Skip to content

Phase 2i-a — Element-wise array arithmetic

Summary

Matching 1d array[N, int] / array[N, float] operands support +, -, *, / with a new array result; sum(a * b) lowers via a temp product array.

Agent continuation

  1. Read li-tests/math_linalg/elementwise_mul_float.li, sum_elementwise_product.li.
  2. Run cmake --build build && ./li-tests/run_all.sh (expect 165 pass).
  3. Then merge PR stack through #142; optional SIMD for ArrayBinOpF64 lanes.
  4. Blocked on broadcast / mismatched-length promotion (not in v1).

Changed

Path What
compiler/types/typecheck.cpp Element-wise array binop types
compiler/mir/ ArrayBinOpF64, ArrayBinOpI64; sum(expr)
compiler/codegen/emit.cpp Per-element loops
li-tests/math_linalg/elementwise_*.li, sum_elementwise_product.li

Not changed

  • ** on arrays; % / // on arrays.
  • 2d matrix element-wise.
  • @vectorized auto-SIMD.

Breaking

N/A.

Security

N/A.

Performance

Scalar element loops; LLVM may autovec at -O3.

Downstream

  • docs/guide/math-hpc-examples.md — sum(a * b) row updated.
  • Tier 1 simd_dot could use sum(a * b) style in a follow-up bench refresh.