Fixed point Vs Floating point

Full cycle tables are useful only of you have gates, cells, blocks of ram or rom to piss away. Other than that a quadrant and tiny bit of program or logic is quite sufficient.

will

or

with

Quadrants are nice if program loop time (ie output frequency) is not an issue. Add a few quadrant manipulation instructions to a tight loop and the performance suffers drastically (like 10x worse!).

None of the ones that you named do it--certainly not any Intel ones. Support for denormals is provided by handling the EM_DENORMAL exception in software, or by flushing to zero.

Nice bluff, though.

Cheers,

Phil Hobbs

Plus there is another interesting technique called block floating point with one exponent and a block of mantissas.

Hmm. Though it was some time ago i had studied that databook rather closely. IMM it did act properly, which at the time included throwing interrupts (exceptions) {By the way there is a difference}. Perhaps that property was lost when the FPU was brought onboard in the 486 and never restored.

And what part of "it is a time - memory space" tradeoff did you not understand? Nor is the performance hit necessarily that bad, even in assembler.

You sure are making an effort to be obnoxious lately. What's going on?

John

I think JosephKK is ill... he's been unusually cranky with me, too.

...Jim Thompson

-- | James E.Thompson, P.E. | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona 85048 Skype: Contacts Only | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

I looked in the latest version of the Intel software optimization manual,

Cheers,

Phil Hobbs

I will gladly take a draw rather than drag this out.