Home | History | Annotate | Download | only in LoopStrengthReduce
      1 ; RUN: opt -loop-reduce -S < %s | FileCheck %s
      2 ;
      3 ; <rdar://10619599> "SelectionDAGBuilder shouldn't visit PHI nodes!" assert.
      4 ; <rdar://10655343> SCEVExpander segfault on simple test case
      5 
      6 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-f128:128:128-n8:16:32"
      7 target triple = "i386-apple-darwin"
      8 
      9 ; LSR should convert the inner loop (bb7.us) IV (j.01.us) into float*.
     10 ; This involves a nested AddRec, the outer AddRec's loop invariant components
     11 ; cannot find a preheader, so they should be expanded in the loop header
     12 ; (bb7.lr.ph.us) below the existing phi i.12.us.
     13 ; Currently, LSR won't kick in on such loops.
     14 ; CHECK: @nopreheader
     15 ; CHECK: bb7.us:
     16 ; CHECK-NOT: phi float*
     17 ; CHECK: %j.01.us = phi i32
     18 ; CHECK-NOT: phi float*
     19 define void @nopreheader(float* nocapture %a, i32 %n) nounwind {
     20 entry:
     21   %0 = sdiv i32 %n, undef
     22   indirectbr i8* undef, [label %bb10.preheader]
     23 
     24 bb10.preheader:                                   ; preds = %bb4
     25   indirectbr i8* undef, [label %bb8.preheader.lr.ph, label %return]
     26 
     27 bb8.preheader.lr.ph:                              ; preds = %bb10.preheader
     28   indirectbr i8* null, [label %bb7.lr.ph.us, label %bb9]
     29 
     30 bb7.lr.ph.us:                                     ; preds = %bb9.us, %bb8.preheader.lr.ph
     31   %i.12.us = phi i32 [ %2, %bb9.us ], [ 0, %bb8.preheader.lr.ph ]
     32   %tmp30 = mul i32 %0, %i.12.us
     33   indirectbr i8* undef, [label %bb7.us]
     34 
     35 bb7.us:                                           ; preds = %bb7.lr.ph.us, %bb7.us
     36   %j.01.us = phi i32 [ 0, %bb7.lr.ph.us ], [ %1, %bb7.us ]
     37   %tmp31 = add i32 %tmp30, %j.01.us
     38   %scevgep9 = getelementptr float* %a, i32 %tmp31
     39   store float undef, float* %scevgep9, align 1
     40   %1 = add nsw i32 %j.01.us, 1
     41   indirectbr i8* undef, [label %bb9.us, label %bb7.us]
     42 
     43 bb9.us:                                           ; preds = %bb7.us
     44   %2 = add nsw i32 %i.12.us, 1
     45   indirectbr i8* undef, [label %bb7.lr.ph.us, label %return]
     46 
     47 bb9:                                              ; preds = %bb9, %bb8.preheader.lr.ph
     48   indirectbr i8* undef, [label %bb9, label %return]
     49 
     50 return:                                           ; preds = %bb9, %bb9.us, %bb10.preheader
     51   ret void
     52 }
     53 
     54 ; In this case, SCEVExpander simply cannot materialize the AddRecExpr
     55 ; that LSR picks. We must detect that %bb8.preheader does not have a
     56 ; preheader and avoid performing LSR on %bb7.
     57 ; CHECK: @nopreheader2
     58 ; CHECK: bb7:
     59 ; CHECK: %indvar = phi i32
     60 define fastcc void @nopreheader2([200 x i32]* nocapture %Array2) nounwind {
     61 entry:
     62   indirectbr i8* undef, [label %bb]
     63 
     64 bb:                                               ; preds = %bb, %entry
     65   indirectbr i8* undef, [label %bb3, label %bb]
     66 
     67 bb3:                                              ; preds = %bb3, %bb
     68   indirectbr i8* undef, [label %bb8.preheader, label %bb3]
     69 
     70 bb8.preheader:                                    ; preds = %bb9, %bb3
     71   %indvar5 = phi i32 [ %indvar.next6, %bb9 ], [ 0, %bb3 ]
     72   %tmp26 = add i32 %indvar5, 13
     73   indirectbr i8* null, [label %bb7]
     74 
     75 bb7:                                              ; preds = %bb8.preheader, %bb7
     76   %indvar = phi i32 [ 0, %bb8.preheader ], [ %indvar.next, %bb7 ]
     77   %scevgep = getelementptr [200 x i32]* %Array2, i32 %tmp26, i32 %indvar
     78   store i32 undef, i32* %scevgep, align 4
     79   %indvar.next = add i32 %indvar, 1
     80   indirectbr i8* undef, [label %bb9, label %bb7]
     81 
     82 bb9:                                              ; preds = %bb7
     83   %indvar.next6 = add i32 %indvar5, 1
     84   indirectbr i8* undef, [label %return, label %bb8.preheader]
     85 
     86 return:                                           ; preds = %bb9
     87   ret void
     88 }
     89