| 1 |
/* PR target/99195. */ |
| 2 |
/* Check that we take advantage of 64-bit Advanced SIMD operations clearing |
| 3 |
the top half of the vector register and no explicit zeroing instructions |
| 4 |
are emitted. */ |
| 5 |
/* { dg-do compile } */ |
| 6 |
/* { dg-options "-O -march=armv8.2-a+fp16" } */ |
| 7 |
|
| 8 |
#include <arm_neon.h> |
| 9 |
|
| 10 |
#define BINARY(OT,IT,OP,S) \ |
| 11 |
OT \ |
| 12 |
foo_##OP##_##S (IT a, IT b) \ |
| 13 |
{ \ |
| 14 |
IT zeros = vcreate_##S (0); \ |
| 15 |
return vcombine_##S (v##OP##_##S (a, b), zeros); \ |
| 16 |
} |
| 17 |
|
| 18 |
#define FUNC(T,IS,OS,OP,S) BINARY (T##x##OS##_t, T##x##IS##_t, OP, S) |
| 19 |
|
| 20 |
#define OPTWO(T,IS,OS,S,OP1,OP2) \ |
| 21 |
FUNC (T, IS, OS, OP1, S) \ |
| 22 |
FUNC (T, IS, OS, OP2, S) |
| 23 |
|
| 24 |
#define OPTHREE(T, IS, OS, S, OP1, OP2, OP3) \ |
| 25 |
FUNC (T, IS, OS, OP1, S) \ |
| 26 |
OPTWO (T, IS, OS, S, OP2, OP3) |
| 27 |
|
| 28 |
#define OPFOUR(T,IS,OS,S,OP1,OP2,OP3,OP4) \ |
| 29 |
FUNC (T, IS, OS, OP1, S) \ |
| 30 |
OPTHREE (T, IS, OS, S, OP2, OP3, OP4) |
| 31 |
|
| 32 |
#define OPFIVE(T,IS,OS,S,OP1,OP2,OP3,OP4, OP5) \ |
| 33 |
FUNC (T, IS, OS, OP1, S) \ |
| 34 |
OPFOUR (T, IS, OS, S, OP2, OP3, OP4, OP5) |
| 35 |
|
| 36 |
#define OPSIX(T,IS,OS,S,OP1,OP2,OP3,OP4,OP5,OP6) \ |
| 37 |
FUNC (T, IS, OS, OP1, S) \ |
| 38 |
OPFIVE (T, IS, OS, S, OP2, OP3, OP4, OP5, OP6) |
| 39 |
|
| 40 |
#define OPSEVEN(T,IS,OS,S,OP1,OP2,OP3,OP4,OP5,OP6,OP7) \ |
| 41 |
FUNC (T, IS, OS, OP1, S) \ |
| 42 |
OPSIX (T, IS, OS, S, OP2, OP3, OP4, OP5, OP6, OP7) |
| 43 |
|
| 44 |
#define OPEIGHT(T,IS,OS,S,OP1,OP2,OP3,OP4,OP5,OP6,OP7,OP8) \ |
| 45 |
OPTHREE (T, IS, OS, S, OP1, OP2, OP3) \ |
| 46 |
OPFIVE (T, IS, OS, S, OP4, OP5, OP6, OP7, OP8) |
| 47 |
|
| 48 |
#define OPTEN(T,IS,OS,S,OP1,OP2,OP3,OP4,OP5,OP6,OP7,OP8,OP9,OP10) \ |
| 49 |
OPFIVE (T, IS, OS, S, OP1, OP2, OP3, OP4, OP5) \ |
| 50 |
OPFIVE (T, IS, OS, S, OP6, OP7, OP8, OP9, OP10) |
| 51 |
|
| 52 |
#define OPELEVEN(T,IS,OS,S,OP1,OP2,OP3,OP4,OP5,OP6,OP7,OP8,OP9,OP10,OP11) \ |
| 53 |
OPFIVE (T, IS, OS, S, OP1, OP2, OP3, OP4, OP5) \ |
| 54 |
OPSIX (T, IS, OS, S, OP6, OP7, OP8, OP9, OP10, OP11) |
| 55 |
|
| 56 |
#define OPTHIRTEEN(T,IS,OS,S,OP1,OP2,OP3,OP4,OP5,OP6,OP7,OP8,OP9,OP10,OP11,OP12,OP13) \ |
| 57 |
OPSIX (T, IS, OS, S, OP1, OP2, OP3, OP4, OP5, OP6) \ |
| 58 |
OPSEVEN (T, IS, OS, S, OP7, OP8, OP9, OP10, OP11, OP12, OP13) |
| 59 |
|
| 60 |
OPTHIRTEEN (float16, 4, 8, f16, add, sub, mul, div, max, maxnm, min, minnm, abd, pmax, pmin, pmaxnm, pminnm) |
| 61 |
|
| 62 |
#define UNARY(OT,IT,OP,S) \ |
| 63 |
OT \ |
| 64 |
foo_##IT##OP##_##S (IT a) \ |
| 65 |
{ \ |
| 66 |
IT zeros = vcreate_##S (0); \ |
| 67 |
return vcombine_##S ((IT) v##OP##_##S (a), zeros); \ |
| 68 |
} |
| 69 |
|
| 70 |
#undef FUNC |
| 71 |
#define FUNC(T,IS,OS,OP,S) UNARY (T##x##OS##_t, T##x##IS##_t, OP, S) |
| 72 |
|
| 73 |
OPTEN (float16, 4, 8, f16, neg, abs, sqrt, rnd, rndi, rndm, rnda, rndn, rndp, rndx) |
| 74 |
/* { dg-final { scan-assembler-not {\tfmov\t} } } */ |
| 75 |
/* { dg-final { scan-assembler-not {\tmov\t} } } */ |
| 76 |
|