| Message ID | 54a5acc7-0629-32c1-c20e-b922e8809aaa@codesourcery.com |
|---|---|
| State | New |
| Headers | show |
| Series | Fix AArch64 ILP ICE | expand |
On Sep 25 2018, Andrew Stubbs <ams@codesourcery.com> wrote: > Ensure that the address really is the correct mode for an address. > > 2018-09-25 Andrew Stubbs <ams@codesourcery.com> > > gcc/ > * builtins.c (get_builtin_sync_mem): Force address mode conversion. This has survived bootstrap so far. Andreas.
On Tue, Sep 25, 2018 at 4:25 PM Andrew Stubbs <ams@codesourcery.com> wrote: > > On 22/09/18 19:51, Andreas Schwab wrote: > > That breaks aarch64 ILP32. > > The problem is that the mode given to expand_expr is just a "hint", > apparently, and it's being ignored. > > I'm testing the attached patch for GCN. It fixes the ICE for AArch64 > just fine. > > OK? OK. > Andrew
On 26/09/18 14:38, Richard Biener wrote:
> OK.
Committed, thanks.
Andrew
Fix AArch64 ILP32 ICE. Ensure that the address really is the correct mode for an address. 2018-09-25 Andrew Stubbs <ams@codesourcery.com> gcc/ * builtins.c (get_builtin_sync_mem): Force address mode conversion. diff --git a/gcc/builtins.c b/gcc/builtins.c index 1d4de09..956f872 100644 --- a/gcc/builtins.c +++ b/gcc/builtins.c @@ -5869,6 +5869,7 @@ get_builtin_sync_mem (tree loc, machine_mode mode) scalar_int_mode addr_mode = targetm.addr_space.address_mode (addr_space); addr = expand_expr (loc, NULL_RTX, addr_mode, EXPAND_SUM); + addr = convert_memory_address (addr_mode, addr); /* Note that we explicitly do not want any alias information for this memory, so that we kill all other live memories. Otherwise we don't
On 22/09/18 19:51, Andreas Schwab wrote: > That breaks aarch64 ILP32. The problem is that the mode given to expand_expr is just a "hint", apparently, and it's being ignored. I'm testing the attached patch for GCN. It fixes the ICE for AArch64 just fine. OK? Andrew