cuda_transpose_kernel_impl() [2/4]

void syten::CudaDenseTensorImpl::cuda_transpose_kernel_impl	(	std::uint32_t	rank,
		std::size_t	sz,
		const float *	inp,
		float *	out,
		cukrn_transpose_array const &	old_dim,
		cukrn_transpose_array const &	new_dim,
		cukrn_transpose_array const &	ar_perm,
		void *	str,
		bool	do_conj
	)

Launcher for the CUDA tensor transposition kernel, float version.

Parameters

rank	number of tensor indices
sz	number of elements in the tensor
inp	input tensor array, pointer to device memory
out	output tensor array, pointer to device memory
old_dim	dimensions of the input tensor as a simple POD
new_dim	dimensions of the output tensor as a simple POD
ar_perm	permutation, perm[i] = j puts the old leg j at position i+1
str	CUDA stream in which the computation will take place
do_conj	whether to conjugate every entry (ignored)