On Mon, Mar 16, 2015 at 1:54 AM, Davide Italiano <davide at freebsd.org> wrote:> > Shankar's parallel for per-se didn't introduce any performance benefit > (or regression). > If the change I propose is safe, I would like to see Shankar's change > in (and this on top of it). > I have other related changes coming next, but I would like to tackle > them one at a time. >Here's an update. After http://reviews.llvm.org/D8372 , I updated the profiling data. https://people.freebsd.org/~davide/llvm/lld-03162015.svg It seems now 85% of CPU time is spent inside FileArchive::buildTableOfContents(). In particular, 35% of the samples are spent inserting into unordered_map, so there's maybe something we can do differently there (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). Thanks, -- Davide "There are no solved problems; there are only problems that are more or less solved" -- Henri Poincare
On Mon, Mar 16, 2015 at 10:52 PM, Davide Italiano <davide at freebsd.org> wrote:> On Mon, Mar 16, 2015 at 1:54 AM, Davide Italiano <davide at freebsd.org> > wrote: > > > > Shankar's parallel for per-se didn't introduce any performance benefit > > (or regression). > > If the change I propose is safe, I would like to see Shankar's change > > in (and this on top of it). > > I have other related changes coming next, but I would like to tackle > > them one at a time. > > > > Here's an update. > > After http://reviews.llvm.org/D8372 , I updated the profiling data. > > https://people.freebsd.org/~davide/llvm/lld-03162015.svg > It seems now 85% of CPU time is spent inside > FileArchive::buildTableOfContents(). > In particular, 35% of the samples are spent inserting into > unordered_map, so there's maybe something we can do differently there > (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). >Anyone tried a DenseMap instead of an unordered_map? If you need pointer validity to the elements, a DenseMap with unique_ptrs rather than direct values could be an option. Chandler's usual argument here is that walking the map is cheap with high locality (as in a DenseMap) even if the nodes themselves involve indirection. Could be worth an experiment.> > Thanks, > > -- > Davide > > "There are no solved problems; there are only problems that are more > or less solved" -- Henri Poincare > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150316/06782a1d/attachment.html>
On Mon, Mar 16, 2015 at 10:52 PM, Davide Italiano <davide at freebsd.org> wrote:> On Mon, Mar 16, 2015 at 1:54 AM, Davide Italiano <davide at freebsd.org> > wrote: > > > > Shankar's parallel for per-se didn't introduce any performance benefit > > (or regression). > > If the change I propose is safe, I would like to see Shankar's change > > in (and this on top of it). > > I have other related changes coming next, but I would like to tackle > > them one at a time. > > > > Here's an update. > > After http://reviews.llvm.org/D8372 , I updated the profiling data. > > https://people.freebsd.org/~davide/llvm/lld-03162015.svg > It seems now 85% of CPU time is spent inside > FileArchive::buildTableOfContents(). >I'm rather amazed that that patch changed the total CPU time. Just doing the work in parallel shouldn't reduce the total CPU time spent on the task. A reduction in CPU time would happen though if parallelizing it increased the single-threaded performance of the tasks being done in parallel. Perhaps using multiple cores means we are using multiple caches, so each thread is getting much better single-threaded performance due to reduced memory bottlenecking? -- Sean Silva> In particular, 35% of the samples are spent inserting into > unordered_map, so there's maybe something we can do differently there > (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). > > Thanks, > > -- > Davide > > "There are no solved problems; there are only problems that are more > or less solved" -- Henri Poincare > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150316/f3d90ad6/attachment.html>
On Tue, Mar 17, 2015 at 7:17 AM, Sean Silva <chisophugis at gmail.com> wrote:> > > On Mon, Mar 16, 2015 at 10:52 PM, Davide Italiano <davide at freebsd.org> > wrote: >> >> On Mon, Mar 16, 2015 at 1:54 AM, Davide Italiano <davide at freebsd.org> >> wrote: >> > >> > Shankar's parallel for per-se didn't introduce any performance benefit >> > (or regression). >> > If the change I propose is safe, I would like to see Shankar's change >> > in (and this on top of it). >> > I have other related changes coming next, but I would like to tackle >> > them one at a time. >> > >> >> Here's an update. >> >> After http://reviews.llvm.org/D8372 , I updated the profiling data. >> >> https://people.freebsd.org/~davide/llvm/lld-03162015.svg >> It seems now 85% of CPU time is spent inside >> FileArchive::buildTableOfContents(). > > > I'm rather amazed that that patch changed the total CPU time. Just doing the > work in parallel shouldn't reduce the total CPU time spent on the task. A > reduction in CPU time would happen though if parallelizing it increased the > single-threaded performance of the tasks being done in parallel. Perhaps > using multiple cores means we are using multiple