1 ; RUN: opt %loadNPMPolly \
2 ; RUN: -polly-pattern-matching-based-opts=true \
3 ; RUN: '-passes=polly-optree,polly-delicm,polly-simplify,polly-opt-isl' \
4 ; RUN: -polly-tc-opt=true -debug -disable-output < %s 2>&1 \
8 ; Check that the pattern matching detects the matrix multiplication pattern
9 ; after a full run of -polly-optree and -polly-delicm, where the write access
10 ; is not through the original memory access, but trough a PHI node that was
11 ; delicmed. This test covers the polybench 2mm and 3mm cases.
13 ; This test case generates the following schedule, which contains filters:
15 ; domain: "{ Stmt_for_body8[i0, i1, i2] : 0 <= i0 <= 1599 and
18 ; Stmt_for_body3[i0, i1] : 0 <= i0 <= 1599 and
20 ; Stmt_for_body3_last[i0, i1] : 0 <= i0 <= 1599 and
24 ; - filter: "{ Stmt_for_body3[i0, i1] }"
26 ; schedule: "[{ Stmt_for_body3[i0, i1] -> [(i0)] }, { Stmt_for_body3[i0, i1] -> [(i1)] }]"
28 ; coincident: [ 1, 1 ]
29 ; - filter: "{ Stmt_for_body3_last[i0, i1] }"
31 ; schedule: "[{ Stmt_for_body3_last[i0, i1] -> [(i0)] }, { Stmt_for_body3_last[i0, i1] -> [(i1)] }]"
33 ; coincident: [ 1, 1 ]
34 ; - filter: "{ Stmt_for_body8[i0, i1, i2] }"
36 ; schedule: "[{ Stmt_for_body8[i0, i1, i2] -> [(i0)] },
37 ; { Stmt_for_body8[i0, i1, i2] -> [(i1)] },
38 ; { Stmt_for_body8[i0, i1, i2] -> [(i2)] }]"
40 ; coincident: [ 1, 1, 0 ]
42 ; CHECK: The tensor contraction pattern was detected
43 ; CHECK: The matrix multiplication pattern was detected
46 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
47 target triple = "x86_64-unknown-linux-gnu"
49 ; Function Attrs: norecurse nounwind uwtable
50 define void @kernel_2mm(i32 %ni, i32 %nj, i32 %nk, i32 %nl, double %alpha, double %beta, ptr nocapture %tmp, ptr nocapture readonly %A, ptr nocapture readonly %B, ptr nocapture readnone %C, ptr nocapture readnone %D) local_unnamed_addr #0 {
54 entry.split: ; preds = %entry
57 for.body: ; preds = %for.inc25, %entry.split
58 %indvars.iv50 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next51, %for.inc25 ]
61 for.body3: ; preds = %for.inc22, %for.body
62 %indvars.iv46 = phi i64 [ 0, %for.body ], [ %indvars.iv.next47, %for.inc22 ]
63 %arrayidx5 = getelementptr inbounds [1800 x double], ptr %tmp, i64 %indvars.iv50, i64 %indvars.iv46
64 store double 0.000000e+00, ptr %arrayidx5, align 8, !tbaa !2
67 for.body8: ; preds = %for.body8, %for.body3
68 %0 = phi double [ 0.000000e+00, %for.body3 ], [ %add, %for.body8 ]
69 %indvars.iv = phi i64 [ 0, %for.body3 ], [ %indvars.iv.next, %for.body8 ]
70 %arrayidx12 = getelementptr inbounds [2200 x double], ptr %A, i64 %indvars.iv50, i64 %indvars.iv
71 %1 = load double, ptr %arrayidx12, align 8, !tbaa !2
72 %mul = fmul double %1, %alpha
73 %arrayidx16 = getelementptr inbounds [1800 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv46
74 %2 = load double, ptr %arrayidx16, align 8, !tbaa !2
75 %mul17 = fmul double %mul, %2
76 %add = fadd double %0, %mul17
77 store double %add, ptr %arrayidx5, align 8, !tbaa !2
78 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
79 %exitcond = icmp eq i64 %indvars.iv.next, 2200
80 br i1 %exitcond, label %for.inc22, label %for.body8
82 for.inc22: ; preds = %for.body8
83 %indvars.iv.next47 = add nuw nsw i64 %indvars.iv46, 1
84 %exitcond48 = icmp eq i64 %indvars.iv.next47, 1800
85 br i1 %exitcond48, label %for.inc25, label %for.body3
87 for.inc25: ; preds = %for.inc22
88 %indvars.iv.next51 = add nuw nsw i64 %indvars.iv50, 1
89 %exitcond52 = icmp eq i64 %indvars.iv.next51, 1600
90 br i1 %exitcond52, label %for.end27, label %for.body
92 for.end27: ; preds = %for.inc25
96 attributes #0 = { norecurse nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="haswell" "target-features"="+aes,+avx,+avx2,+bmi,+bmi2,+cmov,+cx16,+f16c,+fma,+fsgsbase,+fxsr,+lzcnt,+mmx,+movbe,+pclmul,+popcnt,+rdrnd,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt,-adx,-avx512bw,-avx512cd,-avx512dq,-avx512er,-avx512f,-avx512ifma,-avx512pf,-avx512vbmi,-avx512vl,-avx512vpopcntdq,-clflushopt,-clwb,-clzero,-fma4,-lwp,-mwaitx,-pku,-prefetchwt1,-prfchw,-rdseed,-rtm,-sgx,-sha,-sse4a,-tbm,-xop,-xsavec,-xsaves" "unsafe-fp-math"="false" "use-soft-float"="false" }
98 !llvm.module.flags = !{!0}
101 !0 = !{i32 1, !"wchar_size", i32 4}
102 !1 = !{!"clang version 6.0.0 (trunk 309912) (llvm/trunk 309933)"}
103 !2 = !{!3, !3, i64 0}
104 !3 = !{!"double", !4, i64 0}
105 !4 = !{!"omnipotent char", !5, i64 0}
106 !5 = !{!"Simple C/C++ TBAA"}