Example how to load and call fcn from linked code

apex/apex.cpp

@@ -45,7 +45,7 @@ bool APEXPass::runOnModule(Module &M) {
   // Remove all functions (along with dependencies) that are not in the @path.
   moduleRemoveFunctionsNotInPath(M, apex_dg, path);
 
-  // Swap call to @target with return.
+  // Put exit() call before @target function call.
   logPrintUnderline("Getting/picking callsite for @" + ARG_TARGET_FCN);
   {
     Function *target_fcn = M.getFunction(ARG_TARGET_FCN);
@@ -104,6 +104,23 @@ bool APEXPass::runOnModule(Module &M) {
     }
     logPrint("- call instruction to @" + ARG_TARGET_FCN + " found");
 
+    {
+      logPrint("\nTEST: calling libprint() from lib.c.");
+      // TODO: Test externally linked code.
+      // TODO: Getting: LLVM ERROR: Program used external function
+      //       'libprint' which could not be resolved!
+
+      Constant *_lib_print = M.getOrInsertFunction(
+          "libprint", Type::getVoidTy(M.getContext()), (Type *)0);
+      Function *lib_print_fcn = cast<Function>(_lib_print);
+      logPrint("- libprint() from lib:");
+      lib_print_fcn->dump();
+      Instruction *lib_print_call_inst =
+          CallInst::Create(lib_print_fcn, "", chosen_instruction);
+      logPrintFlat("- call instruction to libprint(): ");
+      lib_print_call_inst->dump();
+    }
+
     logPrint(
         "\nInserting call instruction to exit() before chosen instruction.");
     // Create vector of types and put one integer into this vector.
@@ -141,21 +158,6 @@ bool APEXPass::runOnModule(Module &M) {
     }
     logPrintFlat("- exit call instruction created: ");
     exit_call_inst->dump();
-
-    // TODO: Test externally linked code.
-    // TODO: Getting: LLVM ERROR: Program used external function
-    //       'libprint' which could not be resolved!
-
-    logPrint("\nWIP: Testing lib loadout.");
-    Constant *_lib_print = M.getOrInsertFunction(
-        "libprint", Type::getVoidTy(M.getContext()), (Type *)0);
-    Function *lib_print_fcn = cast<Function>(_lib_print);
-    logPrint("- libprint() from lib:");
-    lib_print_fcn->dump();
-    Instruction *lib_print_call_inst =
-        CallInst::Create(lib_print_fcn, "", chosen_instruction);
-    logPrintFlat("- call instruction to libprint(): ");
-    lib_print_call_inst->dump();
   }
 
   logPrintUnderline("\nFinal Module Dump:");
@@ -1008,6 +1010,7 @@ void APEXPass::apexDgFindDataDependencies(
 
 /// Goes over all functions in module. If there is some function in module
 /// that is not in the @path, put it for removal along it all its dependencies.
+/// Note: Functions that are in @PROTECTED_FCNS, will NOT be removed.
 void APEXPass::moduleRemoveFunctionsNotInPath(Module &M,
                                               APEXDependencyGraph &apex_dg,
                                               std::vector<Function *> &path) {
@@ -1025,15 +1028,29 @@ void APEXPass::moduleRemoveFunctionsNotInPath(Module &M,
           module_fcn_to_be_removed = false;
         }
       }
+      // @module_fcn is inside @PROTECTED_FCNS, do not remove it.
+      // Inside @PROTECTED_FCNS could be some function from external library,
+      // or debug functions or whatever that we do not want to optimize out.
+      for (std::string &protected_fcn_id : PROTECTED_FCNS) {
+        if (protected_fcn_id == module_fcn.getGlobalIdentifier()) {
+          module_fcn_to_be_removed = false;
+        }
+      }
+
       if (true == module_fcn_to_be_removed) {
         functions_to_be_removed.push_back(&module_fcn);
       }
     }
 
     logPrintFlat("- path: ");
     functionVectorFlatPrint(path);
-    logPrintFlat("- functions_to_be_removed: ");
+    logPrintFlat("- functions to be removed: ");
     functionVectorFlatPrint(functions_to_be_removed);
+    logPrintFlat("- protected functions: ");
+    for (std::string &protected_fcn_id : PROTECTED_FCNS) {
+      logPrintFlat(protected_fcn_id + ", ");
+    }
+    logPrint("");
   }
 
   logPrint("\nCollecting instruction dependencies for each function that has "

apex/apex.h

@@ -19,7 +19,6 @@
 #include <llvm/Analysis/CallGraph.h>
 #include <llvm/IR/Instructions.h>
 #include <llvm/IR/LegacyPassManager.h>
-
 #include <llvm/Support/CommandLine.h>
 #include <llvm/Transforms/IPO/PassManagerBuilder.h>
 
@@ -31,6 +30,9 @@
 using namespace llvm;
 using namespace dg;
 
+/// Protected functions IDs. These will not be removed by APEXPass.
+std::vector<std::string> PROTECTED_FCNS = {"libprint", "printf"};
+
 /// Command line arguments for opt.
 cl::opt<std::string> ARG_SOURCE_FCN("source",
                                     cl::desc("Specify source function."),

build_and_run.sh

@@ -19,11 +19,11 @@ INPUT_LIB="lib.c"
 
