Adaptive address translation method for high bandwidth and low IR concurrently and memory controller using the same
An adaptive memory address translation method includes the following steps. Multiple request instructions are received. A memory address corresponding to each request instruction includes a bank address. The memory addresses corresponding to the request instructions are translated, such that the bank addresses corresponding to at least one part of the any two adjacent request instructions are different. A numerical translation is utilized to translate the memory addresses corresponding to the request instructions, such that the memory addresses corresponding to the any two adjacent request instructions have less different bits.
1. An adaptive memory address translation method, comprising:
receiving a plurality of request instructions, wherein a memory address corresponding to each of the request instructions comprises a bank address;
analyzing the memory addresses corresponding to the request instructions to obtain a translation frequency of each bit in the memory addresses, wherein the translation frequency of each bit represents how frequent the bit changes value in the memory addresses;
determining a linear translation formula according to the obtained translation frequency of bits in the memory addresses;
performing a first translation of the memory addresses corresponding to the request instructions according to the linear translation formula, such that the number of identical bank addresses between two adjacent request instructions is fewer than that before the first translation; and
utilizing a numerical translation mechanism to perform a second translation of the memory addresses corresponding to the request instructions, such that the memory addresses corresponding to the any two adjacent request instructions have fewer different bits than those before the second translation.
2. The method according to claim 1 , wherein the linear translation formula relates to a memory bit having a highest translation frequency in the memory addresses corresponding to the request instructions.
3. The method according to claim 1 , wherein the numerical translation mechanism is a Gray code translation.
4. A memory controller, comprising:
a first translation unit for receiving a plurality of request instructions, wherein a memory address corresponding to each of the request instructions comprises a bank address, the first translation unit analyzes the memory addresses corresponding to the request instructions to obtain a translation frequency of each bit in the memory addresses, determines a linear translation formula according to the obtained translation frequency of bits in the memory addresses, and performs a first translation of the memory addresses corresponding to the request instructions according to the linear translation formula, such that the number of identical bank addresses between two adjacent request instructions is fewer than that before the first translation; and
a second translation unit for utilizing a numerical translation mechanism to perform a second translation of the memory addresses corresponding to the request instructions, such that the memory addresses corresponding to the any two adjacent request instructions have fewer different bits than those before the second translation,
wherein the translation frequency of each bit represents how frequent the bit changes value in the memory addresses.
5. The memory controller according to claim 4 , wherein the linear translation formula relates to a memory bit having a highest translation frequency in the memory addresses corresponding to the request instructions.
6. The memory controller according to claim 4 , wherein the numerical translation mechanism utilized by the second translation unit is a Gray code translation.