Buffer account's fields for hash

[HaoranYi]

Problem

Hashing individual fields of the account can be optimized by buffering those
fields and hash them together. Given larger buffer, the hasher can do better
with SIMD parallelism.

https://github.com/HaoranYi/my_bench/blob/master/benches/mybench.rs

     Running benches/mybench.rs (target/release/deps/mybench-5340cc83872f1fb6)
h1                      time:   [185.50 ns 185.78 ns 186.07 ns]

h2                      time:   [222.28 ns 222.62 ns 223.02 ns]

Summary of Changes

buffer individual accounts fields together for hashing

Fixes #

[codecov]

Codecov Report

Merging #33788 (7664ebc) into master (673a38c) will decrease coverage by 0.1%.
Report is 42 commits behind head on master.
The diff coverage is 100.0%.

@@            Coverage Diff            @@
##           master   #33788     +/-   ##
=========================================
- Coverage    81.8%    81.8%   -0.1%     
=========================================
  Files         806      806             
  Lines      217859   217865      +6     
=========================================
- Hits       178239   178229     -10     
- Misses      39620    39636     +16     

[brooksprumo]

Neat idea. Since the buffer is only 80 bytes, we could probably allocate the buffer on the stack and that should shave more time off too (avoid the heap allocation).

[brooksprumo]

Also, are the computed hash values equivalent?

[HaoranYi]

yeah. good idea. pushed a commit to use smallvec for the buffer.

yeah. the hash matches.
https://github.com/HaoranYi/my_bench/blob/master/src/main.rs

[brooksprumo]

Why 128 bytes? Looks like the largest we need is 65 bytes? Does SmallVec require the value to be a power of 2?

Looking at the source, I think SmallVec will accept 96 here too. Alternatively, we can use the const_generics feature and that looks like it'll support arrays of any size:

https://docs.rs/smallvec/latest/src/smallvec/lib.rs.html#772-778

[HaoranYi]

yup. 96 can be used.

#[cfg(any(not(feature = "const_generics"), doc))]
[impl_array!](https://docs.rs/smallvec/latest/src/smallvec/lib.rs.html#2406-2416)(
    0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25,
    26, 27, 28, 29, 30, 31, 32, 36, 0x40, 0x60, 0x80, 0x100, 0x200, 0x400, 0x600, 0x800, 0x1000,
    0x2000, 0x4000, 0x6000, 0x8000, 0x10000, 0x20000, 0x40000, 0x60000, 0x80000, 0x10_0000
);

[HaoranYi]

With 128-byte buffer, we can optimize for small accounts, where data + fields fits into the buffer - hash them all at once.

[brooksprumo]

Nicely done. Lgtm

[mergify]

⚠️ The sha of the head commit of this PR conflicts with #33846. Mergify cannot evaluate rules on this PR. ⚠️