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
= false;
797 features
->inheritedConditionalRendering
= false;
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;
836 return tu_GetPhysicalDeviceFeatures(physicalDevice
, &pFeatures
->features
);
840 tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice
,
841 VkPhysicalDeviceProperties
*pProperties
)
843 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
844 VkSampleCountFlags sample_counts
=
845 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
847 /* I have no idea what the maximum size is, but the hardware supports very
848 * large numbers of descriptors (at least 2^16). This limit is based on
849 * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
850 * we don't have to think about what to do if that overflows, but really
851 * nothing is likely to get close to this.
853 const size_t max_descriptor_set_size
= (1 << 28) / A6XX_TEX_CONST_DWORDS
;
855 VkPhysicalDeviceLimits limits
= {
856 .maxImageDimension1D
= (1 << 14),
857 .maxImageDimension2D
= (1 << 14),
858 .maxImageDimension3D
= (1 << 11),
859 .maxImageDimensionCube
= (1 << 14),
860 .maxImageArrayLayers
= (1 << 11),
861 .maxTexelBufferElements
= 128 * 1024 * 1024,
862 .maxUniformBufferRange
= MAX_UNIFORM_BUFFER_RANGE
,
863 .maxStorageBufferRange
= MAX_STORAGE_BUFFER_RANGE
,
864 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
865 .maxMemoryAllocationCount
= UINT32_MAX
,
866 .maxSamplerAllocationCount
= 64 * 1024,
867 .bufferImageGranularity
= 64, /* A cache line */
868 .sparseAddressSpaceSize
= 0xffffffffu
, /* buffer max size */
869 .maxBoundDescriptorSets
= MAX_SETS
,
870 .maxPerStageDescriptorSamplers
= max_descriptor_set_size
,
871 .maxPerStageDescriptorUniformBuffers
= max_descriptor_set_size
,
872 .maxPerStageDescriptorStorageBuffers
= max_descriptor_set_size
,
873 .maxPerStageDescriptorSampledImages
= max_descriptor_set_size
,
874 .maxPerStageDescriptorStorageImages
= max_descriptor_set_size
,
875 .maxPerStageDescriptorInputAttachments
= MAX_RTS
,
876 .maxPerStageResources
= max_descriptor_set_size
,
877 .maxDescriptorSetSamplers
= max_descriptor_set_size
,
878 .maxDescriptorSetUniformBuffers
= max_descriptor_set_size
,
879 .maxDescriptorSetUniformBuffersDynamic
= MAX_DYNAMIC_UNIFORM_BUFFERS
,
880 .maxDescriptorSetStorageBuffers
= max_descriptor_set_size
,
881 .maxDescriptorSetStorageBuffersDynamic
= MAX_DYNAMIC_STORAGE_BUFFERS
,
882 .maxDescriptorSetSampledImages
= max_descriptor_set_size
,
883 .maxDescriptorSetStorageImages
= max_descriptor_set_size
,
884 .maxDescriptorSetInputAttachments
= MAX_RTS
,
885 .maxVertexInputAttributes
= 32,
886 .maxVertexInputBindings
= 32,
887 .maxVertexInputAttributeOffset
= 4095,
888 .maxVertexInputBindingStride
= 2048,
889 .maxVertexOutputComponents
= 128,
890 .maxTessellationGenerationLevel
= 64,
891 .maxTessellationPatchSize
= 32,
892 .maxTessellationControlPerVertexInputComponents
= 128,
893 .maxTessellationControlPerVertexOutputComponents
= 128,
894 .maxTessellationControlPerPatchOutputComponents
= 120,
895 .maxTessellationControlTotalOutputComponents
= 4096,
896 .maxTessellationEvaluationInputComponents
= 128,
897 .maxTessellationEvaluationOutputComponents
= 128,
898 .maxGeometryShaderInvocations
= 32,
899 .maxGeometryInputComponents
= 64,
900 .maxGeometryOutputComponents
= 128,
901 .maxGeometryOutputVertices
= 256,
902 .maxGeometryTotalOutputComponents
= 1024,
903 .maxFragmentInputComponents
= 124,
904 .maxFragmentOutputAttachments
= 8,
905 .maxFragmentDualSrcAttachments
= 1,
906 .maxFragmentCombinedOutputResources
= 8,
907 .maxComputeSharedMemorySize
= 32768,
908 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
909 .maxComputeWorkGroupInvocations
= 2048,
910 .maxComputeWorkGroupSize
= { 2048, 2048, 2048 },
911 .subPixelPrecisionBits
= 8,
912 .subTexelPrecisionBits
= 8,
913 .mipmapPrecisionBits
= 8,
914 .maxDrawIndexedIndexValue
= UINT32_MAX
,
915 .maxDrawIndirectCount
= UINT32_MAX
,
916 .maxSamplerLodBias
= 4095.0 / 256.0, /* [-16, 15.99609375] */
917 .maxSamplerAnisotropy
= 16,
918 .maxViewports
= MAX_VIEWPORTS
,
919 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
920 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
921 .viewportSubPixelBits
= 8,
922 .minMemoryMapAlignment
= 4096, /* A page */
923 .minTexelBufferOffsetAlignment
= 64,
924 .minUniformBufferOffsetAlignment
= 64,
925 .minStorageBufferOffsetAlignment
= 64,
926 .minTexelOffset
= -16,
927 .maxTexelOffset
= 15,
928 .minTexelGatherOffset
= -32,
929 .maxTexelGatherOffset
= 31,
930 .minInterpolationOffset
= -0.5,
931 .maxInterpolationOffset
= 0.4375,
932 .subPixelInterpolationOffsetBits
= 4,
933 .maxFramebufferWidth
= (1 << 14),
934 .maxFramebufferHeight
= (1 << 14),
935 .maxFramebufferLayers
= (1 << 10),
936 .framebufferColorSampleCounts
= sample_counts
,
937 .framebufferDepthSampleCounts
= sample_counts
,
938 .framebufferStencilSampleCounts
= sample_counts
,
939 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
940 .maxColorAttachments
= MAX_RTS
,
941 .sampledImageColorSampleCounts
= sample_counts
,
942 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
943 .sampledImageDepthSampleCounts
= sample_counts
,
944 .sampledImageStencilSampleCounts
= sample_counts
,
945 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
946 .maxSampleMaskWords
= 1,
947 .timestampComputeAndGraphics
= true,
948 .timestampPeriod
= 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
949 .maxClipDistances
= 8,
950 .maxCullDistances
= 8,
951 .maxCombinedClipAndCullDistances
= 8,
952 .discreteQueuePriorities
= 1,
953 .pointSizeRange
= { 1, 4092 },
954 .lineWidthRange
= { 0.0, 7.9921875 },
955 .pointSizeGranularity
= 0.0625,
956 .lineWidthGranularity
= (1.0 / 128.0),
957 .strictLines
= false, /* FINISHME */
958 .standardSampleLocations
= true,
959 .optimalBufferCopyOffsetAlignment
= 128,
960 .optimalBufferCopyRowPitchAlignment
= 128,
961 .nonCoherentAtomSize
= 64,
964 *pProperties
= (VkPhysicalDeviceProperties
) {
965 .apiVersion
= tu_physical_device_api_version(pdevice
),
966 .driverVersion
= vk_get_driver_version(),
967 .vendorID
= 0, /* TODO */
969 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
971 .sparseProperties
= { 0 },
974 strcpy(pProperties
->deviceName
, pdevice
->name
);
975 memcpy(pProperties
->pipelineCacheUUID
, pdevice
->cache_uuid
, VK_UUID_SIZE
);
979 tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice
,
980 VkPhysicalDeviceProperties2
*pProperties
)
982 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
983 tu_GetPhysicalDeviceProperties(physicalDevice
, &pProperties
->properties
);
985 vk_foreach_struct(ext
, pProperties
->pNext
)
987 switch (ext
->sType
) {
988 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR
: {
989 VkPhysicalDevicePushDescriptorPropertiesKHR
*properties
=
990 (VkPhysicalDevicePushDescriptorPropertiesKHR
*) ext
;
991 properties
->maxPushDescriptors
= MAX_PUSH_DESCRIPTORS
;
994 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES
: {
995 VkPhysicalDeviceIDProperties
*properties
=
996 (VkPhysicalDeviceIDProperties
*) ext
;
997 memcpy(properties
->driverUUID
, pdevice
->driver_uuid
, VK_UUID_SIZE
);
998 memcpy(properties
->deviceUUID
, pdevice
