| 1 |
/* { dg-do assemble { target aarch64_asm_sve_ok } } */ |
| 2 |
/* { dg-options "-O -msve-vector-bits=2048 -mbig-endian --save-temps" } */ |
| 3 |
/* { dg-final { check-function-bodies "**" "" } } */ |
| 4 |
|
| 5 |
typedef unsigned char v128qi __attribute__((vector_size(128))); |
| 6 |
typedef unsigned char v64qi __attribute__((vector_size(64))); |
| 7 |
typedef unsigned short v64hi __attribute__((vector_size(128))); |
| 8 |
typedef _Float16 v64hf __attribute__((vector_size(128))); |
| 9 |
typedef __bf16 v64bf __attribute__((vector_size(128))); |
| 10 |
|
| 11 |
#define PERM0(B) B + 1, B |
| 12 |
#define PERM1(B) PERM0 (B), PERM0 (B + 2) |
| 13 |
#define PERM2(B) PERM1 (B), PERM1 (B + 4) |
| 14 |
#define PERM3(B) PERM2 (B), PERM2 (B + 8) |
| 15 |
#define PERM4(B) PERM3 (B), PERM3 (B + 16) |
| 16 |
#define PERM5(B) PERM4 (B), PERM4 (B + 32) |
| 17 |
#define PERM6(B) PERM5 (B), PERM5 (B + 64) |
| 18 |
|
| 19 |
/* |
| 20 |
** qi_revh_s: |
| 21 |
** ptrue (p[0-7])\.b, vl256 |
| 22 |
** ld1b (z[0-9]+)\.h, \1/z, \[x0\] |
| 23 |
** revh (z[0-9]+)\.s, \1/m, \2\.s |
| 24 |
** st1b \3\.h, \1, \[x8\] |
| 25 |
** ret |
| 26 |
*/ |
| 27 |
v128qi |
| 28 |
qi_revh_s (v128qi x) |
| 29 |
{ |
| 30 |
return __builtin_shuffle (x, x, (v128qi) { PERM6 (0) }); |
| 31 |
} |
| 32 |
|
| 33 |
/* |
| 34 |
** qi_revw_d: |
| 35 |
** ptrue (p[0-7])\.b, vl256 |
| 36 |
** ld1b (z[0-9]+)\.s, \1/z, \[x0\] |
| 37 |
** revw (z[0-9]+)\.d, \1/m, \2\.d |
| 38 |
** st1b \3\.s, \1, \[x8\] |
| 39 |
** ret |
| 40 |
*/ |
| 41 |
v64qi |
| 42 |
qi_revw_d (v64qi x) |
| 43 |
{ |
| 44 |
return __builtin_shuffle (x, x, (v64qi) { PERM5 (0) }); |
| 45 |
} |
| 46 |
|
| 47 |
/* |
| 48 |
** hi_revw_d: |
| 49 |
** ptrue (p[0-7])\.b, vl256 |
| 50 |
** ld1h (z[0-9]+)\.s, \1/z, \[x0\] |
| 51 |
** revw (z[0-9]+)\.d, \1/m, \2\.d |
| 52 |
** st1h \3\.s, \1, \[x8\] |
| 53 |
** ret |
| 54 |
*/ |
| 55 |
v64hi |
| 56 |
hi_revw_d (v64hi x) |
| 57 |
{ |
| 58 |
return __builtin_shuffle (x, x, (v64hi) { PERM5 (0) }); |
| 59 |
} |
| 60 |
|
| 61 |
/* |
| 62 |
** hf_revw_d: |
| 63 |
** ptrue (p[0-7])\.b, vl256 |
| 64 |
** ld1h (z[0-9]+)\.s, \1/z, \[x0\] |
| 65 |
** revw (z[0-9]+)\.d, \1/m, \2\.d |
| 66 |
** st1h \3\.s, \1, \[x8\] |
| 67 |
** ret |
| 68 |
*/ |
| 69 |
v64hf |
| 70 |
hf_revw_d (v64hf x) |
| 71 |
{ |
| 72 |
return __builtin_shuffle (x, x, (v64hi) { PERM5 (0) }); |
| 73 |
} |
| 74 |
|
| 75 |
/* |
| 76 |
** bf_revw_d: |
| 77 |
** ptrue (p[0-7])\.b, vl256 |
| 78 |
** ld1h (z[0-9]+)\.s, \1/z, \[x0\] |
| 79 |
** revw (z[0-9]+)\.d, \1/m, \2\.d |
| 80 |
** st1h \3\.s, \1, \[x8\] |
| 81 |
** ret |
| 82 |
*/ |
| 83 |
v64bf |
| 84 |
bf_revw_d (v64bf x) |
| 85 |
{ |
| 86 |
return __builtin_shuffle (x, x, (v64hi) { PERM5 (0) }); |
| 87 |
} |
| 88 |
|
| 89 |
#undef PERM1 |
| 90 |
#define PERM1(B) PERM0 (B + 2), PERM0 (B) |
| 91 |
|
| 92 |
/* |
| 93 |
** qi_revh_d: |
| 94 |
** ptrue (p[0-7])\.b, vl256 |
| 95 |
** ld1b (z[0-9]+)\.h, \1/z, \[x0\] |
| 96 |
** revh (z[0-9]+)\.d, \1/m, \2\.d |
| 97 |
** st1b \3\.h, \1, \[x8\] |
| 98 |
** ret |
| 99 |
*/ |
| 100 |
v128qi |
| 101 |
qi_revh_d (v128qi x) |
| 102 |
{ |
| 103 |
return __builtin_shuffle (x, x, (v128qi) { PERM6 (0) }); |
| 104 |
} |
| 105 |
|
| 106 |
v64qi |
| 107 |
qi_revw_q (v64qi x) |
| 108 |
{ |
| 109 |
return __builtin_shuffle (x, x, (v64qi) { PERM5 (0) }); |
| 110 |
} |
| 111 |
|
| 112 |
v64hi |
| 113 |
hi_revw_q (v64hi x) |
| 114 |
{ |
| 115 |
return __builtin_shuffle (x, x, (v64hi) { PERM5 (0) }); |
| 116 |
} |
| 117 |
|
| 118 |
#undef PERM2 |
| 119 |
#define PERM2(B) PERM0 (B + 4), PERM0 (B) |
| 120 |
|
| 121 |
v128qi |
| 122 |
qi_revh_q (v128qi x) |
| 123 |
{ |
| 124 |
return __builtin_shuffle (x, x, (v128qi) { PERM6 (0) }); |
| 125 |
} |
| 126 |
|
| 127 |
/* { dg-final { scan-assembler-times {\trev.\t} 6 } } */ |