In trace-driven mode, I found memory address for inst about local memory is all the same in a warp, like
4080 ffffffff 1 R252 LDL.LU 1 R1 4 1 0xfff72c 0
But actually, local memory is private for each thread. When the inst is sent to gpgpu-sim at here, it treats global memory and local memory access as the same. So it may not properly simulate inst about local memory. This confused me. Looking forward to your reply.
Hi
In trace-driven mode, I found memory address for inst about local memory is all the same in a warp, like
4080 ffffffff 1 R252 LDL.LU 1 R1 4 1 0xfff72c 0
But actually, local memory is private for each thread. When the inst is sent to gpgpu-sim at here, it treats global memory and local memory access as the same. So it may not properly simulate inst about local memory. This confused me. Looking forward to your reply.Thanks!