->device_uuid
, VK_UUID_SIZE
);
999 properties
->deviceLUIDValid
= false;
1002 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES
: {
1003 VkPhysicalDeviceMultiviewProperties
*properties
=
1004 (VkPhysicalDeviceMultiviewProperties
*) ext
;
1005 properties
->maxMultiviewViewCount
= MAX_VIEWS
;
1006 properties
->maxMultiviewInstanceIndex
= INT_MAX
;
1009 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES
: {
1010 VkPhysicalDevicePointClippingProperties
*properties
=
1011 (VkPhysicalDevicePointClippingProperties
*) ext
;
1012 properties
->pointClippingBehavior
=
1013 VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES
;
1016 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES
: {
1017 VkPhysicalDeviceMaintenance3Properties
*properties
=
1018 (VkPhysicalDeviceMaintenance3Properties
*) ext
;
1019 /* Make sure everything is addressable by a signed 32-bit int, and
1020 * our largest descriptors are 96 bytes. */
1021 properties
->maxPerSetDescriptors
= (1ull << 31) / 96;
1022 /* Our buffer size fields allow only this much */
1023 properties
->maxMemoryAllocationSize
= 0xFFFFFFFFull
;
1026 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT
: {
1027 VkPhysicalDeviceTransformFeedbackPropertiesEXT
*properties
=
1028 (VkPhysicalDeviceTransformFeedbackPropertiesEXT
*)ext
;
1030 properties
->maxTransformFeedbackStreams
= IR3_MAX_SO_STREAMS
;
1031 properties
->maxTransformFeedbackBuffers
= IR3_MAX_SO_BUFFERS
;
1032 properties
->maxTransformFeedbackBufferSize
= UINT32_MAX
;
1033 properties
->maxTransformFeedbackStreamDataSize
= 512;
1034 properties
->maxTransformFeedbackBufferDataSize
= 512;
1035 properties
->maxTransformFeedbackBufferDataStride
= 512;
1036 properties
->transformFeedbackQueries
= true;
1037 properties
->transformFeedbackStreamsLinesTriangles
= false;
1038 properties
->transformFeedbackRasterizationStreamSelect
= false;
1039 properties
->transformFeedbackDraw
= true;
1042 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT
: {
1043 VkPhysicalDeviceSampleLocationsPropertiesEXT
*properties
=
1044 (VkPhysicalDeviceSampleLocationsPropertiesEXT
*)ext
;
1045 properties
->sampleLocationSampleCounts
= 0;
1046 if (pdevice
->supported_extensions
.EXT_sample_locations
) {
1047 properties
->sampleLocationSampleCounts
=
1048 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
1050 properties
->maxSampleLocationGridSize
= (VkExtent2D
) { 1 , 1 };
1051 properties
->sampleLocationCoordinateRange
[0] = 0.0f
;
1052 properties
->sampleLocationCoordinateRange
[1] = 0.9375f
;
1053 properties
->sampleLocationSubPixelBits
= 4;
1054 properties
->variableSampleLocations
= true;
1057 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES
: {
1058 VkPhysicalDeviceSamplerFilterMinmaxProperties
*properties
=
1059 (VkPhysicalDeviceSamplerFilterMinmaxProperties
*)ext
;
1060 properties
->filterMinmaxImageComponentMapping
= true;
1061 properties
->filterMinmaxSingleComponentFormats
= true;
1064 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES
: {
1065 VkPhysicalDeviceSubgroupProperties
*properties
=
1066 (VkPhysicalDeviceSubgroupProperties
*)ext
;
1067 properties
->subgroupSize
= 64;
1068 properties
->supportedStages
= VK_SHADER_STAGE_COMPUTE_BIT
;
1069 properties
->supportedOperations
= VK_SUBGROUP_FEATURE_BASIC_BIT
|
1070 VK_SUBGROUP_FEATURE_VOTE_BIT
;
1071 properties
->quadOperationsInAllStages
= false;
1074 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT
: {
1075 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*props
=
1076 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*)ext
;
1077 props
->maxVertexAttribDivisor
= UINT32_MAX
;
1086 static const VkQueueFamilyProperties tu_queue_family_properties
= {
1088 VK_QUEUE_GRAPHICS_BIT
| VK_QUEUE_COMPUTE_BIT
| VK_QUEUE_TRANSFER_BIT
,
1090 .timestampValidBits
= 48,
1091 .minImageTransferGranularity
= { 1, 1, 1 },
1095 tu_GetPhysicalDeviceQueueFamilyProperties(
1096 VkPhysicalDevice physicalDevice
,
1097 uint32_t *pQueueFamilyPropertyCount
,
1098 VkQueueFamilyProperties
*pQueueFamilyProperties
)
1100 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1102 vk_outarray_append(&out
, p
) { *p
= tu_queue_family_properties
; }
1106 tu_GetPhysicalDeviceQueueFamilyProperties2(
1107 VkPhysicalDevice physicalDevice
,
1108 uint32_t *pQueueFamilyPropertyCount
,
1109 VkQueueFamilyProperties2
*pQueueFamilyProperties
)
1111 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1113 vk_outarray_append(&out
, p
)
1115 p
->queueFamilyProperties
= tu_queue_family_properties
;
1120 tu_get_system_heap_size()
1122 struct sysinfo info
;
1125 uint64_t total_ram
= (uint64_t) info
.totalram
* (uint64_t) info
.mem_unit
;
1127 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
1128 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
1130 uint64_t available_ram
;
1131 if (total_ram
<= 4ull * 1024ull * 1024ull * 1024ull)
1132 available_ram
= total_ram
/ 2;
1134 available_ram
= total_ram
* 3 / 4;
1136 return available_ram
;
1140 tu_GetPhysicalDeviceMemoryProperties(
1141 VkPhysicalDevice physicalDevice
,
1142 VkPhysicalDeviceMemoryProperties
*pMemoryProperties
)
1144 pMemoryProperties
->memoryHeapCount
= 1;
1145 pMemoryProperties
->memoryHeaps
[0].size
= tu_get_system_heap_size();
1146 pMemoryProperties
->memoryHeaps
[0].flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
;
1148 pMemoryProperties
->memoryTypeCount
= 1;
1149 pMemoryProperties
->memoryTypes
[0].propertyFlags
=
1150 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
1151 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
1152 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
;
1153 pMemoryProperties
->memoryTypes
[0].heapIndex
= 0;
1157 tu_GetPhysicalDeviceMemoryProperties2(
1158 VkPhysicalDevice physicalDevice
,
1159 VkPhysicalDeviceMemoryProperties2
*pMemoryProperties
)
1161 return tu_GetPhysicalDeviceMemoryProperties(
1162 physicalDevice
, &pMemoryProperties
->memoryProperties
);
1166 tu_queue_init(struct tu_device
*device
,
1167 struct tu_queue
*queue
,
1168 uint32_t queue_family_index
,
1170 VkDeviceQueueCreateFlags flags
)
1172 vk_object_base_init(&device
->vk
, &queue
->base
, VK_OBJECT_TYPE_QUEUE
);
1174 queue
->device
= device
;
1175 queue
->queue_family_index
= queue_family_index
;
1176 queue
->queue_idx
= idx
;
1177 queue
->flags
= flags
;
1179 int ret
= tu_drm_submitqueue_new(device
, 0, &queue
->msm_queue_id
);
1181 return VK_ERROR_INITIALIZATION_FAILED
;
1183 tu_fence_init(&queue
->submit_fence
, false);
1189 tu_queue_finish(struct tu_queue
*queue
)
1191 tu_fence_finish(&queue
->submit_fence
);
1192 tu_drm_submitqueue_close(queue
->device
, queue
->msm_queue_id
);
1196 tu_get_device_extension_index(const char *name
)
1198 for (unsigned i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; ++i
) {
1199 if (strcmp(name
, tu_device_extensions
[i
].extensionName
) == 0)
1205 struct PACKED bcolor_entry
{
1217 uint32_t z24
; /* also s8? */
1218 uint16_t srgb
[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
1220 } border_color
[] = {
1221 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = {},
1222 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = {},
1223 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = {
