103 self mlir gpu blob

// Test 390_103: self|mlir[gpu] — the kernel RUNS on the GPU (GPU rung C).
//
// The [gpu] parameter routes the SAME generated kernel through the GPU
// lowering: the transform emits a gpu.module (block-per-element indexing,
// StorageBuffer memref, entry-point ABI), Stage D lowers it via
// convert-gpu-to-spirv and serializes a .spv blob, and spirv-val gates the
// build — an invalid blob is a loud compile failure, never a shipped one.
//
// Then it EXECUTES: the host binary @imports the emitted koru_gpu_dispatch
// runtime, which uploads the buffer, dispatches the SPIR-V blob on the local
// Vulkan device (Apple M2 Pro via MoltenVK), and reads the result back — so
// the numbers below are computed ON THE GPU OF THE MACHINE THAT COMPILED THE
// KERNEL. The buffer is mutated in place, same contract as the CPU |mlir path.
//
// f32 on purpose: Apple GPUs have no Float64 — f64 shapes under [gpu] are a
// loud KORU123 wall, the first concrete instance of the capability lattice.
//
// RESIDUAL: the call-site `[gpu]` parameter is scaffolding — target selection
// belongs to the (separate) per-event selection-language endeavor; this knob
// dies the day that lands.

import std/kernel
import std/io

std/kernel:shape(Body) {
    mass: f32,
}

std/kernel:init(Body) {
    { mass: 1.0 },
    { mass: 2.0 },
    { mass: 3.0 },
}
| kernel k |> std/kernel:self|mlir[gpu] { k.mass *= 2.0 }
| computed c |> std/io:print.blk {
        mass[0]={{ c[0].mass:f }} mass[1]={{ c[1].mass:f }} mass[2]={{ c[2].mass:f }}
    }