From bf0173bf38a4bafee9a00cb99d4ce6696ed8ac67 Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Fri, 21 Jul 2017 16:42:28 +0300
Subject: [PATCH 1/9] ts: update valgrind suppressions
---
cmake/OpenCVUtils.cmake | 6 +
modules/calib3d/src/stereobm.cpp | 2 +-
modules/core/src/stat.cpp | 2 +-
modules/ts/misc/run.py | 2 +-
modules/ts/misc/run_suite.py | 11 +-
platforms/scripts/valgrind.supp | 198 ++++++++++++++++++++++-
platforms/scripts/valgrind_3rdparty.supp | 113 +++++++++++++
7 files changed, 324 insertions(+), 10 deletions(-)
create mode 100644 platforms/scripts/valgrind_3rdparty.supp
diff --git a/cmake/OpenCVUtils.cmake b/cmake/OpenCVUtils.cmake
index d91d1a62d6..8ef7205009 100644
--- a/cmake/OpenCVUtils.cmake
+++ b/cmake/OpenCVUtils.cmake
@@ -533,6 +533,12 @@ macro(ocv_finalize_status)
execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different "${OPENCV_BUILD_INFO_FILE}" "${OPENCV_MODULE_opencv_core_BINARY_DIR}/version_string.inc" OUTPUT_QUIET)
endif()
endif()
+
+ if(UNIX)
+ install(FILES "${OpenCV_SOURCE_DIR}/platforms/scripts/valgrind.supp"
+ "${OpenCV_SOURCE_DIR}/platforms/scripts/valgrind_3rdparty.supp"
+ DESTINATION "${OPENCV_OTHER_INSTALL_PATH}" COMPONENT "dev")
+ endif()
endmacro()
diff --git a/modules/calib3d/src/stereobm.cpp b/modules/calib3d/src/stereobm.cpp
index 5ed45a4884..f6d8213545 100644
--- a/modules/calib3d/src/stereobm.cpp
+++ b/modules/calib3d/src/stereobm.cpp
@@ -197,7 +197,7 @@ prefilterXSobel( const Mat& src, Mat& dst, int ftzero )
{
int x, y;
const int OFS = 256*4, TABSZ = OFS*2 + 256;
- uchar tab[TABSZ];
+ uchar tab[TABSZ] = { 0 };
Size size = src.size();
for( x = 0; x < TABSZ; x++ )
diff --git a/modules/core/src/stat.cpp b/modules/core/src/stat.cpp
index e324f85dac..13d8c215cf 100644
--- a/modules/core/src/stat.cpp
+++ b/modules/core/src/stat.cpp
@@ -1547,7 +1547,7 @@ static bool ocl_meanStdDev( InputArray _src, OutputArray _mean, OutputArray _sdv
bool haveMask = _mask.kind() != _InputArray::NONE;
int nz = haveMask ? -1 : (int)_src.total();
- Scalar mean, stddev;
+ Scalar mean(0), stddev(0);
const int cn = _src.channels();
if (cn > 4)
return false;
diff --git a/modules/ts/misc/run.py b/modules/ts/misc/run.py
index 0f2116e6af..0befa79ea9 100755
--- a/modules/ts/misc/run.py
+++ b/modules/ts/misc/run.py
@@ -37,7 +37,7 @@ if __name__ == "__main__":
# Valgrind
parser.add_argument("--valgrind", action="store_true", default=False, help="Run C++ tests in valgrind")
- parser.add_argument("--valgrind_supp", metavar="FILE", help="Path to valgrind suppression file (example: --valgrind_supp opencv/platforms/scripts/valgrind.supp)")
+ parser.add_argument("--valgrind_supp", metavar="FILE", action='append', help="Path to valgrind suppression file (example: --valgrind_supp opencv/platforms/scripts/valgrind.supp)")
parser.add_argument("--valgrind_opt", metavar="OPT", action="append", default=[], help="Add command line option to valgrind (example: --valgrind_opt=--leak-check=full)")
# Android
diff --git a/modules/ts/misc/run_suite.py b/modules/ts/misc/run_suite.py
index 311f415240..8242c3a1db 100644
--- a/modules/ts/misc/run_suite.py
+++ b/modules/ts/misc/run_suite.py
@@ -103,10 +103,15 @@ class TestSuite(object):
def wrapInValgrind(self, cmd = []):
if self.options.valgrind:
res = ['valgrind']
- if self.options.valgrind_supp:
- res.append("--suppressions=%s" % self.options.valgrind_supp)
+ supp = self.options.valgrind_supp or []
+ for f in supp:
+ if os.path.isfile(f):
+ res.append("--suppressions=%s" % f)
+ else:
+ print("WARNING: Valgrind suppression file is missing, SKIP: %s" % f)
res.extend(self.options.valgrind_opt)
- return res + cmd + [longTestFilter(LONG_TESTS_DEBUG_VALGRIND)]
+ has_gtest_filter = next((True for x in cmd if x.startswith('--gtest_filter=')), False)
+ return res + cmd + ([longTestFilter(LONG_TESTS_DEBUG_VALGRIND)] if not has_gtest_filter else [])
return cmd
def tryCommand(self, cmd):
diff --git a/platforms/scripts/valgrind.supp b/platforms/scripts/valgrind.supp
index 54833e08bd..e78d047c58 100644
--- a/platforms/scripts/valgrind.supp
+++ b/platforms/scripts/valgrind.supp
@@ -1,13 +1,203 @@
{
- IPP static init
+ OpenCV-IPP static init
Memcheck:Cond
fun:ippicvGetCpuFeatures
fun:ippicvStaticInit
}
{
- TBB - allocate_via_handler_v3 issue
+ OpenCV-getInitializationMutex
Memcheck:Leak
- fun:malloc
- fun:_ZN3tbb8internal23allocate_via_handler_v3Em
+ ...
