diff --git a/Makefile b/Makefile
index da4c5e9519..a6154561d6 100644
--- a/Makefile
+++ b/Makefile
@@ -80,6 +80,8 @@ endif
 
 all:	test_x86 test_shm test_python27 ready $(PROGS) afl-as test_build all_done
 
+tests:	source-only
+	@cd test ; ./test.sh
 
 help:
 	@echo "HELP --- the following make targets exist:"
@@ -90,6 +92,8 @@ help:
 	@echo "distrib: everything (for both binary-only and source code fuzzing)"
 	@echo "install: installs everything you have compiled with the build option above"
 	@echo "clean: cleans everything. for qemu_mode and unicorn_mode it means it deletes all downloads as well"
+	@echo "tests: this runs the test framework. It is more catered for the developers, but if you run into problems this helps pinpointing the problem"
+	@echo "document: creates afl-fuzz-document which will only do one run and save all manipulated inputs into out/queue/mutations"
 	@echo "help: shows these build options :-)"
 	@echo "=========================================="
 	@echo "Recommended: \"distrib\" or \"source-only\", then \"install\""
@@ -149,31 +153,36 @@ afl-as: src/afl-as.c include/afl-as.h $(COMM_HDR) | test_x86
 	$(CC) $(CFLAGS) src/$@.c -o $@ $(LDFLAGS)
 	ln -sf afl-as as
 
-afl-common.o : src/afl-common.c include/common.h
-	$(CC) $(CFLAGS) -c src/afl-common.c
+src/afl-common.o : src/afl-common.c include/common.h
+	$(CC) $(CFLAGS) -c src/afl-common.c -o src/afl-common.o
 
-afl-forkserver.o : src/afl-forkserver.c include/forkserver.h
-	$(CC) $(CFLAGS) -c src/afl-forkserver.c
+src/afl-forkserver.o : src/afl-forkserver.c include/forkserver.h
+	$(CC) $(CFLAGS) -c src/afl-forkserver.c -o src/afl-forkserver.o
 
-afl-sharedmem.o : src/afl-sharedmem.c include/sharedmem.h
-	$(CC) $(CFLAGS) -c src/afl-sharedmem.c
+src/afl-sharedmem.o : src/afl-sharedmem.c include/sharedmem.h
+	$(CC) $(CFLAGS) -c src/afl-sharedmem.c -o src/afl-sharedmem.o
 
-afl-fuzz: include/afl-fuzz.h $(AFL_FUZZ_FILES) afl-common.o afl-sharedmem.o afl-forkserver.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) $(AFL_FUZZ_FILES) afl-common.o afl-sharedmem.o afl-forkserver.o -o $@ $(LDFLAGS) $(PYFLAGS)
+afl-fuzz: include/afl-fuzz.h $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o $(COMM_HDR) | test_x86
+	$(CC) $(CFLAGS) $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o -o $@ $(LDFLAGS) $(PYFLAGS)
 
-afl-showmap: src/afl-showmap.c afl-common.o afl-sharedmem.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) src/$@.c afl-common.o afl-sharedmem.o -o $@ $(LDFLAGS)
+afl-showmap: src/afl-showmap.c src/afl-common.o src/afl-sharedmem.o $(COMM_HDR) | test_x86
+	$(CC) $(CFLAGS) src/$@.c src/afl-common.o src/afl-sharedmem.o -o $@ $(LDFLAGS)
 
-afl-tmin: src/afl-tmin.c afl-common.o afl-sharedmem.o afl-forkserver.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) src/$@.c afl-common.o afl-sharedmem.o afl-forkserver.o -o $@ $(LDFLAGS)
+afl-tmin: src/afl-tmin.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o $(COMM_HDR) | test_x86
+	$(CC) $(CFLAGS) src/$@.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o -o $@ $(LDFLAGS)
 
-afl-analyze: src/afl-analyze.c afl-common.o afl-sharedmem.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) src/$@.c afl-common.o afl-sharedmem.o -o $@ $(LDFLAGS)
+afl-analyze: src/afl-analyze.c src/afl-common.o src/afl-sharedmem.o $(COMM_HDR) | test_x86
+	$(CC) $(CFLAGS) src/$@.c src/afl-common.o src/afl-sharedmem.o -o $@ $(LDFLAGS)
 
 afl-gotcpu: src/afl-gotcpu.c $(COMM_HDR) | test_x86
 	$(CC) $(CFLAGS) src/$@.c -o $@ $(LDFLAGS)
 
 
+# document all mutations and only do one run (use with only one input file!)
+document: include/afl-fuzz.h $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o $(COMM_HDR) | test_x86
+	$(CC) $(CFLAGS) $(AFL_FUZZ_FILES) -D_AFL_DOCUMENT_MUTATIONS src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o -o afl-fuzz-document $(LDFLAGS) $(PYFLAGS)
+
+
 code-format:
 	./.custom-format.py -i src/*.c
 	./.custom-format.py -i include/*.h
@@ -219,7 +228,7 @@ all_done: test_build
 .NOTPARALLEL: clean
 
 clean:
-	rm -f $(PROGS) afl-as as afl-g++ afl-clang afl-clang++ *.o *~ a.out core core.[1-9][0-9]* *.stackdump test .test .test1 .test2 test-instr .test-instr0 .test-instr1 qemu_mode/qemu-3.1.1.tar.xz afl-qemu-trace afl-gcc-fast afl-gcc-pass.so afl-gcc-rt.o afl-g++-fast *.so unicorn_mode/24f55a7973278f20f0de21b904851d99d4716263.tar.gz *.8
+	rm -f $(PROGS) afl-as as afl-g++ afl-clang afl-clang++ *.o src/*.o *~ a.out core core.[1-9][0-9]* *.stackdump test .test .test1 .test2 test-instr .test-instr0 .test-instr1 qemu_mode/qemu-3.1.1.tar.xz afl-qemu-trace afl-gcc-fast afl-gcc-pass.so afl-gcc-rt.o afl-g++-fast *.so unicorn_mode/24f55a7973278f20f0de21b904851d99d4716263.tar.gz *.8
 	rm -rf out_dir qemu_mode/qemu-3.1.1 unicorn_mode/unicorn
 	$(MAKE) -C llvm_mode clean
 	$(MAKE) -C libdislocator clean
@@ -280,6 +289,8 @@ endif
 	if [ -f compare-transform-pass.so ]; then set -e; install -m 755 compare-transform-pass.so $${DESTDIR}$(HELPER_PATH); fi
 	if [ -f split-compares-pass.so ]; then set -e; install -m 755 split-compares-pass.so $${DESTDIR}$(HELPER_PATH); fi
 	if [ -f split-switches-pass.so ]; then set -e; install -m 755 split-switches-pass.so $${DESTDIR}$(HELPER_PATH); fi
+	if [ -f libdislocator.so ]; then set -e; install -m 755 libdislocator.so $${DESTDIR}$(HELPER_PATH); fi
+	if [ -f libtokencap.so ]; then set -e; install -m 755 libtokencap.so $${DESTDIR}$(HELPER_PATH); fi
 	if [ -f libcompcov.so ]; then set -e; install -m 755 libcompcov.so $${DESTDIR}$(HELPER_PATH); fi
 
 	set -e; ln -sf afl-gcc $${DESTDIR}$(BIN_PATH)/afl-g++
@@ -294,7 +305,7 @@ endif
 	cp -r testcases/ $${DESTDIR}$(MISC_PATH)
 	cp -r dictionaries/ $${DESTDIR}$(MISC_PATH)
 
-publish: clean
+#publish: clean
 #	test "`basename $$PWD`" = "afl" || exit 1
 #	test -f ~/www/afl/releases/$(PROGNAME)-$(VERSION).tgz; if [ "$$?" = "0" ]; then echo; echo "Change program version in config.h, mmkay?"; echo; exit 1; fi
 #	cd ..; rm -rf $(PROGNAME)-$(VERSION); cp -pr $(PROGNAME) $(PROGNAME)-$(VERSION); \
diff --git a/TODO b/TODO
index 11dc523afa..4aa3dc8b5b 100644
--- a/TODO
+++ b/TODO
@@ -1,14 +1,3 @@
-Roadmap 2.54d:
-==============
-
-afl-fuzz:
- - enable python mutator for MOpt
- - enable custom mutator for MOpt
- - add superion?
-
-remote feature
-
-
 
 Roadmap 2.55d:
 ==============
@@ -26,6 +15,9 @@ qemu_mode:
    Idea: The static analyzer outputs a map in which each edge that must be
    skipped is marked with 1. QEMU loads it at startup in the parent process.
 
