| 414 |
|
* unblocked threads to the point that we know they are available) |
| 415 |
|
* leading to more situations requiring more threads, and so |
| 416 |
|
* on. This aspect of control can be seen as an (analytically |
| 417 |
< |
* intractible) game with an opponent that may choose the worst |
| 417 |
> |
* intractable) game with an opponent that may choose the worst |
| 418 |
|
* (for us) active thread to stall at any time. We take several |
| 419 |
|
* precautions to bound losses (and thus bound gains), mainly in |
| 420 |
|
* methods tryCompensate and awaitJoin: (1) We only try |
| 1400 |
|
* range). If no queue exists at the index, one is created. If |
| 1401 |
|
* the queue is busy, another index is randomly chosen. The |
| 1402 |
|
* submitMask bounds the effective number of queues to the |
| 1403 |
< |
* (nearest poswer of two for) parallelism level. |
| 1403 |
> |
* (nearest power of two for) parallelism level. |
| 1404 |
|
* |
| 1405 |
|
* @param task the task. Caller must ensure non-null. |
| 1406 |
|
*/ |
| 1758 |
|
* adds a new thread if no idle workers are available and either |
| 1759 |
|
* pool would become completely starved or: (at least half |
| 1760 |
|
* starved, and fewer than 50% spares exist, and there is at least |
| 1761 |
< |
* one task apparently available). Even though the availablity |
| 1761 |
> |
* one task apparently available). Even though the availability |
| 1762 |
|
* check requires a full scan, it is worthwhile in reducing false |
| 1763 |
|
* alarms. |
| 1764 |
|
* |
