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
);
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
= false,
585 .fullDrawIndexUint32
= true,
586 .imageCubeArray
= true,
587 .independentBlend
= true,
588 .geometryShader
= true,
589 .tessellationShader
= false,
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
= false;
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
= 4 /* FIXME */,
800 .mipmapPrecisionBits
= 4 /* FIXME */,
801 .maxDrawIndexedIndexValue
= UINT32_MAX
,
802 .maxDrawIndirectCount
= UINT32_MAX
,
803 .maxSamplerLodBias
= 16,
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
= -32,
814 .maxTexelOffset
= 31,
815 .minTexelGatherOffset
= -32,
816 .maxTexelGatherOffset
= 31,
817 .minInterpolationOffset
= -2,
818 .maxInterpolationOffset
= 2,
819 .subPixelInterpolationOffsetBits
= 8,
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;
958 static const VkQueueFamilyProperties tu_queue_family_properties
= {
960 VK_QUEUE_GRAPHICS_BIT
| VK_QUEUE_COMPUTE_BIT
| VK_QUEUE_TRANSFER_BIT
,
962 .timestampValidBits
= 48,
963 .minImageTransferGranularity
= { 1, 1, 1 },
967 tu_GetPhysicalDeviceQueueFamilyProperties(
968 VkPhysicalDevice physicalDevice
,
969 uint32_t *pQueueFamilyPropertyCount
,
970 VkQueueFamilyProperties
*pQueueFamilyProperties
)
972 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
974 vk_outarray_append(&out
, p
) { *p
= tu_queue_family_properties
; }
978 tu_GetPhysicalDeviceQueueFamilyProperties2(
979 VkPhysicalDevice physicalDevice
,
980 uint32_t *pQueueFamilyPropertyCount
,
981 VkQueueFamilyProperties2
*pQueueFamilyProperties
)
983 VK_OUTARRAY_MAKE(out
, pQueueFamilyProperties
, pQueueFamilyPropertyCount
);
985 vk_outarray_append(&out
, p
)
987 p
->queueFamilyProperties
= tu_queue_family_properties
;
992 tu_get_system_heap_size()
997 uint64_t total_ram
= (uint64_t) info
.totalram
* (uint64_t) info
.mem_unit
;
999 /* We don't want to burn too much ram with the GPU. If the user has 4GiB
1000 * or less, we use at most half. If they have more than 4GiB, we use 3/4.
1002 uint64_t available_ram
;
1003 if (total_ram
<= 4ull * 1024ull * 1024ull * 1024ull)
1004 available_ram
= total_ram
/ 2;
1006 available_ram
= total_ram
* 3 / 4;
1008 return available_ram
;
1012 tu_GetPhysicalDeviceMemoryProperties(
1013 VkPhysicalDevice physicalDevice
,
1014 VkPhysicalDeviceMemoryProperties
*pMemoryProperties
)
1016 pMemoryProperties
->memoryHeapCount
= 1;
1017 pMemoryProperties
->memoryHeaps
[0].size
= tu_get_system_heap_size();
1018 pMemoryProperties
->memoryHeaps
[0].flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
;
1020 pMemoryProperties
->memoryTypeCount
= 1;
1021 pMemoryProperties
->memoryTypes
[0].propertyFlags
=
1022 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
1023 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
1024 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
;
1025 pMemoryProperties
->memoryTypes
[0].heapIndex
= 0;
1029 tu_GetPhysicalDeviceMemoryProperties2(
1030 VkPhysicalDevice physicalDevice
,
1031 VkPhysicalDeviceMemoryProperties2
*pMemoryProperties
)
1033 return tu_GetPhysicalDeviceMemoryProperties(
1034 physicalDevice
, &pMemoryProperties
->memoryProperties
);
1038 tu_queue_init(struct tu_device
*device
,
1039 struct tu_queue
*queue
,
1040 uint32_t queue_family_index
,
1042 VkDeviceQueueCreateFlags flags
)
1044 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1045 queue
->device
= device
;
1046 queue
->queue_family_index
= queue_family_index
;
1047 queue
->queue_idx
= idx
;
1048 queue
->flags
= flags
;
1050 int ret
= tu_drm_submitqueue_new(device
, 0, &queue
->msm_queue_id
);
1052 return VK_ERROR_INITIALIZATION_FAILED
;
1054 tu_fence_init(&queue
->submit_fence
, false);
1060 tu_queue_finish(struct tu_queue
*queue
)
1062 tu_fence_finish(&queue
->submit_fence
);
1063 tu_drm_submitqueue_close(queue
->device
, queue
->msm_queue_id
);
1067 tu_get_device_extension_index(const char *name
)
1069 for (unsigned i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; ++i
) {
1070 if (strcmp(name
, tu_device_extensions
[i
].extensionName
) == 0)
1076 struct PACKED bcolor_entry
{
1088 uint32_t z24
; /* also s8? */
1089 uint16_t srgb
[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
1091 } border_color
[] = {
1092 [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
] = {},
1093 [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
] = {},
1094 [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
] = {
1095 .fp32
[3] = 0x3f800000,
1103 .rgb10a2
= 0xc0000000,
1106 [VK_BORDER_COLOR_INT_OPAQUE_BLACK
] = {
1110 [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
] = {
1111 .fp32
[0 ... 3] = 0x3f800000,
1112 .ui16
[0 ... 3] = 0xffff,
1113 .si16
[0 ... 3] = 0x7fff,
1114 .fp16
[0 ... 3] = 0x3c00,
1118 .ui8
[0 ... 3] = 0xff,
1119 .si8
[0 ... 3] = 0x7f,
1120 .rgb10a2
= 0xffffffff,
1122 .srgb
[0 ... 3] = 0x3c00,
1124 [VK_BORDER_COLOR_INT_OPAQUE_WHITE
] = {
1132 tu_CreateDevice(VkPhysicalDevice physicalDevice
