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_device_get_cache_uuid(uint16_t family
, void *uuid
)
54 uint32_t mesa_timestamp
;
56 memset(uuid
, 0, VK_UUID_SIZE
);
57 if (!disk_cache_get_function_timestamp(tu_device_get_cache_uuid
,
61 memcpy(uuid
, &mesa_timestamp
, 4);
62 memcpy((char *) uuid
+ 4, &f
, 2);
63 snprintf((char *) uuid
+ 6, VK_UUID_SIZE
- 10, "tu");
68 tu_bo_init(struct tu_device
*dev
,
73 uint64_t iova
= tu_gem_info_iova(dev
, gem_handle
);
75 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
77 *bo
= (struct tu_bo
) {
78 .gem_handle
= gem_handle
,
87 tu_bo_init_new(struct tu_device
*dev
, struct tu_bo
*bo
, uint64_t size
)
89 /* TODO: Choose better flags. As of 2018-11-12, freedreno/drm/msm_bo.c
90 * always sets `flags = MSM_BO_WC`, and we copy that behavior here.
92 uint32_t gem_handle
= tu_gem_new(dev
, size
, MSM_BO_WC
);
94 return vk_error(dev
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
96 VkResult result
= tu_bo_init(dev
, bo
, gem_handle
, size
);
97 if (result
!= VK_SUCCESS
) {
98 tu_gem_close(dev
, gem_handle
);
99 return vk_error(dev
->instance
, result
);
106 tu_bo_init_dmabuf(struct tu_device
*dev
,
111 uint32_t gem_handle
= tu_gem_import_dmabuf(dev
, fd
, size
);
113 return vk_error(dev
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE
);
115 VkResult result
= tu_bo_init(dev
, bo
, gem_handle
, size
);
116 if (result
!= VK_SUCCESS
) {
117 tu_gem_close(dev
, gem_handle
);
118 return vk_error(dev
->instance
, result
);
125 tu_bo_export_dmabuf(struct tu_device
*dev
, struct tu_bo
*bo
)
127 return tu_gem_export_dmabuf(dev
, bo
->gem_handle
);
131 tu_bo_map(struct tu_device
*dev
, struct tu_bo
*bo
)
136 uint64_t offset
= tu_gem_info_offset(dev
, bo
->gem_handle
);
138 return vk_error(dev
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
140 /* TODO: Should we use the wrapper os_mmap() like Freedreno does? */
141 void *map
= mmap(0, bo
->size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
142 dev
->physical_device
->local_fd
, offset
);
143 if (map
== MAP_FAILED
)
144 return vk_error(dev
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
151 tu_bo_finish(struct tu_device
*dev
, struct tu_bo
*bo
)
153 assert(bo
->gem_handle
);
156 munmap(bo
->map
, bo
->size
);
158 tu_gem_close(dev
, bo
->gem_handle
);
162 tu_physical_device_init(struct tu_physical_device
*device
,
163 struct tu_instance
*instance
,
164 drmDevicePtr drm_device
)
166 const char *path
= drm_device
->nodes
[DRM_NODE_RENDER
];
167 VkResult result
= VK_SUCCESS
;
168 drmVersionPtr version
;
172 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
174 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
175 "failed to open device %s", path
);
178 /* Version 1.3 added MSM_INFO_IOVA. */
179 const int min_version_major
= 1;
180 const int min_version_minor
= 3;
182 version
= drmGetVersion(fd
);
185 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
186 "failed to query kernel driver version for device %s",
190 if (strcmp(version
->name
, "msm")) {
191 drmFreeVersion(version
);
193 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
194 "device %s does not use the msm kernel driver", path
);
197 if (version
->version_major
!= min_version_major
||
198 version
->version_minor
< min_version_minor
) {
199 result
= vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
200 "kernel driver for device %s has version %d.%d, "
201 "but Vulkan requires version >= %d.%d",
202 path
, version
->version_major
, version
->version_minor
,
203 min_version_major
, min_version_minor
);
204 drmFreeVersion(version
);
209 drmFreeVersion(version
);
211 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
212 tu_logi("Found compatible device '%s'.", path
);
214 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
215 device
->instance
= instance
;
216 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
217 strncpy(device
->path
, path
, ARRAY_SIZE(device
->path
));
219 if (instance
->enabled_extensions
.KHR_display
) {
221 open(drm_device
->nodes
[DRM_NODE_PRIMARY
], O_RDWR
| O_CLOEXEC
);
222 if (master_fd
>= 0) {
223 /* TODO: free master_fd is accel is not working? */
227 device
->master_fd
= master_fd
;
228 device
->local_fd
= fd
;
230 if (tu_drm_get_gpu_id(device
, &device
->gpu_id
)) {
231 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
232 tu_logi("Could not query the GPU ID");
233 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
234 "could not get GPU ID");
238 if (tu_drm_get_gmem_size(device
, &device
->gmem_size
)) {
239 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
240 tu_logi("Could not query the GMEM size");
241 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
242 "could not get GMEM size");
246 if (tu_drm_get_gmem_base(device
, &device
->gmem_base
)) {
247 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
248 tu_logi("Could not query the GMEM size");
249 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
250 "could not get GMEM size");
254 memset(device
->name
, 0, sizeof(device
->name
));
255 sprintf(device
->name
, "FD%d", device
->gpu_id
);
257 switch (device
->gpu_id
) {
259 device
->ccu_offset_gmem
= 0x7c000; /* 0x7e000 in some cases? */
260 device
->ccu_offset_bypass
= 0x10000;
261 device
->tile_align_w
= 64;
262 device
->magic
.PC_UNKNOWN_9805
= 0x0;
263 device
->magic
.SP_UNKNOWN_A0F8
= 0x0;
267 device
->ccu_offset_gmem
= 0xf8000;
268 device
->ccu_offset_bypass
= 0x20000;
269 device
->tile_align_w
= 64;
270 device
->magic
.PC_UNKNOWN_9805
= 0x1;
271 device
->magic
.SP_UNKNOWN_A0F8
= 0x1;
274 device
->ccu_offset_gmem
= 0x114000;
275 device
->ccu_offset_bypass
= 0x30000;
276 device
->tile_align_w
= 96;
277 device
->magic
.PC_UNKNOWN_9805
= 0x2;
278 device
->magic
.SP_UNKNOWN_A0F8
= 0x2;
281 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
282 "device %s is unsupported", device
->name
);
285 if (tu_device_get_cache_uuid(device
->gpu_id
, device
->cache_uuid
)) {
286 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
287 "cannot generate UUID");
291 /* The gpu id is already embedded in the uuid so we just pass "tu"
292 * when creating the cache.
294 char buf
[VK_UUID_SIZE
* 2 + 1];
295 disk_cache_format_hex_id(buf
, device
->cache_uuid
, VK_UUID_SIZE
* 2);
296 device
->disk_cache
= disk_cache_create(device
->name
, buf
, 0);
298 fprintf(stderr
, "WARNING: tu is not a conformant vulkan implementation, "
299 "testing use only.\n");
301 fd_get_driver_uuid(device
->driver_uuid
);
302 fd_get_device_uuid(device
->device_uuid
, device
->gpu_id
);
304 tu_physical_device_get_supported_extensions(device
, &device
->supported_extensions
);
306 if (result
!= VK_SUCCESS
) {
307 vk_error(instance
, result
);
311 result
= tu_wsi_init(device
);
312 if (result
!= VK_SUCCESS
) {
313 vk_error(instance
, result
);
327 tu_physical_device_finish(struct tu_physical_device
*device
)
329 tu_wsi_finish(device
);
331 disk_cache_destroy(device
->disk_cache
);
332 close(device
->local_fd
);
333 if (device
->master_fd
!= -1)
334 close(device
->master_fd
);
337 static VKAPI_ATTR
void *
338 default_alloc_func(void *pUserData
,
341 VkSystemAllocationScope allocationScope
)
346 static VKAPI_ATTR
void *
347 default_realloc_func(void *pUserData
,
351 VkSystemAllocationScope allocationScope
)
353 return realloc(pOriginal
, size
);
356 static VKAPI_ATTR
void
357 default_free_func(void *pUserData
, void *pMemory
)
362 static const VkAllocationCallbacks default_alloc
= {
364 .pfnAllocation
= default_alloc_func
,
365 .pfnReallocation
= default_realloc_func
,
366 .pfnFree
= default_free_func
,
369 static const struct debug_control tu_debug_options
[] = {
370 { "startup", TU_DEBUG_STARTUP
},
371 { "nir", TU_DEBUG_NIR
},
372 { "ir3", TU_DEBUG_IR3
},
373 { "nobin", TU_DEBUG_NOBIN
},
374 { "sysmem", TU_DEBUG_SYSMEM
},
375 { "forcebin", TU_DEBUG_FORCEBIN
},
376 { "noubwc", TU_DEBUG_NOUBWC
},
381 tu_get_debug_option_name(int id
)
383 assert(id
< ARRAY_SIZE(tu_debug_options
) - 1);
384 return tu_debug_options
[id
].string
;
388 tu_get_instance_extension_index(const char *name
)
390 for (unsigned i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; ++i
) {
391 if (strcmp(name
, tu_instance_extensions
[i
].extensionName
) == 0)
398 tu_CreateInstance(const VkInstanceCreateInfo
*pCreateInfo
,
399 const VkAllocationCallbacks
*pAllocator
,
400 VkInstance
*pInstance
)
402 struct tu_instance
*instance
;
405 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
407 uint32_t client_version
;
408 if (pCreateInfo
