1; RUN: opt -mtriple=amdgcn-amd-amdhsa -basicaa -load-store-vectorizer -S -o - %s | FileCheck %s
2
3declare i64 @_Z12get_local_idj(i32)
4
5declare i64 @_Z12get_group_idj(i32)
6
7declare double @llvm.fmuladd.f64(double, double, double)
8
9; CHECK-LABEL: @factorizedVsNonfactorizedAccess(
10; CHECK: load <2 x float>
11; CHECK: store <2 x float>
12define amdgpu_kernel void @factorizedVsNonfactorizedAccess(float addrspace(1)* nocapture %c) {
13entry:
14  %call = tail call i64 @_Z12get_local_idj(i32 0)
15  %call1 = tail call i64 @_Z12get_group_idj(i32 0)
16  %div = lshr i64 %call, 4
17  %div2 = lshr i64 %call1, 3
18  %mul = shl i64 %div2, 7
19  %rem = shl i64 %call, 3
20  %mul3 = and i64 %rem, 120
21  %add = or i64 %mul, %mul3
22  %rem4 = shl i64 %call1, 7
23  %mul5 = and i64 %rem4, 896
24  %mul6 = shl nuw nsw i64 %div, 3
25  %add7 = add nuw i64 %mul5, %mul6
26  %mul9 = shl i64 %add7, 10
27  %add10 = add i64 %mul9, %add
28  %arrayidx = getelementptr inbounds float, float addrspace(1)* %c, i64 %add10
29  %load1 = load float, float addrspace(1)* %arrayidx, align 4
30  %conv = fpext float %load1 to double
31  %mul11 = fmul double %conv, 0x3FEAB481D8F35506
32  %conv12 = fptrunc double %mul11 to float
33  %conv18 = fpext float %conv12 to double
34  %storeval1 = tail call double @llvm.fmuladd.f64(double 0x3FF4FFAFBBEC946A, double 0.000000e+00, double %conv18)
35  %cstoreval1 = fptrunc double %storeval1 to float
36  store float %cstoreval1, float addrspace(1)* %arrayidx, align 4
37
38  %add23 = or i64 %add10, 1
39  %arrayidx24 = getelementptr inbounds float, float addrspace(1)* %c, i64 %add23
40  %load2 = load float, float addrspace(1)* %arrayidx24, align 4
41  %conv25 = fpext float %load2 to double
42  %mul26 = fmul double %conv25, 0x3FEAB481D8F35506
43  %conv27 = fptrunc double %mul26 to float
44  %conv34 = fpext float %conv27 to double
45  %storeval2 = tail call double @llvm.fmuladd.f64(double 0x3FF4FFAFBBEC946A, double 0.000000e+00, double %conv34)
46  %cstoreval2 = fptrunc double %storeval2 to float
47  store float %cstoreval2, float addrspace(1)* %arrayidx24, align 4
48  ret void
49}