1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -mtriple=thumbv8.1m.main -mattr=+mve %s -S -loop-reduce -o - | FileCheck %s
3 target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
4 target triple = "thumbv8.1m-arm-none-eabi"
6 define float @vctp8(float* %0, i32 %1) {
8 ; CHECK-NEXT: [[TMP3:%.*]] = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
9 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x i32>, i32 } [[TMP3]], 0
10 ; CHECK-NEXT: [[TMP5:%.*]] = add nsw i32 [[TMP1:%.*]], -1
11 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint float* [[TMP0:%.*]] to i32
12 ; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> undef, i32 [[TMP6]], i32 0
13 ; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], <i32 -32, i32 undef, i32 undef, i32 undef>
14 ; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <4 x i32> [[TMP8]], <4 x i32> undef, <4 x i32> zeroinitializer
15 ; CHECK-NEXT: [[TMP10:%.*]] = add <4 x i32> [[TMP4]], [[TMP9]]
16 ; CHECK-NEXT: br label [[TMP11:%.*]]
18 ; CHECK-NEXT: [[TMP12:%.*]] = phi i32 [ [[TMP5]], [[TMP2:%.*]] ], [ [[TMP21:%.*]], [[TMP11]] ]
19 ; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x float> [ zeroinitializer, [[TMP2]] ], [ [[TMP19:%.*]], [[TMP11]] ]
20 ; CHECK-NEXT: [[TMP14:%.*]] = phi <4 x i32> [ [[TMP10]], [[TMP2]] ], [ [[TMP17:%.*]], [[TMP11]] ]
21 ; CHECK-NEXT: [[TMP15:%.*]] = tail call <16 x i1> @llvm.arm.mve.vctp8(i32 [[TMP12]])
22 ; CHECK-NEXT: [[MASK:%.*]] = tail call <4 x i1> @v16i1_to_v4i1(<16 x i1> [[TMP15]])
23 ; CHECK-NEXT: [[TMP16:%.*]] = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> [[TMP14]], i32 32, <4 x i1> [[MASK]])
24 ; CHECK-NEXT: [[TMP17]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 1
25 ; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 0
26 ; CHECK-NEXT: [[TMP19]] = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> [[TMP13]], <4 x float> [[TMP18]], <4 x i1> [[MASK]], <4 x float> [[TMP13]])
27 ; CHECK-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP12]], 4
28 ; CHECK-NEXT: [[TMP21]] = add i32 [[TMP12]], -4
29 ; CHECK-NEXT: br i1 [[TMP20]], label [[TMP11]], label [[TMP22:%.*]]
31 ; CHECK-NEXT: [[TMP23:%.*]] = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> [[TMP19]])
32 ; CHECK-NEXT: [[TMP24:%.*]] = sitofp i32 [[TMP23]] to float
33 ; CHECK-NEXT: [[TMP25:%.*]] = tail call float @llvm.fabs.f32(float [[TMP24]])
34 ; CHECK-NEXT: ret float [[TMP25]]
36 %3 = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
37 %4 = extractvalue { <4 x i32>, i32 } %3, 0
38 %5 = add nsw i32 %1, -1
39 %6 = ptrtoint float* %0 to i32
40 %7 = insertelement <4 x i32> undef, i32 %6, i32 0
41 %8 = add <4 x i32> %7, <i32 -32, i32 undef, i32 undef, i32 undef>
42 %9 = shufflevector <4 x i32> %8, <4 x i32> undef, <4 x i32> zeroinitializer
43 %10 = add <4 x i32> %4, %9
47 %12 = phi i32 [ %5, %2 ], [ %20, %11 ]
48 %13 = phi <4 x float> [ zeroinitializer, %2 ], [ %19, %11 ]
49 %14 = phi <4 x i32> [ %10, %2 ], [ %17, %11 ]
50 %15 = tail call <16 x i1> @llvm.arm.mve.vctp8(i32 %12)
51 %mask = tail call <4 x i1> @v16i1_to_v4i1(<16 x i1> %15)
52 %16 = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> %14, i32 32, <4 x i1> %mask)
53 %17 = extractvalue { <4 x float>, <4 x i32> } %16, 1
54 %18 = extractvalue { <4 x float>, <4 x i32> } %16, 0
55 %19 = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> %13, <4 x float> %18, <4 x i1> %mask, <4 x float> %13)
56 %20 = add nsw i32 %12, -4
57 %21 = icmp sgt i32 %12, 4
58 br i1 %21, label %11, label %22
61 %23 = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> %19)
62 %24 = sitofp i32 %23 to float
63 %25 = tail call float @llvm.fabs.f32(float %24)
67 define float @vctp16(float* %0, i32 %1) {
