# -*- coding: utf-8 -*- """ This python module implements MPI simulator class Copyright (C) 2018 SINTEF Digital This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see . """ import logging import numpy as np from mpi4py import MPI import time import pycuda.driver as cuda from GPUSimulators.simulator import BaseSimulator, BoundaryCondition #import nvtx from GPUSimulators.simulator import boundary def get_grid(num_nodes, num_dims): if not isinstance(num_nodes, int): raise TypeError("Parameter `num_nodes` is not a an integer.") if not isinstance(num_dims, int): raise TypeError("Parameter `num_dims` is not a an integer.") # Adapted from https://stackoverflow.com/questions/28057307/factoring-a-number-into-roughly-equal-factors # Original code by https://stackoverflow.com/users/3928385/ishamael # Factorizes a number into n roughly equal factors #Dictionary to remember already computed permutations memo = {} def dp(n, left): # returns tuple (cost, [factors]) """ Recursively searches through all factorizations """ #Already tried: return an existing result if (n, left) in memo: return memo[(n, left)] #Spent all factors: return number itself if left == 1: return (n, [n]) #Find a new factor i = 2 best = n best_tuple = [n] while i * i < n: #If a factor found if n % i == 0: #Factorize remainder rem = dp(n // i, left - 1) #If new permutation better, save it if rem[0] + i < best: best = rem[0] + i best_tuple = [i] + rem[1] i += 1 #Store calculation memo[(n, left)] = (best, best_tuple) return memo[(n, left)] grid = dp(num_nodes, num_dims)[1] if len(grid) < num_dims: #Split problematic 4 if 4 in grid: grid.remove(4) grid.append(2) grid.append(2) #Pad with ones to guarantee num_dims grid = grid + [1]*(num_dims - len(grid)) #Sort in descending order grid = np.sort(grid) grid = grid[::-1] # XXX: We only use vertical (north-south) partitioning for now grid[0] = 1 grid[1] = num_nodes return grid class MPIGrid(object): """ Class which represents an MPI grid of nodes. Facilitates easy communication between neighboring nodes """ def __init__(self, comm, ndims=2): self.logger = logging.getLogger(__name__) if ndims != 2: raise ValueError("Unsupported number of dimensions. Must be two at the moment") if comm.size < 1: raise ValueError("Must have at least one node") self.grid = get_grid(comm.size, ndims) self.comm = comm self.logger.debug(f"Created MPI grid: {self.grid}. Rank {self.comm.rank} has coordinate {self.get_coordinate()}") def get_coordinate(self, rank=None): if rank is None: rank = self.comm.rank i = (rank % self.grid[0]) j = (rank // self.grid[0]) return i, j def get_rank(self, i, j): return j*self.grid[0] + i def get_east(self): i, j = self.get_coordinate(self.comm.rank) i = (i+1) % self.grid[0] return self.get_rank(i, j) def get_west(self): i, j = self.get_coordinate(self.comm.rank) i = (i+self.grid[0]-1) % self.grid[0] return self.get_rank(i, j) def get_north(self): i, j = self.get_coordinate(self.comm.rank) j = (j+1) % self.grid[1] return self.get_rank(i, j) def get_south(self): i, j = self.get_coordinate(self.comm.rank) j = (j+self.grid[1]-1) % self.grid[1] return self.get_rank(i, j) def gather(self, data, root=0): out_data = None if self.comm.rank == root: out_data = np.empty([self.comm.size] + list(data.shape), dtype=data.dtype) self.comm.Gather(data, out_data, root) return out_data def get_local_rank(self): """ Returns the local rank on this node for this MPI process """ # This function has been adapted from # https://github.com/SheffieldML/PyDeepGP/blob/master/deepgp/util/parallel.py # by Zhenwen Dai released under BSD 3-Clause "New" or "Revised" License: # # Copyright (c) 2016, Zhenwen Dai # All rights reserved. # # Redistribution and use in source and binary forms, with or without # modification, are permitted provided that the following conditions are met: # # * Redistributions of source code must retain