/* * nextpnr -- Next Generation Place and Route * * Copyright (C) 2018 David Shah * Copyright (C) 2018 Eddie Hung * * Permission to use, copy, modify, and/or distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. * */ #include "timing.h" #include #include #include #include #include #include #include "log.h" #include "util.h" NEXTPNR_NAMESPACE_BEGIN namespace { struct ClockEvent { IdString clock; ClockEdge edge; bool operator==(const ClockEvent &other) const { return clock == other.clock && edge == other.edge; } }; struct ClockPair { ClockEvent start, end; bool operator==(const ClockPair &other) const { return start == other.start && end == other.end; } }; } // namespace NEXTPNR_NAMESPACE_END namespace std { template <> struct hash { std::size_t operator()(const NEXTPNR_NAMESPACE_PREFIX ClockEvent &obj) const noexcept { std::size_t seed = 0; boost::hash_combine(seed, hash()(obj.clock)); boost::hash_combine(seed, hash()(int(obj.edge))); return seed; } }; template <> struct hash { std::size_t operator()(const NEXTPNR_NAMESPACE_PREFIX ClockPair &obj) const noexcept { std::size_t seed = 0; boost::hash_combine(seed, hash()(obj.start)); boost::hash_combine(seed, hash()(obj.start)); return seed; } }; } // namespace std NEXTPNR_NAMESPACE_BEGIN typedef std::vector PortRefVector; typedef std::map DelayFrequency; struct CriticalPath { PortRefVector ports; delay_t path_delay; delay_t path_period; }; typedef std::unordered_map CriticalPathMap; struct Timing { Context *ctx; bool net_delays; bool update; delay_t min_slack; CriticalPathMap *crit_path; DelayFrequency *slack_histogram; IdString async_clock; struct TimingData { TimingData() : max_arrival(), max_path_length(), min_remaining_budget() {} TimingData(delay_t max_arrival) : max_arrival(max_arrival), max_path_length(), min_remaining_budget() {} delay_t max_arrival; unsigned max_path_length = 0; delay_t min_remaining_budget; bool false_startpoint = false; std::unordered_map arrival_time; }; Timing(Context *ctx, bool net_delays, bool update, CriticalPathMap *crit_path = nullptr, DelayFrequency *slack_histogram = nullptr) : ctx(ctx), net_delays(net_delays), update(update), min_slack(1.0e12 / ctx->target_freq), crit_path(crit_path), slack_histogram(slack_histogram), async_clock(ctx->id("$async$")) { } delay_t walk_paths() { const auto clk_period = delay_t(1.0e12 / ctx->target_freq); // First, compute the topographical order of nets to walk through the circuit, assuming it is a _acyclic_ graph // TODO(eddieh): Handle the case where it is cyclic, e.g. combinatorial loops std::vector topographical_order; std::unordered_map> net_data; // In lieu of deleting edges from the graph, simply count the number of fanins to each output port std::unordered_map port_fanin; std::vector input_ports; std::vector output_ports; for (auto &cell : ctx->cells) { input_ports.clear(); output_ports.clear(); for (auto &port : cell.second->ports) { if (!port.second.net) continue; if (port.second.type == PORT_OUT) output_ports.push_back(&port.second); else input_ports.push_back(port.first); } for (auto o : output_ports) { int clocks = 0; TimingPortClass portClass = ctx->getPortTimingClass(cell.second.get(), o->name, clocks); // If output port is influenced by a clock (e.g. FF output) then add it to the ordering as a timing // start-point if (portClass == TMG_REGISTER_OUTPUT) { topographical_order.emplace_back(o->net); for (int i = 0; i < clocks; i++) { TimingClockingInfo clkInfo = ctx->getPortClockingInfo(cell.second.get(), o->name, i); const NetInfo *clknet = get_net_or_empty(cell.second.get(), clkInfo.clock_port); IdString