117 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 62; RUN: opt -p loop-vectorize -S %s | FileCheck %s3 4define void @low_trip_count_small(i32 %x, ptr %dst) {5; CHECK-LABEL: define void @low_trip_count_small(6; CHECK-SAME: i32 [[X:%.*]], ptr [[DST:%.*]]) {7; CHECK-NEXT: [[ENTRY:.*:]]8; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[X]], i32 1)9; CHECK-NEXT: [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[SMAX]], i32 4)10; CHECK-NEXT: [[TMP0:%.*]] = zext nneg i32 [[SMAX]] to i6411; CHECK-NEXT: [[UMIN1:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP0]], i64 4)12; CHECK-NEXT: br label %[[VECTOR_PH:.*]]13; CHECK: [[VECTOR_PH]]:14; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[UMIN1]], 115; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 016; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer17; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]18; CHECK: [[VECTOR_BODY]]:19; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[DST]], i64 020; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[DST]], i64 121; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[DST]], i64 222; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[DST]], i64 323; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x ptr> poison, ptr [[NEXT_GEP]], i32 024; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x ptr> [[TMP1]], ptr [[NEXT_GEP2]], i32 125; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x ptr> [[TMP2]], ptr [[NEXT_GEP3]], i32 226; CHECK-NEXT: [[TMP4:%.*]] = insertelement <4 x ptr> [[TMP3]], ptr [[NEXT_GEP4]], i32 327; CHECK-NEXT: [[TMP5:%.*]] = icmp ule <4 x i64> <i64 0, i64 1, i64 2, i64 3>, [[BROADCAST_SPLAT]]28; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i1> [[TMP5]], i32 029; CHECK-NEXT: br i1 [[TMP6]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]]30; CHECK: [[PRED_STORE_IF]]:31; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 132; CHECK-NEXT: store i8 0, ptr [[TMP7]], align 133; CHECK-NEXT: br label %[[PRED_STORE_CONTINUE]]34; CHECK: [[PRED_STORE_CONTINUE]]:35; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i1> [[TMP5]], i32 136; CHECK-NEXT: br i1 [[TMP8]], label %[[PRED_STORE_IF5:.*]], label %[[PRED_STORE_CONTINUE6:.*]]37; CHECK: [[PRED_STORE_IF5]]:38; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i64 139; CHECK-NEXT: store i8 0, ptr [[TMP9]], align 140; CHECK-NEXT: br label %[[PRED_STORE_CONTINUE6]]41; CHECK: [[PRED_STORE_CONTINUE6]]:42; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP5]], i32 243; CHECK-NEXT: br i1 [[TMP10]], label %[[PRED_STORE_IF7:.*]], label %[[PRED_STORE_CONTINUE8:.*]]44; CHECK: [[PRED_STORE_IF7]]:45; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP3]], i64 146; CHECK-NEXT: store i8 0, ptr [[TMP11]], align 147; CHECK-NEXT: br label %[[PRED_STORE_CONTINUE8]]48; CHECK: [[PRED_STORE_CONTINUE8]]:49; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x i1> [[TMP5]], i32 350; CHECK-NEXT: br i1 [[TMP12]], label %[[PRED_STORE_IF9:.*]], label %[[PRED_STORE_CONTINUE10:.*]]51; CHECK: [[PRED_STORE_IF9]]:52; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[NEXT_GEP4]], i64 153; CHECK-NEXT: store i8 0, ptr [[TMP13]], align 154; CHECK-NEXT: br label %[[PRED_STORE_CONTINUE10]]55; CHECK: [[PRED_STORE_CONTINUE10]]:56; CHECK-NEXT: br label %[[MIDDLE_BLOCK:.*]]57; CHECK: [[MIDDLE_BLOCK]]:58; CHECK-NEXT: br label %[[EXIT:.*]]59; CHECK: [[EXIT]]:60; CHECK-NEXT: ret void61;62entry:63 %smax = call i32 @llvm.smax.i32(i32 %x, i32 1)64 %umin = call i32 @llvm.umin.i32(i32 %smax, i32 4)65 br label %loop66 67loop: ; preds = %loop, %entry68 %ptr = phi ptr [ %dst, %entry ], [ %ptr.next, %loop ]69 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]70 %ptr.next = getelementptr i8, ptr %ptr, i64 171 store i8 0, ptr %ptr.next72 %iv.next = add i32 %iv, 173 %exitcond = icmp eq i32 %iv.next, %umin74 br i1 %exitcond, label %exit, label %loop75 76exit:77 ret void78}79 80 81define ptr @low_trip_count_small_with_live_out(i32 %x, ptr %dst) {82; CHECK-LABEL: define ptr @low_trip_count_small_with_live_out(83; CHECK-SAME: i32 [[X:%.*]], ptr [[DST:%.*]]) {84; CHECK-NEXT: [[ENTRY:.*]]:85; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[X]], i32 1)86; CHECK-NEXT: [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[SMAX]], i32 4)87; CHECK-NEXT: br label %[[LOOP:.*]]88; CHECK: [[LOOP]]:89; CHECK-NEXT: [[PTR:%.*]] = phi ptr [ [[DST]], %[[ENTRY]] ], [ [[PTR_NEXT:%.*]], %[[LOOP]] ]90; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]91; CHECK-NEXT: [[PTR_NEXT]] = getelementptr i8, ptr [[PTR]], i64 192; CHECK-NEXT: store i8 0, ptr [[PTR_NEXT]], align 193; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 194; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], [[UMIN]]95; CHECK-NEXT: br i1 [[EXITCOND]], label %[[EXIT:.*]], label %[[LOOP]]96; CHECK: [[EXIT]]:97; CHECK-NEXT: [[PTR_NEXT_LCSSA:%.*]] = phi ptr [ [[PTR_NEXT]], %[[LOOP]] ]98; CHECK-NEXT: ret ptr [[PTR_NEXT_LCSSA]]99;100entry:101 %smax = call i32 @llvm.smax.i32(i32 %x, i32 1)102 %umin = call i32 @llvm.umin.i32(i32 %smax, i32 4)103 br label %loop104 105loop:106 %ptr = phi ptr [ %dst, %entry ], [ %ptr.next, %loop ]107 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]108 %ptr.next = getelementptr i8, ptr %ptr, i64 1109 store i8 0, ptr %ptr.next110 %iv.next = add i32 %iv, 1111 %exitcond = icmp eq i32 %iv.next, %umin112 br i1 %exitcond, label %exit, label %loop113 114exit:115 ret ptr %ptr.next116}117