Hi, This patch stores the functional unit that the instruction is in in or32.c rather than setting it in *every* instruction executed, wether it is needed or not. I don't know how to clasify the vector instructions so I just used it_unknown. Whenever they get implemented then they can be corrected. The double precision floating point instructions I just clasified as it_float, again this can be corrected when they get implemented. The rotate instructions I clasified as it_shift. The rest stayed as they were in the origional code. In l_invalid there was this comment: /* It would be hard to handle this case for statistics; we skip it since it should not occur anyway: IFF (config.cpu.dependstats) current->func_unit = it_unknown; */ The comment is rather misleading as iqueue gets cleared, thus setting current->func_unit to 0, which just happens to be it_unknown. ChangeLog: * Avoid doing a store in *every* instruction executed by storeing the instruction function unit in or32_opcodes. nog. -------------- next part -------------- A non-text attachment was scrubbed... Name: patch4.gz Type: application/x-gunzip Size: 8429 bytes Desc: not available Url : http://www.opencores.org/forums/openrisc/attachments/20050125/47b250f0/patch4.bin