https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122635

--- Comment #18 from Jeffrey A. Law <law at gcc dot gnu.org> ---
We discussed this during the patchwork call today.  The thinking is:

Try to kill the zero length stores in DSE.  The hope/expectation is the
following DCE call after DSE would zap any dead code exposed by removal of the
dead stores.

For loads, if the result was used by a removed store, then the load would be
eliminated by DCE.  There's a sense that we may not have the necessary support
to remove these loads in DCE, but that's the place where we think we should
focus.

So I think the recommendation would be to try Robin's DSE patch, then look and
see what things look like after DSE and DCE.  So perhaps try Robin's DSE patch
and post the resulting DSE/DCE dumps if we're still seeing dead code after DCE.

Reply via email to