[llvm-dev] Rewriting LLVM IR intrinsic functions

[David Chisnall via llvm-dev]

On 14 Sep 2015, at 06:58, Dipanjan Das via llvm-dev <llvm-dev at lists.llvm.org> wrote:
> 
> 
> Hi David,
> 
> I think you people can help me better if I explain what I am trying to achieve.
> 	• I have compiled gmp with LLVM 2.9
> 	• Linked the library statically with a stub program (gmp_demo.bc)
> 	• Transformed the bitcode to LLVM IR: llvm-dis gmp_demo.bc -o gmp_demo.ll
> 	• Trying to convert the IR to high-level C code: llc -march=c gmp_demo.ll 
> Though I know that llc C backend has many issues and has been removed from LLVM >=3.1, still I am desperate to make it work. The error I receive in the last step is the following: "LLVM ERROR: Code generator does not support intrinsic function 'llvm.uadd.with.overflow.i64'!"
> 
> I was planning to rewrite LLVM intrinsics not supported by llc to equivalent LLVM IR supported by llc. Is this approach feasible?

It would still help to understand exactly why you’re trying to do this.  It sounds as if you’re using LLVM as a source-to-source translator, with C as the input and output language.  If this is the case, then the Clang rewriter infrastructure might be more useful to you.

The intrinsic that it is currently failing on is (as its name implies) an overflow-checked addition.  The reason that the C backed does not understand this builtin is simple: There is no way of (directly) expressing this in C, without extensions.  

You could potentially expand this in IR to a signed addition and a check that the sign bit had not changed, but it would be cleaner to simply modify the C backend to emit the Clang / GCC __builtin function that corresponds to the intrinsic.  You’re likely to find a number of other intrinsics that are not supported: the reason that the C backend was removed was that it never worked very well.  For example, anything involving exceptions can’t be implemented in C, so the C backend lowers it to setjmp / longjmp, which gives you a different ABI.

David