cmd/evm: benchmarking via statetest command + filter by name, index and fork
#30442
Conversation
jwasinger
requested review from
holiman,
MariusVanDerWijden and
lightclient
as code owners
There was a problem hiding this comment.
Checked to see if there was any noticeable slowdown.
Master:
$ time yes ~/go/src/github.com/holiman/goevmlab/examples/sstore_static/sstore_test.json | head -n 1000 | go run ./cmd/evm/ statetest 2>/dev/null 1>/dev/null
real 0m3.389s
user 0m3.688s
sys 0m1.008s
This PR
$ time yes ~/go/src/github.com/holiman/goevmlab/examples/sstore_static/sstore_test.json | head -n 1000 | go run ./cmd/evm/ statetest 2>/dev/null 1>/dev/null
real 0m3.409s
user 0m3.927s
sys 0m1.042s
So 👍
Just some minor nits, otherwise lgtm
holiman
changed the title
--bench flag support to include statetest subcommand. introduce test name, fork, index filtering flags for state test subcommand.statetest command + filter by name, index and fork
cmd/evm/staterunner.go
Outdated
| } else if len(matchingTests) != 1 { | ||
| return fmt.Errorf("can only benchmark single state test case (more than one matching params)") | ||
| } | ||
| var gasUsed uint64 | ||
| result := testing.Benchmark(func(b *testing.B) { | ||
| for i := 0; i < b.N; i++ { | ||
| test := matchingTests[0] | ||
| _, _, gasUsed, _ = test.test.RunNoVerify(test.st, cfg, false, rawdb.HashScheme) | ||
| } | ||
| }) | ||
| var stats execStats | ||
| // Get the average execution time from the benchmarking result. | ||
| // There are other useful stats here that could be reported. | ||
| stats.time = time.Duration(result.NsPerOp()) | ||
| stats.allocs = result.AllocsPerOp() | ||
| stats.bytesAllocated = result.AllocedBytesPerOp() | ||
| fmt.Fprintf(os.Stderr, `EVM gas used: %d | ||
| execution time: %v | ||
| allocations: %d | ||
| allocated bytes: %d | ||
| `, gasUsed, stats.time, stats.allocs, stats.bytesAllocated) |
There was a problem hiding this comment.
I would move this into its own function and do something like this
if bench {
runBenchmark()
}
return nil
which makes it a big cleaner imo
cmd/evm/staterunner.go
Outdated
| } else if len(matchingTests) != 1 { | ||
| return fmt.Errorf("can only benchmark single state test case (more than one matching params)") | ||
| } | ||
| var gasUsed uint64 | ||
| result := testing.Benchmark(func(b *testing.B) { | ||
| for i := 0; i < b.N; i++ { | ||
| test := matchingTests[0] | ||
| _, _, gasUsed, _ = test.test.RunNoVerify(test.st, cfg, false, rawdb.HashScheme) | ||
| } | ||
| }) | ||
| var stats execStats | ||
| // Get the average execution time from the benchmarking result. | ||
| // There are other useful stats here that could be reported. | ||
| stats.time = time.Duration(result.NsPerOp()) | ||
| stats.allocs = result.AllocsPerOp() | ||
| stats.bytesAllocated = result.AllocedBytesPerOp() | ||
| fmt.Fprintf(os.Stderr, `EVM gas used: %d | ||
| execution time: %v | ||
| allocations: %d | ||
| allocated bytes: %d | ||
| `, gasUsed, stats.time, stats.allocs, stats.bytesAllocated) |
cmd/evm/staterunner.go
Outdated
| if !bench { | ||
| return nil | ||
| } else if len(matchingTests) != 1 { | ||
| return fmt.Errorf("can only benchmark single state test case (more than one matching params)") |
There was a problem hiding this comment.
Why though?
Wouldn't it make more sense to put the benchmarking into the loop above, and after each test.test.Run, you do a bench run, if so desired.
You might even consider integrating the bench stats into the results, but now I'm just thinking aloud... ?
There was a problem hiding this comment.
Yes. Benchmark every test unless otherwise-specified sounds like fine default behavior to me.
There was a problem hiding this comment.
SO drop this whole if-clause?
|
Sorry -- I know I approved it earlier, and I came around to merge it now, but then I found some more things that could be improved :) |
|
Tried to re-run the exact test you posted from Running a separate test shows no measurable performance difference between the latest changes and master branch. |
cmd/evm/runner.go
Outdated
| time time.Duration // The execution time. | ||
| allocs int64 // The number of heap allocations during execution. | ||
| bytesAllocated int64 // The cumulative number of bytes allocated during execution. | ||
| type ExecStats struct { |
There was a problem hiding this comment.
Does this need to be exported? I don't see why
There was a problem hiding this comment.
Ah I think I got confused because non-exported members of structs are not serialized in JSON so I assumed that the type also had to be exported 🤷♂️
There was a problem hiding this comment.
Wait but how can StateTestResult have a public field with a private type...
cmd/evm/staterunner.go
Outdated
| if !bench { | ||
| return nil | ||
| } else if len(matchingTests) != 1 { | ||
| return fmt.Errorf("can only benchmark single state test case (more than one matching params)") |
There was a problem hiding this comment.
SO drop this whole if-clause?
When
evm statetest --benchis specified, benchmark the execution similarly toevm run. It will explicitly error if there is more than one matched subtest to be executed (newstatetestflags are added to filter which tests are ran).Breaks the state test runner API. But I figure that we are the only consumers of this API, so it's not a big problem?