Algebraically simplify add/sub with carry/overflow

Previously, the `{add,sub}{Int,Word}C#` PrimOps weren't handled
in PrelRules (constant folding and algebraic simplification) at all.
This implements the necessary logic, so that using these primitives
isn't too punishing compared to their well-optimised, overflow-unaware
counterparts.

This is so that using these primitives in `enumFromThenTo @Int` can
be optimized by constant folding, reducing closure sizes.

Reviewers: bgamari, simonpj, hsyl20

Reviewed By: bgamari, simonpj

Subscribers: AndreasK, thomie, carter

GHC Trac Issues: #8763

Differential Revision: https://phabricator.haskell.org/D4605

4 files changed, 36 insertions, 8 deletions

diff --git a/testsuite/tests/numeric/should_run/T10962.hs b/testsuite/tests/numeric/should_run/T10962.hs
index 896c9e987f..435f3637d7 100644
--- a/testsuite/tests/numeric/should_run/T10962.hs
+++ b/testsuite/tests/numeric/should_run/T10962.hs
@@ -1,16 +1,32 @@
 {-# LANGUAGE MagicHash     #-}
 {-# LANGUAGE UnboxedTuples #-}
 
-module Main where
+module Main (main) where
 
 import GHC.Base
 
+unW# :: Word -> Word#
+unW# (W# w) = w
+
+type WordOpC = Word# -> Word# -> (# Word#, Int# #)
+
+check :: WordOpC -> Word# -> Word# -> IO ()
+check op a b = do
+  let (# w, c #) = op a b
+  print (W# w, I# c)
+
+checkSubInlNoInl :: WordOpC -> Word# -> Word# -> IO ()
+checkSubInlNoInl op a b = do
+  inline check op a b   -- constant folding
+  noinline check op a b -- lowering of PrimOp
+{-# INLINE checkSubInlNoInl #-}
+
 main :: IO ()
 main = do
   -- Overflow.
-  let (# w1, i1 #) = subWordC# 1## 3##
-  print (W# w1, I# i1)
+  checkSubInlNoInl subWordC# 1## 3##
+  checkSubInlNoInl addWordC# (unW# (inline maxBound)) 3##
 
   -- No overflow.
-  let (# w2, i2 #) = subWordC# 3## 1##
-  print (W# w2, I# i2)
+  checkSubInlNoInl subWordC# 5## 2##
+  checkSubInlNoInl addWordC# (unW# (inline maxBound-1)) 1##
diff --git a/testsuite/tests/numeric/should_run/T10962.stdout-ws-32 b/testsuite/tests/numeric/should_run/T10962.stdout-ws-32
index a1dec8410a..605265305d 100644
--- a/testsuite/tests/numeric/should_run/T10962.stdout-ws-32
+++ b/testsuite/tests/numeric/should_run/T10962.stdout-ws-32
@@ -1,2 +1,8 @@
 (4294967294,1)
-(2,0)
+(4294967294,1)
+(2,1)
+(2,1)
+(3,0)
+(3,0)
+(4294967295,0)
+(4294967295,0)
diff --git a/testsuite/tests/numeric/should_run/T10962.stdout-ws-64 b/testsuite/tests/numeric/should_run/T10962.stdout-ws-64
index 853bf94a61..d36f660eb1 100644
--- a/testsuite/tests/numeric/should_run/T10962.stdout-ws-64
+++ b/testsuite/tests/numeric/should_run/T10962.stdout-ws-64
@@ -1,2 +1,8 @@
 (18446744073709551614,1)
-(2,0)
+(18446744073709551614,1)
+(2,1)
+(2,1)
+(3,0)
+(3,0)
+(18446744073709551615,0)
+(18446744073709551615,0)
diff --git a/testsuite/tests/numeric/should_run/all.T b/testsuite/tests/numeric/should_run/all.T
index 37fff44bde..691fc26f7a 100644
--- a/testsuite/tests/numeric/should_run/all.T
+++ b/testsuite/tests/numeric/should_run/all.T
@@ -62,6 +62,6 @@ test('CarryOverflow', omit_ways(['ghci']), compile_and_run, [''])
 test('T9407', normal, compile_and_run, [''])
 test('T9810', normal, compile_and_run, [''])
 test('T10011', normal, compile_and_run, [''])
-test('T10962', omit_ways(['ghci']), compile_and_run, [''])
+test('T10962', omit_ways(['ghci']), compile_and_run, ['-O2'])
 test('T11702', extra_ways(['optasm']), compile_and_run, [''])
 test('T12136', normal, compile_and_run, [''])