->pApplicationInfo
&&
409 pCreateInfo
->pApplicationInfo
->apiVersion
!= 0) {
410 client_version
= pCreateInfo
->pApplicationInfo
->apiVersion
;
412 tu_EnumerateInstanceVersion(&client_version
);
415 instance
= vk_zalloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
416 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
418 return vk_error(NULL
, VK_ERROR_OUT_OF_HOST_MEMORY
);
420 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
423 instance
->alloc
= *pAllocator
;
425 instance
->alloc
= default_alloc
;
427 instance
->api_version
= client_version
;
428 instance
->physical_device_count
= -1;
430 instance
->debug_flags
=
431 parse_debug_string(getenv("TU_DEBUG"), tu_debug_options
);
433 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
434 tu_logi("Created an instance");
436 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
437 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
438 int index
= tu_get_instance_extension_index(ext_name
);
440 if (index
< 0 || !tu_instance_extensions_supported
.extensions
[index
]) {
441 vk_free2(&default_alloc
, pAllocator
, instance
);
442 return vk_error(instance
, VK_ERROR_EXTENSION_NOT_PRESENT
);
445 instance
->enabled_extensions
.extensions
[index
] = true;
448 result
= vk_debug_report_instance_init(&instance
->debug_report_callbacks
);
449 if (result
!= VK_SUCCESS
) {
450 vk_free2(&default_alloc
, pAllocator
, instance
);
451 return vk_error(instance
, result
);
454 glsl_type_singleton_init_or_ref();
456 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
458 *pInstance
= tu_instance_to_handle(instance
);
464 tu_DestroyInstance(VkInstance _instance
,
465 const VkAllocationCallbacks
*pAllocator
)
467 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
472 for (int i
= 0; i
< instance
->physical_device_count
; ++i
) {
473 tu_physical_device_finish(instance
->physical_devices
+ i
);
476 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
478 glsl_type_singleton_decref();
480 vk_debug_report_instance_destroy(&instance
->debug_report_callbacks
);
482 vk_free(&instance
->alloc
, instance
);
486 tu_enumerate_devices(struct tu_instance
*instance
)
488 /* TODO: Check for more devices ? */
489 drmDevicePtr devices
[8];
490 VkResult result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
493 instance
->physical_device_count
= 0;
495 max_devices
= drmGetDevices2(0, devices
, ARRAY_SIZE(devices
));
497 if (instance
->debug_flags
& TU_DEBUG_STARTUP
)
498 tu_logi("Found %d drm nodes", max_devices
);
501 return vk_error(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
);
503 for (unsigned i
= 0; i
< (unsigned) max_devices
; i
++) {
504 if (devices
[i
]->available_nodes
& 1 << DRM_NODE_RENDER
&&
505 devices
[i
]->bustype
== DRM_BUS_PLATFORM
) {
507 result
= tu_physical_device_init(
508 instance
->physical_devices
+ instance
->physical_device_count
,
509 instance
, devices
[i
]);
510 if (result
== VK_SUCCESS
)
511 ++instance
->physical_device_count
;
512 else if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
516 drmFreeDevices(devices
, max_devices
);
522 tu_EnumeratePhysicalDevices(VkInstance _instance
,
523 uint32_t *pPhysicalDeviceCount
,
524 VkPhysicalDevice
*pPhysicalDevices
)
526 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
527 VK_OUTARRAY_MAKE(out
, pPhysicalDevices
, pPhysicalDeviceCount
);
531 if (instance
->physical_device_count
< 0) {
532 result
= tu_enumerate_devices(instance
);
533 if (result
!= VK_SUCCESS
&& result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
537 for (uint32_t i
= 0; i
< instance
->physical_device_count
; ++i
) {
538 vk_outarray_append(&out
, p
)
540 *p
= tu_physical_device_to_handle(instance
->physical_devices
+ i
);
544 return vk_outarray_status(&out
);
548 tu_EnumeratePhysicalDeviceGroups(
549 VkInstance _instance
,
550 uint32_t *pPhysicalDeviceGroupCount
,
551 VkPhysicalDeviceGroupProperties
*pPhysicalDeviceGroupProperties
)
553 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
554 VK_OUTARRAY_MAKE(out
, pPhysicalDeviceGroupProperties
,
555 pPhysicalDeviceGroupCount
);
558 if (instance
->physical_device_count
< 0) {
559 result
= tu_enumerate_devices(instance
);
560 if (result
!= VK_SUCCESS
&& result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
564 for (uint32_t i
= 0; i
< instance
->physical_device_count
; ++i
) {
565 vk_outarray_append(&out
, p
)
567 p
->physicalDeviceCount
= 1;
568 p
->physicalDevices
[0] =
569 tu_physical_device_to_handle(instance
->physical_devices
+ i
);
570 p
->subsetAllocation
= false;
574 return vk_outarray_status(&out
);
578 tu_GetPhysicalDeviceFeatures(VkPhysicalDevice physicalDevice
,
579 VkPhysicalDeviceFeatures
*pFeatures
)
581 memset(pFeatures
, 0, sizeof(*pFeatures
));
583 *pFeatures
= (VkPhysicalDeviceFeatures
) {
584 .robustBufferAccess
= true,
585 .fullDrawIndexUint32
= true,
586 .imageCubeArray
= true,
587 .independentBlend
= true,
588 .geometryShader
= true,
589 .tessellationShader
= true,
590 .sampleRateShading
= true,
591 .dualSrcBlend
= true,
593 .multiDrawIndirect
= true,
594 .drawIndirectFirstInstance
= true,
596 .depthBiasClamp
= false,
597 .fillModeNonSolid
= false,
600 .largePoints
= false,
602 .multiViewport
= false,
603 .samplerAnisotropy
= true,
604 .textureCompressionETC2
= true,
605 .textureCompressionASTC_LDR
= true,
606 .textureCompressionBC
= true,
607 .occlusionQueryPrecise
= true,
608 .pipelineStatisticsQuery
= false,
609 .vertexPipelineStoresAndAtomics
= false,
610 .fragmentStoresAndAtomics
= false,
611 .shaderTessellationAndGeometryPointSize
= false,
612 .shaderImageGatherExtended
= false,
613 .shaderStorageImageExtendedFormats
= false,
614 .shaderStorageImageMultisample
= false,
615 .shaderUniformBufferArrayDynamicIndexing
= false,
616 .shaderSampledImageArrayDynamicIndexing
= false,
617 .shaderStorageBufferArrayDynamicIndexing
= false,
618 .shaderStorageImageArrayDynamicIndexing
= false,
619 .shaderStorageImageReadWithoutFormat
= false,
620 .shaderStorageImageWriteWithoutFormat
= false,
621 .shaderClipDistance
= false,
622 .shaderCullDistance
= false,
623 .shaderFloat64
= false,
624 .shaderInt64
= false,
625 .shaderInt16
= false,
626 .sparseBinding
= false,
627 .variableMultisampleRate
= false,
628 .inheritedQueries
= false,
633 tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice
,
634 VkPhysicalDeviceFeatures2
*pFeatures
)
636 vk_foreach_struct(ext
, pFeatures
->pNext
)
638 switch (ext
->sType
) {
639 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES
:
640 *((VkPhysicalDeviceVulkan11Features
*) ext
) = (VkPhysicalDeviceVulkan11Features
) {
641 .storageBuffer16BitAccess
= false,
642 .uniformAndStorageBuffer16BitAccess
= false,
643 .storagePushConstant16
= false,
644 .storageInputOutput16
= false,
646 .multiviewGeometryShader
= false,
647 .multiviewTessellationShader
= false,
648 .variablePointersStorageBuffer
= false,
649 .variablePointers
= false,
650 .protectedMemory
= false,
651 .samplerYcbcrConversion
= true,
652 .shaderDrawParameters
= true,
655 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES
: {
656 VkPhysicalDeviceVariablePointersFeatures
*features
= (void *) ext
;
657 features
->variablePointersStorageBuffer
= false;
658 features
->variablePointers
= false;
661 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES
: {
662 VkPhysicalDeviceMultiviewFeatures
*features
=
663 (VkPhysicalDeviceMultiviewFeatures
*) ext
;
664 features
->multiview
= false;
665 features
->multiviewGeometryShader
= false;
666 features
->multiviewTessellationShader
= false;
669 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES
: {
670 VkPhysicalDeviceShaderDrawParametersFeatures
*features
=
671 (VkPhysicalDeviceShaderDrawParametersFeatures
*) ext
;
672 features
->shaderDrawParameters
= true;
675 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES
: {
676 VkPhysicalDeviceProtectedMemoryFeatures
*features
=
677 (VkPhysicalDeviceProtectedMemoryFeatures
*) ext
;
678 features
->protectedMemory
= false;
681 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES
: {
682 VkPhysicalDevice16BitStorageFeatures
*features
=
683 (VkPhysicalDevice16BitStorageFeatures
*) ext
;
684 features
->storageBuffer16BitAccess
= false;
685 features
->uniformAndStorageBuffer16BitAccess
= false;
686 features
->storagePushConstant16
= false;
687 features
->storageInputOutput16
= false;
690 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES
: {
691 VkPhysicalDeviceSamplerYcbcrConversionFeatures
*features
=
692 (VkPhysicalDeviceSamplerYcbcrConversionFeatures