the above copyright notice, this # list of conditions and the following disclaimer. # # * Redistributions in binary form must reproduce the above copyright notice, # this list of conditions and the following disclaimer in the documentation # and/or other materials provided with the distribution. # # * Neither the name of DGP nor the names of its # contributors may be used to endorse or promote products derived from # this software without specific prior written permission. # # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" # AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE # DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR # SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER # CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, # OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #Get this ranks unique (physical) node name node_name = MPI.Get_processor_name() #Gather the list of all node names on all nodes node_names = self.comm.allgather(node_name) #Loop over all node names up until our rank #and count how many duplicates of our nodename we find local_rank = len([0 for name in node_names[:self.comm.rank] if name==node_name]) return local_rank class MPISimulator(BaseSimulator): """ Class which handles communication between simulators on different MPI nodes """ def __init__(self, sim, grid): self.profiling_data_mpi = {'start': {}, 'end': {}} self.profiling_data_mpi["start"]["t_mpi_halo_exchange"] = 0 self.profiling_data_mpi["end"]["t_mpi_halo_exchange"] = 0 self.profiling_data_mpi["start"]["t_mpi_halo_exchange_download"] = 0 self.profiling_data_mpi["end"]["t_mpi_halo_exchange_download"] = 0 self.profiling_data_mpi["start"]["t_mpi_halo_exchange_upload"] = 0 self.profiling_data_mpi["end"]["t_mpi_halo_exchange_upload"] = 0 self.profiling_data_mpi["start"]["t_mpi_halo_exchange_sendreceive"] = 0 self.profiling_data_mpi["end"]["t_mpi_halo_exchange_sendreceive"] = 0 self.profiling_data_mpi["start"]["t_mpi_step"] = 0 self.profiling_data_mpi["end"]["t_mpi_step"] = 0 self.profiling_data_mpi["n_time_steps"] = 0 self.logger = logging.getLogger(__name__) autotuner = sim.context.autotuner sim.context.autotuner = None boundary_conditions = sim.get_boundary_conditions() super().__init__(sim.context, sim.nx, sim.ny, sim.dx, sim.dy, boundary_conditions, sim.cfl_scale, sim.num_substeps, sim.block_size[0], sim.block_size[1]) sim.context.autotuner = autotuner self.sim = sim self.grid = grid #Get neighbor node ids self.east = grid.get_east() self.west = grid.get_west() self.north = grid.get_north() self.south = grid.get_south() #Get coordinate of this node #and handle global boundary conditions new_boundary_conditions = BoundaryCondition({ 'north': BoundaryCondition.Type.Dirichlet, 'south': BoundaryCondition.Type.Dirichlet, 'east': BoundaryCondition.Type.Dirichlet, 'west': BoundaryCondition.Type.Dirichlet }) gi, gj = grid.get_coordinate() #print("gi: " + str(gi) + ", gj: " + str(gj)) if gi == 0 and boundary_conditions.west != BoundaryCondition.Type.Periodic: self.west = None new_boundary_conditions.west = boundary_conditions.west if gj == 0 and boundary_conditions.south != BoundaryCondition.Type.Periodic: self.south = None new_boundary_conditions.south = boundary_conditions.south if gi == grid.grid[0]-1 and boundary_conditions.east != BoundaryCondition.Type.Periodic: self.east = None new_boundary_conditions.east = boundary_conditions.east if gj == grid.grid[1]-1 and boundary_conditions.north != BoundaryCondition.Type.Periodic: self.north = None new_boundary_conditions.north = boundary_conditions.north sim.set_boundary_conditions(new_boundary_conditions) #Get number of variables self.nvars = len(self.get_output().gpu_variables) #Shorthands for computing extents and sizes gc_x = int(self.sim.get_output()[0].x_halo) gc_y = int(self.sim.get_output()[0].y_halo) nx = int(self.sim.nx) ny = int(self.sim.ny) #Set regions for ghost cells to read from #These have the format [x0, y0, width, height] self.read_e = np.array([ nx, 0, gc_x, ny + 2*gc_y]) self.read_w = np.array([gc_x, 0, gc_x, ny + 2*gc_y]) self.read_n = np.array([gc_x, ny, nx, gc_y]) self.read_s = np.array([gc_x, gc_y, nx, gc_y]) #Set regions for ghost cells to write to self.write_e = self.read_e + np.array([gc_x, 0, 0, 0]) self.write_w = self.read_w - np.array([gc_x, 0, 0, 0]) self.write_n = self.read_n + np.array([0, gc_y, 0, 0]) self.write_s = self.read_s - np.array([0, gc_y, 0, 0]) #Allocate data for receiving #Note that east and west also transfer ghost cells #whilst north/south only transfer internal cells #Reuses the width/height defined in the read-extets above self.in_e = cuda.pagelocked_empty((int(self.nvars), int(self.read_e[3]), int(self.read_e[2])), dtype=np.float32) #np.empty((self.nvars, self.read_e[3], self.read_e[2]), dtype=np.float32) self.in_w = cuda.pagelocked_empty((int(self.nvars), int(self.read_w[3]), int(self.read_w[2])), dtype=np.float32) #np.empty((self.nvars, self.read_w[3], self.read_w[2]), dtype=np.float32) self.in_n = cuda.pagelocked_empty((int(self.nvars), int(self.read_n[3]), int(self.read_n[2])), dtype=np.float32) #np.empty((self.nvars, self.read_n[3], self.read_n[2]), dtype=np.float32) self.in_s = cuda.pagelocked_empty((int(self.nvars), int(self.read_s[3]), int(self.read_s[2])), dtype=np.float32) #np.empty((self.nvars, self.read_s[3], self.read_s[2]), dtype=np.float32) #Allocate data for sending self.out_e = cuda.pagelocked_empty((int(self.nvars), int(self.read_e[3]), int(self.read_e[2])), dtype=np.float32) #np.empty_like(self.in_e) self.out_w = cuda.pagelocked_empty((int(self.nvars), int(self.read_w[3]), int(self.read_w[2])), dtype=np.float32) #np.empty_like(self.in_w) self.out_n = cuda.pagelocked_empty((int(self.nvars), int(self.read_n[3]), int(self.read_n[2])), dtype=np.float32) #np.empty_like(self.in_n) self.out_s = cuda.pagelocked_empty((int(self.nvars), int(self.read_s[3]), int(self.read_s[2])), dtype=np.float32) #np.empty_like(self.in_s) self.logger.debug(f"Simulator rank {self.grid.comm.rank} initialized on {MPI.Get_processor_name()}") self.full_exchange() sim.context.synchronize() def substep(self, dt, step_number): #nvtx.mark("substep start", color="yellow") self.profiling_data_mpi["start"]["t_mpi_step"] += time.time() #nvtx.mark("substep external", color="blue") self.sim.substep(dt, step_number, external=True, internal=False) # only "internal ghost cells" #nvtx.mark("substep internal", color="red") self.sim.substep(dt, step_number, internal=True, external=False) # "internal ghost cells" excluded #nvtx.mark("substep full", color="blue") #self.sim.substep(dt, step_number, external=True, internal=True) self.sim.swap_buffers() self.profiling_data_mpi["end"]["t_mpi_step"] += time.time() #nvtx.mark("exchange", color="blue") self.full_exchange() #nvtx.mark("sync start", color="blue") self.sim.stream.synchronize() self.sim.internal_stream.synchronize() #nvtx.mark("sync end", color="blue") self.profiling_data_mpi["n_time_steps"] += 1 def get_output(self): return self.sim.get_output() def synchronize(self): self.sim.synchronize() def check(self): return self.sim.check() def compute_dt(self): local_dt = np.array([np.float32(self.sim.compute_dt())]) global_dt = np.empty(1, dtype=np.float32) self.grid.comm.Allreduce(local_dt, global_dt, op=MPI.MIN) self.logger.debug(f"Local dt: {local_dt[0]}, global dt: {global_dt[0]}") return global_dt[0] def get_extent(self): """ Function which returns the extent of node with rank in the grid """ width = self.sim.nx*self.sim.dx height = self.sim.ny*self.sim.dy i, j = self.grid.get_coordinate() x0 = i * width y0 = j * height x1 = x0 + width y1 = y0 + height return [x0, x1, y0, y1] def full_exchange(self): #### # First transfer internal cells north-south #### #Download from