);
1198 tu_drm_submitqueue_close(queue
->device
, queue
->msm_queue_id
);
1202 tu_get_device_extension_index(const char *name
)
1204 for (unsigned i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; ++i
) {
1205 if (strcmp(name
, tu_device_extensions
[i
].extensionName
) == 0)
1211 struct PACKED bcolor_entry
{
1223 uint32_t z24
; /* also s8? */
1224 uint16_t srgb
[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
1226 } border_color
[] = {
1227 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = {},
1228 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = {},
1229 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = {
1230 .fp32
[3] = 0x3f800000,
1238 .rgb10a2
= 0xc0000000,
1241 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = {
1245 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = {
1246 .fp32
[0 ... 3] = 0x3f800000,
1247 .ui16
[0 ... 3] = 0xffff,
1248 .si16
[0 ... 3] = 0x7fff,
1249 .fp16
[0 ... 3] = 0x3c00,
1253 .ui8
[0 ... 3] = 0xff,
1254 .si8
[0 ... 3] = 0x7f,
1255 .rgb10a2
= 0xffffffff,
1257 .srgb
[0 ... 3] = 0x3c00,
1259 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = {
1266 tu_CreateDevice(VkPhysicalDevice physicalDevice
,
1267 const VkDeviceCreateInfo
*pCreateInfo
,
1268 const VkAllocationCallbacks
*pAllocator
,
1271 TU_FROM_HANDLE(tu_physical_device
, physical_device
, physicalDevice
);
1273 struct tu_device
*device
;
1275 /* Check enabled features */
1276 if (pCreateInfo
->pEnabledFeatures
) {
1277 VkPhysicalDeviceFeatures supported_features
;
1278 tu_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
1279 VkBool32
*supported_feature
= (VkBool32
*) &supported_features
;
1280 VkBool32
*enabled_feature
= (VkBool32
*) pCreateInfo
->pEnabledFeatures
;
1281 unsigned num_features
=
1282 sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
1283 for (uint32_t i
= 0; i
< num_features
; i
++) {
1284 if (enabled_feature
[i
] && !supported_feature
[i
])
1285 return vk_error(physical_device
->instance
,
1286 VK_ERROR_FEATURE_NOT_PRESENT
);
1290 device
= vk_zalloc2(&physical_device
->instance
->alloc
, pAllocator
,
1291 sizeof(*device
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1293 return vk_error(physical_device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1295 vk_device_init(&device
->vk
, pCreateInfo
,
1296 &physical_device
->instance
->alloc
, pAllocator
);
1298 device
->instance
= physical_device
->instance
;
1299 device
->physical_device
= physical_device
;
1300 device
->_lost
= false;
1302 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
1303 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
1304 int index
= tu_get_device_extension_index(ext_name
);
1306 !physical_device
->supported_extensions
.extensions
[index
]) {
1307 vk_free(&device
->vk
.alloc
, device
);
1308 return vk_error(physical_device
->instance
,
1309 VK_ERROR_EXTENSION_NOT_PRESENT
);
1312 device
->enabled_extensions
.extensions
[index
] = true;
1315 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
1316 const VkDeviceQueueCreateInfo
*queue_create
=
1317 &pCreateInfo
->pQueueCreateInfos
[i
];
1318 uint32_t qfi
= queue_create
->queueFamilyIndex
;
1319 device
->queues
[qfi
] = vk_alloc(
1320 &device
->vk
.alloc
, queue_create
->queueCount
* sizeof(struct tu_queue
),
1321 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1322 if (!device
->queues
[qfi
]) {
1323 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
1327 memset(device
->queues
[qfi
], 0,
1328 queue_create
->queueCount
* sizeof(struct tu_queue
));
1330 device
->queue_count
[qfi
] = queue_create
->queueCount
;
1332 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
1333 result
= tu_queue_init(device
, &device
->queues
[qfi
][q
], qfi
, q
,
1334 queue_create
->flags
);
1335 if (result
!= VK_SUCCESS
)
1340 device
->compiler
= ir3_compiler_create(NULL
, physical_device
->gpu_id
);
1341 if (!device
->compiler
)
1344 /* initial sizes, these will increase if there is overflow */
1345 device
->vsc_draw_strm_pitch
= 0x1000 + VSC_PAD
;
1346 device
->vsc_prim_strm_pitch
= 0x4000 + VSC_PAD
;
1348 STATIC_ASSERT(sizeof(border_color
) == sizeof(((struct tu6_global
*) 0)->border_color
));
1349 result
= tu_bo_init_new(device
, &device
->global_bo
, sizeof(struct tu6_global
));
1350 if (result
!= VK_SUCCESS
)
1351 goto fail_global_bo
;
1353 result
= tu_bo_map(device
, &device
->global_bo
);
1354 if (result
!= VK_SUCCESS
)
1355 goto fail_global_bo_map
;
1357 struct tu6_global
*global
= device
->global_bo
.map
;
1358 memcpy(global
->border_color
, border_color
, sizeof(border_color
));
1359 global
->predicate
= 0;
1360 tu_init_clear_blit_shaders(global
);
1362 VkPipelineCacheCreateInfo ci
;
1363 ci
.sType
= VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
;
1366 ci
.pInitialData
= NULL
;
1367 ci
.initialDataSize
= 0;
1370 tu_CreatePipelineCache(tu_device_to_handle(device
), &ci
, NULL
, &pc
);
1371 if (result
!= VK_SUCCESS
)
1372 goto fail_pipeline_cache
;
1374 device
->mem_cache
= tu_pipeline_cache_from_handle(pc
);
1376 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++)
1377 mtx_init(&device
->scratch_bos
[i
].construct_mtx
, mtx_plain
);
1379 mtx_init(&device
->vsc_pitch_mtx
, mtx_plain
);
1381 *pDevice
= tu_device_to_handle(device
);
1384 fail_pipeline_cache
:
1386 tu_bo_finish(device
, &device
->global_bo
);
1389 ralloc_free(device
->compiler
);
1392 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1393 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1394 tu_queue_finish(&device
->queues
[i
][q
]);
1395 if (device
->queue_count
[i
])
1396 vk_object_free(&device
->vk
, NULL
, device
->queues
[i
]);
1399 vk_free(&device
->vk
.alloc
, device
);
1404 tu_DestroyDevice(VkDevice _device
, const VkAllocationCallbacks
*pAllocator
)
1406 TU_FROM_HANDLE(tu_device
, device
, _device
);
1411 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1412 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1413 tu_queue_finish(&device
->queues
[i
][q
]);
1414 if (device
->queue_count
[i
])
1415 vk_object_free(&device
->vk
, NULL
, device
->queues
[i
]);
1418 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++) {
1419 if (device
->scratch_bos
[i
].initialized
)
1420 tu_bo_finish(device
, &device
->scratch_bos
[i
].bo
);
1423 ir3_compiler_destroy(device
->compiler
);
1425 VkPipelineCache pc
= tu_pipeline_cache_to_handle(device
->mem_cache
);
1426 tu_DestroyPipelineCache(tu_device_to_handle(device
), pc
, NULL
);
1428 vk_free(&device
->vk
.alloc
, device
);
1432 _tu_device_set_lost(struct tu_device
*device
,
1433 const char *file
, int line
,
1434 const char *msg
, ...)
