diff options
author | Christian Sarofeen <csarofeen@nvidia.com> | 2017-04-18 20:37:25 +0300 |
---|---|---|
committer | soumith <soumith@fb.com> | 2017-04-18 21:07:58 +0300 |
commit | ea15d0d649edc52fc3e25a26a14c9bcc0070339a (patch) | |
tree | 2610b52ca85e81b78a64e95c4c4e5ae29364c0b9 | |
parent | 956008538a0a1d165be64b063292dc7600f35e6a (diff) |
Fused RNN kernel remove explicit instantiation, isn't needed.
-rw-r--r-- | lib/THCUNN/generic/FusedRNNKernel.cu | 62 |
1 files changed, 2 insertions, 60 deletions
diff --git a/lib/THCUNN/generic/FusedRNNKernel.cu b/lib/THCUNN/generic/FusedRNNKernel.cu index 63c4ff9..17a6563 100644 --- a/lib/THCUNN/generic/FusedRNNKernel.cu +++ b/lib/THCUNN/generic/FusedRNNKernel.cu @@ -439,66 +439,8 @@ __global__ void } } -// *********** START Generate specializations *************** // -#define EXPAND_FUNCTION(ITYPE, DIM) \ - template __global__ void THNN_(GRUForward)<DATATYPE, ITYPE, DIM> \ - (TensorInfo<DATATYPE, ITYPE> inputI, \ - TensorInfo<DATATYPE, ITYPE> hiddenI, \ - TensorInfo<DATATYPE, ITYPE> bias1I, \ - TensorInfo<DATATYPE, ITYPE> bias2I, \ - TensorInfo<DATATYPE, ITYPE> hxI, \ - TensorInfo<DATATYPE, ITYPE> hyI, \ - ITYPE hsz, \ - ITYPE totalElements); \ - \ - template void __global__ THNN_(GRUBackward)<DATATYPE, ITYPE, DIM> \ - (TensorInfo<DATATYPE, ITYPE> inputI, \ - TensorInfo<DATATYPE, ITYPE> hiddenI, \ - TensorInfo<DATATYPE, ITYPE> gradoutputI, \ - TensorInfo<DATATYPE, ITYPE> gradinputI, \ - ITYPE hsz, \ - ITYPE totalElements); \ - \ - template void __global__ THNN_(LSTMForward)<DATATYPE, ITYPE, DIM> \ - (TensorInfo<DATATYPE, ITYPE> inputI, \ - TensorInfo<DATATYPE, ITYPE> hiddenI, \ - TensorInfo<DATATYPE, ITYPE> bias1I, \ - TensorInfo<DATATYPE, ITYPE> bias2I, \ - TensorInfo<DATATYPE, ITYPE> cxI, \ - TensorInfo<DATATYPE, ITYPE> hyI, \ - TensorInfo<DATATYPE, ITYPE> cyI, \ - ITYPE hsz, \ - ITYPE totalElements); \ - \ - template void __global__ THNN_(LSTMBackward)<DATATYPE, ITYPE, DIM> \ - (TensorInfo<DATATYPE, ITYPE> inputI, \ - TensorInfo<DATATYPE, ITYPE> hiddenI, \ - TensorInfo<DATATYPE, ITYPE> cxI, \ - TensorInfo<DATATYPE, ITYPE> cyI, \ - TensorInfo<DATATYPE, ITYPE> gradoutputI, \ - TensorInfo<DATATYPE, ITYPE> gradoutputcellI, \ - TensorInfo<DATATYPE, ITYPE> gradinputI, \ - ITYPE hsz, \ - ITYPE totalElements); \ - - -#define EXPAND_DIM(ITYPE) \ - EXPAND_FUNCTION(ITYPE, -2) \ - EXPAND_FUNCTION(ITYPE, -1) \ - EXPAND_FUNCTION(ITYPE, 1) \ - EXPAND_FUNCTION(ITYPE, 2) \ - - -#define EXPAND_TYPE \ - EXPAND_DIM(unsigned int) \ - EXPAND_DIM(unsigned long) \ - - -EXPAND_TYPE - -// ************ END generating specializations ************** // - -// ************ START Create actual function calls ********** // + +// ************ START Create function calls ********** // #define FILL_FUNCTION(ITYPE, DIM, FUNCTION) FUNCTION(ITYPE, DIM) #define FILL_DIM(ITYPE, DIM, FUNCTION) \ |