About the bug of function is_compute_kernel in HolisticTraceAnalysis/hta/utils /utils.py #220

🐛 Describe the bug

NCCL_COMPUTE_KERNEL_RE = re.compile(r"(^nccl.Kernel)|(.(Memcpy)|(Memset))|(.*Sync)")

def is_compute_kernel(name: str) -> bool:
"""
Check if a given GPU kernel is a computation kernel.
Args:
name (str): name of the GPU kernel.
Returns:
A boolean indicating if the kernel is a computation kernel.
"""
return not NCCL_COMPUTE_KERNEL_RE.match(name)

eg：cudaMemsetsync will be considered as the compute kernel

Steps to reproduce

NCCL_COMPUTE_KERNEL_RE = re.compile(r"(^nccl.Kernel)|(.(Memcpy)|(Memset))|(.*Sync)")

Expected behavior

NCCL_COMPUTE_KERNEL_RE = re.compile(r"(^nccl.Kernel)|(.(Memcpy)|(.*Memset))|(.*Sync)")

Environment

master

Additional Info

No response