aboutsummaryrefslogtreecommitdiffhomepage
path: root/unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h
diff options
context:
space:
mode:
authorGravatar Yuefeng Zhou <yuefengz@google.com>2018-02-20 16:50:05 -0800
committerGravatar Yuefeng Zhou <yuefengz@google.com>2018-02-20 16:50:05 -0800
commit1eff6cf8a77f1b8699671d31f8f307a6fd9170ea (patch)
treef327431b9b918148d548b849cd24a3633bd50274 /unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h
parentadb134d47e70906ed15d877d7a0fe03f3f665c02 (diff)
Use device's allocate function instead of internal::aligned_malloc. This would make it easier to track memory usage in device instances.
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h')
-rw-r--r--unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h4
1 files changed, 2 insertions, 2 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h
index d30cc96ab..6fb69910e 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionThreadPool.h
@@ -377,7 +377,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
divup<size_t>(bm_ * bk_ * sizeof(LhsScalar), align) * align;
size_t rhs_size =
divup<size_t>(bn_ * bk_ * sizeof(RhsScalar), align) * align;
- packed_mem_ = static_cast<char*>(internal::aligned_malloc(
+ packed_mem_ = static_cast<char*>(device_.allocate(
(nm0_ * lhs_size + nn0_ * rhs_size) * std::min<size_t>(nk_, P - 1)));
char* mem = static_cast<char*>(packed_mem_);
for (Index x = 0; x < numext::mini<Index>(nk_, P - 1); x++) {
@@ -399,7 +399,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
for (Index m = 0; m < nm_; m++) delete[] state_kernel_[x][m];
delete[] state_kernel_[x];
}
- internal::aligned_free(packed_mem_);
+ device_.deallocate(packed_mem_);
}
void run() {