Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-10 12:53:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

Correct binding table prefetch

Browse Source 
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
Related-To: NEO-6075

Binding table entry count was zeroed even when
ForceBtpPrefetchMode debug flag was enabled
This commit is contained in:
Kamil Kopryk
2022-09-13 12:47:58 +00:00
committed by Compute-Runtime-Automation
parent 88fe22fc49
commit 410fd7d909
9 changed files with 42 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@ -32,33 +32,38 @@ HWCMDTEST_F(IGFX_GEN8_CORE, CommandListAppendLaunchKernel, givenFunctionWhenBind
using MEDIA_INTERFACE_DESCRIPTOR_LOAD = typename FamilyType::MEDIA_INTERFACE_DESCRIPTOR_LOAD;
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
createKernel();
for (auto debugKey : {-1, 0, 1}) {
DebugManagerStateRestore restore;
DebugManager.flags.ForceBtpPrefetchMode.set(debugKey);
ze_group_count_t groupCount{1, 1, 1};
ze_result_t returnValue;
std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
CmdListKernelLaunchParams launchParams = {};
commandList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams);
createKernel();
auto commandStream = commandList->commandContainer.getCommandStream();
ze_group_count_t groupCount{1, 1, 1};
ze_result_t returnValue;
std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
CmdListKernelLaunchParams launchParams = {};
commandList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams);
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(cmdList, commandStream->getCpuBase(), commandStream->getUsed()));
auto commandStream = commandList->commandContainer.getCommandStream();
auto itorMIDL = find<MEDIA_INTERFACE_DESCRIPTOR_LOAD *>(cmdList.begin(), cmdList.end());
ASSERT_NE(itorMIDL, cmdList.end());
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(cmdList, commandStream->getCpuBase(), commandStream->getUsed()));
auto cmd = genCmdCast<MEDIA_INTERFACE_DESCRIPTOR_LOAD *>(*itorMIDL);
ASSERT_NE(cmd, nullptr);
auto itorMIDL = find<MEDIA_INTERFACE_DESCRIPTOR_LOAD *>(cmdList.begin(), cmdList.end());
ASSERT_NE(itorMIDL, cmdList.end());
auto dsh = NEO::ApiSpecificConfig::getBindlessConfiguration() ? device->getNEODevice()->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : commandList->commandContainer.getIndirectHeap(NEO::HeapType::DYNAMIC_STATE);
auto idd = static_cast<INTERFACE_DESCRIPTOR_DATA *>(ptrOffset(dsh->getCpuBase(), cmd->getInterfaceDescriptorDataStartAddress()));
auto cmd = genCmdCast<MEDIA_INTERFACE_DESCRIPTOR_LOAD *>(*itorMIDL);
ASSERT_NE(cmd, nullptr);
if (NEO::EncodeSurfaceState<FamilyType>::doBindingTablePrefetch()) {
uint32_t numArgs = kernel->kernelImmData->getDescriptor().payloadMappings.bindingTable.numEntries;
EXPECT_EQ(numArgs, idd->getBindingTableEntryCount());
} else {
EXPECT_EQ(0u, idd->getBindingTableEntryCount());
auto dsh = NEO::ApiSpecificConfig::getBindlessConfiguration() ? device->getNEODevice()->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : commandList->commandContainer.getIndirectHeap(NEO::HeapType::DYNAMIC_STATE);
auto idd = static_cast<INTERFACE_DESCRIPTOR_DATA *>(ptrOffset(dsh->getCpuBase(), cmd->getInterfaceDescriptorDataStartAddress()));
if (NEO::EncodeSurfaceState<FamilyType>::doBindingTablePrefetch()) {
uint32_t numArgs = kernel->kernelImmData->getDescriptor().payloadMappings.bindingTable.numEntries;
EXPECT_EQ(numArgs, idd->getBindingTableEntryCount());
} else {
EXPECT_EQ(0u, idd->getBindingTableEntryCount());
}
}
}

