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
= false,
594 .drawIndirectFirstInstance
= false,
596 .depthBiasClamp
= false,
597 .fillModeNonSolid
= false,
598 .depthBounds
= 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_VARIABLE_POINTERS_FEATURES
: {
640 VkPhysicalDeviceVariablePointersFeatures
*features
= (void *) ext
;
641 features
->variablePointersStorageBuffer
= false;
642 features
->variablePointers
= false;
645 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES
: {
646 VkPhysicalDeviceMultiviewFeatures
*features
=
647 (VkPhysicalDeviceMultiviewFeatures
*) ext
;
648 features
->multiview
= false;
649 features
->multiviewGeometryShader
= false;
650 features
->multiviewTessellationShader
= false;
653 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES
: {
654 VkPhysicalDeviceShaderDrawParametersFeatures
*features
=
655 (VkPhysicalDeviceShaderDrawParametersFeatures
*) ext
;
656 features
->shaderDrawParameters
= false;
659 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES
: {
660 VkPhysicalDeviceProtectedMemoryFeatures
*features
=
661 (VkPhysicalDeviceProtectedMemoryFeatures
*) ext
;
662 features
->protectedMemory
= false;
665 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES
: {
666 VkPhysicalDevice16BitStorageFeatures
*features
=
667 (VkPhysicalDevice16BitStorageFeatures
*) ext
;
668 features
->storageBuffer16BitAccess
= false;
669 features
->uniformAndStorageBuffer16BitAccess
= false;
670 features
->storagePushConstant16
= false;
671 features
->storageInputOutput16
= false;
674 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES
: {
675 VkPhysicalDeviceSamplerYcbcrConversionFeatures
*features
=
676 (VkPhysicalDeviceSamplerYcbcrConversionFeatures
*) ext
;
677 features
->samplerYcbcrConversion
= true;
680 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT
: {
681 VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*features
=
682 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*) ext
;
683 features
->shaderInputAttachmentArrayDynamicIndexing
= false;
684 features
->shaderUniformTexelBufferArrayDynamicIndexing
= false;
685 features
->shaderStorageTexelBufferArrayDynamicIndexing
= false;
686 features
->shaderUniformBufferArrayNonUniformIndexing
= false;
687 features
->shaderSampledImageArrayNonUniformIndexing
= false;
688 features
->shaderStorageBufferArrayNonUniformIndexing
= false;
689 features
->shaderStorageImageArrayNonUniformIndexing
= false;
690 features
->shaderInputAttachmentArrayNonUniformIndexing
= false;
691 features
->shaderUniformTexelBufferArrayNonUniformIndexing
= false;
692 features
->shaderStorageTexelBufferArrayNonUniformIndexing
= false;
693 features
->descriptorBindingUniformBufferUpdateAfterBind
= false;
694 features
->descriptorBindingSampledImageUpdateAfterBind
= false;
695 features
->descriptorBindingStorageImageUpdateAfterBind
= false;
696 features
->descriptorBindingStorageBufferUpdateAfterBind
= false;
697 features
->descriptorBindingUniformTexelBufferUpdateAfterBind
= false;
698 features
->descriptorBindingStorageTexelBufferUpdateAfterBind
= false;
699 features
->descriptorBindingUpdateUnusedWhilePending
= false;
700 features
->descriptorBindingPartiallyBound
= false;
701 features
->descriptorBindingVariableDescriptorCount
= false;
702 features
->runtimeDescriptorArray
= false;
705 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT
: {
706 VkPhysicalDeviceConditionalRenderingFeaturesEXT
*features
=
707 (VkPhysicalDeviceConditionalRenderingFeaturesEXT
*) ext
;
708 features
->conditionalRendering
= false;
709 features
->inheritedConditionalRendering
= false;
712 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT
: {
713 VkPhysicalDeviceTransformFeedbackFeaturesEXT
*features
=
714 (VkPhysicalDeviceTransformFeedbackFeaturesEXT
*) ext
;
715 features
->transformFeedback
= true;
716 features
->geometryStreams
= false;
723 return tu_GetPhysicalDeviceFeatures(physicalDevice
, &pFeatures
->features
);
727 tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice
,
728 VkPhysicalDeviceProperties
*pProperties
)
730 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
731 VkSampleCountFlags sample_counts
=
732 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
734 /* I have no idea what the maximum size is, but the hardware supports very
735 * large numbers of descriptors (at least 2^16). This limit is based on
736 * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
737 * we don't have to think about what to do if that overflows, but really
738 * nothing is likely to get close to this.
740 const size_t max_descriptor_set_size
= (1 << 28) / A6XX_TEX_CONST_DWORDS
;
742 VkPhysicalDeviceLimits limits
= {
743 .maxImageDimension1D
= (1 << 14),
744 .maxImageDimension2D
= (1 << 14),
745 .maxImageDimension3D
= (1 << 11),
746 .maxImageDimensionCube
= (1 << 14),
747 .maxImageArrayLayers
= (1 << 11),
748 .maxTexelBufferElements
= 128 * 1024 * 1024,
749 .maxUniformBufferRange
= MAX_UNIFORM_BUFFER_RANGE
,
750 .maxStorageBufferRange
= MAX_STORAGE_BUFFER_RANGE
,
751 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
752 .maxMemoryAllocationCount
= UINT32_MAX
,
753 .maxSamplerAllocationCount
= 64 * 1024,
754 .bufferImageGranularity
= 64, /* A cache line */
755 .sparseAddressSpaceSize
= 0xffffffffu
, /* buffer max size */
756 .maxBoundDescriptorSets
= MAX_SETS
,
757 .maxPerStageDescriptorSamplers
= max_descriptor_set_size
,
758 .maxPerStageDescriptorUniformBuffers
= max_descriptor_set_size
,
759 .maxPerStageDescriptorStorageBuffers
= max_descriptor_set_size
,
760 .maxPerStageDescriptorSampledImages
= max_descriptor_set_size
,
761 .maxPerStageDescriptorStorageImages
= max_descriptor_set_size
,
762 .maxPerStageDescriptorInputAttachments
= MAX_RTS
,
763 .maxPerStageResources
= max_descriptor_set_size
,
764 .maxDescriptorSetSamplers
= max_descriptor_set_size
,
765 .maxDescriptorSetUniformBuffers
= max_descriptor_set_size
,
766 .maxDescriptorSetUniformBuffersDynamic
= MAX_DYNAMIC_UNIFORM_BUFFERS
,
767 .maxDescriptorSetStorageBuffers
= max_descriptor_set_size
,
768 .maxDescriptorSetStorageBuffersDynamic
= MAX_DYNAMIC_STORAGE_BUFFERS
,
769 .maxDescriptorSetSampledImages
= max_descriptor_set_size
,
770 .maxDescriptorSetStorageImages
= max_descriptor_set_size
