[PATCH] D12535: [PR24643] Speeding up FoldingSetNodeID::AddPointer

[Ben Craig via llvm-commits]

bcraig added a comment.

I can confirm that on my machine, the addresses that were typically passed to AddPointer were "small".  They tended to fit in 32-bits.  Stack addresses tended to be large.

I will also say that the generated assembly on Linux for my patch is significantly larger / more complicated than the original code.  Despite the increased complexity, it does seem to run faster.  There's also the substantial benefits that I get on Windows by avoiding the memmove.


http://reviews.llvm.org/D12535