HeAP: Make HeAP placer optional

A CMake option 'BUILD_HEAP' (default on) configures building of the
HeAP placer and the associated Eigen3 dependency.

Default for the iCE40 is SA placer, with --heap-placer to use HeAP

Default for the ECP5 is HeAP placer, as SA placer can take 1hr+ for
large ECP5 designs and HeAP tends to give better QoR. --sa-placer can
be used to use SA instead, and auto-fallback to SA if HeAP not built.

Signed-off-by: David Shah <dave@ds0.me>

6 files changed, 71 insertions, 15 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 69089c4c..ade76d60 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -5,6 +5,8 @@ project(nextpnr)
 option(BUILD_GUI "Build GUI" ON)
 option(BUILD_PYTHON "Build Python Integration" ON)
 option(BUILD_TESTS "Build GUI" OFF)
+option(BUILD_HEAP "Build HeAP analytic placer" ON)
+option(USE_OPENMP "Use OpenMP to accelerate analytic placer" OFF)
 option(COVERAGE "Add code coverage info" OFF)
 option(STATIC_BUILD "Create static build" OFF)
 option(EXTERNAL_CHIPDB "Create build with pre-built chipdb binaries" OFF)
@@ -53,12 +55,16 @@ endforeach()
 
 set(CMAKE_CXX_STANDARD 11)
 if (MSVC)
-set(CMAKE_CONFIGURATION_TYPES "Debug;Release" CACHE STRING "" FORCE)
-set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /D_DEBUG /W4 /wd4100 /wd4244 /wd4125 /wd4800 /wd4456 /wd4458 /wd4305 /wd4459 /wd4121 /wd4996")
-set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /W4 /wd4100 /wd4244 /wd4125 /wd4800 /wd4456 /wd4458 /wd4305 /wd4459 /wd4121 /wd4996 /wd4127")
+    set(CMAKE_CONFIGURATION_TYPES "Debug;Release" CACHE STRING "" FORCE)
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /D_DEBUG /W4 /wd4100 /wd4244 /wd4125 /wd4800 /wd4456 /wd4458 /wd4305 /wd4459 /wd4121 /wd4996")
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /W4 /wd4100 /wd4244 /wd4125 /wd4800 /wd4456 /wd4458 /wd4305 /wd4459 /wd4121 /wd4996 /wd4127")
 else()
-set(CMAKE_CXX_FLAGS_DEBUG "-Wall -fPIC -ggdb -pipe")
-set(CMAKE_CXX_FLAGS_RELEASE "-Wall -fPIC -O3 -g -pipe -fopenmp")
+    set(CMAKE_CXX_FLAGS_DEBUG "-Wall -fPIC -ggdb -pipe")
+    if (USE_OPENMP)
+        set(CMAKE_CXX_FLAGS_RELEASE "-Wall -fPIC -O3 -g -pipe -fopenmp")
+    else()
+        set(CMAKE_CXX_FLAGS_RELEASE "-Wall -fPIC -O3 -g -pipe")
+    endif()
 endif()
 set(CMAKE_DEFIN)
 
@@ -180,10 +186,15 @@ if (BUILD_PYTHON)
     endif ()
 endif()
 
-find_package (Eigen3 REQUIRED NO_MODULE)
+include_directories(common/ json/ ${Boost_INCLUDE_DIRS} ${PYTHON_INCLUDE_DIRS})
+
+if(BUILD_HEAP)
+    find_package (Eigen3 REQUIRED NO_MODULE)
+    include_directories(${EIGEN3_INCLUDE_DIRS})
+    add_definitions(${EIGEN3_DEFINITIONS})
+    add_definitions(-DWITH_HEAP)
+endif()
 
-include_directories(common/ json/ ${Boost_INCLUDE_DIRS} ${PYTHON_INCLUDE_DIRS} ${EIGEN3_INCLUDE_DIRS})
-add_definitions(${EIGEN3_DEFINITIONS})
 aux_source_directory(common/ COMMON_SRC_FILES)
 aux_source_directory(json/ JSON_PARSER_FILES)
 set(COMMON_FILES ${COMMON_SRC_FILES} ${JSON_PARSER_FILES})
diff --git a/common/placer_heap.cc b/common/placer_heap.cc
index b6913473..255a3f54 100644
--- a/common/placer_heap.cc
+++ b/common/placer_heap.cc
@@ -31,6 +31,8 @@
  *   - To make the placer timing-driven, the bound2bound weights are multiplied by (1 + 10 * crit^2)
  */
 
+#ifdef WITH_HEAP
+
 #include <Eigen/Core>
 #include <Eigen/IterativeLinearSolvers>
 #include <boost/optional.hpp>
@@ -1509,5 +1511,19 @@ class HeAPPlacer
 int HeAPPlacer::CutSpreader::seq = 0;
 
 bool placer_heap(Context *ctx) { return HeAPPlacer(ctx).place(); }
+NEXTPNR_NAMESPACE_END
+
+#else
+
+#include "log.h"
+#include "nextpnr.h"
 
