vscode-amiga-debug now supports Linux, MacOS

[gigabates]

Ok I can see what it is. The parser extracts the values by looking at the character offsets in the header row, but this row is thrown out of alignment by the long 'c-1048' values. We'll need to switch to another approach e.g. just splitting on whitespace and counting the values. I'll give it a go.

   LOC   CFA      r2    r3    r4    r13   r14   ra    
00001d7c r15+4    u     u     u     u     u     c-4   
00001d80 r13+8    u     u     u     c-8   u     c-4   
00001d84 r13+8    c-1048 c-1044 c-1040 c-8   c-1036 c-4

[girv]

I've no idea what those are Is the elf just bigger than expected or something like that?

[Bartman]

Originally Posted by gigabates

Ok I can see what it is. The parser extracts the values by looking at the character offsets in the header row, but this row is thrown out of alignment by the long 'c-1048' values. We'll need to switch to another approach e.g. just splitting on whitespace and counting the values. I'll give it a go.

   LOC   CFA      r2    r3    r4    r13   r14   ra    
00001d7c r15+4    u     u     u     u     u     c-4   
00001d80 r13+8    u     u     u     c-8   u     c-4   
00001d84 r13+8    c-1048 c-1044 c-1040 c-8   c-1036 c-4

Yep, see https://github.com/BartmanAbyss/vsco...ebug/issues/35
never got round to it.

[Bartman]

Originally Posted by girv

I've no idea what those are Is the elf just bigger than expected or something like that?

Some functions have a stack that is larger than 1000 bytes. No problem with your ELF file, but just my parser

[Bartman]

Originally Posted by girv

I've been waiting a year or more for this Since doing my own faffing around getting Bartman's GCC 10 and 11 to build on macOS.

The other thing I'm wondering about - are there any benchmarks comparing bebbo to gcc12 for C99 code? I'd been mostly using bebbo up until now and supporting both in the code is kind of a pain.

Probably varies from case to case, but in our case, there was like 20% increase from gcc6->gcc7, and 10% from gcc7->gcc8. Performance was the main reason we wanted to go to a newer gcc.

Although, I have to admit, my port doesn't have the bebbo specific 68000 optimizations.

[gigabates]

I've opened a PR with a fix for this.

@girv if you want to test you can try my branch here.

[girv]

Originally Posted by Bartman

Although, I have to admit, my port doesn't have the bebbo specific 68000 optimizations.

That's the bit I'm curious about

gcc12 also seems to make larger amiga executables.

         -Ofast    -O3        -O2        -Os        
gcc12   0x14368    0x14368    0x0EED8    0x0C780    
bebbo   0x0ED3C    0x0ECBC    0x0DA6C    0x0C1F4

[girv]

Originally Posted by gigabates

@girv if you want to test you can try my branch here.

I noticed it had been merged to master

It works! Thanks!
https://i.imgur.com/kE3aEYI.png (large image!)

[gigabates]

Originally Posted by girv

It works! Thanks!
https://i.imgur.com/kE3aEYI.png (large image!)

That's great news! One thing I do notice on the screenshot is some glitching on the flame graph. I've seen this before too but haven't been able to reliably reproduce it. @Bartman have you seen this issue before?

[Bartman]

Originally Posted by gigabates

That's great news! One thing I do notice on the screenshot is some glitching on the flame graph. I've seen this before too but haven't been able to reliably reproduce it. @Bartman have you seen this issue before?

Nopes, never seen before. Is this only on MacOS? The graph is rendered with WebGL, the code came directly from Microsoft. https://github.com/microsoft/vscode-...-profile-flame

I heard MacOS and OpenGL is not a good combination. Could you try if Microsoft's extension has the same problems? If not, it may be my modifications or they have fixed the issue.

[gigabates]

It looks like this fix is related. The code has diverged a bit, but I'll try to backport the changes and see if it makes a difference.

[emiespo]

Hi, tried the latest version 1.6.6 (MacOs), it seems FsUAE won't start now (tried to launch it from the cli and it complains about missing SDL libraries).

I manually installed all the libraries via brew, and it came back to life, but I assume this could be a bug?

[Bartman]

Originally Posted by emiespo

Hi, tried the latest version 1.6.6 (MacOs), it seems FsUAE won't start now (tried to launch it from the cli and it complains about missing SDL libraries).

I manually installed all the libraries via brew, and it came back to life, but I assume this could be a bug?

Thanks for trying. Can you file a bug report here? https://github.com/BartmanAbyss/vsco...a-debug/issues

[Bartman]

Originally Posted by gigabates

It looks like this fix is related. The code has diverged a bit, but I'll try to backport the changes and see if it makes a difference.

thanks!

[walkero]

Originally Posted by Bartman

sure, here are the scripts: https://github.com/BartmanAbyss/vsco...tree/master/ci

Thank you so much for those.

[girv]

FWIW, and with other caveats around benchmarking, I did some comparisons of gcc12 vs. bebbo at different optimisation levels on the main cpu-bound "update" phase of my project.

Results: https://imgur.com/a/u0CEYOk

The figures are elapsed CIA ticks on a cycle-exact A500 emulation.

For this C99 code, Bebbo has a solid performance lead of about 3% over gcc12 and produces much smaller executables. YMMV.

[Bartman]

wow! that's surprising. We saw at least 30% performance improvement over bebbo. I guess the increased file size comes from LTO.

[carrion]

I started using this extension on M1 mac. Works great. Thanks for doing this. It's really amazing with all these features.
I switch now from my own VBCC+VSCode setup to this one, but it already looks impressive. Thanks Again.

[warci]

Hi all, is there a guide somewhere on how to get this working on OSX?