[llvm-bugs] [Bug 34111] New: Compiler no longer optimizing a vector add+shuffle to identity to a horizontal add

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=34111

            Bug ID: 34111
           Summary: Compiler no longer optimizing a vector add+shuffle to
                    identity to a horizontal add
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: douglas_yung at playstation.sony.com
                CC: llvm-bugs at lists.llvm.org

It seems commit r309812 has broken the compiler's ability to optimize a vector
add+shuffle to identity to a simple horizontal add instruction. Consider the
following code:

__m128 add_ps_001(__m128 a, __m128 b) {
  __m128 r = (__m128){ a[0] + a[1], a[2] + a[3], b[0] + b[1], b[2] + b[3] };
  return __builtin_shufflevector(r, a, -1, -1, 2, 3);
}

Prior to commit r309812, when targeting x86+avx with optimizations (clang -S
-O2 -mavx), the compiler would generate the following code:

        .cfi_startproc
# BB#0:                                 # %entry
        vhaddps %xmm1, %xmm0, %xmm0
        retq

But starting with commit r309812, the compiler is now generating less efficient
code:

        .cfi_startproc
# BB#0:                                 # %entry
        vpermilps       $232, %xmm1, %xmm0 # xmm0 = xmm1[0,2,2,3]
        vpermilps       $237, %xmm1, %xmm1 # xmm1 = xmm1[1,3,2,3]
        vaddps  %xmm1, %xmm0, %xmm0
        vmovddup        %xmm0, %xmm0    # xmm0 = xmm0[0,0]
        retq

-- 
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/20170808/0fe2c1e7/attachment.html>