Skip to content

vulkan: fix storageBuffer16BitAccess detection on some adreno driver #8837

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

vulkan: fix storageBuffer16BitAccess detection on some adreno driver #8837

wants to merge 1 commit into from

Conversation

rhjdvsgsgks
Copy link
Contributor

on some adreno driver lacking VK_KHR_16bit_storage. storage_16bit.storageBuffer16BitAccess would return false but vk11_features.storageBuffer16BitAccess would still true. add a check for both of them.

i also removed vk11_features in ggml_vk_print_gpu_info. since it is unused

@github-actions github-actions bot added Vulkan Issues specific to the Vulkan backend ggml changes relating to the ggml tensor library for machine learning labels Aug 2, 2024
@0cc4m 0cc4m self-assigned this Aug 3, 2024
@0cc4m
Copy link
Collaborator

0cc4m commented Aug 5, 2024

Can you upload and link the vulkaninfo output of such a device?

@mofosyne mofosyne added Review Complexity : Low Trivial changes to code that most beginner devs (or those who want a break) can tackle. e.g. UI fix bugfix fixes an issue or bug labels Aug 5, 2024
@rhjdvsgsgks
Copy link
Contributor Author 

==========
VULKANINFO
==========

Vulkan Instance Version: 1.1.290


Instance Extensions: count = 10
===============================
	VK_EXT_debug_report                    : extension revision 9
	VK_EXT_swapchain_colorspace            : extension revision 4
	VK_KHR_android_surface                 : extension revision 6
	VK_KHR_device_group_creation           : extension revision 1
	VK_KHR_external_fence_capabilities     : extension revision 1
	VK_KHR_external_memory_capabilities    : extension revision 1
	VK_KHR_external_semaphore_capabilities : extension revision 1
	VK_KHR_get_physical_device_properties2 : extension revision 1
	VK_KHR_get_surface_capabilities2       : extension revision 1
	VK_KHR_surface                         : extension revision 25

Layers:
=======
Device Properties and Extensions:
=================================
GPU0:
VkPhysicalDeviceProperties:
---------------------------
	apiVersion        = 1.1.87 (4198487)
	driverVersion     = 0.415.0 (2149183488)
	vendorID          = 0x5143
	deviceID          = 0x5040001
	deviceType        = PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
	deviceName        = Adreno (TM) 540
	pipelineCacheUUID = 92b92a0f-4351-0000-0000-010004050000