,
1133 const VkDeviceCreateInfo
*pCreateInfo
,
1134 const VkAllocationCallbacks
*pAllocator
,
1137 TU_FROM_HANDLE(tu_physical_device
, physical_device
, physicalDevice
);
1139 struct tu_device
*device
;
1141 /* Check enabled features */
1142 if (pCreateInfo
->pEnabledFeatures
) {
1143 VkPhysicalDeviceFeatures supported_features
;
1144 tu_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
1145 VkBool32
*supported_feature
= (VkBool32
*) &supported_features
;
1146 VkBool32
*enabled_feature
= (VkBool32
*) pCreateInfo
->pEnabledFeatures
;
1147 unsigned num_features
=
1148 sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
1149 for (uint32_t i
= 0; i
< num_features
; i
++) {
1150 if (enabled_feature
[i
] && !supported_feature
[i
])
1151 return vk_error(physical_device
->instance
,
1152 VK_ERROR_FEATURE_NOT_PRESENT
);
1156 device
= vk_zalloc2(&physical_device
->instance
->alloc
, pAllocator
,
1157 sizeof(*device
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1159 return vk_error(physical_device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1161 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1162 device
->instance
= physical_device
->instance
;
1163 device
->physical_device
= physical_device
;
1166 device
->alloc
= *pAllocator
;
1168 device
->alloc
= physical_device
->instance
->alloc
;
1170 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
1171 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
1172 int index
= tu_get_device_extension_index(ext_name
);
1174 !physical_device
->supported_extensions
.extensions
[index
]) {
1175 vk_free(&device
->alloc
, device
);
1176 return vk_error(physical_device
->instance
,
1177 VK_ERROR_EXTENSION_NOT_PRESENT
);
1180 device
->enabled_extensions
.extensions
[index
] = true;
1183 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
1184 const VkDeviceQueueCreateInfo
*queue_create
=
1185 &pCreateInfo
->pQueueCreateInfos
[i
];
1186 uint32_t qfi
= queue_create
->queueFamilyIndex
;
1187 device
->queues
[qfi
] = vk_alloc(
1188 &device
->alloc
, queue_create
->queueCount
* sizeof(struct tu_queue
),
1189 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1190 if (!device
->queues
[qfi
]) {
1191 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
1195 memset(device
->queues
[qfi
], 0,
1196 queue_create
->queueCount
* sizeof(struct tu_queue
));
1198 device
->queue_count
[qfi
] = queue_create
->queueCount
;
1200 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
1201 result
= tu_queue_init(device
, &device
->queues
[qfi
][q
], qfi
, q
,
1202 queue_create
->flags
);
1203 if (result
!= VK_SUCCESS
)
1208 device
->compiler
= ir3_compiler_create(NULL
, physical_device
->gpu_id
);
1209 if (!device
->compiler
)
1212 #define VSC_DRAW_STRM_SIZE(pitch) ((pitch) * 32 + 0x100) /* extra size to store VSC_SIZE */
1213 #define VSC_PRIM_STRM_SIZE(pitch) ((pitch) * 32)
1215 device
->vsc_draw_strm_pitch
= 0x440 * 4;
1216 device
->vsc_prim_strm_pitch
= 0x1040 * 4;
1218 result
= tu_bo_init_new(device
, &device
->vsc_draw_strm
, VSC_DRAW_STRM_SIZE(device
->vsc_draw_strm_pitch
));
1219 if (result
!= VK_SUCCESS
)
1222 result
= tu_bo_init_new(device
, &device
->vsc_prim_strm
, VSC_PRIM_STRM_SIZE(device
->vsc_prim_strm_pitch
));
1223 if (result
!= VK_SUCCESS
)
1224 goto fail_vsc_data2
;
1226 STATIC_ASSERT(sizeof(struct bcolor_entry
) == 128);
1227 result
= tu_bo_init_new(device
, &device
->border_color
, sizeof(border_color
));
1228 if (result
!= VK_SUCCESS
)
1229 goto fail_border_color
;
1231 result
= tu_bo_map(device
, &device
->border_color
);
1232 if (result
!= VK_SUCCESS
)
1233 goto fail_border_color_map
;
1235 memcpy(device
->border_color
.map
, border_color
, sizeof(border_color
));
1237 VkPipelineCacheCreateInfo ci
;
1238 ci
.sType
= VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
;
1241 ci
.pInitialData
= NULL
;
1242 ci
.initialDataSize
= 0;
1245 tu_CreatePipelineCache(tu_device_to_handle(device
), &ci
, NULL
, &pc
);
1246 if (result
!= VK_SUCCESS
)
1247 goto fail_pipeline_cache
;
1249 device
->mem_cache
= tu_pipeline_cache_from_handle(pc
);
1251 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++)
1252 mtx_init(&device
->scratch_bos
[i
].construct_mtx
, mtx_plain
);
1254 *pDevice
= tu_device_to_handle(device
);
1257 fail_pipeline_cache
:
1258 fail_border_color_map
:
1259 tu_bo_finish(device
, &device
->border_color
);
1262 tu_bo_finish(device
, &device
->vsc_prim_strm
);
1265 tu_bo_finish(device
, &device
->vsc_draw_strm
);
1268 ralloc_free(device
->compiler
);
1271 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1272 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1273 tu_queue_finish(&device
->queues
[i
][q
]);
1274 if (device
->queue_count
[i
])
1275 vk_free(&device
->alloc
, device
->queues
[i
]);
1278 vk_free(&device
->alloc
, device
);
1283 tu_DestroyDevice(VkDevice _device
, const VkAllocationCallbacks
*pAllocator
)
1285 TU_FROM_HANDLE(tu_device
, device
