xref: /llvm-project/llvm/test/CodeGen/X86/uint64-to-float.ll (revision 11fb09ec0afa8620bba44f374763e2ad64332e2e)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86
3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64
4; RUN: llc < %s -mtriple=i686-windows -mattr=+sse2 | FileCheck %s --check-prefix=X86-WIN
5; RUN: llc < %s -mtriple=x86_64-windows -mattr=+sse2 | FileCheck %s --check-prefix=X64-WIN
6
7; Verify that we are using the efficient uitofp --> sitofp lowering illustrated
8; by the compiler_rt implementation of __floatundisf.
9; <rdar://problem/8493982>
10
11define float @test(i64 %a) nounwind {
12; X86-LABEL: test:
13; X86:       # %bb.0: # %entry
14; X86-NEXT:    pushl %ebp
15; X86-NEXT:    movl %esp, %ebp
16; X86-NEXT:    andl $-8, %esp
17; X86-NEXT:    subl $16, %esp
18; X86-NEXT:    movl 12(%ebp), %eax
19; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
20; X86-NEXT:    movlps %xmm0, {{[0-9]+}}(%esp)
21; X86-NEXT:    shrl $31, %eax
22; X86-NEXT:    fildll {{[0-9]+}}(%esp)
23; X86-NEXT:    fadds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)
24; X86-NEXT:    fstps {{[0-9]+}}(%esp)
25; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
26; X86-NEXT:    movss %xmm0, (%esp)
27; X86-NEXT:    flds (%esp)
28; X86-NEXT:    movl %ebp, %esp
29; X86-NEXT:    popl %ebp
30; X86-NEXT:    retl
31;
32; X64-LABEL: test:
33; X64:       # %bb.0: # %entry
34; X64-NEXT:    testq %rdi, %rdi
35; X64-NEXT:    js .LBB0_1
36; X64-NEXT:  # %bb.2: # %entry
37; X64-NEXT:    cvtsi2ss %rdi, %xmm0
38; X64-NEXT:    retq
39; X64-NEXT:  .LBB0_1:
40; X64-NEXT:    movq %rdi, %rax
41; X64-NEXT:    shrq %rax
42; X64-NEXT:    andl $1, %edi
43; X64-NEXT:    orq %rax, %rdi
44; X64-NEXT:    cvtsi2ss %rdi, %xmm0
45; X64-NEXT:    addss %xmm0, %xmm0
46; X64-NEXT:    retq
47;
48; X86-WIN-LABEL: test:
49; X86-WIN:       # %bb.0: # %entry
50; X86-WIN-NEXT:    pushl %ebp
51; X86-WIN-NEXT:    movl %esp, %ebp
52; X86-WIN-NEXT:    andl $-8, %esp
53; X86-WIN-NEXT:    subl $24, %esp
54; X86-WIN-NEXT:    movl 12(%ebp), %eax
55; X86-WIN-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
56; X86-WIN-NEXT:    movlps %xmm0, {{[0-9]+}}(%esp)
57; X86-WIN-NEXT:    shrl $31, %eax
58; X86-WIN-NEXT:    fildll {{[0-9]+}}(%esp)
59; X86-WIN-NEXT:    fnstcw {{[0-9]+}}(%esp)
60; X86-WIN-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
61; X86-WIN-NEXT:    orl $768, %ecx # imm = 0x300
62; X86-WIN-NEXT:    movw %cx, {{[0-9]+}}(%esp)
63; X86-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
64; X86-WIN-NEXT:    fadds __real@5f80000000000000(,%eax,4)
65; X86-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
66; X86-WIN-NEXT:    fstps {{[0-9]+}}(%esp)
67; X86-WIN-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
68; X86-WIN-NEXT:    movss %xmm0, {{[0-9]+}}(%esp)
69; X86-WIN-NEXT:    flds {{[0-9]+}}(%esp)
70; X86-WIN-NEXT:    movl %ebp, %esp
71; X86-WIN-NEXT:    popl %ebp
72; X86-WIN-NEXT:    retl
73;
74; X64-WIN-LABEL: test:
75; X64-WIN:       # %bb.0: # %entry
76; X64-WIN-NEXT:    testq %rcx, %rcx
77; X64-WIN-NEXT:    js .LBB0_1
78; X64-WIN-NEXT:  # %bb.2: # %entry
79; X64-WIN-NEXT:    cvtsi2ss %rcx, %xmm0
80; X64-WIN-NEXT:    retq
81; X64-WIN-NEXT:  .LBB0_1:
82; X64-WIN-NEXT:    movq %rcx, %rax
83; X64-WIN-NEXT:    shrq %rax
84; X64-WIN-NEXT:    andl $1, %ecx
85; X64-WIN-NEXT:    orq %rax, %rcx
86; X64-WIN-NEXT:    cvtsi2ss %rcx, %xmm0
87; X64-WIN-NEXT:    addss %xmm0, %xmm0
88; X64-WIN-NEXT:    retq
89entry:
90  %b = uitofp i64 %a to float
91  ret float %b
92}
93