[llvm-bugs] [Bug 37763] New: [X86] Investigate vectorization of the overflow add/sub nodes to PADD+PADDS+PCMPEQ etc.

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=37763

            Bug ID: 37763
           Summary: [X86] Investigate vectorization of the overflow
                    add/sub nodes to PADD+PADDS+PCMPEQ etc.
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: craig.topper at gmail.com, deadalnix at gmail.com,
                    lebedev.ri at gmail.com, llvm-bugs at lists.llvm.org,
                    spatel+llvm at rotateright.com

We may be able to efficiently vectorize some SADDO/UADDO (etc.) nodes (vXi8 and
vXi16) by performing both a PADDW and PADDSW (etc.) and comparing the results.

e.g.

#include <x86intrin.h>

struct Res {
  __v8hi value;
  __v8hi overflow;
};

struct Res saddo16(__v8hi x, __v8hi y) {
    struct Res r;
    r.value = _mm_add_epi16(x, y);
    r.overflow = (r.value != (__v8hi)_mm_adds_epi16(x, y));
    return r;
}

https://godbolt.org/g/4xkauv

-- 
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/20180610/5d0d4825/attachment.html>