lenary added a comment.
I'm not convinced we should be leaving any of the other Target Parser
information in Support, if we are doing this, though this creates layering
issues which I've been trying to unpick.
If you try to take `llvm/Support/*TargetParser*`, you find a few places where
there are dependencies from these files to other bits of Support:
- `llvm/Support/Host` has lots of logic for turning cpu information into a
target, to handle `-mcpu=native`. My opinion is that should also move into
target parser. There are some parts of this file that can move into other parts
of `llvm/Support`, like the introspection about number of cores.
- `llvm/{Support/ADT}Triple` depends on the Arm/AArch64 target parsers (because
arm triples canonically look like `armv7a-…` or similar, and we use the
targetparser to help canonicalise the triple). My opinion is that all this code
can move into the target parser.
There are quite a few small cleanup patches to enable this. One of the biggest
is that `llvm/Support/CommmandLine` wants to know the current triple and CPU
for `--version`, even in e.g. tablegen. This was the change I most struggled
with, but I think I have a solution that does the job fine.
I'm going to try to polish and post my patch set this week which does this
split, as a comparison to this one.
Re the name: "TargetParser" seems reasonable to me, given that's what we've
called the files so far anyway.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D137516/new/
https://reviews.llvm.org/D137516
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
