[compiler-v2] Optimize stackless-bytecode assign instructions

[vineethk]

Description

When translating an Assign(dst, src) stackless bytecode instruction to file format bytecode, we copy or move src onto the stack, then eagerly store it to a local corresponding to dst.

Instead, now, when the AVOID_STORE_IN_ASSIGNS experiment is enabled, once we copy or move src onto the stack, we just rename that value to dst (in many cases, this avoids a store).

Another way to think about this optimization: when we have dst = src;, it is now treated similar to dst = noop_function(src);: we push src as if it were an argument, "consume" it (leading to popping it off the abstract stack), and then push dst onto the abstract stack as a result of this noop_function call.

• With this optimization on, the number of bytecode instructions produced by v2 compiler for the aptos-framework (and all of its dependencies) is about 1% lower than v1 compiler!
• Compared to without this optimization, v2 compiler with this optimization produces 2.2% fewer instructions when compiling aptos-framework and its dependencies.

The max gas taken for executing gov. proposal also goes down (by about 2.5%) with compiler v2 compared to compiler v1.

How Has This Been Tested?

• Existing tests, with updated baselines.
• All but one test (which is a corner case that has an issue open to fix it - [Feature Request][compiler-v2] Dead store elimination optimization algorithm should be enhanced to eliminate cycles #12400) produces fewer or equal instructions with this optimization.

Key Areas to Review

• Correctness of the optimization

Type of Change

• Performance improvement

Which Components or Systems Does This Change Impact?

• Move Compiler

[trunk-io]

⏱️ 4h 48m total CI duration on this PR

Slowest 15 Jobs	Cumulative Duration	Recent Runs
execution-performance / single-node-performance	1h 31m	🟩 🟩 🟩 🟩
test-target-determinator	17m	🟩 🟩 🟩 🟩
execution-performance / test-target-determinator	17m	🟩 🟩 🟩 🟩
check	15m	🟩 🟩 🟩 🟩
rust-cargo-deny	14m	⬜ 🟩 🟩 🟩 🟩 (+3 more)
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
check-dynamic-deps	9m	🟩 🟩 🟩 🟩 🟩 (+3 more)
rust-move-tests	8m	🟥
fetch-last-released-docker-image-tag	6m	🟩 🟩 🟩 🟩
rust-doc-tests	6m	🟩

_{settings ⋅ feedback ⋅ docs ⋅ learn more about trunk.io}

[vineethk]

• [compiler-v2] Optimize stackless-bytecode assign instructions #15445 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[vineethk]

This is the only test case where we produce more instructions.

[github-actions]

✅ Forge suite realistic_env_max_load success on 20449fa043da5443e728e3c4677ab9e719e79db1

two traffics test: inner traffic : committed: 14395.99 txn/s, latency: 2764.41 ms, (p50: 2700 ms, p70: 2700, p90: 2800 ms, p99: 3200 ms), latency samples: 5473620
two traffics test : committed: 100.02 txn/s, latency: 1956.10 ms, (p50: 1700 ms, p70: 2000, p90: 2100 ms, p99: 18000 ms), latency samples: 1680
Latency breakdown for phase 0: ["MempoolToBlockCreation: max: 2.226, avg: 1.332", "ConsensusProposalToOrdered: max: 0.317, avg: 0.287", "ConsensusOrderedToCommit: max: 0.300, avg: 0.292", "ConsensusProposalToCommit: max: 0.586, avg: 0.579"]
Max non-epoch-change gap was: 0 rounds at version 0 (avg 0.00) [limit 4], 1.01s no progress at version 2479353 (avg 0.20s) [limit 15].
Max epoch-change gap was: 0 rounds at version 0 (avg 0.00) [limit 4], 15.90s no progress at version 2479351 (avg 15.90s) [limit 16].
Test Ok

• Grafana dashboard
• Humio Logs
• Axiom Logs
• Validator CPU Profile
• Fullnode CPU Profile
• Test runner output
• Test run is land-blocking

[github-actions]

✅ Forge suite framework_upgrade success on 3527aa2e299553b759c515d9843586bad48c802c ==> 20449fa043da5443e728e3c4677ab9e719e79db1

