On 10/08/2015 09:06 PM, Richard Biener wrote:
On Thu, Oct 8, 2015 at 6:59 AM, Richard Henderson <r...@redhat.com> wrote:
I started with Armin Rigo's patch, from back in July,
https://gcc.gnu.org/ml/gcc/2015-07/msg00125.html
but then wound up changing all of it. To wit:
(1) If we use more sensible defaults for address spaces, we no longer
require 8 backend functions and 1 new hook. Indeed, basic support for
x86 segmentation via address spaces then requires no overriding of any
hooks.
(2) In order to do anything useful with x86 segmentation, we need to
acknowledge that the most common usage puts a valid object at %seg:0.
Which means that we do need a new hook to disable several optimizations.
We could of course simply make all but the default address-space fall
under -fno-delete-null-pointer-check rules...
Yes, but that's probably less than optimal for targets whose non-default
address space is simply "far", using a wider pointer mode.
r~
