[PATCH] D113959: [llvm][RISC-V] Use floating-point ABI by default if possible

[Jessica Clarke via Phabricator via llvm-commits]

jrtc27 added a comment.

All the preparatory changes to the RUN lines should probably be committed separately from the functional change. How did you update the tests; did you search for all that enabled f and d, or did you just look for ones that failed? Just wondering if there's any subtlety there about tests that appear to be unchanged but end up following slightly different code paths from what they intend to test.



================
Comment at: llvm/lib/Target/RISCV/MCTargetDesc/RISCVBaseInfo.cpp:37
+  bool IsRVD = FeatureBits[RISCV::FeatureStdExtD];
+  bool IsRVF = FeatureBits[RISCV::FeatureStdExtF];
 
----------------
I'd put these two declarations the other way round so they match the canonical order; only the uses need to be in "reverse" order to ensure we check from most to least specific.


================
Comment at: llvm/lib/Target/RISCV/MCTargetDesc/RISCVBaseInfo.cpp:61
 
   // For now, default to the ilp32/ilp32e/lp64 ABI if no explicit ABI is given
+  // or an invalid/unrecognised string is given.
----------------
This comment isn't really correct any more, even with your added sentence; clearly you can't use an lp64 ABI on rv32, the whole point of it was stating that a soft-float ABI is the default.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113959/new/

https://reviews.llvm.org/D113959