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; Check IndexedReference::computeRefCost can handle type differences between 7; Stride and TripCount 8 9; CHECK: Loop 'for.cond' has cost = 64 10 11%struct._Handleitem = type { ptr } 12 13define void @handle_to_ptr(ptr %blocks) { 14; Preheader: 15entry: 16 br label %for.cond 17 18; Loop: 19for.cond: ; preds = %for.body, %entry 20 %i.0 = phi i32 [ 1, %entry ], [ %inc, %for.body ] 21 %cmp = icmp ult i32 %i.0, 1024 22 br i1 %cmp, label %for.body, label %for.end 23 24for.body: ; preds = %for.cond 25 %idxprom = zext i32 %i.0 to i64 26 %arrayidx = getelementptr inbounds ptr, ptr %blocks, i64 %idxprom 27 store ptr null, ptr %arrayidx, align 8 28 %inc = add nuw nsw i32 %i.0, 1 29 br label %for.cond 30 31; Exit blocks 32for.end: ; preds = %for.cond 33 ret void 34} 35 36; Check IndexedReference::computeRefCost can handle type differences between 37; Coeff and ElemSize. 38 39; CHECK: Loop 'for.cond' has cost = 100000000 40; CHECK: Loop 'for.cond1' has cost = 1000000 41; CHECK: Loop 'for.cond5' has cost = 40000 42 43@data = external dso_local global [2 x [4 x [18 x i32]]], align 1 44 45define dso_local void @handle_to_ptr_2(i1 %b0, i1 %b1, i1 %b2) { 46entry: 47 br label %for.cond 48 49for.cond: 50 %i.0 = phi i16 [ 0, %entry ], [ %inc18, %for.inc17 ] 51 %idxprom = zext i16 %i.0 to i32 52 br i1 %b2, label %for.end19, label %for.cond1 53 54for.cond1: 55 %j.0 = phi i16 [ %inc15, %for.inc14 ], [ 0, %for.cond ] 56 br i1 %b1, label %for.inc17, label %for.cond5.preheader 57 58for.cond5.preheader: 59 %idxprom10 = zext i16 %j.0 to i32 60 br label %for.cond5 61 62for.cond5: 63 %k.0 = phi i16 [ %inc, %for.inc ], [ 0, %for.cond5.preheader ] 64 br i1 %b0, label %for.inc14, label %for.inc 65 66for.inc: 67 %idxprom12 = zext i16 %k.0 to i32 68 %arrayidx13 = getelementptr inbounds [2 x [4 x [18 x i32]]], ptr @data, i32 0, i32 %idxprom, i32 %idxprom10, i32 %idxprom12 69 store i32 7, ptr %arrayidx13, align 1 70 %inc = add nuw nsw i16 %k.0, 1 71 br label %for.cond5 72 73for.inc14: 74 %inc15 = add nuw nsw i16 %j.0, 1 75 br label %for.cond1 76 77for.inc17: 78 %inc18 = add nuw nsw i16 %i.0, 1 79 br label %for.cond 80 81for.end19: 82 ret void 83} 84 85; Check IndexedReference::computeRefCost can handle negative stride 86 87; CHECK: Loop 'for.neg.cond' has cost = 64 88 89define void @handle_to_ptr_neg_stride(ptr %blocks) { 90; Preheader: 91entry: 92 br label %for.neg.cond 93 94; Loop: 95for.neg.cond: ; preds = %for.neg.body, %entry 96 %i.0 = phi i32 [ 1023, %entry ], [ %dec, %for.neg.body ] 97 %cmp = icmp sgt i32 %i.0, 0 98 br i1 %cmp, label %for.neg.body, label %for.neg.end 99 100for.neg.body: ; preds = %for.neg.cond 101 %idxprom = zext i32 %i.0 to i64 102 %arrayidx = getelementptr inbounds ptr, ptr %blocks, i64 %idxprom 103 store ptr null, ptr %arrayidx, align 8 104 %dec = add nsw i32 %i.0, -1 105 br label %for.neg.cond 106 107; Exit blocks 108for.neg.end: ; preds = %for.neg.cond 109 ret void 110} 111 112 113 114; for (int i = 40960; i > 0; i--) 115; B[i] = B[40960 - i]; 116 117; FIXME: Currently negative access functions are treated the same as positive 118; access functions. When this is fixed this testcase should have a cost 119; approximately 2x higher. 120 121; CHECK: Loop 'for.cond2' has cost = 2561 122define void @Test2(ptr %B) { 123entry: 124 br label %for.cond2 125 126for.cond2: ; preds = %for.body, %entry 127 %i.0 = phi i32 [ 40960, %entry ], [ %dec, %for.body ] 128 %cmp = icmp sgt i32 %i.0, 0 129 br i1 %cmp, label %for.body, label %for.end 130 131for.body: ; preds = %for.cond 132 %sub = sub nsw i32 40960, %i.0 133 %idxprom = sext i32 %sub to i64 134 %arrayidx = getelementptr inbounds double, ptr %B, i64 %idxprom 135 %0 = load double, ptr %arrayidx, align 8 136 %idxprom1 = sext i32 %i.0 to i64 137 %arrayidx2 = getelementptr inbounds double, ptr %B, i64 %idxprom1 138 store double %0, ptr %arrayidx2, align 8 139 %dec = add nsw i32 %i.0, -1 140 br label %for.cond2 141 142for.end: ; preds = %for.cond 143 ret void 144} 145 146 147 148; for (i = 40960; i > 0; i--) 149; C[i] = C[i]; 150 151; CHECK: Loop 'for.cond3' has cost = 2561 152define void @Test3(ptr %C) { 153entry: 154 br label %for.cond3 155 156for.cond3: ; preds = %for.body, %entry 157 %i.0 = phi i32 [ 40960, %entry ], [ %dec, %for.body ] 158 %cmp = icmp sgt i32 %i.0, 0 159 br i1 %cmp, label %for.body, label %for.end 160 161for.body: ; preds = %for.cond 162 %idxprom = sext i32 %i.0 to i64 163 %arrayidx = getelementptr inbounds ptr, ptr %C, i64 %idxprom 164 %0 = load ptr, ptr %arrayidx, align 8 165 %idxprom1 = sext i32 %i.0 to i64 166 %arrayidx2 = getelementptr inbounds ptr, ptr %C, i64 %idxprom1 167 store ptr %0, ptr %arrayidx2, align 8 168 %dec = add nsw i32 %i.0, -1 169 br label %for.cond3 170 171for.end: ; preds = %for.cond 172 ret void 173} 174 175 176 177; for (i = 0; i < 40960; i++) 178; D[i] = D[i]; 179 180; CHECK: Loop 'for.cond4' has cost = 2561 181define void @Test4(ptr %D) { 182entry: 183 br label %for.cond4 184 185for.cond4: ; preds = %for.body, %entry 186 %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 187 %cmp = icmp slt i32 %i.0, 40960 188 br i1 %cmp, label %for.body, label %for.end 189 190for.body: ; preds = %for.cond 191 %idxprom = sext i32 %i.0 to i64 192 %arrayidx = getelementptr inbounds ptr, ptr %D, i64 %idxprom 193 %0 = load ptr, ptr %arrayidx, align 8 194 %idxprom1 = sext i32 %i.0 to i64 195 %arrayidx2 = getelementptr inbounds ptr, ptr %D, i64 %idxprom1 196 store ptr %0, ptr %arrayidx2, align 8 197 %inc = add nsw i32 %i.0, 1 198 br label %for.cond4 199 200for.end: ; preds = %for.cond 201 ret void 202} 203