1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -enable-arm-maskedldst -enable-mem-access-versioning=false -tail-predication=force-enabled %s -o - | FileCheck %s
4 define dso_local void @mve_gather_qi_wb(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, ptr noalias nocapture %C, i32 %n, i32 %m, i32 %l) {
5 ; CHECK-LABEL: mve_gather_qi_wb:
6 ; CHECK: @ %bb.0: @ %entry
7 ; CHECK-NEXT: .save {r4, lr}
8 ; CHECK-NEXT: push {r4, lr}
9 ; CHECK-NEXT: add.w r4, r0, r3, lsl #2
10 ; CHECK-NEXT: adr r0, .LCPI0_0
11 ; CHECK-NEXT: vldrw.u32 q1, [r0]
12 ; CHECK-NEXT: vmov.i32 q0, #0x0
13 ; CHECK-NEXT: vadd.i32 q1, q1, r1
14 ; CHECK-NEXT: adds r1, r3, #4
15 ; CHECK-NEXT: dlstp.32 lr, r3
16 ; CHECK-NEXT: .LBB0_1: @ %vector.body
17 ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
18 ; CHECK-NEXT: vldrw.u32 q2, [r4], #16
19 ; CHECK-NEXT: vldrw.u32 q3, [q1, #80]!
20 ; CHECK-NEXT: vmul.i32 q2, q3, q2
21 ; CHECK-NEXT: vadd.i32 q0, q0, q2
22 ; CHECK-NEXT: letp lr, .LBB0_1
23 ; CHECK-NEXT: @ %bb.2: @ %middle.block
24 ; CHECK-NEXT: vaddv.u32 r0, q0
25 ; CHECK-NEXT: str.w r0, [r2, r1, lsl #2]
26 ; CHECK-NEXT: pop {r4, pc}
27 ; CHECK-NEXT: .p2align 4
28 ; CHECK-NEXT: @ %bb.3:
29 ; CHECK-NEXT: .LCPI0_0:
30 ; CHECK-NEXT: .long 4294967228 @ 0xffffffbc
31 ; CHECK-NEXT: .long 4294967248 @ 0xffffffd0
32 ; CHECK-NEXT: .long 4294967268 @ 0xffffffe4
33 ; CHECK-NEXT: .long 4294967288 @ 0xfffffff8
34 entry: ; preds = %middle.
35 %add.us.us = add i32 4, %n
36 %arrayidx.us.us = getelementptr inbounds i32, ptr %C, i32 %add.us.us
38 vector.body: ; preds = %vector.body, %entry
39 %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
40 %vec.phi = phi <4 x i32> [ zeroinitializer, %entry ], [ %7, %vector.body ]
41 %vec.ind = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %entry ], [ %vec.ind.next, %vector.body ]
42 %0 = add i32 %index, %n
43 %1 = getelementptr inbounds i32, ptr %A, i32 %0
44 %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 %n)
45 %2 = bitcast ptr %1 to ptr
46 %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %2, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
47 %3 = mul <4 x i32> %vec.ind, <i32 5, i32 5, i32 5, i32 5>
48 %4 = add <4 x i32> %3, <i32 3, i32 3, i32 3, i32 3>
49 %5 = getelementptr inbounds i32, ptr %B, <4 x i32> %4
50 %wide.masked.gather = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %5, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
51 %6 = mul nsw <4 x i32> %wide.masked.gather, %wide.masked.load
52 %7 = add <4 x i32> %vec.phi, %6
53 %index.next = add i32 %index, 4
