nextpnr/common/place/placer_heap.cc
myrtle 75d2ce6a92
heap: Fix ripup criterea (#1378)
Signed-off-by: gatecat <gatecat@ds0.me>
2024-10-02 22:36:57 +02:00

1855 lines
80 KiB
C++

/*
* nextpnr -- Next Generation Place and Route
*
* Copyright (C) 2019 gatecat <gatecat@ds0.me>
*
* 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.
*
* [[cite]] HeAP
* Analytical Placement for Heterogeneous FPGAs, Marcel Gort and Jason H. Anderson
* https://janders.eecg.utoronto.ca/pdfs/marcelfpl12.pdf
*
* [[cite]] SimPL
* SimPL: An Effective Placement Algorithm, Myung-Chul Kim, Dong-Jin Lee and Igor L. Markov
* http://www.ece.umich.edu/cse/awards/pdfs/iccad10-simpl.pdf
*
* Notable changes from the original algorithm
* - Following the other nextpnr placer, Bels are placed rather than CLBs. This means a strict legalisation pass is
* added in addition to coarse legalisation (referred to as "spreading" to avoid confusion with strict legalisation)
* as described in HeAP to ensure validity. This searches random bels in the vicinity of the position chosen by
* spreading, with diameter increasing over iterations, with a heuristic to prefer lower wirelength choices.
* - To make the placer timing-driven, the bound2bound weights are multiplied by (1 + 10 * crit^2)
*/
#include "placer_heap.h"
#include <Eigen/Core>
#include <Eigen/IterativeLinearSolvers>
#include <boost/optional.hpp>
#include <chrono>
#include <deque>
#include <fstream>
#include <numeric>
#include <queue>
#include <tuple>
#include "fast_bels.h"
#include "log.h"
#include "nextpnr.h"
#include "parallel_refine.h"
#include "place_common.h"
#include "placer1.h"
#include "scope_lock.h"
#include "timing.h"
#include "util.h"
NEXTPNR_NAMESPACE_BEGIN
namespace {
// A simple internal representation for a sparse system of equations Ax = rhs
// This is designed to decouple the functions that build the matrix to the engine that
// solves it, and the representation that requires
template <typename T> struct EquationSystem
{
EquationSystem(size_t rows, size_t cols)
{
A.resize(cols);
rhs.resize(rows);
}
// Simple sparse format, easy to convert to CCS for solver
std::vector<std::vector<std::pair<int, T>>> A; // col -> (row, x[row, col]) sorted by row
std::vector<T> rhs; // RHS vector
void reset()
{
for (auto &col : A)
col.clear();
std::fill(rhs.begin(), rhs.end(), T());
}
void add_coeff(int row, int col, T val)
{
auto &Ac = A.at(col);
// Binary search
int b = 0, e = int(Ac.size()) - 1;
while (b <= e) {
int i = (b + e) / 2;
if (Ac.at(i).first == row) {
Ac.at(i).second += val;
return;
}
if (Ac.at(i).first > row)
e = i - 1;
else
b = i + 1;
}
Ac.insert(Ac.begin() + b, std::make_pair(row, val));
}
void add_rhs(int row, T val) { rhs[row] += val; }
void solve(std::vector<T> &x, float tolerance)
{
using namespace Eigen;
if (x.empty())
return;
NPNR_ASSERT(x.size() == A.size());
VectorXd vx(x.size()), vb(rhs.size());
SparseMatrix<T> mat(A.size(), A.size());
std::vector<int> colnnz;
for (auto &Ac : A)
colnnz.push_back(int(Ac.size()));
mat.reserve(colnnz);
for (int col = 0; col < int(A.size()); col++) {
auto &Ac = A.at(col);
for (auto &el : Ac)
mat.insert(el.first, col) = el.second;
}
for (int i = 0; i < int(x.size()); i++)
vx[i] = x.at(i);
for (int i = 0; i < int(rhs.size()); i++)
vb[i] = rhs.at(i);
ConjugateGradient<SparseMatrix<T>, Lower | Upper> solver;
solver.setTolerance(tolerance);
VectorXd xr = solver.compute(mat).solveWithGuess(vb, vx);
for (int i = 0; i < int(x.size()); i++)
x.at(i) = xr[i];
// for (int i = 0; i < int(x.size()); i++)
// log_info("x[%d] = %f\n", i, x.at(i));
}
};
} // namespace
class HeAPPlacer
{
public:
HeAPPlacer(Context *ctx, PlacerHeapCfg cfg)
: ctx(ctx), cfg(cfg), fast_bels(ctx, /*check_bel_available=*/true, -1), tmg(ctx)
{
Eigen::initParallel();
tmg.setup_only = true;
tmg.setup();
for (auto &cell : ctx->cells)
if (!cell.second->isPseudo() && cell.second->cluster != ClusterId())
cluster2cells[cell.second->cluster].push_back(cell.second.get());
}
bool place()
{
auto startt = std::chrono::high_resolution_clock::now();
ScopeLock<Context> lock(ctx);
place_constraints();
build_fast_bels();
seed_placement();
update_all_chains();
wirelen_t hpwl = total_hpwl();
log_info("Creating initial analytic placement for %d cells, random placement wirelen = %d.\n",
int(place_cells.size()), int(hpwl));
for (int i = 0; i < 4; i++) {
setup_solve_cells();
auto solve_startt = std::chrono::high_resolution_clock::now();
#ifdef NPNR_DISABLE_THREADS
build_solve_direction(false, -1);
build_solve_direction(true, -1);
#else
boost::thread xaxis([&]() { build_solve_direction(false, -1); });
build_solve_direction(true, -1);
xaxis.join();
#endif
auto solve_endt = std::chrono::high_resolution_clock::now();
solve_time += std::chrono::duration<double>(solve_endt - solve_startt).count();
update_all_chains();
hpwl = total_hpwl();
log_info(" at initial placer iter %d, wirelen = %d\n", i, int(hpwl));
}
wirelen_t solved_hpwl = 0, spread_hpwl = 0, legal_hpwl = 0, best_hpwl = std::numeric_limits<wirelen_t>::max();
int iter = 0, stalled = 0;
std::vector<std::tuple<CellInfo *, BelId, PlaceStrength>> solution;
std::vector<pool<BelBucketId>> heap_runs;
pool<BelBucketId> all_buckets;
dict<BelBucketId, int> bucket_count;
for (auto cell : place_cells) {
BelBucketId bucket = ctx->getBelBucketForCellType(cell->type);
if (!all_buckets.count(bucket)) {
heap_runs.push_back(pool<BelBucketId>{bucket});
all_buckets.insert(bucket);
}
bucket_count[bucket]++;
}
// If more than 98% of cells are one cell type, always solve all at once
// Otherwise, follow full HeAP strategy of rotate&all
for (auto &c : bucket_count) {
if (c.second >= 0.98 * int(place_cells.size())) {
heap_runs.clear();
break;
}
}
if (cfg.placeAllAtOnce) {
// Never want to deal with LUTs, FFs, MUXFxs separately,
// for now disable all single-cell-type runs and only have heterogeneous
// runs
heap_runs.clear();
}
heap_runs.push_back(all_buckets);
// The main HeAP placer loop
if (cfg.cell_placement_timeout > 0)
log_info("Running main analytical placer, max placement attempts per cell = %d.\n",
cfg.cell_placement_timeout);
else
log_info("Running main analytical placer.\n");
while (stalled < 5 && (solved_hpwl <= legal_hpwl * 0.8)) {
// Alternate between particular bel types and all bels
for (auto &run : heap_runs) {
auto run_startt = std::chrono::high_resolution_clock::now();
setup_solve_cells(&run);
if (solve_cells.empty())
continue;
// Heuristic: don't bother with threading below a certain size
auto solve_startt = std::chrono::high_resolution_clock::now();
// Build the connectivity matrix and run the solver; multithreaded between x and y axes if applicable
#ifndef NPNR_DISABLE_THREADS
if (solve_cells.size() >= 500) {
boost::thread xaxis([&]() { build_solve_direction(false, (iter == 0) ? -1 : iter); });
build_solve_direction(true, (iter == 0) ? -1 : iter);
xaxis.join();
} else
#endif
{
build_solve_direction(false, (iter == 0) ? -1 : iter);
build_solve_direction(true, (iter == 0) ? -1 : iter);
}
auto solve_endt = std::chrono::high_resolution_clock::now();
solve_time += std::chrono::duration<double>(solve_endt - solve_startt).count();
update_all_chains();
solved_hpwl = total_hpwl();
update_all_chains();
// Run the spreader
for (const auto &group : cfg.cellGroups)
CutSpreader(this, group).run();
for (auto type : run)
if (std::all_of(cfg.cellGroups.begin(), cfg.cellGroups.end(),
[type](const pool<BelBucketId> &grp) { return !grp.count(type); }))
CutSpreader(this, {type}).run();
// Run strict legalisation to find a valid bel for all cells
update_all_chains();
spread_hpwl = total_hpwl();
legalise_placement_strict(true);
update_all_chains();
legal_hpwl = total_hpwl();
auto run_stopt = std::chrono::high_resolution_clock::now();
