Merge branch 'incoming'

Conflicts:
	AUTHORS

61 files changed, 5074 insertions, 585 deletions

diff --git a/AUTHORS b/AUTHORS
index b1003dbd..3a7ceb32 100644
--- a/AUTHORS
+++ b/AUTHORS
@@ -48,4 +48,6 @@ a license to everyone to use it as detailed in LICENSE.)
 * Jasper St. Pierre <jstpierre@mecheye.net>
 * Manuel Schölling <manuel.schoelling@gmx.de>
 * Bruce Mitchener, Jr. <bruce.mitchener@gmail.com>
+* Michael Bishop <mbtyke@gmail.com>
+* Roger Braun <roger@rogerbraun.net>
 
diff --git a/emcc b/emcc
index ffce7363..2f4bad2b 100755
--- a/emcc
+++ b/emcc
@@ -119,11 +119,20 @@ if len(sys.argv) == 1:
   exit(1)
 
 if sys.argv[1] == '--version':
-  print '''emcc (Emscripten GCC-like replacement) 2.0
-Copyright (C) 2012 the Emscripten authors.
+  revision = '(unknown revision)'
+  here = os.getcwd()
+  os.chdir(shared.path_from_root())
+  try:
+    revision = execute(['git', 'show'], stdout=PIPE, stderr=PIPE)[0].split('\n')[0]
+  except:
+    pass
+  finally:
+    os.chdir(here)
+  print '''emcc (Emscripten GCC-like replacement) %s (%s)
+Copyright (C) 2013 the Emscripten authors (see AUTHORS.txt)
 This is free and open source software under the MIT license.
 There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
-  '''
+  ''' % (shared.EMSCRIPTEN_VERSION, revision)
   exit(0)
 elif sys.argv[1] == '--help':
   this = os.path.basename('em++' if os.environ.get('EMMAKEN_CXX') else 'emcc')
@@ -148,14 +157,19 @@ Options that are modified or new in %s include:
                            compiling to JavaScript, not to intermediate
                            bitcode.
   -O2                      As -O1, plus the relooper (loop recreation),
-                           plus closure compiler advanced opts, plus
-                           LLVM -O2 optimizations
-                           Warning: Compiling with this takes a long time!
+                           plus LLVM -O2 optimizations
   -O3                      As -O2, plus dangerous optimizations that may
-                           break the generated code! This is not
-                           recommended at all, see the wiki for more
-                           details (you can try -O2 and then add
-                           dangerous optimizations one by one).
+                           break the generated code! This adds
+
+                              -s INLINING_LIMIT=0
+                              -s DOUBLE_MODE=0
+                              -s PRECISE_I64_MATH=0
+                              --closure 1
+
+                           This is not recommended at all. A better idea
+                           is to try each of these separately on top of
+                           -O2 to see what works. See the wiki for more
+                           information.
 
   -s OPTION=VALUE          JavaScript code generation option passed
                            into the emscripten compiler. For the
@@ -177,6 +191,12 @@ Options that are modified or new in %s include:
                            the last compilation phase from bitcode to
                            JavaScript, or else we will remove it by
                            default in -O1 and above.
+                           In -O0, line numbers wil be shown in the
+                           generated code. In -O1 and above, the optimizer
+                           removes those comments. This flag does however
+                           have the effect of disabling anything that
+                           causes name mangling or minification (closure
+                           or the registerize pass).
 
   --typed-arrays <mode>    0: No typed arrays
                            1: Parallel typed arrays
@@ -193,8 +213,17 @@ Options that are modified or new in %s include:
                            (see --llvm-opts), setting this to 1 has no
                            effect.
 
-  --closure <on>           0: No closure compiler (default in -O0, -O1)
-                           1: Run closure compiler (default in -O2, -O3)
+  --closure <on>           0: No closure compiler (default in -O2 and below)
+                           1: Run closure compiler. This greatly reduces
+                           code size and may in some cases increase
+                           runtime speed (although the opposite can also
+                           occur). Note that it takes time to run, and
+                           may require some changes to the code. This
+                           is run by default in -O3.
+
+                           Note: If closure compiler hits an out-of-memory,
+                           try adjusting JAVA_HEAP_SIZE in the environment
+                           (for example, to 4096m for 4GB).
 
   --js-transform <cmd>     <cmd> will be called on the generated code
                            before it is optimized. This lets you modify
@@ -600,7 +629,8 @@ try:
   ignore_dynamic_linking = False
   shell_path = shared.path_from_root('src', 'shell.html')
   js_libraries = []
-  keep_debug = False
+  keep_llvm_debug = False
+  keep_js_debug = False
   bind = False
   jcache = False
   if use_cxx:
@@ -616,7 +646,8 @@ try:
   for i in range(len(newargs)):
     newargs[i] = newargs[i].strip() # On Windows Vista (and possibly others), excessive spaces in the command line leak into the items in this array, so trim e.g. 'foo.cpp ' -> 'foo.cpp'
     if newargs[i].startswith('-O'):
-      requested_level = newargs[i][2]
+      # Let -O default to -O2, which is what gcc does.
+      requested_level = newargs[i][2:] or '2'
       if requested_level == 's':
         print >> sys.stderr, 'emcc: warning: -Os is ignored (use -O0, -O1, -O2)'
       else:
@@ -667,7 +698,8 @@ try:
       newargs[i] = ''
       newargs[i+1] = ''
     elif newargs[i] == '-g':
-      keep_debug = True
+      keep_llvm_debug = True
+      keep_js_debug = True
     elif newargs[i] == '--bind':
       bind = True
       newargs[i] = ''
@@ -738,8 +770,9 @@ try:
 
   if llvm_opts is None: llvm_opts = LLVM_OPT_LEVEL[opt_level]
   if llvm_lto is None: llvm_lto = llvm_opts > 0
-  if closure is None: closure = 1 if opt_level >= 2 else 0
-  if opt_level <= 0: keep_debug = True # always keep debug in -O0
+  if opt_level <= 0: keep_llvm_debug = keep_js_debug = True # always keep debug in -O0
+  if opt_level > 0: keep_llvm_debug = False # JS optimizer wipes out llvm debug info from being visible
+  if closure is None and opt_level == 3: closure = True
 
   if DEBUG: start_time = time.time() # done after parsing arguments, which might affect debug state
 
@@ -879,7 +912,11 @@ try:
       shared.Settings.CORRECT_OVERFLOWS = 1
 
   if shared.Settings.CORRECT_SIGNS >= 2 or shared.Settings.CORRECT_OVERFLOWS >= 2 or shared.Settings.CORRECT_ROUNDINGS >= 2:
-    keep_debug = True # must keep debug info to do line-by-line operations 
+    keep_llvm_debug = True # must keep debug info to do line-by-line operations 
+
+  if (keep_llvm_debug or keep_js_debug) and closure:
+    print >> sys.stderr, 'emcc: warning: disabling closure because debug info was requested'
+    closure = False
 
   if minify_whitespace is None:
     minify_whitespace = closure # if closure is run, minify whitespace
@@ -984,7 +1021,7 @@ try:
     def create_libcxx():
       if DEBUG: print >> sys.stderr, 'emcc: building libcxx for cache'
       os = []
-      for src in ['algorithm.cpp', 'condition_variable.cpp', 'future.cpp', 'iostream.cpp', 'memory.cpp', 'random.cpp', 'stdexcept.cpp', 'system_error.cpp', 'utility.cpp', 'bind.cpp', 'debug.cpp', 'hash.cpp', 'mutex.cpp', 'string.cpp', 'thread.cpp', 'valarray.cpp', 'chrono.cpp', 'exception.cpp', 'ios.cpp', 'locale.cpp', 'regex.cpp', 'strstream.cpp', 'typeinfo.cpp']:
+      for src in ['algorithm.cpp', 'condition_variable.cpp', 'future.cpp', 'iostream.cpp', 'memory.cpp', 'random.cpp', 'stdexcept.cpp', 'system_error.cpp', 'utility.cpp', 'bind.cpp', 'debug.cpp', 'hash.cpp', 'mutex.cpp', 'string.cpp', 'thread.cpp', 'valarray.cpp', 'chrono.cpp', 'exception.cpp', 'ios.cpp', 'locale.cpp', 'regex.cpp', 'strstream.cpp']:
         o = in_temp(src + '.o')
         execute([shared.PYTHON, shared.EMXX, shared.path_from_root('system', 'lib', 'libcxx', src), '-o', o], stdout=stdout, stderr=stderr)
         os.append(o)
@@ -1003,7 +1040,7 @@ try:
     def create_libcxxabi():
       if DEBUG: print >> sys.stderr, 'emcc: building libcxxabi for cache'
       os = []
-      for src in ['private_typeinfo.cpp']:
+      for src in ['private_typeinfo.cpp', 'typeinfo.cpp']:
         o = in_temp(src + '.o')
         execute([shared.PYTHON, shared.EMXX, shared.path_from_root('system', 'lib', 'libcxxabi', 'src', src), '-o', o], stdout=stdout, stderr=stderr)
         os.append(o)
@@ -1017,29 +1054,32 @@ try:
     libcxxabi_symbols = filter(lambda symbol: symbol not in libc_symbols, libcxxabi_symbols)
     libcxxabi_symbols = set(libcxxabi_symbols)
 
-    force = False # If we have libcxx, we must force inclusion of libc, since libcxx uses new internally. Note: this is kind of hacky
-
+    # If we have libcxx, we must force inclusion of libc, since libcxx uses new internally. Note: this is kind of hacky
+    # Settings this in the environment will avoid checking dependencies and make building big projects a little faster
+    force = os.environ.get('EMCC_FORCE_STDLIBS')
+    has = need = None
     for name, create, fix, library_symbols in [('libcxx',    create_libcxx,    fix_libcxx,    libcxx_symbols),
                                                ('libcxxabi', create_libcxxabi, fix_libcxxabi, libcxxabi_symbols),
                                                ('libc',      create_libc,      fix_libc,      libc_symbols)]:
-      need = set()
-      has = set()
-      for temp_file in temp_files:
-        symbols = shared.Building.llvm_nm(temp_file)
-        for library_symbol in library_symbols:
-          if library_symbol in symbols.undefs:
-            need.add(library_symbol)
-          if library_symbol in symbols.defs:
-            has.add(library_symbol)
-      for haz in has: # remove symbols that are supplied by another of the inputs
-        if haz in need:
-          need.remove(haz)
-      if DEBUG: print >> sys.stderr, 'emcc: considering including %s: we need %s and have %s' % (name, str(need), str(has))
+      if not force:
+        need = set()
+        has = set()
+        for temp_file in temp_files:
+          symbols = shared.Building.llvm_nm(temp_file)
+          for library_symbol in library_symbols:
+            if library_symbol in symbols.undefs:
+              need.add(library_symbol)
+            if library_symbol in symbols.defs:
+              has.add(library_symbol)
+        for haz in has: # remove symbols that are supplied by another of the inputs
+          if haz in need:
+            need.remove(haz)
+        if DEBUG: print >> sys.stderr, 'emcc: considering including %s: we need %s and have %s' % (name, str(need), str(has))
       if force or len(need) > 0:
         # We need to build and link the library in
         if DEBUG: print >> sys.stderr, 'emcc: including %s' % name
         libfile = shared.Cache.get(name, create)
-        if len(has) > 0:
+        if has and len(has) > 0:
           # remove the symbols we do not need
           fixed = in_temp(uniquename(libfile)) + '.bc'
           shutil.copyfile(libfile, fixed)
@@ -1049,7 +1089,7 @@ try:
           libfile = fixed
         extra_files_to_link.append(libfile)
         force = True
-        if fix:
+        if fix and need:
           fix(need)
 
   # First, combine the bitcode files if there are several. We must also link if we have a singleton .a
@@ -1084,7 +1124,7 @@ try:
 
   # Optimize, if asked to
   if not LEAVE_INPUTS_RAW:
-    link_opts = [] if keep_debug else ['-strip-debug']
+    link_opts = [] if keep_llvm_debug else ['-strip-debug'] # remove LLVM debug info in -O1+, since the optimizer removes it anyhow
     if llvm_opts > 0:
       shared.Building.llvm_opt(in_temp(target_basename + '.bc'), llvm_opts)
       if DEBUG: save_intermediate('opt', 'bc')
@@ -1216,8 +1256,9 @@ try:
     if DEBUG: print >> sys.stderr, 'emcc: running closure'
     final = shared.Building.closure_compiler(final)
     if DEBUG: save_intermediate('closure')
-  elif shared.Settings.ASM_JS and shared.Settings.RELOOP:
-    js_optimizer_queue += ['registerize'] # we can't use closure in asm, but this does much of the same
+  elif shared.Settings.RELOOP and not closure and not keep_js_debug:
+    # do this if closure is not enabled (it gives similar speedups), and we do not need to keep debug info around
+    js_optimizer_queue += ['registerize']
 
   if opt_level >= 1:
     if DEBUG: print >> sys.stderr, 'emcc: running post-closure post-opts'
diff --git a/emscripten.py b/emscripten.py
index af762a21..b3e153c1 100755
--- a/emscripten.py
+++ b/emscripten.py
@@ -322,10 +322,7 @@ def emscript(infile, settings, outfile, libraries=[]):
     function_tables_defs = '\n'.join([info[0] for info in infos] + [info[1] for info in infos])
 
     asm_setup = ''
-    maths = ['Math.' + func for func in ['floor', 'abs', 'sqrt', 'pow', 'cos', 'sin', 'tan', 'acos', 'asin', 'atan', 'atan2', 'exp', 'log', 'ceil']]
-    if settings['USE_MATH_IMUL']:
-      maths += ['Math.imul']
-      asm_setup += 'if (!Math.imul) Math.imul = function(x, y) { return (x*y)|0 }; // # not a real polyfill since semantics not identical, but close and fairly fast\n'
+    maths = ['Math.' + func for func in ['floor', 'abs', 'sqrt', 'pow', 'cos', 'sin', 'tan', 'acos', 'asin', 'atan', 'atan2', 'exp', 'log', 'ceil', 'imul']]
     fundamentals = ['Math', 'Int8Array', 'Int16Array', 'Int32Array', 'Uint8Array', 'Uint16Array', 'Uint32Array', 'Float32Array', 'Float64Array']
     math_envs = ['Runtime.bitshift64', 'Math.min'] # TODO: move min to maths
     asm_setup += '\n'.join(['var %s = %s;' % (f.replace('.', '_'), f) for f in math_envs])
diff --git a/src/analyzer.js b/src/analyzer.js
index 1c53b76c..c930231f 100644
--- a/src/analyzer.js
+++ b/src/analyzer.js
@@ -223,6 +223,9 @@ function analyzer(data, sidePass) {
                   for (var i = 0; i < item.params.length; i++) {
                     if (item.params[i].type == 'i64') item.params[i].type = 'i32';
                   }
+                } else if (item.intertype == 'inttoptr') {
+                  var input = item.params[0];
+                  if (input.type == 'i64') input.type = 'i32'; // inttoptr can only care about 32 bits anyhow since pointers are 32-bit
                 }
                 if (isIllegalType(item.valueType) || isIllegalType(item.type)) {
                   isIllegal = true;
@@ -681,9 +684,9 @@ function analyzer(data, sidePass) {
                       params: [(signed && j + whole > sourceElements.length) ? signedKeepAlive : null],
                       type: 'i32',
                     };
-                    if (j == 0 && isUnsignedOp(value.op) && sourceBits < 32) {
+                    if (j == 0 && sourceBits < 32) {
                       // zext sign correction
-                      result.ident = makeSignOp(result.ident, 'i' + sourceBits, 'un', 1, 1);
+                      result.ident = makeSignOp(result.ident, 'i' + sourceBits, isUnsignedOp(value.op) ? 'un' : 're', 1, 1);
                     }
                     if (fraction != 0) {
                       var other = {
diff --git a/src/compiler.js b/src/compiler.js
index 25c306cf..0b43842e 100644
--- a/src/compiler.js
+++ b/src/compiler.js
@@ -160,12 +160,6 @@ if (SAFE_HEAP >= 2) {
   SAFE_HEAP_LINES = set(SAFE_HEAP_LINES); // for fast checking
 }
 
