P4 to Git Change 1998899 by gandryey@gera-win10 on 2019/09/16 15:49:58
SWDEV-86035 - Swtich PAL to the new interface 533 Affected files ... ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbedefs#48 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#161 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.hpp#46 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#151 edit
Bu işleme şunda yer alıyor:
foreman
@@ -656,7 +656,7 @@ void NullDevice::fillDeviceInfo(const Pal::DeviceProperties& palProp,
|
||||
info_.numAsyncQueues_ = numComputeRings;
|
||||
|
||||
info_.numRTQueues_ = numExclusiveComputeRings;
|
||||
info_.numRTCUs_ = palProp.engineProperties[Pal::EngineTypeExclusiveCompute].maxNumDedicatedCu;
|
||||
info_.numRTCUs_ = palProp.engineProperties[Pal::EngineTypeCompute].maxNumDedicatedCu;
|
||||
|
||||
info_.threadTraceEnable_ = settings().threadTraceEnable_;
|
||||
|
||||
@@ -787,7 +787,6 @@ Device::Device()
|
||||
xferRead_(nullptr),
|
||||
mapCache_(nullptr),
|
||||
resourceCache_(nullptr),
|
||||
numComputeEngines_(0),
|
||||
numDmaEngines_(0),
|
||||
heapInitComplete_(false),
|
||||
xferQueue_(nullptr),
|
||||
@@ -904,22 +903,28 @@ bool Device::create(Pal::IDevice* device) {
|
||||
}
|
||||
|
||||
// Find the number of available engines
|
||||
numComputeEngines_ = properties().engineProperties[Pal::EngineTypeCompute].engineCount;
|
||||
if (properties().engineProperties[Pal::EngineTypeExclusiveCompute].maxNumDedicatedCu > 0) {
|
||||
for (uint i = 0; i < properties().engineProperties[Pal::EngineTypeExclusiveCompute].engineCount;
|
||||
if (properties().engineProperties[Pal::EngineTypeCompute].maxNumDedicatedCu > 0) {
|
||||
for (uint i = 0; i < properties().engineProperties[Pal::EngineTypeCompute].engineCount;
|
||||
++i) {
|
||||
if (properties().engineProperties[Pal::EngineTypeExclusiveCompute].engineSubType[i] ==
|
||||
Pal::EngineSubType::RtCuHighCompute) {
|
||||
if (exclusiveComputeEnginesId_.find(ExclusiveQueueType::RealTime0) !=
|
||||
if (properties().engineProperties[Pal::EngineTypeCompute].capabilities[i].flags.exclusive) {
|
||||
if (properties()
|
||||
.engineProperties[Pal::EngineTypeCompute]
|
||||
.capabilities[i]
|
||||
.queuePrioritySupport == Pal::SupportQueuePriorityRealtime) {
|
||||
if (exclusiveComputeEnginesId_.find(ExclusiveQueueType::RealTime0) !=
|
||||
exclusiveComputeEnginesId_.end()) {
|
||||
exclusiveComputeEnginesId_.insert({ExclusiveQueueType::RealTime1, i});
|
||||
} else {
|
||||
exclusiveComputeEnginesId_.insert({ExclusiveQueueType::RealTime0, i});
|
||||
exclusiveComputeEnginesId_.insert({ExclusiveQueueType::RealTime1, i});
|
||||
} else {
|
||||
exclusiveComputeEnginesId_.insert({ExclusiveQueueType::RealTime0, i});
|
||||
}
|
||||
} else if (properties()
|
||||
.engineProperties[Pal::EngineTypeCompute]
|
||||
.capabilities[i]
|
||||
.queuePrioritySupport == Pal::SupportQueuePriorityMedium) {
|
||||
exclusiveComputeEnginesId_.insert({ExclusiveQueueType::Medium, i});
|
||||
}
|
||||
}
|
||||
if (properties().engineProperties[Pal::EngineTypeExclusiveCompute].engineSubType[i] ==
|
||||
Pal::EngineSubType::RtCuMedCompute) {
|
||||
exclusiveComputeEnginesId_.insert({ExclusiveQueueType::Medium, i});
|
||||
} else {
|
||||
computeEnginesId_.push_back(i);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -957,7 +962,7 @@ bool Device::create(Pal::IDevice* device) {
|
||||
return false;
|
||||
}
|
||||
|
||||
numComputeEngines_ = std::min(numComputeEngines_, settings().numComputeRings_);
|
||||
computeEnginesId_.resize(std::min(numComputeEngines(), settings().numComputeRings_));
|
||||
|
||||
amd::Context::Info info = {0};
|
||||
std::vector<amd::Device*> devices;
|
||||
@@ -1140,11 +1145,14 @@ bool Device::initializeHeapResources() {
|
||||
Pal::DeviceFinalizeInfo finalizeInfo = {};
|
||||
|
||||
// Request all compute engines
|
||||
finalizeInfo.requestedEngineCounts[Pal::EngineTypeCompute].engines =
|
||||
((1 << numComputeEngines_) - 1);
|
||||
for (const auto& it : computeEnginesId_) {
|
||||
// Request real time compute engines
|
||||
finalizeInfo.requestedEngineCounts[Pal::EngineTypeCompute].engines |= (1 << it);
|
||||
}
|
||||
|
||||
for (const auto& it : exclusiveComputeEnginesId_) {
|
||||
// Request real time compute engines
|
||||
finalizeInfo.requestedEngineCounts[Pal::EngineTypeExclusiveCompute].engines |=