1224 .fp32
[3] = 0x3f800000,
1232 .rgb10a2
= 0xc0000000,
1235 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = {
1239 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = {
1240 .fp32
[0 ... 3] = 0x3f800000,
1241 .ui16
[0 ... 3] = 0xffff,
1242 .si16
[0 ... 3] = 0x7fff,
1243 .fp16
[0 ... 3] = 0x3c00,
1247 .ui8
[0 ... 3] = 0xff,
1248 .si8
[0 ... 3] = 0x7f,
1249 .rgb10a2
= 0xffffffff,
1251 .srgb
[0 ... 3] = 0x3c00,
1253 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = {
1260 tu_CreateDevice(VkPhysicalDevice physicalDevice
,
1261 const VkDeviceCreateInfo
*pCreateInfo
,
1262 const VkAllocationCallbacks
*pAllocator
,
1265 TU_FROM_HANDLE(tu_physical_device
, physical_device
, physicalDevice
);
1267 struct tu_device
*device
;
1269 /* Check enabled features */
1270 if (pCreateInfo
->pEnabledFeatures
) {
1271 VkPhysicalDeviceFeatures supported_features
;
1272 tu_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
1273 VkBool32
*supported_feature
= (VkBool32
*) &supported_features
;
1274 VkBool32
*enabled_feature
= (VkBool32
*) pCreateInfo
->pEnabledFeatures
;
1275 unsigned num_features
=
1276 sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
1277 for (uint32_t i
= 0; i
< num_features
; i
++) {
1278 if (enabled_feature
[i
] && !supported_feature
[i
])
1279 return vk_error(physical_device
->instance
,
1280 VK_ERROR_FEATURE_NOT_PRESENT
);
1284 device
= vk_zalloc2(&physical_device
->instance
->alloc
, pAllocator
,
1285 sizeof(*device
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1287 return vk_error(physical_device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1289 vk_device_init(&device
->vk
, pCreateInfo
,
1290 &physical_device
->instance
->alloc
, pAllocator
);
1292 device
->instance
= physical_device
->instance
;
1293 device
->physical_device
= physical_device
;
1294 device
->_lost
= false;
1296 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
1297 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
1298 int index
= tu_get_device_extension_index(ext_name
);
1300 !physical_device
->supported_extensions
.extensions
[index
]) {
1301 vk_free(&device
->vk
.alloc
, device
);
1302 return vk_error(physical_device
->instance
,
1303 VK_ERROR_EXTENSION_NOT_PRESENT
);
1306 device
->enabled_extensions
.extensions
[index
] = true;
1309 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
1310 const VkDeviceQueueCreateInfo
*queue_create
=
1311 &pCreateInfo
->pQueueCreateInfos
[i
];
1312 uint32_t qfi
= queue_create
->queueFamilyIndex
;
1313 device
->queues
[qfi
] = vk_alloc(
1314 &device
->vk
.alloc
, queue_create
->queueCount
* sizeof(struct tu_queue
),
1315 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1316 if (!device
->queues
[qfi
]) {
1317 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
1321 memset(device
->queues
[qfi
], 0,
1322 queue_create
->queueCount
* sizeof(struct tu_queue
));
1324 device
->queue_count
[qfi
] = queue_create
->queueCount
;
1326 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
1327 result
= tu_queue_init(device
, &device
->queues
[qfi
][q
], qfi
, q
,
1328 queue_create
->flags
);
1329 if (result
!= VK_SUCCESS
)
1334 device
->compiler
= ir3_compiler_create(NULL
, physical_device
->gpu_id
);
1335 if (!device
->compiler
)
1338 /* initial sizes, these will increase if there is overflow */
1339 device
->vsc_draw_strm_pitch
= 0x1000 + VSC_PAD
;
1340 device
->vsc_prim_strm_pitch
= 0x4000 + VSC_PAD
;
1342 STATIC_ASSERT(sizeof(border_color
) == sizeof(((struct tu6_global
*) 0)->border_color
));
1343 result
= tu_bo_init_new(device
, &device
->global_bo
, sizeof(struct tu6_global
));
1344 if (result
!= VK_SUCCESS
)
1345 goto fail_global_bo
;
1347 result
= tu_bo_map(device
, &device
->global_bo
);
1348 if (result
!= VK_SUCCESS
)
1349 goto fail_global_bo_map
;
1351 memcpy(device
->global_bo
.map
+ gb_offset(border_color
), border_color
, sizeof(border_color
));
1352 tu_init_clear_blit_shaders(device
->global_bo
.map
);
1354 VkPipelineCacheCreateInfo ci
;
1355 ci
.sType
= VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
;
1358 ci
.pInitialData
= NULL
;
1359 ci
.initialDataSize
= 0;
1362 tu_CreatePipelineCache(tu_device_to_handle(device
), &ci
, NULL
, &pc
);
1363 if (result
!= VK_SUCCESS
)
1364 goto fail_pipeline_cache
;
1366 device
->mem_cache
= tu_pipeline_cache_from_handle(pc
);
1368 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++)
1369 mtx_init(&device
->scratch_bos
[i
].construct_mtx
, mtx_plain
);
1371 mtx_init(&device
->vsc_pitch_mtx
, mtx_plain
);
1373 *pDevice
= tu_device_to_handle(device
);
1376 fail_pipeline_cache
:
1378 tu_bo_finish(device
, &device
->global_bo
);
1381 ralloc_free(device
->compiler
);
1384 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1385 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1386 tu_queue_finish(&device
->queues
[i
][q
]);
1387 if (device
->queue_count
[i
])
1388 vk_object_free(&device
->vk
, NULL
, device
->queues
[i
]);
1391 vk_free(&device
->vk
.alloc
, device
);
1396 tu_DestroyDevice(VkDevice _device
, const VkAllocationCallbacks
*pAllocator
)
1398 TU_FROM_HANDLE(tu_device
, device
, _device
);
1403 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1404 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1405 tu_queue_finish(&device
->queues
[i
][q
]);
1406 if (device
->queue_count
[i
])
1407 vk_object_free(&device
->vk
, NULL
, device
->queues
[i
]);
1410 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++) {
1411 if (device
->scratch_bos
[i
].initialized
)
1412 tu_bo_finish(device
, &device
->scratch_bos
[i
].bo
);
1415 ir3_compiler_destroy(device
->compiler
);
1417 VkPipelineCache pc
= tu_pipeline_cache_to_handle(device
->mem_cache
);
1418 tu_DestroyPipelineCache(tu_device_to_handle(device
), pc
, NULL
);
1420 vk_free(&device
->vk
.alloc
, device
);
1424 _tu_device_set_lost(struct tu_device
*device
,
1425 const char *file
, int line
,
1426 const char *msg
, ...)
1428 /* Set the flag indicating that waits should return in finite time even
1429 * after device loss.
1431 p_atomic_inc(&device
->_lost
);
1433 /* TODO: Report the log message through VkDebugReportCallbackEXT instead */
1434 fprintf(stderr
, "%s:%d: ", file
, line
);
1437 vfprintf(stderr
, msg
, ap
);
1440 if (env_var_as_boolean("TU_ABORT_ON_DEVICE_LOSS", false))
1443 return VK_ERROR_DEVICE_LOST
;
1447 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
)
1449 unsigned size_log2
= MAX2(util_logbase2_ceil64(size
), MIN_SCRATCH_BO_SIZE_LOG2
);
1450 unsigned index
= size_log2
- MIN_SCRATCH_BO_SIZE_LOG2
;
1451 assert(index
< ARRAY_SIZE(dev
->scratch_bos
));
1453 for (unsigned i
= index
; i
< ARRAY_SIZE(dev
->scratch_bos
); i
++) {
1454 if (p_atomic_read(&dev
->scratch_bos
[i
].initialized
)) {
1455 /* Fast path: just return the already-allocated BO. */
1456 *bo
= &dev
->scratch_bos
[i
].bo
;
1461 /* Slow path: actually allocate the BO. We take a lock because the process
1462 * of allocating it is slow, and we don't want to block the CPU while it
1465 mtx_lock(&dev
->scratch_bos
[index
].construct_mtx
);
1467 /* Another thread may have allocated it already while we were waiting on
1468 * the lock. We need to check this in order to avoid double-allocating.