+ fun:_ZN2cv22getInitializationMutexEv
+}
+
+{
+ OpenCV-getStdAllocator
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3Mat15getStdAllocatorEv
+}
+
+{
+ OpenCV-getOpenCLAllocator
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl18getOpenCLAllocatorEv
+}
+
+{
+ OpenCV-getCoreTlsData
+ Memcheck:Leak
+ fun:_Znwm
+ fun:_ZN2cv14getCoreTlsDataEv
+}
+
+{
+ OpenCV-TLS-getTlsStorage
+ Memcheck:Leak
+ ...
+ fun:_ZN2cvL13getTlsStorageEv
+}
+
+{
+ OpenCV-TLS-getData()
+ Memcheck:Leak
+ ...
+ fun:*setData*
+ fun:_ZNK2cv16TLSDataContainer7getDataEv
+}
+
+{
+ OpenCV-parallel_for
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv13ThreadManager8initPoolEv*
+}
+
+{
+ OpenCV-parallel_for
+ Memcheck:Leak
+ fun:_Znwm
+ fun:*instance*
+ fun:_ZN2cv21parallel_for_pthreadsERKNS_5RangeERKNS_16ParallelLoopBodyEd
+ fun:_ZN2cv13parallel_for_ERKNS_5RangeERKNS_16ParallelLoopBodyEd
+}
+
+{
+ OpenCV-parallel_for-ThreadManager::TLS
+ Memcheck:Leak
+ fun:_Znwm
+ fun:_ZNK2cv7TLSDataINS_13ThreadManager13work_thread_tEE18createDataInstanceEv
+}
+
+{
+ OpenCV-parallel_for-setNumThreads()
+ Memcheck:Leak
+ fun:_Znwm
+ fun:_ZN2cv13ThreadManager8instanceEv
+ fun:_ZN2cv33parallel_pthreads_set_threads_numEi
+ fun:_ZN2cv13setNumThreadsEi
+}
+
+{
+ OpenCV-parallel_for-getNumThreads()
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv13getNumThreadsEv
+}
+
+{
+ OpenCV-getIPPSingelton
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ippL15getIPPSingeltonEv
+}
+
+{
+ OpenCV-getGlobalMatOpInitializer
+ Memcheck:Leak
+ fun:_Znwm
+ fun:_ZN2cvL25getGlobalMatOpInitializerEv
+}
+
+{
+ OpenCV-CoreTLSData
+ Memcheck:Leak
+ ...
+ fun:_ZNK2cv7TLSDataINS_11CoreTLSDataEE3getEv
+}
+
+{
+ OpenCV-ThreadID
+ Memcheck:Leak
+ fun:_Znwm
+ fun:_ZNK2cv7TLSDataINS_12_GLOBAL__N_18ThreadIDEE18createDataInstanceEv
+}
+
+{
+ OpenCV-ThreadID-TLS
+ Memcheck:Leak
+ fun:_Znwm
+ fun:getThreadIDTLS
+}
+
+{
+ OpenCV-CoreTLS
+ Memcheck:Leak
+ fun:_Znwm
+ fun:_ZNK2cv7TLSDataINS_11CoreTLSDataEE18createDataInstanceEv
+}
+
+{
+ OpenCV-haveOpenCL
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl10haveOpenCLEv
+}
+
+{
+ OpenCV-DNN-getLayerFactoryMutex
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3dnn*L20getLayerFactoryMutexEv
+}
+
+{
+ OpenCV-ocl::Context
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl7Context10getDefaultEb
+}
+
+{
+ OpenCV-ocl::Device
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl6Device10getDefaultEv
+}
+
+{
+ OpenCV-ocl::Queue
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl5Queue6createERKNS0_7ContextERKNS0_6DeviceE
+}
+
+{
+ OpenCV-ocl::Program
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl6Kernel6createEPKcRKNS0_7ProgramE
+}
+
+{
+ OpenCV-ocl::ProgramEntry
+ Memcheck:Leak
+ ...
+ fun:_ZNK2cv3ocl8internal12ProgramEntrycvRNS0_13ProgramSourceEEv
+}
+
+{
+ OpenCV-ocl::Context::getProg
+ Memcheck:Leak
+ ...
+ fun:_ZN2cv3ocl7Context7getProgERKNS0_13ProgramSourceERKNS_6StringERS5_
+}
+
+
+{
+ OpenCV-ITT
+ Memcheck:Leak
+ ...
+ fun:__itt_*create*
+}
+
+{
+ OpenCV-FFmpeg-swsscale
+ Memcheck:Addr16
+ ...
+ fun:sws_scale
+ fun:_ZN20CvVideoWriter_FFMPEG10writeFrameEPKhiiiii
+ fun:cvWriteFrame_FFMPEG
}
diff --git a/platforms/scripts/valgrind_3rdparty.supp b/platforms/scripts/valgrind_3rdparty.supp
new file mode 100644
index 0000000000..7b6472d827
--- /dev/null
+++ b/platforms/scripts/valgrind_3rdparty.supp
@@ -0,0 +1,113 @@
+{
+ IPP static init
+ Memcheck:Cond
+ fun:ippicvGetCpuFeatures
+ fun:ippicvStaticInit
+}
+
+{
+ TBB - allocate_via_handler_v3 issue
+ Memcheck:Leak
+ fun:malloc
+ fun:_ZN3tbb8internal23allocate_via_handler_v3Em
+}
+
+{
+ GTest
+ Memcheck:Cond
+ fun:_ZN7testing8internal11CmpHelperLEIddEENS_15AssertionResultEPKcS4_RKT_RKT0_
+}
+
+{
+ OpenCL
+ Memcheck:Cond
+ ...
+ obj:**/libOpenCL.so*
+}
+
+{
+ OpenCL-Intel
+ Memcheck:Cond
+ ...
