Method and system for specualtively sending processor-issued store operations to a store queue with full signal asserted

• US 8,352,712 B2
• Filed: 05/06/2004
• Issued: 01/08/2013
• Est. Priority Date: 05/06/2004
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Bell, Robert H. Jr., Capasso, Thomas Michael, Guthrie, Guy Lynn, Shen, Hugh, Stuecheli, Jeffrey Adam
• Primary Examiner(s): FENNEMA, ROBERT E
• Application Number:
  US10/840,560
• Publication Number:
  US 20050251660A1
• Time in Patent Office: 3,169 Days
• Field of Search: 712/225
• US Class Current: 712/225
• CPC Class Codes: G06F 9/3824 Operand accessing
• CPC Class Codes: G06F 9/3842 Speculative instruction execution
• CPC Class Codes: G06F 9/3861 Recovery, e.g. branch miss-prediction, exception handling error detection or correction G06F11/00

Abstract

A method and processor chip design for enabling a processor core to continue sending store operations speculatively to the store queue after the core receives indication that the store queue is full. The processor core is configured with speculative store logic that enables the processor core to continue issuing store operations while the store queue full signal is asserted. A copy of the speculatively issued store operation is placed within a speculative store buffer. The core waits for a signal from the store queue indicating the store operation was accepted into the store queue. When the speculatively-issued store operation is accepted within the store queue, the copy is discarded from the buffer. However, when the store operation is rejected, the speculative store logic re-issues the store operation ahead of normal store operations.

Claims

1. In a data processing system with a processor chip that includes a processor core and a store queue with multiple entries, a method for speculatively issuing a store operation from the processor core to the store queue, said method comprising:

determining when the store queue is full, wherein said store queue has multiple entries with at least one entry capable of concurrently holding multiple store operations, and the store queue is “

full”

when all entries of the store queue currently holds at least one store operation issued from the processor core; and

while said store queue is full, speculatively issuing a next store operation to the store queue as a first speculative store operation, wherein the next store operation is issued as a normal, non-speculative store operation when said store queue is not full;

wherein the at least one entry of the store queue is designed to gather multiple store operations when the store operations target portions of the same cache line; and

wherein said first speculative store operation is scheduled for re-issue to the store queue if (a) the first speculative store operation is not gathered by one of the entries within the store queue, and (b) no entry within the store queue becomes empty before the first speculative store operations arrives at the store queue.

2. The method of claim 1, further comprising:

creating and holding a copy of said first speculative store operation in a buffer;

when said first speculative store operation is gathered into or allocated an entry of said store queue, deleting said copy of said speculative store operation from said buffer; and

when said first speculative store operation is not gathered into or allocated an entry of said store queue, scheduling said copy of said first speculative store operation to be reissued to said store queue, wherein the copy is removed from the buffer and re-sent by the processor core to the store queue when the store queue is no longer full.

3. The method of claim 2, wherein said scheduling includes, when the store queue is no longer full, scheduling said copy of said first speculative store operation to be issued to the store queue ahead of other later store operations being subsequently scheduled by a store operation generation pipeline of said processor core.

4. The method of claim 1, further comprising:

speculatively issuing a second speculative store operation to said store queue if said store queue is still full and said first speculative store operation is gathered into or allocated an entry of said store queue.

5. The method of claim 4 further comprising:

determining when a newly generated store operation targets the same cache line address as the first speculative store operation;

when said first speculative store operation is gathered into or allocated an entry of said store queue, issuing said newly generated store operation to said store queue, wherein said newly generated store operation is speculatively issued as a second speculative store operation if the store queue is still full and issued as a normal store operation if the store queue is no longer full; and

when said first speculative store operation is not gathered into or allocated an entry of said store queue, withholding the issuance of said newly generated store operation until after the copy of said first speculative store operation within said buffer is reissued to said store queue.

6. The method of claim 5, wherein said withholding step further withholds the issuance of said newly generated store operation until after the store queue is no longer full and the first speculative store operation has been reissued and gathered into or allocated an entry of the store queue.

7. The method of claim 4, wherein when said buffer is a multiple entry buffer capable of holding copies of more than one speculatively issued store operation, said method further comprises:

holding the copy of the first speculative store operation in a register of the buffer;

when the first speculative store operation is not accepted by the store queue, determining when the newly generated store operation targets a different cache line from said first speculatively store operation;

issuing said newly generated store operation as a normal store operation to said store queue when said store queue is no longer full; and

when said store queue is still full and the newly generated store operation targets a different cache line from said first speculatively store operation;

speculatively issuing said newly generated store operation as a second speculative store operation independent of said first speculative store operation; and

creating and storing a copy of said second speculative store operation in a second register of the speculative store buffer.

8. The method of claim 7, wherein said buffer is a first in first out buffer and said copy of said first speculative store is reissued by the processor core to the store queue prior to a reissuing of said second speculative store operation.

