1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
2 // RUN: %clang_cc1 -triple thumbv8.1m.main-none-none-eabi -target-feature +mve.fp -mfloat-abi hard -O0 -disable-O0-optnone -emit-llvm -o - %s | opt -S -passes=mem2reg | FileCheck %s
3 // RUN: %clang_cc1 -DPOLYMORPHIC -triple thumbv8.1m.main-none-none-eabi -target-feature +mve.fp -mfloat-abi hard -O0 -disable-O0-optnone -emit-llvm -o - %s | opt -S -passes=mem2reg | FileCheck %s
5 // REQUIRES: aarch64-registered-target || arm-registered-target
9 // CHECK-LABEL: @test_vcvtq_f16_s16(
11 // CHECK-NEXT: [[TMP0:%.*]] = sitofp <8 x i16> [[A:%.*]] to <8 x half>
12 // CHECK-NEXT: ret <8 x half> [[TMP0]]
14 float16x8_t
test_vcvtq_f16_s16(int16x8_t a
)
18 #else /* POLYMORPHIC */
19 return vcvtq_f16_s16(a
);
20 #endif /* POLYMORPHIC */
23 // CHECK-LABEL: @test_vcvtq_f16_u16(
25 // CHECK-NEXT: [[TMP0:%.*]] = uitofp <8 x i16> [[A:%.*]] to <8 x half>
26 // CHECK-NEXT: ret <8 x half> [[TMP0]]
28 float16x8_t
test_vcvtq_f16_u16(uint16x8_t a
)
32 #else /* POLYMORPHIC */
33 return vcvtq_f16_u16(a
);
34 #endif /* POLYMORPHIC */
37 // CHECK-LABEL: @test_vcvtq_f32_s32(
39 // CHECK-NEXT: [[TMP0:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
40 // CHECK-NEXT: ret <4 x float> [[TMP0]]
42 float32x4_t
test_vcvtq_f32_s32(int32x4_t a
)
46 #else /* POLYMORPHIC */
47 return vcvtq_f32_s32(a
);
48 #endif /* POLYMORPHIC */
51 // CHECK-LABEL: @test_vcvtq_f32_u32(
53 // CHECK-NEXT: [[TMP0:%.*]] = uitofp <4 x i32> [[A:%.*]] to <4 x float>
54 // CHECK-NEXT: ret <4 x float> [[TMP0]]
56 float32x4_t
test_vcvtq_f32_u32(uint32x4_t a
)
60 #else /* POLYMORPHIC */
61 return vcvtq_f32_u32(a
);
62 #endif /* POLYMORPHIC */
65 // CHECK-LABEL: @test_vcvtq_s16_f16(
67 // CHECK-NEXT: [[TMP0:%.*]] = fptosi <8 x half> [[A:%.*]] to <8 x i16>
68 // CHECK-NEXT: ret <8 x i16> [[TMP0]]
70 int16x8_t
test_vcvtq_s16_f16(float16x8_t a
)
72 return vcvtq_s16_f16(a
);
75 // CHECK-LABEL: @test_vcvtq_s32_f32(
77 // CHECK-NEXT: [[TMP0:%.*]] = fptosi <4 x float> [[A:%.*]] to <4 x i32>
78 // CHECK-NEXT: ret <4 x i32> [[TMP0]]
80 int32x4_t
test_vcvtq_s32_f32(float32x4_t a
)
82 return vcvtq_s32_f32(a
);
85 // CHECK-LABEL: @test_vcvtq_u16_f16(
87 // CHECK-NEXT: [[TMP0:%.*]] = fptoui <8 x half> [[A:%.*]] to <8 x i16>
88 // CHECK-NEXT: ret <8 x i16> [[TMP0]]
90 uint16x8_t
test_vcvtq_u16_f16(float16x8_t a
)
92 return vcvtq_u16_f16(a
);
95 // CHECK-LABEL: @test_vcvtq_u32_f32(
97 // CHECK-NEXT: [[TMP0:%.*]] = fptoui <4 x float> [[A:%.*]] to <4 x i32>
98 // CHECK-NEXT: ret <4 x i32> [[TMP0]]
100 uint32x4_t
test_vcvtq_u32_f32(float32x4_t a
)
102 return vcvtq_u32_f32(a
);
105 // CHECK-LABEL: @test_vcvtq_m_f16_s16(
106 // CHECK-NEXT: entry:
107 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
108 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
109 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fp.int.predicated.v8f16.v8i16.v8i1(<8 x i16> [[A:%.*]], i32 0, <8 x i1> [[TMP1]], <8 x half> [[INACTIVE:%.*]])
110 // CHECK-NEXT: ret <8 x half> [[TMP2]]
112 float16x8_t
test_vcvtq_m_f16_s16(float16x8_t inactive
, int16x8_t a
, mve_pred16_t p
)
115 return vcvtq_m(inactive
, a
, p
);
116 #else /* POLYMORPHIC */
117 return vcvtq_m_f16_s16(inactive
, a
, p
);
118 #endif /* POLYMORPHIC */
121 // CHECK-LABEL: @test_vcvtq_m_f16_u16(
122 // CHECK-NEXT: entry:
123 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
124 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
