1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-12-17 03:04:41 +00:00

Logging pulled in from dirichlet branch

This commit is contained in:
Peter Boyle
2022-11-30 15:55:17 -05:00
parent d8c29f5fcf
commit 3791bc527b
6 changed files with 83 additions and 23 deletions

View File

@@ -3,8 +3,13 @@
#warning "Using explicit device memory copies"
NAMESPACE_BEGIN(Grid);
//#define dprintf(...) printf ( __VA_ARGS__ ); fflush(stdout);
#define dprintf(...)
#define MAXLINE 512
static char print_buffer [ MAXLINE ];
#define mprintf(...) snprintf (print_buffer,MAXLINE, __VA_ARGS__ ); std::cout << GridLogMemory << print_buffer;
#define dprintf(...) snprintf (print_buffer,MAXLINE, __VA_ARGS__ ); std::cout << GridLogMemory << print_buffer;
//#define dprintf(...)
////////////////////////////////////////////////////////////
@@ -104,7 +109,7 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
///////////////////////////////////////////////////////////
assert(AccCache.state!=Empty);
dprintf("MemoryManager: Discard(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr);
mprintf("MemoryManager: Discard(%lx) %lx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr);
assert(AccCache.accLock==0);
assert(AccCache.cpuLock==0);
assert(AccCache.CpuPtr!=(uint64_t)NULL);
@@ -112,7 +117,7 @@ void MemoryManager::AccDiscard(AcceleratorViewEntry &AccCache)
AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
DeviceBytes -=AccCache.bytes;
LRUremove(AccCache);
dprintf("MemoryManager: Free(%llx) LRU %lld Total %lld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);
dprintf("MemoryManager: Free(%lx) LRU %ld Total %ld\n",(uint64_t)AccCache.AccPtr,DeviceLRUBytes,DeviceBytes);
}
uint64_t CpuPtr = AccCache.CpuPtr;
EntryErase(CpuPtr);
@@ -126,9 +131,11 @@ void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
///////////////////////////////////////////////////////////////////////////
assert(AccCache.state!=Empty);
dprintf("MemoryManager: Evict(%llx) %llx\n",(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr);
assert(AccCache.accLock==0);
assert(AccCache.cpuLock==0);
mprintf("MemoryManager: Evict cpu %lx acc %lx cpuLock %ld accLock %ld\n",
(uint64_t)AccCache.CpuPtr,(uint64_t)AccCache.AccPtr,
(uint64_t)AccCache.cpuLock,(uint64_t)AccCache.accLock);
if (AccCache.accLock!=0) return;
if (AccCache.cpuLock!=0) return;
if(AccCache.state==AccDirty) {
Flush(AccCache);
}
@@ -137,7 +144,7 @@ void MemoryManager::Evict(AcceleratorViewEntry &AccCache)
AcceleratorFree((void *)AccCache.AccPtr,AccCache.bytes);
DeviceBytes -=AccCache.bytes;
LRUremove(AccCache);
dprintf("MemoryManager: Free(%llx) footprint now %lld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);
dprintf("MemoryManager: Free(%lx) footprint now %ld \n",(uint64_t)AccCache.AccPtr,DeviceBytes);
}
uint64_t CpuPtr = AccCache.CpuPtr;
EntryErase(CpuPtr);
@@ -150,7 +157,7 @@ void MemoryManager::Flush(AcceleratorViewEntry &AccCache)
assert(AccCache.AccPtr!=(uint64_t)NULL);
assert(AccCache.CpuPtr!=(uint64_t)NULL);
acceleratorCopyFromDevice((void *)AccCache.AccPtr,(void *)AccCache.CpuPtr,AccCache.bytes);
dprintf("MemoryManager: Flush %llx -> %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
mprintf("MemoryManager: Flush %lx -> %lx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
DeviceToHostBytes+=AccCache.bytes;
DeviceToHostXfer++;
AccCache.state=Consistent;
@@ -165,7 +172,7 @@ void MemoryManager::Clone(AcceleratorViewEntry &AccCache)
AccCache.AccPtr=(uint64_t)AcceleratorAllocate(AccCache.bytes);
DeviceBytes+=AccCache.bytes;
}
dprintf("MemoryManager: Clone %llx <- %llx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
mprintf("MemoryManager: Clone %lx <- %lx\n",(uint64_t)AccCache.AccPtr,(uint64_t)AccCache.CpuPtr); fflush(stdout);
acceleratorCopyToDevice((void *)AccCache.CpuPtr,(void *)AccCache.AccPtr,AccCache.bytes);
HostToDeviceBytes+=AccCache.bytes;
HostToDeviceXfer++;
@@ -191,6 +198,7 @@ void MemoryManager::CpuDiscard(AcceleratorViewEntry &AccCache)
void MemoryManager::ViewClose(void* Ptr,ViewMode mode)
{
if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
dprintf("AcceleratorViewClose %lx\n",(uint64_t)Ptr);
AcceleratorViewClose((uint64_t)Ptr);
