Verbose Mode

NVPL BLAS can be instructed to print information about each interface function invocation that includes the parameters passed (the output aligned with the signature of a function) as well as the time it takes to run. In this case the library also one time prints information about the system it is running on. When verbose mode is enabled the library prints the information to stdout. Each line starts with NVPL_BLAS_VERBOSE:.

To use the verbose mode, set the environment variable:

export NVPL_BLAS_VERBOSE=1

An example of output in verbose mode:

NVPL_BLAS_VERBOSE: NVPL BLAS version 0.1.0
NVPL_BLAS_VERBOSE: Platform: Neoverse V2, cores:72 sve_width:128. Cache: L1:64 KB (cl:64 ways:4 sets:256) L2:1024 KB (cl:64 ways:8 sets:2048) L3:122880 KB (cl:64 ways:12 sets:163840)
NVPL_BLAS_VERBOSE: dgemm_(N,N,128,256,123,2,0xfffdddb62000,128,0xfffdddb12000,123,-1,0xfffdddac2000,128) time_us:4742 int:lp64 max_nthr:72 tid:fffdddeb0020
NVPL_BLAS_VERBOSE: cscal_(1024,0xaaab41ed69e0,0xaaab41ee5000,1) time_us:5.76 int:lp64 max_nthr:72 tid:fffdddeb0020

Warning

Verbose mode has non-negligible performance impact.