Machines are benchmarked by code, not algorithms
This addresses reproducibility issues for benchmark providers and users in machine evaluation, but it is incremental as it builds on existing concerns about code-level variations.
The article tackles the problem of benchmark reproducibility by showing that small changes in source code or compilation can affect machine evaluation results, using color-to-grayscale image conversion as an example.
This article highlights how small modifications to either the source code of a benchmark program or the compilation options may impact its behavior on a specific machine. It argues that for evaluating machines, benchmark providers and users be careful to ensure reproducibility of results based on the machine code actually running on the hardware and not just source code. The article uses color to grayscale conversion of digital images as a running example.