"Performance aware programming" course
My solutions and sandbox for course: https://www.computerenhance.com/
Solutions mostly written in zig or C/C++ (using zig build system).
zig version 0.11.0 unless specified otherwise.
Haversine distance
How to run
If you don't have data set, first generate it:
zig build -Doptimize=ReleaseFast run -- -f 10m.json -g 10000000Run processing & benchmarking:
zig build -Doptimize=ReleaseFast run -- -f 10m.json -i 5 -rNotes
- Generator (40x speedup) from buffering file io. C/C++ standard library does it by default (if not disabled) however zig gives you full control over it therefore requires few extra lines of code.
Results
All tests/benchmarking done using ~5 years old machine:
CPU: (Skylake) Intel i7-6700 CPU @ 3.40GHz
turbo boost up to 4Ghz
4 cores / 8 threads
Cache L1: 64K (per core)
Cache L2: 256K (per core)
Cache L3: 8MB (shared)
RAM: 32GB dual-channel (Crucial CT16G4DFRA266 2 x 16 GB DDR4 2666 MHz)
Generate
Generates 10M random coordinate pairs and prints them out in json format. 8.84s
Basic implementation
- read whole file in memory
- parse (using default std.json.parser) into a struct
- loop over and do the math Only optimization more or less is just to inline haversine function. Let the compiler optimize.
1.4 million haversines/second
zig version 0.11.0-dev.1593+d24ebf1d1 before this change
BEST SUB-RESULTS:
(test-iterations: 20, input size: 10000000, float: f64)
avg harvestine: 10008.47710
read time: 828.909ms
parse time: 5.415s
math time: 654.993ms
total time: 7.046s
throughput: 1419120 haversines/second
zig version 0.11.0
BEST SUB-RESULTS:
(test-iterations: 20, input size: 10000000, float: f64)
avg haversine: 10006.62310
read time: 1.281s
parse time: 4.794s
math time: 650.867ms
total time: 6.834s
throughput: 1463088 haversines/second