Sorry for the patch spamming, but I find easier to break down the
various work I am doing.
This is the first patch that introduces some new functionality.
by using the kernel decorator one could rewrite this code
mod = SourceModule("""
__global__ void doublify(float *a)
{
int idx = threadIdx.x + threadIdx.y*4;
a[idx] *= 2;
}
""")
func = mod.get_function("doublify")
func(a_gpu, block=(4,4,1))
in
from pycuda.decorators import kernel
@kernel
def doublify(a='float *'):
"""
int idx = threadIdx.x + threadIdx.y*4;
a[idx] *= 2;
"""
doublify( a_gpu, block=(4,4,1))
you can pass any flag that SourceModule takes with
@kernel(options=['-O2'])
Of course there are caveats:
- the name of the input variables must correspond to the one inside the doc body
- the default value is a string that specifies the type of the var
- # includes are supported for now, all other stuff that should go before the
- no need to specify the __global__ declaration, it gets generated
see the tests for some example
I think this addition to pycuda would be nice, and the code would look
some what cleaner
The real power will be obvious when you see the element-wise kernel
decorator ...
Comments? ideas? Flames?
Fabrizio
--------------------------
Luck favors the prepared mind. (Pasteur)
_______________________________________________
PyCUDA mailing list
pyc...@host304.hostmonster.com
http://host304.hostmonster.com/mailman/listinfo/pycuda_tiker.net
