This repo is an update to the reference Implementation of the HPL-AI Benchmark with PAPI instrumentation of FP64, FP32, FP16, and BF16 FLOPS.

To boostrap the configuration files make sure you have GNU Autotools installed and type:

```
autoreconf -ivf
```

To configure the build and prepare for compilation run:

```
./configure
```

Use `--help` option to see the available settings.

To compile and run the benchmark executable:

```
$ make
$ ./hpl-ai <n> <max_iter>
```
Where n is the size of linear system to solve.
max_iter is the maximum number of iterations for GMRES iterative solver.

Example output:

```
Time spent in conversion to single:        0.000 second
Time spent in factorization       :        0.004 second
Time spent in solve               :        0.000 second
Time spent in conversion to double:        0.000 second
Residual norm at the beginning of GMRES: 1.176789e-08
Estimated residual norm at the 1-th iteration of GMRES: 2.550870e-15
Estimated residual norm at the 2-th iteration of GMRES: 5.056252e-22
Time spent in GMRES               :        0.001 second
Total time                        :        0.005 second
Effective operation per sec       :     0.140091 GFLOPs
The following scaled residual check will be computed:
||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )
The relative machine precision (eps) is taken to be: 1.110223e-16
Computational tests pass if scaled residuals are less than 16.0
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 4.531977 ...PASSED
```

SPDX-License-Identifier: BSD 3-Clause Clear License