dead code elimination about llvm-mctoll HOT 3 OPEN

zext 0 also seems to be common:

%1 = zext i32 0 to i64
store i64 %1, i64* %0

from llvm-mctoll.

Indeed the code generated by llvm-mctoll, at present, is not optimal. No effort is yet expended in making raised LLVM IR optimal. One train of thought - for not yet running any optimization passes on the Module Functions - is that the raised output is expected to be consumed and consequently optimized by the clang/llvm to generate native code (potentially for a non-x64 target). The thought is that any optimizations, such as DCE, would be part of this compilation process and is not (yet?) critical for the raised code.

In situations where the raised IR is not intended to be compiled to a native target, I'd expect that one could use LLVM opt tool to run LLVM passes as desired to get a more optimized version.

from llvm-mctoll.

True, that also occurred to me. But you gotta be careful, opt -S -O3 easily optimizes away everything 😄
Even opt -S -instcombine -dce already removes quite some code. Could maybe be added to the readme.
Btw, maybe there could be a mctoll mode that outputs to stdout so it can be directly piped into opt.

from llvm-mctoll.