API

SPIR-V aggregate type.

Equality is defined in terms of identity, since different aggregate types have in principle different semantics.

32-bit floating-point representation of π.

Module annotated with instruction ranges for each logical SPIR-V section, suitable for read-only operations and analyses.

Any desired modifications of annotated modules should be staged and applied via a Diff.

Control tree.

The leaves are labeled as REGION_BLOCK regions, with the distinguishing property that they have no children.

Children nodes of any given subtree are in reverse postorder according to the original control-flow graph.

Graph whose vertices and edges remain identical after deletion of other vertices.

x*F
(x)F

Converts an input x to a Float32 when F is right-multiplied with it.

Abstract type expressing SPIR-V which capabilities and extensions are supported.

Intermediate representation of SPIR-V modules.

The types and constants mappings can be updated at any time without explicitly notifying of the mutation behavior (i.e. functions may not end with !), so long as additions only are performed. Such additions are currently done upon construction of GlobalsInfo. The ID counter can also be incremented without notice.

SPIR-V image type.

Type parameters:

• Format: SPIR-V ImageFormat enumerated value.
• Dim: SPIR-V Dim enumerated value.
• Depth: 64-bit integer with value 0 (not a depth image), 1 (depth image) or 2 (unknown).
• Arrayed: Bool indicating whether the image is a layer of an image array.
• MS: Bool indicating whether the image is multisampled.
• Sampled: 64-bit integer with value 0 (unknown), 1 (may be sampled) or 2 (read-write, no sampling).

Parsed SPIR-V instruction. It represents an instruction of the form %result_id = %opcode(%arguments...)::%type_id.

Layout strategy used to compute alignments, offsets and strides.

" Julia layout, with a special handling of mutable fields for composite types.

Mutable fields are stored as 8-byte pointers in Julia, but in SPIR-V there is no such concept of pointer. Therefore, we treat mutable fields as if they were fully inlined - that imposes a few restrictions on the behavior of such mutable objects, but is necessary to keep some form of compatibility with GPU representations. The alternative would be to completely disallow structs which contain mutable fields.

Layout strategy which assumes no padding is required at all. This might be useful for maximally packing data when serialized, reducing size; or to improve performance by avoiding padding when not needed, e.g. if you already pad your structures manually upfront (only do that if you know what you're doing).

SPIR-V instruction in binary format.

SPIR-V module, as a series of headers followed by a stream of instructions. The header embeds two magic numbers, one for the module itself and one for the tool that generated it (e.g. glslang). It also contains the version of the specification applicable to the module, the maximum ID number and an optional instruction schema.

Result ID used in a SPIR-V context, following single static assignment rules for valid modules.

SPIR-V target for compilation through the Julia frontend.

ShaderCompilationCache()

Create a cache that may be used to associate shaders to be compiled (ShaderInfo) with compiled code (ShaderSource).

Although this cache will already significantly speed up shader creation by compiling only what is required, performance-critical applications will additionally want to use a high-level cache that maps a ShaderSource to a driver-dependent object generated by the graphics API chosen for that application. For Vulkan, that would be a Vk.ShaderModule, for example. It is then recommended to use an IdDict to perform the caching, to avoid hashing the whole source code.

Shader-compatible layout strategy, where layout information is strictly read from shader decorations.

SPIR-V shader code, with stage and entry point information.

Extensions and capabilities supported by a client API.

Type metadata meant to be analyzed and modified to generate appropriate decorations.

x*U
(x)U

Converts an input x to an UInt32 when U is right-multiplied with it.

Vulkan-compatible layout strategy.

alignment(layout, T)

Memory alignment which must be respected when allocating a value of the given type.

This refers to an external alignment, i.e. this value is of interest when allocating a value of type T within a struct or array. In particular, this value does not indicate what should be the alignment of any struct member or the stride of any array element contained in T, if T is a struct or an array.

datasize(layout, T)

This is the total space occupied by a value of type T, including padding between struct elements and array strides.

disassemble(io, spir_module)

Transform the content of spir_module into a human-readable format and prints it to io.

Get a SPIR-V type from a Julia type, caching the mapping in the IR if one is provided.

@any_hit [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a any_hit shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing any_hit-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@callable [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a callable shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing callable-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@closest_hit [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a closest_hit shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing closest_hit-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@compute [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a compute shader.

The supported parameters are the following:

• options = ComputeExecutionOptions(): a ComputeExecutionOptions structure providing compute-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@for i in iter begin #= ... do something with i ... =# end @for i in iter1, j in iter2 begin #= ... do something with i and j ... =# end

Define a C-style for loop.