9. A processor chip for utilization within a data processing system, said processor chip comprising:

a store queue having multiple entries, each entry capable of gathering multiple processor issued store operations targeting the same address, and the store queue is “

full”

when all entries of the store queue currently holds at least one store operation issued from the processor core;

tracking logic for determining when all the entries within the store queue have been allocated to at least one processor-issued store operation, indicating that the store queue is full; and

a processor core having normal logic for issuing store operations as a normal store when the store queue is not full and speculative store logic for speculatively issuing store operations when the store queue is full;

wherein a speculatively issued store operation is scheduled for re-issue to the store queue if the speculatively issued store operation is not gathered by one of the entries within the store queue, and if no entry within the store queue becomes empty before the speculatively issued store operations arrives at the store queue.

10. The processor chip of claim 9, wherein the processor core further comprises:

a speculative store buffer that holds a copy of the speculatively issued store operation; and

wherein said speculative store logic includes logic for;

determining when the store queue is full;

while said store queue is full, speculatively issuing a next store operation to the store queue as a first speculative store operation, wherein the next store operation is issued as a normal, non-speculative store operation when said store queue is not full; and

enabling reissue of the copy of the speculatively issued store operation when the speculatively issued store operation is not accepted by the store queue, wherein the copy is removed from the buffer and re-sent by the processor core to the store queue when the store queue is no longer full.

11. The processor chip of claim 10, wherein said logic for enabling reissue further comprises:

selection logic that selects the copy of said speculatively issued store for reissuing ahead of a normal store operation.

12. The processor chip of claim 11, wherein said selection logic is a speculatively issued store operation selection multiplexer (MUX) having a data input of both said normal issued store operation and the copy of said speculatively issued store operation and a select input that selects the copy of the speculatively issued store whenever that copy is available, wherein the copy is available once the speculatively issued store is not accepted at the store queue.

13. The processor chip of claim 12, wherein said buffer is a first in first out buffer that includes multiple registers that may each hold a copy of a different one of multiple speculatively issued store operations, said speculative store logic further comprising:

logic for determining when a second store operation being issued while the store queue is full targets a different cache line address from a previously issued speculative store operation that was not accepted at the store queue, wherein said second store operation is speculatively issued and a copy of said speculatively issued second store operation is placed in a second register within the store queue; and

logic for reissuing the copies of the speculatively issued stores in the order in which they are placed within the buffer.

14. The processor chip of claim 9, further comprising:

a core interface unit (CIU), which houses the tracking logic and is coupled to said processor core and said store queue via a set of signal wires, including;

a store request wire and a full signal wire connecting said CIU to said processor core;

an entry pop wire;

a store_gathered wire; and

a spec_gathered wire connecting said CIU to logic of said store queue, wherein said spec_gathered wire further connects said store queue logic to said processor core, and wherein the spec gathered wire signals the processor core when a first speculatively-issued store operation gathers into or is assigned an entry in the store queue, wherein the processor core responds by deleting the copy of the first speculatively-issued store from the buffer, wherein said copy of the speculative store operation is not re-issued.

15. The processor chip of claim 9, further comprising a store queue (STQ) controller, wherein said tracking logic is associated with said STQ controller, and said tracking logic is coupled to said processor core via a full signal wire and a spec_gathered wire, said full signal wire and said spec_gathered wire respectively providing said core with a signal including when a store operation is being speculatively issued and when the speculatively issued store operation has been accepted into an entry of the store queue, wherein the processor core responds to a signal on the spec-gathered wire by deleting the copy of the first speculatively-issued store from the buffer, wherein said copy of the speculative store operation is not re-issued.

16. The processor chip of claim 9, wherein said store queue includes control logic and wherein said tracking logic is located within the processor core, said processor chip further comprising an entry pop wire, a store_gathered wire, and a spec_gathered wire interconnecting said store queue control logic and said tracking logic, wherein said tracking logic receives signals via said wires to determine when a store operation is being speculatively issued and when a speculatively issued store operation has been accepted into an entry of said store queue, wherein the processor core responds to a signal on the spec-gathered wire by ignoring the copy of the first speculatively-issued store from the buffer during a subsequent scheduling of a store operation, whereby the first speculatively-issued store operation is discarded and not re-issued.

17. The processor chip of claim 9, further comprising:

entry dispatching logic associated with the store queue for selectively dispatching an entry of said store queue to enable the entry to be re-allocated to a next set of store operations; and

signaling logic associated with said entry dispatch logic for asserting a pop signal to said tracking logic.

18. The processor chip of claim 9, further comprising:

store queue (STQ) controller logic for determining when a store operation has been gathered into or allocated to an entry of said store queue; and

signaling logic that asserts a store_gathered signal for normal store operations that are gathered and a spec_gathered signal for speculatively issued store operations that are gather or allocated to an entry of the store queue.

