It would be better if here
bench.runner(func(row, col) {}
row and col could be integer indices rather than Text. Because in most benchmarks the function will convert the Text back to index. The instructions required for that conversion are counted. This extra overhead distorts benchmarks that have low instruction count.