Parallel processing architecture with block move support
Techniques for task processing are disclosed. An array of compute elements is accessed. Each compute element within the array is known to a compiler and is coupled to its neighboring compute elements. The array of compute elements is coupled to at least one data cache. The data cache provides memory storage for the array. Control for the compute elements is provided on a cycle-by-cycle basis. Control is enabled by a stream of wide control words generated by the compiler. A load address and a store address are generated. The load and the store addresses comprise memory block move addresses. The memory block move addresses point to memory storage locations in the data cache. A memory block move is executed, based on the memory block move addresses. The data for the memory block move is transferred outside of the array.
1 . A processor-implemented method for task processing comprising:
accessing an array of compute elements, wherein each compute element within the array of compute elements is known to a compiler and is coupled to its neighboring compute elements within the array of compute elements, wherein the array of compute elements is coupled to at least one data cache, wherein the data cache provides memory storage for the array of compute elements;
providing control for the array of compute elements on a cycle-by-cycle basis, wherein the control is enabled by a stream of wide control words generated by the compiler;
generating a load address and a store address, wherein the load address and the store address comprise memory block move addresses, and wherein the memory block move addresses point to memory storage locations in the at least one data cache;
coupling load buffers located adjacent to at least one edge of the array of compute elements, wherein the load buffers provide storage for data obtained from the load address and a dataless store address; and
executing a memory block move, based on the memory block move addresses, wherein data for the memory block move is transferred outside of the array of compute elements.
2 . The method of claim 1 wherein the load address and the store address are generated in a same cycle.
3 . The method of claim 1 wherein the memory block move comprises a data cache to data cache transfer.
4 . The method of claim 1 wherein a control word from the stream of wide control words includes a load target start address, a store target start address, a block size, and a stride.
5 . The method of claim 4 wherein the generating a load address and a store address encompasses physical address translation of the load target start address and the store target start address, respectively.
6 . The method of claim 1 wherein the memory block move is executed as a pseudo-atomic operation.
7 . The method of claim 6 wherein the pseudo-atomic operation uses memory hazard detection and mitigation.
8 . The method of claim 1 wherein the memory block move that is transferred outside of the array of compute elements is enabled by the load buffers.
9 . The method of claim 1 wherein the load buffers are located adjacent to two opposite edges of the array of compute elements.
10 . The method of claim 1 further comprising coupling a crossbar switch between the load buffers and the at least one data cache.
11 . The method of claim 10 wherein the crossbar switch enables memory access anywhere within the at least one data cache.
12 . The method of claim 1 wherein the array of compute elements comprises a two-dimensional (2D) array.
13 . The method of claim 12 wherein the 2D array includes rows of compute elements and columns of compute elements.
14 . The method of claim 13 wherein the generating a load address and a store address is performed by one or more compute elements within a column of compute elements.
15 . The method of claim 1 wherein successful completion of the memory block move occurs within one architectural cycle.
16 . The method of claim 15 wherein the architectural cycle includes a plurality of clock cycles.
17 . The method of claim 1 wherein the memory block move implements a load-to-store forwarding operation.
18 . The method of claim 17 wherein the load-to-store forwarding operation enables hazard detection and mitigation.
19 . The method of claim 1 wherein the stream of wide control words comprises variable length control words generated by the compiler.
20 . A computer program product embodied in a non-transitory computer readable medium for task processing, the computer program product comprising code which causes one or more processors to perform operations of:
accessing an array of compute elements, wherein each compute element within the array of compute elements is known to a compiler and is coupled to its neighboring compute elements within the array of compute elements, wherein the array of compute elements is coupled to at least one data cache, wherein the data cache provides memory storage for the array of compute elements;
providing control for the array of compute elements on a cycle-by-cycle basis, wherein the control is enabled by a stream of wide control words generated by the compiler;
generating a load address and a store address, wherein the load address and the store address comprise memory block move addresses, and wherein the memory block move addresses point to memory storage locations in the at least one data cache;
coupling load buffers located adjacent to at least one edge of the array of compute elements, wherein the load buffers provide storage for data obtained from the load address and a dataless store address; and
executing a memory block move, based on the memory block move addresses, wherein data for the memory block move is transferred outside of the array of compute elements.
21 . A computer system for task processing comprising:
a memory which stores instructions;
one or more processors coupled to the memory, wherein the one or more processors, when executing the instructions which are stored, are configured to:
access an array of compute elements, wherein each compute element within the array of compute elements is known to a compiler and is coupled to its neighboring compute elements within the array of compute elements, wherein the array of compute elements is coupled to at least one data cache, wherein the data cache provides memory storage for the array of compute elements;
provide control for the array of compute elements on a cycle-by-cycle basis, wherein the control is enabled by a stream of wide control words generated by the compiler;
generate a load address and a store address, wherein the load address and the store address comprise memory block move addresses, and wherein the memory block move addresses point to memory storage locations in the at least one data cache;
couple load buffers located adjacent to at least one edge of the array of compute elements, wherein the load buffers provide storage for data obtained from the load address and a dataless store address; and
execute a memory block move, based on the memory block move addresses, wherein data for the memory block move is transferred outside of the array of compute elements.