Optimize host calls in the wasmi interpreter

[Robbepop]

This commit does two things to speed-up host calls:

• Reuse heap allocations for consecutive host calls to hold arguments.
• More efficient copying of stack values to host argument buffer.

This optimization was carved out from #287 and it would be nice to have it included so that both v0 and v1 implementation have a more comparable performance once wasmi benchmarks are more precise.

Benchmarks

No noticeable performance impact was measured, however, that's mainly because the current set of benchmarks do not cover the particular scenario of performing lots of host function calls.

Before

test bench_regex_redux ... bench:   1,831,339 ns/iter (+/- 153,861)
test bench_rev_comp    ... bench:   1,661,062 ns/iter (+/- 146,230)
test bench_tiny_keccak ... bench:   1,276,610 ns/iter (+/- 132,734)
test fac_opt           ... bench:      23,268 ns/iter (+/- 4,547)
test fac_recursive     ... bench:      24,568 ns/iter (+/- 1,908)
test recursive_ok      ... bench:     615,655 ns/iter (+/- 43,826)
test recursive_trap    ... bench:      79,457 ns/iter (+/- 4,446)

After

test bench_regex_redux ... bench:   1,802,265 ns/iter (+/- 130,377)
test bench_rev_comp    ... bench:   1,656,515 ns/iter (+/- 126,081)
test bench_tiny_keccak ... bench:   1,260,390 ns/iter (+/- 118,418)
test fac_opt           ... bench:      22,251 ns/iter (+/- 1,458)
test fac_recursive     ... bench:      23,990 ns/iter (+/- 2,073)
test recursive_ok      ... bench:     593,989 ns/iter (+/- 86,742)
test recursive_trap    ... bench:      78,235 ns/iter (+/- 5,871)

[codecov-commenter]

Codecov Report

Merging #291 (3b9109b) into master (9556783) will decrease coverage by 0.45%.
The diff coverage is 42.59%.

@@            Coverage Diff             @@
##           master     #291      +/-   ##
==========================================
- Coverage   80.02%   79.56%   -0.46%     
==========================================
  Files          30       31       +1     
  Lines        4696     4738      +42     
==========================================
+ Hits         3758     3770      +12     
- Misses        938      968      +30     

Impacted Files	Coverage Δ
benches/src/lib.rs	0.00% <0.00%> (ø)
src/runner.rs	92.30% <85.18%> (-0.27%)	⬇️

---

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 9556783...3b9109b. Read the comment docs.

[Robbepop]

Added benchmarks in 1455e9a.

Run the benchmark using:

cargo bench --manifest-path benches/Cargo.toml host_call

Before

running 1 test
test host_calls        ... bench:     135,455 ns/iter (+/- 19,820)

After

running 1 test
test host_calls        ... bench:      95,340 ns/iter (+/- 13,254)

So we can measure a speed-up of roughly 42% for the benchmark.

[athei]

I would suggest removing the as here. In my mind that means returning the whole thing as slice and not only parts of it.

[Robbepop]

Good suggestion, will do this!