*) ext
;
693 features
->samplerYcbcrConversion
= true;
696 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT
: {
697 VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*features
=
698 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*) ext
;
699 features
->shaderInputAttachmentArrayDynamicIndexing
= false;
700 features
->shaderUniformTexelBufferArrayDynamicIndexing
= false;
701 features
->shaderStorageTexelBufferArrayDynamicIndexing
= false;
702 features
->shaderUniformBufferArrayNonUniformIndexing
= false;
703 features
->shaderSampledImageArrayNonUniformIndexing
= false;
704 features
->shaderStorageBufferArrayNonUniformIndexing
= false;
705 features
->shaderStorageImageArrayNonUniformIndexing
= false;
706 features
->shaderInputAttachmentArrayNonUniformIndexing
= false;
707 features
->shaderUniformTexelBufferArrayNonUniformIndexing
= false;
708 features
->shaderStorageTexelBufferArrayNonUniformIndexing
= false;
709 features
->descriptorBindingUniformBufferUpdateAfterBind
= false;
710 features
->descriptorBindingSampledImageUpdateAfterBind
= false;
711 features
->descriptorBindingStorageImageUpdateAfterBind
= false;
712 features
->descriptorBindingStorageBufferUpdateAfterBind
= false;
713 features
->descriptorBindingUniformTexelBufferUpdateAfterBind
= false;
714 features
->descriptorBindingStorageTexelBufferUpdateAfterBind
= false;
715 features
->descriptorBindingUpdateUnusedWhilePending
= false;
716 features
->descriptorBindingPartiallyBound
= false;
717 features
->descriptorBindingVariableDescriptorCount
= false;
718 features
->runtimeDescriptorArray
= false;
721 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT
: {
722 VkPhysicalDeviceConditionalRenderingFeaturesEXT
*features
=
723 (VkPhysicalDeviceConditionalRenderingFeaturesEXT
*) ext
;
724 features
->conditionalRendering
= false;
725 features
->inheritedConditionalRendering
= false;
728 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT
: {
729 VkPhysicalDeviceTransformFeedbackFeaturesEXT
*features
=
730 (VkPhysicalDeviceTransformFeedbackFeaturesEXT
*) ext
;
731 features
->transformFeedback
= true;
732 features
->geometryStreams
= false;
735 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT
: {
736 VkPhysicalDeviceIndexTypeUint8FeaturesEXT
*features
=
737 (VkPhysicalDeviceIndexTypeUint8FeaturesEXT
*)ext
;
738 features
->indexTypeUint8
= true;
741 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT
: {
742 VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*features
=
743 (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*)ext
;
744 features
->vertexAttributeInstanceRateDivisor
= true;
745 features
->vertexAttributeInstanceRateZeroDivisor
= true;
752 return tu_GetPhysicalDeviceFeatures(physicalDevice
, &pFeatures
->features
);
756 tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice
,
757 VkPhysicalDeviceProperties
*pProperties
)
759 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
760 VkSampleCountFlags sample_counts
=
761 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
763 /* I have no idea what the maximum size is, but the hardware supports very
764 * large numbers of descriptors (at least 2^16). This limit is based on
765 * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
766 * we don't have to think about what to do if that overflows, but really
767 * nothing is likely to get close to this.
769 const size_t max_descriptor_set_size
= (1 << 28) / A6XX_TEX_CONST_DWORDS
;
771 VkPhysicalDeviceLimits limits
= {
772 .maxImageDimension1D
= (1 << 14),
773 .maxImageDimension2D
= (1 << 14),
774 .maxImageDimension3D
= (1 << 11),
775 .maxImageDimensionCube
= (1 << 14),
776 .maxImageArrayLayers
= (1 << 11),
777 .maxTexelBufferElements
= 128 * 1024 * 1024,
778 .maxUniformBufferRange
= MAX_UNIFORM_BUFFER_RANGE
,
779 .maxStorageBufferRange
= MAX_STORAGE_BUFFER_RANGE
,
780 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
781 .maxMemoryAllocationCount
= UINT32_MAX
,
782 .maxSamplerAllocationCount
= 64 * 1024,
783 .bufferImageGranularity
= 64, /* A cache line */
784 .sparseAddressSpaceSize
= 0xffffffffu
, /* buffer max size */
785 .maxBoundDescriptorSets
= MAX_SETS
,
786 .maxPerStageDescriptorSamplers
= max_descriptor_set_size
,
787 .maxPerStageDescriptorUniformBuffers
= max_descriptor_set_size
,
788 .maxPerStageDescriptorStorageBuffers
= max_descriptor_set_size
,
789 .maxPerStageDescriptorSampledImages
= max_descriptor_set_size
,
790 .maxPerStageDescriptorStorageImages
= max_descriptor_set_size
,
791 .maxPerStageDescriptorInputAttachments
= MAX_RTS
,
792 .maxPerStageResources
= max_descriptor_set_size
,
793 .maxDescriptorSetSamplers
= max_descriptor_set_size
,
794 .maxDescriptorSetUniformBuffers
= max_descriptor_set_size
,
795 .maxDescriptorSetUniformBuffersDynamic
= MAX_DYNAMIC_UNIFORM_BUFFERS
,
796 .maxDescriptorSetStorageBuffers
= max_descriptor_set_size
,
797 .maxDescriptorSetStorageBuffersDynamic
= MAX_DYNAMIC_STORAGE_BUFFERS
,
798 .maxDescriptorSetSampledImages
= max_descriptor_set_size
,
799 .maxDescriptorSetStorageImages
= max_descriptor_set_size
,
800 .maxDescriptorSetInputAttachments
= MAX_RTS
,
801 .maxVertexInputAttributes
= 32,
802 .maxVertexInputBindings
= 32,
803 .maxVertexInputAttributeOffset
= 4095,
804 .maxVertexInputBindingStride
= 2048,
805 .maxVertexOutputComponents
= 128,
806 .maxTessellationGenerationLevel
= 64,
807 .maxTessellationPatchSize
= 32,
808 .maxTessellationControlPerVertexInputComponents
= 128,
809 .maxTessellationControlPerVertexOutputComponents
= 128,
810 .maxTessellationControlPerPatchOutputComponents
= 120,
811 .maxTessellationControlTotalOutputComponents
= 4096,
812 .maxTessellationEvaluationInputComponents
= 128,
813 .maxTessellationEvaluationOutputComponents
= 128,
814 .maxGeometryShaderInvocations
= 32,
815 .maxGeometryInputComponents
= 64,
816 .maxGeometryOutputComponents
= 128,
817 .maxGeometryOutputVertices
= 256,
818 .maxGeometryTotalOutputComponents
= 1024,
819 .maxFragmentInputComponents
= 124,
820 .maxFragmentOutputAttachments
= 8,
821 .maxFragmentDualSrcAttachments
= 1,
822 .maxFragmentCombinedOutputResources
= 8,
823 .maxComputeSharedMemorySize
= 32768,
824 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
825 .maxComputeWorkGroupInvocations
= 2048,
826 .maxComputeWorkGroupSize
= { 2048, 2048, 2048 },
827 .subPixelPrecisionBits
= 8,
828 .subTexelPrecisionBits
= 8,
829 .mipmapPrecisionBits
= 8,
830 .maxDrawIndexedIndexValue
= UINT32_MAX
,
831 .maxDrawIndirectCount
= UINT32_MAX
,
832 .maxSamplerLodBias
= 4095.0 / 256.0, /* [-16, 15.99609375] */
833 .maxSamplerAnisotropy
= 16,
834 .maxViewports
= MAX_VIEWPORTS
,
835 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
836 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
837 .viewportSubPixelBits
= 8,
838 .minMemoryMapAlignment
= 4096, /* A page */
839 .minTexelBufferOffsetAlignment
= 64,
840 .minUniformBufferOffsetAlignment
= 64,
841 .minStorageBufferOffsetAlignment
= 64,
842 .minTexelOffset
= -16,
843 .maxTexelOffset
= 15,
844 .minTexelGatherOffset
= -32,
845 .maxTexelGatherOffset
= 31,
846 .minInterpolationOffset
= -0.5,
847 .maxInterpolationOffset
= 0.4375,
848 .subPixelInterpolationOffsetBits
= 4,
849 .maxFramebufferWidth
= (1 << 14),
850 .maxFramebufferHeight
= (1 << 14),
851 .maxFramebufferLayers
= (1 << 10),
852 .framebufferColorSampleCounts
= sample_counts
,
853 .framebufferDepthSampleCounts
= sample_counts
,
854 .framebufferStencilSampleCounts
= sample_counts
,
855 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
856 .maxColorAttachments
= MAX_RTS
,
857 .sampledImageColorSampleCounts
= sample_counts
,
858 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
859 .sampledImageDepthSampleCounts
= sample_counts
,
860 .sampledImageStencilSampleCounts
= sample_counts
,
861 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
862 .maxSampleMaskWords
= 1,
863 .timestampComputeAndGraphics
= true,
864 .timestampPeriod
= 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
865 .maxClipDistances
= 8,
866 .maxCullDistances
= 8,
867 .maxCombinedClipAndCullDistances
= 8,
868 .discreteQueuePriorities
= 1,
869 .pointSizeRange
= { 0.125, 255.875 },
870 .lineWidthRange
= { 0.0, 7.9921875 },
871 .pointSizeGranularity
= (1.0 / 8.0),
872 .lineWidthGranularity
= (1.0 / 128.0),
873 .strictLines
= false, /* FINISHME */
874 .standardSampleLocations
= true,
875 .optimalBufferCopyOffsetAlignment
= 128,
876 .optimalBufferCopyRowPitchAlignment
= 128,
877 .nonCoherentAtomSize
= 64,
880 *pProperties
= (VkPhysicalDeviceProperties
