[llvm-bugs] [Bug 34382] New: [X86] Not taking advantage of the permute-in-lane instruction (vpermilps)

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=34382

            Bug ID: 34382
           Summary: [X86] Not taking advantage of the permute-in-lane
                    instruction (vpermilps)
           Product: libraries
           Version: trunk
          Hardware: All
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: ayman.musa at intel.com
                CC: llvm-bugs at lists.llvm.org

define <16 x float> @test_16xfloat_perm_mask0(<16 x float> %vec) {
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32
1, i32 1, i32 3, i32 0, i32 6, i32 4, i32 5, i32 7, i32 8, i32 8, i32 9, i32 9,
i32 15, i32 14, i32 14, i32 12>
  ret <16 x float> %res 
}

LLVM emits (showing 2.86 throughput on IACA tool):
    vmovaps .LCPI142_0(%rip), %zmm1 # zmm1 =
[1,1,3,0,6,4,5,7,8,8,9,9,15,14,14,12]
    vpermps %zmm0, %zmm1, %zmm0

While it could have emitted (showing 1.00 throughput on IACA tool):
    vpermilps .LCPI142_0(%rip), %zmm0, %zmm0

* LCPI142_0 holds the needed indexes for each permute instruction.
** Throughput results from IACA tool => lower is better.

-- 
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/20170830/7f9a0f66/attachment.html>