Methods for accessing memory in a two-dimensional main memory having a plurality of memory slices
In one embodiment of the invention, a memory module is disclosed including a printed circuit board with an edge connector; an address controller coupled to the printed circuit board; and a plurality of memory slices. Each of the plurality of memory slices of the memory module includes one or more memory integrated circuits coupled to the printed circuit board, and a slave memory controller coupled to the printed circuit board and the one or more memory integrated circuits. The slave memory controller receives memory access requests for the memory module from the address controller. The slave memory controller selectively activates one or more of the one or more memory integrated circuits in the respective memory slice in response to the address received from the address controller to read data from or write data into selected memory locations in the memory integrated circuits.
1. A method for accessing memory in a two-dimensional main memory having a plurality of memory slices, the method comprising:
receiving a plurality of block memory requests to access a main memory having a plurality of memory slices, wherein each memory slice includes a portion of memory of a plurality of memory modules in the two-dimensional main memory;
combining the plurality of block memory requests into a compound memory request that can concurrently access data in different memory slices over a plurality of memory modules in the two-dimensional main memory;
aggregating data together in response to the compound memory request; and
transposing data between rows of data for the main memory and columns of data for one or more processors in response to the plurality of block memory requests.
2. The method of claim 1 , wherein
the compound memory request is a write request, and
memory block data is aggregated and transposed into a row of data to write into the plurality of memory slices in the main memory.
3. The method of claim 2 , further comprising:
activating the plurality of memory slices within a two-dimensional array of memory modules in the main memory to write the row of data into memory devices of selectively activated memory modules.
4. The method of claim 2 , wherein
the transposing of data includes
storing data from the one or more processors by column into a transpositional buffer, and
reading data out from the transpositional buffer by row for writing into the main memory.
5. The method of claim 1 , wherein
the compound memory request is a read request, and
memory block data is aggregated together from the plurality of memory slices to read a row of data and transposed to provide the data to the one or more processors.
6. The method of claim 5 , further comprising:
activating the plurality of memory slices within a two-dimensional array of memory modules in the main memory to read the row of data out from memory devices of selectively activated memory modules.
7. The method of claim 5 , wherein
the transposing of data includes
storing data from main memory by row into a transpositional buffer, and
reading data out from the transpositional buffer by column for the one or more processors.
8. A method for reading data out of a two-dimensional main memory having a plurality of memory slices, the method comprising:
receiving a plurality of block memory read requests to access a main memory having a plurality of memory slices, wherein each memory slice includes a portion of memory of a plurality of memory modules in the two-dimensional main memory;
combining the plurality of block memory read requests into a compound memory slice request, wherein the compound memory slice request is a read request that can concurrently access data in different memory slices over a plurality of memory modules in the two-dimensional main memory;
executing the compound memory slice request with the main memory, wherein memory block data is aggregated together from the plurality of memory slices to read out a row of data in response to the compound memory slice request;
transposing rows of data from the main memory into columns of data for one or more processors; and
providing a row of data to the one or more processors to process data.
9. The method of claim 8 , further comprising:
activating the plurality of memory slices within a two-dimensional array of memory modules in the main memory to read the row of data out from memory devices of selectively activated memory modules.
10. The method of claim 8 , wherein
the transposing of rows of data includes
storing data from main memory by row into a transpositional buffer, and
reading data out from the transpositional buffer by column for the one or more processors.
11. A method for writing data into a two-dimensional main memory having a plurality of memory slices, the method comprising:
receiving a plurality of block memory write requests to access a main memory having a plurality of memory slices, wherein each memory slice includes a portion of memory of a plurality of memory modules in the two-dimensional main memory;
combining the plurality of block memory write requests into a compound memory slice request, wherein the compound memory slice request is a write request that can concurrently access data in different memory slices over a plurality of memory modules in the two-dimensional main memory;
aggregating memory block data together associated with the plurality of block memory write requests of the compound memory slice request;
executing the compound memory slice request with the main memory;
transposing columns of data associated with the compound memory slice request into a row of data to write into the plurality of memory slices in the main memory; and
providing the row of data to the plurality of memory slices for storage in the main memory.
12. The method of claim 11 , further comprising:
activating the plurality of memory slices within a two-dimensional array of memory modules in the main memory to write the row of data into memory devices of selectively activated memory modules.
13. The method of claim 11 , wherein
the transposing of columns of data includes
storing data from the one or more processors by column into a transpositional buffer, and
reading data out from the transpositional buffer by row for writing into the main memory.