299 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 52; RUN: opt < %s -passes=loop-vectorize,dce,instcombine -force-vector-interleave=1 -force-vector-width=4 -enable-if-conversion -S | FileCheck %s3 4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"5 6; This is the loop in this example:7;8;int function0(int *a, int *b, int start, int end) {9;10; for (int i=start; i<end; ++i) {11; unsigned k = a[i];12;13; if (a[i] > b[i]) <------ notice the IF inside the loop.14; k = k * 5 + 3;15;16; a[i] = k; <---- K is a phi node that becomes vector-select.17; }18;}19 20define void @function0(ptr nocapture %a, ptr nocapture %b, i32 %start, i32 %end) nounwind uwtable ssp {21; CHECK-LABEL: define void @function0(22; CHECK-SAME: ptr captures(none) [[A:%.*]], ptr captures(none) [[B:%.*]], i32 [[START:%.*]], i32 [[END:%.*]]) #[[ATTR0:[0-9]+]] {23; CHECK-NEXT: [[ENTRY:.*:]]24; CHECK-NEXT: [[CMP16:%.*]] = icmp slt i32 [[START]], [[END]]25; CHECK-NEXT: br i1 [[CMP16]], label %[[FOR_BODY_LR_PH:.*]], label %[[FOR_END:.*]]26; CHECK: [[FOR_BODY_LR_PH]]:27; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[START]] to i6428; CHECK-NEXT: [[TMP1:%.*]] = xor i32 [[START]], -129; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[END]], [[TMP1]]30; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i6431; CHECK-NEXT: [[TMP4:%.*]] = add nuw nsw i64 [[TMP3]], 132; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP2]], 333; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]34; CHECK: [[VECTOR_MEMCHECK]]:35; CHECK-NEXT: [[TMP5:%.*]] = shl nsw i64 [[TMP0]], 236; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP5]]37; CHECK-NEXT: [[TMP6:%.*]] = xor i32 [[START]], -138; CHECK-NEXT: [[TMP7:%.*]] = add i32 [[END]], [[TMP6]]39; CHECK-NEXT: [[TMP8:%.*]] = zext i32 [[TMP7]] to i6440; CHECK-NEXT: [[TMP9:%.*]] = add nsw i64 [[TMP0]], [[TMP8]]41; CHECK-NEXT: [[TMP10:%.*]] = shl nsw i64 [[TMP9]], 242; CHECK-NEXT: [[TMP11:%.*]] = add nsw i64 [[TMP10]], 443; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP11]]44; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]]45; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP11]]46; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]]47; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]]48; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]49; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]50; CHECK: [[VECTOR_PH]]:51; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP4]], 858993458852; CHECK-NEXT: [[TMP12:%.*]] = add nsw i64 [[N_VEC]], [[TMP0]]53; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]54; CHECK: [[VECTOR_BODY]]:55; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]56; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 [[INDEX]], [[TMP0]]57; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[OFFSET_IDX]]58; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP13]], align 4, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]59; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[OFFSET_IDX]]60; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i32>, ptr [[TMP14]], align 4, !alias.scope [[META3]]61; CHECK-NEXT: [[DOTNOT:%.*]] = icmp sgt <4 x i32> [[WIDE_LOAD]], [[WIDE_LOAD4]]62; CHECK-NEXT: [[TMP15:%.*]] = mul <4 x i32> [[WIDE_LOAD]], splat (i32 5)63; CHECK-NEXT: [[TMP16:%.*]] = add <4 x i32> [[TMP15]], splat (i32 3)64; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[DOTNOT]], <4 x i32> [[TMP16]], <4 x i32> [[WIDE_LOAD]]65; CHECK-NEXT: store <4 x i32> [[PREDPHI]], ptr [[TMP13]], align 4, !alias.scope [[META0]], !noalias [[META3]]66; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 467; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]68; CHECK-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]69; CHECK: [[MIDDLE_BLOCK]]:70; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP4]], [[N_VEC]]71; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END_LOOPEXIT:.*]], label %[[SCALAR_PH]]72; CHECK: [[SCALAR_PH]]:73; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP12]], %[[MIDDLE_BLOCK]] ], [ [[TMP0]], %[[FOR_BODY_LR_PH]] ], [ [[TMP0]], %[[VECTOR_MEMCHECK]] ]74; CHECK-NEXT: br label %[[FOR_BODY:.*]]75; CHECK: [[FOR_BODY]]:76; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[IF_END:.*]] ]77; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]]78; CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[ARRAYIDX]], align 479; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDVARS_IV]]80; CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[ARRAYIDX4]], align 481; CHECK-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]82; CHECK-NEXT: br i1 [[CMP5]], label %[[IF_THEN:.