| 1 |
/* { dg-do compile { target c++11 } } */ |
| 2 |
/* { dg-options "-O2 -msse2 -fdump-tree-optimized" } */ |
| 3 |
|
| 4 |
template <class T> |
| 5 |
using V [[gnu::vector_size(16)]] = T; |
| 6 |
|
| 7 |
template <class T, unsigned M = sizeof(V<T>)> |
| 8 |
V<T> load(const void *p) { |
| 9 |
using W = V<T>; |
| 10 |
W r; |
| 11 |
__builtin_memcpy(&r, p, M); |
| 12 |
return r; |
| 13 |
} |
| 14 |
|
| 15 |
// movq or movsd |
| 16 |
template V<char> load<char, 8>(const void *); // bad |
| 17 |
template V<short> load<short, 8>(const void *); // bad |
| 18 |
template V<int> load<int, 8>(const void *); // bad |
| 19 |
template V<long> load<long, 8>(const void *); // good |
| 20 |
// the following is disabled because V2SF isn't a supported mode |
| 21 |
// template V<float> load<float, 8>(const void *); // bad |
| 22 |
template V<double> load<double, 8>(const void *); // good (movsd?) |
| 23 |
|
| 24 |
// movd or movss |
| 25 |
template V<char> load<char, 4>(const void *); // bad |
| 26 |
template V<short> load<short, 4>(const void *); // bad |
| 27 |
template V<int> load<int, 4>(const void *); // good |
| 28 |
template V<float> load<float, 4>(const void *); // good |
| 29 |
|
| 30 |
/* We should end up with one load and one insert for each function. */ |
| 31 |
/* { dg-final { scan-tree-dump-times "BIT_INSERT_EXPR" 9 "optimized" } } */ |
| 32 |
/* { dg-final { scan-tree-dump-times "MEM" 9 "optimized" } } */ |