In fact the problem is a little bit different. BP doesn't generate
true FP opcodes at all (except for no-wait control instructions like
FNINIT beginning with FN...). It encodes them as an emulating
interrupt call. At run time (if a copro is present) the opcodes will
be patched by a patch module which hooks the relevant interrupts to
regular FPU opcodes before they are executed for the first time. As
the emulation opcode is always one byte longer than the regular FPU
opcode, the empty place is filled with a one byte WAIT instruction.
This WAIT is inserted _in_ _front_ of the FPU opcode.

        asm fldz end;

will compile to

        CD 35 EE   { int $33 , byte 2 of opcode }

this will be patched before the first execution to

        9B D9 EE   { WAIT  FLDZ }

On modern copros like yours this won't slow down the execution because
they automatically skip any WAIT in front of a
non-control-instruction, i.e. they don't _execute_ it - although the
opcode has to be read and decoded by the CPU - but this can't be
avoided, Well, only by writing external procedures with an assembler
and linking them into the program or by coding your FPU instructions
via db ;-)

By the way: TP6.0 has an undocumented quirk. If the switch {$G+} is
set, BASM would produce true opcodes without leading WAITs.

PROGRAM.3:  asm fldz end                                    
  cs:000D CD35EE         fldz                                
PROGRAM.4: end.                                              

after run time patch

PROGRAM.3:  asm fldz end                                    
  cs:000D 9B             wait                                
  cs:000E D9EE           fldz                                
PROGRAM.4: end.                                              

Or did'nt you realize that the code

        CD33EE           fldz

which is produced by the FLDZ instruction and which is displayed by TD
as "fldz" is only an emulator-fldz (i.e. an interrupt call
CD33=int$33) which will change immediately to

        9A               wait
        D9EE             fldz

if you step through the instruction because the called interrupt
patches the code and executes it afterwards as 'wait fldz' ?

FCLEX
FDISI
FENI
FINIT
FSTCW
FSTENV
FSAVE
FSTSW

(Use the FN.. ones)

AFAIK, TP/BP7 always generate emulated instructions, resulting in the WAIT.

However, by using Norbert Juffa's TPL60N19.ZIP or (I don't use it, so not
sure) BPL70N16.ZIP, you can get rid of the WAITs, but you'll get NOPs
instead.

HTH,

Robert
--
Robert AH Prins
prin...@williscorroon.com