diff options
| author |  Rasmus Munk Larsen <rmlarsen@google.com> | 2017-09-08 15:49:55 -0700 |
|---|---|---|
| committer | Rasmus Munk Larsen <rmlarsen@google.com> | 2017-09-08 15:49:55 -0700 |
| commit | 94e2213b3823ec1141077a7a7cc4f0b0af11876b (patch) | |
| tree | 246855911c9845989f9240ba1765332cebe423d6 /unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h | |
| parent | 6d42309f1394396c984212abad73469e6a50c3a3 (diff) | |
Avoid undefined behavior in Eigen::TensorCostModel::numThreads.
If the cost is large enough then the thread count can be larger than the maximum
representable int, so just casting it to an int is undefined behavior.
Contributed by phurst@google.com.
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h')
| -rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h b/unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h index 83c449cf1..ac6717977 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorCostModel.h @@ -174,8 +174,9 @@ class TensorCostModel { static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE int numThreads( double output_size, const TensorOpCost& cost_per_coeff, int max_threads) { double cost = totalCost(output_size, cost_per_coeff); - int threads = (cost - kStartupCycles) / kPerThreadCycles + 0.9; - return numext::mini(max_threads, numext::maxi(1, threads)); + // Make sure we don't invoke undefined behavior when we convert to an int. + threads = numext::mini<double>(threads, GenericNumTraits<int>::highest()); + return numext::mini(max_threads, numext::maxi<int>(1, threads)); } // taskSize assesses parallel task size. |