54 %vec.ind.next = add <4 x i32> %vec.ind, <i32 4, i32 4, i32 4, i32 4>
55 %8 = icmp eq i32 %index.next, 5000
56 br i1 %8, label %middle.block, label %vector.body
57 middle.block: ; preds = %vector.body
58 %9 = select <4 x i1> %active.lane.mask, <4 x i32> %7, <4 x i32> %vec.phi
59 %10 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %7)
60 store i32 %10, ptr %arrayidx.us.us, align 4
61 %inc21.us.us = add nuw i32 4, 1
62 %exitcond81.not = icmp eq i32 %inc21.us.us, %n
64 end: ; preds = %middle.block
68 define dso_local void @mve_gatherscatter_offset(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, ptr noalias nocapture %C, i32 %n, i32 %m, i32 %l) {
69 ; CHECK-LABEL: mve_gatherscatter_offset:
70 ; CHECK: @ %bb.0: @ %entry
71 ; CHECK-NEXT: .save {r4, lr}
72 ; CHECK-NEXT: push {r4, lr}
73 ; CHECK-NEXT: .vsave {d8, d9}
74 ; CHECK-NEXT: vpush {d8, d9}
75 ; CHECK-NEXT: add.w r4, r0, r3, lsl #2
76 ; CHECK-NEXT: adr r0, .LCPI1_0
77 ; CHECK-NEXT: vldrw.u32 q2, [r0]
78 ; CHECK-NEXT: add.w r12, r3, #4
79 ; CHECK-NEXT: vmov.i32 q0, #0x0
80 ; CHECK-NEXT: vmov.i32 q1, #0x14
81 ; CHECK-NEXT: dlstp.32 lr, r3
82 ; CHECK-NEXT: .LBB1_1: @ %vector.body
83 ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
84 ; CHECK-NEXT: vldrw.u32 q3, [r1, q2, uxtw #2]
85 ; CHECK-NEXT: vldrw.u32 q4, [r4], #16
86 ; CHECK-NEXT: vmul.i32 q3, q3, q4
87 ; CHECK-NEXT: vstrw.32 q3, [r1, q2, uxtw #2]
88 ; CHECK-NEXT: vadd.i32 q2, q2, q1
89 ; CHECK-NEXT: vadd.i32 q0, q0, q3
90 ; CHECK-NEXT: letp lr, .LBB1_1
91 ; CHECK-NEXT: @ %bb.2: @ %middle.block
92 ; CHECK-NEXT: vaddv.u32 r0, q0
93 ; CHECK-NEXT: str.w r0, [r2, r12, lsl #2]
94 ; CHECK-NEXT: vpop {d8, d9}
95 ; CHECK-NEXT: pop {r4, pc}
96 ; CHECK-NEXT: .p2align 4
97 ; CHECK-NEXT: @ %bb.3:
98 ; CHECK-NEXT: .LCPI1_0:
99 ; CHECK-NEXT: .long 3 @ 0x3
100 ; CHECK-NEXT: .long 8 @ 0x8
101 ; CHECK-NEXT: .long 13 @ 0xd
102 ; CHECK-NEXT: .long 18 @ 0x12
103 entry: ; preds = %middle.
104 %add.us.us = add i32 4, %n
105 %arrayidx.us.us = getelementptr inbounds i32, ptr %C, i32 %add.us.us
106 br label %vector.body
107 vector.body: ; preds = %vector.body, %entry
108 %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
109 %vec.phi = phi <4 x i32> [ zeroinitializer, %entry ], [ %7, %vector.body ]
110 %vec.ind = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %entry ], [ %vec.ind.next, %vector.body ]
111 %0 = add i32 %index, %n
112 %1 = getelementptr inbounds i32, ptr %A, i32 %0
113 %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 %n)
114 %2 = bitcast ptr %1 to ptr
