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 DEALINGS
34 #include "radv_private.h"
35 #include "util/strtod.h"
39 #include <amdgpu_drm.h>
40 #include "amdgpu_id.h"
41 #include "winsys/amdgpu/radv_amdgpu_winsys_public.h"
42 #include "ac_llvm_util.h"
43 #include "vk_format.h"
45 #include "util/debug.h"
46 struct radv_dispatch_table dtable
;
49 radv_get_function_timestamp(void *ptr
, uint32_t* timestamp
)
53 if (!dladdr(ptr
, &info
) || !info
.dli_fname
) {
56 if (stat(info
.dli_fname
, &st
)) {
59 *timestamp
= st
.st_mtim
.tv_sec
;
64 radv_device_get_cache_uuid(enum radeon_family family
, void *uuid
)
66 uint32_t mesa_timestamp
, llvm_timestamp
;
68 memset(uuid
, 0, VK_UUID_SIZE
);
69 if (radv_get_function_timestamp(radv_device_get_cache_uuid
, &mesa_timestamp
) ||
70 radv_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo
, &llvm_timestamp
))
73 memcpy(uuid
, &mesa_timestamp
, 4);
74 memcpy((char*)uuid
+ 4, &llvm_timestamp
, 4);
75 memcpy((char*)uuid
+ 8, &f
, 2);
76 snprintf((char*)uuid
+ 10, VK_UUID_SIZE
- 10, "radv");
80 static const VkExtensionProperties instance_extensions
[] = {
82 .extensionName
= VK_KHR_SURFACE_EXTENSION_NAME
,
85 #ifdef VK_USE_PLATFORM_XCB_KHR
87 .extensionName
= VK_KHR_XCB_SURFACE_EXTENSION_NAME
,
91 #ifdef VK_USE_PLATFORM_XLIB_KHR
93 .extensionName
= VK_KHR_XLIB_SURFACE_EXTENSION_NAME
,
97 #ifdef VK_USE_PLATFORM_WAYLAND_KHR
99 .extensionName
= VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME
,
105 static const VkExtensionProperties common_device_extensions
[] = {
107 .extensionName
= VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME
,
111 .extensionName
= VK_KHR_SWAPCHAIN_EXTENSION_NAME
,
115 .extensionName
= VK_AMD_DRAW_INDIRECT_COUNT_EXTENSION_NAME
,
119 .extensionName
= VK_AMD_NEGATIVE_VIEWPORT_HEIGHT_EXTENSION_NAME
,
125 radv_extensions_register(struct radv_instance
*instance
,
126 struct radv_extensions
*extensions
,
127 const VkExtensionProperties
*new_ext
,
131 VkExtensionProperties
*new_ptr
;
133 assert(new_ext
&& num_ext
> 0);
136 return VK_ERROR_INITIALIZATION_FAILED
;
138 new_size
= (extensions
->num_ext
+ num_ext
) * sizeof(VkExtensionProperties
);
139 new_ptr
= vk_realloc(&instance
->alloc
, extensions
->ext_array
,
140 new_size
, 8, VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
142 /* Old array continues to be valid, update nothing */
144 return VK_ERROR_OUT_OF_HOST_MEMORY
;
146 memcpy(&new_ptr
[extensions
->num_ext
], new_ext
,
147 num_ext
* sizeof(VkExtensionProperties
));
148 extensions
->ext_array
= new_ptr
;
149 extensions
->num_ext
+= num_ext
;
155 radv_extensions_finish(struct radv_instance
*instance
,
156 struct radv_extensions
*extensions
)
161 radv_loge("Attemted to free invalid extension struct\n");
163 if (extensions
->ext_array
)
164 vk_free(&instance
->alloc
, extensions
->ext_array
);
168 is_extension_enabled(const VkExtensionProperties
*extensions
,
172 assert(extensions
&& name
);
174 for (uint32_t i
= 0; i
< num_ext
; i
++) {
175 if (strcmp(name
, extensions
[i
].extensionName
) == 0)
183 radv_physical_device_init(struct radv_physical_device
*device
,
184 struct radv_instance
*instance
,
188 drmVersionPtr version
;
191 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
193 return VK_ERROR_INCOMPATIBLE_DRIVER
;
195 version
= drmGetVersion(fd
);
198 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
199 "failed to get version %s: %m", path
);
202 if (strcmp(version
->name
, "amdgpu")) {
203 drmFreeVersion(version
);
205 return VK_ERROR_INCOMPATIBLE_DRIVER
;
207 drmFreeVersion(version
);
209 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
210 device
->instance
= instance
;
211 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
212 strncpy(device
->path
, path
, ARRAY_SIZE(device
->path
));
214 device
->ws
= radv_amdgpu_winsys_create(fd
);
216 result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
219 device
->ws
->query_info(device
->ws
, &device
->rad_info
);
220 result
= radv_init_wsi(device
);
221 if (result
!= VK_SUCCESS
) {
222 device
->ws
->destroy(device
->ws
);
226 if (radv_device_get_cache_uuid(device
->rad_info
.family
, device
->uuid
)) {
227 radv_finish_wsi(device
);
228 device
->ws
->destroy(device
->ws
);
229 result
= vk_errorf(VK_ERROR_INITIALIZATION_FAILED
,
230 "cannot generate UUID");
234 result
= radv_extensions_register(instance
,
236 common_device_extensions
,
237 ARRAY_SIZE(common_device_extensions
));
238 if (result
!= VK_SUCCESS
)
241 fprintf(stderr
, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
242 device
->name
= device
->rad_info
.name
;
252 radv_physical_device_finish(struct radv_physical_device
*device
)
254 radv_extensions_finish(device
->instance
, &device
->extensions
);
255 radv_finish_wsi(device
);
256 device
->ws
->destroy(device
->ws
);
261 default_alloc_func(void *pUserData
, size_t size
, size_t align
,
262 VkSystemAllocationScope allocationScope
)
268 default_realloc_func(void *pUserData
, void *pOriginal
, size_t size
,
269 size_t align
, VkSystemAllocationScope allocationScope
)
271 return realloc(pOriginal
, size
);
275 default_free_func(void *pUserData
, void *pMemory
)
280 static const VkAllocationCallbacks default_alloc
= {
282 .pfnAllocation
= default_alloc_func
,
283 .pfnReallocation
= default_realloc_func
,
284 .pfnFree
= default_free_func
,
287 static const struct debug_control radv_debug_options
[] = {
288 {"fastclears", RADV_DEBUG_FAST_CLEARS
},
289 {"nodcc", RADV_DEBUG_NO_DCC
},
290 {"shaders", RADV_DEBUG_DUMP_SHADERS
},
291 {"nocache", RADV_DEBUG_NO_CACHE
},
292 {"shaderstats", RADV_DEBUG_DUMP_SHADER_STATS
},
293 {"nohiz", RADV_DEBUG_NO_HIZ
},
294 {"nocompute", RADV_DEBUG_NO_COMPUTE_QUEUE
},
295 {"unsafemath", RADV_DEBUG_UNSAFE_MATH
},
299 VkResult
radv_CreateInstance(
300 const VkInstanceCreateInfo
* pCreateInfo
,
301 const VkAllocationCallbacks
* pAllocator
,
302 VkInstance
* pInstance
)
304 struct radv_instance
*instance
;
306 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
308 uint32_t client_version
;
309 if (pCreateInfo
->pApplicationInfo
&&
310 pCreateInfo
->pApplicationInfo
->apiVersion
!= 0) {
311 client_version
= pCreateInfo
->pApplicationInfo
->apiVersion
;
313 client_version
= VK_MAKE_VERSION(1, 0, 0);
316 if (VK_MAKE_VERSION(1, 0, 0) > client_version
||
317 client_version
> VK_MAKE_VERSION(1, 0, 0xfff)) {
318 return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER
,
319 "Client requested version %d.%d.%d",
320 VK_VERSION_MAJOR(client_version
),
321 VK_VERSION_MINOR(client_version
),
322 VK_VERSION_PATCH(client_version
));
325 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
326 if (!is_extension_enabled(instance_extensions
,
327 ARRAY_SIZE(instance_extensions
),
328 pCreateInfo
->ppEnabledExtensionNames
[i
]))
329 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
332 instance
= vk_alloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
333 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
335 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
337 memset(instance
, 0, sizeof(*instance
));
339 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
342 instance
->alloc
= *pAllocator
;
344 instance
->alloc
= default_alloc
;
346 instance
->apiVersion
