Pinched the following code from somewhere:
#include <dos.h>////      Save FPU state.////      This uses a magic spell taken from Intel application note AP-113.//void fpusave (char far* t){    //_ES = FP_SEG (t);    //_BX = FP_OFF (t);    __asm les bx,t    __asm  fnstcw  es:[bx]     // save IEM bit status    __asm  nop                 // delay while control word saved    __asm  fndisi              // disable BUSY signal    __asm  mov     ax, es:[bx] // get original control word in AX    __asm  fsave   es:[bx]     // save FPU state    __asm  fwait               // wait for save to complete    __asm  mov     es:[bx],ax  // put original control word in saved state}
////      Restore FPU state.//void fpurest (char far* t){    //_ES = FP_SEG (t);    //_BX = FP_OFF (t);    __asm les bx,t    __asm frstor es:[bx]}
Found the compiler cannot recognise fndisi instruction.  Compiles with MS VC++ 1.5 and above and BCC as well.

When Obj2asmed the code generated by BC or VC,

fndisi   (db e1)

is being intepreted as

fdisi    (9b db e1)

Don't ask me what they mean as I haven't read into the Intel data book yet...

The following code will now generate identical obj file by MSVC, DMC/SC and BC:
void fpusave (char far* t){#if defined(__BORLANDC__)    _ES = FP_SEG (t);    _BX = FP_OFF (t);#else     // MSVC or DMC/SC    __asm les bx,t#endif    __asm fnstcw  es:[bx]     // save IEM bit status    __asm nop                 // delay while control word saved
#if defined(__SC__)    __emit__(0xdb,0xe1);#else    __asm fndisi              // disable BUSY signal#endif
    __asm mov     ax, es:[bx] // get original control word in AX    __asm fsave   es:[bx]     // save FPU state    __asm fwait               // wait for save to complete    __asm mov     es:[bx],ax  // put original control word in saved state}
void fpurest (char far* t){#if defined(__BORLANDC__)    _ES = FP_SEG (t);    _BX = FP_OFF (t);#else     // MS or SC    __asm les bx,t#endif
    __asm frstor es:[bx]}

You're right. That instruction only does anything on the original 8087, and isn't even documented by later processor manuals, when the inline assembler was written <g>. I'll see about adding it. In the meantime, you can get by with:

    db 0xDB, 0xE1

  "KarL" <someone@somewhere.org> wrote in message news:b979vu$1nkh$1@digitaldaemon.com...
  Pinched the following code from somewhere:
#include <dos.h>////      Save FPU state.////      This uses a magic spell taken from Intel application note AP-113.//void fpusave (char far* t){    //_ES = FP_SEG (t);    //_BX = FP_OFF (t);    __asm les bx,t    __asm  fnstcw  es:[bx]     // save IEM bit status    __asm  nop                 // delay while control word saved    __asm  fndisi              // disable BUSY signal    __asm  mov     ax, es:[bx] // get original control word in AX    __asm  fsave   es:[bx]     // save FPU state    __asm  fwait               // wait for save to complete    __asm  mov     es:[bx],ax  // put original control word in saved state}////      Restore FPU state.//void fpurest (char far* t){    //_ES = FP_SEG (t);    //_BX = FP_OFF (t);    __asm les bx,t    __asm frstor es:[bx]}Found the compiler cannot recognise fndisi instruction.  Compiles with MS VC++ 1.5 and above and BCC as well.

  When Obj2asmed the code generated by BC or VC,

  fndisi   (db e1)

  is being intepreted as

  fdisi    (9b db e1)

  Don't ask me what they mean as I haven't read into the Intel data book yet...

I had a follow-up post on this already ....

Thanks Walter....

--------------8<------- Quote from the documentation-----------
Zortech C++ allowed instructions to be assembled from integers using the asm() pseudo-function.  For example:
asm (0x8C, 0x96, 0xCC, 0xFE);
DigiMars C++ does not provide this function. Therefore, you need to replace asm() calls with calls to __emit__.  For more information, see "Using Assembly Language Functions."
--------------8<------- Quote from the documentation-----------
  "Walter" <walter@digitalmars.com> wrote in message news:b97k4c$2174$1@digitaldaemon.com...
  You're right. That instruction only does anything on the original 8087, and isn't even documented by later processor manuals, when the inline assembler was written <g>. I'll see about adding it. In the meantime, you can get by with:

      db 0xDB, 0xE1

"KarL" <someone@somewhere.org> wrote:

> Found the compiler cannot recognise fndisi instruction.

Then why don't you use "fnsave" instead?

--gv