Random Numbers

AMDGPU.jl generates random numbers on the device through AMD's rocRAND. The quickest way to get a random ROCArray is the package-level rand/randn, which mirror Base:

using AMDGPU

AMDGPU.rand(Float32, 1024)        # uniform in [0, 1)
AMDGPU.randn(Float32, 8, 8)       # standard normal
AMDGPU.rand(Int32, 100)           # uniform integers

The standard Random in-place interface works on existing ROCArrays and fills them on the GPU:

using Random

a = ROCArray{Float32}(undef, 1024)
Random.rand!(a)
Random.randn!(a)

rocRAND handles rand! for UInt8, UInt16, UInt32, Float16, Float32 and Float64, and randn! for the three float types. Every other element type goes to the kernel-based GPUArrays generator instead: rand! accepts any isbits type Random can sample, such as Int32, tuples, or structs with a custom sampler, and randn! accepts any float or complex float type, such as ComplexF32. The mean and stddev keywords of randn! work on both paths.

Seeding

Seed the device generators for reproducible results with AMDGPU.seed!:

AMDGPU.seed!(1234)

This seeds both the rocRAND generator (used by AMDGPU.rand/randn) and the GPUArrays fallback generator, so results are reproducible regardless of which path a given operation takes.