) {
881 .apiVersion
= tu_physical_device_api_version(pdevice
),
882 .driverVersion
= vk_get_driver_version(),
883 .vendorID
= 0, /* TODO */
885 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
887 .sparseProperties
= { 0 },
890 strcpy(pProperties
->deviceName
, pdevice
->name
);
891 memcpy(pProperties
->pipelineCacheUUID
, pdevice
->cache_uuid
, VK_UUID_SIZE
);
895 tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice
,
896 VkPhysicalDeviceProperties2
*pProperties
)
898 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
899 tu_GetPhysicalDeviceProperties(physicalDevice
, &pProperties
->properties
);
901 vk_foreach_struct(ext
, pProperties
->pNext
)
903 switch (ext
->sType
) {
904 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR
: {
905 VkPhysicalDevicePushDescriptorPropertiesKHR
*properties
=
906 (VkPhysicalDevicePushDescriptorPropertiesKHR
*) ext
;
907 properties
->maxPushDescriptors
= MAX_PUSH_DESCRIPTORS
;
910 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES
: {
911 VkPhysicalDeviceIDProperties
*properties
=
912 (VkPhysicalDeviceIDProperties
*) ext
;
913 memcpy(properties
->driverUUID
, pdevice
->driver_uuid
, VK_UUID_SIZE
);
914 memcpy(properties
->deviceUUID
, pdevice
->device_uuid
, VK_UUID_SIZE
);
915 properties
->deviceLUIDValid
= false;
918 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES
: {
919 VkPhysicalDeviceMultiviewProperties
*properties
=
920 (VkPhysicalDeviceMultiviewProperties
*) ext
;
921 properties
->maxMultiviewViewCount
= MAX_VIEWS
;
922 properties
->maxMultiviewInstanceIndex
= INT_MAX
;
925 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES
: {
926 VkPhysicalDevicePointClippingProperties
*properties
=
927 (VkPhysicalDevicePointClippingProperties
*) ext
;
928 properties
->pointClippingBehavior
=
929 VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES
;
932 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES
: {
933 VkPhysicalDeviceMaintenance3Properties
*properties
=
934 (VkPhysicalDeviceMaintenance3Properties
*) ext
;
935 /* Make sure everything is addressable by a signed 32-bit int, and
936 * our largest descriptors are 96 bytes. */
937 properties
->maxPerSetDescriptors
= (1ull << 31) / 96;
938 /* Our buffer size fields allow only this much */
939 properties
->maxMemoryAllocationSize
= 0xFFFFFFFFull
;
942 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT
: {
943 VkPhysicalDeviceTransformFeedbackPropertiesEXT
*properties
=
944 (VkPhysicalDeviceTransformFeedbackPropertiesEXT
*)ext
;
946 properties
->maxTransformFeedbackStreams
= IR3_MAX_SO_STREAMS
;
947 properties
->maxTransformFeedbackBuffers
= IR3_MAX_SO_BUFFERS
;
948 properties
->maxTransformFeedbackBufferSize
= UINT32_MAX
;
949 properties
->maxTransformFeedbackStreamDataSize
= 512;
950 properties
->maxTransformFeedbackBufferDataSize
= 512;
951 properties
->maxTransformFeedbackBufferDataStride
= 512;
952 properties
->transformFeedbackQueries
= true;
953 properties
->transformFeedbackStreamsLinesTriangles
= false;
954 properties
->transformFeedbackRasterizationStreamSelect
= false;
955 properties
->transformFeedbackDraw
= true;
958 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT
: {
959 VkPhysicalDeviceSampleLocationsPropertiesEXT
*properties
=
960 (VkPhysicalDeviceSampleLocationsPropertiesEXT
*)ext
;
961 properties
->sampleLocationSampleCounts
= 0;
962 if (pdevice
->supported_extensions
.EXT_sample_locations
) {
963 properties
->sampleLocationSampleCounts
=
964 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
966 properties
->maxSampleLocationGridSize
= (VkExtent2D
) { 1 , 1 };
967 properties
->sampleLocationCoordinateRange
[0] = 0.0f
;
968 properties
->sampleLocationCoordinateRange
[1] = 0.9375f
;
969 properties
->sampleLocationSubPixelBits
= 4;
970 properties
->variableSampleLocations
= true;
973 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES
: {
974 VkPhysicalDeviceSamplerFilterMinmaxProperties
*properties
=
975 (VkPhysicalDeviceSamplerFilterMinmaxProperties
*)ext
;
976 properties
->filterMinmaxImageComponentMapping
= true;
977 properties
->filterMinmaxSingleComponentFormats
= true;
980 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES
: {
981 VkPhysicalDeviceSubgroupProperties
*properties
=
982 (VkPhysicalDeviceSubgroupProperties
*)ext
;
983 properties
->subgroupSize
= 64;
984 properties
->supportedStages
= VK_SHADER_STAGE_COMPUTE_BIT
;
985 properties
->supportedOperations
= VK_SUBGROUP_FEATURE_BASIC_BIT
|
986 VK_SUBGROUP_FEATURE_VOTE_BIT
;
987 properties
->quadOperationsInAllStages
= false;
990 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT
: {
991 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*props
=
992 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*)ext
;
993 props
->maxVertexAttribDivisor
= UINT32_MAX
;
1002 static const VkQueueFamilyProperties tu_queue_family_properties
= {
1004 VK_QUEUE_GRAPHICS_BIT
| VK_QUEUE_COMPUTE_BIT
| VK_QUEUE_TRANSFER_BIT
,
1006 .timestampValidBits
= 48,
1007 .minImageTransferGranularity
= { 1, 1, 1 },
1011 tu_GetPhysicalDeviceQueueFamilyProperties(
1012 VkPhysicalDevice physicalDevice
,
1013 uint32_t *pQueueFamilyPropertyCount
,
1014 VkQueueFamilyProperties
*pQueueFamilyProperties
)
1016 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1018 vk_outarray_append(&out
, p
) { *p
= tu_queue_family_properties
; }
1022 tu_GetPhysicalDeviceQueueFamilyProperties2(
1023 VkPhysicalDevice physicalDevice
,
1024 uint32_t *pQueueFamilyPropertyCount
,
1025 VkQueueFamilyProperties2
*pQueueFamilyProperties
)
1027 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
1029 vk_outarray_append(&out
, p
)
1031 p
->queueFamilyProperties
= tu_queue_family_properties
;
1036 tu_get_system_heap_size()
1038 struct sysinfo info
;
1041 uint64_t total_ram
= (uint64_t) info
.totalram
* (uint64_t) info
.mem_unit
;
1043 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
1044 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
1046 uint64_t available_ram
;
1047 if (total_ram
<= 4ull * 1024ull * 1024ull * 1024ull)
1048 available_ram
= total_ram
/ 2;
1050 available_ram
= total_ram
* 3 / 4;
1052 return available_ram
;
1056 tu_GetPhysicalDeviceMemoryProperties(
1057 VkPhysicalDevice physicalDevice
,
1058 VkPhysicalDeviceMemoryProperties
*pMemoryProperties
)
1060 pMemoryProperties
->memoryHeapCount
= 1;
1061 pMemoryProperties
->memoryHeaps
[0].size
= tu_get_system_heap_size();
1062 pMemoryProperties
->memoryHeaps
[0].flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
;
1064 pMemoryProperties
->memoryTypeCount
= 1;
1065 pMemoryProperties
->memoryTypes
[0].propertyFlags
=
1066 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
1067 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
1068 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
;
1069 pMemoryProperties
->memoryTypes
[0].heapIndex
= 0;
1073 tu_GetPhysicalDeviceMemoryProperties2(
1074 VkPhysicalDevice physicalDevice
,
1075 VkPhysicalDeviceMemoryProperties2
*pMemoryProperties
)
1077 return tu_GetPhysicalDeviceMemoryProperties(
1078 physicalDevice
, &pMemoryProperties
->memoryProperties
);
1082 tu_queue_init(struct tu_device
*device
,
1083 struct tu_queue
*queue
,
1084 uint32_t queue_family_index
,
1086 VkDeviceQueueCreateFlags flags
)
1088 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1089 queue
->device
= device
;
1090 queue
->queue_family_index
= queue_family_index
;
1091 queue
->queue_idx
= idx
;
1092 queue
->flags
= flags
;
1094 int ret
= tu_drm_submitqueue_new(device
, 0, &queue
->msm_queue_id
);
1096 return VK_ERROR_INITIALIZATION_FAILED
;
1098 tu_fence_init(&queue
->submit_fence
, false);
1104 tu_queue_finish(struct tu_queue
*queue
)
1106 tu_fence_finish(&queue
->submit_fence
);
1107 tu_drm_submitqueue_close(queue
->device
, queue
->msm_queue_id
);
1111 tu_get_device_extension_index(const char *name
)
1113 for (unsigned i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; ++i
) {
1114 if (strcmp(name
, tu_device_extensions
[i
].extensionName
) == 0)
1120 struct PACKED bcolor_entry
{
1132 uint32_t z24
; /* also s8? */
1133 uint16_t srgb
[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
1135 } border_color
[] = {
1136 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = {},
1137 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = {},
1138 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = {
1139 .fp32
[3] = 0x3f800000,
1147 .rgb10a2
= 0xc0000000,
1150 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = {
1154 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = {
1155 .fp32
[0 ... 3] = 0x3f800000,
1156 .ui16
[0 ... 3] = 0xffff,
1157 .si16
[0 ... 3] = 0x7fff,
1158 .fp16
[0 ... 3] = 0x3c00,
1162 .ui8
[0 ... 3] = 0xff,
1163 .si8
[0 ... 3] = 0x7f,
1164 .rgb10a2
= 0xffffffff,
1166 .srgb
[0 ... 3] = 0x3c00,
1168 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = {
1176 tu_CreateDevice(VkPhysicalDevice physicalDevice
,
1177 const VkDeviceCreateInfo
*pCreateInfo
,
1178 const VkAllocationCallbacks
*pAllocator
,
1181 TU_FROM_HANDLE(tu_physical_device
, physical_device
, physicalDevice
);
1183 struct tu_device
*device
;
1185 /* Check enabled features */
1186 if (pCreateInfo
->pEnabledFeatures
) {
1187 VkPhysicalDeviceFeatures supported_features
;
1188 tu_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
1189 VkBool32
*supported_feature
= (VkBool32
*) &supported_features
;
1190 VkBool32
*enabled_feature
= (VkBool32
*) pCreateInfo
->pEnabledFeatures
;
1191 unsigned num_features
=
1192 sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
1193 for (uint32_t i
= 0; i
< num_features
; i
++) {
1194 if (enabled_feature
[i
] && !supported_feature
[i
])
1195 return vk_error(physical_device
->instance
,
1196 VK_ERROR_FEATURE_NOT_PRESENT
);
1200 device
= vk_zalloc2(&physical_device
->instance
->alloc
, pAllocator
,
1201 sizeof(*device
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1203 return vk_error(physical_device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1205 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1206 device
->instance
= physical_device
->instance
;
1207 device
->physical_device
= physical_device
;
1210 device
->alloc
= *pAllocator
;
1212 device
->alloc
= physical_device
->instance
->alloc
;
1214 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
1215 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
1216 int index
= tu_get_device_extension_index(ext_name
);
1218 !physical_device
->supported_extensions
.extensions
[index
]) {
1219 vk_free(&device
->alloc
, device
);
1220 return vk_error(physical_device
->instance
,
1221 VK_ERROR_EXTENSION_NOT_PRESENT
);
1224 device
->enabled_extensions
.extensions
[index
] = true;
1227 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
1228 const VkDeviceQueueCreateInfo
*queue_create
=
1229 &pCreateInfo
->pQueueCreateInfos
[i
];
1230 uint32_t qfi
= queue_create
->queueFamilyIndex
;
1231 device
->queues
[qfi
] = vk_alloc(
1232 &device
->alloc
, queue_create
->queueCount
* sizeof(struct tu_queue
),
1233 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1234 if (!device
->queues
[qfi
]) {
1235 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
1239 memset(device
->queues
[qfi
], 0,
1240 queue_create
->queueCount
* sizeof(struct tu_queue
));
1242 device
->queue_count
[qfi
] = queue_create
->queueCount
;
1244 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
1245 result
= tu_queue_init(device
, &device
->queues
[qfi
][q
], qfi
, q
,
1246 queue_create
->flags
);
1247 if (result
!= VK_SUCCESS
)
1252 device
->compiler
= ir3_compiler_create(NULL
, physical_device
->gpu_id
);
1253 if (!device
->compiler
)
1256 #define VSC_DRAW_STRM_SIZE(pitch) ((pitch) * 32 + 0x100) /* extra size to store VSC_SIZE */
1257 #define VSC_PRIM_STRM_SIZE(pitch) ((pitch) * 32)
1259 device
->vsc_draw_strm_pitch
= 0x440 * 4;
1260 device
->vsc_prim_strm_pitch
= 0x1040 * 4;
1262 result
= tu_bo_init_new(device
, &device
->vsc_draw_strm
, VSC_DRAW_STRM_SIZE(device
->vsc_draw_strm_pitch
));
1263 if (result
!= VK_SUCCESS
)
1266 result
= tu_bo_init_new(device
, &device
->vsc_prim_strm
, VSC_PRIM_STRM_SIZE(device
->vsc_prim_strm_pitch
));
1267 if (result
!= VK_SUCCESS
)
1268 goto fail_vsc_data2
;
1270 STATIC_ASSERT(sizeof(struct bcolor_entry
) == 128);
1271 result
= tu_bo_init_new(device
, &device
->border_color
, sizeof(border_color
));
1272 if (result
!= VK_SUCCESS
)
1273 goto fail_border_color
;
1275 result
= tu_bo_map(device
, &device
->border_color
);
1276 if (result
!= VK_SUCCESS
)
1277 goto fail_border_color_map
;
1279 memcpy(device
->border_color
.map
, border_color
, sizeof(border_color
));
1281 VkPipelineCacheCreateInfo ci
;
1282 ci
.sType
= VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
;
1285 ci
.pInitialData
= NULL
;
1286 ci
.initialDataSize
= 0;
1289 tu_CreatePipelineCache(tu_device_to_handle(device
), &ci
, NULL
, &pc
);
1290 if (result
!= VK_SUCCESS
)
1291 goto fail_pipeline_cache
;
1293 device
->mem_cache
= tu_pipeline_cache_from_handle(pc
);
1295 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++)
1296 mtx_init(&device
->scratch_bos
[i
].construct_mtx
, mtx_plain
);
1298 *pDevice
= tu_device_to_handle(device
);
1301 fail_pipeline_cache
:
1302 fail_border_color_map
:
1303 tu_bo_finish(device
, &device
->border_color
);
1306 tu_bo_finish(device
, &device
->vsc_prim_strm
);
1309 tu_bo_finish(device
, &device
->vsc_draw_strm
);
1312 ralloc_free(device
->compiler
);
1315 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1316 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1317 tu_queue_finish(&device
->queues
[i
][q
]);
1318 if (device
->queue_count
[i
])
1319 vk_free(&device
->alloc
, device
->queues
[i
]);
1322 vk_free(&device
->alloc
, device
);
1327 tu_DestroyDevice(VkDevice _device
, const VkAllocationCallbacks
*pAllocator
)
1329 TU_FROM_HANDLE(tu_device
, device
, _device
);
1334 tu_bo_finish(device
, &device
->vsc_draw_strm
);
1335 tu_bo_finish(device
, &device
->vsc_prim_strm
);
1337 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1338 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1339 tu_queue_finish(&device
->queues
[i
][q
]);
1340 if (device
->queue_count
[i
])
1341 vk_free(&device
->alloc
, device
->queues
[i
]);
1344 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++) {
1345 if (device
->scratch_bos
[i
].initialized
)
1346 tu_bo_finish(device
, &device
->scratch_bos
[i
].bo
);
1349 /* the compiler does not use pAllocator */
1350 ralloc_free(device
->compiler
);
1352 VkPipelineCache pc
= tu_pipeline_cache_to_handle(device
->mem_cache
);
1353 tu_DestroyPipelineCache(tu_device_to_handle(device
), pc
, NULL
);
1355 vk_free(&device
->alloc
, device
);
1359 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
)
1361 unsigned size_log2
= MAX2(util_logbase2_ceil64(size
), MIN_SCRATCH_BO_SIZE_LOG2
);
1362 unsigned index
= size_log2
- MIN_SCRATCH_BO_SIZE_LOG2
;
1363 assert(index
< ARRAY_SIZE(dev
->scratch_bos
));
1365 for (unsigned i
= index
; i
< ARRAY_SIZE(dev
->scratch_bos
); i
++) {
1366 if (p_atomic_read(&dev
->scratch_bos
[i
].initialized
)) {
1367 /* Fast path: just return the already-allocated BO. */
1368 *bo
= &dev
->scratch_bos
[i
].bo
;
1373 /* Slow path: actually allocate the BO. We take a lock because the process
1374 * of allocating it is slow, and we don't want to block the CPU while it
1377 mtx_lock(&dev
->scratch_bos
[index
].construct_mtx
);
1379 /* Another thread may have allocated it already while we were waiting on
1380 * the lock. We need to check this in order to avoid double-allocating.
1382 if (dev
->scratch_bos
[index
].initialized
) {
1383 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1384 *bo
= &dev
->scratch_bos
[index
].bo
;
1388 unsigned bo_size
= 1ull << size_log2
;
1389 VkResult result
= tu_bo_init_new(dev
, &dev
->scratch_bos
[index
].bo
, bo_size
);
1390 if (result
!= VK_SUCCESS
) {
1391 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1395 p_atomic_set(&dev
->scratch_bos
[index
].initialized
, true);
1397 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1399 *bo
= &dev
->scratch_bos
[index
].bo
;
1404 tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount
,
1405 VkLayerProperties
*pProperties
)
1407 *pPropertyCount
= 0;
1412 tu_EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice
,
1413 uint32_t *pPropertyCount
,
1414 VkLayerProperties
*pProperties
)
1416 *pPropertyCount
= 0;
1421 tu_GetDeviceQueue2(VkDevice _device
,
1422 const VkDeviceQueueInfo2
*pQueueInfo
,
1425 TU_FROM_HANDLE(tu_device
, device
, _device
);
1426 struct tu_queue
*queue
;
1429 &device
->queues
[pQueueInfo
->queueFamilyIndex
][pQueueInfo
->queueIndex
];
1430 if (pQueueInfo
->flags
!= queue
->flags
) {
1431 /* From the Vulkan 1.1.70 spec:
1433 * "The queue returned by vkGetDeviceQueue2 must have the same
1434 * flags value from this structure as that used at device
1435 * creation time in a VkDeviceQueueCreateInfo instance. If no
1436 * matching flags were specified at device creation time then
1437 * pQueue will return VK_NULL_HANDLE."
