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
|
; RUN: llc -march=hexagon -O2 < %s -pipeliner-experimental-cg=true | FileCheck %s
; We do not pipeline sigma yet, but the non-pipelined version
; with good scheduling is pretty fast. The compiler generates
; 18 packets, and the assembly version is 16.
; CHECK: loop0(.LBB0_[[LOOP:.]],
; CHECK: .LBB0_[[LOOP]]:
; CHECK-COUNT-17: }
; CHECK: }{{[ \t]*}}:endloop
@g0 = external constant [10 x i16], align 128
declare i32 @llvm.hexagon.S2.vsplatrb(i32) #0
declare <16 x i32> @llvm.hexagon.V6.vshuffh(<16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.lvsplatw(i32) #0
declare <16 x i32> @llvm.hexagon.V6.vd0() #0
declare <32 x i32> @llvm.hexagon.V6.vsububh(<16 x i32>, <16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32>, <16 x i32>) #0
declare <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32>, <16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1>, <16 x i32>, <16 x i32>) #0
declare <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1>, <16 x i32>, <16 x i32>) #0
declare <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32>, <16 x i32>) #0
declare <32 x i32> @llvm.hexagon.V6.vmpabus.acc(<32 x i32>, <32 x i32>, i32) #0
declare <16 x i32> @llvm.hexagon.V6.vlalignbi(<16 x i32>, <16 x i32>, i32) #0
declare <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32>, <16 x i32>, i32) #0
declare <32 x i32> @llvm.hexagon.V6.vlutvwh(<16 x i32>, <16 x i32>, i32) #0
declare <16 x i32> @llvm.hexagon.V6.vmpyhvsrs(<16 x i32>, <16 x i32>) #0
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 <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32>, <16 x i32>) #0
define void @f0(i8* nocapture readonly %a0, i32 %a1, i32 %a2, i32 %a3, i8 zeroext %a4, i8* nocapture %a5) #1 {
b0:
%v0 = add nsw i32 %a3, -1
%v1 = icmp sgt i32 %v0, 1
br i1 %v1, label %b1, label %b8
b1: ; preds = %b0
%v2 = mul i32 %a1, 2
%v3 = load <16 x i32>, <16 x i32>* bitcast ([10 x i16]* @g0 to <16 x i32>*), align 128
%v4 = tail call <16 x i32> @llvm.hexagon.V6.vshuffh(<16 x i32> %v3) #2
%v5 = zext i8 %a4 to i32
%v6 = tail call i32 @llvm.hexagon.S2.vsplatrb(i32 %v5) #2
%v7 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 %v6) #2
%v8 = tail call <16 x i32> @llvm.hexagon.V6.vd0() #2
%v9 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 16843009) #2
%v10 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 33686018) #2
%v11 = icmp sgt i32 %a2, 64
%v12 = add i32 %a1, 64
%v13 = add i32 %v12, %a1
%v14 = icmp sgt i32 %a2, 0
%v15 = add i32 %a3, -2
%v16 = bitcast i8* %a0 to <16 x i32>*
%v17 = load <16 x i32>, <16 x i32>* %v16, align 64
br label %b2
b2: ; preds = %b7, %b1
%v18 = phi <16 x i32> [ %v17, %b1 ], [ %v28, %b7 ]
%v19 = phi i8* [ %a0, %b1 ], [ %v23, %b7 ]
%v20 = phi i8* [ %a5, %b1 ], [ %v22, %b7 ]
%v21 = phi i32 [ 1, %b1 ], [ %v118, %b7 ]
%v22 = getelementptr inbounds i8, i8* %v20, i32 %a1
%v23 = getelementptr inbounds i8, i8* %v19, i32 %a1
%v24 = bitcast i8* %v23 to <16 x i32>*
%v25 = getelementptr inbounds i8, i8* %v19, i32 %v2
%v26 = bitcast i8* %v25 to <16 x i32>*
%v27 = bitcast i8* %v22 to <16 x i32>*
%v28 = load <16 x i32>, <16 x i32>* %v24, align 64
%v29 = load <16 x i32>, <16 x i32>* %v26, align 64
br i1 %v11, label %b3, label %b4
b3: ; preds = %b2
%v30 = getelementptr inbounds i8, i8* %v19, i32 64
%v31 = getelementptr inbounds i8, i8* %v19, i32 %v12
%v32 = bitcast i8* %v31 to <16 x i32>*
%v33 = getelementptr inbounds i8, i8* %v19, i32 %v13
%v34 = bitcast i8* %v33 to <16 x i32>*