= client_version
;
347 instance
->physicalDeviceCount
= -1;
351 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
353 instance
->debug_flags
= parse_debug_string(getenv("RADV_DEBUG"),
356 *pInstance
= radv_instance_to_handle(instance
);
361 void radv_DestroyInstance(
362 VkInstance _instance
,
363 const VkAllocationCallbacks
* pAllocator
)
365 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
367 for (int i
= 0; i
< instance
->physicalDeviceCount
; ++i
) {
368 radv_physical_device_finish(instance
->physicalDevices
+ i
);
371 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
375 vk_free(&instance
->alloc
, instance
);
378 VkResult
radv_EnumeratePhysicalDevices(
379 VkInstance _instance
,
380 uint32_t* pPhysicalDeviceCount
,
381 VkPhysicalDevice
* pPhysicalDevices
)
383 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
386 if (instance
->physicalDeviceCount
< 0) {
388 instance
->physicalDeviceCount
= 0;
389 for (unsigned i
= 0; i
< RADV_MAX_DRM_DEVICES
; i
++) {
390 snprintf(path
, sizeof(path
), "/dev/dri/renderD%d", 128 + i
);
391 result
= radv_physical_device_init(instance
->physicalDevices
+
392 instance
->physicalDeviceCount
,
394 if (result
== VK_SUCCESS
)
395 ++instance
->physicalDeviceCount
;
396 else if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
401 if (!pPhysicalDevices
) {
402 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
404 *pPhysicalDeviceCount
= MIN2(*pPhysicalDeviceCount
, instance
->physicalDeviceCount
);
405 for (unsigned i
= 0; i
< *pPhysicalDeviceCount
; ++i
)
406 pPhysicalDevices
[i
] = radv_physical_device_to_handle(instance
->physicalDevices
+ i
);
409 return *pPhysicalDeviceCount
< instance
->physicalDeviceCount
? VK_INCOMPLETE
413 void radv_GetPhysicalDeviceFeatures(
414 VkPhysicalDevice physicalDevice
,
415 VkPhysicalDeviceFeatures
* pFeatures
)
417 // RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
419 memset(pFeatures
, 0, sizeof(*pFeatures
));
421 *pFeatures
= (VkPhysicalDeviceFeatures
) {
422 .robustBufferAccess
= true,
423 .fullDrawIndexUint32
= true,
424 .imageCubeArray
= true,
425 .independentBlend
= true,
426 .geometryShader
= false,
427 .tessellationShader
= false,
428 .sampleRateShading
= false,
429 .dualSrcBlend
= true,
431 .multiDrawIndirect
= true,
432 .drawIndirectFirstInstance
= true,
434 .depthBiasClamp
= true,
435 .fillModeNonSolid
= true,
440 .multiViewport
= false,
441 .samplerAnisotropy
= true,
442 .textureCompressionETC2
= false,
443 .textureCompressionASTC_LDR
= false,
444 .textureCompressionBC
= true,
445 .occlusionQueryPrecise
= true,
446 .pipelineStatisticsQuery
= false,
447 .vertexPipelineStoresAndAtomics
= true,
448 .fragmentStoresAndAtomics
= true,
449 .shaderTessellationAndGeometryPointSize
= true,
450 .shaderImageGatherExtended
= true,
451 .shaderStorageImageExtendedFormats
= true,
452 .shaderStorageImageMultisample
= false,
453 .shaderUniformBufferArrayDynamicIndexing
= true,
454 .shaderSampledImageArrayDynamicIndexing
= true,
455 .shaderStorageBufferArrayDynamicIndexing
= true,
456 .shaderStorageImageArrayDynamicIndexing
= true,
457 .shaderStorageImageReadWithoutFormat
= false,
458 .shaderStorageImageWriteWithoutFormat
= false,
459 .shaderClipDistance
= true,
460 .shaderCullDistance
= true,
461 .shaderFloat64
= false,
462 .shaderInt64
= false,
463 .shaderInt16
= false,
465 .variableMultisampleRate
= false,
466 .inheritedQueries
= false,
470 void radv_GetPhysicalDeviceProperties(
471 VkPhysicalDevice physicalDevice
,
472 VkPhysicalDeviceProperties
* pProperties
)
474 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
475 VkSampleCountFlags sample_counts
= 0xf;
476 VkPhysicalDeviceLimits limits
= {
477 .maxImageDimension1D
= (1 << 14),
478 .maxImageDimension2D
= (1 << 14),
479 .maxImageDimension3D
= (1 << 11),
480 .maxImageDimensionCube
= (1 << 14),
481 .maxImageArrayLayers
= (1 << 11),
482 .maxTexelBufferElements
= 128 * 1024 * 1024,
483 .maxUniformBufferRange
= UINT32_MAX
,
484 .maxStorageBufferRange
= UINT32_MAX
,
485 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
486 .maxMemoryAllocationCount
= UINT32_MAX
,
487 .maxSamplerAllocationCount
= 64 * 1024,
488 .bufferImageGranularity
= 64, /* A cache line */
489 .sparseAddressSpaceSize
= 0,
490 .maxBoundDescriptorSets
= MAX_SETS
,
491 .maxPerStageDescriptorSamplers
= 64,
492 .maxPerStageDescriptorUniformBuffers
= 64,
493 .maxPerStageDescriptorStorageBuffers
= 64,
494 .maxPerStageDescriptorSampledImages
= 64,
495 .maxPerStageDescriptorStorageImages
= 64,
496 .maxPerStageDescriptorInputAttachments
= 64,
497 .maxPerStageResources
= 128,
498 .maxDescriptorSetSamplers
= 256,
499 .maxDescriptorSetUniformBuffers
= 256,
500 .maxDescriptorSetUniformBuffersDynamic
= 256,
501 .maxDescriptorSetStorageBuffers
= 256,
502 .maxDescriptorSetStorageBuffersDynamic
= 256,
503 .maxDescriptorSetSampledImages
= 256,
504 .maxDescriptorSetStorageImages
= 256,
505 .maxDescriptorSetInputAttachments
= 256,
506 .maxVertexInputAttributes
= 32,
507 .maxVertexInputBindings
= 32,
508 .maxVertexInputAttributeOffset
= 2047,
509 .maxVertexInputBindingStride
= 2048,
510 .maxVertexOutputComponents
= 128,
511 .maxTessellationGenerationLevel
= 0,
512 .maxTessellationPatchSize
= 0,
513 .maxTessellationControlPerVertexInputComponents
= 0,
514 .maxTessellationControlPerVertexOutputComponents
= 0,
515 .maxTessellationControlPerPatchOutputComponents
= 0,
516 .maxTessellationControlTotalOutputComponents
= 0,
517 .maxTessellationEvaluationInputComponents
= 0,
518 .maxTessellationEvaluationOutputComponents
= 0,
519 .maxGeometryShaderInvocations
= 32,
520 .maxGeometryInputComponents
= 64,
521 .maxGeometryOutputComponents
= 128,
522 .maxGeometryOutputVertices
= 256,
523 .maxGeometryTotalOutputComponents
= 1024,
524 .maxFragmentInputComponents
= 128,
525 .maxFragmentOutputAttachments
= 8,
526 .maxFragmentDualSrcAttachments
= 1,
527 .maxFragmentCombinedOutputResources
= 8,
528 .maxComputeSharedMemorySize
= 32768,
529 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
530 .maxComputeWorkGroupInvocations
= 2048,
531 .maxComputeWorkGroupSize
= {
536 .subPixelPrecisionBits
= 4 /* FIXME */,
537 .subTexelPrecisionBits
= 4 /* FIXME */,
538 .mipmapPrecisionBits
= 4 /* FIXME */,
539 .maxDrawIndexedIndexValue
= UINT32_MAX
,
540 .maxDrawIndirectCount
= UINT32_MAX
,
541 .maxSamplerLodBias
= 16,
542 .maxSamplerAnisotropy
= 16,
543 .maxViewports
= MAX_VIEWPORTS
,
544 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
545 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
546 .viewportSubPixelBits
= 13, /* We take a float? */
547 .minMemoryMapAlignment
= 4096, /* A page */
548 .minTexelBufferOffsetAlignment
= 1,
549 .minUniformBufferOffsetAlignment
= 4,
550 .minStorageBufferOffsetAlignment
= 4,
551 .minTexelOffset
= -32,
552 .maxTexelOffset
= 31,
553 .minTexelGatherOffset
= -32,
554 .maxTexelGatherOffset
= 31,
555 .minInterpolationOffset
= -2,
556 .maxInterpolationOffset
= 2,
557 .subPixelInterpolationOffsetBits
= 8,
558 .maxFramebufferWidth
= (1 << 14),
559 .maxFramebufferHeight
= (1 << 14),
560 .maxFramebufferLayers
= (1 << 10),
561 .framebufferColorSampleCounts
= sample_counts
,
562 .framebufferDepthSampleCounts
= sample_counts
,
563 .framebufferStencilSampleCounts