VkPhysicalDeviceLimits:
-----------------------
	maxImageDimension1D                             = 16384
	maxImageDimension2D                             = 16384
	maxImageDimension3D                             = 2048
	maxImageDimensionCube                           = 16384
	maxImageArrayLayers                             = 2048
	maxTexelBufferElements                          = 65536
	maxUniformBufferRange                           = 65536
	maxStorageBufferRange                           = 2147483647
	maxPushConstantsSize                            = 128
	maxMemoryAllocationCount                        = 4096
	maxSamplerAllocationCount                       = 4000
	bufferImageGranularity                          = 0x00000001
	sparseAddressSpaceSize                          = 0x00000000
	maxBoundDescriptorSets                          = 4
	maxPerStageDescriptorSamplers                   = 16
	maxPerStageDescriptorUniformBuffers             = 14
	maxPerStageDescriptorStorageBuffers             = 24
	maxPerStageDescriptorSampledImages              = 128
	maxPerStageDescriptorStorageImages              = 4
	maxPerStageDescriptorInputAttachments           = 8
	maxPerStageResources                            = 158
	maxDescriptorSetSamplers                        = 96
	maxDescriptorSetUniformBuffers                  = 84
	maxDescriptorSetUniformBuffersDynamic           = 8
	maxDescriptorSetStorageBuffers                  = 24
	maxDescriptorSetStorageBuffersDynamic           = 4
	maxDescriptorSetSampledImages                   = 768
	maxDescriptorSetStorageImages                   = 24
	maxDescriptorSetInputAttachments                = 8
	maxVertexInputAttributes                        = 32
	maxVertexInputBindings                          = 32
	maxVertexInputAttributeOffset                   = 4096
	maxVertexInputBindingStride                     = 2048
	maxVertexOutputComponents                       = 128
	maxTessellationGenerationLevel                  = 0
	maxTessellationPatchSize                        = 0
	maxTessellationControlPerVertexInputComponents  = 0
	maxTessellationControlPerVertexOutputComponents = 0
	maxTessellationControlPerPatchOutputComponents  = 0
	maxTessellationControlTotalOutputComponents     = 0
	maxTessellationEvaluationInputComponents        = 0
	maxTessellationEvaluationOutputComponents       = 0
	maxGeometryShaderInvocations                    = 0
	maxGeometryInputComponents                      = 0
	maxGeometryOutputComponents                     = 0
	maxGeometryOutputVertices                       = 0
	maxGeometryTotalOutputComponents                = 0
	maxFragmentInputComponents                      = 112
	maxFragmentOutputAttachments                    = 8
	maxFragmentDualSrcAttachments                   = 1
	maxFragmentCombinedOutputResources              = 72
	maxComputeSharedMemorySize                      = 32768
	maxComputeWorkGroupCount: count = 3
		65535
		65535
		65535
	maxComputeWorkGroupInvocations                  = 512
	maxComputeWorkGroupSize: count = 3
		1024
		1024
		64
	subPixelPrecisionBits                           = 8
	subTexelPrecisionBits                           = 8
	mipmapPrecisionBits                             = 8
	maxDrawIndexedIndexValue                        = 4294967295
	maxDrawIndirectCount                            = 4294967295
	maxSamplerLodBias                               = 15.9961
	maxSamplerAnisotropy                            = 16
	maxViewports                                    = 1
	maxViewportDimensions: count = 2
		16384
		16384
	viewportBoundsRange: count = 2
		-32768
		32767
	viewportSubPixelBits                            = 8
	minMemoryMapAlignment                           = 64
	minTexelBufferOffsetAlignment                   = 0x00000040
	minUniformBufferOffsetAlignment                 = 0x00000040
	minStorageBufferOffsetAlignment                 = 0x00000040
	minTexelOffset                                  = -8
	maxTexelOffset                                  = 7
	minTexelGatherOffset                            = -32
	maxTexelGatherOffset                            = 31
	minInterpolationOffset                          = -0.5
	maxInterpolationOffset                          = 0.4375
	subPixelInterpolationOffsetBits                 = 4
	maxFramebufferWidth                             = 16384
	maxFramebufferHeight                            = 16384
	maxFramebufferLayers                            = 2048
	framebufferColorSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	framebufferDepthSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	framebufferStencilSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	framebufferNoAttachmentsSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	maxColorAttachments                             = 8
	sampledImageColorSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	sampledImageIntegerSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	sampledImageDepthSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	sampledImageStencilSampleCounts: count = 3
		SAMPLE_COUNT_1_BIT
		SAMPLE_COUNT_2_BIT
		SAMPLE_COUNT_4_BIT
	storageImageSampleCounts: count = 1
		SAMPLE_COUNT_1_BIT
	maxSampleMaskWords                              = 1
	timestampComputeAndGraphics                     = true
	timestampPeriod                                 = 52.0833
	maxClipDistances                                = 8
	maxCullDistances                                = 8
	maxCombinedClipAndCullDistances                 = 8
	discreteQueuePriorities                         = 3
	pointSizeRange: count = 2
		1
		1023
	lineWidthRange: count = 2
		1
		1
	pointSizeGranularity                            = 0.0625
	lineWidthGranularity                            = 0
	strictLines                                     = false
	standardSampleLocations                         = true
	optimalBufferCopyOffsetAlignment                = 0x00000040
	optimalBufferCopyRowPitchAlignment              = 0x00000040
	nonCoherentAtomSize                             = 0x00000001

VkPhysicalDeviceSparseProperties:
---------------------------------
	residencyStandard2DBlockShape            = false
	residencyStandard2DMultisampleBlockShape = false
	residencyStandard3DBlockShape            = false
	residencyAlignedMipSize                  = false
	residencyNonResidentStrict               = false

