|
| Resource () |
| default constructor More...
|
|
template<typename xpu , typename DType > |
mshadow::Random< xpu, DType > * | get_random (mshadow::Stream< xpu > *stream) const |
| Get random number generator. More...
|
|
template<typename xpu , typename DType > |
common::random::RandGenerator< xpu, DType > * | get_parallel_random () const |
| Get parallel random number generator. More...
|
|
template<typename xpu , int ndim> |
mshadow::Tensor< xpu, ndim, real_t > | get_space (mshadow::Shape< ndim > shape, mshadow::Stream< xpu > *stream) const |
| Get space requested as mshadow Tensor. The caller can request arbitrary size. More...
|
|
template<int ndim> |
mshadow::Tensor< cpu, ndim, real_t > | get_host_space (mshadow::Shape< ndim > shape) const |
| Get cpu space requested as mshadow Tensor. The caller can request arbitrary size. More...
|
|
template<typename xpu , int ndim, typename DType > |
mshadow::Tensor< xpu, ndim, DType > | get_space_typed (mshadow::Shape< ndim > shape, mshadow::Stream< xpu > *stream) const |
| Get space requested as mshadow Tensor in specified type. The caller can request arbitrary size. More...
|
|
template<int ndim, typename DType > |
mshadow::Tensor< cpu, ndim, DType > | get_host_space_typed (mshadow::Shape< ndim > shape) const |
| Get CPU space as mshadow Tensor in specified type. The caller can request arbitrary size. More...
|
|
void * | get_space_internal (size_t size) const |
| internal function to get space from resources. More...
|
|
void * | get_host_space_internal (size_t size) const |
| internal function to get cpu space from resources. More...
|
|
Resources used by mxnet operations. A resource is something special other than NDArray, but will still participate.
template<typename xpu , int ndim>
mshadow::Tensor<xpu, ndim, real_t> mxnet::Resource::get_space |
( |
mshadow::Shape< ndim > |
shape, |
|
|
mshadow::Stream< xpu > * |
stream |
|
) |
| const |
|
inline |
Get space requested as mshadow Tensor. The caller can request arbitrary size.
This space can be shared with other calls to this->get_space. So the caller need to serialize the calls when using the conflicted space. The old space can get freed, however, this will incur a synchronization, when running on device, so the launched kernels that depend on the temp space can finish correctly.
- Parameters
-
shape | the Shape of returning tensor. |
stream | the stream of retruning tensor. |
- Returns
- the mshadow tensor requested.
- Template Parameters
-
xpu | the device type of random number generator. |
ndim | the number of dimension of the tensor requested. |