+custom_mutators:
+ - rip what Superion is doing into custom mutators for js, php, etc.
+
 unit testing / or large testcase campaign
 
 
diff --git a/docs/ChangeLog b/docs/ChangeLog
index 4ea4e4b23b..0f5bb99c21 100644
--- a/docs/ChangeLog
+++ b/docs/ChangeLog
@@ -22,6 +22,15 @@ Version ++2.54d (dev):
     add AFL_CUSTOM_MUTATOR_ONLY (that will trigger the previous behaviour)
   - no more unlinking the input file, this way the input file can also be a
     FIFO or disk partition
+  - setting LLVM_CONFIG for llvm_mode will now again switch to the selected
+    llvm version. If you setup is correct.
+  - fuzzing strategy yields for custom mutator were missing from the UI, added them :)
+  - added "make tests" which will perform checks to see that all functionality
+    is working as expected. this is currently the starting point, its not complete :)
+  - added mutation documentation feature ("make document"), creates afl-fuzz-document
+    and saves all mutations of the first run on the first file into out/queue/mutations
+  - libtokencap and libdislocator now compile to the afl_root directory and are
+    installed to the .../lib/afl directory when present during make install
   - reducing duplicate code in afl-fuzz
   - added "make help"
   - removed compile warnings from python internal stuff
diff --git a/docs/env_variables.txt b/docs/env_variables.txt
index d568893993..e763118e3d 100644
--- a/docs/env_variables.txt
+++ b/docs/env_variables.txt
@@ -97,7 +97,8 @@ Then there are a few specific features that are only available in llvm_mode:
 
     - Setting AFL_LLVM_LAF_TRANSFORM_COMPARES will split string compare functions
 
-    - Setting AFL_LLVM_LAF_SPLIT_COMPARES will split > 8 bit CMP instructions
+    - Setting AFL_LLVM_LAF_SPLIT_COMPARES will split all floating point and
+      64, 32 and 16 bit integer CMP instructions
 
     See llvm_mode/README.laf-intel for more information. 
 
diff --git a/experimental/persistent_demo/persistent_demo.c b/experimental/persistent_demo/persistent_demo.c
index 0b1e29897b..d091febed2 100644
--- a/experimental/persistent_demo/persistent_demo.c
+++ b/experimental/persistent_demo/persistent_demo.c
@@ -33,6 +33,7 @@
 
 int main(int argc, char** argv) {
 
+  ssize_t len;   /* how much input did we read? */
   char buf[100]; /* Example-only buffer, you'd replace it with other global or
                     local variables appropriate for your use case. */
 
@@ -57,11 +58,15 @@ int main(int argc, char** argv) {
                Beware of reading from buffered FILE* objects such as stdin. Use
                raw file descriptors or call fopen() / fdopen() in every pass. */
 
-    read(0, buf, 100);
+    len = read(0, buf, 100);
 
     /* STEP 3: This is where we'd call the tested library on the read data.
                We just have some trivial inline code that faults on 'foo!'. */
 
+    /* do we have enough data? */
+    if (len < 4)
+      return 0;
+
     if (buf[0] == 'f') {
       printf("one\n");
       if (buf[1] == 'o') {
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 9536e06a70..4912b3f016 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -682,5 +682,10 @@ static u64 get_cur_time_us(void) {
 
 }
 
+#ifdef _AFL_DOCUMENT_MUTATIONS
+  extern u8 do_document;
+  extern u32 document_counter;
+#endif
+
 #endif
 
diff --git a/libdislocator/Makefile b/libdislocator/Makefile
index 236667ec29..cbaa05ea5e 100644
--- a/libdislocator/Makefile
+++ b/libdislocator/Makefile
@@ -24,15 +24,15 @@ CFLAGS      += -Wall -D_FORTIFY_SOURCE=2 -g -Wno-pointer-sign
 all: libdislocator.so
 
 libdislocator.so: libdislocator.so.c ../config.h
-	$(CC) $(CFLAGS) -shared -fPIC $< -o $@ $(LDFLAGS)
+	$(CC) $(CFLAGS) -shared -fPIC $< -o ../$@ $(LDFLAGS)
 
 .NOTPARALLEL: clean
 
 clean:
 	rm -f *.o *.so *~ a.out core core.[1-9][0-9]*
-	rm -f libdislocator.so
+	rm -f ../libdislocator.so
 
 install: all
-	install -m 755 libdislocator.so $${DESTDIR}$(HELPER_PATH)
+	install -m 755 ../libdislocator.so $${DESTDIR}$(HELPER_PATH)
 	install -m 644 README.dislocator $${DESTDIR}$(HELPER_PATH)
 
diff --git a/libtokencap/Makefile b/libtokencap/Makefile
index ec4c8f95fc..0faad511c8 100644
--- a/libtokencap/Makefile
+++ b/libtokencap/Makefile
@@ -24,15 +24,15 @@ CFLAGS      += -Wall -D_FORTIFY_SOURCE=2 -g -Wno-pointer-sign
 all: libtokencap.so
 
 libtokencap.so: libtokencap.so.c ../config.h
-	$(CC) $(CFLAGS) -shared -fPIC $< -o $@ $(LDFLAGS)
+	$(CC) $(CFLAGS) -shared -fPIC $< -o ../$@ $(LDFLAGS)
 
 .NOTPARALLEL: clean
 
 clean:
 	rm -f *.o *.so *~ a.out core core.[1-9][0-9]*
-	rm -f libtokencap.so
+	rm -f ../libtokencap.so
 
 install: all
-	install -m 755 libtokencap.so $${DESTDIR}$(HELPER_PATH)
+	install -m 755 ../libtokencap.so $${DESTDIR}$(HELPER_PATH)
 	install -m 644 README.tokencap $${DESTDIR}$(HELPER_PATH)
 
diff --git a/llvm_mode/Makefile b/llvm_mode/Makefile
index f635ca8375..2d35505eb4 100644
--- a/llvm_mode/Makefile
+++ b/llvm_mode/Makefile
@@ -37,7 +37,8 @@ endif
 
 LLVMVER  = $(shell $(LLVM_CONFIG) --version)
 LLVM_UNSUPPORTED = $(shell $(LLVM_CONFIG) --version | egrep -q '^[12]|^3\.0|^1[0-9]' && echo 1 || echo 0 )
-LLVM_MAJOR = ($shell $(LLVM_CONFIG) --version | sed 's/\..*//')
+LLVM_MAJOR = $(shell $(LLVM_CONFIG) --version | sed 's/\..*//')
+LLVM_BINDIR = $(shell $(LLVM_CONFIG) --bindir)
 
 ifeq "$(LLVM_UNSUPPORTED)" "1"
   $(warn llvm_mode only supports versions 3.8.0 up to 9)
@@ -77,11 +78,18 @@ endif
 # this seems to be busted on some distros, so using the one in $PATH is
 # probably better.
 
-ifeq "$(origin CC)" "default"
-  ifeq "$(shell uname)" "OpenBSD"
-    CC         = $(BIN_PATH)/clang
-    CXX        = $(BIN_PATH)/clang++
+CC         = $(LLVM_BINDIR)/clang
+CXX        = $(LLVM_BINDIR)/clang++
+
+ifeq "$(shell test -e $(CC) || echo 1 )" "1"
+  # llvm-config --bindir is not providing a valid path, so ...
+  ifeq "$(shell test -e "$(BIN_DIR)/clang" && echo 1)" "1"
+    # we found one in the local install directory, lets use these
+    CC         = $(BIN_DIR)/clang
+    CXX        = $(BIN_DIR)/clang++
   else
+    # hope for the best
+    $(warn we have trouble finding clang/clang++ - llvm-config is not helping us)
     CC         = clang
     CXX        = clang++
   endif
diff --git a/llvm_mode/README.md b/llvm_mode/README.md
index c7ef4b4526..f196395b16 100644
--- a/llvm_mode/README.md
+++ b/llvm_mode/README.md
@@ -41,6 +41,10 @@ In order to leverage this mechanism, you need to have clang installed on your
 system. You should also make sure that the llvm-config tool is in your path
 (or pointed to via LLVM_CONFIG in the environment).
 
