228 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 -force-vector-width=4 -S %s | FileCheck %s3 4define void @call_loop_invariant_operand_bundle(ptr %dst, {float, float} %sv) {5; CHECK-LABEL: define void @call_loop_invariant_operand_bundle(6; CHECK-SAME: ptr [[DST:%.*]], { float, float } [[SV:%.*]]) {7; CHECK-NEXT: [[ENTRY:.*:]]8; CHECK-NEXT: br label %[[VECTOR_PH:.*]]9; CHECK: [[VECTOR_PH]]:10; CHECK-NEXT: [[TMP0:%.*]] = extractvalue { float, float } [[SV]], 011; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[TMP0]], i64 012; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer13; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { float, float } [[SV]], 114; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x float> poison, float [[TMP1]], i64 015; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT1]], <4 x float> poison, <4 x i32> zeroinitializer16; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]17; CHECK: [[VECTOR_BODY]]:18; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]19; CHECK-NEXT: [[TMP2:%.*]] = getelementptr float, ptr [[DST]], i32 [[INDEX]]20; CHECK-NEXT: [[TMP3:%.*]] = call <4 x float> @llvm.pow.v4f32(<4 x float> [[BROADCAST_SPLAT]], <4 x float> [[BROADCAST_SPLAT2]]) [ "deopt"(float 1.000000e+01) ]21; CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP2]], align 422; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 423; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 100024; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]25; CHECK: [[MIDDLE_BLOCK]]:26; CHECK-NEXT: br label %[[EXIT:.*]]27; CHECK: [[EXIT]]:28; CHECK-NEXT: ret void29;30entry:31 br label %loop32 33loop:34 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]35 %a = extractvalue { float, float } %sv, 036 %b = extractvalue { float, float } %sv, 137 %addr = getelementptr float, ptr %dst, i32 %iv38 %p = call float @llvm.pow.f32(float %a, float %b) [ "deopt"(float 10.0) ]39 store float %p, ptr %addr40 %iv.next = add nsw i32 %iv, 141 %cond = icmp ne i32 %iv.next, 100042 br i1 %cond, label %loop, label %exit43 44exit:45 ret void46}47 48define void @call_unknown_operand_bundle(ptr %dst, {float, float} %sv) {49; CHECK-LABEL: define void @call_unknown_operand_bundle(50; CHECK-SAME: ptr [[DST:%.*]], { float, float } [[SV:%.*]]) {51; CHECK-NEXT: [[ENTRY:.*:]]52; CHECK-NEXT: br label %[[VECTOR_PH:.*]]53; CHECK: [[VECTOR_PH]]:54; CHECK-NEXT: [[TMP0:%.*]] = extractvalue { float, float } [[SV]], 055; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[TMP0]], i64 056; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer57; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { float, float } [[SV]], 158; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x float> poison, float [[TMP1]], i64 059; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT1]], <4 x float> poison, <4 x i32> zeroinitializer60; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]61; CHECK: [[VECTOR_BODY]]:62; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]63; CHECK-NEXT: [[TMP2:%.*]] = getelementptr float, ptr [[DST]], i32 [[INDEX]]64; CHECK-NEXT: [[TMP3:%.*]] = call <4 x float> @llvm.pow.v4f32(<4 x float> [[BROADCAST_SPLAT]], <4 x float> [[BROADCAST_SPLAT2]]) [ "unknown"(ptr null) ]65; CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP2]], align 466; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 467; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 100068; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]69; CHECK: [[MIDDLE_BLOCK]]:70; CHECK-NEXT: br label %[[EXIT:.*]]71; CHECK: [[EXIT]]:72; CHECK-NEXT: ret void73;74entry:75 br label %loop76 77loop:78 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]79 %a = extractvalue { float, float } %sv, 080 %b = extractvalue { float, float } %sv, 181 %addr = getelementptr float, ptr %dst, i32 %iv82 %p = call float @llvm.pow.f32(float %a, float %b) [ "unknown"(ptr null) ]83 store float %p, ptr %addr84 %iv.next = add nsw i32 %iv, 185 %cond = icmp ne i32 %iv.next, 100086 br i1 %cond, label %loop, label %exit87 88exit:89 ret void90}91 92define void @call_cold_operand_bundle(ptr %dst, {float, float} %sv) {93; CHECK-LABEL: define void @call_cold_operand_bundle(94; CHECK-SAME: ptr [[DST:%.*]], { float, float } [[SV:%.*]]) {95; CHECK-NEXT: [[ENTRY:.*:]]96; CHECK-NEXT: br label %[[VECTOR_PH:.*]]97; CHECK: [[VECTOR_PH]]:98; CHECK-NEXT: [[TMP0:%.*]] = extractvalue { float, float } [[SV]], 099; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[TMP0]], i64 0100; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer101; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { float, float } [[SV]], 1102; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x float> poison, float [[TMP1]], i64 0103; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT1]], <4 x float> poison, <4 x i32> zeroinitializer104; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]105; CHECK: [[VECTOR_BODY]]:106; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]107; CHECK-NEXT: [[TMP2:%.*]] = getelementptr float, ptr [[DST]], i32 [[INDEX]]108; CHECK-NEXT: [[TMP3:%.