Bugfixes, benchmarks and improvements to FlatMap#1882

Merged

kennyweiss merged 28 commits into

developfrom

feature/kweiss/flatmap-improvements

Jun 18, 2026

Commits on Jun 18, 2026

Fix FlatMap copy assignment -- need to compare addresses, not values
kennyweiss
committed
Remove FlatMap's const operator[], which inserts for missing keys
kennyweiss
committed
DeviceHash: hash in 64 bits regardless of IndexType width
kennyweiss
committed
DeviceHash: hash floating-point keys by bit pattern, not value
kennyweiss
committed
FlatTable: wrap probe advance with a mask, not a signed division
kennyweiss
committed
Adds initial benchmark for flatmap vs map vs unordered_map vs sparsehash
kennyweiss
committed
Improves performance of FlatMap batched insertion for SEQ policy
kennyweiss
committed
Adds FlatMap benchmarks for hits and misses of precached entities
kennyweiss
committed
Exploring faster hash functions
kennyweiss
committed
Adds benchmark for flatmap load factor
kennyweiss
committed
Benchmark: decouple lookup order from insertion order in FlatMap suite
kennyweiss
committed
FlatMap: force-inline the lookup hot path
kennyweiss
committed
FlatMap: hash once and avoid FP division in getEmplacePos
kennyweiss
committed
Benchmark: report realized load factor in target-load-factor scenarios
kennyweiss
committed
FlatTable: honor visitor early-exit in scalar visitHashBucket
kennyweiss
committed
Fixes hip build via missing AXOM_HOST_DEVICE
kennyweiss
committed
FlatMap: Fuse the find and empty-slot probes in getEmplacePos()
kennyweiss
committed
FlatMap: Keep move semantics during batch insertion
kennyweiss
committed
Improves FlatMap benchmark
kennyweiss
committed
FlatMap: Device hash type must be 64 bits
kennyweiss
committed
Moves AXOM_FORCE_INLINE to core's Macros.hpp
kennyweiss
committed
Adds utility function for initializing initial probe group via bitshift and masking
kennyweiss
committed
FlatMap: Improves documentation and testing of find_with_hash
kennyweiss
committed
Adds benchmarks for device contruction and lookup
kennyweiss
committed
FlatMap: Generalizes the device benchmarks to other execution spaces, including omp
kennyweiss
committed
Add number of threads to omp benchmarks
kennyweiss
committed
Updates RELEASE-NOTES
kennyweiss
committed
Bugfix for rzvector -- `if constexpr` needs an `else`
kennyweiss
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Bugfixes, benchmarks and improvements to FlatMap#1882

Bugfixes, benchmarks and improvements to FlatMap#1882
kennyweiss merged 28 commits into
developfrom
feature/kweiss/flatmap-improvements

Commits on Jun 18, 2026

Fix FlatMap copy assignment -- need to compare addresses, not values

Remove FlatMap's const operator[], which inserts for missing keys

DeviceHash: hash in 64 bits regardless of IndexType width

DeviceHash: hash floating-point keys by bit pattern, not value

FlatTable: wrap probe advance with a mask, not a signed division

Adds initial benchmark for flatmap vs map vs unordered_map vs sparsehash

Improves performance of FlatMap batched insertion for SEQ policy

Adds FlatMap benchmarks for hits and misses of precached entities

Exploring faster hash functions

Adds benchmark for flatmap load factor

Benchmark: decouple lookup order from insertion order in FlatMap suite

FlatMap: force-inline the lookup hot path

FlatMap: hash once and avoid FP division in getEmplacePos

Benchmark: report realized load factor in target-load-factor scenarios

FlatTable: honor visitor early-exit in scalar visitHashBucket

Fixes hip build via missing AXOM_HOST_DEVICE

FlatMap: Fuse the find and empty-slot probes in getEmplacePos()

FlatMap: Keep move semantics during batch insertion

Improves FlatMap benchmark

FlatMap: Device hash type must be 64 bits

Moves AXOM_FORCE_INLINE to core's Macros.hpp

Adds utility function for initializing initial probe group via bitshift and masking

FlatMap: Improves documentation and testing of find_with_hash

Adds benchmarks for device contruction and lookup

FlatMap: Generalizes the device benchmarks to other execution spaces, including omp

Add number of threads to omp benchmarks

Updates RELEASE-NOTES

Bugfix for rzvector -- `if constexpr` needs an `else`