Closed xdoardo closed 2 weeks ago
I'm closing as the evidence shows that the number above is correct. We compiled coremark
from scratch and ran it with wasmer
3.3.0, 5.0.0 and the latest wasmtime
. We confirm that in wasmer
3.3.0 coremark
gives a result around ~30000, while latest wasmer
and wasmtime
score ~11500.
This, I believe, means that we're not missing any obvious optimization on our cranelift backend.
We recently ran benchmarks as part of the latest release. The
coremark
benchmark result with the cranelift backend is suspiciously different from LLVM's:We should investigate if we are missing any optimization that can impact this result.