clksig = clknet ? clknet->name : async_clock; net_data[o->net][ClockEvent{clksig, clknet ? clkInfo.edge : RISING_EDGE}] = TimingData{clkInfo.clockToQ.maxDelay()}; } } else { if (portClass == TMG_STARTPOINT || portClass == TMG_GEN_CLOCK || portClass == TMG_IGNORE) { topographical_order.emplace_back(o->net); TimingData td; td.false_startpoint = (portClass == TMG_GEN_CLOCK || portClass == TMG_IGNORE); td.max_arrival = 0; net_data[o->net][ClockEvent{async_clock, RISING_EDGE}] = td; } // Don't analyse paths from a clock input to other pins - they will be considered by the // special-case handling register input/output class ports if (portClass == TMG_CLOCK_INPUT) continue; // Otherwise, for all driven input ports on this cell, if a timing arc exists between the input and // the current output port, increment fanin counter for (auto i : input_ports) { DelayInfo comb_delay; bool is_path = ctx->getCellDelay(cell.second.get(), i, o->name, comb_delay); if (is_path) port_fanin[o]++; } } } } std::deque queue(topographical_order.begin(), topographical_order.end()); // Now walk the design, from the start points identified previously, building up a topographical order while (!queue.empty()) { const auto net = queue.front(); queue.pop_front(); for (auto &usr : net->users) { int user_clocks; TimingPortClass usrClass = ctx->getPortTimingClass(usr.cell, usr.port, user_clocks); if (usrClass == TMG_IGNORE || usrClass == TMG_CLOCK_INPUT) continue; for (auto &port : usr.cell->ports) { if (port.second.type != PORT_OUT || !port.second.net) continue; int port_clocks; TimingPortClass portClass = ctx->getPortTimingClass(usr.cell, port.first, port_clocks); // Skip if this is a clocked output (but allow non-clocked ones) if (portClass == TMG_REGISTER_OUTPUT || portClass == TMG_STARTPOINT || portClass == TMG_IGNORE || portClass == TMG_GEN_CLOCK) continue; DelayInfo comb_delay; bool is_path = ctx->getCellDelay(usr.cell, usr.port, port.first, comb_delay); if (!is_path) continue; // Decrement the fanin count, and only add to topographical order if all its fanins have already // been visited auto it = port_fanin.find(&port.second); NPNR_ASSERT(it != port_fanin.end()); if (--it->second == 0) { topographical_order.emplace_back(port.second.net); queue.emplace_back(port.second.net); port_fanin.erase(it); } } } } // Sanity check to ensure that all ports where fanins were recorded were indeed visited if (!port_fanin.empty()) { for (auto fanin : port_fanin) { NetInfo *net = fanin.first->net; if (net != nullptr) { log_info(" remaining fanin includes %s (net %s)\n", fanin.first->name.c_str(ctx), net->name.c_str(ctx)); if (net->driver.cell != nullptr) log_info(" driver = %s.%s\n", net->driver.cell->name.c_str(ctx), net->driver.port.c_str(ctx)); for (auto net_user : net->users) log_info(" user: %s.%s\n", net_user.cell->name.c_str(ctx), net_user.port.c_str(ctx)); } else { log_info(" remaining fanin includes %s (no net)\n", fanin.first->name.c_str(ctx)); } } if (ctx->force) log_warning("timing analysis failed due to presence of combinatorial loops, incomplete specification of timing ports, etc.\n"); else log_error("timing analysis failed due to presence of combinatorial loops, incomplete specification of timing ports, etc.