[llvm-bugs] [Bug 49616] New: Failure to remove before/after reverse shuffles from loop with decrementing iterator

via llvm-bugs llvm-bugs at lists.llvm.org
Wed Mar 17 09:07:12 PDT 2021


https://bugs.llvm.org/show_bug.cgi?id=49616

            Bug ID: 49616
           Summary: Failure to remove before/after reverse shuffles from
                    loop with decrementing iterator
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: anton.a.afanasyev at gmail.com, florian_hahn at apple.com,
                    llvm-bugs at lists.llvm.org, spatel+llvm at rotateright.com

https://godbolt.org/z/dWTd9e

void loop_dec(unsigned short *p, int max, int n) {
    for (int i = 0; i < (n & 31); i++) {
        unsigned m = *--p;
        *p = (unsigned short)(m >= max ? m-max : 0);
    }
}

By decrementing the pointer in the loop we end up with loop bodies like this:

52:
  %53 = phi i64 [ %41, %40 ], [ %64, %52 ]
  %54 = sub i64 0, %53
  %55 = getelementptr inbounds i16, i16* %51, i64 %54
  %56 = bitcast i16* %55 to <4 x i16>*
  %57 = load <4 x i16>, <4 x i16>* %56, align 2, !tbaa !2
  %58 = shufflevector <4 x i16> %57, <4 x i16> poison, <4 x i32> <i32 3, i32 2,
i32 1, i32 0>
  %59 = zext <4 x i16> %58 to <4 x i32>
  %60 = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> %59, <4 x i32> %50)
  %61 = trunc <4 x i32> %60 to <4 x i16>
  %62 = shufflevector <4 x i16> %61, <4 x i16> poison, <4 x i32> <i32 3, i32 2,
i32 1, i32 0>
  %63 = bitcast i16* %55 to <4 x i16>*
  store <4 x i16> %62, <4 x i16>* %63, align 2, !tbaa !2
  %64 = add i64 %53, 4
  %65 = icmp eq i64 %64, %45
  br i1 %65, label %66, label %52, !llvm.loop !9 

AFAICT we should be able to remove both these 'reverse' shufflevectors.

I'm not sure if we should be trying to fix this in InstCombine/VectorCombine or
inside the LoopVectorizer.

-- 
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/20210317/9fd4a5ce/attachment.html>


More information about the llvm-bugs mailing list