mirror of
https://github.com/intel/compute-runtime.git
synced 2025-12-26 15:03:02 +08:00
b16bd14f37
Change-Id: I4f1f59ecb51b95041dc6dcc6c606b94595813f53 Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
107 lines
3.7 KiB
C++
107 lines
3.7 KiB
C++
/*
|
|
* Copyright (C) 2018 Intel Corporation
|
|
*
|
|
* SPDX-License-Identifier: MIT
|
|
*
|
|
*/
|
|
|
|
#pragma once
|
|
#include "runtime/command_queue/hardware_interface.h"
|
|
|
|
namespace OCLRT {
|
|
|
|
template <typename GfxFamily>
|
|
inline void HardwareInterface<GfxFamily>::getDefaultDshSpace(
|
|
const size_t &offsetInterfaceDescriptorTable,
|
|
CommandQueue &commandQueue,
|
|
const MultiDispatchInfo &multiDispatchInfo,
|
|
size_t &totalInterfaceDescriptorTableSize,
|
|
Kernel *parentKernel,
|
|
IndirectHeap *dsh,
|
|
LinearStream *commandStream) {
|
|
|
|
size_t numDispatches = multiDispatchInfo.size();
|
|
totalInterfaceDescriptorTableSize *= numDispatches;
|
|
|
|
if (!parentKernel) {
|
|
dsh->getSpace(totalInterfaceDescriptorTableSize);
|
|
} else {
|
|
dsh->getSpace(commandQueue.getContext().getDefaultDeviceQueue()->getDshOffset() - dsh->getUsed());
|
|
}
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
inline typename HardwareInterface<GfxFamily>::INTERFACE_DESCRIPTOR_DATA *
|
|
HardwareInterface<GfxFamily>::obtainInterfaceDescriptorData(
|
|
WALKER_HANDLE pCmdData) {
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
inline void HardwareInterface<GfxFamily>::setOffsetCrossThreadData(
|
|
WALKER_HANDLE pCmdData,
|
|
size_t &offsetCrossThreadData,
|
|
uint32_t &interfaceDescriptorIndex) {
|
|
|
|
WALKER_TYPE<GfxFamily> *pCmd = static_cast<WALKER_TYPE<GfxFamily> *>(pCmdData);
|
|
pCmd->setIndirectDataStartAddress(static_cast<uint32_t>(offsetCrossThreadData));
|
|
pCmd->setInterfaceDescriptorOffset(interfaceDescriptorIndex++);
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
inline void HardwareInterface<GfxFamily>::dispatchWorkarounds(
|
|
LinearStream *commandStream,
|
|
CommandQueue &commandQueue,
|
|
Kernel &kernel,
|
|
const bool &enable) {
|
|
|
|
if (enable) {
|
|
PreemptionHelper::applyPreemptionWaCmdsBegin<GfxFamily>(commandStream, commandQueue.getDevice());
|
|
// Implement enabling special WA DisableLSQCROPERFforOCL if needed
|
|
GpgpuWalkerHelper<GfxFamily>::applyWADisableLSQCROPERFforOCL(commandStream, kernel, enable);
|
|
} else {
|
|
// Implement disabling special WA DisableLSQCROPERFforOCL if needed
|
|
GpgpuWalkerHelper<GfxFamily>::applyWADisableLSQCROPERFforOCL(commandStream, kernel, enable);
|
|
PreemptionHelper::applyPreemptionWaCmdsEnd<GfxFamily>(commandStream, commandQueue.getDevice());
|
|
}
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
inline void HardwareInterface<GfxFamily>::dispatchProfilingPerfStartCommands(
|
|
const DispatchInfo &dispatchInfo,
|
|
const MultiDispatchInfo &multiDispatchInfo,
|
|
HwTimeStamps *hwTimeStamps,
|
|
HwPerfCounter *hwPerfCounter,
|
|
LinearStream *commandStream,
|
|
CommandQueue &commandQueue) {
|
|
|
|
if (&dispatchInfo == &*multiDispatchInfo.begin()) {
|
|
// If hwTimeStampAlloc is passed (not nullptr), then we know that profiling is enabled
|
|
if (hwTimeStamps != nullptr) {
|
|
GpgpuWalkerHelper<GfxFamily>::dispatchProfilingCommandsStart(*hwTimeStamps, commandStream);
|
|
}
|
|
if (hwPerfCounter != nullptr) {
|
|
GpgpuWalkerHelper<GfxFamily>::dispatchPerfCountersCommandsStart(commandQueue, *hwPerfCounter, commandStream);
|
|
}
|
|
}
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
inline void HardwareInterface<GfxFamily>::dispatchProfilingPerfEndCommands(
|
|
HwTimeStamps *hwTimeStamps,
|
|
HwPerfCounter *hwPerfCounter,
|
|
LinearStream *commandStream,
|
|
CommandQueue &commandQueue) {
|
|
|
|
// If hwTimeStamps is passed (not nullptr), then we know that profiling is enabled
|
|
if (hwTimeStamps != nullptr) {
|
|
GpgpuWalkerHelper<GfxFamily>::dispatchProfilingCommandsEnd(*hwTimeStamps, commandStream);
|
|
}
|
|
if (hwPerfCounter != nullptr) {
|
|
GpgpuWalkerHelper<GfxFamily>::dispatchPerfCountersCommandsEnd(commandQueue, *hwPerfCounter, commandStream);
|
|
}
|
|
}
|
|
|
|
} // namespace OCLRT
|