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
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -verify-machineinstrs %s -o - | FileCheck %s
define arm_aapcs_vfpcc i16 @reduce_v16i16_shift_mul(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-LABEL: reduce_v16i16_shift_mul:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmullt.u8 q2, q0, q1
; CHECK-NEXT: vmullb.u8 q0, q0, q1
; CHECK-NEXT: vshr.s16 q2, q2, #14
; CHECK-NEXT: vshr.s16 q0, q0, #14
; CHECK-NEXT: vaddv.u16 r0, q2
; CHECK-NEXT: vaddva.u16 r0, q0
; CHECK-NEXT: bx lr
entry:
%s0s = zext <16 x i8> %s0 to <16 x i16>
%s1s = zext <16 x i8> %s1 to <16 x i16>
%m = mul <16 x i16> %s0s, %s1s
%sh = ashr <16 x i16> %m, <i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14>
%result = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %sh)
ret i16 %result
}
define arm_aapcs_vfpcc i16 @reduce_v8i16_shift_mul(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-LABEL: reduce_v8i16_shift_mul:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmullb.u8 q0, q0, q1
; CHECK-NEXT: vshr.s16 q0, q0, #14
; CHECK-NEXT: vaddv.u16 r0, q0
; CHECK-NEXT: bx lr
entry:
%s0s = zext <8 x i8> %s0 to <8 x i16>
%s1s = zext <8 x i8> %s1 to <8 x i16>
%m = mul <8 x i16> %s0s, %s1s
%sh = ashr <8 x i16> %m, <i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14>
%result = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> %sh)
ret i16 %result
}
define arm_aapcs_vfpcc i16 @reduce_v16i16_shift_sub(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-LABEL: reduce_v16i16_shift_sub:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmovlt.u8 q2, q1
; CHECK-NEXT: vmovlt.u8 q3, q0
; CHECK-NEXT: vsub.i16 q2, q3, q2
; CHECK-NEXT: vmovlb.u8 q1, q1
; CHECK-NEXT: vmovlb.u8 q0, q0
; CHECK-NEXT: vshr.s16 q2, q2, #14
; CHECK-NEXT: vsub.i16 q0, q0, q1
; CHECK-NEXT: vaddv.u16 r0, q2
; CHECK-NEXT: vshr.s16 q0, q0, #14
; CHECK-NEXT: vaddva.u16 r0, q0
; CHECK-NEXT: bx lr
entry:
%s0s = zext <16 x i8> %s0 to <16 x i16>
%s1s = zext <16 x i8> %s1 to <16 x i16>
%m = sub <16 x i16> %s0s, %s1s
%sh = ashr <16 x i16> %m, <i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14, i16 14>
%result = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %sh)
ret i16 %result
}
define arm_aapcs_vfpcc i32 @mlapred_v4i32_v4i64_zext(<8 x i16> %x, <8 x i16> %y, <8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: mlapred_v4i32_v4i64_zext:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEXT: .pad #32
; CHECK-NEXT: sub sp, #32
; CHECK-NEXT: vorr q2, q2, q3
; CHECK-NEXT: mov r0, sp
; CHECK-NEXT: vstrw.32 q2, [r0]
; CHECK-NEXT: vmov.i8 q3, #0xff
; CHECK-NEXT: vldrh.u32 q2, [r0, #8]
; CHECK-NEXT: vldrh.u32 q5, [r0]
; CHECK-NEXT: add r0, sp, #16
; CHECK-NEXT: vcmp.i32 eq, q2, zr
; CHECK-NEXT: vmov.i8 q2, #0x0
; CHECK-NEXT: vpsel q4, q3, q2
; CHECK-NEXT: vcmp.i32 eq, q5, zr
; CHECK-NEXT: vpsel q2, q3, q2
; CHECK-NEXT: vstrh.32 q4, [r0, #8]
; CHECK-NEXT: vstrh.32 q2, [r0]
; CHECK-NEXT: vldrw.u32 q2, [r0]