+ obj:**/libigdrcl.so
+}
+
+{
+ OpenCL-Intel
+ Memcheck:Leak
+ ...
+ obj:*/libigdrcl.so*
+}
+
+{
+ OpenCL
+ Memcheck:Param
+ ioctl(generic)
+ ...
+ fun:clGetPlatformIDs
+}
+
+{
+ OpenCL-Init
+ Memcheck:Leak
+ ...
+ fun:clGetPlatformIDs
+}
+
+{
+ glib
+ Memcheck:Leak
+ fun:*alloc
+ obj:*/libglib*
+}
+
+{
+ gcrypt
+ Memcheck:Leak
+ ...
+ obj:*/libgcrypt*
+}
+
+{
+ p11-kit
+ Memcheck:Leak
+ fun:*alloc
+ obj:*/libp11-kit*
+}
+
+{
+ gobject
+ Memcheck:Leak
+ fun:*alloc
+ ...
+ obj:*/libgobject*
+}
+
+{
+ tasn
+ Memcheck:Leak
+ fun:*alloc
+ obj:*/libtasn*.so*
+}
+
+{
+ dl_init
+ Memcheck:Leak
+ ...
+ fun:_dl_init
+}
+
+{
+ dl_open
+ Memcheck:Leak
+ ...
+ fun:_dl_open
+}
+
+{
+ GDAL
+ Memcheck:Leak
+ fun:*alloc
+ ...
+ obj:/usr/lib/libgdal.so.1.17.1
+}
From 1516103a15c629ee011e146405cf40946bb0b3dd Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Mon, 24 Jul 2017 11:36:21 +0300
Subject: [PATCH 2/9] cmake: fix compiler flags
---
cmake/OpenCVCompilerOptions.cmake | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/cmake/OpenCVCompilerOptions.cmake b/cmake/OpenCVCompilerOptions.cmake
index 2f2e3f741f..2fcc37345a 100644
--- a/cmake/OpenCVCompilerOptions.cmake
+++ b/cmake/OpenCVCompilerOptions.cmake
@@ -203,7 +203,10 @@ if(CMAKE_COMPILER_IS_GNUCXX)
endif()
set(OPENCV_EXTRA_FLAGS_RELEASE "${OPENCV_EXTRA_FLAGS_RELEASE} -DNDEBUG")
- set(OPENCV_EXTRA_FLAGS_DEBUG "${OPENCV_EXTRA_FLAGS_DEBUG} -O0 -DDEBUG -D_DEBUG")
+ if(NOT " ${CMAKE_CXX_FLAGS} ${CMAKE_CXX_FLAGS_DEBUG} " MATCHES "-O")
+ set(OPENCV_EXTRA_FLAGS_DEBUG "${OPENCV_EXTRA_FLAGS_DEBUG} -O0")
+ endif()
+ set(OPENCV_EXTRA_FLAGS_DEBUG "${OPENCV_EXTRA_FLAGS_DEBUG} -DDEBUG -D_DEBUG")
endif()
if(MSVC)
From d17b099994ef86d7f223a091536482737ef5fd5a Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Mon, 24 Jul 2017 19:24:46 +0300
Subject: [PATCH 3/9] ts: don't run DNN tests with large models (valgrind)
---
modules/ts/misc/run_long.py | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/modules/ts/misc/run_long.py b/modules/ts/misc/run_long.py
index 6ae76aec0f..d2ef41fe30 100644
--- a/modules/ts/misc/run_long.py
+++ b/modules/ts/misc/run_long.py
@@ -7,6 +7,7 @@ from pprint import PrettyPrinter as PP
LONG_TESTS_DEBUG_VALGRIND = [
('calib3d', 'Calib3d_InitUndistortRectifyMap.accuracy', 2017.22),
+ ('dnn', 'Reproducibility*', 1000), # large DNN models
('features2d', 'Features2d_Feature2d.no_crash', 1235.68),
('ml', 'ML_RTrees.regression', 1423.47),
('optflow', 'DenseOpticalFlow_DeepFlow.ReferenceAccuracy', 1360.95),
@@ -43,10 +44,8 @@ LONG_TESTS_DEBUG_VALGRIND = [
]
-def longTestFilter(data):
- res = ['*', '-']
- for _, v, _ in data:
- res.append(v)
+def longTestFilter(data, module = None):
+ res = ['*', '-'] + [v for _, v, m in data if module is None or m == module]
return '--gtest_filter={}'.format(':'.join(res))
From 1650c664bc6693e52e36a3ec53fc96cd8a762e34 Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Mon, 24 Jul 2017 23:27:14 +0300
Subject: [PATCH 4/9] ts: don't run imgcodecs tests on large images (valgrind)
---
modules/ts/misc/run_long.py | 2 ++
1 file changed, 2 insertions(+)
diff --git a/modules/ts/misc/run_long.py b/modules/ts/misc/run_long.py
index d2ef41fe30..356244595f 100644
--- a/modules/ts/misc/run_long.py
+++ b/modules/ts/misc/run_long.py
@@ -9,6 +9,8 @@ LONG_TESTS_DEBUG_VALGRIND = [
('calib3d', 'Calib3d_InitUndistortRectifyMap.accuracy', 2017.22),
('dnn', 'Reproducibility*', 1000), # large DNN models
('features2d', 'Features2d_Feature2d.no_crash', 1235.68),
+ ('imgcodecs', 'Imgcodecs_Png.write_big', 1000), # memory limit
+ ('imgcodecs', 'Imgcodecs_Tiff.decode_tile16384x16384', 1000), # memory limit
('ml', 'ML_RTrees.regression', 1423.47),
('optflow', 'DenseOpticalFlow_DeepFlow.ReferenceAccuracy', 1360.95),
('optflow', 'DenseOpticalFlow_DeepFlow_perf.perf/0', 1881.59),
From aad6d28e133934220e800a5553369edca55f1fee Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Tue, 25 Jul 2017 20:16:32 +0300
Subject: [PATCH 5/9] ts: don't run large videoio test (valgrind)
---
modules/ts/misc/run_long.py | 1 +
1 file changed, 1 insertion(+)
diff --git a/modules/ts/misc/run_long.py b/modules/ts/misc/run_long.py
index 356244595f..d820f9780a 100644
--- a/modules/ts/misc/run_long.py
+++ b/modules/ts/misc/run_long.py
@@ -26,6 +26,7 @@ LONG_TESTS_DEBUG_VALGRIND = [
('shape', 'Shape_SCD.regression', 3311.46),
('tracking', 'AUKF.br_mean_squared_error', 10764.6),
('tracking', 'UKF.br_mean_squared_error', 5228.27),
+ ('videoio', 'Videoio_Video.ffmpeg_writebig', 1000),
('xfeatures2d', 'Features2d_RotationInvariance_Descriptor_BoostDesc_LBGM.regression', 1124.51),
