Multi-granularity parallel storage system and storage
A multi-granularity parallel storage system includes an R/W port and a memory. The memory includes W memory blocks and a data gating network. Each of the memory blocks is a 2D array consisting of multiple memory units, and each memory row of the 2D array includes W memory units. For each memory block, one memory row can be read/written at a time, W is the nth power of 2, and n is a natural number.
1. A multi-granularity parallel storage system comprising:
a memory R/W port having a bit width of W, W is the nth power of 2, and n is a natural number; and
a memory comprising:
W memory, each is a 2-dimension (2D) array consisting of multiple memory units and each memory row of a 2D array comprises W memory units, for each memory block, one memory row is read/written at a time and a data gating network is configured to select W memory units, as read/write targets, from the W memory blocks based on a R/W address and a R/W granularity g, wherein the R/W granularity g represents a number of memory units having consecutive addresses, and g=2 k , k is a natural number, 0≦k≦K, K=log 2 W.
2. The storage system of claim 1 , wherein every g adjacent memory blocks are concatenated into one logic Bank, and all logic Banks have a same start address.
3. The storage system of claim 2 , wherein start addresses of memory blocks in each logic Bank are consecutive, addressing range of each logic Bank is 0−(gN−1), and addressing range of the entire memory is 0−(gN−1), where N denotes size of one memory block.
4. The storage system of claim 3 , wherein in a read operation, the memory is configured to transmit a R/W address and a R/W granularity to each logic Bank, each logic Bank being configured to read and transfer g memory units to the memory R/W port via the data gating network, and wherein data read by W/g logic Banks are concatenated from left to right into output data of a bit width W.
5. The storage system of claim 3 , wherein in a write operation, the memory is configured to divide data transferred from the memory R/W port into W/g portions, each portion of data having a bit width of g; and the memory is further configured to transmit the ith portion of data to the ith logic Bank, 0≦i<W/g, via the data gating network, and simultaneously transmit a R/W address and a R/W granularity to each logic Bank, wherein each logic Bank is configured to write g memory units.
6. A multi-granularity parallel memory comprising:
W memory blocks, W is the nth power of 2, n is a natural number, each of the memory blocks is a 2-dimension (2D) array consisting of multiple memory units, and each memory row of a 2D array comprises W memory units, for each memory block, one memory row can be read/written at a time;
a data gating network configured to select W memory units, as read/write targets, from the W memory blocks based on a R/W address and a R/W granularity g, wherein the R/W granularity g represents a number of memory units having consecutive addresses, and g=2 k , k is a natural number, 0≦k≦K, K=log 2 W.
7. The memory of claim 6 , wherein every g adjacent memory blocks are concatenated into one logic Bank, and all logic Banks have a same start address.
8. The memory of claim 7 , wherein start addresses of memory blocks in each logic Bank are consecutive, addressing range of each logic Bank is 0−(gN−1), and the addressing range of the entire memory is 0−(gN−1), where N denotes size of one memory block.
9. The memory of claim 8 , wherein in a read operation, a RIW address and a R/W granularity are transmitted to each logic Bank, each logic Bank is configured to read and transfer g memory units to the memory RIW port via the data gating network; and data read by W/g logic Banks are concatenated from left to right into output data of a bit width W.
10. The memory of claim 8 , wherein in a write operation, data transferred from a memory R/W port are divided into W/g portions, each portion of data having a bit width of g; the ith portion of data are transmitted to the ith logic Bank, 0≦i<W/g, via the data gating network, and simultaneously a RJW address and a RIW granularity are transmitted to each logic Bank; and each logic Bank writes g memory units.