1439 *pQueue
= VK_NULL_HANDLE
;
1443 *pQueue
= tu_queue_to_handle(queue
);
1447 tu_GetDeviceQueue(VkDevice _device
,
1448 uint32_t queueFamilyIndex
,
1449 uint32_t queueIndex
,
1452 const VkDeviceQueueInfo2 info
=
1453 (VkDeviceQueueInfo2
) { .sType
= VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2
,
1454 .queueFamilyIndex
= queueFamilyIndex
,
1455 .queueIndex
= queueIndex
};
1457 tu_GetDeviceQueue2(_device
, &info
, pQueue
);
1461 tu_QueueSubmit(VkQueue _queue
,
1462 uint32_t submitCount
,
1463 const VkSubmitInfo
*pSubmits
,
1466 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1468 for (uint32_t i
= 0; i
< submitCount
; ++i
) {
1469 const VkSubmitInfo
*submit
= pSubmits
+ i
;
1470 const bool last_submit
= (i
== submitCount
- 1);
1471 struct tu_bo_list bo_list
;
1472 tu_bo_list_init(&bo_list
);
1474 uint32_t entry_count
= 0;
1475 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1476 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1477 entry_count
+= cmdbuf
->cs
.entry_count
;
1480 struct drm_msm_gem_submit_cmd cmds
[entry_count
];
1481 uint32_t entry_idx
= 0;
1482 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1483 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1484 struct tu_cs
*cs
= &cmdbuf
->cs
;
1485 for (unsigned i
= 0; i
< cs
->entry_count
; ++i
, ++entry_idx
) {
1486 cmds
[entry_idx
].type
= MSM_SUBMIT_CMD_BUF
;
1487 cmds
[entry_idx
].submit_idx
=
1488 tu_bo_list_add(&bo_list
, cs
->entries
[i
].bo
,
1489 MSM_SUBMIT_BO_READ
| MSM_SUBMIT_BO_DUMP
);
1490 cmds
[entry_idx
].submit_offset
= cs
->entries
[i
].offset
;
1491 cmds
[entry_idx
].size
= cs
->entries
[i
].size
;
1492 cmds
[entry_idx
].pad
= 0;
1493 cmds
[entry_idx
].nr_relocs
= 0;
1494 cmds
[entry_idx
].relocs
= 0;
1497 tu_bo_list_merge(&bo_list
, &cmdbuf
->bo_list
);
1500 uint32_t flags
= MSM_PIPE_3D0
;
1502 flags
|= MSM_SUBMIT_FENCE_FD_OUT
;
1505 struct drm_msm_gem_submit req
= {
1507 .queueid
= queue
->msm_queue_id
,
1508 .bos
= (uint64_t)(uintptr_t) bo_list
.bo_infos
,
1509 .nr_bos
= bo_list
.count
,
1510 .cmds
= (uint64_t)(uintptr_t)cmds
,
1511 .nr_cmds
= entry_count
,
1514 int ret
= drmCommandWriteRead(queue
->device
->physical_device
->local_fd
,
1518 fprintf(stderr
, "submit failed: %s\n", strerror(errno
));
1522 tu_bo_list_destroy(&bo_list
);
1525 /* no need to merge fences as queue execution is serialized */
1526 tu_fence_update_fd(&queue
->submit_fence
, req
.fence_fd
);
1530 if (_fence
!= VK_NULL_HANDLE
) {
1531 TU_FROM_HANDLE(tu_fence
, fence
, _fence
);
1532 tu_fence_copy(fence
, &queue
->submit_fence
);
1539 tu_QueueWaitIdle(VkQueue _queue
)
1541 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1543 tu_fence_wait_idle(&queue
->submit_fence
);
1549 tu_DeviceWaitIdle(VkDevice _device
)
1551 TU_FROM_HANDLE(tu_device
, device
, _device
);
1553 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1554 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
1555 tu_QueueWaitIdle(tu_queue_to_handle(&device
->queues
[i
][q
]));
1562 tu_ImportSemaphoreFdKHR(VkDevice _device
,
1563 const VkImportSemaphoreFdInfoKHR
*pImportSemaphoreFdInfo
)
1571 tu_GetSemaphoreFdKHR(VkDevice _device
,
1572 const VkSemaphoreGetFdInfoKHR
*pGetFdInfo
,
1581 tu_ImportFenceFdKHR(VkDevice _device
,
1582 const VkImportFenceFdInfoKHR
*pImportFenceFdInfo
)
1590 tu_GetFenceFdKHR(VkDevice _device
,
1591 const VkFenceGetFdInfoKHR
*pGetFdInfo
,
1600 tu_EnumerateInstanceExtensionProperties(const char *pLayerName
,
1601 uint32_t *pPropertyCount
,
1602 VkExtensionProperties
*pProperties
)
1604 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1606 /* We spport no lyaers */
1608 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1610 for (int i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; i
++) {
1611 if (tu_instance_extensions_supported
.extensions
[i
]) {
1612 vk_outarray_append(&out
, prop
) { *prop
= tu_instance_extensions
[i
]; }
1616 return vk_outarray_status(&out
);
1620 tu_EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice
,
1621 const char *pLayerName
,
1622 uint32_t *pPropertyCount
,
1623 VkExtensionProperties
*pProperties
)
1625 /* We spport no lyaers */
1626 TU_FROM_HANDLE(tu_physical_device
, device
, physicalDevice
);
1627 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1629 /* We spport no lyaers */
1631 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1633 for (int i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; i
++) {
1634 if (device
->supported_extensions
.extensions
[i
]) {
1635 vk_outarray_append(&out
, prop
) { *prop
= tu_device_extensions
[i
]; }
1639 return vk_outarray_status(&out
);
1643 tu_GetInstanceProcAddr(VkInstance _instance
, const char *pName
)
1645 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
1647 return tu_lookup_entrypoint_checked(
1648 pName
, instance
? instance
->api_version
: 0,
1649 instance
? &instance
->enabled_extensions
: NULL
, NULL
);
1652 /* The loader wants us to expose a second GetInstanceProcAddr function
1653 * to work around certain LD_PRELOAD issues seen in apps.
1656 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1657 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
);
1660 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1661 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
)
1663 return tu_GetInstanceProcAddr(instance
, pName
);
1667 tu_GetDeviceProcAddr(VkDevice _device
, const char *pName
)
1669 TU_FROM_HANDLE(tu_device
, device
, _device
);
1671 return tu_lookup_entrypoint_checked(pName
, device
->instance
->api_version
,
1672 &device
->instance
->enabled_extensions
,
1673 &device
->enabled_extensions
);
1677 tu_alloc_memory(struct tu_device
*device
,
1678 const VkMemoryAllocateInfo
*pAllocateInfo
,
1679 const VkAllocationCallbacks
*pAllocator
,
1680 VkDeviceMemory
*pMem
)
1682 struct tu_device_memory
*mem
;
1685 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1687 if (pAllocateInfo
->allocationSize
== 0) {
1688 /* Apparently, this is allowed */
1689 *pMem
= VK_NULL_HANDLE
;
1693 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1694 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1696 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1698 const VkImportMemoryFdInfoKHR
*fd_info
=
1699 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
1700 if (fd_info
&& !fd_info
->handleType
)
1704 assert(fd_info
->handleType
==
1705 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
1706 fd_info
->handleType
==
1707 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
1710 * TODO Importing the same fd twice gives us the same handle without
1711 * reference counting. We need to maintain a per-instance handle-to-bo
1712 * table and add reference count to tu_bo.