*]], label %[[IF_END]]83; CHECK: [[IF_THEN]]:84; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TMP18]], 585; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], 386; CHECK-NEXT: br label %[[IF_END]]87; CHECK: [[IF_END]]:88; CHECK-NEXT: [[K_0:%.*]] = phi i32 [ [[ADD]], %[[IF_THEN]] ], [ [[TMP18]], %[[FOR_BODY]] ]89; CHECK-NEXT: store i32 [[K_0]], ptr [[ARRAYIDX]], align 490; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 191; CHECK-NEXT: [[TMP20:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i3292; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[END]], [[TMP20]]93; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP8:![0-9]+]]94; CHECK: [[FOR_END_LOOPEXIT]]:95; CHECK-NEXT: br label %[[FOR_END]]96; CHECK: [[FOR_END]]:97; CHECK-NEXT: ret void98;99entry:100 %cmp16 = icmp slt i32 %start, %end101 br i1 %cmp16, label %for.body.lr.ph, label %for.end102 103for.body.lr.ph:104 %0 = sext i32 %start to i64105 br label %for.body106 107for.body:108 %indvars.iv = phi i64 [ %0, %for.body.lr.ph ], [ %indvars.iv.next, %if.end ]109 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv110 %1 = load i32, ptr %arrayidx, align 4111 %arrayidx4 = getelementptr inbounds i32, ptr %b, i64 %indvars.iv112 %2 = load i32, ptr %arrayidx4, align 4113 %cmp5 = icmp sgt i32 %1, %2114 br i1 %cmp5, label %if.then, label %if.end115 116if.then:117 %mul = mul i32 %1, 5118 %add = add i32 %mul, 3119 br label %if.end120 121if.end:122 %k.0 = phi i32 [ %add, %if.then ], [ %1, %for.body ]123 store i32 %k.0, ptr %arrayidx, align 4124 %indvars.iv.next = add i64 %indvars.iv, 1125 %3 = trunc i64 %indvars.iv.next to i32126 %cmp = icmp slt i32 %3, %end127 br i1 %cmp, label %for.body, label %for.end128 129for.end:130 ret void131}132 133 134 135; int func(int *A, int n) {136; unsigned sum = 0;137; for (int i = 0; i < n; ++i)138; if (A[i] > 30)139; sum += A[i] + 2;140;141; return sum;142; }143 144define i32 @reduction_func(ptr nocapture %A, i32 %n) nounwind uwtable readonly ssp {145; CHECK-LABEL: define i32 @reduction_func(146; CHECK-SAME: ptr captures(none) [[A:%.*]], i32 [[N:%.*]]) #[[ATTR1:[0-9]+]] {147; CHECK-NEXT: [[ENTRY:.*]]:148; CHECK-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[N]], 0149; CHECK-NEXT: br i1 [[CMP10]], label %[[FOR_BODY_PREHEADER:.*]], label %[[FOR_END:.*]]150; CHECK: [[FOR_BODY_PREHEADER]]:151; CHECK-NEXT: [[TMP0:%.*]] = zext nneg i32 [[N]] to i64152; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4153; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]154; CHECK: [[VECTOR_PH]]:155; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP0]], 2147483644156; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]157; CHECK: [[VECTOR_BODY]]:158; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]159; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[PREDPHI:%.*]], %[[VECTOR_BODY]] ]160; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]]161; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP1]], align 4162; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt <4 x i32> [[WIDE_LOAD]], splat (i32 30)163; CHECK-NEXT: [[TMP3:%.*]] = add <4 x i32> [[VEC_PHI]], splat (i32 2)164; CHECK-NEXT: [[TMP4:%.*]] = add <4 x i32> [[TMP3]], [[WIDE_LOAD]]165; CHECK-NEXT: [[PREDPHI]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP4]], <4 x i32> [[VEC_PHI]]166; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4167; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]168; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]169; CHECK: [[MIDDLE_BLOCK]]:170; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PREDPHI]])171; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP0]]172; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END_LOOPEXIT:.*]], label %[[SCALAR_PH]]173; CHECK: [[SCALAR_PH]]:174; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]175; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP6]], %[[MIDDLE_BLOCK]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]176; CHECK-NEXT: br label %[[FOR_BODY:.*]]177; CHECK: [[FOR_BODY]]:178; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_INC:.*]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]179; CHECK-NEXT: [[SUM_011:%.*]] = phi i32 [ [[SUM_1:%.*]], %[[FOR_INC]] ], [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ]180; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]]181; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX]], align 4182; CHECK-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[TMP7]], 30183; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[FOR_INC]]184; CHECK: [[IF_THEN]]:185; CHECK-NEXT: [[ADD:%.