This avoids a crash for scalable vectors and or scalarization for
fixed vectors.
The algorithm is different enough that I don't think it makes sense
to merge with ceil/floor/trunc. Algorithm is adapted from gcc's X86
SSE2 output.
Differential D117247
[RISCV] Add inline expansion for vector fround. craig.topper on Jan 13 2022, 1:17 PM. Authored by
Details This avoids a crash for scalable vectors and or scalarization for The algorithm is different enough that I don't think it makes sense
Diff Detail
Event Timeline
Comment Actions Sorry for dropping the ball on this: floating-point isn't in my comfort zone. It sounds like it should do the right thing to me. I'll try to run this over our downstream (OpenCL) testing and see if it passes.
Comment Actions LGTM, our testing shows a nice improvement with this. Do you think we should backport this to LLVM 14, given it fixes a scalable-vector crash?
|