1436 /* Set the flag indicating that waits should return in finite time even
1437 * after device loss.
1439 p_atomic_inc(&device
->_lost
);
1441 /* TODO: Report the log message through VkDebugReportCallbackEXT instead */
1442 fprintf(stderr
, "%s:%d: ", file
, line
);
1445 vfprintf(stderr
, msg
, ap
);
1448 if (env_var_as_boolean("TU_ABORT_ON_DEVICE_LOSS", false))
1451 return VK_ERROR_DEVICE_LOST
;
1455 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
)
1457 unsigned size_log2
= MAX2(util_logbase2_ceil64(size
), MIN_SCRATCH_BO_SIZE_LOG2
);
1458 unsigned index
= size_log2
- MIN_SCRATCH_BO_SIZE_LOG2
;
1459 assert(index
< ARRAY_SIZE(dev
->scratch_bos
));
1461 for (unsigned i
= index
; i
< ARRAY_SIZE(dev
->scratch_bos
); i
++) {
1462 if (p_atomic_read(&dev
->scratch_bos
[i
].initialized
)) {
1463 /* Fast path: just return the already-allocated BO. */
1464 *bo
= &dev
->scratch_bos
[i
].bo
;
1469 /* Slow path: actually allocate the BO. We take a lock because the process
1470 * of allocating it is slow, and we don't want to block the CPU while it
1473 mtx_lock(&dev
->scratch_bos
[index
].construct_mtx
);
1475 /* Another thread may have allocated it already while we were waiting on
1476 * the lock. We need to check this in order to avoid double-allocating.
1478 if (dev
->scratch_bos
[index
].initialized
) {
1479 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1480 *bo
= &dev
->scratch_bos
[index
].bo
;
1484 unsigned bo_size
= 1ull << size_log2
;
1485 VkResult result
= tu_bo_init_new(dev
, &dev
->scratch_bos
[index
].bo
, bo_size
);
1486 if (result
!= VK_SUCCESS
) {
1487 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1491 p_atomic_set(&dev
->scratch_bos
[index
].initialized
, true);
1493 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1495 *bo
= &dev
->scratch_bos
[index
].bo
;
1500 tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount
,
1501 VkLayerProperties
*pProperties
)
1503 *pPropertyCount
= 0;
1508 tu_EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice
,
1509 uint32_t *pPropertyCount
,
1510 VkLayerProperties
*pProperties
)
1512 *pPropertyCount
= 0;
1517 tu_GetDeviceQueue2(VkDevice _device
,
1518 const VkDeviceQueueInfo2
*pQueueInfo
,
1521 TU_FROM_HANDLE(tu_device
, device
, _device
);
1522 struct tu_queue
*queue
;
1525 &device
->queues
[pQueueInfo
->queueFamilyIndex
][pQueueInfo
->queueIndex
];
1526 if (pQueueInfo
->flags
!= queue
->flags
) {
1527 /* From the Vulkan 1.1.70 spec:
1529 * "The queue returned by vkGetDeviceQueue2 must have the same
1530 * flags value from this structure as that used at device
1531 * creation time in a VkDeviceQueueCreateInfo instance. If no
1532 * matching flags were specified at device creation time then
1533 * pQueue will return VK_NULL_HANDLE."
1535 *pQueue
= VK_NULL_HANDLE
;
1539 *pQueue
= tu_queue_to_handle(queue
);
1543 tu_GetDeviceQueue(VkDevice _device
,
1544 uint32_t queueFamilyIndex
,
1545 uint32_t queueIndex
,
1548 const VkDeviceQueueInfo2 info
=
1549 (VkDeviceQueueInfo2
) { .sType
= VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2
,
1550 .queueFamilyIndex
= queueFamilyIndex
,
1551 .queueIndex
= queueIndex
};
1553 tu_GetDeviceQueue2(_device
, &info
, pQueue
);
1557 tu_get_semaphore_syncobjs(const VkSemaphore
*sems
,
1560 struct drm_msm_gem_submit_syncobj
**out
,
1561 uint32_t *out_count
)
1563 uint32_t syncobj_count
= 0;
1564 struct drm_msm_gem_submit_syncobj
*syncobjs
;
1566 for (uint32_t i
= 0; i
< sem_count
; ++i
) {
1567 TU_FROM_HANDLE(tu_semaphore
, sem
, sems
[i
]);
1569 struct tu_semaphore_part
*part
=
1570 sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
?
1571 &sem
->temporary
: &sem
->permanent
;
1573 if (part
->kind
== TU_SEMAPHORE_SYNCOBJ
)
1578 *out_count
= syncobj_count
;
1582 *out
= syncobjs
= calloc(syncobj_count
, sizeof (*syncobjs
));
1584 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1586 for (uint32_t i
= 0, j
= 0; i
< sem_count
; ++i
) {
1587 TU_FROM_HANDLE(tu_semaphore
, sem
, sems
[i
]);
1589 struct tu_semaphore_part
*part
=
1590 sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
?
