From a25f02d73e33ac95829f958a832c715c612994dc Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Antonio=20S=C3=A1nchez?= <cantonios@google.com>
Date: Mon, 6 Nov 2023 17:10:16 +0000
Subject: [PATCH] Fix int overflow causing cxx11_tensor_gpu_1 to fail.

---
 unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h b/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h
index 4eebbe70f..e8fb85d83 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h
@@ -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>),