That is a good question. I seem to recall that when it was compiled for
Next hardware that it used the DSP by default, but when compiled for
other platforms it used their libraries. My guess is the Intel library
used floating point.
It fits in with other narratives about the Pentium chip, one of the
biggest speed boosts was the new floating point system which was a
different design over the 80386 type FPU that was in the 486 (the 486
was faster than the 386 for FPP because the unit was mounted on the die
as opposed to a separate chip).
C
On 3/14/2023 1:51 PM, Tapley, Mark B. wrote:
Chris,
one question on the conclusion: was the Mandelbrot program set to use
floating point, or fixed-point arithmetic? I’m pretty sure the DSP
version was fixed-point (integer, scaled) arithmetic to make it run
faster. The conclusion might apply to the Pentium’s performance in
integer tasks but not be relevant to floating-point tasks.