Current benchmark data was generated on Thu Feb 01 2024, full log can be found HERE
CONTRIBUTIONS are WELCOME!
* -m in a file name stands for multi-threading or multi-processing
* -i in a file name stands for direct intrinsics usage. (Usage of simd intrinsics via libraries is not counted)
* -ffi in a file name stands for non-stdlib FFI usage
* (You may find time < time(user) + time(sys) for some non-parallelized programs, the overhead is from GC or JIT compiler, which are allowed to take advantage of multi-cores as that's more close to real-world scenarios.)
lang | code | time | stddev | peak-mem mem | time(user) | time(sys) | compiler compiler/runtime |
---|---|---|---|---|---|---|---|
v | 1.v | 251ms | 3.2ms | 1.4MB | 240ms | 0ms | v/clang 0.4.4 |
v | 1.v | 261ms | 2.8ms | 2.3MB | 250ms | 0ms | v/clang+gc 0.4.4 |
lang | code | time | stddev | peak-mem mem | time(user) | time(sys) | compiler compiler/runtime |
---|---|---|---|---|---|---|---|
perl | 4.pl | timeout | 0.0ms | 13.8MB | 19783ms | 10ms | perl 5.38.2 |
v | 1.v | timeout | 0.0ms | 1.8MB | 4990ms | 0ms | v/clang 0.4.4 |
v | 1.v | timeout | 0.0ms | 3.0MB | 4987ms | 0ms | v/clang+gc 0.4.4 |