Miscellaneous performance improvements

CMakeLists.txt

@@ -1,10 +1,10 @@
 cmake_minimum_required(VERSION 3.14)
 project(
   Codon
-  VERSION "0.20.2"
+  VERSION "0.20.3"
   HOMEPAGE_URL "https://github.com/exaloop/codon"
   DESCRIPTION "high-performance, extensible Python compiler")
-set(CODON_JIT_PYTHON_VERSION "0.5.2")
+set(CODON_JIT_PYTHON_VERSION "0.5.3")
 configure_file("${PROJECT_SOURCE_DIR}/cmake/config.h.in"
                "${PROJECT_SOURCE_DIR}/codon/config/config.h")
 configure_file("${PROJECT_SOURCE_DIR}/cmake/config.py.in"
@@ -265,6 +265,7 @@ set(CODON_HPPFILES
   codon/cir/transform/parallel/schedule.h
   codon/cir/transform/pass.h
   codon/cir/transform/pythonic/dict.h
+  codon/cir/transform/pythonic/enumerate.h
   codon/cir/transform/pythonic/format.h
   codon/cir/transform/pythonic/generator.h
   codon/cir/transform/pythonic/io.h
@@ -375,6 +376,7 @@ set(CODON_CPPFILES
   codon/cir/transform/parallel/schedule.cpp
   codon/cir/transform/pass.cpp
   codon/cir/transform/pythonic/dict.cpp
+  codon/cir/transform/pythonic/enumerate.cpp
   codon/cir/transform/pythonic/format.cpp
   codon/cir/transform/pythonic/generator.cpp
   codon/cir/transform/pythonic/io.cpp
@@ -537,14 +539,24 @@ set(CODON_TEST_CPPFILES
     test/cir/instr.cpp
     test/cir/llvm/optimize.cpp
     test/cir/module.cpp
+    test/cir/transform/enumerate.cpp
     test/cir/transform/format.cpp
     test/cir/transform/manager.cpp
     test/cir/type.cpp
     test/cir/util/matching.cpp
     test/cir/value.cpp
     test/cir/var.cpp
+    test/dsl/plugins.cpp
     test/types.cpp)
+add_library(codon_test_compiler_plugin SHARED EXCLUDE_FROM_ALL test/dsl/library.cpp)
+target_compile_definitions(codon_test_compiler_plugin PRIVATE CODON_TEST_COMPILER_PLUGIN)
+target_include_directories(codon_test_compiler_plugin PRIVATE
+                "${CMAKE_CURRENT_SOURCE_DIR}" ${LLVM_INCLUDE_DIRS}
+                "${peglib_SOURCE_DIR}")
+target_link_libraries(codon_test_compiler_plugin PRIVATE codonc fmt)
+add_library(codon_test_runtime_plugin SHARED EXCLUDE_FROM_ALL test/dsl/library.cpp)
 add_executable(codon_test ${CODON_TEST_CPPFILES})
+add_dependencies(codon_test codon codon_test_compiler_plugin codon_test_runtime_plugin)
 target_include_directories(codon_test PRIVATE test/cir "${gc_SOURCE_DIR}/include")
 target_link_libraries(codon_test PRIVATE fmt codonc codonrt gtest_main)
 if(ASAN)
@@ -556,7 +568,10 @@ if(ASAN)
                        "-fsanitize-recover=address")
 endif()
 target_compile_definitions(codon_test
-                           PRIVATE TEST_DIR="${CMAKE_CURRENT_SOURCE_DIR}/test")
+                           PRIVATE TEST_DIR="${CMAKE_CURRENT_SOURCE_DIR}/test"
+                           TEST_CODON="$<TARGET_FILE:codon>"
+                           TEST_PLUGIN_COMPILER="$<TARGET_FILE:codon_test_compiler_plugin>"
+                           TEST_PLUGIN_RUNTIME="$<TARGET_FILE:codon_test_runtime_plugin>")
 
 if(APPLE)
   set_target_properties(

codon/cir/llvm/llvisitor.cpp

@@ -535,7 +535,7 @@ void LLVMVisitor::writeToExecutable(const std::string &filename,
 
   if (plugins) {
     for (auto *plugin : *plugins) {
-      auto dylibPath = plugin->info.dylibPath;
+      const auto &dylibPath = plugin->info.getRuntimeDylibPath();
       if (dylibPath.empty())
         continue;
 
