Function scalarF16Range

scalarF16Range(counts?): number[]
Parameters
- counts: {
      neg_norm?: number;
      neg_sub?: number;
      pos_norm: number;
      pos_sub: number;
  } = ...
  structure param with 4 entries indicating the number of entries to be generated each region, entries must be 0 or greater.
  - Optional neg_norm?: number
  - Optional neg_sub?: number
  - pos_norm: number
  - pos_sub: number
Returns number[]
an ascending sorted array of numbers spread over the entire range of 16-bit floats

Numbers are divided into 4 regions: negative normals, negative subnormals, positive subnormals & positive normals. Zero is included.

Numbers are generated via taking a linear spread of the bit field representations of the values in each region. This means that number of precise f16 values between each returned value in a region should be about the same. This allows for a wide range of magnitudes to be generated, instead of being extremely biased towards the edges of the f16 range.

This function is intended to provide dense coverage of the f16 range, for a minimal list of values to use to cover f16 behaviour, use sparseScalarF16Range instead.
- Defined in src/webgpu/util/math.ts:1038

Generated using TypeDoc