IdString bucket_name = ctx->getBelBucketName(*run.begin());
log_info(" at iteration #%d, type %s: wirelen solved = %d, spread = %d, legal = %d; time = %.02fs\n",
iter + 1, (run.size() > 1 ? "ALL" : bucket_name.c_str(ctx)), int(solved_hpwl),
int(spread_hpwl), int(legal_hpwl),
std::chrono::duration<double>(run_stopt - run_startt).count());
}
// Update timing weights
if (cfg.timing_driven)
tmg.run();
if (legal_hpwl < best_hpwl) {
best_hpwl = legal_hpwl;
stalled = 0;
// Save solution
solution.clear();
for (auto &cell : ctx->cells) {
if (cell.second->isPseudo())
continue;
solution.emplace_back(cell.second.get(), cell.second->bel, cell.second->belStrength);
}
} else {
++stalled;
}
for (auto &cl : cell_locs) {
cl.second.legal_x = cl.second.x;
cl.second.legal_y = cl.second.y;
}
ctx->yield();
++iter;
}
// Remove any previous binding
for (auto &sc : solution) {
CellInfo *cell = std::get<0>(sc);
if (cell->bel != BelId())
ctx->unbindBel(cell->bel);
}
// Apply saved solution
for (auto &sc : solution) {
CellInfo *cell;
BelId bel;
PlaceStrength strength;
std::tie(cell, bel, strength) = sc;
// Just skip unbound cells here, these errors are handled just after
if (bel == BelId())
continue;
ctx->bindBel(bel, cell, strength);
}
// Find and display all errors to help in finding the root cause of issues
unsigned num_errors = 0;
for (auto &cell : ctx->cells) {
if (cell.second->isPseudo())
continue;
if (cell.second->bel == BelId()) {
log_nonfatal_error("Found unbound cell '%s' of type '%s'\n", cell.first.c_str(ctx),
cell.second->type.c_str(ctx));
num_errors++;
} else if (ctx->getBoundBelCell(cell.second->bel) != cell.second.get()) {
log_nonfatal_error("Found mismatched binding for '%s' or type '%s'\n", cell.first.c_str(ctx),
cell.second->type.c_str(ctx));
num_errors++;
} else if (ctx->debug)
log_info("AP soln: %s -> %s\n", cell.first.c_str(ctx), ctx->nameOfBel(cell.second->bel));
}
if (num_errors > 0) {
log_error("Stopping the program after %u errors found\n", num_errors);
}
bool any_bad_placements = false;
for (auto bel : ctx->getBels()) {
CellInfo *cell = ctx->getBoundBelCell(bel);
if (!ctx->isBelLocationValid(bel, /* explain_invalid */ true)) {
std::string cell_text = "no cell";
if (cell != nullptr)
cell_text = std::string("cell '") + ctx->nameOf(cell) + "'";
log_warning("post-placement validity check failed for Bel '%s' "
"(%s)\n",
ctx->nameOfBel(bel), cell_text.c_str());
any_bad_placements = true;
}
}
if (any_bad_placements) {
return false;
}
auto endtt = std::chrono::high_resolution_clock::now();
log_info("HeAP Placer Time: %.02fs\n", std::chrono::duration<double>(endtt - startt).count());
log_info(" of which solving equations: %.02fs\n", solve_time);
log_info(" of which spreading cells: %.02fs\n", cl_time);
log_info(" of which strict legalisation: %.02fs\n", sl_time);
ctx->check();
lock.unlock_early();
#if !defined(NPNR_DISABLE_THREADS)
if (cfg.parallelRefine) {
if (!parallel_refine(ctx, ParallelRefineCfg(ctx))) {
return false;
}
} else
#endif
{
auto placer1_cfg = Placer1Cfg(ctx);
placer1_cfg.hpwl_scale_x = cfg.hpwl_scale_x;
placer1_cfg.hpwl_scale_y = cfg.hpwl_scale_y;
placer1_cfg.netShareWeight = cfg.netShareWeight;
if (!placer1_refine(ctx, placer1_cfg)) {
return false;
}
}
return true;
}
private:
Context *ctx;
PlacerHeapCfg cfg;
int max_x = 0, max_y = 0;
FastBels fast_bels;
dict<IdString, std::tuple<int, int>> bel_types;
TimingAnalyser tmg;
dict<IdString, BoundingBox> constraint_region_bounds;
// In some cases, we can't use bindBel because we allow overlap in the earlier stages. So we use this custom
// structure instead
struct CellLocation
{
int x, y;
int legal_x, legal_y;
double rawx, rawy;
bool locked, global;
};
dict<IdString, CellLocation> cell_locs;
// The set of cells that we will actually place. This excludes locked cells and children cells of macros/chains
// (only the root of each macro is placed.)
std::vector<CellInfo *> place_cells;
// The cells in the current equation being solved (a subset of place_cells in some cases, where we only place
// cells of a certain type)
std::vector<CellInfo *> solve_cells;
dict<ClusterId, std::vector<CellInfo *>> cluster2cells;
dict<ClusterId, int> chain_size;
// Performance counting
double solve_time = 0, cl_time = 0, sl_time = 0;
// Place cells with the BEL attribute set to constrain them
void place_constraints()
{
size_t placed_cells = 0;
// Initial constraints placer
for (auto &cell_entry : ctx->cells) {
CellInfo *cell = cell_entry.second.get();
if (cell->isPseudo())
continue;
auto loc = cell->attrs.find(ctx->id("BEL"));
if (loc != cell->attrs.end()) {
std::string loc_name = loc->second.as_string();
BelId bel = ctx->getBelByNameStr(loc_name);
if (bel == BelId()) {
log_error("No Bel named \'%s\' located for "
"this chip (processing BEL attribute on \'%s\')\n",
loc_name.c_str(), cell->name.c_str(ctx));
}
if (!ctx->isValidBelForCellType(cell->type, bel)) {
IdString bel_type = ctx->getBelType(bel);
log_error("Bel \'%s\' of type \'%s\' does not match cell "
"\'%s\' of type \'%s\'\n",
loc_name.c_str(), bel_type.c_str(ctx), cell->name.c_str(ctx), cell->type.c_str(ctx));
}
auto bound_cell = ctx->getBoundBelCell(bel);
if (bound_cell) {
log_error("Cell \'%s\' cannot be bound to bel \'%s\' since it is already bound to cell \'%s\'\n",
cell->name.c_str(ctx), loc_name.c_str(), bound_cell->name.c_str(ctx));
}
ctx->bindBel(bel, cell, STRENGTH_USER);
if (!ctx->isBelLocationValid(bel, /* explain_invalid */ true)) {
IdString bel_type = ctx->getBelType(bel);
log_error("Bel \'%s\' of type \'%s\' is not valid for cell "
"\'%s\' of type \'%s\'\n",
loc_name.c_str(), bel_type.c_str(ctx), cell->name.c_str(ctx), cell->type.c_str(ctx));
}
placed_cells++;
}
}
log_info("Placed %d cells based on constraints.\n", int(placed_cells));
ctx->yield();
}
void build_fast_bels()
{
for (auto bel : ctx->getBels()) {
if (!ctx->checkBelAvail(bel))
continue;
Loc loc = ctx->getBelLocation(bel);
max_x = std::max(max_x, loc.x);
max_y = std::max(max_y, loc.y);
}
pool<IdString> cell_types_in_use;
pool<BelBucketId> buckets_in_use;
for (auto &cell : ctx->cells) {
if (cell.second->isPseudo())
continue;
IdString cell_type = cell.second->type;
cell_types_in_use.insert(cell_type);
BelBucketId bucket = ctx->getBelBucketForCellType(cell_type);
buckets_in_use.insert(bucket);
}
for (auto cell_type : cell_types_in_use) {
fast_bels.addCellType(cell_type);
}
for (auto bucket : buckets_in_use) {
fast_bels.addBelBucket(bucket);
}
// Determine bounding boxes of region constraints
for (auto &region : ctx->region) {
Region *r = region.second.get();
BoundingBox bb;
if (r->constr_bels) {
bb.x0 = std::numeric_limits<int>::max();
bb.x1 = std::numeric_limits<int>::min();
bb.y0 = std::numeric_limits<int>::max();
bb.y1 = std::numeric_limits<int>::min();
for (auto bel : r->bels) {
Loc loc = ctx->getBelLocation(bel);
bb.x0 = std::min(bb.x0, loc.x);
bb.x1 = std::max(bb.x1, loc.x);
bb.y0 = std::min(bb.y0, loc.y);
bb.y1 = std::max(bb.y1, loc.y);
}
} else {
bb.x0 = 0;
bb.y0 = 0;
bb.x1 = max_x;
bb.y1 = max_y;
}
constraint_region_bounds[r->name] = bb;
}
}
// Build and solve in one direction
void build_solve_direction(bool yaxis, int iter)
{
for (int i = 0; i < 5; i++) {
EquationSystem<double> esx(solve_cells.size(), solve_cells.size());
build_equations(esx, yaxis, iter);
solve_equations(esx, yaxis);
}
}
// Check if a cell has any meaningful connectivity
bool has_connectivity(CellInfo *cell)
{
for (auto port : cell->ports) {
if (port.second.net != nullptr && port.second.net->driver.cell != nullptr &&
!port.second.net->users.empty())
return true;
}
return false;
}
// Build up a random initial placement, without regard to legality
// FIXME: Are there better approaches to the initial placement (e.g. greedy?)
