1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -pass-remarks-missed='loop-interchange' -verify-loop-lcssa -S | FileCheck %s 3; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -da-disable-delinearization-checks -pass-remarks-missed='loop-interchange' -verify-loop-lcssa -S | FileCheck -check-prefix=CHECK-DELIN %s 4 5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 6 7; void foo(int n, int m) { 8; int temp[16][16]; 9; int res[16][16]; 10; for(int i = 0; i < n; i++) { 11; for(int j = 0; j < m; j++) 12; res[j][i] = temp[j][i]; 13; } 14; } 15 16;; This loop can be interchanged with -da-disable-delinearization-checks, otherwise it cannot 17;; be interchanged due to dependence. 18define void @lcssa_08(i32 %n, i32 %m) {; 19; CHECK-DELIN-LABEL: @lcssa_08( 20; CHECK-DELIN-NEXT: entry: 21; CHECK-DELIN-NEXT: [[TEMP:%.*]] = alloca [16 x [16 x i32]], align 4 22; CHECK-DELIN-NEXT: [[RES:%.*]] = alloca [16 x [16 x i32]], align 4 23; CHECK-DELIN-NEXT: [[CMP24:%.*]] = icmp sgt i32 [[N:%.*]], 0 24; CHECK-DELIN-NEXT: br i1 [[CMP24]], label [[INNER_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]] 25; CHECK-DELIN: outer.preheader: 26; CHECK-DELIN-NEXT: br label [[OUTER_HEADER:%.*]] 27; CHECK-DELIN: outer.header: 28; CHECK-DELIN-NEXT: [[INDVARS_IV27:%.*]] = phi i64 [ 0, [[OUTER_PREHEADER:%.*]] ], [ [[INDVARS_IV_NEXT28:%.*]], [[OUTER_LATCH:%.*]] ] 29; CHECK-DELIN-NEXT: [[CMP222:%.*]] = icmp sgt i32 [[M:%.*]], 0 30; CHECK-DELIN-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[M]] to i64 31; CHECK-DELIN-NEXT: br i1 [[CMP222]], label [[INNER_FOR_BODY_SPLIT1:%.*]], label [[INNER_FOR_BODY_SPLIT:%.*]] 32; CHECK-DELIN: inner.preheader: 33; CHECK-DELIN-NEXT: [[WIDE_TRIP_COUNT29:%.*]] = zext i32 [[N]] to i64 34; CHECK-DELIN-NEXT: br label [[INNER_FOR_BODY:%.*]] 35; CHECK-DELIN: inner.for.body: 36; CHECK-DELIN-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[INNER_PREHEADER]] ], [ [[TMP1:%.*]], [[INNER_FOR_BODY_SPLIT]] ] 37; CHECK-DELIN-NEXT: br label [[OUTER_PREHEADER]] 38; CHECK-DELIN: inner.for.body.split1: 39; CHECK-DELIN-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds [16 x [16 x i32]], ptr [[TEMP]], i64 0, i64 [[INDVARS_IV]], i64 [[INDVARS_IV27]] 40; CHECK-DELIN-NEXT: [[TMP0:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4 41; CHECK-DELIN-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds [16 x [16 x i32]], ptr [[RES]], i64 0, i64 [[INDVARS_IV]], i64 [[INDVARS_IV27]] 42; CHECK-DELIN-NEXT: store i32 [[TMP0]], ptr [[ARRAYIDX8]], align 4 43; CHECK-DELIN-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1 44; CHECK-DELIN-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 45; CHECK-DELIN-NEXT: br label [[INNER_CRIT_EDGE:%.*]] 46; CHECK-DELIN: inner.for.body.split: 47; CHECK-DELIN-NEXT: [[WIDE_TRIP_COUNT_LCSSA:%.