1470 if (dev
->scratch_bos
[index
].initialized
) {
1471 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1472 *bo
= &dev
->scratch_bos
[index
].bo
;
1476 unsigned bo_size
= 1ull << size_log2
;
1477 VkResult result
= tu_bo_init_new(dev
, &dev
->scratch_bos
[index
].bo
, bo_size
);
1478 if (result
!= VK_SUCCESS
) {
1479 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1483 p_atomic_set(&dev
->scratch_bos
[index
].initialized
, true);
1485 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1487 *bo
= &dev
->scratch_bos
[index
].bo
;
1492 tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount
,
1493 VkLayerProperties
*pProperties
)
1495 *pPropertyCount
= 0;
1500 tu_EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice
,
1501 uint32_t *pPropertyCount
,
1502 VkLayerProperties
*pProperties
)
1504 *pPropertyCount
= 0;
1509 tu_GetDeviceQueue2(VkDevice _device
,
1510 const VkDeviceQueueInfo2
*pQueueInfo
,
1513 TU_FROM_HANDLE(tu_device
, device
, _device
);
1514 struct tu_queue
*queue
;
1517 &device
->queues
[pQueueInfo
->queueFamilyIndex
][pQueueInfo
->queueIndex
];
1518 if (pQueueInfo
->flags
!= queue
->flags
) {
1519 /* From the Vulkan 1.1.70 spec:
1521 * "The queue returned by vkGetDeviceQueue2 must have the same
1522 * flags value from this structure as that used at device
1523 * creation time in a VkDeviceQueueCreateInfo instance. If no
1524 * matching flags were specified at device creation time then
1525 * pQueue will return VK_NULL_HANDLE."
1527 *pQueue
= VK_NULL_HANDLE
;
1531 *pQueue
= tu_queue_to_handle(queue
);
1535 tu_GetDeviceQueue(VkDevice _device
,
1536 uint32_t queueFamilyIndex
,
1537 uint32_t queueIndex
,
1540 const VkDeviceQueueInfo2 info
=
1541 (VkDeviceQueueInfo2
) { .sType
= VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2
,
1542 .queueFamilyIndex
= queueFamilyIndex
,
1543 .queueIndex
= queueIndex
};
1545 tu_GetDeviceQueue2(_device
, &info
, pQueue
);
1549 tu_get_semaphore_syncobjs(const VkSemaphore
*sems
,
1552 struct drm_msm_gem_submit_syncobj
**out
,
1553 uint32_t *out_count
)
1555 uint32_t syncobj_count
= 0;
1556 struct drm_msm_gem_submit_syncobj
*syncobjs
;
1558 for (uint32_t i
= 0; i
< sem_count
; ++i
) {
1559 TU_FROM_HANDLE(tu_semaphore
, sem
, sems
[i
]);
1561 struct tu_semaphore_part
*part
=
1562 sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
?
1563 &sem
->temporary
: &sem
->permanent
;
1565 if (part
->kind
== TU_SEMAPHORE_SYNCOBJ
)
1570 *out_count
= syncobj_count
;
1574 *out
= syncobjs
= calloc(syncobj_count
, sizeof (*syncobjs
));
1576 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1578 for (uint32_t i
= 0, j
= 0; i
< sem_count
; ++i
) {
1579 TU_FROM_HANDLE(tu_semaphore
, sem
, sems
[i
]);
1581 struct tu_semaphore_part
*part
=
1582 sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
?
1583 &sem
->temporary
: &sem
->permanent
;
1585 if (part
->kind
== TU_SEMAPHORE_SYNCOBJ
) {
1586 syncobjs
[j
].handle
= part
->syncobj
;
1587 syncobjs
[j
].flags
= wait
? MSM_SUBMIT_SYNCOBJ_RESET
: 0;
1597 tu_semaphores_remove_temp(struct tu_device
*device
,
1598 const VkSemaphore
*sems
,
1601 for (uint32_t i
= 0; i
< sem_count
; ++i
) {
1602 TU_FROM_HANDLE(tu_semaphore
, sem
, sems
[i
]);
1603 tu_semaphore_remove_temp(device
, sem
);
1608 tu_QueueSubmit(VkQueue _queue
,
1609 uint32_t submitCount
,
1610 const VkSubmitInfo
*pSubmits
,
1613 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1616 for (uint32_t i
= 0; i
< submitCount
; ++i
) {
1617 const VkSubmitInfo
*submit
= pSubmits
+ i
;
1618 const bool last_submit
= (i
== submitCount
- 1);
1619 struct drm_msm_gem_submit_syncobj
*in_syncobjs
= NULL
, *out_syncobjs
= NULL
;
1620 uint32_t nr_in_syncobjs
, nr_out_syncobjs
;
1621 struct tu_bo_list bo_list
;
1622 tu_bo_list_init(&bo_list
);
1624 result
= tu_get_semaphore_syncobjs(pSubmits
[i
].pWaitSemaphores
,
1625 pSubmits
[i
].waitSemaphoreCount
,
1626 false, &in_syncobjs
, &nr_in_syncobjs
);
1627 if (result
!= VK_SUCCESS
) {
1628 return tu_device_set_lost(queue
->device
,
1629 "failed to allocate space for semaphore submission\n");
1632 result
= tu_get_semaphore_syncobjs(pSubmits
[i
].pSignalSemaphores
,
1633 pSubmits
[i
].signalSemaphoreCount
,
1634 false, &out_syncobjs
, &nr_out_syncobjs
);
1635 if (result
!= VK_SUCCESS
) {
1637 return tu_device_set_lost(queue
->device
,
1638 "failed to allocate space for semaphore submission\n");
1641 uint32_t entry_count
= 0;
1642 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1643 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1644 entry_count
+= cmdbuf
->cs
.entry_count
;
1647 struct drm_msm_gem_submit_cmd cmds
[entry_count
];
1648 uint32_t entry_idx
= 0;
1649 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1650 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1651 struct tu_cs
*cs
= &cmdbuf
->cs
;
1652 for (unsigned i
= 0; i
< cs
->entry_count
; ++i
, ++entry_idx
) {
1653 cmds
[entry_idx
].type
= MSM_SUBMIT_CMD_BUF
;
1654 cmds
[entry_idx
].submit_idx
=
1655 tu_bo_list_add(&bo_list
, cs
->entries
[i
].bo
,
1656 MSM_SUBMIT_BO_READ
| MSM_SUBMIT_BO_DUMP
);
1657 cmds
[entry_idx
].submit_offset
= cs
->entries
[i
].offset
;
1658 cmds
[entry_idx
].size
= cs
->entries
[i
].size
;
1659 cmds
[entry_idx
].pad
= 0;
1660 cmds
[entry_idx
].nr_relocs
= 0;
1661 cmds
[entry_idx
].relocs
= 0;
1664 tu_bo_list_merge(&bo_list
, &cmdbuf
->bo_list
);
1667 uint32_t flags
= MSM_PIPE_3D0
;
1668 if (nr_in_syncobjs
) {
1669 flags
|= MSM_SUBMIT_SYNCOBJ_IN
;
1671 if (nr_out_syncobjs
) {
1672 flags
|= MSM_SUBMIT_SYNCOBJ_OUT
;
1676 flags
|= MSM_SUBMIT_FENCE_FD_OUT
;
1679 struct drm_msm_gem_submit req
= {
1681 .queueid
= queue
->msm_queue_id
,
1682 .bos
= (uint64_t)(uintptr_t) bo_list
.bo_infos
,
1683 .nr_bos
= bo_list
.count
,
1684 .cmds
= (uint64_t)(uintptr_t)cmds
,
1685 .nr_cmds
= entry_count
,
1686 .in_syncobjs
= (uint64_t)(uintptr_t)in_syncobjs
,
1687 .out_syncobjs
= (uint64_t)(uintptr_t)out_syncobjs
,
1688 .nr_in_syncobjs
= nr_in_syncobjs
,
1689 .nr_out_syncobjs
= nr_out_syncobjs
,
1690 .syncobj_stride
= sizeof(struct drm_msm_gem_submit_syncobj
),
1693 int ret
= drmCommandWriteRead(queue
->device
->physical_device
->local_fd
,
1699 return tu_device_set_lost(queue
->device
, "submit failed: %s\n",
1703 tu_bo_list_destroy(&bo_list
);
1707 tu_semaphores_remove_temp(queue
->device
, pSubmits
[i
].pWaitSemaphores
,
1708 pSubmits
[i
].waitSemaphoreCount
);
1710 /* no need to merge fences as queue execution is serialized */
1711 tu_fence_update_fd(&queue
->submit_fence
, req
.fence_fd
);
1712 } else if (last_submit
) {
1713 close(req
.fence_fd
);
1717 if (_fence
!= VK_NULL_HANDLE
) {
1718 TU_FROM_HANDLE(tu_fence
, fence
, _fence
);
1719 tu_fence_copy(fence
, &queue
->submit_fence
);
1726 tu_QueueWaitIdle(VkQueue _queue
)
1728 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1730 if (tu_device_is_lost(queue
->device
))
1731 return VK_ERROR_DEVICE_LOST
;
1733 tu_fence_wait_idle(&queue
->submit_fence
);
1739 tu_DeviceWaitIdle(VkDevice _device
)
1741 TU_FROM_HANDLE(tu_device
, device
, _device
);
1743 if (tu_device_is_lost(device
))
1744 return VK_ERROR_DEVICE_LOST
;
1746 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1747 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
1748 tu_QueueWaitIdle(tu_queue_to_handle(&device
->queues
[i
][q
]));
1755 tu_EnumerateInstanceExtensionProperties(const char *pLayerName
,
1756 uint32_t *pPropertyCount
,
1757 VkExtensionProperties
*pProperties
)
