One-line prove-time report fragment shared by [`shard_esp`] and [`shard_esp_cap`]: total steps across all shards plus the wall-clock estimate at the requested `parallelism` (and the sequential baseline).
( profile: &BlockProfile, shard_of: &[u32], manifest: &ShardManifest, parallelism: usize, )
| 1763 | let fb: f64 = fr[k] |
| 1764 | .keys() |
| 1765 | .map(|&p| f64::from(profile.block(p).serialized_size)) |
| 1766 | .sum(); |
| 1767 | owned[k] + STATIC_FRONTIER_PER_BYTE * fb |
| 1768 | }) |
| 1769 | .collect() |
| 1770 | } |
| 1771 | |
| 1772 | /// Partition a STATIC block profile (reference graph + sizes, no kernel |
| 1773 | /// run — see `ix shard graph` / the FFI static-profile builder) into |
| 1774 | /// `num_shards` shards: byte-balanced min-cut over the walk-edge nets, |
| 1775 | /// then the [`rebalance_static`] post-pass toward equal predicted cost. |
| 1776 | /// Writes the same `.ixes` manifest as [`shard_esp`]. This is the |
| 1777 | /// `ix shard <env.ixe>` (no `--profile`) strategy; measured against the |
| 1778 | /// profiled strategy it cut mean shard FFT ~30% and the max shard |
| 1779 | /// 44–51% on the Init(8)/Std(24) harnesses. |
| 1780 | pub fn shard_static( |
| 1781 | profile: &BlockProfile, |
| 1782 | num_shards: usize, |
| 1783 | balance: f64, |
| 1784 | out_path: Option<&str>, |
nothing calls this directly
no test coverage detected