,
771 .maxDescriptorSetInputAttachments
= MAX_RTS
,
772 .maxVertexInputAttributes
= 32,
773 .maxVertexInputBindings
= 32,
774 .maxVertexInputAttributeOffset
= 4095,
775 .maxVertexInputBindingStride
= 2048,
776 .maxVertexOutputComponents
= 128,
777 .maxTessellationGenerationLevel
= 64,
778 .maxTessellationPatchSize
= 32,
779 .maxTessellationControlPerVertexInputComponents
= 128,
780 .maxTessellationControlPerVertexOutputComponents
= 128,
781 .maxTessellationControlPerPatchOutputComponents
= 120,
782 .maxTessellationControlTotalOutputComponents
= 4096,
783 .maxTessellationEvaluationInputComponents
= 128,
784 .maxTessellationEvaluationOutputComponents
= 128,
785 .maxGeometryShaderInvocations
= 32,
786 .maxGeometryInputComponents
= 64,
787 .maxGeometryOutputComponents
= 128,
788 .maxGeometryOutputVertices
= 256,
789 .maxGeometryTotalOutputComponents
= 1024,
790 .maxFragmentInputComponents
= 124,
791 .maxFragmentOutputAttachments
= 8,
792 .maxFragmentDualSrcAttachments
= 1,
793 .maxFragmentCombinedOutputResources
= 8,
794 .maxComputeSharedMemorySize
= 32768,
795 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
796 .maxComputeWorkGroupInvocations
= 2048,
797 .maxComputeWorkGroupSize
= { 2048, 2048, 2048 },
798 .subPixelPrecisionBits
= 8,
799 .subTexelPrecisionBits
= 8,
800 .mipmapPrecisionBits
= 8,
801 .maxDrawIndexedIndexValue
= UINT32_MAX
,
802 .maxDrawIndirectCount
= UINT32_MAX
,
803 .maxSamplerLodBias
= 4095.0 / 256.0, /* [-16, 15.99609375] */
804 .maxSamplerAnisotropy
= 16,
805 .maxViewports
= MAX_VIEWPORTS
,
806 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
807 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
808 .viewportSubPixelBits
= 8,
809 .minMemoryMapAlignment
= 4096, /* A page */
810 .minTexelBufferOffsetAlignment
= 64,
811 .minUniformBufferOffsetAlignment
= 64,
812 .minStorageBufferOffsetAlignment
= 64,
813 .minTexelOffset
= -16,
814 .maxTexelOffset
= 15,
815 .minTexelGatherOffset
= -32,
816 .maxTexelGatherOffset
= 31,
817 .minInterpolationOffset
= -0.5,
818 .maxInterpolationOffset
= 0.4375,
819 .subPixelInterpolationOffsetBits
= 4,
820 .maxFramebufferWidth
= (1 << 14),
821 .maxFramebufferHeight
= (1 << 14),
822 .maxFramebufferLayers
= (1 << 10),
823 .framebufferColorSampleCounts
= sample_counts
,
824 .framebufferDepthSampleCounts
= sample_counts
,
825 .framebufferStencilSampleCounts
= sample_counts
,
826 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
827 .maxColorAttachments
= MAX_RTS
,
828 .sampledImageColorSampleCounts
= sample_counts
,
829 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
830 .sampledImageDepthSampleCounts
= sample_counts
,
831 .sampledImageStencilSampleCounts
= sample_counts
,
832 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
833 .maxSampleMaskWords
= 1,
834 .timestampComputeAndGraphics
= true,
835 .timestampPeriod
= 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
836 .maxClipDistances
= 8,
837 .maxCullDistances
= 8,
838 .maxCombinedClipAndCullDistances
= 8,
839 .discreteQueuePriorities
= 1,
840 .pointSizeRange
= { 0.125, 255.875 },
841 .lineWidthRange
= { 0.0, 7.9921875 },
842 .pointSizeGranularity
= (1.0 / 8.0),
843 .lineWidthGranularity
= (1.0 / 128.0),
844 .strictLines
= false, /* FINISHME */
845 .standardSampleLocations
= true,
846 .optimalBufferCopyOffsetAlignment
= 128,
847 .optimalBufferCopyRowPitchAlignment
= 128,
848 .nonCoherentAtomSize
= 64,
851 *pProperties
= (VkPhysicalDeviceProperties
) {
852 .apiVersion
= tu_physical_device_api_version(pdevice
),
853 .driverVersion
= vk_get_driver_version(),
854 .vendorID
= 0, /* TODO */
856 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
858 .sparseProperties
= { 0 },
861 strcpy(pProperties
->deviceName
, pdevice
->name
);
862 memcpy(pProperties
->pipelineCacheUUID
, pdevice
->cache_uuid
, VK_UUID_SIZE
);
866 tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice
,
867 VkPhysicalDeviceProperties2
*pProperties
)
869 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
870 tu_GetPhysicalDeviceProperties(physicalDevice
, &pProperties
->properties
);
872 vk_foreach_struct(ext
, pProperties
->pNext
)
874 switch (ext
->sType
) {
875 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR
: {
876 VkPhysicalDevicePushDescriptorPropertiesKHR
*properties
=
877 (VkPhysicalDevicePushDescriptorPropertiesKHR
*) ext
;
878 properties
->maxPushDescriptors
= MAX_PUSH_DESCRIPTORS
;
881 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES
: {
882 VkPhysicalDeviceIDProperties
*properties
=
883 (VkPhysicalDeviceIDProperties
*) ext
;
884 memcpy(properties
->driverUUID
, pdevice
->driver_uuid
, VK_UUID_SIZE
);
885 memcpy(properties
->deviceUUID
, pdevice
->device_uuid
, VK_UUID_SIZE
);
886 properties
->deviceLUIDValid
= false;
889 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES
: {
890 VkPhysicalDeviceMultiviewProperties
*properties
=
891 (VkPhysicalDeviceMultiviewProperties
*) ext
;
892 properties
->maxMultiviewViewCount
= MAX_VIEWS
;
893 properties
->maxMultiviewInstanceIndex
= INT_MAX
;
896 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES
: {
897 VkPhysicalDevicePointClippingProperties
*properties
=
898 (VkPhysicalDevicePointClippingProperties
*) ext
;
899 properties
->pointClippingBehavior
=
900 VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES
;
903 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES
: {
904 VkPhysicalDeviceMaintenance3Properties
*properties
=
905 (VkPhysicalDeviceMaintenance3Properties
*) ext
;
906 /* Make sure everything is addressable by a signed 32-bit int, and
907 * our largest descriptors are 96 bytes. */
908 properties
->maxPerSetDescriptors
= (1ull << 31) / 96;
909 /* Our buffer size fields allow only this much */
910 properties
->maxMemoryAllocationSize
= 0xFFFFFFFFull
;
913 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT
: {
914 VkPhysicalDeviceTransformFeedbackPropertiesEXT
*properties
=
915 (VkPhysicalDeviceTransformFeedbackPropertiesEXT
*)ext
;
917 properties
->maxTransformFeedbackStreams
= IR3_MAX_SO_STREAMS
;
918 properties
->maxTransformFeedbackBuffers
= IR3_MAX_SO_BUFFERS
;
919 properties
->maxTransformFeedbackBufferSize
= UINT32_MAX
;
920 properties
->maxTransformFeedbackStreamDataSize
= 512;
921 properties
