1 // RUN: %clang_cc1 -no-opaque-pointers -triple arm64-apple-ios7 -target-feature +neon -target-abi darwinpcs -ffreestanding -emit-llvm -w -o - %s | FileCheck %s --check-prefixes=CHECK,CHECK-LE
2 // RUN: %clang_cc1 -no-opaque-pointers -triple aarch64_be-none-linux-gnu -target-feature +neon -target-abi darwinpcs -ffreestanding -emit-llvm -w -o - %s | FileCheck %s --check-prefixes=CHECK,CHECK-BE
3
4 // REQUIRES: aarch64-registered-target || arm-registered-target
5
6 // CHECK: define{{.*}} signext i8 @f0()
f0(void)7 char f0(void) {
8 return 0;
9 }
10
11 // Struct as return type. Aggregates <= 16 bytes are passed directly. For BE,
12 // return values are round up to 64 bits.
13 //
14 // CHECK-LE: define{{.*}} i8 @f1()
15 // CHECK-BE: define{{.*}} i64 @f1()
16 struct s1 { char f0; };
f1(void)17 struct s1 f1(void) {}
18
19 // CHECK-LE: define{{.*}} i16 @f2()
20 // CHECK-BE: define{{.*}} i64 @f2()
21 struct s2 { short f0; };
f2(void)22 struct s2 f2(void) {}
23
24 // CHECK-LE: define{{.*}} i32 @f3()
25 // CHECK-BE: define{{.*}} i64 @f3()
26 struct s3 { int f0; };
f3(void)27 struct s3 f3(void) {}
28
29 // CHECK-LE: define{{.*}} i32 @f4()
30 // CHECK-BE: define{{.*}} i64 @f4()
31 struct s4 { struct s4_0 { int f0; } f0; };
f4(void)32 struct s4 f4(void) {}
33
34 // CHECK-LE: define{{.*}} i32 @f5()
35 // CHECK-BE: define{{.*}} i64 @f5()
36 struct s5 { struct { } f0; int f1; };
f5(void)37 struct s5 f5(void) {}
38
39 // CHECK-LE: define{{.*}} i32 @f6()
40 // CHECK-BE: define{{.*}} i64 @f6()
41 struct s6 { int f0[1]; };
f6(void)42 struct s6 f6(void) {}
43
44 // CHECK: define{{.*}} void @f7()
45 struct s7 { struct { int : 0; } f0; };
f7(void)46 struct s7 f7(void) {}
47
48 // CHECK: define{{.*}} void @f8()
49 struct s8 { struct { int : 0; } f0[1]; };
f8(void)50 struct s8 f8(void) {}
51
52 // CHECK-LE: define{{.*}} i32 @f9()
53 // CHECK-BE: define{{.*}} i64 @f9()
54 struct s9 { int f0; int : 0; };
f9(void)55 struct s9 f9(void) {}
56
57 // CHECK-LE: define{{.*}} i32 @f10()
58 // CHECK-BE: define{{.*}} i64 @f10()
59 struct s10 { int f0; int : 0; int : 0; };
f10(void)60 struct s10 f10(void) {}
61
62 // CHECK-LE: define{{.*}} i32 @f11()
63 // CHECK-BE: define{{.*}} i64 @f11()
64 struct s11 { int : 0; int f0; };
f11(void)65 struct s11 f11(void) {}
66
67 // CHECK-LE: define{{.*}} i24 @f11_packed()
68 // CHECK-BE: define{{.*}} i64 @f11_packed()
69 struct s11_packed { char c; short s } __attribute__((packed));
f11_packed(void)70 struct s11_packed f11_packed(void) { }
71
72 // CHECK-LE: define{{.*}} i32 @f11_not_packed()
73 // CHECK-BE: define{{.*}} i64 @f11_not_packed()
74 struct s11_not_packed { char c; short s; };
f11_not_packed(void)75 struct s11_not_packed f11_not_packed(void) { }
76
77 // CHECK-LE: define{{.*}} i32 @f12()
78 // CHECK-BE: define{{.*}} i64 @f12()
79 union u12 { char f0; short f1; int f2; };
f12(void)80 union u12 f12(void) {}
81
82 // Homogeneous Aggregate as return type will be passed directly.
