[bug] Fix bug that kernel names are not correctly captured by the profiler

[PGZXB]

Related issue = fixes #5643

[netlify]

✅ Deploy Preview for docsite-preview canceled.

Name	Link
🔨 Latest commit	dee298d
🔍 Latest deploy log	https://app.netlify.com/sites/docsite-preview/deploys/62ed3f4b0e71a90009acc303

[lin-hitonami]

Per offline discussion, it's hard to determine whether a task is cached because the profiler cannot access the kernel information. It needs many refactoring to let the profiler know whether the task is cached. This is a temporary solution that enables the profiler to deduce the real name of a kernel from the mangled name. Special characters cannot be used because they cause error on windows. See

taichi/taichi/runtime/cuda/jit_cuda.cpp

Lines 52 to 72 in 28c7cc8

	std::string convert(std::string new_name) {
	// Evil C++ mangling on Windows will lead to "unsupported characters in
	// symbol" error in LLVM PTX printer. Convert here.
	for (int i = 0; i < (int)new_name.size(); i++) {
	if (new_name[i] == '@')
	new_name.replace(i, 1, "_at_");
	if (new_name[i] == '?')
	new_name.replace(i, 1, "_qm_");
	if (new_name[i] == '$')
	new_name.replace(i, 1, "_dl_");
	if (new_name[i] == '<')
	new_name.replace(i, 1, "_lb_");
	if (new_name[i] == '>')
	new_name.replace(i, 1, "_rb_");
	TI_ASSERT(std::isalpha(new_name[i]) || std::isdigit(new_name[i]) ||
	new_name[i] == '_' || new_name[i] == '.');
	}
	if (!new_name.empty())
	TI_ASSERT(isalpha(new_name[0]) || new_name[0] == '_' || new_name[0] == '.');
	return new_name;
	}

. Therefore for now we use the checksum to determine whether the name is mangled or not.

[ailzhang]

@PGZXB could you cherrypick this commit into rc-v1.1.0 and send another PR against that branch as well? Thanks! :D