1759 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1761 /* We spport no lyaers */
1763 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1765 for (int i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; i
++) {
1766 if (tu_instance_extensions_supported
.extensions
[i
]) {
1767 vk_outarray_append(&out
, prop
) { *prop
= tu_instance_extensions
[i
]; }
1771 return vk_outarray_status(&out
);
1775 tu_EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice
,
1776 const char *pLayerName
,
1777 uint32_t *pPropertyCount
,
1778 VkExtensionProperties
*pProperties
)
1780 /* We spport no lyaers */
1781 TU_FROM_HANDLE(tu_physical_device
, device
, physicalDevice
);
1782 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1784 /* We spport no lyaers */
1786 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1788 for (int i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; i
++) {
1789 if (device
->supported_extensions
.extensions
[i
]) {
1790 vk_outarray_append(&out
, prop
) { *prop
= tu_device_extensions
[i
]; }
1794 return vk_outarray_status(&out
);
1798 tu_GetInstanceProcAddr(VkInstance _instance
, const char *pName
)
1800 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
1802 return tu_lookup_entrypoint_checked(
1803 pName
, instance
? instance
->api_version
: 0,
1804 instance
? &instance
->enabled_extensions
: NULL
, NULL
);
1807 /* The loader wants us to expose a second GetInstanceProcAddr function
1808 * to work around certain LD_PRELOAD issues seen in apps.
1811 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1812 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
);
1815 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1816 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
)
1818 return tu_GetInstanceProcAddr(instance
, pName
);
1822 tu_GetDeviceProcAddr(VkDevice _device
, const char *pName
)
1824 TU_FROM_HANDLE(tu_device
, device
, _device
);
1826 return tu_lookup_entrypoint_checked(pName
, device
->instance
->api_version
,
1827 &device
->instance
->enabled_extensions
,
1828 &device
->enabled_extensions
);
1832 tu_alloc_memory(struct tu_device
*device
,
1833 const VkMemoryAllocateInfo
*pAllocateInfo
,
1834 const VkAllocationCallbacks
*pAllocator
,
1835 VkDeviceMemory
*pMem
)
1837 struct tu_device_memory
*mem
;
1840 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1842 if (pAllocateInfo
->allocationSize
== 0) {
1843 /* Apparently, this is allowed */
1844 *pMem
= VK_NULL_HANDLE
;
1848 mem
= vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*mem
),
1849 VK_OBJECT_TYPE_DEVICE_MEMORY
);
1851 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1853 const VkImportMemoryFdInfoKHR
*fd_info
=
1854 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
1855 if (fd_info
&& !fd_info
->handleType
)
1859 assert(fd_info
->handleType
==
1860 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
1861 fd_info
->handleType
==
1862 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
1865 * TODO Importing the same fd twice gives us the same handle without
1866 * reference counting. We need to maintain a per-instance handle-to-bo
1867 * table and add reference count to tu_bo.
1869 result
= tu_bo_init_dmabuf(device
, &mem
->bo
,
1870 pAllocateInfo
->allocationSize
, fd_info
->fd
);
1871 if (result
== VK_SUCCESS
) {
1872 /* take ownership and close the fd */
1877 tu_bo_init_new(device
, &mem
->bo
, pAllocateInfo
->allocationSize
);
1880 if (result
!= VK_SUCCESS
) {
1881 vk_object_free(&device
->vk
, pAllocator
, mem
);
1885 mem
->size
= pAllocateInfo
->allocationSize
;
1886 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1889 mem
->user_ptr
= NULL
;
1891 *pMem
= tu_device_memory_to_handle(mem
);
1897 tu_AllocateMemory(VkDevice _device
,
1898 const VkMemoryAllocateInfo
*pAllocateInfo
,
1899 const VkAllocationCallbacks
*pAllocator
,
1900 VkDeviceMemory
*pMem
)
1902 TU_FROM_HANDLE(tu_device
, device
, _device
);
1903 return tu_alloc_memory(device
, pAllocateInfo
, pAllocator
, pMem
);
1907 tu_FreeMemory(VkDevice _device
,
1908 VkDeviceMemory _mem
,
1909 const VkAllocationCallbacks
*pAllocator
)
1911 TU_FROM_HANDLE(tu_device
, device
, _device
);
1912 TU_FROM_HANDLE(tu_device_memory
, mem
, _mem
);
1917 tu_bo_finish(device
, &mem
->bo
);
1918 vk_object_free(&device
->vk
, pAllocator
, mem
);
1922 tu_MapMemory(VkDevice _device
,
1923 VkDeviceMemory _memory
,
1924 VkDeviceSize offset
,
1926 VkMemoryMapFlags flags
,
1929 TU_FROM_HANDLE(tu_device
, device
, _device
);
1930 TU_FROM_HANDLE(tu_device_memory
, mem
, _memory
);
1938 if (mem
->user_ptr
) {
1939 *ppData
= mem
->user_ptr
;
1940 } else if (!mem
->map
) {
1941 result
= tu_bo_map(device
, &mem
->bo
);
1942 if (result
!= VK_SUCCESS
)
1944 *ppData
= mem
->map
= mem
->bo
.map
;
1953 return vk_error(device
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
1957 tu_UnmapMemory(VkDevice _device
, VkDeviceMemory _memory
)
1959 /* I do not see any unmapping done by the freedreno Gallium driver. */
1963 tu_FlushMappedMemoryRanges(VkDevice _device
,
1964 uint32_t memoryRangeCount
,
1965 const VkMappedMemoryRange
*pMemoryRanges
)
1971 tu_InvalidateMappedMemoryRanges(VkDevice _device
,
1972 uint32_t memoryRangeCount
,
1973 const VkMappedMemoryRange
*pMemoryRanges
)
1979 tu_GetBufferMemoryRequirements(VkDevice _device
,
1981 VkMemoryRequirements
*pMemoryRequirements
)
1983 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
1985 pMemoryRequirements
->memoryTypeBits
= 1;
1986 pMemoryRequirements
->alignment
= 64;
1987 pMemoryRequirements
->size
=
1988 align64(buffer
->size
, pMemoryRequirements
->alignment
);
1992 tu_GetBufferMemoryRequirements2(
1994 const VkBufferMemoryRequirementsInfo2
*pInfo
,
1995 VkMemoryRequirements2
*pMemoryRequirements
)
1997 tu_GetBufferMemoryRequirements(device
, pInfo
->buffer
,
1998 &pMemoryRequirements
->memoryRequirements
);
2002 tu_GetImageMemoryRequirements(VkDevice _device
,
2004 VkMemoryRequirements
*pMemoryRequirements
)
2006 TU_FROM_HANDLE(tu_image
, image
, _image
);
2008 pMemoryRequirements
->memoryTypeBits
= 1;
2009 pMemoryRequirements
->size
= image
->total_size
;
2010 pMemoryRequirements
->alignment
= image
->layout
[0].base_align
;
2014 tu_GetImageMemoryRequirements2(VkDevice device
,
2015 const VkImageMemoryRequirementsInfo2
*pInfo
,
2016 VkMemoryRequirements2
*pMemoryRequirements
)
2018 tu_GetImageMemoryRequirements(device
, pInfo
->image
,
2019 &pMemoryRequirements
->memoryRequirements
);
2023 tu_GetImageSparseMemoryRequirements(
2026 uint32_t *pSparseMemoryRequirementCount
,
2027 VkSparseImageMemoryRequirements
*pSparseMemoryRequirements
)
2033 tu_GetImageSparseMemoryRequirements2(
2035 const VkImageSparseMemoryRequirementsInfo2
*pInfo
,
2036 uint32_t *pSparseMemoryRequirementCount
,
2037 VkSparseImageMemoryRequirements2
*pSparseMemoryRequirements
)
2043 tu_GetDeviceMemoryCommitment(VkDevice device
,
2044 VkDeviceMemory memory
,
2045 VkDeviceSize
*pCommittedMemoryInBytes
)
2047 *pCommittedMemoryInBytes
= 0;
2051 tu_BindBufferMemory2(VkDevice device
,
2052 uint32_t bindInfoCount
,
2053 const VkBindBufferMemoryInfo
*pBindInfos
)
2055 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
2056 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
2057 TU_FROM_HANDLE(tu_buffer
, buffer
, pBindInfos
[i
].buffer
);
2060 buffer
->bo
= &mem
->bo
;
2061 buffer
->bo_offset
= pBindInfos
[i
].memoryOffset
;
2070 tu_BindBufferMemory(VkDevice device
,
2072 VkDeviceMemory memory
,
2073 VkDeviceSize memoryOffset
)
