From: Richard Guenther <richard.guenther@gmail.com>
To: Mark Heffernan <meheff@google.com>
Cc: Xinliang David Li <davidxl@google.com>,
GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [google] pessimize stack accounting during inlining
Date: Fri, 10 Jun 2011 09:14:00 -0000 [thread overview]
Message-ID: <BANLkTiksqbosGkJv2QTrqeBvN4nLLD+dpw@mail.gmail.com> (raw)
In-Reply-To: <BANLkTim-NWd+=mS6KOAzfHTWp8cyZO00VJO0OpD0zEDd68zZUA@mail.gmail.com>
On Fri, Jun 10, 2011 at 1:45 AM, Mark Heffernan <meheff@google.com> wrote:
> On Wed, Jun 8, 2011 at 1:54 AM, Richard Guenther
> <richard.guenther@gmail.com> wrote:
>> Well, it's still not a hard limit as we can't tell how many spill slots
>> or extra call argument or return value slots we need.
>
> Agreed. It's not perfect. But I've found this does a reasonable job
> of preventing the inliner from pushing the frame size much beyond the
> imposed limit especially if the limit is large (eg, many K) relative
> to the typical total size of spill slots, arguments, etc.
Do you have a testcase?
Richard.
> Mark
>
>>
>> Richard.
>>
>> > Thanks,
>> >
>> > David
>> >
>> > On Tue, Jun 7, 2011 at 4:29 PM, Mark Heffernan <meheff@google.com> wrote:
>> >> This patch pessimizes stack accounting during inlining. This enables
>> >> setting a firm stack size limit (via parameters "large-stack-frame" and
>> >> "large-stack-frame-growth"). Without this patch the inliner is overly
>> >> optimistic about potential stack reuse resulting in actual stack frames much
>> >> larger than the parameterized limits.
>> >> Internal benchmarks show minor performance differences with non-fdo and
>> >> lipo, but overall neutral. Tested/bootstrapped on x86-64.
>> >> Ok for google-main?
>> >> Mark
>> >>
>> >> 2011-06-07 Mark Heffernan <meheff@google.com>
>> >> * cgraph.h (cgraph_global_info): Remove field.
>> >> * ipa-inline.c (cgraph_clone_inlined_nodes): Change
>> >> stack frame computation.
>> >> (cgraph_check_inline_limits): Ditto.
>> >> (compute_inline_parameters): Remove dead initialization.
>> >>
>> >> Index: gcc/cgraph.h
>> >> ===================================================================
>> >> --- gcc/cgraph.h (revision 174512)
>> >> +++ gcc/cgraph.h (working copy)
>> >> @@ -136,8 +136,6 @@ struct GTY(()) cgraph_local_info {
>> >> struct GTY(()) cgraph_global_info {
>> >> /* Estimated stack frame consumption by the function. */
>> >> HOST_WIDE_INT estimated_stack_size;
>> >> - /* Expected offset of the stack frame of inlined function. */
>> >> - HOST_WIDE_INT stack_frame_offset;
>> >>
>> >> /* For inline clones this points to the function they will be
>> >> inlined into. */
>> >> Index: gcc/ipa-inline.c
>> >> ===================================================================
>> >> --- gcc/ipa-inline.c (revision 174512)
>> >> +++ gcc/ipa-inline.c (working copy)
>> >> @@ -229,8 +229,6 @@ void
>> >> cgraph_clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
>> >> bool update_original)
>> >> {
>> >> - HOST_WIDE_INT peak;
>> >> -
>> >> if (duplicate)
>> >> {
>> >> /* We may eliminate the need for out-of-line copy to be output.
>> >> @@ -279,13 +277,13 @@ cgraph_clone_inlined_nodes (struct cgrap
>> >> e->callee->global.inlined_to = e->caller->global.inlined_to;
>> >> else
>> >> e->callee->global.inlined_to = e->caller;
>> >> - e->callee->global.stack_frame_offset
>> >> - = e->caller->global.stack_frame_offset
>> >> - + inline_summary (e->caller)->estimated_self_stack_size;
>> >> - peak = e->callee->global.stack_frame_offset
>> >> - + inline_summary (e->callee)->estimated_self_stack_size;
>> >> - if (e->callee->global.inlined_to->global.estimated_stack_size < peak)
>> >> - e->callee->global.inlined_to->global.estimated_stack_size = peak;
>> >> +
>> >> + /* Pessimistically assume no sharing of stack space. That is, the
>> >> + frame size of a function is estimated as the original frame size
>> >> + plus the sum of the frame sizes of all inlined callees. */
>> >> + e->callee->global.inlined_to->global.estimated_stack_size +=
>> >> + inline_summary (e->callee)->estimated_self_stack_size;
>> >> +
>> >> cgraph_propagate_frequency (e->callee);
>> >>
>> >> /* Recursively clone all bodies. */
>> >> @@ -430,8 +428,7 @@ cgraph_check_inline_limits (struct cgrap
>> >>
>> >> stack_size_limit += stack_size_limit * PARAM_VALUE
>> >> (PARAM_STACK_FRAME_GROWTH) / 100;
>> >>
>> >> - inlined_stack = (to->global.stack_frame_offset
>> >> - + inline_summary (to)->estimated_self_stack_size
>> >> + inlined_stack = (to->global.estimated_stack_size
>> >> + what->global.estimated_stack_size);
>> >> if (inlined_stack > stack_size_limit
>> >> && inlined_stack > PARAM_VALUE (PARAM_LARGE_STACK_FRAME))
>> >> @@ -2064,7 +2061,6 @@ compute_inline_parameters (struct cgraph
>> >> self_stack_size = optimize ? estimated_stack_frame_size (node) : 0;
>> >> inline_summary (node)->estimated_self_stack_size = self_stack_size;
>> >> node->global.estimated_stack_size = self_stack_size;
>> >> - node->global.stack_frame_offset = 0;
>> >>
>> >> /* Can this function be inlined at all? */
>> >> node->local.inlinable = tree_inlinable_function_p (node->decl);
>> >>
>> >
>
prev parent reply other threads:[~2011-06-10 9:05 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <BANLkTik0OK=0ksWUosRPGW3x23-OAA34ujryD7iimFZH51x58w@mail.gmail.com>
2011-06-08 1:06 ` Xinliang David Li
2011-06-08 9:11 ` Richard Guenther
2011-06-10 0:08 ` Mark Heffernan
2011-06-10 9:14 ` Richard Guenther [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=BANLkTiksqbosGkJv2QTrqeBvN4nLLD+dpw@mail.gmail.com \
--to=richard.guenther@gmail.com \
--cc=davidxl@google.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=meheff@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).