Instant quiescing of an accelerator
A system architecture is provided and includes an on-chip coherency unit, a processing unit, an accelerator and dedicated wiring. The processing unit is communicative with the on-chip coherency unit via a first interface. The accelerator is communicative with the on-chip coherency unit via a second interface. The accelerator is configured to be receptive of a request to execute lossless data compression or decompression from the processing unit and to responsively execute the lossless data compression or decompression faster than the processing unit. The processing unit and the accelerator are directly communicative by way of the dedicated wiring.
1. A system architecture, comprising:
an on-chip coherency unit;
a processing unit communicative with the on-chip coherency unit via a first interface;
an accelerator communicative with the on-chip coherency unit via a second interface, the accelerator being configured to be receptive of a request to execute lossless data compression or decompression from the processing unit and to responsively execute the lossless data compression or decompression faster than the processing unit; and
dedicated wiring by which the processing unit and the accelerator are directly communicative, wherein the dedicated wiring comprises:
a first signal wire disposed and configured to issue at least a pause signal from the processing unit to the accelerator; and
a second signal wire disposed and configured to issue at least a paused signal from the accelerator to the processing unit.
2. The system architecture according to claim 1 , wherein the processing unit is provided as one of multiple processing units respectively communicative with the on-chip coherency unit via first interfaces.
3. The system architecture according to claim 1 , wherein the on-chip coherency unit is communicative with external processors and controllers.
4. The system architecture according to claim 1 , further comprising peripheral component interconnect express (PCIe) bridge units (PBUs) respectively communicative with the on-chip coherency unit via second interfaces and a memory core.
5. The system architecture according to claim 1 , wherein the second interface comprises a direct memory access (DMA)-like interface.
6. The system architecture according to claim 1 , wherein:
the processing unit comprises a first communication control element that controls which signals are sent by the processing unit to the accelerator and when those signals are sent, and
the accelerator comprises a second communication control element that controls which signals are sent by the accelerator to the processing unit and when those signals are sent.
7. The system architecture according to claim 1 , wherein the accelerator is configured to:
delay performance of a data update for a direct memory access (DMA) store until a target address thereof reaches an exclusive coherency state; and,
in an event a pause signal is received via the dedicated wiring:
continue performance of the data update where the target address thereof already reached the exclusive coherency state; and
cancel the performance of the data update where the target address thereof has not reached the exclusive coherency state and issue a paused signal via the dedicated wiring accordingly.
8. The system architecture according to claim 7 , wherein the accelerator is further configured to resume or reroll the performance of the data update once a pause indicated by the pause signal is de-asserted or released.
9. A system architecture, comprising:
one or more drawers, each of which comprises one or more central processors (CPs) and one or more system controllers (SCs) and each of the one or more CPs comprising:
an on-chip coherency unit;
a processing unit communicative with the on-chip coherency unit via a first interface;
an accelerator communicative with the on-chip coherency unit via a second interface, the accelerator being configured to be receptive of a request to execute lossless data compression or decompression from the processing unit and to responsively execute the lossless data compression or decompression faster than the processing unit; and
dedicated wiring by which the processing unit and the accelerator are directly communicative, wherein the dedicated wiring of each of the one or more CPs of each of the one or more drawers comprises:
a first signal wire disposed and configured to issue at least a pause signal from the processing unit to the accelerator; and
a second signal wire disposed and configured to issue at least a paused signal from the accelerator to the processing unit.
10. The system architecture according to claim 9 , wherein:
the one or more drawers are communitive with each other, and
the on-chip coherency unit of each of the one or more CPs of each of the one or more drawers is communicative with the one or more CPs and the one or more SCs of the corresponding drawer.
11. The system architecture according to claim 9 , wherein the processing unit of each of the one or more CPs of each of the one or more drawers is provided as one of multiple processing units respectively communicative with the corresponding on-chip coherency unit via first interfaces.
12. The system architecture according to claim 9 , wherein each of the one or more CPs of each of the one or more drawers further comprises peripheral component interconnect express (PCIe) bridge units (PBUs) respectively communicative with the corresponding on-chip coherency unit via second interfaces and a memory core.
13. The system architecture according to claim 9 , wherein the second interface of each of the one or more CPs of each of the one or more drawers comprises a direct memory access (DMA)-like interface.
14. The system architecture according to claim 9 , wherein:
the processing unit of each of the one or more CPs of each of the one or more drawers comprises a first communication control element that controls which signals are sent by the processing unit to the accelerator and when those signals are sent, and
the accelerator of each of the one or more CPs of each of the one or more drawers comprises a second communication control element that controls which signals are sent by the accelerator to the processing unit and when those signals are sent.
15. The system architecture according to claim 9 , wherein the accelerator of each of the one or more CPs of each of the one or more drawers is configured to:
delay performance of a data update for a direct memory access (DMA) store until a target address thereof reaches an exclusive coherency state; and,
in an event a pause signal is received via the dedicated wiring:
continue performance of the data update where the target address thereof already reached the exclusive coherency state; and
cancel the performance of the data update where the target address thereof has not reached the exclusive coherency state and issue a paused signal via the dedicated wiring accordingly.
16. The system architecture according to claim 15 , wherein the accelerator of each of the one or more CPs of each of the one or more drawers is further configured to resume or reroll the performance of the data update once a pause indicated by the pause signal is de-asserted or released.
17. A system architecture, comprising:
an on-chip coherency unit;
a processing unit communicative with the on-chip coherency unit via a first interface;
an accelerator communicative with the on-chip coherency unit via a second interface, the accelerator being configured to be receptive of a request to execute lossless data compression or decompression from the processing unit and to responsively execute the lossless data compression or decompression faster than the processing unit; and
dedicated wiring by which the processing unit and the accelerator are directly communicative, wherein the accelerator is configured to:
delay performance of a data update for a direct memory access (DMA) store until a target address thereof reaches an exclusive coherency state; and,
in an event a pause signal is received via the dedicated wiring:
continue performance of the data update where the target address thereof already reached the exclusive coherency state; and
cancel the performance of the data update where the target address thereof has not reached the exclusive coherency state and issue a paused signal via the dedicated wiring accordingly.
18. The system architecture according to claim 17 , wherein the accelerator is further configured to resume or reroll the performance of the data update once a pause indicated by the pause signal is de-asserted or released.