void seed_placement()
{
pool<IdString> cell_types;
for (const auto &cell : ctx->cells) {
if (cell.second->isPseudo())
continue;
cell_types.insert(cell.second->type);
}
pool<BelId> bels_used;
dict<IdString, std::deque<BelId>> available_bels;
for (auto bel : ctx->getBels()) {
if (!ctx->checkBelAvail(bel)) {
continue;
}
for (auto cell_type : cell_types) {
if (ctx->isValidBelForCellType(cell_type, bel)) {
available_bels[cell_type].push_back(bel);
}
}
}
for (auto &t : available_bels) {
ctx->shuffle(t.second.begin(), t.second.end());
}
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->isPseudo()) {
Loc loc = ci->pseudo_cell->getLocation();
cell_locs[cell.first].x = loc.x;
cell_locs[cell.first].y = loc.y;
cell_locs[cell.first].locked = true;
cell_locs[cell.first].global = false;
continue;
}
if (ci->bel != BelId()) {
Loc loc = ctx->getBelLocation(ci->bel);
cell_locs[cell.first].x = loc.x;
cell_locs[cell.first].y = loc.y;
cell_locs[cell.first].locked = true;
cell_locs[cell.first].global = ctx->getBelGlobalBuf(ci->bel);
} else if (ci->cluster == ClusterId() || ctx->getClusterRootCell(ci->cluster) == ci) {
bool placed = false;
int attempt_count = 0;
while (!placed) {
++attempt_count;
if (attempt_count > 25000) {
log_error("Unable to find a placement location for cell '%s'\n", ci->name.c_str(ctx));
}
// Make sure this cell type is in the available BEL map at
// all.
if (!available_bels.count(ci->type)) {
log_error("Unable to place cell '%s', no BELs remaining to implement cell type '%s'\n",
ci->name.c_str(ctx), ci->type.c_str(ctx));
}
// Find an unused BEL from bels_for_cell_type.
auto &bels_for_cell_type = available_bels.at(ci->type);
BelId bel;
while (true) {
if (bels_for_cell_type.empty()) {
log_error("Unable to place cell '%s', no BELs remaining to implement cell type '%s'\n",
ci->name.c_str(ctx), ci->type.c_str(ctx));
}
BelId candidate_bel = bels_for_cell_type.back();
bels_for_cell_type.pop_back();
if (bels_used.count(candidate_bel)) {
// candidate_bel has already been used by another
// cell type, skip it.
continue;
}
bel = candidate_bel;
break;
}
Loc loc = ctx->getBelLocation(bel);
cell_locs[cell.first].x = loc.x;
cell_locs[cell.first].y = loc.y;
cell_locs[cell.first].locked = false;
cell_locs[cell.first].global = ctx->getBelGlobalBuf(bel);
// FIXME
if (has_connectivity(cell.second.get()) && !cfg.ioBufTypes.count(ci->type)) {
bels_used.insert(bel);
place_cells.push_back(ci);
placed = true;
} else {
ctx->bindBel(bel, ci, STRENGTH_STRONG);
if (ctx->isBelLocationValid(bel)) {
cell_locs[cell.first].locked = true;
placed = true;
bels_used.insert(bel);
} else {
ctx->unbindBel(bel);
available_bels.at(ci->type).push_front(bel);
}
}
}
}
}
}
// Setup the cells to be solved, returns the number of rows
int setup_solve_cells(pool<BelBucketId> *buckets = nullptr)
{
int row = 0;
solve_cells.clear();
// First clear the udata of all cells
for (auto &cell : ctx->cells) {
cell.second->udata = dont_solve;
}
// Then update cells to be placed, which excludes cell children
for (auto cell : place_cells) {
if (buckets && !buckets->count(ctx->getBelBucketForCellType(cell->type)))
continue;
cell->udata = row++;
solve_cells.push_back(cell);
}
// Finally, update the udata of children
for (auto &cluster : cluster2cells)
for (auto child : cluster.second)
child->udata = ctx->getClusterRootCell(cluster.first)->udata;
return row;
}
// Update all chains
void update_all_chains()
{
for (auto cell : place_cells) {
chain_size[cell->name] = 1;
if (cell->cluster != ClusterId()) {
const auto base = cell_locs[cell->name];
for (auto child : cluster2cells.at(cell->cluster)) {
if (child != cell)
chain_size[cell->name]++;
Loc offset = ctx->getClusterOffset(child);
cell_locs[child->name].x = std::max(0, std::min(max_x, base.x + offset.x));
cell_locs[child->name].y = std::max(0, std::min(max_y, base.y + offset.y));
}
}
}
}
// Run a function on all ports of a net - including the driver and all users
template <typename Tf> void foreach_port(NetInfo *net, Tf func)
{
if (net->driver.cell != nullptr)
func(net->driver, store_index<PortRef>());
for (auto usr : net->users.enumerate())
func(usr.value, usr.index);
}
// Build the system of equations for either X or Y
void build_equations(EquationSystem<double> &es, bool yaxis, int iter = -1)
{
// Return the x or y position of a cell, depending on ydir
auto cell_pos = [&](CellInfo *cell) { return yaxis ? cell_locs.at(cell->name).y : cell_locs.at(cell->name).x; };
auto legal_pos = [&](CellInfo *cell) {
return yaxis ? cell_locs.at(cell->name).legal_y : cell_locs.at(cell->name).legal_x;
};
es.reset();
for (auto &net : ctx->nets) {
NetInfo *ni = net.second.get();
if (ni->driver.cell == nullptr)
continue;
if (ni->users.empty())
continue;
if (cell_locs.at(ni->driver.cell->name).global)
continue;
// Find the bounds of the net in this axis, and the ports that correspond to these bounds
PortRef *lbport = nullptr, *ubport = nullptr;
int lbpos = std::numeric_limits<int>::max(), ubpos = std::numeric_limits<int>::min();
foreach_port(ni, [&](PortRef &port, store_index<PortRef> user_idx) {
int pos = cell_pos(port.cell);
if (pos < lbpos) {
lbpos = pos;
lbport = &port;
}
if (pos > ubpos) {
ubpos = pos;
ubport = &port;
}
});
NPNR_ASSERT(lbport != nullptr);
NPNR_ASSERT(ubport != nullptr);
auto stamp_equation = [&](PortRef &var, PortRef &eqn, double weight) {
if (eqn.cell->udata == dont_solve)
return;
int row = eqn.cell->udata;
int v_pos = cell_pos(var.cell);
if (var.cell->udata != dont_solve) {
es.add_coeff(row, var.cell->udata, weight);
} else {
es.add_rhs(row, -v_pos * weight);
}
if (var.cell->cluster != ClusterId()) {
Loc offset = ctx->getClusterOffset(var.cell);
es.add_rhs(row, -(yaxis ? offset.y : offset.x) * weight);
}
};
// Add all relevant connections to the matrix
foreach_port(ni, [&](PortRef &port, store_index<PortRef> user_idx) {
int this_pos = cell_pos(port.cell);
auto process_arc = [&](PortRef *other) {
if (other == &port)
return;
int o_pos = cell_pos(other->cell);
double weight = 1.0 / (ni->users.entries() *
std::max<double>(1, (yaxis ? cfg.hpwl_scale_y : cfg.hpwl_scale_x) *
std::abs(o_pos - this_pos)));
if (user_idx) {
weight *= (1.0 + cfg.timingWeight * std::pow(tmg.get_criticality(CellPortKey(port)),
cfg.criticalityExponent));
}
// If cell 0 is not fixed, it will stamp +w on its equation and -w on the other end's equation,
// if the other end isn't fixed
stamp_equation(port, port, weight);
stamp_equation(port, *other, -weight);
stamp_equation(*other, *other, weight);
stamp_equation(*other, port, -weight);
};
process_arc(lbport);
process_arc(ubport);
});
}
if (iter != -1) {
float alpha = cfg.alpha;
for (size_t row = 0; row < solve_cells.size(); row++) {
int l_pos = legal_pos(solve_cells.at(row));
int c_pos = cell_pos(solve_cells.at(row));
double weight =
alpha * iter /
std::max<double>(1, (yaxis ? cfg.hpwl_scale_y : cfg.hpwl_scale_x) * std::abs(l_pos - c_pos));
// Add an arc from legalised to current position
es.add_coeff(row, row, weight);
es.add_rhs(row, weight * l_pos);
}
}
}
// Build the system of equations for either X or Y
void solve_equations(EquationSystem<double> &es, bool yaxis)
{
// Return the x or y position of a cell, depending on ydir
auto cell_pos = [&](CellInfo *cell) { return yaxis ? cell_locs.at(cell->name).y : cell_locs.at(cell->name).x; };
std::vector<double> vals;
std::transform(solve_cells.begin(), solve_cells.end(), std::back_inserter(vals), cell_pos);
es.solve(vals, cfg.solverTolerance);
for (size_t i = 0; i < vals.size(); i++)
if (yaxis) {
cell_locs.at(solve_cells.at(i)->name).rawy = vals.at(i);
