latqcdtools.base.speedify

compile(func)

compileCUDA(func)

get_optimal_block_size():
"""
Returns an optimal block size based on the current CUDA device.
Defaults to 256 if device information cannot be obtained.

Returns:
    int: Optimal threads per block
"""

numbaList(inList):
"""
Turn a list into List that numba can parse. 
"""

numbaOFF():
"""
Turn off numba compilation for small functions. 
"""

numbaON():
"""
Use numba wherever possible. By default it is turned off, since compilation takes some time,
and hence you will only see a performance boost for particularly long-running functions. Must be
called at the beginning of your code. 
"""

parallel_function_eval(function, input_array, args=(), nproc=6, parallelizer='pathos.pools'):
"""
Parallelize a function over an input_array. Effectively this can replace a loop over an array and should
lead to a performance boost.

Args:
    function (func): to-be-parallelized function 
    input_array (array-like): array over which it should run 
    nproc (int): number of processes 

Returns:
    array-like: func(input_array)
"""

parallel_reduce(function, input_array, args=(), nproc=6, parallelizer='pathos.pools') -> float:
"""
Parallelize a function over an input_array, then sum over the input_array elements. 

Args:
    function (func): to-be-parallelized function 
    input_array (array-like): array over which it should run 
    nproc (int): number of processes 

Returns:
    float
"""

class ComputationClass(function, input_array, args, nproc, parallelizer):