68 ; CHECK-LABEL: @vctp16(
69 ; CHECK-NEXT: [[TMP3:%.*]] = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
70 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x i32>, i32 } [[TMP3]], 0
71 ; CHECK-NEXT: [[TMP5:%.*]] = add nsw i32 [[TMP1:%.*]], -1
72 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint float* [[TMP0:%.*]] to i32
73 ; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> undef, i32 [[TMP6]], i32 0
74 ; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], <i32 -32, i32 undef, i32 undef, i32 undef>
75 ; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <4 x i32> [[TMP8]], <4 x i32> undef, <4 x i32> zeroinitializer
76 ; CHECK-NEXT: [[TMP10:%.*]] = add <4 x i32> [[TMP4]], [[TMP9]]
77 ; CHECK-NEXT: br label [[TMP11:%.*]]
79 ; CHECK-NEXT: [[TMP12:%.*]] = phi i32 [ [[TMP5]], [[TMP2:%.*]] ], [ [[TMP21:%.*]], [[TMP11]] ]
80 ; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x float> [ zeroinitializer, [[TMP2]] ], [ [[TMP19:%.*]], [[TMP11]] ]
81 ; CHECK-NEXT: [[TMP14:%.*]] = phi <4 x i32> [ [[TMP10]], [[TMP2]] ], [ [[TMP17:%.*]], [[TMP11]] ]
82 ; CHECK-NEXT: [[TMP15:%.*]] = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 [[TMP12]])
83 ; CHECK-NEXT: [[MASK:%.*]] = tail call <4 x i1> @v8i1_to_v4i1(<8 x i1> [[TMP15]])
84 ; CHECK-NEXT: [[TMP16:%.*]] = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> [[TMP14]], i32 32, <4 x i1> [[MASK]])
85 ; CHECK-NEXT: [[TMP17]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 1
86 ; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 0
87 ; CHECK-NEXT: [[TMP19]] = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> [[TMP13]], <4 x float> [[TMP18]], <4 x i1> [[MASK]], <4 x float> [[TMP13]])
88 ; CHECK-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP12]], 4
89 ; CHECK-NEXT: [[TMP21]] = add i32 [[TMP12]], -4
90 ; CHECK-NEXT: br i1 [[TMP20]], label [[TMP11]], label [[TMP22:%.*]]
92 ; CHECK-NEXT: [[TMP23:%.*]] = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> [[TMP19]])
93 ; CHECK-NEXT: [[TMP24:%.*]] = sitofp i32 [[TMP23]] to float
94 ; CHECK-NEXT: [[TMP25:%.*]] = tail call float @llvm.fabs.f32(float [[TMP24]])
95 ; CHECK-NEXT: ret float [[TMP25]]
97 %3 = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
98 %4 = extractvalue { <4 x i32>, i32 } %3, 0
99 %5 = add nsw i32 %1, -1
100 %6 = ptrtoint float* %0 to i32
101 %7 = insertelement <4 x i32> undef, i32 %6, i32 0
102 %8 = add <4 x i32> %7, <i32 -32, i32 undef, i32 undef, i32 undef>
103 %9 = shufflevector <4 x i32> %8, <4 x i32> undef, <4 x i32> zeroinitializer
104 %10 = add <4 x i32> %4, %9
107 11: ; preds = %11, %2
108 %12 = phi i32 [ %5, %2 ], [ %20, %11 ]
109 %13 = phi <4 x float> [ zeroinitializer, %2 ], [ %19, %11 ]
110 %14 = phi <4 x i32> [ %10, %2 ], [ %17, %11 ]
111 %15 = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 %12)
112 %mask = tail call <4 x i1> @v8i1_to_v4i1(<8 x i1> %15)
113 %16 = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> %14, i32 32, <4 x i1> %mask)
114 %17 = extractvalue { <4 x float>, <4 x i32> } %16, 1
115 %18 = extractvalue { <4 x float>, <4 x i32> } %16, 0
116 %19 = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> %13, <4 x float> %18, <4 x i1> %mask, <4 x float> %13)
117 %20 = add nsw i32 %12, -4
118 %21 = icmp sgt i32 %12, 4
119 br i1 %21, label %11, label %22
122 %23 = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> %19)
123 %24 = sitofp i32 %23 to float
124 %25 = tail call float @llvm.fabs.f32(float %24)
128 define float @vctpi32(float* %0, i32 %1) {
129 ; CHECK-LABEL: @vctpi32(
130 ; CHECK-NEXT: [[TMP3:%.*]] = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
