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
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve.fp -o - %s | FileCheck --check-prefix=CHECK %s
define void @tailpred(half* nocapture readonly %pSrcA, half* nocapture readonly %pSrcB, half* nocapture %pDst, i32 %blockSize) {
; CHECK-LABEL: tailpred:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r7, lr}
; CHECK-NEXT: push {r4, r5, r7, lr}
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: beq .LBB0_6
; CHECK-NEXT: @ %bb.1: @ %vector.memcheck
; CHECK-NEXT: add.w r5, r2, r3, lsl #1
; CHECK-NEXT: add.w r4, r1, r3, lsl #1
; CHECK-NEXT: cmp r5, r1
; CHECK-NEXT: cset r12, hi
; CHECK-NEXT: cmp r4, r2
; CHECK-NEXT: cset lr, hi
; CHECK-NEXT: cmp r5, r0
; CHECK-NEXT: add.w r5, r0, r3, lsl #1
; CHECK-NEXT: cset r4, hi
; CHECK-NEXT: cmp r5, r2
; CHECK-NEXT: cset r5, hi
; CHECK-NEXT: tst r5, r4
; CHECK-NEXT: it eq
; CHECK-NEXT: andseq.w r5, lr, r12
; CHECK-NEXT: beq .LBB0_4
; CHECK-NEXT: @ %bb.2: @ %while.body.preheader
; CHECK-NEXT: dls lr, r3
; CHECK-NEXT: .LBB0_3: @ %while.body
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vldr.16 s0, [r0]
; CHECK-NEXT: vldr.16 s2, [r1]
; CHECK-NEXT: adds r1, #2
; CHECK-NEXT: adds r0, #2
; CHECK-NEXT: vadd.f16 s0, s2, s0
; CHECK-NEXT: vstr.16 s0, [r2]
; CHECK-NEXT: adds r2, #2
; CHECK-NEXT: le lr, .LBB0_3
; CHECK-NEXT: b .LBB0_6
; CHECK-NEXT: .LBB0_4: @ %vector.ph
; CHECK-NEXT: dlstp.16 lr, r3
; CHECK-NEXT: .LBB0_5: @ %vector.body
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vldrh.u16 q0, [r0], #16
; CHECK-NEXT: vldrh.u16 q1, [r1], #16
; CHECK-NEXT: vadd.f16 q0, q1, q0
; CHECK-NEXT: vstrh.16 q0, [r2], #16
; CHECK-NEXT: letp lr, .LBB0_5
; CHECK-NEXT: .LBB0_6: @ %while.end
; CHECK-NEXT: pop {r4, r5, r7, pc}
entry:
%cmp.not6 = icmp eq i32 %blockSize, 0
br i1 %cmp.not6, label %while.end, label %vector.memcheck
vector.memcheck: ; preds = %entry
%scevgep = getelementptr half, half* %pDst, i32 %blockSize
%scevgep14 = getelementptr half, half* %pSrcA, i32 %blockSize
%scevgep17 = getelementptr half, half* %pSrcB, i32 %blockSize
%bound0 = icmp ugt half* %scevgep14, %pDst
%bound1 = icmp ugt half* %scevgep, %pSrcA
%found.conflict = and i1 %bound0, %bound1
%bound019 = icmp ugt half* %scevgep17, %pDst
%bound120 = icmp ugt half* %scevgep, %pSrcB
%found.conflict21 = and i1 %bound019, %bound120
%conflict.rdx = or i1 %found.conflict, %found.conflict21
br i1 %conflict.rdx, label %while.body, label %vector.ph
vector.ph: ; preds = %vector.memcheck
%n.rnd.up = add i32 %blockSize, 7
%n.vec = and i32 %n.rnd.up, -8
br label %vector.body
vector.body: ; preds = %vector.body, %vector.ph
%index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
%next.gep = getelementptr half, half* %pSrcA, i32 %index
%next.gep28 = getelementptr half, half* %pDst, i32 %index
%next.gep29 = getelementptr half, half* %pSrcB, i32 %index
%active.lane.mask = call <8 x i1> @llvm.get.active.lane.mask.v8i1.i32(i32 %index, i32 %blockSize)
%0 = bitcast half* %next.gep to <8 x half>*
%wide.masked.load = call <8 x half> @llvm.masked.load.v8f16.p0v8f16(<8 x half>* %0, i32 2, <8 x i1> %active.lane.mask, <8 x half> undef)
%1 = bitcast half* %next.gep29 to <8 x half>*
%wide.masked.load32 = call <8 x half> @llvm.masked.load.v8f16.p0v8f16(<8 x half>* %1, i32 2, <8 x i1> %active.lane.mask, <8 x half> undef)
%2 = fadd fast <8 x half> %wide.masked.load32, %wide.masked.load
%3 = bitcast half* %next.gep28 to <8 x half>*
call void @llvm.masked.store.v8f16.p0v8f16(<8 x half> %2, <8 x half>* %3, i32 2, <8 x i1> %active.lane.mask)
%index.next = add i32 %index, 8
