Introduce qimalloc as an allocator option

[s1na]

Closes #46.

So far it's a really dumb allocator. It simply tries to allocate enough wasm pages everytime the alloc method is called (i.e. pages are not filled with subsequent alloc calls). It also doesn't dealloc. I tested it with runevm+hera and it passes some tests.

[s1na]

Some notes:

• The allocation algorithm is a rough translation of @poemm's C malloc
• It's been only implemented for arch wasm32
• I tried it with a few simple test cases, but not fully sure about correctness

From a rust developer PoV this code is probably an abomination 😄

[poemm]

The C version of dumb malloc() is untested for edge cases and assumes things about LLVM compiler output.

The good news: dumb malloc() compiles to wasm in 15 bytes (or 60 bytes if the preprocessor flag GROWABLE_MEMORY is set to true). This mostly solves the problem of memory allocation in Ewasm -- no host system library is needed for malloc() since its size is nearly negligible.

[axic]

Changed this to use the new crate: https://github.com/wasmx/qimalloc

[s1na]

Looks good to me, can't approve my own PR though :)

[jakelang]

I am pretty sure we don't need the cfg_if dependency here. We can do something like #[cfg(feature = "wee_alloc")] static ALLOC: wee_alloc::WeeAlloc = wee_alloc::WeeAlloc::INIT;

[axic]

It also has/needs that #[global_allocator] annotation. Will that work?

[jakelang]

I think so, attributes apply to the item below. If the item is not compiled in, then the attribute isn't either.

[jakelang]

This looks good enough. We'll re-address the cfg-if question later.

[axic]

Released qimalloc so we can refer to a version and not just a random repo in this crate.