weaselab/conflict-set
1
Fork 1
Code Issues 6 Pull Requests Actions 1 Packages Releases 13 Activity

Improve perf/codegen #94

Merged
andrew merged 5 commits from perf into main 2026-10-27 20:45:53 +00:00
Conversation 0 Commits 5 Files Changed 4
+51 -24
2 changed files with 51 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Showing only changes of commit dabddbcb56 - Show all commits

No files matched your search

ConflictSet.cpp
+30 -3
View File
@@ -116,13 +116,40 @@ struct InternalVersionT {
constexpr InternalVersionT() = default; constexpr InternalVersionT() = default;
constexpr explicit InternalVersionT(int64_t value) : value(value) {} constexpr explicit InternalVersionT(int64_t value) : value(value) {}
constexpr int64_t toInt64() const { return value; } // GCOVR_EXCL_LINE constexpr int64_t toInt64() const { return value; } // GCOVR_EXCL_LINE
constexpr auto operator<=>(const InternalVersionT &rhs) const { constexpr bool operator<(const InternalVersionT &rhs) const {
#if USE_64_BIT #if USE_64_BIT
return value <=> rhs.value; return value < rhs.value;
#else #else
// Maintains ordering after overflow, as long as the full-precision versions // Maintains ordering after overflow, as long as the full-precision versions
// are within `kMaxCorrectVersionWindow` of eachother. // are within `kMaxCorrectVersionWindow` of eachother.
return int32_t(value - rhs.value) <=> 0; return int32_t(value - rhs.value) < 0;
#endif
}
constexpr bool operator<=(const InternalVersionT &rhs) const {
#if USE_64_BIT
return value <= rhs.value;
#else
// Maintains ordering after overflow, as long as the full-precision versions
// are within `kMaxCorrectVersionWindow` of eachother.
return int32_t(value - rhs.value) <= 0;
#endif
}
constexpr bool operator>=(const InternalVersionT &rhs) const {
#if USE_64_BIT
return value >= rhs.value;
#else
// Maintains ordering after overflow, as long as the full-precision versions
// are within `kMaxCorrectVersionWindow` of eachother.
return int32_t(value - rhs.value) >= 0;
#endif
}
constexpr bool operator>(const InternalVersionT &rhs) const {
#if USE_64_BIT
return value > rhs.value;
#else
// Maintains ordering after overflow, as long as the full-precision versions
// are within `kMaxCorrectVersionWindow` of eachother.
return int32_t(value - rhs.value) > 0;
#endif #endif
} }
constexpr bool operator==(const InternalVersionT &) const = default; constexpr bool operator==(const InternalVersionT &) const = default;
README.md
+21 -21
View File
@@ -19,30 +19,30 @@ InstalledDir: /usr/lib/llvm-21/bin
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark | ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:---------- |--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 165.35 | 6,047,750.85 | 0.1% | 3,107.03 | 608.25 | 5.108 | 558.59 | 0.0% | 1.98 | `point reads` | 168.06 | 5,950,172.07 | 0.0% | 3,107.03 | 615.58 | 5.047 | 558.59 | 0.0% | 2.01 | `point reads`
| 161.75 | 6,182,478.30 | 0.2% | 3,036.76 | 594.75 | 5.106 | 539.35 | 0.0% | 1.94 | `prefix reads` | 164.93 | 6,063,229.09 | 0.1% | 3,036.76 | 604.27 | 5.025 | 539.35 | 0.0% | 1.97 | `prefix reads`
| 250.88 | 3,985,953.69 | 0.4% | 3,722.71 | 922.20 | 4.037 | 692.00 | 0.0% | 2.98 | `range reads` | 242.64 | 4,121,298.49 | 0.1% | 3,722.71 | 891.62 | 4.175 | 692.00 | 0.0% | 2.90 | `range reads`
| 358.34 | 2,790,617.48 | 0.2% | 4,523.64 | 1,317.11 | 3.435 | 720.22 | 2.1% | 4.28 | `point writes` | 361.34 | 2,767,480.63 | 0.0% | 4,523.64 | 1,328.61 | 3.405 | 720.22 | 2.1% | 4.31 | `point writes`
| 349.16 | 2,864,048.29 | 0.9% | 4,484.57 | 1,279.36 | 3.505 | 705.00 | 1.9% | 4.16 | `prefix writes` | 349.52 | 2,861,056.44 | 0.1% | 4,484.57 | 1,279.55 | 3.505 | 705.00 | 1.9% | 4.19 | `prefix writes`
| 192.06 | 5,206,715.53 | 0.0% | 2,224.10 | 706.55 | 3.148 | 377.17 | 3.3% | 2.30 | `range writes` | 194.68 | 5,136,536.22 | 0.1% | 2,224.10 | 712.83 | 3.120 | 377.17 | 3.3% | 2.33 | `range writes`