@@ -556,7 +556,7 @@ void LLVMVisitor::writeToExecutable(const std::string &filename,
   if (plugins) {
     for (auto *plugin : *plugins) {
       if (plugin->info.linkArgs.empty()) {
-        auto dylibPath = plugin->info.dylibPath;
+        const auto &dylibPath = plugin->info.getRuntimeDylibPath();
         if (dylibPath.empty())
           continue;
 
@@ -1203,6 +1203,10 @@ void LLVMVisitor::run(const std::vector<std::string> &args,
   runLLVMPipeline();
 
   Timer t1("llvm/jitlink");
+  if (plugins) {
+    if (auto error = plugins->loadRuntimeLibraries())
+      compilationError(llvm::toString(std::move(error)));
+  }
   for (auto &lib : libs) {
     std::string err;
     if (llvm::sys::DynamicLibrary::LoadLibraryPermanently(lib.c_str(), &err)) {
@@ -2616,6 +2620,7 @@ void LLVMVisitor::visit(const ForFlow *x) {
 
   auto *condBlock = llvm::BasicBlock::Create(*context, "for.cond", func);
   auto *bodyBlock = llvm::BasicBlock::Create(*context, "for.body", func);
+  auto *checkBlock = llvm::BasicBlock::Create(*context, "for.check", func);
   auto *cleanupBlock = llvm::BasicBlock::Create(*context, "for.cleanup", func);
   auto *exitBlock = llvm::BasicBlock::Create(*context, "for.exit", func);
 
@@ -2633,7 +2638,10 @@ void LLVMVisitor::visit(const ForFlow *x) {
   process(x->getIter());
   auto *iter = value;
   B->SetInsertPoint(block);
-  B->CreateBr(condBlock);
+  B->CreateCondBr(B->CreateCall(coroDone, iter), exitBlock, condBlock);
+
+  B->SetInsertPoint(checkBlock);
+  B->CreateCondBr(B->CreateCall(coroDone, iter), exitBlock, condBlock);
 
   block = condBlock;
   call(coroResume, {iter});
@@ -2653,11 +2661,11 @@ void LLVMVisitor::visit(const ForFlow *x) {
 
   block = bodyBlock;
   enterLoop(
-      {/*breakBlock=*/exitBlock, /*continueBlock=*/condBlock, /*loopId=*/x->getId()});
+      {/*breakBlock=*/exitBlock, /*continueBlock=*/checkBlock, /*loopId=*/x->getId()});
   process(x->getBody());
   exitLoop();
   B->SetInsertPoint(block);
-  B->CreateBr(condBlock);
+  B->CreateBr(checkBlock);
 
   B->SetInsertPoint(cleanupBlock);
   B->CreateCall(coroDestroy, iter);
@@ -3139,6 +3147,7 @@ void LLVMVisitor::codegenPipeline(
 
     auto *condBlock = llvm::BasicBlock::Create(*context, "pipeline.cond", func);
     auto *bodyBlock = llvm::BasicBlock::Create(*context, "pipeline.body", func);
+    auto *checkBlock = llvm::BasicBlock::Create(*context, "pipeline.check", func);
     auto *cleanupBlock = llvm::BasicBlock::Create(*context, "pipeline.cleanup", func);
     auto *exitBlock = llvm::BasicBlock::Create(*context, "pipeline.exit", func);
 
@@ -3155,7 +3164,10 @@ void LLVMVisitor::codegenPipeline(
 
     auto *iter = value;
     B->SetInsertPoint(block);
-    B->CreateBr(condBlock);
+    B->CreateCondBr(B->CreateCall(coroDone, iter), exitBlock, condBlock);
+
+    B->SetInsertPoint(checkBlock);
+    B->CreateCondBr(B->CreateCall(coroDone, iter), exitBlock, condBlock);
 
     block = condBlock;
     call(coroResume, {iter});
@@ -3175,7 +3187,7 @@ void LLVMVisitor::codegenPipeline(
     codegenPipeline(stages, where + 1);
 
     B->SetInsertPoint(block);
-    B->CreateBr(condBlock);
+    B->CreateBr(checkBlock);
 
     B->SetInsertPoint(cleanupBlock);
     B->CreateCall(coroDestroy, iter);

codon/cir/llvm/llvm.h

@@ -18,6 +18,7 @@
 #include "llvm/Analysis/RegionPass.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/AsmParser/Parser.h"
 #include "llvm/Bitcode/BitcodeWriter.h"
 #include "llvm/CodeGen/CommandFlags.h"