+Note that if you have several LLVM versions installed, pointing LLVM_CONFIG
+to the version you want to use will switch compiling to this specific
+version - if you installation is set up correctly :-)
+
 Unfortunately, some systems that do have clang come without llvm-config or the
 LLVM development headers; one example of this is FreeBSD. FreeBSD users will
 also run into problems with clang being built statically and not being able to
diff --git a/llvm_mode/afl-llvm-pass.so.cc b/llvm_mode/afl-llvm-pass.so.cc
index 5d531a8769..58acd9be52 100644
--- a/llvm_mode/afl-llvm-pass.so.cc
+++ b/llvm_mode/afl-llvm-pass.so.cc
@@ -105,9 +105,7 @@ bool AFLCoverage::runOnModule(Module &M) {
 
     SAYF(cCYA "afl-llvm-pass" VERSION cRST " by <lszekeres@google.com>\n");
 
-  } else
-
-    be_quiet = 1;
+  } else if (getenv("AFL_QUIET")) be_quiet = 1;
 
   /* Decide instrumentation ratio */
 
diff --git a/llvm_mode/split-compares-pass.so.cc b/llvm_mode/split-compares-pass.so.cc
index 1e9d654206..c5da42c0e1 100644
--- a/llvm_mode/split-compares-pass.so.cc
+++ b/llvm_mode/split-compares-pass.so.cc
@@ -1,5 +1,6 @@
 /*
  * Copyright 2016 laf-intel
+ * extended for floating point by Heiko Eißfeldt
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,6 +22,7 @@
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/IR/Verifier.h"
 #include "llvm/IR/Module.h"
+#include "llvm/ADT/APFloat.h"
 
 #include "llvm/IR/IRBuilder.h"
 
@@ -49,9 +51,11 @@ class SplitComparesTransform : public ModulePass {
   }
 
  private:
-  bool splitCompares(Module &M, unsigned bitw);
+  size_t splitIntCompares(Module &M, unsigned bitw);
+  size_t splitFPCompares(Module &M);
   bool simplifyCompares(Module &M);
-  bool simplifySignedness(Module &M);
+  bool simplifyIntSignedness(Module &M);
+  size_t nextPowerOfTwo(size_t in);
 
 };
 
@@ -65,6 +69,7 @@ bool SplitComparesTransform::simplifyCompares(Module &M) {
 
   LLVMContext &              C = M.getContext();
   std::vector<Instruction *> icomps;
+  std::vector<Instruction *> fcomps;
   IntegerType *              Int1Ty = IntegerType::getInt1Ty(C);
 
   /* iterate over all functions, bbs and instruction and add
@@ -79,25 +84,41 @@ bool SplitComparesTransform::simplifyCompares(Module &M) {
 
         if ((selectcmpInst = dyn_cast<CmpInst>(&IN))) {
 
-          if (selectcmpInst->getPredicate() != CmpInst::ICMP_UGE &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_SGE &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_ULE &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_SLE) {
+          if (selectcmpInst->getPredicate() == CmpInst::ICMP_UGE ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_SGE ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_ULE ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_SLE) {
 
-            continue;
+            auto op0 = selectcmpInst->getOperand(0);
+            auto op1 = selectcmpInst->getOperand(1);
+
+            IntegerType *intTyOp0 = dyn_cast<IntegerType>(op0->getType());
+            IntegerType *intTyOp1 = dyn_cast<IntegerType>(op1->getType());
+
+            /* this is probably not needed but we do it anyway */
+            if (!intTyOp0 || !intTyOp1) { continue; }
+
+            icomps.push_back(selectcmpInst);
 
           }
 
-          auto op0 = selectcmpInst->getOperand(0);
-          auto op1 = selectcmpInst->getOperand(1);
+          if (selectcmpInst->getPredicate() == CmpInst::FCMP_OGE ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_UGE ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_OLE ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_ULE) {
 
-          IntegerType *intTyOp0 = dyn_cast<IntegerType>(op0->getType());
-          IntegerType *intTyOp1 = dyn_cast<IntegerType>(op1->getType());
+            auto op0 = selectcmpInst->getOperand(0);
+            auto op1 = selectcmpInst->getOperand(1);
 
-          /* this is probably not needed but we do it anyway */
-          if (!intTyOp0 || !intTyOp1) { continue; }
+            Type *TyOp0 = op0->getType();
+            Type *TyOp1 = op1->getType();
 
-          icomps.push_back(selectcmpInst);
+            /* this is probably not needed but we do it anyway */
+            if (TyOp0 != TyOp1) { continue; }
+
+            fcomps.push_back(selectcmpInst);
+
+          }
 
         }
 
@@ -107,7 +128,7 @@ bool SplitComparesTransform::simplifyCompares(Module &M) {
 
   }
 
-  if (!icomps.size()) { return false; }
+  if (!icomps.size() && !fcomps.size()) { return false; }
 
   for (auto &IcmpInst : icomps) {
 
@@ -173,18 +194,83 @@ bool SplitComparesTransform::simplifyCompares(Module &M) {
 
   }
 
+  /* now for floating point */
+  for (auto &FcmpInst : fcomps) {
+
+    BasicBlock *bb = FcmpInst->getParent();
+
+    auto op0 = FcmpInst->getOperand(0);
+    auto op1 = FcmpInst->getOperand(1);
+
+    /* find out what the new predicate is going to be */
+    auto               pred = dyn_cast<CmpInst>(FcmpInst)->getPredicate();
+    CmpInst::Predicate new_pred;
+    switch (pred) {
+
+      case CmpInst::FCMP_UGE: new_pred = CmpInst::FCMP_UGT; break;
+      case CmpInst::FCMP_OGE: new_pred = CmpInst::FCMP_OGT; break;
+      case CmpInst::FCMP_ULE: new_pred = CmpInst::FCMP_ULT; break;
+      case CmpInst::FCMP_OLE: new_pred = CmpInst::FCMP_OLT; break;
+      default:  // keep the compiler happy
+        continue;
+
+    }
+
+    /* split before the icmp instruction */
+    BasicBlock *end_bb = bb->splitBasicBlock(BasicBlock::iterator(FcmpInst));
+
+    /* the old bb now contains a unconditional jump to the new one (end_bb)
+     * we need to delete it later */
+
+    /* create the ICMP instruction with new_pred and add it to the old basic
+     * block bb it is now at the position where the old IcmpInst was */
+    Instruction *fcmp_np;
+    fcmp_np = CmpInst::Create(Instruction::FCmp, new_pred, op0, op1);
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), fcmp_np);
+
+    /* create a new basic block which holds the new EQ fcmp */
+    Instruction *fcmp_eq;
+    /* insert middle_bb before end_bb */
+    BasicBlock *middle_bb =
+        BasicBlock::Create(C, "injected", end_bb->getParent(), end_bb);
+    fcmp_eq = CmpInst::Create(Instruction::FCmp, CmpInst::FCMP_OEQ, op0, op1);
+    middle_bb->getInstList().push_back(fcmp_eq);
+    /* add an unconditional branch to the end of middle_bb with destination
+     * end_bb */
+    BranchInst::Create(end_bb, middle_bb);
+
+    /* replace the uncond branch with a conditional one, which depends on the
+     * new_pred icmp. True goes to end, false to the middle (injected) bb */
+    auto term = bb->getTerminator();
+    BranchInst::Create(end_bb, middle_bb, fcmp_np, bb);
+    term->eraseFromParent();
+
+    /* replace the old IcmpInst (which is the first inst in end_bb) with a PHI
+     * inst to wire up the loose ends */
+    PHINode *PN = PHINode::Create(Int1Ty, 2, "");
+    /* the first result depends on the outcome of icmp_eq */
+    PN->addIncoming(fcmp_eq, middle_bb);
+    /* if the source was the original bb we know that the icmp_np yielded true
+     * hence we can hardcode this value */
+    PN->addIncoming(ConstantInt::get(Int1Ty, 1), bb);
+    /* replace the old IcmpInst with our new and shiny PHI inst */
+    BasicBlock::iterator ii(FcmpInst);
+    ReplaceInstWithInst(FcmpInst->getParent()->getInstList(), ii, PN);
+
+  }
+
   return true;
 