+NEXTPNR_NAMESPACE_BEGIN
+bool placer_heap(Context *ctx)
+{
+    log_error("nextpnr was built without the HeAP placer\n");
+    return false;
+}
 NEXTPNR_NAMESPACE_END
+
+#endif
diff --git a/ecp5/arch.cc b/ecp5/arch.cc
index 5ea6a7c3..8385e57b 100644
--- a/ecp5/arch.cc
+++ b/ecp5/arch.cc
@@ -457,6 +457,7 @@ delay_t Arch::estimateDelay(WireId src, WireId dst) const
     auto src_loc = est_location(src), dst_loc = est_location(dst);
 
     int dx = abs(src_loc.first - dst_loc.first), dy = abs(src_loc.second - dst_loc.second);
+
     return (130 - 25 * args.speed) *
            (6 + std::max(dx - 5, 0) + std::max(dy - 5, 0) + 2 * (std::min(dx, 5) + std::min(dy, 5)));
 
@@ -486,6 +487,7 @@ delay_t Arch::predictDelay(const NetInfo *net_info, const PortRef &sink) const
     }
 
     int dx = abs(driver_loc.x - sink_loc.x), dy = abs(driver_loc.y - sink_loc.y);
+
     return (130 - 25 * args.speed) *
            (6 + std::max(dx - 5, 0) + std::max(dy - 5, 0) + 2 * (std::min(dx, 5) + std::min(dy, 5)));
 }
@@ -505,7 +507,22 @@ bool Arch::getBudgetOverride(const NetInfo *net_info, const PortRef &sink, delay
 
 // -----------------------------------------------------------------------
 
-bool Arch::place() { return placer_heap(getCtx()); }
+bool Arch::place()
+{
+    // HeAP is the default unless overriden or not built
+#ifdef WITH_HEAP
+    if (bool_or_default(settings, id("sa_placer"), false)) {
+#endif
+        if (!placer1(getCtx(), Placer1Cfg(getCtx())))
+            return false;
+#ifdef WITH_HEAP
+    } else {
+        if (!placer_heap(getCtx()))
+            return false;
+    }
+#endif
+    return true;
+}
 
 bool Arch::route()
 {
diff --git a/ecp5/main.cc b/ecp5/main.cc
index 15027a5a..de279e63 100644
--- a/ecp5/main.cc
+++ b/ecp5/main.cc
@@ -59,6 +59,8 @@ po::options_description ECP5CommandHandler::getArchOptions()
     specific.add_options()("um5g-45k", "set device type to LFE5UM5G-45F");
     specific.add_options()("um5g-85k", "set device type to LFE5UM5G-85F");
 
+    specific.add_options()("sa-placer", "use pure simulated annealing placer instead of HeAP analytic placer");
+
     specific.add_options()("package", po::value<std::string>(), "select device package (defaults to CABGA381)");
     specific.add_options()("speed", po::value<int>(), "select device speedgrade (6, 7 or 8)");
 
@@ -149,8 +151,12 @@ std::unique_ptr<Context> ECP5CommandHandler::createContext()
             chipArgs.speed = ArchArgs::SPEED_6;
         }
     }
+    auto ctx = std::unique_ptr<Context>(new Context(chipArgs));
+
+    if (vm.count("sa-placer"))
+        ctx->settings[ctx->id("sa_placer")] = "1";
 
-    return std::unique_ptr<Context>(new Context(chipArgs));
+    return ctx;
 }
 
 void ECP5CommandHandler::customAfterLoad(Context *ctx)
diff --git a/ice40/arch.cc b/ice40/arch.cc
index 5688b6e6..09e64b16 100644
--- a/ice40/arch.cc
+++ b/ice40/arch.cc
@@ -671,10 +671,13 @@ bool Arch::getBudgetOverride(const NetInfo *net_info, const PortRef &sink, delay
 
 bool Arch::place()
 {
-    // if (!placer1(getCtx(), Placer1Cfg(getCtx())))
-    //    return false;
-    if (!placer_heap(getCtx()))
-        return false;
+    if (bool_or_default(settings, id("heap_placer"), false)) {
+        if (!placer_heap(getCtx()))
+            return false;
+    } else {
+        if (!placer1(getCtx(), Placer1Cfg(getCtx())))
+            return false;
+    }
     if (bool_or_default(settings, id("opt_timing"), false)) {
         TimingOptCfg tocfg(getCtx());
         tocfg.cellTypes.insert(id_ICESTORM_LC);
diff --git a/ice40/main.cc b/ice40/main.cc
index 2313c2ae..7233f169 100644
--- a/ice40/main.cc
+++ b/ice40/main.cc
@@ -69,6 +69,8 @@ po::options_description Ice40CommandHandler::getArchOptions()
     specific.add_options()("promote-logic",
                            "enable promotion of 'logic' globals (in addition to clk/ce/sr by default)");
     specific.add_options()("no-promote-globals", "disable all global promotion");
+    specific.add_options()("heap-placer",
+                           "use HeAP analytic placer instead of simulated annealing (faster, experimental)");
     specific.add_options()("opt-timing", "run post-placement timing optimisation pass (experimental)");
     specific.add_options()("tmfuzz", "run path delay estimate fuzzer");
     specific.add_options()("pcf-allow-unconstrained", "don't require PCF to constrain all IO");
@@ -176,7 +178,8 @@ std::unique_ptr<Context> Ice40CommandHandler::createContext()
         ctx->settings[ctx->id("opt_timing")] = "1";
     if (vm.count("pcf-allow-unconstrained"))
         ctx->settings[ctx->id("pcf_allow_unconstrained")] = "1";
-
+    if (vm.count("heap-placer"))
+        ctx->settings[ctx->id("heap_placer")] = "1";
     return ctx;
 }