intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 05:56:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimize Level Zero indirect allocations handling.

Browse Source 
Make them resident directly instead of populating residency container
Remove finds, not needed, CSR resolves duplicates at makeResident calls
Observed gain is 32x for 10k indirect allocations.


Co-authored-by: Michal Mrozek <michal.mrozek@intel.com>

Co-authored-by: Dominik Dabek <dominik.dabek@intel.com>

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This commit is contained in:
Michal Mrozek
2022-01-14 13:56:53 +00:00
committed by Compute-Runtime-Automation
parent 513e6a2ee4
commit 3ecbc55ba9
9 changed files with 192 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
shared/test/common/mocks/mock_command_stream_receiver.h
View File

@@ -252,6 +252,12 @@ class MockCsrHw2 : public CommandStreamReceiverHw<GfxFamily> {
return taskCount;
}
void makeSurfacePackNonResident(ResidencyContainer &allocationsForResidency) override {
for (auto &surface : allocationsForResidency) {
rememberedResidencies.push_back(surface);
}
CommandStreamReceiver::makeSurfacePackNonResident(allocationsForResidency);
}
void programHardwareContext(LinearStream &cmdStream) override {
programHardwareContextCalled = true;
}
@@ -261,6 +267,7 @@ class MockCsrHw2 : public CommandStreamReceiverHw<GfxFamily> {
std::unique_ptr<uint8_t> storedTaskStream;
size_t storedTaskStreamSize = 0;
ResidencyContainer rememberedResidencies;
int flushCalledCount = 0;
std::unique_ptr<CommandBuffer> recordedCommandBuffer = nullptr;
ResidencyContainer copyOfAllocations;