Closed Munksgaard closed 3 years ago
Tuning nw increases the performance on the tiny dataset:
[jxk588@gpu04:~/src/futhark]$ futhark bench --backend=opencl futhark-benchmarks/rodinia/nw/nw.fut Compiling futhark-benchmarks/rodinia/nw/nw.fut... Reporting average runtime of 10 runs for each dataset. futhark-benchmarks/rodinia/nw/nw.fut (using nw.fut.tuning): data/large.in: 3038μs (RSD: 0.008; min: -0%; max: +2%) data/tiny.in: 40μs (RSD: 0.141; min: -15%; max: +36%) data/small.in: 193μs (RSD: 0.020; min: -2%; max: +3%) data/medium.in: 1986μs (RSD: 0.002; min: -0%; max: +0%) [jxk588@gpu04:~/src/futhark]$ futhark bench --no-tuning --backend=opencl futhark-benchmarks/rodinia/nw/nw.fut Compiling futhark-benchmarks/rodinia/nw/nw.fut... Reporting average runtime of 10 runs for each dataset. futhark-benchmarks/rodinia/nw/nw.fut: data/large.in: 3038μs (RSD: 0.008; min: -0%; max: +2%) data/tiny.in: 499μs (RSD: 0.027; min: -4%; max: +5%) data/small.in: 191μs (RSD: 0.013; min: -2%; max: +2%) data/medium.in: 1986μs (RSD: 0.002; min: -0%; max: +0%)
Tuning nw increases the performance on the tiny dataset: