xref: /llvm-project/llvm/test/Analysis/LoopCacheAnalysis/interchange-cost-beneficial.ll (revision 670259466b238176ac302c8dedf806d2b2be7e0c)
1; RUN: opt <  %s  -cache-line-size=64 -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck  %s
2
3;; This test checks the effect of rounding cache cost to 1 when it is
4;; evaluated to 0 because at least 1 cache line is accessed by the loopnest.
5;; It does not make sense to output that zero cache lines are used.
6;; The cost of reference group for B[j], C[j], D[j] and E[j] were
7;; calculted 0 before but now they are 1 which makes each loop cost more reasonable.
8;
9; void test(int n, int m, int o, int A[2][3], int B[2], int C[2], int D[2], int E[2]) {
10;   for (int i = 0; i < 3; i++)
11;     for (int j = 0; j < 2; j++)
12;        A[j][i] = 1;
13;        B[j] = 1;
14;        C[j] = 1;
15;        D[j] = 1
16;        E[j] = 1
17; }
18
19; CHECK: Loop 'for.j' has cost = 18
20; CHECK-NEXT: Loop 'for.i' has cost = 10
21
22define void @test(ptr %A, ptr %B, ptr %C, ptr %D, ptr %E) {
23
24entry:
25  br label %for.i.preheader.split
26
27for.i.preheader.split:                            ; preds = %for.i.preheader
28  br label %for.i
29
30for.i:                                            ; preds = %for.inci, %for.i.preheader.split
31  %i = phi i64 [ %inci, %for.inci ], [ 0, %for.i.preheader.split ]
32  br label %for.j
33
34for.j:                                            ; preds = %for.incj, %for.i
35  %j = phi i64 [ %incj, %for.j ], [ 0, %for.i ]
36  %mul_j = mul nsw i64 %j, 3
37  %index_j = add i64 %mul_j, %i
38  %arrayidxA = getelementptr inbounds [2 x [ 3 x i32]], ptr %A, i64 %j, i64 %i
39  store i32 1, ptr %arrayidxA, align 4
40  %arrayidxB = getelementptr inbounds i32, ptr %B, i64 %j
41  store i32 1, ptr %arrayidxB, align 4
42  %arrayidxC = getelementptr inbounds i32, ptr %C, i64 %j
43  store i32 1, ptr %arrayidxC, align 4
44  %arrayidxD = getelementptr inbounds i32, ptr %D, i64 %j
45  store i32 1, ptr %arrayidxD, align 4
46  %arrayidxE = getelementptr inbounds i32, ptr %E, i64 %j
47  store i32 1, ptr %arrayidxE, align 4
48  %incj = add nsw i64 %j, 1
49  %exitcond.us = icmp eq i64 %incj, 2
50  br i1 %exitcond.us, label %for.inci, label %for.j
51
52for.inci:                                         ; preds = %for.incj
53  %inci = add nsw i64 %i, 1
54  %exitcond55.us = icmp eq i64 %inci, 3
55  br i1 %exitcond55.us, label %for.end.loopexit, label %for.i
56
57for.end.loopexit:                                 ; preds = %for.inci
58  br label %for.end
59
60for.end:                                          ; preds = %for.end.loopexit, %for.cond1.preheader.lr.ph, %entry
61  ret void
62}
63