*]] = call <4 x float> @llvm.pow.v4f32(<4 x float> [[BROADCAST_SPLAT]], <4 x float> [[BROADCAST_SPLAT2]]) [ "cold"() ]109; CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP2]], align 4110; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4111; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000112; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]113; CHECK: [[MIDDLE_BLOCK]]:114; CHECK-NEXT: br label %[[EXIT:.*]]115; CHECK: [[EXIT]]:116; CHECK-NEXT: ret void117;118entry:119 br label %loop120 121loop:122 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]123 %a = extractvalue { float, float } %sv, 0124 %b = extractvalue { float, float } %sv, 1125 %addr = getelementptr float, ptr %dst, i32 %iv126 %p = call float @llvm.pow.f32(float %a, float %b) [ "cold"() ]127 store float %p, ptr %addr128 %iv.next = add nsw i32 %iv, 1129 %cond = icmp ne i32 %iv.next, 1000130 br i1 %cond, label %loop, label %exit131 132exit:133 ret void134}135 136define void @assume_loop_variant_operand_bundle(ptr noalias %a, ptr noalias %b) {137; CHECK-LABEL: define void @assume_loop_variant_operand_bundle(138; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]]) {139; CHECK-NEXT: [[ENTRY:.*:]]140; CHECK-NEXT: br label %[[VECTOR_PH:.*]]141; CHECK: [[VECTOR_PH]]:142; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]143; CHECK: [[VECTOR_BODY]]:144; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]145; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0146; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1147; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2148; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3149; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP0]]150; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP8]], align 4151; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "align"(ptr [[A]], i64 [[TMP0]]) ]152; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "align"(ptr [[A]], i64 [[TMP1]]) ]153; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "align"(ptr [[A]], i64 [[TMP2]]) ]154; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "align"(ptr [[A]], i64 [[TMP3]]) ]155; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x float> [[WIDE_LOAD]], splat (float 1.000000e+00)156; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP0]]157; CHECK-NEXT: store <4 x float> [[TMP5]], ptr [[TMP10]], align 4158; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4159; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1600160; CHECK-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]161; CHECK: [[MIDDLE_BLOCK]]:162; CHECK-NEXT: br label %[[EXIT:.*]]163; CHECK: [[EXIT]]:164; CHECK-NEXT: ret void165;166entry:167 br label %loop168 169loop:170 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]171 %arrayidx = getelementptr inbounds float, ptr %b, i64 %iv172 %0 = load float, ptr %arrayidx, align 4173 %cmp1 = fcmp ogt float %0, 1.000000e+02174 tail call void @llvm.assume(i1 true) [ "align"(ptr %a, i64 %iv) ]175 %add = fadd float %0, 1.000000e+00176 %arrayidx5 = getelementptr inbounds float, ptr %a, i64 %iv177 store float %add, ptr %arrayidx5, align 4178 %iv.next = add nuw nsw i64 %iv, 1179 %exitcond = icmp eq i64 %iv, 1599180 br i1 %exitcond, label %exit, label %loop181 182exit:183 ret void184}185 186define void @assume_cold_operand_bundle(ptr noalias %a, ptr noalias %b) {187; CHECK-LABEL: define void @assume_cold_operand_bundle(188; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]]) {189; CHECK-NEXT: [[ENTRY:.*:]]190; CHECK-NEXT: br label %[[VECTOR_PH:.*]]191; CHECK: [[VECTOR_PH]]:192; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "cold"() ]193; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]194; CHECK: [[VECTOR_BODY]]:195; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]196; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]]197; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP0]], align 4198; CHECK-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[WIDE_LOAD]], splat (float 1.000000e+00)199; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]]200; CHECK-NEXT: store <4 x float> [[TMP1]], ptr [[TMP2]], align 4201; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4202; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1600203; CHECK-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]204; CHECK: [[MIDDLE_BLOCK]]:205; CHECK-NEXT: br label %[[EXIT:.*]]206; CHECK: [[EXIT]]:207; CHECK-NEXT: ret void208;209entry:210 br label %loop211 212loop:213 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]214 %arrayidx = getelementptr inbounds float, ptr %b, i64 %iv215 %0 = load float, ptr %arrayidx, align 4216 %cmp1 = fcmp ogt float %0, 1.000000e+02217 tail call void @llvm.assume(i1 true) [ "cold"() ]218 %add = fadd float %0, 1.000000e+00219 %arrayidx5 = getelementptr inbounds float, ptr %a, i64 %iv220 store float %add, ptr %arrayidx5, align 4221 %iv.next = add nuw nsw i64 %iv, 1222 %exitcond = icmp eq i64 %iv, 1599223 br i1 %exitcond, label %exit, label %loop224 225exit:226 ret void227}228