[PATCH] D55281: debuginfo: Use symbol difference for CU length to simplify assembly reading/editing

Tue Dec 18 08:44:58 PST 2018

dblaikie added a comment.

In D55281#1334452 <https://reviews.llvm.org/D55281#1334452>, @probinson wrote:

> In D55281#1333802 <https://reviews.llvm.org/D55281#1333802>, @dblaikie wrote:
>
> > Still a bit curious about the other discussion about flags, etc.
>
>
> Well, you have better access to Eric than the rest of us....

Useful as a tie breaker, but I'm not sure it's that big of a deal we can't decide amongst ourselves. But yeah, since he approved the patches originally, might poke him about it at some point.

> Certainly having cl::opt flags makes it easier to test things in a non-platform-dependent way, if nothing else.  I can see putting all the ptxas-special stuff under a single "use-ptxas-format" flag instead of several different ones.  It makes the control less feature-oriented, but more obvious about the purpose/intent.

I'm just still not sure the extra cl::opt is worth it, though - are you? Is the extra coverage significant/important?
I think the only reason we have a flag for the debugger tuning is that it can't always be deduced from the triple (eg: folks using LLDB on Linux, etc) - but this case can, it's always NVPTX, there's nothing else currently possible to do with that target other than emit assembly that goes to ptxas.
But yeah, as a developer/testing-only feature I don't object too much to a use-ptxas-format flag to cover all this, but I still think it opens up a "wait, is there some other way/reason to be using ptxas other than for the NVPTX target?".

Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D55281/new/

https://reviews.llvm.org/D55281