Compatibility test results for 3527aa2e299553b759c515d9843586bad48c802c ==> 20449fa043da5443e728e3c4677ab9e719e79db1 (PR)
Upgrade the nodes to version: 20449fa043da5443e728e3c4677ab9e719e79db1
framework_upgrade::framework-upgrade::full-framework-upgrade : committed: 457.35 txn/s, submitted: 485.22 txn/s, failed submission: 0.88 txn/s, expired: 27.87 txn/s, latency: 4679.04 ms, (p50: 2100 ms, p70: 3800, p90: 15700 ms, p99: 19000 ms), latency samples: 41567
framework_upgrade::framework-upgrade::full-framework-upgrade : committed: 490.01 txn/s, submitted: 522.80 txn/s, failed submission: 1.16 txn/s, expired: 32.78 txn/s, latency: 6729.48 ms, (p50: 1500 ms, p70: 2100, p90: 26700 ms, p99: 27800 ms), latency samples: 33930
5. check swarm health
Compatibility test for 3527aa2e299553b759c515d9843586bad48c802c ==> 20449fa043da5443e728e3c4677ab9e719e79db1 passed
Upgrade the remaining nodes to version: 20449fa043da5443e728e3c4677ab9e719e79db1
framework_upgrade::framework-upgrade::full-framework-upgrade : committed: 1253.89 txn/s, submitted: 1256.32 txn/s, failed submission: 2.43 txn/s, expired: 2.43 txn/s, latency: 2365.96 ms, (p50: 2000 ms, p70: 2400, p90: 4200 ms, p99: 6100 ms), latency samples: 113400
Test Ok

• Grafana dashboard
• Humio Logs
• Axiom Logs
• Validator CPU Profile
• Fullnode CPU Profile
• Test runner output
• Test run is land-blocking

[github-actions]

✅ Forge suite compat success on 3527aa2e299553b759c515d9843586bad48c802c ==> 20449fa043da5443e728e3c4677ab9e719e79db1

Compatibility test results for 3527aa2e299553b759c515d9843586bad48c802c ==> 20449fa043da5443e728e3c4677ab9e719e79db1 (PR)
1. Check liveness of validators at old version: 3527aa2e299553b759c515d9843586bad48c802c
compatibility::simple-validator-upgrade::liveness-check : committed: 14247.80 txn/s, latency: 2422.05 ms, (p50: 2100 ms, p70: 2200, p90: 2600 ms, p99: 7600 ms), latency samples: 531140
2. Upgrading first Validator to new version: 20449fa043da5443e728e3c4677ab9e719e79db1
compatibility::simple-validator-upgrade::single-validator-upgrading : committed: 7230.55 txn/s, latency: 3809.58 ms, (p50: 3800 ms, p70: 4500, p90: 5200 ms, p99: 5700 ms), latency samples: 134620
compatibility::simple-validator-upgrade::single-validator-upgrade : committed: 7424.95 txn/s, latency: 4377.34 ms, (p50: 4500 ms, p70: 4700, p90: 6200 ms, p99: 6400 ms), latency samples: 248280
3. Upgrading rest of first batch to new version: 20449fa043da5443e728e3c4677ab9e719e79db1
compatibility::simple-validator-upgrade::half-validator-upgrading : committed: 6932.96 txn/s, latency: 4096.36 ms, (p50: 4400 ms, p70: 4600, p90: 5300 ms, p99: 5700 ms), latency samples: 130340
compatibility::simple-validator-upgrade::half-validator-upgrade : committed: 7172.87 txn/s, latency: 4534.61 ms, (p50: 4800 ms, p70: 5000, p90: 5900 ms, p99: 6600 ms), latency samples: 245460
4. upgrading second batch to new version: 20449fa043da5443e728e3c4677ab9e719e79db1
compatibility::simple-validator-upgrade::rest-validator-upgrading : committed: 1310.57 txn/s, latency: 17544.44 ms, (p50: 18600 ms, p70: 24200, p90: 27900 ms, p99: 28900 ms), latency samples: 56760
compatibility::simple-validator-upgrade::rest-validator-upgrade : committed: 8016.65 txn/s, latency: 4052.66 ms, (p50: 2600 ms, p70: 2700, p90: 11700 ms, p99: 12300 ms), latency samples: 259740
5. check swarm health
Compatibility test for 3527aa2e299553b759c515d9843586bad48c802c ==> 20449fa043da5443e728e3c4677ab9e719e79db1 passed
Test Ok

• Grafana dashboard
• Humio Logs
• Axiom Logs
• Validator CPU Profile
• Fullnode CPU Profile
• Test runner output
• Test run is land-blocking