-if (PGO) { // by default, correct everything during PGO
-  CORRECT_SIGNS = CORRECT_SIGNS || 1;
-  CORRECT_OVERFLOWS = CORRECT_OVERFLOWS || 1;
-  CORRECT_ROUNDINGS = CORRECT_ROUNDINGS || 1;
-}
-
 EXPORTED_FUNCTIONS = set(EXPORTED_FUNCTIONS);
 EXPORTED_GLOBALS = set(EXPORTED_GLOBALS);
 EXCEPTION_CATCHING_WHITELIST = set(EXCEPTION_CATCHING_WHITELIST);
@@ -185,7 +179,7 @@ assert(!(!NAMED_GLOBALS && BUILD_AS_SHARED_LIB)); // shared libraries must have
 
 if (phase == 'pre') {
   if (!MICRO_OPTS || !RELOOP || ASSERTIONS || CHECK_SIGNS || CHECK_OVERFLOWS || INIT_STACK || INIT_HEAP ||
-      !SKIP_STACK_IN_SMALL || SAFE_HEAP || PGO || PROFILE || !DISABLE_EXCEPTION_CATCHING) {
+      !SKIP_STACK_IN_SMALL || SAFE_HEAP || !DISABLE_EXCEPTION_CATCHING) {
     print('// Note: Some Emscripten settings will significantly limit the speed of the generated code.');
   } else {
     print('// Note: For maximum-speed code, see "Optimizing Code" on the Emscripten wiki, https://github.com/kripken/emscripten/wiki/Optimizing-Code');
diff --git a/src/corruptionCheck.js b/src/corruptionCheck.js
new file mode 100644
index 00000000..315f5cf0
--- /dev/null
+++ b/src/corruptionCheck.js
@@ -0,0 +1,98 @@
+
+// See settings.js, CORRUPTION_CHECK
+
+var CorruptionChecker = {
+  BUFFER_FACTOR: Math.round({{{ CORRUPTION_CHECK }}}),
+
+  ptrs: {},
+  checks: 0,
+  checkFrequency: 1,
+
+  init: function() {
+    this.realMalloc = _malloc;
+    _malloc = Module['_malloc'] = this.malloc;
+
+    this.realFree = _free;
+    _free = Module['_free'] = this.free;
+
+    if (typeof _realloc != 'undefined') {
+      this.realRealloc = _realloc;
+      _realloc = Module['_realloc'] = this.realloc;
+    }
+
+    __ATEXIT__.push({ func: function() {
+      Module.printErr('No corruption detected, ran ' + CorruptionChecker.checks + ' checks.');
+    } });
+  },
+  malloc: function(size) {
+    if (size <= 0) size = 1; // malloc(0) sometimes happens - just allocate a larger area, no harm
+    CorruptionChecker.checkAll();
+    size = (size+7)&(~7);
+    var allocation = CorruptionChecker.realMalloc(size*(1+2*CorruptionChecker.BUFFER_FACTOR));
+    var ptr = allocation + size*CorruptionChecker.BUFFER_FACTOR;
+    assert(!CorruptionChecker.ptrs[ptr]);
+    CorruptionChecker.ptrs[ptr] = size;
+    CorruptionChecker.fillBuffer(allocation, size*CorruptionChecker.BUFFER_FACTOR);
+    CorruptionChecker.fillBuffer(allocation + size*(1+CorruptionChecker.BUFFER_FACTOR), size*CorruptionChecker.BUFFER_FACTOR);
+    //Module.printErr('malloc ' + size + ' ==> ' + [ptr, allocation]);
+    return ptr;
+  },
+  free: function(ptr) {
+    if (!ptr) return; // ok to free(NULL), does nothing
+    CorruptionChecker.checkAll();
+    var size = CorruptionChecker.ptrs[ptr];
+    //Module.printErr('free ' + ptr + ' of size ' + size);
+    assert(size);
+    var allocation = ptr - size*CorruptionChecker.BUFFER_FACTOR;
+    //Module.printErr('free ' + ptr + ' of size ' + size + ' and allocation ' + allocation);
+    delete CorruptionChecker.ptrs[ptr];
+    CorruptionChecker.realFree(allocation);
+  },
+  realloc: function(ptr, newSize) {
+    //Module.printErr('realloc ' + ptr + ' to size ' + newSize);
+    if (newSize <= 0) newSize = 1; // like in malloc
+    if (!ptr) return CorruptionChecker.malloc(newSize); // realloc(NULL, size) forwards to malloc according to the spec
+    var size = CorruptionChecker.ptrs[ptr];
+    assert(size);
+    var allocation = ptr - size*CorruptionChecker.BUFFER_FACTOR;
+    var newPtr = CorruptionChecker.malloc(newSize);
+    //Module.printErr('realloc ' + ptr + ' to size ' + newSize + ' is now ' + newPtr);
+    var newAllocation = newPtr + newSize*CorruptionChecker.BUFFER_FACTOR;
+    HEAPU8.set(HEAPU8.subarray(ptr, ptr + Math.min(size, newSize)), newPtr);
+    CorruptionChecker.free(ptr);
+    return newPtr;
+  },
+  canary: function(x) {
+    return (x&127) + 10;
+  },
+  fillBuffer: function(buffer, size) {
+    for (var x = buffer; x < buffer + size; x++) {
+      {{{ makeSetValue('x', 0, 'CorruptionChecker.canary(x)', 'i8') }}};
+    }
+  },
+  checkBuffer: function(buffer, size) {
+    for (var x = buffer; x < buffer + size; x++) {
+      if (({{{ makeGetValue('x', 0, 'i8') }}}&255) != CorruptionChecker.canary(x)) {
+        assert(0, 'Heap corruption detected!' + [x, buffer, size, {{{ makeGetValue('x', 0, 'i8') }}}&255, CorruptionChecker.canary(x)]);
+      }
+    }
+  },
+  checkPtr: function(ptr) {
+    var size = CorruptionChecker.ptrs[ptr];
+    assert(size);
+    var allocation = ptr - size*CorruptionChecker.BUFFER_FACTOR;
+    CorruptionChecker.checkBuffer(allocation, size*CorruptionChecker.BUFFER_FACTOR);
+    CorruptionChecker.checkBuffer(allocation + size*(1+CorruptionChecker.BUFFER_FACTOR), size*CorruptionChecker.BUFFER_FACTOR);
+  },
+  checkAll: function(force) {
+    CorruptionChecker.checks++;
+    if (!force && CorruptionChecker.checks % CorruptionChecker.checkFrequency != 0) return;
+    //Module.printErr('checking for corruption ' + (CorruptionChecker.checks/CorruptionChecker.checkFrequency));
+    for (var ptr in CorruptionChecker.ptrs) {
+      CorruptionChecker.checkPtr(ptr, false);
+    }
+  },
+};
+
+CorruptionChecker.init();
+
diff --git a/src/jsifier.js b/src/jsifier.js
index 761a5fec..4af522b4 100644
--- a/src/jsifier.js
+++ b/src/jsifier.js
@@ -478,8 +478,7 @@ function JSify(data, functionsOnly, givenFunctions) {
           ident = '_' + ident;
         }
         var depsText = (deps ? '\n' + deps.map(addFromLibrary).filter(function(x) { return x != '' }).join('\n') : '');
-        // redirected idents just need a var, but no value assigned to them - it would be unused
-        var contentText = isFunction ? snippet : ('var ' + ident + (redirectedIdent ? '' : '=' + snippet) + ';');
+        var contentText = isFunction ? snippet : ('var ' + ident + '=' + snippet + ';');
         if (ASM_JS) {
           var sig = LibraryManager.library[ident.substr(1) + '__sig'];
           if (isFunction && sig && LibraryManager.library[ident.substr(1) + '__asm']) {
@@ -508,7 +507,7 @@ function JSify(data, functionsOnly, givenFunctions) {
         item.JS = addFromLibrary(shortident);
       } else {
         item.JS = 'var ' + item.ident + '; // stub for ' + item.ident;
-        if (WARN_ON_UNDEFINED_SYMBOLS) {
+        if (WARN_ON_UNDEFINED_SYMBOLS || ASM_JS) { // always warn on undefs in asm, since it breaks validation
           warn('Unresolved symbol: ' + item.ident);
         }
       }
@@ -631,15 +630,6 @@ function JSify(data, functionsOnly, givenFunctions) {
         }
       }
 
-      if (PROFILE) {
-        func.JS += '  if (PROFILING) { '
-                +      'var __parentProfilingNode__ = PROFILING_NODE; PROFILING_NODE = PROFILING_NODE.children["' + func.ident + '"]; '
-                +      'if (!PROFILING_NODE) __parentProfilingNode__.children["' + func.ident + '"] = PROFILING_NODE = { time: 0, children: {}, calls: 0 };'
-                +      'PROFILING_NODE.calls++; '
-                +      'var __profilingStartTime__ = Date.now() '
-                +    '}\n';
-      }
-
       if (true) { // TODO: optimize away when not needed
         if (CLOSURE_ANNOTATIONS) func.JS += '/** @type {number} */';
         func.JS += '  var label = 0;\n';
@@ -1145,12 +1135,6 @@ function JSify(data, functionsOnly, givenFunctions) {
   });
   makeFuncLineActor('return', function(item) {
     var ret = RuntimeGenerator.stackExit(item.funcData.initialStack, item.funcData.otherStackAllocations) + ';\n';
-    if (PROFILE) {
-      ret += 'if (PROFILING) { '
-          +    'PROFILING_NODE.time += Date.now() - __profilingStartTime__; '
-          +    'PROFILING_NODE = __parentProfilingNode__ '
-          +  '}\n';
-    }
     if (LABEL_DEBUG && functionNameFilterTest(item.funcData.ident)) {
       ret += "Module.print(INDENT + 'Exiting: " + item.funcData.ident + "');\n"
           +  "INDENT = INDENT.substr(0, INDENT.length-2);\n";
@@ -1527,7 +1511,7 @@ function JSify(data, functionsOnly, givenFunctions) {
         print('// ASM_LIBRARY FUNCTIONS');
         function fix(f) { // fix indenting to not confuse js optimizer
           f = f.substr(f.indexOf('f')); // remove initial spaces before 'function'
-          f = f.substr(0, f.lastIndexOf('\n')+1); // remove spaces and last }
+          f = f.substr(0, f.lastIndexOf('\n')+1); // remove spaces and last }  XXX assumes function has multiple lines
           return f + '}'; // add unindented } to match function
         }
         print(asmLibraryFunctions.map(fix).join('\n'));
@@ -1547,6 +1531,10 @@ function JSify(data, functionsOnly, givenFunctions) {
     // This is the main 'post' pass. Print out the generated code that we have here, together with the
     // rest of the output that we started to print out earlier (see comment on the
     // "Final shape that will be created").
+    if (CORRUPTION_CHECK) {
+      assert(!ASM_JS); // cannot monkeypatch asm!
+      print(processMacros(read('corruptionCheck.js')));
+    }
     if (PRECISE_I64_MATH && Types.preciseI64MathUsed) {
       print(read('long.js'));
     } else {
diff --git a/src/library.js b/src/library.js
index 1e026937..d0f73fdd 100644
--- a/src/library.js
+++ b/src/library.js
@@ -624,7 +624,12 @@ LibraryManager.library = {
   // dirent.h
   // ==========================================================================
 
-  __dirent_struct_layout: Runtime.generateStructInfo(['d_ino', 'd_name', 'd_off', 'd_reclen', 'd_type'], '%struct.dirent'),
+  __dirent_struct_layout: Runtime.generateStructInfo([
+    ['i32', 'd_ino'],
+    ['b1024', 'd_name'],
+    ['i32', 'd_off'],
+    ['i32', 'd_reclen'],
+    ['i32', 'd_type']]),
   opendir__deps: ['$FS', '__setErrNo', '$ERRNO_CODES', '__dirent_struct_layout'],
   opendir: function(dirname) {
     // DIR *opendir(const char *dirname);
@@ -786,7 +791,9 @@ LibraryManager.library = {
   // utime.h
   // ==========================================================================
 
-  __utimbuf_struct_layout: Runtime.generateStructInfo(['actime', 'modtime'], '%struct.utimbuf'),
+  __utimbuf_struct_layout: Runtime.generateStructInfo([
+    ['i32', 'actime'],
+    ['i32', 'modtime']]),
   utime__deps: ['$FS', '__setErrNo', '$ERRNO_CODES', '__utimbuf_struct_layout'],
   utime: function(path, times) {
     // int utime(const char *path, const struct utimbuf *times);
@@ -877,23 +884,23 @@ LibraryManager.library = {
   // ==========================================================================
 
   __stat_struct_layout: Runtime.generateStructInfo([
-    'st_dev',
-    'st_ino',
-    'st_mode',
-    'st_nlink',
-    'st_uid',
-    'st_gid',
-    'st_rdev',
-    'st_size',
-    'st_atime',
-    'st_spare1',
-    'st_mtime',
-    'st_spare2',
-    'st_ctime',
-    'st_spare3',
-    'st_blksize',
-    'st_blocks',
-    'st_spare4'], '%struct.stat'),
+    ['i32', 'st_dev'],
+    ['i32', 'st_ino'],
+    ['i32', 'st_mode'],
+    ['i32', 'st_nlink'],
+    ['i32', 'st_uid'],
+    ['i32', 'st_gid'],
+    ['i32', 'st_rdev'],
+    ['i32', 'st_size'],
+    ['i32', 'st_atime'],
+    ['i32', 'st_spare1'],
+    ['i32', 'st_mtime'],
+    ['i32', 'st_spare2'],
+    ['i32', 'st_ctime'],
+    ['i32', 'st_spare3'],
+    ['i32', 'st_blksize'],
+    ['i32', 'st_blocks'],
+    ['i32', 'st_spare4']]),
   stat__deps: ['$FS', '__stat_struct_layout'],
   stat: function(path, buf, dontResolveLastLink) {
     // http://pubs.opengroup.org/onlinepubs/7908799/xsh/stat.html
@@ -1065,17 +1072,17 @@ LibraryManager.library = {
   // ==========================================================================
 
