+import csv
+from typing import Dict, List, TextIO, Tuple
+from httomo.runner.monitoring_interface import MonitoringInterface
+from mpi4py import MPI
+from collections import OrderedDict
+
+
+[docs]class BenchmarkMonitoring(MonitoringInterface):
+
def __init__(self, comm: MPI.Comm) -> None:
+
self._comm = comm
+
self._data: List[Dict[str, str]] = []
+
self._total = 0.0
+
+
[docs] def report_method_block(
+
self,
+
method_name: str,
+
module_path: str,
+
task_id: str,
+
slicing_dim: int,
+
block_dims: Tuple[int, int, int],
+
block_idx_chunk: Tuple[int, int, int],
+
block_idx_global: Tuple[int, int, int],
+
cpu_time: float,
+
gpu_kernel_time: float = 0.0,
+
gpu_h2d_time: float = 0.0,
+
gpu_d2h_time: float = 0.0,
+
):
+
self._data.append(
+
OrderedDict(
+
[
+
("Type", "method"),
+
("Rank", str(self._comm.rank)),
+
("Name", method_name),
+
("Task id", task_id),
+
("Module", module_path),
+
("Slicing dim", str(slicing_dim)),
+
("Block offset (chunk)", str(block_idx_chunk[slicing_dim])),
+
("Block offset (global)", str(block_idx_global[slicing_dim])),
+
("Block dim z", str(block_dims[0])),
+
("Block dim y", str(block_dims[1])),
+
("Block dim x", str(block_dims[2])),
+
("CPU time", str(cpu_time)),
+
("GPU kernel time", str(gpu_kernel_time)),
+
("GPU H2D time", str(gpu_h2d_time)),
+
("GPU D2H time", str(gpu_d2h_time)),
+
]
+
)
+
)
+
+
[docs] def report_source_block(
+
self,
+
name: str,
+
first_task_id: str,
+
slicing_dim: int,
+
block_dims: Tuple[int, int, int],
+
block_idx_chunk: Tuple[int, int, int],
+
block_idx_global: Tuple[int, int, int],
+
cpu_time: float,
+
):
+
self._data.append(
+
OrderedDict(
+
[
+
("Type", "source"),
+
("Rank", str(self._comm.rank)),
+
("Name", name),
+
("Task id", first_task_id),
+
("Module", ""),
+
("Slicing dim", str(slicing_dim)),
+
("Block offset (chunk)", str(block_idx_chunk[slicing_dim])),
+
("Block offset (global)", str(block_idx_global[slicing_dim])),
+
("Block dim z", str(block_dims[0])),
+
("Block dim y", str(block_dims[1])),
+
("Block dim x", str(block_dims[2])),
+
("CPU time", str(cpu_time)),
+
("GPU kernel time", "0.0"),
+
("GPU H2D time", "0.0"),
+
("GPU D2H time", "0.0"),
+
]
+
)
+
)
+
+
[docs] def report_sink_block(
+
self,
+
name: str,
+
last_task_id: str,
+
slicing_dim: int,
+
block_dims: Tuple[int, int, int],
+
block_idx_chunk: Tuple[int, int, int],
+
block_idx_global: Tuple[int, int, int],
+
cpu_time: float,
+
):
+
self._data.append(
+
OrderedDict(
+
[
+
("Type", "sink"),
+
("Rank", str(self._comm.rank)),
+
("Name", name),
+
("Task id", last_task_id),
+
("Module", ""),
+
("Slicing dim", str(slicing_dim)),
+
("Block offset (chunk)", str(block_idx_chunk[slicing_dim])),
+
("Block offset (global)", str(block_idx_global[slicing_dim])),
+
("Block dim z", str(block_dims[0])),
+
("Block dim y", str(block_dims[1])),
+
("Block dim x", str(block_dims[2])),
+
("CPU time", str(cpu_time)),
+
("GPU kernel time", "0.0"),
+
("GPU H2D time", "0.0"),
+
("GPU D2H time", "0.0"),
+
]
+
)
+
)
+
+
[docs] def report_total_time(self, cpu_time: float):
+
self._data.append(
+
OrderedDict(
+
[
+
("Type", "total"),
+
("Rank", str(self._comm.rank)),
+
("Name", ""),
+
("Task id", ""),
+
("Module", ""),
+
("Slicing dim", ""),
+
("Block offset (chunk)", ""),
+
("Block offset (global)", ""),
+
("Block dim z", ""),
+
("Block dim y", ""),
+
("Block dim x", ""),
+
("CPU time", str(cpu_time)),
+
("GPU kernel time", "0.0"),
+
("GPU H2D time", "0.0"),
+
("GPU D2H time", "0.0"),
+
]
+
)
+
)
+
+
[docs] def write_results(self, dest: TextIO):
+
self._aggregate_mpi()
+
if self._comm.rank == 0:
+
writer = csv.DictWriter(dest, fieldnames=self._data[0].keys())
+
writer.writeheader()
+
writer.writerows(self._data)
+
+
def _aggregate_mpi(self):
+
alldata = self._comm.gather(self._data)
+
if self._comm.rank == 0:
+
self._data = sum(alldata, [])
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+