diff options
author | Mehdi Goli <mehdi.goli@codeplay.com> | 2017-03-07 14:27:10 +0000 |
---|---|---|
committer | Mehdi Goli <mehdi.goli@codeplay.com> | 2017-03-07 14:27:10 +0000 |
commit | f84963ed95ff277bf3abb2e2517b3017a25ccf3f (patch) | |
tree | b9616be8fe4f8048287a147d070288701457ea3c /unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h | |
parent | 8296b87d7bd98c19c6064241880691f164790ede (diff) |
Adding TensorIndexTuple and TensorTupleReduceOP backend (ArgMax/Min) for sycl; fixing the address space issue for const TensorMap; converting all discard_write to write due to data missmatch.
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h')
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h index fcd7d4d00..5b4c3c5bd 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h @@ -11,7 +11,7 @@ // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. /***************************************************************** - * TensorSyclConvertToDeviceExpression.h + * TensorTensorContractionsycl.h * * \brief: * TensorContractionsycl @@ -389,9 +389,9 @@ template< typename Self, typename OutScalar, typename ContractT, typename LeftNo cl::sycl::range<2>(LocalThreadSizeM, LocalThreadSizeN)), KernelConstructor<HostExpr, OutScalar, LhsScalar, RhsScalar, LHSFunctorExpr, RHSFunctorExpr, LhsLocalAcc, RhsLocalAcc, OutAccessor, Index, ContractT, LeftNocontractT, RightNocontractT, lhs_inner_dim_contiguous, rhs_inner_dim_contiguous, rhs_inner_dim_reordered, TileSizeDimM, TileSizeDimN, TileSizeDimK, - WorkLoadPerThreadM, WorkLoadPerThreadN, LocalThreadSizeM, LocalThreadSizeN, LoadPerThreadLhs, LoadPerThreadRhs, LHSTupleType, RHSTupleType, Eigen::DefaultDevice>(lhs_functors, rhs_functors, + WorkLoadPerThreadM, WorkLoadPerThreadN, LocalThreadSizeM, LocalThreadSizeN, LoadPerThreadLhs, LoadPerThreadRhs, LHSTupleType, RHSTupleType, Eigen::SyclKernelDevice>(lhs_functors, rhs_functors, localLhs, localRhs, out_res, out_offset, roundUpK, M, N, K, m_k_strides, m_left_contracting_strides, m_right_contracting_strides,m_i_strides, m_j_strides, - m_left_nocontract_strides,m_right_nocontract_strides, left_tuple_of_accessors, right_tuple_of_accessors, Eigen::DefaultDevice())); + m_left_nocontract_strides,m_right_nocontract_strides, left_tuple_of_accessors, right_tuple_of_accessors, Eigen::SyclKernelDevice())); }); self.device().asynchronousExec(); } |