inline template functions

[Martin Kaul <martin@xxxxxxxxxxxxxxx>]

Hi

when I compile the following piece of code:

-------------------------------------
template< typename T >
inline
T max( T x1, T x2 )
{
  return ( x1 >= x2 ) ? x1 : x2;
}

int testFunction( int x1 )
{
  return max( x1, 20 );
}
-------------------------------------

The gcc (arm-elf-gcc 3.4.2) do not copy the content of the function 
max() into testFunction() as expected cause of inlining.

  14              	_Z12testFunctioni:
  21 0000 0DC0A0E1 		mov	ip, sp
  23 0004 00D82DE9 		stmfd	sp!, {fp, ip, lr, pc}
  27 0008 1410A0E3 		mov	r1, #20
  30 000c 04B04CE2 		sub	fp, ip, #4

  34 0010 FEFFFFEB 		bl	_Z3maxIiET_S0_S0_

  37 0014 00689DE8 		ldmfd	sp, {fp, sp, lr}
  38 0018 1EFF2FE1 		bx	lr

  45              	_Z3maxIiET_S0_S0_:
  53 0000 010050E1 		cmp	r0, r1
  54 0004 0100A0B1 		movlt	r0, r1
  58 0008 1EFF2FE1 		bx	lr

The assembler code above shows that _Z3maxIiET_S0_S0_ is called instead 
of putting the two lines (cmp & movlt) directly into Z12testFunctioni.

When I remove the template Parameter and change max() to

inline
int max( int x1, int x2 )
{
  return ( x1 >= x2 ) ? x1 : x2;
}

the inlining works perfectly.

  14              	_Z12testFunctioni:
  23 0000 140050E3 		cmp	r0, #20
  24 0004 1400A0B3 		movlt	r0, #20
  28 0008 1EFF2FE1 		bx	lr

12 Bytes of code instead of 40 Bytes code :-/


I looked into the manual and found the option 
-fno-implicit-inline-templates but both variants 
(-fno-implicit-inline-templates and -fimplicit-inline-templates) 
produces the same code.

I compiled with -O2

how can I avoid this non-optimal "feature" of gcc ;-)

regards
  Martin Kaul