From b33cf92878a57ec86d5e5715e7cde3a0cd360fd6 Mon Sep 17 00:00:00 2001 From: Benoit Steiner Date: Tue, 18 Nov 2014 14:32:41 -0800 Subject: Fixed the evaluation of expressions involving tensors of 2 or 3 elements on CUDA devices. --- unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'unsupported/Eigen') diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h b/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h index 4fa8e83ef..f27f643c1 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h @@ -168,11 +168,10 @@ __launch_bounds__(1024) const Index PacketSize = unpacket_traits::size; const Index vectorized_step_size = step_size * PacketSize; const Index vectorized_size = (size / PacketSize) * PacketSize; - Index i = first_index * PacketSize; - for ( ; i < vectorized_size; i += vectorized_step_size) { + for (Index i = first_index * PacketSize; i < vectorized_size; i += vectorized_step_size) { eval.evalPacket(i); } - for ( ; i < size; i += step_size) { + for (Index i = vectorized_size + first_index; i < size; i += step_size) { eval.evalScalar(i); } } -- cgit v1.2.3