83 // CHECK: define{{.*}} %struct.s13 @f13()
84 struct s13 { float f0; };
f13(void)85 struct s13 f13(void) {}
86 // CHECK: define{{.*}} %union.u14 @f14()
87 union u14 { float f0; };
f14(void)88 union u14 f14(void) {}
89
90 // CHECK: define{{.*}} void @f15()
f15(struct s7 a0)91 void f15(struct s7 a0) {}
92
93 // CHECK: define{{.*}} void @f16()
f16(struct s8 a0)94 void f16(struct s8 a0) {}
95
96 // CHECK-LE: define{{.*}} i32 @f17()
97 // CHECK-BE: define{{.*}} i64 @f17()
98 struct s17 { short f0 : 13; char f1 : 4; };
f17(void)99 struct s17 f17(void) {}
100
101 // CHECK-LE: define{{.*}} i32 @f18()
102 // CHECK-BE: define{{.*}} i64 @f18()
103 struct s18 { short f0; char f1 : 4; };
f18(void)104 struct s18 f18(void) {}
105
106 // CHECK-LE: define{{.*}} i32 @f19()
107 // CHECK-BE: define{{.*}} i64 @f19()
108 struct s19 { int f0; struct s8 f1; };
f19(void)109 struct s19 f19(void) {}
110
111 // CHECK-LE: define{{.*}} i32 @f20()
112 // CHECK-BE: define{{.*}} i64 @f20()
113 struct s20 { struct s8 f1; int f0; };
f20(void)114 struct s20 f20(void) {}
115
116 // CHECK-LE: define{{.*}} i32 @f21()
117 // CHECK-BE: define{{.*}} i64 @f21()
118 struct s21 { struct {} f1; int f0 : 4; };
f21(void)119 struct s21 f21(void) {}
120
121 // CHECK-LE: define{{.*}} i16 @f22()
122 // CHECK-LE: define{{.*}} i32 @f23()
123 // CHECK-BE: define{{.*}} i64 @f22()
124 // CHECK-BE: define{{.*}} i64 @f23()
125 // CHECK: define{{.*}} i64 @f24()
126 // CHECK: define{{.*}} [2 x i64] @f25()
127 // CHECK: define{{.*}} { float, float } @f26()
128 // CHECK: define{{.*}} { double, double } @f27()
f22(void)129 _Complex char f22(void) {}
f23(void)130 _Complex short f23(void) {}
f24(void)131 _Complex int f24(void) {}
f25(void)132 _Complex long long f25(void) {}
f26(void)133 _Complex float f26(void) {}
f27(void)134 _Complex double f27(void) {}
135
136 // CHECK-LE: define{{.*}} i16 @f28()
137 // CHECK-BE: define{{.*}} i64 @f28()
138 struct s28 { _Complex char f0; };
f28()139 struct s28 f28() {}
140
141 // CHECK-LE: define{{.*}} i32 @f29()
142 // CHECK-BE: define{{.*}} i64 @f29()
143 struct s29 { _Complex short f0; };
f29()144 struct s29 f29() {}
145
146 // CHECK: define{{.*}} i64 @f30()
147 struct s30 { _Complex int f0; };
f30()148 struct s30 f30() {}
149
150 struct s31 { char x; };
f31(struct s31 s)151 void f31(struct s31 s) { }
152 // CHECK: define{{.*}} void @f31(i64 %s.coerce)
153 // CHECK: %s = alloca %struct.s31, align 1
154 // CHECK-BE: %coerce.highbits = lshr i64 %s.coerce, 56
155 // CHECK-BE: trunc i64 %coerce.highbits to i8
156 // CHECK-LE: trunc i64 %s.coerce to i8
157 // CHECK: store i8 %{{.*}},
158
159 struct s32 { double x; };
f32(struct s32 s)160 void f32(struct s32 s) { }
161 // CHECK: @f32([1 x double] %{{.*}})
162
163 // A composite type larger than 16 bytes should be passed indirectly.
164 struct s33 { char buf[32*32]; };
f33(struct s33 s)165 void f33(struct s33 s) { }
166 // CHECK: define{{.*}} void @f33(%struct.s33* noundef %s)
167
168 struct s34 { char c; };
169 void f34(struct s34 s);
g34(struct s34 * s)170 void g34(struct s34 *s) { f34(*s); }
171 // CHECK: @g34(%struct.s34* noundef %s)
172 // CHECK: %[[a:.*]] = load i8, i8* %{{.*}}
173 // CHECK: zext i8 %[[a]] to i64
174 // CHECK: call void @f34(i64 %{{.*}})
175
176 /*
177 * Check that va_arg accesses stack according to ABI alignment
178 */
t1(int i,...)179 long long t1(int i, ...) {
180 // CHECK: t1
181 __builtin_va_list ap;
182 __builtin_va_start(ap, i);
183 // CHECK-NOT: add i32 %{{.*}} 7
184 // CHECK-NOT: and i32 %{{.*}} -8
185 long long ll = __builtin_va_arg(ap, long long);
186 __builtin_va_end(ap);
187 return ll;
188 }
t2(int i,...)189 double t2(int i, ...) {
190 // CHECK: t2
191 __builtin_va_list ap;
192 __builtin_va_start(ap, i);
193 // CHECK-NOT: add i32 %{{.*}} 7
194 // CHECK-NOT: and i32 %{{.*}} -8
195 double ll = __builtin_va_arg(ap, double);
196 __builtin_va_end(ap);
197 return ll;
198 }
t3(int i,...)199 _Bool t3(int i, ...) {
200 // CHECK: t3
201 __builtin_va_list ap;
202 __builtin_va_start(ap, i);
203 // CHECK: %0 = va_arg {{.*}}* %ap, i8
204 // CHECK-NEXT: store i8 %0, i8* %varet, align 1
205 _Bool b = __builtin_va_arg(ap, _Bool);
206 __builtin_va_end(ap);
207 return b;
208 }
209
210 #include <arm_neon.h>
211
212 // Homogeneous Vector Aggregate as return type and argument type.
213 // CHECK: define{{.*}} %struct.int8x16x2_t @f0_0(<16 x i8> noundef %{{.*}}, <16 x i8> noundef %{{.*}})
f0_0(int8x16_t a0,int8x16_t a1)214 int8x16x2_t f0_0(int8x16_t a0, int8x16_t a1) {
215 return vzipq_s8(a0, a1);
216 }
217
218 // Test direct vector passing.
219 typedef float T_float32x2 __attribute__ ((__vector_size__ (8)));
220 typedef float T_float32x4 __attribute__ ((__vector_size__ (16)));
221 typedef float T_float32x8 __attribute__ ((__vector_size__ (32)));
222 typedef float T_float32x16 __attribute__ ((__vector_size__ (64)));
223
224 // CHECK: define{{.*}} <2 x float> @f1_0(<2 x float> noundef %{{.*}})
f1_0(T_float32x2 a0)225 T_float32x2 f1_0(T_float32x2 a0) { return a0; }
226 // CHECK: define{{.*}} <4 x float> @f1_1(<4 x float> noundef %{{.*}})
f1_1(T_float32x4 a0)227 T_float32x4 f1_1(T_float32x4 a0) { return a0; }
228 // Vector with length bigger than 16-byte is illegal and is passed indirectly.
