On 11/06/2015 03:19 PM, H.J. Lu wrote:
On Fri, Nov 6, 2015 at 7:31 AM, Uros Bizjak <ubiz...@gmail.com> wrote:
On Fri, Nov 6, 2015 at 3:07 PM, Yulia Koval <vaalfr...@gmail.com> wrote:
Hi,
I updated and reposted the patch. Regtested/bootstraped on
x86_64/Linux and i686/Linux. Ok for trunk?
This version still emits insns from ix86_function_arg, so NAK.
Uros.
Hi Uros,
The ix86_function_arg change only applies the interrupt function body,
which has fixed parameters and won't be called by any functions directly.
Uros was pretty clear that generating code from within ix86_function_arg
was a NAK from his side (and I'd agree with that assessment from Uros).
Generating code like that from within FUNCTION_ARG is pretty bad.
Can you arrange to load up the arguments from within the prologue so
that they look like they were passed in? Alternately you have to make
FUNCTION_ARG return 0 for those cases and arrange to define all the
other argument passing offsets appropriately so those arguments can be
found in the stack.
Jeff
