JIT: Normalize optimizations for ConditionalSelect and BlendVariable

[saucecontrol]

This fills in some optimizations present for ConditionalSelect but not BlendVariable and vice-versa on xarch.

---

Enables folding for BlendVariable with constant mask.

static Vector512<int> BlendMaskAllOnes(Vector512<int> v, int n) =>
    Avx512F.BlendVariable(
        Vector512.Create(n),
        v,
        Vector512.Create(-1));

-       vpbroadcastd zmm0, r8d
-       kmovq    k1, qword ptr [reloc @RWD00]
-       vpblendmd zmm0 {k1}, zmm0, zmmword ptr [rdx]
+       vmovups  zmm0, zmmword ptr [rdx]
        vmovups  zmmword ptr [rcx], zmm0
        mov      rax, rcx
 
        vzeroupper 
        ret      
-RWD00  	dq	000000000000FFFFh
-; Total bytes of code: 34
+; Total bytes of code: 19

---

Folds ConvertVectorToMask(CNS_VEC) to CNS_MASK

static Vector512<int> MultiInsert(Vector512<int> v, int n) =>
    Vector512.ConditionalSelect(
        Vector512.Create(0, -1, 0, 0, 0, -1, 0, 0, 0, -1, 0, 0, 0, -1, 0, 0),
        v,
        Vector512.Create(n));

-       vbroadcastf32x4 zmm0, xmmword ptr [reloc @RWD00]
-       vpbroadcastd zmm1, r8d
-       vpmovd2m k1, zmm0
-       vpblendmd zmm0 {k1}, zmm1, zmmword ptr [rdx]
+       vpbroadcastd zmm0, r8d
+       kmovq    k1, qword ptr [reloc @RWD00]
+       vpblendmd zmm0 {k1}, zmm0, zmmword ptr [rdx]
        vmovups  zmmword ptr [rcx], zmm0
        mov      rax, rcx

        vzeroupper 
        ret      
-RWD00  	dq	FFFFFFFF00000000h, 0000000000000000h
+RWD00  	dq	0000000000002222h
-; Total bytes of code: 41
+; Total bytes of code: 34
 

---

Enables EVEX embedded zero in more cases.

static Vector512<int> MultiInsertZero(Vector512<int> v) =>
    Avx512F.BlendVariable(
        v,
        Vector512<int>.Zero,
        Vector512.Create(0, -1, 0, 0, 0, -1, 0, 0, 0, -1, 0, 0, 0, -1, 0, 0));

-       vmovups  zmm0, zmmword ptr [rdx]
-       vxorps   ymm1, ymm1, ymm1
        kmovq    k1, qword ptr [reloc @RWD00]
-       vpblendmd zmm0 {k1}, zmm0, zmm1
+       vpblendmd zmm0 {k1}{z}, zmm0, zmmword ptr [rdx]
        vmovups  zmmword ptr [rcx], zmm0
        mov      rax, rcx
 
        vzeroupper 
        ret      
-RWD00  	dq	0000000000002222h
+RWD00  	dq	000000000000DDDDh
-; Total bytes of code: 38
+; Total bytes of code: 28
 

---

Optimizes blend with zero to AND/ANDN in more cases

static Vector256<int> MultiInsertZero(Vector256<int> v) =>
    Avx2.BlendVariable(
        v,
        Vector256<int>.Zero,
        Vector256.Create(0, -1, 0, 0, 0, -1, 0, 0));

-       vmovups  ymm0, ymmword ptr [rdx]
-       vxorps   ymm1, ymm1, ymm1
-       vbroadcastf128 ymm2, xmmword ptr [reloc @RWD00]
-       vpblendvb ymm0, ymm0, ymm1, ymm2
+       vbroadcastf128 ymm0, xmmword ptr [reloc @RWD00]
+       vpandn   ymm0, ymm0, ymmword ptr [rdx]
        vmovups  ymmword ptr [rcx], ymm0
        mov      rax, rcx

        vzeroupper 
        ret      
 RWD00  	dq	FFFFFFFF00000000h, 0000000000000000h
-; Total bytes of code: 34
+; Total bytes of code: 24
  

More improvements in the full diffs

[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 normalizes and consolidates optimizations for ConditionalSelect and BlendVariable intrinsics on x86/x64 architectures. The changes enable several new optimization patterns and refactor existing code to share common logic between similar operations.

Changes:

• Adds constant mask folding for BlendVariable intrinsics (all zeros → select first operand, all ones → select second operand)
• Enables folding of ConvertVectorToMask(CNS_VEC) to constant masks at lowering time
• Consolidates blend-with-zero optimizations into AND/AND_NOT operations across both ConditionalSelect and BlendVariable
• Refactors mask inversion logic into a reusable TryInvertMask helper function to reduce code duplication

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/coreclr/jit/valuenum.cpp	Adds value numbering optimizations for BlendVariable intrinsics with constant masks
src/coreclr/jit/rationalize.cpp	Removes obsolete small type normalization logic for BlendVariable
src/coreclr/jit/lowerxarch.cpp	Adds lowering optimizations for BlendVariable/ConvertVectorToMask, refactors mask inversion into TryInvertMask helper, and consolidates ConditionalSelect lowering
src/coreclr/jit/lower.h	Adds declaration for new TryInvertMask helper method
src/coreclr/jit/gentree.cpp	Fixes comment typo and adds constant folding for BlendVariable in morph phase

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

[saucecontrol]

@dotnet/jit-contrib this is ready for review

[JulieLeeMSFT]

@kg, PTAL. This is a community PR.

[kg]

LGTM. Any concerns @tannergooding ?

[tannergooding]

LGTM (thought I had already signed off on this one)

[tannergooding]

Updating CI since its been a month, should auto merge when it completes