= sample_counts
,
564 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
565 .maxColorAttachments
= MAX_RTS
,
566 .sampledImageColorSampleCounts
= sample_counts
,
567 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
568 .sampledImageDepthSampleCounts
= sample_counts
,
569 .sampledImageStencilSampleCounts
= sample_counts
,
570 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
571 .maxSampleMaskWords
= 1,
572 .timestampComputeAndGraphics
= false,
573 .timestampPeriod
= 100000.0 / pdevice
->rad_info
.clock_crystal_freq
,
574 .maxClipDistances
= 8,
575 .maxCullDistances
= 8,
576 .maxCombinedClipAndCullDistances
= 8,
577 .discreteQueuePriorities
= 1,
578 .pointSizeRange
= { 0.125, 255.875 },
579 .lineWidthRange
= { 0.0, 7.9921875 },
580 .pointSizeGranularity
= (1.0 / 8.0),
581 .lineWidthGranularity
= (1.0 / 128.0),
582 .strictLines
= false, /* FINISHME */
583 .standardSampleLocations
= true,
584 .optimalBufferCopyOffsetAlignment
= 128,
585 .optimalBufferCopyRowPitchAlignment
= 128,
586 .nonCoherentAtomSize
= 64,
589 *pProperties
= (VkPhysicalDeviceProperties
) {
590 .apiVersion
= VK_MAKE_VERSION(1, 0, 5),
593 .deviceID
= pdevice
->rad_info
.pci_id
,
594 .deviceType
= VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
,
596 .sparseProperties
= {0}, /* Broadwell doesn't do sparse. */
599 strcpy(pProperties
->deviceName
, pdevice
->name
);
600 memcpy(pProperties
->pipelineCacheUUID
, pdevice
->uuid
, VK_UUID_SIZE
);
603 void radv_GetPhysicalDeviceQueueFamilyProperties(
604 VkPhysicalDevice physicalDevice
,
606 VkQueueFamilyProperties
* pQueueFamilyProperties
)
608 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
609 int num_queue_families
= 1;
611 if (pdevice
->rad_info
.compute_rings
> 0 &&
612 pdevice
->rad_info
.chip_class
>= CIK
&&
613 !(pdevice
->instance
->debug_flags
& RADV_DEBUG_NO_COMPUTE_QUEUE
))
614 num_queue_families
++;
616 if (pQueueFamilyProperties
== NULL
) {
617 *pCount
= num_queue_families
;
626 pQueueFamilyProperties
[idx
] = (VkQueueFamilyProperties
) {
627 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
628 VK_QUEUE_COMPUTE_BIT
|
629 VK_QUEUE_TRANSFER_BIT
,
631 .timestampValidBits
= 64,
632 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
637 if (pdevice
->rad_info
.compute_rings
> 0 &&
638 pdevice
->rad_info
.chip_class
>= CIK
&&
639 !(pdevice
->instance
->debug_flags
& RADV_DEBUG_NO_COMPUTE_QUEUE
)) {
641 pQueueFamilyProperties
[idx
] = (VkQueueFamilyProperties
) {
642 .queueFlags
= VK_QUEUE_COMPUTE_BIT
| VK_QUEUE_TRANSFER_BIT
,
643 .queueCount
= pdevice
->rad_info
.compute_rings
,
644 .timestampValidBits
= 64,
645 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
653 void radv_GetPhysicalDeviceMemoryProperties(
654 VkPhysicalDevice physicalDevice
,
655 VkPhysicalDeviceMemoryProperties
* pMemoryProperties
)
657 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
659 STATIC_ASSERT(RADV_MEM_TYPE_COUNT
<= VK_MAX_MEMORY_TYPES
);
661 pMemoryProperties
->memoryTypeCount
= RADV_MEM_TYPE_COUNT
;
662 pMemoryProperties
->memoryTypes
[RADV_MEM_TYPE_VRAM
] = (VkMemoryType
) {
663 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
,
664 .heapIndex
= RADV_MEM_HEAP_VRAM
,
666 pMemoryProperties
->memoryTypes
[RADV_MEM_TYPE_GTT_WRITE_COMBINE
] = (VkMemoryType
) {
667 .propertyFlags
= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
668 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
669 .heapIndex
= RADV_MEM_HEAP_GTT
,
671 pMemoryProperties
->memoryTypes
[RADV_MEM_TYPE_VRAM_CPU_ACCESS
] = (VkMemoryType
) {
672 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
673 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
674 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
675 .heapIndex
= RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
677 pMemoryProperties
->memoryTypes
[RADV_MEM_TYPE_GTT_CACHED
] = (VkMemoryType
) {
678 .propertyFlags
= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
679 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
680 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
,
681 .heapIndex
= RADV_MEM_HEAP_GTT
,
684 STATIC_ASSERT(RADV_MEM_HEAP_COUNT
<= VK_MAX_MEMORY_HEAPS
);
686 pMemoryProperties
->memoryHeapCount
= RADV_MEM_HEAP_COUNT
;
687 pMemoryProperties
->memoryHeaps
[RADV_MEM_HEAP_VRAM
] = (VkMemoryHeap
) {
688 .size
= physical_device
->rad_info
.vram_size
-
689 physical_device
->rad_info
.visible_vram_size
,
690 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
692 pMemoryProperties
->memoryHeaps
[RADV_MEM_HEAP_VRAM_CPU_ACCESS
] = (VkMemoryHeap
) {
693 .size
= physical_device
->rad_info
.visible_vram_size
,
694 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
696 pMemoryProperties
->memoryHeaps
[RADV_MEM_HEAP_GTT
] = (VkMemoryHeap
) {
697 .size
= physical_device
->rad_info
.gart_size
,
703 radv_queue_init(struct radv_device
*device
, struct radv_queue
*queue
,
704 int queue_family_index
, int idx
)
706 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
707 queue
->device
= device
;
708 queue
->queue_family_index
= queue_family_index
;
709 queue
->queue_idx
= idx
;
711 queue
->hw_ctx
= device
->ws
->ctx_create(device
->ws
);
713 return VK_ERROR_OUT_OF_HOST_MEMORY
;
719 radv_queue_finish(struct radv_queue
*queue
)
722 queue
->device
->ws
->ctx_destroy(queue
->hw_ctx
);
725 VkResult
radv_CreateDevice(
726 VkPhysicalDevice physicalDevice
,
727 const VkDeviceCreateInfo
* pCreateInfo
,
728 const VkAllocationCallbacks
* pAllocator
,
731 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
733 struct radv_device
*device
;
735 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
736 if (!is_extension_enabled(physical_device
->extensions
.ext_array
,
737 physical_device
->extensions
.num_ext
,
738 pCreateInfo
->ppEnabledExtensionNames
[i
]))
739 return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT
);
742 device
= vk_alloc2(&physical_device
->instance
->alloc
, pAllocator
,
744 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
746 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
748 memset(device
, 0, sizeof(*device
));
750 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
751 device
->instance
= physical_device
->instance
;
752 device
->physical_device
= physical_device
;
754 device
->debug_flags
= device
->instance
->debug_flags
;
756 device
->ws
= physical_device
->ws
;
758 device
->alloc
= *pAllocator
;
760 device
->alloc
= physical_device
->instance
->alloc
;
762 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
763 const VkDeviceQueueCreateInfo
*queue_create
= &pCreateInfo
->pQueueCreateInfos
[i
];
764 uint32_t qfi
= queue_create
->queueFamilyIndex
;
766 device
->queues
[qfi
] = vk_alloc(&device
->alloc
,
767 queue_create
->queueCount
* sizeof(struct radv_queue
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
768 if (!device
->queues
[qfi
]) {
769 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
773 memset(device
->queues
[qfi
], 0, queue_create
->queueCount
* sizeof(struct radv_queue
));
775 device
->queue_count
[qfi
] = queue_create
->queueCount
;
777 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
778 result
= radv_queue_init(device
, &device
->queues
[qfi
][q
], qfi
, q
);
779 if (result
