Cold-Air-Networks/mirror-llvm-project
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 7

[OFFLOAD][L0] Fix zero argurment kernel launch (#212296)

Browse source 
PR #205224 changed the path for kernels with no arguments to not use
appendLaunchKernelWithArgs, while at the same time it removed
zeKernelSetGroupSize which resulted in incorrect sizing of the kernel.

This PR removes the alternate path and makes sure all launches go
through appendLaunchKernelWithArgs.

Alternatively we can restore zeKernelSetGroupSize on the alternate path.
This commit is contained in:
Alex Duran 2026-07-27 21:44:06 +02:00 committed by GitHub
parent 63e60a2853
commit 7677166579
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 4 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
offload/plugins-nextgen/level_zero/include/L0Kernel.h
View file

@ -42,15 +42,13 @@ struct L0LaunchEnvTy {
ze_group_size_t GroupSizes = {0, 0, 0};
KernelPropertiesTy &KernelPR;
bool IsCooperative = false;
bool IsPtrArg = false;
void **ArgPtrs = nullptr;
std::unique_lock<std::mutex> Lock;
L0LaunchEnvTy(KernelPropertiesTy &KernelPR, KernelArgsTy &KernelArgs,
KernelLaunchParamsTy LaunchParams)
: KernelPR(KernelPR), IsCooperative(KernelArgs.Flags.Cooperative),
IsPtrArg(LaunchParams.Args != nullptr), ArgPtrs(LaunchParams.Args),
Lock(KernelPR.Mtx, std::defer_lock) {}
ArgPtrs(LaunchParams.Args), Lock(KernelPR.Mtx, std::defer_lock) {}
};
class L0KernelTy : public GenericKernelTy {

11
offload/plugins-nextgen/level_zero/src/L0Queue.cpp
View file

@ -52,14 +52,9 @@ Error L0QueueTy::dispatchLaunchKernel(ze_kernel_handle_t Kernel,
ze_event_handle_t *WaitEvents) {
// Unlock KEnv lock after launching the kernel.
llvm::scope_exit UnlockGuard([&KEnv]() { KEnv.Lock.unlock(); });
if (KEnv.IsPtrArg)
return CmdList->appendLaunchKernelWithArgs(
Kernel, &KEnv.GroupCounts, &KEnv.GroupSizes, KEnv.ArgPtrs, SignalEvent,
NumWaitEvents, WaitEvents, KEnv.IsCooperative);
return CmdList->appendLaunchKernel(Kernel, &KEnv.GroupCounts, SignalEvent,
NumWaitEvents, WaitEvents,
KEnv.IsCooperative);
return CmdList->appendLaunchKernelWithArgs(
Kernel, &KEnv.GroupCounts, &KEnv.GroupSizes, KEnv.ArgPtrs, SignalEvent,
NumWaitEvents, WaitEvents, KEnv.IsCooperative);
}
Error L0QueueTy::memoryFill(void *Ptr, const void *Pattern, size_t PatternSize,