[Wasm RyuJIT] Start generating relocations for call_indirect

[kg]

• Add various relocation types from the wasm linking spec to CorInfoReloc and RelocType
• Add new instruction formats for relocated constant payloads like IF_CALL and IF_MEMADDR
• Use relocations in emitIns_Call
• Add pseudo-instructions for relocated i32_const
• Make WasmTypeNode an ISymbolNode so it can be a reloc target

[dotnet-policy-service]

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

[Copilot]

Pull request overview

This PR updates the WASM JIT emitter to begin supporting relocatable immediates for call_indirect, by representing certain immediates as “padded reloc” placeholders and centralizing constant emission logic.

Changes:

• Use getWasmTypeSymbol and store the returned handle in the call_indirect immediate as a relocatable constant.
• Teach instrDesc::idCodeSize() to account for padded reloc sizes for call_indirect immediates.
• Add helper routines to emit either normal (S)LEB128 constants or fixed-size padded relocation placeholders.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/coreclr/jit/emitwasm.h	Declares new helper APIs for emitting padded relocations/constants.
src/coreclr/jit/emitwasm.cpp	Implements padded-reloc emission and switches call_indirect immediate handling to use a relocatable constant path.

[kg]

Current status:

Generating: N003 (???,???) [000001] --CXG+-----                         *  CALL r2r_ind void   Program:voidFunc()
Added IP mapping: 0x0001 STACK_EMPTY CALL_INSTRUCTION (G_M45341_IG02,ins#4,ofs#8)
IN0005:             i32.const 0
IN0006:             i32.load 0 0
IN0007:             call_indirect 4611686018431713352 0      ;; Program:voidFunc()

~~~ snip ~~~

IN0005: 00001E      i32.const 0
IN0006: 000020      i32.load 0 0
recordRelocation: 000001D2A75F2E7C (rw: 000001D2A75F2E7C) => 4000000000420048, type 28 (relocType), delta 0
Unhandled exception. ILCompiler.CodeGenerationFailedException: Code generation failed for method '[wasm-test]Program.callVoidFunc()'
 ---> System.ArgumentException: Unsupported relocation type: R_WASM_TYPE_INDEX_LEB
   at Internal.JitInterface.CorInfoImpl.GetRelocType(CorInfoReloc)
   at Internal.JitInterface.CorInfoImpl.recordRelocation(Void*, Void*, Void*, CorInfoReloc, Int32)
   at Internal.JitInterface.CorInfoImpl._recordRelocation(IntPtr, IntPtr*, Void*, Void*, Void*, CorInfoReloc, Int32)
   --- End of inner exception stack trace ---
   at Internal.JitInterface.CorInfoImpl.CompileMethodInternal(IMethodNode, MethodIL)

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 5 comments.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

[kg]

cc @dotnet/jit-contrib

I'm not sure we need to actually generate a valid ULEB/SLEB fill pattern here because R2R is going to be replacing the fill pattern with the real value.
I'm also not sure we need to pad relocations in the first place, since we care about code size ideally when R2R replaces the placeholders with the relocated values it should be shrinking the relocations as necessary.

[SingleAccretion]

int32 ... what do you think?

I don't see any strong reason to do hacks here. The relocation write support will also likely be used in the object writer to actually resolve the relocations (similar to e. g. https://github.com/dotnet/runtimelab/blob/c1a9777ae47f937d4e1a5484247ce73c86f45030/src/coreclr/tools/aot/ILCompiler.LLVM/CodeGen/WasmObjectWriter.cs#L310-L318). That needs proper [S|U]LEBs.

Edit: for that kind of use case, all relocs (not just those with addends) will need WriteValue implemented.

[SingleAccretion]

LGTM module a few comments and the proper LEB management in Relocation.cs.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 2 comments.

[kg]

It pads the (U|S)LEB values to 5 bytes now using the algorithm from runtimelab. For completeness's sake I'm manually validating them for the whole 32-bit space right now since I can't manually test them inside of crossgen yet (we're not generating that type of reloc anywhere.)

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 1 comment.

[SingleAccretion]

LGTM with minor comments.

[AndyAyersMS]

LGTM

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 4 comments.