Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC Commits: 45305ab8 by sheaf at 2025-09-10T10:43:29-04:00 Make rationalTo{Float,Double} inline in phase 0 We hold off on inlining these until phase 0 to allow constant-folding rules to fire. However, once we get to phase 0, we should inline them, e.g. to expose unboxing opportunities. See CLC proposal #356. - - - - - 2 changed files: - libraries/base/changelog.md - libraries/ghc-internal/src/GHC/Internal/Float.hs Changes: ===================================== libraries/base/changelog.md ===================================== @@ -4,6 +4,7 @@ * Remove deprecated, unstable heap representation details from `GHC.Exts` ([CLC proposal #212](https://github.com/haskell/core-libraries-committee/issues/212)) * Add `Data.List.NonEmpty.mapMaybe`. ([CLC proposal #337](https://github.com/haskell/core-libraries-committee/issues/337)) * Fix issues with toRational for types capable to represent infinite and not-a-number values ([CLC proposal #338](https://github.com/haskell/core-libraries-committee/issues/338)) + * Ensure that `rationalToFloat` and `rationalToDouble` always inline in the end. ([CLC proposal #356](https://github.com/haskell/core-libraries-committee/issues/356)) * Modify the implementation of `Data.List.sortOn` to use `(>)` instead of `compare`. ([CLC proposal #332](https://github.com/haskell/core-libraries-committee/issues/332)) * `GHC.Exts.IOPort#` and its related operations have been removed ([CLC #213](https://github.com/haskell/core-libraries-committee/issues/213)) * Add `thenA` and `thenM`. ([CLC proposal #351](https://github.com/haskell/core-libraries-committee/issues/351)) ===================================== libraries/ghc-internal/src/GHC/Internal/Float.hs ===================================== @@ -466,8 +466,8 @@ instance Fractional Float where recip x = 1.0 / x rationalToFloat :: Integer -> Integer -> Float -{-# NOINLINE [0] rationalToFloat #-} --- Re NOINLINE pragma, see Note [realToFrac natural-to-float] +{-# INLINE [0] rationalToFloat #-} +-- Re INLINE pragma, see Note [realToFrac natural-to-float] rationalToFloat n 0 | n == 0 = 0/0 | n < 0 = (-1)/0 @@ -718,8 +718,8 @@ instance Fractional Double where recip x = 1.0 / x rationalToDouble :: Integer -> Integer -> Double -{-# NOINLINE [0] rationalToDouble #-} --- Re NOINLINE pragma, see Note [realToFrac natural-to-float] +{-# INLINE [0] rationalToDouble #-} +-- Re INLINE pragma, see Note [realToFrac natural-to-float] rationalToDouble n 0 | n == 0 = 0/0 | n < 0 = (-1)/0 @@ -1673,7 +1673,11 @@ Now we'd have a BUILTIN constant folding rule for rationalToFloat; but to allow that rule to fire reliably we should delay inlining rationalToFloat until stage 0. (It may get an inlining from CPR analysis.) -Hence the NOINLINE[0] rationalToFloat, and similarly rationalToDouble. +Hence the INLINE[0] rationalToFloat, and similarly for rationalToDouble. +This activation means: + + - we don't inline until phase 0 (solving the above) + - we do inline starting at phase 0 (because we do want it inlining in the end) -} -- Utils View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/45305ab8bc002b0e9d482e43552b4f64... -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/45305ab8bc002b0e9d482e43552b4f64... You're receiving this email because of your account on gitlab.haskell.org.