This loop does not rely on the standard iteration protocol in Julia, and will instead mimic how loops work in C: a loop entry point usually defines loop variables, a condition guards the loop body, and a continue block usually increments loop variables on subsequent iterations.

iter must implement pre_loop, cond_loop and post_loop.

@fragment [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a fragment shader.

The supported parameters are the following:

• options = FragmentExecutionOptions(): a FragmentExecutionOptions structure providing fragment-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@geometry [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a geometry shader.

The supported parameters are the following:

• options = GeometryExecutionOptions(): a GeometryExecutionOptions structure providing geometry-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@intersection [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a intersection shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing intersection-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@load address::T
@load address[index]::T

Load a value of type T, either directly (if no index is specified) or at offset = index - 1 elements from address. address should be a device address, i.e. a UInt64 value representing the address of a physical storage buffer.

@mesh [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a mesh shader.

The supported parameters are the following:

• options = MeshExecutionOptions(): a MeshExecutionOptions structure providing mesh-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@miss [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a miss shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing miss-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@ray_generation [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a ray_generation shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing ray_generation-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@store value address::T
@store value address[index]::T
@store address::T = value
@store address[index]::T = value

Store a value of type T at the given address, either directly (if no index is specified) or at an offset of index - 1 elements from address.

If value is not of type T, a conversion will be attempted.

address should be a device address, i.e. a UInt64 value representing the address of a physical storage buffer.

@task [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a task shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing task-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@tessellation_control [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a tessellation_control shader.

The supported parameters are the following:

• options = TessellationExecutionOptions(): a TessellationExecutionOptions structure providing tessellation_control-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@tessellation_evaluation [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a tessellation_evaluation shader.

The supported parameters are the following:

• options = TessellationExecutionOptions(): a TessellationExecutionOptions structure providing tessellation_evaluation-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @vertex, @geometry, @tessellation_control, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

@vertex [parameters] f(::Type1::StorageClass1[{...}], ...)

Compile the provided signature f(args...) into a vertex shader.

The supported parameters are the following:

• options = CommonExecutionOptions(): a CommonExecutionOptions structure providing vertex-specific options.
• features = AllSupported(): a SPIRV.FeatureSupport structure informing the SPIR-V compiler

what capabilities and extensions are allowed. In application code, this should generally be a SupportedFeatures structure coming from the client API. For instance, a package extension for Vulkan.jl exists which provides SupportedFeatures(physical_device, api_version, device_extensions, device_features).

• layout = VulkanLayout(): a VulkanLayout to use to compute alignments, strides and offsets in structures and arrays.
• cache = nothing: an optional ShaderCompilationCache used to cache ShaderSources to prevent repeated compilation of SPIR-V modules.
• interpreter = SPIRVInterpreter(): a SPIRVInterpreter containing method tables and optimization options for the computation. If a cache is provided, this interpreter must be identical to the one used to compile all previous cache entries.
• assemble = false: a boolean flag indicating whether to assemble the resulting module into a ShaderSource, or leave it as a Shader for further introspection. If a cache is provided, assemble must be set to true for caching to be possible.

See also: @geometry, @tessellation_control, @tessellation_evaluation, @fragment, @compute, @ray_generation, @intersection, @closest_hit, @any_hit, @miss, @callable, @task, @mesh

Extended help

Syntax reference

The following syntax allows one to annotate function arguments with shader-specific information:

f(::Type::StorageClass{[BuiltIn], [@Decoration [= value]...]}, ...)

For each argument, the following is to be provided:

• REQUIRED An argument type.
• REQUIRED A SPIR-V storage class (required).
• OPTIONAL A SPIR-V built-in category, see the Vulkan built-in variables.
• OPTIONAL One or more SPIR-V decorations. Arguments to such decorations may be interpolated with $.

This syntax is processed via SPIRV.compile_shader.

Examples

Here are a few examples using the described syntax:

@vertex [parameters] f(::Mutable{Vec4}::Output{Position}, ::UInt32::Input{VertexIndex})

@fragment [parameters] f(::Vec4::Input, ::InvocationData::PushConstant)

@fragment [parameters] f(::Mutable{Vec4}::Output, ::Vec4::Input{FragCoord}, ::Vec2::Input{@Flat})

@compute [parameters] f(::UInt32::Workgroup)

@compute [parameters] f(::Mutable{Arr{256, Float32}}::Workgroup, ::UInt32::Input{LocalInvocationIndex})

Logical representation of a SPIR-V module.

Sequence of blocks u ─→ [v, vs...] ─→ w

Enumerated value representing the type of an instruction. See https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html#instructions3 for a list of SPIR-V opcodes (excluding extended instruction sets).