Fix top LLVM: renamed NVPTX barrier intrinsics.

[mcourteaux]

llvm/llvm-project#140615 was temporarily reverted (but we haven't pulled yet) and will be relanded soon here llvm/llvm-project#141143.
This PR introduces a bit of logic that handles both intrinsic naming possibilities, which does not depend on LLVM version, but just tries both, to assure this doesn't break in any LLVM version. I tried locally on my non-updated LLVM that this still works. We can remove this change once we drop support for LLVM 19.

This patch is confirmed to work in the build bots here: #8629, but I'm opening a separate PR to separate the concerns.

Can be merged without testing I believe. @steven-johnson or @abadams, add the build bots for review is you want to.

[steven-johnson]

Suggestion: add a comment that specifically says "this can be removed once we no longer support LLVM19 or earlier"

[mcourteaux]

Did that, and reordered the flow of the checks to favor the up-to-date check first.

[abadams]

I'm surprised this worked, because I thought module->getFunction will get declarations, regardless of whether or not the llvm backend will eventually provide definitions for them. Is it possible the new name has worked for a while?

[alexreinking]

Shouldn't this be guarded by preprocessor macros, anyway?

[mcourteaux]

I'm surprised this worked, because I thought module->getFunction will get declarations, regardless of whether or not the llvm backend will eventually provide definitions for them. Is it possible the new name has worked for a while?

Well, the original code on main fails with that function returning null. So it seems that this is not considering the declarations we have in ptx_dev.ll. But on the other hand... it still didn't return the new intrinsic when I didn't have it declared in ptx_dev.ll. Based on what I just experienced, it seemed like llvm::Module::getFunction() only returns a function if it both is declared and it has an implementation.

[mcourteaux]

Under LLVM 19, the invalid PTX is indeed quite invalid...

	{ // callseq 2, 0
	.param .b32 param0;
	st.param.b32 	[param0], 0;
	call.uni 
	llvm.nvvm.barrier.cta.sync.aligned.all, 
	(
	param0
	);
	} // callseq 2

Seems like it does return the declaration...