19. A data processing system comprising:

a processor chip that includes;

a store queue having multiple entries, each entry capable of gathering multiple processor issued store operations targeting a same cache line address, and the store queue is “

full”

when all entries of the store queue currently holds at least one store operation issued from the processor core;

tracking logic for determining when all the entries within the store queue have been allocated to at least one processor-issued store operation, indicating that the store queue is full; and

a processor core having normal logic for issuing store operations as a normal store when the store queue is not full and speculative store logic for speculatively issuing store operations when the store queue is full; and

a memory hierarchy coupled to said processor chip;

wherein a speculatively issued store operation is scheduled for re-issue to the store queue if the speculatively issued store operation is not gathered by one of the entries within the store queue, and if no entry within the store queue becomes empty before the speculatively issued store operations arrives at the store queue.

20. The data processing system of claim 19, wherein the processor core further comprises:

a speculative store buffer that holds a copy of the speculatively issued store operation; and

wherein said speculative store logic includes logic for;

determining when the store queue is full;

while said store queue is full, speculatively issuing a next store operation to the store queue as a first speculative store operation, wherein the next store operation is issued as a normal, non-speculative store operation when said store queue is not full; and

enabling reissue of the copy of the speculatively issued store operation when the speculatively issued store operation is not accepted by the store queue, wherein the copy is removed from the buffer and re-sent by the processor core to the store queue when the store queue is no longer full.

21. The data processing system of claim 20, wherein said logic for enabling reissue further comprises:

selection logic that selects the copy of said speculatively issued store for reissuing ahead of a normal store operation.

22. The data processing system of claim 21, wherein said selection logic is a speculatively issued store operation selection multiplexer (MUX) having a data input of both said normal issued store operation and the copy of said speculatively issued store operation and a select input that selects the copy of the speculatively issued store whenever that copy is available within the buffer, wherein the copy is available within the buffer once the speculatively issued store is not accepted at the store queue.

23. The data processing system of claim 22, wherein said buffer is a first in first out buffer that includes multiple registers that may each hold a copy of a different one of multiple speculatively issued store operations, said speculative store logic further comprising:

logic for determining when a second store operation being issued while the store queue is full targets a different cache line address from a previously issued speculative store operation that was not accepted at the store queue, wherein said second store operation is speculatively issued and a copy of said speculatively issued second store operation is placed in a second register within the store queue; and

logic for reissuing the copies of the speculatively issued stores in the order in which they are placed within the buffer.

24. The data processing system of claim 20, said processor chip further comprising:

a core interface unit (CIU), which houses the tracking logic and is coupled to said processor core and said store queue via a set of signal wires, including;

a store request wire and a full signal wire connecting said CIU to said processor core;

an entry pop wire, which informs the CIU when an entry of the store queue becomes available for assigning to a next store operation;

a store_gathered wire, which informs the CIU when a normal store operation is gathered into or assigned an entry of the store queue; and

a spec_gathered wire connecting said CIU to logic of said store queue, wherein said spec_gathered wire signals the CIU when a first speculatively-issued store operation gathers into or is assigned an entry in the store queue, wherein the CIU responds by triggering the processor core to initiate a deletion of the copy of the first speculatively-issued store from the buffer.

25. The data processing system of claim 20, said processor chip further comprising a store queue (STQ) controller, wherein said tracking logic is associated with said STQ controller, and said tracking logic is coupled to said processor core via a full signal wire and a spec_gathered wire, said full signal wire and said spec_gathered wire respectively providing said core with a signal indicating when a store operation is being speculatively issued and when the speculatively issued store operation has been accepted into an entry of the store queue, wherein said core responds to a signal on the spec_gathered wire by deleting said first speculative store operation from the buffer, wherein said copy of the speculative store operation is not re-issued.

26. The data processing system of claim 20, wherein said store queue includes control logic and wherein said tracking logic is located within the processor core, said processor chip further comprising an entry pop wire, a store_gathered wire, and a spec_gathered wire interconnecting said store queue control logic and said tracking logic, wherein said tracking logic receives signals via said wires to determine when a store operation is being speculatively issued and when a speculatively issued store operation has been accepted into an entry of said store queue, wherein the processor core responds to a signal on the spec-gathered wire by ignoring the copy of the first speculatively-issued store from the buffer during a subsequent scheduling of a store operation, whereby the first speculatively-issued store is discarded and not re-issued.

27. The data processing system of claim 20, said processor chip further comprising:

entry dispatching logic associated with the store queue for selectively dispatching an entry of said store queue to enable the entry to be re-allocated to a next set of store operations; and

signaling logic associated with said entry dispatch logic for asserting a pop signal to said tracking logic.

28. The data processing system of claim 20, said processor chip further comprising:

store queue (STQ) controller logic for determining when a store operation has been gathered into or allocated to an entry of said store queue; and

signaling logic that asserts a store_gathered signal for normal store operations that are gathered and a spec_gathered signal for speculatively issued store operations that are gather or allocated to an entry of the store queue, wherein said core responds to the spec_gathered signal by deleting said first speculative store operation from the buffer, wherein said copy of the speculative store operation is not re-issued.