summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Analysis/InstructionSimplify.cpp
diff options
context:
space:
mode:
authorRoman Lebedev <lebedev.ri@gmail.com>2019-09-08 20:14:15 +0000
committerRoman Lebedev <lebedev.ri@gmail.com>2019-09-08 20:14:15 +0000
commit6e2c5c87103d76e03cd6f1c5d297bddcad4cf7e1 (patch)
tree27827da05e5634e37a0e4c842f9142d5e8dfb1cf /llvm/lib/Analysis/InstructionSimplify.cpp
parent354a46444ce92be73f3859d93a8c6fb237b9e7eb (diff)
downloadbcm5719-llvm-6e2c5c87103d76e03cd6f1c5d297bddcad4cf7e1.tar.gz
bcm5719-llvm-6e2c5c87103d76e03cd6f1c5d297bddcad4cf7e1.zip
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246)
Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
Diffstat (limited to 'llvm/lib/Analysis/InstructionSimplify.cpp')
-rw-r--r--llvm/lib/Analysis/InstructionSimplify.cpp33
1 files changed, 24 insertions, 9 deletions
diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index df87b941039..7960635d009 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -1371,7 +1371,8 @@ Value *llvm::SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact,
/// Commuted variants are assumed to be handled by calling this function again
/// with the parameters swapped.
static Value *simplifyUnsignedRangeCheck(ICmpInst *ZeroICmp,
- ICmpInst *UnsignedICmp, bool IsAnd) {
+ ICmpInst *UnsignedICmp, bool IsAnd,
+ const DataLayout &DL) {
Value *X, *Y;
ICmpInst::Predicate EqPred;
@@ -1395,6 +1396,18 @@ static Value *simplifyUnsignedRangeCheck(ICmpInst *ZeroICmp,
if (UnsignedPred == ICmpInst::ICMP_ULT && EqPred == ICmpInst::ICMP_NE)
return IsAnd ? UnsignedICmp : ZeroICmp;
+ // X <= Y && Y != 0 --> X <= Y iff X != 0
+ // X <= Y || Y != 0 --> Y != 0 iff X != 0
+ if (UnsignedPred == ICmpInst::ICMP_ULE && EqPred == ICmpInst::ICMP_NE &&
+ isKnownNonZero(X, DL))
+ return IsAnd ? UnsignedICmp : ZeroICmp;
+
+ // X > Y && Y == 0 --> Y == 0 iff X != 0
+ // X > Y || Y == 0 --> X > Y iff X != 0
+ if (UnsignedPred == ICmpInst::ICMP_UGT && EqPred == ICmpInst::ICMP_EQ &&
+ isKnownNonZero(X, DL))
+ return IsAnd ? ZeroICmp : UnsignedICmp;
+
// X >= Y || Y != 0 --> true
// X >= Y || Y == 0 --> X >= Y
if (UnsignedPred == ICmpInst::ICMP_UGE && !IsAnd) {
@@ -1587,10 +1600,11 @@ static Value *simplifyAndOfICmpsWithAdd(ICmpInst *Op0, ICmpInst *Op1,
}
static Value *simplifyAndOfICmps(ICmpInst *Op0, ICmpInst *Op1,
- const InstrInfoQuery &IIQ) {
- if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/true))
+ const InstrInfoQuery &IIQ,
+ const DataLayout &DL) {
+ if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/true, DL))
return X;
- if (Value *X = simplifyUnsignedRangeCheck(Op1, Op0, /*IsAnd=*/true))
+ if (Value *X = simplifyUnsignedRangeCheck(Op1, Op0, /*IsAnd=*/true, DL))
return X;
if (Value *X = simplifyAndOfICmpsWithSameOperands(Op0, Op1))
@@ -1660,10 +1674,11 @@ static Value *simplifyOrOfICmpsWithAdd(ICmpInst *Op0, ICmpInst *Op1,
}
static Value *simplifyOrOfICmps(ICmpInst *Op0, ICmpInst *Op1,
- const InstrInfoQuery &IIQ) {
- if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/false))
+ const InstrInfoQuery &IIQ,
+ const DataLayout &DL) {
+ if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/false, DL))
return X;
- if (Value *X = simplifyUnsignedRangeCheck(Op1, Op0, /*IsAnd=*/false))
+ if (Value *X = simplifyUnsignedRangeCheck(Op1, Op0, /*IsAnd=*/false, DL))
return X;
if (Value *X = simplifyOrOfICmpsWithSameOperands(Op0, Op1))
@@ -1738,8 +1753,8 @@ static Value *simplifyAndOrOfCmps(const SimplifyQuery &Q,
auto *ICmp0 = dyn_cast<ICmpInst>(Op0);
auto *ICmp1 = dyn_cast<ICmpInst>(Op1);
if (ICmp0 && ICmp1)
- V = IsAnd ? simplifyAndOfICmps(ICmp0, ICmp1, Q.IIQ)
- : simplifyOrOfICmps(ICmp0, ICmp1, Q.IIQ);
+ V = IsAnd ? simplifyAndOfICmps(ICmp0, ICmp1, Q.IIQ, Q.DL)
+ : simplifyOrOfICmps(ICmp0, ICmp1, Q.IIQ, Q.DL);
auto *FCmp0 = dyn_cast<FCmpInst>(Op0);
auto *FCmp1 = dyn_cast<FCmpInst>(Op1);
OpenPOWER on IntegriCloud