, _device
);
1290 tu_bo_finish(device
, &device
->vsc_draw_strm
);
1291 tu_bo_finish(device
, &device
->vsc_prim_strm
);
1293 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1294 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1295 tu_queue_finish(&device
->queues
[i
][q
]);
1296 if (device
->queue_count
[i
])
1297 vk_free(&device
->alloc
, device
->queues
[i
]);
1300 for (unsigned i
= 0; i
< ARRAY_SIZE(device
->scratch_bos
); i
++) {
1301 if (device
->scratch_bos
[i
].initialized
)
1302 tu_bo_finish(device
, &device
->scratch_bos
[i
].bo
);
1305 /* the compiler does not use pAllocator */
1306 ralloc_free(device
->compiler
);
1308 VkPipelineCache pc
= tu_pipeline_cache_to_handle(device
->mem_cache
);
1309 tu_DestroyPipelineCache(tu_device_to_handle(device
), pc
, NULL
);
1311 vk_free(&device
->alloc
, device
);
1315 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
)
1317 unsigned size_log2
= MAX2(util_logbase2_ceil64(size
), MIN_SCRATCH_BO_SIZE_LOG2
);
1318 unsigned index
= size_log2
- MIN_SCRATCH_BO_SIZE_LOG2
;
1319 assert(index
< ARRAY_SIZE(dev
->scratch_bos
));
1321 for (unsigned i
= index
; i
< ARRAY_SIZE(dev
->scratch_bos
); i
++) {
1322 if (p_atomic_read(&dev
->scratch_bos
[i
].initialized
)) {
1323 /* Fast path: just return the already-allocated BO. */
1324 *bo
= &dev
->scratch_bos
[i
].bo
;
1329 /* Slow path: actually allocate the BO. We take a lock because the process
1330 * of allocating it is slow, and we don't want to block the CPU while it
1333 mtx_lock(&dev
->scratch_bos
[index
].construct_mtx
);
1335 /* Another thread may have allocated it already while we were waiting on
1336 * the lock. We need to check this in order to avoid double-allocating.
1338 if (dev
->scratch_bos
[index
].initialized
) {
1339 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1340 *bo
= &dev
->scratch_bos
[index
].bo
;
1344 unsigned bo_size
= 1ull << size_log2
;
1345 VkResult result
= tu_bo_init_new(dev
, &dev
->scratch_bos
[index
].bo
, bo_size
);
1346 if (result
!= VK_SUCCESS
) {
1347 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1351 p_atomic_set(&dev
->scratch_bos
[index
].initialized
, true);
1353 mtx_unlock(&dev
->scratch_bos
[index
].construct_mtx
);
1355 *bo
= &dev
->scratch_bos
[index
].bo
;
1360 tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount
,
1361 VkLayerProperties
*pProperties
)
1363 *pPropertyCount
= 0;
1368 tu_EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice
,
1369 uint32_t *pPropertyCount
,
1370 VkLayerProperties
*pProperties
)
1372 *pPropertyCount
= 0;
1377 tu_GetDeviceQueue2(VkDevice _device
,
1378 const VkDeviceQueueInfo2
*pQueueInfo
,
1381 TU_FROM_HANDLE(tu_device
, device
, _device
);
1382 struct tu_queue
*queue
;
1385 &device
->queues
[pQueueInfo
->queueFamilyIndex
][pQueueInfo
->queueIndex
];
1386 if (pQueueInfo
->flags
!= queue
->flags
) {
1387 /* From the Vulkan 1.1.70 spec:
1389 * "The queue returned by vkGetDeviceQueue2 must have the same
1390 * flags value from this structure as that used at device
1391 * creation time in a VkDeviceQueueCreateInfo instance. If no
1392 * matching flags were specified at device creation time then
1393 * pQueue will return VK_NULL_HANDLE."
1395 *pQueue
= VK_NULL_HANDLE
;
1399 *pQueue
= tu_queue_to_handle(queue
);
1403 tu_GetDeviceQueue(VkDevice _device
,
1404 uint32_t queueFamilyIndex
,
1405 uint32_t queueIndex
,
1408 const VkDeviceQueueInfo2 info
=
1409 (VkDeviceQueueInfo2
) { .sType
= VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2
,
1410 .queueFamilyIndex
= queueFamilyIndex
,
1411 .queueIndex
= queueIndex
};
1413 tu_GetDeviceQueue2(_device
, &info
, pQueue
);
1417 tu_QueueSubmit(VkQueue _queue
,
1418 uint32_t submitCount
,
1419 const VkSubmitInfo
*pSubmits
,
1422 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1424 for (uint32_t i
= 0; i
< submitCount
; ++i
) {
1425 const VkSubmitInfo
*submit
= pSubmits
+ i
;
1426 const bool last_submit
= (i
== submitCount
- 1);
1427 struct tu_bo_list bo_list
;
1428 tu_bo_list_init(&bo_list
);
1430 uint32_t entry_count
= 0;
1431 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1432 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1433 entry_count
+= cmdbuf
->cs
.entry_count
;
1436 struct drm_msm_gem_submit_cmd cmds
[entry_count
];
1437 uint32_t entry_idx
= 0;
1438 for (uint32_t j
= 0; j
< submit
->commandBufferCount
; ++j
) {
1439 TU_FROM_HANDLE(tu_cmd_buffer
, cmdbuf
, submit
->pCommandBuffers
[j
]);
1440 struct tu_cs
*cs
= &cmdbuf
->cs
;
1441 for (unsigned i
= 0; i
< cs
->entry_count
; ++i
, ++entry_idx
) {
1442 cmds
[entry_idx
].type
= MSM_SUBMIT_CMD_BUF
;
1443 cmds
[entry_idx
].submit_idx
=
1444 tu_bo_list_add(&bo_list
, cs
->entries
[i
].bo
,
1445 MSM_SUBMIT_BO_READ
| MSM_SUBMIT_BO_DUMP
);
1446 cmds
[entry_idx
].submit_offset
= cs
->entries
[i
].offset
;
1447 cmds
[entry_idx
].size
= cs
->entries
[i
].size
;
1448 cmds
[entry_idx
].pad
= 0;
1449 cmds
[entry_idx
].nr_relocs
= 0;
1450 cmds
[entry_idx
