[llvm-bugs] [Bug 37849] New: Missed optimization: invariant code motion

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=37849

            Bug ID: 37849
           Summary: Missed optimization: invariant code motion
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: m-takahiro at jp.fujitsu.com
                CC: llvm-bugs at lists.llvm.org

I compiled the following code at -O3 with loop unrolling disabled to make it
easy to see generated IRs.

---------------------
void foo(float * restrict a, float * restrict b, const int N) {
  for (int i = 1; i < N; ++i)
    a[i] = a[0] + b[i];
}
---------------------

Clang version is as follows:

---------------------
$ clang --version
clang version 7.0.0 (trunk 335036)
---------------------


And generated IRs are as follows:

---------------------
; <label>:8:                                      ; pre
  %9 = phi i64 [ 1, %5 ], [ %15, %8 ]
  %10 = load float, float* %0, align 4, !tbaa !2
  %11 = getelementptr inbounds float, float* %1, i64 %9
  %12 = load float, float* %11, align 4, !tbaa !2
  %13 = fadd float %10, %12
  %14 = getelementptr inbounds float, float* %0, i64 %9
  store float %13, float* %14, align 4, !tbaa !2
  %15 = add nuw nsw i64 %9, 1
  %16 = icmp eq i64 %15, %6
  br i1 %16, label %7, label %8
---------------------

'a[0]' (%10) is invariant inside the loop because an initial value of induction
variable is 1 and its value is incremented by 1. So, I think invariant code
motion can be applied.

-- 
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/20180619/4f2791f5/attachment.html>