Efficient load value prediction

Patent metadata

Field	Value
Publication number	US-11243772-B2
Application number	US-201916414415-A
Country	US
Kind code	B2
Filing date	May 16, 2019
Priority date	May 16, 2019
Publication date	Feb 8, 2022
Grant date	Feb 8, 2022

Abstract

Certain aspects of the present disclosure provide techniques for training load value predictors, comprising: determining if a prediction has been made by one or more of a plurality of load value predictors; determining a misprediction has been made by one or more load value predictors of the plurality of load value predictors; training each of the one or more load value predictors that made the misprediction; and resetting a confidence value associated with each of the one or more load value predictors that made the misprediction.

First claim

What is claimed is:

1. A method for training load value predictors, comprising:
determining that a plurality of load value predictors have made a plurality of respective predictions that predict a value of respective load instructions;
determining that the plurality of respective predictions comprises:
one or more incorrect predictions; and
one or more correct predictions;

training each load value predictor associated with the one or more incorrect predictions;
resetting a confidence value associated with each load value predictor associated with the one or more incorrect predictions;
classifying a first load value predictor of the plurality of load value predictors as a donor based on determining a first amount of used predictions for the first load value predictor not meeting a threshold;
classifying a second load value predictor of the plurality of load value predictors as a receiver based on determining a second amount of used predictions for the second load value predictor meeting the threshold;
converting a storage table of the first load value predictor to a storage table of the second load value predictor based on classifying the first load value predictor as the donor and classifying the second load value predictor as the receiver; and
not training at least one load value predictor associated with the one or more correct predictions.

2. The method of claim 1 , wherein the plurality of load value predictors comprise:
a last value predictor (LVP);
a context value predictor (CVP);
a stride address predictor (SAP); and
a context address predictor (CAP).

3. The method of claim 2 , further comprising:
determining that two or more load value predictors of the plurality of load value predictors made the one or more correct predictions; and
training only one load value predictor of the two or more load value predictors that has a lowest cost value.

4. The method of claim 3 , wherein:
the LVP has a first cost value C LVP ,
the CVP has a second cost value C CVP ,
the SAP has a third cost value C SAP ,
the CAP has a fourth cost value C CAP , and
C LVP <C CVP <C SAP <C CAP .

5. The method of claim 4 , further comprising:
determining that the SAP is one of the two or more load value predictors that made the one or more correct predictions;
determining not to train the SAP; and
invalidating the SAP.

6. The method of claim 2 , further comprising:
determining that one or more load value predictors of the plurality of load value predictors has made no prediction; and
training the one or more load value predictors of the plurality of load value predictors that has made no prediction.

7. The method of claim 1 , wherein training each load value predictor associated with the one or more incorrect predictions comprises training only one load value predictor associated with the one or more incorrect predictions at a time.

8. A processing system for training load value predictors, comprising:
a memory comprising computer-executable instructions;
a processor configured to execute the computer-executable instructions and cause the processing system to:
determine that a plurality of load value predictors have made a plurality of respective predictions that predict a value of respective load instructions;
determine that the plurality of respective predictions comprises:
one or more incorrect predictions; and
one or more correct predictions;

train each load value predictor associated with the one or more incorrect predictions;
reset a confidence value associated with each load value predictor associated with the one or more incorrect predictions;
classify a first load value predictor of the plurality of load value predictors as a donor based on determining a first amount of used predictions for the first load value predictor not meeting a threshold;
classify a second load value predictor of the plurality of load value predictors as a receiver based on determining a second amount of used predictions for the second load value predictor meeting the threshold;
convert a storage table of the first load value predictor to a storage table of the second load value predictor based on classifying the first load value predictor as the donor and classifying the second load value predictor as the receiver; and
not train at least one load value predictor associated with the one or more correct predictions.

9. The processing system of claim 8 , wherein the plurality of load value predictors comprise:
a last value predictor (LVP);
a context value predictor (CVP);
a stride address predictor (SAP); and
a context address predictor (CAP), and
wherein:
the LVP has a first cost value C LVP ,
the CVP has a second cost value C CVP ,
the SAP has a third cost value C SAP ,
the CAP has a fourth cost value C CAP , and
C LVP <C CVP <C SAP <C CAP .

10. The processing system of claim 9 , wherein the processor is further configured to cause the processing system to:
determine that two or more load value predictors of the plurality of load value predictors made the one or more correct predictions:
train only one load value predictor of the two or more load value predictors that has a lowest cost value.

11. The processing system of claim 10 , wherein the processor is further configured to cause the processing system to:
determine that the SAP is one of the two or more load value predictors that made the one or more correct predictions;
determine not to train the SAP; and
invalidate the SAP.

12. The processing system of claim 8 , wherein the processor is further configured to cause the processing system to:
determine that one or more load value predictors of the plurality of load value predictors has made no prediction; and
train the one or more load value predictors of the plurality of load value predictors that has made no prediction.

13. The processing system of claim 8 , wherein the processor is further configured to cause the processing system to train only one load value predictor associated with the one or more incorrect predictions at a time.

14. A non-transitory computer-readable medium comprising instructions that, when executed by a processor of a processing system, cause the processing system to perform a method for training a plurality of load value predictors, the method comprising:
determining that the plurality of load value predictors have made a plurality of respective predictions that predict a value of respective load instruction;
determining that the plurality of respective predictions comprises:
one or more incorrect predictions, and
one or more correct predictions;

training each load value predictor associated with one of the one or more incorrect predictions;
resetting a confidence value associated with each load value predictor associated with the one or more incorrect predictions;
classifying a first load value predictor of the plurality of load value predictors as a donor based on determining a first amount of used predictions for the first load value predictor not meeting a threshold;
classifying a second load value predictor of the plurality of load value predictors as a receiver based on determining a second amount of used predictions for the second load value predictor meeting the threshold;
converting a storage table of the first load value predictor to a storage table of the second load value predictor based on classifying the first load value predictor as the donor and classifying the second load value predictor as the receiver; and
not training at least one load value predictor associated with one of the one or more correct predictions.

Assignees

• Qualcomm Inc

Inventors

• Al Sheikh Rami Mohammad A
• Hower Derek Robert

Classifications

• G06F9/3832 Primary

  Value prediction for operands; operand history buffers · CPC title

• G06F9/3806

  using address prediction, e.g. return stack, branch history buffer · CPC title

• G06F9/3842

  Speculative instruction execution · CPC title

• G06F9/3838

  Dependency mechanisms, e.g. register scoreboarding · CPC title

• G06F9/383

  Operand prefetching (cache prefetching G06F12/0862) · CPC title