!= VK_SUCCESS
)
784 result
= radv_device_init_meta(device
);
785 if (result
!= VK_SUCCESS
)
788 radv_device_init_msaa(device
);
790 for (int family
= 0; family
< RADV_MAX_QUEUE_FAMILIES
; ++family
) {
791 device
->empty_cs
[family
] = device
->ws
->cs_create(device
->ws
, family
);
793 case RADV_QUEUE_GENERAL
:
794 radeon_emit(device
->empty_cs
[family
], PKT3(PKT3_CONTEXT_CONTROL
, 1, 0));
795 radeon_emit(device
->empty_cs
[family
], CONTEXT_CONTROL_LOAD_ENABLE(1));
796 radeon_emit(device
->empty_cs
[family
], CONTEXT_CONTROL_SHADOW_ENABLE(1));
798 case RADV_QUEUE_COMPUTE
:
799 radeon_emit(device
->empty_cs
[family
], PKT3(PKT3_NOP
, 0, 0));
800 radeon_emit(device
->empty_cs
[family
], 0);
803 device
->ws
->cs_finalize(device
->empty_cs
[family
]);
806 if (getenv("RADV_TRACE_FILE")) {
807 device
->trace_bo
= device
->ws
->buffer_create(device
->ws
, 4096, 8,
808 RADEON_DOMAIN_VRAM
, RADEON_FLAG_CPU_ACCESS
);
809 if (!device
->trace_bo
)
812 device
->trace_id_ptr
= device
->ws
->buffer_map(device
->trace_bo
);
813 if (!device
->trace_id_ptr
)
817 *pDevice
= radv_device_to_handle(device
);
821 if (device
->trace_bo
)
822 device
->ws
->buffer_destroy(device
->trace_bo
);
824 for (unsigned i
= 0; i
< RADV_MAX_QUEUE_FAMILIES
; i
++) {
825 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
826 radv_queue_finish(&device
->queues
[i
][q
]);
827 if (device
->queue_count
[i
])
828 vk_free(&device
->alloc
, device
->queues
[i
]);
831 vk_free(&device
->alloc
, device
);
835 void radv_DestroyDevice(
837 const VkAllocationCallbacks
* pAllocator
)
839 RADV_FROM_HANDLE(radv_device
, device
, _device
);
841 if (device
->trace_bo
)
842 device
->ws
->buffer_destroy(device
->trace_bo
);
844 for (unsigned i
= 0; i
< RADV_MAX_QUEUE_FAMILIES
; i
++) {
845 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
846 radv_queue_finish(&device
->queues
[i
][q
]);
847 if (device
->queue_count
[i
])
848 vk_free(&device
->alloc
, device
->queues
[i
]);
850 radv_device_finish_meta(device
);
852 vk_free(&device
->alloc
, device
);
855 VkResult
radv_EnumerateInstanceExtensionProperties(
856 const char* pLayerName
,
857 uint32_t* pPropertyCount
,
858 VkExtensionProperties
* pProperties
)
860 if (pProperties
== NULL
) {
861 *pPropertyCount
= ARRAY_SIZE(instance_extensions
);
865 *pPropertyCount
= MIN2(*pPropertyCount
, ARRAY_SIZE(instance_extensions
));
866 typed_memcpy(pProperties
, instance_extensions
, *pPropertyCount
);
868 if (*pPropertyCount
< ARRAY_SIZE(instance_extensions
))
869 return VK_INCOMPLETE
;
874 VkResult
radv_EnumerateDeviceExtensionProperties(
875 VkPhysicalDevice physicalDevice
,
876 const char* pLayerName
,
877 uint32_t* pPropertyCount
,
878 VkExtensionProperties
* pProperties
)
880 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
882 if (pProperties
== NULL
) {
883 *pPropertyCount
= pdevice
->extensions
.num_ext
;
887 *pPropertyCount
= MIN2(*pPropertyCount
, pdevice
->extensions
.num_ext
);
888 typed_memcpy(pProperties
, pdevice
->extensions
.ext_array
, *pPropertyCount
);
890 if (*pPropertyCount
< pdevice
->extensions
.num_ext
)
891 return VK_INCOMPLETE
;
896 VkResult
radv_EnumerateInstanceLayerProperties(
897 uint32_t* pPropertyCount
,
898 VkLayerProperties
* pProperties
)
900 if (pProperties
== NULL
) {
905 /* None supported at this time */
906 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
909 VkResult
radv_EnumerateDeviceLayerProperties(
910 VkPhysicalDevice physicalDevice
,
911 uint32_t* pPropertyCount
,
912 VkLayerProperties
* pProperties
)
914 if (pProperties
== NULL
) {
919 /* None supported at this time */
920 return vk_error(VK_ERROR_LAYER_NOT_PRESENT
);
923 void radv_GetDeviceQueue(
925 uint32_t queueFamilyIndex
,
929 RADV_FROM_HANDLE(radv_device
, device
, _device
);
931 *pQueue
= radv_queue_to_handle(&device
->queues
[queueFamilyIndex
][queueIndex
]);
934 static void radv_dump_trace(struct radv_device
*device
,
935 struct radeon_winsys_cs
*cs
)
937 const char *filename
= getenv("RADV_TRACE_FILE");
938 FILE *f
= fopen(filename
, "w");
940 fprintf(stderr
, "Failed to write trace dump to %s\n", filename
);
944 fprintf(f
, "Trace ID: %x\n", *device
->trace_id_ptr
);
945 device
->ws
->cs_dump(cs
, f
, *device
->trace_id_ptr
);
949 VkResult
radv_QueueSubmit(
951 uint32_t submitCount
,
952 const VkSubmitInfo
* pSubmits
,
955 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
956 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
957 struct radeon_winsys_fence
*base_fence
= fence
? fence
->fence
: NULL
;
958 struct radeon_winsys_ctx
*ctx
= queue
->hw_ctx
;
960 uint32_t max_cs_submission
= queue
->device
->trace_bo
? 1 : UINT32_MAX
;
962 for (uint32_t i
= 0; i
< submitCount
; i
++) {
963 struct radeon_winsys_cs
**cs_array
;
964 bool can_patch
= true;
967 if (!pSubmits
[i
].commandBufferCount
)
970 cs_array
= malloc(sizeof(struct radeon_winsys_cs
*) *
971 pSubmits
[i
].commandBufferCount
);
973 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
974 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
,
975 pSubmits
[i
].pCommandBuffers
[j
]);
976 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
978 cs_array
[j
] = cmd_buffer
->cs
;
979 if ((cmd_buffer
->usage_flags
& VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT
))
983 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
+= advance
) {
984 advance
= MIN2(max_cs_submission
,
985 pSubmits
[i
].commandBufferCount
- j
);
987 bool e
= j
+ advance
== pSubmits
[i
].commandBufferCount
;
989 if (queue
->device
->trace_bo
)
990 *queue
->device
->trace_id_ptr
= 0;
992 ret
= queue
->device
->ws
->cs_submit(ctx
, queue
->queue_idx
, cs_array
,
993 pSubmits
[i
].commandBufferCount
,
994 (struct radeon_winsys_sem
**)pSubmits
[i
].pWaitSemaphores
,
995 b
? pSubmits
[i
].waitSemaphoreCount
: 0,
996 (struct radeon_winsys_sem
**)pSubmits
[i
].pSignalSemaphores
,
997 e
? pSubmits
[i
].signalSemaphoreCount
: 0,
998 can_patch
, base_fence
);
1001 radv_loge("failed to submit CS %d\n", i
);
1004 if (queue
->device
->trace_bo
) {
1005 bool success
= queue
->device
->ws
->ctx_wait_idle(
1007 radv_queue_family_to_ring(
1008 queue
->queue_family_index
),
1011 if (!success
) { /* Hang */
1012 radv_dump_trace(queue
->device
, cs_array
[j
]);
1022 ret
= queue
->device
->ws
->cs_submit(ctx
, queue
->queue_idx
,
1023 &queue
->device
->empty_cs
[queue
->queue_family_index
],
1024 1, NULL
, 0, NULL
, 0, false, base_fence
);
1026 fence
->submitted
= true;
1032 VkResult
radv_QueueWaitIdle(
1035 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
1037 queue
->device
->ws
->ctx_wait_idle(queue
->hw_ctx
,
1038 radv_queue_family_to_ring(queue
->queue_family_index
),
1043 VkResult
radv_DeviceWaitIdle(
1046 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1048 for (unsigned i
= 0; i
< RADV_MAX_QUEUE_FAMILIES
; i
++) {
1049 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
1050 radv_QueueWaitIdle(radv_queue_to_handle(&device
->queues
[i
][q
]));
1056 PFN_vkVoidFunction
radv_GetInstanceProcAddr(
1057 VkInstance instance
,
1060 return radv_lookup_entrypoint(pName
);
1063 /* The loader wants us to expose a second GetInstanceProcAddr function
1064 * to work around certain LD_PRELOAD issues seen in apps.
