Make the functor test that compares different SIMD backends more tolerant of numerical differences (!2217) · Merge requests · LHCb / Rec

Olli Lupton requested to merge olupton_functor_test_tolerance into master Sep 21, 2020

Allows larger differences between the numerical results obtained when using different SIMD backends. Closes #160 (closed).

(Background: with gcc9 then it seems that less ordering/contraction of floating point operations is done with avx512 than with scalar/sse/avx2, which leads to some small numerical differences)

cc: @ahennequ @dovombru

Make the functor test that compares different SIMD backends more tolerant of numerical differences

Merge request reports