Bart/bullet3
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

parallel solver: various changes

Browse Source 
- threading: adding btSequentialImpulseConstraintSolverMt
 - task scheduler: added parallelSum so that parallel solver can compute residuals
 - CommonRigidBodyMTBase: add slider for solver least squares residual and allow multithreading without needing OpenMP, TBB, or PPL
 - taskScheduler: don't wait for workers to sleep/signal at the end of each parallel block
 - parallel solver: convertContacts split into an allocContactConstraints and setupContactConstraints stage, the latter of which is done in parallel
 - parallel solver: rolling friction is now interleaved along with normal friction
 - parallel solver: batchified split impulse solving + some cleanup
 - parallel solver: sorting batches from largest to smallest
 - parallel solver: added parallel batch creation
 - parallel solver: added warmstartingWriteBackContacts func + other cleanup
 - task scheduler: truncate low bits to preserve determinism with parallelSum
 - parallel solver: reducing dynamic mem allocs and trying to parallelize more of the batch setup
 - parallel solver: parallelize updating constraint batch ids for merging
 - parallel solver: adding debug visualization
 - task scheduler: make TBB task scheduler parallelSum deterministic
 - parallel solver: split batch gen code into separate file; allow selection of batch gen method
 - task scheduler: add sleepWorkerThreadsHint() at end of simulation
 - parallel solver: added grain size per phase
 - task Scheduler: fix for strange threading issue; also no need for main thread to wait for workers to sleep
 - base constraint solver: break out joint setup into separate function for profiling/overriding
 - parallel solver: allow different batching method for contacts vs joints
 - base constraint solver: add convertJoint and convertBodies to make it possible to parallelize joint and body conversion
 - parallel solver: convert joints and bodies in parallel now
 - parallel solver: speed up batch creation with run-length encoding
 - parallel solver: batch gen: run-length expansion in parallel; collect constraint info in parallel
 - parallel solver: adding spatial grid batching method
 - parallel solver: enhancements to spatial grid batching
 - sequential solver: moving code for writing back into functions that derived classes can call
 - parallel solver: do write back of bodies and joints in parallel
 - parallel solver: removed all batching methods except for spatial grid (others were ineffective)
 - parallel solver: added 2D or 3D grid batching options; and a bit of cleanup
 - move btDefaultTaskScheduler into LinearMath project
This commit is contained in:
Lunkhound
2017-06-04 17:57:25 -07:00
parent 94bc897067
commit b8720f2161
25 changed files with 5236 additions and 767 deletions
Download Patch File Download Diff File Expand all files Collapse all files

