[PATCH] D113703: [RISCV] Reduce duplicate FP test cases.

Alex Bradbury via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Nov 12 02:32:57 PST 2021


asb added a comment.

To say a little more about naming and how it got this way (no argument that it's confusing / could be improved!):
While implementing the backend I came at FP support from two directions - firstly going through the ISA manual and ensuring patterns were in place demonstrating the generation of each instruction (the *-arith.ll files), and then going through the LLVM IR manual and checking other backends test files to ensure we have coverage of all intrinsics (the *-intrinsics.ll files) - some being directly lowered to instructions, others appropriately expanded. The duplication is inelegant, but I figured it may be confusing if trying to validate those files were complete to find them incomplete. Of course that's all an artifact of what I personally found useful when bringing the backend up, and needn't constrain what we want to do moving forwards. It does still make some logical sense in my mind to group together tests that are effectively written alongside the selectiondag patterns though.

I've got no objections to the changes proposed in the patch (and the suggestion to split out fma to separate file seems fine), but perhaps the header comment could also be updated or just removed.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113703/new/

https://reviews.llvm.org/D113703



More information about the llvm-commits mailing list