mirror of
https://github.com/opencv/opencv.git
synced 2026-07-31 00:03:03 +04:00
--perf_instrument parameter now has int type and 0, 1, 2 modes (1 - simple trees, 2 - expanded trees for functions with same name but different calling address);
Maximum depth limit var was added to the instrumentation structure; Trace names output console output fix: improper tree formatting could happen; Output in case of error was added; Custom regions improvements; Improved timing and weight calculation for parallel regions; New TC (threads counter) value to indicate how many different threads accessed particular node; parallel_for, warnings fixes and ReturnAddress code from Alexander Alekhin;
This commit is contained in:
Pavel Vlasov
@@ -457,10 +457,11 @@ class InstrStruct
|
||||
public:
|
||||
InstrStruct()
|
||||
{
|
||||
useInstr = false;
|
||||
enableMapping = true;
|
||||
useInstr = false;
|
||||
flags = FLAGS_MAPPING;
|
||||
maxDepth = 0;
|
||||
|
||||
rootNode.m_payload = NodeData("ROOT", NULL, 0, TYPE_GENERAL, IMPL_PLAIN);
|
||||
rootNode.m_payload = NodeData("ROOT", NULL, 0, NULL, false, TYPE_GENERAL, IMPL_PLAIN);
|
||||
tlsStruct.get()->pCurrentNode = &rootNode;
|
||||
}
|
||||
|
||||
@@ -468,7 +469,8 @@ public:
|
||||
Mutex mutexCount;
|
||||
|
||||
bool useInstr;
|
||||
bool enableMapping;
|
||||
int flags;
|
||||
int maxDepth;
|
||||
InstrNode rootNode;
|
||||
TLSData<InstrTLSStruct> tlsStruct;
|
||||
};
|
||||
@@ -476,7 +478,7 @@ public:
|
||||
class CV_EXPORTS IntrumentationRegion
|
||||
{
|
||||
public:
|
||||
IntrumentationRegion(const char* funName, const char* fileName, int lineNum, TYPE instrType = TYPE_GENERAL, IMPL implType = IMPL_PLAIN);
|
||||
IntrumentationRegion(const char* funName, const char* fileName, int lineNum, void *retAddress, bool alwaysExpand, TYPE instrType = TYPE_GENERAL, IMPL implType = IMPL_PLAIN);
|
||||
~IntrumentationRegion();
|
||||
|
||||
private:
|
||||
@@ -484,20 +486,28 @@ private:
|
||||
uint64 m_regionTicks;
|
||||
};
|
||||
|
||||
InstrStruct& getInstrumentStruct();
|
||||
InstrTLSStruct& getInstrumentTLSStruct();
|
||||
CV_EXPORTS InstrNode* getCurrentNode();
|
||||
CV_EXPORTS InstrStruct& getInstrumentStruct();
|
||||
InstrTLSStruct& getInstrumentTLSStruct();
|
||||
CV_EXPORTS InstrNode* getCurrentNode();
|
||||
}
|
||||
}
|
||||
|
||||
///// General instrumentation
|
||||
#ifdef _WIN32
|
||||
#define CV_INSTRUMENT_GET_RETURN_ADDRESS _ReturnAddress()
|
||||
#else
|
||||
#define CV_INSTRUMENT_GET_RETURN_ADDRESS __builtin_extract_return_addr(__builtin_return_address(0))
|
||||
#endif
|
||||
|
||||
// Instrument region
|
||||
#define CV_INSTRUMENT_REGION_META(NAME, TYPE, IMPL) ::cv::instr::IntrumentationRegion __instr_region__(NAME, __FILE__, __LINE__, TYPE, IMPL);
|
||||
#define CV_INSTRUMENT_REGION_META(NAME, ALWAYS_EXPAND, TYPE, IMPL) ::cv::instr::IntrumentationRegion __instr_region__(NAME, __FILE__, __LINE__, CV_INSTRUMENT_GET_RETURN_ADDRESS, ALWAYS_EXPAND, TYPE, IMPL);
|
||||
#define CV_INSTRUMENT_REGION_CUSTOM_META(NAME, ALWAYS_EXPAND, TYPE, IMPL)\
|
||||
void *__curr_address__ = [&]() {return CV_INSTRUMENT_GET_RETURN_ADDRESS;}();\
|
||||
::cv::instr::IntrumentationRegion __instr_region__(NAME, __FILE__, __LINE__, __curr_address__, false, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN);
|
||||
// Instrument functions with non-void return type
|
||||
#define CV_INSTRUMENT_FUN_RT_META(TYPE, IMPL, ERROR_COND, FUN, ...) ([&]()\
|
||||
{\
|
||||
if(::cv::instr::useInstrumentation()){\
|
||||
::cv::instr::IntrumentationRegion __instr__(#FUN, __FILE__, __LINE__, TYPE, IMPL);\
|
||||
::cv::instr::IntrumentationRegion __instr__(#FUN, __FILE__, __LINE__, NULL, false, TYPE, IMPL);\
|
||||
try{\
|
||||
auto status = ((FUN)(__VA_ARGS__));\
|
||||
if(ERROR_COND){\
|
||||
@@ -518,7 +528,7 @@ CV_EXPORTS InstrNode* getCurrentNode();
|
||||
#define CV_INSTRUMENT_FUN_RV_META(TYPE, IMPL, FUN, ...) ([&]()\
|
||||
{\
|
||||
if(::cv::instr::useInstrumentation()){\
|
||||
::cv::instr::IntrumentationRegion __instr__(#FUN, __FILE__, __LINE__, TYPE, IMPL);\
|
||||
::cv::instr::IntrumentationRegion __instr__(#FUN, __FILE__, __LINE__, NULL, false, TYPE, IMPL);\
|
||||
try{\
|
||||
(FUN)(__VA_ARGS__);\
|
||||
}catch(...){\
|
||||
@@ -531,17 +541,19 @@ CV_EXPORTS InstrNode* getCurrentNode();
|
||||
}\
|
||||
}())
|
||||
// Instrumentation information marker
|
||||
#define CV_INSTRUMENT_MARK_META(IMPL, NAME, ...) {::cv::instr::IntrumentationRegion __instr_mark__(NAME, __FILE__, __LINE__, ::cv::instr::TYPE_MARKER, IMPL);}
|
||||
#define CV_INSTRUMENT_MARK_META(IMPL, NAME, ...) {::cv::instr::IntrumentationRegion __instr_mark__(NAME, __FILE__, __LINE__, NULL, false, ::cv::instr::TYPE_MARKER, IMPL);}
|
||||
|
||||
///// General instrumentation
|
||||
// General OpenCV region instrumentation macro
|
||||
#define CV_INSTRUMENT_REGION() CV_INSTRUMENT_REGION_META(__FUNCTION__, cv::instr::TYPE_GENERAL, cv::instr::IMPL_PLAIN)
|
||||
// Parallel OpenCV region instrumentation macro
|
||||
#define CV_INSTRUMENT_REGION_MT() CV_INSTRUMENT_REGION_MT_META(cv::instr::TYPE_GENERAL, cv::instr::IMPL_PLAIN)
|
||||
#define CV_INSTRUMENT_REGION() CV_INSTRUMENT_REGION_META(__FUNCTION__, false, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN)
|
||||
// Custom OpenCV region instrumentation macro
|
||||
#define CV_INSTRUMENT_REGION_NAME(NAME) CV_INSTRUMENT_REGION_CUSTOM_META(NAME, false, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN)
|
||||
// Instrumentation for parallel_for_ or other regions which forks and gathers threads
|
||||
#define CV_INSTRUMENT_REGION_MT_FORK() CV_INSTRUMENT_REGION_META(__FUNCTION__, true, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN);
|
||||
|
||||
///// IPP instrumentation
|
||||
// Wrapper region instrumentation macro
|
||||
#define CV_INSTRUMENT_REGION_IPP() CV_INSTRUMENT_REGION_META(__FUNCTION__, ::cv::instr::TYPE_WRAPPER, ::cv::instr::IMPL_IPP)
|
||||
#define CV_INSTRUMENT_REGION_IPP() CV_INSTRUMENT_REGION_META(__FUNCTION__, false, ::cv::instr::TYPE_WRAPPER, ::cv::instr::IMPL_IPP)
|
||||
// Function instrumentation macro
|
||||
#define CV_INSTRUMENT_FUN_IPP(FUN, ...) CV_INSTRUMENT_FUN_RT_META(::cv::instr::TYPE_FUN, ::cv::instr::IMPL_IPP, status < 0, FUN, __VA_ARGS__)
|
||||
// Diagnostic markers
|
||||
@@ -549,26 +561,28 @@ CV_EXPORTS InstrNode* getCurrentNode();
|
||||
|
||||
///// OpenCL instrumentation
|
||||
// Wrapper region instrumentation macro
|
||||
#define CV_INSTRUMENT_REGION_OPENCL() CV_INSTRUMENT_REGION_META(__FUNCTION__, ::cv::instr::TYPE_WRAPPER, ::cv::instr::IMPL_OPENCL)
|
||||
#define CV_INSTRUMENT_REGION_OPENCL_(NAME) CV_INSTRUMENT_REGION_META(NAME, ::cv::instr::TYPE_WRAPPER, ::cv::instr::IMPL_OPENCL)
|
||||
// Function instrumentation macro
|
||||
#define CV_INSTRUMENT_FUN_OPENCL_KERNEL(FUN, ...) CV_INSTRUMENT_FUN_RT_META(::cv::instr::TYPE_FUN, ::cv::instr::IMPL_OPENCL, status == 0, FUN, __VA_ARGS__)
|
||||
#define CV_INSTRUMENT_REGION_OPENCL() CV_INSTRUMENT_REGION_META(__FUNCTION__, false, ::cv::instr::TYPE_WRAPPER, ::cv::instr::IMPL_OPENCL)
|
||||
// OpenCL kernel compilation wrapper
|
||||
#define CV_INSTRUMENT_REGION_OPENCL_COMPILE(NAME) CV_INSTRUMENT_REGION_META(NAME, false, ::cv::instr::TYPE_WRAPPER, ::cv::instr::IMPL_OPENCL)
|
||||
// OpenCL kernel run wrapper
|
||||
#define CV_INSTRUMENT_REGION_OPENCL_RUN(NAME) CV_INSTRUMENT_REGION_META(NAME, false, ::cv::instr::TYPE_FUN, ::cv::instr::IMPL_OPENCL)
|
||||
// Diagnostic markers
|
||||
#define CV_INSTRUMENT_MARK_OPENCL(NAME) CV_INSTRUMENT_MARK_META(::cv::instr::IMPL_OPENCL, NAME)
|
||||
#else
|
||||
#define CV_INSTRUMENT_REGION_META(...)
|
||||
|
||||
#define CV_INSTRUMENT_REGION()
|
||||
#define CV_INSTRUMENT_REGION_MT()
|
||||
#define CV_INSTRUMENT_REGION_NAME(...)
|
||||
#define CV_INSTRUMENT_REGION_MT_FORK()
|
||||
|
||||
#define CV_INSTRUMENT_REGION_IPP()
|
||||
#define CV_INSTRUMENT_FUN_IPP(FUN, ...) ((FUN)(__VA_ARGS__))
|
||||
#define CV_INSTRUMENT_MARK_IPP(NAME)
|
||||
#define CV_INSTRUMENT_MARK_IPP(...)
|
||||
|
||||
#define CV_INSTRUMENT_REGION_OPENCL()
|
||||
#define CV_INSTRUMENT_REGION_OPENCL_(...)
|
||||
#define CV_INSTRUMENT_FUN_OPENCL_KERNEL(FUN, ...) ((FUN)(__VA_ARGS__))
|
||||
#define CV_INSTRUMENT_MARK_OPENCL(NAME)
|
||||
#define CV_INSTRUMENT_REGION_OPENCL_COMPILE(...)
|
||||
#define CV_INSTRUMENT_REGION_OPENCL_RUN(...)
|
||||
#define CV_INSTRUMENT_MARK_OPENCL(...)
