Method and apparatus for reading cache data, and storage medium

An apparatus for reading cache data, comprising a graphics processing unit (GPU) and a computer readable non-transitory storage medium which stores a computer program thereon, wherein the GPU comprises a Coupled Instruction Cache (CIC);

• wherein the computer is configured to initiate a read instruction to the GPU, wherein the (CIC) of the GPU is configured to simultaneously read cache data for at least two waves in the GPU, the read instruction carrying at least two types of address offset positions corresponding to the at least two waves;
• wherein the CIC is further configured to convert the at least two types of address offset positions corresponding to the read instruction into coupled address offset positions according to a preset rule;
• the CIC comprises matching control logic configured to perform matching operations in data groups of the CIC according to the coupled address offset positions which correspond to the at least two types of address offset positions and obtaining corresponding cache data for the at least two waves;
• and
• wherein the CIC is further configured to read the cache data for the at least two waves obtained through the matching control logic from corresponding cache lines of CRAMs of the CIC;
• wherein each data group of the CIC stores at least two cache data types which correspond to the at least two waves, and each cache data type is composed of a plurality of sets of cache data, each set occupying a preset length of bytes;
• each data group comprises at least two coupling serial number types, and each coupling serial number type is composed of a preset quantity of coupling serial numbers;
• the at least two cache data types have a one-to-one correspondence with the at least two coupling serial number types, and respective sets of cache data have a one-to-one correspondence with coupling serial numbers of a corresponding coupling serial number type;
• wherein the CIC is configured to convert, according to the preset rule, the at least two types of address offset positions corresponding to the read instruction into the coupled address offset positions by:
• determining a first value corresponding to an address offset position based on an original serial number corresponding to the address offset position and a total amount of all coupling serial numbers in the data group;
• determining a second value corresponding to the address offset position based on the original serial number corresponding to the address offset position and a quantity of coupling serial number types;
• determining a third value corresponding to the address offset position based on the original serial number corresponding to the address offset position and the total amount of all coupling serial numbers in the data group;
• and
• determining a coupled address offset position corresponding to the address offset position based on the first value, the second value and the third value corresponding to the address offset position and the preset length of bytes.

A computer readable non-transitory storage medium, which stores a computer program thereon, wherein the computer program, when executed by a graphics processing unit (GPU) initiates a read instruction wherein a Coupled Instruction Cache (CIC) of the GPU is configured to simultaneously read cache data for at least two waves in the GPU, the read instruction carrying at least two types of address offset positions corresponding to the at least two waves;

• wherein the CIC is further configured to convert, according to a preset rule, the at least two types of offset positions corresponding to the read instruction into coupled address offset positions;
• wherein the CIC comprises matching control logic configured to perform matching operations in data groups of the CIC according to the coupled address offset positions which correspond to the at least two types of address offset positions, to obtain corresponding cache data;
• and
• wherein the CIC is further configured to read the cache data obtained through matching;
• wherein each data group of the CIC stores at least two cache data types, which correspond to the at least two waves, and each cache data type is composed of a plurality of sets of cache data, each set occupying a preset length of bytes;
• each data group comprises at least two coupling serial number types, and each coupling serial number type is composed of a preset quantity of coupling serial numbers;
• the at least two cache data types have a one-to-one correspondence with the at least two coupling serial number types, and respective sets of cache data have a one-to-one correspondence with coupling serial numbers of a corresponding coupling serial number type;
• wherein the CIC is configured to convert, according to the preset rule, the at least two types of address offset positions corresponding to the read instruction into the coupled address offset positions by:
• determining a first value corresponding to an address offset position based on an original serial number corresponding to the address offset position and a total amount of all coupling serial numbers in the data group;
• determining a second value corresponding to the address offset position based on the original serial number corresponding to the address offset position and a quantity of coupling serial number types;
• determining a third value corresponding to the address offset position based on the original serial number corresponding to the address offset position and the total amount of all coupling serial numbers in the data group;
• and
• determining a coupled address offset position corresponding to the address offset position based on the first value, the second value and the third value corresponding to the address offset position and the preset length of bytes.