Compiler Folks are very busy people as there aren’t as much of us unfortunately so no need to apologize. I’ve yet to heard from someone on the GCC side and will wait until after GCC 11 is released due to this. Also not to mention the health issues of Coronavirus-19. I was researching this on and off for the last few months in terms of figuring out how to make the pass manager itself async. Its not easy and I’m not even sure if that’s possible. Not sure about GIMPLE as I would have to ask the middle end maintainer on the GCC side but LLVM IR does not seem to have shared state detection or the core classes and same for the back ends. So yes this would interest me. The first place to start with is which data structures are shared for sure. The biggest ones seem to be basic blocks and function definitions in terms of shared state, as those would be shared by passes running on each function. We should start looking at implementing here locks or ref counting here first if your OK with that. It also allows me to understand a little more concrete the linkage between the core classes as would be required for multi threading LLVM. In addition, it allows us to look into partitioning issues with threads at the same thing in terms of how to do it. If others want to chip it as I’ve CCed the list, that’s fine as well and this should be up for discussion with the whole community. I’ve given up on the idea of a async pass manager as it seems to require IR level detection of changed state between passes but maybe I’m wrong, Nick
(this thread seems to have been split from the original thread in gmail at least - which might make it hard to keep track of the context)
Ah, it was a different thread: http://lists.llvm.org/pipermail/llvm-dev/2020-March/139606.html - might be best to pick up from that thread instead of starting a new one.
Hey,
Apologies for the wait, everything right now is going crazy…
Compiler Folks are very busy people as there aren’t as much of us unfortunately so no need to
apologize. I’ve yet to heard from someone on the GCC side and will wait until after GCC 11
is released due to this. Also not to mention the health issues of Coronavirus-19.I think we should early in move this conversation on the llvm Dev list but generally speaking we can see three options here:
- parallelize single passes or a subset of passes that are known to not interfer, e.g. the attributor,
- parallelize analysis pass execution before a transformation that needs them,
- investigate what needs to be done for a parallel execution of many passes, e.g. How can we avoid races on shared structure such as the constant pool.
I was researching this on and off for the last few months in terms of figuring out how to make the pass manager itself async. Its not easy and I’m not even
sure if that’s possible. Not sure about GIMPLE as I would have to ask the middle end maintainer on the GCC side but LLVM IR does not seem to have shared
state detection or the core classes and same for the back ends. So yes this would interest me.The first place to start with is which data structures are shared for sure. The biggest ones seem to be basic blocks and function definitions in terms of shared state, as
those would be shared by passes running on each function. We should start looking at implementing here locks or ref counting here first if your OK with that.
It also allows me to understand a little more concrete the linkage between the core classes as would be required for multi threading LLVM. In addition,
it allows us to look into partitioning issues with threads at the same thing in terms of how to do it.
As was discussed on the previous thread - generally the assumption is that one wouldn’t try to run two function optimizations on the same function at the same time, but, for instance - run function optimizations on unrelated functions at the same time (or CGSCC passes on distinct CGSCCs). But this is difficult in LLVM IR because use lists are shared - so if two functions use the same global variable or call the same 3rd function, optimizing out a function call from each of those functions becomes a write to shared state when trying to update the use list of that 3rd function. MLIR apparently has a different design in this regard that is intended to be more amenable to these situations.
Hi Nick,
I was researching this on and off for the last few months in terms of figuring out how to make the pass manager itself async. Its not easy and I'm not even
sure if that's possible. Not sure about GIMPLE as I would have to ask the middle end maintainer on the GCC side but LLVM IR does not seem to have shared
state detection or the core classes and same for the back ends. So yes this would interest me.The first place to start with is which data structures are shared for sure. The biggest ones seem to be basic blocks and function definitions in terms of shared state, as
those would be shared by passes running on each function. We should start looking at implementing here locks or ref counting here first if your OK with that.
It also allows me to understand a little more concrete the linkage between the core classes as would be required for multi threading LLVM. In addition,
it allows us to look into partitioning issues with threads at the same thing in terms of how to do it.
I think it's great that you're looking into this!
As David already mentioned, the gnarliest immediate issue is that
constant values (which includes global variables and functions) track
uses via a linked list. So in the current design, being able to do
concurrent transformations within the same LLVMContext would require
taking a lock whenever a use of a constant is added or removed. That's
obviously too much overhead.
So one of the first steps you'd have to do would be to change that.
That may seem a bit daunting initially, but it's feasible and seems
like a good idea to evolve LLVM IR in the direction of what MLIR does
here.
Cheers,
Nicolai