Algebraic-Programming / ALP

Home of ALP/GraphBLAS and ALP/Pregel, featuring shared- and distributed-memory auto-parallelisation of linear algebraic and vertex-centric programs. Soon with more to come!
Apache License 2.0
25 stars 4 forks source link

Wrong fma perf test #335

Open alberto-scolari opened 6 months ago

alberto-scolari commented 6 months ago

The fma test is wrong, reporting

Benchmark label: grb::eWiseApply (axpy) of size 10
Info: grb::init (reference) called.
1 (expected) does not equal 2 (template optimised) at position 0.
Outer iteration #0 timings (io, preamble, useful, postamble, time since epoch): 0.000000, 0.010252, 0.000185, 0.000000, 1712665968227
1 (expected) does not equal 2 (template optimised) at position 0.
Outer iteration #1 timings (io, preamble, useful, postamble, time since epoch): 0.000000, 0.013413, 0.000264, 0.000000, 1712665969227
1 (expected) does not equal 2 (template optimised) at position 0.
Outer iteration #2 timings (io, preamble, useful, postamble, time since epoch): 0.000000, 0.020564, 0.000199, 0.000000, 1712665970227
...

The problem is probably due to the change about the in-place semantics.

The very problem is that the error went long time unnoticed.