125 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fp.int.predicated.v8f16.v8i16.v8i1(<8 x i16> [[A:%.*]], i32 1, <8 x i1> [[TMP1]], <8 x half> [[INACTIVE:%.*]])
126 // CHECK-NEXT: ret <8 x half> [[TMP2]]
128 float16x8_t
test_vcvtq_m_f16_u16(float16x8_t inactive
, uint16x8_t a
, mve_pred16_t p
)
131 return vcvtq_m(inactive
, a
, p
);
132 #else /* POLYMORPHIC */
133 return vcvtq_m_f16_u16(inactive
, a
, p
);
134 #endif /* POLYMORPHIC */
137 // CHECK-LABEL: @test_vcvtq_m_f32_s32(
138 // CHECK-NEXT: entry:
139 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
140 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
141 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fp.int.predicated.v4f32.v4i32.v4i1(<4 x i32> [[A:%.*]], i32 0, <4 x i1> [[TMP1]], <4 x float> [[INACTIVE:%.*]])
142 // CHECK-NEXT: ret <4 x float> [[TMP2]]
144 float32x4_t
test_vcvtq_m_f32_s32(float32x4_t inactive
, int32x4_t a
, mve_pred16_t p
)
147 return vcvtq_m(inactive
, a
, p
);
148 #else /* POLYMORPHIC */
149 return vcvtq_m_f32_s32(inactive
, a
, p
);
150 #endif /* POLYMORPHIC */
153 // CHECK-LABEL: @test_vcvtq_m_f32_u32(
154 // CHECK-NEXT: entry:
155 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
156 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
157 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fp.int.predicated.v4f32.v4i32.v4i1(<4 x i32> [[A:%.*]], i32 1, <4 x i1> [[TMP1]], <4 x float> [[INACTIVE:%.*]])
158 // CHECK-NEXT: ret <4 x float> [[TMP2]]
160 float32x4_t
test_vcvtq_m_f32_u32(float32x4_t inactive
, uint32x4_t a
, mve_pred16_t p
)
163 return vcvtq_m(inactive
, a
, p
);
164 #else /* POLYMORPHIC */
165 return vcvtq_m_f32_u32(inactive
, a
, p
);
166 #endif /* POLYMORPHIC */
169 // CHECK-LABEL: @test_vcvtq_m_s16_f16(
170 // CHECK-NEXT: entry:
171 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
172 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
173 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fp.int.predicated.v8i16.v8f16.v8i1(<8 x half> [[A:%.*]], i32 0, <8 x i1> [[TMP1]], <8 x i16> [[INACTIVE:%.*]])
174 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
176 int16x8_t
test_vcvtq_m_s16_f16(int16x8_t inactive
, float16x8_t a
, mve_pred16_t p
)
179 return vcvtq_m(inactive
, a
, p
);
180 #else /* POLYMORPHIC */
181 return vcvtq_m_s16_f16(inactive
, a
, p
);
182 #endif /* POLYMORPHIC */
185 // CHECK-LABEL: @test_vcvtq_m_s32_f32(
186 // CHECK-NEXT: entry:
187 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
188 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
189 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fp.int.predicated.v4i32.v4f32.v4i1(<4 x float> [[A:%.*]], i32 0, <4 x i1> [[TMP1]], <4 x i32> [[INACTIVE:%.*]])
190 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
192 int32x4_t
test_vcvtq_m_s32_f32(int32x4_t inactive
, float32x4_t a
, mve_pred16_t p
)
195 return vcvtq_m(inactive
, a
, p
);
196 #else /* POLYMORPHIC */
197 return vcvtq_m_s32_f32(inactive
, a
, p
);
198 #endif /* POLYMORPHIC */
201 // CHECK-LABEL: @test_vcvtq_m_u16_f16(
202 // CHECK-NEXT: entry:
203 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
204 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
205 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fp.int.predicated.v8i16.v8f16.v8i1(<8 x half> [[A:%.*]], i32 1, <8 x i1> [[TMP1]], <8 x i16> [[INACTIVE:%.*]])
206 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
