batch_identity()

__kernel void stan::math::opencl_kernels::batch_identity	(	__global double *	A,
		unsigned int	batch_rows,
		unsigned int	size
	)

Makes a batch of smaller identity matrices inside the input matrix.

This kernel operates inplace on the matrix A, filling it with smaller identity matrices with a size of batch_rows x batch_rows. This kernel expects a 3D organization of threads: 1st dim: the number of matrices in the batch. 2nd dim: the number of cols/rows in batch matrices. 3rd dim: the number of cols/rows in batch matrices. Each thread in the organization assigns a single value in the batch. In order to create a batch of 3 matrices the size of NxN you need to run the kernel batch_identity(A, N, 3*N*N) with (3, N, N) threads. The special case of batch_identity(A, N, N*N) executed on (1, N, N) threads creates a single identity matrix the size of NxN and is therefore equal to the basic identity kernel.

Parameters

[in,out]	A	The batched identity matrix output.
	batch_rows	The number of rows/cols for the smaller matrices in the batch
	size	The size of A.

Note

Code is a const char* held in identity_kernel_code. This kernel uses the helper macros available in helpers.cl.

Definition at line 40 of file batch_identity.hpp.