tzislam/libnvcd

libNVCD: A Flexible, Adaptive, and User-friendly Performance Measurement Tool for GPUs

libnvcd's Issues

Segmentation fault on V100

Hi,

I followed the procedure outlined on github page and successfully compiled libnvcd.
However, running nvcdrun produced a segmentation fault on our V100 system.

• Spec:
  CentOS Linux release 7.9.2009
  GPU: V100-SMX2 (Driver 510.47.03)
  Modules: gcc/8.3.0, cuda/10.1

Wed Oct 25 10:08:04 2023       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.47.03    Driver Version: 510.47.03    CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla V100-SXM2...  On   | 00000000:3A:00.0 Off |                    0 |
| N/A   28C    P0    42W / 300W |      0MiB / 32768MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

• Compilation:

$ export CUDA_VISIBLE_DEVICES=0 
$ export export CUDA_HOME=/apps/cuda/10.1 
$ export CUDA_ARCH_SM=sm_70 
$ make DEBUG=1 libnvcdhook.so nvcdinfo nvcdrun
$ ls bin/* 
bin/nvcdinfo*  bin/nvcdrun*  bin/libnvcdhook.so*  bin/libnvcd.so

• Runing nvcdinfo

$ export NVCDINFO_DEVICE_ID=0 
$ export NVCDINFO_GROUP_SIZE=5
$ LD_LIBRARY_PATH=$PWD/bin:$LD_LIBRARY_PATH make nvcdinfo_generate_csv 
/scratch/optpar01/work/2023/17-libnvcd/libnvcd/bin/nvcdinfo -d 0 -n 5
GPU 0
	gpu_name = Tesla V100-SXM2-32GB
	gpu_uuid = GPU-8cfe493c-fe94-943f-4f3d-99abd2ba7fa3
=======multiplex=========
|INFO|Processing domain: domain_a
|INFO|	Number of events available in this domain: 24
|INFO|Processing domain: domain_b
|INFO|	Number of events available in this domain: 4
|INFO|Processing domain: domain_d
|INFO|	Number of events available in this domain: 30
|INFO|Processing domain: domain_e
|INFO|	Number of events available in this domain: 20
|INFO|Processing domain: domain_p
|INFO|	Number of events available in this domain: 2
|INFO|Processing domain: domain_s
|INFO|	Number of events available in this domain: 2

Could you kindly clarify what the meaning of 'a, b, d, e, p, s' are ? (Notably, 'c' is missing)

• Running nvcdrun

$ export BENCH_EVENTS=$(head -n1 cupti_group_info/device_0/domain_a.csv)
$ LD_LIBRARY_PATH=$PWD/bin/:$LD_LIBRARY_PATH LD_PRELOAD=$LD_PRELOAD:$PWD/bin/libnvcdhook.so bin/nvcdrun  
TEST MODE: MULTI-THREADED
CUDA RUNTIME: /scratch/optpar01/work/2023/17-libnvcd/libnvcd/nvcdrun/src/main.c:64:'cudaSetDevice(device)' failed. [Reason] cudaErrorInvalidDevice:invalid device ordinal
CUDA RUNTIME: /scratch/optpar01/work/2023/17-libnvcd/libnvcd/nvcdrun/src/main.c:64:'cudaSetDevice(device)' failed. [Reason] cudaErrorInvalidDevice:invalid device ordinal
CUDA RUNTIME: /scratch/optpar01/work/2023/17-libnvcd/libnvcd/nvcdrun/src/main.c:64:'cudaSetDevice(device)' failed. [Reason] cudaErrorInvalidDevice:invalid device ordinal
*** Error in `bin/nvcdrun': double free or corruption (fasttop): 0x0000000000659790 ***

Regards.