('xfeatures2d', 'Features2d_RotationInvariance_Descriptor_VGG120.regression', 2198.1),
('xfeatures2d', 'Features2d_RotationInvariance_Descriptor_VGG48.regression', 1958.52),
From 3f102e5d3a0b36a46a43024d6787c25f730f414d Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Tue, 25 Jul 2017 16:13:56 +0300
Subject: [PATCH 6/9] dnn: protobuf shutdown
---
modules/dnn/src/init.cpp | 17 +++++++++++++++++
1 file changed, 17 insertions(+)
diff --git a/modules/dnn/src/init.cpp b/modules/dnn/src/init.cpp
index 9222d5ecb1..97ea169f81 100644
--- a/modules/dnn/src/init.cpp
+++ b/modules/dnn/src/init.cpp
@@ -42,6 +42,8 @@
#include "precomp.hpp"
#include <opencv2/dnn/layer.details.hpp>
+#include <google/protobuf/stubs/common.h>
+
namespace cv {
namespace dnn {
CV__DNN_EXPERIMENTAL_NS_BEGIN
@@ -56,11 +58,26 @@ Mutex& getInitializationMutex()
// force initialization (single-threaded environment)
Mutex* __initialization_mutex_initializer = &getInitializationMutex();
+namespace {
+using namespace google::protobuf;
+class ProtobufShutdown {
+public:
+ bool initialized;
+ ProtobufShutdown() : initialized(true) {}
+ ~ProtobufShutdown()
+ {
+ initialized = false;
+ google::protobuf::ShutdownProtobufLibrary();
+ }
+};
+} // namespace
void initializeLayerFactory()
{
CV_TRACE_FUNCTION();
+ static ProtobufShutdown protobufShutdown; (void)protobufShutdown;
+
CV_DNN_REGISTER_LAYER_CLASS(Slice, SliceLayer);
CV_DNN_REGISTER_LAYER_CLASS(Split, SplitLayer);
CV_DNN_REGISTER_LAYER_CLASS(Concat, ConcatLayer);
From 2e17251160761d48809c80ef9ea256143b7d06a9 Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Tue, 25 Jul 2017 19:58:00 +0300
Subject: [PATCH 7/9] calib3d: fix invalid memory access
---
modules/calib3d/src/stereobm.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/modules/calib3d/src/stereobm.cpp b/modules/calib3d/src/stereobm.cpp
index f6d8213545..f6708c47d9 100644
--- a/modules/calib3d/src/stereobm.cpp
+++ b/modules/calib3d/src/stereobm.cpp
@@ -227,7 +227,7 @@ prefilterXSobel( const Mat& src, Mat& dst, int ftzero )
v_int16x8 ftz2 = v_setall_s16((short)(ftzero*2));
v_int16x8 z = v_setzero_s16();
- for(; x <= size.width-8; x += 8 )
+ for(; x <= (size.width - 1) - 8; x += 8 )
{
v_int16x8 s00 = v_reinterpret_as_s16(v_load_expand(srow0 + x + 1));
v_int16x8 s01 = v_reinterpret_as_s16(v_load_expand(srow0 + x - 1));
From caa5e3b4c5c82dac993ba2acacf0413458a7bffd Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Wed, 26 Jul 2017 13:11:31 +0300
Subject: [PATCH 8/9] imgproc: fix vectorized code of accumulate
---
modules/imgproc/src/accum.simd.hpp | 197 +++++++++++------------------
1 file changed, 73 insertions(+), 124 deletions(-)
diff --git a/modules/imgproc/src/accum.simd.hpp b/modules/imgproc/src/accum.simd.hpp
index e2be2c952e..7a29447497 100644
--- a/modules/imgproc/src/accum.simd.hpp
+++ b/modules/imgproc/src/accum.simd.hpp
@@ -425,9 +425,7 @@ void acc_simd_(const ushort* src, float* dst, const uchar* mask, int len, int cn
v_uint16x8 v_0 = v_setall_u16(0);
for ( ; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 _v_mask = v_load(mask + x);
- v_uint16x8 v_mask, dummy;
- v_expand(_v_mask, v_mask, dummy);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
v_mask = ~(v_mask == v_0);
v_uint16x8 v_src = v_load(src + x);
v_src = v_src & v_mask;
@@ -443,9 +441,7 @@ void acc_simd_(const ushort* src, float* dst, const uchar* mask, int len, int cn
v_uint16x8 v_0 = v_setall_u16(0);
for ( ; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 _v_mask = v_load(mask + x);
- v_uint16x8 v_mask, dummy;
- v_expand(_v_mask, v_mask, dummy);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
v_mask = ~(v_mask == v_0);
v_uint16x8 v_src0, v_src1, v_src2;
v_load_deinterleave(src + x * cn, v_src0, v_src1, v_src2);
@@ -491,8 +487,7 @@ void acc_simd_(const float* src, float* dst, const uchar* mask, int len, int cn)
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 v_masku16, dummy0;
- v_expand(v_load(mask + x), v_masku16, dummy0);
+ v_uint16x8 v_masku16 = v_load_expand(mask + x);
v_uint32x4 v_masku320, v_masku321;
v_expand(v_masku16, v_masku320, v_masku321);
v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_masku320 == v_reinterpret_as_u32(v_0)));
@@ -506,8 +501,7 @@ void acc_simd_(const float* src, float* dst, const uchar* mask, int len, int cn)
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 v_masku16, dummy0;
- v_expand(v_load(mask + x), v_masku16, dummy0);
+ v_uint16x8 v_masku16 = v_load_expand(mask + x);
v_uint32x4 v_masku320, v_masku321;
