Module copy 

GPU-accelerated strided copy for making tensors contiguous.

Copies a 2D strided tensor to a contiguous layout: dst[row * cols + col] = src[row * stride_row + col * stride_col]

Used after transpose/permute operations to produce contiguous memory.

Structs

StridedCopyParams

Parameters for a strided copy operation.

Statics

COPY_SHADER_SOURCE

MSL source for the strided copy kernel (embedded at compile time).

Functions

dispatch_copy_f32

Copy count f32 elements from src[src_offset..] to dst[dst_offset..].

dispatch_strided_copy_f32

Dispatch a strided copy operation on the GPU.

register

Register strided copy shader source with the given kernel registry.