Skip to content

Latest commit

 

History

History
162 lines (110 loc) · 7.4 KB

atomics.adoc

File metadata and controls

162 lines (110 loc) · 7.4 KB
Raw

Atomics

The purpose of this chapter is to help users understand the various features Vulkan exposes for atomic operations.

Variations of Atomics

To better understand the different extensions, it is first important to be aware of the various types of atomics exposed.

  • Type

    • float

    • int

  • Width

    • 16 bit

    • 32 bit

    • 64 bit

  • Operations

    • loads

    • stores

    • exchange

    • add

    • min

    • max

    • etc.

  • Storage Class

    • StorageBuffer or Uniform (buffer)

    • Workgroup (shared memory)

    • Image (image or sparse image)

Baseline Support

With Vulkan 1.0 and no extensions, an application is allowed to use 32-bit int type for atomics. This can be used for all supported SPIR-V operations (load, store, exchange, etc). SPIR-V contains some atomic operations that are guarded with the Kernel capability and are not currently allowed in Vulkan.

Atomic Counters

While both GLSL and SPIR-V support the use of atomic counters, Vulkan does not expose the AtomicStorage SPIR-V capability needed to use the AtomicCounter storage class. It was decided that an app can just use OpAtomicIAdd and OpAtomicISub with the value 1 to achieve the same results.

Expanding Atomic support

The current extensions that expose additional support for atomics are:

Each explained in more details below.

VK_KHR_shader_atomic_int64

Note

Promoted to core in Vulkan 1.2

This extension allows for 64-bit int atomic operations for buffers and shared memory. If the Int64Atomics SPIR-V capability is declared, all supported SPIR-V operations can be used with 64-bit int.

The two feature bits, shaderBufferInt64Atomics and shaderSharedInt64Atomics, are used to query what storage classes are supported for 64-bit int atomics.

  • shaderBufferInt64Atomics - buffers

  • shaderSharedInt64Atomics - shared memory

The shaderBufferInt64Atomics is always guaranteed to be supported if using Vulkan 1.2+ or the extension is exposed.

VK_EXT_shader_image_atomic_int64

This extension allows for 64-bit int atomic operations for images and sparse images. If the Int64Atomics and Int64ImageEXT SPIR-V capability is declared, all supported SPIR-V operations can be used with 64-bit int on images.

Image vs Sparse Image support

This extension exposes both a shaderImageInt64Atomics and sparseImageInt64Atomics feature bit. The sparseImage* feature is an additional feature bit and is only allowed to be used if the shaderImage* bit is enabled as well. Some hardware has a hard time doing atomics on images with sparse resources, therefor the atomic feature is split up to allow sparse images as an additional feature an implementation can expose.

VK_EXT_shader_atomic_float

This extension allows for float atomic operations for buffers, shared memory, images, and sparse images. Only a subset of operations is supported for float types with this extension.

The extension lists many feature bits. One way to group them is by *Float*Atomics and *Float*AtomicAdd:

  • The *Float*Atomics features allow for the use of OpAtomicStore, OpAtomicLoad, and OpAtomicExchange for float types.

    • Note the OpAtomicCompareExchange “exchange” operation is not included as the SPIR-V spec only allows int types for it.

  • The *Float*AtomicAdd features allow the use of the two extended SPIR-V operations AtomicFloat32AddEXT and AtomicFloat64AddEXT.

From here the rest of the permutations of features fall into the grouping of 32-bit float support:

  • shaderBufferFloat32* - buffers

  • shaderSharedFloat32* - shared memory

  • shaderImageFloat32* - images

  • sparseImageFloat32* - sparse images

and 64-bit float support:

  • shaderBufferFloat64* - buffers

  • shaderSharedFloat64* - shared memory

Note

OpenGLES OES_shader_image_atomic allowed the use of atomics on r32f for imageAtomicExchange. For porting, an application will want to check for shaderImageFloat32Atomics support to be able to do the same in Vulkan.

VK_EXT_shader_atomic_float2

This extension adds 2 additional sets of features missing in VK_EXT_shader_atomic_float

First, it adds 16-bit floats for both buffers and shared memory in the same fashion as found above for VK_EXT_shader_atomic_float.

  • shaderBufferFloat16* - buffers

  • shaderSharedFloat16* - shared memory

Second, it adds float support for min and max atomic operations (OpAtomicFMinEXT and OpAtomicFMaxEXT)

For 16-bit float support (with AtomicFloat16MinMaxEXT capability):

  • shaderBufferFloat16AtomicMinMax - buffers

  • shaderSharedFloat16AtomicMinMax - shared memory

For 32-bit float support (with AtomicFloat32MinMaxEXT capability):

  • shaderBufferFloat32AtomicMinMax - buffers

  • shaderSharedFloat32AtomicMinMax - shared memory

  • shaderImageFloat32AtomicMinMax - images

  • sparseImageFloat32AtomicMinMax - sparse images

For 64-bit float support (with AtomicFloat64MinMaxEXT capability):

  • shaderBufferFloat64AtomicMinMax - buffers

  • shaderSharedFloat64AtomicMinMax - shared memory