code wiki / (root) / nx_compute_runner_test.nx

nx_compute_runner_test.nx

buildroot/runtime/nx_compute_runner_test.nx

7267 B182 linesdepth 8pulls 12 transitivereach 0 importersview sourcekind gate/prooftopic compute
docsdependenciesstructsconstsfunctions

about

nx_compute_runner_test.nx -- end-to-end loop closure. Build a real graph -> run it -> get real tensor output -> verify via numeric oracle vs hand-computed reference. All pure NishiLang. This is the "world class without comfyui annoyance but with the power and more" deliverable: typed DAG + content-addressed nodes + kernel dispatch + numerical verification, in one .nx program.

dependencies 8 imports · 0 importers

nx_syscalls.nx nx_tier.nx nx_tensor.nx nx_compute_node.nx nx_compute_graph.nx nx_blas_i64.nx nx_compute_runner.nx nx_numeric_oracle.nx nx_compute_runner_test.nx

imports: nx_syscalls.nxnx_tier.nxnx_tensor.nxnx_compute_node.nxnx_compute_graph.nxnx_blas_i64.nxnx_compute_runner.nxnx_numeric_oracle.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main sys_mmap nx_cg_alloc sys_mmap ↻ nx_cn_alloc sys_mmap ↻ nx_cg_add_node nx_cn_set_input nx_cg_validate nx_cg_get_node nx_cn_node_kind_is_valid nx_cn_op_is_valid nx_cn_kernel_is_valid nx_cg_toposort sys_mmap ↻ nx_cg_get_node ↻ nx_t_alloc nx_dt_is_valid nx_dt_element_bytes sys_mmap ↻ nx_t_compute_strides_rowma nx_t_fill_zero nx_t_set_i64 nx_t_flat_offset nx_cr_alloc sys_mmap ↻ nx_cr_bind_input nx_cr_bind_const nx_cr_run sys_mmap ↻ nx_cg_topo_at _runner_exec_node nx_cg_get_node ↻ nx_cr_store_set nx_cr_store_get _runner_run_relu _runner_alloc_like sys_mmap ↻ nx_t_alloc ↻ _runner_run_add

structs

none

consts

none

functions

18func main() -> nx_int