131 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x i32>, i32 } [[TMP3]], 0
132 ; CHECK-NEXT: [[TMP5:%.*]] = add nsw i32 [[TMP1:%.*]], -1
133 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint float* [[TMP0:%.*]] to i32
134 ; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> undef, i32 [[TMP6]], i32 0
135 ; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], <i32 -32, i32 undef, i32 undef, i32 undef>
136 ; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <4 x i32> [[TMP8]], <4 x i32> undef, <4 x i32> zeroinitializer
137 ; CHECK-NEXT: [[TMP10:%.*]] = add <4 x i32> [[TMP4]], [[TMP9]]
138 ; CHECK-NEXT: br label [[TMP11:%.*]]
140 ; CHECK-NEXT: [[TMP12:%.*]] = phi i32 [ [[TMP5]], [[TMP2:%.*]] ], [ [[TMP21:%.*]], [[TMP11]] ]
141 ; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x float> [ zeroinitializer, [[TMP2]] ], [ [[TMP19:%.*]], [[TMP11]] ]
142 ; CHECK-NEXT: [[TMP14:%.*]] = phi <4 x i32> [ [[TMP10]], [[TMP2]] ], [ [[TMP17:%.*]], [[TMP11]] ]
143 ; CHECK-NEXT: [[TMP15:%.*]] = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 [[TMP12]])
144 ; CHECK-NEXT: [[TMP16:%.*]] = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> [[TMP14]], i32 32, <4 x i1> [[TMP15]])
145 ; CHECK-NEXT: [[TMP17]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 1
146 ; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 0
147 ; CHECK-NEXT: [[TMP19]] = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> [[TMP13]], <4 x float> [[TMP18]], <4 x i1> [[TMP15]], <4 x float> [[TMP13]])
148 ; CHECK-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP12]], 4
149 ; CHECK-NEXT: [[TMP21]] = add i32 [[TMP12]], -4
150 ; CHECK-NEXT: br i1 [[TMP20]], label [[TMP11]], label [[TMP22:%.*]]
152 ; CHECK-NEXT: [[TMP23:%.*]] = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> [[TMP19]])
153 ; CHECK-NEXT: [[TMP24:%.*]] = sitofp i32 [[TMP23]] to float
154 ; CHECK-NEXT: [[TMP25:%.*]] = tail call float @llvm.fabs.f32(float [[TMP24]])
155 ; CHECK-NEXT: ret float [[TMP25]]
157 %3 = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
158 %4 = extractvalue { <4 x i32>, i32 } %3, 0
159 %5 = add nsw i32 %1, -1
160 %6 = ptrtoint float* %0 to i32
161 %7 = insertelement <4 x i32> undef, i32 %6, i32 0
162 %8 = add <4 x i32> %7, <i32 -32, i32 undef, i32 undef, i32 undef>
163 %9 = shufflevector <4 x i32> %8, <4 x i32> undef, <4 x i32> zeroinitializer
164 %10 = add <4 x i32> %4, %9
167 11: ; preds = %11, %2
168 %12 = phi i32 [ %5, %2 ], [ %20, %11 ]
169 %13 = phi <4 x float> [ zeroinitializer, %2 ], [ %19, %11 ]
170 %14 = phi <4 x i32> [ %10, %2 ], [ %17, %11 ]
171 %15 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %12)
172 %16 = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> %14, i32 32, <4 x i1> %15)
173 %17 = extractvalue { <4 x float>, <4 x i32> } %16, 1
174 %18 = extractvalue { <4 x float>, <4 x i32> } %16, 0
175 %19 = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> %13, <4 x float> %18, <4 x i1> %15, <4 x float> %13)
176 %20 = add nsw i32 %12, -4
177 %21 = icmp sgt i32 %12, 4
178 br i1 %21, label %11, label %22
181 %23 = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> %19)
182 %24 = sitofp i32 %23 to float
183 %25 = tail call float @llvm.fabs.f32(float %24)
188 define float @vctpi64(float* %0, i32 %1) {
189 ; CHECK-LABEL: @vctpi64(
190 ; CHECK-NEXT: [[TMP3:%.*]] = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
191 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x i32>, i32 } [[TMP3]], 0
192 ; CHECK-NEXT: [[TMP5:%.*]] = add nsw i32 [[TMP1:%.*]], -1
193 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint float* [[TMP0:%.*]] to i32
194 ; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> undef, i32 [[TMP6]], i32 0
