User-defined reduction kernel.
This class can be used to define a reduction kernel with or without broadcasting.
The kernel is compiled at an invocation of the
__call__()method, which is cached for each device. The compiled binary is also cached into a file under the
$HOME/.cupy/kernel_cache/directory with a hashed file name. The cached binary is reused by other processes.
- in_params (str) – Input argument list.
- out_params (str) – Output argument list.
- map_expr (str) – Mapping expression for input values.
- reduce_expr (str) – Reduction expression.
- post_map_expr (str) – Mapping expression for reduced values.
- identity (str) – Identity value for starting the reduction.
- name (str) – Name of the kernel function. It should be set for readability of the performance profiling.
- reduce_type (str) – Type of values to be used for reduction. This type
is used to store the special variables
- reduce_dims (bool) – If
True, input arrays are reshaped without copy to smaller dimensions for efficiency.
- preamble (str) – Fragment of the CUDA-C/C++ code that is inserted at the top of the cu file.
- options (tuple of str) – Additional compilation options.
__call__(self, *args, **kwargs)¶
Compiles and invokes the reduction kernel.
The compilation runs only if the kernel is not cached. Note that the kernels with different argument dtypes, ndims, or axis are not compatible. It means that single ReductionKernel object may be compiled into multiple kernel binaries.
Arrays are returned according to the
out_paramsargument of the