This makes sure to perform final value replacement of constants
when we also are sure to propagate those, like in VRP. This avoids
spurious diagnostics when doing so only from within SCCP which can
leave unreachable loops in the IL triggering bogus diagnostics.
The choice is VRP because it alread
On 5/28/21 11:25 AM, Richard Biener wrote:
This makes sure to perform final value replacement of constants
when we also are sure to propagate those, like in VRP. This avoids
spurious diagnostics when doing so only from within SCCP which can
leave unreachable loops in the IL triggering bogus diag
On Mon, 31 May 2021, Andrew MacLeod wrote:
> On 5/28/21 11:25 AM, Richard Biener wrote:
> > This makes sure to perform final value replacement of constants
> > when we also are sure to propagate those, like in VRP. This avoids
> > spurious diagnostics when doing so only from within SCCP which can
On 6/1/21 6:01 AM, Richard Biener wrote:
On Mon, 31 May 2021, Andrew MacLeod wrote:
On 5/28/21 11:25 AM, Richard Biener wrote:
This makes sure to perform final value replacement of constants
when we also are sure to propagate those, like in VRP. This avoids
spurious diagnostics when doing so
I have some old numbers from late April.
VRP vs ranger was more difficult to compare than evrp, since the
gimple is different (ASSERT_EXPRs). What I did was run a late evrp
pass before each VRP pass and compared branches that were folded, for
an estimate.
On it's own VRP1 could fold 5482 branche