| 408.59 | 2,447,415.85 | 0.9% | 6,855.96 | 1,503.60 | 4.560 | 1,227.82 | 1.4% | 0.05 | `monotonic increasing point writes` | 416.48 | 2,401,076.73 | 1.0% | 6,855.96 | 1,522.78 | 4.502 | 1,227.82 | 1.4% | 0.05 | `monotonic increasing point writes`
| 136,536.00 | 7,324.08 | 2.0% | 812,045.00 | 501,448.67 | 1.619 | 151,246.43 | 0.9% | 0.01 | `worst case for radix tree` | 131,614.80 | 7,597.93 | 1.0% | 812,044.75 | 484,589.00 | 1.676 | 151,246.50 | 0.1% | 0.01 | `worst case for radix tree`
| 37.84 | 26,426,456.12 | 0.8% | 701.00 | 139.27 | 5.034 | 102.00 | 0.0% | 0.01 | `create and destroy` | 38.24 | 26,148,048.33 | 1.5% | 701.00 | 140.19 | 5.001 | 102.00 | 0.0% | 0.01 | `create and destroy`
## Radix tree (this implementation) ## Radix tree (this implementation)
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark | ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:---------- |--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 12.44 | 80,368,377.55 | 0.2% | 237.48 | 45.74 | 5.192 | 31.23 | 0.8% | 0.15 | `point reads` | 12.15 | 82,299,987.93 | 0.3% | 233.49 | 44.55 | 5.241 | 31.23 | 0.7% | 0.15 | `point reads`
| 14.59 | 68,537,989.62 | 0.6% | 291.66 | 53.69 | 5.432 | 40.16 | 0.5% | 0.17 | `prefix reads` | 14.31 | 69,893,290.91 | 0.2% | 283.68 | 52.40 | 5.414 | 40.16 | 0.4% | 0.17 | `prefix reads`
| 33.91 | 29,487,389.82 | 0.1% | 759.22 | 124.80 | 6.084 | 101.27 | 0.2% | 0.41 | `range reads` | 33.44 | 29,901,162.28 | 0.2% | 714.70 | 122.88 | 5.816 | 100.97 | 0.2% | 0.40 | `range reads`
| 19.54 | 51,172,096.42 | 0.0% | 369.51 | 71.92 | 5.138 | 47.92 | 0.5% | 0.23 | `point writes` | 19.73 | 50,684,375.53 | 0.1% | 369.51 | 72.61 | 5.089 | 47.92 | 0.5% | 0.24 | `point writes`
| 37.87 | 26,402,961.91 | 0.1% | 662.02 | 139.34 | 4.751 | 100.26 | 0.4% | 0.45 | `prefix writes` | 38.10 | 26,248,730.55 | 0.0% | 662.02 | 140.17 | 4.723 | 100.26 | 0.3% | 0.45 | `prefix writes`
| 39.11 | 25,569,264.79 | 0.1% | 725.90 | 143.87 | 5.046 | 111.09 | 0.1% | 0.48 | `range writes` | 39.76 | 25,152,283.94 | 0.0% | 725.90 | 146.25 | 4.964 | 111.09 | 0.1% | 0.48 | `range writes`
| 77.70 | 12,869,910.98 | 0.5% | 1,430.41 | 286.41 | 4.994 | 271.78 | 0.1% | 0.01 | `monotonic increasing point writes` | 80.73 | 12,387,533.79 | 4.3% | 1,424.45 | 296.99 | 4.796 | 271.47 | 0.1% | 0.01 | `monotonic increasing point writes`
| 319,309.00 | 3,131.76 | 0.5% | 4,181,645.50 | 1,178,098.50 | 3.549 | 728,511.00 | 0.1% | 0.01 | `worst case for radix tree` | 310,683.00 | 3,218.71 | 2.9% | 3,866,495.00 | 1,131,201.00 | 3.418 | 713,374.00 | 0.1% | 0.01 | `worst case for radix tree`
| 97.89 | 10,215,764.97 | 0.3% | 1,775.00 | 360.36 | 4.926 | 288.00 | 0.0% | 0.01 | `create and destroy` | 101.32 | 9,869,495.84 | 0.9% | 1,775.00 | 372.75 | 4.762 | 288.00 | 0.0% | 0.01 | `create and destroy`
# "Real data" test # "Real data" test
@@ -51,13 +51,13 @@ Point queries only. Gc ratio is the ratio of time spent doing garbage collection
## skip list ## skip list
``` ```
Check: 4.7186 seconds, 345.557 MB/s, Add: 3.3305 seconds, 168.339 MB/s, Gc ratio: 38.6509%, Peak idle memory: 5.51852e+06 Check: 4.65794 seconds, 350.058 MB/s, Add: 3.30748 seconds, 169.511 MB/s, Gc ratio: 38.2502%, Peak idle memory: 5.51852e+06
``` ```
## radix tree ## radix tree
``` ```
Check: 0.992421 seconds, 1643 MB/s, Add: 1.21864 seconds, 460.066 MB/s, Gc ratio: 38.8025%, Peak idle memory: 2.0226e+06 Check: 0.984726 seconds, 1655.84 MB/s, Add: 1.22965 seconds, 455.946 MB/s, Gc ratio: 38.8292%, Peak idle memory: 2.0226e+06
``` ```
## hash table ## hash table
@@ -65,6 +65,6 @@ Check: 0.992421 seconds, 1643 MB/s, Add: 1.21864 seconds, 460.066 MB/s, Gc ratio
(The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be) (The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be)
``` ```
Check: 0.855247 seconds, 1906.52 MB/s, Add: 0.63138 seconds, 887.98 MB/s, Gc ratio: 41.2278%, Peak idle memory: 0 Check: 0.864098 seconds, 1886.99 MB/s, Add: 0.636882 seconds, 880.31 MB/s, Gc ratio: 41.0631%, Peak idle memory: 0
``` ```