If you define INLINER this way, it works:
#define INLINER static inline __attribute__((always_inline))
For that matter, this works too:
#define INLINER static inline
Without 'static' in there, the function is exported. If the
function is exported, its implementation has to exist so that
other translation units may call it. The compiler could still
inline it at some call sites and leave the out of line
implementation in place, but that'd duplicate code and I'm not
sure if the compiler ever thinks that would be a win.
If you're worried about making this change in tons of files, the
easiest fix is this:
#define inline static inline
Put that in a prefix file and you're all set.
Hope this helps,
Eric
On Oct 6, 2006, at 8:37 PM, Chris Edgington wrote:
Thanks for the quick responses ... I'll answer your questions ...
1) The code is straight C
2) I've tried lots of different things - static inline, extern
inline, compiler options, ...
3) I understand that inline isn't the performance gain that it
used to be - but this is code that has evolved over the last 6-8
years
4) I suspect that static inline is ultimately the best
workaround - the problem is that all the functions that are
inlined also have a prototype that does not include the static
keyword - so when I try to use static inline the compiler
complains that the function doesn't match the prototype.
Again - the goal here is to get this code to compile/work
without changing it - 'cause there are over a HUNDRED files that
use this format. So, if I can accomplish this by redefining
__inline as a macro that resolves to something else - than I'm
home free without having to change lots of source.
Now - on to my research this evening. It looks like I cannot get
gcc to inline a function, no matter what I do - and that the
errors I'm seeing are likely a result of that fact. Here is some
code:
----------
#include <stdio.h>
#define INLINER inline __attribute__((always_inline))
INLINER void macfn1(void)
{
printf("macfn1-->\n");
return;
}
void macfn2(void)
{
printf("macfn2-->\n");
macfn1();
return;
}
----------
Here is the assembly produced by compiling that code:
.cstring
LC0:
.ascii "macfn1-->\0"
.text
.globl _macfn1
_macfn1:
pushl %ebp
movl %esp, %ebp
pushl %ebx
subl $20, %esp
call ___i686.get_pc_thunk.bx
"L00000000001$pb":
leal LC0-"L00000000001$pb"(%ebx), %eax
movl %eax, (%esp)
call L_puts$stub
addl $20, %esp
popl %ebx
popl %ebp
ret
.cstring
LC1:
.ascii "macfn2-->\0"
.text
.globl _macfn2
_macfn2:
pushl %ebp
movl %esp, %ebp
pushl %ebx
subl $20, %esp
call ___i686.get_pc_thunk.bx
"L00000000002$pb":
leal LC1-"L00000000002$pb"(%ebx), %eax
movl %eax, (%esp)
call L_puts$stub
call _macfn1
addl $20, %esp
popl %ebx
popl %ebp
ret
.section __IMPORT,__jump_table,symbol_stubs,self_modifying_code
+pure_instructions,5
L_puts$stub:
.indirect_symbol _puts
hlt ; hlt ; hlt ; hlt ; hlt
.subsections_via_symbols
.section __TEXT,__textcoal_nt,coalesced,pure_instructions
.weak_definition ___i686.get_pc_thunk.bx
.private_extern ___i686.get_pc_thunk.bx
___i686.get_pc_thunk.bx:
movl (%esp), %ebx
ret
-----------
As you can see, the call to macfn1 is not inlined - even though
I'm clearly defining the function as having the always_inline
attribute.
I'm using the stock gcc in latest OSX dev environment (Tiger
10.4.8).
Here's the output during the make - so you can see what the
compiler is saying:
gcc -c -S -v macinline1.c
Using built-in specs.
Target: i686-apple-darwin8
Configured with: /private/var/tmp/gcc/gcc-5363.obj~28/src/
configure --disable-checking -enable-werror --prefix=/usr --
mandir=/share/man --enable-languages=c,objc,c++,obj-c++ --
program-transform-name=/^[cg][^.-]*$/s/$/-4.0/ --with-gxx-
include-dir=/include/c++/4.0.0 --with-slibdir=/usr/lib --
build=powerpc-apple-darwin8 --with-arch=nocona --with-
tune=generic --program-prefix= --host=i686-apple-darwin8 --
target=i686-apple-darwin8
Thread model: posix
gcc version 4.0.1 (Apple Computer, Inc. build 5363)
/usr/libexec/gcc/i686-apple-darwin8/4.0.1/cc1 -quiet -v -
D__DYNAMIC__ macinline1.c -fPIC -quiet -dumpbase macinline1.c -
march=nocona -auxbase macinline1 -version -o macinline1.s
ignoring nonexistent directory "/usr/local/include"
ignoring nonexistent directory "/usr/lib/gcc/i686-apple-
darwin8/4.0.1/../../../../i686-apple-darwin8/include"
#include "..." search starts here:
#include <...> search starts here:
/usr/lib/gcc/i686-apple-darwin8/4.0.1/include
/usr/include
/System/Library/Frameworks
/Library/Frameworks
End of search list.
GNU C version 4.0.1 (Apple Computer, Inc. build 5363) (i686-
apple-darwin8)
compiled by GNU C version 4.0.1 (Apple Computer, Inc.
build 5363).
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-
heapsize=131072
Compiler executable checksum: 117f4efba60aa7f11e5d2cf30dda5d40
Any more thoughts?
Thanks,
-Chris
On 6 Oct 2006, at 16:48, Chris Edgington wrote:
I'm porting a large application from Windows to OSX. I'm
trying to minimize changes to my clients codebase. One area
that is giving me problems is inline functions. All over the
place the client has functions in header files marked as
inline (this is a performance-critical app). These header
files get included by multiple .C files. When building with
gcc on OSX - what ends up happening is that the compiler puts
a copy of each inlined function in each .o file, then when I
goto link I get linker errors because of multiple definitions
of the same symbol.
Anybody have any thoughts on how to resolve this?
Thanks,
-Chris