; CHECK-NEXT: vpt.i16 ne, q2, zr
; CHECK-NEXT: vmlavt.u16 r0, q0, q1
; CHECK-NEXT: add sp, #32
; CHECK-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEXT: bx lr
entry:
%aa = zext <8 x i16> %a to <8 x i32>
%bb = zext <8 x i16> %b to <8 x i32>
%c1 = icmp eq <8 x i32> %aa, zeroinitializer
%c2 = icmp eq <8 x i32> %bb, zeroinitializer
%c = and <8 x i1> %c1, %c2
%xx = zext <8 x i16> %x to <8 x i32>
%yy = zext <8 x i16> %y to <8 x i32>
%m = mul <8 x i32> %xx, %yy
%s = select <8 x i1> %c, <8 x i32> %m, <8 x i32> zeroinitializer
%z = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> %s)
ret i32 %z
}
define void @correlate(ptr nocapture noundef readonly %ID, ptr nocapture noundef writeonly %ACD, i16 noundef signext %DS, i16 noundef signext %Ls, i16 noundef signext %S) {
; CHECK-LABEL: correlate:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEXT: push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEXT: .pad #12
; CHECK-NEXT: sub sp, #12
; CHECK-NEXT: cmp r3, #1
; CHECK-NEXT: strd r0, r1, [sp] @ 8-byte Folded Spill
; CHECK-NEXT: mov r1, r3
; CHECK-NEXT: str r3, [sp, #8] @ 4-byte Spill
; CHECK-NEXT: blt .LBB4_12
; CHECK-NEXT: @ %bb.1: @ %for.body.lr.ph
; CHECK-NEXT: ldr r1, [sp, #48]
; CHECK-NEXT: add.w r12, r2, #3
; CHECK-NEXT: ldr.w r11, [sp] @ 4-byte Reload
; CHECK-NEXT: mov.w r10, #0
; CHECK-NEXT: mov r8, r2
; CHECK-NEXT: mov r0, r2
; CHECK-NEXT: uxth r3, r1
; CHECK-NEXT: b .LBB4_4
; CHECK-NEXT: .LBB4_2: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: movs r6, #0
; CHECK-NEXT: .LBB4_3: @ %for.end
; CHECK-NEXT: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: ldr r7, [sp, #4] @ 4-byte Reload
; CHECK-NEXT: lsrs r2, r6, #16
; CHECK-NEXT: sub.w r12, r12, #1
; CHECK-NEXT: add.w r11, r11, #2
; CHECK-NEXT: sub.w r8, r8, #1
; CHECK-NEXT: strh.w r2, [r7, r10, lsl #1]
; CHECK-NEXT: add.w r10, r10, #1
; CHECK-NEXT: ldr r2, [sp, #8] @ 4-byte Reload
; CHECK-NEXT: cmp r10, r2
; CHECK-NEXT: mov r2, r0
; CHECK-NEXT: beq .LBB4_12
; CHECK-NEXT: .LBB4_4: @ %for.body
; CHECK-NEXT: @ =>This Loop Header: Depth=1
; CHECK-NEXT: @ Child Loop BB4_8 Depth 2
; CHECK-NEXT: @ Child Loop BB4_11 Depth 2
; CHECK-NEXT: cmp r2, r10
; CHECK-NEXT: ble .LBB4_2
; CHECK-NEXT: @ %bb.5: @ %vector.main.loop.iter.check
; CHECK-NEXT: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: sub.w r4, r2, r10
; CHECK-NEXT: cmp r4, #8
; CHECK-NEXT: bhs .LBB4_7
; CHECK-NEXT: @ %bb.6: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: movs r6, #0
; CHECK-NEXT: mov.w r9, #0
; CHECK-NEXT: b .LBB4_10
; CHECK-NEXT: .LBB4_7: @ %vector.ph
; CHECK-NEXT: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: bic r2, r8, #7
; CHECK-NEXT: movs r7, #1
; CHECK-NEXT: subs r2, #8
; CHECK-NEXT: bic r9, r4, #7
; CHECK-NEXT: movs r6, #0
; CHECK-NEXT: mov r5, r11
; CHECK-NEXT: add.w lr, r7, r2, lsr #3
; CHECK-NEXT: ldr r2, [sp] @ 4-byte Reload
; CHECK-NEXT: .LBB4_8: @ %vector.body
; CHECK-NEXT: @ Parent Loop BB4_4 Depth=1
