| // We need separate THC_STATE from LIBRARY_STATE because this is called for copy functions. |
| // Copy functions have a THCState if either the source or destination are GPU types, |
| // but expand functions don't have this property. So, if we have a source CPU type |
| // and destination GPU type, we will have a THCState but shouldn't pass it to the expand |
| // function. |
| |
| #define IMPLEMENT_NEWFOREXPAND(TYPEC, THC_STATE_NOARGS) \ |
| template<> \ |
| TH##TYPEC##Tensor *newForExpand(LIBRARY_STATE_TYPE_NOARGS) { \ |
| return TH##TYPEC##Tensor_new(THC_STATE_NOARGS); \ |
| } |
| |
| IMPLEMENT_NEWFOREXPAND(Byte,) |
| IMPLEMENT_NEWFOREXPAND(Char,) |
| IMPLEMENT_NEWFOREXPAND(Short,) |
| IMPLEMENT_NEWFOREXPAND(Int,) |
| IMPLEMENT_NEWFOREXPAND(Long,) |
| IMPLEMENT_NEWFOREXPAND(Float,) |
| IMPLEMENT_NEWFOREXPAND(Double,) |
| IMPLEMENT_NEWFOREXPAND(Half,) |
| |
| #ifdef CUDA_EXPAND |
| IMPLEMENT_NEWFOREXPAND(CudaByte, LIBRARY_STATE_NOARGS) |
| IMPLEMENT_NEWFOREXPAND(CudaChar, LIBRARY_STATE_NOARGS) |
| IMPLEMENT_NEWFOREXPAND(CudaShort, LIBRARY_STATE_NOARGS) |
| IMPLEMENT_NEWFOREXPAND(CudaInt, LIBRARY_STATE_NOARGS) |
| IMPLEMENT_NEWFOREXPAND(CudaLong, LIBRARY_STATE_NOARGS) |
| IMPLEMENT_NEWFOREXPAND(Cuda, LIBRARY_STATE_NOARGS) |
| IMPLEMENT_NEWFOREXPAND(CudaDouble, LIBRARY_STATE_NOARGS) |
| #ifdef CUDA_HALF_TENSOR |
| IMPLEMENT_NEWFOREXPAND(CudaHalf, LIBRARY_STATE_NOARGS) |
| #endif |
| #endif |
| |
| #undef IMPLEMENT_NEWFOREXPAND |
| |
| #define IMPLEMENT_EXPAND(TYPEC, THC_STATE) \ |
| template<> \ |
| void expand(LIBRARY_STATE_TYPE TH##TYPEC##Tensor *r, TH##TYPEC##Tensor *tensor, \ |
| THLongStorage *sizes) { \ |
| TH##TYPEC##Tensor_expand(THC_STATE r, tensor, sizes); \ |
| } |
| |
| IMPLEMENT_EXPAND(Byte,) |
| IMPLEMENT_EXPAND(Char,) |
| IMPLEMENT_EXPAND(Short,) |
| IMPLEMENT_EXPAND(Int,) |
| IMPLEMENT_EXPAND(Long,) |
| IMPLEMENT_EXPAND(Float,) |
| IMPLEMENT_EXPAND(Double,) |
| IMPLEMENT_EXPAND(Half,) |
| |
| #ifdef CUDA_EXPAND |
| IMPLEMENT_EXPAND(CudaByte, LIBRARY_STATE) |
| IMPLEMENT_EXPAND(CudaChar, LIBRARY_STATE) |
| IMPLEMENT_EXPAND(CudaShort, LIBRARY_STATE) |
| IMPLEMENT_EXPAND(CudaInt, LIBRARY_STATE) |
| IMPLEMENT_EXPAND(CudaLong, LIBRARY_STATE) |
| IMPLEMENT_EXPAND(Cuda, LIBRARY_STATE) |
| IMPLEMENT_EXPAND(CudaDouble, LIBRARY_STATE) |
| #ifdef CUDA_HALF_TENSOR |
| IMPLEMENT_EXPAND(CudaHalf, LIBRARY_STATE) |
| #endif |
| #endif |
| |
| #undef IMPLEMENT_EXPAND |