1067 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
1068 VkInstance instance
,
1072 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
1073 VkInstance instance
,
1076 return radv_GetInstanceProcAddr(instance
, pName
);
1079 PFN_vkVoidFunction
radv_GetDeviceProcAddr(
1083 return radv_lookup_entrypoint(pName
);
1086 VkResult
radv_AllocateMemory(
1088 const VkMemoryAllocateInfo
* pAllocateInfo
,
1089 const VkAllocationCallbacks
* pAllocator
,
1090 VkDeviceMemory
* pMem
)
1092 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1093 struct radv_device_memory
*mem
;
1095 enum radeon_bo_domain domain
;
1097 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
1099 if (pAllocateInfo
->allocationSize
== 0) {
1100 /* Apparently, this is allowed */
1101 *pMem
= VK_NULL_HANDLE
;
1105 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
1106 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1108 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1110 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
1111 if (pAllocateInfo
->memoryTypeIndex
== RADV_MEM_TYPE_GTT_WRITE_COMBINE
||
1112 pAllocateInfo
->memoryTypeIndex
== RADV_MEM_TYPE_GTT_CACHED
)
1113 domain
= RADEON_DOMAIN_GTT
;
1115 domain
= RADEON_DOMAIN_VRAM
;
1117 if (pAllocateInfo
->memoryTypeIndex
== RADV_MEM_TYPE_VRAM
)
1118 flags
|= RADEON_FLAG_NO_CPU_ACCESS
;
1120 flags
|= RADEON_FLAG_CPU_ACCESS
;
1122 if (pAllocateInfo
->memoryTypeIndex
== RADV_MEM_TYPE_GTT_WRITE_COMBINE
)
1123 flags
|= RADEON_FLAG_GTT_WC
;
1125 mem
->bo
= device
->ws
->buffer_create(device
->ws
, alloc_size
, 32768,
1129 result
= VK_ERROR_OUT_OF_DEVICE_MEMORY
;
1132 mem
->type_index
= pAllocateInfo
->memoryTypeIndex
;
1134 *pMem
= radv_device_memory_to_handle(mem
);
1139 vk_free2(&device
->alloc
, pAllocator
, mem
);
1144 void radv_FreeMemory(
1146 VkDeviceMemory _mem
,
1147 const VkAllocationCallbacks
* pAllocator
)
1149 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1150 RADV_FROM_HANDLE(radv_device_memory
, mem
, _mem
);
1155 device
->ws
->buffer_destroy(mem
->bo
);
1158 vk_free2(&device
->alloc
, pAllocator
, mem
);
1161 VkResult
radv_MapMemory(
1163 VkDeviceMemory _memory
,
1164 VkDeviceSize offset
,
1166 VkMemoryMapFlags flags
,
1169 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1170 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1177 *ppData
= device
->ws
->buffer_map(mem
->bo
);
1183 return VK_ERROR_MEMORY_MAP_FAILED
;
1186 void radv_UnmapMemory(
1188 VkDeviceMemory _memory
)
1190 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1191 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1196 device
->ws
->buffer_unmap(mem
->bo
);
1199 VkResult
radv_FlushMappedMemoryRanges(
1201 uint32_t memoryRangeCount
,
1202 const VkMappedMemoryRange
* pMemoryRanges
)
1207 VkResult
radv_InvalidateMappedMemoryRanges(
1209 uint32_t memoryRangeCount
,
1210 const VkMappedMemoryRange
* pMemoryRanges
)
1215 void radv_GetBufferMemoryRequirements(
1218 VkMemoryRequirements
* pMemoryRequirements
)
1220 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1222 pMemoryRequirements
->memoryTypeBits
= (1u << RADV_MEM_TYPE_COUNT
) - 1;
1224 pMemoryRequirements
->size
= buffer
->size
;
1225 pMemoryRequirements
->alignment
= 16;
1228 void radv_GetImageMemoryRequirements(
1231 VkMemoryRequirements
* pMemoryRequirements
)
1233 RADV_FROM_HANDLE(radv_image
, image
, _image
);
1235 pMemoryRequirements
->memoryTypeBits
= (1u << RADV_MEM_TYPE_COUNT
) - 1;
1237 pMemoryRequirements
->size
= image
->size
;
1238 pMemoryRequirements
->alignment
= image
->alignment
;
1241 void radv_GetImageSparseMemoryRequirements(
1244 uint32_t* pSparseMemoryRequirementCount
,
1245 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
1250 void radv_GetDeviceMemoryCommitment(
1252 VkDeviceMemory memory
,
1253 VkDeviceSize
* pCommittedMemoryInBytes
)
1255 *pCommittedMemoryInBytes
= 0;
1258 VkResult
radv_BindBufferMemory(
1261 VkDeviceMemory _memory
,
1262 VkDeviceSize memoryOffset
)
1264 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1265 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1268 buffer
->bo
= mem
->bo
;
1269 buffer
->offset
= memoryOffset
;
1278 VkResult
radv_BindImageMemory(
1281 VkDeviceMemory _memory
,
1282 VkDeviceSize memoryOffset
)
1284 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
1285 RADV_FROM_HANDLE(radv_image
, image
, _image
);
1288 image
->bo
= mem
->bo
;
1289 image
->offset
= memoryOffset
;
1298 VkResult
radv_QueueBindSparse(
1300 uint32_t bindInfoCount
,
1301 const VkBindSparseInfo
* pBindInfo
,
1304 stub_return(VK_ERROR_INCOMPATIBLE_DRIVER
);
1307 VkResult
radv_CreateFence(
1309 const VkFenceCreateInfo
* pCreateInfo
,
1310 const VkAllocationCallbacks
* pAllocator
,
1313 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1314 struct radv_fence
*fence
= vk_alloc2(&device
->alloc
, pAllocator
,
1316 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1319 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1321 memset(fence
, 0, sizeof(*fence
));
1322 fence
->submitted
= false;
1323 fence
->signalled
= !!(pCreateInfo
->flags
& VK_FENCE_CREATE_SIGNALED_BIT
);
1324 fence
->fence
= device
->ws
->create_fence();
1325 if (!fence
->fence
) {
1326 vk_free2(&device
->alloc
, pAllocator
, fence
);
1327 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1330 *pFence
= radv_fence_to_handle(fence
);
1335 void radv_DestroyFence(
1338 const VkAllocationCallbacks
* pAllocator
)
1340 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1341 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
1345 device
->ws
->destroy_fence(fence
->fence
);
1346 vk_free2(&device
->alloc
, pAllocator
, fence
);
1349 static uint64_t radv_get_absolute_timeout(uint64_t timeout
)
1351 uint64_t current_time
;
1354 clock_gettime(CLOCK_MONOTONIC
, &tv
);
1355 current_time
= tv
.tv_nsec
+ tv
.tv_sec
*1000000000ull;
1357 timeout
= MIN2(UINT64_MAX
- current_time
, timeout
);
1359 return current_time
+ timeout
;
1362 VkResult
radv_WaitForFences(
1364 uint32_t fenceCount
,
1365 const VkFence
* pFences
,
1369 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1370 timeout
= radv_get_absolute_timeout(timeout
);
1372 if (!waitAll
&& fenceCount
> 1) {
1373 fprintf(stderr
, "radv: WaitForFences without waitAll not implemented yet\n");
1376 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
1377 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
1378 bool expired
= false;
1380 if (fence
->signalled
)
1383 if (!fence
->submitted
)
1386 expired
= device
->ws
->fence_wait(device
->ws
, fence
->fence
, true, timeout
);
1390 fence
->signalled
= true;
1396 VkResult
radv_ResetFences(VkDevice device
,
1397 uint32_t fenceCount
,
1398 const VkFence
*pFences
)
1400 for (unsigned i
= 0; i
< fenceCount