].relocs
= 0;
1453 tu_bo_list_merge(&bo_list
, &cmdbuf
->bo_list
);
1456 uint32_t flags
= MSM_PIPE_3D0
;
1458 flags
|= MSM_SUBMIT_FENCE_FD_OUT
;
1461 struct drm_msm_gem_submit req
= {
1463 .queueid
= queue
->msm_queue_id
,
1464 .bos
= (uint64_t)(uintptr_t) bo_list
.bo_infos
,
1465 .nr_bos
= bo_list
.count
,
1466 .cmds
= (uint64_t)(uintptr_t)cmds
,
1467 .nr_cmds
= entry_count
,
1470 int ret
= drmCommandWriteRead(queue
->device
->physical_device
->local_fd
,
1474 fprintf(stderr
, "submit failed: %s\n", strerror(errno
));
1478 tu_bo_list_destroy(&bo_list
);
1481 /* no need to merge fences as queue execution is serialized */
1482 tu_fence_update_fd(&queue
->submit_fence
, req
.fence_fd
);
1486 if (_fence
!= VK_NULL_HANDLE
) {
1487 TU_FROM_HANDLE(tu_fence
, fence
, _fence
);
1488 tu_fence_copy(fence
, &queue
->submit_fence
);
1495 tu_QueueWaitIdle(VkQueue _queue
)
1497 TU_FROM_HANDLE(tu_queue
, queue
, _queue
);
1499 tu_fence_wait_idle(&queue
->submit_fence
);
1505 tu_DeviceWaitIdle(VkDevice _device
)
1507 TU_FROM_HANDLE(tu_device
, device
, _device
);
1509 for (unsigned i
= 0; i
< TU_MAX_QUEUE_FAMILIES
; i
++) {
1510 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
1511 tu_QueueWaitIdle(tu_queue_to_handle(&device
->queues
[i
][q
]));
1518 tu_ImportSemaphoreFdKHR(VkDevice _device
,
1519 const VkImportSemaphoreFdInfoKHR
*pImportSemaphoreFdInfo
)
1527 tu_GetSemaphoreFdKHR(VkDevice _device
,
1528 const VkSemaphoreGetFdInfoKHR
*pGetFdInfo
,
1537 tu_ImportFenceFdKHR(VkDevice _device
,
1538 const VkImportFenceFdInfoKHR
*pImportFenceFdInfo
)
1546 tu_GetFenceFdKHR(VkDevice _device
,
1547 const VkFenceGetFdInfoKHR
*pGetFdInfo
,
1556 tu_EnumerateInstanceExtensionProperties(const char *pLayerName
,
1557 uint32_t *pPropertyCount
,
1558 VkExtensionProperties
*pProperties
)
1560 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1562 /* We spport no lyaers */
1564 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1566 for (int i
= 0; i
< TU_INSTANCE_EXTENSION_COUNT
; i
++) {
1567 if (tu_instance_extensions_supported
.extensions
[i
]) {
1568 vk_outarray_append(&out
, prop
) { *prop
= tu_instance_extensions
[i
]; }
1572 return vk_outarray_status(&out
);
1576 tu_EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice
,
1577 const char *pLayerName
,
1578 uint32_t *pPropertyCount
,
1579 VkExtensionProperties
*pProperties
)
1581 /* We spport no lyaers */
1582 TU_FROM_HANDLE(tu_physical_device
, device
, physicalDevice
);
1583 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
1585 /* We spport no lyaers */
1587 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1589 for (int i
= 0; i
< TU_DEVICE_EXTENSION_COUNT
; i
++) {
1590 if (device
->supported_extensions
.extensions
[i
]) {
1591 vk_outarray_append(&out
, prop
) { *prop
= tu_device_extensions
[i
]; }
1595 return vk_outarray_status(&out
);
1599 tu_GetInstanceProcAddr(VkInstance _instance
, const char *pName
)
1601 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
1603 return tu_lookup_entrypoint_checked(
1604 pName
, instance
? instance
->api_version
: 0,
1605 instance
? &instance
->enabled_extensions
: NULL
, NULL
);
1608 /* The loader wants us to expose a second GetInstanceProcAddr function
1609 * to work around certain LD_PRELOAD issues seen in apps.
1612 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1613 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
);
1616 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
1617 vk_icdGetInstanceProcAddr(VkInstance instance
, const char *pName
)
1619 return tu_GetInstanceProcAddr(instance
, pName
);
1623 tu_GetDeviceProcAddr(VkDevice _device
, const char *pName
)
1625 TU_FROM_HANDLE(tu_device
, device
, _device
);
1627 return tu_lookup_entrypoint_checked(pName
, device
->instance
->api_version
,
1628 &device
->instance
->enabled_extensions
,
1629 &device
->enabled_extensions
);
1633 tu_alloc_memory(struct tu_device
*device
,
1634 const VkMemoryAllocateInfo
*pAllocateInfo
,
1635 const VkAllocationCallbacks
*pAllocator
,
1636 VkDeviceMemory
*pMem
)
1638 struct tu_device_memory
*mem
;
1641 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1643 if (pAllocateInfo
->allocationSize
== 0) {
1644 /* Apparently, this is allowed */
1645 *pMem
= VK_NULL_HANDLE
;
1649 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1650 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1652 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1654 const VkImportMemoryFdInfoKHR
*fd_info
=
1655 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
1656 if (fd_info
&& !fd_info
->handleType
)
1660 assert(fd_info
->handleType
==
1661 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
1662 fd_info
->handleType
==
1663 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
1666 * TODO Importing the same fd twice gives us the same handle without
1667 * reference counting. We need to maintain a per-instance handle-to-bo
1668 * table and add reference count to tu_bo.
