Merge pull request #447 from whitequark/wasi

Port nextpnr-{ice40,ecp5} to WASI

CMakeLists.txt

@@ -14,6 +14,24 @@ option(SERIALIZE_CHIPDB "Never build chipdb in parallel to reduce peak memory us
 
 set(Boost_NO_BOOST_CMAKE ON)
 
+if(WASI)
+    set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -lwasi-emulated-mman")
+    set(USE_THREADS OFF)
+    add_definitions(
+        -DBOOST_EXCEPTION_DISABLE
+        -DBOOST_NO_EXCEPTIONS
+        -DBOOST_SP_NO_ATOMIC_ACCESS
+        -DBOOST_AC_DISABLE_THREADS
+        -DBOOST_NO_CXX11_HDR_MUTEX
+    )
+else()
+    set(USE_THREADS ON)
+endif()
+
+if (NOT USE_THREADS)
+    add_definitions(-DNPNR_DISABLE_THREADS)
+endif()
+
 set(link_param "")
 if (STATIC_BUILD)
     set(Boost_USE_STATIC_LIBS   ON)
@@ -86,7 +104,6 @@ else()
         set(CMAKE_CXX_FLAGS_RELEASE "-Wall -fPIC -O3 -g -pipe")
     endif()
 endif()
-set(CMAKE_DEFIN)
 
 set(CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/3rdparty/sanitizers-cmake/cmake;." ${CMAKE_MODULE_PATH})
 
@@ -101,7 +118,10 @@ endif()
 find_package(Sanitizers)
 
 # List of Boost libraries to include
-set(boost_libs filesystem thread program_options iostreams system)
+set(boost_libs filesystem program_options iostreams system)
+if (USE_THREADS)
+    list(APPEND boost_libs thread)
+endif()
 
 if (BUILD_GUI AND NOT BUILD_PYTHON)
     message(FATAL_ERROR "GUI requires Python to build")
@@ -231,6 +251,10 @@ foreach (family ${ARCH})
 
     # Add the CLI binary target
     add_executable(${PROGRAM_PREFIX}nextpnr-${family} ${COMMON_FILES} ${${ufamily}_FILES})
+    if (WASI)
+        # set(CMAKE_EXECUTABLE_SUFFIX) breaks CMake tests for some reason
+        set_property(TARGET ${PROGRAM_PREFIX}nextpnr-${family} PROPERTY SUFFIX ".wasm")
+    endif()
     install(TARGETS ${PROGRAM_PREFIX}nextpnr-${family} RUNTIME DESTINATION bin)
     target_compile_definitions(${PROGRAM_PREFIX}nextpnr-${family} PRIVATE MAIN_EXECUTABLE)
 

common/nextpnr.cc

@@ -23,6 +23,25 @@
 #include "log.h"
 #include "util.h"
 
+#if defined(__wasm)
+extern "C" {
+    // FIXME: WASI does not currently support exceptions.
+    void* __cxa_allocate_exception(size_t thrown_size) throw() {
+        return malloc(thrown_size);
+    }
+    bool __cxa_uncaught_exception() throw();
+    void __cxa_throw(void* thrown_exception, struct std::type_info * tinfo, void (*dest)(void*)) {
+        std::terminate();
+    }
+}
+
+namespace boost {
+    void throw_exception( std::exception const & e ) {
+        NEXTPNR_NAMESPACE::log_error("boost::exception(): %s\n", e.what());
+    }
+}
+#endif
+
 NEXTPNR_NAMESPACE_BEGIN
 
 assertion_failure::assertion_failure(std::string msg, std::string expr_str, std::string filename, int line)

common/nextpnr.h

@@ -33,7 +33,9 @@
 #include <boost/functional/hash.hpp>
 #include <boost/lexical_cast.hpp>
 #include <boost/range/adaptor/reversed.hpp>
+#ifndef NPNR_DISABLE_THREADS
 #include <boost/thread.hpp>
+#endif
 
 #ifndef NEXTPNR_H
 #define NEXTPNR_H
@@ -647,6 +649,7 @@ struct DeterministicRNG
 
 struct BaseCtx
 {
+#ifndef NPNR_DISABLE_THREADS
     // Lock to perform mutating actions on the Context.
     std::mutex mutex;
     boost::thread::id mutex_owner;
@@ -655,6 +658,7 @@ struct BaseCtx
     // method will lock/unlock it when its' released the main mutex to make
     // sure the UI is not starved.
     std::mutex ui_mutex;
+#endif
 
