AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-301667 - Refactor graph code

Browse Source 
- Remove Last graph node optimization and instead submit a barrier NOP
packet always. This simplifies the code.

Change-Id: Ied443173ba47a08b6df148ac7e3ead712acda11c
This commit is contained in:
Saleel Kudchadker
2024-05-09 19:13:34 +00:00
committed by Maneesh Gupta
parent 3d540ec113
commit badf2b0880
3 changed files with 14 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/platform/command.hpp
+2 -7
View File
@@ -1251,8 +1251,6 @@ class Marker : public Command {
class AccumulateCommand : public Command {
private:
uint8_t* lastPacket_;
//! Kernel names and timestamps list for activity profiling
std::vector<std::string> kernelNames_;
std::vector<std::pair<uint64_t, uint64_t>> tsList_;
@@ -1260,12 +1258,9 @@ class AccumulateCommand : public Command {
public:
//! Create a new Marker
AccumulateCommand(HostQueue& queue, const EventWaitList& eventWaitList = nullWaitList,
const Event* waitingEvent = nullptr, uint8_t* lastPacket = nullptr)
: Command(queue, CL_COMMAND_TASK, eventWaitList, 0, waitingEvent),
lastPacket_(lastPacket)
const Event* waitingEvent = nullptr)
: Command(queue, CL_COMMAND_TASK, eventWaitList, 0, waitingEvent)
{}
//! Return last packet
uint8_t* getLastPacket() const { return lastPacket_; }
//! Add kernel name to the list if available
void addKernelName(const std::string& kernelName) {