JIT: Include auxiliary types in hardware intrinsic value numbering

[tannergooding]

Hardware intrinsic value numbering omitted the auxiliary type even when it changes the operation's semantics. For example, AVX2 gathers using the same index bits as int and long indices received the same liberal VN, allowing the JIT to incorrectly fold a comparison of their different results.

Pack the auxiliary type into the upper byte of the existing SIMD base-type constant and update decoding and diagnostic output. SimdType retains two operands, and intrinsics without an auxiliary type retain their original type constant, avoiding an additional VN argument and constant lookup. A compile-time assertion bounds the encoding.

Add focused regressions for gather index width on xarch and saturating-add operand signedness on Arm64.

Validation: checked native and cross-target JIT builds; 368 regression tests passed; the gather repro fails against the baseline and passes with the fix, including under JitStress2; JIT formatting passed. Arm64 crossgen2 output preserves distinct sqadd and suqadd instructions, but the Arm64 test was not executed on matching hardware.

Resolves #134486

Note

This PR description and changes were prepared with GitHub Copilot.

[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.

[dotnet-policy-service]

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

[Copilot]

Note

Copilot was unable to run its full agentic suite in this review.

Copilot review overview

Review effort: Lite
Findings: 2 · 1

Open (3)

• This masking introduces a new dependency on SimdTypeMask inside a FEATURE_HW_INTRINSICS region.… · New
• SimdTypeBits/SimdTypeMask are defined under FEATURE_SIMD, but they’re now used by… · New
• The signature change adds auxiliaryType, but the header comment still describes this as a generic… · New

What changed in this PR

This PR updates JIT value numbering for HW intrinsics to distinguish overloads that share the same operand bits (via an “auxiliary type” encoded into the SIMD type VN) and adds a regression test to validate the behavior for gather and AddSaturateScalar.

Changes:

• Encode an additional auxiliaryType into VNF_SimdType value numbers and update related decoding/dumping.
• Update HW intrinsic value numbering to incorporate GetAuxiliaryType() into the result type VN.
• Add a JIT regression test covering AVX2 gather index width and Arm64 AddSaturateScalar overload distinction.

File	Description
src/​tests/​JIT/​Regression_ro_2/​Runtime_134486.cs	Adds regression coverage to catch incorrect VN/CSE across intrinsic overloads.
src/​coreclr/​jit/​valuenum.h	Introduces SIMD type encoding constants and extends the VNForSimdType API to accept an auxiliary type.
src/​coreclr/​jit/​valuenum.cpp	Implements auxiliary-type encoding/decoding, updates simd-type dumping, and uses auxiliary type for HW intrinsic VNs.