# XLT examples — what to learn here XLT (`.xlt`) is SLeeLa's **CPU-model sheet** layer — the hardware counterpart of Nordshrift's system sheets. An `.xlt` file declaratively *assembles a machine* from registered catalog parts: a CPU SKU, memory, refresh cadence, storage, a bus, optional capabilities, memory tiers, and even a program to run on the result. These examples teach the model format from a two-line machine up to a tiered machine that runs a program. Build the validator once, then check each model: ```sh make -C impl/xlt impl/xlt/build/xltc lib/examples/xlt/BASIC-cpu-memory.xlt ``` Read the examples in order. ## The three cases | Level | File | The one new idea | |---|---|---| | Basic | [`BASIC-cpu-memory.xlt`](BASIC-cpu-memory.xlt) | The minimum model: a CPU SKU, a memory preset + capacity, and a target. | | Moderate | [`MODERATE-storage-bus-capabilities.xlt`](MODERATE-storage-bus-capabilities.xlt) | Add cadence, storage, a bus, and the optional `gpu;`/`dma;` capability directives. | | Advanced | [`ADVANCED-tiered-memory.xlt`](ADVANCED-tiered-memory.xlt) (+ [`ADVANCED-app.sleela`](ADVANCED-app.sleela)) | Two memory tiers (usual/accurate) and running a program under a resource grant. | The original [`CLASSIC-epyc-ddr2-model.xlt`](CLASSIC-epyc-ddr2-model.xlt) is the full-featured starter. ## What we want you to take away 1. **A machine is a parts list.** Each line fills exactly one slot (`cpu`, `memory`, `cadence`, `storage`, `bus`, …), and every ref must be a *registered catalog part* — `xltc` rejects an unknown SKU or preset. Reading a model tells you the whole machine. (Valid refs include CPUs like `SLAMDEPYC9755`, memory presets like `ddr4_3200`/`ddr5_6400`, storage like `nvmeGen4`, buses like `pcieGen5x16`.) 2. **Capabilities are bare directives.** `gpu;`, `dma;`, and `nativeram;` request runtime features; each may appear at most once and shows up in the build list on success. They are optional — the basic model omits them entirely. 3. **Memory can be tiered.** The advanced model declares a cheap `usual` tier (DDR2) and a fast `accurate` tier (DDR5), and picks the default with `maintier`. At the language level, functions choose a tier with the `usualmem` / `accuratemem` modifiers; unmarked code inherits the main tier. 4. **A model can run a program.** With `program "" ;` and a `grant ;`, the model builds the machine *and* admits and runs the program through the built CPU's fetch-decode-execute loop under that resource budget — one declarative document that describes the hardware and calls your code. ## A note on the tier modifiers The `usualmem` / `accuratemem` **function** modifiers are an XLT-aware language feature recognized by the toolchain (both `make compiler` and the current bundled `working-binaries/sleela`, which now supports syntax 1.0 – 3.1, accept them). `ADVANCED-app.sleela` is nonetheless kept as plain SLeeLa 1.7 that validates on older toolchains too, with the memory tiers declared at the **model** level in the `.xlt` sheet. For a model-level tiered example from the library, see `lib/xlt/examples/epyc-tiered-memory.xlt`. ## Validate them ```sh impl/xlt/build/xltc lib/examples/xlt/MODERATE-storage-bus-capabilities.xlt impl/xlt/build/xltc lib/examples/xlt/ADVANCED-tiered-memory.xlt ``` Each prints `BUILD SUCCESS` and lists the assembled parts (the advanced model lists 10, including the program ready to run under its grant).