the GPU self.profiling_data_mpi["start"]["t_mpi_halo_exchange_download"] += time.time() if self.north is not None: for k in range(self.nvars): self.sim.u0[k].download(self.sim.stream, cpu_data=self.out_n[k,:,:], asynch=True, extent=self.read_n) if self.south is not None: for k in range(self.nvars): self.sim.u0[k].download(self.sim.stream, cpu_data=self.out_s[k,:,:], asynch=True, extent=self.read_s) self.sim.stream.synchronize() self.profiling_data_mpi["end"]["t_mpi_halo_exchange_download"] += time.time() #Send/receive to north/south neighbors self.profiling_data_mpi["start"]["t_mpi_halo_exchange_sendreceive"] += time.time() comm_send = [] comm_recv = [] if self.north is not None: comm_send += [self.grid.comm.Isend(self.out_n, dest=self.north, tag=4*self.nt + 0)] comm_recv += [self.grid.comm.Irecv(self.in_n, source=self.north, tag=4*self.nt + 1)] if self.south is not None: comm_send += [self.grid.comm.Isend(self.out_s, dest=self.south, tag=4*self.nt + 1)] comm_recv += [self.grid.comm.Irecv(self.in_s, source=self.south, tag=4*self.nt + 0)] #Wait for incoming transfers to complete for comm in comm_recv: comm.wait() self.profiling_data_mpi["end"]["t_mpi_halo_exchange_sendreceive"] += time.time() #Upload to the GPU self.profiling_data_mpi["start"]["t_mpi_halo_exchange_upload"] += time.time() if self.north is not None: for k in range(self.nvars): self.sim.u0[k].upload(self.sim.stream, self.in_n[k,:,:], extent=self.write_n) if self.south is not None: for k in range(self.nvars): self.sim.u0[k].upload(self.sim.stream, self.in_s[k,:,:], extent=self.write_s) self.profiling_data_mpi["end"]["t_mpi_halo_exchange_upload"] += time.time() #Wait for sending to complete self.profiling_data_mpi["start"]["t_mpi_halo_exchange_sendreceive"] += time.time() for comm in comm_send: comm.wait() self.profiling_data_mpi["end"]["t_mpi_halo_exchange_sendreceive"] += time.time() #### # Then transfer east-west including ghost cells that have been filled in by north-south transfer above #### #Download from the GPU self.profiling_data_mpi["start"]["t_mpi_halo_exchange_download"] += time.time() if self.east is not None: for k in range(self.nvars): self.sim.u0[k].download(self.sim.stream, cpu_data=self.out_e[k,:,:], asynch=True, extent=self.read_e) if self.west is not None: for k in range(self.nvars): self.sim.u0[k].download(self.sim.stream, cpu_data=self.out_w[k,:,:], asynch=True, extent=self.read_w) self.sim.stream.synchronize() self.profiling_data_mpi["end"]["t_mpi_halo_exchange_download"] += time.time() #Send/receive to east/west neighbors self.profiling_data_mpi["start"]["t_mpi_halo_exchange_sendreceive"] += time.time() comm_send = [] comm_recv = [] if self.east is not None: comm_send += [self.grid.comm.Isend(self.out_e, dest=self.east, tag=4*self.nt + 2)] comm_recv += [self.grid.comm.Irecv(self.in_e, source=self.east, tag=4*self.nt + 3)] if self.west is not None: comm_send += [self.grid.comm.Isend(self.out_w, dest=self.west, tag=4*self.nt + 3)] comm_recv += [self.grid.comm.Irecv(self.in_w, source=self.west, tag=4*self.nt + 2)] #Wait for incoming transfers to complete for comm in comm_recv: comm.wait() self.profiling_data_mpi["end"]["t_mpi_halo_exchange_sendreceive"] += time.time() #Upload to the GPU self.profiling_data_mpi["start"]["t_mpi_halo_exchange_upload"] += time.time() if self.east is not None: for k in range(self.nvars): self.sim.u0[k].upload(self.sim.stream, self.in_e[k,:,:], extent=self.write_e) if self.west is not None: for k in range(self.nvars): self.sim.u0[k].upload(self.sim.stream, self.in_w[k,:,:], extent=self.write_w) self.profiling_data_mpi["end"]["t_mpi_halo_exchange_upload"] += time.time() #Wait for sending to complete self.profiling_data_mpi["start"]["t_mpi_halo_exchange_sendreceive"] += time.time() for comm in comm_send: comm.wait() self.profiling_data_mpi["end"]["t_mpi_halo_exchange_sendreceive"] += time.time()