[Wasm RyuJIT] Block stores

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.

[kg]

This is sufficient to compile these three managed methods to wasm functions with native memory.fill and memory.copy instructions:

structS{publicintA,B,C,D;}structS2{publicstringA,B,C,D;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidzeroStruct(S*result){*result=default;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticunsafevoidcopyStruct(S*a,S*b){*a=*b;}[MethodImpl(MethodImplOptions.NoInlining|MethodImplOptions.NoOptimization)]staticvoidzeroStructWithRefs(refS2result){result=default;}

[dotnet-policy-service]

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

[kg]

Not sure what i'm doing wrong here yet.

*************** Starting PHASE Lowering nodeinfo
compEnregLocals() is false, setting doNotEnreg flag for all locals.
Local V00 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V01 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V02 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
Local V03 should not be enregistered because: opts.compFlags & CLFLG_REGVAR is not set
BB02: already in WASM value stack order
lowering store lcl var/field (before):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
/--* t2 int N002 (???,???) [000003] DA---+----- * STORE_LCL_VAR struct<Program+S, 16> V03 loc1 Local V03 should not be enregistered because: written/read in a block op
lowering store lcl var/field (after):
N001 (???,???) [000002] -----+----- t2 = CNS_INT int 0
[000012] D---------- t12 = LCL_ADDR byref V03 loc1 [+0]
/--* t12 byref +--* t2 int N002 (???,???) [000003] sA---+----- * STORE_BLK struct<Program+S, 16> (init)
Local V03 should not be enregistered because: written/read in a block op
lowering return node
N001 (???,???) [000007] -----+----- * RETURN void ============
Program:copyStruct(ptr) - NYI (Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476 - NYI_WASM: IR not in a stackified form)
Z:\runtime\src\coreclr\jit\lowerwasm.cpp:476
Assertion failed 'NYI_WASM: IR not in a stackified form' in 'Program:copyStruct(ptr)' during 'Lowering nodeinfo' (IL size 17; hash 0xce1e49ee; MinOpts)

[Copilot]

Pull request overview

This PR implements infrastructure for WebAssembly block store operations in the RyuJIT compiler. It adds support for lowering block copy and initialization operations to use WASM-specific memory.copy and memory.fill instructions.

Changes:

• Adds LowerBlockStore implementation for WASM target to handle block copy and initialization operations
• Introduces two new WASM instructions (memory_copy and memory_fill) and their associated instruction format (IF_MEMCPY)
• Adds emitter support for encoding and displaying the new IF_MEMCPY instruction format

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/coreclr/jit/lowerwasm.cpp	Implements LowerBlockStore to handle GT_STORE_BLK and GT_INIT_BLK operations, with partial infrastructure for memory.copy and memory.fill
src/coreclr/jit/instrswasm.h	Adds memory_copy and memory_fill instruction definitions with their opcodes
src/coreclr/jit/emitfmtswasm.h	Defines the IF_MEMCPY instruction format for memory operations with two memory indices
src/coreclr/jit/emitwasm.cpp	Implements encoding, size calculation, and display logic for IF_MEMCPY instruction format

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[kg]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

[AndyAyersMS]

I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

That's probably my fault: I revised that code to try and handle the return buffer placement.

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

[MichalStrehovsky]

@MichalStrehovsky in the managed type system signatures is the this always explicitly represented?

If you mean the MethodSignature class, it represents what the input ECMA-335 has. So:

• !MethodSignature.IsStatic && !MethodSignature.IsExplicitThis: there's an implicit this parameter not listed in the signature
• !MethodSignature.IsStatic && MethodSignature.IsExplicitThis: this is part of the signature (can be hit in obscure cases involving C++/CLI when the compiler wants to attach modifiers to this, likely not a concern for WASM targets)
• MethodSignature.IsStatic: static method, no this for obvious reasons

[adamperlin]

cc @adamperlin see the WasmLowering.GetSignature changes. I couldn't figure out what was wrong with the existing implementation so I rewrote it, but the rewrite isn't complete. It was generating signatures with zeroes in them and that produced invalid wasm modules.

The re-write looks good to me, appending to a list seems like an easier solution to ensure correctness. My only feedback would just be that we should probably add a comment (maybe in the header) explaining what the maximum number of parameters we might add when lowering the signature is (explaining the + 4 in the allocated signature list size).

[kg]

@dotnet/jit-contrib PTAL

[AndyAyersMS]

Looks good overall, just curious about the GC disabling part.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[kg]

Resolving all comments to merge as-is. Issues will be fixed in follow-up PRs.