; ++i
) {
1401 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
1402 fence
->submitted
= fence
->signalled
= false;
1408 VkResult
radv_GetFenceStatus(VkDevice _device
, VkFence _fence
)
1410 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1411 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
1413 if (fence
->signalled
)
1415 if (!fence
->submitted
)
1416 return VK_NOT_READY
;
1418 if (!device
->ws
->fence_wait(device
->ws
, fence
->fence
, false, 0))
1419 return VK_NOT_READY
;
1425 // Queue semaphore functions
1427 VkResult
radv_CreateSemaphore(
1429 const VkSemaphoreCreateInfo
* pCreateInfo
,
1430 const VkAllocationCallbacks
* pAllocator
,
1431 VkSemaphore
* pSemaphore
)
1433 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1434 struct radeon_winsys_sem
*sem
;
1436 sem
= device
->ws
->create_sem(device
->ws
);
1438 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1440 *pSemaphore
= (VkSemaphore
)sem
;
1444 void radv_DestroySemaphore(
1446 VkSemaphore _semaphore
,
1447 const VkAllocationCallbacks
* pAllocator
)
1449 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1450 struct radeon_winsys_sem
*sem
;
1454 sem
= (struct radeon_winsys_sem
*)_semaphore
;
1455 device
->ws
->destroy_sem(sem
);
1458 VkResult
radv_CreateEvent(
1460 const VkEventCreateInfo
* pCreateInfo
,
1461 const VkAllocationCallbacks
* pAllocator
,
1464 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1465 struct radv_event
*event
= vk_alloc2(&device
->alloc
, pAllocator
,
1467 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1470 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1472 event
->bo
= device
->ws
->buffer_create(device
->ws
, 8, 8,
1474 RADEON_FLAG_CPU_ACCESS
);
1476 vk_free2(&device
->alloc
, pAllocator
, event
);
1477 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
1480 event
->map
= (uint64_t*)device
->ws
->buffer_map(event
->bo
);
1482 *pEvent
= radv_event_to_handle(event
);
1487 void radv_DestroyEvent(
1490 const VkAllocationCallbacks
* pAllocator
)
1492 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1493 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1497 device
->ws
->buffer_destroy(event
->bo
);
1498 vk_free2(&device
->alloc
, pAllocator
, event
);
1501 VkResult
radv_GetEventStatus(
1505 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1507 if (*event
->map
== 1)
1508 return VK_EVENT_SET
;
1509 return VK_EVENT_RESET
;
1512 VkResult
radv_SetEvent(
1516 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1522 VkResult
radv_ResetEvent(
1526 RADV_FROM_HANDLE(radv_event
, event
, _event
);
1532 VkResult
radv_CreateBuffer(
1534 const VkBufferCreateInfo
* pCreateInfo
,
1535 const VkAllocationCallbacks
* pAllocator
,
1538 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1539 struct radv_buffer
*buffer
;
1541 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1543 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
1544 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1546 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1548 buffer
->size
= pCreateInfo
->size
;
1549 buffer
->usage
= pCreateInfo
->usage
;
1553 *pBuffer
= radv_buffer_to_handle(buffer
);
1558 void radv_DestroyBuffer(
1561 const VkAllocationCallbacks
* pAllocator
)
1563 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1564 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
1569 vk_free2(&device
->alloc
, pAllocator
, buffer
);
1572 static inline unsigned
1573 si_tile_mode_index(const struct radv_image
*image
, unsigned level
, bool stencil
)
1576 return image
->surface
.stencil_tiling_index
[level
];
1578 return image
->surface
.tiling_index
[level
];
1582 radv_initialise_color_surface(struct radv_device
*device
,
1583 struct radv_color_buffer_info
*cb
,
1584 struct radv_image_view
*iview
)
1586 const struct vk_format_description
*desc
;
1587 unsigned ntype
, format
, swap
, endian
;
1588 unsigned blend_clamp
= 0, blend_bypass
= 0;
1589 unsigned pitch_tile_max
, slice_tile_max
, tile_mode_index
;
1591 const struct radeon_surf
*surf
= &iview
->image
->surface
;
1592 const struct radeon_surf_level
*level_info
= &surf
->level
[iview
->base_mip
];
1594 desc
= vk_format_description(iview
->vk_format
);
1596 memset(cb
, 0, sizeof(*cb
));
1598 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1599 va
+= level_info
->offset
;
1600 cb
->cb_color_base
= va
>> 8;
1602 /* CMASK variables */
1603 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1604 va
+= iview
->image
->cmask
.offset
;
1605 cb
->cb_color_cmask
= va
>> 8;
1606 cb
->cb_color_cmask_slice
= iview
->image
->cmask
.slice_tile_max
;
1608 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1609 va
+= iview
->image
->dcc_offset
;
1610 cb
->cb_dcc_base
= va
>> 8;
1612 cb
->cb_color_view
= S_028C6C_SLICE_START(iview
->base_layer
) |
1613 S_028C6C_SLICE_MAX(iview
->base_layer
+ iview
->extent
.depth
- 1);
1615 cb
->micro_tile_mode
= iview
->image
->surface
.micro_tile_mode
;
1616 pitch_tile_max
= level_info
->nblk_x
/ 8 - 1;
1617 slice_tile_max
= (level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1;
1618 tile_mode_index
= si_tile_mode_index(iview
->image
, iview
->base_mip
, false);
1620 cb
->cb_color_pitch
= S_028C64_TILE_MAX(pitch_tile_max
);
1621 cb
->cb_color_slice
= S_028C68_TILE_MAX(slice_tile_max
);
1623 /* Intensity is implemented as Red, so treat it that way. */
1624 cb
->cb_color_attrib
= S_028C74_FORCE_DST_ALPHA_1(desc
->swizzle
[3] == VK_SWIZZLE_1
) |
1625 S_028C74_TILE_MODE_INDEX(tile_mode_index
);
1627 if (iview
->image
->samples
> 1) {
1628 unsigned log_samples
= util_logbase2(iview
->image
->samples
);
1630 cb
->cb_color_attrib
|= S_028C74_NUM_SAMPLES(log_samples
) |
1631 S_028C74_NUM_FRAGMENTS(log_samples
);
1634 if (iview
->image
->fmask
.size
) {
1635 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
+ iview
->image
->fmask
.offset
;
1636 if (device
->physical_device
->rad_info
.chip_class
>= CIK
)
1637 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(iview
->image
->fmask
.pitch_in_pixels
/ 8 - 1);
1638 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(iview
->image
->fmask
.tile_mode_index
);
1639 cb
->cb_color_fmask
= va
>> 8;
1640 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(iview
->image
->fmask
.slice_tile_max
);
1642 /* This must be set for fast clear to work without FMASK. */
1643 if (device
->physical_device
->rad_info
.chip_class
>= CIK
)
1644 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(pitch_tile_max
);
1645 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index
);
1646 cb
->cb_color_fmask
= cb
->cb_color_base
;
1647 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(slice_tile_max
);
1650 ntype
= radv_translate_color_numformat(iview
->vk_format
,