208 uint16x8_t
test_vcvtq_m_u16_f16(uint16x8_t inactive
, float16x8_t a
, mve_pred16_t p
)
211 return vcvtq_m(inactive
, a
, p
);
212 #else /* POLYMORPHIC */
213 return vcvtq_m_u16_f16(inactive
, a
, p
);
214 #endif /* POLYMORPHIC */
217 // CHECK-LABEL: @test_vcvtq_m_u32_f32(
218 // CHECK-NEXT: entry:
219 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
220 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
221 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fp.int.predicated.v4i32.v4f32.v4i1(<4 x float> [[A:%.*]], i32 1, <4 x i1> [[TMP1]], <4 x i32> [[INACTIVE:%.*]])
222 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
224 uint32x4_t
test_vcvtq_m_u32_f32(uint32x4_t inactive
, float32x4_t a
, mve_pred16_t p
)
227 return vcvtq_m(inactive
, a
, p
);
228 #else /* POLYMORPHIC */
229 return vcvtq_m_u32_f32(inactive
, a
, p
);
230 #endif /* POLYMORPHIC */
233 // CHECK-LABEL: @test_vcvtq_x_f16_s16(
234 // CHECK-NEXT: entry:
235 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
236 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
237 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fp.int.predicated.v8f16.v8i16.v8i1(<8 x i16> [[A:%.*]], i32 0, <8 x i1> [[TMP1]], <8 x half> undef)
238 // CHECK-NEXT: ret <8 x half> [[TMP2]]
240 float16x8_t
test_vcvtq_x_f16_s16(int16x8_t a
, mve_pred16_t p
)
243 return vcvtq_x(a
, p
);
244 #else /* POLYMORPHIC */
245 return vcvtq_x_f16_s16(a
, p
);
246 #endif /* POLYMORPHIC */
249 // CHECK-LABEL: @test_vcvtq_x_f16_u16(
250 // CHECK-NEXT: entry:
251 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
252 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
253 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fp.int.predicated.v8f16.v8i16.v8i1(<8 x i16> [[A:%.*]], i32 1, <8 x i1> [[TMP1]], <8 x half> undef)
254 // CHECK-NEXT: ret <8 x half> [[TMP2]]
256 float16x8_t
test_vcvtq_x_f16_u16(uint16x8_t a
, mve_pred16_t p
)
259 return vcvtq_x(a
, p
);
260 #else /* POLYMORPHIC */
261 return vcvtq_x_f16_u16(a
, p
);
262 #endif /* POLYMORPHIC */
265 // CHECK-LABEL: @test_vcvtq_x_f32_s32(
266 // CHECK-NEXT: entry:
267 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
268 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
269 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fp.int.predicated.v4f32.v4i32.v4i1(<4 x i32> [[A:%.*]], i32 0, <4 x i1> [[TMP1]], <4 x float> undef)
270 // CHECK-NEXT: ret <4 x float> [[TMP2]]
272 float32x4_t
test_vcvtq_x_f32_s32(int32x4_t a
, mve_pred16_t p
)
275 return vcvtq_x(a
, p
);
276 #else /* POLYMORPHIC */
277 return vcvtq_x_f32_s32(a
, p
);
278 #endif /* POLYMORPHIC */
281 // CHECK-LABEL: @test_vcvtq_x_f32_u32(
282 // CHECK-NEXT: entry:
283 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
284 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
285 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fp.int.predicated.v4f32.v4i32.v4i1(<4 x i32> [[A:%.*]], i32 1, <4 x i1> [[TMP1]], <4 x float> undef)
286 // CHECK-NEXT: ret <4 x float> [[TMP2]]
288 float32x4_t
test_vcvtq_x_f32_u32(uint32x4_t a
, mve_pred16_t p
)
291 return vcvtq_x(a
, p
);
292 #else /* POLYMORPHIC */
293 return vcvtq_x_f32_u32(a
, p
);
294 #endif /* POLYMORPHIC */
297 // CHECK-LABEL: @test_vcvtq_x_s16_f16(
298 // CHECK-NEXT: entry:
299 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
300 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
