Date: Sun, 27 Apr 2003 11:31:03 -0700 (PDT) From: Marcel Moolenaar <marcel@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 29883 for review Message-ID: <200304271831.h3RIV3L7077081@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=29883 Change 29883 by marcel@marcel_nfs on 2003/04/27 11:30:56 Prefer ar.bsp over ar.bspstore. The value of ar.bspstore is supposed to be equal to the value of ar.bsp due to flushing the register stack, but a PAL bug existed where a PMI could cause RSE spills or fills to fail to complete. This makes me wonder: why use a register that has a more dynamic nature than one that is invariant over the course of this function and has the value we're interested in? This change doesn't fix a problem (alas, because it means I still have to fix the problem I hoped this would fix :-) and is probably gratuitous for all practical purposes but it has higher quality this way. For example: ar.bsp may have faster access times because it's invariant. The RSE has to synchronize access to ar.bspstore with RSE operation... Affected files ... .. //depot/projects/ia64_epc/sys/ia64/ia64/context.s#13 edit Differences ... ==== //depot/projects/ia64_epc/sys/ia64/ia64/context.s#13 (text+ko) ==== @@ -197,7 +197,7 @@ } { .mmi st8 [r31]=r16,16 // pr - mov r17=ar.bspstore + mov r17=ar.bsp mov r16=ar.pfs ;; }
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200304271831.h3RIV3L7077081>