*]] = phi i64 [ [[WIDE_TRIP_COUNT]], [[OUTER_LATCH]] ], [ [[WIDE_TRIP_COUNT]], [[OUTER_HEADER]] ] 48; CHECK-DELIN-NEXT: [[TMP1]] = add nuw nsw i64 [[INDVARS_IV]], 1 49; CHECK-DELIN-NEXT: [[TMP2:%.*]] = icmp ne i64 [[TMP1]], [[WIDE_TRIP_COUNT_LCSSA]] 50; CHECK-DELIN-NEXT: br i1 [[TMP2]], label [[INNER_FOR_BODY]], label [[OUTER_CRIT_EDGE:%.*]] 51; CHECK-DELIN: inner.crit_edge: 52; CHECK-DELIN-NEXT: br label [[OUTER_LATCH]] 53; CHECK-DELIN: outer.latch: 54; CHECK-DELIN-NEXT: [[INDVARS_IV_NEXT28]] = add nuw nsw i64 [[INDVARS_IV27]], 1 55; CHECK-DELIN-NEXT: [[EXITCOND30:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT28]], [[WIDE_TRIP_COUNT29]] 56; CHECK-DELIN-NEXT: br i1 [[EXITCOND30]], label [[OUTER_HEADER]], label [[INNER_FOR_BODY_SPLIT]] 57; CHECK-DELIN: outer.crit_edge: 58; CHECK-DELIN-NEXT: br label [[FOR_COND_CLEANUP]] 59; CHECK-DELIN: for.cond.cleanup: 60; CHECK-DELIN-NEXT: ret void 61; 62entry: 63 %temp = alloca [16 x [16 x i32]], align 4 64 %res = alloca [16 x [16 x i32]], align 4 65 %cmp24 = icmp sgt i32 %n, 0 66 br i1 %cmp24, label %outer.preheader, label %for.cond.cleanup 67 68outer.preheader: ; preds = %entry 69 %wide.trip.count29 = zext i32 %n to i64 70 br label %outer.header 71 72outer.header: ; preds = %outer.preheader, %outer.latch 73 %indvars.iv27 = phi i64 [ 0, %outer.preheader ], [ %indvars.iv.next28, %outer.latch ] 74 %cmp222 = icmp sgt i32 %m, 0 75 br i1 %cmp222, label %inner.preheader, label %outer.latch 76 77inner.preheader: ; preds = %outer.header 78 ; When inner.preheader becomes the outer preheader, do not move 79 ; %wide.trip.count into the inner loop header lest LCSSA break 80 ; (if %wide.trip.count gets moved, its use is now outside the inner loop). 81 %wide.trip.count = zext i32 %m to i64 82 br label %inner.for.body 83 84inner.for.body: ; preds = %inner.preheader, %inner.for.body 85 %indvars.iv = phi i64 [ 0, %inner.preheader ], [ %indvars.iv.next, %inner.for.body ] 86 %arrayidx6 = getelementptr inbounds [16 x [16 x i32]], ptr %temp, i64 0, i64 %indvars.iv, i64 %indvars.iv27 87 %0 = load i32, ptr %arrayidx6, align 4 88 %arrayidx8 = getelementptr inbounds [16 x [16 x i32]], ptr %res, i64 0, i64 %indvars.iv, i64 %indvars.iv27 89 store i32 %0, ptr %arrayidx8, align 4 90 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 91 %exitcond = icmp ne i64 %indvars.iv.next, %wide.trip.count 92 br i1 %exitcond, label %inner.for.body, label %inner.crit_edge 93 94inner.crit_edge: ; preds = %inner.for.body 95 br label %outer.latch 96 97outer.latch: ; preds = %inner.crit_edge, %outer.header 98 %indvars.iv.next28 = add nuw nsw i64 %indvars.iv27, 1 99 %exitcond30 = icmp ne i64 %indvars.iv.next28, %wide.trip.count29 100 br i1 %exitcond30, label %outer.header, label %outer.crit_edge 101 102outer.crit_edge: ; preds = %outer.latch 103 br label %for.cond.cleanup 104 105for.cond.cleanup: ; preds = %outer.crit_edge, %entry 106 ret void 107} 108 109@global = external local_unnamed_addr global [4 x [4 x [2 x i16]]] align 16 110 111; %N.ext is defined in the outer loop header and used in the inner loop. After 112; interchanging, it will be defined in the new inner loop and used in the new; 113; outer latch, so we need to create a new LCSSA phi node for it. 114 115define void @test2(i32 %N) { 116; CHECK-LABEL: @test2( 117; CHECK-NEXT: bb: 118; CHECK-NEXT: br label [[INNER_PREHEADER:%.*]] 119; CHECK: outer.header.preheader: 120; CHECK-NEXT: br label [[OUTER_HEADER:%.*]] 121; CHECK: outer.header: 122; CHECK-NEXT: [[OUTER_IV:%.*]] = phi i64 [ [[OUTER_IV_NEXT:%.*]], [[OUTER_LATCH:%.*]] ], [ 0, [[OUTER_HEADER_PREHEADER:%.*]] ] 123; CHECK-NEXT: [[N_EXT:%.*]] = sext i32 [[N:%.*]] to i64 124; CHECK-NEXT: br label [[INNER_SPLIT1:%.*]] 125; CHECK: inner.preheader: 126; CHECK-NEXT: br label [[INNER:%.*]] 127; CHECK: inner: 128; CHECK-NEXT: [[INNER_IV:%.*]] = phi i64 [ [[TMP0:%.*]], [[INNER_SPLIT:%.*]] ], [ 0, [[INNER_PREHEADER]] ] 129; CHECK-NEXT: br label [[OUTER_HEADER_PREHEADER]] 130; CHECK: inner.split1: 131; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [4 x [4 x [2 x i16]]], ptr @global, i64 0, i64 [[INNER_IV]], i64 [[OUTER_IV]], i64 0 132; CHECK-NEXT: [[INNER_IV_NEXT:%.*]] = add nsw i64 [[INNER_IV]], 1 133; CHECK-NEXT: [[C_1:%.*]] = icmp ne i64 [[INNER_IV_NEXT]], [[N_EXT]] 134; CHECK-NEXT: br label [[OUTER_LATCH]] 135; CHECK: inner.split: 136; CHECK-NEXT: [[N_EXT_LCSSA:%.*]] = phi i64 [ [[N_EXT]], [[OUTER_LATCH]] ] 137; CHECK-NEXT: [[TMP0]] = add nsw i64 [[INNER_IV]], 1 138; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i64 [[TMP0]], [[N_EXT_LCSSA]] 139; CHECK-NEXT: br i1 [[TMP1]], label [[INNER]], label [[EXIT:%.*]] 140; CHECK: outer.latch: 141; CHECK-NEXT: [[OUTER_IV_NEXT]] = add nsw i64 [[OUTER_IV]], 1 142; CHECK-NEXT: [[C_2:%.*]] = icmp ne i64 [[OUTER_IV]], [[N_EXT]] 143; CHECK-NEXT: br i1 [[C_2]], label [[OUTER_HEADER]], label [[INNER_SPLIT]] 144; CHECK: exit: 145; CHECK-NEXT: ret void 146; 147bb: 148 br label %outer.header 149 150outer.header: ; preds = %bb11, %bb2 151 %outer.iv = phi i64 [ 0, %bb ], [ %outer.iv.next, %outer.latch ] 152 %N.ext = sext i32 %N to i64 153 br label %inner 154 155inner: ; preds = %bb6, %bb4 156 %inner.iv = phi i64 [ 0, %outer.header ], [ %inner.iv.next, %inner ] 157 %tmp8 = getelementptr inbounds [4 x [4 x [2 x i16]]], ptr @global, i64 0, i64 %inner.iv, i64 %outer.iv, i64 0 158 %inner.iv.next = add nsw i64 %inner.iv, 1 159 %c.1 = icmp ne i64 %inner.iv.next, %N.ext 160 br i1 %c.1, label %inner, label %outer.latch 161 162outer.latch: ; preds = %bb6 163 %outer.iv.next = add nsw i64 %outer.iv, 1 164 %c.2 = icmp ne i64 %outer.iv, %N.ext 165 br i1 %c.2 , label %outer.header, label %exit 166 167exit: ; preds = %bb11 168 ret void 169} 170