1591 &sem
->temporary
: &sem
->permanent
;
1593 if (part
->kind
== TU_SEMAPHORE_SYNCOBJ
) {
1594 syncobjs
[j
].handle
= part
->syncobj
;
1595 syncobjs
[j
].flags
= wait
? MSM_SUBMIT_SYNCOBJ_RESET
: 0;
1605 tu_semaphores_remove_temp(struct tu_device
*device
,
1606 const VkSemaphore
*sems
,
1609 for (uint32_t i
= 0; i
< sem_count
; ++i
) {
1610 TU_FROM_HANDLE(tu_semaphore
, sem
, sems
[i
]);
1611 tu_semaphore_remove_temp(device
, sem
);
1616 tu_QueueSubmit(VkQueue _queue
,
1617 uint32_t submitCount
,
1618 const VkSubmitInfo
*pSubmits
,
1621 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1624 for (uint32_t i
= 0; i
< submitCount
; ++i
) {
1625 const VkSubmitInfo
*submit
= pSubmits
+ i
;
1626 const bool last_submit
= (i
== submitCount
- 1);
1627 struct drm_msm_gem_submit_syncobj
*in_syncobjs
= NULL
, *out_syncobjs
= NULL
;
1628 uint32_t nr_in_syncobjs
, nr_out_syncobjs
;
1629 struct tu_bo_list bo_list
;
1630 tu_bo_list_init(&bo_list
);
1632 result
= tu_get_semaphore_syncobjs(pSubmits
[i
].pWaitSemaphores
,
1633 pSubmits
[i
].waitSemaphoreCount
,
1634 false, &in_syncobjs
, &nr_in_syncobjs
);
1635 if (result
!= VK_SUCCESS
) {
1636 return tu_device_set_lost(queue
->device
,
1637 "failed to allocate space for semaphore submission\n");
1640 result
= tu_get_semaphore_syncobjs(pSubmits
[i
].pSignalSemaphores
,
1641 pSubmits
[i
].signalSemaphoreCount
,
1642 false, &out_syncobjs
, &nr_out_syncobjs
);
1643 if (result
!= VK_SUCCESS
) {
1645 return tu_device_set_lost(queue
->device
,
1646 "failed to allocate space for semaphore submission\n");
1649 uint32_t entry_count
= 0;
1650 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1651 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1652 entry_count
+= cmdbuf
->cs
.entry_count
;
1655 struct drm_msm_gem_submit_cmd cmds
[entry_count
];
1656 uint32_t entry_idx
= 0;
1657 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1658 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1659 struct tu_cs
*cs
= &cmdbuf
->cs
;
1660 for (unsigned i
= 0; i
< cs
->entry_count
; ++i
, ++entry_idx
) {
1661 cmds
[entry_idx
].type
= MSM_SUBMIT_CMD_BUF
;
1662 cmds
[entry_idx
].submit_idx
=
1663 tu_bo_list_add(&bo_list
, cs
->entries
[i
].bo
,
1664 MSM_SUBMIT_BO_READ
| MSM_SUBMIT_BO_DUMP
);
1665 cmds
[entry_idx
].submit_offset
= cs
->entries
[i
].offset
;
1666 cmds
[entry_idx
].size
= cs
->entries
[i
].size
;
1667 cmds
[entry_idx
].pad
= 0;
1668 cmds
[entry_idx
].nr_relocs
= 0;
1669 cmds
[entry_idx
].relocs
= 0;
1672 tu_bo_list_merge(&bo_list
, &cmdbuf
->bo_list
);
1675 uint32_t flags
= MSM_PIPE_3D0
;
1676 if (nr_in_syncobjs
) {
1677 flags
|= MSM_SUBMIT_SYNCOBJ_IN
;
1679 if (nr_out_syncobjs
) {
1680 flags
|= MSM_SUBMIT_SYNCOBJ_OUT
;
1684 flags
|= MSM_SUBMIT_FENCE_FD_OUT
;
1687 struct drm_msm_gem_submit req
= {
1689 .queueid
= queue
->msm_queue_id
,
1690 .bos
= (uint64_t)(uintptr_t) bo_list
.bo_infos
,
1691 .nr_bos
= bo_list
.count
,
1692 .cmds
= (uint64_t)(uintptr_t)cmds
,
1693 .nr_cmds
= entry_count
,
1694 .in_syncobjs
= (uint64_t)(uintptr_t)in_syncobjs
,
1695 .out_syncobjs
= (uint64_t)(uintptr_t)out_syncobjs
,
1696 .nr_in_syncobjs
= nr_in_syncobjs
,
1697 .nr_out_syncobjs
= nr_out_syncobjs
,
1698 .syncobj_stride
= sizeof(struct drm_msm_gem_submit_syncobj
),
1701 int ret
= drmCommandWriteRead(queue
->device
->physical_device
->local_fd
,
1707 return tu_device_set_lost(queue
->device
, "submit failed: %s\n",
1711 tu_bo_list_destroy(&bo_list
);
1715 tu_semaphores_remove_temp(queue
->device
, pSubmits
[i
].pWaitSemaphores
,
1716 pSubmits
[i
].waitSemaphoreCount
);
1718 /* no need to merge fences as queue execution is serialized */
1719 tu_fence_update_fd(&queue
->submit_fence
, req
.fence_fd
);
1720 } else if (last_submit
) {
1721 close(req
.fence_fd
);
1725 if (_fence
!= VK_NULL_HANDLE
) {
1726 TU_FROM_HANDLE(tu_fence
, fence
, _fence
);
1727 tu_fence_copy(fence
, &queue
->submit_fence
);
1734 tu_QueueWaitIdle(VkQueue _queue
)
1736 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1738 if (tu_device_is_lost(queue
->device
))
1739 return VK_ERROR_DEVICE_LOST
;
1741 tu_fence_wait_idle(&queue
->submit_fence
);
1747 tu_DeviceWaitIdle(VkDevice _device
)
1749 TU_FROM_HANDLE(tu_device
, device
, _device
);
1751 if (tu_device_is_lost(device
))
1752 return VK_ERROR_DEVICE_LOST
;
1754 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1755 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
1756 tu_QueueWaitIdle(tu_queue_to_handle(&device
->queues
[i
][q
]));
1763 tu_EnumerateInstanceExtensionProperties(const char *pLayerName
,
1764 uint32_t *pPropertyCount
,
1765 VkExtensionProperties
*pProperties
)
1767 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1769 /* We spport no lyaers */
1771 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1773 for (int i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; i
++) {
1774 if (tu_instance_extensions_supported
.extensions
[i
]) {
1775 vk_outarray_append(&out
, prop
) { *prop
= tu_instance_extensions
[i
]; }
1779 return vk_outarray_status(&out
);
1783 tu_EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice
,
1784 const char *pLayerName
,
1785 uint32_t *pPropertyCount
,
1786 VkExtensionProperties
*pProperties
)
1788 /* We spport no lyaers */
1789 TU_FROM_HANDLE(tu_physical_device
, device
, physicalDevice
);
1790 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1792 /* We spport no lyaers */
1794 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1796 for (int i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; i
++) {
1797 if (device
->supported_extensions
.extensions
[i
]) {
1798 vk_outarray_append(&out
, prop
) { *prop
= tu_device_extensions
[i
]; }
1802 return vk_outarray_status(&out
);
1806 tu_GetInstanceProcAddr(VkInstance _instance
, const char *pName
)
1808 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
1810 return tu_lookup_entrypoint_checked(
1811 pName
, instance
? instance
->api_version
: 0,
1812 instance
? &instance
->enabled_extensions
: NULL
, NULL
);
1815 /* The loader wants us to expose a second GetInstanceProcAddr function
1816 * to work around certain LD_PRELOAD issues seen in apps.