   __statvfs_struct_layout: Runtime.generateStructInfo([
-    'f_bsize',
-    'f_frsize',
-    'f_blocks',
-    'f_bfree',
-    'f_bavail',
-    'f_files',
-    'f_ffree',
-    'f_favail',
-    'f_fsid',
-    'f_flag',
-    'f_namemax'], '%struct.statvfs'),
+    ['i32', 'f_bsize'],
+    ['i32', 'f_frsize'],
+    ['i32', 'f_blocks'],
+    ['i32', 'f_bfree'],
+    ['i32', 'f_bavail'],
+    ['i32', 'f_files'],
+    ['i32', 'f_ffree'],
+    ['i32', 'f_favail'],
+    ['i32', 'f_fsid'],
+    ['i32', 'f_flag'],
+    ['i32', 'f_namemax']]),
   statvfs__deps: ['$FS', '__statvfs_struct_layout'],
   statvfs: function(path, buf) {
     // http://pubs.opengroup.org/onlinepubs/7908799/xsh/stat.html
@@ -1110,12 +1117,12 @@ LibraryManager.library = {
   // ==========================================================================
 
   __flock_struct_layout: Runtime.generateStructInfo([
-    'l_type',
-    'l_whence',
-    'l_start',
-    'l_len',
-    'l_pid',
-    'l_xxx'], '%struct.flock'),
+    ['i16', 'l_type'],
+    ['i16', 'l_whence'],
+    ['i32', 'l_start'],
+    ['i32', 'l_len'],
+    ['i16', 'l_pid'],
+    ['i16', 'l_xxx']]),
   open__deps: ['$FS', '__setErrNo', '$ERRNO_CODES', '__dirent_struct_layout'],
   open: function(path, oflag, varargs) {
     // int open(const char *path, int oflag, ...);
@@ -1336,7 +1343,10 @@ LibraryManager.library = {
   // poll.h
   // ==========================================================================
 
-  __pollfd_struct_layout: Runtime.generateStructInfo(['fd', 'events', 'revents'], '%struct.pollfd'),
+  __pollfd_struct_layout: Runtime.generateStructInfo([
+    ['i32', 'fd'],
+    ['i16', 'events'],
+    ['i16', 'revents']]),
   poll__deps: ['$FS', '__pollfd_struct_layout'],
   poll: function(fds, nfds, timeout) {
     // int poll(struct pollfd fds[], nfds_t nfds, int timeout);
@@ -2481,6 +2491,17 @@ LibraryManager.library = {
         continue;
       }
 
+      // TODO: Support strings like "%5c" etc.
+      if (format[formatIndex] === '%' && format[formatIndex+1] == 'c') {
+        var argPtr = {{{ makeGetValue('varargs', 'argIndex', 'void*') }}};
+        argIndex += Runtime.getNativeFieldSize('void*');
+        fields++;
+        next = get();
+        {{{ makeSetValue('argPtr', 0, 'next', 'i8') }}}
+        formatIndex += 2;
+        continue;
+      }
+
       // remove whitespace
       while (1) {
         next = get();
@@ -4202,6 +4223,8 @@ LibraryManager.library = {
     return 1;
   },
 
+  arc4random: 'rand',
+
   // ==========================================================================
   // string.h
   // ==========================================================================
@@ -4502,11 +4525,16 @@ LibraryManager.library = {
     return 0;
   },
 
+  memcmp__asm: 'true',
+  memcmp__sig: 'iiii',
   memcmp: function(p1, p2, num) {
-    for (var i = 0; i < num; i++) {
-      var v1 = {{{ makeGetValue('p1', 'i', 'i8', 0, 1) }}};
-      var v2 = {{{ makeGetValue('p2', 'i', 'i8', 0, 1) }}};
-      if (v1 != v2) return v1 > v2 ? 1 : -1;
+    p1 = p1|0; p2 = p2|0; num = num|0;
+    var i = 0, v1 = 0, v2 = 0;
+    while ((i|0) < (num|0)) {
+      var v1 = {{{ makeGetValueAsm('p1', 'i', 'i8', true) }}};
+      var v2 = {{{ makeGetValueAsm('p2', 'i', 'i8', true) }}};
+      if ((v1|0) != (v2|0)) return ((v1|0) > (v2|0) ? 1 : -1)|0;
+      i = (i+1)|0;
     }
     return 0;
   },
@@ -4607,10 +4635,8 @@ LibraryManager.library = {
 
   __strtok_state: 0,
   strtok__deps: ['__strtok_state', 'strtok_r'],
+  strtok__postset: '___strtok_state = Runtime.staticAlloc(4);',
   strtok: function(s, delim) {
-    if (!___strtok_state) {
-      ___strtok_state = _malloc(4);
-    }
     return _strtok_r(s, delim, ___strtok_state);
   },
 
@@ -5070,6 +5096,7 @@ LibraryManager.library = {
   },
 
   __cxa_call_unexpected: function(exception) {
+    Module.printErr('Unexpected exception thrown, this is not properly supported - aborting');
     ABORT = true;
     throw exception;
   },
@@ -5619,11 +5646,11 @@ LibraryManager.library = {
   // ==========================================================================
 
   __utsname_struct_layout: Runtime.generateStructInfo([
-    'sysname',
-    'nodename',
-    'release',
-    'version',
-    'machine'], '%struct.utsname'),
+	  ['b32', 'sysname'],
+	  ['b32', 'nodename'],
+	  ['b32', 'release'],
+	  ['b32', 'version'],
+	  ['b32', 'machine']]),
   uname__deps: ['__utsname_struct_layout'],
   uname: function(name) {
     // int uname(struct utsname *name);
@@ -5823,17 +5850,17 @@ LibraryManager.library = {
   },
 
   __tm_struct_layout: Runtime.generateStructInfo([
-    'tm_sec',
-    'tm_min',
-    'tm_hour',
-    'tm_mday',
-    'tm_mon',
-    'tm_year',
-    'tm_wday',
-    'tm_yday',
-    'tm_isdst',
-    'tm_gmtoff',
-    'tm_zone'], '%struct.tm'),
+    ['i32', 'tm_sec'],
+    ['i32', 'tm_min'],
+    ['i32', 'tm_hour'],
+    ['i32', 'tm_mday'],
+    ['i32', 'tm_mon'],
+    ['i32', 'tm_year'],
+    ['i32', 'tm_wday'],
+    ['i32', 'tm_yday'],
+    ['i32', 'tm_isdst'],
+    ['i32', 'tm_gmtoff'],
+    ['i32', 'tm_zone']]),
   // Statically allocated time struct.
   __tm_current: 'allocate({{{ Runtime.QUANTUM_SIZE }}}*26, "i8", ALLOC_STACK)',
   // Statically allocated timezone strings.
@@ -6032,7 +6059,9 @@ LibraryManager.library = {
   // sys/time.h
   // ==========================================================================
 
-  __timespec_struct_layout: Runtime.generateStructInfo(['tv_sec', 'tv_nsec'], '%struct.timespec'),
+  __timespec_struct_layout: Runtime.generateStructInfo([
+    ['i32', 'tv_sec'],
+    ['i32', 'tv_nsec']]),
   // TODO: Implement these for real.
   clock_gettime__deps: ['__timespec_struct_layout'],
   clock_gettime: function(clk_id, tp) {
@@ -6089,10 +6118,10 @@ LibraryManager.library = {
   // ==========================================================================
 
   __tms_struct_layout: Runtime.generateStructInfo([
-    'tms_utime',
-    'tms_stime',
-    'tms_cutime',
-    'tms_cstime'], '%struct.tms'),
+    ['i32', 'tms_utime'],
+    ['i32', 'tms_stime'],
+    ['i32', 'tms_cutime'],
+    ['i32', 'tms_cstime']]),
   times__deps: ['__tms_struct_layout', 'memset'],
   times: function(buffer) {
     // clock_t times(struct tms *buffer);
@@ -6614,7 +6643,9 @@ LibraryManager.library = {
   // ==========================================================================
 
   // TODO: Implement for real.
-  __rlimit_struct_layout: Runtime.generateStructInfo(['rlim_cur', 'rlim_max'], '%struct.rlimit'),
+  __rlimit_struct_layout: Runtime.generateStructInfo([
+    ['i32', 'rlim_cur'],
+    ['i32', 'rlim_max']]),
   getrlimit__deps: ['__rlimit_struct_layout'],
   getrlimit: function(resource, rlp) {
     // int getrlimit(int resource, struct rlimit *rlp);
@@ -6630,22 +6661,22 @@ LibraryManager.library = {
 
   // TODO: Implement for real. We just do time used, and no useful data
   __rusage_struct_layout: Runtime.generateStructInfo([
-    'ru_utime',
-    'ru_stime',
-    'ru_maxrss',
-    'ru_ixrss',
-    'ru_idrss',
-    'ru_isrss',
-    'ru_minflt',
-    'ru_majflt',
-    'ru_nswap',
-    'ru_inblock',
-    'ru_oublock',
-    'ru_msgsnd',
-    'ru_msgrcv',
-    'ru_nsignals',
-    'ru_nvcsw',
-    'ru_nivcsw'], '%struct.rusage'),
+    ['i64', 'ru_utime'],
+    ['i64', 'ru_stime'],
+    ['i32', 'ru_maxrss'],
+    ['i32', 'ru_ixrss'],
+    ['i32', 'ru_idrss'],
+    ['i32', 'ru_isrss'],
+    ['i32', 'ru_minflt'],
+    ['i32', 'ru_majflt'],
+    ['i32', 'ru_nswap'],
+    ['i32', 'ru_inblock'],
+    ['i32', 'ru_oublock'],
+    ['i32', 'ru_msgsnd'],
+    ['i32', 'ru_msgrcv'],
+    ['i32', 'ru_nsignals'],
+    ['i32', 'ru_nvcsw'],
+    ['i32', 'ru_nivcsw']]),
   getrusage__deps: ['__rusage_struct_layout'],
   getrusage: function(resource, rlp) {
     // %struct.timeval = type { i32, i32 }
@@ -7289,36 +7320,6 @@ LibraryManager.library = {
   emscripten_random: function() {
     return Math.random();
   },
-
-  $Profiling: {
-    max_: 0,
-    times: null,
-    invalid: 0,
-    dump: function() {
-      if (Profiling.invalid) {
-        Module.printErr('Invalid # of calls to Profiling begin and end!');
-        return;
-      }
-      Module.printErr('Profiling data:')
-      for (var i = 0; i < Profiling.max_; i++) {
-        Module.printErr('Block ' + i + ': ' + Profiling.times[i]);
-      }
-    }
-  },
-  EMSCRIPTEN_PROFILE_INIT__deps: ['$Profiling'],
-  EMSCRIPTEN_PROFILE_INIT: function(max_) {
-    Profiling.max_ = max_;
-    Profiling.times = new Array(max_);
-    for (var i = 0; i < max_; i++) Profiling.times[i] = 0;
-  },
-  EMSCRIPTEN_PROFILE_BEGIN__inline: function(id) {
-    return 'Profiling.times[' + id + '] -= Date.now();'
-         + 'Profiling.invalid++;'
-  },
-  EMSCRIPTEN_PROFILE_END__inline: function(id) {
-    return 'Profiling.times[' + id + '] += Date.now();'
-         + 'Profiling.invalid--;'
-  }
 };
 
 function autoAddDeps(object, name) {
diff --git a/src/library_browser.js b/src/library_browser.js
index d16fbc0b..5b19a360 100644
--- a/src/library_browser.js
+++ b/src/library_browser.js
@@ -68,6 +68,7 @@ mergeInto(LibraryManager.library, {
       function getMimetype(name) {
         return {
           'jpg': 'image/jpeg',
+          'jpeg': 'image/jpeg',
           'png': 'image/png',
           'bmp': 'image/bmp',
           'ogg': 'audio/ogg',
@@ -81,7 +82,7 @@ mergeInto(LibraryManager.library, {
 
       var imagePlugin = {};
       imagePlugin['canHandle'] = function(name) {
-        return name.substr(-4) in { '.jpg': 1, '.png': 1, '.bmp': 1 };
+        return !Module.noImageDecoding && /\.(jpg|jpeg|png|bmp)$/.exec(name);
       };
       imagePlugin['handle'] = function(byteArray, name, onload, onerror) {
         var b = null;
@@ -123,7 +124,7 @@ mergeInto(LibraryManager.library, {
 
       var audioPlugin = {};
       audioPlugin['canHandle'] = function(name) {
-        return name.substr(-4) in { '.ogg': 1, '.wav': 1, '.mp3': 1 };
+        return !Module.noAudioDecoding && name.substr(-4) in { '.ogg': 1, '.wav': 1, '.mp3': 1 };
       };
       audioPlugin['handle'] = function(byteArray, name, onload, onerror) {
         var done = false;
@@ -200,8 +201,18 @@ mergeInto(LibraryManager.library, {
         return null;
       }
 #endif
+      var ctx;
       try {
-        var ctx = canvas.getContext(useWebGL ? 'experimental-webgl' : '2d');
+        if (useWebGL) {
+          ctx = canvas.getContext('experimental-webgl', {
+            alpha: false,
+#if GL_TESTING
+            preserveDrawingBuffer: true
+#endif
+          });
+        } else {
+          ctx = canvas.getContext('2d');
+        }
         if (!ctx) throw ':(';
       } catch (e) {
         Module.print('Could not create canvas - ' + e);
@@ -262,7 +273,7 @@ mergeInto(LibraryManager.library, {
       }
       return ctx;
     },
-
+    destroyContext: function(canvas, useWebGL, setInModule) {},
     requestFullScreen: function() {
       var canvas = Module['canvas'];
       function fullScreenChange() {
diff --git a/src/library_egl.js b/src/library_egl.js
index a9eb37dd..271ea29e 100644
--- a/src/library_egl.js
+++ b/src/library_egl.js
@@ -83,6 +83,17 @@ var LibraryEGL = {
     return 1;
   },
   
+// EGLAPI EGLBoolean EGLAPIENTRY eglTerminate(EGLDisplay dpy);
+  eglTerminate: function(display) {
+    if (display != 62000 /* Magic ID for Emscripten 'default display' */) {
+      EGL.setErrorCode(0x3008 /* EGL_BAD_DISPLAY */);
+      return 0;
+    }
+    // TODO: Tear down EGL here. Currently a no-op since we don't need to actually do anything here for the browser.
+    EGL.setErrorCode(0x3000 /* EGL_SUCCESS */);
+    return 1;
+  },
+
   // EGLAPI EGLBoolean EGLAPIENTRY eglGetConfigs(EGLDisplay dpy, EGLConfig *configs, EGLint config_size, EGLint *num_config);
   eglGetConfigs: function(display, configs, config_size, numConfigs) { 
     return EGL.chooseConfig(display, 0, configs, config_size, numConfigs);
@@ -225,6 +236,20 @@ var LibraryEGL = {
     return 62006; /* Magic ID for Emscripten 'default surface' */
   },
 
+  // EGLAPI EGLBoolean EGLAPIENTRY eglDestroySurface(EGLDisplay display, EGLSurface surface);
+  eglDestroySurface: function(display, surface) { 
+    if (display != 62000 /* Magic ID for Emscripten 'default display' */) {
+      EGL.setErrorCode(0x3008 /* EGL_BAD_DISPLAY */);
+      return 0; 
+    }
+    if (surface != 62006 /* Magic ID for the only EGLSurface supported by Emscripten */) {
+      EGL.setErrorCode(0x300D /* EGL_BAD_SURFACE */);
+      return 1;
+    }
+    EGL.setErrorCode(0x3000 /* EGL_SUCCESS */);
+    return 1; /* Magic ID for Emscripten 'default surface' */
+  },
+
   eglCreateContext__deps: ['glutCreateWindow', '$GL'],
   