115 %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %2, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
116 %3 = mul <4 x i32> %vec.ind, <i32 5, i32 5, i32 5, i32 5>
117 %4 = add <4 x i32> %3, <i32 3, i32 3, i32 3, i32 3>
118 %5 = getelementptr inbounds i32, ptr %B, <4 x i32> %4
119 %wide.masked.gather = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %5, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
120 %6 = mul nsw <4 x i32> %wide.masked.gather, %wide.masked.load
121 call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> %6, <4 x ptr> %5, i32 4, <4 x i1> %active.lane.mask)
122 %7 = add <4 x i32> %vec.phi, %6
123 %index.next = add i32 %index, 4
124 %vec.ind.next = add <4 x i32> %vec.ind, <i32 4, i32 4, i32 4, i32 4>
125 %8 = icmp eq i32 %index.next, 5000
126 br i1 %8, label %middle.block, label %vector.body
127 middle.block: ; preds = %vector.body
128 %9 = select <4 x i1> %active.lane.mask, <4 x i32> %7, <4 x i32> %vec.phi
129 %10 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %7)
130 store i32 %10, ptr %arrayidx.us.us, align 4
131 %inc21.us.us = add nuw i32 4, 1
132 %exitcond81.not = icmp eq i32 %inc21.us.us, %n
134 end: ; preds = %middle.block
138 define dso_local void @mve_scatter_qi(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, ptr noalias nocapture %C, i32 %n, i32 %m, i32 %l) {
139 ; CHECK-LABEL: mve_scatter_qi:
140 ; CHECK: @ %bb.0: @ %entry
141 ; CHECK-NEXT: .save {r4, lr}
142 ; CHECK-NEXT: push {r4, lr}
143 ; CHECK-NEXT: add.w r4, r0, r3, lsl #2
144 ; CHECK-NEXT: adr r0, .LCPI2_0
145 ; CHECK-NEXT: vldrw.u32 q1, [r0]
146 ; CHECK-NEXT: add.w r12, r3, #4
147 ; CHECK-NEXT: vmov.i32 q0, #0x0
148 ; CHECK-NEXT: movw lr, #1250
149 ; CHECK-NEXT: vadd.i32 q1, q1, r1
150 ; CHECK-NEXT: movs r1, #3
151 ; CHECK-NEXT: .LBB2_1: @ %vector.body
152 ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
153 ; CHECK-NEXT: vctp.32 r3
154 ; CHECK-NEXT: subs r3, #4
156 ; CHECK-NEXT: vldrwt.u32 q2, [r4], #16
157 ; CHECK-NEXT: vmul.i32 q3, q2, r1
158 ; CHECK-NEXT: vmla.i32 q0, q2, r1
160 ; CHECK-NEXT: vstrwt.32 q3, [q1, #80]!
161 ; CHECK-NEXT: le lr, .LBB2_1
162 ; CHECK-NEXT: @ %bb.2: @ %middle.block
163 ; CHECK-NEXT: vaddv.u32 r0, q0
164 ; CHECK-NEXT: str.w r0, [r2, r12, lsl #2]
165 ; CHECK-NEXT: pop {r4, pc}
166 ; CHECK-NEXT: .p2align 4
167 ; CHECK-NEXT: @ %bb.3:
168 ; CHECK-NEXT: .LCPI2_0:
169 ; CHECK-NEXT: .long 4294967228 @ 0xffffffbc
170 ; CHECK-NEXT: .long 4294967248 @ 0xffffffd0
171 ; CHECK-NEXT: .long 4294967268 @ 0xffffffe4
172 ; CHECK-NEXT: .long 4294967288 @ 0xfffffff8
173 entry: ; preds = %middle.
