Hi. It would be nice if only to allow us to scale back precission (after user confirmation) when memory is low. Though I don't know how to actually check for...
Hi Harry, Adding support for fp16 (1.5.10 s.exp.frac) sounds good to me. I can even add native fp16 support to FM when I add the true float (32-bit) and long...
Hi Alex, Adding support for fp16 (1.5.10 s.exp.frac) sounds good to me. I can even add native fp16 support to FM when I add the true float (32-bit) and long...
Hi Roberto, If I understand it correctly the 16bit float format can store values from are 2.98×10-8 to 65504 (plus the same range with a minus sign) and has a...
Hi Bill, I think the crash that you describe has nothing to do with FM. I experienced it myself some time ago. Seems to be a problem with Photoshop itself. ...
My question is whether other compilers handle these? The Wiki doesn't say anything!!!! Double in and out but store in half. Come on guys. Personally, I'd...
Hi Harry, "Probably not that good for storing 16-bit IMAGE information". True, but I was trying to explain it might be a welcome thing for non-imaging purposes...
Hi Ron, Guess you have a point with that :-P Nevertheless, "double in and out and store in half" still save a whole half block of memory and maybe even...
Hi. Just noticed fprintf prints out floating point numbers using the locale. For Dutch this means floats are formatted as "1,2" instead of "1.2". Can we have...
I don't have a problem after FM 1 is released of adding new features when it is demonstrated they are necessary. If 16bit float were standard in compilers or...
Hi Ron, If it was my decision, we would already be having Version 3 or 4 of FilterMeister. Although I basically do not mind if a FM version is called "Beta 8"...
After all the discussion on this topic, I have to chime in. This appears to me as another time consuming activity that detracts from what should be the main...
I largely agree with Harry on this and in fact was just composing a mail responding the the obsession with a version "1". I'm guessing it's symbolic for a...
Hi Roberto, However, I think now I actually misunderstood the format of floating point. Just checked wikipedia for the normalized 32-bit floating point format...
Hi Bill, This appears to me as another time consuming activity that detracts from what should be the main directions of FM: You should know that I do not earn...
Everyone knows my position so I won't beat a dead horse. I will add that the push in most usages for FM and the plug-in created is resolution which usually...
Martijn, I agree on the price. I looked at all the alternatives in 1999, and at that time would have paid 4x the price for FilterMeister. Also agree with...
... Harry, Does this mean higher processing speed in a CPU, or are these only available in graphics processing units? Regards, Bill ... From: Harald Heim To:...
Hi Ron, I need 32bit float for storage and 64bit (double) for computation. That's what the latest FM release already provides. I also have camera data that is...
Hi Bill, As far as I know 16bit floats are not supported by CPUs. Converting from 16bit float to 32bit float when reading a value from memory is quite fast, ...
... Harry, I guess I've never successfully stated the merits of separately defining the size and location of the input, output, and mask tiles. How about...
Hi Bill, I will have a look at the tile stuff. It is probably not that much work to allow the user to specify the tile width. The tile height can already be ...
Harry, That would be good. If you could do it in the context of the of the FilterRecord, the required fields have been around since version 2.5, which is a...
With 1s5e10m I would think the range lies between: ±1024^32 for max. values downto ±0,000(+31x0)1 for min. value. .... Still seems a darn high range with a...
Hi Harry, ... Please don't jump off on this on your own, until you and I have had a chance to talk. I've already done quite a bit of thinking about how to...
Hi Roberto, Not to keep beating a dead horse, but... :-) The range of fp16 (i.e., 1.5.10 with excess 15 exponent) is +/- 6x10e-8 to +/- 65504 (if denormals are...
Hi Martijn, Thanks for bringing this up. I think we went through this issue some time ago with strtod() and atof(). I think in the meantime Harry has added...
Hi Harry, Yep, I'm definitely the one who is always trying to put the brakes on the formal release. :-) mea culpa To me, the most essential missing component...
Alex, Have you recieved the follow-up I sent to you and Harry yesterday? It's a bug (the behaviour doesn't appear in compiled plug-ins, only in FM itself) and...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
