The team will post updates and news about our project here
ApolloOS Release6 Major Performance Increase
Gunnar von Boehn (Apollo Team Member) Posts 6254 12 Apr 2021 11:59
Good news!
We started adding 68k optimizations to GCC. And we started to enable AMMX in GCC. While we just started with this work and there is more todo, we can already see the huge benefit of this.
We benchmarked that many OS operations already run 30-40% faster thanks to the improved compiler.
Adam A
Posts 130 12 Apr 2021 12:03
Great news, Thank you.
Chastanier Cclecle
Posts 19 12 Apr 2021 12:07
Is it with bebbo GCC toolchain ? If yes: -> I see last commit is 20 days old, so you have your hown fork ? -> Where is the repo / can we test it ? -> is it still amigaos compatible ?
Gunnar von Boehn (Apollo Team Member) Posts 6254 12 Apr 2021 12:30
Chastanier Cclecle wrote:
Is it with bebbo GCC toolchain ?
We have our own version. But we work closely with Bebbo together.
I see think there is bug potential in GCC tuning and I would not be surprised of ApolloOS will run twice as fast at some point.
Roger Andre Lassen
Posts 150 12 Apr 2021 15:01
Awesome news!
Lord Aga (Apollo Team Member) Posts 119 12 Apr 2021 15:27
Woohoo, great progress! :)
DiscreetFX Studios
Posts 155 12 Apr 2021 20:23
Awesome news!
Jean-Baptiste Bolcato
Posts 63 13 Apr 2021 00:50
Sounds great! That is for all Vamps right? (not just the V4)
Kamelito Loveless
Posts 261 13 Apr 2021 08:00
Good news but one need a set of automated tests to ensure that there’s no régressions before using it in production. A debugger being GDB or an other one should be able to debug program using the new instructions too.
Gunnar von Boehn (Apollo Team Member) Posts 6254 13 Apr 2021 08:26
Kamelito Loveless wrote:
Good news but one need a set of automated tests
And regression tests and performance test is exact what we coded for this.
But sometimes improving GCC is even simpler. If you take a deeper look into the GCC source tree then you can easily see that some of the 68040/60 backend was done by "arm-chair coders" which wrote the code for 68K by looking at "books" without ever testing their code on real 68K CPUs.
For example the GCC did wrongly turn OFF some instructions for 68040. I think the only explanation for this can be that the GCC coder did misread some parts in the Motorola manuals and took some typos in the manual for reality without ever verifying it on a real CPU.
Its very easy to see such mistakes in GCC and fix them - if you are an 68K expert.
Robo Kupka
Posts 51 13 Apr 2021 08:32
Will you release the buffed up AMMX GCC on aminet ?
Gunnar von Boehn (Apollo Team Member) Posts 6254 13 Apr 2021 08:42
Robo Kupka wrote:
Will you release the buffed up AMMX GCC on aminet ?
Improving GCC is a lot of work and doing this properly is a very a big task. This will certainly take many month. Bebbo did already a lot work here and we are on close contact with him. I think it makes a lot sense to make one good release when its done.
Kamelito Loveless
Posts 261 13 Apr 2021 09:44
This is a huge task but one that should be done, great!
Ray Couzens
Posts 93 14 Apr 2021 09:35
Great news! Not only is the Vampire Amiga itself being improved but also - in the near future - we will be getting a Vampire optimized C compiler!
I never expected this when buying my V4, it just keeps getting better.
Roland V
Posts 7 17 Apr 2021 18:44
Will this progress also be back ported to V1200V2?
Kamelito Loveless
Posts 261 19 Apr 2021 15:10
My understanding is that : As V4 and V2 share the same CPU then the compiler will produce a code usable for both cards so no need to back port the compiler.
My understanding is that : As V4 and V2 share the same CPU then the compiler will produce a code usable for both cards so no need to back port the compiler.
I think they mean will the improvements come to the V2 and the team have said yes once the V4 is perfected