1670 result
= tu_bo_init_dmabuf(device
, &mem
->bo
,
1671 pAllocateInfo
->allocationSize
, fd_info
->fd
);
1672 if (result
== VK_SUCCESS
) {
1673 /* take ownership and close the fd */
1678 tu_bo_init_new(device
, &mem
->bo
, pAllocateInfo
->allocationSize
);
1681 if (result
!= VK_SUCCESS
) {
1682 vk_free2(&device
->alloc
, pAllocator
, mem
);
1686 mem
->size
= pAllocateInfo
->allocationSize
;
1687 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1690 mem
->user_ptr
= NULL
;
1692 *pMem
= tu_device_memory_to_handle(mem
);
1698 tu_AllocateMemory(VkDevice _device
,
1699 const VkMemoryAllocateInfo
*pAllocateInfo
,
1700 const VkAllocationCallbacks
*pAllocator
,
1701 VkDeviceMemory
*pMem
)
1703 TU_FROM_HANDLE(tu_device
, device
, _device
);
1704 return tu_alloc_memory(device
, pAllocateInfo
, pAllocator
, pMem
);
1708 tu_FreeMemory(VkDevice _device
,
1709 VkDeviceMemory _mem
,
1710 const VkAllocationCallbacks
*pAllocator
)
1712 TU_FROM_HANDLE(tu_device
, device
, _device
);
1713 TU_FROM_HANDLE(tu_device_memory
, mem
, _mem
);
1718 tu_bo_finish(device
, &mem
->bo
);
1719 vk_free2(&device
->alloc
, pAllocator
, mem
);
1723 tu_MapMemory(VkDevice _device
,
1724 VkDeviceMemory _memory
,
1725 VkDeviceSize offset
,
1727 VkMemoryMapFlags flags
,
1730 TU_FROM_HANDLE(tu_device
, device
, _device
);
1731 TU_FROM_HANDLE(tu_device_memory
, mem
, _memory
);
1739 if (mem
->user_ptr
) {
1740 *ppData
= mem
->user_ptr
;
1741 } else if (!mem
->map
) {
1742 result
= tu_bo_map(device
, &mem
->bo
);
1743 if (result
!= VK_SUCCESS
)
1745 *ppData
= mem
->map
= mem
->bo
.map
;
1754 return vk_error(device
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
1758 tu_UnmapMemory(VkDevice _device
, VkDeviceMemory _memory
)
1760 /* I do not see any unmapping done by the freedreno Gallium driver. */
1764 tu_FlushMappedMemoryRanges(VkDevice _device
,
1765 uint32_t memoryRangeCount
,
1766 const VkMappedMemoryRange
*pMemoryRanges
)
1772 tu_InvalidateMappedMemoryRanges(VkDevice _device
,
1773 uint32_t memoryRangeCount
,
1774 const VkMappedMemoryRange
*pMemoryRanges
)
1780 tu_GetBufferMemoryRequirements(VkDevice _device
,
1782 VkMemoryRequirements
*pMemoryRequirements
)
1784 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
1786 pMemoryRequirements
->memoryTypeBits
= 1;
1787 pMemoryRequirements
->alignment
= 64;
1788 pMemoryRequirements
->size
=
1789 align64(buffer
->size
, pMemoryRequirements
->alignment
);
1793 tu_GetBufferMemoryRequirements2(
1795 const VkBufferMemoryRequirementsInfo2
*pInfo
,
1796 VkMemoryRequirements2
*pMemoryRequirements
)
1798 tu_GetBufferMemoryRequirements(device
, pInfo
->buffer
,
1799 &pMemoryRequirements
->memoryRequirements
);
1803 tu_GetImageMemoryRequirements(VkDevice _device
,
1805 VkMemoryRequirements
*pMemoryRequirements
)
1807 TU_FROM_HANDLE(tu_image
, image
, _image
);
1809 pMemoryRequirements
->memoryTypeBits
= 1;
1810 pMemoryRequirements
->size
= image
->layout
.size
;
1811 pMemoryRequirements
->alignment
= image
->layout
.base_align
;
1815 tu_GetImageMemoryRequirements2(VkDevice device
,
1816 const VkImageMemoryRequirementsInfo2
*pInfo
,
1817 VkMemoryRequirements2
*pMemoryRequirements
)
1819 tu_GetImageMemoryRequirements(device
, pInfo
->image
,
1820 &pMemoryRequirements
->memoryRequirements
);
1824 tu_GetImageSparseMemoryRequirements(
1827 uint32_t *pSparseMemoryRequirementCount
,
1828 VkSparseImageMemoryRequirements
*pSparseMemoryRequirements
)
1834 tu_GetImageSparseMemoryRequirements2(
1836 const VkImageSparseMemoryRequirementsInfo2
*pInfo
,
1837 uint32_t *pSparseMemoryRequirementCount
,
1838 VkSparseImageMemoryRequirements2
*pSparseMemoryRequirements
)
1844 tu_GetDeviceMemoryCommitment(VkDevice device
,
1845 VkDeviceMemory memory
,
1846 VkDeviceSize
*pCommittedMemoryInBytes
)
1848 *pCommittedMemoryInBytes
= 0;
1852 tu_BindBufferMemory2(VkDevice device
,
1853 uint32_t bindInfoCount
,
1854 const VkBindBufferMemoryInfo
*pBindInfos
)
1856 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
1857 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
1858 TU_FROM_HANDLE(tu_buffer
, buffer
, pBindInfos
[i
].buffer
);
1861 buffer
->bo
= &mem
->bo
;
1862 buffer
->bo_offset
= pBindInfos
[i
].memoryOffset
;
1871 tu_BindBufferMemory(VkDevice device
,
1873 VkDeviceMemory memory
,
1874 VkDeviceSize memoryOffset
)
1876 const VkBindBufferMemoryInfo info
= {
1877 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
1880 .memoryOffset
= memoryOffset
1883 return tu_BindBufferMemory2(device
, 1, &info
);
1887 tu_BindImageMemory2(VkDevice device
,
1888 uint32_t bindInfoCount
,
1889 const VkBindImageMemoryInfo
*pBindInfos
)
1891 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
1892 TU_FROM_HANDLE(tu_image
, image
, pBindInfos
[i
].image
);
