Richard Biener <richard.guent...@gmail.com> writes: > On Thu, Jan 9, 2014 at 9:06 PM, Richard Sandiford > <rdsandif...@googlemail.com> wrote: >> I got an off-list request to backport the bswap patterns to 4.8. >> They've been in trunk for a while without any problems being reported >> and they should be relatively safe. >> >> Here's what I applied after testing on mips64-linux-gnu (with >> --with-arch=mips64r2). The only difference is the use of: >> >> [(set_attr "length" "8")] >> >> rather than trunk's: >> >> [(set_attr "insn_count" "2")] > > FYI, just a clarification from a RM perspective - it's ok for target > maintainers > to backport patches like this to the newest maintained release given they > only touch target specific code (even if they are not regressions). We've > always had the policy to allow HW enablement on release branches and > the target maintainers decide whether it's safe and non-intrusive enough.
OK, thanks. That's how I thought it worked and was why I didn't ask explicitly. Hope I didn't rock the boat here. (Sounds like there might have been an IRC discussion or something. :-)) I assumed the patch I just applied to add extra cases to the /proc/cpuinfo detection for Loongson came under the same category. I put that one in 4.7 too because it should be ultra safe and because it's a host thing rather than a target thing (someone upgrading their kernel might otherwise find that gcc suddenly changes behaviour). Thanks, Richard