VkPhysicalDeviceDriverPropertiesKHR:
------------------------------------
	driverID        = DRIVER_ID_QUALCOMM_PROPRIETARY
	driverName      = Qualcomm Technologies Inc. Adreno Vulkan Driver
	driverInfo      = Driver Build: f2ab992, I401605978b, 1569657027
Date: 09/28/19
Compiler Version: EV031.27.05.01
Driver Branch: 

	conformanceVersion:
		major    = 1
		minor    = 1
		subminor = 2
		patch    = 1

VkPhysicalDeviceFragmentDensityMapPropertiesEXT:
------------------------------------------------
	minFragmentDensityTexelSize:
		width  = 32
		height = 32
	maxFragmentDensityTexelSize:
		width  = 256
		height = 256
	fragmentDensityInvocations = false

VkPhysicalDevicePushDescriptorPropertiesKHR:
--------------------------------------------
	maxPushDescriptors = 32

VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT:
-------------------------------------------------
	filterMinmaxSingleComponentFormats = true
	filterMinmaxImageComponentMapping  = true

Device Extensions: count = 34
	VK_ANDROID_external_memory_android_hardware_buffer : extension revision 2
	VK_EXT_fragment_density_map                        : extension revision 1
	VK_EXT_global_priority                             : extension revision 2
	VK_EXT_hdr_metadata                                : extension revision 2
	VK_EXT_queue_family_foreign                        : extension revision 1
	VK_EXT_sampler_filter_minmax                       : extension revision 1
	VK_GOOGLE_display_timing                           : extension revision 1
	VK_KHR_bind_memory2                                : extension revision 1
	VK_KHR_create_renderpass2                          : extension revision 1
	VK_KHR_dedicated_allocation                        : extension revision 1
	VK_KHR_descriptor_update_template                  : extension revision 1
	VK_KHR_device_group                                : extension revision 2
	VK_KHR_driver_properties                           : extension revision 1
	VK_KHR_external_fence                              : extension revision 1
	VK_KHR_external_fence_fd                           : extension revision 1
	VK_KHR_external_memory                             : extension revision 1
	VK_KHR_external_memory_fd                          : extension revision 1
	VK_KHR_external_semaphore                          : extension revision 1
	VK_KHR_external_semaphore_fd                       : extension revision 1
	VK_KHR_get_memory_requirements2                    : extension revision 1
	VK_KHR_incremental_present                         : extension revision 1
	VK_KHR_maintenance1                                : extension revision 1
	VK_KHR_maintenance2                                : extension revision 1
	VK_KHR_maintenance3                                : extension revision 1
	VK_KHR_multiview                                   : extension revision 1
	VK_KHR_push_descriptor                             : extension revision 1
	VK_KHR_relaxed_block_layout                        : extension revision 1
	VK_KHR_sampler_mirror_clamp_to_edge                : extension revision 1
	VK_KHR_sampler_ycbcr_conversion                    : extension revision 1
	VK_KHR_shader_draw_parameters                      : extension revision 1
	VK_KHR_shared_presentable_image                    : extension revision 1
	VK_KHR_storage_buffer_storage_class                : extension revision 1
	VK_KHR_swapchain                                   : extension revision 70
	VK_KHR_variable_pointers                           : extension revision 1

VkQueueFamilyProperties:
========================
	queueProperties[0]:
	-------------------
		minImageTransferGranularity = (1,1,1)
		queueCount                  = 3
		queueFlags                  = QUEUE_GRAPHICS_BIT | QUEUE_COMPUTE_BIT | QUEUE_PROTECTED_BIT
		timestampValidBits          = 48
		present support             = false

