Apply val_cnt and pop_cnt to pop_label()

[zhouwfang]

• Apply val_cnt and pop_cnt.
• Remove Label by introducing Frame::labels_cnt.

There is not much performance improvement on Linux due to this PR, which replaces

let values_cnt: usize = frame.labels[i ..].iter().map(|label| label.values_cnt).sum();

in pop_label() by pop_cnt + val_cnt.

#46

[zhouwfang]

@ia0 Do you expect this to be removed eventually? Only from the performance perspective, I'm not sure how much gain there would be by removing it. WDYT?

[ia0]

Yes, I expect this to be eventually removed (as well as push_label()). This is not only cleaner, but also useful for when we'll write a compiler to our own bytecode (or an ISA) because it means we don't need to compile instructions like Block or Loop that don't do anything except indicate a scope.

But this can be done in a separate PR.

[ia0]

Thanks! That's a really nice cleanup. I'll push a small review commit and merge.

[ia0]

Yes, I expect this to be eventually removed (as well as push_label()). This is not only cleaner, but also useful for when we'll write a compiler to our own bytecode (or an ISA) because it means we don't need to compile instructions like Block or Loop that don't do anything except indicate a scope.

But this can be done in a separate PR.

[ia0]

Before change:

Peak RAM usage: 6692
CoreMark result: 0.13914786 (in 144.034s)

After change:

Peak RAM usage: 6244
CoreMark result: 0.1541212 (in 130.039s)

[zhouwfang]

@ia0 Could you also get the current code size? Thanks!

[ia0]

@ia0 Could you also get the current code size? Thanks!

You just need to run rust-size ../../target/thumbv7em-none-eabi/release/wasm-bench (you might need cargo install cargo-binutils first) after ./run.sh nordic base. For this PR, it went from 144KiB to 141KiB.