Question about missing SHFL latency in RTX 3070 GPGPU-Sim configuration

[PDH-gimon]

Hi,
I have a question about the integer opcode latency configuration for the RTX 3070.
In the RTX 3070 gpgpusim.config, the integer opcode latency and initiation interval are specified with five values

However, when running GPGPU-Sim, the option description indicates that these parameters correspond to six integer operation types

I also checked the Accel-Sim GPU microbenchmark/tuner code (GPU_Microbenchmark/ubench/core/config_int/config_int.cu). From what I understand, the tuner generates only five values for ptx_opcode_latency_int and does not appear to measure an architecture-specific SHFL latency

On the other hand, I noticed that some older GPGPU-Sim configurations explicitly contain a sixth value for SHFL. For example, the Titan V configuration contains

This led me to the following questions:

1. How were the explicit SHFL latency/initiation values in configurations such as Titan V (32 cycles latency and 4 cycles initiation interval) determined? Were they obtained from a microbenchmark, an NVIDIA document, or inherited from GPGPU-Sim's default values?
2. Why does the RTX 3070 configuration contain only five integer opcode values while GPGPU-Sim currently expects six, including SHFL? Is the SHFL entry intentionally omitted, or is it simply missing from this configuration?
3. If there is no architecture-specific SHFL measurement for RTX 3070, would using the GPGPU-Sim default values (SHFL latency = 32, SHFL initiation interval = 4) be the recommended approach?

I am trying to determine an appropriate SHFL timing value for the RTX 3070 configuration and would appreciate any information about the origin of these parameters.

Thank you!