br label %b5
b4: ; preds = %b2
br i1 %v14, label %b5, label %b7
b5: ; preds = %b4, %b3
%v35 = phi <16 x i32>* [ %v26, %b4 ], [ %v34, %b3 ]
%v36 = phi <16 x i32>* [ %v24, %b4 ], [ %v32, %b3 ]
%v37 = phi i8* [ %v19, %b4 ], [ %v30, %b3 ]
%v38 = bitcast i8* %v37 to <16 x i32>*
br label %b6
b6: ; preds = %b6, %b5
%v39 = phi <16 x i32>* [ %v108, %b6 ], [ %v27, %b5 ]
%v40 = phi <16 x i32>* [ %v115, %b6 ], [ %v35, %b5 ]
%v41 = phi <16 x i32>* [ %v114, %b6 ], [ %v36, %b5 ]
%v42 = phi <16 x i32>* [ %v113, %b6 ], [ %v38, %b5 ]
%v43 = phi i32 [ %v116, %b6 ], [ %a2, %b5 ]
%v44 = phi <16 x i32> [ %v45, %b6 ], [ %v8, %b5 ]
%v45 = phi <16 x i32> [ %v50, %b6 ], [ %v18, %b5 ]
%v46 = phi <16 x i32> [ %v47, %b6 ], [ %v8, %b5 ]
%v47 = phi <16 x i32> [ %v51, %b6 ], [ %v28, %b5 ]
%v48 = phi <16 x i32> [ %v49, %b6 ], [ %v8, %b5 ]
%v49 = phi <16 x i32> [ %v52, %b6 ], [ %v29, %b5 ]
%v50 = load <16 x i32>, <16 x i32>* %v42, align 64
%v51 = load <16 x i32>, <16 x i32>* %v41, align 64
%v52 = load <16 x i32>, <16 x i32>* %v40, align 64
%v53 = tail call <32 x i32> @llvm.hexagon.V6.vsububh(<16 x i32> %v8, <16 x i32> %v47) #2
%v54 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v45, <16 x i32> %v47) #2
%v55 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v49, <16 x i32> %v47) #2
%v56 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v54, <16 x i32> %v7) #2
%v57 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v55, <16 x i32> %v7) #2
%v58 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v56, <16 x i32> %v9, <16 x i32> %v10) #2
%v59 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v57, <16 x i32> %v58, <16 x i32> %v9) #2
%v60 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v56, <16 x i32> %v8, <16 x i32> %v45) #2
%v61 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v57, <16 x i32> %v8, <16 x i32> %v49) #2
%v62 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v61, <16 x i32> %v60) #2
%v63 = tail call <32 x i32> @llvm.hexagon.V6.vmpabus.acc(<32 x i32> %v53, <32 x i32> %v62, i32 -1) #2
%v64 = tail call <16 x i32> @llvm.hexagon.V6.vlalignbi(<16 x i32> %v45, <16 x i32> %v44, i32 1) #2
%v65 = tail call <16 x i32> @llvm.hexagon.V6.vlalignbi(<16 x i32> %v49, <16 x i32> %v48, i32 1) #2
%v66 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v64, <16 x i32> %v47) #2
%v67 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v65, <16 x i32> %v47) #2
%v68 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v66, <16 x i32> %v7) #2
%v69 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v67, <16 x i32> %v7) #2
%v70 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v68, <16 x i32> %v59, <16 x i32> %v9) #2
%v71 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v69, <16 x i32> %v70, <16 x i32> %v9) #2
%v72 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v68, <16 x i32> %v8, <16 x i32> %v64) #2
%v73 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v69, <16 x i32> %v8, <16 x i32> %v65) #2
%v74 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v73, <16 x i32> %v72) #2
%v75 = tail call <32 x i32> @llvm.hexagon.V6.vmpabus.acc(<32 x i32> %v63, <32 x i32> %v74, i32 -1) #2
%v76 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %v50, <16 x i32> %v45, i32 1) #2
%v77 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %v52, <16 x i32> %v49, i32 1) #2
%v78 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v76, <16 x i32> %v47) #2
%v79 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v77, <16 x i32> %v47) #2