; CHECK-NEXT: @ => This Inner Loop Header: Depth=2
; CHECK-NEXT: vldrh.u16 q0, [r2], #16
; CHECK-NEXT: vldrh.u16 q1, [r5], #16
; CHECK-NEXT: rsbs r7, r3, #0
; CHECK-NEXT: vmullb.s16 q2, q1, q0
; CHECK-NEXT: vmullt.s16 q0, q1, q0
; CHECK-NEXT: vshl.s32 q2, r7
; CHECK-NEXT: vshl.s32 q0, r7
; CHECK-NEXT: vaddva.u32 r6, q2
; CHECK-NEXT: vaddva.u32 r6, q0
; CHECK-NEXT: le lr, .LBB4_8
; CHECK-NEXT: @ %bb.9: @ %middle.block
; CHECK-NEXT: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: cmp r4, r9
; CHECK-NEXT: beq .LBB4_3
; CHECK-NEXT: .LBB4_10: @ %vec.epilog.ph
; CHECK-NEXT: @ in Loop: Header=BB4_4 Depth=1
; CHECK-NEXT: ldr r1, [sp] @ 4-byte Reload
; CHECK-NEXT: add.w r2, r9, r10
; CHECK-NEXT: sub.w r5, r8, r9
; CHECK-NEXT: add.w r7, r1, r9, lsl #1
; CHECK-NEXT: add.w r2, r1, r2, lsl #1
; CHECK-NEXT: dlstp.32 lr, r5
; CHECK-NEXT: .LBB4_11: @ %vec.epilog.vector.body
; CHECK-NEXT: @ Parent Loop BB4_4 Depth=1
; CHECK-NEXT: @ => This Inner Loop Header: Depth=2
; CHECK-NEXT: rsbs r4, r3, #0
; CHECK-NEXT: vldrh.s32 q0, [r7], #8
; CHECK-NEXT: vldrh.s32 q1, [r2], #8
; CHECK-NEXT: vmul.i32 q0, q1, q0
; CHECK-NEXT: vshl.s32 q0, r4
; CHECK-NEXT: vaddva.u32 r6, q0
; CHECK-NEXT: letp lr, .LBB4_11
; CHECK-NEXT: b .LBB4_3
; CHECK-NEXT: .LBB4_12: @ %for.end17
; CHECK-NEXT: add sp, #12
; CHECK-NEXT: pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}
entry:
%conv = sext i16 %Ls to i32
%cmp31 = icmp sgt i16 %Ls, 0
br i1 %cmp31, label %for.body.lr.ph, label %for.end17
for.body.lr.ph: ; preds = %entry
%conv2 = sext i16 %DS to i32
%conv1027 = zext i16 %S to i32
%broadcast.splatinsert = insertelement <8 x i32> poison, i32 %conv1027, i64 0
%broadcast.splat = shufflevector <8 x i32> %broadcast.splatinsert, <8 x i32> poison, <8 x i32> zeroinitializer
%broadcast.splatinsert40 = insertelement <4 x i32> poison, i32 %conv1027, i64 0
%broadcast.splat41 = shufflevector <4 x i32> %broadcast.splatinsert40, <4 x i32> poison, <4 x i32> zeroinitializer
br label %for.body
for.body: ; preds = %for.body.lr.ph, %for.end
%lag.032 = phi i32 [ 0, %for.body.lr.ph ], [ %inc16, %for.end ]
%0 = sub i32 %conv2, %lag.032
%cmp428 = icmp slt i32 %lag.032, %conv2
br i1 %cmp428, label %vector.main.loop.iter.check, label %for.end
vector.main.loop.iter.check: ; preds = %for.body
%min.iters.check = icmp ult i32 %0, 8
br i1 %min.iters.check, label %vec.epilog.ph, label %vector.ph
vector.ph: ; preds = %vector.main.loop.iter.check
%n.vec = and i32 %0, -8
br label %vector.body
vector.body: ; preds = %vector.body, %vector.ph
%index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
%vec.phi = phi i32 [ 0, %vector.ph ], [ %9, %vector.body ]
%1 = getelementptr inbounds i16, ptr %ID, i32 %index
%wide.load = load <8 x i16>, ptr %1, align 2
%2 = sext <8 x i16> %wide.load to <8 x i32>
%3 = add nuw nsw i32 %index, %lag.032
%4 = getelementptr inbounds i16, ptr %ID, i32 %3
%wide.load34 = load <8 x i16>, ptr %4, align 2
%5 = sext <8 x i16> %wide.load34 to <8 x i32>
%6 = mul nsw <8 x i32> %5, %2
