Macro-op cache data entry pointers distributed as initial pointers held in tag array and next pointers held in data array for efficient and performant variable length macro-op cache entries
A microprocessor includes a macro-op (MOP) cache (MOC) comprising a set-associative MOC tag RAM (MTR) and a MOC data RAM (MDR) managed as a pool of MDR entries. A MOC entry (ME) comprises one MTR entry and one or more MDR entries that hold the MOPs of the ME. The MDR entries of the ME have a program order. Each MDR entry holds MOPs and a next MDR entry pointer. Each MTR entry holds initial MDR entry pointers and specifies the number of the MDR entries of the ME. During ME allocation, the MOC populates the MDR entry pointers to point to the MDR entries based on the program order. In response to an access that hits upon an MTR entry, the MOC fetches the MDR entries according to the program order initially using the initial pointers and subsequently using the next pointers.
1. A microprocessor, comprising:
a macro-op (MOP) cache (MOC) comprising:
a MOC tag RAM (MTR) arranged as a set-associative cache of MTR entries; and
a MOC data RAM (MDR) managed as a pool of MDR entries;
wherein a MOC entry (ME) comprises:
one MTR entry; and
one or more MDR entries that hold the MOPs of the ME, wherein the MDR entries of the ME have a program order;
wherein each MDR entry is configured to hold:
one or more MOPs of the ME; and
a next MDR entry pointer;
wherein each MTR entry is configured to hold:
a length that specifies the number of the MDR entries of the ME; and
one or more initial MDR entry pointers;
wherein the MOC is configured to:
during allocation of a ME into the MOC, populate the one or more initial MDR entry pointers and the next MDR entry pointers to point to the one or more MDR entries of the ME based on the program order; and
in response to an access of the MOC that hits upon the MTR entry of an ME, fetch the MDR entries of the ME according to the program order:
initially using the one or more initial MDR entry pointers of the hit upon MTR entry; and
subsequently using the next MDR entry pointers of fetched MDR entries of the ME, until all the MDR entries of the ME have been fetched from the MDR.
2. The microprocessor of claim 1 ,
wherein the one or more initial MDR entry pointers is a predetermined number; and
wherein the number of MDR entries of a ME is constrained by a size of the pool of MDR entries but is not constrained by the predetermined number of the one or more initial MDR entry pointers because of the next MDR entry pointers.
3. The microprocessor of claim 1 ,
wherein each MDR entry is configured to hold up to a predetermined number Q of MOPs,
wherein Q is greater than zero.
4. The microprocessor of claim 3 ,
wherein Q is three.
5. The microprocessor of claim 1 ,
wherein the MDR comprises a pipeline having a fetch latency of S clock cycles, wherein S is at least two;
wherein the one or more initial MDR entry pointers comprise at least S initial MDR entry pointers; and
wherein the S initial MDR entry pointers are used to fetch first through Sth MDR entries of the ME during first through Sth adjacent clock cycles such that pipeline bubbles are avoided during transition from fetching the MDR entries of the ME initially using the one or more initial MDR entry pointers provided by the hit upon MTR entry to fetching the remaining MDR entries of the ME using the next MDR entry pointers provided by the MDR.
6. The microprocessor of claim 1 ,
wherein the one or more initial MDR entry pointers comprise B initial MDR entry pointers for concurrently fetching B MDR entries from the MDR, wherein B is greater than one.
7. The microprocessor of claim 6 ,
wherein the MDR comprises B read ports for concurrently fetching the B MDR entries.
8. The microprocessor of claim 6 ,
wherein the number of MDR entries of a ME need not be allocated in quanta of B MDR entries but is instead allocatable in quanta of less than B MDR entries.
9. The microprocessor of claim 1 ,
wherein the MDR comprises a pipeline having a fetch latency of S clock cycles, wherein S is at least two;
wherein the one or more initial MDR entry pointers comprise S*B initial MDR entry pointers; and
wherein the S*B initial MDR entry pointers are used to fetch first through S*Bth MDR entries of the ME during first through Sth adjacent clock cycles such that pipeline bubbles are avoided during transition from fetching the MDR entries of the ME initially using the one or more initial MDR entry pointers provided by the hit upon MTR entry to fetching the remaining MDR entries of the ME using the next MDR entry pointers provided by the MDR.
10. The microprocessor of claim 1 ,
wherein each MDR entry is configured to hold up to a predetermined number Q of MOPs,
wherein Q is greater than one.
11. The microprocessor of claim 10 ,
wherein Q is three.
12. The microprocessor of claim 1 ,
wherein the MOC allocates the MDR entries of the pool such that the initial MDR entry pointers and the next MDR entry pointers may point to any MDR entry of the pool.
13. The microprocessor of claim 1 , further comprising:
a fusion engine configured to:
fuse MOPs decoded from architectural instructions of a plurality of fetch blocks into the MOPs of a ME; and
allocate the ME into the MOC;
wherein a fetch block comprises a sequential run of architectural instructions in a program instruction stream.
