Closed GoogleCodeExporter closed 9 years ago
I modified a few things to get it to work.
In b40c_kernel_utils.cu, line 278:
template <int Value>
__global__ void FlushKernel()
{
}
This function is called twice in radixsort_api.cu, lines 470, 502:
FlushKernel<0> <<<_grid_size, RADIXSORT_THREADS,
scan_scatter_attrs.sharedSizeBytes>>>();
FlushKernel<0><<<_grid_size, RADIXSORT_THREADS,
scan_scatter_attrs.sharedSizeBytes>>>();
In file kernel/radixsort_kernel_common.cu, I changed the IsKeysOnly function to
specialize using a class template, just after line 130.
In file kernel/radixsort_spine_kernel.cu, line 116. I made it a template
function. This isn't the best way since it could be given incorrect template
arguments, but it does compile.
In file radixsort_api.cu, to accomodate the templateness of the spine kernel,
line 313 had to be changed.
cudaFuncGetAttributes(&_spine_scan_kernel_attrs, SrtsScanSpine<unsigned int>);
With these changes I've successfully able to compile multiple objects and then
link them using the example files above.
Original comment by scott.ro...@gmail.com
on 12 Aug 2010 at 3:24
Attachments:
Hi Scott,
Thanks for the report. I encountered that problem yesterday as well when
integrating Duane's code into Thrust. Can you confirm that r201 resolves the
problem?
Original comment by wnbell
on 12 Aug 2010 at 11:16
Hey Nathan,
just one more multiple definition:
line 99 in radixsort_api.cu,
bool RADIXSORT_DEBUG = false;
making it const is one fix.
Original comment by scott.ro...@gmail.com
on 13 Aug 2010 at 4:42
Ah, I forgot about that one :)
Should be resolved in r206
Original comment by wnbell
on 13 Aug 2010 at 9:30
Original issue reported on code.google.com by
scott.ro...@gmail.com
on 11 Aug 2010 at 11:58Attachments: