veclib.zig: SIMD based Vector Computing Library

WARNING This project is still under development.

veclib utilizes Zig standard SIMD vector calculation. Aside from the standard, veclib supports scalar-vector mixed calculation and runtime-known length slice handling.

Features

• Nullary Function
  • iota
• Element-wise Unary Function (f(comptime T: type, arg: anytype, out: []T) void)
  • copy
  • sqrt
  • sin, cos, tan
  • exp, exp2, exp1m
  • log, log2, log10, log1p
  • abs
  • floor, ceil, trunc, round
  • byteSwap, bitReverse
  • countLeadingZeros, countTrailingZeros, popCount
• Element-wise Binary Function (f(comptime T: type, arg1: anytype, arg2: anytype, out: []T) void)
  • add, wrapAdd, saturateAdd
  • sub, wrapSub, saturateSub
  • mul, wrapMul, saturateMul,
  • div, divFloor, divTrunc, rem, mod
  • bitAnd, bitOr, bitXor
  • min, max
• Element-wise Binary Comparison (f(comptime T: type, arg1: anytype, arg2: anytype, out: [] bool) void)
  • eq, neq, gt, gte, lt, lte,
• Ternary Function (f(comptime T: type, arg1: anytype, arg2: anytype, arg3: anytype, out: []T))
  • mulAdd
  • clip
• Reduction Function (f(comptime T: type, arg: []const T) T)
  • sum, wrapSum, saturateSum,
  • prod, wrapProd, saturateProd,
  • smallest, largest,
  • all, any

We use anytype for type inference, however, there some limitations.

• arg at element-wise functions can be scalar or slice.
  • scalar value will be broadcsted (not to slice size, but to SIMD vector size)
• Not all types are supported at some functions.
  • e.g. Bit-wise operations are only supported for integer types.