\n"); } // Go forwards topographically to find the maximum arrival time and max path length for each net for (auto net : topographical_order) { if (!net_data.count(net)) continue; auto &nd_map = net_data.at(net); for (auto &startdomain : nd_map) { ClockEvent start_clk = startdomain.first; auto &nd = startdomain.second; if (nd.false_startpoint) continue; const auto net_arrival = nd.max_arrival; const auto net_length_plus_one = nd.max_path_length + 1; nd.min_remaining_budget = clk_period; for (auto &usr : net->users) { int port_clocks; TimingPortClass portClass = ctx->getPortTimingClass(usr.cell, usr.port, port_clocks); auto net_delay = net_delays ? ctx->getNetinfoRouteDelay(net, usr) : delay_t(); auto usr_arrival = net_arrival + net_delay; if (portClass == TMG_REGISTER_INPUT || portClass == TMG_ENDPOINT || portClass == TMG_IGNORE || portClass == TMG_CLOCK_INPUT) { // Skip } else { auto budget_override = ctx->getBudgetOverride(net, usr, net_delay); // Iterate over all output ports on the same cell as the sink for (auto port : usr.cell->ports) { if (port.second.type != PORT_OUT || !port.second.net) continue; DelayInfo comb_delay; // Look up delay through this path bool is_path = ctx->getCellDelay(usr.cell, usr.port, port.first, comb_delay); if (!is_path) continue; auto &data = net_data[port.second.net][start_clk]; auto &arrival = data.max_arrival; arrival = std::max(arrival, usr_arrival + comb_delay.maxDelay()); if (!budget_override) { // Do not increment path length if budget overriden since it doesn't // require a share of the slack auto &path_length = data.max_path_length; path_length = std::max(path_length, net_length_plus_one); } } } } } } std::unordered_map> crit_nets; // Now go backwards topographically to determine the minimum path slack, and to distribute all path slack evenly // between all nets on the path for (auto net : boost::adaptors::reverse(topographical_order)) { if (!net_data.count(net)) continue; auto &nd_map = net_data.at(net); for (auto &startdomain : nd_map) { auto &nd = startdomain.second; // Ignore false startpoints if (nd.false_startpoint) continue; const delay_t net_length_plus_one = nd.max_path_length + 1; auto &net_min_remaining_budget = nd.min_remaining_budget; for (auto &usr : net->users) { auto net_delay = net_delays ? ctx->getNetinfoRouteDelay(net, usr) : delay_t(); auto budget_override = ctx->getBudgetOverride(net, usr, net_delay); int port_clocks; TimingPortClass portClass = ctx->getPortTimingClass(usr.cell, usr.port, port_clocks); if (portClass == TMG_REGISTER_INPUT || portClass == TMG_ENDPOINT) { auto process_endpoint = [&](IdString clksig, ClockEdge edge, delay_t setup) { const auto net_arrival = nd.max_arrival; const auto endpoint_arrival = net_arrival + net_delay + setup; auto path_budget = clk_period - endpoint_arrival; delay_t period; if (edge == startdomain.first.edge) { period = clk_period; } else { period = clk_period / 2; } if (update) { auto budget_share = budget_override ? 0 : path_budget / net_length_plus_one; usr.budget = std::min(usr.budget, net_delay + budget_share); net_min_remaining_budget = std::min(net_min_remaining_budget, path_budget - budget_share); } if (path_budget < min_slack) min_slack = path_budget; if (slack_histogram) { int slack_ps = ctx->getDelayNS(path_budget) * 1000; (*slack_histogram)[slack_ps]++; } ClockEvent dest_ev{clksig, edge}; ClockPair clockPair{startdomain.first, dest_ev}; nd.arrival_time[dest_ev] = std::max(nd.arrival_time[dest_ev], endpoint_arrival); if (crit_path) { if (!crit_nets.count(clockPair) || crit_nets.at(clockPair).first < endpoint_arrival) { crit_nets[clockPair] = std::make_pair(endpoint_arrival, net); (*crit_path)[clockPair].path_delay = endpoint_arrival; (*crit_path)[clockPair].path_period = clk_period; (*crit_path)[clockPair].ports.clear(); (*crit_path)[clockPair].ports.push_back(&usr); } } }; if (portClass == TMG_REGISTER_INPUT) { for (int i = 0; i < port_clocks; i++) { TimingClockingInfo clkInfo = ctx->getPortClockingInfo(usr.cell, usr.port, i); const NetInfo *clknet = get_net_or_empty(usr.cell, clkInfo.clock_port); IdString clksig = clknet ? clknet->name : async_clock; process_endpoint(clksig, clknet ? clkInfo.edge : RISING_EDGE, clkInfo.setup.maxDelay()); } } else { process_endpoint(async_clock, RISING_EDGE, 0); } } else if (update) { // Iterate over all output ports on the same cell as the sink for (const auto &port : usr.cell->ports) { if (port.second.type != PORT_OUT || !port.second.net) continue; DelayInfo comb_delay; bool is_path = ctx->getCellDelay(usr.cell, usr.port, port.first, comb_delay); if (!is_path) continue; if (net_data.count(port.second.net) && net_data.at(port.second.net).count(startdomain.first)) { auto path_budget = net_data.at(port.second.net).at(startdomain.first).min_remaining_budget; auto budget_share = budget_override ? 0 : path_budget / net_length_plus_one; usr.budget = std::min(usr.budget, net_delay + budget_share); net_min_remaining_budget = std::min(net_min_remaining_budget, path_budget - budget_share); } } } } } } if (crit_path) { // Walk backwards from the most critical net for (auto crit_pair : crit_nets) { NetInfo *crit_net = crit_pair.second.second; auto &cp_ports = (*crit_path)[crit_pair.first].ports; while (crit_net) { const PortInfo *crit_ipin = nullptr; delay_t max_arrival = std::numeric_limits::min(); // Look at all input ports on its driving cell for (const auto &port : crit_net->driver.cell->ports) { if (port.second.type != PORT_IN || !port.second.net) continue; DelayInfo comb_delay; bool is_path = ctx->getCellDelay(crit_net->driver.cell, port.first, crit_net->driver.port, comb_delay); if (!is_path) continue; // If input port is influenced by a clock, skip int port_clocks; TimingPortClass portClass = ctx->getPortTimingClass(crit_net->driver.cell, port.first, port_clocks); if (portClass == TMG_REGISTER_INPUT || portClass == TMG_CLOCK_INPUT || portClass == TMG_ENDPOINT || portClass == TMG_IGNORE) continue; // And find the fanin net with the latest arrival time if (net_data.count(port.second.net) && net_data.at(port.second.net).count(crit_pair.first.start)) { const auto net_arrival = net_data.at(port.second.net).at(crit_pair.first.start).max_arrival; if (net_arrival > max_arrival) { max_arrival = net_arrival; crit_ipin = &port.second; } } } if (!crit_ipin) break; // Now convert PortInfo* into a PortRef* for (auto &usr : crit_ipin->net->users) { if (usr.cell->name == crit_net->driver.cell->name && usr.port == crit_ipin->name) { cp_ports.push_back(&usr); break; } } crit_net = crit_ipin->net; } std::reverse(cp_ports.begin(), cp_ports.end()); } } return min_slack; } void assign_budget() { // Clear delays to a very high value first for (auto &net : ctx->nets) { for (auto &usr : net.second->users) { usr.budget = std::numeric_limits::max(); } } walk_paths(); } }; void assign_budget(Context *ctx, bool quiet) { if (!quiet) { log_break(); log_info("Annotating ports with timing budgets for target frequency %.2f MHz\n", ctx->target_freq / 1e6); } Timing timing(ctx, ctx->slack_redist_iter > 0 /* net_delays */, true /* update */); timing.assign_budget(); if (!quiet || ctx->verbose) { for (auto &net : ctx->nets) { for (auto &user : net.second->users) { // Post-update check if (!ctx->auto_freq && user.budget < 0) log_warning("port %s.%s, connected to net '%s', has negative " "timing budget of %fns\n", user.cell->name.c_str(ctx), user.port.c_str(ctx), net.first.c_str(ctx), ctx->getDelayNS(user.budget)); else if (ctx->verbose) log_info("port %s.