Wave-DAG is a preprocessor for accelerating structured wavefront patterns on multi core architectures built on top of OpenMP 5.0, ensuing future portability across platforms.
The test cases used in Wave-DAG were taken from the Wavebench project. For more information about Wavebench see wavebench/README.md.
For more information about the underlying library see wavebench-dag/include/Readme.md.
Basic Smith-Waterman serial code.
for(int i = 1; i < n; ++i)
for(int j = ; j < m; ++j) {
int score = (A[i - 1] == B[j - 1])? match : miss;
M[i * m + j] = max(M[(i - 1) * m + (j - 1)] + score,
max(M[i * m +(j - 1)] + gap, M[(i - 1) * m + j] + gap));
}
Basic Smith-Waterman with Wave-DAG pragmas added.
#pragma omp dag coarsening(block, 512, 512)
for(int i = 1; i < n; ++i)
for(int j = ; j < m; ++j) {
#pragma omp dag task depend({(i + 1) * m + j + 1,((i + 1) < n) && ((j + 1) < m)}, \
{(i + 1) * m + j,(i + 1) < n}, \
{i * m + j + 1,(j + 1) < m})
{
int score = (A[i - 1] == B[j - 1]) ? match : miss;
M[i * m + j] = max(M[(i - 1) * m + (j - 1)] + score,
max(M[i * m +(j - 1)] + gap, M[(i - 1) * m + j] + gap));
}
}
The resulting code then will be preprocessod by the Wave-DAG preprocessor and output a C++ OpenMP 5.0 compliant parallel code.
Lower the better: Execution time for the Smith-Waterman Wavebench benchmark on Perlmutter NVIDIA A100 GPUs and AMD EPYC 7763 CPUs. Both nvc-mp
& gcc-omp-dag
were executed using 64 threads.
To build Wave-DAG you need GCC 9.2 and in order to run the original wavebench code you need a working PGI compiler. To compile the tests use:
make omp # Compile the Wave-DAG version of the test cases
make serial # Compile a serial version of the test cases
make acc-gpu # Compile the Wavebench GPU version of the test cases
To run the samples use:
bash run.sh
For more information see LICENSE.