174 %add.us.us = add i32 4, %n
175 %arrayidx.us.us = getelementptr inbounds i32, ptr %C, i32 %add.us.us
176 br label %vector.body
177 vector.body: ; preds = %vector.body, %entry
178 %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
179 %vec.phi = phi <4 x i32> [ zeroinitializer, %entry ], [ %7, %vector.body ]
180 %vec.ind = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %entry ], [ %vec.ind.next, %vector.body ]
181 %0 = add i32 %index, %n
182 %1 = getelementptr inbounds i32, ptr %A, i32 %0
183 %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 %n)
184 %2 = bitcast ptr %1 to ptr
185 %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %2, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
186 %3 = mul <4 x i32> %vec.ind, <i32 5, i32 5, i32 5, i32 5>
187 %4 = add <4 x i32> %3, <i32 3, i32 3, i32 3, i32 3>
188 %5 = getelementptr inbounds i32, ptr %B, <4 x i32> %4
189 %6 = mul nsw <4 x i32> <i32 3, i32 3, i32 3, i32 3>, %wide.masked.load
190 call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> %6, <4 x ptr> %5, i32 4, <4 x i1> %active.lane.mask)
191 %7 = add <4 x i32> %vec.phi, %6
192 %index.next = add i32 %index, 4
193 %vec.ind.next = add <4 x i32> %vec.ind, <i32 4, i32 4, i32 4, i32 4>
194 %8 = icmp eq i32 %index.next, 5000
195 br i1 %8, label %middle.block, label %vector.body
196 middle.block: ; preds = %vector.body
197 %9 = select <4 x i1> %active.lane.mask, <4 x i32> %7, <4 x i32> %vec.phi
198 %10 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %7)
199 store i32 %10, ptr %arrayidx.us.us, align 4
200 %inc21.us.us = add nuw i32 4, 1
201 %exitcond81.not = icmp eq i32 %inc21.us.us, %n
203 end: ; preds = %middle.block
207 define void @justoffsets(ptr noalias nocapture readonly %r, ptr noalias nocapture %w, i32 %N) {
208 ; CHECK-LABEL: justoffsets:
209 ; CHECK: @ %bb.0: @ %entry
210 ; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
211 ; CHECK-NEXT: push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}
212 ; CHECK-NEXT: .pad #4
213 ; CHECK-NEXT: sub sp, #4
214 ; CHECK-NEXT: .vsave {d8, d9, d10, d11, d12, d13}
215 ; CHECK-NEXT: vpush {d8, d9, d10, d11, d12, d13}
216 ; CHECK-NEXT: cmp r2, #0
217 ; CHECK-NEXT: beq .LBB3_3
218 ; CHECK-NEXT: @ %bb.1: @ %vector.ph
219 ; CHECK-NEXT: adr r5, .LCPI3_1
220 ; CHECK-NEXT: adr r4, .LCPI3_0
221 ; CHECK-NEXT: vldrw.u32 q0, [r5]
222 ; CHECK-NEXT: adr r5, .LCPI3_2
223 ; CHECK-NEXT: movw r9, #47888
224 ; CHECK-NEXT: movw r10, #50417
225 ; CHECK-NEXT: vldrw.u32 q1, [r5]
226 ; CHECK-NEXT: vldrw.u32 q2, [r4]
227 ; CHECK-NEXT: movw r4, #32769
228 ; CHECK-NEXT: movw r12, #7471
229 ; CHECK-NEXT: mov.w r3, #32768
230 ; CHECK-NEXT: movw r11, #38470
231 ; CHECK-NEXT: movw r8, #19595
232 ; CHECK-NEXT: movt r9, #65535
233 ; CHECK-NEXT: movt r10, #65535
234 ; CHECK-NEXT: movw r7, #32767
235 ; CHECK-NEXT: movt r4, #65535
236 ; CHECK-NEXT: movw r5, #13282
237 ; CHECK-NEXT: movw r6, #19485
238 ; CHECK-NEXT: dlstp.32 lr, r2
239 ; CHECK-NEXT: .LBB3_2: @ %vector.body
240 ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
