Use hardware concurrency as tbb grain size during memcpy and memset

diff --git a/common/memcpy.h b/common/memcpy.h
index 530f6172b681da5f7f9bab4ce6166f80119885e9..caf403e58b4374eb5002dafd49b9212e92450697 100644 (file)
--- a/common/memcpy.h
+++ b/common/memcpy.h
@@ -25,12 +25,14 @@
 
 #include <tbb/parallel_for.h>
 
+#include <boost/thread/thread.hpp>
+
 namespace caspar {
 
 static void fast_memcpy(void* dest, const void* src, std::size_t size)
 {
        tbb::affinity_partitioner partitioner;
-       tbb::parallel_for(tbb::blocked_range<std::size_t>(0, size, size / 128), [&](const tbb::blocked_range<size_t>& range)
+       tbb::parallel_for(tbb::blocked_range<std::size_t>(0, size, size / boost::thread::hardware_concurrency()), [&](const tbb::blocked_range<size_t>& range)
        {
                A_memcpy(
                                reinterpret_cast<std::uint8_t*>(dest) + range.begin(),

diff --git a/common/memset.h b/common/memset.h
index a53b0b2d4e2f019c5c2a073fdd740ed4f08036f7..f3b126f7878f43ab6ce06c564c8c1dc554417eb7 100644 (file)
--- a/common/memset.h
+++ b/common/memset.h
@@ -25,12 +25,14 @@
 
 #include <tbb/parallel_for.h>
 
+#include <boost/thread/thread.hpp>
+
 namespace caspar {
 
 static void fast_memset(void* dest, int c, std::size_t size)
 {
        tbb::affinity_partitioner partitioner;
-       tbb::parallel_for(tbb::blocked_range<std::size_t>(0, size, size / 128), [&](const tbb::blocked_range<size_t>& range)
+       tbb::parallel_for(tbb::blocked_range<std::size_t>(0, size, size / boost::thread::hardware_concurrency()), [&](const tbb::blocked_range<size_t>& range)
        {
                A_memset(
                                reinterpret_cast<std::uint8_t*>(dest) + range.begin(),