Python Parallel Programming Cookbook

The MapReduce operation with PyCUDA

PyCUDA provides a functionality to perform reduction operations on the GPU. This is possible with the pycuda.reduction.ReductionKernel method:

ReductionKernel(dtype_out, arguments, map_expr ,reduce_expr, 
                name,optional_parameters)  

Here, we note that:

• dtype_out: This is the output's data type. It must be specified by the numpy.dtype data type.
• arguments: This is a C argument list of all the parameters involved in the reduction's operation.
• map_expr: This is a string that represents the mapping operation. Each vector in this expression must be referenced with the variable i.
• reduce_expr: This is a string that represents the reduction operation. The operands in this expression are indicated by lowercase letters, such ...