     // ID String database.
     mutable std::unordered_map<std::string, int> *idstring_str_to_idx;
@@ -706,28 +710,36 @@ struct BaseCtx
     // Must be called before performing any mutating changes on the Ctx/Arch.
     void lock(void)
     {
+#ifndef NPNR_DISABLE_THREADS
         mutex.lock();
         mutex_owner = boost::this_thread::get_id();
+#endif
     }
 
     void unlock(void)
     {
+#ifndef NPNR_DISABLE_THREADS
         NPNR_ASSERT(boost::this_thread::get_id() == mutex_owner);
         mutex.unlock();
+#endif
     }
 
     // Must be called by the UI before rendering data. This lock will be
     // prioritized when processing code calls yield().
     void lock_ui(void)
     {
+#ifndef NPNR_DISABLE_THREADS
         ui_mutex.lock();
         mutex.lock();
+#endif
     }
 
     void unlock_ui(void)
     {
+#ifndef NPNR_DISABLE_THREADS
         mutex.unlock();
         ui_mutex.unlock();
+#endif
     }
 
     // Yield to UI by unlocking the main mutex, flashing the UI mutex and
@@ -737,10 +749,12 @@ struct BaseCtx
     // Must be called with the main lock taken.
     void yield(void)
     {
+#ifndef NPNR_DISABLE_THREADS
         unlock();
         ui_mutex.lock();
         ui_mutex.unlock();
         lock();
+#endif
     }
 
     IdString id(const std::string &s) const { return IdString(this, s); }

common/placer_heap.cc

@@ -37,7 +37,6 @@
 #include <Eigen/Core>
 #include <Eigen/IterativeLinearSolvers>
 #include <boost/optional.hpp>
-#include <boost/thread.hpp>
 #include <chrono>
 #include <deque>
 #include <fstream>
@@ -154,9 +153,14 @@ class HeAPPlacer
         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();
 
@@ -211,13 +215,16 @@ class HeAPPlacer
                 // Heuristic: don't bother with threading below a certain size
                 auto solve_startt = std::chrono::high_resolution_clock::now();
 
-                if (solve_cells.size() < 500) {
+#ifndef NPNR_DISABLE_THREADS
-                    build_solve_direction(false, (iter == 0) ? -1 : iter);
+                if (solve_cells.size() >= 500) {
-                    build_solve_direction(true, (iter == 0) ? -1 : iter);
-                } else {
                     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();

common/router2.cc

@@ -33,7 +33,6 @@
 #include <deque>
 #include <fstream>
 #include <queue>
-#include <thread>
 #include "log.h"
 #include "nextpnr.h"
 #include "router1.h"
@@ -985,8 +984,22 @@ struct Router2
         }
         if (ctx->verbose)
             log_info("%d/%d nets not multi-threadable\n", int(tcs.at(N).route_nets.size()), int(route_queue.size()));
+#ifdef NPNR_DISABLE_THREADS
+        // Singlethreaded routing - quadrants
+        for (int i = 0; i < Nq; i++) {
+            router_thread(tcs.at(i));
+        }
+        // Vertical splits
+        for (int i = Nq; i < Nq + Nv; i++) {
+            router_thread(tcs.at(i));
+        }
+        // Horizontal splits
+        for (int i = Nq + Nv; i < Nq + Nv + Nh; i++) {
+            router_thread(tcs.at(i));
+        }
+#else
         // Multithreaded part of routing - quadrants
-        std::vector<std::thread> threads;
+        std::vector<boost::thread> threads;
         for (int i = 0; i < Nq; i++) {
             threads.emplace_back([this, &tcs, i]() { router_thread(tcs.at(i)); });
         }
@@ -1007,6 +1020,7 @@ struct Router2
         for (auto &t : threads)
             t.join();
         threads.clear();
+#endif
         // Singlethreaded part of routing - nets that cross partitions
         // or don't fit within bounding box
         for (auto st_net : tcs.at(N).route_nets)
@@ -1130,4 +1144,4 @@ Router2Cfg::Router2Cfg(Context *ctx)
     perf_profile = ctx->setting<float>("router2/perfProfile", false);
 }
 
-NEXTPNR_NAMESPACE_END
+NEXTPNR_NAMESPACE_END

ecp5/arch.cc

@@ -71,12 +71,13 @@ const char *chipdb_blob_25k = nullptr;
 const char *chipdb_blob_45k = nullptr;
 const char *chipdb_blob_85k = nullptr;
 
