I can only find the total size of LDS. how is it organized?
How many banks? how is the bank width, 128bit or 32bit?
I think it should be important to get this information before writing a effcient program.
Evergreen LDS has 32-banks, each bank is 32-bit wide. For more information read this :
http://developer.amd.com/gpu/ATIStreamSDK/assets/ATI_Stream_SDK_Performance_Notes.pdf
ATI_Stream_SDK_Performance_Notes.pdf says :
"For current AMD Evergreen GPUs, local memory is 32 kB, organized into 32 banks; each bank is 4 bytes."
...?