Debian Codegraph @

gcc/testsuite/gcc.target/aarch64/bfloat16_simd_3.c

raw
1 /* { dg-do assemble { target { aarch64*-*-* } } } */
2 /* { dg-require-effective-target arm_v8_2a_bf16_neon_ok } */
3 /* { dg-additional-options "-march=armv8.2-a -O3 --save-temps" } */
4 /* { dg-final { check-function-bodies "**" "" } } */
5
6 #include <arm_neon.h>
7
8 /*
9 **stacktest1:
10 ** sub sp, sp, #16
11 ** str h0, \[sp, 14\]
12 ** ldr h0, \[sp, 14\]
13 ** add sp, sp, 16
14 ** ret
15 */
16 bfloat16_t stacktest1 (bfloat16_t __a)
17 {
18 volatile bfloat16_t b = __a;
19 return b;
20 }
21
22 /*
23 **stacktest2:
24 ** sub sp, sp, #16
25 ** str d0, \[sp, 8\]
26 ** ldr d0, \[sp, 8\]
27 ** add sp, sp, 16
28 ** ret
29 */
30 bfloat16x4_t stacktest2 (bfloat16x4_t __a)
31 {
32 volatile bfloat16x4_t b = __a;
33 return b;
34 }
35
36 /*
37 **stacktest3:
38 ** sub sp, sp, #16
39 ** str q0, \[sp\]
40 ** ldr q0, \[sp\]
41 ** add sp, sp, 16
42 ** ret
43 */
44 bfloat16x8_t stacktest3 (bfloat16x8_t __a)
45 {
46 volatile bfloat16x8_t b = __a;
47 return b;
48 }
49
50 /* Test compilation of __attribute__ vectors of 8, 16, 32, etc. BFloats. */
51 typedef bfloat16_t v8bf __attribute__((vector_size(16)));
52 typedef bfloat16_t v16bf __attribute__((vector_size(32)));
53 typedef bfloat16_t v32bf __attribute__((vector_size(64)));
54 typedef bfloat16_t v64bf __attribute__((vector_size(128)));
55 typedef bfloat16_t v128bf __attribute__((vector_size(256)));
56
57 v8bf stacktest4 (v8bf __a)
58 {
59 volatile v8bf b = __a;
60 return b;
61 }
62
63 v16bf stacktest5 (v16bf __a)
64 {
65 volatile v16bf b = __a;
66 return b;
67 }
68
69 v32bf stacktest6 (v32bf __a)
70 {
71 volatile v32bf b = __a;
72 return b;
73 }
74
75 v64bf stacktest7 (v64bf __a)
76 {
77 volatile v64bf b = __a;
78 return b;
79 }
80
81 v128bf stacktest8 (v128bf __a)
82 {
83 volatile v128bf b = __a;
84 return b;
85 }
86
87 /* Test use of constant values to assign values to vectors. */
88
89 typedef bfloat16_t v2bf __attribute__((vector_size(4)));
90 v2bf c2 (void) { return (v2bf) 0x12345678; }
91
92 bfloat16x4_t c3 (void) { return (bfloat16x4_t) 0x1234567812345678; }