[llvm-commits] [Patch] Make TableGen put chains on more realistic instrucitons

[Jakob Stoklund Olesen]

On Jun 24, 2012, at 11:10 AM, Tim Northover <t.p.northover at gmail.com> wrote:

> Hi,
> 
> Recently on the dev list an issue came up where TableGen wasn't giving
> load instructions (as an example) chains when their only use was in
> Pat instantiations. TableGen's logic for this decision currently
> relies on the Pattern field being set in the destination Instruction.
> 
> This patch includes a sensible subset of the possible flags in the
> decision: load/store, side effects and the (oddly unused right now)
> hasCtrlDep which seems to be exactly what we want.
> 
> Including those extra flags almost certainly reduces the risk of wrong
> code generation, though it does limit the flexibility of the system:
> if anyone wanted a loading instruction which didn't take a chain they
> would need to make the logic more flexible. Since we couldn't come up
> with an actual user of this functionality, it's probably best not to
> complicate things for now.
> 
> Ok to commit?

Yes, LGTM. Thanks.

/jakob