 }
 
 /* this function transforms signed compares to equivalent unsigned compares */
-bool SplitComparesTransform::simplifySignedness(Module &M) {
+bool SplitComparesTransform::simplifyIntSignedness(Module &M) {
 
   LLVMContext &              C = M.getContext();
   std::vector<Instruction *> icomps;
   IntegerType *              Int1Ty = IntegerType::getInt1Ty(C);
 
-  /* iterate over all functions, bbs and instruction and add
+  /* iterate over all functions, bbs and instructions and add
    * all signed compares to icomps vector */
   for (auto &F : M) {
 
@@ -196,26 +282,24 @@ bool SplitComparesTransform::simplifySignedness(Module &M) {
 
         if ((selectcmpInst = dyn_cast<CmpInst>(&IN))) {
 
-          if (selectcmpInst->getPredicate() != CmpInst::ICMP_SGT &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_SLT) {
+          if (selectcmpInst->getPredicate() == CmpInst::ICMP_SGT ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_SLT) {
 
-            continue;
+            auto op0 = selectcmpInst->getOperand(0);
+            auto op1 = selectcmpInst->getOperand(1);
 
-          }
-
-          auto op0 = selectcmpInst->getOperand(0);
-          auto op1 = selectcmpInst->getOperand(1);
+            IntegerType *intTyOp0 = dyn_cast<IntegerType>(op0->getType());
+            IntegerType *intTyOp1 = dyn_cast<IntegerType>(op1->getType());
 
-          IntegerType *intTyOp0 = dyn_cast<IntegerType>(op0->getType());
-          IntegerType *intTyOp1 = dyn_cast<IntegerType>(op1->getType());
+            /* see above */
+            if (!intTyOp0 || !intTyOp1) { continue; }
 
-          /* see above */
-          if (!intTyOp0 || !intTyOp1) { continue; }
+            /* i think this is not possible but to lazy to look it up */
+            if (intTyOp0->getBitWidth() != intTyOp1->getBitWidth()) { continue; }
 
-          /* i think this is not possible but to lazy to look it up */
-          if (intTyOp0->getBitWidth() != intTyOp1->getBitWidth()) { continue; }
+            icomps.push_back(selectcmpInst);
 
-          icomps.push_back(selectcmpInst);
+          }
 
         }
 
@@ -328,8 +412,333 @@ bool SplitComparesTransform::simplifySignedness(Module &M) {
 
