stop using custom wasm simd for slide_hash

[folkertdev]

Fixes #301

On current stable rust, this produces (from what I can tell) better and also shorter code.

https://godbolt.org/z/b3dh3n8Ma

One of these day's I'll get wasmtime to work on godbolt. It should be possible (to compile with llvm for wasm and then use wasmtime to get the generated assembly). I did it manually for now and the assembly also did look better and shorter.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Flag	Coverage Δ
fuzz-compress	41.78% <0.00%> (-0.06%)	⬇️
fuzz-decompress	30.14% <0.00%> (-0.05%)	⬇️
test-aarch64-apple-darwin	90.30% <0.00%> (+<0.01%)	⬆️
test-x86_64-apple-darwin	88.42% <100.00%> (+0.01%)	⬆️
test-x86_64-unknown-linux-gnu	89.94% <0.00%> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
zlib-rs/src/deflate/slide_hash.rs	98.96% <100.00%> (+0.01%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[bjorn3]

Yeah, looks better. The new code only has a single loop while the old code has two loops. And the new code folds the offsets into the vector load and store instructions while the old code emits separate addition instructions.

Output of wasm-decompile from the WebAssembly Binary Toolkit (WABT)

export memory memory(initial: 16, max: 0);

global stack_pointer:int = 1048576;
export global data_end:int = 1048576;
export global heap_base:int = 1048576;

table T_a:funcref(min: 1, max: 1);

export function old(a:simd_ptr@1, b:simd_ptr@1, c:int) {
  var d:int = b & -8;
  if (eqz(d)) goto B_a;
  var e:simd = i16x8_splat(c);
  c = b + -8;
  b = (c >> 3) + 1 & 3;
  if (eqz(b)) goto B_b;
  d = d - (b << 3);
  loop L_c {
    a[0] = a[0] i16x8_sub_sat_u e;
    a = a + 16;
    b = b + -1;
    if (b) continue L_c;
  }
  label B_b:
  if (c < 24) goto B_a;
  loop L_d {
    a[0] = a[0] i16x8_sub_sat_u e;
    b = a + 16;
    b[0] = b[0] i16x8_sub_sat_u e;
    b = a + 32;
    b[0] = b[0] i16x8_sub_sat_u e;
    b = a + 48;
    b[0] = b[0] i16x8_sub_sat_u e;
    a = a + 64;
    d = d + -32;
    if (d) continue L_d;
  }
  label B_a:
}

export function new(a:simd_ptr@2, b:int, c:int) {
  b = b & -32;
  if (eqz(b)) goto B_a;
  b = 0 - b;
  var d:simd = i16x8_splat(c);
  loop L_b {
    a[0] = (a[0] i16x8_max_u d) i16x8_sub d;
    a[8] = (a[8] i16x8_max_u d) i16x8_sub d;
    a[16] = (a[16] i16x8_max_u d) i16x8_sub d;
    a[24] = (a[24] i16x8_max_u d) i16x8_sub d;
    a = a + 64;
    b = b + 32;
    if (b) continue L_b;
  }
  label B_a:
}