Automatic asynchronous offload to many-core coprocessors
View Patent ↗Methods and systems for asynchronous offload to many-core coprocessors include splitting a loop in an input source code into a sampling sub-part, a many integrated core (MIC) sub-part, and a central processing unit (CPU) sub-part; executing the sampling sub-part with a processor to determine loop characteristics including memory- and processor-operations executed by the loop; identifying optimal split boundaries based on the loop characteristics such that the MIC sub-part will complete in a same amount of time when executed on a MIC processor as the CPU sub-part will take when executed on a CPU; and modifying the input source code to split the loop at the identified boundaries, such that the MIC sub-part is executed on a MIC processor and the CPU sub-part is concurrently executed on a CPU.
1. A method for automating asynchronous offload to many-core coprocessors, comprising:
splitting, by a processor, a parallelizable loop in an input source code into a sampling sub-loop, a many integrated core (MIC) sub-loop, and a central processing unit (CPU) sub-loop;
executing, by the processor, the sampling sub-loop to determine loop characteristics including memory-operations and processor-operations executed by the parallelizable loop;
identifying, by the processor, optimal split boundaries based on the loop characteristics such that the MIC sub-loop will complete in a same amount of time when executed on a MIC processor as the CPU sub-loop will take when executed on a CPU; and
modifying, by the processor, the input source code to split the parallelizable loop at the identified optimal split boundaries, such that the MIC sub-loop is executed on a MIC processor and the CPU sub-loop is concurrently executed on a CPU.
2. The method of claim 1 , wherein modifying the input source code further comprises:
removing an original loop code block;
adding a MIC loop code block that executes the original loop code block on the MIC processor; and
adding a CPU loop code block that executes the original loop code block on the CPU.
3. The method of claim 2 , wherein modifying the input source code further comprises inserting preprocessor MIC offloading statements.
4. The method of claim 1 , wherein modifying the input source code further comprises handling reduction variables by replacing variable names in the MIC sub-loops and the CPU sub-loops to prevent conflicts.
5. The method of claim 1 , wherein modifying the input source code further comprises handling shared memory in critical sections using horizontal integration and vertical integration.
6. The method of claim 5 , wherein modifying the input source code further comprises performing the horizontal integration by overlapping memory regions.
7. The method of claim 5 , wherein modifying the input source code further comprises performing the vertical integration by appending memory regions to one another.
8. A system for automating asynchronous offload to many-core coprocessors, comprising:
a processor configured to split a parallelizable loop in an input source code into a sampling sub-loop, a many integrated core (MIC) sub-loop, and a central processing unit (CPU) sub-loop;
a loop sampling module configured to execute the sampling sub-loop to determine loop characteristics including memory-operations and processor-operations executed by the parallelizable loop and to identify optimal split boundaries based on the loop characteristics such that the MIC sub-loop will complete in a same amount of time when executed on a MIC processor as the CPU sub-loop will take when executed on a CPU; and
a code modifying module configured to modify the input source code to split the parallelizable loop at the identified optimal split boundaries, such that the MIC sub-loop is executed on a MIC processor and the CPU sub-loop is concurrently executed on a CPU.
9. The system of claim 8 , wherein the code modifying module is further configured to remove an original loop code block, to add a MIC loop code block that executes the original loop code block on the MIC processor, and to add a CPU loop code block that executes the original loop code block on the CPU.
10. The system of claim 9 , wherein the code modifying module is further configured to insert preprocessor MIC offloading statements.
11. The system of claim 8 , wherein the code modifying module is further configured to handle reduction variables by replacing variable names in the MIC sub-loops and the CPU sub-loops to prevent conflicts.
12. The system of claim 8 , wherein the code modifying module is further configured to handle shared memory in critical sections using horizontal integration and vertical integration.
13. The system of claim 12 , wherein the code modifying module is further configured to perform the horizontal integration by overlapping memory regions.
14. The system of claim 12 , wherein the code modifying module is further configured to perform the vertical integration by appending memory regions to one another.