| 3568 | } |
| 3569 | #ifdef DLIB_USE_CUDA |
| 3570 | void test_copy_tensor_gpu() |
| 3571 | { |
| 3572 | using namespace dlib::tt; |
| 3573 | print_spinner(); |
| 3574 | resizable_tensor dest(10, 9, 7, 15); |
| 3575 | resizable_tensor src1(10, 3, 7, 15); |
| 3576 | resizable_tensor src2(10, 3, 7, 15); |
| 3577 | resizable_tensor src3(10, 9, 7, 15); |
| 3578 | tt::tensor_rand rnd; |
| 3579 | rnd.fill_gaussian(dest); |
| 3580 | rnd.fill_gaussian(src1); |
| 3581 | rnd.fill_gaussian(src2); |
| 3582 | rnd.fill_gaussian(src3); |
| 3583 | cuda::copy_tensor(false, dest, 0, src1, 0, src1.k()); //full copy src1->dest |
| 3584 | cuda::copy_tensor(false, dest, src1.k(), src2, 0, src2.k()); //full copy src2->dest with offset of src1 |
| 3585 | cuda::copy_tensor(false, dest, src1.k() + src2.k(), src3, 3, 3); //partial copy src3 into the rest place of dest |
| 3586 | |
| 3587 | |
| 3588 | for (long i = 0; i < dest.num_samples(); ++i) |
| 3589 | { |
| 3590 | for (long k = 0; k < dest.k(); ++k) |
| 3591 | { |
| 3592 | for (long r = 0; r < dest.nr(); ++r) |
| 3593 | { |
| 3594 | for (long c = 0; c < dest.nc(); ++c) |
| 3595 | { |
| 3596 | float dest_value = tensor_read_cpu(dest, i, k, r, c); |
| 3597 | // first part is from src1 |
| 3598 | if (k < src1.k()) |
| 3599 | { |
| 3600 | float src_value = tensor_read_cpu(src1, i, k, r, c); |
| 3601 | DLIB_TEST(src_value == dest_value); |
| 3602 | } |
| 3603 | // second part is from src2 |
| 3604 | else if (k < src1.k() + src2.k()) |
| 3605 | { |
| 3606 | float src_value = tensor_read_cpu(src2, i, k - src1.k(), r, c); |
| 3607 | DLIB_TEST(src_value == dest_value); |
| 3608 | } |
| 3609 | // third part is from src3 |
| 3610 | else |
| 3611 | { |
| 3612 | float src_value = tensor_read_cpu(src3, i, k - src1.k() - src2.k() + 3, r, c); |
| 3613 | DLIB_TEST(src_value == dest_value); |
| 3614 | } |
| 3615 | } |
| 3616 | } |
| 3617 | } |
| 3618 | } |
| 3619 | } |
| 3620 | void test_copy_tensor_add_to_gpu() |
| 3621 | { |
| 3622 | using namespace dlib::tt; |
no test coverage detected