cell_locs.at(solve_cells.at(i)->name).y = std::min(max_y, std::max(0, int(vals.at(i))));
if (solve_cells.at(i)->region != nullptr)
cell_locs.at(solve_cells.at(i)->name).y =
limit_to_reg(solve_cells.at(i)->region, cell_locs.at(solve_cells.at(i)->name).y, true);
} else {
cell_locs.at(solve_cells.at(i)->name).rawx = vals.at(i);
cell_locs.at(solve_cells.at(i)->name).x = std::min(max_x, std::max(0, int(vals.at(i))));
if (solve_cells.at(i)->region != nullptr)
cell_locs.at(solve_cells.at(i)->name).x =
limit_to_reg(solve_cells.at(i)->region, cell_locs.at(solve_cells.at(i)->name).x, false);
}
}
// Compute HPWL
wirelen_t total_hpwl()
{
wirelen_t hpwl = 0;
for (auto &net : ctx->nets) {
NetInfo *ni = net.second.get();
if (ni->driver.cell == nullptr)
continue;
CellLocation &drvloc = cell_locs.at(ni->driver.cell->name);
if (drvloc.global)
continue;
int xmin = drvloc.x, xmax = drvloc.x, ymin = drvloc.y, ymax = drvloc.y;
for (auto &user : ni->users) {
CellLocation &usrloc = cell_locs.at(user.cell->name);
xmin = std::min(xmin, usrloc.x);
xmax = std::max(xmax, usrloc.x);
ymin = std::min(ymin, usrloc.y);
ymax = std::max(ymax, usrloc.y);
}
hpwl += cfg.hpwl_scale_x * (xmax - xmin) + cfg.hpwl_scale_y * (ymax - ymin);
}
return hpwl;
}
// Strict placement legalisation, performed after the initial HeAP spreading
void legalise_placement_strict(bool require_validity = false)
{
auto startt = std::chrono::high_resolution_clock::now();
// Unbind all cells placed in this solution
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->bel != BelId() &&
(ci->udata != dont_solve ||
(ci->cluster != ClusterId() && ctx->getClusterRootCell(ci->cluster)->udata != dont_solve)))
ctx->unbindBel(ci->bel);
}
// At the moment we don't follow the full HeAP algorithm using cuts for legalisation, instead using
// the simple greedy largest-macro-first approach.
std::priority_queue<std::pair<int, IdString>> remaining;
for (auto cell : solve_cells) {
remaining.emplace(chain_size[cell->name], cell->name);
}
int ripup_radius = 2;
int total_iters = 0;
int total_iters_noreset = 0;
while (!remaining.empty()) {
auto top = remaining.top();
remaining.pop();
CellInfo *ci = ctx->cells.at(top.second).get();
// Was now placed, ignore
if (ci->bel != BelId())
continue;
// log_info(" Legalising %s (%s) %d\n", top.second.c_str(ctx), ci->type.c_str(ctx), top.first);
FastBels::FastBelsData *fb;
fast_bels.getBelsForCellType(ci->type, &fb);
int radius = 0;
int iter = 0;
int iter_at_radius = 0;
int total_iters_for_cell = 0;
bool placed = false;
BelId bestBel;
int best_inp_len = std::numeric_limits<int>::max();
total_iters++;
total_iters_noreset++;
if (total_iters > int(solve_cells.size())) {
total_iters = 0;
ripup_radius = std::min(std::max(max_x, max_y), ripup_radius * 2);
}
if (total_iters_noreset > std::max(5000, 8 * int(ctx->cells.size()))) {
log_error("Unable to find legal placement for all cells, design is probably at utilisation limit.\n");
}
while (!placed) {
if (cfg.cell_placement_timeout > 0 && total_iters_for_cell > cfg.cell_placement_timeout)
log_error("Unable to find legal placement for cell '%s' of type '%s' after %d attempts, check "
"constraints and "
"utilisation. Use `--placer-heap-cell-placement-timeout` to change the number of "
"attempts.\n",
ctx->nameOf(ci), ci->type.c_str(ctx), total_iters_for_cell);
// Determine a search radius around the solver location (which increases over time) that is clamped to
// the region constraint for the cell (if applicable)
int rx = radius, ry = radius;
if (ci->region != nullptr) {
rx = std::min(radius, (constraint_region_bounds[ci->region->name].x1 -
constraint_region_bounds[ci->region->name].x0) /
2 +
1);
ry = std::min(radius, (constraint_region_bounds[ci->region->name].y1 -
constraint_region_bounds[ci->region->name].y0) /
2 +
1);
}
// Pick a random X and Y location within our search radius
int nx = ctx->rng(2 * rx + 1) + std::max(cell_locs.at(ci->name).x - rx, 0);
int ny = ctx->rng(2 * ry + 1) + std::max(cell_locs.at(ci->name).y - ry, 0);
iter++;
iter_at_radius++;
if (iter >= (10 * (radius + 1))) {
// No luck yet, increase radius
radius = std::min(std::max(max_x, max_y), radius + 1);
while (radius < std::max(max_x, max_y)) {
// Keep increasing the radius until it will actually increase the number of cells we are
// checking (e.g. BRAM and DSP will not be in all cols/rows), so we don't waste effort
for (int x = std::max(0, cell_locs.at(ci->name).x - radius);
x <= std::min(max_x, cell_locs.at(ci->name).x + radius); x++) {
if (x >= int(fb->size()))
break;
for (int y = std::max(0, cell_locs.at(ci->name).y - radius);
y <= std::min(max_y, cell_locs.at(ci->name).y + radius); y++) {
if (y >= int(fb->at(x).size()))
break;
if (fb->at(x).at(y).size() > 0)
goto notempty;
}
}
radius = std::min(std::max(max_x, max_y), radius + 1);
}
notempty:
iter_at_radius = 0;
iter = 0;
}
// If our randomly chosen cooridnate is out of bounds; or points to a tile with no relevant bels; ignore
// it
if (nx < 0 || nx > max_x)
continue;
if (ny < 0 || ny > max_y)
continue;
if (nx >= int(fb->size()))
continue;
if (ny >= int(fb->at(nx).size()))
continue;
if (fb->at(nx).at(ny).empty())
continue;
// The number of attempts to find a location to try
int need_to_explore = 2 * radius;
// If we have found at least one legal location; and made enough attempts; assume it's good enough and
// finish
if (iter_at_radius >= need_to_explore && bestBel != BelId()) {
CellInfo *bound = ctx->getBoundBelCell(bestBel);
if (bound != nullptr) {
ctx->unbindBel(bound->bel);
remaining.emplace(chain_size[bound->name], bound->name);
}
ctx->bindBel(bestBel, ci, STRENGTH_WEAK);
placed = true;
Loc loc = ctx->getBelLocation(bestBel);
cell_locs[ci->name].x = loc.x;
cell_locs[ci->name].y = loc.y;
break;
}
if (ci->cluster == ClusterId()) {
// The case where we have no relative constraints
for (auto sz : fb->at(nx).at(ny)) {
// Look through all bels in this tile; checking region constraint if applicable
if (!ci->testRegion(sz))
continue;
// Prefer available bels; unless we are dealing with a wide radius (e.g. difficult control sets)
// or occasionally trigger a tiebreaker
if (ctx->checkBelAvail(sz) || (radius > ripup_radius || ctx->rng(20000) < 10)) {
CellInfo *bound = ctx->getBoundBelCell(sz);
if (bound != nullptr) {
// Only rip up cells without constraints
if (bound->cluster != ClusterId() || bound->belStrength > STRENGTH_WEAK)
continue;
ctx->unbindBel(bound->bel);
}
// Provisionally bind the bel
ctx->bindBel(sz, ci, STRENGTH_WEAK);
if (require_validity && !ctx->isBelLocationValid(sz)) {
// New location is not legal; unbind the cell (and rebind the cell we ripped up if
// applicable)
ctx->unbindBel(sz);
if (bound != nullptr)
ctx->bindBel(sz, bound, STRENGTH_WEAK);
} else if (iter_at_radius < need_to_explore) {
// It's legal, but we haven't tried enough locations yet
ctx->unbindBel(sz);
if (bound != nullptr)
ctx->bindBel(sz, bound, STRENGTH_WEAK);
int input_len = 0;
// Compute a fast input wirelength metric at this bel; and save if better than our last
// try
for (auto &port : ci->ports) {
auto &p = port.second;
if (p.type != PORT_IN || p.net == nullptr || p.net->driver.cell == nullptr)
continue;
CellInfo *drv = p.net->driver.cell;
auto drv_loc = cell_locs.find(drv->name);
if (drv_loc == cell_locs.end())
continue;
if (drv_loc->second.global)
continue;
input_len += std::abs(drv_loc->second.x - nx) + std::abs(drv_loc->second.y - ny);
}
if (input_len < best_inp_len) {
best_inp_len = input_len;
bestBel = sz;
}
break;
} else {
// It's legal, and we've tried enough. Finish.
