1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
| ; RUN: llc -march=hexagon -O2 -mcpu=hexagonv60 -hexagon-initial-cfg-cleanup=0 --stats -o - 2>&1 < %s | FileCheck %s
; This was aborting while processing SUnits.
; REQUIRES: asserts
; CHECK: vmem
; CHECK-NOT: Number of node order issues found
; CHECK: Number of loops software pipelined
; CHECK-NOT: Number of node order issues found
target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
target triple = "hexagon-unknown--elf"
declare <16 x i32> @llvm.hexagon.V6.lo(<32 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #0
declare <32 x i32> @llvm.hexagon.V6.vshuffvdd(<16 x i32>, <16 x i32>, i32) #0
declare <32 x i32> @llvm.hexagon.V6.vdealvdd(<16 x i32>, <16 x i32>, i32) #0
declare <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32>, <16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.vshufeh(<16 x i32>, <16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32>, <16 x i32>) #0
declare <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32>, <16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32>, <16 x i32>, i32) #0
define void @f0() #1 {
b0:
%v0 = load i16*, i16** undef, align 4
%v1 = load i32*, i32** undef, align 4
br label %b1
b1: ; preds = %b3, %b0
%v2 = phi i32 [ 0, %b0 ], [ %v129, %b3 ]
%v3 = mul nuw nsw i32 %v2, 768
%v4 = add nuw nsw i32 %v3, 32
%v5 = add nuw nsw i32 %v3, 64
%v6 = add nuw nsw i32 %v3, 96
br label %b2
b2: ; preds = %b2, %b1
%v7 = phi i32* [ %v1, %b1 ], [ %v127, %b2 ]
%v8 = phi i16* [ %v0, %b1 ], [ %v128, %b2 ]
%v9 = phi i32 [ 0, %b1 ], [ %v125, %b2 ]
%v10 = mul nuw nsw i32 %v9, 32
%v11 = bitcast i32* %v7 to <16 x i32>*
%v12 = load <16 x i32>, <16 x i32>* %v11, align 64, !tbaa !1
%v13 = add nuw nsw i32 %v10, 16
%v14 = getelementptr inbounds i32, i32* %v1, i32 %v13
%v15 = bitcast i32* %v14 to <16 x i32>*
%v16 = load <16 x i32>, <16 x i32>* %v15, align 64, !tbaa !1
%v17 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v16, <16 x i32> %v12)
%v18 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v17) #2
%v19 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v17) #2
%v20 = tail call <32 x i32> @llvm.hexagon.V6.vdealvdd(<16 x i32> %v19, <16 x i32> %v18, i32 -4) #2
%v21 = bitcast i16* %v8 to <16 x i32>*
%v22 = load <16 x i32>, <16 x i32>* %v21, align 64, !tbaa !4
%v23 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v20) #2
%v24 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v20) #2
%v25 = tail call <16 x i32> @llvm.hexagon.V6.vshufeh(<16 x i32> %v24, <16 x i32> %v23) #2
%v26 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v24, <16 x i32> %v23) #2
%v27 = tail call <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32> %v25, <16 x i32> %v22) #2
%v28 = tail call <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32> %v26, <16 x i32> %v22) #2
%v29 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v27) #2
%v30 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v28) #2
%v31 = tail call <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32> %v29, <16 x i32> %v30, i32 16) #2
%v32 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v28) #2
%v33 = tail call <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32> undef, <16 x i32> %v32, i32 16) #2
%v34 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v33, <16 x i32> %v31) #2
%v35 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v34) #2
%v36 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v34) #2
%v37 = tail call <32 x i32> @llvm.hexagon.V6.vshuffvdd(<16 x i32> %v36, <16 x i32> %v35, i32 -4) #2
%v38 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v37)
%v39 = add nuw nsw i32 %v10, %v3
%v40 = getelementptr inbounds i32, i32* undef, i32 %v39
%v41 = bitcast i32* %v40 to <16 x i32>*
store <16 x i32> %v38, <16 x i32>* %v41, align 64, !tbaa !6
%v42 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v37)
store <16 x i32> %v42, <16 x i32>* undef, align 64, !tbaa !6
%v43 = getelementptr i32, i32* %v7, i32 32
%v44 = getelementptr i16, i16* %v8, i32 32
%v45 = bitcast i32* %v43 to <16 x i32>*
%v46 = load <16 x i32>, <16 x i32>* %v45, align 64, !tbaa !1
%v47 = add nuw nsw i32 %v10, 48
%v48 = getelementptr inbounds i32, i32* %v1, i32 %v47
%v49 = bitcast i32* %v48 to <16 x i32>*
%v50 = load <16 x i32>, <16 x i32>* %v49, align 64, !tbaa !1
%v51 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v50, <16 x i32> %v46)
%v52 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v51) #2
%v53 = tail call <32 x i32> @llvm.hexagon.V6.vdealvdd(<16 x i32> undef, <16 x i32> %v52, i32 -4) #2
%v54 = bitcast i16* %v44 to <16 x i32>*
%v55 = load <16 x i32>, <16 x i32>* %v54, align 64, !tbaa !4
%v56 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v53) #2
%v57 = tail call <16 x i32> @llvm.hexagon.V6.vshufeh(<16 x i32> undef, <16 x i32> %v56) #2
%v58 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> undef, <16 x i32> %v56) #2
%v59 = tail call <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32> %v57, <16 x i32> %v55) #2
%v60 = tail call <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32> %v58, <16 x i32> %v55) #2
%v61 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v59) #2
%v62 = tail call <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32> %v61, <16 x i32> undef, i32 16) #2
%v63 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v59) #2
%v64 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v60) #2
%v65 = tail call <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32> %v63, <16 x i32> %v64, i32 16) #2
%v66 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v65, <16 x i32> %v62) #2
%v67 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v66) #2
%v68 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v66) #2
%v69 = tail call <32 x i32> @llvm.hexagon.V6.vshuffvdd(<16 x i32> %v68, <16 x i32> %v67, i32 -4) #2
%v70 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v69)
%v71 = add nuw nsw i32 %v4, %v10
%v72 = getelementptr inbounds i32, i32* undef, i32 %v71
%v73 = bitcast i32* %v72 to <16 x i32>*
store <16 x i32> %v70, <16 x i32>* %v73, align 64, !tbaa !6
%v74 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v69)
%v75 = add nuw nsw i32 %v71, 16
%v76 = getelementptr inbounds i32, i32* undef, i32 %v75
%v77 = bitcast i32* %v76 to <16 x i32>*
store <16 x i32> %v74, <16 x i32>* %v77, align 64, !tbaa !6
%v78 = getelementptr i32, i32* %v7, i32 64
%v79 = getelementptr i16, i16* %v8, i32 64
%v80 = bitcast i32* %v78 to <16 x i32>*
%v81 = load <16 x i32>, <16 x i32>* %v80, align 64, !tbaa !1
%v82 = add nuw nsw i32 %v10, 80
%v83 = getelementptr inbounds i32, i32* %v1, i32 %v82
%v84 = bitcast i32* %v83 to <16 x i32>*
%v85 = load <16 x i32>, <16 x i32>* %v84, align 64, !tbaa !1
%v86 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v85, <16 x i32> %v81)
%v87 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v86) #2
%v88 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v86) #2
%v89 = tail call <32 x i32> @llvm.hexagon.V6.vdealvdd(<16 x i32> %v88, <16 x i32> %v87, i32 -4) #2
%v90 = bitcast i16* %v79 to <16 x i32>*
%v91 = load <16 x i32>, <16 x i32>* %v90, align 64, !tbaa !4
%v92 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v89) #2
%v93 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v89) #2
%v94 = tail call <16 x i32> @llvm.hexagon.V6.vshufeh(<16 x i32> %v93, <16 x i32> %v92) #2
%v95 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v93, <16 x i32> %v92) #2
%v96 = tail call <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32> %v94, <16 x i32> %v91) #2
%v97 = tail call <32 x i32> @llvm.hexagon.V6.vmpyuhv(<16 x i32> %v95, <16 x i32> %v91) #2
%v98 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v97) #2
%v99 = tail call <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32> undef, <16 x i32> %v98, i32 16) #2
%v100 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v96) #2
%v101 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v97) #2
%v102 = tail call <16 x i32> @llvm.hexagon.V6.vaslw.acc(<16 x i32> %v100, <16 x i32> %v101, i32 16) #2
%v103 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v102, <16 x i32> %v99) #2
%v104 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v103) #2
%v105 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v103) #2
%v106 = tail call <32 x i32> @llvm.hexagon.V6.vshuffvdd(<16 x i32> %v105, <16 x i32> %v104, i32 -4) #2
%v107 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v106)
%v108 = add nuw nsw i32 %v5, %v10
%v109 = getelementptr inbounds i32, i32* undef, i32 %v108
%v110 = bitcast i32* %v109 to <16 x i32>*
store <16 x i32> %v107, <16 x i32>* %v110, align 64, !tbaa !6
%v111 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v106)
%v112 = add nuw nsw i32 %v108, 16
%v113 = getelementptr inbounds i32, i32* undef, i32 %v112
%v114 = bitcast i32* %v113 to <16 x i32>*
store <16 x i32> %v111, <16 x i32>* %v114, align 64, !tbaa !6
%v115 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> undef) #2
%v116 = tail call <32 x i32> @llvm.hexagon.V6.vshuffvdd(<16 x i32> undef, <16 x i32> %v115, i32 -4) #2
%v117 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v116)
%v118 = add nuw nsw i32 %v6, %v10
%v119 = getelementptr inbounds i32, i32* undef, i32 %v118
%v120 = bitcast i32* %v119 to <16 x i32>*
store <16 x i32> %v117, <16 x i32>* %v120, align 64, !tbaa !6
%v121 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v116)
%v122 = add nuw nsw i32 %v118, 16
%v123 = getelementptr inbounds i32, i32* undef, i32 %v122
%v124 = bitcast i32* %v123 to <16 x i32>*
store <16 x i32> %v121, <16 x i32>* %v124, align 64, !tbaa !6
%v125 = add nuw nsw i32 %v9, 4
%v126 = icmp eq i32 %v125, 24
%v127 = getelementptr i32, i32* %v7, i32 128
%v128 = getelementptr i16, i16* %v8, i32 128
br i1 %v126, label %b3, label %b2
b3: ; preds = %b2
%v129 = add nuw nsw i32 %v2, 1
br label %b1
}
attributes #0 = { nounwind readnone }
attributes #1 = { "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
attributes #2 = { nounwind }
!llvm.module.flags = !{!0}
!0 = !{i32 2, !"halide_mattrs", !"+hvx"}
!1 = !{!2, !2, i64 0}
!2 = !{!"in_u32", !3}
!3 = !{!"Halide buffer"}
!4 = !{!5, !5, i64 0}
!5 = !{!"in_u16", !3}
!6 = !{!7, !7, i64 0}
!7 = !{!"op_vmpy_v__uh_v__uh__1", !3}
|