Questions on inlining of code

Questions on inlining of code

[Steve Snyder]

Two question on inlining of code in egcs/pgcc:

1. Is it possible to disable automatic inlining (compiler switches -O3 or 
-finline) while still respecting the inline declaration in the source code?
I'd like to compile with max optimization (-O6) while avoiding the bloat 
that comes with aggressive inlining of code.  At the same time, though, I 
don't want to disabled the inlining of code explicitly declared as such.

2. Does aggressive inlining of code make any sense on a Pentium+ CPU?
It obviously helps on a 386/486 by avoiding the call/return instructions. 
I wonder, though, if the inlining of code doesn't just thrash the L2 cache 
on more recent processors.  Given the prevalence these days of 512kb L2 
caches, maybe inlining should not be automatically enabled by higher levels
of optimization?

Thank you.


*** Steve Snyder ***

Re: Questions on inlining of code

[Jeffrey A Law]

  In message < 199903181530.KAA01307@indy3.indy.net >you write:
  > Two question on inlining of code in egcs/pgcc:
  > 
  > 1. Is it possible to disable automatic inlining (compiler switches -O3 or 
  > -finline) while still respecting the inline declaration in the source code?
  > I'd like to compile with max optimization (-O6) while avoiding the bloat 
  > that comes with aggressive inlining of code.  At the same time, though, I 
  > don't want to disabled the inlining of code explicitly declared as such.
The only difference between -O2 and -On for n > 2 is automatic function
inlining.  So, just use -O2.


  > 2. Does aggressive inlining of code make any sense on a Pentium+ CPU?
Maybe.  Depends on the precise situation.

jeff

Questions on inlining of code

[Steve Snyder]

Two question on inlining of code in egcs/pgcc:

1. Is it possible to disable automatic inlining (compiler switches -O3 or 
-finline) while still respecting the inline declaration in the source code?
I'd like to compile with max optimization (-O6) while avoiding the bloat 
that comes with aggressive inlining of code.  At the same time, though, I 
don't want to disabled the inlining of code explicitly declared as such.

2. Does aggressive inlining of code make any sense on a Pentium+ CPU?
It obviously helps on a 386/486 by avoiding the call/return instructions. 
I wonder, though, if the inlining of code doesn't just thrash the L2 cache 
on more recent processors.  Given the prevalence these days of 512kb L2 
caches, maybe inlining should not be automatically enabled by higher levels
of optimization?

Thank you.


*** Steve Snyder ***

Re: Questions on inlining of code

[Jeffrey A Law]

  In message < 199903181530.KAA01307@indy3.indy.net >you write:
  > Two question on inlining of code in egcs/pgcc:
  > 
  > 1. Is it possible to disable automatic inlining (compiler switches -O3 or 
  > -finline) while still respecting the inline declaration in the source code?
  > I'd like to compile with max optimization (-O6) while avoiding the bloat 
  > that comes with aggressive inlining of code.  At the same time, though, I 
  > don't want to disabled the inlining of code explicitly declared as such.
The only difference between -O2 and -On for n > 2 is automatic function
inlining.  So, just use -O2.


  > 2. Does aggressive inlining of code make any sense on a Pentium+ CPU?
Maybe.  Depends on the precise situation.

jeff

Re: Questions on inlining of code

[Jason Merrill]

>>>>> Steve Snyder <ssnyder@indy.net> writes:

 > Two question on inlining of code in egcs/pgcc:
 > 1. Is it possible to disable automatic inlining (compiler switches -O3 or 
 > -finline) while still respecting the inline declaration in the source code?

The switch you want is -fno-inline-functions; -fno-inline disables all
inlining.

Jason

Re: Questions on inlining of code

[N8TM]

In a message dated 3/18/99 7:31:44 AM Pacific Standard Time, ssnyder@indy.net
writes:

> 2. Does aggressive inlining of code make any sense on a Pentium+ CPU?
>  It obviously helps on a 386/486 by avoiding the call/return instructions. 
>  I wonder, though, if the inlining of code doesn't just thrash the L2 cache 
>  on more recent processors.  Given the prevalence these days of 512kb L2

I suspect your context is different from mine, but I would answer that in-
lining is likely to be useful only in inner loops.  I would point out also
that there is little benefit in going beyond -Os with the "stable release"
compilers, and seldom more than 10% with the snapshots.  Depending on your
source code, you may be able to exert some control over excessive in-lining by
splitting it into separately compiled files, using includes as necessary.

Re: Questions on inlining of code

[N8TM]

In a message dated 3/18/99 7:31:44 AM Pacific Standard Time, ssnyder@indy.net
writes:

> 2. Does aggressive inlining of code make any sense on a Pentium+ CPU?
>  It obviously helps on a 386/486 by avoiding the call/return instructions. 
>  I wonder, though, if the inlining of code doesn't just thrash the L2 cache 
>  on more recent processors.  Given the prevalence these days of 512kb L2

I suspect your context is different from mine, but I would answer that in-
lining is likely to be useful only in inner loops.  I would point out also
that there is little benefit in going beyond -Os with the "stable release"
compilers, and seldom more than 10% with the snapshots.  Depending on your
source code, you may be able to exert some control over excessive in-lining by
splitting it into separately compiled files, using includes as necessary.

Re: Questions on inlining of code

[Jason Merrill]

>>>>> Steve Snyder <ssnyder@indy.net> writes:

 > Two question on inlining of code in egcs/pgcc:
 > 1. Is it possible to disable automatic inlining (compiler switches -O3 or 
 > -finline) while still respecting the inline declaration in the source code?

The switch you want is -fno-inline-functions; -fno-inline disables all
inlining.

Jason