For A100 and RTX3070 config, the -ptx_opcode_latency_int is missing 6th opcode latency, causing the maximum integer pipeline depth to be undefined, see:
https://github.com/accel-sim/gpgpu-sim_distribution/blob/b18ee3977962921c49fabe06d26ae19694497c26/src/gpgpu-sim/shader.cc#L3667
Since creating the int pipeline, the shader core will initialize an array of warp_inst_t with length determined by the pipeline depth, which causes huge memory overhead in certain cases.
Running l1_lat benchmark requires ~10GB of memory:

For A100 and RTX3070 config, the
-ptx_opcode_latency_intis missing 6th opcode latency, causing the maximum integer pipeline depth to be undefined, see:https://github.com/accel-sim/gpgpu-sim_distribution/blob/b18ee3977962921c49fabe06d26ae19694497c26/src/gpgpu-sim/shader.cc#L3667
Since creating the int pipeline, the shader core will initialize an array of
warp_inst_twith length determined by the pipeline depth, which causes huge memory overhead in certain cases.Running

l1_latbenchmark requires ~10GB of memory: