zhangengru/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2025-03-20 07:40:02 +08:00
Code Issues Projects Wiki Activity

Merge remote-tracking branch 'origin/master' into hw

Browse Source
This commit is contained in:
Bartosz Taudul
2021-06-02 01:12:28 +02:00
parent 688a9722c7 f4d80a4f5f
commit b7c5939bb8
21 changed files with 112 additions and 104 deletions
Download Patch File Download Diff File Expand all files Collapse all files
client/TracyAlloc.cpp
+35
View File
@@ -0,0 +1,35 @@
#ifdef TRACY_ENABLE
#include <atomic>
#include "../common/TracyAlloc.hpp"
#include "../common/TracyYield.hpp"
namespace tracy
{
extern std::atomic<int> RpInitDone;
extern std::atomic<int> RpInitLock;
TRACY_API void InitRpmallocPlumbing()
{
const auto done = RpInitDone.load( std::memory_order_acquire );
if( !done )
{
int expected = 0;
while( !RpInitLock.compare_exchange_weak( expected, 1, std::memory_order_release, std::memory_order_relaxed ) ) { expected = 0; YieldThread(); }
const auto done = RpInitDone.load( std::memory_order_acquire );
if( !done )
{
rpmalloc_initialize();
RpInitDone.store( 1, std::memory_order_release );
}
RpInitLock.store( 0, std::memory_order_release );
}
rpmalloc_thread_initialize();
RpThreadInitDone = true;
}
}
#endif
client/TracyProfiler.cpp
+9 -70
View File
@@ -64,6 +64,7 @@
#include "../common/TracyAlign.hpp"
#include "../common/TracySocket.hpp"
#include "../common/TracySystem.hpp"
#include "../common/TracyYield.hpp"
#include "../common/tracy_lz4.hpp"
#include "tracy_rpmalloc.hpp"
#include "TracyCallstack.hpp"
@@ -118,45 +119,6 @@ extern "C" typedef BOOL (WINAPI *t_GetLogicalProcessorInformationEx)( LOGICAL_PR
namespace tracy
{
namespace
{
# if ( defined _WIN32 || defined __CYGWIN__ ) && _WIN32_WINNT >= _WIN32_WINNT_VISTA
BOOL CALLBACK InitOnceCallback( PINIT_ONCE /*initOnce*/, PVOID /*Parameter*/, PVOID* /*Context*/)
{
rpmalloc_initialize();
return TRUE;
}
INIT_ONCE InitOnce = INIT_ONCE_STATIC_INIT;
# elif defined __linux__
void InitOnceCallback()
{
rpmalloc_initialize();
}
pthread_once_t once_control = PTHREAD_ONCE_INIT;
# else
void InitOnceCallback()
{
rpmalloc_initialize();
}
std::once_flag once_flag;
# endif
}
struct RPMallocInit
{
RPMallocInit()
{
# if ( defined _WIN32 || defined __CYGWIN__ ) && _WIN32_WINNT >= _WIN32_WINNT_VISTA
InitOnceExecuteOnce( &InitOnce, InitOnceCallback, nullptr, nullptr );
# elif defined __linux__
pthread_once( &once_control, InitOnceCallback );
# else
std::call_once( once_flag, InitOnceCallback );
# endif
rpmalloc_thread_initialize();
}
};
#ifndef TRACY_DELAYED_INIT
struct InitTimeWrapper
@@ -964,12 +926,6 @@ TRACY_API int64_t GetFrequencyQpc()
#ifdef TRACY_DELAYED_INIT
struct ThreadNameData;
TRACY_API moodycamel::ConcurrentQueue<QueueItem>& GetQueue();
TRACY_API void InitRPMallocThread();
void InitRPMallocThread()
{
RPMallocInit rpinit;
}
struct ProfilerData
{
@@ -991,7 +947,6 @@ struct ProducerWrapper
struct ProfilerThreadData
{
ProfilerThreadData( ProfilerData& data ) : token( data ), gpuCtx( { nullptr } ) {}
RPMallocInit rpmalloc_init;
ProducerWrapper token;
GpuCtxWrapper gpuCtx;
# ifdef TRACY_ON_DEMAND
@@ -999,11 +954,14 @@ struct ProfilerThreadData
# endif
};
std::atomic<int> RpInitDone { 0 };
std::atomic<int> RpInitLock { 0 };
thread_local bool RpThreadInitDone = false;
# ifdef TRACY_MANUAL_LIFETIME
ProfilerData* s_profilerData = nullptr;
TRACY_API void StartupProfiler()
{
RPMallocInit init;
s_profilerData = (ProfilerData*)tracy_malloc( sizeof( ProfilerData ) );
new (s_profilerData) ProfilerData();
s_profilerData->profiler.SpawnWorkerThreads();
@@ -1030,11 +988,10 @@ static ProfilerData& GetProfilerData()
if( !ptr )
{
int expected = 0;
while( !profilerDataLock.compare_exchange_strong( expected, 1, std::memory_order_release, std::memory_order_relaxed ) ) { expected = 0; }
while( !profilerDataLock.compare_exchange_weak( expected, 1, std::memory_order_release, std::memory_order_relaxed ) ) { expected = 0; YieldThread(); }
ptr = profilerData.load( std::memory_order_acquire );
if( !ptr )
{
RPMallocInit init;
ptr = (ProfilerData*)tracy_malloc( sizeof( ProfilerData ) );
new (ptr) ProfilerData();
profilerData.store( ptr, std::memory_order_release );
@@ -1071,7 +1028,6 @@ public:
void* p = pthread_getspecific(m_key);
if (!p)
{
RPMallocInit init;
p = (ProfilerThreadData*)tracy_malloc( sizeof( ProfilerThreadData ) );
new (p) ProfilerThreadData(GetProfilerData());
pthread_setspecific(m_key, p);
@@ -1123,18 +1079,12 @@ namespace
# endif
#else
TRACY_API void InitRPMallocThread()
{
rpmalloc_thread_initialize();
}
// MSVC static initialization order solution. gcc/clang uses init_order() to avoid all this.