} else if( (mode==CpuRead)||(mode==CpuWrite)){
CpuViewClose((uint64_t)Ptr);
@@ -202,6 +210,7 @@ void *MemoryManager::ViewOpen(void* _CpuPtr,size_t bytes,ViewMode mode,ViewAdvis
{
uint64_t CpuPtr = (uint64_t)_CpuPtr;
if( (mode==AcceleratorRead)||(mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard) ){
dprintf("AcceleratorViewOpen %lx\n",(uint64_t)CpuPtr);
return (void *) AcceleratorViewOpen(CpuPtr,bytes,mode,hint);
} else if( (mode==CpuRead)||(mode==CpuWrite)){
return (void *)CpuViewOpen(CpuPtr,bytes,mode,hint);
@@ -241,11 +250,12 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
assert(AccCache.cpuLock==0); // Programming error
if(AccCache.state!=Empty) {
dprintf("ViewOpen found entry %llx %llx : %lld %lld\n",
dprintf("ViewOpen found entry %lx %lx : %ld %ld accLock %ld\n",
(uint64_t)AccCache.CpuPtr,
(uint64_t)CpuPtr,
(uint64_t)AccCache.bytes,
(uint64_t)bytes);
(uint64_t)bytes,
(uint64_t)AccCache.accLock);
assert(AccCache.CpuPtr == CpuPtr);
assert(AccCache.bytes ==bytes);
}
@@ -280,6 +290,7 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
AccCache.state = Consistent; // Empty + AccRead => Consistent
}
AccCache.accLock= 1;
dprintf("Copied Empty entry into device accLock= %d\n",AccCache.accLock);
} else if(AccCache.state==CpuDirty ){
if(mode==AcceleratorWriteDiscard) {
CpuDiscard(AccCache);
@@ -292,21 +303,21 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
AccCache.state = Consistent; // CpuDirty + AccRead => Consistent
}
AccCache.accLock++;
dprintf("Copied CpuDirty entry into device accLock %d\n",AccCache.accLock);
dprintf("CpuDirty entry into device ++accLock= %d\n",AccCache.accLock);
} else if(AccCache.state==Consistent) {
if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
AccCache.state = AccDirty; // Consistent + AcceleratorWrite=> AccDirty
else
AccCache.state = Consistent; // Consistent + AccRead => Consistent
AccCache.accLock++;
dprintf("Consistent entry into device accLock %d\n",AccCache.accLock);
dprintf("Consistent entry into device ++accLock= %d\n",AccCache.accLock);
} else if(AccCache.state==AccDirty) {
if((mode==AcceleratorWrite)||(mode==AcceleratorWriteDiscard))
AccCache.state = AccDirty; // AccDirty + AcceleratorWrite=> AccDirty
else
AccCache.state = AccDirty; // AccDirty + AccRead => AccDirty
AccCache.accLock++;
dprintf("AccDirty entry into device accLock %d\n",AccCache.accLock);
dprintf("AccDirty entry ++accLock= %d\n",AccCache.accLock);
} else {
assert(0);
}
@@ -314,6 +325,7 @@ uint64_t MemoryManager::AcceleratorViewOpen(uint64_t CpuPtr,size_t bytes,ViewMod
// If view is opened on device remove from LRU
if(AccCache.LRU_valid==1){
// must possibly remove from LRU as now locked on GPU
dprintf("AccCache entry removed from LRU \n");
LRUremove(AccCache);
}
@@ -334,10 +346,12 @@ void MemoryManager::AcceleratorViewClose(uint64_t CpuPtr)
assert(AccCache.accLock>0);
AccCache.accLock--;
// Move to LRU queue if not locked and close on device
if(AccCache.accLock==0) {
dprintf("AccleratorViewClose %lx AccLock decremented to %ld move to LRU queue\n",(uint64_t)CpuPtr,(uint64_t)AccCache.accLock);
LRUinsert(AccCache);
} else {
dprintf("AccleratorViewClose %lx AccLock decremented to %ld\n",(uint64_t)CpuPtr,(uint64_t)AccCache.accLock);
}
}
void MemoryManager::CpuViewClose(uint64_t CpuPtr)
@@ -473,6 +487,29 @@ int MemoryManager::isOpen (void* _CpuPtr)
return 0;
}
}
void MemoryManager::Audit(std::string s)
{
for(auto it=AccViewTable.begin();it!=AccViewTable.end();it++){
auto &AccCache = it->second;
std::string str;
if ( AccCache.state==Empty ) str = std::string("Empty");
if ( AccCache.state==CpuDirty ) str = std::string("CpuDirty");
if ( AccCache.state==AccDirty ) str = std::string("AccDirty");
if ( AccCache.state==Consistent)str = std::string("Consistent");
if ( AccCache.cpuLock || AccCache.accLock ) {
std::cout << GridLogError << s<< "\n\t 0x"<<std::hex<<AccCache.CpuPtr<<std::dec
<< "\t0x"<<std::hex<<AccCache.AccPtr<<std::dec<<"\t" <<str
<< "\t cpuLock " << AccCache.cpuLock
<< "\t accLock " << AccCache.accLock
<< "\t LRUvalid " << AccCache.LRU_valid<<std::endl;
}
assert( AccCache.cpuLock== 0 ) ;
assert( AccCache.accLock== 0 ) ;
}
}
void MemoryManager::PrintState(void* _CpuPtr)
{