[Wasm RyuJit] add length prefix to each jit code contribution by AndyAyersMS · Pull Request #127773 · dotnet/runtime

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-05-04T21:19:46Z

And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

AndyAyersMS · 2026-05-04T21:21:52Z

@adamperlin PTAL
fyi @dotnet/jit-contrib

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

Copilot

Pull request overview

This PR updates the Wasm RyuJIT emission pipeline so each JIT-produced function/funclet contribution includes its own length prefix, shifting the responsibility away from host-side prefixing/parsing and enabling simpler host splitting of funclets.

Changes:

Emit a new Wasm pseudo-instruction (INS_code_size) at the start of each method/funclet body to write a fixed-size (5-byte) length prefix.
Move funclet start/end location computation earlier (into genEmitMachineCode) and stop doing it in unwindEmit.
Update WasmObjectWriter to stop auto-length-prefixing the Wasm code section, while explicitly prefixing manually injected stubs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/tools/Common/Compiler/ObjectWriter/WasmObjectWriter.cs	Stops section-writer auto prefixing for code section and explicitly writes length prefixes for inserted stubs.
src/coreclr/jit/unwindwasm.cpp	Removes `emitUpdateFuncletLocations()` call from `unwindEmit` now that locations are computed earlier.
src/coreclr/jit/instrswasm.h	Adds `code_size` instruction definition (format `IF_CODE_SIZE`).
src/coreclr/jit/emitwasm.h	Declares `emitOutputULEB128Padded` for emitting fixed-size prefixes.
src/coreclr/jit/emitwasm.cpp	Implements padded ULEB emission and adds encoding/printing support for `IF_CODE_SIZE`.
src/coreclr/jit/emitfmtswasm.h	Introduces `CODE_SIZE` instruction format.
src/coreclr/jit/codegenwasm.cpp	Emits `INS_code_size` at the start of root and funclet prologs.
src/coreclr/jit/codegencommon.cpp	Computes funclet locations earlier for Wasm so sizes/offsets are available during emission.

dotnet-policy-service · 2026-05-05T10:10:07Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

adamperlin

This looks good to me. We'll also need to add a length prefix to each code blob created by the WasmEmitter in crossgen, but we can do that as a follow up!

adamperlin · 2026-05-05T17:29:26Z

Note we currently emit a maximal-length value. We could shrink this down as the JIT supports instructions that get smaller once the lengths of other instructions are known, but it is a bit clumsy as it is the first value in the encoded stream and so forces everything else to shift. It would be easier perhaps for the host to do this.

I think we can potentially have the host do this. We already need to implement padded relocation shrinkage and some other size optimizations on the host side, so this could be part of that work.

AndyAyersMS · 2026-05-05T19:47:36Z

/ba-g infrastructure issues

#127773 adds length prefixes directly to emitted Wasm code in the JIT. This change is a follow up which adds length prefixes directly to generated wasm import thunks and stubs in crossgen. It also removes the logic in the Object Writer around handling length prefixes, since these will now be encoded in the `ObjectData` itself.

> [!NOTE] > This PR was created with assistance from Github Copilot This PR adds handling to the Wasm Object Writer to extract funclets attached to a compiled method as separate Wasm functions. To do this, an `INodeWithFunclets` interface was added for ReadyToRun and NativeAOT so that we can generically access EHInfo and FrameInfo fields attached to nodes which represent managed method definitions. We lay funclets out sequentially in the object file and they are added to the Wasm table contiguously. This is required by #128222 and lets us grab the funclet index using an offset from the index of the method it is attached to, i.e., funclet `i` for method M is at `table_index(M) + i`. Follow up on #127773 and other Wasm EH work. --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

[Wasm RyuJit] add length prefix to each jit code contribution
df3dc82
And no longer do this on the host side. This will let the host split out funclets without having to parse the JIT-generated code.

CopilotAI review requested due to automatic review settings May 4, 2026 21:19

AndyAyersMS requested a review from MichalStrehovsky as a code owner May 4, 2026 21:19

Copilotstarted reviewing on behalf of AndyAyersMS May 4, 2026 21:19 View session

github-actionsBot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label May 4, 2026

dotnet-policy-serviceBot assigned AndyAyersMS May 4, 2026

AndyAyersMS requested a review from adamperlin May 4, 2026 21:22

CopilotAI reviewed May 4, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/codegencommon.cpp

This was referenced May 5, 2026
slow macOS - "##[error]The job running on agent Azure Pipelines 9 ran longer than the maximum time of 60 minutes." dotnet/dnceng#1883
Open
The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008
Open

kg approved these changes May 5, 2026
View reviewed changes

am11 added the arch-wasm WebAssembly architecture label May 5, 2026

adamperlin approved these changes May 5, 2026
View reviewed changes

AndyAyersMS removed the request for review from MichalStrehovsky May 5, 2026 19:46

AndyAyersMS merged commit 0f84a10 into dotnet:mainMay 5, 2026
138 of 144 checks passed

steveisok mentioned this pull request May 6, 2026
Set per-target compile PDB for cDAC descriptor object libraries #127836
Merged

dotnet-maestroBot mentioned this pull request May 6, 2026
[main] Source code updates from dotnet/runtime dotnet/dotnet#6423
Merged

kg mentioned this pull request May 6, 2026
emitwasm.cpp:604:14: error: comparison of integer expressions of different signedness: ‘int’ and ‘unsigned int’ #127873
Closed

adamperlin mentioned this pull request May 8, 2026
crossgen2: Add Length Prefix to Crossgen-Generated Wasm #127936
Merged

AndyAyersMS mentioned this pull request May 14, 2026
[Wasm RyuJIT] Basic Control Flow #121178
Closed
12 tasks

adamperlin mentioned this pull request May 29, 2026
crossgen2: Extract funclets in WasmObjectWriter #128736
Merged

github-actionsBot locked and limited conversation to collaborators Jun 5, 2026