[llvm-dev] [EXT] Please expose predicates to MachineVerifier

Eli Friedman via llvm-dev llvm-dev at lists.llvm.org
Mon Apr 1 11:33:26 PDT 2019

There's already an existing hook for adding target-specific verification logic in the target's TargetInstrInfo::verifyInstruction.  That doesn't really scale to checking that you're using an appropriate subtarget for every instruction, though.  (You could, but it would involve a lot of redundant hand-written code for many targets.)

We can't really use any of the existing code generated by TableGen to perform the checks in question; I don't think we currently expose the relevant predicates in any useful way.  But it would probably be appropriate to extend TableGen to generate code for those checks, I think.  It would probably also require some modifications to the TableGen files themselves; currently, at least for some targets, we use Requires predicates on instructions which are legal, but we don't want the TableGen'ed isel to select for some reason.


From: llvm-dev <llvm-dev-bounces at lists.llvm.org> On Behalf Of Mark Schimmel via llvm-dev
Sent: Monday, April 1, 2019 11:21 AM
To: llvm-dev at lists.llvm.org
Subject: [EXT] [llvm-dev] Please expose predicates to MachineVerifier

Could we expose predicates defined in the target InstrInfo.td file to the MachineVerifier? We use BuildMI() to create many instructions after ISEL, but the predicates are not being checked at this point. Thus, I could forget to check the target and build an instruction that is illegal for a specific configuration. In such a case it would be nice if the MachineVerifier could detect this for me.

Example predicate IsCore8 usage:

def ADDx: BINOP<...>, Requires<[IsCore8]>;

let Predicates=[IsCore8] in {
def : SUBX: BINOP<...>;

The predicates are encoded into the targets GenDAGIsel file but not anywhere else that I can find:
  OPC_CheckPatternPredicate, 4, // (Subtarget.isCore8())


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20190401/b159c012/attachment.html>

More information about the llvm-dev mailing list