Method and apparatus for leveraging simultaneous multithreading for bulk compute operations
Apparatus and method for leveraging simultaneous multithreading for bulk compute operations. For example, one embodiment of a processor comprises: a plurality of cores including a first core to simultaneously process instructions of a plurality of threads; a cache hierarchy coupled to the first core and the memory, the cache hierarchy comprising a Level 1 (L1) cache, a Level 2 (L2) cache, and a Level 3 (L3) cache; and a plurality of compute units coupled to the first core including a first compute unit associated with the L1 cache, a second compute unit associated with the L2 cache, and a third compute unit associated with the L3 cache, wherein the first core is to offload instructions for execution by the compute units, the first core to offload instructions from a first thread to the first compute unit, instructions from a second thread to the second compute unit, and instructions from a third thread to the third compute unit.
1. A processor comprising:
a plurality of cores including a first core to simultaneously process instructions of a plurality of threads;
a cache hierarchy coupled to the first core and the memory, the cache hierarchy comprising a Level 1 (L1) cache, a Level 2 (L2) cache, and a Level 3 (L3) cache; and
a plurality of compute units coupled to the first core including a first compute unit associated with the L1 cache, a second compute unit associated with the L2 cache, and a third compute unit associated with the L3 cache,
wherein the first core is to offload instructions for execution by the compute units, the first core to offload instructions from a first thread to the first compute unit, instructions from a second thread to the second compute unit, and instructions from a third thread to the third compute unit.
2. The processor of claim 1 wherein the instructions offloaded by the core comprise instructions associated with structured, repetitive loops with a plurality of iterations.
3. The processor of claim 2 wherein the instructions offloaded by the core comprise instructions to perform matrix operations and matrix-vector operations.
4. The processor of claim 2 wherein the instructions offloaded by the core comprise instructions of a Deep Neural Network kernel.
5. The processor of claim 1 wherein the first compute unit is bound exclusively to the first thread, the second compute unit is bound exclusively to the second thread, and the third compute unit is bound exclusively to the third thread, and wherein the first core is to process instructions from each of the first thread, second thread, and third thread.
6. The processor of claim 5 wherein a compute unit of the first compute unit, second compute unit, and third compute unit comprises:
a code register file comprising a first plurality of entries to store instructions to be executed by the compute unit;
a scheduler to schedule the instructions for execution;
a data register file comprising a second plurality of entries to store data associated with the instructions; and
compute circuitry to execute the instructions and store results in the data register file.
7. The processor of claim 6 wherein the compute unit further comprises:
an address generation unit (AGU) to determine virtual addresses required by the compute circuitry to execute the instructions; and
a translation cache (TC) to store translations from the virtual addresses to physical addresses of system memory accessible by the first core.
8. The processor of claim 7 further comprising:
an in-order compute issue queue coupled between the scheduler and the compute circuitry and configured to store compute instructions scheduled by the scheduler for execution by the compute circuitry.
9. The processor of claim 7 further comprising:
an in-order load/store issue queue coupled between the scheduler and the AGU and configured to store load/store instructions scheduled by the scheduler for processing by the AGU.
10. The processor of claim 1 wherein the first compute unit is to directly access the L1 cache and not directly access the L2 cache or the L3 cache; the second compute unit is to directly access the L2 cache and not directly access the L1 cache or L3 cache; and the third compute unit is to directly access the L3 cache and not directly access the L1 cache or L2 cache.
11. The processor of claim 10 further comprising:
cache allocation hardware logic to allocate a specified portion of the L3 cache to the third compute unit.
12. The processor of claim 10 wherein the L2 cache is to store a first plurality of cache lines, each cache line to include a bit to indicate whether the cache line is owned by the L2 cache or the L1 cache.
13. A system comprising:
a memory to store instructions of a plurality of threads;
a plurality of cores including a first core to simultaneously process instructions of a plurality of threads;
a cache hierarchy coupled to the first core and the memory, the cache hierarchy comprising a Level 1 (L1) cache, a Level 2 (L2) cache, and a Level 3 (L3) cache; and
a plurality of compute units coupled to the first core including a first compute unit associated with the L1 cache, a second compute unit associated with the L2 cache, and a third compute unit associated with the L3 cache,
wherein the first core is to offload instructions for execution by the compute units, the first core to offload instructions from a first thread to the first compute unit, instructions from a second thread to the second compute unit, and instructions from a third thread to the third compute unit.
14. The system of claim 13 wherein the instructions offloaded by the core comprise instructions associated with structured, repetitive loops with a plurality of iterations.
15. The system of claim 14 wherein the instructions offloaded by the core comprise instructions to perform matrix operations and matrix-vector operations.
16. The system of claim 14 wherein the instructions offloaded by the core comprise instructions of a Deep Neural Network kernel.
17. The system of claim 13 wherein the first compute unit is bound exclusively to the first thread, the second compute unit is bound exclusively to the second thread, and the third compute unit is bound exclusively to the third thread, and wherein the first core is to process instructions from each of the first thread, second thread, and third thread.
18. The system of claim 17 wherein a compute unit of the first compute unit, second compute unit, and third compute unit comprises:
a code register file comprising a first plurality of entries to store instructions to be executed by the compute unit;
a scheduler to schedule the instructions for execution;
a data register file comprising a second plurality of entries to store data associated with the instructions; and
compute circuitry to execute the instructions and store results in the data register file.
19. The system of claim 18 wherein the compute unit further comprises:
an address generation unit (AGU) to determine virtual addresses required by the compute circuitry to execute the instructions; and
a translation cache (TC) to store translations from the virtual addresses to physical addresses of system memory accessible by the first core.
20. The system of claim 19 further comprising:
an in-order compute issue queue coupled between the scheduler and the compute circuitry and configured to store compute instructions scheduled by the scheduler for execution by the compute circuitry.
21. The system of claim 19 further comprising:
an in-order load/store issue queue coupled between the scheduler and the AGU and configured to store load/store instructions scheduled by the scheduler for processing by the AGU.
22. The system of claim 13 wherein the first compute unit is to directly access the L1 cache and not directly access the L2 cache or the L3 cache; the second compute unit is to directly access the L2 cache and not directly access the L1 cache or L3 cache; and the third compute unit is to directly access the L3 cache and not directly access the L1 cache or L2 cache.
23. The system of claim 22 further comprising:
cache allocation hardware logic to allocate a specified portion of the L3 cache to the third compute unit.
24. The system of claim 22 wherein the L2 cache is to store a first plurality of cache lines, each cache line to include a bit to indicate whether the cache line is owned by the L2 cache or the L1 cache.