Hello, I've made more progress in my quest for improving performance of poppler.
1. https://bugs.freedesktop.org/show_bug.cgi?id=8111 Currently every Dict::find(char *) (which are a majority) needs to convert char * to UGooString, which involves memory allocations. I've implemented ability to directly find(char *) by adding UGooString::cmp(char *) that has the same semantics as UGooString::cmp(UGooString(char*)). Also converting tabs to spaces so that it always has correct indentation (regardless of your current tab settings) and small refactoring to use dictLookupBool() and dictLookupInt() for simpler code. 2. https://bugs.freedesktop.org/show_bug.cgi?id=8112 Currently during Parser::getObj() and when adding values to dictionaries poppler makes unnecessary copies of objects. When objects are strings, this means reallocating memory. Since Parser::getObj() is called (literally) in N*100k times, those memory allocations contribute significantly to the execution time (e.g. delete is 4th most expensive function). I've made really only two small changes: a) avoid memory copy by giving the caller an ownership of the object and nulling the object in one code path b) optimizing UGooString usage of key variable (it was always constructed but is not always needed) Also adds helper Dict::add* functions to make b) (and maybe other future changes like that) possible. This change gives me a consistent ~7% speedup for PDF loading stage. Patches are attached to respective bugs. Or you can get this and my other improvements from http://blog.kowalczyk.info/software/sumatrapdf/develop.html I think there are more opportunities for improvements like 2) although even more improvements would come from improving various Stream::getChar() methods (currently Lexter::getChar(), EmbedStream::getChar() and FlateStream::getChar() are in top 5 of most exensive methods during loading. I haven't yet found a way to improve that. -- kjk Sumatra (PDF Viewer for Windows): http://blog.kowalczyk.info/software/sumatrapdf _______________________________________________ poppler mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/poppler
