diff options
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h')
| -rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h index b9956cd43..6d3aa24c8 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h @@ -505,9 +505,9 @@ template<typename Scalar, typename Index, typename LhsMapper, __global__ void #if defined(EIGEN_HIPCC) __launch_bounds__(512, 1) -#else +#else __launch_bounds__(512) -#endif +#endif EigenContractionKernel(const LhsMapper lhs, const RhsMapper rhs, const OutputMapper output, const Index m_size, const Index n_size, const Index k_size) { @@ -698,7 +698,7 @@ EigenFloatContractionKernelInternal16x16(const LhsMapper lhs, const RhsMapper rh #undef prefetch_lhs #undef add_vals - + Index horiz_base = threadIdx.y*4+base_n; if (!CHECK_LHS_BOUNDARY && !CHECK_RHS_BOUNDARY) { for (int i = 0; i < 4; i++) { @@ -1137,7 +1137,7 @@ template<typename Index, typename LhsMapper, __global__ void #if defined(EIGEN_HIPCC) __launch_bounds__(256, 1) -#else +#else __launch_bounds__(256) #endif EigenFloatContractionKernel(const LhsMapper lhs, const RhsMapper rhs, @@ -1184,7 +1184,7 @@ template<typename Index, typename LhsMapper, __global__ void #if defined(EIGEN_HIPCC) __launch_bounds__(256, 1) -#else +#else __launch_bounds__(256) #endif EigenFloatContractionKernel16x16(const LhsMapper lhs, const RhsMapper rhs, |