Returning a Float from Assembly to C

MORE INFORMATION

The following sample code shows two methods to return a floating-point value from assembly code to C code. The first method (function Add1), returns the address of a storage location set up by the assembly module. This method is fairly straightforward. The sample uses large-model code, so far pointers are used. If small- or medium-model code is selected, the segment values are not needed.

There is a problem with this technique if the assembly module is contained in a Windows DLL. The data segment in the DLL is not accessible by the calling program, so returning the address as shown in Add1 does not work. An alternative technique (function Add1Again) is also illustrated in the sample code. In this method, the calling program passes an address of a floating-point variable, and the function places the return data at that address. The sample is intended to be built as an MS-DOS program (or a QuickWin), but Add1Again can be used in a DLL.

Sample Build

      rem MAKEFTST.BAT
      rem Batch program to build the test files
   cl /c /AL /Od /Zi floatstc.c
   ml /c /FPi /Zi floatsta.asm
      rem /FPi allows the MASM module to use the C floating point emulator
   link /CO floatstc + floatsta;
      rem Resulting program is FLOATSTC.EXE and can be test in CodeView
      rem if desired
				

Sample Code

   ;assembly language routines
   .MODEL LARGE,C
   .8087

   PUBLIC Add1
   PUBLIC Add1Again

   .FARDATA
   temp REAL4 ?    ; temporary storage for return value

   .CODE

       ; procedure to return a float
   Add1 PROC USES es, y:REAL4
       ; y is the float passed from C
   ASSUME ES:SEG temp
       mov ax, SEG temp
       mov es, ax          ; load the far data segment
       fld DWORD PTR y     ; load the coprocessor
       fld1
       faddp ST(1),ST      ; add 1 to the value passed in
       fstp DWORD PTR es:temp  ; store the value to temporary storage
       mov dx, SEG temp        ; load DX:AX with the address of temp
       mov ax, OFFSET temp
           ; this is what the C module expects as a return value
       ret
   Add1 ENDP

       ; procedure to do the same things
       ; but return a value at the passed address
   Add1Again PROC USES es bx, y:REAL4, z:DWORD
       ; y is the float passed from C
       ;z is the address of a float in the C module
   ASSUME ES:NOTHING
       les bx, z       ; load ES:BX with the address of the return location
       fld DWORD PTR y
       fld1
       faddp st(1),st
       fstp DWORD PTR es:[bx]
           ; store the final value to the C modules data area
       ret
   Add1Again ENDP

   END

/****************************************
 * C module to test the assembly routines
 ****************************************/ 

   #include <stdio.h>

   float Add1(float i);
   void Add1Again(float i, float *j);

   float p = 10;
   float y = 0;
   float z = 0;

   void main()
   {
       y = Add1(p);
       printf("p = %f\t\ty = %f\n",p,y);
       Add1Again(p, &z);
       printf("p = %f\t\tz = %f\n",p,z);
   }