%s, connected to net '%s', has " "timing budget of %fns\n", user.cell->name.c_str(ctx), user.port.c_str(ctx), net.first.c_str(ctx), ctx->getDelayNS(user.budget)); } } } // For slack redistribution, if user has not specified a frequency dynamically adjust the target frequency to be the // currently achieved maximum if (ctx->auto_freq && ctx->slack_redist_iter > 0) { delay_t default_slack = delay_t((1.0e9 / ctx->getDelayNS(1)) / ctx->target_freq); ctx->target_freq = 1.0e9 / ctx->getDelayNS(default_slack - timing.min_slack); if (ctx->verbose) log_info("minimum slack for this assign = %.2f ns, target Fmax for next " "update = %.2f MHz\n", ctx->getDelayNS(timing.min_slack), ctx->target_freq / 1e6); } if (!quiet) log_info("Checksum: 0x%08x\n", ctx->checksum()); } void timing_analysis(Context *ctx, bool print_histogram, bool print_fmax, bool print_path) { auto format_event = [ctx](const ClockEvent &e, int field_width = 0) { std::string value; if (e.clock == ctx->id("$async$")) value = std::string(""); else value = (e.edge == FALLING_EDGE ? std::string("negedge ") : std::string("posedge ")) + e.clock.str(ctx); if (int(value.length()) < field_width) value.insert(value.length(), field_width - int(value.length()), ' '); return value; }; CriticalPathMap crit_paths; DelayFrequency slack_histogram; Timing timing(ctx, true /* net_delays */, false /* update */, (print_path || print_fmax) ? &crit_paths : nullptr, print_histogram ? &slack_histogram : nullptr); auto min_slack = timing.walk_paths(); std::map> clock_reports; std::vector xclock_paths; if (print_path || print_fmax) { for (auto path : crit_paths) { const ClockEvent &a = path.first.start; const ClockEvent &b = path.first.end; if (a.clock != b.clock || a.clock == ctx->id("$async$")) continue; delay_t slack = path.second.path_period - path.second.path_delay; if (!clock_reports.count(a.clock) || slack < (clock_reports.at(a.clock).second.path_period - clock_reports.at(a.clock).second.path_delay)) { clock_reports[a.clock] = path; } } for (auto &path : crit_paths) { const ClockEvent &a = path.first.start; const ClockEvent &b = path.first.end; if (a.clock == b.clock && a.clock != ctx->id("$async$")) continue; xclock_paths.push_back(path.first); } if (clock_reports.empty()) { log_warning("No clocks found in design"); } std::sort(xclock_paths.begin(), xclock_paths.end(), [ctx](const ClockPair &a, const ClockPair &b) { if (a.start.clock.str(ctx) < b.start.clock.str(ctx)) return true; if (a.start.clock.str(ctx) > b.start.clock.str(ctx)) return false; if (a.start.edge < b.start.edge) return true; if (a.start.edge > b.start.edge) return false; if (a.end.clock.str(ctx) < b.end.clock.str(ctx)) return true; if (a.end.clock.str(ctx) > b.end.clock.str(ctx)) return false; if (a.end.edge < b.end.edge) return true; return false; }); } if (print_path) { auto print_path_report = [ctx](ClockPair &clocks, PortRefVector &crit_path) { delay_t total = 0; auto &front = crit_path.front(); auto &front_port = front->cell->ports.at(front->port); auto &front_driver = front_port.net->driver; int port_clocks; auto portClass = ctx->getPortTimingClass(front_driver.cell, front_driver.port, port_clocks); IdString last_port = front_driver.port; if (portClass == TMG_REGISTER_OUTPUT) { for (int i = 0; i < port_clocks; i++) { TimingClockingInfo clockInfo = ctx->getPortClockingInfo(front_driver.cell, front_driver.port, i); const NetInfo *clknet = get_net_or_empty(front_driver.cell, clockInfo.clock_port); if (clknet != nullptr && clknet->name == clocks.start.clock && clockInfo.edge == clocks.start.edge) { } } } for (auto sink : crit_path) { auto sink_cell = sink->cell; auto &port = sink_cell->ports.at(sink->port); auto net = port.net; auto &driver = net->driver; auto driver_cell = driver.cell; DelayInfo comb_delay; if (last_port == driver.port) { // Case where we start with a STARTPOINT etc comb_delay.delay = 0; } else { ctx->getCellDelay(sink_cell, last_port, driver.port, comb_delay); } total += comb_delay.maxDelay(); log_info("%4.1f %4.1f Source %s.