DSP Robotics Support

Let's repeat just one word: AVX (256-bit instruction set extension)

schty wrote:Let's repeat just one word: AVX (256-bit instruction set extension)

We haven't even reached full SSE2 yet! But since AVX is not on most CPUs yet, so we'll need some way of finding the CPU abilities.

I think there may be a use-case for SSE int maths also, but I need to see use-cases.

infuzion wrote:But since AVX is not on most CPUs yet...

AVX was first introduced already back in 2008 and implemented in 2011 (Intel Sandy Bridge processor, Q1 2011, Sandy Bridge E processor, Q4 2011, Ivy Bridge processor, Q1 2012 and AMD: Bulldozer processor, Q3 2011) and quickly becoming mainstream now.

TheAudiophileDutchman wrote:AVX was first introduced already back in 2008 and implemented in 2011

Trouble is that there is no way to know the instruction capabilities of the CPU on which the program is running - so that an alternative code version can be substituted when AVX etc. is not available. The SSE2 stuff I would say was pretty safe - most machines that don't support that are probably dead by now.
However, I'm sure that there are many folks out there who, like me, don't run off to the store the moment a new CPU architecture is introduced - and I am certain that this will include many of the educational users that are prime customers for FS. I'm on only my 3rd desktop machine in nearly 20 years, and have no intention of buying a new one until my current MoBo has gone to Silicon Heaven!
And that would make for big problems on the forums, and for plugin distribution - every single assembly module out there would have to come complete with a "minimum system requirements" spec' - so I would say to be very careful what you wish for! Even more so when there are more pressing things for the dev's to be bringing up to date, e.g. 64bit VSTs so that no "bridging" isn't needed, VST3 etc.
DSPr are also busy trying to generate a user base for their new software, and catering to the whims of a niche within a niche doesn't seem sensible when they are primarily selling the software on the promise of ease of use. They have already been good enough to add some features that weren't on their original road-map just to keep us whiners happy - so lets not get too greedy!

Trouble is that there is no way to know the instruction capabilities of the CPU on which the program is running - so that an alternative code version can be substituted when AVX etc. is not available

Well, CPU-Z can obviously detect AVX, it's just not implemented in FS yet if that's what you mean. Nevertheless, I agree that full SSE2 and 64-bit support are long overdue and therefore of higher priority.

TheAudiophileDutchman wrote:Well, CPU-Z can obviously detect AVX, it's just not implemented in FS yet if that's what you mean

The data are available for sure - and for Code and Primitives I guess it could work similar to the existing SSE/SSE2 compatibility settings.
But even if the information were made available within the schematic, it would require assembly coders to write multiple versions of their code if they wished to retain back-compatibility. For 'lone wolf' users, making stuff only for their own use, not a problem of course - but I have used a couple of forums in the past where some very clever and generous users have ended up puling their hair out because of "bug reports" from people without the PC "know how" or common sense to realise that their setup did not meet the necessary standards to use posted code, or just expressing their annoyance at being "excluded". Bad enough for a well-paid tech' support team, but extremely demoralising for "community spirited" amateurs.
I have no objection to adding every opcode under the sun for any technical reason, more that I am wary of splitting FS users into the "haves" and the "have nots" based on whether they can "keep up with the Joneses" at the PC store. The current FS minimum spec's are very broad, and IMHO, this is a very good thing.

AVX would be cool, but some commands like the fused ones are a bit more complicated to throw in. SSE2-4 would be easier & better supported.