VkPhysicalDeviceMemoryProperties:
=================================
memoryHeaps: count = 2
	memoryHeaps[0]:
		size   = 3838820352 (0xe4cfc000) (3.58 GiB)
		flags: count = 1
			MEMORY_HEAP_DEVICE_LOCAL_BIT
	memoryHeaps[1]:
		size   = 268435456 (0x10000000) (256.00 MiB)
		flags: count = 1
			MEMORY_HEAP_DEVICE_LOCAL_BIT
memoryTypes: count = 6
	memoryTypes[0]:
		heapIndex     = 0
		propertyFlags = 0x0001: count = 1
			MEMORY_PROPERTY_DEVICE_LOCAL_BIT
		usable for:
			IMAGE_TILING_OPTIMAL:
				color images
				FORMAT_D16_UNORM
				FORMAT_X8_D24_UNORM_PACK32
				FORMAT_D32_SFLOAT
				FORMAT_D24_UNORM_S8_UINT
				(non-sparse)
			IMAGE_TILING_LINEAR:
				color images
				(non-sparse)
	memoryTypes[1]:
		heapIndex     = 0
		propertyFlags = 0x000b: count = 3
			MEMORY_PROPERTY_DEVICE_LOCAL_BIT
			MEMORY_PROPERTY_HOST_VISIBLE_BIT
			MEMORY_PROPERTY_HOST_CACHED_BIT
		usable for:
			IMAGE_TILING_OPTIMAL:
				None
			IMAGE_TILING_LINEAR:
				None
	memoryTypes[2]:
		heapIndex     = 0
		propertyFlags = 0x000f: count = 4
			MEMORY_PROPERTY_DEVICE_LOCAL_BIT
			MEMORY_PROPERTY_HOST_VISIBLE_BIT
			MEMORY_PROPERTY_HOST_COHERENT_BIT
			MEMORY_PROPERTY_HOST_CACHED_BIT
		usable for:
			IMAGE_TILING_OPTIMAL:
				None
			IMAGE_TILING_LINEAR:
				color images
				(non-sparse)
	memoryTypes[3]:
		heapIndex     = 0
		propertyFlags = 0x0001: count = 1
			MEMORY_PROPERTY_DEVICE_LOCAL_BIT
		usable for:
			IMAGE_TILING_OPTIMAL:
				None
			IMAGE_TILING_LINEAR:
				None
	memoryTypes[4]:
		heapIndex     = 0
		propertyFlags = 0x0007: count = 3
			MEMORY_PROPERTY_DEVICE_LOCAL_BIT
			MEMORY_PROPERTY_HOST_VISIBLE_BIT
			MEMORY_PROPERTY_HOST_COHERENT_BIT
		usable for:
			IMAGE_TILING_OPTIMAL:
				None
			IMAGE_TILING_LINEAR:
				None
	memoryTypes[5]:
		heapIndex     = 1
		propertyFlags = 0x0021: count = 2
			MEMORY_PROPERTY_DEVICE_LOCAL_BIT
			MEMORY_PROPERTY_PROTECTED_BIT
		usable for:
			IMAGE_TILING_OPTIMAL:
				None
			IMAGE_TILING_LINEAR:
				None

VkPhysicalDeviceFeatures:
=========================
	robustBufferAccess                      = true
	fullDrawIndexUint32                     = true
	imageCubeArray                          = true
	independentBlend                        = true
	geometryShader                          = false
	tessellationShader                      = false
	sampleRateShading                       = true
	dualSrcBlend                            = true
	logicOp                                 = false
	multiDrawIndirect                       = true
	drawIndirectFirstInstance               = false
	depthClamp                              = true
	depthBiasClamp                          = true
	fillModeNonSolid                        = true
	depthBounds                             = false
	wideLines                               = false
	largePoints                             = true
	alphaToOne                              = true
	multiViewport                           = false
	samplerAnisotropy                       = true
	textureCompressionETC2                  = true
	textureCompressionASTC_LDR              = true
	textureCompressionBC                    = false
	occlusionQueryPrecise                   = false
	pipelineStatisticsQuery                 = false
	vertexPipelineStoresAndAtomics          = true
	fragmentStoresAndAtomics                = true
	shaderTessellationAndGeometryPointSize  = false
	shaderImageGatherExtended               = true
	shaderStorageImageExtendedFormats       = false
	shaderStorageImageMultisample           = false
	shaderStorageImageReadWithoutFormat     = false
	shaderStorageImageWriteWithoutFormat    = true
	shaderUniformBufferArrayDynamicIndexing = true
	shaderSampledImageArrayDynamicIndexing  = true
	shaderStorageBufferArrayDynamicIndexing = true
	shaderStorageImageArrayDynamicIndexing  = true
	shaderClipDistance                      = true
	shaderCullDistance                      = true
	shaderFloat64                           = false
	shaderInt64                             = false
	shaderInt16                             = true
	shaderResourceResidency                 = false
	shaderResourceMinLod                    = false
	sparseBinding                           = false
	sparseResidencyBuffer                   = false
	sparseResidencyImage2D                  = false
	sparseResidencyImage3D                  = false
	sparseResidency2Samples                 = false
	sparseResidency4Samples                 = false
	sparseResidency8Samples                 = false
	sparseResidency16Samples                = false
	sparseResidencyAliased                  = false
	variableMultisampleRate                 = false
	inheritedQueries                        = true

