uber / h3-go
100%
master: 100%

LAST BUILD BRANCH: bench/typed-sinks
DEFAULT BRANCH: master
Repo Added
Files 19
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

LAST BUILD ON BRANCH bench/typed-sinks
branch: SELECT
CHANGE BRANCH
x

coverage: 100.0%. Remained the same
36505723603

Pull #144

github

justinhwang 
bench: replace the interface sink with a typed noinline sink in paritytest

The cross-implementation benchmarks kept results alive by assigning them to a
package-level `any`. That boxes every Cell, float64, LatLng and int64 result on
its way into the interface, which adds a small heap allocation and several
nanoseconds to every row and reports 1 alloc/op for functions that do not
allocate. On the smallest operations the box was a third or more of the
measured time, and it masked allocation changes in the library.

Replace the sink with a generic, noinline function. The result is materialized
and passed in its own type, so the call cannot be eliminated and nothing is
boxed. No benchmark body changes beyond the sink call.

benchstat, Go implementation, old harness vs new, five interleaved full-suite
rounds on an Apple M3 Max (five samples is below benchstat's threshold for a
confidence interval, so only the p-values are shown). Selected rows; the PR
carries the full tables for both implementations:

                                       old       new   vs base
NewLatLng-16                       11.250n    1.959n   -82.59% (p=0.008 n=5)
CellToLatLng-16                     86.39n    70.77n   -18.08% (p=0.008 n=5)
CellToString-16                     32.96n    20.18n   -38.77% (p=0.008 n=5)
CellMarshalText-16                  52.26n    19.79n   -62.13% (p=0.008 n=5)
CellFromString-16                   33.62n    26.05n   -22.52% (p=0.008 n=5)
Resolution-16                       2.058n    1.964n   -4.57% (p=0.008 n=5)
NumCells-16                         8.674n    1.978n   -77.20% (p=0.008 n=5)
Parent-16                          13.650n    5.874n   -56.97% (p=0.008 n=5)
ImmediateParent-16                 11.470n    3.919n   -65.83% (p=0.008 n=5)
CenterChild-16                      9.530n    2.226n   -76.64% (p=0.008 n=5)
ChildPosToCell-16                  15.850n    8.004n   -49.50% (p=0.008 n=5)
CellAreaRads2-16                    500.2n    488.8n   -2.28% (... (continued)
Pull Request #144: bench: replace the interface sink with a typed noinline sink in paritytest

4309 of 4309 relevant lines covered (100.0%)

2903949.97 hits per line

Relevant lines Covered
4309 RELEVANT LINES 4309 COVERED LINES
2903949.97 HITS PER LINE
Source Files on master

Recent builds

Builds Branch Commit Type Ran Committer Via Coverage
36505723603 bench/typed-sinks bench: replace the interface sink with a typed noinline sink in paritytest The cross-implementation benchmarks kept results alive by assigning them to a package-level `any`. That boxes every Cell, float64, LatLng and int64 result on its way into ... Pull #144 justinhwang github
100.0
36499045753 master perf: reuse per-vertex trig in cell area and hoist boundary tables in x/h3go (#143) Profile CellArea and EdgeLength on the pure-Go path. The premise that they allocate no longer holds: Boundary is inlined into both and the compiler keeps the boun... push web-flow github
100.0
36497520575 perf/cell-area-boundary perf: reuse per-vertex trig in cell area and hoist boundary tables in x/h3go Profile CellArea and EdgeLength on the pure-Go path. The premise that they allocate no longer holds: Boundary is inlined into both and the compiler keeps the boundary sl... Pull #143 justinhwang github
100.0
36494950939 master perf: value-receiver coordIJK ops and deferred normalize in x/h3go (#142) Profile GridDistance and GridPath on the pure-Go path. The time was not in the algorithm but in how the coordIJK helpers compiled: normalize, upAp7, downAp7 and friends wer... push web-flow github
100.0
36494016491 perf/grid-distance perf: value-receiver coordIJK ops and deferred normalize in x/h3go Profile GridDistance and GridPath on the pure-Go path. The time was not in the algorithm but in how the coordIJK helpers compiled: normalize, upAp7, downAp7 and friends were point... Pull #142 justinhwang github
100.0
36480078553 master perf: generic flat hash table and buffer reuse for CellsToMultiPolygon in x/h3go (#141) Generalize the open-addressing cell set from PR 140 into flatTable[K Index, V], a linear-probing table keyed by any H3 index type with an inline value. cellSe... push web-flow github
100.0
36477088470 perf/multipolygon-flat-index perf: generic flat hash table and buffer reuse for CellsToMultiPolygon in x/h3go Generalize the open-addressing cell set from PR 140 into flatTable[K Index, V], a linear-probing table keyed by any H3 index type with an inline value. cellSet becom... Pull #141 justinhwang github
100.0
36476861668 perf/multipolygon-flat-index perf: generic flat hash table and buffer reuse for CellsToMultiPolygon in x/h3go Generalize the open-addressing cell set from PR 140 into flatTable[K Index, V], a linear-probing table keyed by any H3 index type with an inline value. cellSet becom... Pull #141 justinhwang github
100.0
36476859224 perf/multipolygon-flat-index perf: generic flat hash table and buffer reuse for CellsToMultiPolygon in x/h3go Generalize the open-addressing cell set from PR 140 into flatTable[K Index, V], a linear-probing table keyed by any H3 index type with an inline value. cellSet becom... push justinhwang github
100.0
36474828248 perf/multipolygon-flat-index perf: generic flat hash table and buffer reuse for CellsToMultiPolygon in x/h3go Generalize the open-addressing cell set from PR 140 into flatTable[K Index, V], a linear-probing table keyed by any H3 index type with an inline value. cellSet becom... Pull #141 justinhwang github
100.0
See All Builds (412)