301 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fp.int.predicated.v8i16.v8f16.v8i1(<8 x half> [[A:%.*]], i32 0, <8 x i1> [[TMP1]], <8 x i16> undef)
302 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
304 int16x8_t
test_vcvtq_x_s16_f16(float16x8_t a
, mve_pred16_t p
)
306 return vcvtq_x_s16_f16(a
, p
);
309 // CHECK-LABEL: @test_vcvtq_x_s32_f32(
310 // CHECK-NEXT: entry:
311 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
312 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
313 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fp.int.predicated.v4i32.v4f32.v4i1(<4 x float> [[A:%.*]], i32 0, <4 x i1> [[TMP1]], <4 x i32> undef)
314 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
316 int32x4_t
test_vcvtq_x_s32_f32(float32x4_t a
, mve_pred16_t p
)
318 return vcvtq_x_s32_f32(a
, p
);
321 // CHECK-LABEL: @test_vcvtq_x_u16_f16(
322 // CHECK-NEXT: entry:
323 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
324 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
325 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fp.int.predicated.v8i16.v8f16.v8i1(<8 x half> [[A:%.*]], i32 1, <8 x i1> [[TMP1]], <8 x i16> undef)
326 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
328 uint16x8_t
test_vcvtq_x_u16_f16(float16x8_t a
, mve_pred16_t p
)
330 return vcvtq_x_u16_f16(a
, p
);
333 // CHECK-LABEL: @test_vcvtq_x_u32_f32(
334 // CHECK-NEXT: entry:
335 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
336 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
337 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fp.int.predicated.v4i32.v4f32.v4i1(<4 x float> [[A:%.*]], i32 1, <4 x i1> [[TMP1]], <4 x i32> undef)
338 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
340 uint32x4_t
test_vcvtq_x_u32_f32(float32x4_t a
, mve_pred16_t p
)
342 return vcvtq_x_u32_f32(a
, p
);
345 // CHECK-LABEL: @test_vcvttq_f16_f32(
346 // CHECK-NEXT: entry:
347 // CHECK-NEXT: [[TMP0:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.narrow(<8 x half> [[A:%.*]], <4 x float> [[B:%.*]], i32 1)
348 // CHECK-NEXT: ret <8 x half> [[TMP0]]
350 float16x8_t
test_vcvttq_f16_f32(float16x8_t a
, float32x4_t b
)
352 return vcvttq_f16_f32(a
, b
);
355 // CHECK-LABEL: @test_vcvttq_m_f16_f32(
356 // CHECK-NEXT: entry:
357 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
358 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
359 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.narrow.predicated(<8 x half> [[A:%.*]], <4 x float> [[B:%.*]], i32 1, <4 x i1> [[TMP1]])
360 // CHECK-NEXT: ret <8 x half> [[TMP2]]
362 float16x8_t
test_vcvttq_m_f16_f32(float16x8_t a
, float32x4_t b
, mve_pred16_t p
)
364 return vcvttq_m_f16_f32(a
, b
, p
);
367 // CHECK-LABEL: @test_vcvtq_n_f16_s16(
368 // CHECK-NEXT: entry:
369 // CHECK-NEXT: [[TMP0:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fix.v8f16.v8i16(i32 0, <8 x i16> [[A:%.*]], i32 1)
370 // CHECK-NEXT: ret <8 x half> [[TMP0]]
372 float16x8_t
test_vcvtq_n_f16_s16(int16x8_t a
)
375 return vcvtq_n(a
, 1);
377 return vcvtq_n_f16_s16(a
, 1);
381 // CHECK-LABEL: @test_vcvtq_n_f16_u16(
382 // CHECK-NEXT: entry:
383 // CHECK-NEXT: [[TMP0:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fix.v8f16.v8i16(i32 1, <8 x i16> [[A:%.*]], i32 2)
384 // CHECK-NEXT: ret <8 x half> [[TMP0]]
386 float16x8_t
test_vcvtq_n_f16_u16(uint16x8_t a
)
389 return vcvtq_n(a
, 2);
391 return vcvtq_n_f16_u16(a
, 2);
