JIT: expand multi-dimensional array intrinsics for large and struct elements

[Copilot]

On platforms without a JIT (e.g. iOS with R2R), a multi-dimensional array access that impArrayAccessIntrinsic declines to expand becomes a real call to the Array intrinsic method, which falls back to the interpreter. Two of those bailouts are unnecessary.

Widen gtArrElemSize

GenTreeArrElem::gtArrElemSize was an unsigned char, so the importer bailed out whenever the element size exceeded 255 bytes. The field (and the constructor parameter) is now unsigned, and the FitsIn<unsigned char> check is removed. All consumers already read it into an unsigned.

Support struct-typed NI_Array_Set

a[i, j, k] = someStruct was rejected as an implementation limitation. It is now imported the same way CEE_STOBJ is:

case NI_Array_Set:
    if (varTypeIsStruct(elemType))
    {
        arrElem = gtNewStoreValueNode(elemLayout, arrElem, val);
        arrElem = impStoreStruct(arrElem, CHECK_SPILL_ALL);
    }
    else
    {
        arrElem = gtNewStoreIndNode(elemType, arrElem, val);
    }
    break;

Value-type arrays are invariant, so the covariance check that guards the varTypeIsGC path is not needed here.

Test

JIT/Regression/JitBlue/Runtime_131371 covers get/set/address on rank-2 and rank-3 arrays for a small struct, a 304-byte struct, and a 304-byte struct containing a GC reference, plus bounds-check behavior.

For a 304-byte element type, SetLarge now imports as STORE_BLK struct<LargeStruct, 304> over an ARR_ELEM[,,] instead of emitting a call. Validated on linux-x64 under JitStress 1/2/0x1F/0x3F and GCStress=0xC; existing JIT/Methodical groups (which include the MDArray tests) pass. Not exercised on arm64 or iOS.

• Fixes [iOS] CoreCLR R2R is dramatically slower than Mono AOT: 60 FPS drops to 7.4 FPS in real-time game rendering #131371

[Copilot]

Copilot was unable to review this pull request because there is no eligible user to bill. To allow Copilot reviews on bot-authored pull requests, enable direct organization billing in your organization's Copilot settings.

[azure-pipelines]

Azure Pipelines:
16 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Copilot was unable to review this pull request because there is no eligible user to bill. To allow Copilot reviews on bot-authored pull requests, enable direct organization billing in your organization's Copilot settings.

[dotnet-policy-service]

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

[Copilot]

Copilot was unable to review this pull request because there is no eligible user to bill. To allow Copilot reviews on bot-authored pull requests, enable direct organization billing in your organization's Copilot settings.

[Copilot]

Copilot was unable to review this pull request because there is no eligible user to bill. To allow Copilot reviews on bot-authored pull requests, enable direct organization billing in your organization's Copilot settings.

[AndyAyersMS]

If we're going to do this we might also want to consider #129455.

[jakobbotsch]

If we're going to do this we might also want to consider #129455.

The main motivation here is to fix the regression reported from #131371 because of the interpreter fallback on platforms without JIT, but I suppose it will improve performance in general on other targets too.

[azure-pipelines]

Azure Pipelines:
Successfully started running 5 pipeline(s).
11 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[jakobbotsch]

PTAL @dotnet/jit-contrib