4 ms·
Could LLVM be speeded up by passing it the data in a more efficient structure? Or by slimming down the data it's passed in some way?
by AndrewDucker 3y ago
Could LLVM be speeded up by passing it the data in a more efficient structure? Or by slimming down the data it's passed in some way?
- Sharlin 3y agoIt’s well known that the LLVM IR bytecode generated by rustc is terribly verbose – Rust constructs mostly get lowered very naively to a buttload of redundant IR, to be then pruned and condensed by the backend. This is by design, as it helps keep the frontend as simple and fast as possible, but there are certainly cases where it would be a net benefit to move some of the complexity to rustc in order to lighten LLVM’s workload.