From a95bbd0bdf06d7d61b0bef6de77b59ca31b2c32d Mon Sep 17 00:00:00 2001
From: Sebastian Graf <sebastian.graf@kit.edu>
Date: Thu, 14 May 2020 14:40:41 +0200
Subject: [PATCH] Make `Int`'s `mod` and `rem` strict in their first arguments
They used to be strict until 4d2ac2d (9 years ago).
It's obviously better to be strict for performance reasons.
It also blocks #18067.
NoFib results:
```
--------------------------------------------------------------------------------
Program Allocs Instrs
--------------------------------------------------------------------------------
integer -1.1% +0.4%
wheel-sieve2 +21.2% +20.7%
--------------------------------------------------------------------------------
Min -1.1% -0.0%
Max +21.2% +20.7%
Geometric Mean +0.2% +0.2%
```
The regression in `wheel-sieve2` is due to reboxing that likely will go
away with the resolution of #18067. See !3282 for details.
Fixes #18187.
---
libraries/base/GHC/Real.hs | 19 +++++++++++--------
1 file changed, 11 insertions(+), 8 deletions(-)
diff --git a/libraries/base/GHC/Real.hs b/libraries/base/GHC/Real.hs
index 38a3aa2870d..1425f8c306e 100644
--- a/libraries/base/GHC/Real.hs
+++ b/libraries/base/GHC/Real.hs
@@ -334,11 +334,8 @@ instance Integral Int where
-- in GHC.Int
| otherwise = a `quotInt` b
- a `rem` b
+ !a `rem` b -- See Note [Special case of mod and rem is lazy]
| b == 0 = divZeroError
- -- The quotRem CPU instruction fails for minBound `quotRem` -1,
- -- but minBound `rem` -1 is well-defined (0). We therefore
- -- special-case it.
| b == (-1) = 0
| otherwise = a `remInt` b
@@ -348,11 +345,8 @@ instance Integral Int where
-- in GHC.Int
| otherwise = a `divInt` b
- a `mod` b
+ !a `mod` b -- See Note [Special case of mod and rem is lazy]
| b == 0 = divZeroError
- -- The divMod CPU instruction fails for minBound `divMod` -1,
- -- but minBound `mod` -1 is well-defined (0). We therefore
- -- special-case it.
| b == (-1) = 0
| otherwise = a `modInt` b
@@ -368,6 +362,15 @@ instance Integral Int where
| b == (-1) && a == minBound = (overflowError, 0)
| otherwise = a `divModInt` b
+{- Note [Special case of mod and rem is lazy]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+The `quotRem`/`divMod` CPU instruction fails for minBound `quotRem` -1, but
+minBound `rem` -1 is well-defined (0). We therefore special-case for `b == -1`,
+but not for `a == minBound` because of Note [Order of tests] in GHC.Int. But
+now we have to make sure the function stays strict in a, to guarantee unboxing.
+Hence the bang on a, see #18187.
+-}
+
--------------------------------------------------------------
-- Instances for @Word@
--------------------------------------------------------------
--
GitLab