On 1/4/21 2:20 PM, Jakub Jelinek wrote:
> On Mon, Jan 04, 2021 at 02:10:39PM -0700, Jeff Law wrote:
>>> I explained what the code handles and when in the pipeline in
>>> the discussion of the previous patch:
>>> https://gcc.gnu.org/pipermail/gcc-patches/2020-November/559770.html
>> Right, but that message talks about GC. This is not a GC issue.
>>
>> This feels like we need a SAVE_EXPR to me to ensure single evaluation
>> and an unshare_expr to avoid problems with destructive gimplification.
> unshare_expr will not duplicate SAVE_EXPRs.
> So, one would need to unshare with special handling of SAVE_EXPRs that would
> throw them away (for the simple arguments case) rather than handling them
> normally.
My mental model of how this works must be broken then. I thought we
would need to unshare the expression, then wrap it in the SAVE_EXPR. It
seems like you're saying that we've already got the SAVE_EXPR and that
unshare_expr won't traverse into it. That would indeed be problematical.
jeff