121
src/LinearMath/btThreads.cpp
View File

@@ -453,6 +453,33 @@ void btParallelFor( int iBegin, int iEnd, int grainSize, const btIParallelForBod
#endif// #if BT_THREADSAFE
}
btScalar btParallelSum( int iBegin, int iEnd, int grainSize, const btIParallelSumBody& body )
{
#if BT_THREADSAFE
#if BT_DETECT_BAD_THREAD_INDEX
if ( !btThreadsAreRunning() )
{
// clear out thread ids
for ( int i = 0; i < BT_MAX_THREAD_COUNT; ++i )
{
gDebugThreadIds[ i ] = kInvalidThreadId;
}
}
#endif // #if BT_DETECT_BAD_THREAD_INDEX
btAssert( gBtTaskScheduler != NULL ); // call btSetTaskScheduler() with a valid task scheduler first!
return gBtTaskScheduler->parallelSum( iBegin, iEnd, grainSize, body );
#else // #if BT_THREADSAFE
// non-parallel version of btParallelSum
btAssert( !"called btParallelFor in non-threadsafe build. enable BT_THREADSAFE" );
return body.sumLoop( iBegin, iEnd );
#endif //#else // #if BT_THREADSAFE
}
///
/// btTaskSchedulerSequential -- non-threaded implementation of task scheduler
@@ -470,6 +497,11 @@ public:
BT_PROFILE( "parallelFor_sequential" );
body.forLoop( iBegin, iEnd );
}
virtual btScalar parallelSum( int iBegin, int iEnd, int grainSize, const btIParallelSumBody& body ) BT_OVERRIDE
{
BT_PROFILE( "parallelSum_sequential" );
return body.sumLoop( iBegin, iEnd );
}
};
@@ -514,11 +546,25 @@ public:
#pragma omp parallel for schedule( static, 1 )
for ( int i = iBegin; i < iEnd; i += grainSize )
{
BT_PROFILE( "OpenMP_job" );
BT_PROFILE( "OpenMP_forJob" );
body.forLoop( i, ( std::min )( i + grainSize, iEnd ) );
}
btPopThreadsAreRunning();
}
virtual btScalar parallelSum( int iBegin, int iEnd, int grainSize, const btIParallelSumBody& body ) BT_OVERRIDE
{
BT_PROFILE( "parallelFor_OpenMP" );
btPushThreadsAreRunning();
btScalar sum = btScalar( 0 );
#pragma omp parallel for schedule( static, 1 ) reduction(+:sum)
for ( int i = iBegin; i < iEnd; i += grainSize )
{
BT_PROFILE( "OpenMP_sumJob" );
sum += body.sumLoop( i, ( std::min )( i + grainSize, iEnd ) );
}
btPopThreadsAreRunning();
return sum;
}
};
#endif // #if BT_USE_OPENMP && BT_THREADSAFE
@@ -571,22 +617,21 @@ public:
btResetThreadIndexCounter();
}
}
struct BodyAdapter
struct ForBodyAdapter
{
const btIParallelForBody* mBody;
ForBodyAdapter( const btIParallelForBody* body ) : mBody( body ) {}
void operator()( const tbb::blocked_range<int>& range ) const
{
BT_PROFILE( "TBB_job" );
BT_PROFILE( "TBB_forJob" );
mBody->forLoop( range.begin(), range.end() );
}
};
virtual void parallelFor( int iBegin, int iEnd, int grainSize, const btIParallelForBody& body ) BT_OVERRIDE
{
BT_PROFILE( "parallelFor_TBB" );
// TBB dispatch
BodyAdapter tbbBody;
tbbBody.mBody = &body;
ForBodyAdapter tbbBody( &body );
btPushThreadsAreRunning();
tbb::parallel_for( tbb::blocked_range<int>( iBegin, iEnd, grainSize ),
tbbBody,
@@ -594,6 +639,29 @@ public:
);
btPopThreadsAreRunning();
}
struct SumBodyAdapter
{
const btIParallelSumBody* mBody;
btScalar mSum;
SumBodyAdapter( const btIParallelSumBody* body ) : mBody( body ), mSum( btScalar( 0 ) ) {}
SumBodyAdapter( const SumBodyAdapter& src, tbb::split ) : mBody( src.mBody ), mSum( btScalar( 0 ) ) {}
void join( const SumBodyAdapter& src ) { mSum += src.mSum; }
void operator()( const tbb::blocked_range<int>& range )
{
BT_PROFILE( "TBB_sumJob" );
mSum += mBody->sumLoop( range.begin(), range.end() );
}
};
virtual btScalar parallelSum( int iBegin, int iEnd, int grainSize, const btIParallelSumBody& body ) BT_OVERRIDE
{
BT_PROFILE( "parallelSum_TBB" );
SumBodyAdapter tbbBody( &body );
btPushThreadsAreRunning();
tbb::parallel_deterministic_reduce( tbb::blocked_range<int>( iBegin, iEnd, grainSize ), tbbBody );
btPopThreadsAreRunning();
return tbbBody.mSum;
}
};
#endif // #if BT_USE_TBB && BT_THREADSAFE
@@ -605,6 +673,7 @@ public:
class btTaskSchedulerPPL : public btITaskScheduler
{
int m_numThreads;
concurrency::combinable<btScalar> m_sum; // for parallelSum
public:
btTaskSchedulerPPL() : btITaskScheduler( "PPL" )
{
@@ -644,15 +713,16 @@ public:
btResetThreadIndexCounter();
}
}
struct BodyAdapter
struct ForBodyAdapter
{
const btIParallelForBody* mBody;
int mGrainSize;
int mIndexEnd;
ForBodyAdapter( const btIParallelForBody* body, int grainSize, int end ) : mBody( body ), mGrainSize( grainSize ), mIndexEnd( end ) {}
void operator()( int i ) const
{
BT_PROFILE( "PPL_job" );
BT_PROFILE( "PPL_forJob" );
mBody->forLoop( i, ( std::min )( i + mGrainSize, mIndexEnd ) );
}
};
@@ -660,10 +730,7 @@ public:
{
BT_PROFILE( "parallelFor_PPL" );
// PPL dispatch
BodyAdapter pplBody;
pplBody.mBody = &body;
pplBody.mGrainSize = grainSize;
pplBody.mIndexEnd = iEnd;
ForBodyAdapter pplBody( &body, grainSize, iEnd );
btPushThreadsAreRunning();
// note: MSVC 2010 doesn't support partitioner args, so avoid them
concurrency::parallel_for( iBegin,
@@ -673,6 +740,36 @@ public:
);
btPopThreadsAreRunning();
}
struct SumBodyAdapter
{
const btIParallelSumBody* mBody;
concurrency::combinable<btScalar>* mSum;
int mGrainSize;
int mIndexEnd;
SumBodyAdapter( const btIParallelSumBody* body, concurrency::combinable<btScalar>* sum, int grainSize, int end ) : mBody( body ), mSum(sum), mGrainSize( grainSize ), mIndexEnd( end ) {}
void operator()( int i ) const
{
BT_PROFILE( "PPL_sumJob" );
mSum->local() += mBody->sumLoop( i, ( std::min )( i + mGrainSize, mIndexEnd ) );
}
};
static btScalar sumFunc( btScalar a, btScalar b ) { return a + b; }
virtual btScalar parallelSum( int iBegin, int iEnd, int grainSize, const btIParallelSumBody& body ) BT_OVERRIDE
{
BT_PROFILE( "parallelSum_PPL" );
m_sum.clear();
SumBodyAdapter pplBody( &body, &m_sum, grainSize, iEnd );
btPushThreadsAreRunning();
// note: MSVC 2010 doesn't support partitioner args, so avoid them
concurrency::parallel_for( iBegin,
iEnd,
grainSize,
pplBody
);
btPopThreadsAreRunning();
return m_sum.combine( sumFunc );
}
};
#endif // #if BT_USE_PPL && BT_THREADSAFE