intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-25 21:42:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: Add heapless mode programming in ocl

Browse Source 
Related-To: NEO-7621
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
This commit is contained in:
Kamil Kopryk
2023-11-23 13:58:58 +00:00
committed by Compute-Runtime-Automation
parent c35b13ccae
commit ce7298d512
82 changed files with 1927 additions and 1224 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -37,7 +37,8 @@ constexpr size_t TimestampDestinationAddressAlignment = 16;
constexpr size_t ImmWriteDestinationAddressAlignment = 8;
template <typename Family>
void EncodeDispatchKernel<Family>::setGrfInfo(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, uint32_t numGrf,
template <typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::setGrfInfo(InterfaceDescriptorType *pInterfaceDescriptor, uint32_t numGrf,
const size_t &sizeCrossThreadData, const size_t &sizePerThreadData,
const RootDeviceEnvironment &rootDeviceEnvironment) {
}
@@ -77,9 +78,6 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
EncodeDispatchKernel<Family>::setGrfInfo(&idd, kernelDescriptor.kernelAttributes.numGrfRequired, sizeCrossThreadData,
sizePerThreadData, rootDeviceEnvironment);
auto &productHelper = args.device->getProductHelper();
productHelper.updateIddCommand(&idd, kernelDescriptor.kernelAttributes.numGrfRequired,
kernelDescriptor.kernelAttributes.threadArbitrationPolicy);
bool localIdsGenerationByRuntime = args.dispatchInterface->requiresGenerationOfLocalIdsByRuntime();
auto requiredWorkgroupOrder = args.dispatchInterface->getRequiredWorkgroupOrder();
@@ -105,17 +103,18 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
hwInfo);
auto &gfxCoreHelper = args.device->getGfxCoreHelper();
auto slmSize = static_cast<SHARED_LOCAL_MEMORY_SIZE>(
auto slmSize = static_cast<uint32_t>(
gfxCoreHelper.computeSlmValues(hwInfo, args.dispatchInterface->getSlmTotalSize()));
if (DebugManager.flags.OverrideSlmAllocationSize.get() != -1) {
slmSize = static_cast<SHARED_LOCAL_MEMORY_SIZE>(DebugManager.flags.OverrideSlmAllocationSize.get());
slmSize = static_cast<uint32_t>(DebugManager.flags.OverrideSlmAllocationSize.get());
}
idd.setSharedLocalMemorySize(slmSize);
auto bindingTableStateCount = kernelDescriptor.payloadMappings.bindingTable.numEntries;
bool skipSshProgramming = false;
auto &productHelper = args.device->getProductHelper();
if (productHelper.isSkippingStatefulInformationRequired(kernelDescriptor)) {
bindingTableStateCount = 0u;
skipSshProgramming = true;
@@ -272,7 +271,9 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
args.useGlobalAtomics, // useGlobalAtomics
args.partitionCount > 1, // multiOsContextCapable
args.isRcs, // isRcs
container.doubleSbaWaRef()}; // doubleSbaWa
container.doubleSbaWaRef(), // doubleSbaWa
false, // heaplessModeEnabled
};
EncodeStateBaseAddress<Family>::encode(encodeStateBaseAddressArgs);
container.setDirtyStateForAllHeaps(false);
}
@@ -392,7 +393,8 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
}
template <typename Family>
inline void EncodeDispatchKernel<Family>::setupPostSyncMocs(WALKER_TYPE &walkerCmd, const RootDeviceEnvironment &rootDeviceEnvironment, bool dcFlush) {
template <typename WalkerType>
inline void EncodeDispatchKernel<Family>::setupPostSyncMocs(WalkerType &walkerCmd, const RootDeviceEnvironment &rootDeviceEnvironment, bool dcFlush) {
auto &postSyncData = walkerCmd.getPostSync();
auto gmmHelper = rootDeviceEnvironment.getGmmHelper();