WGSLLanguageFeatures

Available features

The following WGSL language extensions are defined at WGSL language extensions in the WGSL specification. Bear in mind that the exact set of features available will vary across implementations and physical devices, and may change over time.

packed_4x8_integer_dot_product

Allows DP4a (Dot Product of 4 Elements and Accumulate) GPU instructions to be used via your WGSL code. These efficiently perform 8-bit integer dot products to accelerate computation, saving memory and network bandwidth and improving performance compared with the equivalent f32 versions. They are commonly used in machine learning models in inferencing, within AI frameworks.

Specifically, when packed_4x8_integer_dot_product is available, WGSL code can use:

• 32-bit integer scalars packing 4-component vectors of 8-bit integers to be used as inputs to dot product instructions (via the dot4U8Packed() and dot4I8Packed() built-in functions).
• Packing and unpacking instructions with packed 4-component vectors of 8-bit integers (via built-in functions such as pack4xI8() and pack4xI8Clamp()).

pointer_composite_access

Enables WGSL shader code to access components of complex data types using the same dot (.) syntax whether you're working directly with the data or with a pointer to it.

When pointer_composite_access is available:

• If foo is a pointer: foo.bar is available as a more convenient way to write (*foo).bar. The asterisk (*) would normally be needed to turn the pointer into a "reference" that can be dereferenced, but now both pointers and references are almost interchangeable.
• If foo is not a pointer: The dot (.) operator works exactly as you're used to for directly accessing members.
• if pa is a pointer that stores the starting address of an array, then pa[i] gives you direct access to the memory location where the ith element of that array is stored.

See Syntax sugar for dereferencing composites in WGSL for further details and an example.

readonly_and_readwrite_storage_textures

When available, allows the "read-only" and "read-write" storageTexture.access values to be set when specifying storage texture bind group entry types in a bind group layout. These enable WGSL code to read storage textures, and read/write storage textures, respectively.

subgroup_id

When available, the subgroup_id and num_subgroups WGSL built-in values are usable in compute shaders. These improve the process of scheduling work across subgroups, by indexing memory to avoid overlapping memory accesses. See WGSL subgroup_id extension for more details.

Note: For the subgroup_id WGSL feature to be usable, the subgroups extension needs to be enabled in the GPUDevice (see GPUSupportedFeatures).

subgroup_uniformity

When available, changes the scope of the uniformity analysis for the subgroup and quad built-in functions to the subgroup level (instead of the workgroup level). This feature allows subgroup functionality to be considered uniform in more cases, improving usability for developers and reducing the need to disable uniformity checks altogether. A practical implication is that more values are considered subgroup-uniform, such as the subgroup_id built-in value. See WGSL subgroup_uniformity extension for more details.

uniform_buffer_standard_layout

When available, uniform buffers use the same memory layout constraints as storage buffers, which makes it easier to share data structures across both kinds of buffers. This means uniform buffers are no longer required to have 16-byte alignment on array elements, or to pad nested structure offsets to a multiple of 16 bytes.

See WGSL uniform_buffer_standard_layout extension for more details.

unrestricted_pointer_parameters

Loosens restrictions on pointers being passed to WGSL functions. When available, the following are allowed:

• Parameter pointers to storage, uniform, and workgroup address spaces being passed to user-declared functions.

• Pointers to structure members and array elements being passed to user-declared functions.

  See Pointers As Function Parameters for more details.

Instance properties

The following property is available to all read-only setlike objects:

size

Returns the number of values in the set.

Instance methods

The following methods are available to all read-only setlike objects:

has()

Returns a boolean asserting whether or not an element with the given value is present in the set.

values()

Returns a new iterator object that yields values for each element in the set in insertion order.

keys()

An alias for values().

entries()

Returns a new iterator object that contains an array of [value, value] for each element in the set in insertion order.

forEach()

Calls the provided callback function once for each value present in the set in insertion order.

Examples

Check whether an extension is available

if (
  navigator.gpu.wgslLanguageFeatures.has(
    "readonly_and_readwrite_storage_textures",
  )
) {
  console.log("Read-only and read-write storage textures are available");
}

Return set size and iterate through values

const wgslFeatures = navigator.gpu.wgslLanguageFeatures;

// Return the size of the set
console.log(wgslFeatures.size);

// Iterate through all the set values using values()
const valueIterator = wgslFeatures.values();
for (const value of valueIterator) {
  console.log(value);
}

// …

Specifications

Browser compatibility

See also

• WebGPU API