Source code for pyfr.backends.cuda.provider

# -*- coding: utf-8 -*-

from pycuda import compiler, driver

from pyfr.backends.base import (BaseKernelProvider,
                                BasePointwiseKernelProvider, ComputeKernel)
import pyfr.backends.cuda.generator as generator
from pyfr.util import memoize


def get_grid_for_block(block, nrow, ncol=1):
    return (int((nrow + (-nrow % block[0])) // block[0]),
            int((ncol + (-ncol % block[1])) // block[1]))


class CUDAKernelProvider(BaseKernelProvider):
    @memoize
    def _build_kernel(self, name, src, argtypes):
        # Compile the source code and retrieve the kernel
        fun = compiler.SourceModule(src).get_function(name)

        # Prepare the kernel for execution
        fun.prepare(argtypes)

        # Declare a preference for L1 cache over shared memory
        fun.set_cache_config(driver.func_cache.PREFER_L1)

        return fun


[docs]class CUDAPointwiseKernelProvider(CUDAKernelProvider, BasePointwiseKernelProvider): kernel_generator_cls = generator.CUDAKernelGenerator
[docs] def _instantiate_kernel(self, dims, fun, arglst): cfg = self.backend.cfg # Determine the block size if len(dims) == 1: block = (cfg.getint('backend-cuda', 'block-1d', '64'), 1, 1) else: block = cfg.getliteral('backend-cuda', 'block-2d', '128, 1') block += (1,) # Use this to compute the grid size grid = get_grid_for_block(block, *dims[::-1]) class PointwiseKernel(ComputeKernel): def run(self, queue, **kwargs): narglst = [kwargs.get(ka, ka) for ka in arglst] fun.prepared_async_call(grid, block, queue.cuda_stream_comp, *narglst) return PointwiseKernel()