395 // CHECK-LABEL: @test_vcvtq_n_f32_s32(
396 // CHECK-NEXT: entry:
397 // CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fix.v4f32.v4i32(i32 0, <4 x i32> [[A:%.*]], i32 3)
398 // CHECK-NEXT: ret <4 x float> [[TMP0]]
400 float32x4_t
test_vcvtq_n_f32_s32(int32x4_t a
)
403 return vcvtq_n(a
, 3);
405 return vcvtq_n_f32_s32(a
, 3);
409 // CHECK-LABEL: @test_vcvtq_n_f32_u32(
410 // CHECK-NEXT: entry:
411 // CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fix.v4f32.v4i32(i32 1, <4 x i32> [[A:%.*]], i32 32)
412 // CHECK-NEXT: ret <4 x float> [[TMP0]]
414 float32x4_t
test_vcvtq_n_f32_u32(uint32x4_t a
)
417 return vcvtq_n(a
, 32);
419 return vcvtq_n_f32_u32(a
, 32);
423 // CHECK-LABEL: @test_vcvtq_n_s16_f16(
424 // CHECK-NEXT: entry:
425 // CHECK-NEXT: [[TMP0:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fix.v8i16.v8f16(i32 0, <8 x half> [[A:%.*]], i32 1)
426 // CHECK-NEXT: ret <8 x i16> [[TMP0]]
428 int16x8_t
test_vcvtq_n_s16_f16(float16x8_t a
)
430 return vcvtq_n_s16_f16(a
, 1);
433 // CHECK-LABEL: @test_vcvtq_n_u16_f16(
434 // CHECK-NEXT: entry:
435 // CHECK-NEXT: [[TMP0:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fix.v8i16.v8f16(i32 1, <8 x half> [[A:%.*]], i32 2)
436 // CHECK-NEXT: ret <8 x i16> [[TMP0]]
438 uint16x8_t
test_vcvtq_n_u16_f16(float16x8_t a
)
440 return vcvtq_n_u16_f16(a
, 2);
443 // CHECK-LABEL: @test_vcvtq_n_s32_f32(
444 // CHECK-NEXT: entry:
445 // CHECK-NEXT: [[TMP0:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fix.v4i32.v4f32(i32 0, <4 x float> [[A:%.*]], i32 3)
446 // CHECK-NEXT: ret <4 x i32> [[TMP0]]
448 int32x4_t
test_vcvtq_n_s32_f32(float32x4_t a
)
450 return vcvtq_n_s32_f32(a
, 3);
453 // CHECK-LABEL: @test_vcvtq_n_u32_f32(
454 // CHECK-NEXT: entry:
455 // CHECK-NEXT: [[TMP0:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fix.v4i32.v4f32(i32 1, <4 x float> [[A:%.*]], i32 32)
456 // CHECK-NEXT: ret <4 x i32> [[TMP0]]
458 uint32x4_t
test_vcvtq_n_u32_f32(float32x4_t a
)
460 return vcvtq_n_u32_f32(a
, 32);
463 // CHECK-LABEL: @test_vcvtq_m_n_f16_s16(
464 // CHECK-NEXT: entry:
465 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
466 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
467 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fix.predicated.v8f16.v8i16.v8i1(i32 0, <8 x half> [[INACTIVE:%.*]], <8 x i16> [[A:%.*]], i32 1, <8 x i1> [[TMP1]])
468 // CHECK-NEXT: ret <8 x half> [[TMP2]]
470 float16x8_t
test_vcvtq_m_n_f16_s16(float16x8_t inactive
, int16x8_t a
, mve_pred16_t p
)
473 return vcvtq_m_n(inactive
, a
, 1, p
);
475 return vcvtq_m_n_f16_s16(inactive
, a
, 1, p
);
479 // CHECK-LABEL: @test_vcvtq_m_n_f16_u16(
480 // CHECK-NEXT: entry:
481 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
482 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
483 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fix.predicated.v8f16.v8i16.v8i1(i32 1, <8 x half> [[INACTIVE:%.*]], <8 x i16> [[A:%.*]], i32 2, <8 x i1> [[TMP1]])
484 // CHECK-NEXT: ret <8 x half> [[TMP2]]
486 float16x8_t
test_vcvtq_m_n_f16_u16(float16x8_t inactive
, uint16x8_t a
, mve_pred16_t p
)
489 return vcvtq_m_n(inactive
, a
, 2, p
);
491 return vcvtq_m_n_f16_u16(inactive
, a
, 2, p
);
495 // CHECK-LABEL: @test_vcvtq_m_n_f32_s32(
496 // CHECK-NEXT: entry:
497 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
498 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