1819 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1820 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
);
1823 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1824 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
)
1826 return tu_GetInstanceProcAddr(instance
, pName
);
1830 tu_GetDeviceProcAddr(VkDevice _device
, const char *pName
)
1832 TU_FROM_HANDLE(tu_device
, device
, _device
);
1834 return tu_lookup_entrypoint_checked(pName
, device
->instance
->api_version
,
1835 &device
->instance
->enabled_extensions
,
1836 &device
->enabled_extensions
);
1840 tu_alloc_memory(struct tu_device
*device
,
1841 const VkMemoryAllocateInfo
*pAllocateInfo
,
1842 const VkAllocationCallbacks
*pAllocator
,
1843 VkDeviceMemory
*pMem
)
1845 struct tu_device_memory
*mem
;
1848 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1850 if (pAllocateInfo
->allocationSize
== 0) {
1851 /* Apparently, this is allowed */
1852 *pMem
= VK_NULL_HANDLE
;
1856 mem
= vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*mem
),
1857 VK_OBJECT_TYPE_DEVICE_MEMORY
);
1859 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1861 const VkImportMemoryFdInfoKHR
*fd_info
=
1862 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
1863 if (fd_info
&& !fd_info
->handleType
)
1867 assert(fd_info
->handleType
==
1868 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
1869 fd_info
->handleType
==
1870 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
1873 * TODO Importing the same fd twice gives us the same handle without
1874 * reference counting. We need to maintain a per-instance handle-to-bo
1875 * table and add reference count to tu_bo.
1877 result
= tu_bo_init_dmabuf(device
, &mem
->bo
,
1878 pAllocateInfo
->allocationSize
, fd_info
->fd
);
1879 if (result
== VK_SUCCESS
) {
1880 /* take ownership and close the fd */
1885 tu_bo_init_new(device
, &mem
->bo
, pAllocateInfo
->allocationSize
);
1888 if (result
!= VK_SUCCESS
) {
1889 vk_object_free(&device
->vk
, pAllocator
, mem
);
1893 mem
->size
= pAllocateInfo
->allocationSize
;
1894 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1897 mem
->user_ptr
= NULL
;
1899 *pMem
= tu_device_memory_to_handle(mem
);
1905 tu_AllocateMemory(VkDevice _device
,
1906 const VkMemoryAllocateInfo
*pAllocateInfo
,
1907 const VkAllocationCallbacks
*pAllocator
,
1908 VkDeviceMemory
*pMem
)
1910 TU_FROM_HANDLE(tu_device
, device
, _device
);
1911 return tu_alloc_memory(device
, pAllocateInfo
, pAllocator
, pMem
);
1915 tu_FreeMemory(VkDevice _device
,
1916 VkDeviceMemory _mem
,
1917 const VkAllocationCallbacks
*pAllocator
)
1919 TU_FROM_HANDLE(tu_device
, device
, _device
);
1920 TU_FROM_HANDLE(tu_device_memory
, mem
, _mem
);
1925 tu_bo_finish(device
, &mem
->bo
);
1926 vk_object_free(&device
->vk
, pAllocator
, mem
);
1930 tu_MapMemory(VkDevice _device
,
1931 VkDeviceMemory _memory
,
1932 VkDeviceSize offset
,
1934 VkMemoryMapFlags flags
,
1937 TU_FROM_HANDLE(tu_device
, device
, _device
);
1938 TU_FROM_HANDLE(tu_device_memory
, mem
, _memory
);
1946 if (mem
->user_ptr
) {
1947 *ppData
= mem
->user_ptr
;
1948 } else if (!mem
->map
) {
1949 result
= tu_bo_map(device
, &mem
->bo
);
1950 if (result
!= VK_SUCCESS
)
1952 *ppData
= mem
->map
= mem
->bo
.map
;
1961 return vk_error(device
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
1965 tu_UnmapMemory(VkDevice _device
, VkDeviceMemory _memory
)
1967 /* I do not see any unmapping done by the freedreno Gallium driver. */
1971 tu_FlushMappedMemoryRanges(VkDevice _device
,
1972 uint32_t memoryRangeCount
,
1973 const VkMappedMemoryRange
*pMemoryRanges
)
1979 tu_InvalidateMappedMemoryRanges(VkDevice _device
,
1980 uint32_t memoryRangeCount
,
1981 const VkMappedMemoryRange
*pMemoryRanges
)
1987 tu_GetBufferMemoryRequirements(VkDevice _device
,
1989 VkMemoryRequirements
*pMemoryRequirements
)
1991 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
1993 pMemoryRequirements
->memoryTypeBits
= 1;
1994 pMemoryRequirements
->alignment
= 64;
1995 pMemoryRequirements
->size
=
1996 align64(buffer
->size
, pMemoryRequirements
->alignment
);
2000 tu_GetBufferMemoryRequirements2(
2002 const VkBufferMemoryRequirementsInfo2
*pInfo
,
2003 VkMemoryRequirements2
*pMemoryRequirements
)
2005 tu_GetBufferMemoryRequirements(device
, pInfo
->buffer
,
2006 &pMemoryRequirements
->memoryRequirements
);
2010 tu_GetImageMemoryRequirements(VkDevice _device
,
2012 VkMemoryRequirements
*pMemoryRequirements
)
2014 TU_FROM_HANDLE(tu_image
, image
, _image
);
2016 pMemoryRequirements
->memoryTypeBits
= 1;
2017 pMemoryRequirements
->size
= image
->total_size
;
2018 pMemoryRequirements
->alignment
= image
->layout
[0].base_align
;
2022 tu_GetImageMemoryRequirements2(VkDevice device
,
2023 const VkImageMemoryRequirementsInfo2
*pInfo
,
2024 VkMemoryRequirements2
*pMemoryRequirements
)
2026 tu_GetImageMemoryRequirements(device
, pInfo
->image
,
2027 &pMemoryRequirements
->memoryRequirements
);
2031 tu_GetImageSparseMemoryRequirements(
