If you need help, try to find the answer through the following path: `Help` in the console ⇒ [Red by Example](https://www.red-by-example.org/) ⇒ [Official Red Docs](https://github.com/red/docs/blob/master/en/SUMMARY.adoc) ⇒ [Rebol Core Manual](http://www.rebol.com/docs/core23/rebolcore.html ⇒ [Red Wiki](https://github.com/red/red/wiki ⇒ Net search "redlang + your-question" ⇒ Ask us, we are here to help!
To create a function that returns unset! instead of
a traditional return value, use exit.
>> example: func [value] [value]
== func [value][value]
>> example "value"
== "value"
>> example: func [value] [value exit]
== func [value][value exit]
>> example "value"
>>
>> type? example "value"
== unset!
>>unset! in general, and why it exists?
unset! there would be a hole in the language, several fundamental semantic rules would be collapsing, e.g. reduce [1 print ""] => [1] (reducing 2 expressions would return 1 expression).
- Any ETA on concurrency support in Red?
- Is it possible (or will be possible) to call SIMD instructions from R/S? Or work directly with
xmm[0-7]registers (currently this feature limited to IA32)? - Aside from concurrency and data level parallelism, how can I exploit multi-channel commands (i.e. how can I make RAM bus very busy) in Red or R/S? Maybe
system/fpuand floating-point bit diddling will do? - In R/S,
print system/fpu/typereturns1, honestly I don't know how to interpret it :( (see R/S specs). Does it mean x87 FPU or SSE?
- SIMD: we certainly want it, might happen in a 0.9.x version, or post 1.0. Direct access to XMM registers will be possible, though you'll have specific R/S types to abstract SIMD operations.
- Multi-channel: no plans for that before 1.0 (not sure what are the
multi-channel commandsyou're referring to). - Search in the source code, definitions are there (we should probably replace it with an #enum): https://github.com/red/red/blob/master/system/runtime/system.reds#L14. Currently Red relies only on the x87 FPU, SSE is not yet used for general floats support, though such support should happen before 1.0.
@dockimbel thanks Nenad!
By "multi-channel" I meant instructions which use multiple processing units at once (frankly I'm not completely sure what I'm talking about :D guh, noobiness). I think in real worlds it's called parallelism ;) Perhaps what I want in my case is something that can be considered "costly" operation which fully loads primary system bus and avoids caching mechanism, like SIMD MOVNTDQ instruction.
About codebase: are there any tips for newcomers on how to navigate it? Sometimes I know that my answer is in the sources (like with fpu/type above), but I'm not sure where to find it :(
grep is your friend on Unix, or any equivalent feature in your Code editor. I have a "Find in files" feature (<=> grep) in my main code editor, which I use dozens of time everyday to navigate through the source code.
make RAM bus very busy should be achieved by any tight loop intensively reading/writing 32-bit values in memory (until we get 64-bit support). Such operations will go through the caches, there is no specific cache control instructions emitted by the compiler. Such optimizations are planned for 2.0.