 # Generated bytecodes during all stages of processing.
 BYTECODE_FROM_INPUT="bytecode_from_input.bc" # No optimizations.
-BYTECODE_FROM_INPUT_BASIC_OPTS="bytecode_from_input_basic_opts.bc" # Some basic optimizations (see BASIC_OPTS).
-BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX="bytecode_from_input_basic_opts_and_apex.bc" # BASIC_OPTS + APEXPass.
-BYTECODE_FROM_LIB="bytecode_from_lib.bc"
-BYTECODE_LINKED="bytecode_linked.bc"
-LINKED="linked.ll"
+#BYTECODE_FROM_INPUT_BASIC_OPTS="bytecode_from_input_basic_opts.bc" # Some basic optimizations (see BASIC_OPTS).
+BYTECODE_FROM_LIB="bytecode_from_lib.bc" # Compiled libraries from ${INPUT_LIB}
+BYTECODE_FROM_LINKED_INPUT_LIB="bytecode_from_linked.bc" # Linked input and lib bytecode. APEXPass will work with this.
+LINKED_DISASSEMBLY="bytecode_from_linked.ll"
+BYTECODE_FROM_APEX="bytecode_from_apex.bc" # APEXPass bytecode.
 
 # Optimizations that are ran before APEXPass
 BASIC_OPTS="-dce -loop-simplify -simplifycfg"
@@ -74,17 +74,19 @@ clang -O0 -c -emit-llvm c-code/${INPUT_LIB} -o build/${BYTECODE_FROM_LIB}
 # compile without anything
 clang -O0 -c -emit-llvm c-code/${INPUT_SOURCE} -o build/${BYTECODE_FROM_INPUT}
 
-llvm-link build/${BYTECODE_FROM_LIB} build/${BYTECODE_FROM_INPUT} -S -o=build/${LINKED}
-llvm-as build/${LINKED} -o build/${BYTECODE_LINKED}
+llvm-link build/${BYTECODE_FROM_LIB} build/${BYTECODE_FROM_INPUT} -S -o=build/${LINKED_DISASSEMBLY}
+llvm-as build/${LINKED_DISASSEMBLY} -o build/${BYTECODE_FROM_LINKED_INPUT_LIB}
 
 echo "========= Running APEXPass."
 
 # run some basic optimizations, so the IR is cleaner
 #opt -o build/${BYTECODE_FROM_INPUT_BASIC_OPTS} ${BASIC_OPTS} build/${BYTECODE_FROM_INPUT} > /dev/null
 
-# run our pass
-opt -o build/${BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX} -load build/apex/libAPEXPass.so -apex -source=${SOURCE_FCN} -target=${TARGET_FCN} < build/${BYTECODE_LINKED}  > /dev/null
-#opt -o build/${BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX} -load build/apex/libAPEXPass.so -apex -source=${SOURCE_FCN} -target=${TARGET_FCN} < build/${BYTECODE_FROM_INPUT_BASIC_OPTS}  > /dev/null
+# run APEXPass without additional opts
+opt -o build/${BYTECODE_FROM_APEX} -load build/apex/libAPEXPass.so -apex -source=${SOURCE_FCN} -target=${TARGET_FCN} < build/${BYTECODE_FROM_LINKED_INPUT_LIB}  > /dev/null
+# run APEXPass with ${BASIC_OPTS}
+#opt -o build/${BYTECODE_FROM_APEX} -load build/apex/libAPEXPass.so -apex ${BASIC_OPTS} -source=${SOURCE_FCN} -target=${TARGET_FCN} < build/${BYTECODE_FROM_LINKED_INPUT_LIB}  > /dev/null
+
 
 # ==============================================================================
 # Generate callgraphs before and after =========================================
@@ -105,7 +107,7 @@ dot callgraph_no_opt.dot -Tsvg -O
 #mv callgraph.dot callgraph_default_opt.dot
 #dot callgraph_default_opt.dot -Tsvg -O
 
-opt -dot-callgraph ../${BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX} > /dev/null
+opt -dot-callgraph ../${BYTECODE_FROM_APEX} > /dev/null
 mv callgraph.dot callgraph_apex_opt.dot
 dot callgraph_apex_opt.dot -Tsvg -O
 
@@ -151,7 +153,7 @@ echo "========= Disassembling llvm bytecode."
 
 llvm-dis build/${BYTECODE_FROM_INPUT}
 #llvm-dis build/${BYTECODE_FROM_INPUT_BASIC_OPTS}
-llvm-dis build/${BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX}
+llvm-dis build/${BYTECODE_FROM_APEX}
 llvm-dis build/${BYTECODE_FROM_LIB}
 
 
@@ -167,5 +169,5 @@ lli build/${BYTECODE_FROM_INPUT}
 #lli build/${BYTECODE_FROM_INPUT_BASIC_OPTS}
 
 # run pass and execute modified bytecode
-echo "========= Running ${BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX}"
-lli build/${BYTECODE_FROM_INPUT_BASIC_OPTS_AND_APEX}
+echo "========= Running ${BYTECODE_FROM_APEX}"
+lli build/${BYTECODE_FROM_APEX}