[LLVMbugs] [Bug 7518] New: terrible codegen of <2 x float>

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=7518

           Summary: terrible codegen of <2 x float>
           Product: libraries
           Version: trunk
          Platform: PC
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: clattner at apple.com
                CC: llvmbugs at cs.uiuc.edu


The X86 backend is generating awful code for 2x float.  This is an important
type (which should use the low 8 bytes of an XMM register) because they x86-64
ABI can be most efficiently expressed with 2 x float for structs that contain
multiple floats.  Something like  "struct {float a,b,c,d;}" end up being passed
as two <2x float>.  Anyway, this:


define void @_Z3barR6Floats(<2 x float> %Q, float *%P2) nounwind {
  %a = extractelement <2 x float> %Q, i32 0
  %b = extractelement <2 x float> %Q, i32 1
  %c = fadd float %a, %b

  store float %c, float* %P2
  ret void
}

Compiles to:


__Z3barR6Floats:                        ## @_Z3barR6Floats
## BB#0:
    movq    %xmm0, -16(%rsp)
    movq    %xmm0, -8(%rsp)
    movss    -16(%rsp), %xmm0
    addss    -4(%rsp), %xmm0
    movss    %xmm0, (%rdi)
    ret

But this (which is the same semantics):

define void @_Z3barR6Floats(<4 x float> %Q, float *%P2) nounwind {
  %a = extractelement <4 x float> %Q, i32 0
  %b = extractelement <4 x float> %Q, i32 1
  %c = fadd float %a, %b

  store float %c, float* %P2
  ret void
}

compiles to:


__Z3barR6Floats:                        ## @_Z3barR6Floats
## BB#0:
    pshufd    $1, %xmm0, %xmm1
    addss    %xmm0, %xmm1
    movss    %xmm1, (%rdi)
    ret

Hopefully this is a simple type legalizer issue?

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.