->maxTransformFeedbackBufferDataSize
= 512;
922 properties
->maxTransformFeedbackBufferDataStride
= 512;
923 properties
->transformFeedbackQueries
= true;
924 properties
->transformFeedbackStreamsLinesTriangles
= false;
925 properties
->transformFeedbackRasterizationStreamSelect
= false;
926 properties
->transformFeedbackDraw
= true;
929 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT
: {
930 VkPhysicalDeviceSampleLocationsPropertiesEXT
*properties
=
931 (VkPhysicalDeviceSampleLocationsPropertiesEXT
*)ext
;
932 properties
->sampleLocationSampleCounts
= 0;
933 if (pdevice
->supported_extensions
.EXT_sample_locations
) {
934 properties
->sampleLocationSampleCounts
=
935 VK_SAMPLE_COUNT_1_BIT
| VK_SAMPLE_COUNT_2_BIT
| VK_SAMPLE_COUNT_4_BIT
;
937 properties
->maxSampleLocationGridSize
= (VkExtent2D
) { 1 , 1 };
938 properties
->sampleLocationCoordinateRange
[0] = 0.0f
;
939 properties
->sampleLocationCoordinateRange
[1] = 0.9375f
;
940 properties
->sampleLocationSubPixelBits
= 4;
941 properties
->variableSampleLocations
= true;
944 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES
: {
945 VkPhysicalDeviceSamplerFilterMinmaxProperties
*properties
=
946 (VkPhysicalDeviceSamplerFilterMinmaxProperties
*)ext
;
947 properties
->filterMinmaxImageComponentMapping
= true;
948 properties
->filterMinmaxSingleComponentFormats
= true;
951 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES
: {
952 VkPhysicalDeviceSubgroupProperties
*properties
=
953 (VkPhysicalDeviceSubgroupProperties
*)ext
;
954 properties
->subgroupSize
= 64;
955 properties
->supportedStages
= VK_SHADER_STAGE_COMPUTE_BIT
;
956 properties
->supportedOperations
= VK_SUBGROUP_FEATURE_BASIC_BIT
|
957 VK_SUBGROUP_FEATURE_VOTE_BIT
;
958 properties
->quadOperationsInAllStages
= false;
968 static const VkQueueFamilyProperties tu_queue_family_properties
= {
970 VK_QUEUE_GRAPHICS_BIT
| VK_QUEUE_COMPUTE_BIT
| VK_QUEUE_TRANSFER_BIT
,
972 .timestampValidBits
= 48,
973 .minImageTransferGranularity
= { 1, 1, 1 },
977 tu_GetPhysicalDeviceQueueFamilyProperties(
978 VkPhysicalDevice physicalDevice
,
979 uint32_t *pQueueFamilyPropertyCount
,
980 VkQueueFamilyProperties
*pQueueFamilyProperties
)
982 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
984 vk_outarray_append(&out
, p
) { *p
= tu_queue_family_properties
; }
988 tu_GetPhysicalDeviceQueueFamilyProperties2(
989 VkPhysicalDevice physicalDevice
,
990 uint32_t *pQueueFamilyPropertyCount
,
991 VkQueueFamilyProperties2
*pQueueFamilyProperties
)
993 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
995 vk_outarray_append(&out
, p
)
997 p
->queueFamilyProperties
= tu_queue_family_properties
;
1002 tu_get_system_heap_size()
1004 struct sysinfo info
;
1007 uint64_t total_ram
= (uint64_t) info
.totalram
* (uint64_t) info
.mem_unit
;
1009 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
1010 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
1012 uint64_t available_ram
;
1013 if (total_ram
<= 4ull * 1024ull * 1024ull * 1024ull)
1014 available_ram
= total_ram
/ 2;
1016 available_ram
= total_ram
* 3 / 4;
1018 return available_ram
;
1022 tu_GetPhysicalDeviceMemoryProperties(
1023 VkPhysicalDevice physicalDevice
,
1024 VkPhysicalDeviceMemoryProperties
*pMemoryProperties
)
1026 pMemoryProperties
->memoryHeapCount
= 1;
1027 pMemoryProperties
->memoryHeaps
[0].size
= tu_get_system_heap_size();
1028 pMemoryProperties
->memoryHeaps
[0].flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
;
1030 pMemoryProperties
->memoryTypeCount
= 1;
1031 pMemoryProperties
->memoryTypes
[0].propertyFlags
=
1032 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
1033 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
1034 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
;
1035 pMemoryProperties
->memoryTypes
[0].heapIndex
= 0;
1039 tu_GetPhysicalDeviceMemoryProperties2(
1040 VkPhysicalDevice physicalDevice
,
1041 VkPhysicalDeviceMemoryProperties2
*pMemoryProperties
)
1043 return tu_GetPhysicalDeviceMemoryProperties(
1044 physicalDevice
, &pMemoryProperties
->memoryProperties
);
1048 tu_queue_init(struct tu_device
*device
,
1049 struct tu_queue
*queue
,
1050 uint32_t queue_family_index
,
1052 VkDeviceQueueCreateFlags flags
)
1054 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1055 queue
->device
= device
;
1056 queue
->queue_family_index
= queue_family_index
;
1057 queue
->queue_idx
= idx
;
1058 queue
->flags
= flags
;
1060 int ret
= tu_drm_submitqueue_new(device
, 0, &queue
->msm_queue_id
);
1062 return VK_ERROR_INITIALIZATION_FAILED
;
1064 tu_fence_init(&queue
->submit_fence
, false);
1070 tu_queue_finish(struct tu_queue
*queue
)
1072 tu_fence_finish(&queue
->submit_fence
);
1073 tu_drm_submitqueue_close(queue
->device
, queue
->msm_queue_id
);
1077 tu_get_device_extension_index(const char *name
)
1079 for (unsigned i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; ++i
) {
1080 if (strcmp(name
, tu_device_extensions
[i
].extensionName
) == 0)
1086 struct PACKED bcolor_entry
{
1098 uint32_t z24
; /* also s8? */
1099 uint16_t srgb
[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
1101 } border_color
[] = {
1102 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = {},
1103 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = {},
1104 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = {
1105 .fp32
[3] = 0x3f800000,
1113 .rgb10a2
= 0xc0000000,
1116 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = {
1120 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = {
1121 .fp32
[0 ... 3] = 0x3f800000,
1122 .ui16
[0 ... 3] = 0xffff,
1123 .si16
[0 ... 3] = 0x7fff,
1124 .fp16
[0 ... 3] = 0x3c00,
1128 .ui8
[0 ... 3] = 0xff,
1129 .si8
[0 ... 3] = 0x7f,
1130 .rgb10a2
= 0xffffffff,
1132 .srgb
[0 ... 3] = 0x3c00,
1134 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = {
1142 tu_CreateDevice(VkPhysicalDevice physicalDevice
,
1143 const VkDeviceCreateInfo
*pCreateInfo
,
1144 const VkAllocationCallbacks
*pAllocator
,
1147 TU_FROM_HANDLE(tu_physical_device
, physical_device
, physicalDevice
);
1149 struct tu_device
*device
;
1151 /* Check enabled features */
1152 if (pCreateInfo
->pEnabledFeatures
) {
1153 VkPhysicalDeviceFeatures supported_features
;
1154 tu_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
