1 ; RUN: llc < %s -march=x86-64 | FileCheck %s
3 ; Full strength reduction wouldn't reduce register pressure, so LSR should
4 ; stick with indexing here.
6 ; CHECK: movaps (%rsi,%rax,4), %xmm3
7 ; CHECK: movaps %xmm3, (%rdi,%rax,4)
9 ; CHECK: cmpl %eax, (%rdx)
12 define void @vvfloorf(float* nocapture %y, float* nocapture %x, i32* nocapture %n) nounwind {
14 %0 = load i32* %n, align 4
15 %1 = icmp sgt i32 %0, 0
16 br i1 %1, label %bb, label %return
19 %indvar = phi i64 [ %indvar.next, %bb ], [ 0, %entry ]
20 %tmp = shl i64 %indvar, 2
21 %scevgep = getelementptr float* %y, i64 %tmp
22 %scevgep9 = bitcast float* %scevgep to <4 x float>*
23 %scevgep10 = getelementptr float* %x, i64 %tmp
24 %scevgep1011 = bitcast float* %scevgep10 to <4 x float>*
25 %2 = load <4 x float>* %scevgep1011, align 16
26 %3 = bitcast <4 x float> %2 to <4 x i32>
27 %4 = and <4 x i32> %3, <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>
28 %5 = bitcast <4 x i32> %4 to <4 x float>
29 %6 = and <4 x i32> %3, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
30 %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
31 %tmp.i4 = bitcast <4 x float> %7 to <4 x i32>
32 %8 = xor <4 x i32> %tmp.i4, <i32 -1, i32 -1, i32 -1, i32 -1>
33 %9 = and <4 x i32> %8, <i32 1258291200, i32 1258291200, i32 1258291200, i32 1258291200>
34 %10 = or <4 x i32> %9, %6
35 %11 = bitcast <4 x i32> %10 to <4 x float>
36 %12 = fadd <4 x float> %2, %11
37 %13 = fsub <4 x float> %12, %11
38 %14 = tail call <4 x float> @llvm.x86.sse.cmp.ps(<4 x float> %2, <4 x float> %13, i8 1) nounwind
39 %15 = bitcast <4 x float> %14 to <4 x i32>
40 %16 = tail call <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32> %15) nounwind readnone
41 %17 = fadd <4 x float> %13, %16
42 %tmp.i = bitcast <4 x float> %17 to <4 x i32>
43 %18 = or <4 x i32> %tmp.i, %6
44 %19 = bitcast <4 x i32> %18 to <4 x float>
45 store <4 x float> %19, <4 x float>* %scevgep9, align 16
46 %tmp12 = add i64 %tmp, 4
47 %tmp13 = trunc i64 %tmp12 to i32
48 %20 = load i32* %n, align 4
49 %21 = icmp sgt i32 %20, %tmp13
50 %indvar.next = add i64 %indvar, 1
51 br i1 %21, label %bb, label %return
57 declare <4 x float> @llvm.x86.sse.cmp.ps(<4 x float>, <4 x float>, i8) nounwind readnone
59 declare <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32>) nounwind readnone