%s\n", ctx->getDelayNS(comb_delay.maxDelay()), ctx->getDelayNS(total), driver_cell->name.c_str(ctx), driver.port.c_str(ctx)); auto net_delay = ctx->getNetinfoRouteDelay(net, *sink); total += net_delay; auto driver_loc = ctx->getBelLocation(driver_cell->bel); auto sink_loc = ctx->getBelLocation(sink_cell->bel); log_info("%4.1f %4.1f Net %s budget %f ns (%d,%d) -> (%d,%d)\n", ctx->getDelayNS(net_delay), ctx->getDelayNS(total), net->name.c_str(ctx), ctx->getDelayNS(sink->budget), driver_loc.x, driver_loc.y, sink_loc.x, sink_loc.y); log_info(" Sink %s.%s\n", sink_cell->name.c_str(ctx), sink->port.c_str(ctx)); last_port = sink->port; } }; for (auto &clock : clock_reports) { log_break(); log_info("Critical path report for clock '%s':\n", clock.first.c_str(ctx)); log_info("curr total\n"); auto &crit_path = clock.second.second.ports; print_path_report(clock.second.first, crit_path); } for (auto &xclock : xclock_paths) { log_break(); std::string start = format_event(xclock.start); std::string end = format_event(xclock.end); log_info("Critical path report for cross-domain path '%s' -> '%s':\n", start.c_str(), end.c_str()); log_info("curr total\n"); auto &crit_path = crit_paths.at(xclock).ports; print_path_report(xclock, crit_path); } } if (print_fmax) { log_break(); for (auto &clock : clock_reports) { double Fmax; if (clock.second.first.start.edge == clock.second.first.end.edge) Fmax = 1000 / ctx->getDelayNS(clock.second.second.path_delay); else Fmax = 500 / ctx->getDelayNS(clock.second.second.path_delay); log_info("Max frequency for clock '%s': %.02f MHz\n", clock.first.c_str(ctx), Fmax); } log_break(); int start_field_width = 0, end_field_width = 0; for (auto &xclock : xclock_paths) { start_field_width = std::max((int)format_event(xclock.start).length(), start_field_width); end_field_width = std::max((int)format_event(xclock.end).length(), end_field_width); } for (auto &xclock : xclock_paths) { const ClockEvent &a = xclock.start; const ClockEvent &b = xclock.end; auto &path = crit_paths.at(xclock); auto ev_a = format_event(a, start_field_width), ev_b = format_event(b, end_field_width); log_info("Max delay %s -> %s: %0.02f ns\n", ev_a.c_str(), ev_b.c_str(), ctx->getDelayNS(path.path_delay)); } log_break(); } if (print_histogram && slack_histogram.size() > 0) { unsigned num_bins = 20; unsigned bar_width = 60; auto min_slack = slack_histogram.begin()->first; auto max_slack = slack_histogram.rbegin()->first; auto bin_size = std::max(1u, (max_slack - min_slack) / num_bins); num_bins = std::min((max_slack - min_slack) / bin_size, num_bins) + 1; std::vector bins(num_bins); unsigned max_freq = 0; for (const auto &i : slack_histogram) { auto &bin = bins[(i.first - min_slack) / bin_size]; bin += i.second; max_freq = std::max(max_freq, bin); } bar_width = std::min(bar_width, max_freq); log_break(); log_info("Slack histogram:\n"); log_info(" legend: * represents %d endpoint(s)\n", max_freq / bar_width); log_info(" + represents [1,%d) endpoint(s)\n", max_freq / bar_width); for (unsigned i = 0; i < num_bins; ++i) log_info("[%6d, %6d) |%s%c\n", min_slack + bin_size * i, min_slack + bin_size * (i + 1), std::string(bins[i] * bar_width / max_freq, '*').c_str(), (bins[i] * bar_width) % max_freq > 0 ? '+' : ' '); } } NEXTPNR_NAMESPACE_END