From: nimbus@nospicedham.XXX.invalid   
      
   Frank Kotler dit dans news:ois44e$njp$1@dont-email.me:   
      
   > Well, just for completeness... The Fortran code is not mine.   
   Got it off   
   > the Forum.   
      
   >        program testassembly   
   >        implicit none   
   >        integer*8 summ,a,b   
   >        a=2000000000   
   >        b=2000000000   
   >        write(6,*)summ(a,b)   
   >        end   
      
   > The poster was trying to link against 64-bit code (as I   
   recall he was   
   > getting a segfault). This is my 32-bit version.   
      
   > section .text   
   > global summ_  ; leading underscore too for Windows   
   >   
   > summ_:   
   >      push ebp   
   >      mov ebp, esp   
   >   
   >      mov eax, [ebp + 8] ; address of first parameter   
   >      mov ecx, [ebp + 12] ; address of second parameter   
   >      mov ebx, [eax+4] ; first parameter   
   >      mov eax, [eax]   
   >      mov edx, [ecx+4] ; second parameter   
   >      mov ecx, [ecx]   
   >      add eax, ecx ; add 'em up   
   >      adc edx, ebx   
   > ; return result in edx:eax?   
   >   
   >      mov esp, ebp   
   >      pop ebp   
   >      ret   
      
   Looks fine, save maybe - not sure if trashing EBX in a   
   function is safe - conventionnally use of EAX,ECX,EDX is   
   allowed. Here's FWIW my version of summing two integer*8,   
   based of Bogdan's integer*4 one upthread reads :   
   ____________________________________________________   
   	section .text use32   
   	global	_sum64_   
   	_sum64_:		;  _2_  _underscores_ for Win32   
      
   	%idefine	a	[ebp+8]   
   	%idefine	b	[ebp+12]   
      
   		push	ebp   
   		mov	ebp, esp   
      
   	; assume both arguments of integer*8 type (64 bits)   
   	; Return sum in (EDX,EAX)   
   		mov	ecx, a		; ECX = address of param1   
   		mov	eax, [ecx]	; EAX = param1 dereferenced (LO   
   DW)   
   		 mov	edx,[ecx+4]; ECX= param1 (hi DW)   
   		mov	ecx, b   
   		add	eax, [ecx]   
   		 adc	edx,[ecx+4]   
   	; LEAVE = mov esp, ebp / pop ebp   
   		leave   
   		ret   
   ____________________________________________________   
      
      
   > I assemble the .asm file first, then "g77 callsum.f summ.o -   
   o callsum"   
   > (no Makefile for this). As you can see, the .asm file   
   "expects" 64 bits   
   > and attempts to return 64 bits. This "seems to work" but g77   
   complains   
   > if I use a larger integer.   
      
   Do you mean using an integer /constant/ larger, in absolute   
   value, than 31-bits in the Fortran source ? That's a   
   misfeature of g77, you /cannot specify/ such a constant   
   directly to the compiler. Need to use a variable and either   
   have the large integer computed at run time, or play some   
   trickery with  Fortran EQUIVALENCE...   
      
   > Perhaps "implicit none" is a problem? I don't   
   > even know how to comment it out!   
      
   'Implicit none' is accepted by g77. As for commenting a line   
   out, in Fortran there are several ways, historically it has   
   been by "punching" a 'C' in the first column of the "card"   
   (source line)  :=)   
      
   Regards   
      
   --   
   NimbUs   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)