PING: [df] Always mark the stack pointer as defined on entry.

PING: [df] Always mark the stack pointer as defined on entry.

[Richard Sandiford]

http://gcc.gnu.org/ml/gcc-patches/2008-08/msg02236.html
Tell DF that the incoming stack pointer is always important.
The current code assumes that the initial value of the stack
pointer isn't relevant once the prologue has been generated.

Richard

Re: PING: [df] Always mark the stack pointer as defined on entry.

[Richard Sandiford]

"H.J. Lu" <hjl.tools@gmail.com> writes:
> On Wed, Sep 3, 2008 at 12:06 PM, Richard Sandiford
> <rdsandiford@googlemail.com> wrote:
>> http://gcc.gnu.org/ml/gcc-patches/2008-08/msg02236.html
>> Tell DF that the incoming stack pointer is always important.
>> The current code assumes that the initial value of the stack
>> pointer isn't relevant once the prologue has been generated.
>>
>
> Do you have any testcases to show how your change affects
> generated codes?

The testcase is in the R10K speculation patch:

    http://gcc.gnu.org/ml/gcc-patches/2008-09/msg00041.html

(Sorry, although I referred vaguely to that in the original submission,
I should have mentioned it by name now that the R10K patch has been
posted.)

> How will your change interact with dynamic stack alignment?

I don't believe it will.  This patch only affects the behaviour
_after_ prologue and epilogue generation.  My understanding is
that all dsa-specific stuff will have been done by then.

Richard

Re: PING: [df] Always mark the stack pointer as defined on entry.

[H.J. Lu]

On Wed, Sep 3, 2008 at 12:06 PM, Richard Sandiford
<rdsandiford@googlemail.com> wrote:
> http://gcc.gnu.org/ml/gcc-patches/2008-08/msg02236.html
> Tell DF that the incoming stack pointer is always important.
> The current code assumes that the initial value of the stack
> pointer isn't relevant once the prologue has been generated.
>

Do you have any testcases to show how your change affects
generated codes? How will your change interact with dynamic
stack alignment?


-- 
H.J.

Re: PING: [df] Always mark the stack pointer as defined on entry.

[Richard Henderson]

Richard Sandiford wrote:
> http://gcc.gnu.org/ml/gcc-patches/2008-08/msg02236.html

Ok.


r~

RE: PING: [df] Always mark the stack pointer as defined on entry.

[Ye, Joey]

Richard Wrote:
> "H.J. Lu" <hjl.tools@gmail.com> writes:
>> How will your change interact with dynamic stack alignment?
>
> I don't believe it will.  This patch only affects the behaviour
> _after_ prologue and epilogue generation.  My understanding is
> that all dsa-specific stuff will have been done by then.
Looks OK for stack alignment.

Thanks - Joey