 }
 
+size_t SplitComparesTransform::nextPowerOfTwo(size_t in) {
+  --in;
+  in |= in >> 1;
+  in |= in >> 2;
+  in |= in >> 4;
+//  in |= in >> 8;
+//  in |= in >> 16;
+  return in + 1;
+}
+
+/* splits fcmps into two nested fcmps with sign compare and the rest */
+size_t SplitComparesTransform::splitFPCompares(Module &M) {
+  size_t count = 0;
+
+  LLVMContext &C = M.getContext();
+
+  const DataLayout &dl = M.getDataLayout();
+
+  /* define unions with floating point and (sign, exponent, mantissa)  triples */
+  if (dl.isLittleEndian()) {
+  }
+  else if (dl.isBigEndian()) {
+  }
+  else {
+    return count;
+  }
+
+  std::vector<CmpInst *> fcomps;
+
+  /* get all EQ, NE, GT, and LT fcmps. if the other two
+   * functions were executed only these four predicates should exist */
+  for (auto &F : M) {
+
+    for (auto &BB : F) {
+
+      for (auto &IN : BB) {
+
+        CmpInst *selectcmpInst = nullptr;
+
+        if ((selectcmpInst = dyn_cast<CmpInst>(&IN))) {
+
+          if (selectcmpInst->getPredicate() == CmpInst::FCMP_OEQ ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_ONE ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_UNE ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_OGT ||
+              selectcmpInst->getPredicate() == CmpInst::FCMP_OLT) {
+
+            auto op0 = selectcmpInst->getOperand(0);
+            auto op1 = selectcmpInst->getOperand(1);
+
+            Type *TyOp0 = op0->getType();
+            Type *TyOp1 = op1->getType();
+
+            if (TyOp0 != TyOp1) { continue; }
+
+            fcomps.push_back(selectcmpInst);
+
+          }
+
+        }
+
+      }
+
+    }
+
+  }
+  if (!fcomps.size()) { return count; }
+
+  IntegerType *Int1Ty = IntegerType::getInt1Ty(C);
+
+  for (auto &FcmpInst : fcomps) {
+
+    BasicBlock *bb = FcmpInst->getParent();
+
+    auto op0 = FcmpInst->getOperand(0);
+    auto op1 = FcmpInst->getOperand(1);
+
+    unsigned op0_size, op1_size;
+    op0_size = op0->getType()->getPrimitiveSizeInBits();
+    op1_size = op1->getType()->getPrimitiveSizeInBits();
+
+    if (op0_size != op1_size) {
+      continue;
+    }
+
+    const unsigned int precision  = llvm::APFloatBase::semanticsPrecision(op0->getType()->getFltSemantics());
+    const unsigned int sizeInBits = llvm::APFloatBase::semanticsSizeInBits(op0->getType()->getFltSemantics());
+
+
+    const unsigned shiftR_exponent = precision - 1;
+    const unsigned long long mask_fraction = ((1 << (precision - 2))) | ((1 << (precision - 2)) - 1);
+    const unsigned long long mask_exponent = (1 << (sizeInBits - precision)) - 1;
+
+    // round up sizes to the next power of two
+    // this should help with integer compare splitting
+    size_t exTySizeBytes = ((sizeInBits - precision + 7) >> 3);
+    size_t frTySizeBytes = ((precision - 1          + 7) >> 3);
+
+    IntegerType *IntExponentTy = IntegerType::get(C, nextPowerOfTwo(exTySizeBytes) << 3);
+    IntegerType *IntFractionTy = IntegerType::get(C, nextPowerOfTwo(frTySizeBytes) << 3);
+
+    BasicBlock *end_bb = bb->splitBasicBlock(BasicBlock::iterator(FcmpInst));
+
+    /* create the integers from floats directly */
+    Instruction *b_op0, *b_op1;
+    b_op0 = CastInst::Create(Instruction::BitCast, op0, IntegerType::get(C, op0_size));
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), b_op0);
+
+    b_op1 = CastInst::Create(Instruction::BitCast, op1, IntegerType::get(C, op1_size));
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), b_op1);
+
+    /* isolate signs of value of floating point type */
+
+    /* create a 1 bit compare for the sign bit. to do this shift and trunc
+     * the original operands so only the first bit remains.*/
+    Instruction *s_s0, *t_s0, *s_s1, *t_s1, *icmp_sign_bit;
+
+    s_s0 = BinaryOperator::Create(Instruction::LShr, b_op0,
+                                   ConstantInt::get(b_op0->getType(), op0_size - 1));
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), s_s0);
+    t_s0 = new TruncInst(s_s0, Int1Ty);
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), t_s0);
+
+    s_s1 = BinaryOperator::Create(Instruction::LShr, b_op1,
+                                   ConstantInt::get(b_op1->getType(), op1_size - 1));
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), s_s1);
+    t_s1 = new TruncInst(s_s1, Int1Ty);
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), t_s1);
+
+    /* compare of the sign bits */
+    icmp_sign_bit = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, t_s0, t_s1);
+    bb->getInstList().insert(bb->getTerminator()->getIterator(), icmp_sign_bit);
+
+    /* create a new basic block which is executed if the signedness bits are
+     * equal */
+    BasicBlock * signequal_bb =
+        BasicBlock::Create(C, "signequal", end_bb->getParent(), end_bb);
+
+    BranchInst::Create(end_bb, signequal_bb);
+
+    /* create a new bb which is executed if exponents are equal */
+    BasicBlock * middle_bb =
+        BasicBlock::Create(C, "injected", end_bb->getParent(), end_bb);
+
+    BranchInst::Create(end_bb, middle_bb);
+
+    auto term = bb->getTerminator();
+    /* if the signs are different goto end_bb else to signequal_bb */
+    BranchInst::Create(signequal_bb, end_bb, icmp_sign_bit, bb);
+    term->eraseFromParent();
+
+    /* insert code for equal signs */
+
+    /* isolate the exponents */
+    Instruction *s_e0, *m_e0, *t_e0, *s_e1, *m_e1, *t_e1;
+
+    s_e0 = BinaryOperator::Create(Instruction::LShr, b_op0, ConstantInt::get(b_op0->getType(), shiftR_exponent));
+    s_e1 = BinaryOperator::Create(Instruction::LShr, b_op1, ConstantInt::get(b_op1->getType(), shiftR_exponent));
+    signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), s_e0);
+    signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), s_e1);
+
+    if (sizeInBits - precision < exTySizeBytes * 8) {
+      m_e0 = BinaryOperator::Create(Instruction::And, s_e0, ConstantInt::get(s_e0->getType(), mask_exponent));
+      m_e1 = BinaryOperator::Create(Instruction::And, s_e1, ConstantInt::get(s_e1->getType(), mask_exponent));
+      signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), m_e0);
+      signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), m_e1);
+
+      t_e0 = new TruncInst(m_e0, IntExponentTy);
+      t_e1 = new TruncInst(m_e1, IntExponentTy);
+    } else {
+      t_e0 = new TruncInst(s_e0, IntExponentTy);
+      t_e1 = new TruncInst(s_e1, IntExponentTy);
+    }
+    signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), t_e0);
+    signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), t_e1);
+    /* compare the exponents of the operands */
+    Instruction *icmp_exponent_result;
+    switch (FcmpInst->getPredicate()) {
+      case CmpInst::FCMP_OEQ:
+        icmp_exponent_result =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, t_e0, t_e1);
+      break;
+      case CmpInst::FCMP_ONE:
+      case CmpInst::FCMP_UNE:
+        icmp_exponent_result =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_NE, t_e0, t_e1);
+      break;
+      case CmpInst::FCMP_OGT:
+        Instruction *icmp_exponent;
+        icmp_exponent =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_UGT, t_e0, t_e1);
+        signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), icmp_exponent);
+        icmp_exponent_result = BinaryOperator::Create(Instruction::Xor, icmp_exponent, t_s0);
+      break;
+      case CmpInst::FCMP_OLT:
+        icmp_exponent =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULT, t_e0, t_e1);
+        signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), icmp_exponent);
+        icmp_exponent_result = BinaryOperator::Create(Instruction::Xor, icmp_exponent, t_s0);
+      break;
+      default:
+        continue;
+    }
+    signequal_bb->getInstList().insert(signequal_bb->getTerminator()->getIterator(), icmp_exponent_result);
+
+    {
+    auto term = signequal_bb->getTerminator();
+    /* if the exponents are different do a fraction cmp */
+    BranchInst::Create(middle_bb, end_bb, icmp_exponent_result, signequal_bb);
+    term->eraseFromParent();
+    }
+
+
+    /* isolate the mantissa aka fraction */
+    Instruction *t_f0, *t_f1;
+    bool needTrunc = IntFractionTy->getPrimitiveSizeInBits() < op0_size;
+//errs() << "Fractions: IntFractionTy size " << IntFractionTy->getPrimitiveSizeInBits() << ", op0_size " << op0_size << ", needTrunc " << needTrunc << "\n";
+    if (precision - 1 < frTySizeBytes * 8) {
+      Instruction *m_f0, *m_f1;
+      m_f0 = BinaryOperator::Create(Instruction::And, b_op0, ConstantInt::get(b_op0->getType(), mask_fraction));
+      m_f1 = BinaryOperator::Create(Instruction::And, b_op1, ConstantInt::get(b_op1->getType(), mask_fraction));
+      middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), m_f0);
+      middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), m_f1);
+
+      if (needTrunc) {
+        t_f0 = new TruncInst(m_f0, IntFractionTy);
+        t_f1 = new TruncInst(m_f1, IntFractionTy);
+        middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), t_f0);
+        middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), t_f1);
+      } else {
+        t_f0 = m_f0;
+        t_f1 = m_f1;
+      }
+    } else {
+      if (needTrunc) {
+        t_f0 = new TruncInst(b_op0, IntFractionTy);
+        t_f1 = new TruncInst(b_op1, IntFractionTy);
+        middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), t_f0);
+        middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), t_f1);
+      } else {
+        t_f0 = b_op0;
+        t_f1 = b_op1;
+      }
+    }
+
+    /* compare the fractions of the operands */
+    Instruction *icmp_fraction_result;
+    switch (FcmpInst->getPredicate()) {
+      case CmpInst::FCMP_OEQ:
+        icmp_fraction_result =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, t_f0, t_f1);
+      break;
+      case CmpInst::FCMP_UNE:
+      case CmpInst::FCMP_ONE:
+        icmp_fraction_result =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_NE, t_f0, t_f1);
+      break;
+      case CmpInst::FCMP_OGT:
+        Instruction *icmp_fraction;
+        icmp_fraction =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_UGT, t_f0, t_f1);
+        middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), icmp_fraction);
+        icmp_fraction_result = BinaryOperator::Create(Instruction::Xor, icmp_fraction, t_s0);
+      break;
+      case CmpInst::FCMP_OLT:
+        icmp_fraction =
+            CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULT, t_f0, t_f1);
+        middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), icmp_fraction);
+        icmp_fraction_result = BinaryOperator::Create(Instruction::Xor, icmp_fraction, t_s0);
+      break;
+      default:
+        continue;
+    }
+    middle_bb->getInstList().insert(middle_bb->getTerminator()->getIterator(), icmp_fraction_result);
+
+    PHINode *PN = PHINode::Create(Int1Ty, 3, "");
+
+    switch (FcmpInst->getPredicate()) {
+      case CmpInst::FCMP_OEQ:
+        /* unequal signs cannot be equal values */
+        /* goto false branch */
+        PN->addIncoming(ConstantInt::get(Int1Ty, 0), bb);
+        /* unequal exponents cannot be equal values, too */
+        PN->addIncoming(ConstantInt::get(Int1Ty, 0), signequal_bb);
+        /* fractions comparison */
+        PN->addIncoming(icmp_fraction_result, middle_bb);
+      break;
+      case CmpInst::FCMP_ONE:
+      case CmpInst::FCMP_UNE:
+        /* unequal signs are unequal values */
+        /* goto true branch */
+        PN->addIncoming(ConstantInt::get(Int1Ty, 1), bb);
+        /* unequal exponents are unequal values, too */
+        PN->addIncoming(ConstantInt::get(Int1Ty, 1), signequal_bb);
+        /* fractions comparison */
+        PN->addIncoming(icmp_fraction_result, middle_bb);
+      break;
+      case CmpInst::FCMP_OGT:
+        /* if op1 is negative goto true branch, 
+           else go on comparing */
+        PN->addIncoming(t_s1, bb);
+        PN->addIncoming(icmp_exponent_result, signequal_bb);
+        PN->addIncoming(icmp_fraction_result, middle_bb);
+      break;
+      case CmpInst::FCMP_OLT:
+        /* if op0 is negative goto true branch,
+           else go on comparing */
+        PN->addIncoming(t_s0, bb);
+        PN->addIncoming(icmp_exponent_result, signequal_bb);
+        PN->addIncoming(icmp_fraction_result, middle_bb);
+      break;
+      default:
+        continue;
+    }
+
+    BasicBlock::iterator ii(FcmpInst);
+    ReplaceInstWithInst(FcmpInst->getParent()->getInstList(), ii, PN);
+    ++count;
+  }
+
+  return count;
+
+}
+
 /* splits icmps of size bitw into two nested icmps with bitw/2 size each */
-bool SplitComparesTransform::splitCompares(Module &M, unsigned bitw) {
+size_t SplitComparesTransform::splitIntCompares(Module &M, unsigned bitw) {
+  size_t count = 0;
 
   LLVMContext &C = M.getContext();
 
@@ -339,13 +748,14 @@ bool SplitComparesTransform::splitCompares(Module &M, unsigned bitw) {
 
   std::vector<Instruction *> icomps;
 
-  if (bitw % 2) { return false; }
+  if (bitw % 2) { return 0; }
 
   /* not supported yet */
-  if (bitw > 64) { return false; }
+  if (bitw > 64) { return 0; }
 