caches, so each thread is > getting much better single-threaded performance due to reduced memory > bottlenecking? > > -- Sean Silva > >> >> In particular, 35% of the samples are spent inserting into >> unordered_map, so there's maybe something we can do differently there >> (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). >> >> Thanks, >> >> -- >> Davide >> >> "There are no solved problems; there are only problems that are more >> or less solved" -- Henri Poincare >> _______________________________________________ >> LLVM Developers mailing list >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > >David, Thanks for the input. I'll try DenseMap tomorrow and report results. Sean, I personally was amazed by that too. I cannot exclude some errors in the sampling for hwpmc, I'll try to repeat the profiling and/or use another profiler to see if I can confirm the results. About your other answer, I guess that would require a more fine-grained analysis which includes memory bandwidth, cache misses etc.. I'll try to get to it later this week or in the weekend. For now, I'm just focusing on CPU profiling. Thanks, -- Davide "There are no solved problems; there are only problems that are more or less solved" -- Henri Poincare
On Mon, Mar 16, 2015 at 11:00 PM, David Blaikie <dblaikie at gmail.com> wrote:> > > On Mon, Mar 16, 2015 at 10:52 PM, Davide Italiano <davide at freebsd.org> > wrote: >> >> On Mon, Mar 16, 2015 at 1:54 AM, Davide Italiano <davide at freebsd.org> >> wrote: >> > >> > Shankar's parallel for per-se didn't introduce any performance benefit >> > (or regression). >> > If the change I propose is safe, I would like to see Shankar's change >> > in (and this on top of it). >> > I have other related changes coming next, but I would like to tackle >> > them one at a time. >> > >> >> Here's an update. >> >> After http://reviews.llvm.org/D8372 , I updated the profiling data. >> >> https://people.freebsd.org/~davide/llvm/lld-03162015.svg >> It seems now 85% of CPU time is spent inside >> FileArchive::buildTableOfContents(). >> In particular, 35% of the samples are spent inserting into >> unordered_map, so there's maybe something we can do differently there >> (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). > > > Anyone tried a DenseMap instead of an unordered_map? If you need pointer > validity to the elements, a DenseMap with unique_ptrs rather than direct > values could be an option. Chandler's usual argument here is that walking > the map is cheap with high locality (as in a DenseMap) even if the nodes > themselves involve indirection. Could be worth an experiment. >I did now. It actually makes things slower for the aforementioned workload (linking clang). It was worth trying though. Patch, in case somebody wants to try at home: https://people.freebsd.org/~davide/llvm/densemap_membermap.diff Patched: real 1m27.849s user 2m47.373s sys 0m16.370s real 1m29.583s user 2m47.771s sys 0m16.816s real 1m25.956s user 2m43.397s sys 0m15.254s real 1m29.380s user 2m47.618s sys 0m15.386s real 1m25.426s user 2m43.388s sys 0m16.961s Unpatched: real 1m26.872s user 2m46.999s sys 0m16.540s real 1m28.187s user 2m47.084s sys 0m17.149s real 1m24.814s user 2m43.311s sys 0m16.979s real 1m25.011s user 2m43.184s sys 0m16.975s real 1m25.536s user 2m44.577s sys 0m16.784s -- Davide "There are no solved problems; there are only problems that are more or less solved" -- Henri Poincare
> Here's an update. > > After http://reviews.llvm.org/D8372 , I updated the profiling data. > > https://people.freebsd.org/~davide/llvm/lld-03162015.svg > It seems now 85% of CPU time is spent inside > FileArchive::buildTableOfContents(). > In particular, 35% of the samples are spent inserting into > unordered_map, so there's maybe something we can do differently there > (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). >Why do we even need to build the table from name to member? Can't we just walk "archive->symbols()" and check for each symbol if it is needed by the current link status? Cheers, Rafael
Rafael, Your latest benchmark results look great. LLD took 1.38 seconds where gold --threads takes 0.85 seconds. It needs to be faster, but that's not too bad. On Thu, Mar 19, 2015 at 10:13 AM, Rafael Espíndola < rafael.espindola at gmail.com> wrote:> > Here's an update. > > > > After http://reviews.llvm.org/D8372 , I updated the profiling data. > > > > https://people.freebsd.org/~davide/llvm/lld-03162015.svg > > It seems now 85% of CPU time is spent inside > > FileArchive::buildTableOfContents(). > > In particular, 35% of the samples are spent inserting into > > unordered_map, so there's maybe something we can do differently there > > (e.g. , Rui's proposal of a concurrent map doesn't seem that bad). > > > > Why do we even need to build the table from name to member? > > Can't we just walk "archive->symbols()" and check for each symbol if > it is needed by the current link status?Are you suggesting we do linear search instead of hash table lookup each time ArchiveFile::find(StringRef symbolName) is called? -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150320/8e67ab35/attachment.html>