as of yesterday, i've got a working (=boots into X) amd64 kernel (2.6.35.7) by only patching a hundred lines or so in linux itself (for either linux or clang bugs).
Posted Oct 12, 2010 16:09 UTC (Tue) by bronson (subscriber, #4806)
[Link]
Very cool! Have you formed a gut feeling for performance? Either compile time or generated code?
Just curious about real-world experiences.
LLVM 2.8 is available
Posted Oct 12, 2010 17:17 UTC (Tue) by PaXTeam (subscriber, #24616)
[Link]
> Have you formed a gut feeling for performance? Either compile time or generated code?
i haven't had time for benchmarks (and i don't actually know what would exercise various kernel subsystems/paths well enough, kernel compilation isn't exactly it ;) but based on the generated code and overall code size, clang generates about 3% fatter code than gcc (at least with -Os, i didn't try -O2). as for compile time, i only tested llvm 2.7 last spring and the times for compiling the same kernel/.config were something like this: gcc 4.4.x: 20 mins, llvm-gcc: 18 mins, clang: 16 mins.