-  /* get all EQ, NE, UGT, and ULT icmps of width bitw. if the other two
-   * unctions were executed only these four predicates should exist */
+  /* get all EQ, NE, UGT, and ULT icmps of width bitw. if the 
+   * functions simplifyCompares() and simplifyIntSignedness()
+   * were executed only these four predicates should exist */
   for (auto &F : M) {
 
     for (auto &BB : F) {
@@ -356,33 +766,31 @@ bool SplitComparesTransform::splitCompares(Module &M, unsigned bitw) {
 
         if ((selectcmpInst = dyn_cast<CmpInst>(&IN))) {
 
-          if (selectcmpInst->getPredicate() != CmpInst::ICMP_EQ &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_NE &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_UGT &&
-              selectcmpInst->getPredicate() != CmpInst::ICMP_ULT) {
+          if (selectcmpInst->getPredicate() == CmpInst::ICMP_EQ ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_NE ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_UGT ||
+              selectcmpInst->getPredicate() == CmpInst::ICMP_ULT) {
 
-            continue;
+            auto op0 = selectcmpInst->getOperand(0);
+            auto op1 = selectcmpInst->getOperand(1);
 
-          }
+            IntegerType *intTyOp0 = dyn_cast<IntegerType>(op0->getType());
+            IntegerType *intTyOp1 = dyn_cast<IntegerType>(op1->getType());
 
-          auto op0 = selectcmpInst->getOperand(0);
-          auto op1 = selectcmpInst->getOperand(1);
+            if (!intTyOp0 || !intTyOp1) { continue; }
 
-          IntegerType *intTyOp0 = dyn_cast<IntegerType>(op0->getType());
-          IntegerType *intTyOp1 = dyn_cast<IntegerType>(op1->getType());
+            /* check if the bitwidths are the one we are looking for */
+            if (intTyOp0->getBitWidth() != bitw ||
+                intTyOp1->getBitWidth() != bitw) {
 
-          if (!intTyOp0 || !intTyOp1) { continue; }
+              continue;
 
-          /* check if the bitwidths are the one we are looking for */
-          if (intTyOp0->getBitWidth() != bitw ||
-              intTyOp1->getBitWidth() != bitw) {
+            }
 
-            continue;
+            icomps.push_back(selectcmpInst);
 
           }
 
-          icomps.push_back(selectcmpInst);
-
         }
 
       }
@@ -391,7 +799,7 @@ bool SplitComparesTransform::splitCompares(Module &M, unsigned bitw) {
 
   }
 
-  if (!icomps.size()) { return false; }
+  if (!icomps.size()) { return 0; }
 
   for (auto &IcmpInst : icomps) {
 
@@ -482,7 +890,7 @@ bool SplitComparesTransform::splitCompares(Module &M, unsigned bitw) {
       /* transformations for < and > */
 
       /* create a basic block which checks for the inverse predicate.
-       * if this is true we can go to the end if not we have to got to the
+       * if this is true we can go to the end if not we have to go to the
        * bb which checks the lower half of the operands */
       Instruction *icmp_inv_cmp, *op0_low, *op1_low, *icmp_low;
       BasicBlock * inv_cmp_bb =
@@ -528,10 +936,10 @@ bool SplitComparesTransform::splitCompares(Module &M, unsigned bitw) {
       ReplaceInstWithInst(IcmpInst->getParent()->getInstList(), ii, PN);
 
     }
-
+    ++count;
   }
 
-  return true;
+  return count;
 
 }
 
@@ -545,26 +953,32 @@ bool SplitComparesTransform::runOnModule(Module &M) {
 
   simplifyCompares(M);
 
-  simplifySignedness(M);
+  simplifyIntSignedness(M);
 
   if (getenv("AFL_QUIET") == NULL)
-    errs() << "Split-compare-pass by laf.intel@gmail.com\n";
+    errs() << "Split-compare-pass by laf.intel@gmail.com, extended by heiko@hexco.de\n";
+
+  errs() << "Split-floatingpoint-compare-pass: " << splitFPCompares(M) << " FP comparisons splitted\n";
 
   switch (bitw) {
 
     case 64:
-      errs() << "Running split-compare-pass " << 64 << "\n";
-      splitCompares(M, 64);
+      errs() << "Split-integer-compare-pass " << bitw << "bit: " 
+        << splitIntCompares(M, bitw) << " splitted\n";
 
+      bitw >>= 1;
       [[clang::fallthrough]]; /*FALLTHRU*/                   /* FALLTHROUGH */
     case 32:
-      errs() << "Running split-compare-pass " << 32 << "\n";
-      splitCompares(M, 32);
+      errs() << "Split-integer-compare-pass " << bitw << "bit: " 
+        << splitIntCompares(M, bitw) << " splitted\n";
 
+      bitw >>= 1;
       [[clang::fallthrough]]; /*FALLTHRU*/                   /* FALLTHROUGH */
     case 16:
-      errs() << "Running split-compare-pass " << 16 << "\n";
-      splitCompares(M, 16);
+      errs() << "Split-integer-compare-pass " << bitw << "bit: " 
+        << splitIntCompares(M, bitw) << " splitted\n";
+
+      bitw >>= 1;
       break;
 
     default:
diff --git a/src/afl-fuzz-globals.c b/src/afl-fuzz-globals.c
index 1358a1fb40..01b242b81e 100644
--- a/src/afl-fuzz-globals.c
+++ b/src/afl-fuzz-globals.c
@@ -259,3 +259,7 @@ PyObject *py_functions[PY_FUNC_COUNT];
 
 #endif
 
+#ifdef _AFL_DOCUMENT_MUTATIONS
+  u8 do_document;
+  u32 document_counter;
+#endif
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 4908626a1e..fab82e2db9 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1478,8 +1478,9 @@ void check_cpu_governor(void) {
        "    to make afl-fuzz skip this check - but expect some performance "
        "drop.\n",
        min / 1024, max / 1024);
+  FATAL("Suboptimal CPU scaling governor");
 
-#else
+#elif defined __APPLE__
   u64   min = 0, max = 0;
   size_t mlen = sizeof(min);
   if (getenv("AFL_SKIP_CPUFREQ")) return;
@@ -1510,8 +1511,8 @@ void check_cpu_governor(void) {
        "    to make afl-fuzz skip this check - but expect some performance "
        "drop.\n",
        min / 1024, max / 1024);
-#endif
   FATAL("Suboptimal CPU scaling governor");
+#endif
 }
 
 /* Count the number of logical CPU cores. */
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 59483b8fd0..1824f0b2b6 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -4231,6 +4231,7 @@ u8 common_fuzzing(char** argv, struct MOpt_globals_t MOpt_globals) {
 
 #define core_fuzzing(a) common_fuzzing((a), MOpt_globals_core)
 
+
 void pso_updating(void) {
 
   g_now += 1;
@@ -4310,6 +4311,22 @@ void pso_updating(void) {
 u8 fuzz_one(char** argv) {
 
   int key_val_lv = 0;
+
+#ifdef _AFL_DOCUMENT_MUTATIONS
+  if (do_document == 0) {
+    char *fn = alloc_printf("%s/mutations", out_dir);
+    if (fn) {
+      do_document = mkdir(fn, 0700); // if it exists we do not care
+      do_document = 1;
+      ck_free(fn);
+    } else
+      PFATAL("malloc()");
+  } else {
+    do_document = 2;
+    stop_soon = 2;
+  }
+#endif
+
   if (limit_time_sig == 0) {
 
     key_val_lv = fuzz_one_original(argv);
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index f2f663dcf1..220433fce1 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -251,6 +251,18 @@ void write_to_testcase(void* mem, u32 len) {
 
   s32 fd = out_fd;
 
+#ifdef _AFL_DOCUMENT_MUTATIONS
+  s32 doc_fd;
+  char *fn = alloc_printf("%s/mutations/%09u:%s", out_dir, document_counter++, describe_op(0));
+  if (fn != NULL) {
+    if ((doc_fd = open(fn, O_WRONLY | O_CREAT | O_TRUNC, 0600)) >= 0) {
+      if (write(doc_fd, mem, len) != len) PFATAL("write to mutation file failed: %s", fn);
+      close(doc_fd);
+    }
+    ck_free(fn);
+  }
+#endif
+
   if (out_file) {
 
