• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -march=x86 -mcpu=generic -mattr=+sse < %s | FileCheck %s
2; RUN: llc -march=x86 -mcpu=atom -mattr=+sse < %s | FileCheck -check-prefix=ATOM %s
3
4%vec = type <6 x float>
5; CHECK: divps
6; CHECK: divss
7; CHECK: divss
8
9; Scheduler causes a different instruction order to be produced on Intel Atom
10; ATOM: divps
11; ATOM: divss
12; ATOM: divss
13
14define %vec @vecdiv( %vec %p1, %vec %p2)
15{
16  %result = fdiv %vec %p1, %p2
17  ret %vec %result
18}
19
20@a = constant %vec < float 2.0, float 4.0, float 8.0, float 16.0, float 32.0, float 64.0 >
21@b = constant %vec < float 2.0, float 2.0, float 2.0, float 2.0, float 2.0, float 2.0 >
22
23; Expected result: < 1.0, 2.0, 4.0, ..., 2.0^(n-1) >
24; main() returns 0 if the result is expected and 1 otherwise
25; to execute, use llvm-as < %s | lli
26define i32 @main() nounwind {
27entry:
28  %avec = load %vec* @a
29  %bvec = load %vec* @b
30
31  %res = call %vec @vecdiv(%vec %avec, %vec %bvec)
32  br label %loop
33loop:
34  %idx = phi i32 [0, %entry], [%nextInd, %looptail]
35  %expected = phi float [1.0, %entry], [%nextExpected, %looptail]
36  %elem = extractelement %vec %res, i32 %idx
37  %expcmp = fcmp oeq float %elem, %expected
38  br i1 %expcmp, label %looptail, label %return
39looptail:
40  %nextExpected = fmul float %expected, 2.0
41  %nextInd = add i32 %idx, 1
42  %cmp = icmp slt i32 %nextInd, 6
43  br i1 %cmp, label %loop, label %return
44return:
45  %retval = phi i32 [0, %looptail], [1, %loop]
46  ret i32 %retval
47}
48