LLVM/Clang 3.3 Should Be Close To Building Linux Kernel
Aside from Intel MKL and the state of glibc, another interesting topic at this week's Linux Foundation Collaboration Summit was in regards to LLVM/Clang for building the Linux kernel rather than GCC. Building the Linux kernel with LLVM/Clang has long been pursued by developers and something we have been talking about on Phoronix quite a bit in the past two years or so.
Up to now, for building the Linux kernel with Clang on either ARM or x86 has required patches to both the compiler and kernel itself. Developers have been after "Clang'ing the kernel" due the compilers faster build times over GCC, better debugging support, Clang's static analysis tools, some performance advantages, more modular architectures, and greater code portability when code is not GCC-specific.
As far as the new information shared at LFCS 2013 about LLVM/Clang for the kernel, there's some interesting bits. All of the required patches to LLVM/Clang for building the Linux kernel are now upstream. LLVM/Clang 3.3 will "likely work mostly out-of-the-box for the Linux kernel" but when using their kernel patches. The Clang integrated assembler (IA) is also disabled at this time due to the kernel's inline Assembly syntax.
The developers' kernel patches for enabling Clang usage come down to Kbuild support so it's no longer GCC-specific, explicit Assembly to handle register variables, removing VLAIS (Variable Length Arrays In Structs), and some other small code changes to make the code more portable and compatible with LLVM/Clang.
Open TODO list items are upstreaming the VLAIS patches along with a few of the other patches, enabling Clang Integrated Assembler support, getting the Clang checker to work on the kernel, and other random items.
More details on the state of building the Linux kernel with Clang can be found from the PDF slides. More details and links to the patches and other information is also available from the LLVM Linux project page.