BenchmarkConfig

class brainevent.BenchmarkConfig(name, args, kernel_kwargs=<factory>, data_kwargs=<factory>)

A single benchmark configuration for a primitive.

Returned by def_benchmark_data functions as part of a list.

name

A short descriptive label for this configuration (e.g., "T,homo,bool").

Type:

str

args

Positional arguments (the actual input data) to pass to the primitive’s call function.

Type:

tuple

kernel_kwargs

Keyword arguments passed directly into the kernel call (e.g., block_size, shape, transpose). These may vary across benchmark runs and are forwarded to the call function.

Type:

dict

data_kwargs

Parameters that describe properties of the benchmark data but are not passed into the kernel (e.g., nnz, sparsity, density, sequence_length). These are recorded in the result table for reference only.

Type:

dict