Fix for 127747

[dhartglassMSFT]

fgOptimizeHWIntrinsic does the following transform (note V09 used before it is defined in the 'After')

Before
\--*  HWINTRINSIC simd16 16 byte Add $441
   +--*  HWINTRINSIC simd16 16 byte Negate $480
   |  \--*  HWINTRINSIC simd16 16 byte ShiftLogicalSaturate $440
   |     +--*  COMMA     simd16 $400
   |     |  +--*  STORE_LCL_VAR simd16 V09 cse1         d:1 $VN.Void
   |     |  |  \--*  CNS_VEC   simd16<0x01010101, 0x01010101, 0x01010101, 0x01010101> $400
   |     |  \--*  LCL_VAR   simd16 V09 cse1         u:1 $400
   |     \--*  CNS_VEC   simd16<0x00000000, 0x00000000, 0x00000000, 0x00000000> $401
   \--*  LCL_VAR   simd16 V09 cse1         u:1 $400
After:
\--*  HWINTRINSIC simd16 16 byte Subtract $441
   +--*  LCL_VAR   simd16 V09 cse1         u:1 $400
   \--*  HWINTRINSIC simd16 16 byte ShiftLogicalSaturate $440
      +--*  COMMA     simd16 $400
      |  +--*  STORE_LCL_VAR simd16 V09 cse1         d:1 $VN.Void
      |  |  \--*  CNS_VEC   simd16<0x01010101, 0x01010101, 0x01010101, 0x01010101> $400
      |  \--*  LCL_VAR   simd16 V09 cse1         u:1 $400
      \--*  CNS_VEC   simd16<0x00000000, 0x00000000, 0x00000000, 0x00000000> $401

fgOptimizeAddition already checks "gtCanSwapOrder" for the transform (ADD(NEG(a), b) => SUB(b, a)) - that check was missing from fgOptimizeHWIntrinsic for an identical transform

fixes #127747
"Assertion failed Write to unaliased local overlaps outstanding read during 'Rationalize IR'"

[dotnet-policy-service]

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

[Copilot]

Pull request overview

This PR updates CoreCLR JIT morphing (fgOptimizeHWIntrinsic) to avoid performing a HWIntrinsic ADD optimization that would swap operand evaluation order when such swapping is not safe.

Changes:

• Add a gtCanSwapOrder(op1, op2) guard before rewriting (-v1) + v2 into v2 - v1 for SIMD HWIntrinsics.
• Bail out of that optimization when swapping could change observable behavior (e.g., due to side effects / ordering constraints).

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[dhartglassMSFT]

An earlier run showed green in all the spmi asmdiff tests

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.