1; RUN: llc -regalloc=greedy < %s | FileCheck %s 2 3; LSR shouldn't introduce more induction variables than needed, increasing 4; register pressure and therefore spilling. There is more room for improvement 5; here. 6 7; CHECK: sub sp, #{{32|28|24}} 8 9; CHECK: %for.inc 10; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, # 11; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, # 12; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, # 13; CHECK: add 14 15target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:32:64-v128:32:128-a0:0:32-n32" 16target triple = "thumbv7-apple-macosx10.7.0" 17 18%struct.partition_entry = type { i32, i32, i64, i64 } 19 20define i32 @partition_overlap_check(%struct.partition_entry* nocapture %part, i32 %num_entries) nounwind readonly optsize ssp { 21entry: 22 %cmp79 = icmp sgt i32 %num_entries, 0 23 br i1 %cmp79, label %outer.loop, label %for.end72 24 25outer.loop: ; preds = %for.inc69, %entry 26 %overlap.081 = phi i32 [ %overlap.4, %for.inc69 ], [ 0, %entry ] 27 %0 = phi i32 [ %inc71, %for.inc69 ], [ 0, %entry ] 28 %offset = getelementptr %struct.partition_entry* %part, i32 %0, i32 2 29 %len = getelementptr %struct.partition_entry* %part, i32 %0, i32 3 30 %tmp5 = load i64* %offset, align 4, !tbaa !0 31 %tmp15 = load i64* %len, align 4, !tbaa !0 32 %add = add nsw i64 %tmp15, %tmp5 33 br label %inner.loop 34 35inner.loop: ; preds = %for.inc, %outer.loop 36 %overlap.178 = phi i32 [ %overlap.081, %outer.loop ], [ %overlap.4, %for.inc ] 37 %1 = phi i32 [ 0, %outer.loop ], [ %inc, %for.inc ] 38 %cmp23 = icmp eq i32 %0, %1 39 br i1 %cmp23, label %for.inc, label %if.end 40 41if.end: ; preds = %inner.loop 42 %len39 = getelementptr %struct.partition_entry* %part, i32 %1, i32 3 43 %offset28 = getelementptr %struct.partition_entry* %part, i32 %1, i32 2 44 %tmp29 = load i64* %offset28, align 4, !tbaa !0 45 %tmp40 = load i64* %len39, align 4, !tbaa !0 46 %add41 = add nsw i64 %tmp40, %tmp29 47 %cmp44 = icmp sge i64 %tmp29, %tmp5 48 %cmp47 = icmp slt i64 %tmp29, %add 49 %or.cond = and i1 %cmp44, %cmp47 50 %overlap.2 = select i1 %or.cond, i32 1, i32 %overlap.178 51 %cmp52 = icmp sle i64 %add41, %add 52 %cmp56 = icmp sgt i64 %add41, %tmp5 53 %or.cond74 = and i1 %cmp52, %cmp56 54 %overlap.3 = select i1 %or.cond74, i32 1, i32 %overlap.2 55 %cmp61 = icmp sgt i64 %tmp29, %tmp5 56 %cmp65 = icmp slt i64 %add41, %add 57 %or.cond75 = or i1 %cmp61, %cmp65 58 br i1 %or.cond75, label %for.inc, label %if.then66 59 60if.then66: ; preds = %if.end 61 br label %for.inc 62 63for.inc: ; preds = %if.end, %if.then66, %inner.loop 64 %overlap.4 = phi i32 [ %overlap.178, %inner.loop ], [ 1, %if.then66 ], [ %overlap.3, %if.end ] 65 %inc = add nsw i32 %1, 1 66 %exitcond = icmp eq i32 %inc, %num_entries 67 br i1 %exitcond, label %for.inc69, label %inner.loop 68 69for.inc69: ; preds = %for.inc 70 %inc71 = add nsw i32 %0, 1 71 %exitcond83 = icmp eq i32 %inc71, %num_entries 72 br i1 %exitcond83, label %for.end72, label %outer.loop 73 74for.end72: ; preds = %for.inc69, %entry 75 %overlap.0.lcssa = phi i32 [ 0, %entry ], [ %overlap.4, %for.inc69 ] 76 ret i32 %overlap.0.lcssa 77} 78 79!0 = metadata !{metadata !"long long", metadata !1} 80!1 = metadata !{metadata !"omnipotent char", metadata !2} 81!2 = metadata !{metadata !"Simple C/C++ TBAA", null} 82