scott.smith added inline comments.
================
Comment at: source/Plugins/SymbolFile/DWARF/SymbolFileDWARF.cpp:1995-1996
//----------------------------------------------------------------------
- TaskMapOverInt(0, num_compile_units, extract_fn);
+ llvm::parallel::for_each_n(llvm::parallel::par, 0U, num_compile_units,
+ extract_fn);
----------------
scott.smith wrote:
> zturner wrote:
> > What did you decide about the recursive parallelism? I don't know if that
> > works yet using LLVM's default executor.
> 1. This code doesn't care.
> 2. It looks like it works, since (I think) for_each creates a separate
> TaskGroup for each call.
> 3. However I got a deadlock when using this for parallelizing the dynamic
> library loading itself, which used to work. That could either be due to
> other code changes, some oversight on my part, or it could be that for_each_n
> doesn't actually support recursion - which means that I misunderstood
> for_each_n. So I have more work to do...
On further inspection, llvm::parallel does not support recursion, since
TaskGroup uses a single static Executor, and provides no way to override that
(and besides, there's no way to pass parameters from for_each_n to the
TaskGroup). That's fixable though, by making the default executor a thread
local variable, so that worker threads can enqueue work to a different executor.
Repository:
rL LLVM
https://reviews.llvm.org/D33246
_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
