The embedded high-performance language Accelerate has got a new backend based on LLVM, which generates code for multicore GPUs and CPUs. In a new paper, we describe this code generator, which is —as...