I'd guess that since it enables the compiler to perform speculative
memory accesses, in theory it could cause additional page faults that
were not originally there. So it's probably most appropriate in the
situation where the compiler has profile feedback information that
indicates that the load will eventually be performed.
Regards,
Darryl.
On 3/30/2010 8:36 PM, ольга крыжановская wrote:
Why does this option require -xO5 and not a lesser level like -xO3?
Why is this not the default in ON?
Olga
On Tue, Mar 30, 2010 at 6:45 AM, Darryl Gove<[email protected]> wrote:
Hi,
The flag basically means that memory traps don't get forwarded to the
application. So the kernel still gets to handle the trap, but the app won't
segfault if you access a bad address.
The option is basically to allow the compiler to use speculative memory
accesses.
Regards,
Darryl.
On 3/28/2010 2:42 PM, ольга крыжановская wrote:
Does -xsafe=mem work with mmap() for files? From the documentation it
says that -xsafe=mem turns of the memory traps. Does that include
traps used by the kernel to serve files mapped into mmap() into the
process address space?
Olga
_______________________________________________
tools-compilers mailing list
[email protected]
_______________________________________________
tools-compilers mailing list
[email protected]