1714 result
= tu_bo_init_dmabuf(device
, &mem
->bo
,
1715 pAllocateInfo
->allocationSize
, fd_info
->fd
);
1716 if (result
== VK_SUCCESS
) {
1717 /* take ownership and close the fd */
1722 tu_bo_init_new(device
, &mem
->bo
, pAllocateInfo
->allocationSize
);
1725 if (result
!= VK_SUCCESS
) {
1726 vk_free2(&device
->alloc
, pAllocator
, mem
);
1730 mem
->size
= pAllocateInfo
->allocationSize
;
1731 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1734 mem
->user_ptr
= NULL
;
1736 *pMem
= tu_device_memory_to_handle(mem
);
1742 tu_AllocateMemory(VkDevice _device
,
1743 const VkMemoryAllocateInfo
*pAllocateInfo
,
1744 const VkAllocationCallbacks
*pAllocator
,
1745 VkDeviceMemory
*pMem
)
1747 TU_FROM_HANDLE(tu_device
, device
, _device
);
1748 return tu_alloc_memory(device
, pAllocateInfo
, pAllocator
, pMem
);
1752 tu_FreeMemory(VkDevice _device
,
1753 VkDeviceMemory _mem
,
1754 const VkAllocationCallbacks
*pAllocator
)
1756 TU_FROM_HANDLE(tu_device
, device
, _device
);
1757 TU_FROM_HANDLE(tu_device_memory
, mem
, _mem
);
1762 tu_bo_finish(device
, &mem
->bo
);
1763 vk_free2(&device
->alloc
, pAllocator
, mem
);
1767 tu_MapMemory(VkDevice _device
,
1768 VkDeviceMemory _memory
,
1769 VkDeviceSize offset
,
1771 VkMemoryMapFlags flags
,
1774 TU_FROM_HANDLE(tu_device
, device
, _device
);
1775 TU_FROM_HANDLE(tu_device_memory
, mem
, _memory
);
1783 if (mem
->user_ptr
) {
1784 *ppData
= mem
->user_ptr
;
1785 } else if (!mem
->map
) {
1786 result
= tu_bo_map(device
, &mem
->bo
);
1787 if (result
!= VK_SUCCESS
)
1789 *ppData
= mem
->map
= mem
->bo
.map
;
1798 return vk_error(device
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
1802 tu_UnmapMemory(VkDevice _device
, VkDeviceMemory _memory
)
1804 /* I do not see any unmapping done by the freedreno Gallium driver. */
1808 tu_FlushMappedMemoryRanges(VkDevice _device
,
1809 uint32_t memoryRangeCount
,
1810 const VkMappedMemoryRange
*pMemoryRanges
)
1816 tu_InvalidateMappedMemoryRanges(VkDevice _device
,
1817 uint32_t memoryRangeCount
,
1818 const VkMappedMemoryRange
*pMemoryRanges
)
1824 tu_GetBufferMemoryRequirements(VkDevice _device
,
1826 VkMemoryRequirements
*pMemoryRequirements
)
1828 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
1830 pMemoryRequirements
->memoryTypeBits
= 1;
1831 pMemoryRequirements
->alignment
= 64;
1832 pMemoryRequirements
->size
=
1833 align64(buffer
->size
, pMemoryRequirements
->alignment
);
1837 tu_GetBufferMemoryRequirements2(
1839 const VkBufferMemoryRequirementsInfo2
*pInfo
,
1840 VkMemoryRequirements2
*pMemoryRequirements
)
1842 tu_GetBufferMemoryRequirements(device
, pInfo
->buffer
,
1843 &pMemoryRequirements
->memoryRequirements
);
1847 tu_GetImageMemoryRequirements(VkDevice _device
,
1849 VkMemoryRequirements
*pMemoryRequirements
)
1851 TU_FROM_HANDLE(tu_image
, image
, _image
);
1853 pMemoryRequirements
->memoryTypeBits
= 1;
1854 pMemoryRequirements
->size
= image
->layout
.size
;
1855 pMemoryRequirements
->alignment
= image
->layout
.base_align
;
1859 tu_GetImageMemoryRequirements2(VkDevice device
,
1860 const VkImageMemoryRequirementsInfo2
*pInfo
,
1861 VkMemoryRequirements2
*pMemoryRequirements
)
1863 tu_GetImageMemoryRequirements(device
, pInfo
->image
,
1864 &pMemoryRequirements
->memoryRequirements
);
1868 tu_GetImageSparseMemoryRequirements(
1871 uint32_t *pSparseMemoryRequirementCount
,
1872 VkSparseImageMemoryRequirements
*pSparseMemoryRequirements
)
1878 tu_GetImageSparseMemoryRequirements2(
1880 const VkImageSparseMemoryRequirementsInfo2
*pInfo
,
1881 uint32_t *pSparseMemoryRequirementCount
,
1882 VkSparseImageMemoryRequirements2
*pSparseMemoryRequirements
)
1888 tu_GetDeviceMemoryCommitment(VkDevice device
,
1889 VkDeviceMemory memory
,
1890 VkDeviceSize
*pCommittedMemoryInBytes
)
1892 *pCommittedMemoryInBytes
= 0;
1896 tu_BindBufferMemory2(VkDevice device
,
1897 uint32_t bindInfoCount
,
1898 const VkBindBufferMemoryInfo
*pBindInfos
)
1900 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
1901 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
1902 TU_FROM_HANDLE(tu_buffer
, buffer
, pBindInfos
[i
].buffer
);
1905 buffer
->bo
= &mem
->bo
;
1906 buffer
->bo_offset
= pBindInfos
[i
].memoryOffset
;
1915 tu_BindBufferMemory(VkDevice device
,
1917 VkDeviceMemory memory
,
1918 VkDeviceSize memoryOffset
)
1920 const VkBindBufferMemoryInfo info
= {
1921 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
1924 .memoryOffset
= memoryOffset
1927 return tu_BindBufferMemory2(device
, 1, &info
);
1931 tu_BindImageMemory2(VkDevice device
,
1932 uint32_t bindInfoCount
,
1933 const VkBindImageMemoryInfo
*pBindInfos
)
1935 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
1936 TU_FROM_HANDLE(tu_image
, image
, pBindInfos
[i
].image
);
1937 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
1940 image
->bo
= &mem
->bo
;
1941 image
->bo_offset
= pBindInfos
[i
].memoryOffset
;
1944 image
->bo_offset
= 0;
1952 tu_BindImageMemory(VkDevice device
,
1954 VkDeviceMemory memory
,
1955 VkDeviceSize memoryOffset
)
1957 const VkBindImageMemoryInfo info
= {
1958 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
1961 .memoryOffset
= memoryOffset
1964 return tu_BindImageMemory2(device
, 1, &info
);
1968 tu_QueueBindSparse(VkQueue _queue
,
1969 uint32_t bindInfoCount
,
1970 const VkBindSparseInfo
*pBindInfo
,
1976 // Queue semaphore functions
1979 tu_CreateSemaphore(VkDevice _device
,
1980 const VkSemaphoreCreateInfo
*pCreateInfo
,
1981 const VkAllocationCallbacks
*pAllocator
,
1982 VkSemaphore
*pSemaphore
)
1984 TU_FROM_HANDLE(tu_device
, device
, _device
);
1986 struct tu_semaphore
*sem
=
1987 vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sem
), 8,
1988 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1990 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1992 *pSemaphore
= tu_semaphore_to_handle(sem
);
1997 tu_DestroySemaphore(VkDevice _device
,
1998 VkSemaphore _semaphore
,
1999 const VkAllocationCallbacks
*pAllocator
)
2001 TU_FROM_HANDLE(tu_device
, device
, _device
);
2002 TU_FROM_HANDLE(tu_semaphore
, sem
, _semaphore
);
2006 vk_free2(&device
->alloc
, pAllocator
, sem
);
2010 tu_CreateEvent(VkDevice _device
,
2011 const VkEventCreateInfo
*pCreateInfo
,
2012 const VkAllocationCallbacks
*pAllocator
,
2015 TU_FROM_HANDLE(tu_device
, device
, _device
);
2016 struct tu_event
*event
=
2017 vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*event
), 8,
2018 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2021 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2023 VkResult result
= tu_bo_init_new(device
, &event
->bo
, 0x1000);
2024 if (result
!= VK_SUCCESS
)
2027 result
= tu_bo_map(device
, &event
->bo
);
2028 if (result
!= VK_SUCCESS
)
2031 *pEvent
= tu_event_to_handle(event
);
2036 tu_bo_finish(device
, &event
->bo
);
2038 vk_free2(&device
->alloc
, pAllocator
, event
);
2039 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2043 tu_DestroyEvent(VkDevice _device
,
2045 const VkAllocationCallbacks
*pAllocator
)
2047 TU_FROM_HANDLE(tu_device
, device
, _device
);
2048 TU_FROM_HANDLE(tu_event
, event
, _event
);
2053 tu_bo_finish(device
, &event
->bo
);
2054 vk_free2(&device
->alloc
, pAllocator
, event
);
2058 tu_GetEventStatus(VkDevice _device
, VkEvent _event
)
2060 TU_FROM_HANDLE(tu_event
, event
, _event
);
2062 if (*(uint64_t*) event
->bo
.map
== 1)
2063 return VK_EVENT_SET
;
2064 return VK_EVENT_RESET
;
2068 tu_SetEvent(VkDevice _device
, VkEvent _event
)
2070 TU_FROM_HANDLE(tu_event
, event
, _event
);
2071 *(uint64_t*) event
->bo
.map
= 1;
2077 tu_ResetEvent(VkDevice _device
, VkEvent _event
)
2079 TU_FROM_HANDLE(tu_event
, event
, _event
);
2080 *(uint64_t*) event
->bo
.map
= 0;
2086 tu_CreateBuffer(VkDevice _device
,
2087 const VkBufferCreateInfo
*pCreateInfo
,
2088 const VkAllocationCallbacks
*pAllocator
,
2091 TU_FROM_HANDLE(tu_device
, device
, _device
);
2092 struct tu_buffer
*buffer
;
2094 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
2096 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
2097 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2099 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2101 buffer
->size
= pCreateInfo
->size
;
2102 buffer
->usage
= pCreateInfo
->usage
;
2103 buffer
->flags
= pCreateInfo
->flags
;
2105 *pBuffer
= tu_buffer_to_handle(buffer
);
2111 tu_DestroyBuffer(VkDevice _device
,
2113 const VkAllocationCallbacks
*pAllocator
)
2115 TU_FROM_HANDLE(tu_device
, device
, _device
);
2116 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
2121 vk_free2(&device
->alloc
, pAllocator
, buffer
);
2125 tu_CreateFramebuffer(VkDevice _device
,
2126 const VkFramebufferCreateInfo
*pCreateInfo
,
2127 const VkAllocationCallbacks
*pAllocator
,
2128 VkFramebuffer
*pFramebuffer
)
2130 TU_FROM_HANDLE(tu_device
, device
, _device
);
2131 struct tu_framebuffer
*framebuffer
;
2133 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
2135 size_t size
= sizeof(*framebuffer
) + sizeof(struct tu_attachment_info
) *
2136 pCreateInfo
->attachmentCount
;
2137 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
2138 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2139 if (framebuffer
== NULL
)
2140 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2142 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
2143 framebuffer
->width
= pCreateInfo
->width
;
2144 framebuffer
->height
= pCreateInfo
->height
;
2145 framebuffer
->layers
= pCreateInfo
->layers
;
2146 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
2147 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
2148 struct tu_image_view
*iview
= tu_image_view_from_handle(_iview
);
2149 framebuffer
->attachments
[i
].attachment
= iview
;
2152 *pFramebuffer
= tu_framebuffer_to_handle(framebuffer
);
2157 tu_DestroyFramebuffer(VkDevice _device
,
2159 const VkAllocationCallbacks
*pAllocator
)
2161 TU_FROM_HANDLE(tu_device
, device
, _device
);
2162 TU_FROM_HANDLE(tu_framebuffer
, fb
, _fb
);
2166 vk_free2(&device
->alloc
, pAllocator
, fb
);
2170 tu_init_sampler(struct tu_device
*device
,
2171 struct tu_sampler
*sampler
,
2172 const VkSamplerCreateInfo
*pCreateInfo
)
2174 const struct VkSamplerReductionModeCreateInfo
*reduction
=
2175 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_REDUCTION_MODE_CREATE_INFO
);
2176 const struct VkSamplerYcbcrConversionInfo
*ycbcr_conversion
=
2177 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_YCBCR_CONVERSION_INFO
);
2179 unsigned aniso
= pCreateInfo
->anisotropyEnable
?