499 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fix.predicated.v4f32.v4i32.v4i1(i32 0, <4 x float> [[INACTIVE:%.*]], <4 x i32> [[A:%.*]], i32 3, <4 x i1> [[TMP1]])
500 // CHECK-NEXT: ret <4 x float> [[TMP2]]
502 float32x4_t
test_vcvtq_m_n_f32_s32(float32x4_t inactive
, int32x4_t a
, mve_pred16_t p
)
505 return vcvtq_m_n(inactive
, a
, 3, p
);
507 return vcvtq_m_n_f32_s32(inactive
, a
, 3, p
);
511 // CHECK-LABEL: @test_vcvtq_m_n_f32_u32(
512 // CHECK-NEXT: entry:
513 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
514 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
515 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fix.predicated.v4f32.v4i32.v4i1(i32 1, <4 x float> [[INACTIVE:%.*]], <4 x i32> [[A:%.*]], i32 32, <4 x i1> [[TMP1]])
516 // CHECK-NEXT: ret <4 x float> [[TMP2]]
518 float32x4_t
test_vcvtq_m_n_f32_u32(float32x4_t inactive
, uint32x4_t a
, mve_pred16_t p
)
521 return vcvtq_m_n(inactive
, a
, 32, p
);
523 return vcvtq_m_n_f32_u32(inactive
, a
, 32, p
);
527 // CHECK-LABEL: @test_vcvtq_m_n_s16_f16(
528 // CHECK-NEXT: entry:
529 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
530 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
531 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fix.predicated.v8i16.v8f16.v8i1(i32 0, <8 x i16> [[INACTIVE:%.*]], <8 x half> [[A:%.*]], i32 1, <8 x i1> [[TMP1]])
532 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
534 int16x8_t
test_vcvtq_m_n_s16_f16(int16x8_t inactive
, float16x8_t a
, mve_pred16_t p
)
537 return vcvtq_m_n(inactive
, a
, 1, p
);
539 return vcvtq_m_n_s16_f16(inactive
, a
, 1, p
);
543 // CHECK-LABEL: @test_vcvtq_m_n_u16_f16(
544 // CHECK-NEXT: entry:
545 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
546 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
547 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fix.predicated.v8i16.v8f16.v8i1(i32 1, <8 x i16> [[INACTIVE:%.*]], <8 x half> [[A:%.*]], i32 2, <8 x i1> [[TMP1]])
548 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
550 uint16x8_t
test_vcvtq_m_n_u16_f16(uint16x8_t inactive
, float16x8_t a
, mve_pred16_t p
)
553 return vcvtq_m_n(inactive
, a
, 2, p
);
555 return vcvtq_m_n_u16_f16(inactive
, a
, 2, p
);
559 // CHECK-LABEL: @test_vcvtq_m_n_s32_f32(
560 // CHECK-NEXT: entry:
561 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
562 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
563 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fix.predicated.v4i32.v4f32.v4i1(i32 0, <4 x i32> [[INACTIVE:%.*]], <4 x float> [[A:%.*]], i32 3, <4 x i1> [[TMP1]])
564 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
566 int32x4_t
test_vcvtq_m_n_s32_f32(int32x4_t inactive
, float32x4_t a
, mve_pred16_t p
)
569 return vcvtq_m_n(inactive
, a
, 3, p
);
571 return vcvtq_m_n_s32_f32(inactive
, a
, 3, p
);
575 // CHECK-LABEL: @test_vcvtq_m_n_u32_f32(
576 // CHECK-NEXT: entry:
577 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
578 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
579 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fix.predicated.v4i32.v4f32.v4i1(i32 1, <4 x i32> [[INACTIVE:%.*]], <4 x float> [[A:%.*]], i32 32, <4 x i1> [[TMP1]])
580 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
582 uint32x4_t
test_vcvtq_m_n_u32_f32(uint32x4_t inactive
, float32x4_t a
, mve_pred16_t p
)
585 return vcvtq_m_n(inactive
, a
, 32, p
);
587 return vcvtq_m_n_u32_f32(inactive
, a
, 32, p
);
591 // CHECK-LABEL: @test_vcvtq_x_n_f16_s16(