if (bound != nullptr)
remaining.emplace(chain_size[bound->name], bound->name);
Loc loc = ctx->getBelLocation(sz);
cell_locs[ci->name].x = loc.x;
cell_locs[ci->name].y = loc.y;
placed = true;
break;
}
}
}
} else {
// We do have relative constraints
for (auto sz : fb->at(nx).at(ny)) {
// List of cells and their destination
std::vector<std::pair<CellInfo *, BelId>> targets;
// List of bels we placed things at; and the cell that was there before if applicable
std::vector<std::pair<BelId, CellInfo *>> swaps_made;
if (!ctx->getClusterPlacement(ci->cluster, sz, targets))
continue;
for (auto &target : targets) {
// Check it satisfies the region constraint if applicable
if (!target.first->testRegion(target.second))
goto fail;
CellInfo *bound = ctx->getBoundBelCell(target.second);
// Chains cannot overlap; so if we have to ripup a cell make sure it isn't part of a chain
if (bound != nullptr)
if (bound->cluster != ClusterId() || bound->belStrength > STRENGTH_WEAK)
goto fail;
}
// Actually perform the move; keeping track of the moves we make so we can revert them if needed
for (auto &target : targets) {
CellInfo *bound = ctx->getBoundBelCell(target.second);
if (bound != nullptr)
ctx->unbindBel(target.second);
ctx->bindBel(target.second, target.first, STRENGTH_STRONG);
swaps_made.emplace_back(target.second, bound);
}
// Check that the move we have made is legal
for (auto &sm : swaps_made) {
if (!ctx->isBelLocationValid(sm.first))
goto fail;
}
if (false) {
fail:
// If the move turned out to be illegal; revert all the moves we made
for (auto &swap : swaps_made) {
ctx->unbindBel(swap.first);
if (swap.second != nullptr)
ctx->bindBel(swap.first, swap.second, STRENGTH_WEAK);
}
continue;
}
for (auto &target : targets) {
Loc loc = ctx->getBelLocation(target.second);
cell_locs[target.first->name].x = loc.x;
cell_locs[target.first->name].y = loc.y;
// log_info("%s %d %d %d\n", target.first->name.c_str(ctx), loc.x, loc.y, loc.z);
}
for (auto &swap : swaps_made) {
// Where we have ripped up cells; add them to the queue
if (swap.second != nullptr)
remaining.emplace(chain_size[swap.second->name], swap.second->name);
}
placed = true;
break;
}
}
total_iters_for_cell++;
}
}
auto endt = std::chrono::high_resolution_clock::now();
sl_time += std::chrono::duration<float>(endt - startt).count();
}
// Implementation of the cut-based spreading as described in the HeAP/SimPL papers
template <typename T> T limit_to_reg(Region *reg, T val, bool dir)
{
if (reg == nullptr)
return val;
int limit_low = dir ? constraint_region_bounds[reg->name].y0 : constraint_region_bounds[reg->name].x0;
int limit_high = dir ? constraint_region_bounds[reg->name].y1 : constraint_region_bounds[reg->name].x1;
return std::max<T>(std::min<T>(val, limit_high), limit_low);
}
struct ChainExtent
{
int x0, y0, x1, y1;
};
struct SpreaderRegion
{
int id;
int x0, y0, x1, y1;
std::vector<int> cells, bels;
bool overused(float beta) const
{
for (size_t t = 0; t < cells.size(); t++) {
if (bels.at(t) < 4) {
if (cells.at(t) > bels.at(t))
return true;
} else {
if (cells.at(t) > beta * bels.at(t))
return true;
}
}
return false;
}
};
class CutSpreader
{
public:
CutSpreader(HeAPPlacer *p, const pool<BelBucketId> &buckets) : p(p), ctx(p->ctx), buckets(buckets)
{
// Get fast BELs data for all buckets being Cut/Spread.
size_t idx = 0;
for (BelBucketId bucket : buckets) {
type_index[bucket] = idx;
FastBels::FastBelsData *fast_bels;
p->fast_bels.getBelsForBelBucket(bucket, &fast_bels);
fb.push_back(fast_bels);
++idx;
NPNR_ASSERT(fb.size() == idx);
}
}
static int seq;
void run()
{
auto startt = std::chrono::high_resolution_clock::now();
init();
find_overused_regions();
for (auto &r : regions) {
if (merged_regions.count(r.id))
continue;
#if 0
log_info("%s (%d, %d) |_> (%d, %d) %d/%d\n", beltype.c_str(ctx), r.x0, r.y0, r.x1, r.y1, r.cells,
r.bels);
#endif
}
expand_regions();
std::queue<std::pair<int, bool>> workqueue;
#if 0
std::vector<std::pair<double, double>> orig;
if (ctx->debug)
for (auto c : p->solve_cells)
orig.emplace_back(p->cell_locs[c->name].rawx, p->cell_locs[c->name].rawy);
#endif
for (auto &r : regions) {
if (merged_regions.count(r.id))
continue;
#if 0
for (auto t : sorted(beltype)) {
log_info("%s (%d, %d) |_> (%d, %d) %d/%d\n", t.c_str(ctx), r.x0, r.y0, r.x1, r.y1,
r.cells.at(type_index.at(t)), r.bels.at(type_index.at(t)));
}
#endif
workqueue.emplace(r.id, false);
}
while (!workqueue.empty()) {
auto front = workqueue.front();
workqueue.pop();
auto &r = regions.at(front.first);
if (std::all_of(r.cells.begin(), r.cells.end(), [](int x) { return x == 0; }))
continue;
auto res = cut_region(r, front.second);
if (res) {
workqueue.emplace(res->first, !front.second);
workqueue.emplace(res->second, !front.second);
} else {
// Try the other dir, in case stuck in one direction only
auto res2 = cut_region(r, !front.second);
if (res2) {
workqueue.emplace(res2->first, front.second);
workqueue.emplace(res2->second, front.second);
}
}
}
#if 0
if (ctx->debug) {
std::ofstream sp("spread" + std::to_string(seq) + ".csv");
for (size_t i = 0; i < p->solve_cells.size(); i++) {
auto &c = p->solve_cells.at(i);
if (c->type != beltype)
continue;
sp << orig.at(i).first << "," << orig.at(i).second << "," << p->cell_locs[c->name].rawx << "," << p->cell_locs[c->name].rawy << std::endl;
}
std::ofstream oc("cells" + std::to_string(seq) + ".csv");
for (size_t y = 0; y <= p->max_y; y++) {
for (size_t x = 0; x <= p->max_x; x++) {
oc << cells_at_location.at(x).at(y).size() << ", ";
}
oc << std::endl;
}
++seq;
}
#endif
auto endt = std::chrono::high_resolution_clock::now();
p->cl_time += std::chrono::duration<float>(endt - startt).count();
}
private:
HeAPPlacer *p;
Context *ctx;
pool<BelBucketId> buckets;
dict<BelBucketId, size_t> type_index;
std::vector<std::vector<std::vector<int>>> occupancy;
std::vector<std::vector<int>> groups;
std::vector<std::vector<ChainExtent>> chaines;
std::map<IdString, ChainExtent> cell_extents;
std::vector<std::vector<std::vector<std::vector<BelId>>> *> fb;
std::vector<SpreaderRegion> regions;
pool<int> merged_regions;
// Cells at a location, sorted by real (not integer) x and y
std::vector<std::vector<std::vector<CellInfo *>>> cells_at_location;
