> 2023年4月21日 13:21，Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx> 写道： > > On Fri, Apr 21, 2023 at 12:38:07PM +0100, Jiaxun Yang wrote: >> >> >>> 2023年4月21日 11:48，Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx> 写道： >>> >>> On Sun, Apr 09, 2023 at 11:43:06AM +0100, Jiaxun Yang wrote: >>>> parse_r helper is used to parse register number from oprand, >>>> it only handles oprand which use number to refer register before. >>>> >>>> However compiler may use $gp and $sp to reference register 29 and >>>> 30. Handle this situation by adding relevant match name. >>> >>> are you sure the compiler is doing that ? Which compiler is this ? >>> To me this feels more like missing #defines from asm/regdef.h for >>> some inline assembly stuff... >> >> Yes, clang is doing that, see[1]. >> >> There is no inline assembly mentioning $fp for cfcmsa but clang is still generating $fp for >> cfcmsa. >> >> I was able to reproduce this by dumping assembly generated by clang, but I’m >> unable to reproduce it for now, it seems like clang will only allocate $fp in rare cases. > > so is this clang miss-behaviour ? If yes fix should be done there. I think it is not forbidden. Clang is doing nothing wrong and as long as assembler is ok with that parse_r should able to handle it. Thanks Jiaxun > > Thomas. > > -- > Crap can work. Given enough thrust pigs will fly, but it's not necessarily a > good idea. [ RFC1925, 2.3 ]
