1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
|
#include "../nms.h"
#include <ATen/autocast_mode.h>
#include <torch/library.h>
#include <torch/types.h>
namespace vision {
namespace ops {
namespace {
template <c10::DispatchKey autocast_key, c10::DeviceType device_type>
at::Tensor nms_autocast(
const at::Tensor& dets,
const at::Tensor& scores,
double iou_threshold) {
c10::impl::ExcludeDispatchKeyGuard no_autocast(autocast_key);
return nms(
at::autocast::cached_cast(at::kFloat, dets, device_type),
at::autocast::cached_cast(at::kFloat, scores, device_type),
iou_threshold);
}
} // namespace
TORCH_LIBRARY_IMPL(torchvision, Autocast, m) {
m.impl(
TORCH_SELECTIVE_NAME("torchvision::nms"),
TORCH_FN(
(nms_autocast<c10::DispatchKey::Autocast, c10::DeviceType::CUDA>)));
}
TORCH_LIBRARY_IMPL(torchvision, AutocastCPU, m) {
m.impl(
TORCH_SELECTIVE_NAME("torchvision::nms"),
TORCH_FN(
(nms_autocast<c10::DispatchKey::AutocastCPU, c10::DeviceType::CPU>)));
}
TORCH_LIBRARY_IMPL(torchvision, AutocastXPU, m) {
m.impl(
TORCH_SELECTIVE_NAME("torchvision::nms"),
TORCH_FN(
(nms_autocast<c10::DispatchKey::AutocastXPU, c10::DeviceType::XPU>)));
}
} // namespace ops
} // namespace vision
|