int occ_at(int x, int y, int type) { return occupancy.at(x).at(y).at(type); }
int bels_at(int x, int y, int type)
{
if (x >= int(fb.at(type)->size()) || y >= int(fb.at(type)->at(x).size()))
return 0;
return int(fb.at(type)->at(x).at(y).size());
}
bool is_cell_fixed(const CellInfo &cell) const
{
return buckets.count(ctx->getBelBucketForCellType(cell.type)) == 0;
}
size_t cell_index(const CellInfo &cell) const { return type_index.at(ctx->getBelBucketForCellType(cell.type)); }
void init()
{
occupancy.resize(p->max_x + 1,
std::vector<std::vector<int>>(p->max_y + 1, std::vector<int>(buckets.size(), 0)));
groups.resize(p->max_x + 1, std::vector<int>(p->max_y + 1, -1));
chaines.resize(p->max_x + 1, std::vector<ChainExtent>(p->max_y + 1));
cells_at_location.resize(p->max_x + 1, std::vector<std::vector<CellInfo *>>(p->max_y + 1));
for (int x = 0; x <= p->max_x; x++)
for (int y = 0; y <= p->max_y; y++) {
for (int t = 0; t < int(buckets.size()); t++) {
occupancy.at(x).at(y).at(t) = 0;
}
groups.at(x).at(y) = -1;
chaines.at(x).at(y) = {x, y, x, y};
}
auto set_chain_ext = [&](IdString cell, int x, int y) {
if (!cell_extents.count(cell))
cell_extents[cell] = {x, y, x, y};
else {
cell_extents[cell].x0 = std::min(cell_extents[cell].x0, x);
cell_extents[cell].y0 = std::min(cell_extents[cell].y0, y);
cell_extents[cell].x1 = std::max(cell_extents[cell].x1, x);
cell_extents[cell].y1 = std::max(cell_extents[cell].y1, y);
}
};
for (auto &cell_loc : p->cell_locs) {
IdString cell_name = cell_loc.first;
const CellInfo &cell = *ctx->cells.at(cell_name);
const CellLocation &loc = cell_loc.second;
if (is_cell_fixed(cell)) {
continue;
}
if (cell.belStrength > STRENGTH_STRONG) {
continue;
}
occupancy.at(cell_loc.second.x).at(cell_loc.second.y).at(cell_index(cell))++;
// Compute ultimate extent of each chain root
if (cell.cluster != ClusterId()) {
set_chain_ext(ctx->getClusterRootCell(cell.cluster)->name, loc.x, loc.y);
}
}
for (auto &cell_loc : p->cell_locs) {
IdString cell_name = cell_loc.first;
const CellInfo &cell = *ctx->cells.at(cell_name);
const CellLocation &loc = cell_loc.second;
if (is_cell_fixed(cell)) {
continue;
}
if (cell.belStrength > STRENGTH_STRONG) {
continue;
}
// Transfer chain extents to the actual chains structure
ChainExtent *ce = nullptr;
if (cell.cluster != ClusterId()) {
ce = &(cell_extents.at(ctx->getClusterRootCell(cell.cluster)->name));
}
if (ce) {
auto &lce = chaines.at(loc.x).at(loc.y);
lce.x0 = std::min(lce.x0, ce->x0);
lce.y0 = std::min(lce.y0, ce->y0);
lce.x1 = std::max(lce.x1, ce->x1);
lce.y1 = std::max(lce.y1, ce->y1);
}
}
for (auto cell : p->solve_cells) {
if (is_cell_fixed(*cell)) {
continue;
}
cells_at_location.at(p->cell_locs.at(cell->name).x).at(p->cell_locs.at(cell->name).y).push_back(cell);
}
}
void merge_regions(SpreaderRegion &merged, SpreaderRegion &mergee)
{
// Prevent grow_region from recursing while doing this
for (int x = mergee.x0; x <= mergee.x1; x++) {
for (int y = mergee.y0; y <= mergee.y1; y++) {
// log_info("%d %d\n", groups.at(x).at(y), mergee.id);
NPNR_ASSERT(groups.at(x).at(y) == mergee.id);
groups.at(x).at(y) = merged.id;
for (size_t t = 0; t < buckets.size(); t++) {
merged.cells.at(t) += occ_at(x, y, t);
merged.bels.at(t) += bels_at(x, y, t);
}
}
}
merged_regions.insert(mergee.id);
grow_region(merged, mergee.x0, mergee.y0, mergee.x1, mergee.y1);
}
void grow_region(SpreaderRegion &r, int x0, int y0, int x1, int y1, bool init = false)
{
// log_info("growing to (%d, %d) |_> (%d, %d)\n", x0, y0, x1, y1);
if ((x0 >= r.x0 && y0 >= r.y0 && x1 <= r.x1 && y1 <= r.y1) || init)
return;
int old_x0 = r.x0 + (init ? 1 : 0), old_y0 = r.y0, old_x1 = r.x1, old_y1 = r.y1;
r.x0 = std::min(r.x0, x0);
r.y0 = std::min(r.y0, y0);
r.x1 = std::max(r.x1, x1);
r.y1 = std::max(r.y1, y1);
auto process_location = [&](int x, int y) {
// Merge with any overlapping regions
if (groups.at(x).at(y) == -1) {
for (size_t t = 0; t < buckets.size(); t++) {
r.bels.at(t) += bels_at(x, y, t);
r.cells.at(t) += occ_at(x, y, t);
}
}
if (groups.at(x).at(y) != -1 && groups.at(x).at(y) != r.id)
merge_regions(r, regions.at(groups.at(x).at(y)));
groups.at(x).at(y) = r.id;
// Grow to cover any chains
auto &chaine = chaines.at(x).at(y);
grow_region(r, chaine.x0, chaine.y0, chaine.x1, chaine.y1);
};
for (int x = r.x0; x < old_x0; x++)
for (int y = r.y0; y <= r.y1; y++)
process_location(x, y);
for (int x = old_x1 + 1; x <= x1; x++)
for (int y = r.y0; y <= r.y1; y++)
process_location(x, y);
for (int y = r.y0; y < old_y0; y++)
for (int x = r.x0; x <= r.x1; x++)
process_location(x, y);
for (int y = old_y1 + 1; y <= r.y1; y++)
for (int x = r.x0; x <= r.x1; x++)
process_location(x, y);
}
void find_overused_regions()
{
for (int x = 0; x <= p->max_x; x++)
for (int y = 0; y <= p->max_y; y++) {
// Either already in a group, or not overutilised. Ignore
if (groups.at(x).at(y) != -1)
continue;
bool overutilised = false;
for (size_t t = 0; t < buckets.size(); t++) {
if (occ_at(x, y, t) > bels_at(x, y, t)) {
overutilised = true;
break;
}
}
if (!overutilised)
continue;
// log_info("%d %d %d\n", x, y, occ_at(x, y));
int id = int(regions.size());
groups.at(x).at(y) = id;
SpreaderRegion reg;
reg.id = id;
reg.x0 = reg.x1 = x;
reg.y0 = reg.y1 = y;
for (size_t t = 0; t < buckets.size(); t++) {
reg.bels.push_back(bels_at(x, y, t));
reg.cells.push_back(occ_at(x, y, t));
}
// Make sure we cover carries, etc
grow_region(reg, reg.x0, reg.y0, reg.x1, reg.y1, true);
bool expanded = true;
while (expanded) {
expanded = false;
// Keep trying expansion in x and y, until we find no over-occupancy cells
// or hit grouped cells
// First try expanding in x
if (reg.x1 < p->max_x) {
bool over_occ_x = false;
for (int y1 = reg.y0; y1 <= reg.y1; y1++) {
for (size_t t = 0; t < buckets.size(); t++) {
if (occ_at(reg.x1 + 1, y1, t) > bels_at(reg.x1 + 1, y1, t)) {
over_occ_x = true;
break;
}
}
}
if (over_occ_x) {
expanded = true;
grow_region(reg, reg.x0, reg.y0, reg.x1 + 1, reg.y1);
}
}
if (reg.y1 < p->max_y) {
bool over_occ_y = false;
for (int x1 = reg.x0; x1 <= reg.x1; x1++) {
for (size_t t = 0; t < buckets.size(); t++) {
if (occ_at(x1, reg.y1 + 1, t) > bels_at(x1, reg.y1 + 1, t)) {
over_occ_y = true;
break;
}
}
}
if (over_occ_y) {
