diff options
| author | laanwj <126646+laanwj@users.noreply.github.com> | 2022-05-04 20:19:51 +0200 |
|---|---|---|
| committer | laanwj <126646+laanwj@users.noreply.github.com> | 2022-05-04 20:36:09 +0200 |
| commit | fe6a299fc0020cd62156d4b7dd9c8dac358c69c5 (patch) | |
| tree | 8128addc4daae60885f601485a024cdf78895a53 /src/Makefile.bench.include | |
| parent | 33aaf434af2740d33abc0f45f613a666092630c3 (diff) | |
| parent | 5e61532e72c1021fda9c7b213bd9cf397cb3a802 (diff) | |
Merge bitcoin/bitcoin#24852: util: optimize HexStr
5e61532e72c1021fda9c7b213bd9cf397cb3a802 util: optimizes HexStr (Martin Leitner-Ankerl)
4e2b99f72a90b956f3050095abed4949aff9b516 bench: Adds a benchmark for HexStr (Martin Leitner-Ankerl)
67c8411c37b483caa2fe3f7f4f40b68ed2a9bcf7 test: Adds a test for HexStr that checks all 256 bytes (Martin Leitner-Ankerl)
Pull request description:
In my benchmark, this rewrite improves runtime 27% (g++) to 46% (clang++) for the benchmark `HexStrBench`:
g++ 11.2.0
| ns/byte | byte/s | err% | ins/byte | cyc/byte | IPC | bra/byte | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 0.94 | 1,061,381,310.36 | 0.7% | 12.00 | 3.01 | 3.990 | 1.00 | 0.0% | 0.01 | `HexStrBench` master
| 0.68 | 1,465,366,544.25 | 1.7% | 6.00 | 2.16 | 2.778 | 1.00 | 0.0% | 0.01 | `HexStrBench` branch
clang++ 13.0.1
| ns/byte | byte/s | err% | ins/byte | cyc/byte | IPC | bra/byte | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 0.80 | 1,244,713,415.92 | 0.9% | 10.00 | 2.56 | 3.913 | 0.50 | 0.0% | 0.01 | `HexStrBench` master
| 0.43 | 2,324,188,940.72 | 0.2% | 4.00 | 1.37 | 2.914 | 0.25 | 0.0% | 0.01 | `HexStrBench` branch
Note that the idea for this change comes from denis2342 in #23364. This is a rewrite so no unaligned accesses occur.
Also, the lookup table is now calculated at compile time, which hopefully makes the code a bit easier to review.
ACKs for top commit:
laanwj:
Code review ACK 5e61532e72c1021fda9c7b213bd9cf397cb3a802
aureleoules:
tACK 5e61532e72c1021fda9c7b213bd9cf397cb3a802.
theStack:
ACK 5e61532e72c1021fda9c7b213bd9cf397cb3a802 🚤
Tree-SHA512: 40b53d5908332473ef24918d3a80ad1292b60566c02585fa548eb4c3189754971be5a70325f4968fce6d714df898b52d9357aba14d4753a8c70e6ffd273a2319
Diffstat (limited to 'src/Makefile.bench.include')
| -rw-r--r-- | src/Makefile.bench.include | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/src/Makefile.bench.include b/src/Makefile.bench.include index e90c8530d8..532f668668 100644 --- a/src/Makefile.bench.include +++ b/src/Makefile.bench.include @@ -44,6 +44,7 @@ bench_bench_bitcoin_SOURCES = \ bench/rollingbloom.cpp \ bench/rpc_blockchain.cpp \ bench/rpc_mempool.cpp \ + bench/strencodings.cpp \ bench/util_time.cpp \ bench/verify_script.cpp |