[PATCH] D126364: Fix interaction of pragma FENV_ACCESS with other pragmas

[Eli Friedman via Phabricator via cfe-commits]

efriedma added a comment.

The way I see it, there are two possibilities here:

1. In Sema, we have two rounding modes that correspond to FE_DYNAMIC: llvm::RoundingMode::Dynamic, and llvm::RoundingMode::NearestTiesToEven, plus some boolean to indicate whether the user actually explicitly specified FE_TONEAREST.  (You're currently missing the boolean, which means that currently "#pragma STDC FENV_ACCESS OFF" followed by "#pragma STDC FENV_ACCESS ON" actually mutates the rounding mode.) We juggle the rounding modes and the bit based on whether FENV_ACCESS is currently enabled.
2. We just have one rounding mode in Sema that corresponds to FE_DYNAMIC.  Then in CodeGen, we set the actual rounding mode based on whether FENV_ACCESS is currently enabled.

(2) seems a lot simpler.

> On targets that support static rounding mode (like RISCV) dynamic and constant rounding modes may be different and the behavior changes if default mode is replaced by dynamic.

Whether a target supports static rounding modes on floating-point instructions is completely irrelevant.  The user-visible behavior must be the same either way.  If a target doesn't have specialized instructions, the code generator can save/restore the rounding mode.  This should be transparent to the user; the user can't read the rounding mode in between the save and restore.  (We already do this sort of rounding mode manipulation on x86, to implement float-to-int conversion on targets without SSE.)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D126364/new/

https://reviews.llvm.org/D126364