Investigate increased contract instrumentation costs

[athei]

The weight for contract instrumentation per byte increased by a factor of 20x. I am not sure why and when this happened.

This regression should be bisected and corrected.

[athei]

After some investigation I found out the following:

• The regression only exists in the wasm runtime but not in the native runtime.
• It was introduces by this PR.

This would suggest that the regression is somehow related to the initialization of the RuntimeLogger. However the pwasm-utils which do the instrumentation do not contain code to emit log messages. Also, building the benchmarks with disable-logging does not remove the regression.

I analyzed the PR but cannot find anything that could have causes this regression. Maybe @bkchr has an idea.

[bkchr]

So reverting the commit fixes your problem?

[athei]

I did a bisect and this was the first commit that had the regression.

[athei]

I tracked down the culprit. The regression is gone when I remove the trace! statements in this file. Those are tight loops and it makes total sense to see such a slowdown when those are not compiled out.

Specifying disable-logging does not help. This feature does not activate log/release_max_level_off which could explain the problem. However, even if I add this to sp-io the regression is still there. It seems that the tracing code is not properly removed. Maybe the log feature is not properly propagated (maybe something with the new resolver).

[athei]

The issue is being solved by deactivating logging when a runtime is being build for benchmarking (see linked PR).