[ Reply to This | Parent | # ]

I don't know. I assumed those terms were explained in the specification. I have not read it though. Have you?

Reading through the specification can be surprisingly useless.

In my experience, it is unusual these days to find usable definitions in the specification, as most patents will preface any description there with terms like "this is but one embodiment and is not meant to be limiting". Should you find something to your liking, your opponent will almost certainly claim that is either not definitive, or not relevant, or both. It would require claim construction to extract that definition, and a judge to rule that one side's determination was in fact definitive.

Your surest bet is when only the claims are utilized to decide what has been patented. The specification (and the prosecution history) can be examined for definitions, but that is by no means a slam-dunk, since it often depends on the context surrounding it. You and I reading it can certainly form an opinion, but we should avoid terms like "obviously" or "clearly". Once you leave the claims, nothing is particularly obvious or clear and is often subject to judicial review.

All that said, I read the spec and there does seem to be some useful definitions there about different formats. But since they lie outside the claims, I would consider them more of a discussion point rather than gospel. And more to the original point of the discussion, there is (heh, was, thanks to the court) significant disagreement over whether they constituted transformations in any case.

Clearly, the original poster believes they are. Personally, I agree with the court's finding.

[ Reply to This | Parent | # ]

• Memory register vs a register? - Authored by: Anonymous on Friday, March 29 2013 @ 12:14 AM EDT
  • Memory register vs a register? - Authored by: Anonymous on Friday, March 29 2013 @ 03:28 AM EDT

Parts of this patent actually make sense.

There is a standard floating-point format that is used by most current hardware. That format describes how floating-point numbers are stored in memory. The format is rather complicated, but for good reasons. If you want to add two floating-point numbers, for example, the hardware has to analyze the bits of the two operands in a rather costly way, perform the operation, and transform the result back to the complicated floating-point format.

An alternative is to convert floating-point numbers from the memory format into a different format that is easier to analyze when moving data from memory into a register, and doing the opposite when moving data from a register to memory. The floating-point operation might then be faster, because they don't need to analyze the operands in a complicated format, and don't need to convert results to the complicated memory format. I actually suspect that Intel for example does exactly that in modern processors. So we have two different number formats, one called the "memory" format, which would be visible to any programmer, and one called the "register" format, which would be invisible to any programmer, and only be internal to the processor.

There are a few problems, however. One is that when I read the patent, the memory format seems to be different from any of the IEEE standard formats. As a result, I find a highly unlikely that anyone in the world would infringe on this patent. The register format seems to be highly (and unneccessarily) complicated, which would again make it highly unlikely that anyone would infringe, because the method described is actually not very clever. For exactly the same reasons I would suppose there is no prior art; it must take an extraordinary brain to come up with something badly complicated like that :-)

The biggest problem however is that this invention, even if it were novel, useful, non-obvious etc. etc. would simply not be patentable.

[ Reply to This | Parent | # ]

• The above comment is highly recommended. - Authored by: Anonymous on Saturday, March 30 2013 @ 12:46 PM EDT