1155 VkBool32
*supported_feature
= (VkBool32
*) &supported_features
;
1156 VkBool32
*enabled_feature
= (VkBool32
*) pCreateInfo
->pEnabledFeatures
;
1157 unsigned num_features
=
1158 sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
1159 for (uint32_t i
= 0; i
< num_features
; i
++) {
1160 if (enabled_feature
[i
] && !supported_feature
[i
])
1161 return vk_error(physical_device
->instance
,
1162 VK_ERROR_FEATURE_NOT_PRESENT
);
1166 device
= vk_zalloc2(&physical_device
->instance
->alloc
, pAllocator
,
1167 sizeof(*device
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1169 return vk_error(physical_device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1171 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1172 device
->instance
= physical_device
->instance
;
1173 device
->physical_device
= physical_device
;
1176 device
->alloc
= *pAllocator
;
1178 device
->alloc
= physical_device
->instance
->alloc
;
1180 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
1181 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
1182 int index
= tu_get_device_extension_index(ext_name
);
1184 !physical_device
->supported_extensions
.extensions
[index
]) {
1185 vk_free(&device
->alloc
, device
);
1186 return vk_error(physical_device
->instance
,
1187 VK_ERROR_EXTENSION_NOT_PRESENT
);
1190 device
->enabled_extensions
.extensions
[index
] = true;
1193 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
1194 const VkDeviceQueueCreateInfo
*queue_create
=
1195 &pCreateInfo
->pQueueCreateInfos
[i
];
1196 uint32_t qfi
= queue_create
->queueFamilyIndex
;
1197 device
->queues
[qfi
] = vk_alloc(
1198 &device
->alloc
, queue_create
->queueCount
* sizeof(struct tu_queue
),
1199 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1200 if (!device
->queues
[qfi
]) {
1201 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
1205 memset(device
->queues
[qfi
], 0,
1206 queue_create
->queueCount
* sizeof(struct tu_queue
));
1208 device
->queue_count
[qfi
] = queue_create
->queueCount
;
1210 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
1211 result
= tu_queue_init(device
, &device
->queues
[qfi
][q
], qfi
, q
,
1212 queue_create
->flags
);
1213 if (result
!= VK_SUCCESS
)
1218 device
->compiler
= ir3_compiler_create(NULL
, physical_device
->gpu_id
);
1219 if (!device
->compiler
)
1222 #define VSC_DRAW_STRM_SIZE(pitch) ((pitch) * 32 + 0x100) /* extra size to store VSC_SIZE */
1223 #define VSC_PRIM_STRM_SIZE(pitch) ((pitch) * 32)
1225 device
->vsc_draw_strm_pitch
= 0x440 * 4;
1226 device
->vsc_prim_strm_pitch
= 0x1040 * 4;
1228 result
= tu_bo_init_new(device
, &device
->vsc_draw_strm
, VSC_DRAW_STRM_SIZE(device
->vsc_draw_strm_pitch
));
1229 if (result
!= VK_SUCCESS
)
1232 result
= tu_bo_init_new(device
, &device
->vsc_prim_strm
, VSC_PRIM_STRM_SIZE(device
->vsc_prim_strm_pitch
));
1233 if (result
!= VK_SUCCESS
)
1234 goto fail_vsc_data2
;
1236 STATIC_ASSERT(sizeof(struct bcolor_entry
) == 128);
1237 result
= tu_bo_init_new(device
, &device
->border_color
, sizeof(border_color
));
1238 if (result
!= VK_SUCCESS
)
1239 goto fail_border_color
;
1241 result
= tu_bo_map(device
, &device
->border_color
);
1242 if (result
!= VK_SUCCESS
)
1243 goto fail_border_color_map
;
1245 memcpy(device
->border_color
.map
, border_color
, sizeof(border_color
));
1247 VkPipelineCacheCreateInfo ci
;
1248 ci
.sType
= VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
;
1251 ci
.pInitialData
= NULL
;
1252 ci
.initialDataSize
= 0;
1255 tu_CreatePipelineCache(tu_device_to_handle(device
), &ci
, NULL
, &pc
);
1256 if (result
!= VK_SUCCESS
)
1257 goto fail_pipeline_cache
;
1259 device
->mem_cache
= tu_pipeline_cache_from_handle(pc
);
1261 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++)
1262 mtx_init(&device
->scratch_bos
[i
].construct_mtx
, mtx_plain
);
1264 *pDevice
= tu_device_to_handle(device
);
1267 fail_pipeline_cache
:
1268 fail_border_color_map
:
1269 tu_bo_finish(device
, &device
->border_color
);
1272 tu_bo_finish(device
, &device
->vsc_prim_strm
);
1275 tu_bo_finish(device
, &device
->vsc_draw_strm
);
1278 ralloc_free(device
->compiler
);
1281 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1282 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1283 tu_queue_finish(&device
->queues
[i
][q
]);
1284 if (device
->queue_count
[i
])
1285 vk_free(&device
->alloc
, device
->queues
[i
]);
1288 vk_free(&device
->alloc
, device
);
1293 tu_DestroyDevice(VkDevice _device
, const VkAllocationCallbacks
*pAllocator
)
1295 TU_FROM_HANDLE(tu_device
, device
, _device
);
1300 tu_bo_finish(device
, &device
->vsc_draw_strm
);
1301 tu_bo_finish(device
, &device
->vsc_prim_strm
);
1303 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1304 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1305 tu_queue_finish(&device
->queues
[i
][q
]);
1306 if (device
->queue_count
[i
])
1307 vk_free(&device
->alloc
, device
->queues
[i
]);
1310 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++) {
1311 if (device
->scratch_bos
[i
].initialized
)
1312 tu_bo_finish(device
, &device
->scratch_bos
[i
].bo
);
1315 /* the compiler does not use pAllocator */
1316 ralloc_free(device
->compiler
);
1318 VkPipelineCache pc
= tu_pipeline_cache_to_handle(device
->mem_cache
);
1319 tu_DestroyPipelineCache(tu_device_to_handle(device
), pc
, NULL
);
1321 vk_free(&device
->alloc
, device
);
1325 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
)
1327 unsigned size_log2
= MAX2(util_logbase2_ceil64(size
), MIN_SCRATCH_BO_SIZE_LOG2
);
1328 unsigned index
= size_log2
- MIN_SCRATCH_BO_SIZE_LOG2
;
1329 assert(index
< ARRAY_SIZE(dev
->scratch_bos
));
1331 for (unsigned i
= index
; i
< ARRAY_SIZE(dev
->scratch_bos
); i
++) {
1332 if (p_atomic_read(&dev
->scratch_bos
[i
].initialized
)) {
1333 /* Fast path: just return the already-allocated BO. */
1334 *bo
= &dev
->scratch_bos
[i
].bo
;
1339 /* Slow path: actually allocate the BO. We take a lock because the process
1340 * of allocating it is slow, and we don't want to block the CPU while it
1343 mtx_lock(&dev
->scratch_bos
[index
].construct_mtx
);
1345 /* Another thread may have allocated it already while we were waiting on
1346 * the lock. We need to check this in order to avoid double-allocating.
