152 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck %s3 4define void @signed(ptr %x, ptr %y, i32 %n) {5; CHECK-LABEL: @signed(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[X2:%.*]] = ptrtoint ptr [[X:%.*]] to i648; CHECK-NEXT: [[Y1:%.*]] = ptrtoint ptr [[Y:%.*]] to i649; CHECK-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 010; CHECK-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]11; CHECK: for.body.preheader:12; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i6413; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 414; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]15; CHECK: vector.memcheck:16; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[Y1]], [[X2]]17; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 1618; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]19; CHECK: vector.ph:20; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 421; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]22; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]23; CHECK: vector.body:24; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]25; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDEX]]26; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 427; CHECK-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> [[WIDE_LOAD]])28; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[Y]], i64 [[INDEX]]29; CHECK-NEXT: store <4 x i32> [[TMP4]], ptr [[TMP5]], align 430; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 431; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]32; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]33; CHECK: middle.block:34; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]35; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]36; CHECK: scalar.ph:37; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]38; CHECK-NEXT: br label [[FOR_BODY:%.*]]39; CHECK: for.cond.cleanup.loopexit:40; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]41; CHECK: for.cond.cleanup:42; CHECK-NEXT: ret void43; CHECK: for.body:44; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]45; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV]]46; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr [[ARRAYIDX]], align 447; CHECK-NEXT: [[TMP9:%.*]] = tail call i32 @llvm.fptosi.sat.i32.f32(float [[TMP8]])48; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[Y]], i64 [[INDVARS_IV]]49; CHECK-NEXT: store i32 [[TMP9]], ptr [[ARRAYIDX2]], align 450; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 151; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]52; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]53;54entry:55 %cmp6 = icmp sgt i32 %n, 056 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup57 58for.body.preheader: ; preds = %entry59 %wide.trip.count = zext i32 %n to i6460 br label %for.body61 62for.cond.cleanup: ; preds = %for.body, %entry63 ret void64 65for.body: ; preds = %for.body.preheader, %for.body66 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]67 %arrayidx = getelementptr inbounds float, ptr %x, i64 %indvars.iv68 %0 = load float, ptr %arrayidx, align 469 %1 = tail call i32 @llvm.fptosi.sat.i32.f32(float %0)70 %arrayidx2 = getelementptr inbounds i32, ptr %y, i64 %indvars.iv71 store i32 %1, ptr %arrayidx2, align 472 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 173 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count74 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body75}76 77define void @unsigned(ptr %x, ptr %y, i32 %n) {78; CHECK-LABEL: @unsigned(79; CHECK-NEXT: entry:80; CHECK-NEXT: [[X2:%.*]] = ptrtoint ptr [[X:%.*]] to i6481; CHECK-NEXT: [[Y1:%.*]] = ptrtoint ptr [[Y:%.*]] to i6482; CHECK-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 083; CHECK-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]84; CHECK: for.body.preheader:85; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i6486; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 487; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]88; CHECK: vector.memcheck:89; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[Y1]], [[X2]]90; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 1691; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]92; CHECK: vector.ph:93; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 494; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]95; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]96; CHECK: vector.body:97; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]98; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDEX]]99; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 4100; CHECK-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> [[WIDE_LOAD]])101; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[Y]], i64 [[INDEX]]102; CHECK-NEXT: store <4 x i32> [[TMP4]], ptr [[TMP5]], align 4103; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4104; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]105; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]106; CHECK: middle.block:107; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]108; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]109; CHECK: scalar.ph:110; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]111; CHECK-NEXT: br label [[FOR_BODY:%.*]]112; CHECK: for.cond.cleanup.loopexit:113; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]114; CHECK: for.cond.cleanup:115; CHECK-NEXT: ret void116; CHECK: for.body:117; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]118; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV]]119; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr [[ARRAYIDX]], align 4120; CHECK-NEXT: [[TMP9:%.*]] = tail call i32 @llvm.fptoui.sat.i32.f32(float [[TMP8]])121; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[Y]], i64 [[INDVARS_IV]]122; CHECK-NEXT: store i32 [[TMP9]], ptr [[ARRAYIDX2]], align 4123; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1124; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]125; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]126;127entry:128 %cmp6 = icmp sgt i32 %n, 0129 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup130 131for.body.preheader: ; preds = %entry132 %wide.trip.count = zext i32 %n to i64133 br label %for.body134 135for.cond.cleanup: ; preds = %for.body, %entry136 ret void137 138for.body: ; preds = %for.body.preheader, %for.body139 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]140 %arrayidx = getelementptr inbounds float, ptr %x, i64 %indvars.iv141 %0 = load float, ptr %arrayidx, align 4142 %1 = tail call i32 @llvm.fptoui.sat.i32.f32(float %0)143 %arrayidx2 = getelementptr inbounds i32, ptr %y, i64 %indvars.iv144 store i32 %1, ptr %arrayidx2, align 4145 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1146 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count147 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body148}149 150declare i32 @llvm.fptosi.sat.i32.f32(float)151declare i32 @llvm.fptoui.sat.i32.f32(float)152