mirror of
https://github.com/powdr-labs/powdr.git
synced 2026-04-20 03:03:25 -04:00
ff8f81e8ce
Fixes #1559 (alternative to #1560) With this PR, we always run the "default" sequence iterator, even if we have a cached sequence (which is still run before). This way, if the cached sequence was not sufficient to solve the entire block, the default solving sequence will have another attempt. The reason this doesn't lead to a dramatic performance degradation is because since #1528, we skip identities that have been completed. In the typical case, the cached sequence will have completed most identities. The reason this is better than #1560 is because the cached sequence typically makes progress with every identity, whereas the default iterator does not. ## Benchmark I ran the RISC-V Keccak example 3 times. It looks like the time spent in block machines increases by roughly 20%. Given that this fixes a bug and the overall time spent in block machines is small (even in the bitwise-heavy Keccak example), I think it's worth it! ### Main ``` == Witgen profile (1766802 events) 44.5% ( 4.4s): FixedLookup 36.3% ( 3.6s): Main Machine 9.7% ( 964.7ms): Secondary machine 0: main_binary (BlockMachine) 6.8% ( 669.3ms): witgen (outer code) 2.4% ( 236.0ms): Secondary machine 2: main_shift (BlockMachine) 0.3% ( 32.6ms): Secondary machine 1: main_memory (DoubleSortedWitnesses) 0.0% ( 183.7µs): Secondary machine 3: main_split_gl (BlockMachine) --------------------------- ==> Total: 9.907174375s == Witgen profile (1766802 events) 41.0% ( 3.8s): FixedLookup 39.0% ( 3.6s): Main Machine 10.2% ( 951.7ms): Secondary machine 0: main_binary (BlockMachine) 6.9% ( 644.2ms): witgen (outer code) 2.5% ( 231.5ms): Secondary machine 2: main_shift (BlockMachine) 0.3% ( 32.1ms): Secondary machine 1: main_memory (DoubleSortedWitnesses) 0.0% ( 183.4µs): Secondary machine 3: main_split_gl (BlockMachine) --------------------------- ==> Total: 9.295457333s == Witgen profile (1766802 events) 43.7% ( 4.2s): FixedLookup 37.0% ( 3.6s): Main Machine 10.0% ( 963.6ms): Secondary machine 0: main_binary (BlockMachine) 6.6% ( 636.3ms): witgen (outer code) 2.4% ( 234.7ms): Secondary machine 2: main_shift (BlockMachine) 0.3% ( 29.0ms): Secondary machine 1: main_memory (DoubleSortedWitnesses) 0.0% ( 190.8µs): Secondary machine 3: main_split_gl (BlockMachine) --------------------------- ==> Total: 9.677017958s ``` ### This branch ``` == Witgen profile (1986686 events) 43.3% ( 4.3s): FixedLookup 36.2% ( 3.6s): Main Machine 11.5% ( 1.1s): Secondary machine 0: main_binary (BlockMachine) 6.0% ( 600.2ms): witgen (outer code) 2.7% ( 273.3ms): Secondary machine 2: main_shift (BlockMachine) 0.3% ( 28.6ms): Secondary machine 1: main_memory (DoubleSortedWitnesses) 0.0% ( 203.4µs): Secondary machine 3: main_split_gl (BlockMachine) --------------------------- ==> Total: 9.975125084s == Witgen profile (1986686 events) 40.4% ( 3.9s): FixedLookup 37.2% ( 3.6s): Main Machine 12.1% ( 1.2s): Secondary machine 0: main_binary (BlockMachine) 7.1% ( 687.7ms): witgen (outer code) 2.9% ( 276.7ms): Secondary machine 2: main_shift (BlockMachine) 0.3% ( 30.3ms): Secondary machine 1: main_memory (DoubleSortedWitnesses) 0.0% ( 197.3µs): Secondary machine 3: main_split_gl (BlockMachine) --------------------------- ==> Total: 9.619824375s == Witgen profile (1986686 events) 42.4% ( 4.2s): FixedLookup 36.1% ( 3.6s): Main Machine 11.9% ( 1.2s): Secondary machine 0: main_binary (BlockMachine) 6.6% ( 654.9ms): witgen (outer code) 2.8% ( 276.6ms): Secondary machine 2: main_shift (BlockMachine) 0.3% ( 29.1ms): Secondary machine 1: main_memory (DoubleSortedWitnesses) 0.0% ( 202.3µs): Secondary machine 3: main_split_gl (BlockMachine) --------------------------- ==> Total: 9.957052209s ```