local mmi, mma = math.min, math.max -- use a 1D Seg class ( stage[], not stage[][]) local mfl, mce = math.floor, math.ceil local mmi, mma = math.min, math.max local pow2 = {1} for i = 2, 28 do pow2[i] = pow2[i - 1] * 2 end local SegTree = {} SegTree.updateAll = function(self) for i = self.stagenum - 1, 1, -1 do local cnt = pow2[i] for j = 0, cnt - 1 do self.stage[cnt + j] = self.func(self.stage[(cnt + j) * 2], self.stage[(cnt + j) * 2 + 1]) end end end SegTree.create = function(self, n, func, emptyvalue) self.func, self.emptyvalue = func, emptyvalue local stagenum, mul = 1, 1 self.stage = {} while mul < n do mul, stagenum = mul * 2, stagenum + 1 end self.stagenum = stagenum for i = 1, mul * 2 - 1 do self.stage[i] = emptyvalue end for i = 1, n do self.stage[mul + i - 1] = i end self:updateAll() end SegTree.update = function(self, idx, force) idx = idx + pow2[self.stagenum] - 1 for i = self.stagenum - 1, 1, -1 do local dst = mfl(idx / 2) local rem = dst * 4 + 1 - idx self.stage[dst] = self.func(self.stage[idx], self.stage[rem]) if not force and self.stage[dst] ~= self.stage[idx] then break end idx = dst end end SegTree.new = function(n, func, emptyvalue) local obj = {} setmetatable(obj, {__index = SegTree}) obj:create(n, func, emptyvalue) return obj end local MinCostFlow = {} MinCostFlow.initialize = function(self, n, spos, tpos, inf) self.n = n self.spos, self.tpos = spos, tpos self.inf = inf -- edge_dst[src][i] := dst self.edge_dst = {} -- edge_cost[src][i] := cost from src to edge_dst[src][i] self.edge_cost = {} -- edge_cap[src][i] := capacity from src to edge_dst[src][i] self.edge_cap = {} -- initial capacity. corresponding to edge_cap self.edge_initialcap = {} -- edge_dst_invedge_idx[src][i] := "j" where edge_dst[dst][j] == src -- in this case, edge_dst_invedge_idx[dst][j] should be "i". self.edge_dst_invedge_idx = {} -- len[v] := length from spos. len[spos] := 0 self.len = {} -- sub_graph_flag[v] := temporal flag to restore shortest path self.sub_graph_flag = {} -- sub_graph_v[i] := list of vertexes that are contained in the sub-graph. from tpos to spos. self.sub_graph_v = {} -- sub_graph_edgeidx[i] := edge index from sub_graph_v[i + 1] to sub_graph_v[i] self.sub_graph_edgeidx = {} -- sub_graph_size := the size of sub_graph_v. -- may not equal to #sub_graph_v (because not cleared). self.sub_graph_size = 0 for i = 1, n do self.edge_dst[i] = {} self.edge_cost[i] = {} self.edge_cap[i] = {} self.edge_initialcap[i] = {} self.edge_dst_invedge_idx[i] = {} self.len[i] = 0 self.sub_graph_flag[i] = false end end MinCostFlow.addEdge = function(self, src, dst, cost, cap) table.insert(self.edge_dst[src], dst) table.insert(self.edge_cost[src], cost) table.insert(self.edge_cap[src], cap) table.insert(self.edge_initialcap[src], cap) table.insert(self.edge_dst_invedge_idx[src], 1 + #self.edge_dst[dst]) table.insert(self.edge_dst[dst], src) table.insert(self.edge_cost[dst], -cost) table.insert(self.edge_cap[dst], 0)--invcap table.insert(self.edge_initialcap[dst], 0)--invcap table.insert(self.edge_dst_invedge_idx[dst], #self.edge_dst[src]) end MinCostFlow.invwalk_recursive = function(self) local edge_cap, edge_cost = self.edge_cap, self.edge_cost local len = self.len local sub_graph_flag = self.sub_graph_flag local sub_graph_v = self.sub_graph_v local sub_graph_edgeidx = self.sub_graph_edgeidx local searched_cnt = {0} while true do local invsrc = sub_graph_v[self.sub_graph_size] if invsrc == self.spos then break end local eddsrc = self.edge_dst[invsrc] local eddiisrc = self.edge_dst_invedge_idx[invsrc] local i = searched_cnt[self.sub_graph_size] if i < #eddsrc then i = i + 1 searched_cnt[self.sub_graph_size] = i local invdst = eddsrc[i] local j = eddiisrc[i] if 0 < edge_cap[invdst][j] and len[invdst] + edge_cost[invdst][j] == len[invsrc] and not sub_graph_flag[invdst] then self.sub_graph_size = self.sub_graph_size + 1 sub_graph_v[self.sub_graph_size] = invdst sub_graph_edgeidx[self.sub_graph_size - 1] = j sub_graph_flag[invdst] = true searched_cnt[self.sub_graph_size] = 0 end else self.sub_graph_size = self.sub_graph_size - 1 end end end MinCostFlow.walkDKSeg = function(self) local edge_dst, edge_cost, edge_cap = self.edge_dst, self.edge_cost, self.edge_cap local len = self.len local n = self.n local asked = {} for i = 1, n do asked[i] = false end asked[n + 1] = true local function mergefunc(x, y) if asked[x] then return y elseif asked[y] then return x else return len[x] < len[y] and x or y end end local st = SegTree.new(n, mergefunc, n + 1) local function