1893 TU_FROM_HANDLE(tu_device_memory
, mem
, pBindInfos
[i
].memory
);
1896 image
->bo
= &mem
->bo
;
1897 image
->bo_offset
= pBindInfos
[i
].memoryOffset
;
1900 image
->bo_offset
= 0;
1908 tu_BindImageMemory(VkDevice device
,
1910 VkDeviceMemory memory
,
1911 VkDeviceSize memoryOffset
)
1913 const VkBindImageMemoryInfo info
= {
1914 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO
,
1917 .memoryOffset
= memoryOffset
1920 return tu_BindImageMemory2(device
, 1, &info
);
1924 tu_QueueBindSparse(VkQueue _queue
,
1925 uint32_t bindInfoCount
,
1926 const VkBindSparseInfo
*pBindInfo
,
1932 // Queue semaphore functions
1935 tu_CreateSemaphore(VkDevice _device
,
1936 const VkSemaphoreCreateInfo
*pCreateInfo
,
1937 const VkAllocationCallbacks
*pAllocator
,
1938 VkSemaphore
*pSemaphore
)
1940 TU_FROM_HANDLE(tu_device
, device
, _device
);
1942 struct tu_semaphore
*sem
=
1943 vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sem
), 8,
1944 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1946 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1948 *pSemaphore
= tu_semaphore_to_handle(sem
);
1953 tu_DestroySemaphore(VkDevice _device
,
1954 VkSemaphore _semaphore
,
1955 const VkAllocationCallbacks
*pAllocator
)
1957 TU_FROM_HANDLE(tu_device
, device
, _device
);
1958 TU_FROM_HANDLE(tu_semaphore
, sem
, _semaphore
);
1962 vk_free2(&device
->alloc
, pAllocator
, sem
);
1966 tu_CreateEvent(VkDevice _device
,
1967 const VkEventCreateInfo
*pCreateInfo
,
1968 const VkAllocationCallbacks
*pAllocator
,
1971 TU_FROM_HANDLE(tu_device
, device
, _device
);
1972 struct tu_event
*event
=
1973 vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*event
), 8,
1974 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1977 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1979 VkResult result
= tu_bo_init_new(device
, &event
->bo
, 0x1000);
1980 if (result
!= VK_SUCCESS
)
1983 result
= tu_bo_map(device
, &event
->bo
);
1984 if (result
!= VK_SUCCESS
)
1987 *pEvent
= tu_event_to_handle(event
);
1992 tu_bo_finish(device
, &event
->bo
);
1994 vk_free2(&device
->alloc
, pAllocator
, event
);
1995 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1999 tu_DestroyEvent(VkDevice _device
,
2001 const VkAllocationCallbacks
*pAllocator
)
2003 TU_FROM_HANDLE(tu_device
, device
, _device
);
2004 TU_FROM_HANDLE(tu_event
, event
, _event
);
2009 tu_bo_finish(device
, &event
->bo
);
2010 vk_free2(&device
->alloc
, pAllocator
, event
);
2014 tu_GetEventStatus(VkDevice _device
, VkEvent _event
)
2016 TU_FROM_HANDLE(tu_event
, event
, _event
);
2018 if (*(uint64_t*) event
->bo
.map
== 1)
2019 return VK_EVENT_SET
;
2020 return VK_EVENT_RESET
;
2024 tu_SetEvent(VkDevice _device
, VkEvent _event
)
2026 TU_FROM_HANDLE(tu_event
, event
, _event
);
2027 *(uint64_t*) event
->bo
.map
= 1;
2033 tu_ResetEvent(VkDevice _device
, VkEvent _event
)
2035 TU_FROM_HANDLE(tu_event
, event
, _event
);
2036 *(uint64_t*) event
->bo
.map
= 0;
2042 tu_CreateBuffer(VkDevice _device
,
2043 const VkBufferCreateInfo
*pCreateInfo
,
2044 const VkAllocationCallbacks
*pAllocator
,
2047 TU_FROM_HANDLE(tu_device
, device
, _device
);
2048 struct tu_buffer
*buffer
;
2050 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
2052 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
2053 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2055 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2057 buffer
->size
= pCreateInfo
->size
;
2058 buffer
->usage
= pCreateInfo
->usage
;
2059 buffer
->flags
= pCreateInfo
->flags
;
2061 *pBuffer
= tu_buffer_to_handle(buffer
);
2067 tu_DestroyBuffer(VkDevice _device
,
2069 const VkAllocationCallbacks
*pAllocator
)
2071 TU_FROM_HANDLE(tu_device
, device
, _device
);
2072 TU_FROM_HANDLE(tu_buffer
, buffer
, _buffer
);
2077 vk_free2(&device
->alloc
, pAllocator
, buffer
);
2081 tu_CreateFramebuffer(VkDevice _device
,
2082 const VkFramebufferCreateInfo
*pCreateInfo
,
2083 const VkAllocationCallbacks
*pAllocator
,
2084 VkFramebuffer
*pFramebuffer
)
2086 TU_FROM_HANDLE(tu_device
, device
, _device
);
2087 struct tu_framebuffer
*framebuffer
;
2089 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
2091 size_t size
= sizeof(*framebuffer
) + sizeof(struct tu_attachment_info
) *
2092 pCreateInfo
->attachmentCount
;
2093 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
2094 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2095 if (framebuffer
== NULL
)
2096 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2098 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
2099 framebuffer
->width
= pCreateInfo
->width
;
2100 framebuffer
->height
= pCreateInfo
->height
;
2101 framebuffer
->layers
= pCreateInfo
->layers
;
2102 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
