From f84963ed95ff277bf3abb2e2517b3017a25ccf3f Mon Sep 17 00:00:00 2001 From: Mehdi Goli Date: Tue, 7 Mar 2017 14:27:10 +0000 Subject: 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. --- unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h') diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h index c9c7acfdc..94899252b 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionSycl.h @@ -35,7 +35,7 @@ static void run(OP op, BufferTOut& bufOut, ptrdiff_t out_offset, BufferTIn& bufI /* Two accessors are used: one to the buffer that is being reduced, * and a second to local memory, used to store intermediate data. */ auto aI =bufI.template get_access(h); - auto aOut =bufOut.template get_access(h); + auto aOut =bufOut.template get_access(h); typedef decltype(aI) InputAccessor; typedef decltype(aOut) OutputAccessor; typedef cl::sycl::accessor LocalAccessor; @@ -158,7 +158,7 @@ struct InnerReducer { typedef decltype(TensorSycl::internal::createTupleOfAccessors(cgh, self.impl())) Tuple_of_Acc; // create a tuple of accessors from Evaluator Tuple_of_Acc tuple_of_accessors = TensorSycl::internal::createTupleOfAccessors(cgh, self.impl()); - auto output_accessor = dev.template get_sycl_accessor(cgh, output); + auto output_accessor = dev.template get_sycl_accessor(cgh, output); ptrdiff_t out_offset = dev.get_offset(output); Index red_size = (num_values_to_reduce!=0)? num_values_to_reduce : static_cast(1); cgh.parallel_for( cl::sycl::nd_range<1>(cl::sycl::range<1>(GRange), cl::sycl::range<1>(tileSize)), -- cgit v1.2.3