Improve read speed #361
Merged
Improve read speed #361
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Utilize philhofer/fwd#32 to peek bytes Eliminate bounds checks, mostly on reading sizes, but also integers. ``` benchmark old MB/s new MB/s speedup BenchmarkReadMapHeaderBytes-32 1136.75 1136.89 1.00x BenchmarkReadArrayHeaderBytes-32 1133.52 1149.34 1.01x BenchmarkTestReadBytesHeader-32 1200.18 1171.21 0.98x BenchmarkReadNilByte-32 2907.50 2176.63 0.75x BenchmarkReadFloat64Bytes-32 4308.86 3869.64 0.90x BenchmarkReadFloat32Bytes-32 2354.55 2332.21 0.99x BenchmarkReadBoolBytes-32 863.66 1124.07 1.30x BenchmarkReadTimeBytes-32 3710.27 3668.07 0.99x BenchmarkReadMapHeader-32 228.74 391.76 1.71x BenchmarkReadArrayHeader-32 243.90 454.31 1.86x BenchmarkReadNil-32 127.42 191.55 1.50x BenchmarkReadFloat64-32 1046.08 1142.49 1.09x BenchmarkReadFloat32-32 595.67 669.65 1.12x BenchmarkReadInt64-32 422.29 798.35 1.89x BenchmarkReadUintWithInt64-32 186.59 371.28 1.99x BenchmarkReadUint64-32 282.87 377.76 1.34x BenchmarkReadIntWithUint64-32 465.30 684.49 1.47x BenchmarkRead16Bytes-32 1021.15 1021.38 1.00x BenchmarkRead256Bytes-32 5324.42 6166.06 1.16x BenchmarkRead2048Bytes-32 8405.41 8382.61 1.00x BenchmarkRead16StringAsBytes-32 1102.72 1321.71 1.20x BenchmarkRead256StringAsBytes-32 6959.78 6627.49 0.95x BenchmarkRead16String-32 320.75 378.46 1.18x BenchmarkRead256String-32 2066.40 2242.78 1.09x BenchmarkReadComplex64-32 900.13 1064.21 1.18x BenchmarkReadComplex128-32 1809.88 2028.11 1.12x BenchmarkReadTime-32 1362.46 1425.79 1.05x ``` Apologies for the noisy benchmark. `BenchmarkReadNilByte` and `BenchmarkReadFloat64Byte` does not touch new code, so it seems to be "micro-bench-itis", where random deltas show up. Bench takes ~30m to run for whatever reason. The compiler is not clever enough to track back to `l := len(b)`, so it inserts a bounds check. Also `len(b) < 3` and `big.Uint16(b[1:])` causes a bounds check. So we rewrite those to avoid it. This should cover the lowest hanging fruits.
philhofer
approved these changes
Sep 16, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Utilize philhofer/fwd#32 to peek bytes
Eliminate bounds checks, mostly on reading sizes, but also integers.
Apologies for the noisy benchmark.
BenchmarkReadNilByteandBenchmarkReadFloat64Bytedoes not touch new code, so it seems to be "micro-bench-itis", where random deltas show up. Bench takes ~30m to run for whatever reason.The compiler is not clever enough to track back to
l := len(b), so it inserts a bounds check. Alsolen(b) < 3andbig.Uint16(b[1:])causes a bounds check. So we rewrite those to avoid it.This should cover the lowest hanging fruits.