cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Disabled a power-of-two error check as this may be too strict. See test_dims in autotest.cc which dimensions are guaranteed to work with reductions. Other dimensions may also work but are not yet extensively tested

Browse Source
This commit is contained in:
jpekkila
2019-06-17 18:32:23 +03:00
parent 5e6cc9b8cc
commit f970dca68c
3 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/core/kernels/kernels.cuh
View File

@@ -1025,9 +1025,9 @@ reduce_vec(const cudaStream_t stream,
// NOTE! Also does not work properly with non-power of two mesh dimension
// Issue is with "smem[BLOCK_SIZE];". If you init smem to NANs, you can
// see that uninitialized smem values are used in the comparison
ERRCHK_ALWAYS(is_power_of_two(nx));
ERRCHK_ALWAYS(is_power_of_two(ny));
ERRCHK_ALWAYS(is_power_of_two(nz));
//ERRCHK_ALWAYS(is_power_of_two(nx));
//ERRCHK_ALWAYS(is_power_of_two(ny));
//ERRCHK_ALWAYS(is_power_of_two(nz));
if (rtype == RTYPE_MAX || rtype == RTYPE_MIN) {
kernel_reduce_1of3_vec<dlength_vec><<<bpg, tpb, 0, stream>>>(vtxbuf0, vtxbuf1, vtxbuf2, scratchpad);