I told you guys it would be using LLVM IR. Pretty much every OpenCL compiler already uses Clang/LLVM for its implementation. Standardizing on LLVM makes sense.
>>5
Any language or virtual machine which is not designed to make use of modern hardware features is shit.
I haven't seen a VM, other than LLVM, which exposes a processors SIMD pipeline. I haven't seen a VM which allows you to run code on multiple heterogeneous processing cores, except for OpenCL/SPIR.
Let me know when V8 is ported to run on OpenCL/SPIR. Until then, it's shit.