1348 if (dev
->scratch_bos
[index
].initialized
) {
1349 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1350 *bo
= &dev
->scratch_bos
[index
].bo
;
1354 unsigned bo_size
= 1ull << size_log2
;
1355 VkResult result
= tu_bo_init_new(dev
, &dev
->scratch_bos
[index
].bo
, bo_size
);
1356 if (result
!= VK_SUCCESS
) {
1357 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1361 p_atomic_set(&dev
->scratch_bos
[index
].initialized
, true);
1363 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1365 *bo
= &dev
->scratch_bos
[index
].bo
;
1370 tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount
,
1371 VkLayerProperties
*pProperties
)
1373 *pPropertyCount
= 0;
1378 tu_EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice
,
1379 uint32_t *pPropertyCount
,
1380 VkLayerProperties
*pProperties
)
1382 *pPropertyCount
= 0;
1387 tu_GetDeviceQueue2(VkDevice _device
,
1388 const VkDeviceQueueInfo2
*pQueueInfo
,
1391 TU_FROM_HANDLE(tu_device
, device
, _device
);
1392 struct tu_queue
*queue
;
1395 &device
->queues
[pQueueInfo
->queueFamilyIndex
][pQueueInfo
->queueIndex
];
1396 if (pQueueInfo
->flags
!= queue
->flags
) {
1397 /* From the Vulkan 1.1.70 spec:
1399 * "The queue returned by vkGetDeviceQueue2 must have the same
1400 * flags value from this structure as that used at device
1401 * creation time in a VkDeviceQueueCreateInfo instance. If no
1402 * matching flags were specified at device creation time then
1403 * pQueue will return VK_NULL_HANDLE."
1405 *pQueue
= VK_NULL_HANDLE
;
1409 *pQueue
= tu_queue_to_handle(queue
);
1413 tu_GetDeviceQueue(VkDevice _device
,
1414 uint32_t queueFamilyIndex
,
1415 uint32_t queueIndex
,
1418 const VkDeviceQueueInfo2 info
=
1419 (VkDeviceQueueInfo2
) { .sType
= VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2
,
1420 .queueFamilyIndex
= queueFamilyIndex
,
1421 .queueIndex
= queueIndex
};
1423 tu_GetDeviceQueue2(_device
, &info
, pQueue
);
1427 tu_QueueSubmit(VkQueue _queue
,
1428 uint32_t submitCount
,
1429 const VkSubmitInfo
*pSubmits
,
1432 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1434 for (uint32_t i
= 0; i
< submitCount
; ++i
) {
1435 const VkSubmitInfo
*submit
= pSubmits
+ i
;
1436 const bool last_submit
= (i
== submitCount
- 1);
1437 struct tu_bo_list bo_list
;
1438 tu_bo_list_init(&bo_list
);
1440 uint32_t entry_count
= 0;
1441 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1442 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1443 entry_count
+= cmdbuf
->cs
.entry_count
;
1446 struct drm_msm_gem_submit_cmd cmds
[entry_count
];
1447 uint32_t entry_idx
= 0;
1448 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1449 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1450 struct tu_cs
*cs
= &cmdbuf
->cs
;
1451 for (unsigned i
= 0; i
< cs
->entry_count
; ++i
, ++entry_idx
) {
1452 cmds
[entry_idx
].type
= MSM_SUBMIT_CMD_BUF
;
1453 cmds
[entry_idx
].submit_idx
=
1454 tu_bo_list_add(&bo_list
, cs
->entries
[i
].bo
,
1455 MSM_SUBMIT_BO_READ
| MSM_SUBMIT_BO_DUMP
);
1456 cmds
[entry_idx
].submit_offset
= cs
->entries
[i
].offset
;
1457 cmds
[entry_idx
].size
= cs
->entries
[i
].size
;
1458 cmds
[entry_idx
].pad
= 0;
1459 cmds
[entry_idx
].nr_relocs
= 0;
1460 cmds
[entry_idx
].relocs
= 0;
1463 tu_bo_list_merge(&bo_list
, &cmdbuf
->bo_list
);
1466 uint32_t flags
= MSM_PIPE_3D0
;
1468 flags
|= MSM_SUBMIT_FENCE_FD_OUT
;
1471 struct drm_msm_gem_submit req
= {
1473 .queueid
= queue
->msm_queue_id
,
1474 .bos
= (uint64_t)(uintptr_t) bo_list
.bo_infos
,
1475 .nr_bos
= bo_list
.count
,
1476 .cmds
= (uint64_t)(uintptr_t)cmds
,
1477 .nr_cmds
= entry_count
,
1480 int ret
= drmCommandWriteRead(queue
->device
->physical_device
->local_fd
,
1484 fprintf(stderr
, "submit failed: %s\n", strerror(errno
));
1488 tu_bo_list_destroy(&bo_list
);
1491 /* no need to merge fences as queue execution is serialized */
1492 tu_fence_update_fd(&queue
->submit_fence
, req
.fence_fd
);
1496 if (_fence
!= VK_NULL_HANDLE
) {
1497 TU_FROM_HANDLE(tu_fence
, fence
, _fence
);
1498 tu_fence_copy(fence
, &queue
->submit_fence
);
1505 tu_QueueWaitIdle(VkQueue _queue
)
1507 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1509 tu_fence_wait_idle(&queue
->submit_fence
);
1515 tu_DeviceWaitIdle(VkDevice _device
)
1517 TU_FROM_HANDLE(tu_device
, device
, _device
);
1519 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1520 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
1521 tu_QueueWaitIdle(tu_queue_to_handle(&device
->queues
[i
][q
]));
1528 tu_ImportSemaphoreFdKHR(VkDevice _device
,
1529 const VkImportSemaphoreFdInfoKHR
*pImportSemaphoreFdInfo
)
1537 tu_GetSemaphoreFdKHR(VkDevice _device
,
1538 const VkSemaphoreGetFdInfoKHR
*pGetFdInfo
,
1547 tu_ImportFenceFdKHR(VkDevice _device
,
1548 const VkImportFenceFdInfoKHR
*pImportFenceFdInfo
)
1556 tu_GetFenceFdKHR(VkDevice _device
,
1557 const VkFenceGetFdInfoKHR
*pGetFdInfo
,
1566 tu_EnumerateInstanceExtensionProperties(const char *pLayerName
,
1567 uint32_t *pPropertyCount
,
1568 VkExtensionProperties
*pProperties
)
1570 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1572 /* We spport no lyaers */
1574 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1576 for (int i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; i
++) {
1577 if (tu_instance_extensions_supported
.extensions
[i
]) {
1578 vk_outarray_append(&out
, prop
) { *prop
= tu_instance_extensions
[i
]; }
1582 return vk_outarray_status(&out
);
1586 tu_EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice
,
1587 const char *pLayerName
,
1588 uint32_t *pPropertyCount
,
1589 VkExtensionProperties
*pProperties
)
1591 /* We spport no lyaers */
1592 TU_FROM_HANDLE(tu_physical_device
, device
, physicalDevice
);
1593 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1595 /* We spport no lyaers */
1597 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1599 for (int i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; i
++) {
1600 if (device
->supported_extensions
.extensions
[i
]) {
1601 vk_outarray_append(&out
, prop
) { *prop
= tu_device_extensions
[i
]; }
1605 return vk_outarray_status(&out
);
1609 tu_GetInstanceProcAddr(VkInstance _instance
, const char *pName
)
1611 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
1613 return tu_lookup_entrypoint_checked(
1614 pName
, instance
? instance
->api_version
: 0,
1615 instance
? &instance
->enabled_extensions
: NULL
, NULL
);
1618 /* The loader wants us to expose a second GetInstanceProcAddr function
1619 * to work around certain LD_PRELOAD issues seen in apps.