expanded = true;
grow_region(reg, reg.x0, reg.y0, reg.x1, reg.y1 + 1);
}
}
}
regions.push_back(reg);
}
}
void expand_regions()
{
std::queue<int> overu_regions;
float beta = p->cfg.beta;
for (auto &r : regions) {
if (!merged_regions.count(r.id) && r.overused(beta))
overu_regions.push(r.id);
}
while (!overu_regions.empty()) {
int rid = overu_regions.front();
overu_regions.pop();
if (merged_regions.count(rid))
continue;
auto &reg = regions.at(rid);
while (reg.overused(beta)) {
bool changed = false;
for (int j = 0; j < p->cfg.spread_scale_x; j++) {
if (reg.x0 > 0) {
grow_region(reg, reg.x0 - 1, reg.y0, reg.x1, reg.y1);
changed = true;
if (!reg.overused(beta))
break;
}
if (reg.x1 < p->max_x) {
grow_region(reg, reg.x0, reg.y0, reg.x1 + 1, reg.y1);
changed = true;
if (!reg.overused(beta))
break;
}
}
for (int j = 0; j < p->cfg.spread_scale_y; j++) {
if (reg.y0 > 0) {
grow_region(reg, reg.x0, reg.y0 - 1, reg.x1, reg.y1);
changed = true;
if (!reg.overused(beta))
break;
}
if (reg.y1 < p->max_y) {
grow_region(reg, reg.x0, reg.y0, reg.x1, reg.y1 + 1);
changed = true;
if (!reg.overused(beta))
break;
}
}
if (!changed) {
for (auto bucket : buckets) {
if (reg.cells > reg.bels) {
IdString bucket_name = ctx->getBelBucketName(bucket);
log_error("Failed to expand region (%d, %d) |_> (%d, %d) of %d %ss\n", reg.x0, reg.y0,
reg.x1, reg.y1, reg.cells.at(type_index.at(bucket)), bucket_name.c_str(ctx));
}
}
break;
}
}
}
}
// Implementation of the recursive cut-based spreading as described in the HeAP paper
// Note we use "left" to mean "-x/-y" depending on dir and "right" to mean "+x/+y" depending on dir
std::vector<CellInfo *> cut_cells;
boost::optional<std::pair<int, int>> cut_region(SpreaderRegion &r, bool dir)
{
cut_cells.clear();
auto &cal = cells_at_location;
int total_cells = 0;
for (int x = r.x0; x <= r.x1; x++) {
for (int y = r.y0; y <= r.y1; y++) {
std::copy(cal.at(x).at(y).begin(), cal.at(x).at(y).end(), std::back_inserter(cut_cells));
}
}
for (auto &cell : cut_cells) {
total_cells += p->chain_size.count(cell->name) ? p->chain_size.at(cell->name) : 1;
}
std::sort(cut_cells.begin(), cut_cells.end(), [&](const CellInfo *a, const CellInfo *b) {
return dir ? (p->cell_locs.at(a->name).rawy < p->cell_locs.at(b->name).rawy)
: (p->cell_locs.at(a->name).rawx < p->cell_locs.at(b->name).rawx);
});
if (cut_cells.size() < 2)
return {};
// Find the cells midpoint, counting chains in terms of their total size - making the initial source cut
int pivot_cells = 0;
int pivot = 0;
for (auto &cell : cut_cells) {
pivot_cells += p->chain_size.count(cell->name) ? p->chain_size.at(cell->name) : 1;
if (pivot_cells >= total_cells / 2)
break;
pivot++;
}
if (pivot >= int(cut_cells.size())) {
pivot = int(cut_cells.size()) - 1;
}
// log_info("orig pivot %d/%d lc %d rc %d\n", pivot, int(cut_cells.size()), pivot_cells, total_cells -
// pivot_cells);
// Find the clearance required either side of the pivot
int clearance_l = 0, clearance_r = 0;
for (size_t i = 0; i < cut_cells.size(); i++) {
int size;
if (cell_extents.count(cut_cells.at(i)->name)) {
auto &ce = cell_extents.at(cut_cells.at(i)->name);
size = dir ? (ce.y1 - ce.y0 + 1) : (ce.x1 - ce.x0 + 1);
} else {
size = 1;
}
if (int(i) < pivot)
clearance_l = std::max(clearance_l, size);
else
clearance_r = std::max(clearance_r, size);
}
// Find the target cut that minimises difference in utilisation, whilst trying to ensure that all chains
// still fit
// First trim the boundaries of the region in the axis-of-interest, skipping any rows/cols without any
// bels of the appropriate type
int trimmed_l = dir ? r.y0 : r.x0, trimmed_r = dir ? r.y1 : r.x1;
while (trimmed_l < (dir ? r.y1 : r.x1)) {
bool have_bels = false;
for (int i = dir ? r.x0 : r.y0; i <= (dir ? r.x1 : r.y1); i++) {
for (size_t t = 0; t < buckets.size(); t++) {
if (bels_at(dir ? i : trimmed_l, dir ? trimmed_l : i, t) > 0) {
have_bels = true;
break;
}
}
}
if (have_bels)
break;
trimmed_l++;
}
while (trimmed_r > (dir ? r.y0 : r.x0)) {
bool have_bels = false;
for (int i = dir ? r.x0 : r.y0; i <= (dir ? r.x1 : r.y1); i++) {
for (size_t t = 0; t < buckets.size(); t++) {
if (bels_at(dir ? i : trimmed_r, dir ? trimmed_r : i, t) > 0) {
have_bels = true;
break;
}
}
}
if (have_bels)
break;
trimmed_r--;
}
// log_info("tl %d tr %d cl %d cr %d\n", trimmed_l, trimmed_r, clearance_l, clearance_r);
if ((trimmed_r - trimmed_l + 1) <= std::max(clearance_l, clearance_r))
return {};
// Now find the initial target cut that minimises utilisation imbalance, whilst
// meeting the clearance requirements for any large macros
std::vector<int> left_cells_v(buckets.size(), 0), right_cells_v(buckets.size(), 0);
std::vector<int> left_bels_v(buckets.size(), 0), right_bels_v(r.bels);
for (int i = 0; i <= pivot; i++)
left_cells_v.at(cell_index(*cut_cells.at(i))) +=
p->chain_size.count(cut_cells.at(i)->name) ? p->chain_size.at(cut_cells.at(i)->name) : 1;
for (int i = pivot + 1; i < int(cut_cells.size()); i++)
right_cells_v.at(cell_index(*cut_cells.at(i))) +=
p->chain_size.count(cut_cells.at(i)->name) ? p->chain_size.at(cut_cells.at(i)->name) : 1;
int best_tgt_cut = -1;
double best_deltaU = std::numeric_limits<double>::max();
// std::pair<int, int> target_cut_bels;
std::vector<int> slither_bels(buckets.size(), 0);
for (int i = trimmed_l; i <= trimmed_r; i++) {
for (size_t t = 0; t < buckets.size(); t++)
slither_bels.at(t) = 0;
for (int j = dir ? r.x0 : r.y0; j <= (dir ? r.x1 : r.y1); j++) {
for (size_t t = 0; t < buckets.size(); t++)
slither_bels.at(t) += dir ? bels_at(j, i, t) : bels_at(i, j, t);
}
for (size_t t = 0; t < buckets.size(); t++) {
left_bels_v.at(t) += slither_bels.at(t);
right_bels_v.at(t) -= slither_bels.at(t);
}
if (((i - trimmed_l) + 1) >= clearance_l && ((trimmed_r - i) + 1) >= clearance_r) {
// Solution is potentially valid
double aU = 0;
for (size_t t = 0; t < buckets.size(); t++)
aU += (left_cells_v.at(t) + right_cells_v.at(t)) *
std::abs(double(left_cells_v.at(t)) / double(std::max(left_bels_v.at(t), 1)) -
double(right_cells_v.at(t)) / double(std::max(right_bels_v.at(t), 1)));
if (aU < best_deltaU) {
best_deltaU = aU;
best_tgt_cut = i;
}
}
}
if (best_tgt_cut == -1)
return {};
// left_bels = target_cut_bels.first;