1652 vk_format_get_first_non_void_channel(iview
->vk_format
));
1653 format
= radv_translate_colorformat(iview
->vk_format
);
1654 if (format
== V_028C70_COLOR_INVALID
|| ntype
== ~0u)
1655 radv_finishme("Illegal color\n");
1656 swap
= radv_translate_colorswap(iview
->vk_format
, FALSE
);
1657 endian
= radv_colorformat_endian_swap(format
);
1659 /* blend clamp should be set for all NORM/SRGB types */
1660 if (ntype
== V_028C70_NUMBER_UNORM
||
1661 ntype
== V_028C70_NUMBER_SNORM
||
1662 ntype
== V_028C70_NUMBER_SRGB
)
1665 /* set blend bypass according to docs if SINT/UINT or
1666 8/24 COLOR variants */
1667 if (ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
||
1668 format
== V_028C70_COLOR_8_24
|| format
== V_028C70_COLOR_24_8
||
1669 format
== V_028C70_COLOR_X24_8_32_FLOAT
) {
1674 if ((ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
) &&
1675 (format
== V_028C70_COLOR_8
||
1676 format
== V_028C70_COLOR_8_8
||
1677 format
== V_028C70_COLOR_8_8_8_8
))
1678 ->color_is_int8
= true;
1680 cb
->cb_color_info
= S_028C70_FORMAT(format
) |
1681 S_028C70_COMP_SWAP(swap
) |
1682 S_028C70_BLEND_CLAMP(blend_clamp
) |
1683 S_028C70_BLEND_BYPASS(blend_bypass
) |
1684 S_028C70_SIMPLE_FLOAT(1) |
1685 S_028C70_ROUND_MODE(ntype
!= V_028C70_NUMBER_UNORM
&&
1686 ntype
!= V_028C70_NUMBER_SNORM
&&
1687 ntype
!= V_028C70_NUMBER_SRGB
&&
1688 format
!= V_028C70_COLOR_8_24
&&
1689 format
!= V_028C70_COLOR_24_8
) |
1690 S_028C70_NUMBER_TYPE(ntype
) |
1691 S_028C70_ENDIAN(endian
);
1692 if (iview
->image
->samples
> 1)
1693 if (iview
->image
->fmask
.size
)
1694 cb
->cb_color_info
|= S_028C70_COMPRESSION(1);
1696 if (iview
->image
->cmask
.size
&&
1697 (device
->debug_flags
& RADV_DEBUG_FAST_CLEARS
))
1698 cb
->cb_color_info
|= S_028C70_FAST_CLEAR(1);
1700 if (iview
->image
->surface
.dcc_size
&& level_info
->dcc_enabled
)
1701 cb
->cb_color_info
|= S_028C70_DCC_ENABLE(1);
1703 if (device
->physical_device
->rad_info
.chip_class
>= VI
) {
1704 unsigned max_uncompressed_block_size
= 2;
1705 if (iview
->image
->samples
> 1) {
1706 if (iview
->image
->surface
.bpe
== 1)
1707 max_uncompressed_block_size
= 0;
1708 else if (iview
->image
->surface
.bpe
== 2)
1709 max_uncompressed_block_size
= 1;
1712 cb
->cb_dcc_control
= S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size
) |
1713 S_028C78_INDEPENDENT_64B_BLOCKS(1);
1716 /* This must be set for fast clear to work without FMASK. */
1717 if (!iview
->image
->fmask
.size
&&
1718 device
->physical_device
->rad_info
.chip_class
== SI
) {
1719 unsigned bankh
= util_logbase2(iview
->image
->surface
.bankh
);
1720 cb
->cb_color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(bankh
);
1725 radv_initialise_ds_surface(struct radv_device
*device
,
1726 struct radv_ds_buffer_info
*ds
,
1727 struct radv_image_view
*iview
)
1729 unsigned level
= iview
->base_mip
;
1731 uint64_t va
, s_offs
, z_offs
;
1732 const struct radeon_surf_level
*level_info
= &iview
->image
->surface
.level
[level
];
1733 memset(ds
, 0, sizeof(*ds
));
1734 switch (iview
->vk_format
) {
1735 case VK_FORMAT_D24_UNORM_S8_UINT
:
1736 case VK_FORMAT_X8_D24_UNORM_PACK32
:
1737 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
1738 ds
->offset_scale
= 2.0f
;
1740 case VK_FORMAT_D16_UNORM
:
1741 case VK_FORMAT_D16_UNORM_S8_UINT
:
1742 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
1743 ds
->offset_scale
= 4.0f
;
1745 case VK_FORMAT_D32_SFLOAT
:
1746 case VK_FORMAT_D32_SFLOAT_S8_UINT
:
1747 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
1748 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
1749 ds
->offset_scale
= 1.0f
;
1755 format
= radv_translate_dbformat(iview
->vk_format
);
1756 if (format
== V_028040_Z_INVALID
) {
1757 fprintf(stderr
, "Invalid DB format: %d, disabling DB.\n", iview
->vk_format
);
1760 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
;
1761 s_offs
= z_offs
= va
;
1762 z_offs
+= iview
->image
->surface
.level
[level
].offset
;
1763 s_offs
+= iview
->image
->surface
.stencil_level
[level
].offset
;
1765 ds
->db_depth_view
= S_028008_SLICE_START(iview
->base_layer
) |
1766 S_028008_SLICE_MAX(iview
->base_layer
+ iview
->extent
.depth
- 1);
1767 ds
->db_depth_info
= S_02803C_ADDR5_SWIZZLE_MASK(1);
1768 ds
->db_z_info
= S_028040_FORMAT(format
) | S_028040_ZRANGE_PRECISION(1);
1770 if (iview
->image
->samples
> 1)
1771 ds
->db_z_info
|= S_028040_NUM_SAMPLES(util_logbase2(iview
->image
->samples
));
1773 if (iview
->image
->surface
.flags
& RADEON_SURF_SBUFFER
)
1774 ds
->db_stencil_info
= S_028044_FORMAT(V_028044_STENCIL_8
);
1776 ds
->db_stencil_info
= S_028044_FORMAT(V_028044_STENCIL_INVALID
);
1778 if (device
->physical_device
->rad_info
.chip_class
>= CIK
) {
1779 struct radeon_info
*info
= &device
->physical_device
->rad_info
;
1780 unsigned tiling_index
= iview
->image
->surface
.tiling_index
[level
];
1781 unsigned stencil_index
= iview
->image
->surface
.stencil_tiling_index
[level
];
1782 unsigned macro_index
= iview
->image
->surface
.macro_tile_index
;
1783 unsigned tile_mode
= info
->si_tile_mode_array
[tiling_index
];
1784 unsigned stencil_tile_mode
= info
->si_tile_mode_array
[stencil_index
];
1785 unsigned macro_mode
= info
->cik_macrotile_mode_array
[macro_index
];
1787 ds
->db_depth_info
|=
1788 S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode
)) |
1789 S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode
)) |
1790 S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode
)) |
1791 S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode
)) |
1792 S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode
)) |
1793 S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode
));
1794 ds
->db_z_info
|= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode
));
1795 ds
->db_stencil_info
|= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode
));
1797 unsigned tile_mode_index
= si_tile_mode_index(iview
->image
, level
, false);
1798 ds
->db_z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
1799 tile_mode_index
= si_tile_mode_index(iview
->image
, level
, true);
1800 ds
->db_stencil_info
|= S_028044_TILE_MODE_INDEX(tile_mode_index
);
1803 if (iview
->image
->htile
.size
&& !level
) {
1804 ds
->db_z_info
|= S_028040_TILE_SURFACE_ENABLE(1) |
1805 S_028040_ALLOW_EXPCLEAR(1);
1807 if (iview
->image
->surface
.flags
& RADEON_SURF_SBUFFER
) {
1808 /* Workaround: For a not yet understood reason, the
1809 * combination of MSAA, fast stencil clear and stencil
1810 * decompress messes with subsequent stencil buffer
1811 * uses. Problem was reproduced on Verde, Bonaire,
1812 * Tonga, and Carrizo.
1814 * Disabling EXPCLEAR works around the problem.
1816 * Check piglit's arb_texture_multisample-stencil-clear
1817 * test if you want to try changing this.