241 ; CHECK-NEXT: vldrb.u32 q4, [r0, q0]
242 ; CHECK-NEXT: vldrb.u32 q3, [r0, q1]
243 ; CHECK-NEXT: vldrb.u32 q5, [r0, q2]
244 ; CHECK-NEXT: adds r0, #12
245 ; CHECK-NEXT: vmul.i32 q6, q4, r11
246 ; CHECK-NEXT: vmla.i32 q6, q3, r8
247 ; CHECK-NEXT: vmla.i32 q6, q5, r12
248 ; CHECK-NEXT: vadd.i32 q6, q6, r3
249 ; CHECK-NEXT: vshr.u32 q6, q6, #16
250 ; CHECK-NEXT: vstrb.32 q6, [r1, q1]
251 ; CHECK-NEXT: vmul.i32 q6, q4, r4
252 ; CHECK-NEXT: vmul.i32 q4, q4, r10
253 ; CHECK-NEXT: vmla.i32 q6, q3, r5
254 ; CHECK-NEXT: vmla.i32 q4, q3, r7
255 ; CHECK-NEXT: vmla.i32 q6, q5, r6
256 ; CHECK-NEXT: vmla.i32 q4, q5, r9
257 ; CHECK-NEXT: vadd.i32 q6, q6, r3
258 ; CHECK-NEXT: vadd.i32 q3, q4, r3
259 ; CHECK-NEXT: vshr.u32 q6, q6, #16
260 ; CHECK-NEXT: vshr.u32 q3, q3, #16
261 ; CHECK-NEXT: vstrb.32 q3, [r1, q0]
262 ; CHECK-NEXT: vstrb.32 q6, [r1, q2]
263 ; CHECK-NEXT: adds r1, #12
264 ; CHECK-NEXT: letp lr, .LBB3_2
265 ; CHECK-NEXT: .LBB3_3: @ %for.cond.cleanup
266 ; CHECK-NEXT: vpop {d8, d9, d10, d11, d12, d13}
267 ; CHECK-NEXT: add sp, #4
268 ; CHECK-NEXT: pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}
269 ; CHECK-NEXT: .p2align 4
270 ; CHECK-NEXT: @ %bb.4:
271 ; CHECK-NEXT: .LCPI3_0:
272 ; CHECK-NEXT: .long 2 @ 0x2
273 ; CHECK-NEXT: .long 5 @ 0x5
274 ; CHECK-NEXT: .long 8 @ 0x8
275 ; CHECK-NEXT: .long 11 @ 0xb
276 ; CHECK-NEXT: .LCPI3_1:
277 ; CHECK-NEXT: .long 1 @ 0x1
278 ; CHECK-NEXT: .long 4 @ 0x4
279 ; CHECK-NEXT: .long 7 @ 0x7
280 ; CHECK-NEXT: .long 10 @ 0xa
281 ; CHECK-NEXT: .LCPI3_2:
282 ; CHECK-NEXT: .long 0 @ 0x0
283 ; CHECK-NEXT: .long 3 @ 0x3
284 ; CHECK-NEXT: .long 6 @ 0x6
285 ; CHECK-NEXT: .long 9 @ 0x9
287 %cmp47.not = icmp eq i32 %N, 0
288 br i1 %cmp47.not, label %for.cond.cleanup, label %vector.ph
290 vector.ph: ; preds = %vector.memcheck
291 %n.rnd.up = add i32 %N, 3
292 %n.vec = and i32 %n.rnd.up, -4
293 br label %vector.body
295 vector.body: ; preds = %vector.body, %vector.ph
296 %pointer.phi = phi ptr [ %r, %vector.ph ], [ %ptr.ind, %vector.body ]
297 %pointer.phi55 = phi ptr [ %w, %vector.ph ], [ %ptr.ind56, %vector.body ]
298 %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
299 %l1 = getelementptr i8, ptr %pointer.phi, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
300 %l2 = getelementptr i8, ptr %pointer.phi55, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
301 %l3 = getelementptr inbounds i8, <4 x ptr> %l1, i32 1
302 %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 %N)
303 %wide.masked.gather = call <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr> %l1, i32 1, <4 x i1> %active.lane.mask, <4 x i8> undef)
304 %l4 = getelementptr inbounds i8, <4 x ptr> %l1, i32 2
305 %wide.masked.gather57 = call <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr> %l3, i32 1, <4 x i1> %active.lane.mask, <4 x i8> undef)
306 %wide.masked.gather58 = call <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr> %l4, i32 1, <4 x i1> %active.lane.mask, <4 x i8> undef)
307 %l5 = zext <4 x i8> %wide.masked.gather to <4 x i32>