v_expand(v_masku16, v_masku320, v_masku321);
v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_masku320 == v_reinterpret_as_u32(v_0)));
@@ -770,8 +764,7 @@ void acc_simd_(const ushort* src, double* dst, const uchar* mask, int len, int c
{
for ( ; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 v_mask, dummy;
- v_expand(v_load(mask + x), v_mask, dummy);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
v_mask = ~(v_mask == v_0);
v_uint16x8 v_src = v_load(src + x);
v_src = v_src & v_mask;
@@ -803,8 +796,7 @@ void acc_simd_(const ushort* src, double* dst, const uchar* mask, int len, int c
{
for ( ; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 v_mask, dummy;
- v_expand(v_load(mask + x), v_mask, dummy);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
v_mask = ~(v_mask == v_0);
v_uint16x8 v_src0, v_src1, v_src2;
v_load_deinterleave(src + x * cn, v_src0, v_src1, v_src2);
@@ -871,10 +863,7 @@ void acc_simd_(const float* src, double* dst, const uchar* mask, int len, int cn
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 v_masku16, dummy0;
- v_expand(v_load(mask + x), v_masku16, dummy0);
- v_uint32x4 v_masku32, dummy1;
- v_expand(v_masku16, v_masku32, dummy1);
+ v_uint32x4 v_masku32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
v_expand(v_masku32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
@@ -892,10 +881,7 @@ void acc_simd_(const float* src, double* dst, const uchar* mask, int len, int cn
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 v_masku16, dummy0;
- v_expand(v_load(mask + x), v_masku16, dummy0);
- v_uint32x4 v_masku32, dummy1;
- v_expand(v_masku16, v_masku32, dummy1);
+ v_uint32x4 v_masku32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
v_expand(v_masku32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
@@ -947,10 +933,7 @@ void acc_simd_(const double* src, double* dst, const uchar* mask, int len, int c
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 v_masku16, dummy0;
- v_expand(v_load(mask + x), v_masku16, dummy0);
- v_uint32x4 v_masku32, dummy1;
- v_expand(v_masku16, v_masku32, dummy1);
+ v_uint32x4 v_masku32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
v_expand(v_masku32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
@@ -967,10 +950,7 @@ void acc_simd_(const double* src, double* dst, const uchar* mask, int len, int c
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 v_masku16, dummy0;
- v_expand(v_load(mask + x), v_masku16, dummy0);
- v_uint32x4 v_masku32, dummy1;
- v_expand(v_masku16, v_masku32, dummy1);
+ v_uint32x4 v_masku32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
v_expand(v_masku32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
@@ -1157,9 +1137,9 @@ void accSqr_simd_(const ushort* src, float* dst, const uchar* mask, int len, int
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
+ v_uint16x8 v_mask16 = v_load_expand(mask + x);
v_uint32x4 v_mask0, v_mask1;
- v_expand(stub, v_mask0, v_mask1);
+ v_expand(v_mask16, v_mask0, v_mask1);
v_mask0 = ~(v_mask0 == v_0);
v_mask1 = ~(v_mask1 == v_0);
v_uint16x8 v_src = v_load(src + x);
@@ -1182,9 +1162,9 @@ void accSqr_simd_(const ushort* src, float* dst, const uchar* mask, int len, int
{
for ( ; x <= len - cVectorWidth ; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
+ v_uint16x8 v_mask16 = v_load_expand(mask + x);
v_uint32x4 v_mask0, v_mask1;
- v_expand(stub, v_mask0, v_mask1);
+ v_expand(v_mask16, v_mask0, v_mask1);
v_mask0 = ~(v_mask0 == v_0);
v_mask1 = ~(v_mask1 == v_0);
@@ -1254,11 +1234,11 @@ void accSqr_simd_(const float* src, float* dst, const uchar* mask, int len, int
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 v_stub = v_load_expand(mask + x);
- v_uint32x4 v_stub0, v_stub1;
- v_expand(v_stub, v_stub0, v_stub1);
- v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_stub0 == v_0));
- v_float32x4 v_mask1 = v_reinterpret_as_f32(~(v_stub1 == v_0));
+ v_uint16x8 v_mask16 = v_load_expand(mask + x);
+ v_uint32x4 v_mask_0, v_mask_1;
+ v_expand(v_mask16, v_mask_0, v_mask_1);
+ v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_mask_0 == v_0));
+ v_float32x4 v_mask1 = v_reinterpret_as_f32(~(v_mask_1 == v_0));
v_float32x4 v_src0 = v_load(src + x);
v_float32x4 v_src1 = v_load(src + x + 4);
v_src0 = v_src0 & v_mask0;
@@ -1274,11 +1254,11 @@ void accSqr_simd_(const float* src, float* dst, const uchar* mask, int len, int
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 v_stub = v_load_expand(mask + x);
- v_uint32x4 v_stub0, v_stub1;
- v_expand(v_stub, v_stub0, v_stub1);
