wasmi_v1: improve instruction scheduler

[Robbepop]

This PR changes the instruction scheduler of the wasmi_v1 engine.
The new approach seems promising and also is less affected by the performance regressions with newer rustc versions than the current interpreter instruction scheduler.
Note that the instruction scheduler is on the absolute hot path of the interpreter which is why we see significant changes in performance.

We compare the performance of Rust 1.59 because that is the last version before the performance regressions and we also use Rust 1.62 because it is the current stable Rust version.

Rust 1.59

Test	Before	After	Change
tiny_keccak	970.77	793.95 µs	-17.890%
rev_complement	1.1200	991.54 µs	-11.326%
regex_redux	1.1962	1.0809 ms	-9.5504%
count_until	1.7180	1.4201 ms	-17.382%
factorial_recursive	1.1851	1.1655 µs	-1.6003%
factorial_optimized	740.47	635.99 ns	-13.924%
recursive_ok	312.90	304.17 µs	-3.2702%
recursive_trap	28.212	27.951 µs	-0.8383%
host_calls	45.054	43.287 µs	-3.7734%
fib_recursive	12.633	13.025 ms	+3.0991%

Rust 1.62 (Stable)

Test	Before	After	Change
tiny_keccak	1.4818 ms	916.57 µs	-36.988%
rev_complement	1.4554 ms	1.1097 ms	-23.228%
regex_redux	1.4766 ms	1.1896 ms	-19.362%
count_until	2.2027 ms	1.6776 ms	-23.720%
factorial_recursive	1.4349 µs	1.2398 µs	-14.031%
factorial_optimized	935.66 ns	719.33 ns	-23.144%
recursive_ok	354.32 µs	322.68 µs	-8.6313%
recursive_trap	33.905 µs	29.536 µs	-12.983%
host_calls	53.669 µs	49.136 µs	-8.5481%
fib_recursive	15.320 ms	13.496 ms	-11.907%

Conclusion

All in all a significant improvement across the board. However, even the new instruction scheduler still suffers a performance degradation of roughly 15-20% with a newer rustc version compared to Rust 1.59.

[codecov-commenter]

Codecov Report

Merging #376 (ef3be2a) into master (6133346) will decrease coverage by 0.07%.
The diff coverage is 99.07%.

❗ Current head ef3be2a differs from pull request most recent head 3aaff64. Consider uploading reports for the commit 3aaff64 to get more accurate results

@@            Coverage Diff             @@
##           master     #376      +/-   ##
==========================================
- Coverage   79.39%   79.31%   -0.08%     
==========================================
  Files          97       98       +1     
  Lines       11187    11197      +10     
==========================================
- Hits         8882     8881       -1     
- Misses       2305     2316      +11     

Impacted Files	Coverage Δ
wasmi_v1/src/engine/bytecode/mod.rs	100.00% <ø> (ø)
wasmi_v1/src/engine/bytecode/utils.rs	68.29% <ø> (-10.96%)	⬇️
wasmi_v1/src/engine/code_map.rs	57.62% <66.66%> (-30.81%)	⬇️
wasmi_v1/src/engine/exec_context.rs	97.64% <99.29%> (+1.20%)	⬆️
wasmi_v1/src/engine/call_stack.rs	80.64% <100.00%> (+1.33%)	⬆️
wasmi_v1/src/engine/mod.rs	85.18% <100.00%> (ø)
src/module.rs	73.58% <0.00%> (-0.63%)	⬇️
benches/wasm/wasm_kernel/src/rev_complement.rs	0.00% <0.00%> (ø)
src/runner.rs	92.44% <0.00%> (+0.13%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6133346...3aaff64. Read the comment docs.

[Robbepop]

Note that this PR also fixes this rustc issue in parts since the new instruction dispatcher is no longer as heavily affected by the rustc performance regressions as was the case with the previous (old) instruction dispatcher. Between Rust 1.59 and Rust 1.62 (current stable) the old instruction dispatcher regressed by roughly 50-60% in performance according to our benchmarks whereas the new instruction dispatcher is "only" regressed by roughly 20%. Still 20% is a lot though ...