   // EGLAPI EGLContext EGLAPIENTRY eglCreateContext(EGLDisplay dpy, EGLConfig config, EGLContext share_context, const EGLint *attrib_list);
@@ -234,7 +259,21 @@ var LibraryEGL = {
       return 0;
     }
 
-    _glutCreateWindow();
+    EGL.windowID = _glutCreateWindow();
+    EGL.setErrorCode(0x3000 /* EGL_SUCCESS */);
+    return 62004; // Magic ID for Emscripten EGLContext
+  },
+
+  eglDestroyContext__deps: ['glutDestroyWindow', '$GL'],
+  
+  // EGLAPI EGLBoolean EGLAPIENTRY eglDestroyContext(EGLDisplay dpy, EGLContext context);
+  eglDestroyContext: function(display, context) {
+    if (display != 62000 /* Magic ID for Emscripten 'default display' */) {
+      EGL.setErrorCode(0x3008 /* EGL_BAD_DISPLAY */);
+      return 0;
+    }
+
+    _glutDestroyWindow(EGL.windowID);
     EGL.setErrorCode(0x3000 /* EGL_SUCCESS */);
     return 62004; // Magic ID for Emscripten EGLContext
   },
diff --git a/src/library_gl.js b/src/library_gl.js
index f7966dab..027b2841 100644
--- a/src/library_gl.js
+++ b/src/library_gl.js
@@ -408,7 +408,19 @@ var LibraryGL = {
   },
 
   glReadPixels: function(x, y, width, height, format, type, pixels) {
-    Module.ctx.readPixels(x, y, width, height, format, type, HEAPU8.subarray(pixels));
+    assert(type == 0x1401 /* GL_UNSIGNED_BYTE */);
+    var sizePerPixel;
+    switch (format) {
+      case 0x1907 /* GL_RGB */:
+        sizePerPixel = 3;
+        break;
+      case 0x1908 /* GL_RGBA */:
+        sizePerPixel = 4;
+        break;
+      default: throw 'unsupported glReadPixels format';
+    }
+    var totalSize = width*height*sizePerPixel;
+    Module.ctx.readPixels(x, y, width, height, format, type, HEAPU8.subarray(pixels, pixels + totalSize));
   },
 
   glBindTexture: function(target, texture) {
@@ -908,11 +920,9 @@ var LibraryGL = {
   },
 
   glIsProgram: function(program) {
-    var fb = GL.programs[program];
-    if (typeof(fb) == 'undefined') {
-      return 0;
-    }
-    return Module.ctx.isProgram(fb);
+    var program = GL.programs[program];
+    if (!program) return 0;
+    return Module.ctx.isProgram(program);
   },
 
   glBindAttribLocation__sig: 'viii',
@@ -983,6 +993,11 @@ var LibraryGL = {
     fogMode: 0x0800, // GL_EXP
     fogEnabled: false,
 
+    // VAO support
+    vaos: [],
+    currentVao: null,
+    enabledVertexAttribArrays: {}, // helps with vao cleanups
+
     init: function() {
       GLEmulation.fogColor = new Float32Array(4);
 
@@ -1008,6 +1023,11 @@ var LibraryGL = {
         if (cap == 0x0B60 /* GL_FOG */) {
           GLEmulation.fogEnabled = true;
           return;
+        } else if (cap == 0x0de1 /* GL_TEXTURE_2D */) {
+          // XXX not according to spec, and not in desktop GL, but works in some GLES1.x apparently, so support
+          // it by forwarding to glEnableClientState
+          _glEnableClientState(cap);
+          return;
         } else if (!(cap in validCapabilities)) {
           return;
         }
@@ -1018,6 +1038,11 @@ var LibraryGL = {
         if (cap == 0x0B60 /* GL_FOG */) {
           GLEmulation.fogEnabled = false;
           return;
+        } else if (cap == 0x0de1 /* GL_TEXTURE_2D */) {
+          // XXX not according to spec, and not in desktop GL, but works in some GLES1.x apparently, so support
+          // it by forwarding to glDisableClientState
+          _glDisableClientState(cap);
+          return;
         } else if (!(cap in validCapabilities)) {
           return;
         }
@@ -1052,6 +1077,51 @@ var LibraryGL = {
             return;
           }
           case 0x8871: pname = Module.ctx.MAX_COMBINED_TEXTURE_IMAGE_UNITS /* close enough */; break; // GL_MAX_TEXTURE_COORDS
+          case 0x807A: { // GL_VERTEX_ARRAY_SIZE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.VERTEX];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.size : 0', 'i32') }}};
+            return;
+          }
+          case 0x807B: { // GL_VERTEX_ARRAY_TYPE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.VERTEX];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.type : 0', 'i32') }}};
+            return;
+          }
+          case 0x807C: { // GL_VERTEX_ARRAY_STRIDE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.VERTEX];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.stride : 0', 'i32') }}};
+            return;
+          }
+          case 0x8081: { // GL_COLOR_ARRAY_SIZE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.COLOR];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.size : 0', 'i32') }}};
+            return;
+          }
+          case 0x8082: { // GL_COLOR_ARRAY_TYPE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.COLOR];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.type : 0', 'i32') }}};
+            return;
+          }
+          case 0x8083: { // GL_COLOR_ARRAY_STRIDE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.COLOR];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.stride : 0', 'i32') }}};
+            return;
+          }
+          case 0x8088: { // GL_TEXTURE_COORD_ARRAY_SIZE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.TEXTURE0];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.size : 0', 'i32') }}};
+            return;
+          }
+          case 0x8089: { // GL_TEXTURE_COORD_ARRAY_TYPE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.TEXTURE0];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.type : 0', 'i32') }}};
+            return;
+          }
+          case 0x808A: { // GL_TEXTURE_COORD_ARRAY_STRIDE
+            var attribute = GLImmediate.clientAttributes[GLImmediate.TEXTURE0];
+            {{{ makeSetValue('params', '0', 'attribute ? attribute.stride : 0', 'i32') }}};
+            return;
+          }
         }
         glGetIntegerv(pname, params);
       };
@@ -1267,8 +1337,13 @@ var LibraryGL = {
         glBindBuffer(target, buffer);
         if (target == Module.ctx.ARRAY_BUFFER) {
           GL.currArrayBuffer = buffer;
+          if (GLEmulation.currentVao) {
+            assert(GLEmulation.currentVao.arrayBuffer == buffer || GLEmulation.currentVao.arrayBuffer == 0 || buffer == 0, 'TODO: support for multiple array buffers in vao');
+            GLEmulation.currentVao.arrayBuffer = buffer;
+          }
         } else if (target == Module.ctx.ELEMENT_ARRAY_BUFFER) {
           GL.currElementArrayBuffer = buffer;
+          if (GLEmulation.currentVao) GLEmulation.currentVao.elementArrayBuffer = buffer;
         }
       };
 
@@ -1312,6 +1387,28 @@ var LibraryGL = {
         }
         glHint(target, mode);
       };
+
+      var glEnableVertexAttribArray = _glEnableVertexAttribArray;
+      _glEnableVertexAttribArray = function(index) {
+        glEnableVertexAttribArray(index);
+        GLEmulation.enabledVertexAttribArrays[index] = 1;
+        if (GLEmulation.currentVao) GLEmulation.currentVao.enabledVertexAttribArrays[index] = 1;
+      };
+
+      var glDisableVertexAttribArray = _glDisableVertexAttribArray;
+      _glDisableVertexAttribArray = function(index) {
+        glDisableVertexAttribArray(index);
+        delete GLEmulation.enabledVertexAttribArrays[index];
+        if (GLEmulation.currentVao) delete GLEmulation.currentVao.enabledVertexAttribArrays[index];
+      };
+
+      var glVertexAttribPointer = _glVertexAttribPointer;
+      _glVertexAttribPointer = function(index, size, type, normalized, stride, pointer) {
+        glVertexAttribPointer(index, size, type, normalized, stride, pointer);
+        if (GLEmulation.currentVao) { // TODO: avoid object creation here? likely not hot though
+          GLEmulation.currentVao.vertexAttribPointers[index] = [index, size, type, normalized, stride, pointer];
+        }
+      };
     },
 
     getProcAddress: function(name) {
@@ -1375,6 +1472,9 @@ var LibraryGL = {
         case 'glIsFramebuffer': ret = {{{ Functions.getIndex('_glIsFramebuffer', true) }}}; break;
         case 'glCheckFramebufferStatus': ret = {{{ Functions.getIndex('_glCheckFramebufferStatus', true) }}}; break;
         case 'glRenderbufferStorage': ret = {{{ Functions.getIndex('_glRenderbufferStorage', true) }}}; break;
+        case 'glGenVertexArrays': ret = {{{ Functions.getIndex('_glGenVertexArrays', true) }}}; break;
+        case 'glDeleteVertexArrays': ret = {{{ Functions.getIndex('_glDeleteVertexArrays', true) }}}; break;
+        case 'glBindVertexArray': ret = {{{ Functions.getIndex('_glBindVertexArray', true) }}}; break;
       }
       if (!ret) Module.printErr('WARNING: getProcAddress failed for ' + name);
       return ret;
@@ -1427,6 +1527,20 @@ var LibraryGL = {
     assert(id == 0);
   },
 
+  glGetPointerv: function(name, p) {
+    var attribute;
+    switch(name) {
+      case 0x808E: // GL_VERTEX_ARRAY_POINTER
+        attribute = GLImmediate.clientAttributes[GLImmediate.VERTEX]; break;
+      case 0x8090: // GL_COLOR_ARRAY_POINTER
+        attribute = GLImmediate.clientAttributes[GLImmediate.COLOR]; break;
+      case 0x8092: // GL_TEXTURE_COORD_ARRAY_POINTER
+        attribute = GLImmediate.clientAttributes[GLImmediate.TEXTURE0]; break;
+      default: throw 'TODO: glGetPointerv for ' + name;
+    }
+    {{{ makeSetValue('p', '0', 'attribute ? attribute.pointer : 0', 'i32') }}};
+  },
+
   // GL Immediate mode
 
   $GLImmediate__postset: 'GL.immediate.setupFuncs(); Browser.moduleContextCreatedCallbacks.push(function() { GL.immediate.init() });',
@@ -1501,7 +1615,7 @@ var LibraryGL = {
     },
 
     // Temporary buffers
-    MAX_TEMP_BUFFER_SIZE: 2*1024*1024,
+    MAX_TEMP_BUFFER_SIZE: {{{ GL_MAX_TEMP_BUFFER_SIZE }}},
     tempBufferIndexLookup: null,
     tempVertexBuffers: null,
     tempIndexBuffers: null,
@@ -1784,27 +1898,35 @@ var LibraryGL = {
           }
 
           // If the array buffer is unchanged and the renderer as well, then we can avoid all the work here
-          // XXX We use some heuristics here, and this may not work in all cases. Try disabling this if you
-          // have odd glitches (by setting canSkip always to 0, or even cleaning up the renderer right
-          // after rendering)
+          // XXX We use some heuristics here, and this may not work in all cases. Try disabling GL_UNSAFE_OPTS if you
+          // have odd glitches
+#if GL_UNSAFE_OPTS
           var lastRenderer = GL.immediate.lastRenderer;
           var canSkip = this == lastRenderer &&
                         arrayBuffer == GL.immediate.lastArrayBuffer &&
                         (GL.currProgram || this.program) == GL.immediate.lastProgram &&
                         !GL.immediate.matricesModified;
           if (!canSkip && lastRenderer) lastRenderer.cleanup();
+#endif
           if (!GL.currArrayBuffer) {
             // Bind the array buffer and upload data after cleaning up the previous renderer
+#if GL_UNSAFE_OPTS
+            // Potentially unsafe, since lastArrayBuffer might not reflect the true array buffer in code that mixes immediate/non-immediate
             if (arrayBuffer != GL.immediate.lastArrayBuffer) {
+#endif
               Module.ctx.bindBuffer(Module.ctx.ARRAY_BUFFER, arrayBuffer);
+#if GL_UNSAFE_OPTS
             }
+#endif
             Module.ctx.bufferSubData(Module.ctx.ARRAY_BUFFER, start, GL.immediate.vertexData.subarray(start >> 2, end >> 2));
           }
+#if GL_UNSAFE_OPTS
           if (canSkip) return;
           GL.immediate.lastRenderer = this;
           GL.immediate.lastArrayBuffer = arrayBuffer;
           GL.immediate.lastProgram = GL.currProgram || this.program;
           GL.immediate.matricesModified = false;
+#endif
 
           if (!GL.currProgram) {
             Module.ctx.useProgram(this.program);
@@ -1880,9 +2002,11 @@ var LibraryGL = {
             Module.ctx.bindBuffer(Module.ctx.ARRAY_BUFFER, null);
           }
 
+#if GL_UNSAFE_OPTS
           GL.immediate.lastRenderer = null;
           GL.immediate.lastArrayBuffer = null;
           GL.immediate.lastProgram = null;
+#endif
           GL.immediate.matricesModified = true;
         }
       };
@@ -1997,22 +2121,61 @@ var LibraryGL = {
 #endif
         if (attribute.stride) stride = attribute.stride;
       }
-
-      var bytes = 0;
-      for (var i = 0; i < attributes.length; i++) {
-        var attribute = attributes[i];
-        if (!attribute) break;
-        attribute.offset = attribute.pointer - start;
-        if (attribute.offset > bytes) { // ensure we start where we should
-          assert((attribute.offset - bytes)%4 == 0); // XXX assuming 4-alignment
-          bytes += attribute.offset - bytes;
+      var bytes = 0; // total size in bytes
+      if (!stride && !beginEnd) {
+        // beginEnd can not have stride in the attributes, that is fine. otherwise,
+        // no stride means that all attributes are in fact packed. to keep the rest of
+        // our emulation code simple, we perform unpacking/restriding here. this adds overhead, so
+        // it is a good idea to not hit this!
+#if ASSERTIONS
+        Runtime.warnOnce('Unpacking/restriding attributes, this is not fast');
+#endif
+        if (!GL.immediate.restrideBuffer) GL.immediate.restrideBuffer = _malloc(GL.immediate.MAX_TEMP_BUFFER_SIZE);
+        start = GL.immediate.restrideBuffer;
+#if ASSERTIONS
+        assert(start % 4 == 0);
+#endif
+        // calculate restrided offsets and total size
+        for (var i = 0; i < attributes.length; i++) {
+          var attribute = attributes[i];
+          if (!attribute) break;
+          var size = attribute.size * GL.immediate.byteSizeByType[attribute.type - GL.immediate.byteSizeByTypeRoot];
+          if (size % 4 != 0) size += 4 - (size % 4); // align everything
+          attribute.offset = bytes;
+          bytes += size;
+        }
+#if ASSERTIONS
+        assert(count*bytes <= GL.immediate.MAX_TEMP_BUFFER_SIZE);
+#endif
+        // copy out the data (we need to know the stride for that, and define attribute.pointer
+        for (var i = 0; i < attributes.length; i++) {
+          var attribute = attributes[i];
+          if (!attribute) break;
+          var size4 = Math.floor((attribute.size * GL.immediate.byteSizeByType[attribute.type - GL.immediate.byteSizeByTypeRoot])/4);
+          for (var j = 0; j < count; j++) {
+            for (var k = 0; k < size4; k++) { // copy in chunks of 4 bytes, our alignment makes this possible
+              HEAP32[((start + attribute.offset + bytes*j)>>2) + k] = HEAP32[(attribute.pointer>>2) + j*size4 + k];
+            }
+          }
+          attribute.pointer = start + attribute.offset;
+        }
+      } else {
+        // normal situation, everything is strided and in the same buffer
+        for (var i = 0; i < attributes.length; i++) {
+          var attribute = attributes[i];
+          if (!attribute) break;
+          attribute.offset = attribute.pointer - start;
+          if (attribute.offset > bytes) { // ensure we start where we should
+            assert((attribute.offset - bytes)%4 == 0); // XXX assuming 4-alignment
+            bytes += attribute.offset - bytes;
+          }
+          bytes += attribute.size * GL.immediate.byteSizeByType[attribute.type - GL.immediate.byteSizeByTypeRoot];
+          if (bytes % 4 != 0) bytes += 4 - (bytes % 4); // XXX assuming 4-alignment
+        }
+        assert(beginEnd || bytes <= stride); // if not begin-end, explicit stride should make sense with total byte size
+        if (bytes < stride) { // ensure the size is that of the stride
+          bytes = stride;
         }
-        bytes += attribute.size * GL.immediate.byteSizeByType[attribute.type - GL.immediate.byteSizeByTypeRoot];
-        if (bytes % 4 != 0) bytes += 4 - (bytes % 4); // XXX assuming 4-alignment
-      }
-      assert(stride == 0 || bytes <= stride);
-      if (bytes < stride) { // ensure the size is that of the stride
-        bytes = stride;
       }
       GL.immediate.stride = bytes;
 
@@ -2026,6 +2189,9 @@ var LibraryGL = {
     },
 
     flush: function(numProvidedIndexes, startIndex, ptr) {
+#if ASSERTIONS
+      assert(numProvidedIndexes >= 0 || !numProvidedIndexes);
+#endif
       startIndex = startIndex || 0;
       ptr = ptr || 0;
 
@@ -2085,6 +2251,10 @@ var LibraryGL = {
       if (emulatedElementArrayBuffer) {
         Module.ctx.bindBuffer(Module.ctx.ELEMENT_ARRAY_BUFFER, GL.buffers[GL.currElementArrayBuffer] || null);
       }
+
+#if GL_UNSAFE_OPTS == 0
+      renderer.cleanup();
+#endif
     }
   },
 
@@ -2302,6 +2472,7 @@ var LibraryGL = {
     var attrib;
     switch(cap) {
       case 0x8078: // GL_TEXTURE_COORD_ARRAY
+      case 0x0de1: // GL_TEXTURE_2D - XXX not according to spec, and not in desktop GL, but works in some GLES1.x apparently, so support it
         attrib = GL.immediate.TEXTURE0 + GL.immediate.clientActiveTexture; break;
       case 0x8074: // GL_VERTEX_ARRAY
         attrib = GL.immediate.VERTEX; break;
@@ -2310,14 +2481,19 @@ var LibraryGL = {
       case 0x8076: // GL_COLOR_ARRAY
         attrib = GL.immediate.COLOR; break;
       default:
-        throw 'unhandled clientstate: ' + cap;
+#if ASSERTIONS
+        Module.printErr('WARNING: unhandled clientstate: ' + cap);
+#endif
+        return;
     }
     if (disable && GL.immediate.enabledClientAttributes[attrib]) {
       GL.immediate.enabledClientAttributes[attrib] = false;
       GL.immediate.totalEnabledClientAttributes--;
+      if (GLEmulation.currentVao) delete GLEmulation.currentVao.enabledClientStates[cap];
     } else if (!disable && !GL.immediate.enabledClientAttributes[attrib]) {
       GL.immediate.enabledClientAttributes[attrib] = true;
       GL.immediate.totalEnabledClientAttributes++;
+      if (GLEmulation.currentVao) GLEmulation.currentVao.enabledClientStates[cap] = 1;
     }
     GL.immediate.modifiedClientAttributes = true;
   },
@@ -2344,6 +2520,63 @@ var LibraryGL = {
     GL.immediate.clientActiveTexture = texture - 0x84C0; // GL_TEXTURE0
   },
 
+  // Vertex array object (VAO) support. TODO: when the WebGL extension is popular, use that and remove this code and GL.vaos
+  glGenVertexArrays__deps: ['$GLEMulation'],
+  glGenVertexArrays__sig: ['vii'],
+  glGenVertexArrays: function(n, vaos) {
+    for (var i = 0; i < n; i++) {
+      var id = GL.getNewId(GLEmulation.vaos); 
+      GLEmulation.vaos[id] = {
+        id: id,
+        arrayBuffer: 0,
+        elementArrayBuffer: 0,
+        enabledVertexAttribArrays: {},
+        vertexAttribPointers: {},
+        enabledClientStates: {},
+      };
+      {{{ makeSetValue('vaos', 'i*4', 'id', 'i32') }}};
+    }
+  },
+  glDeleteVertexArrays__sig: ['vii'],
+  glDeleteVertexArrays: function(n, vaos) {
+    for (var i = 0; i < n; i++) {
+      var id = {{{ makeGetValue('vaos', 'i*4', 'i32') }}};
+      GLEmulation.vaos[id] = null;
+      if (GLEmulation.currentVao && GLEmulation.currentVao.id == id) GLEmulation.currentVao = null;
+    }
+  },
+  glBindVertexArray__sig: ['vi'],
+  glBindVertexArray: function(vao) {
+    // undo vao-related things, wipe the slate clean, both for vao of 0 or an actual vao
+    GLEmulation.currentVao = null; // make sure the commands we run here are not recorded
+    if (GL.immediate.lastRenderer) GL.immediate.lastRenderer.cleanup();
+    _glBindBuffer(Module.ctx.ARRAY_BUFFER, 0); // XXX if one was there before we were bound?
+    _glBindBuffer(Module.ctx.ELEMENT_ARRAY_BUFFER, 0);
+    for (var vaa in GLEmulation.enabledVertexAttribArrays) {
+      Module.ctx.disableVertexAttribArray(vaa);
+    }
+    GLEmulation.enabledVertexAttribArrays = {};
+    GL.immediate.enabledClientAttributes = [0, 0];
+    GL.immediate.totalEnabledClientAttributes = 0;
+    GL.immediate.modifiedClientAttributes = true;
+    if (vao) {
+      // replay vao
+      var info = GLEmulation.vaos[vao];
+      _glBindBuffer(Module.ctx.ARRAY_BUFFER, info.arrayBuffer); // XXX overwrite current binding?
+      _glBindBuffer(Module.ctx.ELEMENT_ARRAY_BUFFER, info.elementArrayBuffer);
+      for (var vaa in info.enabledVertexAttribArrays) {
+        _glEnableVertexAttribArray(vaa);
+      }
+      for (var vaa in info.vertexAttribPointers) {
+        _glVertexAttribPointer.apply(null, info.vertexAttribPointers[vaa]);
+      }
+      for (var attrib in info.enabledClientStates) {
+        _glEnableClientState(attrib|0);
+      }
+      GLEmulation.currentVao = info; // set currentVao last, so the commands we ran here were not recorded
+    }
+  },
+
   // OpenGL Immediate Mode matrix routines.
   // Note that in the future we might make these available only in certain modes.
   glMatrixMode__deps: ['$GL', '$GLImmediateSetup', '$GLEmulation'], // emulation is not strictly needed, this is a workaround
@@ -2434,6 +2667,7 @@ var LibraryGL = {
     GL.immediate.matrix.lib.mat4.multiply(GL.immediate.matrix[GL.immediate.currentMatrix],
         GL.immediate.matrix.lib.mat4.frustum(left, right, bottom, top_, nearVal, farVal));
   },
+  glFrustumf: 'glFrustum',
 
   glOrtho: function(left, right, bottom, top_, nearVal, farVal) {
     GL.immediate.matricesModified = true;
@@ -2464,8 +2698,9 @@ var LibraryGL = {
 
   gluPerspective: function(fov, aspect, near, far) {
     GL.immediate.matricesModified = true;
-    GL.immediate.matrix.lib.mat4.multiply(GL.immediate.matrix[GL.immediate.currentMatrix],
-        GL.immediate.matrix.lib.mat4.perspective(fov, aspect, near, far, GL.immediate.currentMatrix));
+    GL.immediate.matrix[GL.immediate.currentMatrix] =
+      GL.immediate.matrix.lib.mat4.perspective(fov, aspect, near, far,
+                                               GL.immediate.matrix[GL.immediate.currentMatrix]);
   },
 
   gluLookAt: function(ex, ey, ez, cx, cy, cz, ux, uy, uz) {
@@ -2534,8 +2769,8 @@ var LibraryGL = {
 
   glTexGeni: function() { throw 'glTexGeni: TODO' },
   glTexGenfv: function() { throw 'glTexGenfv: TODO' },
-  glTexEnvi: function() { throw 'glTexEnvi: TODO' },
-  glTexEnvfv: function() { throw 'glTexEnvfv: TODO' },
+  glTexEnvi: function() { Runtime.warnOnce('glTexEnvi: TODO') },
+  glTexEnvfv: function() { Runtime.warnOnce('glTexEnvfv: TODO') },
 
   glTexImage1D: function() { throw 'glTexImage1D: TODO' },
   glTexCoord3f: function() { throw 'glTexCoord3f: TODO' },
@@ -2548,6 +2783,22 @@ var LibraryGL = {
   glVertexAttribPointer__sig: 'viiiiii',
   glCheckFramebufferStatus__sig: 'ii',
   glRenderbufferStorage__sig: 'viiii',
+
+  // Open GLES1.1 compatibility
+  glGenFramebuffersOES : 'glGenFramebuffers',
+  glGenRenderbuffersOES : 'glGenRenderbuffers',
+  glBindFramebufferOES : 'glBindFramebuffer',
+  glBindRenderbufferOES : 'glBindRenderbuffer',
+  glGetRenderbufferParameterivOES : 'glGetRenderbufferParameteriv',
+  glFramebufferRenderbufferOES : 'glFramebufferRenderbuffer',
+  glRenderbufferStorageOES : 'glRenderbufferStorage',
+  glCheckFramebufferStatusOES : 'glCheckFramebufferStatus',
+  glDeleteFramebuffersOES : 'glDeleteFramebuffers',
+  glDeleteRenderbuffersOES : 'glDeleteRenderbuffers',
+  glGenVertexArraysOES: 'glGenVertexArrays',
+  glDeleteVertexArraysOES: 'glDeleteVertexArrays',
+  glBindVertexArrayOES: 'glBindVertexArray',
+  glFramebufferTexture2DOES: 'glFramebufferTexture2D'
 };
 
 // Simple pass-through functions. Starred ones have return values. [X] ones have X in the C name but not in the JS name
diff --git a/src/library_glut.js b/src/library_glut.js
index 2e662698..bb4dfefa 100644
--- a/src/library_glut.js
+++ b/src/library_glut.js
@@ -381,6 +381,12 @@ var LibraryGLUT = {
     return 1;
   },
 
+  glutDestroyWindow__deps: ['$Browser'],
+  glutDestroyWindow: function(name) {
+    Module.ctx = Browser.destroyContext(Module['canvas'], true, true);
+    return 1;
+  },
+
   glutReshapeWindow__deps: ['$GLUT', 'glutPostRedisplay'],
   glutReshapeWindow: function(width, height) {
     GLUT.cancelFullScreen();
diff --git a/src/library_sdl.js b/src/library_sdl.js
index cfab6410..e02e1e62 100644
--- a/src/library_sdl.js
+++ b/src/library_sdl.js
@@ -607,9 +607,11 @@ var LibrarySDL = {
   SDL_Init: function(what) {
     SDL.startTime = Date.now();
     // capture all key events. we just keep down and up, but also capture press to prevent default actions
-    document.onkeydown = SDL.receiveEvent;
-    document.onkeyup = SDL.receiveEvent;
-    document.onkeypress = SDL.receiveEvent;
+    if (!Module['doNotCaptureKeyboard']) {
+      document.onkeydown = SDL.receiveEvent;
+      document.onkeyup = SDL.receiveEvent;
+      document.onkeypress = SDL.receiveEvent;
+    }
     window.onunload = SDL.receiveEvent;
     SDL.keyboardState = _malloc(0x10000);
     _memset(SDL.keyboardState, 0, 0x10000);
@@ -1077,7 +1079,9 @@ var LibrarySDL = {
     }
     var surf = SDL.makeSurface(raw.width, raw.height, 0, false, 'load:' + filename);
     var surfData = SDL.surfaces[surf];
+    surfData.ctx.globalCompositeOperation = "copy";
     surfData.ctx.drawImage(raw, 0, 0, raw.width, raw.height, 0, 0, raw.width, raw.height);
+    surfData.ctx.globalCompositeOperation = "source-over";
     // XXX SDL does not specify that loaded images must have available pixel data, in fact
     //     there are cases where you just want to blit them, so you just need the hardware
     //     accelerated version. However, code everywhere seems to assume that the pixels
@@ -1176,6 +1180,10 @@ var LibrarySDL = {
 
   // SDL Mixer
 
+  Mix_Init: function(flags) {
+    return 8; /* MIX_INIT_OGG */
+  },
+
   Mix_OpenAudio: function(frequency, format, channels, chunksize) {
     SDL.allocateChannels(32);
     // Just record the values for a later call to Mix_QuickLoad_RAW
diff --git a/src/parseTools.js b/src/parseTools.js
index e081d0de..357d33a1 100644
--- a/src/parseTools.js
+++ b/src/parseTools.js
@@ -103,6 +103,11 @@ function isNiceIdent(ident, loose) {
   }
 }
 
+function isJSVar(ident) {
+  return /^\(?[$_]?[\w$_\d ]*\)+$/.test(ident);
+
+}
+
 function isStructPointerType(type) {
   // This test is necessary for clang - in llvm-gcc, we
   // could check for %struct. The downside is that %1 can
@@ -976,12 +981,6 @@ function checkSafeHeap() {
   return SAFE_HEAP === 1 || checkSpecificSafeHeap();
 }
 
-if (ASM_JS) {
-  var hexMemoryMask = '0x' + (TOTAL_MEMORY-1).toString(16);
-  var decMemoryMask = (TOTAL_MEMORY-1).toString();
-  var memoryMask = hexMemoryMask.length <= decMemoryMask.length ? hexMemoryMask : decMemoryMask;
-}
-
 function getHeapOffset(offset, type, forceAsm) {
   if (USE_TYPED_ARRAYS !== 2) {
     return offset;
@@ -991,11 +990,11 @@ function getHeapOffset(offset, type, forceAsm) {
     }
     var shifts = Math.log(Runtime.getNativeTypeSize(type))/Math.LN2;
     offset = '(' + offset + ')';
-    if (ASM_JS && (phase == 'funcs' || forceAsm)) offset = '(' + offset + '&' + memoryMask + ')';
     if (shifts != 0) {
       return '(' + offset + '>>' + shifts + ')';
     } else {
-      return offset;
+      // we need to guard against overflows here, HEAP[U]8 expects a guaranteed int
+      return isJSVar(offset) ? offset : '(' + offset + '|0)';
     }
   }
 }
@@ -1045,20 +1044,6 @@ function asmCoercion(value, type, signedness) {
   }
 }
 
