Used to reset the input array of algorithms like device_merge_inplace.

Returns tuning parameters and their possible values as an OrderedDict. Each parameter maps to a list of valid values to explore during tuning.

* \brief Get the current architecture in device compilation. * * This function will always return `unknown` when called from the host, host could s

\brief Get trait `float_bit_mask`. \warning You cannot call this function when `is_floating_point()` returns `false`; doing so will result in a compil

Compute the required layout for this type and return it.

Internal which maps a continious named parameter in [0; 1] to it's discrete value.

Searches for specific fallback configuration in tuning results and adds it to the output :param output: list of strings to append to

searches for specific fallback configuration in benchmark results and adds it to the output :param output: list of strings to append

Returns the best performing benchmark from a list of benchmarks. For now, use the items per second as metric. in case the benchmark w

Returns the best performing tuning from a list of tunings. For now, use the items per second as metric. In case the tuning with the

Takes in a list of instantiated types in the form of (name, value)-pairs for some 'name' in the selection types. Returns a na

Takes in a list of instantiated types in the form of (name, value)-pairs for some 'name' in the selection types. Returns a h

Return the name of the cache file based on algo name, arch name and key value types

Return the path of the cache file

Used to accumulate the results of state.run() calls.

Adds base arguments from a config to instances, for get_score_assigner().

\brief Calculate kernel grid dimensions from the total number of blocks and the block size ROCm can launch a kernel with up to 2^32 threads in one di

\brief Count trailing zeroes Count the number of consecutive 0-bits, starting from the least significant bit. \p x must be non-zero.

Extracts the top-level template arguments from a base config class in a C++ struct specialization. Assumes the input contains a line ending

Find the best (lowest) time from results or cache entries.

\brief Returns flat (linear, 1D) thread identifier in a multidimensional tile (block).