JIT: Reduce Vector256/512 Sum to a shared per-lane reduction on x64#127329

Merged

tannergooding merged 9 commits intomainfrom

copilot/fix-vector256-intrinsics-regression

Apr 25, 2026

Commits on Apr 23, 2026

Initial plan
Copilot
authored
Reduce Vector256/512 Sum to single V128 reduction on x64

Copilot
and
tannergooding
authored
Preserve FP sum ordering: reduce at full width, then combine halves

Copilot
and
tannergooding
authored
Simplify ternary and fall through V512->V256->V128 halves-combine

Copilot
and
tannergooding
authored
Align = on op1Shuffled/permIntrinsic declarations

Copilot
and
tannergooding
authored
V512 float Sum: preserve (s0+s1)+(s2+s3) halve-combine grouping

Copilot
and
tannergooding
authored
V512 FP Sum: extract 4 lanes directly via GetLower128/ExtractVector128

Copilot
and
tannergooding
authored

Commits on Apr 24, 2026