308 %l6 = mul nuw nsw <4 x i32> %l5, <i32 19595, i32 19595, i32 19595, i32 19595>
309 %l7 = zext <4 x i8> %wide.masked.gather57 to <4 x i32>
310 %l8 = mul nuw nsw <4 x i32> %l7, <i32 38470, i32 38470, i32 38470, i32 38470>
311 %l9 = zext <4 x i8> %wide.masked.gather58 to <4 x i32>
312 %l10 = mul nuw nsw <4 x i32> %l9, <i32 7471, i32 7471, i32 7471, i32 7471>
313 %l11 = add nuw nsw <4 x i32> %l6, <i32 32768, i32 32768, i32 32768, i32 32768>
314 %l12 = add nuw nsw <4 x i32> %l11, %l8
315 %l13 = add nuw nsw <4 x i32> %l12, %l10
316 %l14 = lshr <4 x i32> %l13, <i32 16, i32 16, i32 16, i32 16>
317 %l15 = trunc <4 x i32> %l14 to <4 x i8>
318 %l16 = mul nuw nsw <4 x i32> %l5, <i32 32767, i32 32767, i32 32767, i32 32767>
319 %l17 = mul nsw <4 x i32> %l7, <i32 -15119, i32 -15119, i32 -15119, i32 -15119>
320 %l18 = mul nsw <4 x i32> %l9, <i32 -17648, i32 -17648, i32 -17648, i32 -17648>
321 %l19 = add nuw nsw <4 x i32> %l16, <i32 32768, i32 32768, i32 32768, i32 32768>
322 %l20 = add nsw <4 x i32> %l19, %l17
323 %l21 = add nsw <4 x i32> %l20, %l18
324 %l22 = lshr <4 x i32> %l21, <i32 16, i32 16, i32 16, i32 16>
325 %l23 = trunc <4 x i32> %l22 to <4 x i8>
326 %l24 = mul nuw nsw <4 x i32> %l5, <i32 13282, i32 13282, i32 13282, i32 13282>
327 %l25 = mul nsw <4 x i32> %l7, <i32 -32767, i32 -32767, i32 -32767, i32 -32767>
328 %l26 = mul nuw nsw <4 x i32> %l9, <i32 19485, i32 19485, i32 19485, i32 19485>
329 %l27 = add nuw nsw <4 x i32> %l24, <i32 32768, i32 32768, i32 32768, i32 32768>
330 %l28 = add nsw <4 x i32> %l27, %l25
331 %l29 = add nsw <4 x i32> %l28, %l26
332 %l30 = lshr <4 x i32> %l29, <i32 16, i32 16, i32 16, i32 16>
333 %l31 = trunc <4 x i32> %l30 to <4 x i8>
334 %l32 = getelementptr inbounds i8, <4 x ptr> %l2, i32 1
335 call void @llvm.masked.scatter.v4i8.v4p0(<4 x i8> %l15, <4 x ptr> %l2, i32 1, <4 x i1> %active.lane.mask)
336 %l33 = getelementptr inbounds i8, <4 x ptr> %l2, i32 2
337 call void @llvm.masked.scatter.v4i8.v4p0(<4 x i8> %l23, <4 x ptr> %l32, i32 1, <4 x i1> %active.lane.mask)
338 call void @llvm.masked.scatter.v4i8.v4p0(<4 x i8> %l31, <4 x ptr> %l33, i32 1, <4 x i1> %active.lane.mask)
339 %index.next = add i32 %index, 4
340 %l34 = icmp eq i32 %index.next, %n.vec
341 %ptr.ind = getelementptr i8, ptr %pointer.phi, i32 12
342 %ptr.ind56 = getelementptr i8, ptr %pointer.phi55, i32 12
343 br i1 %l34, label %for.cond.cleanup, label %vector.body
345 for.cond.cleanup: ; preds = %vector.body, %for.body, %entry
349 declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)
350 declare <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x i32>)
351 declare <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x i8>)
352 declare <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32, i32)
353 declare <4 x i32> @llvm.masked.load.v4i32.p0(ptr, i32, <4 x i1>, <4 x i32>)
354 declare void @llvm.masked.scatter.v4i32.v4p0(<4 x i32>, <4 x ptr>, i32, <4 x i1>)
355 declare void @llvm.masked.scatter.v4i8.v4p0(<4 x i8>, <4 x ptr>, i32, <4 x i1>)