Floating-Point Numbers

Double-as-Supertype

Like in the JS API, double (aka JavaScript's native Number type) is used in several places as a supertype of various numeric values. Such a value will be numerically downcast to the appropriate subtype depending how it is used, as specified in the JS API spec.

• In WGPUColor, as a supertype of f32/u32/i32
  • In WGPURenderPassColorAttachment::clearValue, the type depends on the texture format.
  • In wgpuRenderPassEncoderSetBlendConstant, the type is f32.
• In WGPUConstantEntry::value, as a supertype of all of the overrideable WGSL types (bool/f32/u32/i32/f16 and possibly more).
  • The type depends on the WGSL type of the constant being overridden.

Nullable Floating-Point Type

Floating-point-typed (float/double) values which are nullable or optional use NaN to represent the null value. A value value represents the null value iff isnan(value) != 0. (Do not use an equality check with a NaN constant, because NaN == NaN is false.)

Infinities are invalid. See Non-Finite Float Value Errors.

Non-Finite Float Value Errors

The JavaScript API does not allow non-finite floating-point values (it throws a TypeError).

In webgpu.h, a value is finite iff isfinite(value) != 0. Using a non-finite value (aside from NaN for a Nullable Floating-Point Type) results in a validation Device Error, except on Wasm-on-JS targets where a TypeError thrown by JS may be passed unhandled, interrupting Wasm execution.