Home | History | Annotate | Download | only in AArch64
      1 ; RUN: llc < %s
      2 
      3 ; This used to assert with "Overran sorted position" in AssignTopologicalOrder
      4 ; due to a cycle created in performPostLD1Combine.
      5 
      6 target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
      7 target triple = "arm64-apple-ios7.0.0"
      8 
      9 ; Function Attrs: nounwind ssp
     10 define void @f(double* %P1) #0 {
     11 entry:
     12   %arrayidx4 = getelementptr inbounds double, double* %P1, i64 1
     13   %0 = load double, double* %arrayidx4, align 8, !tbaa !1
     14   %1 = load double, double* %P1, align 8, !tbaa !1
     15   %2 = insertelement <2 x double> undef, double %0, i32 0
     16   %3 = insertelement <2 x double> %2, double %1, i32 1
     17   %4 = fsub <2 x double> zeroinitializer, %3
     18   %5 = fmul <2 x double> undef, %4
     19   %6 = extractelement <2 x double> %5, i32 0
     20   %cmp168 = fcmp olt double %6, undef
     21   br i1 %cmp168, label %if.then172, label %return
     22 
     23 if.then172:                                       ; preds = %cond.end90
     24   %7 = tail call i64 @llvm.objectsize.i64.p0i8(i8* undef, i1 false)
     25   br label %return
     26 
     27 return:                                           ; preds = %if.then172, %cond.end90, %entry
     28   ret void
     29 }
     30 
     31 ; Avoid an assert/bad codegen in LD1LANEPOST lowering by not forming
     32 ; LD1LANEPOST ISD nodes with a non-constant lane index.
     33 define <4 x i32> @f2(i32 *%p, <4 x i1> %m, <4 x i32> %v1, <4 x i32> %v2, i32 %idx) {
     34   %L0 = load i32, i32* %p
     35   %p1 = getelementptr i32, i32* %p, i64 1
     36   %L1 = load i32, i32* %p1
     37   %v = select <4 x i1> %m, <4 x i32> %v1, <4 x i32> %v2
     38   %vret = insertelement <4 x i32> %v, i32 %L0, i32 %idx
     39   store i32 %L1, i32 *%p
     40   ret <4 x i32> %vret
     41 }
     42 
     43 ; Check that a cycle is avoided during isel between the LD1LANEPOST instruction and the load of %L1.
     44 define <4 x i32> @f3(i32 *%p, <4 x i1> %m, <4 x i32> %v1, <4 x i32> %v2) {
     45   %L0 = load i32, i32* %p
     46   %p1 = getelementptr i32, i32* %p, i64 1
     47   %L1 = load i32, i32* %p1
     48   %v = select <4 x i1> %m, <4 x i32> %v1, <4 x i32> %v2
     49   %vret = insertelement <4 x i32> %v, i32 %L0, i32 %L1
     50   ret <4 x i32> %vret
     51 }
     52 
     53 ; Function Attrs: nounwind readnone
     54 declare i64 @llvm.objectsize.i64.p0i8(i8*, i1) #1
     55 
     56 attributes #0 = { nounwind ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
     57 attributes #1 = { nounwind readnone }
     58 
     59 !1 = !{!2, !2, i64 0}
     60 !2 = !{!"double", !3, i64 0}
     61 !3 = !{!"omnipotent char", !4, i64 0}
     62 !4 = !{!"Simple C/C++ TBAA"}
     63