227 lines · plain
1; RUN: opt -mtriple=armv7 -mcpu=cortex-a57 -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=CHECK-NOUNROLL2; RUN: opt -mtriple=thumbv7 -mcpu=cortex-a57 -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=CHECK-NOUNROLL3; RUN: opt -mtriple=thumbv7 -mcpu=cortex-a72 -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=CHECK-NOUNROLL4; RUN: opt -mtriple=thumbv8m -mcpu=cortex-m23 -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=CHECK-UNROLL5; RUN: opt -mtriple=thumbv8m.main -mcpu=cortex-m33 -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=CHECK-UNROLL6; RUN: opt -mtriple=thumbv7em -mcpu=cortex-m7 -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=CHECK-UNROLL7 8; CHECK-LABEL: partial9define arm_aapcs_vfpcc void @partial(ptr nocapture %C, ptr nocapture readonly %A, ptr nocapture readonly %B) local_unnamed_addr #0 {10entry:11 br label %for.body12 13; CHECK-LABEL: for.body14for.body:15 16; CHECK-NOUNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, %entry ], [ [[IV2:%[a-z.0-9]+]], %for.body ]17; CHECK-NOUNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 118; CHECK-NOUNROLL: [[IV2]] = add nuw nsw i32 [[IV0]], 219; CHECK-NOUNROLL: [[CMP:%[a-z.0-9]+]] = icmp eq i32 [[IV2]], 102420; CHECK-NOUNROLL: br i1 [[CMP]], label [[END:%[a-z.]+]], label %for.body21 22; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, %entry ], [ [[IV16:%[a-z.0-9]+]], %for.body ]23; CHECK-UNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 124; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 225; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 326; CHECK-UNROLL: [[IV4:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 427; CHECK-UNROLL: [[IV5:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 528; CHECK-UNROLL: [[IV6:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 629; CHECK-UNROLL: [[IV7:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 730; CHECK-UNROLL: [[IV8:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 831; CHECK-UNROLL: [[IV9:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 932; CHECK-UNROLL: [[IV10:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1033; CHECK-UNROLL: [[IV11:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1134; CHECK-UNROLL: [[IV12:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1235; CHECK-UNROLL: [[IV13:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1336; CHECK-UNROLL: [[IV14:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1437; CHECK-UNROLL: [[IV15:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1538; CHECK-UNROLL: [[IV16]] = add nuw nsw i32 [[IV0]], 1639; CHECK-UNROLL: [[CMP:%[a-z.0-9]+]] = icmp eq i32 [[IV16]], 102440; CHECK-UNROLL: br i1 [[CMP]], label [[END:%[a-z.]+]], label %for.body41 42 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ]43 %arrayidx = getelementptr inbounds i32, ptr %A, i32 %i.0844 %0 = load i32, ptr %arrayidx, align 445 %arrayidx1 = getelementptr inbounds i32, ptr %B, i32 %i.0846 %1 = load i32, ptr %arrayidx1, align 447 %mul = mul nsw i32 %1, %048 %arrayidx2 = getelementptr inbounds i32, ptr %C, i32 %i.0849 store i32 %mul, ptr %arrayidx2, align 450 %inc = add nuw nsw i32 %i.08, 151 %exitcond = icmp eq i32 %inc, 102452 br i1 %exitcond, label %for.cond.cleanup, label %for.body53 54for.cond.cleanup:55 ret void56}57 58; CHECK-LABEL: runtime59define arm_aapcs_vfpcc void @runtime(ptr nocapture %C, ptr nocapture readonly %A, ptr nocapture readonly %B, i32 %N) local_unnamed_addr #0 {60entry:61 %cmp8 = icmp eq i32 %N, 062 br i1 %cmp8, label %for.cond.cleanup, label %for.body63 64; CHECK-LABEL: for.body65for.body:66; CHECK-NOUNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z.0-9]+]] ], [ [[IV2:%[a-z.0-9]+]], %for.body ]67; CHECK-NOUNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 168; CHECK-NOUNROLL: [[IV2]] = add nuw i32 [[IV0]], 269; CHECK-NOUNROLL: br70 71; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z.0-9]+]] ], [ [[IV4:%[a-z.0-9]+]], %for.body ]72; CHECK-UNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 173; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 274; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 375; CHECK-UNROLL: [[IV4]] = add nuw i32 [[IV0]], 476; CHECK-UNROLL: br77 78; CHECK-UNROLL: for.body.epil:79; CHECK-UNROLL: for.body.epil.1:80; CHECK-UNROLL: for.body.epil.2:81 82 %i.09 = phi i32 [ %inc, %for.body ], [ 0, %entry ]83 %arrayidx = getelementptr inbounds i32, ptr %A, i32 %i.0984 %0 = load i32, ptr %arrayidx, align 485 %arrayidx1 = getelementptr inbounds i32, ptr %B, i32 %i.0986 %1 = load i32, ptr %arrayidx1, align 487 %mul = mul nsw i32 %1, %088 %arrayidx2 = getelementptr inbounds i32, ptr %C, i32 %i.0989 store i32 %mul, ptr %arrayidx2, align 490 %inc = add nuw i32 %i.09, 191 %exitcond = icmp eq i32 %inc, %N92 br i1 %exitcond, label %for.cond.cleanup, label %for.body93 94for.cond.cleanup:95 ret void96}97 98; CHECK-LABEL: nested_runtime99define arm_aapcs_vfpcc void @nested_runtime(ptr nocapture %C, ptr nocapture readonly %A, ptr nocapture readonly %B, i32 %N) local_unnamed_addr #0 {100entry:101 %cmp25 = icmp eq i32 %N, 0102 br i1 %cmp25, label %for.cond.cleanup, label %for.body4.lr.ph103 104for.body4.lr.ph:105 %h.026 = phi i32 [ %inc11, %for.cond.cleanup3 ], [ 0, %entry ]106 %mul = mul i32 %h.026, %N107 br label %for.body4108 109for.cond.cleanup:110 ret void111 112for.cond.cleanup3:113 %inc11 = add nuw i32 %h.026, 1114 %exitcond27 = icmp eq i32 %inc11, %N115 br i1 %exitcond27, label %for.cond.cleanup, label %for.body4.lr.ph116 117; CHECK-LABEL: for.body4118for.body4:119; CHECK-NOUNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z0-9.]+]] ], [ [[IV1:%[a-z.0-9]+]], %for.body4 ]120; CHECK-NOUNROLL: [[IV1]] = add nuw i32 [[IV0]], 1121; CHECK-NOUNROLL: br122 123; CHECK-UNROLL: for.body4.epil:124; CHECK-UNROLL: for.body4.epil.1:125; CHECK-UNROLL: for.body4.epil.2:126; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z0-9.]+]] ], [ [[IV4:%[a-z.0-9]+]], %for.body4 ]127; CHECK-UNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1128; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 2129; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 3130; CHECK-UNROLL: [[IV4]] = add nuw i32 [[IV0]], 4131; CHECK-UNROLL: br132 133 %w.024 = phi i32 [ 0, %for.body4.lr.ph ], [ %inc, %for.body4 ]134 %add = add i32 %w.024, %mul135 %arrayidx = getelementptr inbounds i16, ptr %A, i32 %add136 %0 = load i16, ptr %arrayidx, align 2137 %conv = sext i16 %0 to i32138 %arrayidx5 = getelementptr inbounds i16, ptr %B, i32 %w.024139 %1 = load i16, ptr %arrayidx5, align 2140 %conv6 = sext i16 %1 to i32141 %mul7 = mul nsw i32 %conv6, %conv142 %arrayidx8 = getelementptr