- v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_stub0 == v_0));
- v_float32x4 v_mask1 = v_reinterpret_as_f32(~(v_stub1 == v_0));
+ v_uint16x8 v_mask16 = v_load_expand(mask + x);
+ v_uint32x4 v_mask_0, v_mask_1;
+ v_expand(v_mask16, v_mask_0, v_mask_1);
+ v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_mask_0 == v_0));
+ v_float32x4 v_mask1 = v_reinterpret_as_f32(~(v_mask_1 == v_0));
v_float32x4 v_src00, v_src10, v_src20, v_src01, v_src11, v_src21;
v_load_deinterleave(src + x * cn, v_src00, v_src10, v_src20);
@@ -1319,9 +1299,7 @@ void accSqr_simd_(const uchar* src, double* dst, const uchar* mask, int len, int
int size = len * cn;
for (; x <= size - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 v_src = v_load(src + x);
- v_uint16x8 v_int, dummy;
- v_expand(v_src, v_int, dummy);
+ v_uint16x8 v_int = v_load_expand(src + x);
v_uint32x4 v_int0, v_int1;
v_expand(v_int, v_int0, v_int1);
@@ -1353,17 +1331,15 @@ void accSqr_simd_(const uchar* src, double* dst, const uchar* mask, int len, int
}
else
{
- v_uint8x16 v_0 = v_setzero_u8();
+ v_uint16x8 v_0 = v_setzero_u16();
if (cn == 1)
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 v_mask = v_load(mask + x);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
v_mask = ~(v_mask == v_0);
- v_uint8x16 v_src = v_load(src + x);
- v_src = v_src & v_mask;
- v_uint16x8 v_int, dummy;
- v_expand(v_src, v_int, dummy);
+ v_uint16x8 v_src = v_load_expand(src + x);
+ v_uint16x8 v_int = v_src & v_mask;
v_uint32x4 v_int0, v_int1;
v_expand(v_int, v_int0, v_int1);
@@ -1395,19 +1371,19 @@ void accSqr_simd_(const uchar* src, double* dst, const uchar* mask, int len, int
}
else if (cn == 3)
{
- for (; x <= len - cVectorWidth; x += cVectorWidth)
+ for (; x <= len - /*cVectorWidth*/16; x += cVectorWidth)
{
- v_uint8x16 v_mask = v_load(mask + x);
- v_mask = ~(v_mask == v_0);
v_uint8x16 v_src0, v_src1, v_src2;
v_load_deinterleave(src + x * cn, v_src0, v_src1, v_src2);
- v_src0 = v_src0 & v_mask;
- v_src1 = v_src1 & v_mask;
- v_src2 = v_src2 & v_mask;
v_uint16x8 v_int0, v_int1, v_int2, dummy;
v_expand(v_src0, v_int0, dummy);
v_expand(v_src1, v_int1, dummy);
v_expand(v_src2, v_int2, dummy);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
+ v_mask = ~(v_mask == v_0);
+ v_int0 = v_int0 & v_mask;
+ v_int1 = v_int1 & v_mask;
+ v_int2 = v_int2 & v_mask;
v_uint32x4 v_int00, v_int01, v_int10, v_int11, v_int20, v_int21;
v_expand(v_int0, v_int00, v_int01);
@@ -1627,9 +1603,7 @@ void accSqr_simd_(const float* src, double* dst, const uchar* mask, int len, int
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 v_mask, dummy;
- v_expand(stub, v_mask, dummy);
+ v_uint32x4 v_mask = v_load_expand_q(mask + x);;
v_mask = ~(v_mask == v_0);
v_float32x4 v_src = v_load(src + x);
v_src = v_src & v_reinterpret_as_f32(v_mask);
@@ -1646,9 +1620,7 @@ void accSqr_simd_(const float* src, double* dst, const uchar* mask, int len, int
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 v_mask, dummy;
- v_expand(stub, v_mask, dummy);
+ v_uint32x4 v_mask = v_load_expand_q(mask + x);
v_mask = ~(v_mask == v_0);
v_float32x4 v_src0, v_src1, v_src2;
@@ -1709,11 +1681,9 @@ void accSqr_simd_(const double* src, double* dst, const uchar* mask, int len, in
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 stub0, stub1;
- v_expand(stub, stub0, stub1);
+ v_uint32x4 v_mask32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
- v_expand(stub0, v_masku640, v_masku641);
+ v_expand(v_mask32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
v_float64x2 v_mask1 = v_reinterpret_as_f64(~(v_masku641 == v_0));
v_float64x2 v_src0 = v_load(src + x);
@@ -1731,11 +1701,9 @@ void accSqr_simd_(const double* src, double* dst, const uchar* mask, int len, in
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 stub0, stub1;
- v_expand(stub, stub0, stub1);
+ v_uint32x4 v_mask32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
- v_expand(stub0, v_masku640, v_masku641);
+ v_expand(v_mask32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
v_float64x2 v_mask1 = v_reinterpret_as_f64(~(v_masku641 == v_0));
@@ -2059,11 +2027,10 @@ void accProd_simd_(const float* src1, const float* src2, float* dst, const uchar
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 stub0, stub1;
- v_expand(stub, stub0, stub1);
- v_float32x4 v_mask0 = v_reinterpret_as_f32(~(stub0 == v_0));
- v_float32x4 v_mask1 = v_reinterpret_as_f32(~(stub1 == v_0));
+ v_uint32x4 v_mask32_0 = v_load_expand_q(mask + x);
+ v_uint32x4 v_mask32_1 = v_load_expand_q(mask + x + 4);
+ v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_mask32_0 == v_0));
