267 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt < %s -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=2 -S | FileCheck %s3 4define void @test1(ptr noalias nocapture %a, ptr noalias nocapture readonly %b) {5; CHECK-LABEL: define void @test1(6; CHECK-SAME: ptr noalias captures(none) [[A:%.*]], ptr noalias readonly captures(none) [[B:%.*]]) {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: [[TMP0:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]]14; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, ptr [[TMP0]], i64 415; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP0]], align 416; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x float>, ptr [[TMP1]], align 417; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0:![0-9]+]])18; CHECK-NEXT: [[TMP2:%.*]] = fadd <4 x float> [[WIDE_LOAD]], splat (float 1.000000e+00)19; CHECK-NEXT: [[TMP3:%.*]] = fadd <4 x float> [[WIDE_LOAD1]], splat (float 1.000000e+00)20; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]]21; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, ptr [[TMP4]], i64 422; CHECK-NEXT: store <4 x float> [[TMP2]], ptr [[TMP4]], align 423; CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP5]], align 424; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 825; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 160026; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]27; CHECK: [[MIDDLE_BLOCK]]:28; CHECK-NEXT: br label %[[FOR_END:.*]]29; CHECK: [[FOR_END]]:30; CHECK-NEXT: ret void31;32entry:33 br label %for.body34 35 36for.body: ; preds = %for.body, %entry37 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]38 %arrayidx = getelementptr inbounds float, ptr %b, i64 %indvars.iv39 %0 = load float, ptr %arrayidx, align 440 %cmp1 = fcmp ogt float %0, 1.000000e+0241 tail call void @llvm.experimental.noalias.scope.decl(metadata !0)42 %add = fadd float %0, 1.000000e+0043 %arrayidx5 = getelementptr inbounds float, ptr %a, i64 %indvars.iv44 store float %add, ptr %arrayidx5, align 445 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 146 %exitcond = icmp eq i64 %indvars.iv, 159947 br i1 %exitcond, label %for.end, label %for.body48 49for.end: ; preds = %for.body50 ret void51}52 53declare void @llvm.experimental.noalias.scope.decl(metadata)54 55%struct.data = type { ptr, ptr }56 57define void @test2(ptr nocapture readonly %d) {58; CHECK-LABEL: define void @test2(59; CHECK-SAME: ptr readonly captures(none) [[D:%.*]]) {60; CHECK-NEXT: [[ENTRY:.*:]]61; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_DATA:%.*]], ptr [[D]], i64 0, i32 162; CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B]], align 863; CHECK-NEXT: [[PTRINT:%.*]] = ptrtoint ptr [[TMP0]] to i6464; CHECK-NEXT: [[MASKEDPTR:%.*]] = and i64 [[PTRINT]], 3165; CHECK-NEXT: [[MASKCOND:%.*]] = icmp eq i64 [[MASKEDPTR]], 066; CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[D]], align 867; CHECK-NEXT: [[PTRINT2:%.*]] = ptrtoint ptr [[TMP1]] to i6468; CHECK-NEXT: [[MASKEDPTR3:%.*]] = and i64 [[PTRINT2]], 3169; CHECK-NEXT: [[MASKCOND4:%.*]] = icmp eq i64 [[MASKEDPTR3]], 070; CHECK-NEXT: br label %[[VECTOR_MEMCHECK:.*]]71; CHECK: [[VECTOR_MEMCHECK]]:72; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[PTRINT2]], [[PTRINT]]73; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP2]], 3274; CHECK-NEXT: br i1 [[DIFF_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]75; CHECK: [[VECTOR_PH]]:76; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]77; CHECK: [[VECTOR_BODY]]:78; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]79; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0]])80; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, ptr [[TMP0]], i64 [[INDEX]]81; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[TMP3]], i64 482; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP3]], align 483; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x float>, ptr [[TMP4]], align 484; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x float> [[WIDE_LOAD]], splat (float 1.000000e+00)85; CHECK-NEXT: [[TMP6:%.