inbounds i32, ptr %C, i32 %w.024143 %2 = load i32, ptr %arrayidx8, align 4144 %add9 = add nsw i32 %mul7, %2145 store i32 %add9, ptr %arrayidx8, align 4146 %inc = add nuw i32 %w.024, 1147 %exitcond = icmp eq i32 %inc, %N148 br i1 %exitcond, label %for.cond.cleanup3, label %for.body4149}150 151; CHECK-LABEL: loop_call152define arm_aapcs_vfpcc void @loop_call(ptr nocapture %C, ptr nocapture readonly %A, ptr nocapture readonly %B) local_unnamed_addr #1 {153entry:154 br label %for.body155 156for.cond.cleanup:157 ret void158 159; CHECK-LABEL: for.body160for.body:161; CHECK-NOUNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, %entry ], [ [[IV1:%[a-z.0-9]+]], %for.body ]162; CHECK-NOUNROLL: [[IV1]] = add nuw nsw i32 [[IV0]], 1163; CHECK-NOUNROLL: icmp eq i32 [[IV1]], 1024164; CHECK-NOUNROLL: br165 166; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, %entry ], [ [[IV1:%[a-z.0-9]+]], %for.body ]167; CHECK-UNROLL: [[IV1]] = add nuw nsw i32 [[IV0]], 1168; CHECK-UNROLL: icmp eq i32 [[IV1]], 1024169; CHECK-UNROLL: br170 171 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ]172 %arrayidx = getelementptr inbounds i32, ptr %A, i32 %i.08173 %0 = load i32, ptr %arrayidx, align 4174 %arrayidx1 = getelementptr inbounds i32, ptr %B, i32 %i.08175 %1 = load i32, ptr %arrayidx1, align 4176 %call = tail call arm_aapcs_vfpcc i32 @some_func(i32 %0, i32 %1) #3177 %arrayidx2 = getelementptr inbounds i32, ptr %C, i32 %i.08178 store i32 %call, ptr %arrayidx2, align 4179 %inc = add nuw nsw i32 %i.08, 1180 %exitcond = icmp eq i32 %inc, 1024181 br i1 %exitcond, label %for.cond.cleanup, label %for.body182}183 184; CHECK-LABEL: iterate_inc185; CHECK-NOUNROLL: %n.addr.04 = phi ptr [ %1, %while.body ], [ %n, %while.body.preheader ]186; CHECK-NOUNROLL: %tobool = icmp eq ptr %1, null187; CHECK-NOUNROLL: br i1 %tobool188; CHECK-NOUNROLL-NOT: load189 190; CHECK-UNROLL: [[CMP0:%[a-z.0-9]+]] = icmp eq ptr [[VAR0:%[a-z.0-9]+]], null191; CHECK-UNROLL: br i1 [[CMP0]], label [[END:%[a-z.0-9]+]]192; CHECK-UNROLL: [[CMP1:%[a-z.0-9]+]] = icmp eq ptr [[VAR1:%[a-z.0-9]+]], null193; CHECK-UNROLL: br i1 [[CMP1]], label [[END]]194; CHECK-UNROLL: [[CMP2:%[a-z.0-9]+]] = icmp eq ptr [[VAR2:%[a-z.0-9]+]], null195; CHECK-UNROLL: br i1 [[CMP2]], label [[END]]196; CHECK-UNROLL: [[CMP3:%[a-z.0-9]+]] = icmp eq ptr [[VAR3:%[a-z.0-9]+]], null197; CHECK-UNROLL: br i1 [[CMP3]], label [[END]]198; CHECK-UNROLL: [[CMP4:%[a-z.0-9]+]] = icmp eq ptr [[VAR4:%[a-z.0-9]+]], null199; CHECK-UNROLL: br i1 [[CMP4]], label [[END]]200; CHECK-UNROLL-NOT: load201 202%struct.Node = type { ptr, i32 }203 204define arm_aapcscc void @iterate_inc(ptr %n) local_unnamed_addr #0 {205entry:206 %tobool3 = icmp eq ptr %n, null207 br i1 %tobool3, label %while.end, label %while.body.preheader208 209while.body.preheader:210 br label %while.body211 212while.body:213 %n.addr.04 = phi ptr [ %1, %while.body ], [ %n, %while.body.preheader ]214 %val = getelementptr inbounds %struct.Node, ptr %n.addr.04, i32 0, i32 1215 %0 = load i32, ptr %val, align 4216 %add = add nsw i32 %0, 1217 store i32 %add, ptr %val, align 4218 %1 = load ptr, ptr %n.addr.04, align 4219 %tobool = icmp eq ptr %1, null220 br i1 %tobool, label %while.end, label %while.body221 222while.end:223 ret void224}225 226declare arm_aapcs_vfpcc i32 @some_func(i32, i32) local_unnamed_addr #2227