Don't grow memory on first allocation

[athei]

Right now the allocator always grows the memory on the first allocation. This is because it isn't aware of how much heap is available on initialization.

Growing the memory is a very expensive operation. I can imagine that this is especially bad for calls which do little work but still allocate some memory. Not sure if we have such a performance profile in our waterfall. But I can imagine that this impacts getter functions.

This solution avoids this by first using the available heap. It is not perfect as it performs a bit more work per allocation. We could probably get rid of that with enough willpower and unsafe.

[HCastano]

Couple questions around Wasm instructions

[codecov-commenter]

Codecov Report

Merging #1712 (19c420b) into master (3d6784c) will decrease coverage by 0.02%.
The diff coverage is 37.50%.

@@            Coverage Diff             @@
##           master    #1712      +/-   ##
==========================================
- Coverage   52.98%   52.97%   -0.02%     
==========================================
  Files         212      212              
  Lines        6726     6728       +2     
==========================================
  Hits         3564     3564              
- Misses       3162     3164       +2     

Files Changed	Coverage Δ
crates/allocator/src/bump.rs	85.15% <37.50%> (-1.36%)	⬇️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[athei]

Merged master and answered questions. Ready for another round.

[xgreenx]

Did I get right that there is one available page by default, which is why this change will help?

[athei]

Did I get right that there is one available page by default, which is why this change will help?

Yes. But not necessarily one. Could be more. Depends on the program. rustc will request as many initial pages as it needs to place its stack and other data. The point is that the last page will have some free space we can use without doing more allocations.