[llvm-bugs] [Bug 26773] New: SLP should recognise return as entry point for vectorisation

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=26773

            Bug ID: 26773
           Summary: SLP should recognise return as entry point for
                    vectorisation
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: renato.golin at linaro.org
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Created attachment 15957
  --> https://llvm.org/bugs/attachment.cgi?id=15957&action=edit
Multiplication operator example

As described in #21778, small loops get unrolled early on, which prevents the
loop vectoriser to optimise some small loops with a single call to a SIMD
intrinsic, like the SLP vectorizer does.

The attached code has the two test user functions vectorised by the SLP, but
not the out-of-line copy of the operator*.

I believe the SLP should recognise "return" as an entry point, just as valid as
a store, since functions will be called with either the values in registers on
in the stack, which can utilise the target's PCS standard to connect arguments
in a seamless fashion and avoid *any* unnecessary vector-load/store, or at
least detect the local pattern and have only one vector-load/store, instead of
using the back-end to merge loads/stores.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160229/c0286075/attachment.html>