JIT: Accelerate uint->floating casts on pre-AVX-512 x86

[saucecontrol]

This is another one peeled from #116805 (only one more to go after this 😄)

It takes care of uint->floating casts without a helper call on pre-AVX-512 x86. Since baseline x86 is the only platform without native instructions for all 32-bit conversions, adding this allows some more simplification in fgMorphExpandCast

Typical diff:

-       push     eax
-       vzeroupper 
-       push     0
-       push     ecx
-       call     [CORINFO_HELP_LNG2FLT]
-       fstp     dword ptr [esp]
-       vmovss   xmm0, dword ptr [esp]
+       vxorps   xmm0, xmm0, xmm0
+       vcvtsi2sd xmm0, xmm0, ecx
+       vaddsd   xmm1, xmm0, qword ptr [reloc @RWD00]
+       vblendvpd xmm0, xmm0, xmm1, xmm0
+       vcvtpd2ps xmm0, xmm0
        vmovd    eax, xmm0
-       pop      ecx
        ret      

+RWD00  	dq	41F0000000000000h
 
-; Total bytes of code 27
+; Total bytes of code 31

[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 extends the JIT’s cast handling to avoid helper calls for uint → floating-point conversions on pre-AVX-512 x86, while also centralizing the “cast requires helper” decision logic to simplify fgMorphExpandCast.

Changes:

• Add Compiler::fgCastRequiresHelper and use it from importer/morph to consolidate helper-call decisions for casts.
• Implement x86 lowering for uint → float/double casts without helper calls when AVX-512 unsigned-convert is unavailable.
• Adjust cast handling/refactoring in fgMorphExpandCast and AVX-512 long→floating decomposition codepaths.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/coreclr/jit/morph.cpp	Refactors cast expansion to use fgCastRequiresHelper and simplifies helper selection logic.
src/coreclr/jit/lowerxarch.cpp	Adds x86 lowering sequence for uint→floating casts without helper calls (pre-AVX-512 path).
src/coreclr/jit/importer.cpp	Replaces ad-hoc helper decision logic with fgCastRequiresHelper.
src/coreclr/jit/flowgraph.cpp	Introduces fgCastRequiresHelper implementation.
src/coreclr/jit/decomposelongs.cpp	Minor refactor/clarification for AVX-512 long→floating cast decomposition.
src/coreclr/jit/compiler.h	Declares fgCastRequiresHelper.

[saucecontrol]

@dotnet/jit-contrib this is ready for review.

Diffs show an odd throughput regression for Windows x64. Looks like it might just be lack of /LTCG?

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

[JulieLeeMSFT]

@anthonycanino, @kg @tannergooding PTAL.

[kg]

Looks okay to me, going to defer to @tannergooding for the final approval though

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

[saucecontrol]

Updated diffs. The mysterious Windows x64 throughput regression I saw last month has resolved itself.