Re-enabled the optimized reduction CUDA code.

This commit is contained in:
Benoit Steiner
2016-01-11 09:07:14 -08:00
parent b557662e58
commit 780623261e
2 changed files with 3 additions and 5 deletions

View File

@@ -527,7 +527,6 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
}
// Attempt to use an optimized reduction.
#if 0
else if (RunningOnGPU && data && (m_device.majorDeviceVersion() >= 3)) {
bool reducing_inner_dims = true;
for (int i = 0; i < NumReducedDims; ++i) {
@@ -563,7 +562,6 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
return false;
}
}
#endif
return true;
}