229 // CHECK: define{{.*}} void @f1_2(<8 x float>* noalias sret(<8 x float>) align 16 %{{.*}}, <8 x float>* noundef %0)
f1_2(T_float32x8 a0)230 T_float32x8 f1_2(T_float32x8 a0) { return a0; }
231 // CHECK: define{{.*}} void @f1_3(<16 x float>* noalias sret(<16 x float>) align 16 %{{.*}}, <16 x float>* noundef %0)
f1_3(T_float32x16 a0)232 T_float32x16 f1_3(T_float32x16 a0) { return a0; }
233
234 // Testing alignment with aggregates: HFA, aggregates with size <= 16 bytes and
235 // aggregates with size > 16 bytes.
236 struct s35
237 {
238 float v[4]; //Testing HFA.
239 } __attribute__((aligned(16)));
240 typedef struct s35 s35_with_align;
241
242 typedef __attribute__((neon_vector_type(4))) float float32x4_t;
f35(int i,s35_with_align s1,s35_with_align s2)243 float32x4_t f35(int i, s35_with_align s1, s35_with_align s2) {
244 // CHECK: define{{.*}} <4 x float> @f35(i32 noundef %i, [4 x float] %s1.coerce, [4 x float] %s2.coerce)
245 // CHECK: %s1 = alloca %struct.s35, align 16
246 // CHECK: %s2 = alloca %struct.s35, align 16
247 // CHECK: %[[a:.*]] = bitcast %struct.s35* %s1 to <4 x float>*
248 // CHECK: load <4 x float>, <4 x float>* %[[a]], align 16
249 // CHECK: %[[b:.*]] = bitcast %struct.s35* %s2 to <4 x float>*
250 // CHECK: load <4 x float>, <4 x float>* %[[b]], align 16
251 float32x4_t v = vaddq_f32(*(float32x4_t *)&s1,
252 *(float32x4_t *)&s2);
253 return v;
254 }
255
256 struct s36
257 {
258 int v[4]; //Testing 16-byte aggregate.
259 } __attribute__((aligned(16)));
260 typedef struct s36 s36_with_align;
261
262 typedef __attribute__((neon_vector_type(4))) int int32x4_t;
f36(int i,s36_with_align s1,s36_with_align s2)263 int32x4_t f36(int i, s36_with_align s1, s36_with_align s2) {
264 // CHECK: define{{.*}} <4 x i32> @f36(i32 noundef %i, i128 %s1.coerce, i128 %s2.coerce)
265 // CHECK: %s1 = alloca %struct.s36, align 16
266 // CHECK: %s2 = alloca %struct.s36, align 16
267 // CHECK: store i128 %s1.coerce, i128* %{{.*}}, align 16
268 // CHECK: store i128 %s2.coerce, i128* %{{.*}}, align 16
269 // CHECK: %[[a:.*]] = bitcast %struct.s36* %s1 to <4 x i32>*
270 // CHECK: load <4 x i32>, <4 x i32>* %[[a]], align 16
271 // CHECK: %[[b:.*]] = bitcast %struct.s36* %s2 to <4 x i32>*
272 // CHECK: load <4 x i32>, <4 x i32>* %[[b]], align 16
273 int32x4_t v = vaddq_s32(*(int32x4_t *)&s1,
274 *(int32x4_t *)&s2);
275 return v;
276 }
277
278 struct s37
279 {
280 int v[18]; //Testing large aggregate.
281 } __attribute__((aligned(16)));
282 typedef struct s37 s37_with_align;
283
f37(int i,s37_with_align s1,s37_with_align s2)284 int32x4_t f37(int i, s37_with_align s1, s37_with_align s2) {
285 // CHECK: define{{.*}} <4 x i32> @f37(i32 noundef %i, %struct.s37* noundef %s1, %struct.s37* noundef %s2)
286 // CHECK: %[[a:.*]] = bitcast %struct.s37* %s1 to <4 x i32>*
287 // CHECK: load <4 x i32>, <4 x i32>* %[[a]], align 16
288 // CHECK: %[[b:.*]] = bitcast %struct.s37* %s2 to <4 x i32>*
289 // CHECK: load <4 x i32>, <4 x i32>* %[[b]], align 16
290 int32x4_t v = vaddq_s32(*(int32x4_t *)&s1,
291 *(int32x4_t *)&s2);
292 return v;
293 }
294 s37_with_align g37;
caller37()295 int32x4_t caller37() {
296 // CHECK: caller37
297 // CHECK: %[[a:.*]] = alloca %struct.s37, align 16
298 // CHECK: %[[b:.*]] = alloca %struct.s37, align 16
299 // CHECK: call void @llvm.memcpy
300 // CHECK: call void @llvm.memcpy
301 // CHECK: call <4 x i32> @f37(i32 noundef 3, %struct.s37* noundef %[[a]], %struct.s37* noundef %[[b]])
302 return f37(3, g37, g37);
303 }
304
305 // rdar://problem/12648441
306 // Test passing structs with size < 8, < 16 and > 16
307 // with alignment of 16 and without
308
309 // structs with size <= 8 bytes, without alignment attribute
310 // passed as i64 regardless of the align attribute
311 struct s38
312 {
313 int i;
314 short s;
315 };
316 typedef struct s38 s38_no_align;
317 // passing structs in registers
318 __attribute__ ((noinline))
f38(int i,s38_no_align s1,s38_no_align s2)319 int f38(int i, s38_no_align s1, s38_no_align s2) {
320 // CHECK: define{{.*}} i32 @f38(i32 noundef %i, i64 %s1.coerce, i64 %s2.coerce)
321 // CHECK: %s1 = alloca %struct.s38, align 4
322 // CHECK: %s2 = alloca %struct.s38, align 4
323 // CHECK: store i64 %s1.coerce, i64* %{{.*}}, align 4
324 // CHECK: store i64 %s2.coerce, i64* %{{.*}}, align 4
325 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s1, i32 0, i32 0
326 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s2, i32 0, i32 0
327 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s1, i32 0, i32 1
328 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s2, i32 0, i32 1
329 return s1.i + s2.i + i + s1.s + s2.s;
330 }
331 s38_no_align g38;
332 s38_no_align g38_2;
caller38()333 int caller38() {
334 // CHECK: define{{.*}} i32 @caller38()
335 // CHECK: %[[a:.*]] = load i64, i64* bitcast (%struct.s38* @g38 to i64*), align 4
336 // CHECK: %[[b:.*]] = load i64, i64* bitcast (%struct.s38* @g38_2 to i64*), align 4
337 // CHECK: call i32 @f38(i32 noundef 3, i64 %[[a]], i64 %[[b]])