2075 const VkBindBufferMemoryInfo info
= {
2076 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
2079 .memoryOffset
= memoryOffset
2082 return tu_BindBufferMemory2(device
, 1, &info
);
2086 tu_BindImageMemory2(VkDevice device
,
2087 uint32_t bindInfoCount
,
2088 const VkBindImageMemoryInfo
*pBindInfos
)
2090 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
2091 TU_FROM_HANDLE(tu_image
, image
, pBindInfos
[i
].image
);
2092 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
2095 image
->bo
= &mem
->bo
;
2096 image
->bo_offset
= pBindInfos
[i
].memoryOffset
;
2099 image
->bo_offset
= 0;
2107 tu_BindImageMemory(VkDevice device
,
2109 VkDeviceMemory memory
,
2110 VkDeviceSize memoryOffset
)
2112 const VkBindImageMemoryInfo info
= {
2113 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
2116 .memoryOffset
= memoryOffset
2119 return tu_BindImageMemory2(device
, 1, &info
);
2123 tu_QueueBindSparse(VkQueue _queue
,
2124 uint32_t bindInfoCount
,
2125 const VkBindSparseInfo
*pBindInfo
,
2131 // Queue semaphore functions
2135 tu_semaphore_part_destroy(struct tu_device
*device
,
2136 struct tu_semaphore_part
*part
)
2138 switch(part
->kind
) {
2139 case TU_SEMAPHORE_NONE
:
2141 case TU_SEMAPHORE_SYNCOBJ
:
2142 drmSyncobjDestroy(device
->physical_device
->local_fd
, part
->syncobj
);
2145 part
->kind
= TU_SEMAPHORE_NONE
;
2149 tu_semaphore_remove_temp(struct tu_device
*device
,
2150 struct tu_semaphore
*sem
)
2152 if (sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
) {
2153 tu_semaphore_part_destroy(device
, &sem
->temporary
);
2158 tu_CreateSemaphore(VkDevice _device
,
2159 const VkSemaphoreCreateInfo
*pCreateInfo
,
2160 const VkAllocationCallbacks
*pAllocator
,
2161 VkSemaphore
*pSemaphore
)
2163 TU_FROM_HANDLE(tu_device
, device
, _device
);
2165 struct tu_semaphore
*sem
=
2166 vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*sem
),
2167 VK_OBJECT_TYPE_SEMAPHORE
);
2169 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2171 const VkExportSemaphoreCreateInfo
*export
=
2172 vk_find_struct_const(pCreateInfo
->pNext
, EXPORT_SEMAPHORE_CREATE_INFO
);
2173 VkExternalSemaphoreHandleTypeFlags handleTypes
=
2174 export
? export
->handleTypes
: 0;
2176 sem
->permanent
.kind
= TU_SEMAPHORE_NONE
;
2177 sem
->temporary
.kind
= TU_SEMAPHORE_NONE
;
2180 if (drmSyncobjCreate(device
->physical_device
->local_fd
, 0, &sem
->permanent
.syncobj
) < 0) {
2181 vk_free2(&device
->vk
.alloc
, pAllocator
, sem
);
2182 return VK_ERROR_OUT_OF_HOST_MEMORY
;
2184 sem
->permanent
.kind
= TU_SEMAPHORE_SYNCOBJ
;
2186 *pSemaphore
= tu_semaphore_to_handle(sem
);
2191 tu_DestroySemaphore(VkDevice _device
,
2192 VkSemaphore _semaphore
,
2193 const VkAllocationCallbacks
*pAllocator
)
2195 TU_FROM_HANDLE(tu_device
, device
, _device
);
2196 TU_FROM_HANDLE(tu_semaphore
, sem
, _semaphore
);
2200 tu_semaphore_part_destroy(device
, &sem
->permanent
);
2201 tu_semaphore_part_destroy(device
, &sem
->temporary
);
2203 vk_object_free(&device
->vk
, pAllocator
, sem
);
2207 tu_CreateEvent(VkDevice _device
,
2208 const VkEventCreateInfo
*pCreateInfo
,
2209 const VkAllocationCallbacks
*pAllocator
,
2212 TU_FROM_HANDLE(tu_device
, device
, _device
);
2214 struct tu_event
*event
=
2215 vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*event
),
2216 VK_OBJECT_TYPE_EVENT
);
2218 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2220 VkResult result
= tu_bo_init_new(device
, &event
->bo
, 0x1000);
2221 if (result
!= VK_SUCCESS
)
2224 result
= tu_bo_map(device
, &event
->bo
);
2225 if (result
!= VK_SUCCESS
)
2228 *pEvent
= tu_event_to_handle(event
);
2233 tu_bo_finish(device
, &event
->bo
);
2235 vk_object_free(&device
->vk
, pAllocator
, event
);
2236 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2240 tu_DestroyEvent(VkDevice _device
,
2242 const VkAllocationCallbacks
*pAllocator
)
2244 TU_FROM_HANDLE(tu_device
, device
, _device
);
2245 TU_FROM_HANDLE(tu_event
, event
, _event
);
2250 tu_bo_finish(device
, &event
->bo
);
2251 vk_object_free(&device
->vk
, pAllocator
, event
);
2255 tu_GetEventStatus(VkDevice _device
, VkEvent _event
)
2257 TU_FROM_HANDLE(tu_event
, event
, _event
);
2259 if (*(uint64_t*) event
->bo
.map
== 1)
2260 return VK_EVENT_SET
;
2261 return VK_EVENT_RESET
;
2265 tu_SetEvent(VkDevice _device
, VkEvent _event
)
2267 TU_FROM_HANDLE(tu_event
, event
, _event
);
2268 *(uint64_t*) event
->bo
.map
= 1;
2274 tu_ResetEvent(VkDevice _device
, VkEvent _event
)
2276 TU_FROM_HANDLE(tu_event
, event
, _event
);
2277 *(uint64_t*) event
->bo
.map
= 0;
2283 tu_CreateBuffer(VkDevice _device
,
2284 const VkBufferCreateInfo
*pCreateInfo
,
2285 const VkAllocationCallbacks
*pAllocator
,
2288 TU_FROM_HANDLE(tu_device
, device
, _device
);
2289 struct tu_buffer
*buffer
;
2291 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
2293 buffer
= vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*buffer
),
2294 VK_OBJECT_TYPE_BUFFER
);
2296 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2298 buffer
->size
= pCreateInfo
->size
;
2299 buffer
->usage
= pCreateInfo
->usage
;
2300 buffer
->flags
= pCreateInfo
->flags
;
2302 *pBuffer
= tu_buffer_to_handle(buffer
);
2308 tu_DestroyBuffer(VkDevice _device
,
2310 const VkAllocationCallbacks
*pAllocator
)
2312 TU_FROM_HANDLE(tu_device
, device
, _device
);
2313 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
2318 vk_object_free(&device
->vk
, pAllocator
, buffer
);
2322 tu_CreateFramebuffer(VkDevice _device
,
2323 const VkFramebufferCreateInfo
*pCreateInfo
,
2324 const VkAllocationCallbacks
*pAllocator
,
2325 VkFramebuffer
*pFramebuffer
)
2327 TU_FROM_HANDLE(tu_device
, device
, _device
);
2328 TU_FROM_HANDLE(tu_render_pass
, pass
, pCreateInfo
->renderPass
);
2329 struct tu_framebuffer
*framebuffer
;
2331 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
2333 size_t size
= sizeof(*framebuffer
) + sizeof(struct tu_attachment_info
) *
2334 pCreateInfo
->attachmentCount
;
2335 framebuffer
= vk_object_alloc(&device
->vk
, pAllocator
, size
,
2336 VK_OBJECT_TYPE_FRAMEBUFFER
);
2337 if (framebuffer
== NULL
)
2338 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2340 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
2341 framebuffer
->width
= pCreateInfo
->width
;
2342 framebuffer
->height
= pCreateInfo
->height
;
2343 framebuffer
->layers
= pCreateInfo
->layers
;
2344 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
2345 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
2346 struct tu_image_view
*iview
= tu_image_view_from_handle(_iview
);
2347 framebuffer
->attachments
[i
].attachment
= iview
;
2350 tu_framebuffer_tiling_config(framebuffer
, device
, pass
);
2352 *pFramebuffer
= tu_framebuffer_to_handle(framebuffer
);
2357 tu_DestroyFramebuffer(VkDevice _device
,
2359 const VkAllocationCallbacks
*pAllocator
)
2361 TU_FROM_HANDLE(tu_device
, device
, _device
);
2362 TU_FROM_HANDLE(tu_framebuffer
, fb
, _fb
);
2367 vk_object_free(&device
->vk
, pAllocator
, fb
);
2371 tu_init_sampler(struct tu_device
*device
,
2372 struct tu_sampler
*sampler
,
2373 const VkSamplerCreateInfo
*pCreateInfo
)
2375 const struct VkSamplerReductionModeCreateInfo
*reduction
=
2376 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_REDUCTION_MODE_CREATE_INFO
);
2377 const struct VkSamplerYcbcrConversionInfo
*ycbcr_conversion
=
2378 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_YCBCR_CONVERSION_INFO
);
2380 unsigned aniso
= pCreateInfo
->anisotropyEnable
?