2103 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
2104 struct tu_image_view
*iview
= tu_image_view_from_handle(_iview
);
2105 framebuffer
->attachments
[i
].attachment
= iview
;
2108 *pFramebuffer
= tu_framebuffer_to_handle(framebuffer
);
2113 tu_DestroyFramebuffer(VkDevice _device
,
2115 const VkAllocationCallbacks
*pAllocator
)
2117 TU_FROM_HANDLE(tu_device
, device
, _device
);
2118 TU_FROM_HANDLE(tu_framebuffer
, fb
, _fb
);
2122 vk_free2(&device
->alloc
, pAllocator
, fb
);
2125 static enum a6xx_tex_clamp
2126 tu6_tex_wrap(VkSamplerAddressMode address_mode
)
2128 switch (address_mode
) {
2129 case VK_SAMPLER_ADDRESS_MODE_REPEAT
:
2130 return A6XX_TEX_REPEAT
;
2131 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT
:
2132 return A6XX_TEX_MIRROR_REPEAT
;
2133 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE
:
2134 return A6XX_TEX_CLAMP_TO_EDGE
;
2135 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER
:
2136 return A6XX_TEX_CLAMP_TO_BORDER
;
2137 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE
:
2138 /* only works for PoT.. need to emulate otherwise! */
2139 return A6XX_TEX_MIRROR_CLAMP
;
2141 unreachable("illegal tex wrap mode");
2146 static enum a6xx_tex_filter
2147 tu6_tex_filter(VkFilter filter
, unsigned aniso
)
2150 case VK_FILTER_NEAREST
:
2151 return A6XX_TEX_NEAREST
;
2152 case VK_FILTER_LINEAR
:
2153 return aniso
? A6XX_TEX_ANISO
: A6XX_TEX_LINEAR
;
2154 case VK_FILTER_CUBIC_EXT
:
2155 return A6XX_TEX_CUBIC
;
2157 unreachable("illegal texture filter");
2162 static inline enum adreno_compare_func
2163 tu6_compare_func(VkCompareOp op
)
2165 return (enum adreno_compare_func
) op
;
2169 tu_init_sampler(struct tu_device
*device
,
2170 struct tu_sampler
*sampler
,
2171 const VkSamplerCreateInfo
*pCreateInfo
)
2173 const struct VkSamplerReductionModeCreateInfo
*reduction
=
2174 vk_find_struct_const(pCreateInfo
->pNext
, SAMPLER_REDUCTION_MODE_CREATE_INFO
);
2176 unsigned aniso
= pCreateInfo
->anisotropyEnable
?
2177 util_last_bit(MIN2((uint32_t)pCreateInfo
->maxAnisotropy
>> 1, 8)) : 0;
2178 bool miplinear
= (pCreateInfo
->mipmapMode
== VK_SAMPLER_MIPMAP_MODE_LINEAR
);
2180 sampler
->descriptor
[0] =
2181 COND(miplinear
, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR
) |
2182 A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo
->magFilter
, aniso
)) |
2183 A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo
->minFilter
, aniso
)) |
2184 A6XX_TEX_SAMP_0_ANISO(aniso
) |
2185 A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo
->addressModeU
)) |
2186 A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo
->addressModeV
)) |
2187 A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo
->addressModeW
)) |
2188 A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo
->mipLodBias
);
2189 sampler
->descriptor
[1] =
2190 /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
2191 COND(pCreateInfo
->unnormalizedCoordinates
, A6XX_TEX_SAMP_1_UNNORM_COORDS
) |
2192 A6XX_TEX_SAMP_1_MIN_LOD(pCreateInfo
->minLod
) |
2193 A6XX_TEX_SAMP_1_MAX_LOD(pCreateInfo
->maxLod
) |
2194 COND(pCreateInfo
->compareEnable
,
2195 A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo
->compareOp
)));
2196 /* This is an offset into the border_color BO, which we fill with all the
2197 * possible Vulkan border colors in the correct order, so we can just use
2198 * the Vulkan enum with no translation necessary.
2200 sampler
->descriptor
[2] =
2201 A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo
->borderColor
*
2202 sizeof(struct bcolor_entry
));
2203 sampler
->descriptor
[3] = 0;
2206 /* note: vulkan enum matches hw */
2207 sampler
->descriptor
[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(reduction
->reductionMode
);
2211 * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
2216 tu_CreateSampler(VkDevice _device
,
2217 const VkSamplerCreateInfo
*pCreateInfo
,
2218 const VkAllocationCallbacks
*pAllocator
,
2219 VkSampler
*pSampler
)
2221 TU_FROM_HANDLE(tu_device
, device
, _device
);
2222 struct tu_sampler
*sampler
;
2224 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
2226 sampler
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
2227 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2229 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2231 tu_init_sampler(device
, sampler
, pCreateInfo
);
2232 *pSampler
= tu_sampler_to_handle(sampler
);
2238 tu_DestroySampler(VkDevice _device
,
2240 const VkAllocationCallbacks
*pAllocator
)
2242 TU_FROM_HANDLE(tu_device
, device
, _device
);
2243 TU_FROM_HANDLE(tu_sampler
, sampler
, _sampler
);
2247 vk_free2(&device
->alloc
, pAllocator
, sampler
);
2250 /* vk_icd.h does not declare this function, so we declare it here to
2251 * suppress Wmissing-prototypes.
