aboutsummaryrefslogtreecommitdiff
path: root/src/util/strencodings.cpp
diff options
context:
space:
mode:
authorlaanwj <126646+laanwj@users.noreply.github.com>2022-05-04 20:19:51 +0200
committerlaanwj <126646+laanwj@users.noreply.github.com>2022-05-04 20:36:09 +0200
commitfe6a299fc0020cd62156d4b7dd9c8dac358c69c5 (patch)
tree8128addc4daae60885f601485a024cdf78895a53 /src/util/strencodings.cpp
parent33aaf434af2740d33abc0f45f613a666092630c3 (diff)
parent5e61532e72c1021fda9c7b213bd9cf397cb3a802 (diff)
downloadbitcoin-fe6a299fc0020cd62156d4b7dd9c8dac358c69c5.tar.xz
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/util/strencodings.cpp')
-rw-r--r--src/util/strencodings.cpp33
1 files changed, 27 insertions, 6 deletions
diff --git a/src/util/strencodings.cpp b/src/util/strencodings.cpp
index 35f62f0422..bcedd4f517 100644
--- a/src/util/strencodings.cpp
+++ b/src/util/strencodings.cpp
@@ -9,6 +9,7 @@
#include <tinyformat.h>
#include <algorithm>
+#include <array>
#include <cstdlib>
#include <cstring>
#include <limits>
@@ -452,17 +453,37 @@ std::string Capitalize(std::string str)
return str;
}
+namespace {
+
+using ByteAsHex = std::array<char, 2>;
+
+constexpr std::array<ByteAsHex, 256> CreateByteToHexMap()
+{
+ constexpr char hexmap[16] = {'0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'a', 'b', 'c', 'd', 'e', 'f'};
+
+ std::array<ByteAsHex, 256> byte_to_hex{};
+ for (size_t i = 0; i < byte_to_hex.size(); ++i) {
+ byte_to_hex[i][0] = hexmap[i >> 4];
+ byte_to_hex[i][1] = hexmap[i & 15];
+ }
+ return byte_to_hex;
+}
+
+} // namespace
+
std::string HexStr(const Span<const uint8_t> s)
{
std::string rv(s.size() * 2, '\0');
- static constexpr char hexmap[16] = { '0', '1', '2', '3', '4', '5', '6', '7',
- '8', '9', 'a', 'b', 'c', 'd', 'e', 'f' };
- auto it = rv.begin();
+ static constexpr auto byte_to_hex = CreateByteToHexMap();
+ static_assert(sizeof(byte_to_hex) == 512);
+
+ char* it = rv.data();
for (uint8_t v : s) {
- *it++ = hexmap[v >> 4];
- *it++ = hexmap[v & 15];
+ std::memcpy(it, byte_to_hex[v].data(), 2);
+ it += 2;
}
- assert(it == rv.end());
+
+ assert(it == rv.data() + rv.size());
return rv;
}