cuda call doing leakage ?

Accelerated Computing CUDA CUDA Programming and Performance
1

hi all,
using Valgrind I’m getting this kind of reports:

==32626== 21,360 bytes in 5 blocks are still reachable in loss record 25 of 25
==32626== at 0x4A05E03: malloc (vg_replace_malloc.c:195)
==32626== by 0x72536C7: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x7253E12: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x726B445: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x722C1AD: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x7223530: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x727DB45: cuInit (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x76C7914: ??? (in /usr/local/cuda/lib64/libcudart.so.2.3)
==32626== by 0x76A8F53: cudaGetDeviceCount (in /usr/local/cuda/lib64/libcudart.so.2.3)

==32626== 792 bytes in 5 blocks are still reachable in loss record 21 of 25
==32626== at 0x4A05E03: malloc (vg_replace_malloc.c:195)
==32626== by 0x725442F: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x7245818: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x725387B: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x7253E12: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x726B445: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x722C1AD: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x7223530: ??? (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x727DB45: cuInit (in /usr/lib64/libcuda.so.190.32)
==32626== by 0x76C7914: ??? (in /usr/local/cuda/lib64/libcudart.so.2.3)
==32626== by 0x76A8F53: cudaGetDeviceCount (in /usr/local/cuda/lib64/libcudart.so.2.3)

Shall I warry about it ?

Regards

2

It is only about 20kb and it only happens once, so probably not. Incidentally, one of the nvidia people who post here mentioned a little while ago that all of the known cuda valgrind errors have been fixed in Cuda 3.0, so in all likelihood it will be a thing of the past once Cuda 3.0 is released and you upgrade…