Currently llvm crashes on following code, run through llc: declare void @llvm.experimental.stackmap(i64, i32, ...) define void @foo() { tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 0, i32 0, i64 9223372036854775807) ret void } The issue is that 9223372036854775807 (decimal for 0x7fffffffffffffff) is the "empty key" for an int64_t, and in StackMaps::recordStackMapOpers we crash when we try to insert this as a key into ConstPool. The this happens if we change the constant to be the tombstone. Two potential fixes I can think of: 1. have some special logic to remember the offsets for the tombstone and empty i64 constants. This can easily be tracked using two "Optional<int>" fields. 2. change ConstantPool to be use a std::map instead of a llvm::DenseMap as the map in the MapVector. An aside: the same function has this check "((I->Offset + (int64_t(1)<<31)) >> 32)" -- won't this cause a signed overflow (and hence UB) if I->Offset is negative? -- Sanjoy
Apologies, please ignore the aside:> An aside: the same function has this check "((I->Offset + > (int64_t(1)<<31)) >> 32)" -- won't this cause a signed overflow (and > hence UB) if I->Offset is negative?The overflow won't be a *signed overflow*, so we're okay. -- Sanjoy
We also could make the DenseMap just an uint64_t. In this case the empty key and the tombstone would be never encoded into the constant pool, because they can be encoded in the offset field itself.> On Oct 31, 2014, at 11:20 AM, Sanjoy Das <sanjoy at playingwithpointers.com> wrote: > > Currently llvm crashes on following code, run through llc: > > declare void @llvm.experimental.stackmap(i64, i32, ...) > > define void @foo() { > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 0, > i32 0, i64 9223372036854775807) > ret void > } > > The issue is that 9223372036854775807 (decimal for 0x7fffffffffffffff) > is the "empty key" for an int64_t, and in > StackMaps::recordStackMapOpers we crash when we try to insert this as > a key into ConstPool. The this happens if we change the constant to > be the tombstone. > > Two potential fixes I can think of: > > 1. have some special logic to remember the offsets for the tombstone > and empty i64 constants. This can easily be tracked using two > "Optional<int>" fields. > > 2. change ConstantPool to be use a std::map instead of a > llvm::DenseMap as the map in the MapVector. > > An aside: the same function has this check "((I->Offset + > (int64_t(1)<<31)) >> 32)" -- won't this cause a signed overflow (and > hence UB) if I->Offset is negative? >I guess "!isInt<32>(I->Offset)” might be a more readable solution.> -- Sanjoy
> We also could make the DenseMap just an uint64_t. In this case the empty key and the tombstone would be never encoded into the constant pool, because they can be encoded in the offset field itself.That's a great idea, I'll send in a patch soon. -- Sanjoy
Reasonably Related Threads
- [LLVMdev] [RFC] Stackmap and Patchpoint Intrinsic Proposal
- [LLVMdev] [RFC] Stackmap and Patchpoint Intrinsic Proposal
- [LLVMdev] [RFC] Stackmap and Patchpoint Intrinsic Proposal
- [LLVMdev] [RFC] Stackmap and Patchpoint Intrinsic Proposal
- [LLVMdev] [RFC] Stackmap and Patchpoint Intrinsic Proposal