1 ; RUN: llc -march=hexagon -mcpu=hexagonv5 -disable-hsdr < %s | FileCheck %s
3 ; Check that store is post-incremented.
4 ; CHECK: memuh(r{{[0-9]+}} + {{ *}}#6{{ *}})
5 ; CHECK: combine(r{{[0-9]+}}{{ *}},{{ *}}r{{[0-9]+}}{{ *}})
8 target datalayout = "e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:32:32-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a0:0-n16:32"
9 target triple = "hexagon"
11 define void @matrix_add_const(i32 %N, i16* nocapture %A, i16 signext %val) #0 {
13 %cmp5 = icmp eq i32 %N, 0
14 br i1 %cmp5, label %for.end, label %polly.cond
16 for.end.loopexit: ; preds = %polly.stmt.for.body29
19 for.end: ; preds = %for.end.loopexit, %polly.loop_header24.preheader, %entry
22 polly.cond: ; preds = %entry
23 %0 = icmp sgt i32 %N, 3
24 br i1 %0, label %polly.then, label %polly.loop_header24.preheader
26 polly.then: ; preds = %polly.cond
28 %leftover_lb = and i32 %1, -4
29 %2 = icmp sgt i32 %leftover_lb, 0
30 br i1 %2, label %polly.loop_body.lr.ph, label %polly.loop_header24.preheader
32 polly.loop_body.lr.ph: ; preds = %polly.then
33 %3 = insertelement <4 x i16> undef, i16 %val, i32 0
34 %4 = insertelement <4 x i16> %3, i16 %val, i32 1
35 %5 = insertelement <4 x i16> %4, i16 %val, i32 2
36 %6 = insertelement <4 x i16> %5, i16 %val, i32 3
37 br label %polly.loop_body
39 polly.loop_header24.preheader.loopexit: ; preds = %polly.loop_body
40 br label %polly.loop_header24.preheader
42 polly.loop_header24.preheader: ; preds = %polly.loop_header24.preheader.loopexit, %polly.then, %polly.cond
43 %polly.loopiv27.ph = phi i32 [ 0, %polly.cond ], [ %leftover_lb, %polly.then ], [ %leftover_lb, %polly.loop_header24.preheader.loopexit ]
44 %7 = icmp slt i32 %polly.loopiv27.ph, %N
45 br i1 %7, label %polly.stmt.for.body29.preheader, label %for.end
47 polly.stmt.for.body29.preheader: ; preds = %polly.loop_header24.preheader
48 br label %polly.stmt.for.body29
50 polly.loop_body: ; preds = %polly.loop_body.lr.ph, %polly.loop_body
51 %p_arrayidx.phi = phi i16* [ %A, %polly.loop_body.lr.ph ], [ %p_arrayidx.inc, %polly.loop_body ]
52 %polly.loopiv34 = phi i32 [ 0, %polly.loop_body.lr.ph ], [ %polly.next_loopiv, %polly.loop_body ]
53 %polly.next_loopiv = add nsw i32 %polly.loopiv34, 4
54 %vector_ptr = bitcast i16* %p_arrayidx.phi to <4 x i16>*
55 %_p_vec_full = load <4 x i16>, <4 x i16>* %vector_ptr, align 2
56 %addp_vec = add <4 x i16> %_p_vec_full, %6
57 store <4 x i16> %addp_vec, <4 x i16>* %vector_ptr, align 2
58 %8 = icmp slt i32 %polly.next_loopiv, %leftover_lb
59 %p_arrayidx.inc = getelementptr i16, i16* %p_arrayidx.phi, i32 4
60 br i1 %8, label %polly.loop_body, label %polly.loop_header24.preheader.loopexit
62 polly.stmt.for.body29: ; preds = %polly.stmt.for.body29.preheader, %polly.stmt.for.body29
63 %polly.loopiv2733 = phi i32 [ %polly.next_loopiv28, %polly.stmt.for.body29 ], [ %polly.loopiv27.ph, %polly.stmt.for.body29.preheader ]
64 %polly.next_loopiv28 = add nsw i32 %polly.loopiv2733, 1
65 %p_arrayidx30 = getelementptr i16, i16* %A, i32 %polly.loopiv2733
66 %_p_scalar_ = load i16, i16* %p_arrayidx30, align 2
67 %p_add = add i16 %_p_scalar_, %val
68 store i16 %p_add, i16* %p_arrayidx30, align 2
69 %exitcond = icmp eq i32 %polly.next_loopiv28, %N
70 br i1 %exitcond, label %for.end.loopexit, label %polly.stmt.for.body29
73 attributes #0 = { nounwind "fp-contract-model"="standard" "no-frame-pointer-elim-non-leaf" "realign-stack" "relocation-model"="static" "ssp-buffers-size"="8" }