// right_bels = target_cut_bels.second;
for (size_t t = 0; t < buckets.size(); t++) {
left_bels_v.at(t) = 0;
right_bels_v.at(t) = 0;
}
for (int x = r.x0; x <= (dir ? r.x1 : best_tgt_cut); x++)
for (int y = r.y0; y <= (dir ? best_tgt_cut : r.y1); y++) {
for (size_t t = 0; t < buckets.size(); t++) {
left_bels_v.at(t) += bels_at(x, y, t);
}
}
for (int x = dir ? r.x0 : (best_tgt_cut + 1); x <= r.x1; x++)
for (int y = dir ? (best_tgt_cut + 1) : r.y0; y <= r.y1; y++) {
for (size_t t = 0; t < buckets.size(); t++) {
right_bels_v.at(t) += bels_at(x, y, t);
}
}
if (std::accumulate(left_bels_v.begin(), left_bels_v.end(), 0) == 0 ||
std::accumulate(right_bels_v.begin(), right_bels_v.end(), 0) == 0)
return {};
// Perturb the source cut to eliminate overutilisation
auto is_part_overutil = [&](bool r) {
double delta = 0;
for (size_t t = 0; t < left_cells_v.size(); t++) {
delta += double(left_cells_v.at(t)) / double(std::max(left_bels_v.at(t), 1)) -
double(right_cells_v.at(t)) / double(std::max(right_bels_v.at(t), 1));
}
return r ? delta < 0 : delta > 0;
};
while (pivot > 0 && is_part_overutil(false)) {
auto &move_cell = cut_cells.at(pivot);
int size = p->chain_size.count(move_cell->name) ? p->chain_size.at(move_cell->name) : 1;
left_cells_v.at(cell_index(*cut_cells.at(pivot))) -= size;
right_cells_v.at(cell_index(*cut_cells.at(pivot))) += size;
pivot--;
}
while (pivot < int(cut_cells.size()) - 1 && is_part_overutil(true)) {
auto &move_cell = cut_cells.at(pivot + 1);
int size = p->chain_size.count(move_cell->name) ? p->chain_size.at(move_cell->name) : 1;
left_cells_v.at(cell_index(*cut_cells.at(pivot))) += size;
right_cells_v.at(cell_index(*cut_cells.at(pivot))) -= size;
pivot++;
}
// Split regions into bins, and then spread cells by linear interpolation within those bins
auto spread_binlerp = [&](int cells_start, int cells_end, double area_l, double area_r) {
int N = cells_end - cells_start;
if (N <= 2) {
for (int i = cells_start; i < cells_end; i++) {
auto &pos = dir ? p->cell_locs.at(cut_cells.at(i)->name).rawy
: p->cell_locs.at(cut_cells.at(i)->name).rawx;
pos = area_l + i * ((area_r - area_l) / N);
}
return;
}
// Split region into up to 10 (K) bins
int K = std::min<int>(N, 10);
std::vector<std::pair<int, double>> bin_bounds; // [(cell start, area start)]
bin_bounds.emplace_back(cells_start, area_l);
for (int i = 1; i < K; i++)
bin_bounds.emplace_back(cells_start + (N * i) / K, area_l + ((area_r - area_l + 0.99) * i) / K);
bin_bounds.emplace_back(cells_end, area_r + 0.99);
for (int i = 0; i < K; i++) {
auto &bl = bin_bounds.at(i), br = bin_bounds.at(i + 1);
double orig_left = dir ? p->cell_locs.at(cut_cells.at(bl.first)->name).rawy
: p->cell_locs.at(cut_cells.at(bl.first)->name).rawx;
double orig_right = dir ? p->cell_locs.at(cut_cells.at(br.first - 1)->name).rawy
: p->cell_locs.at(cut_cells.at(br.first - 1)->name).rawx;
double m = (br.second - bl.second) / std::max(0.00001, orig_right - orig_left);
for (int j = bl.first; j < br.first; j++) {
Region *cr = cut_cells.at(j)->region;
if (cr != nullptr) {
// Limit spreading bounds to constraint region; if applicable
double brsc = p->limit_to_reg(cr, br.second, dir);
double blsc = p->limit_to_reg(cr, bl.second, dir);
double mr = (brsc - blsc) / std::max(0.00001, orig_right - orig_left);
auto &pos = dir ? p->cell_locs.at(cut_cells.at(j)->name).rawy
: p->cell_locs.at(cut_cells.at(j)->name).rawx;
NPNR_ASSERT(pos >= orig_left && pos <= orig_right);
pos = blsc + mr * (pos - orig_left);
} else {
auto &pos = dir ? p->cell_locs.at(cut_cells.at(j)->name).rawy
: p->cell_locs.at(cut_cells.at(j)->name).rawx;
NPNR_ASSERT(pos >= orig_left && pos <= orig_right);
pos = bl.second + m * (pos - orig_left);
}
}
}
};
spread_binlerp(0, pivot + 1, trimmed_l, best_tgt_cut);
spread_binlerp(pivot + 1, int(cut_cells.size()), best_tgt_cut + 1, trimmed_r);
// Update various data structures
for (int x = r.x0; x <= r.x1; x++)
for (int y = r.y0; y <= r.y1; y++) {
cells_at_location.at(x).at(y).clear();
}
for (auto cell : cut_cells) {
auto &cl = p->cell_locs.at(cell->name);
cl.x = std::min(r.x1, std::max(r.x0, int(cl.rawx)));
cl.y = std::min(r.y1, std::max(r.y0, int(cl.rawy)));
cells_at_location.at(cl.x).at(cl.y).push_back(cell);
}
SpreaderRegion rl, rr;
rl.id = int(regions.size());
rl.x0 = r.x0;
rl.y0 = r.y0;
rl.x1 = dir ? r.x1 : best_tgt_cut;
rl.y1 = dir ? best_tgt_cut : r.y1;
rl.cells = left_cells_v;
rl.bels = left_bels_v;
rr.id = int(regions.size()) + 1;
rr.x0 = dir ? r.x0 : (best_tgt_cut + 1);
rr.y0 = dir ? (best_tgt_cut + 1) : r.y0;
rr.x1 = r.x1;
rr.y1 = r.y1;
rr.cells = right_cells_v;
rr.bels = right_bels_v;
regions.push_back(rl);
regions.push_back(rr);
for (int x = rl.x0; x <= rl.x1; x++)
for (int y = rl.y0; y <= rl.y1; y++)
groups.at(x).at(y) = rl.id;
for (int x = rr.x0; x <= rr.x1; x++)
for (int y = rr.y0; y <= rr.y1; y++)
groups.at(x).at(y) = rr.id;
return std::make_pair(rl.id, rr.id);
};
};
typedef decltype(CellInfo::udata) cell_udata_t;
cell_udata_t dont_solve = std::numeric_limits<cell_udata_t>::max();
};
int HeAPPlacer::CutSpreader::seq = 0;
bool placer_heap(Context *ctx, PlacerHeapCfg cfg) { return HeAPPlacer(ctx, cfg).place(); }
PlacerHeapCfg::PlacerHeapCfg(Context *ctx)
{
alpha = ctx->setting<float>("placerHeap/alpha");
beta = ctx->setting<float>("placerHeap/beta");
criticalityExponent = ctx->setting<int>("placerHeap/criticalityExponent");
timingWeight = ctx->setting<int>("placerHeap/timingWeight");
parallelRefine = ctx->setting<bool>("placerHeap/parallelRefine", false);
netShareWeight = ctx->setting<float>("placerHeap/netShareWeight", 0);
timing_driven = ctx->setting<bool>("timing_driven");
solverTolerance = 1e-5;
placeAllAtOnce = false;
int timeout_divisor = ctx->setting<int>("placerHeap/cellPlacementTimeout", 8);
if (timeout_divisor > 0) {
// Set a conservative default. This is a rather large number and could probably
// be shaved down, but for now it will keep the process from running indefinite.
cell_placement_timeout = std::max(10000, (int(ctx->cells.size()) * int(ctx->cells.size()) / timeout_divisor));
} else {
cell_placement_timeout = 0;
}
hpwl_scale_x = 1;
hpwl_scale_y = 1;
spread_scale_x = 1;
spread_scale_y = 1;
}
NEXTPNR_NAMESPACE_END