[metal] Plug in the SNodeRep structs into codegen

[k-ye]

This is mostly a refactoring of the SNode structs implementation on Metal, with no functional changes.

I added a SNodeRep_* for each snode type supported on Metal. It is very similar to the node_* files in https://github.com/taichi-dev/taichi/tree/master/taichi/runtime/llvm. Each generated SNode structs will own an rep_ object. These SNodeRep_* provides the interfaces for operations like checking activation, activating/deactivating and getting a child.

Advantages:

• SNodeReps make it much easier to implement listgen and struct-for kernels fo checking activation.
• For SNodeOpStmt, before this PR, we had to manually figure out the memory address of the specific cell, then call the global helper function. Now that these functions are implemented inside each rep, we can just call the corresponding interface on the generated SNode structs themselves.

The constructor of a generated SNode takes in, in addition to its memory address, Runtime and MemoryAllocator now. These aren't used for now, but will be for pointer SNodes.

Here's an example:

class S1_ch {
 public:
  S1_ch(device byte *a) : addr_(a) {}
  S2 get0(device Runtime *rtm, device MemoryAllocator *ma) {
    return {addr_ + (0), rtm, ma};
  }

  device byte *addr() { return addr_; }

  constant static constexpr int stride = 0 + S2::stride;
 private:
  device byte *addr_;
};

struct S1 {
  // bitmasked
  constant static constexpr int n = 128;
  constant static constexpr int elem_stride = S1_ch::stride;
  constant static constexpr int stride = elem_stride * n + /*bitmasked=*/16;

  S1(device byte *addr, device Runtime *rtm, device MemoryAllocator *ma) {
    rep_.init(addr, /*meta_offset=*/elem_stride * n);
  }

  S1_ch children(int i) {
    return {rep_.addr() + (i * elem_stride)};
  }

  inline bool is_active(int i) {
    return rep_.is_active(i);
  }

  inline void activate(int i) {
    rep_.activate(i);
  }

  inline void deactivate(int i) {
    rep_.deactivate(i);
  }

 private:
  SNodeRep_bitmasked rep_;
};

Related issue = #1174

[Click here for the format server]

---

[codecov]

Codecov Report

Merging #1480 into master will increase coverage by 1.20%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1480      +/-   ##
==========================================
+ Coverage   66.17%   67.38%   +1.20%     
==========================================
  Files          38       38              
  Lines        5405     5593     +188     
  Branches      976      976              
==========================================
+ Hits         3577     3769     +192     
+ Misses       1660     1658       -2     
+ Partials      168      166       -2     

Impacted Files	Coverage Δ
python/taichi/lang/__init__.py	80.92% <100.00%> (+1.08%)	⬆️
python/taichi/lang/kernel.py	81.57% <0.00%> (+0.07%)	⬆️
python/taichi/lang/impl.py	90.60% <0.00%> (+0.29%)	⬆️
python/taichi/misc/util.py	44.58% <0.00%> (+0.35%)	⬆️
python/taichi/lang/snode.py	93.69% <0.00%> (+0.36%)	⬆️
python/taichi/lang/util.py	62.26% <0.00%> (+0.48%)	⬆️
python/taichi/lang/transformer.py	93.50% <0.00%> (+0.48%)	⬆️
python/taichi/lang/meta.py	64.70% <0.00%> (+0.52%)	⬆️
python/taichi/lang/linalg.py	90.00% <0.00%> (+0.66%)	⬆️
python/taichi/lang/matrix.py	91.50% <0.00%> (+0.75%)	⬆️
... and 7 more

---

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 a6575fb...2147d38. Read the comment docs.

[yuanming-hu]

Looks great! Thank you!