2034 uint32_t *pSparseMemoryRequirementCount
,
2035 VkSparseImageMemoryRequirements
*pSparseMemoryRequirements
)
2041 tu_GetImageSparseMemoryRequirements2(
2043 const VkImageSparseMemoryRequirementsInfo2
*pInfo
,
2044 uint32_t *pSparseMemoryRequirementCount
,
2045 VkSparseImageMemoryRequirements2
*pSparseMemoryRequirements
)
2051 tu_GetDeviceMemoryCommitment(VkDevice device
,
2052 VkDeviceMemory memory
,
2053 VkDeviceSize
*pCommittedMemoryInBytes
)
2055 *pCommittedMemoryInBytes
= 0;
2059 tu_BindBufferMemory2(VkDevice device
,
2060 uint32_t bindInfoCount
,
2061 const VkBindBufferMemoryInfo
*pBindInfos
)
2063 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
2064 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
2065 TU_FROM_HANDLE(tu_buffer
, buffer
, pBindInfos
[i
].buffer
);
2068 buffer
->bo
= &mem
->bo
;
2069 buffer
->bo_offset
= pBindInfos
[i
].memoryOffset
;
2078 tu_BindBufferMemory(VkDevice device
,
2080 VkDeviceMemory memory
,
2081 VkDeviceSize memoryOffset
)
2083 const VkBindBufferMemoryInfo info
= {
2084 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
2087 .memoryOffset
= memoryOffset
2090 return tu_BindBufferMemory2(device
, 1, &info
);
2094 tu_BindImageMemory2(VkDevice device
,
2095 uint32_t bindInfoCount
,
2096 const VkBindImageMemoryInfo
*pBindInfos
)
2098 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
2099 TU_FROM_HANDLE(tu_image
, image
, pBindInfos
[i
].image
);
2100 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
2103 image
->bo
= &mem
->bo
;
2104 image
->bo_offset
= pBindInfos
[i
].memoryOffset
;
2107 image
->bo_offset
= 0;
2115 tu_BindImageMemory(VkDevice device
,
2117 VkDeviceMemory memory
,
2118 VkDeviceSize memoryOffset
)
2120 const VkBindImageMemoryInfo info
= {
2121 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
2124 .memoryOffset
= memoryOffset
2127 return tu_BindImageMemory2(device
, 1, &info
);
2131 tu_QueueBindSparse(VkQueue _queue
,
2132 uint32_t bindInfoCount
,
2133 const VkBindSparseInfo
*pBindInfo
,
2139 // Queue semaphore functions
2143 tu_semaphore_part_destroy(struct tu_device
*device
,
2144 struct tu_semaphore_part
*part
)
2146 switch(part
->kind
) {
2147 case TU_SEMAPHORE_NONE
:
2149 case TU_SEMAPHORE_SYNCOBJ
:
2150 drmSyncobjDestroy(device
->physical_device
->local_fd
, part
->syncobj
);
2153 part
->kind
= TU_SEMAPHORE_NONE
;
2157 tu_semaphore_remove_temp(struct tu_device
*device
,
2158 struct tu_semaphore
*sem
)
2160 if (sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
) {
2161 tu_semaphore_part_destroy(device
, &sem
->temporary
);
2166 tu_CreateSemaphore(VkDevice _device
,
2167 const VkSemaphoreCreateInfo
*pCreateInfo
,
2168 const VkAllocationCallbacks
*pAllocator
,
2169 VkSemaphore
*pSemaphore
)
2171 TU_FROM_HANDLE(tu_device
, device
, _device
);
2173 struct tu_semaphore
*sem
=
2174 vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*sem
),
2175 VK_OBJECT_TYPE_SEMAPHORE
);
2177 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2179 const VkExportSemaphoreCreateInfo
*export
=
2180 vk_find_struct_const(pCreateInfo
->pNext
, EXPORT_SEMAPHORE_CREATE_INFO
);
2181 VkExternalSemaphoreHandleTypeFlags handleTypes
=
2182 export
? export
->handleTypes
: 0;
2184 sem
->permanent
.kind
= TU_SEMAPHORE_NONE
;
2185 sem
->temporary
.kind
= TU_SEMAPHORE_NONE
;
2188 if (drmSyncobjCreate(device
->physical_device
->local_fd
, 0, &sem
->permanent
.syncobj
) < 0) {
2189 vk_free2(&device
->vk
.alloc
, pAllocator
, sem
);
2190 return VK_ERROR_OUT_OF_HOST_MEMORY
;
2192 sem
->permanent
.kind
= TU_SEMAPHORE_SYNCOBJ
;
2194 *pSemaphore
= tu_semaphore_to_handle(sem
);
2199 tu_DestroySemaphore(VkDevice _device
,
2200 VkSemaphore _semaphore
,
2201 const VkAllocationCallbacks
*pAllocator
)
2203 TU_FROM_HANDLE(tu_device
, device
, _device
);
2204 TU_FROM_HANDLE(tu_semaphore
, sem
, _semaphore
);
2208 tu_semaphore_part_destroy(device
, &sem
->permanent
);
2209 tu_semaphore_part_destroy(device
, &sem
->temporary
);
2211 vk_object_free(&device
->vk
, pAllocator
, sem
);
2215 tu_CreateEvent(VkDevice _device
,
2216 const VkEventCreateInfo
*pCreateInfo
,
2217 const VkAllocationCallbacks
*pAllocator
,
2220 TU_FROM_HANDLE(tu_device
, device
, _device
);
2222 struct tu_event
*event
=
2223 vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*event
),
2224 VK_OBJECT_TYPE_EVENT
);
2226 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2228 VkResult result
= tu_bo_init_new(device
, &event
->bo
, 0x1000);
2229 if (result
!= VK_SUCCESS
)
2232 result
= tu_bo_map(device
, &event
->bo
);
2233 if (result
!= VK_SUCCESS
)
2236 *pEvent
= tu_event_to_handle(event
);
2241 tu_bo_finish(device
, &event
->bo
);
2243 vk_object_free(&device
->vk
, pAllocator
, event
);
2244 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2248 tu_DestroyEvent(VkDevice _device
,
2250 const VkAllocationCallbacks
*pAllocator
)
2252 TU_FROM_HANDLE(tu_device
, device
, _device
);
2253 TU_FROM_HANDLE(tu_event
, event
, _event
);
2258 tu_bo_finish(device
, &event
->bo
);
2259 vk_object_free(&device