338 return f38(3, g38, g38_2);
339 }
340 // passing structs on stack
341 __attribute__ ((noinline))
f38_stack(int i,int i2,int i3,int i4,int i5,int i6,int i7,int i8,int i9,s38_no_align s1,s38_no_align s2)342 int f38_stack(int i, int i2, int i3, int i4, int i5, int i6, int i7, int i8,
343 int i9, s38_no_align s1, s38_no_align s2) {
344 // CHECK: define{{.*}} i32 @f38_stack(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i32 noundef %i8, i32 noundef %i9, i64 %s1.coerce, i64 %s2.coerce)
345 // CHECK: %s1 = alloca %struct.s38, align 4
346 // CHECK: %s2 = alloca %struct.s38, align 4
347 // CHECK: store i64 %s1.coerce, i64* %{{.*}}, align 4
348 // CHECK: store i64 %s2.coerce, i64* %{{.*}}, align 4
349 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s1, i32 0, i32 0
350 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s2, i32 0, i32 0
351 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s1, i32 0, i32 1
352 // CHECK: getelementptr inbounds %struct.s38, %struct.s38* %s2, i32 0, i32 1
353 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + s1.s + s2.s;
354 }
caller38_stack()355 int caller38_stack() {
356 // CHECK: define{{.*}} i32 @caller38_stack()
357 // CHECK: %[[a:.*]] = load i64, i64* bitcast (%struct.s38* @g38 to i64*), align 4
358 // CHECK: %[[b:.*]] = load i64, i64* bitcast (%struct.s38* @g38_2 to i64*), align 4
359 // CHECK: call i32 @f38_stack(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i32 noundef 8, i32 noundef 9, i64 %[[a]], i64 %[[b]])
360 return f38_stack(1, 2, 3, 4, 5, 6, 7, 8, 9, g38, g38_2);
361 }
362
363 // structs with size <= 8 bytes, with alignment attribute
364 struct s39
365 {
366 int i;
367 short s;
368 } __attribute__((aligned(16)));
369 typedef struct s39 s39_with_align;
370 // passing aligned structs in registers
371 __attribute__ ((noinline))
f39(int i,s39_with_align s1,s39_with_align s2)372 int f39(int i, s39_with_align s1, s39_with_align s2) {
373 // CHECK: define{{.*}} i32 @f39(i32 noundef %i, i128 %s1.coerce, i128 %s2.coerce)
374 // CHECK: %s1 = alloca %struct.s39, align 16
375 // CHECK: %s2 = alloca %struct.s39, align 16
376 // CHECK: store i128 %s1.coerce, i128* %{{.*}}, align 16
377 // CHECK: store i128 %s2.coerce, i128* %{{.*}}, align 16
378 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s1, i32 0, i32 0
379 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s2, i32 0, i32 0
380 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s1, i32 0, i32 1
381 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s2, i32 0, i32 1
382 return s1.i + s2.i + i + s1.s + s2.s;
383 }
384 s39_with_align g39;
385 s39_with_align g39_2;
caller39()386 int caller39() {
387 // CHECK: define{{.*}} i32 @caller39()
388 // CHECK: %[[a:.*]] = load i128, i128* bitcast (%struct.s39* @g39 to i128*), align 16
389 // CHECK: %[[b:.*]] = load i128, i128* bitcast (%struct.s39* @g39_2 to i128*), align 16
390 // CHECK: call i32 @f39(i32 noundef 3, i128 %[[a]], i128 %[[b]])
391 return f39(3, g39, g39_2);
392 }
393 // passing aligned structs on stack
394 __attribute__ ((noinline))
f39_stack(int i,int i2,int i3,int i4,int i5,int i6,int i7,int i8,int i9,s39_with_align s1,s39_with_align s2)395 int f39_stack(int i, int i2, int i3, int i4, int i5, int i6, int i7, int i8,
396 int i9, s39_with_align s1, s39_with_align s2) {
397 // CHECK: define{{.*}} i32 @f39_stack(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i32 noundef %i8, i32 noundef %i9, i128 %s1.coerce, i128 %s2.coerce)
398 // CHECK: %s1 = alloca %struct.s39, align 16
399 // CHECK: %s2 = alloca %struct.s39, align 16
400 // CHECK: store i128 %s1.coerce, i128* %{{.*}}, align 16
401 // CHECK: store i128 %s2.coerce, i128* %{{.*}}, align 16
402 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s1, i32 0, i32 0
403 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s2, i32 0, i32 0
404 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s1, i32 0, i32 1
405 // CHECK: getelementptr inbounds %struct.s39, %struct.s39* %s2, i32 0, i32 1
406 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + s1.s + s2.s;
407 }
caller39_stack()408 int caller39_stack() {
409 // CHECK: define{{.*}} i32 @caller39_stack()
410 // CHECK: %[[a:.*]] = load i128, i128* bitcast (%struct.s39* @g39 to i128*), align 16
411 // CHECK: %[[b:.*]] = load i128, i128* bitcast (%struct.s39* @g39_2 to i128*), align 16
412 // CHECK: call i32 @f39_stack(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i32 noundef 8, i32 noundef 9, i128 %[[a]], i128 %[[b]])
413 return f39_stack(1, 2, 3, 4, 5, 6, 7, 8, 9, g39, g39_2);
414 }
415
416 // structs with size <= 16 bytes, without alignment attribute
417 struct s40
418 {
419 int i;
420 short s;
421 int i2;
422 short s2;
423 };
424 typedef struct s40 s40_no_align;
425 // passing structs in registers
426 __attribute__ ((noinline))
f40(int i,s40_no_align s1,s40_no_align s2)427 int f40(int i, s40_no_align s1, s40_no_align s2) {