2180 util_last_bit(MIN2((uint32_t)pCreateInfo
->maxAnisotropy
>> 1, 8)) : 0;
2181 bool miplinear
= (pCreateInfo
->mipmapMode
== VK_SAMPLER_MIPMAP_MODE_LINEAR
);
2182 float min_lod
= CLAMP(pCreateInfo
->minLod
, 0.0f
, 4095.0f
/ 256.0f
);
2183 float max_lod
= CLAMP(pCreateInfo
->maxLod
, 0.0f
, 4095.0f
/ 256.0f
);
2185 sampler
->descriptor
[0] =
2186 COND(miplinear
, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR
) |
2187 A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo
->magFilter
, aniso
)) |
2188 A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo
->minFilter
, aniso
)) |
2189 A6XX_TEX_SAMP_0_ANISO(aniso
) |
2190 A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo
->addressModeU
)) |
2191 A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo
->addressModeV
)) |
2192 A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo
->addressModeW
)) |
2193 A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo
->mipLodBias
);
2194 sampler
->descriptor
[1] =
2195 /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
2196 COND(pCreateInfo
->unnormalizedCoordinates
, A6XX_TEX_SAMP_1_UNNORM_COORDS
) |
2197 A6XX_TEX_SAMP_1_MIN_LOD(min_lod
) |
2198 A6XX_TEX_SAMP_1_MAX_LOD(max_lod
) |
2199 COND(pCreateInfo
->compareEnable
,
2200 A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo
->compareOp
)));
2201 /* This is an offset into the border_color BO, which we fill with all the
2202 * possible Vulkan border colors in the correct order, so we can just use
2203 * the Vulkan enum with no translation necessary.
2205 sampler
->descriptor
[2] =
2206 A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo
->borderColor
*
2207 sizeof(struct bcolor_entry
));
2208 sampler
->descriptor
[3] = 0;
2211 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(
2212 tu6_reduction_mode(reduction
->reductionMode
));
2215 sampler
->ycbcr_sampler
= ycbcr_conversion
?
2216 tu_sampler_ycbcr_conversion_from_handle(ycbcr_conversion
->conversion
) : NULL
;
2218 if (sampler
->ycbcr_sampler
&&
2219 sampler
->ycbcr_sampler
->chroma_filter
== VK_FILTER_LINEAR
) {
2220 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_CHROMA_LINEAR
;
2224 * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
2229 tu_CreateSampler(VkDevice _device
,
2230 const VkSamplerCreateInfo
*pCreateInfo
,
2231 const VkAllocationCallbacks
*pAllocator
,
2232 VkSampler
*pSampler
)
2234 TU_FROM_HANDLE(tu_device
, device
, _device
);
2235 struct tu_sampler
*sampler
;
2237 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
2239 sampler
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
2240 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2242 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2244 tu_init_sampler(device
, sampler
, pCreateInfo
);
2245 *pSampler
= tu_sampler_to_handle(sampler
);
2251 tu_DestroySampler(VkDevice _device
,
2253 const VkAllocationCallbacks
*pAllocator
)
2255 TU_FROM_HANDLE(tu_device
, device
, _device
);
2256 TU_FROM_HANDLE(tu_sampler
, sampler
, _sampler
);
2260 vk_free2(&device
->alloc
, pAllocator
, sampler
);
2263 /* vk_icd.h does not declare this function, so we declare it here to
2264 * suppress Wmissing-prototypes.
2266 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2267 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
2269 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2270 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
2272 /* For the full details on loader interface versioning, see
2273 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2274 * What follows is a condensed summary, to help you navigate the large and
2275 * confusing official doc.
2277 * - Loader interface v0 is incompatible with later versions. We don't
2280 * - In loader interface v1:
2281 * - The first ICD entrypoint called by the loader is
2282 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2284 * - The ICD must statically expose no other Vulkan symbol unless it
2285 * is linked with -Bsymbolic.
2286 * - Each dispatchable Vulkan handle created by the ICD must be
2287 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2288 * ICD must initialize VK_LOADER_DATA.loadMagic to
2290 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2291 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2292 * such loader-managed surfaces.
2294 * - Loader interface v2 differs from v1 in:
2295 * - The first ICD entrypoint called by the loader is
2296 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2297 * statically expose this entrypoint.
2299 * - Loader interface v3 differs from v2 in:
2300 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2301 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2302 * because the loader no longer does so.
2304 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);
2309 tu_GetMemoryFdKHR(VkDevice _device
,
2310 const VkMemoryGetFdInfoKHR
*pGetFdInfo
,
2313 TU_FROM_HANDLE(tu_device
, device
, _device
);
2314 TU_FROM_HANDLE(tu_device_memory
, memory
, pGetFdInfo
->memory
);
2316 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
2318 /* At the moment, we support only the below handle types. */
2319 assert(pGetFdInfo
->handleType
==
2320 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
2321 pGetFdInfo
->handleType
==
2322 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2324 int prime_fd
= tu_bo_export_dmabuf(device
, &memory
->bo
);
2326 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2333 tu_GetMemoryFdPropertiesKHR(VkDevice _device
,
2334 VkExternalMemoryHandleTypeFlagBits handleType
,
2336 VkMemoryFdPropertiesKHR
*pMemoryFdProperties
)
2338 assert(handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2339 pMemoryFdProperties
->memoryTypeBits
= 1;
2344 tu_GetPhysicalDeviceExternalSemaphoreProperties(
2345 VkPhysicalDevice physicalDevice
,
2346 const VkPhysicalDeviceExternalSemaphoreInfo
*pExternalSemaphoreInfo
,
2347 VkExternalSemaphoreProperties
*pExternalSemaphoreProperties
)
2349 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= 0;
2350 pExternalSemaphoreProperties
->compatibleHandleTypes
= 0;
2351 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= 0;
2355 tu_GetPhysicalDeviceExternalFenceProperties(
2356 VkPhysicalDevice physicalDevice
,
2357 const VkPhysicalDeviceExternalFenceInfo
*pExternalFenceInfo
,
2358 VkExternalFenceProperties
*pExternalFenceProperties
)
2360 pExternalFenceProperties
->exportFromImportedHandleTypes
= 0;
2361 pExternalFenceProperties
->compatibleHandleTypes
= 0;
2362 pExternalFenceProperties
->externalFenceFeatures
= 0;
2366 tu_CreateDebugReportCallbackEXT(
2367 VkInstance _instance
,
2368 const VkDebugReportCallbackCreateInfoEXT
*pCreateInfo
,
2369 const VkAllocationCallbacks
*pAllocator
,
2370 VkDebugReportCallbackEXT
*pCallback
)
2372 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2373 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
2374 pCreateInfo
, pAllocator
,
2375 &instance
->alloc
, pCallback
);
2379 tu_DestroyDebugReportCallbackEXT(VkInstance _instance
,
2380 VkDebugReportCallbackEXT _callback
,
2381 const VkAllocationCallbacks
*pAllocator
)
2383 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2384 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
2385 _callback
, pAllocator
, &instance
->alloc
);
2389 tu_DebugReportMessageEXT(VkInstance _instance
,
2390 VkDebugReportFlagsEXT flags
,
2391 VkDebugReportObjectTypeEXT objectType
,
2394 int32_t messageCode
,
2395 const char *pLayerPrefix
,
2396 const char *pMessage
)
2398 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2399 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
2400 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
2404 tu_GetDeviceGroupPeerMemoryFeatures(
2407 uint32_t localDeviceIndex
,
2408 uint32_t remoteDeviceIndex
,
2409 VkPeerMemoryFeatureFlags
*pPeerMemoryFeatures
)
2411 assert(localDeviceIndex
== remoteDeviceIndex
);
2413 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
2414 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
2415 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
2416 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
2419 void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
2420 VkPhysicalDevice physicalDevice
,
2421 VkSampleCountFlagBits samples
,
2422 VkMultisamplePropertiesEXT
* pMultisampleProperties
)
2424 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
2426 if (samples
<= VK_SAMPLE_COUNT_4_BIT
&& pdevice
->supported_extensions
.EXT_sample_locations
)
2427 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 1, 1 };
2429 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 0, 0 };