75 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -mtriple=aarch64-unknown -mattr=+sve -S %s | FileCheck %s3 4define ptr @test(ptr %start.1, ptr %start.2, ptr %end) {5; CHECK-LABEL: @test(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[START_22:%.*]] = ptrtoint ptr [[START_2:%.*]] to i648; CHECK-NEXT: [[END1:%.*]] = ptrtoint ptr [[END:%.*]] to i649; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[END1]], -810; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[START_22]]11; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 312; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 113; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()14; CHECK-NEXT: [[TMP5:%.*]] = shl nuw i64 [[TMP4]], 215; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], [[TMP5]]16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]17; CHECK: vector.ph:18; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()19; CHECK-NEXT: [[TMP7:%.*]] = mul nuw i64 [[TMP6]], 420; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], [[TMP7]]21; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]22; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[N_VEC]], 823; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START_1:%.*]], i64 [[TMP8]]24; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[N_VEC]], 825; CHECK-NEXT: [[IND_END3:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[TMP9]]26; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]27; CHECK: vector.body:28; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]29; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 830; CHECK-NEXT: [[TMP30:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[OFFSET_IDX]]31; CHECK-NEXT: [[TMP33:%.*]] = call i64 @llvm.vscale.i64()32; CHECK-NEXT: [[TMP34:%.*]] = shl nuw i64 [[TMP33]], 133; CHECK-NEXT: [[TMP35:%.*]] = getelementptr i64, ptr [[TMP30]], i64 [[TMP34]]34; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[TMP30]], align 835; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[TMP35]], align 836; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP7]]37; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]38; CHECK-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]39; CHECK: middle.block:40; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]41; CHECK-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[IND_END]], i64 -842; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]43; CHECK: scalar.ph:44; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START_1]], [[ENTRY:%.*]] ]45; CHECK-NEXT: [[BC_RESUME_VAL4:%.*]] = phi ptr [ [[IND_END3]], [[MIDDLE_BLOCK]] ], [ [[START_2]], [[ENTRY]] ]46; CHECK-NEXT: br label [[LOOP:%.*]]47; CHECK: loop:48; CHECK-NEXT: [[IV_1:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]49; CHECK-NEXT: [[IV_2:%.*]] = phi ptr [ [[BC_RESUME_VAL4]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]50; CHECK-NEXT: store i64 0, ptr [[IV_2]], align 851; CHECK-NEXT: [[IV_2_NEXT]] = getelementptr inbounds ptr, ptr [[IV_2]], i64 152; CHECK-NEXT: [[IV_1_NEXT]] = getelementptr inbounds ptr, ptr [[IV_1]], i64 153; CHECK-NEXT: [[CMP_I_I_NOT_I:%.*]] = icmp eq ptr [[IV_2_NEXT]], [[END]]54; CHECK-NEXT: br i1 [[CMP_I_I_NOT_I]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]55; CHECK: exit:56; CHECK-NEXT: [[RES_LCSSA:%.*]] = phi ptr [ [[IV_1]], [[LOOP]] ], [ [[IND_ESCAPE]], [[MIDDLE_BLOCK]] ]57; CHECK-NEXT: ret ptr [[RES_LCSSA]]58;59entry:60 br label %loop61 62loop:63 %iv.1 = phi ptr [ %start.1, %entry ], [ %iv.1.next, %loop ]64 %iv.2 = phi ptr [ %start.2, %entry ], [ %iv.2.next, %loop ]65 store i64 0, ptr %iv.266 %iv.2.next = getelementptr inbounds ptr, ptr %iv.2, i64 167 %iv.1.next = getelementptr inbounds ptr, ptr %iv.1, i64 168 %cmp.i.i.not.i = icmp eq ptr %iv.2.next, %end69 br i1 %cmp.i.i.not.i, label %exit, label %loop70 71exit:72 %res.lcssa = phi ptr [ %iv.1, %loop ]73 ret ptr %res.lcssa74}75