%v80 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v78, <16 x i32> %v7) #2
%v81 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v79, <16 x i32> %v7) #2
%v82 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v80, <16 x i32> %v71, <16 x i32> %v9) #2
%v83 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v81, <16 x i32> %v82, <16 x i32> %v9) #2
%v84 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v80, <16 x i32> %v8, <16 x i32> %v76) #2
%v85 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v81, <16 x i32> %v8, <16 x i32> %v77) #2
%v86 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v85, <16 x i32> %v84) #2
%v87 = tail call <32 x i32> @llvm.hexagon.V6.vmpabus.acc(<32 x i32> %v75, <32 x i32> %v86, i32 -1) #2
%v88 = tail call <16 x i32> @llvm.hexagon.V6.vlalignbi(<16 x i32> %v47, <16 x i32> %v46, i32 1) #2
%v89 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %v51, <16 x i32> %v47, i32 1) #2
%v90 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v88, <16 x i32> %v47) #2
%v91 = tail call <16 x i32> @llvm.hexagon.V6.vabsdiffub(<16 x i32> %v89, <16 x i32> %v47) #2
%v92 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v90, <16 x i32> %v7) #2
%v93 = tail call <64 x i1> @llvm.hexagon.V6.vgtub(<16 x i32> %v91, <16 x i32> %v7) #2
%v94 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v92, <16 x i32> %v83, <16 x i32> %v9) #2
%v95 = tail call <16 x i32> @llvm.hexagon.V6.vaddbnq(<64 x i1> %v93, <16 x i32> %v94, <16 x i32> %v9) #2
%v96 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v92, <16 x i32> %v8, <16 x i32> %v88) #2
%v97 = tail call <16 x i32> @llvm.hexagon.V6.vmux(<64 x i1> %v93, <16 x i32> %v8, <16 x i32> %v89) #2
%v98 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %v97, <16 x i32> %v96) #2
%v99 = tail call <32 x i32> @llvm.hexagon.V6.vmpabus.acc(<32 x i32> %v87, <32 x i32> %v98, i32 -1) #2
%v100 = tail call <32 x i32> @llvm.hexagon.V6.vlutvwh(<16 x i32> %v95, <16 x i32> %v4, i32 0) #2
%v101 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v99) #2
%v102 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %v100) #2
%v103 = tail call <16 x i32> @llvm.hexagon.V6.vmpyhvsrs(<16 x i32> %v101, <16 x i32> %v102) #2
%v104 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v99) #2
%v105 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v100) #2
%v106 = tail call <16 x i32> @llvm.hexagon.V6.vmpyhvsrs(<16 x i32> %v104, <16 x i32> %v105) #2
%v107 = tail call <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32> %v106, <16 x i32> %v103) #2
%v108 = getelementptr inbounds <16 x i32>, <16 x i32>* %v39, i32 1
store <16 x i32> %v107, <16 x i32>* %v39, align 64
%v109 = icmp sgt i32 %v43, 128
%v110 = getelementptr inbounds <16 x i32>, <16 x i32>* %v42, i32 1
%v111 = getelementptr inbounds <16 x i32>, <16 x i32>* %v41, i32 1
%v112 = getelementptr inbounds <16 x i32>, <16 x i32>* %v40, i32 1
%v113 = select i1 %v109, <16 x i32>* %v110, <16 x i32>* %v42
%v114 = select i1 %v109, <16 x i32>* %v111, <16 x i32>* %v41
%v115 = select i1 %v109, <16 x i32>* %v112, <16 x i32>* %v40
%v116 = add nsw i32 %v43, -64
%v117 = icmp sgt i32 %v43, 64
br i1 %v117, label %b6, label %b7
b7: ; preds = %b6, %b4
%v118 = add nuw nsw i32 %v21, 1
%v119 = icmp eq i32 %v21, %v15
br i1 %v119, label %b8, label %b2
b8: ; preds = %b7, %b0
ret void
}
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind "target-cpu"="hexagonv62" "target-features"="+hvx,+hvx-length64b" }
attributes #2 = { nounwind }
|
