From ad94d8cc4fd1867ecc37c6eeea883b25939a0c99 Mon Sep 17 00:00:00 2001
From: Dale Phurrough <dale@hidale.com>
Date: Tue, 8 Dec 2020 11:05:14 +0100
Subject: [PATCH] Merge pull request #19029 from
 diablodale:fix19004-memthreadstart

add thread-safe startup of fastMalloc and fastFree

* add perf test core memory allocation

* fix threading in isAlignedAllocationEnabled()

* tweaks requested by maintainer
---
 modules/core/perf/perf_allocation.cpp | 48 +++++++++++++++++++++++++++++++++++
 modules/core/src/alloc.cpp            | 20 ++++++++-------
 2 files changed, 59 insertions(+), 9 deletions(-)
 create mode 100755 modules/core/perf/perf_allocation.cpp

diff --git a/modules/core/perf/perf_allocation.cpp b/modules/core/perf/perf_allocation.cpp
new file mode 100755
index 0000000..2f3bf3e
--- /dev/null
+++ b/modules/core/perf/perf_allocation.cpp
@@ -0,0 +1,48 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#include "perf_precomp.hpp"
+#include <array>
+
+using namespace perf;
+
+#define ALLOC_MAT_SIZES ::perf::szSmall24, ::perf::szSmall32, ::perf::szSmall64, \
+    ::perf::sz5MP, ::perf::sz2K, ::perf::szSmall128, ::perf::szODD, ::perf::szQVGA, \
+    ::perf::szVGA, ::perf::szSVGA, ::perf::sz720p, ::perf::sz1080p, ::perf::sz2160p, \
+    ::perf::sz4320p, ::perf::sz3MP, ::perf::szXGA, ::perf::szSXGA, ::perf::szWQHD, \
+    ::perf::sznHD, ::perf::szqHD
+
+namespace opencv_test
+{
+
+typedef perf::TestBaseWithParam<MatType> MatDepth_tb;
+
+PERF_TEST_P(MatDepth_tb, DISABLED_Allocation_Aligned,
+    testing::Values(CV_8UC1, CV_16SC1, CV_8UC3, CV_8UC4))
+{
+    const int matType = GetParam();
+    const cv::Mat utility(1, 1, matType);
+    const size_t elementBytes = utility.elemSize();
+
+    const std::array<cv::Size, 20> sizes{ALLOC_MAT_SIZES};
+    std::array<size_t, 20> bytes;
+    for (size_t i = 0; i < sizes.size(); ++i)
+    {
+        bytes[i] = sizes[i].width * sizes[i].height * elementBytes;
+    }
+
+    declare.time(60)
+           .iterations(100);
+
+    TEST_CYCLE()
+    {
+        for (int i = 0; i < 100000; ++i)
+        {
+            fastFree(fastMalloc(bytes[i % sizes.size()]));
+        }
+    }
+    SANITY_CHECK_NOTHING();
+}
+
+};
diff --git a/modules/core/src/alloc.cpp b/modules/core/src/alloc.cpp
index 9801299..4b58fb6 100644
--- a/modules/core/src/alloc.cpp
+++ b/modules/core/src/alloc.cpp
@@ -100,25 +100,27 @@ static bool readMemoryAlignmentParameter()
     // TODO add checks for valgrind, ASAN if value == false
     return value;
 }
+
+#if defined _MSC_VER
+#pragma warning(suppress:4714)  // preventive: const marked as __forceinline not inlined
+static __forceinline
+#else
 static inline
+#endif
 bool isAlignedAllocationEnabled()
 {
-    static bool initialized = false;
-    static bool useMemalign = true;
-    if (!initialized)
-    {
-        initialized = true;  // trick to avoid stuck in acquire (works only if allocations are scope based)
-        useMemalign = readMemoryAlignmentParameter();
-    }
+    // use construct on first use idiom https://isocpp.org/wiki/faq/ctors#static-init-order-on-first-use
+    // details: https://github.com/opencv/opencv/issues/15691
+    static bool useMemalign = readMemoryAlignmentParameter();
     return useMemalign;
 }
-// do not use variable directly, details: https://github.com/opencv/opencv/issues/15691
+
+// need for this static const is disputed; retaining as it doesn't cause harm
 static const bool g_force_initialization_memalign_flag
 #if defined __GNUC__
     __attribute__((unused))
 #endif
     = isAlignedAllocationEnabled();
-
 #endif
 
 #ifdef OPENCV_ALLOC_ENABLE_STATISTICS
-- 
2.7.4