llvm.org GIT mirror llvm / a8f5cd3
Fix a crash in which a multiplication was being reported as being both negative and positive: positive, because it could be directly computed to be positive; negative, because the nsw flags means it is either negative or undefined (the multiplication always overflowed). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145104 91177308-0d34-0410-b5e6-96231b3b80d8 Duncan Sands 7 years ago
2 changed file(s) with 24 addition(s) and 2 deletion(s). Raw diff Collapse all Expand all
247247 APInt::getHighBitsSet(BitWidth, LeadZ);
248248 KnownZero &= Mask;
249249
250 if (isKnownNonNegative)
250 // Only make use of no-wrap flags if we failed to compute the sign bit
251 // directly. This matters if the multiplication always overflows, in
252 // which case we prefer to follow the result of the direct computation,
253 // though as the program is invoking undefined behaviour we can choose
254 // whatever we like here.
255 if (isKnownNonNegative && !KnownOne.isNegative())
251256 KnownZero.setBit(BitWidth - 1);
252 else if (isKnownNegative)
257 else if (isKnownNegative && !KnownZero.isNegative())
253258 KnownOne.setBit(BitWidth - 1);
254259
255260 return;
0 ; RUN: opt < %s -instsimplify
1
2 ; The mul can be proved to always overflow (turning a negative value
3 ; into a positive one) and thus results in undefined behaviour. At
4 ; the same time we were deducing from the nsw flag that that mul could
5 ; be assumed to have a negative value (since if not it has an undefined
6 ; value, which can be taken to be negative). We were reporting the mul
7 ; as being both positive and negative, firing an assertion!
8 define i1 @test1(i32 %a) {
9 entry:
10 %0 = or i32 %a, 1
11 %1 = shl i32 %0, 31
12 %2 = mul nsw i32 %1, 4
13 %3 = and i32 %2, -4
14 %4 = icmp ne i32 %3, 0
15 ret i1 %4
16 }