1 //===- Threads.h ------------------------------------------------*- C++ -*-===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // LLD supports threads to distribute workloads to multiple cores. Using
11 // multicore is most effective when more than one core are idle. At the
12 // last step of a build, it is often the case that a linker is the only
13 // active process on a computer. So, we are naturally interested in using
14 // threads wisely to reduce latency to deliver results to users.
16 // That said, we don't want to do "too clever" things using threads.
17 // Complex multi-threaded algorithms are sometimes extremely hard to
18 // reason about and can easily mess up the entire design.
20 // Fortunately, when a linker links large programs (when the link time is
21 // most critical), it spends most of the time to work on massive number of
22 // small pieces of data of the same kind, and there are opportunities for
23 // large parallelism there. Here are examples:
25 // - We have hundreds of thousands of input sections that need to be
26 // copied to a result file at the last step of link. Once we fix a file
27 // layout, each section can be copied to its destination and its
28 // relocations can be applied independently.
30 // - We have tens of millions of small strings when constructing a
31 // mergeable string section.
33 // For the cases such as the former, we can just use parallelForEach
34 // instead of std::for_each (or a plain for loop). Because tasks are
35 // completely independent from each other, we can run them in parallel
36 // without any coordination between them. That's very easy to understand
39 // For the cases such as the latter, we can use parallel algorithms to
40 // deal with massive data. We have to write code for a tailored algorithm
41 // for each problem, but the complexity of multi-threading is isolated in
42 // a single pass and doesn't affect the linker's overall design.
44 // The above approach seems to be working fairly well. As an example, when
45 // linking Chromium (output size 1.6 GB), using 4 cores reduces latency to
46 // 75% compared to single core (from 12.66 seconds to 9.55 seconds) on my
47 // Ivy Bridge Xeon 2.8 GHz machine. Using 40 cores reduces it to 63% (from
48 // 12.66 seconds to 7.95 seconds). Because of the Amdahl's law, the
49 // speedup is not linear, but as you add more cores, it gets faster.
51 // On a final note, if you are trying to optimize, keep the axiom "don't
52 // guess, measure!" in mind. Some important passes of the linker are not
53 // that slow. For example, resolving all symbols is not a very heavy pass,
54 // although it would be very hard to parallelize it. You want to first
55 // identify a slow pass and then optimize it.
57 //===----------------------------------------------------------------------===//
59 #ifndef LLD_COMMON_THREADS_H
60 #define LLD_COMMON_THREADS_H
62 #include "llvm/Support/Parallel.h"
67 extern bool ThreadsEnabled;
69 template <typename R, class FuncTy> void parallelForEach(R &&Range, FuncTy Fn) {
71 for_each(llvm::parallel::par, std::begin(Range), std::end(Range), Fn);
73 for_each(llvm::parallel::seq, std::begin(Range), std::end(Range), Fn);
76 inline void parallelForEachN(size_t Begin, size_t End,
77 std::function<void(size_t)> Fn) {
79 for_each_n(llvm::parallel::par, Begin, End, Fn);
81 for_each_n(llvm::parallel::seq, Begin, End, Fn);