[llvm-dev] Source level code transformation V.S. IR-level code transformation

[Linchuan Chen via llvm-dev]

Dear all,
    I'm working on a simple code transformation problem that can be
described as below:

    for a C++ loop:

    *for (int i = 0; i < N; ++i) {*
*        a = M[i] + b;*
*    }*

    I want to transform it to:

    *int A[4];    *

*    for (int i = 0; i < N; ++i) {*
*        A[0] = M[i] + b;*
*        A[1] = M[i] + b;*
*        A[2] = M[i] + b;*
*        A[3] = M[i] + b;*
*    }*

   The reason I want to do it is to transform my code to a form that SLP
vectorizer is able to vectorize it.

   I know there are a few approaches for transforming the code, the
simplest being a naive text processor for duplicating the assignment
instructions in the loop, but I don't think it a standard way for code
transformation.

   I also got to know a few other alternatives:

   1. using libTooling to edit the AST
   2. transform the IR code of the loop, in a PASS, so that the SLP PASS
   can consume the resulting IR

   Does anyone know which alternative is the most used or most standard way
of doing such a transformation work? Thanks!
-- 
Sincerely,
Linchuan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180114/b850231e/attachment.html>