+ v_float32x4 v_mask1 = v_reinterpret_as_f32(~(v_mask32_1 == v_0));
v_store(dst + x, v_load(dst + x) + ((v_load(src1 + x) * v_load(src2 + x)) & v_mask0));
v_store(dst + x + 4, v_load(dst + x + 4) + ((v_load(src1 + x + 4) * v_load(src2 + x + 4)) & v_mask1));
@@ -2073,11 +2040,10 @@ void accProd_simd_(const float* src1, const float* src2, float* dst, const uchar
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 stub0, stub1;
- v_expand(stub, stub0, stub1);
- v_float32x4 v_mask0 = v_reinterpret_as_f32(~(stub0 == v_0));
- v_float32x4 v_mask1 = v_reinterpret_as_f32(~(stub1 == v_0));
+ v_uint32x4 v_mask32_0 = v_load_expand_q(mask + x);
+ v_uint32x4 v_mask32_1 = v_load_expand_q(mask + x + 4);
+ v_float32x4 v_mask0 = v_reinterpret_as_f32(~(v_mask32_0 == v_0));
+ v_float32x4 v_mask1 = v_reinterpret_as_f32(~(v_mask32_1 == v_0));
v_float32x4 v_1src00, v_1src01, v_1src10, v_1src11, v_1src20, v_1src21;
v_float32x4 v_2src00, v_2src01, v_2src10, v_2src11, v_2src20, v_2src21;
@@ -2109,12 +2075,8 @@ void accProd_simd_(const uchar* src1, const uchar* src2, double* dst, const ucha
int size = len * cn;
for (; x <= size - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 v_1src = v_load(src1 + x);
- v_uint8x16 v_2src = v_load(src2 + x);
-
- v_uint16x8 v_1int, v_2int, dummy;
- v_expand(v_1src, v_1int, dummy);
- v_expand(v_2src, v_2int, dummy);
+ v_uint16x8 v_1int = v_load_expand(src1 + x);
+ v_uint16x8 v_2int = v_load_expand(src2 + x);
v_uint32x4 v_1int_0, v_1int_1, v_2int_0, v_2int_1;
v_expand(v_1int, v_1int_0, v_1int_1);
@@ -2148,19 +2110,15 @@ void accProd_simd_(const uchar* src1, const uchar* src2, double* dst, const ucha
}
else
{
- v_uint8x16 v_0 = v_setzero_u8();
+ v_uint16x8 v_0 = v_setzero_u16();
if (cn == 1)
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 v_mask = v_load(mask + x);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
v_mask = ~(v_mask == v_0);
- v_uint8x16 v_1src = v_load(src1 + x) & v_mask;
- v_uint8x16 v_2src = v_load(src2 + x) & v_mask;
-
- v_uint16x8 v_1int, v_2int, dummy;
- v_expand(v_1src, v_1int, dummy);
- v_expand(v_2src, v_2int, dummy);
+ v_uint16x8 v_1int = v_load_expand(src1 + x) & v_mask;
+ v_uint16x8 v_2int = v_load_expand(src2 + x) & v_mask;
v_uint32x4 v_1int_0, v_1int_1, v_2int_0, v_2int_1;
v_expand(v_1int, v_1int_0, v_1int_1);
@@ -2194,19 +2152,11 @@ void accProd_simd_(const uchar* src1, const uchar* src2, double* dst, const ucha
}
else if (cn == 3)
{
- for (; x <= len - cVectorWidth; x += cVectorWidth)
+ for (; x <= len - /*cVectorWidth*/16; x += cVectorWidth)
{
- v_uint8x16 v_mask = v_load(mask + x);
- v_mask = ~(v_mask == v_0);
v_uint8x16 v_1src0, v_1src1, v_1src2, v_2src0, v_2src1, v_2src2;
v_load_deinterleave(src1 + x * cn, v_1src0, v_1src1, v_1src2);
v_load_deinterleave(src2 + x * cn, v_2src0, v_2src1, v_2src2);
- v_1src0 = v_1src0 & v_mask;
- v_1src1 = v_1src1 & v_mask;
- v_1src2 = v_1src2 & v_mask;
- v_2src0 = v_2src0 & v_mask;
- v_2src1 = v_2src1 & v_mask;
- v_2src2 = v_2src2 & v_mask;
v_uint16x8 v_1int0, v_1int1, v_1int2, v_2int0, v_2int1, v_2int2, dummy;
v_expand(v_1src0, v_1int0, dummy);
@@ -2216,6 +2166,15 @@ void accProd_simd_(const uchar* src1, const uchar* src2, double* dst, const ucha
v_expand(v_2src1, v_2int1, dummy);
v_expand(v_2src2, v_2int2, dummy);
+ v_uint16x8 v_mask = v_load_expand(mask + x);
+ v_mask = ~(v_mask == v_0);
+ v_1int0 = v_1int0 & v_mask;
+ v_1int1 = v_1int1 & v_mask;
+ v_1int2 = v_1int2 & v_mask;
+ v_2int0 = v_2int0 & v_mask;
+ v_2int1 = v_2int1 & v_mask;
+ v_2int2 = v_2int2 & v_mask;
+
v_uint32x4 v_1int00, v_1int01, v_1int10, v_1int11, v_1int20, v_1int21;
v_uint32x4 v_2int00, v_2int01, v_2int10, v_2int11, v_2int20, v_2int21;
v_expand(v_1int0, v_1int00, v_1int01);
@@ -2440,9 +2399,7 @@ void accProd_simd_(const float* src1, const float* src2, double* dst, const ucha
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 v_mask, dummy;
- v_expand(stub, v_mask, dummy);
+ v_uint32x4 v_mask = v_load_expand_q(mask + x);
v_mask = ~(v_mask == v_0);
v_float32x4 v_1src = v_load(src1 + x);
v_float32x4 v_2src = v_load(src2 + x);
@@ -2462,9 +2419,7 @@ void accProd_simd_(const float* src1, const float* src2, double* dst, const ucha
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 v_mask, dummy;
- v_expand(stub, v_mask, dummy);
+ v_uint32x4 v_mask = v_load_expand_q(mask + x);
v_mask = ~(v_mask == v_0);
v_float32x4 v_1src0, v_1src1, v_1src2, v_2src0, v_2src1, v_2src2;
v_load_deinterleave(src1 + x * cn, v_1src0, v_1src1, v_1src2);