walk(src, dst, cost) if len[src] + cost < len[dst] then len[dst] = len[src] + cost st:update(dst) end end for i = 1, n do local src = st.stage[1] if asked[src] then break end if self.inf <= len[src] then break end asked[src] = true st:update(src, true) local eddst, edcap, edcost = edge_dst[src], edge_cap[src], edge_cost[src] for i = 1, #eddst do if 0 < edcap[i] then walk(src, eddst[i], edcost[i]) end end end end MinCostFlow.walkDK = function(self) -- Dijkstra-like walk local edge_dst, edge_cost, edge_cap = self.edge_dst, self.edge_cost, self.edge_cap local len = self.len local tasklim = self.n if not self.taskstate then self.taskstate = {} self.tasks = {} for i = 1, tasklim do self.taskstate[i] = false end end local taskstate = self.taskstate local tasks = self.tasks local tasknum, done = 0, 0 local function addtask(idx) if not taskstate[idx] and idx ~= self.tpos then taskstate[idx] = true tasknum = tasknum + 1 local taskidx = tasknum % tasklim if taskidx == 0 then taskidx = tasklim end tasks[taskidx] = idx end end local function walk(src, dst, cost) if len[src] + cost < len[dst] then len[dst] = len[src] + cost addtask(dst) end end addtask(self.spos) while done < tasknum do done = done + 1 local taskidx = done % tasklim if taskidx == 0 then taskidx = tasklim end local idx = tasks[taskidx] taskstate[idx] = false local eddst, edcap, edcost = edge_dst[idx], edge_cap[idx], edge_cost[idx] for i = 1, #eddst do if 0 < edcap[i] then walk(idx, eddst[i], edcost[i]) end end end end MinCostFlow.walkBF = function(self) --Bellman-Ford walk local edge_dst, edge_cost, edge_cap = self.edge_dst, self.edge_cost, self.edge_cap local len = self.len local n = self.n if not self.updated1 then self.updated1 = {} self.updated2 = {} end local updated1, updated2 = self.updated1, self.updated2 for i = 1, n do updated1[i] = false end updated1[self.spos] = true len[self.spos] = 0 for irp = 1, n do local updsrc = irp % 2 == 1 and updated1 or updated2 local upddst = irp % 2 == 1 and updated2 or updated1 for i = 1, n do upddst[i] = false end local alldone = true for src = 1, n do if updsrc[src] then local eddst, edcap, edcost = edge_dst[src], edge_cap[src], edge_cost[src] for i = 1, #eddst do if 0 < edcap[i] then local dst = eddst[i] local cost = edcost[i] if len[src] + cost < len[dst] then len[dst] = len[src] + cost upddst[dst] = true alldone = false end end end end end if alldone then break end end end MinCostFlow.makeSubGraph = function(self) local inf = self.inf local len = self.len local n = self.n local edge_dst, edge_cap = self.edge_dst, self.edge_cap local edge_dst_invedge_idx = self.edge_dst_invedge_idx local edge_cost = self.edge_cost local sub_graph_v = self.sub_graph_v local sub_graph_edgeidx = self.sub_graph_edgeidx local sub_graph_flag = self.sub_graph_flag for i = 1, n do len[i] = inf sub_graph_flag[i] = false end len[self.spos] = 0 self:walkDKSeg() -- Bellman-Ford is good, but Dijkstra-like is very fast in some case -- self:walkDK() -- self:walkBF() self.sub_graph_size = 0 if inf <= len[self.tpos] then return 0 end -- restore route (from tpos to spos) self.sub_graph_size = 1 sub_graph_v[1] = self.tpos self:invwalk_recursive() local min_capacity = inf for i = self.sub_graph_size, 2, -1 do local src = sub_graph_v[i] local j = sub_graph_edgeidx[i - 1] min_capacity = mmi(min_capacity, edge_cap[src][j]) end return min_capacity end MinCostFlow.flow = function(self, capacity) local edge_dst, edge_cap = self.edge_dst, self.edge_cap local edge_dst_invedge_idx = self.edge_dst_invedge_idx local sub_graph_v = self.sub_graph_v local sub_graph_edgeidx = self.sub_graph_edgeidx for i = self.sub_graph_size, 2, -1 do local src = sub_graph_v[i] local dst = sub_graph_v[i - 1] local j = sub_graph_edgeidx[i - 1] edge_cap[src][j] = edge_cap[src][j] - capacity local k = edge_dst_invedge_idx[src][j] edge_cap[dst][k] = edge_cap[dst][k] + capacity end end MinCostFlow.getMinCostFlow = function(self, amount, invalid) local ret = 0 local cap = self:makeSubGraph() while 0 < cap do cap = mmi(amount, cap) ret = ret + self.len[self.tpos] * cap self:flow(cap) amount = amount - cap if 0 < amount then cap = self:makeSubGraph() else break end end if 0 < amount then return invalid end return ret end local mcf = MinCostFlow local n = io.read("*n") local m = io.read("*n") mcf:initialize(n, 1, n, 1000000007 * 100000) for i = 1, m do local u, v, c, d = io.read("*n", "*n", "*n", "*n") mcf:addEdge(u, v, c, 1) mcf:addEdge(v, u, c, 1) mcf:addEdge(u, v, d, 1) mcf:addEdge(v, u, d, 1) end print(mcf:getMinCostFlow(2))