What is your question?
Hi everyone, I modified the example to JIT mode, i.e., gemm(mA, mB, mC, stream), but observed cache missing among different processes.

In fact, stream is only used in kernel launch, and does not affect the compilation. Adding it as a placeholder in mangle name might help.

Do you have any suggestions? Thanks!

http://github.com/NVIDIA/cutlass/blob/bd96096d58e4886e204cd1d71a385ca73e7719b8/examples/python/CuTeDSL/hopper/dense_gemm.py#L381

http://github.com/NVIDIA/cutlass/blob/bd96096d58e4886e204cd1d71a385ca73e7719b8/python/CuTeDSL/cutlass/base_dsl/dsl.py#L555

Furthermore, could you kindly expose the mangle_name API so that users could check if they need re-compilation (i.e., the AoT)?