Expression ckernels accept multiple input memory addresses as an input, and write to a single memory address, with no state shared between multiple calls of the ckernel.
A simple example of this is an addition ckernel,
with a signature like "(int32, int32) -> int32".
This ckernel would expect the src and src_stride
parameters to be arrays of size two, and all the
data pointers to point to int32_t values.
The function pointer for such a ckernel looks like this:
# include/dynd/kernels/expr_kernels.hpp
/** Typedef for an expression operation on a single element */
typedef void (*expr_single_operation_t)(char *dst, const char *const *src,
ckernel_prefix *extra);
/** Typedef for an expression operation on a strided segment of elements */
typedef void (*expr_strided_operation_t)(char *dst, intptr_t dst_stride,
const char *const *src,
const intptr_t *src_stride,
size_t count, ckernel_prefix *extra);
The number of source addresses is baked into the ckernel prior to getting to this low level of representation, but is available at the deferred ckernel level.