VkPhysicalDeviceFragmentDensityMapFeaturesEXT:
----------------------------------------------
	fragmentDensityMap                    = true
	fragmentDensityMapDynamic             = false
	fragmentDensityMapNonSubsampledImages = true

@0cc4m
Copy link
Collaborator

0cc4m commented Aug 6, 2024

You can upload it as a file or snippet to Github and link it here. The text you posted doesn't contain the relevant parts.

@rhjdvsgsgks
Copy link
Contributor Author

You can upload it as a file or snippet to Github and link it here. The text you posted doesn't contain the relevant parts.

i double checked that it is the full output

@0cc4m
Copy link
Collaborator

0cc4m commented Aug 6, 2024

You can upload it as a file or snippet to Github and link it here. The text you posted doesn't contain the relevant parts.

i double checked that it is the full output

Alright, I am confused cause it shows neither VkPhysicalDeviceVulkan11Features nor VkPhysicalDevice16BitStorageFeatures.

0cc4m
0cc4m reviewed Aug 6, 2024
View reviewed changes
ggml/src/ggml-vulkan.cpp

vkGetPhysicalDeviceFeatures2(device->physical_device, &device_features2);

device->fp16 = device->fp16 && vk12_features.shaderFloat16;

if (!vk11_features.storageBuffer16BitAccess) {
if (!(vk11_features.storageBuffer16BitAccess && storage_16bit.storageBuffer16BitAccess)) {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
if (!(vk11_features.storageBuffer16BitAccess && storage_16bit.storageBuffer16BitAccess)) {
if (!vk11_features.storageBuffer16BitAccess || !storage_16bit.storageBuffer16BitAccess) {

Just a nitpick, but I think it's more readable this way.

ggml/src/ggml-vulkan.cpp
VkPhysicalDevice16BitStorageFeatures storage_16bit;
storage_16bit.pNext = nullptr;
storage_16bit.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES;
vk11_features.pNext = &storage_16bit;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Doing this is against the Vulkan Specification and causes a validation error when run with Validation layers enabled:

vkCreateDevice():  If the pNext chain includes a VkPhysicalDeviceVulkan11Features structure, then it must not include a VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES structure.

I'm not sure how to handle this properly, we could avoid VkPhysicalDeviceVulkan11Features entirely, but I don't want to do that just cause of Qualcomm driver issues. Do you have an idea?

@FranzKafkaYu
Copy link

@rhjdvsgsgks Hi Sir,sorry to bother you.May I ask you which tag or commit do you use for testing in Adreno GPU devices?I filed a issue here #8965 ,and wanna check whether it's the code base too old for this issue.Can you use Vulkan-GPU for acceleration in Android device?if so can you give me some instructions?thanks a lot.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@0cc4m 0cc4m 0cc4m left review comments

At least 1 approving review is required to merge this pull request.

Assignees

@0cc4m 0cc4m

Labels
bugfix fixes an issue or bug ggml changes relating to the ggml tensor library for machine learning Review Complexity : Low Trivial changes to code that most beginner devs (or those who want a break) can tackle. e.g. UI fix Vulkan Issues specific to the Vulkan backend
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@rhjdvsgsgks @0cc4m @FranzKafkaYu @mofosyne