160 lines · plain
1; RUN: opt < %s -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck %s2 3target datalayout = "e-m:e-i64:64-n32:64"4target triple = "powerpc64le-unknown-linux-gnu"5 6; Check delinearization in loop cache analysis can handle fixed-size arrays.7; The IR is copied from llvm/test/Analysis/DependenceAnalysis/SimpleSIVNoValidityCheckFixedSize.ll8 9; CHECK: Loop 'for.body' has cost = 418611610; CHECK-NEXT: Loop 'for.body4' has cost = 13094411 12;; #define N 102413;; #define M 204814;; void t1(int a[N][M]) {15;; for (int i = 0; i < N-1; ++i)16;; for (int j = 2; j < M; ++j)17;; a[i][j] = a[i+1][j-2];18;; }19 20define void @t1(ptr %a) {21entry:22 br label %for.body23 24for.body: ; preds = %entry, %for.inc1125 %indvars.iv4 = phi i64 [ 0, %entry ], [ %indvars.iv.next5, %for.inc11 ]26 br label %for.body427 28for.body4: ; preds = %for.body, %for.body429 %indvars.iv = phi i64 [ 2, %for.body ], [ %indvars.iv.next, %for.body4 ]30 %0 = add nuw nsw i64 %indvars.iv4, 131 %1 = add nsw i64 %indvars.iv, -232 %arrayidx6 = getelementptr inbounds [2048 x i32], ptr %a, i64 %0, i64 %133 %2 = load i32, ptr %arrayidx6, align 434 %a_gep = getelementptr inbounds [2048 x i32], ptr %a, i64 035 %arrayidx10 = getelementptr inbounds [2048 x i32], ptr %a_gep, i64 %indvars.iv4, i64 %indvars.iv36 store i32 %2, ptr %arrayidx10, align 437 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 138 %exitcond = icmp ne i64 %indvars.iv.next, 204839 br i1 %exitcond, label %for.body4, label %for.inc1140 41for.inc11: ; preds = %for.body442 %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 143 %exitcond7 = icmp ne i64 %indvars.iv.next5, 102344 br i1 %exitcond7, label %for.body, label %for.end1345 46for.end13: ; preds = %for.inc1147 ret void48}49 50 51; CHECK: Loop 'for.body' has cost = 418611652; CHECK-NEXT: Loop 'for.body4' has cost = 13094453 54define void @t2(ptr %a) {55entry:56 br label %for.body57 58for.body: ; preds = %entry, %for.inc1159 %indvars.iv4 = phi i64 [ 0, %entry ], [ %indvars.iv.next5, %for.inc11 ]60 br label %for.body461 62for.body4: ; preds = %for.body, %for.body463 %indvars.iv = phi i64 [ 2, %for.body ], [ %indvars.iv.next, %for.body4 ]64 %0 = add nuw nsw i64 %indvars.iv4, 165 %1 = add nsw i64 %indvars.iv, -266 %arrayidx6 = getelementptr inbounds [2048 x i32], ptr %a, i64 %0, i64 %167 %2 = load i32, ptr %arrayidx6, align 468 %call = call ptr @func_with_returned_arg(ptr returned %a)69 %arrayidx10 = getelementptr inbounds [2048 x i32], ptr %call, i64 %indvars.iv4, i64 %indvars.iv70 store i32 %2, ptr %arrayidx10, align 471 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 172 %exitcond = icmp ne i64 %indvars.iv.next, 204873 br i1 %exitcond, label %for.body4, label %for.inc1174 75for.inc11: ; preds = %for.body476 %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 177 %exitcond7 = icmp ne i64 %indvars.iv.next5, 102378 br i1 %exitcond7, label %for.body, label %for.end1379 80for.end13: ; preds = %for.inc1181 ret void82}83 84declare ptr @func_with_returned_arg(ptr returned %arg)85 86; CHECK: Loop 'for.body' has cost = 211212881510400000087; CHECK-NEXT: Loop 'for.body4' has cost = 1676292710400000088; CHECK-NEXT: Loop 'for.body8' has cost = 13096036800000089; CHECK-NEXT: Loop 'for.body12' has cost = 104768294400090; CHECK-NEXT: Loop 'for.body16' has cost = 3277209600091 92;; #define N 12893;; #define M 204894;; void t3(int a[][N][N][N][M]) {95;; for (int i1 = 0; i1 < N-1; ++i1)96;; for (int i2 = 2; i2 < N; ++i2)97;; for (int i3 = 0; i3 < N; ++i3)98;; for (int i4 = 3; i4 < N; ++i4)99;; for (int i5 = 0; i5 < M-2; ++i5)100;; a[i1][i2][i3][i4][i5] = a[i1+1][i2-2][i3][i4-3][i5+2];101;; }102 103define void @t3(ptr %a) {104entry:105 br label %for.body106 107for.body: ; preds = %entry, %for.inc46108 %indvars.iv18 = phi i64 [ 0, %entry ], [ %indvars.iv.next19, %for.inc46 ]109 br label %for.body4110 111for.body4: ; preds = %for.body, %for.inc43112 %indvars.iv14 = phi i64 [ 2, %for.body ], [ %indvars.iv.next15, %for.inc43 ]113 br label %for.body8114 115for.body8: ; preds = %for.body4, %for.inc40116 %indvars.iv11 = phi i64 [ 0, %for.body4 ], [ %indvars.iv.next12, %for.inc40 ]117 br label %for.body12118 119for.body12: ; preds = %for.body8, %for.inc37120 %indvars.iv7 = phi i64 [ 3, %for.body8 ], [ %indvars.iv.next8, %for.inc37 ]121 br label %for.body16122 123for.body16: ; preds = %for.body12, %for.body16124 %indvars.iv = phi i64 [ 0, %for.body12 ], [ %indvars.iv.next, %for.body16 ]125 %0 = add nuw nsw i64 %indvars.iv18, 1126 %1 = add nsw i64 %indvars.iv14, -2127 %2 = add nsw i64 %indvars.iv7, -3128 %3 = add nuw nsw i64 %indvars.iv, 2129 %arrayidx26 = getelementptr inbounds [128 x [128 x [128 x [2048 x i32]]]], ptr %a, i64 %0, i64 %1, i64 %indvars.iv11, i64 %2, i64 %3130 %4 = load i32, ptr %arrayidx26, align 4131 %arrayidx36 = getelementptr inbounds [128 x [128 x [128 x [2048 x i32]]]], ptr %a, i64 %indvars.iv18, i64 %indvars.iv14, i64 %indvars.iv11, i64 %indvars.iv7, i64 %indvars.iv132 store i32 %4, ptr %arrayidx36, align 4133 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1134 %exitcond = icmp ne i64 %indvars.iv.next, 2046135 br i1 %exitcond, label %for.body16, label %for.inc37136 137for.inc37: ; preds = %for.body16138 %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 1139 %exitcond10 = icmp ne i64 %indvars.iv.next8, 128140 br i1 %exitcond10, label %for.body12, label %for.inc40141 142for.inc40: ; preds = %for.inc37143 %indvars.iv.next12 = add nuw nsw i64 %indvars.iv11, 1144 %exitcond13 = icmp ne i64 %indvars.iv.next12, 128145 br i1 %exitcond13, label %for.body8, label %for.inc43146 147for.inc43: ; preds = %for.inc40148 %indvars.iv.next15 = add nuw nsw i64 %indvars.iv14, 1149 %exitcond17 = icmp ne i64 %indvars.iv.next15, 128150 br i1 %exitcond17, label %for.body4, label %for.inc46151 152for.inc46: ; preds = %for.inc43153 %indvars.iv.next19 = add nuw nsw i64 %indvars.iv18, 1154 %exitcond21 = icmp ne i64 %indvars.iv.next19, 127155 br i1 %exitcond21, label %for.body, label %for.end48156 157for.end48: ; preds = %for.inc46158 ret void159}160