Mill Computing, Inc. › Forums › Announcements › Events › Yearly ping and see how things are going thread?
- AuthorPosts
- #3974 |
I don’t want to take up too much of anybody’s precious time (thus the long duration between pings), but I am curious how progress is going. How is 2024 progressing for The Mill team? I miss the regular flow of new presentations
We now have significant C/C++ code compiled with LLVM for a front end, with our own code generation for a back end.
We can now run Coremark, for example, on a functional instruction set simulator of our Architecture.
We still have optimizations to complete implementing in that code generation back end.
I am also working on Hardware generation from our specification machinery.At this point all of the areas we are working on could progress much faster with money for hiring people. That was not always the case…
Also, we could disclose a lot more if we could file patents for what we still have as internal ideas, but that requires money as well.
wow, that sounds exciting!
that would be the first proof of whether the architecture can deliver what it promises.
i think that many people are eagerly awaiting the first results of a coremark and how they will evolve over the course of the development cyclesi am very impressed that after such a long time there will be life on the outside again.
i am very excited!… and the next 8 months around. are there any initial coremark results that might satisfy the curiosity of many followers?
We’ve been busy completing the sim and compiler toolchains. Making good progress but a lot to do still.
The Coremark results so far show that a Mill Gold will be able to perform on par (CoreMarks/MHz) with recent Intel and AMD CPUs. That is about as much as can be expected for Caremark, considering the nature of the benchmark and the level of focus it has been receiving from compiler writers already.
Essentially, Coremark is a small benchmark that can be run entrirely out of L1 cache, and all branches can be (almost) perfectly predicted given enough cache and predictor tables. For OoO CPUs it then becomes a matter of exploiting all instruction-level parallelism in the code, of which there is not a lot.
The Mill essentially can find the same instruction-level parallelism, and statically schedule it. The compiler toolchain (specializer) still leaves some opportunities for parallelisation unused, so there is some opportunity for improvement available, in the order of 10%.
We’re now moving on to (micro) benchmarks where the Mill will truly shine, which will be in areas like code with more instruction-level parallelism, amenable to auto-vectorisation, and a need for stricter security.
- AuthorPosts