%7 = ashr <8 x i32> %6, %broadcast.splat
%8 = tail call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> %7)
%9 = add i32 %8, %vec.phi
%index.next = add nuw i32 %index, 8
%10 = icmp eq i32 %index.next, %n.vec
br i1 %10, label %middle.block, label %vector.body
middle.block: ; preds = %vector.body
%cmp.n = icmp eq i32 %0, %n.vec
br i1 %cmp.n, label %for.end, label %vec.epilog.ph
vec.epilog.ph: ; preds = %middle.block, %vector.main.loop.iter.check
%bc.merge.rdx = phi i32 [ 0, %vector.main.loop.iter.check ], [ %9, %middle.block ]
%vec.epilog.resume.val = phi i32 [ 0, %vector.main.loop.iter.check ], [ %n.vec, %middle.block ]
%n.rnd.up = add i32 %0, 3
%n.vec36 = and i32 %n.rnd.up, -4
br label %vec.epilog.vector.body
vec.epilog.vector.body: ; preds = %vec.epilog.vector.body, %vec.epilog.ph
%index37 = phi i32 [ %vec.epilog.resume.val, %vec.epilog.ph ], [ %index.next42, %vec.epilog.vector.body ]
%vec.phi38 = phi i32 [ %bc.merge.rdx, %vec.epilog.ph ], [ %20, %vec.epilog.vector.body ]
%active.lane.mask = tail call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index37, i32 %0)
%11 = getelementptr inbounds i16, ptr %ID, i32 %index37
%wide.masked.load = tail call <4 x i16> @llvm.masked.load.v4i16.p0(ptr %11, i32 2, <4 x i1> %active.lane.mask, <4 x i16> poison)
%12 = sext <4 x i16> %wide.masked.load to <4 x i32>
%13 = add nuw nsw i32 %index37, %lag.032
%14 = getelementptr inbounds i16, ptr %ID, i32 %13
%wide.masked.load39 = tail call <4 x i16> @llvm.masked.load.v4i16.p0(ptr %14, i32 2, <4 x i1> %active.lane.mask, <4 x i16> poison)
%15 = sext <4 x i16> %wide.masked.load39 to <4 x i32>
%16 = mul nsw <4 x i32> %15, %12
%17 = ashr <4 x i32> %16, %broadcast.splat41
%18 = select <4 x i1> %active.lane.mask, <4 x i32> %17, <4 x i32> zeroinitializer
%19 = tail call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %18)
%20 = add i32 %19, %vec.phi38
%index.next42 = add i32 %index37, 4
%21 = icmp eq i32 %index.next42, %n.vec36
br i1 %21, label %for.end, label %vec.epilog.vector.body
for.end: ; preds = %vec.epilog.vector.body, %middle.block, %for.body
%Accumulator.0.lcssa = phi i32 [ 0, %for.body ], [ %9, %middle.block ], [ %20, %vec.epilog.vector.body ]
%22 = lshr i32 %Accumulator.0.lcssa, 16
%conv13 = trunc i32 %22 to i16
%arrayidx14 = getelementptr inbounds i16, ptr %ACD, i32 %lag.032
store i16 %conv13, ptr %arrayidx14, align 2
%inc16 = add nuw nsw i32 %lag.032, 1
%exitcond33.not = icmp eq i32 %inc16, %conv
br i1 %exitcond33.not, label %for.end17, label %for.body
for.end17: ; preds = %for.end, %entry
ret void
}
declare i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %sh)
declare i16 @llvm.vector.reduce.add.v8i16(<8 x i16> %sh)
declare i32 @llvm.vector.reduce.add.v8i32(<8 x i32>)
declare <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32, i32)
declare <4 x i16> @llvm.masked.load.v4i16.p0(ptr nocapture, i32 immarg, <4 x i1>, <4 x i16>)
declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)
|