428 // CHECK: define{{.*}} i32 @f40(i32 noundef %i, [2 x i64] %s1.coerce, [2 x i64] %s2.coerce)
429 // CHECK: %s1 = alloca %struct.s40, align 4
430 // CHECK: %s2 = alloca %struct.s40, align 4
431 // CHECK: store [2 x i64] %s1.coerce, [2 x i64]* %{{.*}}, align 4
432 // CHECK: store [2 x i64] %s2.coerce, [2 x i64]* %{{.*}}, align 4
433 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s1, i32 0, i32 0
434 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s2, i32 0, i32 0
435 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s1, i32 0, i32 1
436 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s2, i32 0, i32 1
437 return s1.i + s2.i + i + s1.s + s2.s;
438 }
439 s40_no_align g40;
440 s40_no_align g40_2;
caller40()441 int caller40() {
442 // CHECK: define{{.*}} i32 @caller40()
443 // CHECK: %[[a:.*]] = load [2 x i64], [2 x i64]* bitcast (%struct.s40* @g40 to [2 x i64]*), align 4
444 // CHECK: %[[b:.*]] = load [2 x i64], [2 x i64]* bitcast (%struct.s40* @g40_2 to [2 x i64]*), align 4
445 // CHECK: call i32 @f40(i32 noundef 3, [2 x i64] %[[a]], [2 x i64] %[[b]])
446 return f40(3, g40, g40_2);
447 }
448 // passing structs on stack
449 __attribute__ ((noinline))
f40_stack(int i,int i2,int i3,int i4,int i5,int i6,int i7,int i8,int i9,s40_no_align s1,s40_no_align s2)450 int f40_stack(int i, int i2, int i3, int i4, int i5, int i6, int i7, int i8,
451 int i9, s40_no_align s1, s40_no_align s2) {
452 // CHECK: define{{.*}} i32 @f40_stack(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i32 noundef %i8, i32 noundef %i9, [2 x i64] %s1.coerce, [2 x i64] %s2.coerce)
453 // CHECK: %s1 = alloca %struct.s40, align 4
454 // CHECK: %s2 = alloca %struct.s40, align 4
455 // CHECK: store [2 x i64] %s1.coerce, [2 x i64]* %{{.*}}, align 4
456 // CHECK: store [2 x i64] %s2.coerce, [2 x i64]* %{{.*}}, align 4
457 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s1, i32 0, i32 0
458 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s2, i32 0, i32 0
459 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s1, i32 0, i32 1
460 // CHECK: getelementptr inbounds %struct.s40, %struct.s40* %s2, i32 0, i32 1
461 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + s1.s + s2.s;
462 }
caller40_stack()463 int caller40_stack() {
464 // CHECK: define{{.*}} i32 @caller40_stack()
465 // CHECK: %[[a:.*]] = load [2 x i64], [2 x i64]* bitcast (%struct.s40* @g40 to [2 x i64]*), align 4
466 // CHECK: %[[b:.*]] = load [2 x i64], [2 x i64]* bitcast (%struct.s40* @g40_2 to [2 x i64]*), align 4
467 // CHECK: call i32 @f40_stack(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i32 noundef 8, i32 noundef 9, [2 x i64] %[[a]], [2 x i64] %[[b]])
468 return f40_stack(1, 2, 3, 4, 5, 6, 7, 8, 9, g40, g40_2);
469 }
470
471 // structs with size <= 16 bytes, with alignment attribute
472 struct s41
473 {
474 int i;
475 short s;
476 int i2;
477 short s2;
478 } __attribute__((aligned(16)));
479 typedef struct s41 s41_with_align;
480 // passing aligned structs in registers
481 __attribute__ ((noinline))
f41(int i,s41_with_align s1,s41_with_align s2)482 int f41(int i, s41_with_align s1, s41_with_align s2) {
483 // CHECK: define{{.*}} i32 @f41(i32 noundef %i, i128 %s1.coerce, i128 %s2.coerce)
484 // CHECK: %s1 = alloca %struct.s41, align 16
485 // CHECK: %s2 = alloca %struct.s41, align 16
486 // CHECK: store i128 %s1.coerce, i128* %{{.*}}, align 16
487 // CHECK: store i128 %s2.coerce, i128* %{{.*}}, align 16
488 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s1, i32 0, i32 0
489 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s2, i32 0, i32 0
490 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s1, i32 0, i32 1
491 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s2, i32 0, i32 1
492 return s1.i + s2.i + i + s1.s + s2.s;
493 }
494 s41_with_align g41;
495 s41_with_align g41_2;
caller41()496 int caller41() {
497 // CHECK: define{{.*}} i32 @caller41()
498 // CHECK: %[[a:.*]] = load i128, i128* bitcast (%struct.s41* @g41 to i128*), align 16
499 // CHECK: %[[b:.*]] = load i128, i128* bitcast (%struct.s41* @g41_2 to i128*), align 16
500 // CHECK: call i32 @f41(i32 noundef 3, i128 %[[a]], i128 %[[b]])
501 return f41(3, g41, g41_2);
502 }
503 // passing aligned structs on stack
504 __attribute__ ((noinline))
f41_stack(int i,int i2,int i3,int i4,int i5,int i6,int i7,int i8,int i9,s41_with_align s1,s41_with_align s2)505 int f41_stack(int i, int i2, int i3, int i4, int i5, int i6, int i7, int i8,
506 int i9, s41_with_align s1, s41_with_align s2) {
507 // CHECK: define{{.*}} i32 @f41_stack(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i32 noundef %i8, i32 noundef %i9, i128 %s1.coerce, i128 %s2.coerce)
508 // CHECK: %s1 = alloca %struct.s41, align 16
509 // CHECK: %s2 = alloca %struct.s41, align 16
510 // CHECK: store i128 %s1.coerce, i128* %{{.*}}, align 16
511 // CHECK: store i128 %s2.coerce, i128* %{{.*}}, align 16
512 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s1, i32 0, i32 0