%4 = icmp eq i32 %index.next, %n.vec
br i1 %4, label %while.end, label %vector.body
while.body: ; preds = %vector.memcheck, %while.body
%blkCnt.010 = phi i32 [ %dec, %while.body ], [ %blockSize, %vector.memcheck ]
%pSrcA.addr.09 = phi half* [ %incdec.ptr, %while.body ], [ %pSrcA, %vector.memcheck ]
%pDst.addr.08 = phi half* [ %incdec.ptr3, %while.body ], [ %pDst, %vector.memcheck ]
%pSrcB.addr.07 = phi half* [ %incdec.ptr1, %while.body ], [ %pSrcB, %vector.memcheck ]
%incdec.ptr = getelementptr inbounds half, half* %pSrcA.addr.09, i32 1
%5 = load half, half* %pSrcA.addr.09, align 2
%incdec.ptr1 = getelementptr inbounds half, half* %pSrcB.addr.07, i32 1
%6 = load half, half* %pSrcB.addr.07, align 2
%7 = fadd fast half %6, %5
%incdec.ptr3 = getelementptr inbounds half, half* %pDst.addr.08, i32 1
store half %7, half* %pDst.addr.08, align 2
%dec = add i32 %blkCnt.010, -1
%cmp.not = icmp eq i32 %dec, 0
br i1 %cmp.not, label %while.end, label %while.body
while.end: ; preds = %vector.body, %while.body, %entry
ret void
}
define void @notailpred(half* nocapture readonly %pSrcA, half* nocapture readonly %pSrcB, half* nocapture %pDst, i32 %blockSize) {
; CHECK-LABEL: notailpred:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, lr}
; CHECK-NEXT: cbz r3, .LBB1_6
; CHECK-NEXT: @ %bb.1: @ %while.body.preheader
; CHECK-NEXT: cmp r3, #8
; CHECK-NEXT: blo .LBB1_3
; CHECK-NEXT: @ %bb.2: @ %vector.memcheck
; CHECK-NEXT: add.w r5, r2, r3, lsl #1
; CHECK-NEXT: add.w r6, r1, r3, lsl #1
; CHECK-NEXT: cmp r5, r1
; CHECK-NEXT: add.w r4, r0, r3, lsl #1
; CHECK-NEXT: cset r7, hi
; CHECK-NEXT: cmp r6, r2
; CHECK-NEXT: cset r6, hi
; CHECK-NEXT: cmp r5, r0
; CHECK-NEXT: cset r5, hi
; CHECK-NEXT: cmp r4, r2
; CHECK-NEXT: cset r4, hi
; CHECK-NEXT: tst r4, r5
; CHECK-NEXT: it eq
; CHECK-NEXT: andseq.w r7, r7, r6
; CHECK-NEXT: beq .LBB1_7
; CHECK-NEXT: .LBB1_3:
; CHECK-NEXT: mov r5, r3
; CHECK-NEXT: mov r12, r0
; CHECK-NEXT: mov r7, r2
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: .LBB1_4: @ %while.body.preheader31
; CHECK-NEXT: dls lr, r5
; CHECK-NEXT: .LBB1_5: @ %while.body
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vldr.16 s0, [r12]
; CHECK-NEXT: vldr.16 s2, [r4]
; CHECK-NEXT: adds r4, #2
; CHECK-NEXT: add.w r12, r12, #2
; CHECK-NEXT: vadd.f16 s0, s2, s0
; CHECK-NEXT: vstr.16 s0, [r7]
; CHECK-NEXT: adds r7, #2
; CHECK-NEXT: le lr, .LBB1_5
; CHECK-NEXT: .LBB1_6: @ %while.end
; CHECK-NEXT: pop {r4, r5, r6, r7, pc}
; CHECK-NEXT: .LBB1_7: @ %vector.ph
; CHECK-NEXT: bic r6, r3, #7
; CHECK-NEXT: movs r5, #1
; CHECK-NEXT: sub.w r7, r6, #8
; CHECK-NEXT: add.w r4, r1, r6, lsl #1
; CHECK-NEXT: add.w r12, r0, r6, lsl #1
; CHECK-NEXT: add.w lr, r5, r7, lsr #3
; CHECK-NEXT: add.w r7, r2, r6, lsl #1
; CHECK-NEXT: and r5, r3, #7
; CHECK-NEXT: .LBB1_8: @ %vector.body
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vldrh.u16 q0, [r0], #16
; CHECK-NEXT: vldrh.u16 q1, [r1], #16
; CHECK-NEXT: vadd.f16 q0, q1, q0
; CHECK-NEXT: vstrb.8 q0, [r2], #16
; CHECK-NEXT: le lr, .LBB1_8
; CHECK-NEXT: @ %bb.9: @ %middle.block
; CHECK-NEXT: cmp r6, r3
; CHECK-NEXT: bne .LBB1_4
; CHECK-NEXT: b .LBB1_6
entry:
%cmp.not6 = icmp eq i32 %blockSize, 0
br i1 %cmp.not6, label %while.end, label %while.body.preheader
while.body.preheader: ; preds = %entry
%min.iters.check = icmp ult i32 %blockSize, 8
br i1 %min.iters.check, label %while.body.preheader31, label %vector.memcheck
vector.memcheck: ; preds = %while.body.preheader
%scevgep = getelementptr half, half* %pDst, i32 %blockSize
%scevgep14 = getelementptr half, half* %pSrcA, i32 %blockSize