195 ; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], <i32 -32, i32 undef, i32 undef, i32 undef>
196 ; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <4 x i32> [[TMP8]], <4 x i32> undef, <4 x i32> zeroinitializer
197 ; CHECK-NEXT: [[TMP10:%.*]] = add <4 x i32> [[TMP4]], [[TMP9]]
198 ; CHECK-NEXT: br label [[TMP11:%.*]]
200 ; CHECK-NEXT: [[TMP12:%.*]] = phi i32 [ [[TMP5]], [[TMP2:%.*]] ], [ [[TMP21:%.*]], [[TMP11]] ]
201 ; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x float> [ zeroinitializer, [[TMP2]] ], [ [[TMP19:%.*]], [[TMP11]] ]
202 ; CHECK-NEXT: [[TMP14:%.*]] = phi <4 x i32> [ [[TMP10]], [[TMP2]] ], [ [[TMP17:%.*]], [[TMP11]] ]
203 ; CHECK-NEXT: [[TMP15:%.*]] = tail call <4 x i1> @llvm.arm.mve.vctp64(i32 [[TMP12]])
204 ; CHECK-NEXT: [[TMP16:%.*]] = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> [[TMP14]], i32 32, <4 x i1> [[TMP15]])
205 ; CHECK-NEXT: [[TMP17]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 1
206 ; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <4 x float>, <4 x i32> } [[TMP16]], 0
207 ; CHECK-NEXT: [[TMP19]] = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> [[TMP13]], <4 x float> [[TMP18]], <4 x i1> [[TMP15]], <4 x float> [[TMP13]])
208 ; CHECK-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP12]], 4
209 ; CHECK-NEXT: [[TMP21]] = add i32 [[TMP12]], -4
210 ; CHECK-NEXT: br i1 [[TMP20]], label [[TMP11]], label [[TMP22:%.*]]
212 ; CHECK-NEXT: [[TMP23:%.*]] = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> [[TMP19]])
213 ; CHECK-NEXT: [[TMP24:%.*]] = sitofp i32 [[TMP23]] to float
214 ; CHECK-NEXT: [[TMP25:%.*]] = tail call float @llvm.fabs.f32(float [[TMP24]])
215 ; CHECK-NEXT: ret float [[TMP25]]
217 %3 = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 8)
218 %4 = extractvalue { <4 x i32>, i32 } %3, 0
219 %5 = add nsw i32 %1, -1
220 %6 = ptrtoint float* %0 to i32
221 %7 = insertelement <4 x i32> undef, i32 %6, i32 0
222 %8 = add <4 x i32> %7, <i32 -32, i32 undef, i32 undef, i32 undef>
223 %9 = shufflevector <4 x i32> %8, <4 x i32> undef, <4 x i32> zeroinitializer
224 %10 = add <4 x i32> %4, %9
227 11: ; preds = %11, %2
228 %12 = phi i32 [ %5, %2 ], [ %20, %11 ]
229 %13 = phi <4 x float> [ zeroinitializer, %2 ], [ %19, %11 ]
230 %14 = phi <4 x i32> [ %10, %2 ], [ %17, %11 ]
231 %15 = tail call <4 x i1> @llvm.arm.mve.vctp64(i32 %12)
232 %16 = tail call { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32> %14, i32 32, <4 x i1> %15)
233 %17 = extractvalue { <4 x float>, <4 x i32> } %16, 1
234 %18 = extractvalue { <4 x float>, <4 x i32> } %16, 0
235 %19 = tail call <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float> %13, <4 x float> %18, <4 x i1> %15, <4 x float> %13)
236 %20 = add nsw i32 %12, -4
237 %21 = icmp sgt i32 %12, 4
238 br i1 %21, label %11, label %22
241 %23 = tail call i32 bitcast (i32 (...)* @vecAddAcrossF32Mve to i32 (<4 x float>)*)(<4 x float> %19)
242 %24 = sitofp i32 %23 to float
243 %25 = tail call float @llvm.fabs.f32(float %24)
247 declare { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32, i32)
248 declare <16 x i1> @llvm.arm.mve.vctp8(i32)
249 declare <8 x i1> @llvm.arm.mve.vctp16(i32)
250 declare <4 x i1> @llvm.arm.mve.vctp32(i32)
251 declare <4 x i1> @llvm.arm.mve.vctp64(i32)
252 declare { <4 x float>, <4 x i32> } @llvm.arm.mve.vldr.gather.base.wb.predicated.v4f32.v4i32.v4i1(<4 x i32>, i32, <4 x i1>)
253 declare <4 x float> @llvm.arm.mve.add.predicated.v4f32.v4i1(<4 x float>, <4 x float>, <4 x i1>, <4 x float>)
254 declare i32 @vecAddAcrossF32Mve(...)
255 declare <4 x i1> @v8i1_to_v4i1(<8 x i1>)
256 declare <4 x i1> @v16i1_to_v4i1(<16 x i1>)
257 declare float @llvm.fabs.f32(float)