-boost::iostreams::mapped_file_source blob_files[3];
+boost::iostreams::mapped_file blob_files[3];
 
 const char *mmap_file(int index, const char *filename)
 {
     try {
-        blob_files[index].open(filename);
+        // WASI only supports MAP_PRIVATE
+        blob_files[index].open(filename, boost::iostreams::mapped_file::priv);
         if (!blob_files[index].is_open())
             log_error("Unable to read chipdb %s\n", filename);
         return (const char *)blob_files[index].data();

ecp5/family.cmake

@@ -70,8 +70,9 @@ if (NOT EXTERNAL_CHIPDB)
         target_compile_options(ecp5_chipdb PRIVATE -g0 -O0 -w)
         set(PREV_DEV_CC_BBA_DB)
         foreach (dev ${devices})
-            set(DEV_CC_DB ${CMAKE_CURRENT_BINARY_DIR}/ecp5/chipdbs/chipdb-${dev}.cc)
             set(DEV_CC_BBA_DB ${CMAKE_CURRENT_BINARY_DIR}/ecp5/chipdbs/chipdb-${dev}.bba)
+            set(DEV_CC_DB ${CMAKE_CURRENT_BINARY_DIR}/ecp5/chipdbs/chipdb-${dev}.cc)
+            set(DEV_BIN_DB ${CMAKE_CURRENT_BINARY_DIR}/ecp5/chipdbs/chipdb-${dev}.bin)
             set(DEV_CONSTIDS_INC ${CMAKE_CURRENT_SOURCE_DIR}/ecp5/constids.inc)
             set(DEV_GFXH ${CMAKE_CURRENT_SOURCE_DIR}/ecp5/gfx.h)
             if (PREGENERATED_BBA_PATH)
@@ -85,11 +86,19 @@ if (NOT EXTERNAL_CHIPDB)
                     COMMAND mv ${DEV_CC_BBA_DB}.new ${DEV_CC_BBA_DB}
                     DEPENDS ${DB_PY} ${DEV_CONSTIDS_INC} ${DEV_GFXH} ${PREV_DEV_CC_BBA_DB}
                     )
-                add_custom_command(OUTPUT ${DEV_CC_DB}
+                if(USE_C_EMBED)
-                    COMMAND bbasm --c ${BBASM_ENDIAN_FLAG} ${DEV_CC_BBA_DB} ${DEV_CC_DB}.new
+                    add_custom_command(OUTPUT ${DEV_CC_DB}
-                    COMMAND mv ${DEV_CC_DB}.new ${DEV_CC_DB}
+                        COMMAND bbasm --e ${BBASM_ENDIAN_FLAG} ${DEV_CC_BBA_DB} ${DEV_CC_DB}.new ${DEV_BIN_DB}
-                    DEPENDS bbasm ${DEV_CC_BBA_DB}
+                        COMMAND mv ${DEV_CC_DB}.new ${DEV_CC_DB}
-                    )
+                        DEPENDS bbasm ${DEV_CC_BBA_DB}
+                        )
+                else()
+                    add_custom_command(OUTPUT ${DEV_CC_DB}
+                        COMMAND bbasm --c ${BBASM_ENDIAN_FLAG} ${DEV_CC_BBA_DB} ${DEV_CC_DB}.new
+                        COMMAND mv ${DEV_CC_DB}.new ${DEV_CC_DB}
+                        DEPENDS bbasm ${DEV_CC_BBA_DB}
+                        )
+                endif()
             endif()
             if (SERIALIZE_CHIPDB)
                 set(PREV_DEV_CC_BBA_DB ${DEV_CC_BBA_DB})

ice40/arch.cc

@@ -57,12 +57,12 @@ const char *chipdb_blob_5k = nullptr;
 const char *chipdb_blob_u4k = nullptr;
 const char *chipdb_blob_8k = nullptr;
 
-boost::iostreams::mapped_file_source blob_files[5];
+boost::iostreams::mapped_file blob_files[5];
 
 const char *mmap_file(int index, const char *filename)
 {
     try {
-        blob_files[index].open(filename);
+        blob_files[index].open(filename, boost::iostreams::mapped_file::priv);
         if (!blob_files[index].is_open())
             log_error("Unable to read chipdb %s\n", filename);
         return (const char *)blob_files[index].data();