-var TWO_TWENTY = Math.pow(2, 20);
-
-function asmMultiplyI32(a, b) {
-  // special-case: there is no integer multiply in asm, because there is no true integer
-  // multiply in JS. While we wait for Math.imul, do double multiply
-  if ((isNumber(a) && Math.abs(a) < TWO_TWENTY) || (isNumber(b) && Math.abs(b) < TWO_TWENTY)) {
-    return '(((' + a + ')*(' + b + '))&-1)'; // small enough to emit directly as a multiply
-  }
-  if (USE_MATH_IMUL) {
-    return 'Math.imul(' + a + ',' + b + ')';
-  }
-  return '(~~(+((' + a + ')|0) * +((' + b + ')|0)))';
-}
-
 function asmFloatToInt(x) {
   return '(~~(' + x + '))';
 }
@@ -1070,7 +1055,6 @@ function makeGetTempDouble(i, type, forSet) { // get an aliased part of the temp
   var ptr = getFastValue('tempDoublePtr', '+', Runtime.getNativeTypeSize(type)*i);
   var offset;
   if (type == 'double') {
-    if (ASM_JS) ptr = '(' + ptr + ')&' + memoryMask;
     offset = '(' + ptr + ')>>3';
   } else {
     offset = getHeapOffset(ptr, type);
@@ -1153,8 +1137,8 @@ function makeGetValue(ptr, pos, type, noNeedFirst, unsigned, ignore, align, noSa
   }
 }
 
-function makeGetValueAsm(ptr, pos, type) {
-  return makeGetValue(ptr, pos, type, null, null, null, null, null, true);
+function makeGetValueAsm(ptr, pos, type, unsigned) {
+  return makeGetValue(ptr, pos, type, null, unsigned, null, null, null, true);
 }
 
 function indexizeFunctions(value, type) {
@@ -1372,9 +1356,11 @@ function makeHEAPView(which, start, end) {
 var PLUS_MUL = set('+', '*');
 var MUL_DIV = set('*', '/');
 var PLUS_MINUS = set('+', '-');
+var TWO_TWENTY = Math.pow(2, 20);
 
 // Given two values and an operation, returns the result of that operation.
 // Tries to do as much as possible at compile time.
+// Leaves overflows etc. unhandled, *except* for integer multiply, in order to be efficient with Math.imul
 function getFastValue(a, op, b, type) {
   a = a.toString();
   b = b.toString();
@@ -1410,8 +1396,14 @@ function getFastValue(a, op, b, type) {
           return '(' + a + '<<' + shifts + ')';
         }
       }
-      if (ASM_JS && !(type in Runtime.FLOAT_TYPES)) {
-        return asmMultiplyI32(a, b); // unoptimized multiply, do it using asm.js's special multiply operation
+      if (!(type in Runtime.FLOAT_TYPES)) {
+        // if guaranteed small enough to not overflow into a double, do a normal multiply
+        var bits = getBits(type) || 32; // default is 32-bit multiply for things like getelementptr indexes
+        // Note that we can emit simple multiple in non-asm.js mode, but asm.js will not parse "16-bit" multiple, so must do imul there
+        if ((isNumber(a) && Math.abs(a) < TWO_TWENTY) || (isNumber(b) && Math.abs(b) < TWO_TWENTY) || (bits < 32 && !ASM_JS)) {
+          return '(((' + a + ')*(' + b + '))&' + ((Math.pow(2, bits)-1)|0) + ')'; // keep a non-eliminatable coercion directly on this
+        }
+        return 'Math.imul(' + a + ',' + b + ')';
       }
     } else {
       if (a == '0') {
@@ -1717,9 +1709,7 @@ function handleOverflow(text, bits) {
   if (!bits) return text;
   var correct = correctOverflows();
   warnOnce(!correct || bits <= 32, 'Cannot correct overflows of this many bits: ' + bits);
-  if (CHECK_OVERFLOWS) return 'CHECK_OVERFLOW(' + text + ', ' + bits + ', ' + Math.floor(correctSpecificOverflow() && !PGO) + (
-    PGO ? ', "' + Debugging.getIdentifier() + '"' : ''
-  ) + ')';
+  if (CHECK_OVERFLOWS) return 'CHECK_OVERFLOW(' + text + ', ' + bits + ', ' + Math.floor(correctSpecificOverflow()) + ')';
   if (!correct) return text;
   if (bits == 32) {
     return '((' + text + ')|0)';
@@ -1827,9 +1817,7 @@ function makeSignOp(value, type, op, force, ignore) {
   var bits, full;
   if (type in Runtime.INT_TYPES) {
     bits = parseInt(type.substr(1));
-    full = op + 'Sign(' + value + ', ' + bits + ', ' + Math.floor(ignore || (correctSpecificSign() && !PGO)) + (
-      PGO ? ', "' + (ignore ? '' : Debugging.getIdentifier()) + '"' : ''
-    ) + ')';
+    full = op + 'Sign(' + value + ', ' + bits + ', ' + Math.floor(ignore || (correctSpecificSign())) + ')';
     // Always sign/unsign constants at compile time, regardless of CHECK/CORRECT
     if (isNumber(value)) {
       return eval(full).toString();
@@ -2142,14 +2130,7 @@ function processMathop(item) {
     case 'add': return handleOverflow(getFastValue(idents[0], '+', idents[1], item.type), bits);
     case 'sub': return handleOverflow(getFastValue(idents[0], '-', idents[1], item.type), bits);
     case 'sdiv': case 'udiv': return makeRounding(getFastValue(idents[0], '/', idents[1], item.type), bits, op[0] === 's');
-    case 'mul': {
-      if (bits == 32 && PRECISE_I32_MUL) {
-        Types.preciseI64MathUsed = true;
-        return '(i64Math' + (ASM_JS ? '_' : '.') + 'multiply(' + asmCoercion(idents[0], 'i32') + ',0,' + asmCoercion(idents[1], 'i32') + ',0),' + makeGetValue('tempDoublePtr', 0, 'i32') + ')';
-      } else {
-        return '((' +getFastValue(idents[0], '*', idents[1], item.type) + ')&-1)'; // force a non-eliminatable coercion here, to prevent a double result from leaking
-      }
-    }
+    case 'mul': return getFastValue(idents[0], '*', idents[1], item.type); // overflow handling is already done in getFastValue for '*'
     case 'urem': case 'srem': return getFastValue(idents[0], '%', idents[1], item.type);
     case 'or': {
       if (bits > 32) {
diff --git a/src/preamble.js b/src/preamble.js
index aab50e9a..a7731e7f 100644
--- a/src/preamble.js
+++ b/src/preamble.js
@@ -150,52 +150,6 @@ function SAFE_HEAP_COPY_HISTORY(dest, src) {
 //==========================================
 #endif
 
-var CorrectionsMonitor = {
-#if PGO
-  MAX_ALLOWED: Infinity,
-#else
-  MAX_ALLOWED: 0, // XXX
-#endif
-  corrections: 0,
-  sigs: {},
-
-  note: function(type, succeed, sig) {
-    if (!succeed) {
-      this.corrections++;
-      if (this.corrections >= this.MAX_ALLOWED) abort('\n\nToo many corrections!');
-    }
-#if PGO
-    if (!sig)
-      sig = (new Error().stack).toString().split('\n')[2].split(':').slice(-1)[0]; // Spidermonkey-specific FIXME
-    sig = type + '|' + sig;
-    if (!this.sigs[sig]) {
-      //Module.print('Correction: ' + sig);
-      this.sigs[sig] = [0, 0]; // fail, succeed
-    }
-    this.sigs[sig][succeed ? 1 : 0]++;
-#endif
-  },
-
-  print: function() {
-#if PGO
-    var items = [];
-    for (var sig in this.sigs) {
-      items.push({
-        sig: sig,
-        fails: this.sigs[sig][0],
-        succeeds: this.sigs[sig][1],
-        total: this.sigs[sig][0] + this.sigs[sig][1]
-      });
-    }
-    items.sort(function(x, y) { return y.total - x.total; });
-    for (var i = 0; i < items.length; i++) {
-      var item = items[i];
-      Module.print(item.sig + ' : ' + item.total + ' hits, %' + (Math.ceil(100*item.fails/item.total)) + ' failures');
-    }
-#endif
-  }
-};
-
 #if CHECK_OVERFLOWS
 //========================================
 // Debugging tools - Mathop overflows
@@ -207,24 +161,20 @@ function CHECK_OVERFLOW(value, bits, ignore, sig) {
   // For signedness issue here, see settings.js, CHECK_SIGNED_OVERFLOWS
 #if CHECK_SIGNED_OVERFLOWS
   if (value === Infinity || value === -Infinity || value >= twopbits1 || value < -twopbits1) {
-    CorrectionsMonitor.note('SignedOverflow', 0, sig);
-    if (value === Infinity || value === -Infinity || Math.abs(value) >= twopbits) CorrectionsMonitor.note('Overflow');
+    throw 'SignedOverflow';
+    if (value === Infinity || value === -Infinity || Math.abs(value) >= twopbits) throw 'Overflow';
+  }
 #else
   if (value === Infinity || value === -Infinity || Math.abs(value) >= twopbits) {
-    CorrectionsMonitor.note('Overflow', 0, sig);
+    throw 'Overflow';
+  }
 #endif
 #if CORRECT_OVERFLOWS
-    // Fail on >32 bits - we warned at compile time
-    if (bits <= 32) {
-      value = value & (twopbits - 1);
-    }
-#endif
-  } else {
-#if CHECK_SIGNED_OVERFLOWS
-    CorrectionsMonitor.note('SignedOverflow', 1, sig);
-#endif
-    CorrectionsMonitor.note('Overflow', 1, sig);
+  // Fail on >32 bits - we warned at compile time
+  if (bits <= 32) {
+    value = value & (twopbits - 1);
   }
+#endif
   return value;
 }
 #endif
@@ -243,39 +193,6 @@ var INDENT = '';
 var START_TIME = Date.now();
 #endif
 
-#if PROFILE
-var PROFILING = 0;
-var PROFILING_ROOT = { time: 0, children: {}, calls: 0 };
-var PROFILING_NODE;
-
-function startProfiling() {
-  PROFILING_NODE = PROFILING_ROOT;
-  PROFILING = 1;
-}
-Module['startProfiling'] = startProfiling;
-
-function stopProfiling() {
-  PROFILING = 0;
-  assert(PROFILING_NODE === PROFILING_ROOT, 'Must have popped all the profiling call stack');
-}
-Module['stopProfiling'] = stopProfiling;
-
-function printProfiling() {
-  function dumpData(name_, node, indent) {
-    Module.print(indent + ('________' + node.time).substr(-8) + ': ' + name_ + ' (' + node.calls + ')');
-    var children = [];
-    for (var child in node.children) {
-      children.push(node.children[child]);
-      children[children.length-1].name_ = child;
-    }
-    children.sort(function(x, y) { return y.time - x.time });
-    children.forEach(function(child) { dumpData(child.name_, child, indent + '  ') });
-  }
-  dumpData('root', PROFILING_ROOT, ' ');
-}
-Module['printProfiling'] = printProfiling;
-#endif
-
 //========================================
 // Runtime essentials
 //========================================
@@ -655,47 +572,43 @@ function enlargeMemory() {
 #endif
 
 var TOTAL_STACK = Module['TOTAL_STACK'] || {{{ TOTAL_STACK }}};
-#if ASM_JS == 0
 var TOTAL_MEMORY = Module['TOTAL_MEMORY'] || {{{ TOTAL_MEMORY }}};
-#else
-var TOTAL_MEMORY = {{{ TOTAL_MEMORY }}}; // in asm, we hardcode the mask, so cannot adjust memory at runtime
-#endif
 var FAST_MEMORY = Module['FAST_MEMORY'] || {{{ FAST_MEMORY }}};
 
 // Initialize the runtime's memory
 #if USE_TYPED_ARRAYS
 // check for full engine support (use string 'subarray' to avoid closure compiler confusion)
-  assert(!!Int32Array && !!Float64Array && !!(new Int32Array(1)['subarray']) && !!(new Int32Array(1)['set']),
-         'Cannot fallback to non-typed array case: Code is too specialized');
+assert(!!Int32Array && !!Float64Array && !!(new Int32Array(1)['subarray']) && !!(new Int32Array(1)['set']),
+       'Cannot fallback to non-typed array case: Code is too specialized');
 
 #if USE_TYPED_ARRAYS == 1
-  HEAP = IHEAP = new Int32Array(TOTAL_MEMORY);
-  IHEAPU = new Uint32Array(IHEAP.buffer);
+HEAP = IHEAP = new Int32Array(TOTAL_MEMORY);
+IHEAPU = new Uint32Array(IHEAP.buffer);
 #if USE_FHEAP
-  FHEAP = new Float64Array(TOTAL_MEMORY);
+FHEAP = new Float64Array(TOTAL_MEMORY);
 #endif
 #endif
 #if USE_TYPED_ARRAYS == 2
-  var buffer = new ArrayBuffer(TOTAL_MEMORY);
-  HEAP8 = new Int8Array(buffer);
-  HEAP16 = new Int16Array(buffer);
-  HEAP32 = new Int32Array(buffer);
-  HEAPU8 = new Uint8Array(buffer);
-  HEAPU16 = new Uint16Array(buffer);
-  HEAPU32 = new Uint32Array(buffer);
-  HEAPF32 = new Float32Array(buffer);
-  HEAPF64 = new Float64Array(buffer);
-
-  // Endianness check (note: assumes compiler arch was little-endian)
-  HEAP32[0] = 255;
-  assert(HEAPU8[0] === 255 && HEAPU8[3] === 0, 'Typed arrays 2 must be run on a little-endian system');
+var buffer = new ArrayBuffer(TOTAL_MEMORY);
+HEAP8 = new Int8Array(buffer);
+HEAP16 = new Int16Array(buffer);
+HEAP32 = new Int32Array(buffer);
+HEAPU8 = new Uint8Array(buffer);
+HEAPU16 = new Uint16Array(buffer);
+HEAPU32 = new Uint32Array(buffer);
+HEAPF32 = new Float32Array(buffer);
+HEAPF64 = new Float64Array(buffer);
+
+// Endianness check (note: assumes compiler arch was little-endian)
+HEAP32[0] = 255;
+assert(HEAPU8[0] === 255 && HEAPU8[3] === 0, 'Typed arrays 2 must be run on a little-endian system');
 #endif
 #else
-  // Make sure that our HEAP is implemented as a flat array.
-  HEAP = []; // Hinting at the size with |new Array(TOTAL_MEMORY)| should help in theory but makes v8 much slower
-  for (var i = 0; i < FAST_MEMORY; i++) {
-    HEAP[i] = 0; // XXX We do *not* use {{| makeSetValue(0, 'i', 0, 'null') |}} here, since this is done just to optimize runtime speed
-  }
+// Make sure that our HEAP is implemented as a flat array.
+HEAP = []; // Hinting at the size with |new Array(TOTAL_MEMORY)| should help in theory but makes v8 much slower
+for (var i = 0; i < FAST_MEMORY; i++) {
+  HEAP[i] = 0; // XXX We do *not* use {{| makeSetValue(0, 'i', 0, 'null') |}} here, since this is done just to optimize runtime speed
+}
 #endif
 
