[PATCH] D49503: [test-suite] Add Image Processing Kernels Using Benchmark Library: Dither Algorithms
Michael Kruse via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Sep 14 12:51:10 PDT 2018
Meinersbur added inline comments.
================
Comment at: test-suite/trunk/MicroBenchmarks/ImageProcessing/Dither/main.cpp:144-154
+#if MINIMUM_DIM > 128
+BENCHMARK(BENCHMARK_FLOYD_DITHER)
+ ->RangeMultiplier(2)
+ ->Range(128, MINIMUM_DIM)
+ ->Unit(benchmark::kMicrosecond);
+#else
+BENCHMARK(BENCHMARK_FLOYD_DITHER)
----------------
MatzeB wrote:
> I just noticed an odd effect here. A benchmark run now gives us 12 results for the same function running.
>
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/512/3
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/512/2
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/256/3
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/256/2
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/128/2
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/128/3
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/256/4
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/512/4
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/128/4
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/512/8
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/128/8
> test-suite :: MicroBenchmarks/ImageProcessing/Dither/Dither.test:BENCHMARK_ORDERED_DITHER/256/8
>
> I just had a change that was mostly improving things but happened to slightly regress this benchmark. Unfortunately this effect is now multiplied by 12 so this benchmark has a far bigger weight than others...
Do you suggest to reduce the number of variants?
Repository:
rL LLVM
https://reviews.llvm.org/D49503
More information about the llvm-commits
mailing list