141 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 62; RUN: opt -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck %s3 4define i32 @preserve_inbounds(i64 %start, ptr %ptr) {5; CHECK-LABEL: define i32 @preserve_inbounds(6; CHECK-SAME: i64 [[START:%.*]], ptr [[PTR:%.*]]) {7; CHECK-NEXT: [[ENTRY:.*:]]8; CHECK-NEXT: br label %[[VECTOR_PH:.*]]9; CHECK: [[VECTOR_PH]]:10; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]11; CHECK: [[VECTOR_BODY]]:12; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]13; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]14; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 [[START]], [[INDEX]]15; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], -116; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP0]]17; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 018; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i64 -319; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP3]], align 420; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>21; CHECK-NEXT: [[TMP4]] = add <4 x i32> [[REVERSE]], [[VEC_PHI]]22; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 423; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 102424; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]25; CHECK: [[MIDDLE_BLOCK]]:26; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])27; CHECK-NEXT: br label %[[END:.*]]28; CHECK: [[END]]:29; CHECK-NEXT: ret i32 [[TMP6]]30;31entry:32 br label %loop33 34loop:35 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]36 %rev.ind = phi i64 [ %start, %entry ], [ %rev.ind.next, %loop ]37 %redux = phi i32 [ 0, %entry ], [ %redux.next, %loop ]38 %rev.ind.next = add i64 %rev.ind, -139 %gep.ptr.ind = getelementptr inbounds i32, ptr %ptr, i64 %rev.ind.next40 %ld.ptr = load i32, ptr %gep.ptr.ind, align 441 %redux.next = add i32 %ld.ptr, %redux42 %iv.next = add i32 %iv, 143 %exit.cond = icmp ne i32 %iv.next, 102444 br i1 %exit.cond, label %loop, label %end45 46end:47 ret i32 %redux.next48}49 50define i32 @preserve_nusw(i64 %start, ptr %ptr) {51; CHECK-LABEL: define i32 @preserve_nusw(52; CHECK-SAME: i64 [[START:%.*]], ptr [[PTR:%.*]]) {53; CHECK-NEXT: [[ENTRY:.*:]]54; CHECK-NEXT: br label %[[VECTOR_PH:.*]]55; CHECK: [[VECTOR_PH]]:56; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]57; CHECK: [[VECTOR_BODY]]:58; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]59; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]60; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 [[START]], [[INDEX]]61; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], -162; CHECK-NEXT: [[TMP1:%.*]] = getelementptr nusw i32, ptr [[PTR]], i64 [[TMP0]]63; CHECK-NEXT: [[TMP2:%.*]] = getelementptr nusw i32, ptr [[TMP1]], i64 064; CHECK-NEXT: [[TMP3:%.*]] = getelementptr nusw i32, ptr [[TMP2]], i64 -365; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP3]], align 466; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>67; CHECK-NEXT: [[TMP4]] = add <4 x i32> [[REVERSE]], [[VEC_PHI]]68; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 469; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 102470; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]71; CHECK: [[MIDDLE_BLOCK]]:72; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])73; CHECK-NEXT: br label %[[END:.*]]74; CHECK: [[END]]:75; CHECK-NEXT: ret i32 [[TMP6]]76;77entry:78 br label %loop79 80loop:81 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]82 %rev.ind = phi i64 [ %start, %entry ], [ %rev.ind.next, %loop ]83 %redux = phi i32 [ 0, %entry ], [ %redux.next, %loop ]84 %rev.ind.next = add i64 %rev.ind, -185 %gep.ptr.ind = getelementptr nusw i32, ptr %ptr, i64 %rev.ind.next86 %ld.ptr = load i32, ptr %gep.ptr.ind, align 487 %redux.next = add i32 %ld.ptr, %redux88 %iv.next = add i32 %iv, 189 %exit.cond = icmp ne i32 %iv.next, 102490 br i1 %exit.cond, label %loop, label %end91 92end:93 ret i32 %redux.next94}95 96define i32 @drop_nuw(i64 %start, ptr %ptr) {97; CHECK-LABEL: define i32 @drop_nuw(98; CHECK-SAME: i64 [[START:%.*]], ptr [[PTR:%.*]]) {99; CHECK-NEXT: [[ENTRY:.*:]]100; CHECK-NEXT: br label %[[VECTOR_PH:.*]]101; CHECK: [[VECTOR_PH]]:102; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]103; CHECK: [[VECTOR_BODY]]:104; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]105; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]106; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 [[START]], [[INDEX]]107; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], -1108; CHECK-NEXT: [[TMP1:%.*]] = getelementptr nuw i32, ptr [[PTR]], i64 [[TMP0]]109; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], i64 0110; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i32, ptr [[TMP2]], i64 -3111; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4112; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>113; CHECK-NEXT: [[TMP4]] = add <4 x i32> [[REVERSE]], [[VEC_PHI]]114; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4115; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024116; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]117; CHECK: [[MIDDLE_BLOCK]]:118; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])119; CHECK-NEXT: br label %[[END:.*]]120; CHECK: [[END]]:121; CHECK-NEXT: ret i32 [[TMP6]]122;123entry:124 br label %loop125 126loop:127 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]128 %rev.ind = phi i64 [ %start, %entry ], [ %rev.ind.next, %loop ]129 %redux = phi i32 [ 0, %entry ], [ %redux.next, %loop ]130 %rev.ind.next = add i64 %rev.ind, -1131 %gep.ptr.ind = getelementptr nuw i32, ptr %ptr, i64 %rev.ind.next132 %ld.ptr = load i32, ptr %gep.ptr.ind, align 4133 %redux.next = add i32 %ld.ptr, %redux134 %iv.next = add i32 %iv, 1135 %exit.cond = icmp ne i32 %iv.next, 1024136 br i1 %exit.cond, label %loop, label %end137 138end:139 ret i32 %redux.next140}141