xref: /llvm-project/polly/test/ScheduleOptimizer/pattern-matching-based-opts_15.ll (revision e1f056f692d869708c1898d9d65a69ac5584a0ed)
1; RUN: opt %loadNPMPolly -passes=polly-opt-isl -polly-pattern-matching-based-opts=true \
2; RUN: -debug-only=polly-opt-isl -disable-output \
3; RUN: -polly-tc-opt=true < %s 2>&1 | FileCheck %s
4; REQUIRES: asserts
5;
6;    for (i = 0; i < _PB_NI; i++)
7;      for (j = 0; j < _PB_NJ; j++)
8;      {
9;        for (k = 0; k < _PB_NK; k++)
10;        {
11;          double Mul = A[i][k] * B[k][j];
12;          D[i][j][k] += Mul;
13;          C[i][j] += Mul;
14;        }
15;      }
16;
17; CHECK-NOT: The matrix multiplication pattern was detected
18; CHECK-NOT: The tensor contraction pattern was detected
19
20target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
21target triple = "x86_64-unknown-linux-gnu"
22
23define void @kernel_gemm(ptr %C, ptr %A, ptr %B, ptr %D) {
24entry:
25  br label %for.cond1.preheader
26
27for.cond1.preheader:                              ; preds = %for.inc27, %entry
28  %indvars.iv7 = phi i64 [ 0, %entry ], [ %indvars.iv.next8, %for.inc27 ]
29  br label %for.cond4.preheader
30
31for.cond4.preheader:                              ; preds = %for.inc24, %for.cond1.preheader
32  %indvars.iv4 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next5, %for.inc24 ]
33  %arrayidx22 = getelementptr inbounds [1024 x double], ptr %C, i64 %indvars.iv7, i64 %indvars.iv4
34  br label %for.body6
35
36for.body6:                                        ; preds = %for.body6, %for.cond4.preheader
37  %indvars.iv = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next, %for.body6 ]
38  %arrayidx8 = getelementptr inbounds [1024 x double], ptr %A, i64 %indvars.iv7, i64 %indvars.iv
39  %i = load double, ptr %arrayidx8, align 8
40  %arrayidx12 = getelementptr inbounds [1024 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv4
41  %i1 = load double, ptr %arrayidx12, align 8
42  %mul = fmul double %i1, %i
43  %arrayidx18 = getelementptr inbounds [1024 x [1024 x double]], ptr %D, i64 %indvars.iv7, i64 %indvars.iv4, i64 %indvars.iv
44  %i2 = load double, ptr %arrayidx18, align 8
45  %add = fadd double %i2, %mul
46  store double %add, ptr %arrayidx18, align 8
47  %i3 = load double, ptr %arrayidx22, align 8
48  %add23 = fadd double %i3, %mul
49  store double %add23, ptr %arrayidx22, align 8
50  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
51  %exitcond.not = icmp eq i64 %indvars.iv.next, 1024
52  br i1 %exitcond.not, label %for.inc24, label %for.body6
53
54for.inc24:                                        ; preds = %for.body6
55  %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 1
56  %exitcond6.not = icmp eq i64 %indvars.iv.next5, 1024
57  br i1 %exitcond6.not, label %for.inc27, label %for.cond4.preheader
58
59for.inc27:                                        ; preds = %for.inc24
60  %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 1
61  %exitcond9.not = icmp eq i64 %indvars.iv.next8, 1024
62  br i1 %exitcond9.not, label %for.end29, label %for.cond1.preheader
63
64for.end29:                                        ; preds = %for.inc27
65  ret void
66}
67