[PATCH] [Core] Add parallel infrastructure to lld.

[Sean Silva]

================
Comment at: include/lld/Core/Parallel.h:198
@@ +197,3 @@
+  // Do a sequential sort for small inputs.
+  if (std::distance(start, end) < detail::minParallelSize || depth > maxDepth) {
+    std::sort(start, end, comp);
----------------
Michael Spencer wrote:
> Sean Silva wrote:
> > This algorithm can go quadratic. You should fall back to std::sort if the depth becomes larger than logarithmic in the length of the range to be sorted to ensure n*log(n). Also, it would make sense to fall back to std::sort if the depth gets greater than log2(coreCount) to avoid increasing the parallelism beyond the number of cores.
> Good idea on the max depth to use. As for the max task count, I found that limiting the max splits to the number of cores to be slower, as chunks take differing times to complete. Smaller chunks balance better.
> As for the max task count, I found that limiting the max splits to the number of cores to be slower, as chunks take differing times to complete. Smaller chunks balance better.

Ah, ok then.

If you incrementally build up a set of addresses and need to in the end emit them in sorted order, there are trie-like data structures that allow for *extremely* dense encoding of sorted maps/sets (because key prefixes are shared) and constant-time insertion; I believe that these data structures are also very amenable to lock-free variants (one common usage is for kernel IP routing tables). They also have the advantage that they can be reconstituted into a sorted array of addresses in O(n) time. Specialized data structures like these might be something worth looking into as a memory optimization.


http://llvm-reviews.chandlerc.com/D649