1819 if (iview
->image
->samples
<= 1)
1820 ds
->db_stencil_info
|= S_028044_ALLOW_EXPCLEAR(1);
1822 /* Use all of the htile_buffer for depth if there's no stencil. */
1823 ds
->db_stencil_info
|= S_028044_TILE_STENCIL_DISABLE(1);
1825 va
= device
->ws
->buffer_get_va(iview
->bo
) + iview
->image
->offset
+
1826 iview
->image
->htile
.offset
;
1827 ds
->db_htile_data_base
= va
>> 8;
1828 ds
->db_htile_surface
= S_028ABC_FULL_CACHE(1);
1830 ds
->db_htile_data_base
= 0;
1831 ds
->db_htile_surface
= 0;
1834 ds
->db_z_read_base
= ds
->db_z_write_base
= z_offs
>> 8;
1835 ds
->db_stencil_read_base
= ds
->db_stencil_write_base
= s_offs
>> 8;
1837 ds
->db_depth_size
= S_028058_PITCH_TILE_MAX((level_info
->nblk_x
/ 8) - 1) |
1838 S_028058_HEIGHT_TILE_MAX((level_info
->nblk_y
/ 8) - 1);
1839 ds
->db_depth_slice
= S_02805C_SLICE_TILE_MAX((level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1);
1842 VkResult
radv_CreateFramebuffer(
1844 const VkFramebufferCreateInfo
* pCreateInfo
,
1845 const VkAllocationCallbacks
* pAllocator
,
1846 VkFramebuffer
* pFramebuffer
)
1848 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1849 struct radv_framebuffer
*framebuffer
;
1851 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
1853 size_t size
= sizeof(*framebuffer
) +
1854 sizeof(struct radv_attachment_info
) * pCreateInfo
->attachmentCount
;
1855 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
1856 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1857 if (framebuffer
== NULL
)
1858 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1860 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
1861 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
1862 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
1863 struct radv_image_view
*iview
= radv_image_view_from_handle(_iview
);
1864 framebuffer
->attachments
[i
].attachment
= iview
;
1865 if (iview
->aspect_mask
& VK_IMAGE_ASPECT_COLOR_BIT
) {
1866 radv_initialise_color_surface(device
, &framebuffer
->attachments
[i
].cb
, iview
);
1867 } else if (iview
->aspect_mask
& (VK_IMAGE_ASPECT_DEPTH_BIT
| VK_IMAGE_ASPECT_STENCIL_BIT
)) {
1868 radv_initialise_ds_surface(device
, &framebuffer
->attachments
[i
].ds
, iview
);
1872 framebuffer
->width
= pCreateInfo
->width
;
1873 framebuffer
->height
= pCreateInfo
->height
;
1874 framebuffer
->layers
= pCreateInfo
->layers
;
1876 *pFramebuffer
= radv_framebuffer_to_handle(framebuffer
);
1880 void radv_DestroyFramebuffer(
1883 const VkAllocationCallbacks
* pAllocator
)
1885 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1886 RADV_FROM_HANDLE(radv_framebuffer
, fb
, _fb
);
1890 vk_free2(&device
->alloc
, pAllocator
, fb
);
1893 static unsigned radv_tex_wrap(VkSamplerAddressMode address_mode
)
1895 switch (address_mode
) {
1896 case VK_SAMPLER_ADDRESS_MODE_REPEAT
:
1897 return V_008F30_SQ_TEX_WRAP
;
1898 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT
:
1899 return V_008F30_SQ_TEX_MIRROR
;
1900 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE
:
1901 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL
;
1902 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER
:
1903 return V_008F30_SQ_TEX_CLAMP_BORDER
;
1904 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE
:
1905 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
1907 unreachable("illegal tex wrap mode");
1913 radv_tex_compare(VkCompareOp op
)
1916 case VK_COMPARE_OP_NEVER
:
1917 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER
;
1918 case VK_COMPARE_OP_LESS
:
1919 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS
;
1920 case VK_COMPARE_OP_EQUAL
:
1921 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL
;
1922 case VK_COMPARE_OP_LESS_OR_EQUAL
:
1923 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
1924 case VK_COMPARE_OP_GREATER
:
1925 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER
;
1926 case VK_COMPARE_OP_NOT_EQUAL
:
1927 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
1928 case VK_COMPARE_OP_GREATER_OR_EQUAL
:
1929 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
1930 case VK_COMPARE_OP_ALWAYS
:
1931 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
1933 unreachable("illegal compare mode");
1939 radv_tex_filter(VkFilter filter
, unsigned max_ansio
)
1942 case VK_FILTER_NEAREST
:
1943 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_POINT
:
1944 V_008F38_SQ_TEX_XY_FILTER_POINT
);
1945 case VK_FILTER_LINEAR
:
1946 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_BILINEAR
:
1947 V_008F38_SQ_TEX_XY_FILTER_BILINEAR
);
1948 case VK_FILTER_CUBIC_IMG
:
1950 fprintf(stderr
, "illegal texture filter");
1956 radv_tex_mipfilter(VkSamplerMipmapMode mode
)
1959 case VK_SAMPLER_MIPMAP_MODE_NEAREST
:
1960 return V_008F38_SQ_TEX_Z_FILTER_POINT
;
1961 case VK_SAMPLER_MIPMAP_MODE_LINEAR
:
1962 return V_008F38_SQ_TEX_Z_FILTER_LINEAR
;
1964 return V_008F38_SQ_TEX_Z_FILTER_NONE
;
1969 radv_tex_bordercolor(VkBorderColor bcolor
)
1972 case VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
:
1973 case VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
:
1974 return V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
1975 case VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
:
1976 case VK_BORDER_COLOR_INT_OPAQUE_BLACK
:
1977 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK
;
1978 case VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
:
1979 case VK_BORDER_COLOR_INT_OPAQUE_WHITE
:
1980 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE
;
1988 radv_tex_aniso_filter(unsigned filter
)
2002 radv_init_sampler(struct radv_device
*device
,
2003 struct radv_sampler
*sampler
,
2004 const VkSamplerCreateInfo
*pCreateInfo
)
2006 uint32_t max_aniso
= pCreateInfo
->anisotropyEnable
&& pCreateInfo
->maxAnisotropy
> 1.0 ?
2007 (uint32_t) pCreateInfo
->maxAnisotropy
: 0;
2008 uint32_t max_aniso_ratio
= radv_tex_aniso_filter(max_aniso
);
2009 bool is_vi
= (device
->physical_device
->rad_info
.chip_class
>= VI
);
2011 sampler
->state
[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo
->addressModeU
)) |
2012 S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo
->addressModeV
)) |
2013 S_008F30_CLAMP_Z(radv_tex_wrap(pCreateInfo
->addressModeW
)) |
2014 S_008F30_MAX_ANISO_RATIO(max_aniso_ratio
) |
2015 S_008F30_DEPTH_COMPARE_FUNC(radv_tex_compare(pCreateInfo
->compareOp
)) |
2016 S_008F30_FORCE_UNNORMALIZED(pCreateInfo
->unnormalizedCoordinates
? 1 : 0) |
2017 S_008F30_ANISO_THRESHOLD(max_aniso_ratio
>> 1) |
2018 S_008F30_ANISO_BIAS(max_aniso_ratio
) |
2019 S_008F30_DISABLE_CUBE_WRAP(0) |
2020 S_008F30_COMPAT_MODE(is_vi
));
2021 sampler
->state
[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(pCreateInfo
->minLod
, 0, 15), 8)) |
2022 S_008F34_MAX_LOD(S_FIXED(CLAMP(pCreateInfo
->maxLod
, 0, 15), 8)) |
2023 S_008F34_PERF_MIP(max_aniso_ratio
? max_aniso_ratio
+ 6 : 0));
2024 sampler
->state
[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(pCreateInfo
->mipLodBias
, -16, 16), 8)) |
2025 S_008F38_XY_MAG_FILTER(radv_tex_filter(pCreateInfo
->magFilter
, max_aniso
)) |
2026 S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo
->minFilter
, max_aniso
)) |
2027 S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo
->mipmapMode
)) |
2028 S_008F38_MIP_POINT_PRECLAMP(1) |
2029 S_008F38_DISABLE_LSB_CEIL(1) |
2030 S_008F38_FILTER_PREC_FIX(1) |
2031 S_008F38_ANISO_OVERRIDE(is_vi
));
2032 sampler
->state
[3] = (S_008F3C_BORDER_COLOR_PTR(0) |
2033 S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(pCreateInfo
->borderColor
)));
2036 VkResult
radv_CreateSampler(
2038 const VkSamplerCreateInfo
* pCreateInfo
,
2039 const VkAllocationCallbacks
* pAllocator
,
2040 VkSampler
* pSampler
)
2042 RADV_FROM_HANDLE(radv_device
, device
, _device
);
2043 struct radv_sampler
*sampler
;
2045 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
2047 sampler
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
2048 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2050 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2052 radv_init_sampler(device
, sampler
, pCreateInfo
);
2053 *pSampler
= radv_sampler_to_handle(sampler
);
2058 void radv_DestroySampler(
2061 const VkAllocationCallbacks
* pAllocator
)
2063 RADV_FROM_HANDLE(radv_device
, device
, _device
);
2064 RADV_FROM_HANDLE(radv_sampler
, sampler
, _sampler
);
2068 vk_free2(&device
->alloc
, pAllocator
, sampler
);
2072 /* vk_icd.h does not declare this function, so we declare it here to
2073 * suppress Wmissing-prototypes.
2075 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2076 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
2078 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
2079 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
2081 /* For the full details on loader interface versioning, see
2082 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
2083 * What follows is a condensed summary, to help you navigate the large and
2084 * confusing official doc.
2086 * - Loader interface v0 is incompatible with later versions. We don't
2089 * - In loader interface v1:
2090 * - The first ICD entrypoint called by the loader is
2091 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
2093 * - The ICD must statically expose no other Vulkan symbol unless it is
2094 * linked with -Bsymbolic.
2095 * - Each dispatchable Vulkan handle created by the ICD must be
2096 * a pointer to a struct whose first member is VK_LOADER_DATA. The
2097 * ICD must initialize VK_LOADER_DATA.loadMagic to ICD_LOADER_MAGIC.
2098 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
2099 * vkDestroySurfaceKHR(). The ICD must be capable of working with
2100 * such loader-managed surfaces.
2102 * - Loader interface v2 differs from v1 in:
2103 * - The first ICD entrypoint called by the loader is
2104 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
2105 * statically expose this entrypoint.
2107 * - Loader interface v3 differs from v2 in:
2108 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
2109 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
2110 * because the loader no longer does so.
2112 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);