592 // CHECK-NEXT: entry:
593 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
594 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
595 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fix.predicated.v8f16.v8i16.v8i1(i32 0, <8 x half> undef, <8 x i16> [[A:%.*]], i32 1, <8 x i1> [[TMP1]])
596 // CHECK-NEXT: ret <8 x half> [[TMP2]]
598 float16x8_t
test_vcvtq_x_n_f16_s16(int16x8_t a
, mve_pred16_t p
)
601 return vcvtq_x_n(a
, 1, p
);
603 return vcvtq_x_n_f16_s16(a
, 1, p
);
607 // CHECK-LABEL: @test_vcvtq_x_n_f16_u16(
608 // CHECK-NEXT: entry:
609 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
610 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
611 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.vcvt.fix.predicated.v8f16.v8i16.v8i1(i32 1, <8 x half> undef, <8 x i16> [[A:%.*]], i32 2, <8 x i1> [[TMP1]])
612 // CHECK-NEXT: ret <8 x half> [[TMP2]]
614 float16x8_t
test_vcvtq_x_n_f16_u16(uint16x8_t a
, mve_pred16_t p
)
617 return vcvtq_x_n(a
, 2, p
);
619 return vcvtq_x_n_f16_u16(a
, 2, p
);
623 // CHECK-LABEL: @test_vcvtq_x_n_f32_s32(
624 // CHECK-NEXT: entry:
625 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
626 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
627 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fix.predicated.v4f32.v4i32.v4i1(i32 0, <4 x float> undef, <4 x i32> [[A:%.*]], i32 3, <4 x i1> [[TMP1]])
628 // CHECK-NEXT: ret <4 x float> [[TMP2]]
630 float32x4_t
test_vcvtq_x_n_f32_s32(int32x4_t a
, mve_pred16_t p
)
633 return vcvtq_x_n(a
, 3, p
);
635 return vcvtq_x_n_f32_s32(a
, 3, p
);
639 // CHECK-LABEL: @test_vcvtq_x_n_f32_u32(
640 // CHECK-NEXT: entry:
641 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
642 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
643 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.fix.predicated.v4f32.v4i32.v4i1(i32 1, <4 x float> undef, <4 x i32> [[A:%.*]], i32 32, <4 x i1> [[TMP1]])
644 // CHECK-NEXT: ret <4 x float> [[TMP2]]
646 float32x4_t
test_vcvtq_x_n_f32_u32(uint32x4_t a
, mve_pred16_t p
)
649 return vcvtq_x_n(a
, 32, p
);
651 return vcvtq_x_n_f32_u32(a
, 32, p
);
655 // CHECK-LABEL: @test_vcvtq_x_n_s16_f16(
656 // CHECK-NEXT: entry:
657 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
658 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
659 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fix.predicated.v8i16.v8f16.v8i1(i32 0, <8 x i16> undef, <8 x half> [[A:%.*]], i32 1, <8 x i1> [[TMP1]])
660 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
662 int16x8_t
test_vcvtq_x_n_s16_f16(float16x8_t a
, mve_pred16_t p
)
664 return vcvtq_x_n_s16_f16(a
, 1, p
);
667 // CHECK-LABEL: @test_vcvtq_x_n_u16_f16(
668 // CHECK-NEXT: entry:
669 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
670 // CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
671 // CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.arm.mve.vcvt.fix.predicated.v8i16.v8f16.v8i1(i32 1, <8 x i16> undef, <8 x half> [[A:%.*]], i32 2, <8 x i1> [[TMP1]])
672 // CHECK-NEXT: ret <8 x i16> [[TMP2]]
674 uint16x8_t
test_vcvtq_x_n_u16_f16(float16x8_t a
, mve_pred16_t p
)
676 return vcvtq_x_n_u16_f16(a
, 2, p
);
679 // CHECK-LABEL: @test_vcvtq_x_n_s32_f32(
680 // CHECK-NEXT: entry:
681 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
682 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
