julia> a = CUDA.rand(Float16, 4, 4);
julia> b = CUDA.rand(Float16, 4, 4);
julia> c = CUDA.zeros(Float32, 4, 4);
julia> GemmKernels.BLAS.gemmEx!('N', 'N', 1f0, a, b, 0f0, c)
conf.launch_args = (threads = 256, blocks = (0, 0), shmem = 65536)
ERROR: ArgumentError: Grid dimensions should be non-null
Culprit: https://github.com/JuliaGPU/GemmKernels.jl/blob/c376fdca8fa196614b10a683871058dd1be9ce42/src/config.jl#L42=
Should probably be a
cld, but even then the computed values are wrong.