     // unlink(out_file);                                     /* Ignore errors.
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 803faceddd..5e773e965d 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -576,12 +576,13 @@ void show_stats(void) {
                 "  imported : " cRST "%-10s" bSTG       bV "\n",
        tmp, sync_id ? DI(queued_imported) : (u8*)"n/a");
 
-  sprintf(tmp, "%s/%s, %s/%s, %s/%s", DI(stage_finds[STAGE_HAVOC]),
+  sprintf(tmp, "%s/%s, %s/%s, %s/%s, %s/%s", DI(stage_finds[STAGE_HAVOC]),
           DI(stage_cycles[STAGE_HAVOC]), DI(stage_finds[STAGE_SPLICE]),
           DI(stage_cycles[STAGE_SPLICE]), DI(stage_finds[STAGE_PYTHON]),
-          DI(stage_cycles[STAGE_PYTHON]));
+          DI(stage_cycles[STAGE_PYTHON]), DI(stage_finds[STAGE_CUSTOM_MUTATOR]),
+          DI(stage_cycles[STAGE_CUSTOM_MUTATOR]));
 
-  SAYF(bV bSTOP "       havoc : " cRST "%-36s " bSTG bV bSTOP, tmp);
+  SAYF(bV bSTOP "havoc/custom : " cRST "%-36s " bSTG bV bSTOP, tmp);
 