1622 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1623 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
);
1626 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1627 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
)
1629 return tu_GetInstanceProcAddr(instance
, pName
);
1633 tu_GetDeviceProcAddr(VkDevice _device
, const char *pName
)
1635 TU_FROM_HANDLE(tu_device
, device
, _device
);
1637 return tu_lookup_entrypoint_checked(pName
, device
->instance
->api_version
,
1638 &device
->instance
->enabled_extensions
,
1639 &device
->enabled_extensions
);
1643 tu_alloc_memory(struct tu_device
*device
,
1644 const VkMemoryAllocateInfo
*pAllocateInfo
,
1645 const VkAllocationCallbacks
*pAllocator
,
1646 VkDeviceMemory
*pMem
)
1648 struct tu_device_memory
*mem
;
1651 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1653 if (pAllocateInfo
->allocationSize
== 0) {
1654 /* Apparently, this is allowed */
1655 *pMem
= VK_NULL_HANDLE
;
1659 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1660 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1662 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1664 const VkImportMemoryFdInfoKHR
*fd_info
=
1665 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
1666 if (fd_info
&& !fd_info
->handleType
)
1670 assert(fd_info
->handleType
==
1671 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
1672 fd_info
->handleType
==
1673 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
1676 * TODO Importing the same fd twice gives us the same handle without
1677 * reference counting. We need to maintain a per-instance handle-to-bo
1678 * table and add reference count to tu_bo.
1680 result
= tu_bo_init_dmabuf(device
, &mem
->bo
,
1681 pAllocateInfo
->allocationSize
, fd_info
->fd
);
1682 if (result
== VK_SUCCESS
) {
1683 /* take ownership and close the fd */
1688 tu_bo_init_new(device
, &mem
->bo
, pAllocateInfo
->allocationSize
);
1691 if (result
!= VK_SUCCESS
) {
1692 vk_free2(&device
->alloc
, pAllocator
, mem
);
1696 mem
->size
= pAllocateInfo
->allocationSize
;
1697 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1700 mem
->user_ptr
= NULL
;
1702 *pMem
= tu_device_memory_to_handle(mem
);
1708 tu_AllocateMemory(VkDevice _device
,
1709 const VkMemoryAllocateInfo
*pAllocateInfo
,
1710 const VkAllocationCallbacks
*pAllocator
,
1711 VkDeviceMemory
*pMem
)
1713 TU_FROM_HANDLE(tu_device
, device
, _device
);
1714 return tu_alloc_memory(device
, pAllocateInfo
, pAllocator
, pMem
);
1718 tu_FreeMemory(VkDevice _device
,
1719 VkDeviceMemory _mem
,
1720 const VkAllocationCallbacks
*pAllocator
)
1722 TU_FROM_HANDLE(tu_device
, device
, _device
);
1723 TU_FROM_HANDLE(tu_device_memory
, mem
, _mem
);
1728 tu_bo_finish(device
, &mem
->bo
);
1729 vk_free2(&device
->alloc
, pAllocator
, mem
);
1733 tu_MapMemory(VkDevice _device
,
1734 VkDeviceMemory _memory
,
1735 VkDeviceSize offset
,
1737 VkMemoryMapFlags flags
,
1740 TU_FROM_HANDLE(tu_device
, device
, _device
);
1741 TU_FROM_HANDLE(tu_device_memory
, mem
, _memory
);
1749 if (mem
->user_ptr
) {
1750 *ppData
= mem
->user_ptr
;
1751 } else if (!mem
->map
) {
1752 result
= tu_bo_map(device
, &mem
->bo
);
1753 if (result
!= VK_SUCCESS
)
1755 *ppData
= mem
->map
= mem
->bo
.map
;
1764 return vk_error(device
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
1768 tu_UnmapMemory(VkDevice _device
, VkDeviceMemory _memory
)
1770 /* I do not see any unmapping done by the freedreno Gallium driver. */
1774 tu_FlushMappedMemoryRanges(VkDevice _device
,
1775 uint32_t memoryRangeCount
,
1776 const VkMappedMemoryRange
*pMemoryRanges
)
1782 tu_InvalidateMappedMemoryRanges(VkDevice _device
,
1783 uint32_t memoryRangeCount
,
1784 const VkMappedMemoryRange
*pMemoryRanges
)
1790 tu_GetBufferMemoryRequirements(VkDevice _device
,
1792 VkMemoryRequirements
*pMemoryRequirements
)
1794 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
1796 pMemoryRequirements
->memoryTypeBits
= 1;
1797 pMemoryRequirements
->alignment
= 64;
1798 pMemoryRequirements
->size
=
1799 align64(buffer
->size
, pMemoryRequirements
->alignment
);
1803 tu_GetBufferMemoryRequirements2(
1805 const VkBufferMemoryRequirementsInfo2
*pInfo
,
1806 VkMemoryRequirements2
*pMemoryRequirements
)
1808 tu_GetBufferMemoryRequirements(device
, pInfo
->buffer
,
1809 &pMemoryRequirements
->memoryRequirements
);
1813 tu_GetImageMemoryRequirements(VkDevice _device
,
1815 VkMemoryRequirements
*pMemoryRequirements
)
1817 TU_FROM_HANDLE(tu_image
, image
, _image
);
1819 pMemoryRequirements
->memoryTypeBits
= 1;
1820 pMemoryRequirements
->size
= image
->layout
.size
;
1821 pMemoryRequirements
->alignment
= image
->layout
.base_align
;
1825 tu_GetImageMemoryRequirements2(VkDevice device
,
1826 const VkImageMemoryRequirementsInfo2
*pInfo
,
1827 VkMemoryRequirements2
*pMemoryRequirements
)
1829 tu_GetImageMemoryRequirements(device
, pInfo
->image
,
1830 &pMemoryRequirements
->memoryRequirements
);
1834 tu_GetImageSparseMemoryRequirements(
1837 uint32_t *pSparseMemoryRequirementCount
,
1838 VkSparseImageMemoryRequirements
*pSparseMemoryRequirements
)
1844 tu_GetImageSparseMemoryRequirements2(
1846 const VkImageSparseMemoryRequirementsInfo2
*pInfo
,
1847 uint32_t *pSparseMemoryRequirementCount
,
1848 VkSparseImageMemoryRequirements2
*pSparseMemoryRequirements
)
1854 tu_GetDeviceMemoryCommitment(VkDevice device
,
1855 VkDeviceMemory memory
,
1856 VkDeviceSize
*pCommittedMemoryInBytes
)
1858 *pCommittedMemoryInBytes
= 0;
1862 tu_BindBufferMemory2(VkDevice device
,
1863 uint32_t bindInfoCount
,
1864 const VkBindBufferMemoryInfo
*pBindInfos
)
1866 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
1867 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
1868 TU_FROM_HANDLE(tu_buffer
, buffer
, pBindInfos
[i
].buffer
);
1871 buffer
->bo
= &mem
->bo
;
1872 buffer
->bo_offset
= pBindInfos
[i
].memoryOffset
;
1881 tu_BindBufferMemory(VkDevice device
,
1883 VkDeviceMemory memory
,
1884 VkDeviceSize memoryOffset
)
1886 const VkBindBufferMemoryInfo info
= {
1887 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