*]] = fadd <4 x float> [[WIDE_LOAD1]], splat (float 1.000000e+00)86; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META6:![0-9]+]])87; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i64 [[INDEX]]88; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[TMP7]], i64 489; CHECK-NEXT: store <4 x float> [[TMP5]], ptr [[TMP7]], align 490; CHECK-NEXT: store <4 x float> [[TMP6]], ptr [[TMP8]], align 491; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 892; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 160093; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]94; CHECK: [[MIDDLE_BLOCK]]:95; CHECK-NEXT: br label %[[FOR_END:.*]]96; CHECK: [[SCALAR_PH]]:97; CHECK-NEXT: br label %[[FOR_BODY:.*]]98; CHECK: [[FOR_BODY]]:99; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ]100; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0]])101; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[TMP0]], i64 [[INDVARS_IV]]102; CHECK-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX]], align 4103; CHECK-NEXT: [[ADD:%.*]] = fadd float [[TMP10]], 1.000000e+00104; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META6]])105; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i64 [[INDVARS_IV]]106; CHECK-NEXT: store float [[ADD]], ptr [[ARRAYIDX5]], align 4107; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1108; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV]], 1599109; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]110; CHECK: [[FOR_END]]:111; CHECK-NEXT: ret void112;113entry:114 %b = getelementptr inbounds %struct.data, ptr %d, i64 0, i32 1115 %0 = load ptr, ptr %b, align 8116 %ptrint = ptrtoint ptr %0 to i64117 %maskedptr = and i64 %ptrint, 31118 %maskcond = icmp eq i64 %maskedptr, 0119 %1 = load ptr, ptr %d, align 8120 %ptrint2 = ptrtoint ptr %1 to i64121 %maskedptr3 = and i64 %ptrint2, 31122 %maskcond4 = icmp eq i64 %maskedptr3, 0123 br label %for.body124 125 126for.body: ; preds = %for.body, %entry127 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]128 tail call void @llvm.experimental.noalias.scope.decl(metadata !0)129 %arrayidx = getelementptr inbounds float, ptr %0, i64 %indvars.iv130 %2 = load float, ptr %arrayidx, align 4131 %add = fadd float %2, 1.000000e+00132 tail call void @llvm.experimental.noalias.scope.decl(metadata !4)133 %arrayidx5 = getelementptr inbounds float, ptr %1, i64 %indvars.iv134 store float %add, ptr %arrayidx5, align 4135 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1136 %exitcond = icmp eq i64 %indvars.iv, 1599137 br i1 %exitcond, label %for.end, label %for.body138 139for.end: ; preds = %for.body140 ret void141}142 143define void @predicated_noalias_scope_decl(ptr noalias nocapture readonly %a, ptr noalias nocapture %b, i32 %n) {144; Check that the vector.body still contains a llvm.experimental.noalias.scope.decl145; CHECK-LABEL: define void @predicated_noalias_scope_decl(146; CHECK-SAME: ptr noalias readonly captures(none) [[A:%.*]], ptr noalias captures(none) [[B:%.*]], i32 [[N:%.*]]) {147; CHECK-NEXT: [[ENTRY:.*:]]148; CHECK-NEXT: [[CMP15:%.*]] = icmp eq i32 [[N]], 0149; CHECK-NEXT: br i1 [[CMP15]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]150; CHECK: [[FOR_BODY_PREHEADER]]:151; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64152; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 8153; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]154; CHECK: [[VECTOR_PH]]:155; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 8156; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]157; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]158; CHECK: [[VECTOR_BODY]]:159; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]160; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]161; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)162; CHECK-NEXT: [[TMP1:%.*]] = icmp ult <4 x i64> [[VEC_IND]], splat (i64 495616)163; CHECK-NEXT: [[TMP2:%.*]] = icmp ult <4 x i64> [[STEP_ADD]], splat (i64 495616)164; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0]])165; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP1]], <4 x float> splat (float 2.300000e+01), <4 x float> splat (float 4.200000e+01)166; CHECK-NEXT: [[PREDPHI1:%.*]] = select <4 x i1> [[TMP2]], <4 x float> splat (float 2.300000e+01), <4 x float> splat (float 4.200000e+01)167; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]]168; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[TMP3]], i64 4169; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP3]], align 4170; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x float>, ptr [[TMP4]], align 4171; CHECK-NEXT: [[TMP5:%.