ABataev marked an inline comment as done.
ABataev added inline comments.
================
Comment at: clang/include/clang/Basic/DiagnosticParseKinds.td:1230
- "unknown '%0' device kind trait in the 'device' context selector set,
expected"
- " one of 'host', 'nohost', 'cpu', 'gpu' or 'fpga'">;
----------------
jdoerfert wrote:
> ABataev wrote:
> > jdoerfert wrote:
> > > ABataev wrote:
> > > > jdoerfert wrote:
> > > > > I would have expected this error to be still accurate, maybe with the
> > > > > addition ", or quoted versions thereof".
> > > > Currently, we could emit it only in codegen phase to avoid double
> > > > converting from expression to string. Will emit it there.
> > > Why can't we emit this error if the user writes `device={kind(gggggpu)}`
> > > anymore? Even `device={kind("gggggpu")}` should be diagnosable early.
> > The main problem here is the conversion and expression evaluation. We
> > convert the data from the expression to strings at the codegen phase. I
> > don't want to do the same thing for the second time in Sema just for
> > diagnostic.
> We have to convert it during sema already, actually during parsing. I'm
> working on declare variant begin/end support right now (part of TR8) which
> needs the information during parsing.
Hmmm, it is where we need to parse the code only if the context matches the
trait? Ok, will rework to convert it in sema too. Seems to me, we'll need to
match the context in both, codegen and sema. Will try to move the matcher
somewhere to avoid duplication in sema and codege.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D70973/new/
https://reviews.llvm.org/D70973
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