513 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s2, i32 0, i32 0
514 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s1, i32 0, i32 1
515 // CHECK: getelementptr inbounds %struct.s41, %struct.s41* %s2, i32 0, i32 1
516 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + s1.s + s2.s;
517 }
caller41_stack()518 int caller41_stack() {
519 // CHECK: define{{.*}} i32 @caller41_stack()
520 // CHECK: %[[a:.*]] = load i128, i128* bitcast (%struct.s41* @g41 to i128*), align 16
521 // CHECK: %[[b:.*]] = load i128, i128* bitcast (%struct.s41* @g41_2 to i128*), align 16
522 // CHECK: call i32 @f41_stack(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i32 noundef 8, i32 noundef 9, i128 %[[a]], i128 %[[b]])
523 return f41_stack(1, 2, 3, 4, 5, 6, 7, 8, 9, g41, g41_2);
524 }
525
526 // structs with size > 16 bytes, without alignment attribute
527 struct s42
528 {
529 int i;
530 short s;
531 int i2;
532 short s2;
533 int i3;
534 short s3;
535 };
536 typedef struct s42 s42_no_align;
537 // passing structs in registers
538 __attribute__ ((noinline))
f42(int i,s42_no_align s1,s42_no_align s2)539 int f42(int i, s42_no_align s1, s42_no_align s2) {
540 // CHECK: define{{.*}} i32 @f42(i32 noundef %i, %struct.s42* noundef %s1, %struct.s42* noundef %s2)
541 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s1, i32 0, i32 0
542 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s2, i32 0, i32 0
543 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s1, i32 0, i32 1
544 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s2, i32 0, i32 1
545 return s1.i + s2.i + i + s1.s + s2.s;
546 }
547 s42_no_align g42;
548 s42_no_align g42_2;
caller42()549 int caller42() {
550 // CHECK: define{{.*}} i32 @caller42()
551 // CHECK: %[[a:.*]] = alloca %struct.s42, align 4
552 // CHECK: %[[b:.*]] = alloca %struct.s42, align 4
553 // CHECK: %[[c:.*]] = bitcast %struct.s42* %[[a]] to i8*
554 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
555 // CHECK: %[[d:.*]] = bitcast %struct.s42* %[[b]] to i8*
556 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
557 // CHECK: call i32 @f42(i32 noundef 3, %struct.s42* noundef %[[a]], %struct.s42* noundef %[[b]])
558 return f42(3, g42, g42_2);
559 }
560 // passing structs on stack
561 __attribute__ ((noinline))
f42_stack(int i,int i2,int i3,int i4,int i5,int i6,int i7,int i8,int i9,s42_no_align s1,s42_no_align s2)562 int f42_stack(int i, int i2, int i3, int i4, int i5, int i6, int i7, int i8,
563 int i9, s42_no_align s1, s42_no_align s2) {
564 // CHECK: define{{.*}} i32 @f42_stack(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i32 noundef %i8, i32 noundef %i9, %struct.s42* noundef %s1, %struct.s42* noundef %s2)
565 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s1, i32 0, i32 0
566 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s2, i32 0, i32 0
567 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s1, i32 0, i32 1
568 // CHECK: getelementptr inbounds %struct.s42, %struct.s42* %s2, i32 0, i32 1
569 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + s1.s + s2.s;
570 }
caller42_stack()571 int caller42_stack() {
572 // CHECK: define{{.*}} i32 @caller42_stack()
573 // CHECK: %[[a:.*]] = alloca %struct.s42, align 4
574 // CHECK: %[[b:.*]] = alloca %struct.s42, align 4
575 // CHECK: %[[c:.*]] = bitcast %struct.s42* %[[a]] to i8*
576 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
577 // CHECK: %[[d:.*]] = bitcast %struct.s42* %[[b]] to i8*
578 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
579 // CHECK: call i32 @f42_stack(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i32 noundef 8, i32 noundef 9, %struct.s42* noundef %[[a]], %struct.s42* noundef %[[b]])
580 return f42_stack(1, 2, 3, 4, 5, 6, 7, 8, 9, g42, g42_2);
581 }
582
583 // structs with size > 16 bytes, with alignment attribute
584 struct s43
585 {
586 int i;
587 short s;
588 int i2;
589 short s2;
590 int i3;
591 short s3;
592 } __attribute__((aligned(16)));
593 typedef struct s43 s43_with_align;
594 // passing aligned structs in registers
595 __attribute__ ((noinline))
f43(int i,s43_with_align s1,s43_with_align s2)596 int f43(int i, s43_with_align s1, s43_with_align s2) {
597 // CHECK: define{{.*}} i32 @f43(i32 noundef %i, %struct.s43* noundef %s1, %struct.s43* noundef %s2)
598 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s1, i32 0, i32 0
599 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s2, i32 0, i32 0
600 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s1, i32 0, i32 1
601 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s2, i32 0, i32 1
602 return s1.i + s2.i + i + s1.s + s2.s;
603 }
604 s43_with_align g43;
605 s43_with_align g43_2;
caller43()606 int caller43() {
607 // CHECK: define{{.*}} i32 @caller43()
608 // CHECK: %[[a:.*]] = alloca %struct.s43, align 16
609 // CHECK: %[[b:.*]] = alloca %struct.s43, align 16
610 // CHECK: %[[c:.*]] = bitcast %struct.s43* %[[a]] to i8*
611 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