7
opencl/source/helpers/hardware_commands_helper_base.inl
View File

@ -251,10 +251,9 @@ size_t HardwareCommandsHelper<GfxFamily>::sendIndirectState(
uint64_t offsetInterfaceDescriptor = offsetInterfaceDescriptorTable + interfaceDescriptorIndex * sizeof(INTERFACE_DESCRIPTOR_DATA);
auto bindingTablePrefetchSize = std::min(31u, static_cast<uint32_t>(kernel.getNumberOfBindingTableStates()));
if (!EncodeSurfaceState<GfxFamily>::doBindingTablePrefetch()) {
bindingTablePrefetchSize = 0;
auto bindingTablePrefetchSize = 0;
if (EncodeSurfaceState<GfxFamily>::doBindingTablePrefetch()) {
bindingTablePrefetchSize = std::min(31u, static_cast<uint32_t>(kernel.getNumberOfBindingTableStates()));
}
HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(

1
shared/source/command_container/command_encoder.h
View File

@ -304,6 +304,7 @@ struct EncodeSurfaceState {
static void getSshAlignedPointer(uintptr_t &ptr, size_t &offset);
static bool doBindingTablePrefetch();
static bool isBindingTablePrefetchPreferred();
static size_t pushBindingTableAndSurfaceStates(IndirectHeap &dstHeap, size_t bindingTableCount,
const void *srcKernelSsh, size_t srcKernelSshSize,

12
shared/source/command_container/command_encoder.inl
View File

@ -675,12 +675,18 @@ void EncodeIndirectParams<Family>::setWorkDimIndirect(CommandContainer &containe
}
}
template <typename Family>
bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
auto enableBindingTablePrefetech = isBindingTablePrefetchPreferred();
if (DebugManager.flags.ForceBtpPrefetchMode.get() != -1) {
enableBindingTablePrefetech = static_cast<bool>(DebugManager.flags.ForceBtpPrefetchMode.get());
}
return enableBindingTablePrefetech;
}
template <typename Family>
void EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, uint32_t samplerCount, uint32_t bindingTableEntryCount) {
auto enablePrefetch = EncodeSurfaceState<Family>::doBindingTablePrefetch();
if (DebugManager.flags.ForceBtpPrefetchMode.get() != -1) {
enablePrefetch = static_cast<bool>(DebugManager.flags.ForceBtpPrefetchMode.get());
}
if (enablePrefetch) {
interfaceDescriptor.setSamplerCount(static_cast<typename INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT>((samplerCount + 3) / 4));

2
shared/source/command_container/command_encoder_bdw_and_later.inl
View File

@ -482,7 +482,7 @@ inline void EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(EncodeSurface
}
template <typename GfxFamily>
bool EncodeSurfaceState<GfxFamily>::doBindingTablePrefetch() {
bool EncodeSurfaceState<GfxFamily>::isBindingTablePrefetchPreferred() {
return true;
}

2
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@ -628,7 +628,7 @@ size_t EncodeMiFlushDW<Family>::getMiFlushDwWaSize() {
}
template <typename Family>
bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
bool EncodeSurfaceState<Family>::isBindingTablePrefetchPreferred() {
return false;
}

2
shared/source/gen11/command_encoder_gen11.cpp
View File

@ -21,7 +21,7 @@ using Family = NEO::Gen11Family;
namespace NEO {
template <>
bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
bool EncodeSurfaceState<Family>::isBindingTablePrefetchPreferred() {
return false;
}

2
shared/source/gen12lp/command_encoder_gen12lp.cpp
View File

@ -83,7 +83,7 @@ void EncodeSurfaceState<Family>::encodeExtraBufferParams(EncodeSurfaceStateArgs
}
template <>
bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
bool EncodeSurfaceState<Family>::isBindingTablePrefetchPreferred() {
return false;
}

3
shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp
View File

@ -297,9 +297,6 @@ void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(INTERFACE_DESCRIPTO
template <>
void EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, uint32_t samplerCount, uint32_t bindingTableEntryCount) {
auto enablePrefetch = EncodeSurfaceState<Family>::doBindingTablePrefetch();
if (DebugManager.flags.ForceBtpPrefetchMode.get() != -1) {
enablePrefetch = static_cast<bool>(DebugManager.flags.ForceBtpPrefetchMode.get());
}
if (enablePrefetch) {
interfaceDescriptor.setBindingTableEntryCount(std::min(bindingTableEntryCount, 31u));