intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-23 03:01:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

Implement BCS selection heuristic for OpenCL CommandQueue

Browse Source 
Related-To: NEO-6057
Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>
This commit is contained in:
Maciej Dziuban
2021-10-06 17:40:30 +00:00
committed by Compute-Runtime-Automation
parent 3bff852ac0
commit 4c4b37f8d2
6 changed files with 272 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

67
opencl/source/command_queue/command_queue.cpp
View File

@@ -157,16 +157,53 @@ CommandStreamReceiver *CommandQueue::getBcsForAuxTranslation() const {
}
CommandStreamReceiver &CommandQueue::selectCsrForBuiltinOperation(const CsrSelectionArgs &args) const {
const bool blitAllowed = blitEnqueueAllowed(args);
const bool blitPreferred = blitEnqueuePreferred(args);
const bool blitRequired = isCopyOnly;
const bool blit = blitAllowed && (blitPreferred || blitRequired);
if (blit) {
if (isCopyOnly) {
return *getAnyBcs();
} else {
}
if (!blitEnqueueAllowed(args)) {
return getGpgpuCommandStreamReceiver();
}
bool preferBcs = true;
aub_stream::EngineType preferredBcsEngineType = aub_stream::EngineType::NUM_ENGINES;
switch (args.direction) {
case TransferDirection::LocalToLocal: {
const auto &clHwHelper = ClHwHelper::get(device->getHardwareInfo().platform.eRenderCoreFamily);
preferBcs = clHwHelper.preferBlitterForLocalToLocalTransfers();
if (auto flag = DebugManager.flags.PreferCopyEngineForCopyBufferToBuffer.get(); flag != -1) {
preferBcs = static_cast<bool>(flag);
}
if (preferBcs) {
preferredBcsEngineType = aub_stream::EngineType::ENGINE_BCS;
}
break;
}
case TransferDirection::HostToHost:
case TransferDirection::HostToLocal:
case TransferDirection::LocalToHost: {
preferBcs = true;
preferredBcsEngineType = EngineHelpers::getBcsEngineType(device->getHardwareInfo(), device->getDeviceBitfield(),
device->getSelectorCopyEngine(), false);
break;
}
default:
UNRECOVERABLE_IF(true);
}
CommandStreamReceiver *selectedCsr = nullptr;
if (preferBcs) {
selectedCsr = getBcsCommandStreamReceiver(preferredBcsEngineType);
if (selectedCsr == nullptr) {
selectedCsr = getAnyBcs();
}
}
if (selectedCsr == nullptr) {
selectedCsr = &getGpgpuCommandStreamReceiver();
}
UNRECOVERABLE_IF(selectedCsr == nullptr);
return *selectedCsr;
}
Device &CommandQueue::getDevice() const noexcept {
@@ -752,10 +789,6 @@ bool CommandQueue::queueDependenciesClearRequired() const {
}
bool CommandQueue::blitEnqueueAllowed(const CsrSelectionArgs &args) const {
if (getAnyBcs() == nullptr) {
return false;
}
bool blitEnqueueAllowed = getGpgpuCommandStreamReceiver().peekTimestampPacketWriteEnabled() || this->isCopyOnly;
if (DebugManager.flags.EnableBlitterForEnqueueOperations.get() != -1) {
blitEnqueueAllowed = DebugManager.flags.EnableBlitterForEnqueueOperations.get();
@@ -789,18 +822,6 @@ bool CommandQueue::blitEnqueueAllowed(const CsrSelectionArgs &args) const {
}
}
bool CommandQueue::blitEnqueuePreferred(const CsrSelectionArgs &args) const {
if (args.direction == TransferDirection::LocalToLocal) {
if (DebugManager.flags.PreferCopyEngineForCopyBufferToBuffer.get() != -1) {
return static_cast<bool>(DebugManager.flags.PreferCopyEngineForCopyBufferToBuffer.get());
}
const auto &clHwHelper = ClHwHelper::get(device->getHardwareInfo().platform.eRenderCoreFamily);
return clHwHelper.preferBlitterForLocalToLocalTransfers();
}
return true;
}
bool CommandQueue::blitEnqueueImageAllowed(const size_t *origin, const size_t *region, const Image &image) const {
const auto &hwInfo = device->getHardwareInfo();
const auto &hwInfoConfig = HwInfoConfig::get(hwInfo.platform.eProductFamily);