In r247018 I get a ~8x speedup on a synthetic benchmark I tried. Can you validate this fixes the regression?
On Sat, Sep 5, 2015 at 12:56 AM Hans Wennborg <h...@chromium.org> wrote: > On Fri, Aug 14, 2015 at 2:55 AM, Manuel Klimek via cfe-commits > <cfe-commits@lists.llvm.org> wrote: > > Author: klimek > > Date: Fri Aug 14 04:55:36 2015 > > New Revision: 245036 > > > > URL: http://llvm.org/viewvc/llvm-project?rev=245036&view=rev > > Log: > > Add structed way to express command line options in the compilation > database. > > > > Currently, arguments are passed via the string attribute 'command', > > assuming a shell-escaped / quoted command line to extract the original > > arguments. This works well enough on Unix systems, but turns out to be > > problematic for Windows tools to generate. > > > > This CL adds a new attribute 'arguments', an array of strings, which > > specifies the exact command line arguments. If 'arguments' is available > > in the compilation database, it is preferred to 'commands'. > > > > Currently there is no plan to retire 'commands': there are enough > > different use cases where users want to create their own mechanism for > > creating compilation databases, that it doesn't make sense to force them > > all to implement shell command line parsing. > > > > Patch by Daniel Dilts. > > This seems to have caused a bad performance regression for a tool we > use in Chromium. On the file I tried, run-time went from 0.4 s to 3.0 > s (7.5x slow-down), and peak memory usage from 42 MB to 366 MB (8.7x). > > I suspect what's happened is that JSONCompilationDatabase::parse() > became significantly slower because it's now calling > unescapeCommandLine() and allocating a std::vector on each "command" > field during parsing, whereas previously the code wouldn't do that > until JSONCompilationDatabase::getCommands() was called on a specific > file or set of files. > > One idea for fixing this would be to make the second part of > CompileCommandRef be a yaml::Node pointer, and look at the node type: > if it's a ScalarNode, it contains "commands" and needs to be > unescaped; if it's a SequenceNode, it's "arguments". >
_______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
