From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeffrey A Law To: hjl@lucon.org (H.J. Lu) Cc: wilson@cygnus.com, scox@cygnus.com, crux@pool.informatik.rwth-aachen.de, egcs@cygnus.com Subject: Re: More fp bug in egcs Date: Fri, 08 May 1998 16:04:00 -0000 Message-id: <16552.894658448@hurl.cygnus.com> References: X-SW-Source: 1998-05/msg00297.html In message < m0yWSyp-000268C@ocean.lucon.org >you write: > * reload1.c (emit_reload_insns): Don't output the last reload > insn if OLD is not the dest of NSN and is in the src and is > clobbered by INSN. I played with this a little the other day on the assumption that we could consider it as an optimization patch. I did bootstraps with an instrumented compiler to detect when this optimization could be applied. For an x86 -O2 bootstrap it triggered only about 15 times; I analyzed at some (but not all) of the cases where this change was able to optimize code better. They all looked like safe optimizations (which isn't a suprise given the purpose of the patch). Sometimes we deleted a store into memory other times it was just a reg-reg copy. The optimization never triggered during a PA bootstrap. Which, isn't a big suprise. I went ahead and installed a modified version of the patch which only triggers when -fexpensive-optimizations is enabled. jeff