For those who have been following my adventure in the world of IRA, or
just
want to take advantage of some of the time I've spent on this, here's a
detailed recap of my exploits and my current status (which I'm very
happy
with):
I initially thought IRA was calculating costs wrongly for our
Jeff Law wrote:
I had an epiphany this morning and came up with an idea to achieve
the
lookahead I thought I needed, thereby making the costs created by '?'
a
lot more concrete and reliable.
Firstly, I have altered the alt_cost adjustment (for '?') in ira-
costs.c,
so that it only
On 12/02/09 13:29, Ian Bolton wrote:
I had an epiphany this morning and came up with an idea to achieve the
lookahead I thought I needed, thereby making the costs created by '?' a
lot more concrete and reliable.
Firstly, I have altered the alt_cost adjustment (for '?') in ira-costs.c,
so that
Jeff Law wrote:
Ian Bolton wrote:
Initial results showed that IRA was moving input arguments out of
their
BOTTOM_REGS (e.g. $c1) into TOP_CREGS to do work on them, since it
thought TOP_CREGS were less costly to use, despite the cost of the
move
instruction to get the input argument
On 11/19/09 08:40, Ian Bolton wrote:
Jeff Law wrote:
On 11/16/09 10:33, Ian Bolton wrote:
The question is: how to fix this? I have initially put the
REG_ALLOC_ORDER
back to how it was and changed the operand constraints in our MD
file,
so each of the
Jeff Law wrote:
On 11/16/09 10:33, Ian Bolton wrote:
The question is: how to fix this? I have initially put the
REG_ALLOC_ORDER
back to how it was and changed the operand constraints in our MD
file,
so each of the apathetic instructions, will either take a 't'
(TOP_CREG)
or '?b'
Ian Bolton wrote:
Ian Bolton wrote:
Vladimir Makarov wrote:
Ian Bolton wrote:
Yesterday, I wrote:
BTW, today I have achieved some good results with existing IRA
by
doing
the following:
1) Changed the REG_ALLOC_ORDER so that TOP_CREGS are given out
before
Yesterday, I wrote:
BTW, today I have achieved some good results with existing IRA by doing
the following:
1) Changed the REG_ALLOC_ORDER so that TOP_CREGS are given out before
BOTTOM_REGS. My previous hacked version worked by increasing the
priority of those that wanted BOTTOM_REGS, so
On 11/11/09 08:18, Ian Bolton wrote:
Good news! I have been able to remove my total_costs hack
above by instead commenting out the following line from ira_build()
in ira-build.c:
remove_unnecessary_regions (false);
Which is probably an indication of a problem elsewhere. However, it's
Ian Bolton wrote:
Yesterday, I wrote:
BTW, today I have achieved some good results with existing IRA by doing
the following:
1) Changed the REG_ALLOC_ORDER so that TOP_CREGS are given out before
BOTTOM_REGS. My previous hacked version worked by increasing the
priority of those that wanted
Vladimir Makarov wrote:
Ian Bolton wrote:
Yesterday, I wrote:
BTW, today I have achieved some good results with existing IRA by
doing
the following:
1) Changed the REG_ALLOC_ORDER so that TOP_CREGS are given out
before
BOTTOM_REGS. My previous hacked version worked by increasing
Ian Bolton wrote:
Vladimir Makarov wrote:
Ian Bolton wrote:
Yesterday, I wrote:
BTW, today I have achieved some good results with existing IRA by
doing
the following:
1) Changed the REG_ALLOC_ORDER so that TOP_CREGS are given out
before
BOTTOM_REGS. My previous hacked
On Mon, 2009-11-09 at 14:13 +, Ian Bolton wrote:
Dave Hudson wrote:
I've been working on gcc for an ISA (Ubicom32) that seems to have some
similarities to the problem you're seeing (we have some regs that can
be
used for many things but not all) and was seeing a ton a pointless
On 11/06/09 05:53, Dave Hudson wrote:
the most
critical thing was to ensure that REGNO_REG_CLASS was returning a
minimal class correctly.
I believe that's been documented as the right thing to do for about 15
years :-) So, yes, you definitely want REGNO_REG_CLASS to return the
smallest
On 11/06/09 05:53, Dave Hudson wrote:
the most
critical thing was to ensure that REGNO_REG_CLASS was returning a
minimal class correctly.
I believe that's been documented as the right thing to do for about 15
years :-) So, yes, you definitely want REGNO_REG_CLASS to return the
[ Per Ian's request I'm forwarding some discussion we had on IRA to the
list. ]
On 11/04/09 12:24, Ian Bolton wrote:
Hi Jeff and Vladimir,
(I wasn't sure whether to CC the mailing list. Your guidance on this
would be appreciated, since I am very new to the GNU community.)
Your choice.
Dave Hudson wrote:
On Thu, 2009-11-05 at 18:05 +, Ian Bolton wrote:
I think I may have made a breakthrough!
As mentioned above, IRA is correctly increasing the cost for TOP_REGS
when an allocno in region 1 is being used in one of our special
instructions that needs BOTTOM_REGS. We
Jeff Law wrote:
On 11/03/09 09:29, Ian Bolton wrote:
Hi again, Vladimir,
I am pleased to report some performance improvements after altering
ira-costs.c. A key benchmark for us has improved by 5%.
Specifically, in record_reg_classes(), after the alt_cost has been
calculated and it will be
!
Is there an ETA on when reload will be gone away? ;-)
Best regards,
Ian
-Original Message-
From: Jeff Law [mailto:l...@redhat.com]
Sent: 22 October 2009 23:05
To: Ian Bolton
Cc: Vladimir Makarov; gcc@gcc.gnu.org
Subject: Re: Understanding IRA
On 10/19/09 12:30, Ian Bolton wrote:
Hi Jeff
On 11/04/09 10:52, Ian Bolton wrote:
Hi Jeff,
From an empirical perspective, the value of your patch is hard to
determine at this stage - one benchmark improved about 0.5% but others
have marginally regressed.
It's a hack, no doubt about it. Your results are about what I expected,
a few
On 11/04/09 10:14, Vladimir Makarov wrote:
I am agree with Jeff. It is hard to understand what you are doing
without the architecture knowledge and some macro values in your port
(IRA_COVER_CLASSES, MEMORY_MOVE_COST, and REGISTER_MOVE_COST).
I'd also add that besides right macro value
Hi again, Vladimir,
I am pleased to report some performance improvements after altering
ira-costs.c. A key benchmark for us has improved by 5%.
Specifically, in record_reg_classes(), after the alt_cost has been
calculated and it will be applied to pp-mem_cost and pp-cost[k], I
check whether
On 11/03/09 09:29, Ian Bolton wrote:
Hi again, Vladimir,
I am pleased to report some performance improvements after altering
ira-costs.c. A key benchmark for us has improved by 5%.
Specifically, in record_reg_classes(), after the alt_cost has been
calculated and it will be applied to
On 10/19/09 12:30, Ian Bolton wrote:
Hi Jeff and Vladimir.
Jeff: I'd be interested in trying the patch if you can send it my way.
It's nothing special.
/* Return nonzero if REGNO is a particularly bad choice for reloading X. */
static int
ira_bad_reload_regno_1 (int regno, rtx x)
{
int
: Understanding IRA
On 10/16/09 08:53, Vladimir Makarov wrote:
The biggest problem of GCC RA is still reload pass. It frequently
changes decisions of IRA not in a good way.
Agreed. Not only may reload make a bad choice, it's horribly
unpredictable. Trivial changes often lead to drastically different
Ian Bolton wrote:
Hi Jeff and Vladimir.
Jeff: I'd be interested in trying the patch if you can send it my way.
Vladimir: Today I have seen reasonable improvements on
progressively-trimmed-down versions of a *single* test case by applying
some cost adjustments in the case when an operand wants
Hi Vladimir,
I have just begun working for Icera Inc, on a private port of GCC,
and my first task has been to investigate the performance impact of
using the Chaitin-Briggs algorithm of IRA vs the priority algorithm
that we currently have enabled in our production compiler (due to not
defining
27 matches
Mail list logo
