[WLDFA]

Summary

• complete PTO-ISA/pto-spec#398 Atomic TSTORE APIs with supported PE masks and Local CUBE Predicate/PredInv execution masks
• keep Shared Atomic TSTORE restricted to PE participation masks, without Predicate execution masks
• complete PTO-ISA/pto-spec#399 cooperative CUBE-to-Shared ND metadata handling by merging writer valid shapes and publishing them only on LAST
• add positive codegen coverage, negative API coverage, contract tests, and usage documentation

LLVM dependency

Validated with LinxISA/llvm-project dev-llvm15_56 commit 2113a9922e5271592d7022f4888edba3d264d007, which contains the #398 Atomic TSTORE encoding and #399 CUBE capacity/conversion/transport support.

Verification

• python3 test/test_issue_398_399_contract.py: 4 passed
• test/test_predicate_adapters.py: 17 passed
• target objects TStoreAtomic.o and CubeToSharedND.o generated successfully
• focused negative cases reject RowMajor+Predicate, mismatched Predicate shape, Shared+Predicate, invalid PE mask, and invalid cooperative TMOV contracts
• disassembly confirms masked two-source B.IOT, PredInv, Shared PE mask, and assembled valid shape LB0=3, LB1=64, LB2=4
• git diff --check: passed

Existing gate failures

The full target object gate reports 86 passed and 23 unrelated existing failures, including retired fixtures and LLVM Shared-handle backend restrictions. Host make check reaches an existing ElementTile<unsigned long> expectation failure; the focused #398/#399 checks pass.