[Git][ghc/ghc][wip/T16742] PrelRules: Don't break let/app invariant in shiftRule
Ben Gamari
gitlab at gitlab.haskell.org
Tue Jun 11 22:51:10 UTC 2019
Ben Gamari pushed to branch wip/T16742 at Glasgow Haskell Compiler / GHC
Commits:
04196035 by Ben Gamari at 2019-06-11T22:50:52Z
PrelRules: Don't break let/app invariant in shiftRule
Previously shiftRule would rewrite as invalid shift like
```
let x = I# (uncheckedIShiftL# n 80)
in ...
```
to
```
let x = I# (error "invalid shift")
in ...
```
However, this breaks the let/app invariant as `error` is not
okay-for-speculation. There isn't an easy way to avoid this so let's not
try. Instead we just take advantage of the undefined nature of invalid
shifts and return zero.
Fixes #16742.
- - - - -
7 changed files:
- compiler/coreSyn/CoreSyn.hs
- compiler/prelude/PrelRules.hs
- compiler/prelude/PrimOp.hs
- testsuite/tests/codeGen/should_run/T16449_2.hs
- − testsuite/tests/codeGen/should_run/T16449_2.stderr
- + testsuite/tests/codeGen/should_run/T16449_2.stdout
- testsuite/tests/codeGen/should_run/all.T
Changes:
=====================================
compiler/coreSyn/CoreSyn.hs
=====================================
@@ -445,6 +445,9 @@ which will generate a @case@ if necessary
The let/app invariant is initially enforced by mkCoreLet and mkCoreApp in
coreSyn/MkCore.
+For discussion of some implications of the let/app invariant primops see
+Note [Checking versus non-checking primops] in PrimOp.
+
Note [CoreSyn type and coercion invariant]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We allow a /non-recursive/, /non-top-level/ let to bind type and
=====================================
compiler/prelude/PrelRules.hs
=====================================
@@ -475,8 +475,7 @@ shiftRule shift_op
-> return e1
-- See Note [Guarding against silly shifts]
| shift_len < 0 || shift_len > wordSizeInBits dflags
- -> return $ mkRuntimeErrorApp rUNTIME_ERROR_ID wordPrimTy
- ("Bad shift length " ++ show shift_len)
+ -> return $ Lit $ mkLitNumberWrap dflags LitNumInt 0 (exprType e1)
-- Do the shift at type Integer, but shift length is Int
Lit (LitNumber nt x t)
@@ -701,7 +700,27 @@ can't constant fold it, but if it gets to the assember we get
Error: operand type mismatch for `shl'
So the best thing to do is to rewrite the shift with a call to error,
-when the second arg is stupid.
+when the second arg is large. However, in general we cannot do this; consider
+this case
+
+ let x = I# (uncheckedIShiftL# n 80)
+ in ...
+
+Here x contains an invalid shift and consequently we would like to rewrite it
+as follows:
+
+ let x = I# (error "invalid shift)
+ in ...
+
+This was originally done in the fix to #16449 but this breaks the let/app
+invariant (see Note [CoreSyn let/app invariant] in CoreSyn) as noted in #16742.
+For the reasons discussed in Note [Checking versus non-checking primops] (in
+the PrimOp module) there is no safe way rewrite the argument of I# such that
+it bottoms.
+
+Consequently we instead take advantage of the fact that large shifts are
+undefined behavior (see associated documentation in primops.txt.pp) and
+transform the invalid shift into an "obviously incorrect" value.
There are two cases:
=====================================
compiler/prelude/PrimOp.hs
=====================================
@@ -304,6 +304,27 @@ primOpOutOfLine :: PrimOp -> Bool
* *
************************************************************************
+Note [Checking versus non-checking primops]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+ In GHC primops break down into two classes:
+
+ a. Checking primops behave, for instance, like division. In this
+ case the primop may throw an exception (e.g. division-by-zero)
+ and is consequently is marked with the can_fail flag described below.
+ The ability to fail comes at the expense of precluding some optimizations.
+
+ b. Non-checking primops behavior, for instance, like addition. While
+ addition can overflow it does not produce an exception. So can_fail is
+ set to False, and we get more optimisation opportunities. But we must
+ never throw an exception, so we cannot rewrite to a call to error.
+
+ It is important that a non-checking primop never be transformed in a way that
+ would cause it to bottom. Doing so would violate Core's let/app invariant
+ (see Note [CoreSyn let/app invariant] in CoreSyn) which is critical to
+ the simplifier's ability to float without fear of changing program meaning.
+
+
Note [PrimOp can_fail and has_side_effects]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Both can_fail and has_side_effects mean that the primop has
=====================================
testsuite/tests/codeGen/should_run/T16449_2.hs
=====================================
@@ -5,5 +5,9 @@ module Main where
import GHC.Prim
import GHC.Int
+-- Test that large unchecked shifts, which constitute undefined behavior, do
+-- not crash the compiler and instead evaluate to 0.
+-- See Note [Guarding against silly shifts] in PrelRules.
+
-- Shift should be larger than the word size (e.g. 64 on 64-bit) for this test.
main = print (I# (uncheckedIShiftL# 1# 1000#))
=====================================
testsuite/tests/codeGen/should_run/T16449_2.stderr deleted
=====================================
@@ -1 +0,0 @@
-T16449_2: Bad shift length 1000
=====================================
testsuite/tests/codeGen/should_run/T16449_2.stdout
=====================================
@@ -0,0 +1,2 @@
+0
+
=====================================
testsuite/tests/codeGen/should_run/all.T
=====================================
@@ -196,4 +196,4 @@ test('T15892',
extra_run_opts('+RTS -G1 -A32k -RTS') ],
compile_and_run, ['-O'])
test('T16617', normal, compile_and_run, [''])
-test('T16449_2', exit_code(1), compile_and_run, [''])
+test('T16449_2', exit_code(0), compile_and_run, [''])
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/commit/041960356df883d5883f973acefe44f15b678f4e
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/commit/041960356df883d5883f973acefe44f15b678f4e
You're receiving this email because of your account on gitlab.haskell.org.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20190611/657deb31/attachment-0001.html>
More information about the ghc-commits
mailing list