Just to close the loop, Nick complained that this also broke a case were the inline assembly was valid, but just not implemented yet in MC. In the end we went with just propagating -no-integrated-as, so that
* clang -S foo.c: parses inline assembly. * clang -S foo.c -no-integrated-as: doesn't parse inline assembly. I don't like the idea of supporting "inline assembly" without assembly in it. This seems like another case where compilers get smarter over time and see user tricks. If we can get by with just a big on/off switch like -no-integrated-as, I think we should. Cheers, Rafael On 21 February 2014 05:10, Richard Sandiford <[email protected]> wrote: > Reading this back I realise the last part was lacking the rider: > > Richard Sandiford <[email protected]> writes: >> Given that the compiler can't exploit its knowledge the asm contents, > > ...as a substitute for the constraints. > > It's OK to use the integrated assembler to get a more accurate idea > of how long the asm is, or how best to schedule it. But not being > able to parse it for those purposes seems like warning rather than > error territory. > > Thanks, > Richard > > _______________________________________________ > cfe-commits mailing list > [email protected] > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
