Variant of the STREAM benchmark written in CUDA Fortran (and, hence, working on the GPU).
The PGI compiler is required to compile.
The four STREAM benchmarks are run on the GPU, that is (with arrays a, b, c):
- COPY:
c_j = a_j - SCALE:
b_j = scalar * c_j - ADD:
c_j = a_j + b_j - TRIAD:
a_j = b_j + scalar * c_j
For information is available in stream.F90. The Makefile should suffice to call make run for compiling and running the four micro-benchmarks.
See also www.streambench.org, especially for notes on publishing results based on (variants of) the STREAM benchmark.