Branch prediction using loop iteration count
An integrated circuit comprising instruction processing circuitry for processing a plurality of program instructions and instruction prediction circuitry. The instruction prediction circuitry comprises circuitry for detecting successive occurrences of a same program loop sequence of program instructions. The instruction prediction circuitry also comprises circuitry for predicting a number of iterations of the same program loop sequence of program instructions, in response to detecting, by the circuitry for detecting, that a second occurrence of the same program loop sequence of program instructions comprises a same number of iterations as a first occurrence of the same program loop sequence of program instructions.
1. A device comprising:
an instruction pipeline;
a program counter coupled to the instruction pipeline and configured to store a value; and
a branch predictor coupled to the program counter and configured to:
determine that the value corresponds to a beginning of a hyperblock loop by performing a first comparison of the value with a first data set;
in response to the value corresponding to the beginning of the hyperblock loop, determine that a second data set characterizing the hyperblock loop is stored in a memory by performing a second comparison of the value with the second data set, wherein the second data set is different from the first data set; and
predict a number of iterations for the hyperblock loop based on the second data set.
2. The device of claim 1 , wherein the branch predictor is configured to:
determine that the second data set characterizing the hyperblock loop is valid; and
predict the number of iterations in response to determining that the second data set characterizing the hyperblock loop is valid.
3. The device of claim 2 , wherein the branch predictor is configured to determine that the second data set is valid based on a bit stored in the memory.
4. The device of claim 1 , wherein to determine that the second data set is stored in the memory, the branch predictor is configured to match the value to an entry stored the memory.
5. The device of claim 1 , wherein the branch predictor is configured to predict the number of iterations based on a count stored in the memory.
6. The device of claim 1 , wherein the instruction pipeline is configured to process instructions of the hyperblock loop for the number of iterations predicted by the branch predictor.
7. The device of claim 1 , wherein the branch predictor is configured to:
determine that no data set characterizing the hyperblock loop is stored in the memory; and
store a new data set characterizing the hyperblock loop to the memory.
8. The device of claim 7 , wherein the branch predictor is configured to:
determine a count of iterations processed by the instruction pipeline; and
store the count of iterations as part of the new data set in the memory.
9. The device of claim 1 , wherein the branch predictor is configured to determine that the value corresponds to the beginning of the hyperblock loop by at least:
determining that the value corresponds to a beginning of a hyperblock; and
determining that the hyperblock is the hyperblock loop in response to determining that the value corresponds to the beginning of the hyperblock.
10. The device of claim 1 ,
wherein the memory is a buffer in the branch predictor, and
wherein the branch predictor is configured to determine that the second data set is stored in a branch target buffer.
11. A method comprising:
determining that a value stored in a program counter corresponds to a beginning of a hyperblock loop by comparing the value with a first data set;
determining that a second data set characterizing the hyperblock loop is stored in a memory by comparing the value with the second data set, wherein the second data set is different from the first data set; and
predicting a number of iterations for the hyperblock loop based on the second data set.
12. The method of claim 11 , further comprising determining that the second data set characterizing the hyperblock loop is valid,
wherein predicting the number of iterations in response to determining that the second data set characterizing the hyperblock loop is valid.
13. The method of claim 12 , wherein determining that the second data set is valid is based on a bit stored in the memory.
14. The method of claim 11 , wherein determining that the second data set is stored in the memory comprises matching the value to an entry stored the memory.
15. The method of claim 11 , wherein predicting the number of iterations is based on a count stored in the memory.
16. The method of claim 11 , further comprising processing instructions of the hyperblock loop for the predicted number of iterations.
17. The method of claim 11 , further comprising:
determining that a second value stored in the program counter corresponds to a beginning of a second hyperblock loop;
determining that no data set characterizing the second hyperblock loop is stored in the memory; and
storing a new data set characterizing the second hyperblock loop to the memory.
18. The method of claim 17 , further comprising:
determining a count of iterations processed by an instruction pipeline; and
storing the count of iterations as part of the new data set in the memory.
19. The method of claim 11 , wherein determining that the value corresponds to the beginning of the hyperblock loop comprises:
determining that the value corresponds to a beginning of a hyperblock; and
determining that the hyperblock is the hyperblock loop in response to determining that the value corresponds to the beginning of the hyperblock.
20. A device comprising:
a program counter configured to store a value;
a memory; and
a branch predictor coupled to the program counter, and configured to:
determine that the value corresponds to a hyperblock loop by comparing the value with a first data set;
determine that a second data set characterizing the hyperblock loop is stored in the memory by at least matching the value to an entry in the second data set stored in the memory, wherein the second data set is different from the first data set; and
predict a number of iterations for the hyperblock loop based on the second data set.