use std::hash instead of hashing byte by byte
This patch uses libcxx's string_view hashing. The performance is better than the
current implementation, and almost the same as for the hashing implementation amended
by the patch in https://android-review.googlesource.com/404742
The following experiments were conducted on a HiKey aarch64 A53.
No changes:
$ adb push out/target/product/hikey/data/benchmarktest64/ziparchive-benchmarks /data
$ adb shell /data/ziparchive-benchmarks --benchmark_repetitions=5
-----------------------------------------------------------------
Benchmark Time CPU Iterations
-----------------------------------------------------------------
FindEntry_no_match 82279629 ns 81759676 ns 9
FindEntry_no_match 81648056 ns 81627431 ns 9
FindEntry_no_match 81384074 ns 81369057 ns 9
FindEntry_no_match 89618889 ns 82437755 ns 9
FindEntry_no_match 81811389 ns 81785261 ns 9
FindEntry_no_match_mean 83348407 ns 81795836 ns 9
FindEntry_no_match_stddev 3520421 ns 394962 ns 0
Iterate_all_files 137622000 ns 137589032 ns 5
Iterate_all_files 139666333 ns 138409469 ns 5
Iterate_all_files 150070000 ns 140883697 ns 5
Iterate_all_files 138600667 ns 138540646 ns 5
Iterate_all_files 137599833 ns 137567438 ns 5
Iterate_all_files_mean 140711767 ns 138598056 ns 5
Iterate_all_files_stddev 5299929 ns 1354928 ns 0
$ cd system/core/libziparchive
$ git fetch https://android.googlesource.com/platform/system/core refs/changes/42/404742/7 && git cherry-pick FETCH_HEAD
$ mma
$ cd -
$ adb push out/target/product/hikey/data/benchmarktest64/ziparchive-benchmarks /data
$ adb shell /data/ziparchive-benchmarks --benchmark_repetitions=5
-----------------------------------------------------------------
Benchmark Time CPU Iterations
-----------------------------------------------------------------
FindEntry_no_match 53302756 ns 53291178 ns 13
FindEntry_no_match 54314487 ns 54272272 ns 13
FindEntry_no_match 53866923 ns 53851178 ns 13
FindEntry_no_match 53324423 ns 53317296 ns 13
FindEntry_no_match 53289231 ns 53289159 ns 13
FindEntry_no_match_mean 53619564 ns 53604216 ns 13
FindEntry_no_match_stddev 458449 ns 443527 ns 0
Iterate_all_files 128211111 ns 112254783 ns 6
Iterate_all_files 110695000 ns 110677726 ns 6
Iterate_all_files 109351250 ns 109350998 ns 6
Iterate_all_files 109367500 ns 109367796 ns 6
Iterate_all_files 110872222 ns 110591407 ns 6
Iterate_all_files_mean 113699417 ns 110448542 ns 6
Iterate_all_files_stddev 8143723 ns 1194577 ns 0
$ cd system/core/libziparchive
$ git checkout HEAD~
$ git am 0001-use-std-hash.patch
$ mma
$ cd -
$ adb push out/target/product/hikey/data/benchmarktest64/ziparchive-benchmarks /data
$ adb shell /data/ziparchive-benchmarks --benchmark_repetitions=5
-----------------------------------------------------------------
Benchmark Time CPU Iterations
-----------------------------------------------------------------
FindEntry_no_match 55339872 ns 55195112 ns 13
FindEntry_no_match 56232628 ns 56069924 ns 13
FindEntry_no_match 55694103 ns 55255946 ns 13
FindEntry_no_match 55275064 ns 55120136 ns 13
FindEntry_no_match 54971987 ns 54944411 ns 13
FindEntry_no_match_mean 55502731 ns 55317106 ns 13
FindEntry_no_match_stddev 482032 ns 436766 ns 0
Iterate_all_files 114618611 ns 114487804 ns 6
Iterate_all_files 112552917 ns 112229801 ns 6
Iterate_all_files 111288750 ns 111255044 ns 6
Iterate_all_files 111291528 ns 111259045 ns 6
Iterate_all_files 114347222 ns 113677119 ns 6
Iterate_all_files_mean 112819806 ns 112581763 ns 6
Iterate_all_files_stddev 1606214 ns 1454858 ns 0
Change-Id: I1e3413d331bcb460ca38bc2c87e23f89b456cd2f
diff --git a/libziparchive/zip_archive.cc b/libziparchive/zip_archive.cc
index 35d0f0b..6da5c99 100644
--- a/libziparchive/zip_archive.cc
+++ b/libziparchive/zip_archive.cc
@@ -100,6 +100,11 @@
}
static uint32_t ComputeHash(const ZipString& name) {
+#if !defined(_WIN32)
+ return std::hash<std::string_view>{}(
+ std::string_view(reinterpret_cast<const char*>(name.name), name.name_length));
+#else
+ // Remove this code path once the windows compiler knows how to compile the above statement.
uint32_t hash = 0;
uint16_t len = name.name_length;
const uint8_t* str = name.name;
@@ -109,6 +114,7 @@
}
return hash;
+#endif
}
/*