B2G continues to be a major focus of work relating to memory consumption.\u00a0 Even the non-B2G-specific improvements made this fortnight were mostly identified and made to help B2G.<\/p>\n
Gabriele Svelto caused dirty freed pages held onto by jemalloc to be purged when a B2G process experiences memory pressure<\/a>, such as when it goes into the background.\u00a0 This typically reduces a process’ size by about 2 MiB, so it’s a big deal.\u00a0 This was a MemShrink P1 bug.<\/p>\n Benoit Jacob prevented the creation of redundant OpenGL contexts on B2G<\/a>.\u00a0 This saves 750 KiB in the main B2G process, and 750 KiB in every child that uses WebGL.\u00a0 Benoit was able to do this based on data from a B2G-specific heap profiler<\/a> written by Justin Lebar.<\/p>\n I made a large IPC message buffer shrink once all messages from it have been processed<\/a>.\u00a0 This saves either 120 KiB or 248 KiB per child process.<\/p>\n James Lal greatly reduced memory consumption of the ical parser during sync<\/a>.\u00a0 This change also made it much faster.<\/p>\n Justin Lebar and I added a memory reporter for Freetype<\/a>, which uses about 2 MiB of memory in the main B2G process.\u00a0 (This reporter also works on Fennec.)<\/p>\n Andrea Marchesini added a memory reporter for B2G’s gralloc memory<\/a>.<\/p>\n Bill McCloskey made the JavaScript engine clean up more data on memory pressure events<\/a>.<\/p>\n I increased the size of the chunks used by XPT info’s arena allocator<\/a>, which saves about 80 KiB per process on 32-bit, and a bit more on 64-bit.<\/p>\n Randell Jesup fixed a PeerConnection leak in WebRTC code<\/a>.<\/p>\nMiscellaneous<\/h3>\n