diff options
Diffstat (limited to 'src/ExecuteKernelU8S8.cc')
-rw-r--r-- | src/ExecuteKernelU8S8.cc | 28 |
1 files changed, 26 insertions, 2 deletions
diff --git a/src/ExecuteKernelU8S8.cc b/src/ExecuteKernelU8S8.cc index e091a87..b3f8c15 100644 --- a/src/ExecuteKernelU8S8.cc +++ b/src/ExecuteKernelU8S8.cc @@ -14,7 +14,7 @@ double kernel_time = 0.0; double postprocessing_time = 0.0; #endif -namespace fbgemm2 { +namespace fbgemm { template <typename packingAMatrix, typename cT, typename processOutputType> ExecuteKernel< @@ -328,6 +328,18 @@ template class ExecuteKernel< memCopy<>>; template class ExecuteKernel< + PackAWithIm2Col<uint8_t, int16_t>, + PackBMatrix<int8_t, int16_t>, + uint8_t, + ReQuantizeOutput<false>>; + +template class ExecuteKernel< + PackAWithIm2Col<uint8_t, int16_t, 3>, + PackBMatrix<int8_t, int16_t>, + uint8_t, + ReQuantizeOutput<false>>; + +template class ExecuteKernel< PackAWithRowOffset<uint8_t, int32_t>, PackBMatrix<int8_t, int32_t>, int32_t, @@ -346,6 +358,18 @@ template class ExecuteKernel< memCopy<>>; template class ExecuteKernel< + PackAWithIm2Col<uint8_t, int32_t>, + PackBMatrix<int8_t, int32_t>, + uint8_t, + ReQuantizeOutput<false>>; + +template class ExecuteKernel< + PackAWithIm2Col<uint8_t, int32_t, 3>, + PackBMatrix<int8_t, int32_t>, + uint8_t, + ReQuantizeOutput<false>>; + +template class ExecuteKernel< PackAWithQuantRowOffset<uint8_t, int32_t>, PackBMatrix<int8_t, int32_t>, int32_t, @@ -363,4 +387,4 @@ template class ExecuteKernel< int32_t, DoNothing<int32_t, int32_t>>; -} // namespace fbgemm2 +} // namespace fbgemm |