2381 util_last_bit(MIN2((uint32_t)pCreateInfo
->maxAnisotropy
>> 1, 8)) : 0;
2382 bool miplinear
= (pCreateInfo
->mipmapMode
== VK_SAMPLER_MIPMAP_MODE_LINEAR
);
2383 float min_lod
= CLAMP(pCreateInfo
->minLod
, 0.0f
, 4095.0f
/ 256.0f
);
2384 float max_lod
= CLAMP(pCreateInfo
->maxLod
, 0.0f
, 4095.0f
/ 256.0f
);
2386 sampler
->descriptor
[0] =
2387 COND(miplinear
, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR
) |
2388 A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo
->magFilter
, aniso
)) |
2389 A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo
->minFilter
, aniso
)) |
2390 A6XX_TEX_SAMP_0_ANISO(aniso
) |
2391 A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo
->addressModeU
)) |
2392 A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo
->addressModeV
)) |
2393 A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo
->addressModeW
)) |
2394 A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo
->mipLodBias
);
2395 sampler
->descriptor
[1] =
2396 /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
2397 COND(pCreateInfo
->unnormalizedCoordinates
, A6XX_TEX_SAMP_1_UNNORM_COORDS
) |
2398 A6XX_TEX_SAMP_1_MIN_LOD(min_lod
) |
2399 A6XX_TEX_SAMP_1_MAX_LOD(max_lod
) |
2400 COND(pCreateInfo
->compareEnable
,
2401 A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo
->compareOp
)));
2402 /* This is an offset into the border_color BO, which we fill with all the
2403 * possible Vulkan border colors in the correct order, so we can just use
2404 * the Vulkan enum with no translation necessary.
2406 sampler
->descriptor
[2] =
2407 A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo
->borderColor
*
2408 sizeof(struct bcolor_entry
));
2409 sampler
->descriptor
[3] = 0;
2412 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(
2413 tu6_reduction_mode(reduction
->reductionMode
));
2416 sampler
->ycbcr_sampler
= ycbcr_conversion
?
2417 tu_sampler_ycbcr_conversion_from_handle(ycbcr_conversion
->conversion
) : NULL
;
2419 if (sampler
->ycbcr_sampler
&&
2420 sampler
->ycbcr_sampler
->chroma_filter
== VK_FILTER_LINEAR
) {
2421 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_CHROMA_LINEAR
;
2425 * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
2430 tu_CreateSampler(VkDevice _device
,
2431 const VkSamplerCreateInfo
*pCreateInfo
,
2432 const VkAllocationCallbacks
*pAllocator
,
2433 VkSampler
*pSampler
)
2435 TU_FROM_HANDLE(tu_device
, device
, _device
);
2436 struct tu_sampler
*sampler
;
2438 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
2440 sampler
= vk_object_alloc(&device
->vk
, pAllocator
, sizeof(*sampler
),
2441 VK_OBJECT_TYPE_SAMPLER
);
2443 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2445 tu_init_sampler(device
, sampler
, pCreateInfo
);
2446 *pSampler
= tu_sampler_to_handle(sampler
);
2452 tu_DestroySampler(VkDevice _device
,
2454 const VkAllocationCallbacks
*pAllocator
)
2456 TU_FROM_HANDLE(tu_device
, device
, _device
);
2457 TU_FROM_HANDLE(tu_sampler
, sampler
, _sampler
);
2462 vk_object_free(&device
->vk
, pAllocator
, sampler
);
2465 /* vk_icd.h does not declare this function, so we declare it here to
2466 * suppress Wmissing-prototypes.
2468 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2469 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
2471 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2472 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
2474 /* For the full details on loader interface versioning, see
2475 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2476 * What follows is a condensed summary, to help you navigate the large and
2477 * confusing official doc.
2479 * - Loader interface v0 is incompatible with later versions. We don't
2482 * - In loader interface v1:
2483 * - The first ICD entrypoint called by the loader is
2484 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2486 * - The ICD must statically expose no other Vulkan symbol unless it
2487 * is linked with -Bsymbolic.
2488 * - Each dispatchable Vulkan handle created by the ICD must be
2489 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2490 * ICD must initialize VK_LOADER_DATA.loadMagic to
2492 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2493 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2494 * such loader-managed surfaces.
2496 * - Loader interface v2 differs from v1 in:
2497 * - The first ICD entrypoint called by the loader is
2498 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2499 * statically expose this entrypoint.
2501 * - Loader interface v3 differs from v2 in:
2502 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2503 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2504 * because the loader no longer does so.
