The CUDA example "ta_vector_cuda" crashes on V100
victor-anisimov opened this issue · 1 comments
commit f49b725 (HEAD -> master, origin/master, origin/HEAD)
Merge: d51ca60 df73764
Author: Eduard Valeyev eduard@valeyev.net
Date: Thu Jun 9 10:33:56 2022 +0200
examples/cuda/ta_vector_cuda 20 4 40 4 5
created 3 CUDA streams + 2 I/O streams
CUDA {driver,runtime} versions = 11060,11060
CUDA Device Information for MPI Process Rank: 0
Device #0: Tesla V100-SXM2-32GB
managedMemory = 1
singleToDoublePrecisionPerfRatio = 2
attrUnifiedAddressing = 1
attrConcurrentManagedAccess = 1
{total,free} memory = {34089926656,33575600128}
GPU vector operations.
TiledArray: dense matrix vector test...
Number of nodes = 1
Size of Matrix = 20x40 (6.4e-06 GB)
Size of Block = 4x4
N of blocks of C = 50
Average N of blocks of C/node = 50
Add test: a(m,n) + b(m,n)
Iteration 1 wall time: 0.00546789
Iteration 2 wall time: 0.00245285
Iteration 3 wall time: 0.00247002
Iteration 4 wall time: 0.00225306
Iteration 5 wall time: 0.0022161
Average wall time = 0.00298138 sec
Average GFLOPS = 0.000268332
Add scale test: 2a(m,n) + 2b(m,n)
Iteration 1 wall time: 0.00484109
Iteration 2 wall time: 0.00423312
Iteration 3 wall time: 0.00419188
Iteration 4 wall time: 0.00417304
Iteration 5 wall time: 0.00416994
Average wall time = 0.00433102 sec
Average GFLOPS = 0.000554142
Add permute test: 2a(m,n) + 2b(n,m)
!! ERROR TiledArray: The TiledRanges of the left- and right-hand arguments of the binary operation are not equal:
left = ( tiles = [ [0,0], [5,10] ), elements = [ [0,0], [20,40] ) )
right = ( tiles = [ [0,0], [10,5] ), elements = [ [0,0], [40,20] ) )
TiledArray: exception at tiledarray/src/TiledArray/expressions/binary_engine.h(249): The TiledRanges of the left- and right-hand arguments of the binary operation are not equal.
Segmentation fault