• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -O3 -march=hexagon < %s | FileCheck %s
2; Test to ensure LSR does not optimize out addrec of the outerloop.
3; This will help to generate post-increment instructions, otherwise
4; it end up an as extra reg+reg add inside the loop.
5; CHECK:  loop0(.LBB0_[[LOOP:.]],
6; CHECK: .LBB0_[[LOOP]]:
7; CHECK: memuh{{.*}}++
8; CHECK: endloop
9
10
11define dso_local signext i16 @foo(i16* nocapture readonly %filt, i16* nocapture readonly %inp, i32 %c1, i32 %c2) local_unnamed_addr {
12entry:
13  %cmp28 = icmp sgt i32 %c1, 0
14  %cmp221 = icmp sgt i32 %c2, 0
15  %or.cond = and i1 %cmp28, %cmp221
16  br i1 %or.cond, label %for.cond1.preheader.us, label %for.cond.cleanup
17
18for.cond1.preheader.us:                           ; preds = %entry, %for.cond1.for.cond.cleanup3_crit_edge.us
19  %filt.addr.032.us = phi i16* [ %scevgep, %for.cond1.for.cond.cleanup3_crit_edge.us ], [ %filt, %entry ]
20  %inp.addr.031.us = phi i16* [ %scevgep35, %for.cond1.for.cond.cleanup3_crit_edge.us ], [ %inp, %entry ]
21  %l.030.us = phi i32 [ %inc11.us, %for.cond1.for.cond.cleanup3_crit_edge.us ], [ 0, %entry ]
22  %sum0.029.us = phi i16 [ %add8.us, %for.cond1.for.cond.cleanup3_crit_edge.us ], [ 0, %entry ]
23  %scevgep = getelementptr i16, i16* %filt.addr.032.us, i32 %c2
24  br label %for.body4.us
25
26for.body4.us:                                     ; preds = %for.body4.us, %for.cond1.preheader.us
27  %z.025.us = phi i32 [ 0, %for.cond1.preheader.us ], [ %inc.us, %for.body4.us ]
28  %filt.addr.124.us = phi i16* [ %filt.addr.032.us, %for.cond1.preheader.us ], [ %incdec.ptr.us, %for.body4.us ]
29  %inp.addr.123.us = phi i16* [ %inp.addr.031.us, %for.cond1.preheader.us ], [ %incdec.ptr5.us, %for.body4.us ]
30  %sum0.122.us = phi i16 [ %sum0.029.us, %for.cond1.preheader.us ], [ %add8.us, %for.body4.us ]
31  %incdec.ptr.us = getelementptr inbounds i16, i16* %filt.addr.124.us, i32 1
32  %0 = load i16, i16* %filt.addr.124.us, align 2
33  %incdec.ptr5.us = getelementptr inbounds i16, i16* %inp.addr.123.us, i32 1
34  %1 = load i16, i16* %inp.addr.123.us, align 2
35  %add.us = add i16 %0, %sum0.122.us
36  %add8.us = add i16 %add.us, %1
37  %inc.us = add nuw nsw i32 %z.025.us, 1
38  %exitcond = icmp eq i32 %inc.us, %c2
39  br i1 %exitcond, label %for.cond1.for.cond.cleanup3_crit_edge.us, label %for.body4.us
40
41for.cond1.for.cond.cleanup3_crit_edge.us:         ; preds = %for.body4.us
42  %scevgep35 = getelementptr i16, i16* %inp.addr.031.us, i32 %c2
43  %inc11.us = add nuw nsw i32 %l.030.us, 1
44  %exitcond36 = icmp eq i32 %inc11.us, %c1
45  br i1 %exitcond36, label %for.cond.cleanup, label %for.cond1.preheader.us
46
47for.cond.cleanup:                                 ; preds = %for.cond1.for.cond.cleanup3_crit_edge.us, %entry
48  %sum0.0.lcssa = phi i16 [ 0, %entry ], [ %add8.us, %for.cond1.for.cond.cleanup3_crit_edge.us ]
49  ret i16 %sum0.0.lcssa
50}
51