CFD/eigen
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix int overflow causing cxx11_tensor_gpu_1 to fail.

Browse Source
This commit is contained in:
Antonio Sánchez 2023-11-06 17:10:16 +00:00 committed by Rasmus Munk Larsen
parent 6f9ad7da61
commit a25f02d73e
1 changed files with 3 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h
View File

@ -655,13 +655,11 @@ EIGEN_STRONG_INLINE void TensorExecutor<Expression, GpuDevice, Vectorizable, Til
const int block_size = device.maxGpuThreadsPerBlock();
const int max_blocks =
numext::mini<int64_t>(device.getNumGpuMultiProcessors() *
device.maxGpuThreadsPerMultiProcessor(),
NumTraits<StorageIndex>::highest()) /
block_size;
static_cast<int>(numext::mini<int64_t>(device.getNumGpuMultiProcessors() * device.maxGpuThreadsPerMultiProcessor(),
NumTraits<StorageIndex>::highest()) / block_size);
const StorageIndex size = array_prod(evaluator.dimensions());
// Create a least one block to ensure we won't crash when tensorflow calls with tensors of size 0.
const int num_blocks = numext::maxi<int>(numext::mini<int>(max_blocks, numext::div_ceil<int>(size, block_size)), 1);
const int num_blocks = numext::maxi<int>(numext::mini<int>(max_blocks, static_cast<int>(numext::div_ceil<StorageIndex>(size, block_size))), 1);
LAUNCH_GPU_KERNEL(
(EigenMetaKernel<TensorEvaluator<Expression, GpuDevice>, StorageIndex>),