->vk
, pAllocator
, event
);
2263 tu_GetEventStatus(VkDevice _device
, VkEvent _event
)
2265 TU_FROM_HANDLE(tu_event
, event
, _event
);
2267 if (*(uint64_t*) event
->bo
.map
== 1)
2268 return VK_EVENT_SET
;
2269 return VK_EVENT_RESET
;
2273 tu_SetEvent(VkDevice _device
, VkEvent _event
)
2275 TU_FROM_HANDLE(tu_event
, event
, _event
);
2276 *(uint64_t*) event
->bo
.map
= 1;
2282 tu_ResetEvent(VkDevice _device
, VkEvent _event
)
2284 TU_FROM_HANDLE(tu_event
, event
, _event
);
2285 *(uint64_t*) event
->bo
.map
= 0;
2291 tu_CreateBuffer(VkDevice _device
,
2292 const VkBufferCreateInfo
*pCreateInfo
,
2293 const VkAllocationCallbacks
*pAllocator
,
2296 TU_FROM_HANDLE(tu_device
, device
, _device
);
2297 struct tu_buffer
*buffer
;
2299 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
2301 buffer
= vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*buffer
),
2302 VK_OBJECT_TYPE_BUFFER
);
2304 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2306 buffer
->size
= pCreateInfo
->size
;
2307 buffer
->usage
= pCreateInfo
->usage
;
2308 buffer
->flags
= pCreateInfo
->flags
;
2310 *pBuffer
= tu_buffer_to_handle(buffer
);
2316 tu_DestroyBuffer(VkDevice _device
,
2318 const VkAllocationCallbacks
*pAllocator
)
2320 TU_FROM_HANDLE(tu_device
, device
, _device
);
2321 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
2326 vk_object_free(&device
->vk
, pAllocator
, buffer
);
2330 tu_CreateFramebuffer(VkDevice _device
,
2331 const VkFramebufferCreateInfo
*pCreateInfo
,
2332 const VkAllocationCallbacks
*pAllocator
,
2333 VkFramebuffer
*pFramebuffer
)
2335 TU_FROM_HANDLE(tu_device
, device
, _device
);
2336 TU_FROM_HANDLE(tu_render_pass
, pass
, pCreateInfo
->renderPass
);
2337 struct tu_framebuffer
*framebuffer
;
2339 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
2341 size_t size
= sizeof(*framebuffer
) + sizeof(struct tu_attachment_info
) *
2342 pCreateInfo
->attachmentCount
;
2343 framebuffer
= vk_object_alloc(&device
->vk
, pAllocator
, size
,
2344 VK_OBJECT_TYPE_FRAMEBUFFER
);
2345 if (framebuffer
== NULL
)
2346 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2348 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
2349 framebuffer
->width
= pCreateInfo
->width
;
2350 framebuffer
->height
= pCreateInfo
->height
;
2351 framebuffer
->layers
= pCreateInfo
->layers
;
2352 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
2353 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
2354 struct tu_image_view
*iview
= tu_image_view_from_handle(_iview
);
2355 framebuffer
->attachments
[i
].attachment
= iview
;
2358 tu_framebuffer_tiling_config(framebuffer
, device
, pass
);
2360 *pFramebuffer
= tu_framebuffer_to_handle(framebuffer
);
2365 tu_DestroyFramebuffer(VkDevice _device
,
2367 const VkAllocationCallbacks
*pAllocator
)
2369 TU_FROM_HANDLE(tu_device
, device
, _device
);
2370 TU_FROM_HANDLE(tu_framebuffer
, fb
, _fb
);
2375 vk_object_free(&device
->vk
, pAllocator
, fb
);
2379 tu_init_sampler(struct tu_device
*device
,
2380 struct tu_sampler
*sampler
,
2381 const VkSamplerCreateInfo
*pCreateInfo
)
2383 const struct VkSamplerReductionModeCreateInfo
*reduction
=
2384 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_REDUCTION_MODE_CREATE_INFO
);
2385 const struct VkSamplerYcbcrConversionInfo
*ycbcr_conversion
=
2386 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_YCBCR_CONVERSION_INFO
);
2388 unsigned aniso
= pCreateInfo
->anisotropyEnable
?
2389 util_last_bit(MIN2((uint32_t)pCreateInfo
->maxAnisotropy
>> 1, 8)) : 0;
2390 bool miplinear
= (pCreateInfo
->mipmapMode
== VK_SAMPLER_MIPMAP_MODE_LINEAR
);
2391 float min_lod
= CLAMP(pCreateInfo
->minLod
, 0.0f
, 4095.0f
/ 256.0f
);
2392 float max_lod
= CLAMP(pCreateInfo
->maxLod
, 0.0f
, 4095.0f
/ 256.0f
);
2394 sampler
->descriptor
[0] =
2395 COND(miplinear
, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR
) |
2396 A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo
->magFilter
, aniso
)) |
2397 A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo
->minFilter
, aniso
)) |
2398 A6XX_TEX_SAMP_0_ANISO(aniso
) |
2399 A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo
->addressModeU
)) |
2400 A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo
->addressModeV
)) |
2401 A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo
->addressModeW
)) |
2402 A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo
->mipLodBias
);
2403 sampler
->descriptor
[1] =
2404 /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
2405 COND(pCreateInfo
->unnormalizedCoordinates
, A6XX_TEX_SAMP_1_UNNORM_COORDS
) |
2406 A6XX_TEX_SAMP_1_MIN_LOD(min_lod
) |
2407 A6XX_TEX_SAMP_1_MAX_LOD(max_lod
) |
2408 COND(pCreateInfo
->compareEnable
,
2409 A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo
->compareOp
)));
2410 /* This is an offset into the border_color BO, which we fill with all the
2411 * possible Vulkan border colors in the correct order, so we can just use
2412 * the Vulkan enum with no translation necessary.
