intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-19 16:24:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: Affinity mask plus ReturnSubDevicesAsApiDevices

Browse Source 
When using ReturnSubDevicesAsApiDevices=1 to have
sub-devices-as-root-devices, then the driver should read the values
passed in the mask as those corresponding to the physical
sub-devices.

For instance, in a dual system with multi-tile device, we would have:

card 0, tile 0
card 0, tile 1
card 1, tile 0
card 1, tile 1

With:
ReturnSubDevicesAsApiDevices=0
ZE_AFFINITY_MASK=0,1

Then all tiles in card 0 and card 1 need to be exposed.

With:
ReturnSubDevicesAsApiDevices=1
ZE_AFFINITY_MASK=0,3

Then card 0 tile 0, and card 1 tile 1 need to be exposed.

Related-To: NEO-7137

Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
This commit is contained in:
Jaime Arteaga
2023-02-02 02:54:47 +00:00
committed by Compute-Runtime-Automation
parent 275bed2a31
commit 07598fb5e0
5 changed files with 190 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
shared/source/execution_environment/execution_environment.cpp
View File

@@ -147,7 +147,33 @@ void ExecutionEnvironment::parseAffinityMask() {
return;
}
const uint32_t numRootDevices = static_cast<uint32_t>(rootDeviceEnvironments.size());
bool exposeSubDevicesAsApiDevices = false;
if (NEO::DebugManager.flags.ReturnSubDevicesAsApiDevices.get() != -1) {
exposeSubDevicesAsApiDevices = NEO::DebugManager.flags.ReturnSubDevicesAsApiDevices.get();
}
uint32_t numRootDevices = static_cast<uint32_t>(rootDeviceEnvironments.size());
RootDeviceIndicesMap mapOfIndexes;
// Reserve at least for a size equal to rootDeviceEnvironments.size() times four,
// which is enough for typical configurations
size_t reservedSizeForIndices = numRootDevices * 4;
mapOfIndexes.reserve(reservedSizeForIndices);
if (exposeSubDevicesAsApiDevices) {
uint32_t currentDeviceIndex = 0;
for (uint32_t currentRootDevice = 0u; currentRootDevice < static_cast<uint32_t>(rootDeviceEnvironments.size()); currentRootDevice++) {
auto hwInfo = rootDeviceEnvironments[currentRootDevice]->getHardwareInfo();
auto subDevicesCount = GfxCoreHelper::getSubDevicesCount(hwInfo);
uint32_t currentSubDevice = 0;
mapOfIndexes[currentDeviceIndex++] = std::make_tuple(currentRootDevice, currentSubDevice);
for (currentSubDevice = 1; currentSubDevice < subDevicesCount; currentSubDevice++) {
mapOfIndexes[currentDeviceIndex++] = std::make_tuple(currentRootDevice, currentSubDevice);
}
}
numRootDevices = currentDeviceIndex;
UNRECOVERABLE_IF(numRootDevices > reservedSizeForIndices);
}
std::vector<AffinityMaskHelper> affinityMaskHelper(numRootDevices);
@@ -157,6 +183,27 @@ void ExecutionEnvironment::parseAffinityMask() {
auto subEntries = StringHelpers::split(entry, ".");
uint32_t rootDeviceIndex = StringHelpers::toUint32t(subEntries[0]);
// tiles as devices
if (exposeSubDevicesAsApiDevices) {
if (rootDeviceIndex > numRootDevices) {
continue;
}
// ReturnSubDevicesAsApiDevices not supported with AllowSingleTileEngineInstancedSubDevices
// so ignore X.Y
if (subEntries.size() > 1) {
continue;
}
std::tuple<uint32_t, uint32_t> indexKey = mapOfIndexes[rootDeviceIndex];
auto deviceIndex = std::get<0>(indexKey);
auto tileIndex = std::get<1>(indexKey);
affinityMaskHelper[deviceIndex].enableGenericSubDevice(tileIndex);
continue;
}
// cards as devices
if (rootDeviceIndex < numRootDevices) {
auto hwInfo = rootDeviceEnvironments[rootDeviceIndex]->getHardwareInfo();
auto subDevicesCount = GfxCoreHelper::getSubDevicesCount(hwInfo);