Closed tommydcjung closed 1 year ago
@dpetrisko
Hi, I believe the previous behavior was correct (although the name is unintuitive as this ISA manual notes)
I think this is verified by the rv32uf-p-fmadd test here (which may fail now): https://github.com/bespoke-silicon-group/bsg_manycore/tree/d47c40d7d9ed16d0f448686147743e875dd5fa02/software/spmd/bsg_riscv_tests
Never mind guys, got confused by the document.
-Fixing fma ops that were swapped for fnmadd/fnmsub.