// 1a. But s_queue is needed for initialization of variables in point 2.
extern moodycamel::ConcurrentQueue<QueueItem> s_queue;
thread_local RPMallocInit init_order(106) s_rpmalloc_thread_init;
// 2. If these variables would be in the .CRT$XCB section, they would be initialized only in main thread.
thread_local moodycamel::ProducerToken init_order(107) s_token_detail( s_queue );
thread_local ProducerWrapper init_order(108) s_token { s_queue.get_explicit_producer( s_token_detail ) };
@@ -1147,7 +1097,9 @@ thread_local ThreadHandleWrapper init_order(104) s_threadHandle { detail::GetThr
# endif
static InitTimeWrapper init_order(101) s_initTime { SetupHwTimer() };
static RPMallocInit init_order(102) s_rpmalloc_init;
std::atomic<int> init_order(102) RpInitDone( 0 );
std::atomic<int> init_order(102) RpInitLock( 0 );
thread_local bool RpThreadInitDone = false;
moodycamel::ConcurrentQueue<QueueItem> init_order(103) s_queue( QueuePrealloc );
std::atomic<uint32_t> init_order(104) s_lockCounter( 0 );
std::atomic<uint8_t> init_order(104) s_gpuCtxCounter( 0 );
@@ -3613,19 +3565,6 @@ TRACY_API uint64_t ___tracy_alloc_srcloc_name( uint32_t line, const char* source
return tracy::Profiler::AllocSourceLocation( line, source, sourceSz, function, functionSz, name, nameSz );
}
// thread_locals are not initialized on thread creation. At least on GNU/Linux. Instead they are
// initialized on their first ODR-use. This means that the allocator is not automagically
// initialized every time a thread is created. As thus, expose to the C API users a simple API to
// call every time they create a thread. Here we can then put all sorts of per-thread
// initialization.
TRACY_API void ___tracy_init_thread(void) {
#ifdef TRACY_DELAYED_INIT
(void)tracy::GetProfilerThreadData();
#else
(void)tracy::s_rpmalloc_thread_init;
#endif
}
#ifdef __cplusplus
}
#endif
client/TracyProfiler.hpp
-10
View File
@@ -63,7 +63,6 @@ TRACY_API std::atomic<uint32_t>& GetLockCounter();
TRACY_API std::atomic<uint8_t>& GetGpuCtxCounter();
TRACY_API GpuCtxWrapper& GetGpuCtx();
TRACY_API uint64_t GetThreadHandle();
TRACY_API void InitRPMallocThread();
TRACY_API bool ProfilerAvailable();
TRACY_API int64_t GetFrequencyQpc();
@@ -295,7 +294,6 @@ public:
#endif
if( callstack != 0 )
{
InitRPMallocThread();
tracy::GetProfiler().SendCallstack( callstack );
}
@@ -315,7 +313,6 @@ public:
#endif
if( callstack != 0 )
{
InitRPMallocThread();
tracy::GetProfiler().SendCallstack( callstack );
}
@@ -333,7 +330,6 @@ public:
#endif
if( callstack != 0 )
{
InitRPMallocThread();
tracy::GetProfiler().SendCallstack( callstack );
}
@@ -356,7 +352,6 @@ public:
#endif
if( callstack != 0 )
{
InitRPMallocThread();
tracy::GetProfiler().SendCallstack( callstack );
}
@@ -372,7 +367,6 @@ public:
static tracy_force_inline void MessageAppInfo( const char* txt, size_t size )
{
assert( size < std::numeric_limits<uint16_t>::max() );
InitRPMallocThread();
auto ptr = (char*)tracy_malloc( size );
memcpy( ptr, txt, size );
TracyLfqPrepare( QueueType::MessageAppInfo );
@@ -423,7 +417,6 @@ public:
# endif
const auto thread = GetThreadHandle();
InitRPMallocThread();
auto callstack = Callstack( depth );
profiler.m_serialLock.lock();
@@ -445,7 +438,6 @@ public:
# endif
const auto thread = GetThreadHandle();
InitRPMallocThread();
auto callstack = Callstack( depth );
profiler.m_serialLock.lock();
@@ -495,7 +487,6 @@ public:
# endif
const auto thread = GetThreadHandle();
InitRPMallocThread();
auto callstack = Callstack( depth );
profiler.m_serialLock.lock();
@@ -518,7 +509,6 @@ public:
# endif
const auto thread = GetThreadHandle();
InitRPMallocThread();
auto callstack = Callstack( depth );
profiler.m_serialLock.lock();
client/tracy_rpmalloc.cpp
+3 -1
View File
@@ -130,7 +130,9 @@
# include <stdio.h>
# include <stdlib.h>
# if defined(__APPLE__)
# include <mach/mach_vm.h>
# if !TARGET_OS_IPHONE && !TARGET_OS_SIMULATOR
# include <mach/mach_vm.h>
# endif
# include <mach/vm_statistics.h>
# include <pthread.h>
# endif