Wasmi IR v3.0

[Robbepop]

Closes #1642.

New Wasmi IR bytecode that stores immediates inline instead of relying on function local constants.
This new Wasmi IR is generated via build.rs script and thus out of the box way more consistent than its precursor.
The new IR supports encoding the instruction stream into a byte stream and decode it from a byte stream back into instructions.

The following features are supported:

• enum Op: type similar to Wasmi's current Instruction type
• enum OpCode: Op discriminants encoded as u16 - helps with threaded code encoding
• Encode & Decode: safe encoding of Op (and its components) and unsafe but efficient decoding of Op.
  • This encode and decode scheme was designed with direct and indirect threaded code architecture in mind.
• Encoder and Decoder traits that are going to be implemented in the wasmi crate by respective types.

Unblocks the following issues:

• Optimization: no longer copy parameters for function calls #1632: due to having no need for function local constants
• Move function local constants between function locals and temporaries #1598: closed since it was an alternative solution to cope with issues connected to function local constants.
• Experiment with direct threading for Wasmi's execution #1516
• wasmi_ir: see if build.rs script is more efficient than macro_rules for codegen #1485: as this is based on build.rs codegen we found this to be roughly twice as fast as our current macro approach for fresh builds.
• Experiment: Try to share signatures for all Wasmi bytecode executors #1433: the codegen for Decode allows to more easily share signatures.
• wasmi_ir: Add proper IrEncoder to encode Instructions as bytes #1324
• Change Wasmi function call and return ABI #1639: without function local constants this is a no-brainer.

ToDos

• Add missing simd operators.
  • v128.load memarg: (i32) -> (v128)
  • v128.store memarg: (i32 v128) -> ()
  • v128.load8x8_s memarg: (i32) -> (v128)
  • v128.load8x8_u memarg: (i32) -> (v128)
  • v128.load16x4_s memarg: (i32) -> (v128)
  • v128.load16x4_u memarg: (i32) -> (v128)
  • v128.load32x2_s memarg: (i32) -> (v128)
  • v128.load32x2_u memarg: (i32) -> (v128)
  • v128.load8_splat memarg: (i32) -> (v128)
  • v128.load16_splat memarg: (i32) -> (v128)
  • v128.load32_splat memarg: (i32) -> (v128)
  • v128.load64_splat memarg: (i32) -> (v128)
  • v128.load8_lane memarg laneidx: (i32 v128) -> (v128)
  • v128.load16_lane memarg laneidx: (i32 v128) -> (v128)
  • v128.load32_lane memarg laneidx: (i32 v128) -> (v128)
  • v128.load64_lane memarg laneidx: (i32 v128) -> (v128)
  • v128.store8_lane memarg laneidx: (i32 v128) -> ()
  • v128.store16_lane memarg laneidx: (i32 v128) -> ()
  • v128.store32_lane memarg laneidx: (i32 v128) -> ()
  • v128.store64_lane memarg laneidx: (i32 v128) -> ()
  • v128.load32_zero memarg: (i32) -> (v128)
  • v128.load64_zero memarg: (i32) -> (v128)
• Add missing relaxed-simd operators.
  • I16x8RelaxedDotI8x16I7x16S
  • I32x4RelaxedDotI8x16I7x16AddS
  • F32x4RelaxedMadd
  • F32x4RelaxedNmadd
  • F64x2RelaxedMadd
  • F64x2RelaxedNmadd

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.76%. Comparing base (5439ae6) to head (0e1e32d).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1650      +/-   ##
==========================================
- Coverage   70.78%   70.76%   -0.02%     
==========================================
  Files         179      179              
  Lines       15199    15199              
==========================================
- Hits        10758    10755       -3     
- Misses       4441     4444       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.