1; RUN: llc < %s -mtriple=x86_64-linux -mcpu=nehalem | FileCheck %s
2; RUN: llc < %s -mtriple=x86_64-win32 -mcpu=nehalem | FileCheck %s
3
4; Full strength reduction wouldn't reduce register pressure, so LSR should
5; stick with indexing here.
6
7; CHECK: movaps        (%{{rsi|rdx}},%rax,4), [[X3:%xmm[0-9]+]]
8; CHECK: cvtdq2ps
9; CHECK: orps          {{%xmm[0-9]+}}, [[X4:%xmm[0-9]+]]
10; CHECK: movaps        [[X4]], (%{{rdi|rcx}},%rax,4)
11; CHECK: addq  $4, %rax
12; CHECK: cmpl  %eax, (%{{rdx|r8}})
13; CHECK-NEXT: jg
14
15define void @vvfloorf(float* nocapture %y, float* nocapture %x, i32* nocapture %n) nounwind {
16entry:
17  %0 = load i32, i32* %n, align 4
18  %1 = icmp sgt i32 %0, 0
19  br i1 %1, label %bb, label %return
20
21bb:
22  %indvar = phi i64 [ %indvar.next, %bb ], [ 0, %entry ]
23  %tmp = shl i64 %indvar, 2
24  %scevgep = getelementptr float, float* %y, i64 %tmp
25  %scevgep9 = bitcast float* %scevgep to <4 x float>*
26  %scevgep10 = getelementptr float, float* %x, i64 %tmp
27  %scevgep1011 = bitcast float* %scevgep10 to <4 x float>*
28  %2 = load <4 x float>, <4 x float>* %scevgep1011, align 16
29  %3 = bitcast <4 x float> %2 to <4 x i32>
30  %4 = and <4 x i32> %3, <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>
31  %5 = bitcast <4 x i32> %4 to <4 x float>
32  %6 = and <4 x i32> %3, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
33  %7 = tail call <4 x float> @llvm.x86.sse.cmp.ps(<4 x float> %5, <4 x float> <float 8.388608e+06, float 8.388608e+06, float 8.388608e+06, float 8.388608e+06>, i8 5) nounwind
34  %tmp.i4 = bitcast <4 x float> %7 to <4 x i32>
35  %8 = xor <4 x i32> %tmp.i4, <i32 -1, i32 -1, i32 -1, i32 -1>
36  %9 = and <4 x i32> %8, <i32 1258291200, i32 1258291200, i32 1258291200, i32 1258291200>
37  %10 = or <4 x i32> %9, %6
38  %11 = bitcast <4 x i32> %10 to <4 x float>
39  %12 = fadd <4 x float> %2, %11
40  %13 = fsub <4 x float> %12, %11
41  %14 = tail call <4 x float> @llvm.x86.sse.cmp.ps(<4 x float> %2, <4 x float> %13, i8 1) nounwind
42  %15 = bitcast <4 x float> %14 to <4 x i32>
43  %16 = tail call <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32> %15) nounwind readnone
44  %17 = fadd <4 x float> %13, %16
45  %tmp.i = bitcast <4 x float> %17 to <4 x i32>
46  %18 = or <4 x i32> %tmp.i, %6
47  %19 = bitcast <4 x i32> %18 to <4 x float>
48  store <4 x float> %19, <4 x float>* %scevgep9, align 16
49  %tmp12 = add i64 %tmp, 4
50  %tmp13 = trunc i64 %tmp12 to i32
51  %20 = load i32, i32* %n, align 4
52  %21 = icmp sgt i32 %20, %tmp13
53  %indvar.next = add i64 %indvar, 1
54  br i1 %21, label %bb, label %return
55
56return:
57  ret void
58}
59
60declare <4 x float> @llvm.x86.sse.cmp.ps(<4 x float>, <4 x float>, i8) nounwind readnone
61
62declare <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32>) nounwind readnone
63