[am11]

Produce the "x +/- compare" shape that lowering turns into adc/sbb/cinc
in more places:

• If-conversion turns SELECT(cond, x +/- 1, x) and SELECT(cond, k + 1,
  k) into "x +/- cond" and "k + cond" on all targets when cond is a
  compare that lowering can consume as a carry (== 0, != 0 and unsigned
  compares). Inside loops, if-conversion is now allowed when the result
  is such arithmetic instead of a SELECT, so "if (a[i] < p) count++"
  becomes "cmp; adc" instead of a branch.
• Operands without side effects or memory reads may be evaluated
  unconditionally even when the condition has ordering side effects.
• If-converted values stored to a temp whose only use adds or subtracts
  it are forwarded into that use, e.g. "num + (r ? 0 : -1)" becomes
  "cmp r, 1; sbb num, 0" on x64.
• Value numbering can now tell when a value is always 0 or 1: compares,
  0/1 constants, AND/OR/XOR and casts of such values, and phis of them.
  VN-based folding uses this to fold "x != 0" to "x" and "x == 0" to
  "x ^ 1". Values loaded from memory are not affected, so bools holding
  other non-zero values are still handled as ECMA requires.

Note: #135678's commit is included here.

[azure-pipelines]

Azure Pipelines:
Successfully started running 8 pipeline(s).
8 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.

[tannergooding]

There's also #134039 which might overlap and or benefit from the same (and I have my own variant of the optimization too, which I've not looked at in a bit but which I just rebased on dotnet/main and had AI do a fresh pass over: https://github.com/tannergooding/runtime/tree/tannergooding-jit-adc-sbb-carry-pattern).

This is a general pattern and setup that could be broadly impactful with the right overall handling/setup and might warrant some cross scenario measurements and comparisons

[am11]

Sounds good. We could get opus/fable type of model to go over these three/four overlapping PRs and gather valuable parts and supersede. I asked the model to lean towards fewer diffs. In its footnotes, it said something like "there are more phases/places in JIT where this pattern could be applied".