   if (t_bytes)
     sprintf(tmp, "%0.02f%%", stab_ratio);
diff --git a/test/test-compcov.c b/test/test-compcov.c
new file mode 100644
index 0000000000..978d555147
--- /dev/null
+++ b/test/test-compcov.c
@@ -0,0 +1,30 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+int main(int argc, char** argv) {
+  char *input = argv[1], *buf, buffer[20];
+
+  if (argc < 2) {
+    ssize_t ret = read(0, buffer, sizeof(buffer) - 1);
+    buffer[ret] = 0;
+    input = buffer;
+  }
+  
+  if (strcmp(input, "LIBTOKENCAP") == 0)
+    printf("your string was libtokencap\n");
+  else if (strcmp(input, "BUGMENOT") == 0)
+    printf("your string was bugmenot\n");
+  else if (strcmp(input, "BUFFEROVERFLOW") == 0) {
+    buf = malloc(16);
+    strcpy(buf, "TEST");
+    strcat(buf, input);
+    printf("This will only crash with libdislocator: %s\n", buf);
+    return 0;
+  } else
+    printf("I do not know your string\n");
+
+  return 0;
+
+}
diff --git a/test/test.sh b/test/test.sh
new file mode 100755
index 0000000000..d230cf7177
--- /dev/null
+++ b/test/test.sh
@@ -0,0 +1,216 @@
+#!/bin/bash
+
+#
+# Ensure we have: test, type, diff -q, echo -e, grep -aqE, timeout
+#
+test -z "" 2> /dev/null || { echo Error: test command not found ; exit 1 ; }
+GREP=`type grep > /dev/null 2>&1 && echo OK`
+TIMEOUT=`type timeout > /dev/null 2>&1 && echo OK`
+test "$GREP" = OK || { echo Error: grep command not found ; exit 1 ; }
+echo foobar | grep -aqE 'asd|oob' 2> /dev/null || { echo Error: grep command does not support -q, -a and/or -E option ; exit 1 ; }
+echo 1 > test.1
+echo 1 > test.2
+OK=OK
+diff -q test.1 test.2 >/dev/null 2>&1 || OK=
+rm -f test.1 test.2
+test -z "$OK" && { echo Error: diff -q is not working ; exit 1 ; }
+
+ECHO="echo -e"
+$ECHO '\x41' 2>&1 | grep -qE '^A' || {
+  ECHO=
+  test -e /bin/echo && {
+    ECHO="/bin/echo -e"
+    $ECHO '\x41' 2>&1 | grep -qE '^A' || ECHO=
+  }
+}
+test -z "$ECHO" && { echo Error: echo command does not support -e option ; exit 1 ; }
+
+export AFL_EXIT_WHEN_DONE=1
+export AFL_SKIP_CPUFREQ=1
+unset AFL_QUIET
+unset AFL_DEBUG
+unset AFL_HARDEN
+unset AFL_USE_ASAN
+unset AFL_USE_MSAN
+unset AFL_CC
+unset AFL_PRELOAD
+unset AFL_LLVM_WHITELIST
+unset AFL_LLVM_INSTRIM
+unset AFL_LLVM_LAF_SPLIT_SWITCHES
+unset AFL_LLVM_LAF_TRANSFORM_COMPARES
+unset AFL_LLVM_LAF_SPLIT_COMPARES
+
+GREY="\\x1b[1;90m"
+BLUE="\\x1b[1;94m"
+GREEN="\\x1b[0;32m"
+RED="\\x1b[0;31m"
+YELLOW="\\x1b[1;93m"
+RESET="\\x1b[0m"
+
+$ECHO "${RESET}${GREY}[*] starting afl++ test framework ..."
+
+$ECHO "$BLUE[*] Testing: afl-gcc, afl-showmap and afl-fuzz"
+test -e ../afl-gcc -a -e ../afl-showmap -a -e ../afl-fuzz && {
+  ../afl-gcc -o test-instr.plain ../test-instr.c > /dev/null 2>&1
+  AFL_HARDEN=1 ../afl-gcc -o test-instr.harden ../test-instr.c > /dev/null 2>&1
+  test -e test-instr.plain && {
+    $ECHO "$GREEN[+] afl-gcc compilation succeeded"
+    echo 0 | ../afl-showmap -o test-instr.plain.0 -r -- ./test-instr.plain > /dev/null 2>&1
+    ../afl-showmap -o test-instr.plain.1 -r -- ./test-instr.plain < /dev/null > /dev/null 2>&1
+    test -e test-instr.plain.0 -a -e test-instr.plain.1 && {
+      diff -q test-instr.plain.0 test-instr.plain.1 > /dev/null 2>&1 && {
+        $ECHO "$RED[!] afl-gcc instrumentation should be different on different input but is not"
+      } || $ECHO "$GREEN[+] afl-gcc instrumentation present and working correctly"
+    } || $ECHO "$RED[!] afl-gcc instrumentation failed"
+    rm -f test-instr.plain.0 test-instr.plain.1
+  } || $ECHO "$RED[!] afl-gcc failed"
+  test -e test-instr.harden && {
+    grep -qa fstack-protector-all test-instr.harden > /dev/null 2>&1 && {
+      $ECHO "$GREEN[+] afl-gcc hardened mode succeeded and is working"
+    } || $ECHO "$RED[!] afl-gcc hardened mode is not hardened"
+    rm -f test-instr.harden
+  } || $ECHO "$RED[!] afl-gcc hardened mode compilation failed"
+  # now we want to be sure that afl-fuzz is working  
+  test -n "$TIMEOUT" && {
+    mkdir -p in
+    echo 0 > in/in
+    $ECHO "$GREY[*] running afl-fuzz for afl-gcc, this will take approx 10 seconds"
+    {
+      timeout -s KILL 10 ../afl-fuzz -i in -o out -- ./test-instr.plain > /dev/null 2>&1
+    } > /dev/null 2>&1
+    test -n "$( ls out/queue/id:000002* 2> /dev/null )" && {
+      $ECHO "$GREEN[+] afl-fuzz is working correctly with afl-gcc"
+    } || $ECHO "$RED[!] afl-fuzz is not working correctly with afl-gcc"
+    rm -rf in out
+  } || $ECHO "$YELLOW[-] we cannot test afl-fuzz because we are missing the timeout command"
+  rm -f test-instr.plain
+} || $ECHO "$YELLOW[-] afl is not compiled, cannot test"
+
+$ECHO "$BLUE[*] Testing: llvm_mode"
+test -e ../afl-clang-fast && {
+  ../afl-clang-fast -o test-instr.plain ../test-instr.c > /dev/null 2>&1
+  AFL_HARDEN=1 ../afl-clang-fast -o test-compcov.harden test-compcov.c > /dev/null 2>&1
+  test -e test-instr.plain && {
+    $ECHO "$GREEN[+] llvm_mode compilation succeeded"
+    echo 0 | ../afl-showmap -o test-instr.plain.0 -r -- ./test-instr.plain > /dev/null 2>&1
+    ../afl-showmap -o test-instr.plain.1 -r -- ./test-instr.plain < /dev/null > /dev/null 2>&1
+    test -e test-instr.plain.0 -a -e test-instr.plain.1 && {
+      diff -q test-instr.plain.0 test-instr.plain.1 > /dev/null 2>&1 && {
+        $ECHO "$RED[!] llvm_mode instrumentation should be different on different input but is not"
+      } || $ECHO "$GREEN[+] llvm_mode instrumentation present and working correctly"
+    } || $ECHO "$RED[!] llvm_mode instrumentation failed"
+    rm -f test-instr.plain.0 test-instr.plain.1
+  } || $ECHO "$RED[!] llvm_mode failed"
+  test -e test-compcov.harden && {
+    grep -Eqa 'stack_chk_fail|fstack-protector-all|fortified' test-compcov.harden > /dev/null 2>&1 && {
+      $ECHO "$GREEN[+] llvm_mode hardened mode succeeded and is working"
+    } || $ECHO "$RED[!] llvm_mode hardened mode is not hardened"
+    rm -f test-compcov.harden
+  } || $ECHO "$RED[!] llvm_mode hardened mode compilation failed"
+  # now we want to be sure that afl-fuzz is working  
+  test -n "$TIMEOUT" && {
+    mkdir -p in
+    echo 0 > in/in
+    $ECHO "$GREY[*] running afl-fuzz for llvm_mode, this will take approx 10 seconds"
+    {
+      timeout -s KILL 10 ../afl-fuzz -i in -o out -- ./test-instr.plain > /dev/null 2>&1
+    } > /dev/null 2>&1
+    test -n "$( ls out/queue/id:000002* 2> /dev/null )" && {
+      $ECHO "$GREEN[+] afl-fuzz is working correctly with llvm_mode"
+    } || $ECHO "$RED[!] afl-fuzz is not working correctly with llvm_mode"
+    rm -rf in out
+  } || $ECHO "$YELLOW[-] we cannot test afl-fuzz because we are missing the timeout command"
+  rm -f test-instr.plain
+
+  # now for the special llvm_mode things
+  AFL_LLVM_INSTRIM=1 AFL_LLVM_INSTRIM_LOOPHEAD=1 ../afl-clang-fast -o test-compcov.instrim test-compcov.c > /dev/null 2> test.out
+  test -e test-compcov.instrim && {
+    grep -Eq " [1-3] location" test.out && {
+      $ECHO "$GREEN[+] llvm_mode InsTrim feature works correctly"
+    } || $ECHO "$RED[!] llvm_mode InsTrim feature failed"
+  } || $ECHO "$RED[!] llvm_mode InsTrim feature compilation failed"
+  rm -f test-compcov.instrim test.out
+  AFL_LLVM_LAF_SPLIT_SWITCHES=1 AFL_LLVM_LAF_TRANSFORM_COMPARES=1 AFL_LLVM_LAF_SPLIT_COMPARES=1 ../afl-clang-fast -o test-compcov.compcov test-compcov.c > /dev/null 2> test.out
+  test -e test-compcov.compcov && {
+    grep -Eq " [3-9][0-9] location" test.out && {
+      $ECHO "$GREEN[+] llvm_mode laf-intel/compcov feature works correctly"
+    } || $ECHO "$RED[!] llvm_mode laf-intel/compcov feature failed"
+  } || $ECHO "$RED[!] llvm_mode laf-intel/compcov feature compilation failed"
+  rm -f test-compcov.compcov test.out
+  echo foobar.c > whitelist.txt
+  AFL_LLVM_WHITELIST=whitelist.txt ../afl-clang-fast -o test-compcov test-compcov.c > test.out 2>&1
+  test -e test-compcov && {
+    grep -q "No instrumentation targets found" test.out && {
+      $ECHO "$GREEN[+] llvm_mode whitelist feature works correctly"
+    } || $ECHO "$RED[!] llvm_mode whitelist feature failed"
+  } || $ECHO "$RED[!] llvm_mode whitelist feature compilation failed"
+  rm -f test-compcov test.out whitelist.txt
+  ../afl-clang-fast -o test-persistent ../experimental/persistent_demo/persistent_demo.c > /dev/null 2>&1
+  test -e test-persistent && {
+    echo foo | ../afl-showmap -o /dev/null -q -r ./test-persistent && {
+      $ECHO "$GREEN[+] lvm_mode persistent mode feature works correctly"
+    } || $ECHO "$RED[!] llvm_mode persistent mode feature failed to work"
+  } || $ECHO "$RED[!] llvm_mode persistent mode feature compilation failed"
+  rm -f test-persistent
+} || $ECHO "$YELLOW[-] llvm_mode not compiled, cannot test"
+
+$ECHO "$BLUE[*] Testing: shared library extensions"
+gcc -o test-compcov test-compcov.c > /dev/null 2>&1
+test -e ../libtokencap.so && {
+  AFL_TOKEN_FILE=token.out LD_PRELOAD=../libtokencap.so ./test-compcov foobar > /dev/null 2>&1
+  grep -q BUGMENOT token.out > /dev/null 2>&1 && {
+    $ECHO "$GREEN[+] libtokencap did successfully capture tokens"
+  } || $ECHO "$RED[!] libtokencap did not capture tokens"
+  rm -f token.out
+} || $ECHO "$YELLOW[-] libtokencap is not compiled, cannot test"
+test -e ../libdislocator.so && {
+  {
+    ulimit -c 1
+    LD_PRELOAD=../libdislocator.so ./test-compcov BUFFEROVERFLOW > test.out 2> /dev/null
+  } > /dev/null 2>&1
+  grep -q BUFFEROVERFLOW test.out > /dev/null 2>&1 && {
+    $ECHO "$RED[!] libdislocator did not detect the memory corruption"
+  } || $ECHO "$GREEN[+] libdislocator did successfully detect the memory corruption" 
+  rm -f test.out core test-compcov.core core.test-compcov
+} || $ECHO "$YELLOW[-] libdislocator is not compiled, cannot test"
+rm -f test-compcov
+
+$ECHO "$BLUE[*] Testing: qemu_mode"
+test -e ../afl-qemu-trace && {
+  gcc -o test-instr ../test-instr.c
+  gcc -o test-compcov test-compcov.c
+  test -e test-instr -a -e test-compcov && {
+    test -n "$TIMEOUT" && {
+      mkdir -p in
+      echo 0 > in/in
+      $ECHO "$GREY[*] running afl-fuzz for qemu_mode, this will take approx 10 seconds"
+      {
+        timeout -s KILL 10 ../afl-fuzz -Q -i in -o out -- ./test-instr > /dev/null 2>&1
+      } > /dev/null 2>&1
+      test -n "$( ls out/queue/id:000002* 2> /dev/null )" && {
+        $ECHO "$GREEN[+] afl-fuzz is working correctly with qemu_mode"
+      } || $ECHO "$RED[!] afl-fuzz is not working correctly with qemu_mode"
+
+      test -e ../libcompcov.so && {
+        $ECHO "$GREY[*] running afl-fuzz for qemu_mode libcompcov, this will take approx 10 seconds"
+        {
+          export AFL_PRELOAD=../libcompcov.so 
+          export AFL_COMPCOV_LEVEL=2
+          timeout -s KILL 10 ../afl-fuzz -Q -i in -o out -- ./test-compcov > /dev/null 2>&1
+        } > /dev/null 2>&1
+        test -n "$( ls out/queue/id:000002* 2> /dev/null )" && {
+          $ECHO "$GREEN[+] afl-fuzz is working correctly with qemu_mode libcompcov"
+        } || $ECHO "$RED[!] afl-fuzz is not working correctly with qemu_mode libcompcov"
+      } || $ECHO "$YELLOW[-] we cannot test qemu_mode libcompcov because it is not present"
+      rm -rf in out
+    } || $ECHO "$YELLOW[-] we cannot test afl-fuzz because we are missing the timeout command"
+  } || $ECHO "$RED[-] gcc compilation of test targets failed - what is going on??"
+  
+  $ECHO "$YELLOW[?] we need a test case for qemu_mode persistent mode"
+
+  rm -f test-instr test-compcov
+} || $ECHO "$YELLOW[-] qemu_mode is not compiled, cannot test"
+
+$ECHO "$GREY[*] all test cases completed.$RESET"
+
+# unicorn_mode ?