683 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fix.predicated.v4i32.v4f32.v4i1(i32 0, <4 x i32> undef, <4 x float> [[A:%.*]], i32 3, <4 x i1> [[TMP1]])
684 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
686 int32x4_t
test_vcvtq_x_n_s32_f32(float32x4_t a
, mve_pred16_t p
)
688 return vcvtq_x_n_s32_f32(a
, 3, p
);
691 // CHECK-LABEL: @test_vcvtq_x_n_u32_f32(
692 // CHECK-NEXT: entry:
693 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
694 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
695 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vcvt.fix.predicated.v4i32.v4f32.v4i1(i32 1, <4 x i32> undef, <4 x float> [[A:%.*]], i32 32, <4 x i1> [[TMP1]])
696 // CHECK-NEXT: ret <4 x i32> [[TMP2]]
698 uint32x4_t
test_vcvtq_x_n_u32_f32(float32x4_t a
, mve_pred16_t p
)
700 return vcvtq_x_n_u32_f32(a
, 32, p
);
703 // CHECK-LABEL: @test_vcvtbq_f32_f16(
704 // CHECK-NEXT: entry:
705 // CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.widen(<8 x half> [[A:%.*]], i32 0)
706 // CHECK-NEXT: ret <4 x float> [[TMP0]]
708 float32x4_t
test_vcvtbq_f32_f16(float16x8_t a
)
710 return vcvtbq_f32_f16(a
);
713 // CHECK-LABEL: @test_vcvttq_f32_f16(
714 // CHECK-NEXT: entry:
715 // CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.widen(<8 x half> [[A:%.*]], i32 1)
716 // CHECK-NEXT: ret <4 x float> [[TMP0]]
718 float32x4_t
test_vcvttq_f32_f16(float16x8_t a
)
720 return vcvttq_f32_f16(a
);
723 // CHECK-LABEL: @test_vcvtbq_m_f32_f16(
724 // CHECK-NEXT: entry:
725 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
726 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
727 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.widen.predicated(<4 x float> [[INACTIVE:%.*]], <8 x half> [[A:%.*]], i32 0, <4 x i1> [[TMP1]])
728 // CHECK-NEXT: ret <4 x float> [[TMP2]]
730 float32x4_t
test_vcvtbq_m_f32_f16(float32x4_t inactive
, float16x8_t a
, mve_pred16_t p
)
732 return vcvtbq_m_f32_f16(inactive
, a
, p
);
735 // CHECK-LABEL: @test_vcvttq_m_f32_f16(
736 // CHECK-NEXT: entry:
737 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
738 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
739 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.widen.predicated(<4 x float> [[INACTIVE:%.*]], <8 x half> [[A:%.*]], i32 1, <4 x i1> [[TMP1]])
740 // CHECK-NEXT: ret <4 x float> [[TMP2]]
742 float32x4_t
test_vcvttq_m_f32_f16(float32x4_t inactive
, float16x8_t a
, mve_pred16_t p
)
744 return vcvttq_m_f32_f16(inactive
, a
, p
);
747 // CHECK-LABEL: @test_vcvtbq_x_f32_f16(
748 // CHECK-NEXT: entry:
749 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
750 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
751 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.widen.predicated(<4 x float> undef, <8 x half> [[A:%.*]], i32 0, <4 x i1> [[TMP1]])
752 // CHECK-NEXT: ret <4 x float> [[TMP2]]
754 float32x4_t
test_vcvtbq_x_f32_f16(float16x8_t a
, mve_pred16_t p
)
756 return vcvtbq_x_f32_f16(a
, p
);
759 // CHECK-LABEL: @test_vcvttq_x_f32_f16(
760 // CHECK-NEXT: entry:
761 // CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
762 // CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
763 // CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.vcvt.widen.predicated(<4 x float> undef, <8 x half> [[A:%.*]], i32 1, <4 x i1> [[TMP1]])
764 // CHECK-NEXT: ret <4 x float> [[TMP2]]
766 float32x4_t
test_vcvttq_x_f32_f16(float16x8_t a
, mve_pred16_t p
)
768 return vcvttq_x_f32_f16(a
, p
);