AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

First pass at virtualized queue support.

Browse Source 
Also updated stream debug messages to consistently use trace_helper.
This commit is contained in:
Ben Sander
2017-01-04 14:38:18 -06:00
parent fd209f37d9
commit 93fbc9cf7b
4 changed files with 88 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/hip_hcc.cpp
+34 -25
View File
@@ -243,7 +243,7 @@ ihipStream_t::ihipStream_t(ihipCtx_t *ctx, hc::accelerator_view av, unsigned int
_id(0), // will be set by add function.
_flags(flags),
_ctx(ctx),
_criticalData(av)
_criticalData(this, av)
{
unsigned schedBits = ctx->_ctxFlags & hipDeviceScheduleMask;
@@ -256,7 +256,6 @@ ihipStream_t::ihipStream_t(ihipCtx_t *ctx, hc::accelerator_view av, unsigned int
};
tprintf(DB_SYNC, " streamCreate: stream=%s\n", ToString(this).c_str());
};
@@ -271,7 +270,7 @@ ihipStream_t::~ihipStream_t()
void ihipStream_t::wait(LockedAccessor_StreamCrit_t &crit, bool assertQueueEmpty)
{
if (! assertQueueEmpty) {
tprintf (DB_SYNC, "stream %s wait for queue-empty..\n", ToString(this).c_str());
tprintf (DB_SYNC, "%s wait for queue-empty..\n", ToString(this).c_str());
hc::hcWaitMode waitMode = hc::hcWaitModeActive;
if (_scheduleMode == Auto) {
@@ -406,21 +405,21 @@ void ihipStream_t::lockclose_postKernelCommand(const char * kernelName, hc::acce
//=============================================================================
// Recompute the peercnt and the packed _peerAgents whenever a peer is added or deleted.
// The packed _peerAgents can efficiently be used on each memory allocation.
template<>
void ihipCtxCriticalBase_t<CtxMutex>::recomputePeerAgents()
{
_peerCnt = 0;
std::for_each (_peers.begin(), _peers.end(), [this](ihipCtx_t* ctx) {
_peerAgents[_peerCnt++] = ctx->getDevice()->_hsaAgent;
});
}
//=============================================================================
// Recompute the peercnt and the packed _peerAgents whenever a peer is added or deleted.
// The packed _peerAgents can efficiently be used on each memory allocation.
template<>
void ihipCtxCriticalBase_t<CtxMutex>::recomputePeerAgents()
{
_peerCnt = 0;
std::for_each (_peers.begin(), _peers.end(), [this](ihipCtx_t* ctx) {
_peerAgents[_peerCnt++] = ctx->getDevice()->_hsaAgent;
});
}
template<>
bool ihipCtxCriticalBase_t<CtxMutex>::isPeerWatcher(const ihipCtx_t *peer)
template<>
bool ihipCtxCriticalBase_t<CtxMutex>::isPeerWatcher(const ihipCtx_t *peer)
{
auto match = std::find(_peers.begin(), _peers.end(), peer);
return (match != std::end(_peers));
@@ -489,6 +488,7 @@ void ihipCtxCriticalBase_t<CtxMutex>::addStream(ihipStream_t *stream)
{
stream->_id = _streams.size();
_streams.push_back(stream);
tprintf(DB_SYNC, " addStream: %s\n", ToString(stream).c_str());
}
//=============================================================================
@@ -827,11 +827,11 @@ hipError_t ihipDevice_t::initProperties(hipDeviceProp_t* prop)
ihipCtx_t::ihipCtx_t(ihipDevice_t *device, unsigned deviceCnt, unsigned flags) :
_ctxFlags(flags),
_device(device),
_criticalData(deviceCnt)
_criticalData(this, deviceCnt)
{
locked_reset();
tprintf(DB_SYNC, "created ctx with defaultStream=%p\n", _defaultStream);
tprintf(DB_SYNC, "created ctx with defaultStream=%p (%s)\n", _defaultStream, ToString(_defaultStream).c_str());
};
@@ -861,7 +861,7 @@ void ihipCtx_t::locked_reset()
for (auto streamI=crit->const_streams().begin(); streamI!=crit->const_streams().end(); streamI++) {
ihipStream_t *stream = *streamI;
(*streamI)->locked_wait();
tprintf(DB_SYNC, " delete stream=%p\n", stream);
tprintf(DB_SYNC, " delete %s\n", ToString(stream).c_str());
delete stream;
}
@@ -905,15 +905,24 @@ ihipCtx_t::stealActiveQueue(LockedAccessor_CtxCrit_t &ctxCrit, ihipStream_t *nee
for (auto iter=ctxCrit->streams().begin(); iter != ctxCrit->streams().end(); iter++) {
if (*iter != needyStream) {
auto victimCritPtr = (*iter)->_criticalData.mtry_lock();
if (victimCritPtr && victimCritPtr->_hasQueue && (victimCritPtr->_kernelCnt == 0)) {
if (victimCritPtr) {
if (victimCritPtr->_hasQueue && (victimCritPtr->_kernelCnt == 0)) {
victimCritPtr->_hasQueue = false;
victimCritPtr->_hasQueue = false;
tprintf(DB_SYNC, " stealActiveQueue from victim:%s to needy:%s\n",
ToString(*iter).c_str(), ToString(needyStream).c_str());
tprintf(DB_SYNC, " stealActiveQueue move queue from victim:%s to needy:%s\n",
ToString(*iter).c_str(), ToString(needyStream).c_str());
// TODO - cleanup to remove forced setting to N
hc::accelerator_view av = victimCritPtr->_av;
uint64_t *p = (uint64_t*)(&victimCritPtr->_av);
*p = 0; // damage the victim av so attempt to use it will fault.
return victimCritPtr->_av;
(*iter)->_criticalData.munlock();
return av;
} else {
(*iter)->_criticalData.munlock();
}
}
}
}
@@ -1415,7 +1424,7 @@ hipStream_t ihipSyncAndResolveStream(hipStream_t stream)
} else {
// ALl streams have to wait for legacy default stream to be empty:
if (!(stream->_flags & hipStreamNonBlocking)) {
tprintf(DB_SYNC, "stream %p wait default stream\n", stream);
tprintf(DB_SYNC, "%s wait default stream\n", ToString(stream).c_str());
stream->getCtx()->_defaultStream->locked_wait();
}