Content-Type: text/plain; charset=iso-2022-jp Content-Transfer-Encoding: 7bit Hi, I would like to release the alpha level patch to support page granularity hint which means that you will be able to reduce translation buffer miss aggressively with this patch. In my EB164, the performance of a large matrix transpose is 37% faster than the normal linux kernel. This is my second patch to support the page GH, but is my first patch for general use. It seems to work fine for me, but I think it is still alpha state. So please be careful to use this patch. This patch is for the 2.4.0-test1 kernel. Sorry, no 2.2.x patch is exist because I used zone memory allocator introduced in the 2.3.x kernel. You can set GH bits for BRK area or MMAP area separately. If your program calls malloc system call turn on the MMAP, others should turn on the BRK or both of them. You can get the patch from my personal web. http://shimizu-lab.et.u-tokai.ac.jp/~nshimizu/page_gh2.patch You can get matrix transpose benchmark also, http://shimizu-lab.et.u-tokai.ac.jp/~nshimizu/mem2.c This benchmark has two parts which allocate memory by malloc and sbrk. If you turn off one of the option, you can check the performance difference of the GH area and non GH area. Please test and give me some comments. Naohiko Shimizu Dept. Communication Engr./Univ. TOKAI 1117 Kitakaname Hiratsuka 259-12 Japan TEL.+81-463-58-1211(ext. 4084) FAX.+81-463-58-8320 http://shimizu-lab.et.u-tokai.ac.jp/
