Reproducing syrk/syr2k results

[SerenaC94]

Hi,

I was playing around to synthesize syrk and syr2k kernels with scaleHLS, and I discovered that the test_syrk.c and test_syr2k.c files contain a different implementation that the one of the official polybench kernels (see here and here). What is the reason for that?

When I used the "official" implementation, moreover, kernels optimized by scaleHLS resulted in worse performance than the baseline (sizes 2 to 8) or got stuck for days in the loop space exploration (size 16). How did you manage the size 4096 that is reported in the paper?

[hanchenye]

Hi @SerenaC94, thanks for poking me on this. All the experimental results reported in the paper are based on the kernels in our samples folder. The rewriting was intended to accommodate the prototype C/C++ front-end we developed about one year ago. And the samples haven't been updated since then. Currently we are reconstructing the samples folder to include all "official" benchmarks from polybench. Will have you posted once that's done.