[wasm][aot] Optimize 64 bit const shuffles. Otherwise prefer vector swizzle.

[lewing]

Fall back to element access for 64x2 const elements, otherwise prefer the vectorized version.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @vitek-karas
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull Request Overview

This PR optimizes the implementation of the OP_WASM_SIMD_SWIZZLE operation for constant indices. Key changes include a revised handling of constant versus non‐constant swizzle index vectors, the removal of an early bitcast of rhs, and an updated combination of computed index vectors using a bitwise OR instead of addition.

[lewing]

The codegen for the const case is not pretty but it is roughly equivalent to the old codegen.

out of curiosity I checked what the codegen for i64x2 non-const case looks like where llvm synthesizes min

 local $4 v128
 local.get $0
 local.get $1
 v128.load align:4    [SIMD]
 local.get $2
 v128.load align:4    [SIMD]
 local.tee $4
 v128.const 0x00000000000000020000000000000002    [SIMD]
 v128.const 0xffffffffffffffffffffffffffffffff    [SIMD]
 v128.const 0x00000000000000000000000000000000    [SIMD]
 local.get $4
 i64x2.extract.lane 0    [SIMD]
 i64.const 2
 i64.lt.u
 select
 i64.const -1
 i64.const 0
 local.get $4
 i64x2.extract.lane 1    [SIMD]
 i64.const 2
 i64.lt.u
 select
 i64x2.replace.lane 1    [SIMD]
 v128.bitselect    [SIMD]
 i32.const 3
 i8x16.shl    [SIMD]
 v128.const 0x08080808080808080000000000000000    [SIMD]
 i8x16.swizzle    [SIMD]
 v128.const 0x07060504030201000706050403020100    [SIMD]
 v128.or    [SIMD]
 i8x16.swizzle    [SIMD]
 v128.store    [SIMD]

when the intrinsic exists

 local.get $0
 local.get $1
 v128.load align:4    [SIMD]
 local.get $2
 v128.load align:4    [SIMD]
 v128.const 0x00000004000000040000000400000004    [SIMD]
 i32x4.min.u    [SIMD]
 i32.const 2
 i8x16.shl    [SIMD]
 v128.const 0x0c0c0c0c080808080404040400000000    [SIMD]
 i8x16.swizzle    [SIMD]
 v128.const 0x03020100030201000302010003020100    [SIMD]
 v128.or    [SIMD]
 i8x16.swizzle    [SIMD]
 v128.store    [SIMD]

[lewing]

I made it fall back to the old code only for 64x2, but that case should really just be written by hand.