*]] = fmul <4 x float> [[PREDPHI]], [[WIDE_LOAD]]172; CHECK-NEXT: [[TMP6:%.*]] = fmul <4 x float> [[PREDPHI1]], [[WIDE_LOAD2]]173; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]]174; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[TMP7]], i64 4175; CHECK-NEXT: store <4 x float> [[TMP5]], ptr [[TMP7]], align 4176; CHECK-NEXT: store <4 x float> [[TMP6]], ptr [[TMP8]], align 4177; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8178; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[STEP_ADD]], splat (i64 4)179; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]180; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]181; CHECK: [[MIDDLE_BLOCK]]:182; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]183; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]], label %[[SCALAR_PH]]184; CHECK: [[SCALAR_PH]]:185; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]186; CHECK-NEXT: br label %[[FOR_BODY:.*]]187; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT]]:188; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]]189; CHECK: [[FOR_COND_CLEANUP]]:190; CHECK-NEXT: ret void191; CHECK: [[FOR_BODY]]:192; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[IF_END5:.*]] ]193; CHECK-NEXT: [[CMP1:%.*]] = icmp ult i64 [[INDVARS_IV]], 495616194; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_END5]], label %[[IF_ELSE:.*]]195; CHECK: [[IF_ELSE]]:196; CHECK-NEXT: [[CMP2:%.*]] = icmp ult i64 [[INDVARS_IV]], 991232197; CHECK-NEXT: tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0]])198; CHECK-NEXT: br label %[[IF_END5]]199; CHECK: [[IF_END5]]:200; CHECK-NEXT: [[X_0:%.*]] = phi float [ 4.200000e+01, %[[IF_ELSE]] ], [ 2.300000e+01, %[[FOR_BODY]] ]201; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDVARS_IV]]202; CHECK-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX]], align 4203; CHECK-NEXT: [[MUL:%.*]] = fmul float [[X_0]], [[TMP10]]204; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDVARS_IV]]205; CHECK-NEXT: store float [[MUL]], ptr [[ARRAYIDX7]], align 4206; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1207; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[TMP0]]208; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]209;210 211entry:212 %cmp15 = icmp eq i32 %n, 0213 br i1 %cmp15, label %for.cond.cleanup, label %for.body.preheader214 215for.body.preheader: ; preds = %entry216 %0 = zext i32 %n to i64217 br label %for.body218 219for.cond.cleanup.loopexit: ; preds = %if.end5220 br label %for.cond.cleanup221 222for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry223 ret void224 225for.body: ; preds = %for.body.preheader, %if.end5226 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %if.end5 ]227 %cmp1 = icmp ult i64 %indvars.iv, 495616228 br i1 %cmp1, label %if.end5, label %if.else229 230if.else: ; preds = %for.body231 %cmp2 = icmp ult i64 %indvars.iv, 991232232 tail call void @llvm.experimental.noalias.scope.decl(metadata !0)233 br label %if.end5234 235if.end5: ; preds = %for.body, %if.else236 %x.0 = phi float [ 4.200000e+01, %if.else ], [ 2.300000e+01, %for.body ]237 %arrayidx = getelementptr inbounds float, ptr %a, i64 %indvars.iv238 %1 = load float, ptr %arrayidx, align 4239 %mul = fmul float %x.0, %1240 %arrayidx7 = getelementptr inbounds float, ptr %b, i64 %indvars.iv241 store float %mul, ptr %arrayidx7, align 4242 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1243 %cmp = icmp eq i64 %indvars.iv.next, %0244 br i1 %cmp, label %for.cond.cleanup.loopexit, label %for.body245}246 247!0 = !{ !1 }248!1 = distinct !{ !1, !2 }249!2 = distinct !{ !2 }250!3 = distinct !{ !3, !2 }251!4 = !{ !3 }252 253;.254; CHECK: [[META0]] = !{[[META1:![0-9]+]]}255; CHECK: [[META1]] = distinct !{[[META1]], [[META2:![0-9]+]]}256; CHECK: [[META2]] = distinct !{[[META2]]}257; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META4:![0-9]+]], [[META5:![0-9]+]]}258; CHECK: [[META4]] = !{!"llvm.loop.isvectorized", i32 1}259; CHECK: [[META5]] = !{!"llvm.loop.unroll.runtime.disable"}260; CHECK: [[META6]] = !{[[META7:![0-9]+]]}261; CHECK: [[META7]] = distinct !{[[META7]], [[META2]]}262; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META4]], [[META5]]}263; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META4]]}264; CHECK: [[LOOP10]] = distinct !{[[LOOP10]], [[META4]], [[META5]]}265; CHECK: [[LOOP11]] = distinct !{[[LOOP11]], [[META5]], [[META4]]}266;.267