1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-apple-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86
3; RUN: llc < %s -mtriple=x86_64-apple-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64
4
5; Verify that we are using the efficient uitofp --> sitofp lowering illustrated
6; by the compiler_rt implementation of __floatundisf.
7; <rdar://problem/8493982>
8
9define float @test(i64 %a) nounwind {
10; X86-LABEL: test:
11; X86:       # %bb.0: # %entry
12; X86-NEXT:    pushl %ebp
13; X86-NEXT:    movl %esp, %ebp
14; X86-NEXT:    andl $-8, %esp
15; X86-NEXT:    subl $16, %esp
16; X86-NEXT:    movl 12(%ebp), %eax
17; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
18; X86-NEXT:    movlps %xmm0, {{[0-9]+}}(%esp)
19; X86-NEXT:    shrl $31, %eax
20; X86-NEXT:    fildll {{[0-9]+}}(%esp)
21; X86-NEXT:    fadds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)
22; X86-NEXT:    fstps {{[0-9]+}}(%esp)
23; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
24; X86-NEXT:    movss %xmm0, (%esp)
25; X86-NEXT:    flds (%esp)
26; X86-NEXT:    movl %ebp, %esp
27; X86-NEXT:    popl %ebp
28; X86-NEXT:    retl
29;
30; X64-LABEL: test:
31; X64:       # %bb.0: # %entry
32; X64-NEXT:    testq %rdi, %rdi
33; X64-NEXT:    js .LBB0_1
34; X64-NEXT:  # %bb.2: # %entry
35; X64-NEXT:    cvtsi2ss %rdi, %xmm0
36; X64-NEXT:    retq
37; X64-NEXT:  .LBB0_1:
38; X64-NEXT:    movq %rdi, %rax
39; X64-NEXT:    shrq %rax
40; X64-NEXT:    andl $1, %edi
41; X64-NEXT:    orq %rax, %rdi
42; X64-NEXT:    cvtsi2ss %rdi, %xmm0
43; X64-NEXT:    addss %xmm0, %xmm0
44; X64-NEXT:    retq
45entry:
46  %b = uitofp i64 %a to float
47  ret float %b
48}
49