*LOADing GCC binaries into BASIC

I spent a bit of my lunchtime working on getting my VFP/NEON code running as a module. I have the makefile updated and it does build, except for one thing.

module.o uses FPA, whereas I do not and ld complains. So, how do I get cmunge to work for me? Can I convince ld somehow to ignore it?

Edit: I’m going to see if I can bypass the compilation stuff and do that manually myself so I can specify the right flags.

Also, I really should get the DDE at some point, it’s not so much a cost thing, as how useful it all is for me…

write your module with the routines as SWIs

Gee, more application code running as kernel level code… ;-)

Gee, more application code running as kernel level code… ;-)

In this case, it probably does make sense. I’m writing a library to be used from my own C projects and to be called from BASIC where I will write a wrapper for it.

module.o uses FPA, whereas I do not and ld complains. So, how do I get cmunge to work for me? Can I convince ld somehow to ignore it?

I think you’ll struggle with that bit. At the moment the only C library that works with module code is the SCL, and that uses FPA. One of the GCC examples is for a module which doesn’t use the SCL, but I don’t think there’s an option in cmunge for generating VFP/NEON stubs.

Hi Jeffrey,

I have successfully compiled against the SCL with the -mfpu=neon flag. I create an intermediate .s file, before using gcc to convert this to a module.o file and then link against my code.

It’s a bit early to say for sure it’s working until I get my module to actually do something. But at least it builds a working module which I can see with *modules.

PS. I’ve had the SCL and VFP/NEON running already for a few weeks in a regular application. I know it’s true, because when I look at the intermediate .s files the references to SharedCLibrary are there. I don’t think this is anything that unexpected though as it’s documented…

I’ll keep plodding on and see where I end up.

Well, I have a block of SWIs allocated, including names for my init and shutdown. When I call them from BASIC nothing happens, but it’s a start…at least they are recognised.

Yes, of course. Basalt does that for its link.

Exactly this approach? Resources:$.Veneers.<modulename>? Great minds and fools.

An alternative might be to count users and refuse to die

I hate that. As if I’m RMKilling on a whim. Do as you’re told, code. I’m in charge here.

I hate that. As if I’m RMKilling on a whim. Do as you’re told, code. I’m in charge here.

Yeah, but you’re basically God, aren’t you? :-)

It’s dangerous to allow a module with active clients to be killed. We can point at any number of CLib softload mishaps as evidence.

I have no problem with the module offering a *command to discard active clients (useful if something crashes and doesn’t unregister itself), which will then permit the module to die, however if the simple act of randomly killing it stands to cause something to crash…
…training wheels and nappies are sometimes useful.

When I call them from BASIC nothing happens, but it’s a start…

You should make the SWI handler routine at least printf() the SWI number that was called…

Yeah, but you’re basically God, aren’t you? :-)

Oh, Nemo’s well aware of that, but what does it have to do with this thread? :-)
BTW. I think the “basically” may be superfluous.

Exactly this approach? Resources:$.Veneers.?

No, I did not say that. I claim a block in RMA and keep the address in a system variable.

If I could persuade the bits to believe I’m their maker I’ll be happy. The rest of you bugs can find your own debugger.

Apropos of nothing, I’ve just discovered the hard way that of all the multifarious GBPB directory reads, ResourceFS doesn’t support #11. It does the others. Just not the one that would be most useful on ResourceFS of all FSes.

keep the address in a system variable

I want to hate that but I can’t, and it is easier.

Mission accomplished for me. I’ve got a module, with the first ~10 SWIs providing fast 2D primitives including rather nice gouraud triangles. All designed for 24bit colour and although not totally optimised yet, are much quicker than using OS_Plot.

I also have a single *command which I haven’t finished yet which will give a demo.

Thanks for all the help, I’m well on the way now. A gcc compiled module using VFP and callable from BASIC (or anything else).

Yeah, but you’re basically God, aren’t you? :-)

Are you confusing nemo with Boris?

Are you confusing nemo with Boris?

Oh, now that’s uncalled for.

Edit. Re-reading after a delay I can see I ought to have pointed out that anything involving Boris is really an insult so it ought to have a smiley, as should my response to indicate the light-hearted poke.

A gcc compiled module using VFP and callable from BASIC (or anything else).

Great. Will you find some time to package a short example with some explanations?

Great. Will you find some time to package a short example with some explanations?

I can do that, although for some reason my VFP stuff was causing me problems last night and I ended up pushing the VFP context stuff into BASIC (which worked). Once I get it working well, I’ll either release the source or do a tutorial.

Thanks Daryl!