https://github.com/AccelerateHS/accelerate/issues/409

map_1bf1a791bb25369d.ll

; ModuleID = 'map_1bf1a791bb25369d'
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"

; Function Attrs: nounwind
define void @map_1bf1a791bb25369d(i64 %ix.start, i64 %ix.end, double* noalias nocapture %out.ad0, double* noalias nocapture readonly %fv0.ad0) local_unnamed_addr #0 {
entry:
  %0 = icmp slt i64 %ix.start, %ix.end
  br i1 %0, label %while1.top.preheader, label %while1.exit

while1.top.preheader:                             ; preds = %entry
  br label %while1.top

while1.top:                                       ; preds = %while1.top.preheader, %while1.top
  %1 = phi i64 [ %10, %while1.top ], [ %ix.start, %while1.top.preheader ]
  %2 = getelementptr double, double* %fv0.ad0, i64 %1
  %3 = load double, double* %2, align 8
  %4 = fdiv fast double 1.000000e+00, %3
  %5 = fmul fast double %4, %4
  %6 = fsub fast double -0.000000e+00, %5
  %7 = tail call double @tanh(double %6) #1
  %8 = fadd fast double %4, %7
  %9 = getelementptr double, double* %out.ad0, i64 %1
  store double %8, double* %9, align 8
  %10 = add nsw i64 %1, 1
  %exitcond = icmp eq i64 %10, %ix.end
  br i1 %exitcond, label %while1.exit, label %while1.top

while1.exit:                                      ; preds = %while1.top, %entry
  ret void
}

; Function Attrs: nounwind readonly
declare double @tanh(double) local_unnamed_addr #1

attributes #0 = { nounwind }
attributes #1 = { nounwind readonly }

map_1bf1a791bb25369d.s

        .text
        .file   ""
        .section        .rodata.cst8,"aM",@progbits,8
        .p2align        3
.LCPI0_0:
        .quad   4607182418800017408
.LCPI0_1:
        .quad   0
        .text
        .globl  map_1bf1a791bb25369d
        .p2align        4, 0x90
        .type   map_1bf1a791bb25369d,@function
map_1bf1a791bb25369d:
        pushq   %r15
        pushq   %r14
        pushq   %rbx
        subq    $16, %rsp
        movq    %rsi, %r14
        cmpq    %r14, %rdi
        jge     .LBB0_3
        subq    %rdi, %r14
        leaq    (%rdx,%rdi,8), %r15
        leaq    (%rcx,%rdi,8), %rbx
        .p2align        4, 0x90
.LBB0_2:
        vmovsd  .LCPI0_0(%rip), %xmm0
        vdivsd  (%rbx), %xmm0, %xmm0
        vmovsd  %xmm0, 8(%rsp)
        vfnmsub213sd    .LCPI0_1, %xmm0, %xmm0
        callq   tanh
        vaddsd  8(%rsp), %xmm0, %xmm0
        vmovsd  %xmm0, (%r15)
        addq    $8, %r15
        addq    $8, %rbx
        addq    $-1, %r14
        jne     .LBB0_2
.LBB0_3:
        addq    $16, %rsp
        popq    %rbx
        popq    %r14
        popq    %r15
        retq
.Lfunc_end0:
        .size   map_1bf1a791bb25369d, .Lfunc_end0-map_1bf1a791bb25369d


        .section        ".note.GNU-stack","",@progbits
        

$ llc-5.0 --version
LLVM (http://llvm.org/):
  LLVM version 5.0.0

  Optimized build.
  Default target: x86_64-pc-linux-gnu
  Host CPU: broadwell

  Registered Targets:
    [..]