https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104117
Iain Sandoe <iains at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Host|powerpc-apple-darwin9 |
Target Milestone|--- |10.4
Keywords| |wrong-code
Build|powerpc-apple-darwin9 |
Last reconfirmed| |2022-01-19
--- Comment #1 from Iain Sandoe <iains at gcc dot gnu.org> ---
most likely present on earlier branches.
The problem is that we have
double var = (double) 1;
LRA decides that loading a constant integer 1 into a FPR and then using
int->double is good. However, it (a) decides to do this in the prologue
(before we have got a picbase set up) and (b) does not seem to be checking that
the target agrees that the addressing mode is legitimate (it is not).
----
NOTE: incidental changes on 11.x and master have hidden this (but the changes
were not a fix for this bug, so it has most likely simply become latent).