|
||||
#endif
|
||||
|
||||
//! @endcond
|
||||
|
||||
@@ -1029,7 +1029,7 @@ public:
|
||||
|
||||
Node<OBJECT>* findChild(OBJECT& payload) const
|
||||
{
|
||||
for(int i = 0; i < this->m_childs.size(); i++)
|
||||
for(size_t i = 0; i < this->m_childs.size(); i++)
|
||||
{
|
||||
if(this->m_childs[i]->m_payload == payload)
|
||||
return this->m_childs[i];
|
||||
@@ -1039,10 +1039,10 @@ public:
|
||||
|
||||
int findChild(Node<OBJECT> *pNode) const
|
||||
{
|
||||
for (int i = 0; i < this->m_childs.size(); i++)
|
||||
for (size_t i = 0; i < this->m_childs.size(); i++)
|
||||
{
|
||||
if(this->m_childs[i] == pNode)
|
||||
return i;
|
||||
return (int)i;
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
@@ -1059,7 +1059,7 @@ public:
|
||||
|
||||
void removeChilds()
|
||||
{
|
||||
for(int i = 0; i < m_childs.size(); i++)
|
||||
for(size_t i = 0; i < m_childs.size(); i++)
|
||||
{
|
||||
m_childs[i]->m_pParent = 0; // avoid excessive parent vector trimming
|
||||
delete m_childs[i];
|
||||
@@ -1067,6 +1067,14 @@ public:
|
||||
m_childs.clear();
|
||||
}
|
||||
|
||||
int getDepth()
|
||||
{
|
||||
int count = 0;
|
||||
Node *pParent = m_pParent;
|
||||
while(pParent) count++, pParent = pParent->m_pParent;
|
||||
return count;
|
||||
}
|
||||
|
||||
public:
|
||||
OBJECT m_payload;
|
||||
Node<OBJECT>* m_pParent;
|
||||
@@ -1094,10 +1102,19 @@ enum IMPL
|
||||
IMPL_OPENCL,
|
||||
};
|
||||
|
||||
struct NodeDataTls
|
||||
{
|
||||
NodeDataTls()
|
||||
{
|
||||
m_ticksTotal = 0;
|
||||
}
|
||||
uint64 m_ticksTotal;
|
||||
};
|
||||
|
||||
class CV_EXPORTS NodeData
|
||||
{
|
||||
public:
|
||||
NodeData(const char* funName = 0, const char* fileName = NULL, int lineNum = 0, cv::instr::TYPE instrType = TYPE_GENERAL, cv::instr::IMPL implType = IMPL_PLAIN);
|
||||
NodeData(const char* funName = 0, const char* fileName = NULL, int lineNum = 0, void* retAddress = NULL, bool alwaysExpand = false, cv::instr::TYPE instrType = TYPE_GENERAL, cv::instr::IMPL implType = IMPL_PLAIN);
|
||||
NodeData(NodeData &ref);
|
||||
~NodeData();
|
||||
NodeData& operator=(const NodeData&);
|
||||
@@ -1107,17 +1124,18 @@ public:
|
||||
cv::instr::IMPL m_implType;
|
||||
const char* m_fileName;
|
||||
int m_lineNum;
|
||||
|
||||
volatile int m_counter;
|
||||
volatile uint64 m_ticksTotal;
|
||||
|
||||
// No synchronization
|
||||
double getTotalMs() const { return (double)m_ticksTotal * 1000. / cv::getTickFrequency(); }
|
||||
// No synchronization
|
||||
double getMeanMs() const { return (double)m_ticksTotal * 1000. / (m_counter * cv::getTickFrequency()); }
|
||||
|
||||
void* m_retAddress;
|
||||
bool m_alwaysExpand;
|
||||
bool m_funError;
|
||||
bool m_stopPoint;
|
||||
|
||||
volatile int m_counter;
|
||||
volatile uint64 m_ticksTotal;
|
||||
TLSData<NodeDataTls> m_tls;
|
||||
int m_threads;
|
||||
|
||||
// No synchronization
|
||||
double getTotalMs() const { return ((double)m_ticksTotal / cv::getTickFrequency()) * 1000; }
|
||||
double getMeanMs() const { return (((double)m_ticksTotal/m_counter) / cv::getTickFrequency()) * 1000; }
|
||||
};
|
||||
bool operator==(const NodeData& lhs, const NodeData& rhs);
|
||||
|
||||
@@ -1134,8 +1152,9 @@ CV_EXPORTS void resetTrace();
|
||||
|
||||
enum FLAGS
|
||||
{
|
||||
FLAGS_NONE = 0,
|
||||
FLAGS_MAPPING = 1 << 0,
|
||||
FLAGS_NONE = 0,
|
||||
FLAGS_MAPPING = 0x01,
|
||||
FLAGS_EXPAND_SAME_NAMES = 0x02,
|
||||
};
|
||||
|
||||
CV_EXPORTS void setFlags(FLAGS modeFlags);
|
||||
|
||||
Reference in New Issue
Block a user