Can we directly execute LLVM-IR?

4 points by GeorgeTirebiter ↗ HN
LLVM-IR seems to be a kind of "high-level assembly language": no registers (or all registers, depending on Pov), but simple operations. Should we forget about x86, arm, mips, whatever and simply build silicon to run LLVM-IR directly? (test in FPGA first)

1 comment

[ 0.21 ms ] story [ 9.9 ms ] thread
There's already an interpreter for LLVM-IR, lli (https://llvm.org/docs/CommandGuide/lli.html).

A silicon or FPGA implementation would be complicated by the fact that LLVM-IR uses SSA (static single assignment), so common optimizations performed by a compiler backend (e.g. register allocation) would have to be performed in hardware.