14. A method, comprising:
in a microprocessor comprising:
a macro-op (MOP) cache (MOC) comprising:
a MOC tag RAM (MTR) arranged as a set-associative cache of MTR entries; and
a MOC data RAM (MDR) managed as a pool of MDR entries;
wherein a MOC entry (ME) comprises:
one MTR entry; and
one or more MDR entries that hold the MOPs of the ME, wherein the MDR entries of the ME have a program order;
wherein each MDR entry is configured to hold:
one or more MOPs of the ME; and
a next MDR entry pointer;
wherein each MTR entry is configured to hold:
a length that specifies the number of the MDR entries of the ME; and
one or more initial MDR entry pointers;
during allocation of a ME into the MOC, populating by the MOC the one or more initial MDR entry pointers and the next MDR entry pointers to point to the one or more MDR entries of the ME based on the program order; and
in response to an access of the MOC that hits upon the MTR entry of an ME, fetching by the MOC the MDR entries of the ME according to the program order:
initially using the one or more initial MDR entry pointers of the hit upon MTR entry; and
subsequently using the next MDR entry pointers of fetched MDR entries of the ME, until all the MDR entries of the ME have been fetched from the MDR.
15. The method of claim 14 ,
wherein the one or more initial MDR entry pointers is a predetermined number; and
wherein the number of MDR entries of a ME is constrained by a size of the pool of MDR entries but is not constrained by the predetermined number of the one or more initial MDR entry pointers because of the next MDR entry pointers.
16. The method of claim 14 ,
wherein each MDR entry is configured to hold up to a predetermined number Q of MOPs,
wherein Q is greater than zero.
17. The microprocessor of claim 16 ,
wherein Q is three.
18. The method of claim 14 ,
wherein the MDR comprises a pipeline having a fetch latency of S clock cycles, wherein S is at least two; and
wherein the one or more initial MDR entry pointers comprise at least S initial MDR entry pointers;
the method further comprising:
using the S initial MDR entry pointers to fetch first through Sth MDR entries of the ME during first through Sth adjacent clock cycles such that pipeline bubbles are avoided during transition from fetching the MDR entries of the ME initially using the one or more initial MDR entry pointers provided by the hit upon MTR entry to fetching the remaining MDR entries of the ME using the next MDR entry pointers provided by the MDR.
19. The method of claim 14 ,
wherein the one or more initial MDR entry pointers comprise B initial MDR entry pointers for concurrently fetching B MDR entries from the MDR, wherein B is greater than one.
20. The method of claim 19 ,
wherein the MDR comprises B read ports for concurrently fetching the B MDR entries.
21. The method of claim 19 ,
wherein the number of MDR entries of a ME need not be allocated in quanta of B MDR entries but is instead allocatable in quanta of less than B MDR entries.
22. The method of claim 14 ,
wherein the MDR comprises a pipeline having a fetch latency of S clock cycles, wherein S is at least two; and
wherein the one or more initial MDR entry pointers comprise S*B initial MDR entry pointers; and
the method further comprising:
using the S*B initial MDR entry pointers to fetch first through S*Bth MDR entries of the ME during first through Sth adjacent clock cycles such that pipeline bubbles are avoided during transition from fetching the MDR entries of the ME initially using the one or more initial MDR entry pointers provided by the hit upon MTR entry to fetching the remaining MDR entries of the ME using the next MDR entry pointers provided by the MDR.
23. The method of claim 14 ,
wherein each MDR entry is configured to hold up to a predetermined number Q of MOPs,
wherein Q is greater than one.
24. The method of claim 10 ,
wherein Q is three.
25. The method of claim 14 ,
wherein the MOC allocates the MDR entries of the pool such that the initial MDR entry pointers and the next MDR entry pointers may point to any MDR entry of the pool.
26. The method of claim 14 , further comprising:
fusing, by a fusion engine, MOPs decoded from architectural instructions of a plurality of fetch blocks into the MOPs of a ME; and
allocating, by the fusion engine, the ME into the MOC;
wherein a fetch block comprises a sequential run of architectural instructions in a program instruction stream.
27. A non-transitory computer-readable medium having instructions stored thereon that are capable of causing or configuring a microprocessor comprising:
a macro-op (MOP) cache (MOC) comprising:
a MOC tag RAM (MTR) arranged as a set-associative cache of MTR entries; and
a MOC data RAM (MDR) managed as a pool of MDR entries;
wherein a MOC entry (ME) comprises:
one MTR entry; and
one or more MDR entries that hold the MOPs of the ME, wherein the MDR entries of the ME have a program order;
wherein each MDR entry is configured to hold:
one or more MOPs of the ME; and
a next MDR entry pointer;
wherein each MTR entry is configured to hold:
a length that specifies the number of the MDR entries of the ME; and
one or more initial MDR entry pointers;
wherein the MOC is configured to:
during allocation of a ME into the MOC, populate the one or more initial MDR entry pointers and the next MDR entry pointers to point to the one or more MDR entries of the ME based on the program order; and
in response to an access of the MOC that hits upon the MTR entry of an ME, fetch the MDR entries of the ME according to the program order:
initially using the one or more initial MDR entry pointers of the hit upon MTR entry; and
subsequently using the next MDR entry pointers of fetched MDR entries of the ME, until all the MDR entries of the ME have been fetched from the MDR.