serial_executor.hpp File Reference

The executor that does not thread at all. More...

#include <cstdint>
#include <string_view>
#include "orrery/backend/cpu_topology.hpp"
#include "orrery/backend/executor.hpp"
#include "orrery/backend/worker_statistics.hpp"
#include "orrery/core/types.hpp"

Go to the source code of this file.

Classes
class	orrery::backend::SerialExecutor
	Runs the whole range on the calling thread. More...

Detailed Description

The executor that does not thread at all.

It exists for two reasons, and neither is a placeholder.

The first is that it is the baseline. A speedup is a ratio, and the number underneath it has to come from the same kernel compiled the same way and called through the same interface, or the comparison is measuring the difference between two code paths as well as the difference between one thread and eight. Running the identical task function on the calling thread is the only way to be sure of that.

The second is that threading is not always wanted. A unit test asserting an exact acceleration has nothing to gain from eight threads and something to lose in reproducibility of failure. A configuration of a few dozen particles costs less to compute than to hand to a pool. And Phase 9's GPU backend will want the CPU threads out of the way while it measures the device.

It reports statistics like any other executor, with one worker that is busy for the whole region. Those numbers are not interesting on their own; they are what makes the baseline appear in the same table as the schemes being compared against it, rather than as a bare number in the surrounding text.