hpc_bench

Benchmarks custom GPU kernels and operators with the hpc_bench framework using definition.json, workload.jsonl, and solution.json (correctness vs reference + CUDA event timing). Use when the user asks to profile or benchmark a kernel, wire an operator into hpc_bench, run hpc-bench/sol-execbench CLI, add a solution entry point, or debug PASSED/INCORRECT_NUMERICAL traces.