2253 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2254 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
2256 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2257 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
2259 /* For the full details on loader interface versioning, see
2260 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2261 * What follows is a condensed summary, to help you navigate the large and
2262 * confusing official doc.
2264 * - Loader interface v0 is incompatible with later versions. We don't
2267 * - In loader interface v1:
2268 * - The first ICD entrypoint called by the loader is
2269 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2271 * - The ICD must statically expose no other Vulkan symbol unless it
2272 * is linked with -Bsymbolic.
2273 * - Each dispatchable Vulkan handle created by the ICD must be
2274 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2275 * ICD must initialize VK_LOADER_DATA.loadMagic to
2277 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2278 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2279 * such loader-managed surfaces.
2281 * - Loader interface v2 differs from v1 in:
2282 * - The first ICD entrypoint called by the loader is
2283 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2284 * statically expose this entrypoint.
2286 * - Loader interface v3 differs from v2 in:
2287 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2288 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2289 * because the loader no longer does so.
2291 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);
2296 tu_GetMemoryFdKHR(VkDevice _device
,
2297 const VkMemoryGetFdInfoKHR
*pGetFdInfo
,
2300 TU_FROM_HANDLE(tu_device
, device
, _device
);
2301 TU_FROM_HANDLE(tu_device_memory
, memory
, pGetFdInfo
->memory
);
2303 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
2305 /* At the moment, we support only the below handle types. */
2306 assert(pGetFdInfo
->handleType
==
2307 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT
||
2308 pGetFdInfo
->handleType
==
2309 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2311 int prime_fd
= tu_bo_export_dmabuf(device
, &memory
->bo
);
2313 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2320 tu_GetMemoryFdPropertiesKHR(VkDevice _device
,
2321 VkExternalMemoryHandleTypeFlagBits handleType
,
2323 VkMemoryFdPropertiesKHR
*pMemoryFdProperties
)
2325 assert(handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
2326 pMemoryFdProperties
->memoryTypeBits
= 1;
2331 tu_GetPhysicalDeviceExternalSemaphoreProperties(
2332 VkPhysicalDevice physicalDevice
,
2333 const VkPhysicalDeviceExternalSemaphoreInfo
*pExternalSemaphoreInfo
,
2334 VkExternalSemaphoreProperties
*pExternalSemaphoreProperties
)
2336 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= 0;
2337 pExternalSemaphoreProperties
->compatibleHandleTypes
= 0;
2338 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= 0;
2342 tu_GetPhysicalDeviceExternalFenceProperties(
2343 VkPhysicalDevice physicalDevice
,
2344 const VkPhysicalDeviceExternalFenceInfo
*pExternalFenceInfo
,
2345 VkExternalFenceProperties
*pExternalFenceProperties
)
2347 pExternalFenceProperties
->exportFromImportedHandleTypes
= 0;
2348 pExternalFenceProperties
->compatibleHandleTypes
= 0;
2349 pExternalFenceProperties
->externalFenceFeatures
= 0;
2353 tu_CreateDebugReportCallbackEXT(
2354 VkInstance _instance
,
2355 const VkDebugReportCallbackCreateInfoEXT
*pCreateInfo
,
2356 const VkAllocationCallbacks
*pAllocator
,
2357 VkDebugReportCallbackEXT
*pCallback
)
2359 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2360 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
2361 pCreateInfo
, pAllocator
,
2362 &instance
->alloc
, pCallback
);
2366 tu_DestroyDebugReportCallbackEXT(VkInstance _instance
,
2367 VkDebugReportCallbackEXT _callback
,
2368 const VkAllocationCallbacks
*pAllocator
)
2370 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2371 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
2372 _callback
, pAllocator
, &instance
->alloc
);
2376 tu_DebugReportMessageEXT(VkInstance _instance
,
2377 VkDebugReportFlagsEXT flags
,
2378 VkDebugReportObjectTypeEXT objectType
,
2381 int32_t messageCode
,
2382 const char *pLayerPrefix
,
2383 const char *pMessage
)
2385 TU_FROM_HANDLE(tu_instance
, instance
, _instance
);
2386 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
2387 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
2391 tu_GetDeviceGroupPeerMemoryFeatures(
2394 uint32_t localDeviceIndex
,
2395 uint32_t remoteDeviceIndex
,
2396 VkPeerMemoryFeatureFlags
*pPeerMemoryFeatures
)
2398 assert(localDeviceIndex
== remoteDeviceIndex
);
2400 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
2401 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
2402 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
2403 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
2406 void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
2407 VkPhysicalDevice physicalDevice
,
2408 VkSampleCountFlagBits samples
,
2409 VkMultisamplePropertiesEXT
* pMultisampleProperties
)
2411 TU_FROM_HANDLE(tu_physical_device
, pdevice
, physicalDevice
);
2413 if (samples
<= VK_SAMPLE_COUNT_4_BIT
&& pdevice
->supported_extensions
.EXT_sample_locations
)
2414 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 1, 1 };
2416 pMultisampleProperties
->maxSampleLocationGridSize
= (VkExtent2D
){ 0, 0 };