Skip to content

Basic Scalars

The fundamental WGSL types are:

  • i32 - a signed, two's complement, 32-bit integer.
  • u32 - an unsigned 32-bit integer.
  • f32 - an IEEE-754 binary32, floating point.
  • bool - a true or false value, with no specified storage representation.

These types are known as concrete scalar types, and can be used in all evaluation stages.

Sub-16-Bit Scalar Limitations

WGSL does not support low-precision scalar float types like f8, f4, or f2.

  • Booleans as 1-Bit Flags: A bool acts logically as a 1-bit value (conceptually a logical "f1"). However, bool types have no specified physical storage representation and cannot be stored in host-shareable memory buffers (such as uniform or storage buffers).
  • Minimum Floating-Point Precision: The smallest native floating-point types supported by WebGPU are f32 and f16 (the latter requiring the f16 hardware extension).

Bitwise Data Packing for Compact Storage

For applications demanding high-density, low-precision storage (such as packing 3D normal vectors, custom 8-bit color weights, or specular coefficients), developers must pack their data manually.

Instead of relying on nonexistent hardware types like f8 or f4, developers map these values to integers and utilize bitwise shift and mask operations to store multiple components inside a single u32 or i32 scalar.

See the Bitwise Operators & Data Packing guide for a comprehensive tutorial on compression and decompression techniques on the GPU.