 Module['HEAP'] = HEAP;
@@ -722,7 +635,7 @@ STACK_MAX = TOTAL_STACK; // we lose a little stack here, but TOTAL_STACK is nice
 #if USE_TYPED_ARRAYS == 2
 var tempDoublePtr = Runtime.alignMemory(allocate(12, 'i8', ALLOC_STACK), 8);
 assert(tempDoublePtr % 8 == 0);
-function copyTempFloat(ptr) { // functions, because inlining this code is increases code size too much
+function copyTempFloat(ptr) { // functions, because inlining this code increases code size too much
   HEAP8[tempDoublePtr] = HEAP8[ptr];
   HEAP8[tempDoublePtr+1] = HEAP8[ptr+1];
   HEAP8[tempDoublePtr+2] = HEAP8[ptr+2];
@@ -773,9 +686,6 @@ function preMain() {
 }
 function exitRuntime() {
   callRuntimeCallbacks(__ATEXIT__);
-
-  // Print summary of correction activity
-  CorrectionsMonitor.print();
 }
 
 // Tools
@@ -832,6 +742,20 @@ Module['writeArrayToMemory'] = writeArrayToMemory;
 {{{ unSign }}}
 {{{ reSign }}}
 
+#if PRECISE_I32_MUL
+if (!Math.imul) Math.imul = function(a, b) {
+  var ah  = a >>> 16;
+  var al = a & 0xffff;
+  var bh  = b >>> 16;
+  var bl = b & 0xffff;
+  return (al*bl + ((ah*bl + al*bh) << 16))|0;
+};
+#else
+Math.imul = function(a, b) {
+  return (a*b)|0; // fast but imprecise
+};
+#endif
+
 // A counter of dependencies for calling run(). If we need to
 // do asynchronous work before running, increment this and
 // decrement it. Incrementing must happen in a place like
diff --git a/src/runtime.js b/src/runtime.js
index e5b86065..e902d27b 100644
--- a/src/runtime.js
+++ b/src/runtime.js
@@ -242,6 +242,10 @@ var Runtime = {
       } else if (Runtime.isStructType(field)) {
         size = Types.types[field].flatSize;
         alignSize = Types.types[field].alignSize;
+      } else if (field[0] == 'b') {
+        // bN, large number field, like a [N x i8]
+        size = field.substr(1)|0;
+        alignSize = 1;
       } else {
         throw 'Unclear type in struct: ' + field + ', in ' + type.name_ + ' :: ' + dump(Types.types[type.name_]);
       }
@@ -353,7 +357,7 @@ var Runtime = {
   },
 
   addFunction: function(func, sig) {
-    assert(sig);
+    //assert(sig); // TODO: support asm
     var table = FUNCTION_TABLE; // TODO: support asm
     var ret = table.length;
     table.push(func);
@@ -505,26 +509,19 @@ function getRuntime() {
 // example, -1 in int32 would be a very large number as unsigned.
 function unSign(value, bits, ignore, sig) {
   if (value >= 0) {
-#if CHECK_SIGNS
-    if (!ignore) CorrectionsMonitor.note('UnSign', 1, sig);
-#endif
     return value;
   }
 #if CHECK_SIGNS
-  if (!ignore) CorrectionsMonitor.note('UnSign', 0, sig);
+  if (!ignore) throw 'UnSign';
 #endif
   return bits <= 32 ? 2*Math.abs(1 << (bits-1)) + value // Need some trickery, since if bits == 32, we are right at the limit of the bits JS uses in bitshifts
                     : Math.pow(2, bits)         + value;
-  // TODO: clean up previous line
 }
 
 // Converts a value we have as unsigned, into a signed value. For
 // example, 200 in a uint8 would be a negative number.
 function reSign(value, bits, ignore, sig) {
   if (value <= 0) {
-#if CHECK_SIGNS
-    if (!ignore) CorrectionsMonitor.note('ReSign', 1, sig);
-#endif
     return value;
   }
   var half = bits <= 32 ? Math.abs(1 << (bits-1)) // abs is needed if bits == 32
@@ -536,10 +533,7 @@ function reSign(value, bits, ignore, sig) {
                                                        // but, in general there is no perfect solution here. With 64-bit ints, we get rounding and errors
                                                        // TODO: In i64 mode 1, resign the two parts separately and safely
 #if CHECK_SIGNS
-    if (!ignore) {
-      CorrectionsMonitor.note('ReSign', 0, sig);
-      noted = true;
-    }
+    if (!ignore) throw 'ReSign';
 #endif
     value = -2*half + value; // Cannot bitshift half, as it may be at the limit of the bits JS uses in bitshifts
   }
@@ -548,18 +542,9 @@ function reSign(value, bits, ignore, sig) {
   // without CHECK_SIGNS, we would just do the |0 shortcut, so check that that
   // would indeed give the exact same result.
   if (bits === 32 && (value|0) !== value && typeof value !== 'boolean') {
-    if (!ignore) {
-      CorrectionsMonitor.note('ReSign', 0, sig);
-      noted = true;
-    }
+    if (!ignore) throw 'ReSign';
   }
-  if (!noted) CorrectionsMonitor.note('ReSign', 1, sig);
 #endif
   return value;
 }
 
-// Just a stub. We don't care about noting compile-time corrections. But they are called.
-var CorrectionsMonitor = {
-  note: function(){}
-};
-
diff --git a/src/settings.js b/src/settings.js
index 0234d0ca..7caa5b92 100644
--- a/src/settings.js
+++ b/src/settings.js
@@ -92,13 +92,9 @@ var PRECISE_I64_MATH = 1; // If enabled, i64 addition etc. is emulated - which i
                           // that we can't know at compile time that 64-bit math is needed. For example, if you
                           // print 64-bit values with printf, but never add them, we can't know at compile time
                           // and you need to set this to 2.
-var PRECISE_I32_MUL = 0; // If enabled, i64 math is done in i32 multiplication. This is necessary if the values
-                         // exceed the JS double-integer limit of ~52 bits. This option can normally be disabled
-                         // because generally i32 multiplication works ok without it, and enabling it has a big
-                         // impact on performance.
-                         // Note that you can hand-optimize your code to avoid the need for this: If you do
-                         // multiplications that actually need 64-bit precision inside 64-bit values, things
-                         // will work properly. (Unless the LLVM optimizer turns them into 32-bit values?)
+var PRECISE_I32_MUL = 1; // If enabled, i32 multiplication is done with full precision, which means it is
+                         // correct even if the value exceeds the JS double-integer limit of ~52 bits (otherwise,
+                         // rounding will occur above that range).
 
 var CLOSURE_ANNOTATIONS = 0; // If set, the generated code will be annotated for the closure
                              // compiler. This potentially lets closure optimize the code better.
@@ -134,6 +130,16 @@ var SAFE_HEAP_LOG = 0; // Log out all SAFE_HEAP operations
 
 var ASM_HEAP_LOG = 0; // Simple heap logging, like SAFE_HEAP_LOG but cheaper, and in asm.js
 
+var CORRUPTION_CHECK = 0; // When enabled, will emit a buffer area at the beginning and
+                          // end of each allocation on the heap, filled with canary
+                          // values that can be checked later. Corruption is checked for
+                          // at the end of each at each free() (see jsifier to add more, and you
+                          // can add more manual checks by calling CorruptionChecker.checkAll).
+                          // 0 means not enabled, higher values mean the size of the
+                          // buffer areas as a multiple of the allocated area (so
+                          // 1 means 100%, or buffer areas equal to allocated area,
+                          // both before and after). This must be an integer.
+
 var LABEL_DEBUG = 0; // 1: Print out functions as we enter them
                      // 2: Also print out each label as we enter it
 var LABEL_FUNCTION_FILTERS = []; // Filters for function label debug.
@@ -148,11 +154,13 @@ var LIBRARY_DEBUG = 0; // Print out when we enter a library call (library*.js).
                        // Runtime.debug at runtime for logging to cease, and can set it when you
                        // want it back. A simple way to set it in C++ is
                        //   emscripten_run_script("Runtime.debug = ...;");
-var GL_DEBUG = 0; // Print out all calls into WebGL. As with LIBRARY_DEBUG, you can set a runtime
-                  // option, in this case GL.debug.
 var SOCKET_DEBUG = 0; // Log out socket/network data transfer.
 
-var PROFILE_MAIN_LOOP = 0; // Profile the function called in set_main_loop
+var GL_DEBUG = 0; // Print out all calls into WebGL. As with LIBRARY_DEBUG, you can set a runtime
+                  // option, in this case GL.debug.
+var GL_TESTING = 0; // When enabled, sets preserveDrawingBuffer in the context, to allow tests to work (but adds overhead)
+var GL_MAX_TEMP_BUFFER_SIZE = 2097152; // How large GL emulation temp buffers are
+var GL_UNSAFE_OPTS = 1; // Enables some potentially-unsafe optimizations in GL emulation code
 
 var DISABLE_EXCEPTION_CATCHING = 0; // Disables generating code to actually catch exceptions. If the code you
                                     // are compiling does not actually rely on catching exceptions (but the
@@ -203,21 +211,10 @@ var FS_LOG = 0; // Log all FS operations.  This is especially helpful when you'r
                 // a new project and want to see a list of file system operations happening
                 // so that you can create a virtual file system with all of the required files.
 
-var PGO = 0; // Profile-guided optimization.
-             // When run with the CHECK_* options, will not fail on errors. Instead, will
-             // keep a record of which checks succeeded and which failed. On shutdown, will
-             // print out that information. This is useful for knowing which lines need
-             // checking enabled and which do not, that is, this is a way to automate the
-             // generation of line data for CORRECT_*_LINES options.
-             // All CORRECT_* options default to 1 with PGO builds.
-             // See https://github.com/kripken/emscripten/wiki/Optimizing-Code for more info
-
 var NAMED_GLOBALS = 0; // If 1, we use global variables for globals. Otherwise
                        // they are referred to by a base plus an offset (called an indexed global),
                        // saving global variables but adding runtime overhead.
 
-var PROFILE = 0; // Enables runtime profiling. See test_profiling for a usage example.
-
 var EXPORT_ALL = 0; // If true, we export all the symbols
 var EXPORTED_FUNCTIONS = ['_main']; // Functions that are explicitly exported. These functions are kept alive
                                     // through LLVM dead code elimination, and also made accessible outside of
@@ -325,7 +322,6 @@ var BENCHMARK = 0; // If 1, will just time how long main() takes to execute, and
 var ASM_JS = 0; // If 1, generate code in asm.js format. XXX This is highly experimental,
                 // and will not work on most codebases yet. It is NOT recommended that you
                 // try this yet.
-var USE_MATH_IMUL = 0; // If 1, use Math.imul when useful
 
 var EXPLICIT_ZEXT = 0; // If 1, generate an explicit conversion of zext i1 to i32, using ?:
 
diff --git a/system/lib/libc.symbols b/system/lib/libc.symbols
index b98a79e4..d41f4140 100644
--- a/system/lib/libc.symbols
+++ b/system/lib/libc.symbols
@@ -48,8 +48,3 @@ _ZNSt20bad_array_new_lengthC2Ev
 _ZNSt20bad_array_new_lengthD0Ev
 _ZNSt20bad_array_new_lengthD1Ev
 _ZNSt20bad_array_new_lengthD2Ev
-memcpy
-llvm.memcpy.i32
-llvm.memcpy.i64
-llvm.memcpy.p0i8.p0i8.i32
-llvm.memcpy.p0i8.p0i8.i64
diff --git a/system/lib/libcxx/symbols b/system/lib/libcxx/symbols
index 23d4a7a4..0d412de7 100644
--- a/system/lib/libcxx/symbols
+++ b/system/lib/libcxx/symbols
@@ -767,7 +767,6 @@
          W _ZNKSt3__19money_putIwNS_19ostreambuf_iteratorIwNS_11char_traitsIwEEEEE3putES4_bRNS_8ios_baseEwe
          W _ZNKSt3__19money_putIwNS_19ostreambuf_iteratorIwNS_11char_traitsIwEEEEE6do_putES4_bRNS_8ios_baseEwRKNS_12basic_stringIwS3_NS_9allocatorIwEEEE
          W _ZNKSt3__19money_putIwNS_19ostreambuf_iteratorIwNS_11char_traitsIwEEEEE6do_putES4_bRNS_8ios_baseEwe
-         T _ZNKSt8bad_cast4whatEv
          T _ZNKSt9exception4whatEv
          T _ZNSt10bad_typeidC1Ev
          T _ZNSt10bad_typeidC2Ev
@@ -2561,11 +2560,6 @@
          d _ZNSt3__1L7__wcoutE
          d _ZNSt3__1L8__rs_mutE
          W _ZNSt3__1plIcNS_11char_traitsIcEENS_9allocatorIcEEEENS_12basic_stringIT_T0_T1_EEPKS6_RKS9_
-         T _ZNSt8bad_castC1Ev
-         T _ZNSt8bad_castC2Ev
-         T _ZNSt8bad_castD0Ev
-         T _ZNSt8bad_castD1Ev
-         T _ZNSt8bad_castD2Ev
          T _ZNSt9exceptionD0Ev
          T _ZNSt9exceptionD1Ev
          T _ZNSt9exceptionD2Ev
@@ -2719,7 +2713,6 @@
          D _ZTISt15underflow_error
          D _ZTISt16invalid_argument
          D _ZTISt16nested_exception
-         D _ZTISt8bad_cast
          D _ZTISt9exception
          C _ZTSNSt3__110__stdinbufIcEE
          C _ZTSNSt3__110__stdinbufIwEE
@@ -2855,7 +2848,6 @@
          D _ZTSSt15underflow_error
          D _ZTSSt16invalid_argument
          D _ZTSSt16nested_exception
-         D _ZTSSt8bad_cast
          D _ZTSSt9exception
          D _ZTTNSt3__110istrstreamE
          D _ZTTNSt3__110ostrstreamE
@@ -2980,7 +2972,6 @@
          D _ZTVSt15underflow_error
          D _ZTVSt16invalid_argument
          D _ZTVSt16nested_exception
-         D _ZTVSt8bad_cast
          D _ZTVSt9exception
          W _ZThn8_NKSt3__115time_get_bynameIcNS_19istreambuf_iteratorIcNS_11char_traitsIcEEEEE3__XEv
          W _ZThn8_NKSt3__115time_get_bynameIcNS_19istreambuf_iteratorIcNS_11char_traitsIcEEEEE3__cEv
diff --git a/system/lib/libcxxabi/symbols b/system/lib/libcxxabi/symbols
index 8ef205ba..b63f2f82 100644
--- a/system/lib/libcxxabi/symbols
+++ b/system/lib/libcxxabi/symbols
@@ -10,3 +10,10 @@
          D _ZTVN10__cxxabiv121__vmi_class_type_infoE
          D _ZTVN10__cxxabiv123__fundamental_type_infoE
          D _ZTVN10__cxxabiv129__pointer_to_member_type_infoE
+         D _ZTSSt9type_info
+         T _ZNKSt8bad_cast4whatEv
+         T _ZNSt8bad_castC1Ev
+         T _ZNSt8bad_castC2Ev
+         D _ZTISt8bad_cast
+         D _ZTSSt8bad_cast
+         D _ZTVSt8bad_cast
diff --git a/tests/aniso.png b/tests/aniso.png
index 5f5812d2..2bcb2f5f 100644
--- a/tests/aniso.png
+++ b/tests/aniso.png
diff --git a/tests/cubegeom.c b/tests/cubegeom.c
index c137ad80..6158b124 100644
--- a/tests/cubegeom.c
+++ b/tests/cubegeom.c
@@ -207,6 +207,20 @@ int main(int argc, char *argv[])
     glNormalPointer(GL_BYTE, 32, (void*)12);
     glColorPointer(4, GL_UNSIGNED_BYTE, 32, (void*)28);
 