1890 .memoryOffset
= memoryOffset
1893 return tu_BindBufferMemory2(device
, 1, &info
);
1897 tu_BindImageMemory2(VkDevice device
,
1898 uint32_t bindInfoCount
,
1899 const VkBindImageMemoryInfo
*pBindInfos
)
1901 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
1902 TU_FROM_HANDLE(tu_image
, image
, pBindInfos
[i
].image
);
1903 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
1906 image
->bo
= &mem
->bo
;
1907 image
->bo_offset
= pBindInfos
[i
].memoryOffset
;
1910 image
->bo_offset
= 0;
1918 tu_BindImageMemory(VkDevice device
,
1920 VkDeviceMemory memory
,
1921 VkDeviceSize memoryOffset
)
1923 const VkBindImageMemoryInfo info
= {
1924 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
1927 .memoryOffset
= memoryOffset
1930 return tu_BindImageMemory2(device
, 1, &info
);
1934 tu_QueueBindSparse(VkQueue _queue
,
1935 uint32_t bindInfoCount
,
1936 const VkBindSparseInfo
*pBindInfo
,
1942 // Queue semaphore functions
1945 tu_CreateSemaphore(VkDevice _device
,
1946 const VkSemaphoreCreateInfo
*pCreateInfo
,
1947 const VkAllocationCallbacks
*pAllocator
,
1948 VkSemaphore
*pSemaphore
)
1950 TU_FROM_HANDLE(tu_device
, device
, _device
);
1952 struct tu_semaphore
*sem
=
1953 vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sem
), 8,
1954 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1956 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1958 *pSemaphore
= tu_semaphore_to_handle(sem
);
1963 tu_DestroySemaphore(VkDevice _device
,
1964 VkSemaphore _semaphore
,
1965 const VkAllocationCallbacks
*pAllocator
)
1967 TU_FROM_HANDLE(tu_device
, device
, _device
);
1968 TU_FROM_HANDLE(tu_semaphore
, sem
, _semaphore
);
1972 vk_free2(&device
->alloc
, pAllocator
, sem
);
1976 tu_CreateEvent(VkDevice _device
,
1977 const VkEventCreateInfo
*pCreateInfo
,
1978 const VkAllocationCallbacks
*pAllocator
,
1981 TU_FROM_HANDLE(tu_device
, device
, _device
);
1982 struct tu_event
*event
=
1983 vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*event
), 8,
1984 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1987 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1989 VkResult result
= tu_bo_init_new(device
, &event
->bo
, 0x1000);
1990 if (result
!= VK_SUCCESS
)
1993 result
= tu_bo_map(device
, &event
->bo
);
1994 if (result
!= VK_SUCCESS
)
1997 *pEvent
= tu_event_to_handle(event
);
2002 tu_bo_finish(device
, &event
->bo
);
2004 vk_free2(&device
->alloc
, pAllocator
, event
);
2005 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2009 tu_DestroyEvent(VkDevice _device
,
2011 const VkAllocationCallbacks
*pAllocator
)
2013 TU_FROM_HANDLE(tu_device
, device
, _device
);
2014 TU_FROM_HANDLE(tu_event
, event
, _event
);
2019 tu_bo_finish(device
, &event
->bo
);
2020 vk_free2(&device
->alloc
, pAllocator
, event
);
2024 tu_GetEventStatus(VkDevice _device
, VkEvent _event
)
2026 TU_FROM_HANDLE(tu_event
, event
, _event
);
2028 if (*(uint64_t*) event
->bo
.map
== 1)
2029 return VK_EVENT_SET
;
2030 return VK_EVENT_RESET
;
2034 tu_SetEvent(VkDevice _device
, VkEvent _event
)
2036 TU_FROM_HANDLE(tu_event
, event
, _event
);
2037 *(uint64_t*) event
->bo
.map
= 1;
2043 tu_ResetEvent(VkDevice _device
, VkEvent _event
)
2045 TU_FROM_HANDLE(tu_event
, event
, _event
);
2046 *(uint64_t*) event
->bo
.map
= 0;
2052 tu_CreateBuffer(VkDevice _device
,
2053 const VkBufferCreateInfo
*pCreateInfo
,
2054 const VkAllocationCallbacks
*pAllocator
,
2057 TU_FROM_HANDLE(tu_device
, device
, _device
);
2058 struct tu_buffer
*buffer
;
2060 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
2062 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
2063 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2065 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2067 buffer
->size
= pCreateInfo
->size
;
2068 buffer
->usage
= pCreateInfo
->usage
;
2069 buffer
->flags
= pCreateInfo
->flags
;
2071 *pBuffer
= tu_buffer_to_handle(buffer
);
2077 tu_DestroyBuffer(VkDevice _device
,
2079 const VkAllocationCallbacks
*pAllocator
)
2081 TU_FROM_HANDLE(tu_device
, device
, _device
);
2082 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
2087 vk_free2(&device
->alloc
, pAllocator
, buffer
);
2091 tu_CreateFramebuffer(VkDevice _device
,
2092 const VkFramebufferCreateInfo
*pCreateInfo
,
2093 const VkAllocationCallbacks
*pAllocator
,
2094 VkFramebuffer
*pFramebuffer
)
2096 TU_FROM_HANDLE(tu_device
, device
, _device
);
2097 struct tu_framebuffer
*framebuffer
;
2099 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
2101 size_t size
= sizeof(*framebuffer
) + sizeof(struct tu_attachment_info
) *
2102 pCreateInfo
->attachmentCount
;
2103 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
2104 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2105 if (framebuffer
== NULL
)
2106 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2108 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
2109 framebuffer
->width
= pCreateInfo
->width
;
2110 framebuffer
->height
= pCreateInfo
->height
;
2111 framebuffer
->layers
= pCreateInfo
->layers
;
2112 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
2113 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
2114 struct tu_image_view
*iview
= tu_image_view_from_handle(_iview
);
2115 framebuffer
->attachments
[i
].attachment
= iview
;
2118 *pFramebuffer
= tu_framebuffer_to_handle(framebuffer
);
2123 tu_DestroyFramebuffer(VkDevice _device
,
2125 const VkAllocationCallbacks
*pAllocator
)
2127 TU_FROM_HANDLE(tu_device
, device
, _device
);
2128 TU_FROM_HANDLE(tu_framebuffer
, fb
, _fb
);
2132 vk_free2(&device
->alloc
, pAllocator
, fb
);
2136 tu_init_sampler(struct tu_device
*device
,
2137 struct tu_sampler
*sampler
,
2138 const VkSamplerCreateInfo
*pCreateInfo
)
2140 const struct VkSamplerReductionModeCreateInfo
*reduction
=
2141 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_REDUCTION_MODE_CREATE_INFO
);
2142 const struct VkSamplerYcbcrConversionInfo
*ycbcr_conversion
=
2143 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_YCBCR_CONVERSION_INFO
);
2145 unsigned aniso
= pCreateInfo
->anisotropyEnable
?
