75 lines · plain
1; RUN: opt < %s -passes=loop-interchange -verify-dom-info -verify-loop-info \2; RUN: -pass-remarks-output=%t -pass-remarks='loop-interchange' -S3; RUN: cat %t | FileCheck %s4 5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"6 7@D = common global [100 x [100 x [100 x i32]]] zeroinitializer8 9; The outer loop's backedge isn't taken. Check the loop with BTC=0 is considered10; unprofitable, but that we still interchange the two inner loops.11;12; for(int i=0;i<1;i++)13; for(int j=0;j<100;j++)14; for(int k=0;k<100;k++)15; D[i][k][j] = D[i][k][j]+t;16;17 18; CHECK: --- !Analysis19; CHECK-NEXT: Pass: loop-interchange20; CHECK-NEXT: Name: Dependence21; CHECK-NEXT: Function: interchange_i_and_j22; CHECK-NEXT: Args:23; CHECK-NEXT: - String: Computed dependence info, invoking the transform.24; CHECK-NEXT: ...25; CHECK-NEXT: --- !Passed26; CHECK-NEXT: Pass: loop-interchange27; CHECK-NEXT: Name: Interchanged28; CHECK-NEXT: Function: interchange_i_and_j29; CHECK-NEXT: Args:30; CHECK-NEXT: - String: Loop interchanged with enclosing loop.31; CHECK-NEXT: ...32; CHECK-NEXT: --- !Missed33; CHECK-NEXT: Pass: loop-interchange34; CHECK-NEXT: Name: InterchangeNotProfitable35; CHECK-NEXT: Function: interchange_i_and_j36; CHECK-NEXT: Args:37; CHECK-NEXT: - String: Insufficient information to calculate the cost of loop for interchange.38; CHECK-NEXT: ...39 40define void @interchange_i_and_j(i32 %t){41entry:42 br label %outer.header43 44outer.header:45 %i = phi i64 [ 0, %entry ], [ %inc16, %for.inc15 ]46 br label %inner1.header47 48inner1.header:49 %j = phi i64 [ 0, %outer.header ], [ %inc13, %for.inc12 ]50 br label %inner2.body51 52inner2.body:53 %k = phi i64 [ 0, %inner1.header ], [ %inc, %inner2.body ]54 %arrayidx8 = getelementptr inbounds [100 x [100 x i32]], ptr @D, i64 %i, i64 %k, i64 %j55 %0 = load i32, ptr %arrayidx856 %add = add nsw i32 %0, %t57 store i32 %add, ptr %arrayidx858 %inc = add nuw nsw i64 %k, 159 %exitcond = icmp eq i64 %inc, 10060 br i1 %exitcond, label %for.inc12, label %inner2.body61 62for.inc12:63 %inc13 = add nuw nsw i64 %j, 164 %exitcond29 = icmp eq i64 %inc13, 10065 br i1 %exitcond29, label %for.inc15, label %inner1.header66 67for.inc15:68 %inc16 = add nuw nsw i64 %i, 169 %exitcond30 = icmp eq i64 %inc16, 170 br i1 %exitcond30, label %for.end17, label %outer.header71 72for.end17:73 ret void74}75