2414 sampler
->descriptor
[2] =
2415 A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo
->borderColor
*
2416 sizeof(struct bcolor_entry
));
2417 sampler
->descriptor
[3] = 0;
2420 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(
2421 tu6_reduction_mode(reduction
->reductionMode
));
2424 sampler
->ycbcr_sampler
= ycbcr_conversion
?
2425 tu_sampler_ycbcr_conversion_from_handle(ycbcr_conversion
->conversion
) : NULL
;
2427 if (sampler
->ycbcr_sampler
&&
2428 sampler
->ycbcr_sampler
->chroma_filter
== VK_FILTER_LINEAR
) {
2429 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_CHROMA_LINEAR
;
2433 * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
2438 tu_CreateSampler(VkDevice _device
,
2439 const VkSamplerCreateInfo
*pCreateInfo
,
2440 const VkAllocationCallbacks
*pAllocator
,
2441 VkSampler
*pSampler
)
2443 TU_FROM_HANDLE(tu_device
, device
, _device
);
2444 struct tu_sampler
*sampler
;
2446 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
2448 sampler
= vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*sampler
),
2449 VK_OBJECT_TYPE_SAMPLER
);
2451 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2453 tu_init_sampler(device
, sampler
, pCreateInfo
);
2454 *pSampler
= tu_sampler_to_handle(sampler
);
2460 tu_DestroySampler(VkDevice _device
,
2462 const VkAllocationCallbacks
*pAllocator
)
2464 TU_FROM_HANDLE(tu_device
, device
, _device
);
2465 TU_FROM_HANDLE(tu_sampler
, sampler
, _sampler
);
2470 vk_object_free(&device
->vk
, pAllocator
, sampler
);
2473 /* vk_icd.h does not declare this function, so we declare it here to
2474 * suppress Wmissing-prototypes.
2476 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2477 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
2479 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2480 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
2482 /* For the full details on loader interface versioning, see
2483 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2484 * What follows is a condensed summary, to help you navigate the large and
2485 * confusing official doc.
2487 * - Loader interface v0 is incompatible with later versions. We don't
2490 * - In loader interface v1:
2491 * - The first ICD entrypoint called by the loader is
2492 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2494 * - The ICD must statically expose no other Vulkan symbol unless it
2495 * is linked with -Bsymbolic.
2496 * - Each dispatchable Vulkan handle created by the ICD must be
2497 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2498 * ICD must initialize VK_LOADER_DATA.loadMagic to
2500 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2501 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2502 * such loader-managed surfaces.
2504 * - Loader interface v2 differs from v1 in:
2505 * - The first ICD entrypoint called by the loader is
2506 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2507 * statically expose this entrypoint.
2509 * - Loader interface v3 differs from v2 in:
2510 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2511 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2512 * because the loader no longer does so.
2514 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);
2519 tu_GetMemoryFdKHR(VkDevice _device
,
2520 const VkMemoryGetFdInfoKHR
*pGetFdInfo
,
2523 TU_FROM_HANDLE(tu_device
, device
, _device
);
2524 TU_FROM_HANDLE(tu_device_memory
, memory
, pGetFdInfo
->memory
);
2526 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
2528 /* At the moment, we support only the below handle types. */
2529 assert(pGetFdInfo
->handleType
==
2530 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
2531 pGetFdInfo
->handleType
==
2532 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2534 int prime_fd
= tu_bo_export_dmabuf(device
, &memory
->bo
);
2536 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2543 tu_GetMemoryFdPropertiesKHR(VkDevice _device
,
2544 VkExternalMemoryHandleTypeFlagBits handleType
,
2546 VkMemoryFdPropertiesKHR
*pMemoryFdProperties
)
2548 assert(handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2549 pMemoryFdProperties
->memoryTypeBits
= 1;
2554 tu_ImportFenceFdKHR(VkDevice _device
,
2555 const VkImportFenceFdInfoKHR
*pImportFenceFdInfo
)
2563 tu_GetFenceFdKHR(VkDevice _device
,
2564 const VkFenceGetFdInfoKHR
*pGetFdInfo
,
2573 tu_ImportSemaphoreFdKHR(VkDevice _device
,
2574 const VkImportSemaphoreFdInfoKHR
*pImportSemaphoreFdInfo
)
2576 TU_FROM_HANDLE(tu_device
, device
, _device
);
2577 TU_FROM_HANDLE(tu_semaphore
, sem
, pImportSemaphoreFdInfo
->semaphore
);
2579 struct tu_semaphore_part
*dst
= NULL
;
2581 if (pImportSemaphoreFdInfo
->flags
& VK_SEMAPHORE_IMPORT_TEMPORARY_BIT
) {
2582 dst
= &sem
->temporary
;
2584 dst
= &sem
->permanent
;
2587 uint32_t syncobj
= dst
->kind
== TU_SEMAPHORE_SYNCOBJ
? dst
->syncobj
: 0;
2589 switch(pImportSemaphoreFdInfo
->handleType
) {
2590 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
: {
2591 uint32_t old_syncobj
= syncobj
;
2592 ret
= drmSyncobjFDToHandle(device
->physical_device
->local_fd
, pImportSemaphoreFdInfo
->fd
, &syncobj
);
2594 close(pImportSemaphoreFdInfo
->fd
);
2596 drmSyncobjDestroy(device
->physical_device
->local_fd
, old_syncobj
);
2600 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
: {
2602 ret
= drmSyncobjCreate(device
->physical_device
->local_fd
, 0, &syncobj
);
2606 if (pImportSemaphoreFdInfo
->fd
== -1) {
2607 ret
= drmSyncobjSignal(device
->physical_device
->local_fd
, &syncobj
, 1);
2609 ret
= drmSyncobjImportSyncFile(device
->physical_device
->local_fd
, syncobj
, pImportSemaphoreFdInfo
->fd
);
2612 close(pImportSemaphoreFdInfo
->fd
);
2616 unreachable("Unhandled semaphore handle type");
2620 return VK_ERROR_INVALID_EXTERNAL_HANDLE
;
2622 dst
->syncobj
= syncobj
;
2623 dst
->kind
= TU_SEMAPHORE_SYNCOBJ
;
2629 tu_GetSemaphoreFdKHR(VkDevice _device
,
2630 const VkSemaphoreGetFdInfoKHR
*pGetFdInfo
,
2633 TU_FROM_HANDLE(tu_device