*]] = add i32 [[SUM_011]], 2186; CHECK-NEXT: [[ADD4:%.*]] = add i32 [[ADD]], [[TMP7]]187; CHECK-NEXT: br label %[[FOR_INC]]188; CHECK: [[FOR_INC]]:189; CHECK-NEXT: [[SUM_1]] = phi i32 [ [[ADD4]], %[[IF_THEN]] ], [ [[SUM_011]], %[[FOR_BODY]] ]190; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1191; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32192; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[N]], [[LFTR_WIDEIV]]193; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]194; CHECK: [[FOR_END_LOOPEXIT]]:195; CHECK-NEXT: [[SUM_1_LCSSA:%.*]] = phi i32 [ [[SUM_1]], %[[FOR_INC]] ], [ [[TMP6]], %[[MIDDLE_BLOCK]] ]196; CHECK-NEXT: br label %[[FOR_END]]197; CHECK: [[FOR_END]]:198; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[SUM_1_LCSSA]], %[[FOR_END_LOOPEXIT]] ]199; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]]200;201entry:202 %cmp10 = icmp sgt i32 %n, 0203 br i1 %cmp10, label %for.body, label %for.end204 205for.body: ; preds = %entry, %for.inc206 %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]207 %sum.011 = phi i32 [ %sum.1, %for.inc ], [ 0, %entry ]208 %arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv209 %0 = load i32, ptr %arrayidx, align 4210 %cmp1 = icmp sgt i32 %0, 30211 br i1 %cmp1, label %if.then, label %for.inc212 213if.then: ; preds = %for.body214 %add = add i32 %sum.011, 2215 %add4 = add i32 %add, %0216 br label %for.inc217 218for.inc: ; preds = %for.body, %if.then219 %sum.1 = phi i32 [ %add4, %if.then ], [ %sum.011, %for.body ]220 %indvars.iv.next = add i64 %indvars.iv, 1221 %lftr.wideiv = trunc i64 %indvars.iv.next to i32222 %exitcond = icmp eq i32 %lftr.wideiv, %n223 br i1 %exitcond, label %for.end, label %for.body224 225for.end: ; preds = %for.inc, %entry226 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %sum.1, %for.inc ]227 ret i32 %sum.0.lcssa228}229 230@a = common global [1 x ptr] zeroinitializer, align 8231@c = common global ptr null, align 8232 233; We use to if convert this loop. This is not safe because there is a trapping234; constant expression.235; PR16729236 237; Handle PHI with single incoming value having a full mask.238; PR34523239 240define void @PR34523(ptr %p, i16 %val) {241; CHECK-LABEL: define void @PR34523(242; CHECK-SAME: ptr [[P:%.*]], i16 [[VAL:%.*]]) {243; CHECK-NEXT: [[BB1:.*]]:244; CHECK-NEXT: [[TMP0:%.*]] = add i16 [[VAL]], 1245; CHECK-NEXT: [[SMAX:%.*]] = call i16 @llvm.smax.i16(i16 [[TMP0]], i16 2)246; CHECK-NEXT: [[TMP1:%.*]] = xor i16 [[VAL]], -1247; CHECK-NEXT: [[TMP2:%.*]] = add i16 [[SMAX]], [[TMP1]]248; CHECK-NEXT: [[TMP3:%.*]] = zext i16 [[TMP2]] to i32249; CHECK-NEXT: [[TMP4:%.*]] = add nuw nsw i32 [[TMP3]], 1250; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i16 [[TMP2]], 3251; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]252; CHECK: [[VECTOR_PH]]:253; CHECK-NEXT: [[N_VEC:%.*]] = and i32 [[TMP4]], 131068254; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i16255; CHECK-NEXT: [[TMP5:%.*]] = add i16 [[VAL]], [[DOTCAST]]256; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]257; CHECK: [[VECTOR_BODY]]:258; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]259; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4260; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]261; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]262; CHECK: [[MIDDLE_BLOCK]]:263; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP4]], [[N_VEC]]264; CHECK-NEXT: br i1 [[CMP_N]], label %[[BB5:.*]], label %[[SCALAR_PH]]265; CHECK: [[SCALAR_PH]]:266; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ [[TMP5]], %[[MIDDLE_BLOCK]] ], [ [[VAL]], %[[BB1]] ]267; CHECK-NEXT: br label %[[BB2:.*]]268; CHECK: [[BB2]]:269; CHECK-NEXT: [[I:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[_TMP2:%.*]], %[[BB4:.*]] ]270; CHECK-NEXT: br label %[[BB3:.*]]271; CHECK: [[BB3]]:272; CHECK-NEXT: br label %[[BB4]]273; CHECK: [[BB4]]:274; CHECK-NEXT: [[_TMP2]] = add i16 [[I]], 1275; CHECK-NEXT: [[_TMP3:%.*]] = icmp slt i16 [[_TMP2]], 2276; CHECK-NEXT: br i1 [[_TMP3]], label %[[BB2]], label %[[BB5]], !llvm.loop [[LOOP12:![0-9]+]]277; CHECK: [[BB5]]:278; CHECK-NEXT: unreachable279;280bb1:281 br label %bb2282 283bb2: ; preds = %bb4, %bb1284 %i = phi i16 [ %val, %bb1 ], [ %_tmp2, %bb4 ]285 br label %bb3286 287bb3: ; preds = %bb2288 %_tmp1 = phi ptr [ %p, %bb2 ]289 br label %bb4290 291bb4: ; preds = %bb3292 %_tmp2 = add i16 %i, 1293 %_tmp3 = icmp slt i16 %_tmp2, 2294 br i1 %_tmp3, label %bb2, label %bb5295 296bb5: ; preds = %bb4297 unreachable298}299