MCPcopy Create free account
hub / github.com/davisking/dlib / test_copy_tensor_slice_gpu

Function test_copy_tensor_slice_gpu

dlib/test/dnn.cpp:3674–3740  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

3672 }
3673 }
3674 void test_copy_tensor_slice_gpu()
3675 {
3676 using namespace dlib::tt;
3677 print_spinner();
3678 resizable_tensor dest(10, 9, 7, 15);
3679 resizable_tensor src1(10, 3, 7, 15);
3680 resizable_tensor src2(10, 3, 6, 12);
3681 resizable_tensor src3(10, 9, 7, 15);
3682 tt::tensor_rand rnd;
3683 rnd.fill_gaussian(dest);
3684 rnd.fill_gaussian(src1);
3685 rnd.fill_gaussian(src2);
3686 rnd.fill_gaussian(src3);
3687
3688 const resizable_tensor old_dest = dest;
3689
3690 cuda::copy_tensor(false, dest, 0, 0, 0, src1, 0, 0, 0, src1.k(), src1.nr(), src1.nc()); //full copy src1->dest
3691 cuda::copy_tensor(false, dest, src1.k(), 0, 0, src2, 0, 0, 0, src2.k(), src2.nr(), src2.nc()); //full copy src2->dest with offset of src1
3692 cuda::copy_tensor(false, dest, src1.k() + src2.k(), 1, 1, src3, 3, 1, 1, 3, src3.nr()-2, src3.nc()-2); //partial copy src3 into the rest place of dest
3693
3694
3695 for (long i = 0; i < dest.num_samples(); ++i)
3696 {
3697 for (long k = 0; k < dest.k(); ++k)
3698 {
3699 for (long r = 0; r < dest.nr(); ++r)
3700 {
3701 for (long c = 0; c < dest.nc(); ++c)
3702 {
3703 float old_dest_value = tensor_read_cpu(old_dest, i, k, r, c);
3704 float dest_value = tensor_read_cpu(dest, i, k, r, c);
3705 // first part is from src1
3706 if (k < src1.k())
3707 {
3708 float src_value = tensor_read_cpu(src1, i, k, r, c);
3709 DLIB_TEST(src_value == dest_value);
3710 }
3711 // second part is from src2
3712 else if (k < src1.k() + src2.k())
3713 {
3714 if (r < src2.nr() && c < src2.nc())
3715 {
3716 float src_value = tensor_read_cpu(src2, i, k - src1.k(), r, c);
3717 DLIB_TEST(src_value == dest_value);
3718 }
3719 else
3720 {
3721 DLIB_TEST(old_dest_value == dest_value);
3722 }
3723 }
3724 // third part is from src3
3725 else
3726 {
3727 if (r > 0 && c > 0 && r + 1 < src3.nr() && c + 1 < src3.nc())
3728 {
3729 float src_value = tensor_read_cpu(src3, i, k - src1.k() - src2.k() + 3, r, c);
3730 DLIB_TEST(src_value == dest_value);
3731 }

Callers

nothing calls this directly

Calls 7

print_spinnerFunction · 0.85
tensor_read_cpuFunction · 0.85
copy_tensorFunction · 0.50
kMethod · 0.45
nrMethod · 0.45
ncMethod · 0.45
num_samplesMethod · 0.45

Tested by

no test coverage detected