nx_isa_s390x_gate.nx

nx_isa_s390x_gate.nx -- THE s390x CONFORMANCE RULER. WHY IT EXISTS: nx_isa_conform_gate measures rv64, mips64, arm64, riscv32 and ppc64le and names s390x in its own output as UNMEASURED. An emulator nobody runs machine code through is a claim, not a capability. This organ decides "we emulate IBM z/Architecture" by RUNNING s390x machine code, in process, against values taken FROM THE z/ARCHITECTURE PRINCIPLES OF OPERATION -- never from nx_emu_s390x.nx itself, which would calibrate the ruler to its subject. SEPARATE ORGAN, NOT A PATCH TO THE INCUMBENT: five other ISA lanes are editing the estate at the same time, and concurrent edits to one file silently clobber each other here. The SHAPE is copied from nx_isa_conform_gate (word emitters, five-way classification, a value-printing report, negative controls, a fixture-reached tooth, a good-when-higher ratchet); nothing is imported from it, so neither can break the other. FIVE OUTCOMES, NEVER ONE: PASS ran and matched the manual-derived expectation WRONG-ANSWER ran to completion and computed the wrong value UNSUPPORTED hit an instruction class the emulator does not implement RAN-OFF-END the interpreter loop ended without an exit syscall FAULT bad pc, out-of-range access, or the step budget Collapsing these into FAIL is what makes a gap unactionable: an unimplemented instruction class and a miscomputed result need OPPOSITE fixes. WHAT MAKES s390x DIFFERENT FROM EVERY OTHER LANE ON THIS BOARD: it is the only VARIABLE-LENGTH ISA here. The instruction length comes from the top two bits of byte 0 (00 -> 2 bytes, 01/10 -> 4, 11 -> 6), so the emitter cannot be a single put_word helper and every branch displacement is a HALFWORD count, not a byte count. A fixture that assumed a fixed width would mis-place every label and the resulting garbage would read as the emulator being incomplete. So there is an explicit tooth asserting the emitter really returns 2, 4 and 6. BIG-ENDIAN. Every field below is written most-significant byte first. Getting this wrong is the silent-fixture defect: the decoder reads garbage, reports UNSUPPORTED, and the FIXTURE's bug wears the emulator's name. THE TWO NEGATIVE CONTROLS THAT ARE NOT DECORATION: nx_emu_s390x used to set its "handled" flag for a whole opcode family BEFORE looking at the extended opcode, so an unimplemented member of that family executed as a silent