, device
, _device
);
2634 TU_FROM_HANDLE(tu_semaphore
, sem
, pGetFdInfo
->semaphore
);
2636 uint32_t syncobj_handle
;
2638 if (sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
) {
2639 assert(sem
->temporary
.kind
== TU_SEMAPHORE_SYNCOBJ
);
2640 syncobj_handle
= sem
->temporary
.syncobj
;
2642 assert(sem
->permanent
.kind
== TU_SEMAPHORE_SYNCOBJ
);
2643 syncobj_handle
= sem
->permanent
.syncobj
;
2646 switch(pGetFdInfo
->handleType
) {
2647 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
:
2648 ret
= drmSyncobjHandleToFD(device
->physical_device
->local_fd
, syncobj_handle
, pFd
);
2650 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
:
2651 ret
= drmSyncobjExportSyncFile(device
->physical_device
->local_fd
, syncobj_handle
, pFd
);
2653 if (sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
) {
2654 tu_semaphore_part_destroy(device
, &sem
->temporary
);
2656 drmSyncobjReset(device
->physical_device
->local_fd
, &syncobj_handle
, 1);
2661 unreachable("Unhandled semaphore handle type");
2665 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE
);
2670 static bool tu_has_syncobj(struct tu_physical_device
*pdev
)
2673 if (drmGetCap(pdev
->local_fd
, DRM_CAP_SYNCOBJ
, &value
))
2675 return value
&& pdev
->msm_major_version
== 1 && pdev
->msm_minor_version
>= 6;
2679 tu_GetPhysicalDeviceExternalSemaphoreProperties(
2680 VkPhysicalDevice physicalDevice
,
2681 const VkPhysicalDeviceExternalSemaphoreInfo
*pExternalSemaphoreInfo
,
2682 VkExternalSemaphoreProperties
*pExternalSemaphoreProperties
)
2684 TU_FROM_HANDLE(tu_physical_device
, pdev
, physicalDevice
);
2686 if (tu_has_syncobj(pdev
) &&
2687 (pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
||
2688 pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
)) {
2689 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
| VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
;
2690 pExternalSemaphoreProperties
->compatibleHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
| VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
;
2691 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT
|
2692 VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT
;
2694 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= 0;
2695 pExternalSemaphoreProperties
->compatibleHandleTypes
= 0;
2696 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= 0;
2701 tu_GetPhysicalDeviceExternalFenceProperties(
2702 VkPhysicalDevice physicalDevice
,
2703 const VkPhysicalDeviceExternalFenceInfo
*pExternalFenceInfo
,
2704 VkExternalFenceProperties
*pExternalFenceProperties
)
2706 pExternalFenceProperties
->exportFromImportedHandleTypes
= 0;
2707 pExternalFenceProperties
->compatibleHandleTypes
= 0;
2708 pExternalFenceProperties
->externalFenceFeatures
= 0;
2712 tu_CreateDebugReportCallbackEXT(
2713 VkInstance _instance
,
2714 const VkDebugReportCallbackCreateInfoEXT
*pCreateInfo
,
2715 const VkAllocationCallbacks
*pAllocator
,
2716 VkDebugReportCallbackEXT
*pCallback
)
2718 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2719 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
2720 pCreateInfo
, pAllocator
,
2721 &instance
->alloc
, pCallback
);
2725 tu_DestroyDebugReportCallbackEXT(VkInstance _instance
,
2726 VkDebugReportCallbackEXT _callback
,
2727 const VkAllocationCallbacks
*pAllocator
)
2729 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2730 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
2731 _callback
, pAllocator
, &instance
->alloc
);
2735 tu_DebugReportMessageEXT(VkInstance _instance
,
2736 VkDebugReportFlagsEXT flags
,
2737 VkDebugReportObjectTypeEXT objectType
,
2740 int32_t messageCode
,
2741 const char *pLayerPrefix
,
2742 const char *pMessage
)
2744 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2745 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
2746 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
2750 tu_GetDeviceGroupPeerMemoryFeatures(
2753 uint32_t localDeviceIndex
,
2754 uint32_t remoteDeviceIndex
,
2755 VkPeerMemoryFeatureFlags
*pPeerMemoryFeatures
)
2757 assert(localDeviceIndex
== remoteDeviceIndex
);
2759 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
2760 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
2761 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
2762 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
2765 void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
2766 VkPhysicalDevice physicalDevice
,
2767 VkSampleCountFlagBits samples
,
2768 VkMultisamplePropertiesEXT
* pMultisampleProperties
)
2770 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
2772 if (samples
<= VK_SAMPLE_COUNT_4_BIT
&& pdevice
->supported_extensions
.EXT_sample_locations
)
2773 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 1, 1 };
2775 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 0, 0 };
2780 tu_CreatePrivateDataSlotEXT(VkDevice _device
,
2781 const VkPrivateDataSlotCreateInfoEXT
* pCreateInfo
,
2782 const VkAllocationCallbacks
* pAllocator
,
2783 VkPrivateDataSlotEXT
* pPrivateDataSlot
)
2785 TU_FROM_HANDLE(tu_device
, device
, _device
);
2786 return vk_private_data_slot_create(&device
->vk
,
2793 tu_DestroyPrivateDataSlotEXT(VkDevice _device
,
2794 VkPrivateDataSlotEXT privateDataSlot
,
2795 const VkAllocationCallbacks
* pAllocator
)
2797 TU_FROM_HANDLE(tu_device
, device
, _device
);
2798 vk_private_data_slot_destroy(&device
->vk
, privateDataSlot
, pAllocator
);
2802 tu_SetPrivateDataEXT(VkDevice _device
,
2803 VkObjectType objectType
,
2804 uint64_t objectHandle
,
2805 VkPrivateDataSlotEXT privateDataSlot
,
2808 TU_FROM_HANDLE(tu_device
, device
, _device
);
2809 return vk_object_base_set_private_data(&device
->vk
,
2817 tu_GetPrivateDataEXT(VkDevice _device
,
2818 VkObjectType objectType
,
2819 uint64_t objectHandle
,
2820 VkPrivateDataSlotEXT privateDataSlot
,
2823 TU_FROM_HANDLE(tu_device
, device
, _device
);
2824 vk_object_base_get_private_data(&device
->vk
,