2146 util_last_bit(MIN2((uint32_t)pCreateInfo
->maxAnisotropy
>> 1, 8)) : 0;
2147 bool miplinear
= (pCreateInfo
->mipmapMode
== VK_SAMPLER_MIPMAP_MODE_LINEAR
);
2148 float min_lod
= CLAMP(pCreateInfo
->minLod
, 0.0f
, 4095.0f
/ 256.0f
);
2149 float max_lod
= CLAMP(pCreateInfo
->maxLod
, 0.0f
, 4095.0f
/ 256.0f
);
2151 sampler
->descriptor
[0] =
2152 COND(miplinear
, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR
) |
2153 A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo
->magFilter
, aniso
)) |
2154 A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo
->minFilter
, aniso
)) |
2155 A6XX_TEX_SAMP_0_ANISO(aniso
) |
2156 A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo
->addressModeU
)) |
2157 A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo
->addressModeV
)) |
2158 A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo
->addressModeW
)) |
2159 A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo
->mipLodBias
);
2160 sampler
->descriptor
[1] =
2161 /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
2162 COND(pCreateInfo
->unnormalizedCoordinates
, A6XX_TEX_SAMP_1_UNNORM_COORDS
) |
2163 A6XX_TEX_SAMP_1_MIN_LOD(min_lod
) |
2164 A6XX_TEX_SAMP_1_MAX_LOD(max_lod
) |
2165 COND(pCreateInfo
->compareEnable
,
2166 A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo
->compareOp
)));
2167 /* This is an offset into the border_color BO, which we fill with all the
2168 * possible Vulkan border colors in the correct order, so we can just use
2169 * the Vulkan enum with no translation necessary.
2171 sampler
->descriptor
[2] =
2172 A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo
->borderColor
*
2173 sizeof(struct bcolor_entry
));
2174 sampler
->descriptor
[3] = 0;
2177 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(
2178 tu6_reduction_mode(reduction
->reductionMode
));
2181 sampler
->ycbcr_sampler
= ycbcr_conversion
?
2182 tu_sampler_ycbcr_conversion_from_handle(ycbcr_conversion
->conversion
) : NULL
;
2184 if (sampler
->ycbcr_sampler
&&
2185 sampler
->ycbcr_sampler
->chroma_filter
== VK_FILTER_LINEAR
) {
2186 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_CHROMA_LINEAR
;
2190 * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
2195 tu_CreateSampler(VkDevice _device
,
2196 const VkSamplerCreateInfo
*pCreateInfo
,
2197 const VkAllocationCallbacks
*pAllocator
,
2198 VkSampler
*pSampler
)
2200 TU_FROM_HANDLE(tu_device
, device
, _device
);
2201 struct tu_sampler
*sampler
;
2203 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
2205 sampler
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
2206 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2208 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2210 tu_init_sampler(device
, sampler
, pCreateInfo
);
2211 *pSampler
= tu_sampler_to_handle(sampler
);
2217 tu_DestroySampler(VkDevice _device
,
2219 const VkAllocationCallbacks
*pAllocator
)
2221 TU_FROM_HANDLE(tu_device
, device
, _device
);
2222 TU_FROM_HANDLE(tu_sampler
, sampler
, _sampler
);
2226 vk_free2(&device
->alloc
, pAllocator
, sampler
);
2229 /* vk_icd.h does not declare this function, so we declare it here to
2230 * suppress Wmissing-prototypes.
2232 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2233 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
2235 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2236 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
2238 /* For the full details on loader interface versioning, see
2239 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2240 * What follows is a condensed summary, to help you navigate the large and
2241 * confusing official doc.
2243 * - Loader interface v0 is incompatible with later versions. We don't
2246 * - In loader interface v1:
2247 * - The first ICD entrypoint called by the loader is
2248 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2250 * - The ICD must statically expose no other Vulkan symbol unless it
2251 * is linked with -Bsymbolic.
2252 * - Each dispatchable Vulkan handle created by the ICD must be
2253 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2254 * ICD must initialize VK_LOADER_DATA.loadMagic to
2256 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2257 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2258 * such loader-managed surfaces.
2260 * - Loader interface v2 differs from v1 in:
2261 * - The first ICD entrypoint called by the loader is
2262 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2263 * statically expose this entrypoint.
2265 * - Loader interface v3 differs from v2 in:
2266 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2267 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2268 * because the loader no longer does so.
2270 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);
2275 tu_GetMemoryFdKHR(VkDevice _device
,
2276 const VkMemoryGetFdInfoKHR
*pGetFdInfo
,
2279 TU_FROM_HANDLE(tu_device
, device
, _device
);
2280 TU_FROM_HANDLE(tu_device_memory
, memory
, pGetFdInfo
->memory
);
2282 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
2284 /* At the moment, we support only the below handle types. */
2285 assert(pGetFdInfo
->handleType
==
2286 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
2287 pGetFdInfo
->handleType
==
2288 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2290 int prime_fd
= tu_bo_export_dmabuf(device
, &memory
->bo
);
2292 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2299 tu_GetMemoryFdPropertiesKHR(VkDevice _device
,
2300 VkExternalMemoryHandleTypeFlagBits handleType
,
2302 VkMemoryFdPropertiesKHR
*pMemoryFdProperties
)
2304 assert(handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2305 pMemoryFdProperties
->memoryTypeBits
= 1;
2310 tu_GetPhysicalDeviceExternalSemaphoreProperties(
2311 VkPhysicalDevice physicalDevice
,
2312 const VkPhysicalDeviceExternalSemaphoreInfo
*pExternalSemaphoreInfo
,
2313 VkExternalSemaphoreProperties
*pExternalSemaphoreProperties
)
2315 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= 0;
2316 pExternalSemaphoreProperties
->compatibleHandleTypes
= 0;
2317 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= 0;
2321 tu_GetPhysicalDeviceExternalFenceProperties(
2322 VkPhysicalDevice physicalDevice
,
2323 const VkPhysicalDeviceExternalFenceInfo
*pExternalFenceInfo
,
2324 VkExternalFenceProperties
*pExternalFenceProperties
)
2326 pExternalFenceProperties
->exportFromImportedHandleTypes
= 0;
2327 pExternalFenceProperties
->compatibleHandleTypes
= 0;
2328 pExternalFenceProperties
->externalFenceFeatures
= 0;
2332 tu_CreateDebugReportCallbackEXT(
2333 VkInstance _instance
,
2334 const VkDebugReportCallbackCreateInfoEXT
*pCreateInfo
,
2335 const VkAllocationCallbacks
*pAllocator
,
2336 VkDebugReportCallbackEXT
*pCallback
)
2338 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2339 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
2340 pCreateInfo
, pAllocator
,
2341 &instance
->alloc
, pCallback
);
2345 tu_DestroyDebugReportCallbackEXT(VkInstance _instance
,
2346 VkDebugReportCallbackEXT _callback
,
2347 const VkAllocationCallbacks
*pAllocator
)
2349 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2350 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
2351 _callback
, pAllocator
, &instance
->alloc
);
2355 tu_DebugReportMessageEXT(VkInstance _instance
,
2356 VkDebugReportFlagsEXT flags
,
2357 VkDebugReportObjectTypeEXT objectType
,
2360 int32_t messageCode
,
2361 const char *pLayerPrefix
,
2362 const char *pMessage
)
2364 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2365 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
2366 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
2370 tu_GetDeviceGroupPeerMemoryFeatures(
2373 uint32_t localDeviceIndex
,
2374 uint32_t remoteDeviceIndex
,
2375 VkPeerMemoryFeatureFlags
*pPeerMemoryFeatures
)
2377 assert(localDeviceIndex
== remoteDeviceIndex
);
2379 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
2380 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
2381 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
2382 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
2385 void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
2386 VkPhysicalDevice physicalDevice
,
2387 VkSampleCountFlagBits samples
,
2388 VkMultisamplePropertiesEXT
* pMultisampleProperties
)
2390 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
2392 if (samples
<= VK_SAMPLE_COUNT_4_BIT
&& pdevice
->supported_extensions
.EXT_sample_locations
)
2393 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 1, 1 };
2395 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 0, 0 };