xref: /llvm-project/llvm/test/CodeGen/PowerPC/loop-data-prefetch-inner.ll (revision 427fb35192f1f7bb694a5910b05abc5925a798b2)
1; RUN: llc -enable-ppc-prefetching=true -verify-machineinstrs < %s | FileCheck %s
2target datalayout = "E-m:e-i64:64-n32:64"
3target triple = "powerpc64le-unknown-linux"
4
5; Function Attrs: nounwind
6define void @foo(ptr %x, ptr nocapture readonly %y) #0 {
7entry:
8  %scevgep = getelementptr double, ptr %x, i64 1599
9  %scevgep20 = getelementptr double, ptr %y, i64 1599
10  br label %vector.memcheck
11
12vector.memcheck:                                  ; preds = %for.end, %entry
13  %j.015 = phi i32 [ 0, %entry ], [ %inc7, %for.end ]
14  %bound0 = icmp uge ptr %scevgep20, %x
15  %bound1 = icmp uge ptr %scevgep, %y
16  %memcheck.conflict = and i1 %bound0, %bound1
17  br i1 %memcheck.conflict, label %middle.block, label %vector.body
18
19vector.body:                                      ; preds = %vector.memcheck, %vector.body
20  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.memcheck ]
21  %0 = getelementptr inbounds double, ptr %y, i64 %index
22  %wide.load = load <4 x double>, ptr %0, align 8
23  %1 = fadd <4 x double> %wide.load, <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
24  %2 = getelementptr inbounds double, ptr %x, i64 %index
25  store <4 x double> %1, ptr %2, align 8
26  %index.next = add i64 %index, 4
27  %3 = icmp eq i64 %index.next, 1600
28  br i1 %3, label %middle.block, label %vector.body
29
30middle.block:                                     ; preds = %vector.body, %vector.memcheck
31  %resume.val = phi i1 [ false, %vector.memcheck ], [ true, %vector.body ]
32  %trunc.resume.val = phi i64 [ 0, %vector.memcheck ], [ 1600, %vector.body ]
33  br i1 %resume.val, label %for.end, label %for.body3
34
35for.body3:                                        ; preds = %middle.block, %for.body3
36  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body3 ], [ %trunc.resume.val, %middle.block ]
37  %arrayidx = getelementptr inbounds double, ptr %y, i64 %indvars.iv
38  %4 = load double, ptr %arrayidx, align 8
39  %add = fadd double %4, 1.000000e+00
40  %arrayidx5 = getelementptr inbounds double, ptr %x, i64 %indvars.iv
41  store double %add, ptr %arrayidx5, align 8
42  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
43  %exitcond = icmp eq i64 %indvars.iv.next, 1600
44  br i1 %exitcond, label %for.end, label %for.body3
45
46for.end:                                          ; preds = %middle.block, %for.body3
47  tail call void @bar(ptr %x) #2
48  %inc7 = add nuw nsw i32 %j.015, 1
49  %exitcond16 = icmp eq i32 %inc7, 100
50  br i1 %exitcond16, label %for.end8, label %vector.memcheck
51
52for.end8:                                         ; preds = %for.end
53  ret void
54
55; CHECK-LABEL: @foo
56; CHECK: dcbt
57}
58
59declare void @bar(ptr) #1
60
61attributes #0 = { nounwind "target-cpu"="a2q" }
62attributes #1 = { "target-cpu"="a2q" }
63attributes #2 = { nounwind }
64
65