|
||||
finalizeInfo.requestedEngineCounts[Pal::EngineTypeCompute].engines |=
|
||||
(1 << it.second);
|
||||
}
|
||||
// Request all SDMA engines
|
||||
|
||||
@@ -400,7 +400,12 @@ class Device : public NullDevice {
|
||||
ResourceCache& resourceCache() const { return *resourceCache_; }
|
||||
|
||||
//! Returns the number of available compute rings
|
||||
uint numComputeEngines() const { return numComputeEngines_; }
|
||||
uint numComputeEngines() const { return computeEnginesId_.size(); }
|
||||
|
||||
//! Returns the vector of available compute rings with the engine index
|
||||
const std::vector<uint32_t>& computeEnginesId() const {
|
||||
return computeEnginesId_;
|
||||
}
|
||||
|
||||
//! Returns the number of available compute rings
|
||||
uint numExclusiveComputeEngines() const { return exclusiveComputeEnginesId_.size(); }
|
||||
@@ -595,9 +600,9 @@ class Device : public NullDevice {
|
||||
XferBuffers* xferRead_; //!< Transfer buffers read
|
||||
std::vector<amd::Memory*>* mapCache_; //!< Map cache info structure
|
||||
ResourceCache* resourceCache_; //!< Resource cache
|
||||
uint numComputeEngines_; //!< The number of available compute engines
|
||||
std::map<ExclusiveQueueType, uint32_t>
|
||||
exclusiveComputeEnginesId_; //!< The number of available compute engines
|
||||
std::vector<uint32_t> computeEnginesId_; //!< PAL index for compute engine
|
||||
uint numDmaEngines_; //!< The number of available compute engines
|
||||
bool heapInitComplete_; //!< Keep track of initialization status of heap resources
|
||||
VirtualGPU* xferQueue_; //!< Transfer queue
|
||||
|
||||
@@ -50,10 +50,11 @@ VirtualGPU::Queue* VirtualGPU::Queue::Create(const VirtualGPU& gpu, Pal::QueueTy
|
||||
Pal::Result result;
|
||||
Pal::CmdBufferCreateInfo cmdCreateInfo = {};
|
||||
Pal::QueueCreateInfo qCreateInfo = {};
|
||||
qCreateInfo.engineIndex = engineIdx;
|
||||
qCreateInfo.engineIndex =
|
||||
(queueType == Pal::QueueTypeCompute) ? gpu.dev().computeEnginesId()[engineIdx] : engineIdx;
|
||||
qCreateInfo.aqlQueue = true;
|
||||
qCreateInfo.queueType = queueType;
|
||||
qCreateInfo.priority = Pal::QueuePriority::Low;
|
||||
qCreateInfo.priority = Pal::QueuePriority::Normal;
|
||||
|
||||
if (queueType == Pal::QueueTypeDma) {
|
||||
cmdCreateInfo.engineType = qCreateInfo.engineType = Pal::EngineTypeDma;
|
||||
@@ -64,7 +65,7 @@ VirtualGPU::Queue* VirtualGPU::Queue::Create(const VirtualGPU& gpu, Pal::QueueTy
|
||||
if ((priority == amd::CommandQueue::Priority::Medium) &&
|
||||
(amd::CommandQueue::RealTimeDisabled == rtCU)) {
|
||||
it = gpu.dev().exclusiveComputeEnginesId().find(ExclusiveQueueType::Medium);
|
||||
cmdCreateInfo.engineType = qCreateInfo.engineType = Pal::EngineTypeExclusiveCompute;
|
||||
cmdCreateInfo.engineType = qCreateInfo.engineType = Pal::EngineTypeCompute;
|
||||
qCreateInfo.priority = Pal::QueuePriority::Medium;
|
||||
} else if (amd::CommandQueue::RealTimeDisabled != rtCU) {
|
||||
qCreateInfo.numReservedCu = rtCU;
|
||||
@@ -73,18 +74,20 @@ VirtualGPU::Queue* VirtualGPU::Queue::Create(const VirtualGPU& gpu, Pal::QueueTy
|
||||
} else {
|
||||
it = gpu.dev().exclusiveComputeEnginesId().find(ExclusiveQueueType::RealTime0);
|
||||
}
|
||||
cmdCreateInfo.engineType = qCreateInfo.engineType = Pal::EngineTypeExclusiveCompute;
|
||||
cmdCreateInfo.engineType = qCreateInfo.engineType = Pal::EngineTypeCompute;
|
||||
cmdCreateInfo.flags.realtimeComputeUnits = true;
|
||||
qCreateInfo.priority = Pal::QueuePriority::Realtime;
|
||||
}
|
||||
// If the app creates an exclusive compute, then find the engine id
|
||||
if (qCreateInfo.engineType == Pal::EngineTypeExclusiveCompute) {
|
||||
if (it != gpu.dev().exclusiveComputeEnginesId().end()) {
|
||||
qCreateInfo.engineIndex = it->second;
|
||||
} else {
|
||||
return nullptr;
|
||||
|
||||
// If the app creates an exclusive compute, then find the engine id
|
||||
if (qCreateInfo.engineType == Pal::EngineTypeCompute) {
|
||||
if (it != gpu.dev().exclusiveComputeEnginesId().end()) {
|
||||
qCreateInfo.engineIndex = it->second;
|
||||
} else {
|
||||
return nullptr;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Find queue object size
|
||||
size_t qSize = palDev->GetQueueSize(qCreateInfo, &result);
|
||||
if (result != Pal::Result::Success) {
|
||||
|
||||
Yeni konuda referans
Bir kullanıcı engelle