Allocating resources to threads based on speculation metric
View Patent ↗Methods, reservation stations and processors for allocating resources to a plurality of threads based on the extent to which the instructions associated with each of the threads are speculative. The method comprises receiving a speculation metric for each thread at a reservation station. Each speculation metric represents the extent to which the instructions associated with a particular thread are speculative. The more speculative an instruction, the more likely the instruction has been incorrectly predicted by a branch predictor. The reservation station then allocates functional unit resources (e.g. pipelines) to the threads based on the speculation metrics and selects a number of instructions from one or more of the threads based on the allocation. The selected instructions are then issued to the functional unit resources.
1. A method of allocating resources between a plurality of threads in a processor, each thread being associated with a plurality of instructions, the method comprising:
receiving, at a reservation station, a speculation metric for each thread of the plurality of threads, each speculation metric representing an extent to which the instructions associated with the thread are speculative;
allocating, at the reservation station, functional unit resources to the plurality of threads based on the speculation metrics, comprising:
comparing the speculation metrics to identify a thread with less speculative instructions;
determining a difference between the speculation metric for the thread with less speculative instructions and a speculation metric for another thread; and
in response to determining the difference exceeds a threshold, allocating additional functional unit resources to the thread with less speculative instructions;
selecting, at the reservation station, instructions associated with the plurality of threads based on the allocation of functional unit resources; and
issuing, at the reservation station, the selected instructions to the functional unit resources.
2. The method according to claim 1 , further comprising:
receiving at a speculation metric calculator confidence values for the plurality of instructions, each confidence value representing the likelihood that a particular instruction has been correctly predicted;
generating at the speculation metric calculator the speculation metrics based on the confidence values; and
providing the speculation metrics from the speculation metric calculator to the reservation station.
3. The method according to claim 2 , wherein generating a speculation metric for a particular thread comprises summing the confidence values for instructions associated with that thread over a predetermined number of cycles.
4. The method according to claim 2 , wherein generating a speculation metric for a particular thread comprises adding a generated confidence value associated with a particular instruction associated with that thread to the speculation metric for that thread and subtracting the confidence value from the speculation metric for that thread once the associated instruction has been executed.
5. The method according to claim 1 , wherein the allocation of functional unit resources is performed once every clock cycle.
6. The method according to claim 1 , wherein the allocation of functional unit resources is performed periodically.
7. The method according to claim 1 , wherein the functional unit resources comprise a plurality of pipelines, and allocating functional unit resources to the plurality of threads comprises allocating none, one or more pipelines to each thread.
8. The method according to claim 7 , wherein selecting instructions associated with the plurality of threads based on the allocation comprises selecting a number of ready instructions from a particular thread, wherein the number of ready instructions is equal to the number of pipelines allocated to the particular thread.
9. The method according to claim 8 , wherein in response to determining the number of instructions selected based on the allocation is less than the number of pipelines, selecting ready instructions from any thread so that the number of selected instructions is equal to the number of pipelines.
10. An apparatus with a reservation station arranged to allocate resources between a plurality of threads, the reservation station comprising:
a plurality of instruction queues, each instruction queue configured to store instructions for one of the plurality of threads; and
an instruction scheduler in communication with the plurality of instruction queues, the instruction scheduler arranged to:
receive a speculation metric for each thread of the plurality of threads, each speculation metric representing an extent to which the instructions associated with the thread are speculative;
allocate functional unit resources to the plurality of threads based on the speculation metrics, comprising:
comparing the speculation metrics to identify a thread with less speculative instructions;
determining a difference between the speculation metric for the thread with less speculative instructions and a speculation metric for another thread; and
in response to determining the difference exceeds a predetermined threshold, allocating additional functional unit resources to the thread with less speculative instructions;
select instructions from the plurality of instruction queues based on the allocation of functional unit resources; and
issue the selected instructions to the functional unit resources for execution.
11. The apparatus according to claim 10 , wherein the allocation of functional unit resources is performed once every clock cycle.
12. The apparatus according to claim 10 , wherein the allocation of functional unit resources is performed periodically.
13. The apparatus according to claim 10 , wherein the functional unit resources comprise a plurality of pipelines, and allocating functional unit resources to the plurality of threads based on the speculation metrics comprises allocating none, one or more of the pipelines to each thread based on the speculation metrics.
14. The apparatus according to claim 13 , wherein selecting instructions from the plurality of instruction queues based on the allocation comprises selecting a number of ready instructions from a particular instruction queue, wherein the number of ready instructions is equal to the number of pipelines allocated to the thread associated with the particular instruction queue.
15. The apparatus according to claim 14 , wherein in response to determining the number of ready instructions selected based on the allocation is less than the number of pipelines, selecting ready instructions from any instruction queue so that the number of selected instructions is equal to the number of pipelines.
16. An apparatus with a speculation metric calculator arranged to calculate a speculation metric for each of a plurality of threads, the speculation metric calculator comprising:
an input module configured to receive confidence values for a plurality of instructions, each instruction associated with a thread of the plurality of threads, each confidence value representing the likelihood that a particular instruction has been correctly predicted; and
a logic unit in communication with the input module, the logic unit configured to generate the speculation metrics based on the confidence values, each speculation metric representing an extent to which instructions associated with the thread are speculative.
17. The apparatus according to claim 16 , wherein the logic unit is configured to generate a speculation metric for a particular thread by a process comprising summing the confidence values for instructions associated with that thread over a predetermined number of cycles.
18. The apparatus according to claim 16 , wherein the logic unit is configured to generate a speculation metric for a particular thread by a process comprising adding a generated confidence value associated with a particular instruction associated with that thread to the speculation metric for that thread and subtracting the confidence value from the speculation metric for that thread once the associated instruction has been executed.