%scevgep17 = getelementptr half, half* %pSrcB, i32 %blockSize
%bound0 = icmp ugt half* %scevgep14, %pDst
%bound1 = icmp ugt half* %scevgep, %pSrcA
%found.conflict = and i1 %bound0, %bound1
%bound019 = icmp ugt half* %scevgep17, %pDst
%bound120 = icmp ugt half* %scevgep, %pSrcB
%found.conflict21 = and i1 %bound019, %bound120
%conflict.rdx = or i1 %found.conflict, %found.conflict21
br i1 %conflict.rdx, label %while.body.preheader31, label %vector.ph
vector.ph: ; preds = %vector.memcheck
%n.vec = and i32 %blockSize, -8
%ind.end = and i32 %blockSize, 7
%ind.end23 = getelementptr half, half* %pSrcA, i32 %n.vec
%ind.end25 = getelementptr half, half* %pDst, i32 %n.vec
%ind.end27 = getelementptr half, half* %pSrcB, i32 %n.vec
br label %vector.body
vector.body: ; preds = %vector.body, %vector.ph
%index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
%next.gep = getelementptr half, half* %pSrcA, i32 %index
%next.gep28 = getelementptr half, half* %pDst, i32 %index
%next.gep29 = getelementptr half, half* %pSrcB, i32 %index
%0 = bitcast half* %next.gep to <8 x half>*
%wide.load = load <8 x half>, <8 x half>* %0, align 2
%1 = bitcast half* %next.gep29 to <8 x half>*
%wide.load30 = load <8 x half>, <8 x half>* %1, align 2
%2 = fadd fast <8 x half> %wide.load30, %wide.load
%3 = bitcast half* %next.gep28 to <8 x half>*
store <8 x half> %2, <8 x half>* %3, align 2
%index.next = add i32 %index, 8
%4 = icmp eq i32 %index.next, %n.vec
br i1 %4, label %middle.block, label %vector.body
middle.block: ; preds = %vector.body
%cmp.n = icmp eq i32 %n.vec, %blockSize
br i1 %cmp.n, label %while.end, label %while.body.preheader31
while.body.preheader31: ; preds = %middle.block, %vector.memcheck, %while.body.preheader
%blkCnt.010.ph = phi i32 [ %blockSize, %vector.memcheck ], [ %blockSize, %while.body.preheader ], [ %ind.end, %middle.block ]
%pSrcA.addr.09.ph = phi half* [ %pSrcA, %vector.memcheck ], [ %pSrcA, %while.body.preheader ], [ %ind.end23, %middle.block ]
%pDst.addr.08.ph = phi half* [ %pDst, %vector.memcheck ], [ %pDst, %while.body.preheader ], [ %ind.end25, %middle.block ]
%pSrcB.addr.07.ph = phi half* [ %pSrcB, %vector.memcheck ], [ %pSrcB, %while.body.preheader ], [ %ind.end27, %middle.block ]
br label %while.body
while.body: ; preds = %while.body.preheader31, %while.body
%blkCnt.010 = phi i32 [ %dec, %while.body ], [ %blkCnt.010.ph, %while.body.preheader31 ]
%pSrcA.addr.09 = phi half* [ %incdec.ptr, %while.body ], [ %pSrcA.addr.09.ph, %while.body.preheader31 ]
%pDst.addr.08 = phi half* [ %incdec.ptr3, %while.body ], [ %pDst.addr.08.ph, %while.body.preheader31 ]
%pSrcB.addr.07 = phi half* [ %incdec.ptr1, %while.body ], [ %pSrcB.addr.07.ph, %while.body.preheader31 ]
%incdec.ptr = getelementptr inbounds half, half* %pSrcA.addr.09, i32 1
%5 = load half, half* %pSrcA.addr.09, align 2
%incdec.ptr1 = getelementptr inbounds half, half* %pSrcB.addr.07, i32 1
%6 = load half, half* %pSrcB.addr.07, align 2
%7 = fadd fast half %6, %5
%incdec.ptr3 = getelementptr inbounds half, half* %pDst.addr.08, i32 1
store half %7, half* %pDst.addr.08, align 2
%dec = add i32 %blkCnt.010, -1
%cmp.not = icmp eq i32 %dec, 0
br i1 %cmp.not, label %while.end, label %while.body
while.end: ; preds = %while.body, %middle.block, %entry
ret void
}
declare <8 x i1> @llvm.get.active.lane.mask.v8i1.i32(i32, i32) #1
declare <8 x half> @llvm.masked.load.v8f16.p0v8f16(<8 x half>*, i32 immarg, <8 x i1>, <8 x half>) #2
declare void @llvm.masked.store.v8f16.p0v8f16(<8 x half>, <8 x half>*, i32 immarg, <8 x i1>) #3
|