@@ -2522,11 +2477,9 @@ void accProd_simd_(const double* src1, const double* src2, double* dst, const uc
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 stub0, stub1;
- v_expand(stub, stub0, stub1);
+ v_uint32x4 v_mask32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
- v_expand(stub0, v_masku640, v_masku641);
+ v_expand(v_mask32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
v_float64x2 v_mask1 = v_reinterpret_as_f64(~(v_masku641 == v_0));
@@ -2543,11 +2496,9 @@ void accProd_simd_(const double* src1, const double* src2, double* dst, const uc
{
for (; x <= len - cVectorWidth; x += cVectorWidth)
{
- v_uint16x8 stub = v_load_expand(mask + x);
- v_uint32x4 stub0, stub1;
- v_expand(stub, stub0, stub1);
+ v_uint32x4 v_mask32 = v_load_expand_q(mask + x);
v_uint64x2 v_masku640, v_masku641;
- v_expand(stub0, v_masku640, v_masku641);
+ v_expand(v_mask32, v_masku640, v_masku641);
v_float64x2 v_mask0 = v_reinterpret_as_f64(~(v_masku640 == v_0));
v_float64x2 v_mask1 = v_reinterpret_as_f64(~(v_masku641 == v_0));
@@ -2704,12 +2655,10 @@ void accW_simd_(const uchar* src, double* dst, const uchar* mask, int len, int c
int size = len * cn;
for (; x <= size - cVectorWidth; x += cVectorWidth)
{
- v_uint8x16 v_src = v_load(src + x);
- v_uint16x8 v_int, dummy;
- v_expand(v_src, v_int, dummy);
+ v_uint16x8 v_src16 = v_load_expand(src + x);
v_uint32x4 v_int_0, v_int_1;
- v_expand(v_int, v_int_0, v_int_1);
+ v_expand(v_src16, v_int_0, v_int_1);
v_int32x4 v_int0 = v_reinterpret_as_s32(v_int_0);
v_int32x4 v_int1 = v_reinterpret_as_s32(v_int_1);
From 12213f998565c7e34756b9e9f2f4f43e56985181 Mon Sep 17 00:00:00 2001
From: Alexander Alekhin <alexander.alekhin@intel.com>
Date: Wed, 26 Jul 2017 18:16:39 +0300
Subject: [PATCH 9/9] flann: fix out of buffer access
---
.../flann/include/opencv2/flann/lsh_table.h | 25 +++++++++++++++----
1 file changed, 20 insertions(+), 5 deletions(-)
diff --git a/modules/flann/include/opencv2/flann/lsh_table.h b/modules/flann/include/opencv2/flann/lsh_table.h
index 0907dc9fa1..fa7cc71486 100644
--- a/modules/flann/include/opencv2/flann/lsh_table.h
+++ b/modules/flann/include/opencv2/flann/lsh_table.h
@@ -147,6 +147,7 @@ public:
LshTable()
{
key_size_ = 0;
+ feature_size_ = 0;
speed_level_ = kArray;
}
@@ -157,7 +158,7 @@ public:
*/
LshTable(unsigned int feature_size, unsigned int key_size)
{
- (void)feature_size;
+ feature_size_ = feature_size;
(void)key_size;
std::cerr << "LSH is not implemented for that type" << std::endl;
assert(0);
@@ -332,6 +333,8 @@ private:
*/
unsigned int key_size_;
+ unsigned int feature_size_;
+
// Members only used for the unsigned char specialization
/** The mask to apply to a feature to get the hash key
* Only used in the unsigned char case
@@ -345,9 +348,10 @@ private:
template<>
inline LshTable<unsigned char>::LshTable(unsigned int feature_size, unsigned int subsignature_size)
{
+ feature_size_ = feature_size;
initialize(subsignature_size);
// Allocate the mask
- mask_ = std::vector<size_t>((size_t)ceil((float)(feature_size * sizeof(char)) / (float)sizeof(size_t)), 0);
+ mask_ = std::vector<size_t>((feature_size * sizeof(char) + sizeof(size_t) - 1) / sizeof(size_t), 0);
// A bit brutal but fast to code
std::vector<size_t> indices(feature_size * CHAR_BIT);
@@ -392,6 +396,7 @@ inline size_t LshTable<unsigned char>::getKey(const unsigned char* feature) cons
{
// no need to check if T is dividable by sizeof(size_t) like in the Hamming
// distance computation as we have a mask
+ // FIXIT: This is bad assumption, because we reading tail bytes after of the allocated features buffer
const size_t* feature_block_ptr = reinterpret_cast<const size_t*> ((const void*)feature);
// Figure out the subsignature of the feature
@@ -400,10 +405,20 @@ inline size_t LshTable<unsigned char>::getKey(const unsigned char* feature) cons
size_t subsignature = 0;
size_t bit_index = 1;
- for (std::vector<size_t>::const_iterator pmask_block = mask_.begin(); pmask_block != mask_.end(); ++pmask_block) {
+ for (unsigned i = 0; i < feature_size_; i += sizeof(size_t)) {
// get the mask and signature blocks
- size_t feature_block = *feature_block_ptr;
- size_t mask_block = *pmask_block;
+ size_t feature_block;
+ if (i <= feature_size_ - sizeof(size_t))
+ {
+ feature_block = *feature_block_ptr;
+ }
+ else
+ {
+ size_t tmp = 0;
+ memcpy(&tmp, feature_block_ptr, feature_size_ - i); // preserve bytes order
+ feature_block = tmp;
+ }
+ size_t mask_block = mask_[i / sizeof(size_t)];
while (mask_block) {
// Get the lowest set bit in the mask block
size_t lowest_bit = mask_block & (-(ptrdiff_t)mask_block);