612 // CHECK: %[[d:.*]] = bitcast %struct.s43* %[[b]] to i8*
613 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
614 // CHECK: call i32 @f43(i32 noundef 3, %struct.s43* noundef %[[a]], %struct.s43* noundef %[[b]])
615 return f43(3, g43, g43_2);
616 }
617 // passing aligned structs on stack
618 __attribute__ ((noinline))
f43_stack(int i,int i2,int i3,int i4,int i5,int i6,int i7,int i8,int i9,s43_with_align s1,s43_with_align s2)619 int f43_stack(int i, int i2, int i3, int i4, int i5, int i6, int i7, int i8,
620 int i9, s43_with_align s1, s43_with_align s2) {
621 // CHECK: define{{.*}} i32 @f43_stack(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i32 noundef %i8, i32 noundef %i9, %struct.s43* noundef %s1, %struct.s43* noundef %s2)
622 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s1, i32 0, i32 0
623 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s2, i32 0, i32 0
624 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s1, i32 0, i32 1
625 // CHECK: getelementptr inbounds %struct.s43, %struct.s43* %s2, i32 0, i32 1
626 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + s1.s + s2.s;
627 }
caller43_stack()628 int caller43_stack() {
629 // CHECK: define{{.*}} i32 @caller43_stack()
630 // CHECK: %[[a:.*]] = alloca %struct.s43, align 16
631 // CHECK: %[[b:.*]] = alloca %struct.s43, align 16
632 // CHECK: %[[c:.*]] = bitcast %struct.s43* %[[a]] to i8*
633 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
634 // CHECK: %[[d:.*]] = bitcast %struct.s43* %[[b]] to i8*
635 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
636 // CHECK: call i32 @f43_stack(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i32 noundef 8, i32 noundef 9, %struct.s43* noundef %[[a]], %struct.s43* noundef %[[b]])
637 return f43_stack(1, 2, 3, 4, 5, 6, 7, 8, 9, g43, g43_2);
638 }
639
640 // rdar://13668927
641 // We should not split argument s1 between registers and stack.
642 __attribute__ ((noinline))
f40_split(int i,int i2,int i3,int i4,int i5,int i6,int i7,s40_no_align s1,s40_no_align s2)643 int f40_split(int i, int i2, int i3, int i4, int i5, int i6, int i7,
644 s40_no_align s1, s40_no_align s2) {
645 // CHECK: define{{.*}} i32 @f40_split(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, [2 x i64] %s1.coerce, [2 x i64] %s2.coerce)
646 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + s1.s + s2.s;
647 }
caller40_split()648 int caller40_split() {
649 // CHECK: define{{.*}} i32 @caller40_split()
650 // CHECK: call i32 @f40_split(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, [2 x i64] %{{.*}} [2 x i64] %{{.*}})
651 return f40_split(1, 2, 3, 4, 5, 6, 7, g40, g40_2);
652 }
653
654 __attribute__ ((noinline))
f41_split(int i,int i2,int i3,int i4,int i5,int i6,int i7,s41_with_align s1,s41_with_align s2)655 int f41_split(int i, int i2, int i3, int i4, int i5, int i6, int i7,
656 s41_with_align s1, s41_with_align s2) {
657 // CHECK: define{{.*}} i32 @f41_split(i32 noundef %i, i32 noundef %i2, i32 noundef %i3, i32 noundef %i4, i32 noundef %i5, i32 noundef %i6, i32 noundef %i7, i128 %s1.coerce, i128 %s2.coerce)
658 return s1.i + s2.i + i + i2 + i3 + i4 + i5 + i6 + i7 + s1.s + s2.s;
659 }
caller41_split()660 int caller41_split() {
661 // CHECK: define{{.*}} i32 @caller41_split()
662 // CHECK: call i32 @f41_split(i32 noundef 1, i32 noundef 2, i32 noundef 3, i32 noundef 4, i32 noundef 5, i32 noundef 6, i32 noundef 7, i128 %{{.*}}, i128 %{{.*}})
663 return f41_split(1, 2, 3, 4, 5, 6, 7, g41, g41_2);
664 }
665
666 // Handle homogeneous aggregates properly in variadic functions.
667 struct HFA {
668 float a, b, c, d;
669 };
670
test_hfa(int n,...)671 float test_hfa(int n, ...) {
672 // CHECK-LE-LABEL: define{{.*}} float @test_hfa(i32 noundef %n, ...)
673 // CHECK-LE: [[THELIST:%.*]] = alloca i8*
674 // CHECK-LE: [[CURLIST:%.*]] = load i8*, i8** [[THELIST]]
675
676 // HFA is not indirect, so occupies its full 16 bytes on the stack.
677 // CHECK-LE: [[NEXTLIST:%.*]] = getelementptr inbounds i8, i8* [[CURLIST]], i64 16
678 // CHECK-LE: store i8* [[NEXTLIST]], i8** [[THELIST]]
679
680 // CHECK-LE: bitcast i8* [[CURLIST]] to %struct.HFA*
681 __builtin_va_list thelist;
682 __builtin_va_start(thelist, n);
683 struct HFA h = __builtin_va_arg(thelist, struct HFA);
684 return h.d;
685 }
686
test_hfa_call(struct HFA * a)687 float test_hfa_call(struct HFA *a) {
688 // CHECK-LABEL: define{{.*}} float @test_hfa_call(%struct.HFA* noundef %a)
689 // CHECK: call float (i32, ...) @test_hfa(i32 noundef 1, [4 x float] {{.*}})
690 test_hfa(1, *a);
691 }
692
693 struct TooBigHFA {
694 float a, b, c, d, e;
695 };
696
test_toobig_hfa(int n,...)697 float test_toobig_hfa(int n, ...) {
698 // CHECK-LE-LABEL: define{{.*}} float @test_toobig_hfa(i32 noundef %n, ...)
699 // CHECK-LE: [[THELIST:%.*]] = alloca i8*
700 // CHECK-LE: [[CURLIST:%.*]] = load i8*, i8** [[THELIST]]
701
702 // TooBigHFA is not actually an HFA, so gets passed indirectly. Only 8 bytes
703 // of stack consumed.
