Remove async prefetch to avoid regression

Author: riverlijunjie

src/common/util/include/openvino/util/parallel_mem_streambuf.hpp

@@ -31,7 +31,7 @@
 #include "openvino/util/log.hpp"
 #include "openvino/util/parallel_read_streambuf.hpp"
 
-#define ENABLE_OPENVINO_DEBUG 0
+#define ENABLE_BD_PROFILING_LOG 0
 
 namespace ov {
 namespace util {
@@ -242,7 +242,7 @@ class ParallelMemStreamBuf : public std::streambuf {
         madvise(const_cast<char*>(src), size, MADV_WILLNEED);
 #endif
 
-#ifdef ENABLE_OPENVINO_DEBUG
+#if ENABLE_BD_PROFILING_LOG
         const auto t0 = std::chrono::steady_clock::now();
 #endif
 
@@ -252,7 +252,7 @@ class ParallelMemStreamBuf : public std::streambuf {
             std::memcpy(dst + offset, src + offset, copy_size);
         });
 
-#ifdef ENABLE_OPENVINO_DEBUG
+#if ENABLE_BD_PROFILING_LOG
         {
             const auto t1 = std::chrono::steady_clock::now();
             const double elapsed_s = std::chrono::duration<double>(t1 - t0).count();

src/common/util/include/openvino/util/parallel_read_streambuf.hpp

@@ -278,7 +278,7 @@ class ParallelReadStreamBuf : public std::streambuf {
         std::vector<std::future<void>> futures;
         futures.reserve(num_threads);
 
-#ifdef ENABLE_BD_PROFILING_LOG
+#if ENABLE_BD_PROFILING_LOG
         const auto t0 = std::chrono::steady_clock::now();
 #endif
 
@@ -358,7 +358,7 @@ class ParallelReadStreamBuf : public std::streambuf {
         for (auto& f : futures) {
             f.get();
         }
-#ifdef ENABLE_BD_PROFILING_LOG
+#if ENABLE_BD_PROFILING_LOG
         {
             const auto t1 = std::chrono::steady_clock::now();
             const double elapsed_s = std::chrono::duration<double>(t1 - t0).count();

src/common/util/src/os/lin/lin_mmap_object.cpp

@@ -89,11 +89,6 @@ class MapHolder final : public MappedMemory {
                 throw std::runtime_error("Can not create file mapping for " + std::to_string(fd) +
                                          ", err=" + std::strerror(errno));
             }
-            // Hint the kernel to read pages sequentially and start async
-            // prefetch immediately.  MADV_SEQUENTIAL doubles the readahead
-            // window; MADV_WILLNEED initiates I/O in the background so pages
-            // are resident before compile_model accesses individual constants.
-            madvise(m_data, m_size, MADV_SEQUENTIAL | MADV_WILLNEED);
         } else {
             m_data = MAP_FAILED;
         }

src/common/util/src/os/win/win_mmap_object.cpp

@@ -116,10 +116,6 @@ class MapHolder : public ov::MappedMemory {
             if (!m_data) {
                 throw std::runtime_error("Can not create map view for " + util::path_to_string(path));
             }
-            // Hint the OS to start async prefetch of all mapped pages so
-            // they are (mostly) resident by the time constants are accessed.
-            WIN32_MEMORY_RANGE_ENTRY prefetch_range{m_data, m_size};
-            PrefetchVirtualMemory(GetCurrentProcess(), 1, &prefetch_range, 0);
         } else {
             m_data = nullptr;
         }