Standard containers and parsing
Deque/linked lists, ordered maps and sets, priority queues, map deletion, regular expressions, JSON, and generic serialization.
Results, context, and the workloads Prismio cannot run yet.
The maintained suite compares the same algorithm in Prismio, C++, and Rust. It reports medians instead of a single winning number—and marks missing language or standard-library capabilities as unsupported rather than replacing them with stand-ins.
Coverage is part of the result.
57 of 73 cataloged workloads are implemented. The remaining 16 make an absence visible instead of allowing a bespoke benchmark-only substitute.
Category
Implemented
Unsupported
Algorithms
13
1
Data structures
6
5
Compute
14
4
Memory
6
1
I/O and parsing
6
5
Adversarial
12
—
Four readings from the current suite.
These examples deliberately include wins, losses, and parity. Units are elapsed-time medians; they are only meaningful in the context of the full checked-in harness.
| Workload | Prismio | C++20 | Rust | Reading |
|---|---|---|---|---|
tokenization CPU + allocation | 147.8 μs | 265.1 μs | 1.02 ms | Prismio leads this string-scanning workload in the checked-in run. |
recursive_tree_rebuild Allocation + ownership | 393.7 μs | 874.1 μs | 494.7 μs | A reuse-oriented tree update is faster than both comparison arms here. |
binary_search CPU + memory | 89.81 ms | 73.06 ms | 132.71 ms | C++ is faster; Prismio is faster than Rust in this run. |
fibonacci CPU | 12.70 ms | 11.57 ms | 11.52 ms | All three implementations land in the same range. |
Unsupported means exactly that.
The suite does not award a comparison result when Prismio lacks the ordinary feature the workload is meant to evaluate.
Advanced concurrency
User-facing atomics, mutexes, work-stealing pools, and an async I/O runtime. The supported model is native OS threads with blocking typed channels.
System and optimization APIs
Sockets, memory-mapped I/O, explicit SIMD types, and custom collection allocators.
How to read the numbers.
Equivalent programs
The suite requires the three arms to express the same algorithm, not merely return the same checksum. Checksums are validated on each run.
Build commands
The checked-in run uses Prismio’s native build path, Clang with -O3 -std=c++20 -pthread, and Rust with -C opt-level=3.
Caveats matter
Container policies differ across languages; file workloads depend on page cache; and native-task tests include thread creation and joining. A small delta is not a universal language verdict.
Reproduce or challenge a result.
The harness, benchmark sources, unsupported-workload records, and raw checked-in results are part of the Prismio repository.