1; RUN: opt < %s -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck %s 2 3target datalayout = "e-m:e-i64:64-n32:64" 4target triple = "powerpc64le-unknown-linux-gnu" 5 6; void foo(long n, long m, long o, int A[n][m][o]) { 7; for (long i = 0; i < n; i++) 8; for (long j = 0; j < m; j++) 9; for (long k = 0; k < o; k++) 10; A[2*i+3][3*j-4][2*k+7] = 1; 11; } 12 13; CHECK: Loop 'for.i' has cost = 100000000 14; CHECK-NEXT: Loop 'for.j' has cost = 1000000 15; CHECK-NEXT: Loop 'for.k' has cost = 70000 16 17define void @foo(i64 %n, i64 %m, i64 %o, ptr %A) { 18entry: 19 %cmp32 = icmp sgt i64 %n, 0 20 %cmp230 = icmp sgt i64 %m, 0 21 %cmp528 = icmp sgt i64 %o, 0 22 br i1 %cmp32, label %for.cond1.preheader.lr.ph, label %for.end 23 24for.cond1.preheader.lr.ph: ; preds = %entry 25 br i1 %cmp230, label %for.i.preheader, label %for.end 26 27for.i.preheader: ; preds = %for.cond1.preheader.lr.ph 28 br i1 %cmp528, label %for.i.preheader.split, label %for.end 29 30for.i.preheader.split: ; preds = %for.i.preheader 31 br label %for.i 32 33for.i: ; preds = %for.inci, %for.i.preheader.split 34 %i = phi i64 [ %inci, %for.inci ], [ 0, %for.i.preheader.split ] 35 %mul8 = shl i64 %i, 1 36 %add9 = add nsw i64 %mul8, 3 37 %0 = mul i64 %add9, %m 38 %sub = add i64 %0, -4 39 br label %for.j 40 41for.j: ; preds = %for.incj, %for.i 42 %j = phi i64 [ %incj, %for.incj ], [ 0, %for.i ] 43 %mul7 = mul nsw i64 %j, 3 44 %tmp = add i64 %sub, %mul7 45 %tmp27 = mul i64 %tmp, %o 46 br label %for.k 47 48for.k: ; preds = %for.k, %for.j.us 49 %k = phi i64 [ 0, %for.j ], [ %inck, %for.k ] 50 51 %mul = mul nsw i64 %k, 2 52 %arrayidx.sum = add i64 %mul, 7 53 %arrayidx10.sum = add i64 %arrayidx.sum, %tmp27 54 %arrayidx11 = getelementptr inbounds i32, ptr %A, i64 %arrayidx10.sum 55 store i32 1, ptr %arrayidx11, align 4 56 57 %inck = add nsw i64 %k, 1 58 %exitcond.us = icmp eq i64 %inck, %o 59 br i1 %exitcond.us, label %for.incj, label %for.k 60 61for.incj: ; preds = %for.k 62 %incj = add nsw i64 %j, 1 63 %exitcond54.us = icmp eq i64 %incj, %m 64 br i1 %exitcond54.us, label %for.inci, label %for.j 65 66for.inci: ; preds = %for.incj 67 %inci = add nsw i64 %i, 1 68 %exitcond55.us = icmp eq i64 %inci, %n 69 br i1 %exitcond55.us, label %for.end.loopexit, label %for.i 70 71for.end.loopexit: ; preds = %for.inci 72 br label %for.end 73 74for.end: ; preds = %for.end.loopexit, %for.cond1.preheader.lr.ph, %entry 75 ret void 76} 77 78; Loop i is supposed to have the largest cost and be placed 79; as the outermost loop. This test differs from foo() since the 80; loopnest has a suboptimal order j-i-k. 81; After D123400 we ensure that the order of loop cache analysis output 82; is loop i-j-k, despite the suboptimal order in the original loopnest. 83; 84; void foo(long n, long m, long o, int A[n][m][o]) { 85; for (long j = 0; j < m; j++) 86; for (long i = 0; i < n; i++) 87; for (long k = 0; k < o; k++) 88; A[2*i+3][2*j-4][2*k+7] = 1; 89; } 90 91; CHECK: Loop 'for.i' has cost = 100000000 92; CHECK-NEXT: Loop 'for.j' has cost = 1000000 93; CHECK-NEXT: Loop 'for.k' has cost = 70000 94 95define void @foo2(i64 %n, i64 %m, i64 %o, ptr %A) { 96entry: 97 %cmp32 = icmp sgt i64 %n, 0 98 %cmp230 = icmp sgt i64 %m, 0 99 %cmp528 = icmp sgt i64 %o, 0 100 br i1 %cmp32, label %for.cond1.preheader.lr.ph, label %for.end 101 102for.cond1.preheader.lr.ph: ; preds = %entry 103 br i1 %cmp230, label %for.j.preheader, label %for.end 104 105for.j.preheader: ; preds = %for.cond1.preheader.lr.ph 106 br i1 %cmp528, label %for.j.preheader.split, label %for.end 107 108for.j.preheader.split: ; preds = %for.j.preheader 109 br label %for.j 110 111for.i: ; preds = %for.inci, %for.j 112 %i = phi i64 [ %inci, %for.inci ], [ 0, %for.j ] 113 %mul8 = shl i64 %i, 1 114 %add9 = add nsw i64 %mul8, 3 115 %0 = mul i64 %add9, %m 116 %sub = add i64 %0, -4 117 %mul7 = mul nsw i64 %j, 2 118 %tmp = add i64 %sub, %mul7 119 %tmp27 = mul i64 %tmp, %o 120 br label %for.k 121 122for.j: ; preds = %for.incj, %for.j.preheader.split 123 %j = phi i64 [ %incj, %for.incj ], [ 0, %for.j.preheader.split ] 124 br label %for.i 125 126for.k: ; preds = %for.k, %for.i 127 %k = phi i64 [ 0, %for.i ], [ %inck, %for.k ] 128 129 %mul = mul nsw i64 %k, 2 130 %arrayidx.sum = add i64 %mul, 7 131 %arrayidx10.sum = add i64 %arrayidx.sum, %tmp27 132 %arrayidx11 = getelementptr inbounds i32, ptr %A, i64 %arrayidx10.sum 133 store i32 1, ptr %arrayidx11, align 4 134 135 %inck = add nsw i64 %k, 1 136 %exitcond.us = icmp eq i64 %inck, %o 137 br i1 %exitcond.us, label %for.inci, label %for.k 138 139for.incj: ; preds = %for.inci 140 %incj = add nsw i64 %j, 1 141 %exitcond54.us = icmp eq i64 %incj, %m 142 br i1 %exitcond54.us, label %for.end.loopexit, label %for.j 143 144for.inci: ; preds = %for.k 145 %inci = add nsw i64 %i, 1 146 %exitcond55.us = icmp eq i64 %inci, %n 147 br i1 %exitcond55.us, label %for.incj, label %for.i 148 149for.end.loopexit: ; preds = %for.incj 150 br label %for.end 151 152for.end: ; preds = %for.end.loopexit, %for.cond1.preheader.lr.ph, %entry 153 ret void 154} 155