Enum naga::proc::index::BoundsCheckPolicy

pub enum BoundsCheckPolicy {
    Restrict,
    ReadZeroSkipWrite,
    Unchecked,
}

How should code generated by Naga do bounds checks?

When a vector, matrix, or array index is out of bounds—either negative, or greater than or equal to the number of elements in the type—WGSL requires that some other index of the implementation’s choice that is in bounds is used instead. (There are no types with zero elements.)

Similarly, when out-of-bounds coordinates, array indices, or sample indices are presented to the WGSL textureLoad and textureStore operations, the operation is redirected to do something safe.

Different users of Naga will prefer different defaults:

• When used as part of a WebGPU implementation, the WGSL specification requires the Restrict behavior for array, vector, and matrix accesses, and either the Restrict or ReadZeroSkipWrite behaviors for texture accesses.

• When used by the wgpu crate for native development, wgpu selects ReadZeroSkipWrite as its default.

• Naga’s own default is Unchecked, so that shader translations are as faithful to the original as possible.

Sometimes the underlying hardware and drivers can perform bounds checks themselves, in a way that performs better than the checks Naga would inject. If you’re using native checks like this, then having Naga inject its own checks as well would be redundant, and the Unchecked policy is appropriate.

Variants

Restrict

Replace out-of-bounds indexes with some arbitrary in-bounds index.

(This does not necessarily mean clamping. For example, interpreting the index as unsigned and taking the minimum with the largest valid index would also be a valid implementation. That would map negative indices to the last element, not the first.)

ReadZeroSkipWrite

Out-of-bounds reads return zero, and writes have no effect.

When applied to a chain of accesses, like a[i][j].b[k], all index expressions are evaluated, regardless of whether prior or later index expressions were in bounds. But all the accesses per se are skipped if any index is out of bounds.

Unchecked

Naga adds no checks to indexing operations. Generate the fastest code possible. This is the default for Naga, as a translator, but consumers should consider defaulting to a safer behavior.

Trait Implementations

impl Clone for BoundsCheckPolicy

fn clone(&self) -> BoundsCheckPolicy

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for BoundsCheckPolicy

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for BoundsCheckPolicy

The default BoundsCheckPolicy is Unchecked.

fn default() -> Self

Returns the “default value” for a type.

impl Hash for BoundsCheckPolicy

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for BoundsCheckPolicy

fn eq(&self, other: &BoundsCheckPolicy) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for BoundsCheckPolicy

impl Eq for BoundsCheckPolicy

impl StructuralPartialEq for BoundsCheckPolicy