64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add Shell-Patch GPU runtime fast paths

Browse Source
This commit is contained in:
CGH0S7
2026-05-08 09:26:36 +08:00
parent 1064a68d16
commit 063f28b3b4
2 changed files with 583 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
makefile_and_run.py
View File

@@ -160,6 +160,8 @@ def _gpu_runtime_env():
"AMSS_CUDA_AMR_RESTRICT_BATCH": "0",
"AMSS_CUDA_DEVICE_SEGMENT_BATCH": "0",
"AMSS_CUDA_UNCACHED_DEVICE_BUFFERS": "1",
"AMSS_SHELL_FAST_INTERP": "0",
"AMSS_SHELL_PARALLEL_INTERP": "0",
}
if finite_difference in ("2nd-order", "8th-order"):
defaults.update({
@@ -173,6 +175,13 @@ def _gpu_runtime_env():
"AMSS_CUDA_AMR_RESTRICT_BATCH": "1",
"AMSS_CUDA_DEVICE_SEGMENT_BATCH": "1",
})
if getattr(input_data, "basic_grid_set", "") == "Shell-Patch":
defaults.update({
"AMSS_CUDA_AWARE_MPI": "0",
"AMSS_SHELL_FAST_INTERP": "1",
"AMSS_SHELL_PARALLEL_INTERP": "1",
"AMSS_SHELL_INTERP_THREADS": "16",
})
if getattr(input_data, "Equation_Class", "") in ("BSSN", "BSSN-EScalar", "Z4C"):
defaults["AMSS_CUDA_AMR_RESTRICT_DEVICE"] = "1"
if getattr(input_data, "Equation_Class", "") == "Z4C":
@@ -299,6 +308,13 @@ def run_ABE():
mpi_processes = int(z4c_env_np)
elif mpi_processes < 4:
mpi_processes = 4
if (input_data.GPU_Calculation == "yes" and
getattr(input_data, "basic_grid_set", "") == "Shell-Patch"):
shell_env_np = os.environ.get("AMSS_SHELL_GPU_MPI_PROCESSES")
if shell_env_np and int(shell_env_np) > 0:
mpi_processes = int(shell_env_np)
elif mpi_processes < 4:
mpi_processes = 4
if (input_data.GPU_Calculation == "no"):
mpi_command = NUMACTL_CPU_BIND + " mpirun -np " + str(mpi_processes) + " ./ABE"
@@ -330,6 +346,9 @@ def run_ABE():
print(f" AMSS_CUDA_AMR_RESTRICT_BATCH={mpi_env.get('AMSS_CUDA_AMR_RESTRICT_BATCH', '')}")
print(f" AMSS_CUDA_DEVICE_SEGMENT_BATCH={mpi_env.get('AMSS_CUDA_DEVICE_SEGMENT_BATCH', '')}")
print(f" AMSS_CUDA_UNCACHED_DEVICE_BUFFERS={mpi_env.get('AMSS_CUDA_UNCACHED_DEVICE_BUFFERS', '')}")
print(f" AMSS_SHELL_FAST_INTERP={mpi_env.get('AMSS_SHELL_FAST_INTERP', '')}")
print(f" AMSS_SHELL_PARALLEL_INTERP={mpi_env.get('AMSS_SHELL_PARALLEL_INTERP', '')}")
print(f" AMSS_SHELL_INTERP_THREADS={mpi_env.get('AMSS_SHELL_INTERP_THREADS', '')}")
print(f" AMSS_Z4C_CUDA_RESIDENT={mpi_env.get('AMSS_Z4C_CUDA_RESIDENT', '')}")
print(f" AMSS_CONSTRAINT_OUT_EVERY={mpi_env.get('AMSS_CONSTRAINT_OUT_EVERY', '')}")
if "CUDA_MPS_PIPE_DIRECTORY" in mpi_env: