This is very interesting, seems like a promising target between wasm and LLVM in difficulty-to-leverage.
I think one of the missing optimizations in qbe is function inlining, I wonder how much further that will push it in those less favorable benchmarks.
When I checked out QBE, the main aspect that dissuaded me from using it as a backend was the inability to link it as a library with a builder for the IR format. I don’t want to ship a binary and write temporary files.
afaik you don’t need to write a temporary file, you can write to qbe stdin.
I suppose this complaint makes sense if your language is already distributed as a single exe though.
This is very interesting, seems like a promising target between wasm and LLVM in difficulty-to-leverage.
I think one of the missing optimizations in qbe is function inlining, I wonder how much further that will push it in those less favorable benchmarks.
When I checked out QBE, the main aspect that dissuaded me from using it as a backend was the inability to link it as a library with a builder for the IR format. I don’t want to ship a binary and write temporary files.
afaik you don’t need to write a temporary file, you can write to qbe stdin.
I suppose this complaint makes sense if your language is already distributed as a single exe though.