2506 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);
2511 tu_GetMemoryFdKHR(VkDevice _device
,
2512 const VkMemoryGetFdInfoKHR
*pGetFdInfo
,
2515 TU_FROM_HANDLE(tu_device
, device
, _device
);
2516 TU_FROM_HANDLE(tu_device_memory
, memory
, pGetFdInfo
->memory
);
2518 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
2520 /* At the moment, we support only the below handle types. */
2521 assert(pGetFdInfo
->handleType
==
2522 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
2523 pGetFdInfo
->handleType
==
2524 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2526 int prime_fd
= tu_bo_export_dmabuf(device
, &memory
->bo
);
2528 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2535 tu_GetMemoryFdPropertiesKHR(VkDevice _device
,
2536 VkExternalMemoryHandleTypeFlagBits handleType
,
2538 VkMemoryFdPropertiesKHR
*pMemoryFdProperties
)
2540 assert(handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2541 pMemoryFdProperties
->memoryTypeBits
= 1;
2546 tu_ImportFenceFdKHR(VkDevice _device
,
2547 const VkImportFenceFdInfoKHR
*pImportFenceFdInfo
)
2555 tu_GetFenceFdKHR(VkDevice _device
,
2556 const VkFenceGetFdInfoKHR
*pGetFdInfo
,
2565 tu_ImportSemaphoreFdKHR(VkDevice _device
,
2566 const VkImportSemaphoreFdInfoKHR
*pImportSemaphoreFdInfo
)
2568 TU_FROM_HANDLE(tu_device
, device
, _device
);
2569 TU_FROM_HANDLE(tu_semaphore
, sem
, pImportSemaphoreFdInfo
->semaphore
);
2571 struct tu_semaphore_part
*dst
= NULL
;
2573 if (pImportSemaphoreFdInfo
->flags
& VK_SEMAPHORE_IMPORT_TEMPORARY_BIT
) {
2574 dst
= &sem
->temporary
;
2576 dst
= &sem
->permanent
;
2579 uint32_t syncobj
= dst
->kind
== TU_SEMAPHORE_SYNCOBJ
? dst
->syncobj
: 0;
2581 switch(pImportSemaphoreFdInfo
->handleType
) {
2582 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
: {
2583 uint32_t old_syncobj
= syncobj
;
2584 ret
= drmSyncobjFDToHandle(device
->physical_device
->local_fd
, pImportSemaphoreFdInfo
->fd
, &syncobj
);
2586 close(pImportSemaphoreFdInfo
->fd
);
2588 drmSyncobjDestroy(device
->physical_device
->local_fd
, old_syncobj
);
2592 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
: {
2594 ret
= drmSyncobjCreate(device
->physical_device
->local_fd
, 0, &syncobj
);
2598 if (pImportSemaphoreFdInfo
->fd
== -1) {
2599 ret
= drmSyncobjSignal(device
->physical_device
->local_fd
, &syncobj
, 1);
2601 ret
= drmSyncobjImportSyncFile(device
->physical_device
->local_fd
, syncobj
, pImportSemaphoreFdInfo
->fd
);
2604 close(pImportSemaphoreFdInfo
->fd
);
2608 unreachable("Unhandled semaphore handle type");
2612 return VK_ERROR_INVALID_EXTERNAL_HANDLE
;
2614 dst
->syncobj
= syncobj
;
2615 dst
->kind
= TU_SEMAPHORE_SYNCOBJ
;
2621 tu_GetSemaphoreFdKHR(VkDevice _device
,
2622 const VkSemaphoreGetFdInfoKHR
*pGetFdInfo
,
2625 TU_FROM_HANDLE(tu_device
, device
, _device
);
2626 TU_FROM_HANDLE(tu_semaphore
, sem
, pGetFdInfo
->semaphore
);
2628 uint32_t syncobj_handle
;
2630 if (sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
) {
2631 assert(sem
->temporary
.kind
== TU_SEMAPHORE_SYNCOBJ
);
2632 syncobj_handle
= sem
->temporary
.syncobj
;
2634 assert(sem
->permanent
.kind
== TU_SEMAPHORE_SYNCOBJ
);
2635 syncobj_handle
= sem
->permanent
.syncobj
;
2638 switch(pGetFdInfo
->handleType
) {
2639 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
:
2640 ret
= drmSyncobjHandleToFD(device
->physical_device
->local_fd
, syncobj_handle
, pFd
);
2642 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
:
2643 ret
= drmSyncobjExportSyncFile(device
->physical_device
->local_fd
, syncobj_handle
, pFd
);
2645 if (sem
->temporary
.kind
!= TU_SEMAPHORE_NONE
) {
2646 tu_semaphore_part_destroy(device
, &sem
->temporary
);
2648 drmSyncobjReset(device
->physical_device
->local_fd
, &syncobj_handle
, 1);
2653 unreachable("Unhandled semaphore handle type");
2657 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE
);
2662 static bool tu_has_syncobj(struct tu_physical_device
*pdev
)
2665 if (drmGetCap(pdev
->local_fd
, DRM_CAP_SYNCOBJ
, &value
))
2667 return value
&& pdev
->msm_major_version
== 1 && pdev
->msm_minor_version
>= 6;
2671 tu_GetPhysicalDeviceExternalSemaphoreProperties(
2672 VkPhysicalDevice physicalDevice
,
2673 const VkPhysicalDeviceExternalSemaphoreInfo
*pExternalSemaphoreInfo
,
2674 VkExternalSemaphoreProperties
*pExternalSemaphoreProperties
)
2676 TU_FROM_HANDLE(tu_physical_device
, pdev
, physicalDevice
);
2678 if (tu_has_syncobj(pdev
) &&
2679 (pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
||
2680 pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
)) {
2681 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
| VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
;
2682 pExternalSemaphoreProperties
->compatibleHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT
| VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT
;
2683 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT
|
2684 VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT
;
2686 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= 0;
2687 pExternalSemaphoreProperties
->compatibleHandleTypes
= 0;
2688 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= 0;
2693 tu_GetPhysicalDeviceExternalFenceProperties(
2694 VkPhysicalDevice physicalDevice
,
2695 const VkPhysicalDeviceExternalFenceInfo
*pExternalFenceInfo
,
2696 VkExternalFenceProperties
*pExternalFenceProperties
)
2698 pExternalFenceProperties
->exportFromImportedHandleTypes
= 0;
2699 pExternalFenceProperties
->compatibleHandleTypes
= 0;
2700 pExternalFenceProperties
->externalFenceFeatures
= 0;
2704 tu_CreateDebugReportCallbackEXT(
2705 VkInstance _instance
,
2706 const VkDebugReportCallbackCreateInfoEXT
*pCreateInfo
,
2707 const VkAllocationCallbacks
*pAllocator
,
2708 VkDebugReportCallbackEXT
*pCallback
)
2710 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2711 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
2712 pCreateInfo
, pAllocator
,
2713 &instance
->alloc
, pCallback
);
2717 tu_DestroyDebugReportCallbackEXT(VkInstance _instance
,
2718 VkDebugReportCallbackEXT _callback
,
2719 const VkAllocationCallbacks
*pAllocator
)
2721 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2722 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
2723 _callback
, pAllocator
, &instance
->alloc
);
2727 tu_DebugReportMessageEXT(VkInstance _instance
,
2728 VkDebugReportFlagsEXT flags
,
2729 VkDebugReportObjectTypeEXT objectType
,
2732 int32_t messageCode
,
2733 const char *pLayerPrefix
,
2734 const char *pMessage
)
2736 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2737 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
2738 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
2742 tu_GetDeviceGroupPeerMemoryFeatures(
2745 uint32_t localDeviceIndex
,
2746 uint32_t remoteDeviceIndex
,
2747 VkPeerMemoryFeatureFlags
*pPeerMemoryFeatures
)
2749 assert(localDeviceIndex
== remoteDeviceIndex
);
2751 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
2752 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
2753 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
2754 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
2757 void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
2758 VkPhysicalDevice physicalDevice
,
2759 VkSampleCountFlagBits samples
,
2760 VkMultisamplePropertiesEXT
* pMultisampleProperties
)
2762 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
2764 if (samples
<= VK_SAMPLE_COUNT_4_BIT
&& pdevice
->supported_extensions
.EXT_sample_locations
)
2765 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 1, 1 };
2767 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 0, 0 };
2772 tu_CreatePrivateDataSlotEXT(VkDevice _device
,
2773 const VkPrivateDataSlotCreateInfoEXT
* pCreateInfo
,
2774 const VkAllocationCallbacks
* pAllocator
,
2775 VkPrivateDataSlotEXT
* pPrivateDataSlot
)
2777 TU_FROM_HANDLE(tu_device
, device
, _device
);
2778 return vk_private_data_slot_create(&device
->vk
,
2785 tu_DestroyPrivateDataSlotEXT(VkDevice _device
,
2786 VkPrivateDataSlotEXT privateDataSlot
,
2787 const VkAllocationCallbacks
* pAllocator
)
2789 TU_FROM_HANDLE(tu_device
, device
, _device
);
2790 vk_private_data_slot_destroy(&device
->vk
, privateDataSlot
, pAllocator
);
2794 tu_SetPrivateDataEXT(VkDevice _device
,
2795 VkObjectType objectType
,
2796 uint64_t objectHandle
,
2797 VkPrivateDataSlotEXT privateDataSlot
,
2800 TU_FROM_HANDLE(tu_device
, device
, _device
);
2801 return vk_object_base_set_private_data(&device
->vk
,
2809 tu_GetPrivateDataEXT(VkDevice _device
,
2810 VkObjectType objectType
,
2811 uint64_t objectHandle
,
2812 VkPrivateDataSlotEXT privateDataSlot
,
2815 TU_FROM_HANDLE(tu_device
, device
, _device
);
2816 vk_object_base_get_private_data(&device
->vk
,