apeforest commented on a change in pull request #12926: parallelize
NDArray::Copy<cpu, cpu> when data size is large
URL: https://github.com/apache/incubator-mxnet/pull/12926#discussion_r236429880
##########
File path: src/common/utils.h
##########
@@ -713,6 +713,23 @@ inline void EmplaceBackZeros(const NDArrayStorageType
stype, const TShape &shape
}
}
+
+/*!
+ * \brief parallelize copy by OpenMP.
+ */
+template<typename DType>
+inline void ParallelCopy(DType* dst, const DType* src, index_t size) {
+ static index_t copy_block_size =
dmlc::GetEnv("MXNET_CPU_PARALLEL_COPY_SIZE", 200000);
+ if (size >= copy_block_size) {
+ #pragma omp parallel for
num_threads(engine::OpenMP::Get()->GetRecommendedOMPThreadCount())
+ for (index_t i = 0; i < size; ++i) {
+ dst[i] = src[i];
+ }
+ } else {
+ std::memcpy(dst, src, sizeof(DType) * size);
Review comment:
There is a CheckContinuous function call before the memcpy in original
mshadow::Copy function
(https://github.com/dmlc/mshadow/blob/696803bd7723ade8230af878460d96c68a550fbc/mshadow/tensor_cpu-inl.h#L132)
Do we not need this any more?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