704 // CHECK-LE: [[NEXTLIST:%.*]] = getelementptr inbounds i8, i8* [[CURLIST]], i64 8
705 // CHECK-LE: store i8* [[NEXTLIST]], i8** [[THELIST]]
706
707 // CHECK-LE: [[HFAPTRPTR:%.*]] = bitcast i8* [[CURLIST]] to %struct.TooBigHFA**
708 // CHECK-LE: [[HFAPTR:%.*]] = load %struct.TooBigHFA*, %struct.TooBigHFA** [[HFAPTRPTR]]
709 __builtin_va_list thelist;
710 __builtin_va_start(thelist, n);
711 struct TooBigHFA h = __builtin_va_arg(thelist, struct TooBigHFA);
712 return h.d;
713 }
714
715 struct HVA {
716 int32x4_t a, b;
717 };
718
test_hva(int n,...)719 int32x4_t test_hva(int n, ...) {
720 // CHECK-LE-LABEL: define{{.*}} <4 x i32> @test_hva(i32 noundef %n, ...)
721 // CHECK-LE: [[THELIST:%.*]] = alloca i8*
722 // CHECK-LE: [[CURLIST:%.*]] = load i8*, i8** [[THELIST]]
723
724 // HVA is not indirect, so occupies its full 16 bytes on the stack. but it
725 // must be properly aligned.
726 // CHECK-LE: [[ALIGN0:%.*]] = ptrtoint i8* [[CURLIST]] to i64
727 // CHECK-LE: [[ALIGN1:%.*]] = add i64 [[ALIGN0]], 15
728 // CHECK-LE: [[ALIGN2:%.*]] = and i64 [[ALIGN1]], -16
729 // CHECK-LE: [[ALIGNED_LIST:%.*]] = inttoptr i64 [[ALIGN2]] to i8*
730
731 // CHECK-LE: [[NEXTLIST:%.*]] = getelementptr inbounds i8, i8* [[ALIGNED_LIST]], i64 32
732 // CHECK-LE: store i8* [[NEXTLIST]], i8** [[THELIST]]
733
734 // CHECK-LE: bitcast i8* [[ALIGNED_LIST]] to %struct.HVA*
735 __builtin_va_list thelist;
736 __builtin_va_start(thelist, n);
737 struct HVA h = __builtin_va_arg(thelist, struct HVA);
738 return h.b;
739 }
740
741 struct TooBigHVA {
742 int32x4_t a, b, c, d, e;
743 };
744
test_toobig_hva(int n,...)745 int32x4_t test_toobig_hva(int n, ...) {
746 // CHECK-LE-LABEL: define{{.*}} <4 x i32> @test_toobig_hva(i32 noundef %n, ...)
747 // CHECK-LE: [[THELIST:%.*]] = alloca i8*
748 // CHECK-LE: [[CURLIST:%.*]] = load i8*, i8** [[THELIST]]
749
750 // TooBigHVA is not actually an HVA, so gets passed indirectly. Only 8 bytes
751 // of stack consumed.
752 // CHECK-LE: [[NEXTLIST:%.*]] = getelementptr inbounds i8, i8* [[CURLIST]], i64 8
753 // CHECK-LE: store i8* [[NEXTLIST]], i8** [[THELIST]]
754
755 // CHECK-LE: [[HVAPTRPTR:%.*]] = bitcast i8* [[CURLIST]] to %struct.TooBigHVA**
756 // CHECK-LE: [[HVAPTR:%.*]] = load %struct.TooBigHVA*, %struct.TooBigHVA** [[HVAPTRPTR]]
757 __builtin_va_list thelist;
758 __builtin_va_start(thelist, n);
759 struct TooBigHVA h = __builtin_va_arg(thelist, struct TooBigHVA);
760 return h.d;
761 }
762
763 typedef __attribute__((__ext_vector_type__(3))) float float32x3_t;
764 typedef struct { float32x3_t arr[4]; } HFAv3;
765
test_hva_v3(int n,...)766 float32x3_t test_hva_v3(int n, ...) {
767 // CHECK-LE-LABEL: define{{.*}} <3 x float> @test_hva_v3(i32 noundef %n, ...)
768 // CHECK-LE: [[THELIST:%.*]] = alloca i8*
769 // CHECK-LE: [[CURLIST:%.*]] = load i8*, i8** [[THELIST]]
770
771 // HVA is not indirect, so occupies its full 16 bytes on the stack. but it
772 // must be properly aligned.
773 // CHECK-LE: [[ALIGN0:%.*]] = ptrtoint i8* [[CURLIST]] to i64
774 // CHECK-LE: [[ALIGN1:%.*]] = add i64 [[ALIGN0]], 15
775 // CHECK-LE: [[ALIGN2:%.*]] = and i64 [[ALIGN1]], -16
776 // CHECK-LE: [[ALIGNED_LIST:%.*]] = inttoptr i64 [[ALIGN2]] to i8*
777
778 // CHECK-LE: [[NEXTLIST:%.*]] = getelementptr inbounds i8, i8* [[ALIGNED_LIST]], i64 64
779 // CHECK-LE: store i8* [[NEXTLIST]], i8** [[THELIST]]
780
781 // CHECK-LE: bitcast i8* [[ALIGNED_LIST]] to %struct.HFAv3*
782 __builtin_va_list l;
783 __builtin_va_start(l, n);
784 HFAv3 r = __builtin_va_arg(l, HFAv3);
785 return r.arr[2];
786 }
787
test_hva_v3_call(HFAv3 * a)788 float32x3_t test_hva_v3_call(HFAv3 *a) {
789 // CHECK-LABEL: define{{.*}} <3 x float> @test_hva_v3_call(%struct.HFAv3* noundef %a)
790 // CHECK: call <3 x float> (i32, ...) @test_hva_v3(i32 noundef 1, [4 x <4 x float>] {{.*}})
791 return test_hva_v3(1, *a);
792 }
793