+    int temp; // test glGetPointerv, glGetIntegerv
+    glGetPointerv(GL_VERTEX_ARRAY_POINTER, &temp); assert(temp == 0);
+    glGetPointerv(GL_COLOR_ARRAY_POINTER, &temp); assert(temp == 28);
+    glGetPointerv(GL_TEXTURE_COORD_ARRAY_POINTER, &temp); assert(temp == 16);
+    glGetIntegerv(GL_VERTEX_ARRAY_SIZE, &temp); assert(temp == 3);
+    glGetIntegerv(GL_VERTEX_ARRAY_TYPE, &temp); assert(temp == GL_FLOAT);
+    glGetIntegerv(GL_VERTEX_ARRAY_STRIDE, &temp); assert(temp == 32);
+    glGetIntegerv(GL_COLOR_ARRAY_SIZE, &temp); assert(temp == 4);
+    glGetIntegerv(GL_COLOR_ARRAY_TYPE, &temp); assert(temp == GL_UNSIGNED_BYTE);
+    glGetIntegerv(GL_COLOR_ARRAY_STRIDE, &temp); assert(temp == 32);
+    glGetIntegerv(GL_TEXTURE_COORD_ARRAY_SIZE, &temp); assert(temp == 2);
+    glGetIntegerv(GL_TEXTURE_COORD_ARRAY_TYPE, &temp); assert(temp == GL_FLOAT);
+    glGetIntegerv(GL_TEXTURE_COORD_ARRAY_STRIDE, &temp); assert(temp == 32);
+
     glBindTexture(GL_TEXTURE_2D, texture); // diffuse?
     glActiveTexture(GL_TEXTURE0);
     glActiveTexture(GL_TEXTURE1);
diff --git a/tests/cubegeom_pre2_vao.c b/tests/cubegeom_pre2_vao.c
new file mode 100644
index 00000000..98bedd54
--- /dev/null
+++ b/tests/cubegeom_pre2_vao.c
@@ -0,0 +1,380 @@
+/*
+THIS WORK, INCLUDING THE SOURCE CODE, DOCUMENTATION
+AND RELATED MEDIA AND DATA, IS PLACED INTO THE PUBLIC DOMAIN.
+
+THE ORIGINAL AUTHOR IS KYLE FOLEY.
+
+THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
+OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
+MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
+ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
+RESULTING FROM THE USE, MODIFICATION, OR
+REDISTRIBUTION OF THIS SOFTWARE.
+*/
+
+#if !EMSCRIPTEN
+#define USE_GLEW 1
+#endif
+
+#if USE_GLEW
+#include "GL/glew.h"
+#endif
+
+#include "SDL/SDL.h"
+#if !USE_GLEW
+#include "SDL/SDL_opengl.h"
+#endif
+
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+PFNGLGENVERTEXARRAYSPROC glGenVertexArrays_ = NULL;
+PFNGLBINDVERTEXARRAYPROC glBindVertexArray_ = NULL;
+PFNGLDELETEVERTEXARRAYSPROC glDeleteVertexArrays_ = NULL;
+
+void verify() {
+  int width = 640, height = 480;
+  unsigned char *data = (unsigned char*)malloc(width*height*4);
+  glReadPixels(0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, data);
+  int sum = 0;
+  for (int x = 0; x < width*height*4; x++) {
+    if (x % 4 != 3) sum += x * data[x];
+  }
+#if EMSCRIPTEN
+  int result = sum;
+  REPORT_RESULT();
+#endif
+}
+
+int main(int argc, char *argv[])
+{
+    SDL_Surface *screen;
+    if ( SDL_Init(SDL_INIT_VIDEO) != 0 ) {
+        printf("Unable to initialize SDL: %s\n", SDL_GetError());
+        return 1;
+    }
+
+    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 );
+    screen = SDL_SetVideoMode( 640, 480, 24, SDL_OPENGL );
+    if ( !screen ) {
+        printf("Unable to set video mode: %s\n", SDL_GetError());
+        return 1;
+    }
+    
+    glClearColor( 0, 0, 0, 0 );
+    glClear( GL_COLOR_BUFFER_BIT );
+
+    glGenVertexArrays_ =        (PFNGLGENVERTEXARRAYSPROC)     SDL_GL_GetProcAddress("glGenVertexArrays");
+    assert(glGenVertexArrays_);
+    glBindVertexArray_ =        (PFNGLBINDVERTEXARRAYPROC)     SDL_GL_GetProcAddress("glBindVertexArray");
+    assert(glBindVertexArray_);
+    glDeleteVertexArrays_ =     (PFNGLDELETEVERTEXARRAYSPROC)     SDL_GL_GetProcAddress("glDeleteVertexArrays");
+    assert(glDeleteVertexArrays_);
+
+    // Generate a VAO
+    GLuint vao;
+    glGenVertexArrays_(1, &vao);
+    glBindVertexArray_(vao);
+
+    // Create a texture
+
+    GLuint texture;
+    glGenTextures( 1, &texture );
+    glBindTexture( GL_TEXTURE_2D, texture );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+    GLubyte textureData[16*16*4];
+    for (int x = 0; x < 16; x++) {
+      for (int y = 0; y < 16; y++) {
+        *((int*)&textureData[(x*16 + y) * 4]) = x*16 + ((y*16) << 8);
+      }
+    }
+    glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, 16, 16, 0,
+                  GL_RGBA, GL_UNSIGNED_BYTE, textureData );
+
+    // Create a second texture
+
+    GLuint texture2;
+    glGenTextures( 1, &texture2 );
+    glBindTexture( GL_TEXTURE_2D, texture2 );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+    GLubyte texture2Data[] = { 0xff,    0,    0, 0xff,
+                                  0, 0xff,    0, 0xaa,
+                                  0,    0, 0xff, 0x55,
+                               0x80, 0x90, 0x70,    0 };
+    glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, 2, 2, 0,
+                  GL_RGBA, GL_UNSIGNED_BYTE, texture2Data );
+    
+    // BEGIN
+
+#if USE_GLEW
+    glewInit();
+#endif
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    // original: glFrustum(-0.6435469817188064, 0.6435469817188064 ,-0.48266022190470925, 0.48266022190470925 ,0.5400000214576721, 2048);
+    //glFrustum(-0.6435469817188064, 0.1435469817188064 ,-0.48266022190470925, 0.88266022190470925 ,0.5400000214576721, 2048);
+    GLfloat pm[] = { 1.372136116027832, 0, 0, 0, 0, 0.7910231351852417, 0, 0, -0.6352481842041016, 0.29297152161598206, -1.0005275011062622, -1, 0, 0, -1.080284833908081, 0 };
+    glLoadMatrixf(pm);
+
+    glMatrixMode(GL_MODELVIEW);
+    GLfloat matrixData[] = { -1, 0, 0, 0,
+                              0, 0,-1, 0,
+                              0, 1, 0, 0,
+                              0, 0, 0, 1 };
+    glLoadMatrixf(matrixData);
+    //glTranslated(-512,-512,-527); // XXX this should be uncommented, but if it is then nothing is shown
+
+//    glEnable(GL_CULL_FACE);
+  //  glEnable(GL_DEPTH_TEST);
+
+    //glClear(GL_DEPTH_BUFFER_BIT);
+
+//    glEnableClientState(GL_NORMAL_ARRAY);
+  //  glEnableClientState(GL_COLOR_ARRAY);
+    glEnableClientState(GL_TEXTURE_COORD_ARRAY);
+
+    glActiveTexture(GL_TEXTURE0);
+
+    glEnableClientState(GL_VERTEX_ARRAY);
+
+    GLuint arrayBuffer, elementBuffer;
+    glGenBuffers(1, &arrayBuffer);
+    glGenBuffers(1, &elementBuffer);
+
+    GLubyte arrayData[] = {
+/*
+[0, 0,   0, 67] ==>  128 float
+[0, 0, 128, 67] ==>  256 float
+[0, 0,   0, 68] ==>  512 float
+[0, 0, 128, 68] ==> 1024 float
+
+[vertex x        ] [vertex y         ] [vertex z         ] [nr]                [texture u        ] [texture v        ] [lm u   ] [lm v   ] [color r,g,b,a    ] */
+  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  0
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  1
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  2
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  3
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  4
+  0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  5
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  6
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  7
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  8
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  9
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 10
+  0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 11
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 12
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 13
+  0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 14
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 15
+
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128
+    };
+    assert(sizeof(arrayData) == 1408);
+    glBindBuffer(GL_ARRAY_BUFFER, arrayBuffer);
+    glBufferData(GL_ARRAY_BUFFER, sizeof(arrayData), arrayData, GL_STATIC_DRAW);
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+
+    GLushort elementData[] = { 1, 2, 0, 2, 3, 0, 5, 6, 4, 6, 7, 4, 9, 10, 8, 10, 11, 8, 13, 14, 12, 14, 15, 12 };
+    assert(sizeof(elementData) == 48);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, elementBuffer);
+    glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(elementData), elementData, GL_STATIC_DRAW);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+
+    glBindBuffer(GL_ARRAY_BUFFER, arrayBuffer);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, elementBuffer);
+
+    // sauer vertex data is apparently 0-12: V3F, 12: N1B, 16-24: T2F, 24-28: T2S, 28-32: C4B
+    glVertexPointer(3, GL_FLOAT, 32, (void*)0); // all these apply to the ARRAY_BUFFER that is bound
+    glTexCoordPointer(2, GL_FLOAT, 32, (void*)16);
+//    glClientActiveTexture(GL_TEXTURE1); // XXX seems to be ignored in native build
+//    glTexCoordPointer(2, GL_SHORT, 32, (void*)24);
+//    glClientActiveTexture(GL_TEXTURE0); // likely not needed, it is a cleanup
+//    glNormalPointer(GL_BYTE, 32, (void*)12);
+//    glColorPointer(4, GL_UNSIGNED_BYTE, 32, (void*)28);
+
+    glBindTexture(GL_TEXTURE_2D, texture); // diffuse?
+    glActiveTexture(GL_TEXTURE0);
+    glActiveTexture(GL_TEXTURE1);
+    glBindTexture(GL_TEXTURE_2D, texture2); // lightmap?
+    glActiveTexture(GL_TEXTURE0);
+
+    GLint ok;
+
+    const char *vertexShader = "uniform mat4 u_modelView;\n"
+                               "uniform mat4 u_projection;\n"
+                               "varying vec4 v_texCoord0;\n"
+                               "void main(void)\n"
+                               "{\n" // (gl_ProjectionMatrix * gl_ModelViewMatrix * gl_Vertex)
+                                     // (u_projection * u_modelView * a_position)
+                               "    gl_Position = (u_projection * u_modelView * gl_Vertex) + vec4(200, 0, 0, 0);\n"
+                               "    v_texCoord0.xy = gl_MultiTexCoord0.xy/20.0;\n" // added /100 here
+                               "}\n";
+    const char *fragmentShader = "uniform sampler2D diffusemap;\n"
+                                 "varying vec4 v_texCoord0;\n"
+                                 "void main(void)\n"
+                                 "{\n"
+                                 "    vec4 diffuse = texture2D(diffusemap, v_texCoord0.xy);\n"
+                                 "    gl_FragColor = diffuse;\n"
+                                 "}\n";
+
+    GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+    glShaderSource(vs, 1, &vertexShader, NULL);
+    glCompileShader(vs);
+    glGetShaderiv(vs, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+      printf("Shader compilation error with vertex\n");
+      GLint infoLen = 0;
+      glGetShaderiv (vs, GL_INFO_LOG_LENGTH, &infoLen);
+      if (infoLen > 1)
+      {
+         char* infoLog = (char *)malloc(sizeof(char) * infoLen+1);
+         glGetShaderInfoLog(vs, infoLen, NULL, infoLog);
+         printf("Error compiling shader:\n%s\n", infoLog);            
+      }
+    }
+
+    GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+    glShaderSource(fs, 1, &fragmentShader, NULL);
+    glCompileShader(fs);
+    glGetShaderiv(fs, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+      printf("Shader compilation error with fragment\n");
+      GLint infoLen = 0;
+      glGetShaderiv (vs, GL_INFO_LOG_LENGTH, &infoLen);
+      if (infoLen > 1)
+      {
+         char* infoLog = (char *)malloc(sizeof(char) * infoLen+1);
+         glGetShaderInfoLog(vs, infoLen, NULL, infoLog);
+         printf("Error compiling shader:\n%s\n", infoLog);            
+      }
+    }
+
+    GLuint program = glCreateProgram();
+
+    glAttachShader(program, vs);
+    glAttachShader(program, fs);
+    glLinkProgram(program);
+    glGetProgramiv(program, GL_LINK_STATUS, &ok);
+    assert(ok);
+
+    glUseProgram(program);
+
+    //GLint lightmapLocation = glGetUniformLocation(program, "lightmap");
+    //assert(lightmapLocation >= 0);
+    //glUniform1i(lightmapLocation, 1); // sampler2D? Is it the texture unit?
+
+    GLint diffusemapLocation = glGetUniformLocation(program, "diffusemap");
+    assert(diffusemapLocation >= 0);
+    glUniform1i(diffusemapLocation, 0);
+
+    //GLint texgenscrollLocation = glGetUniformLocation(program, "texgenscroll");
+    //assert(texgenscrollLocation >= 0);
+
+    //GLint colorparamsLocation = glGetUniformLocation(program, "colorparams");
+    //assert(colorparamsLocation >= 0);
+
+    //GLfloat texgenscrollData[] = { 0, 0, 0, 0 };
+    //glUniform4fv(texgenscrollLocation, 1, texgenscrollData);
+
+    //GLfloat colorparamsData[] = { 2, 2, 2, 1 };
+    //glUniform4fv(colorparamsLocation, 1, colorparamsData);
+
+    {
+      GLfloat data[16];
+      glGetFloatv(GL_MODELVIEW_MATRIX, data);
+      printf("Modelview: ");
+      for (int i = 0; i < 16; i++) printf("%.3f, ", data[i]);
+      printf("\n");
+      //memset(data, 0, 16*4);
+      GLint modelViewLocation = glGetUniformLocation(program, "u_modelView");
+      assert(modelViewLocation >= 0);
+      glUniformMatrix4fv(modelViewLocation, 1, GL_FALSE, data);
+    }
+    {
+      GLfloat data[16];
+      glGetFloatv(GL_PROJECTION_MATRIX, data);
+      printf("Projection: ");
+      for (int i = 0; i < 16; i++) printf("%.3f, ", data[i]);
+      printf("\n");
+      //memset(data, 0, 16*4);
+      GLint projectionLocation = glGetUniformLocation(program, "u_projection");
+      assert(projectionLocation >= 0);
+      glUniformMatrix4fv(projectionLocation, 1, GL_FALSE, data);
+    }
+
+/*
+    glBindAttribLocation(program, 0, "a_position");
+    glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 32, (void*)0);
+    glEnableVertexAttribArray(0);
+
+    glBindAttribLocation(program, 1, "v_texCoord0");
+    glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 32, (void*)16);
+    glEnableVertexAttribArray(1);
+*/
+
+    // stop recording in the VAO
+
+    glBindVertexArray_(0);
+
+    // unbind all the stuff the VAO would save for us, so this is a valid test
+
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+
+    glDisableClientState(GL_TEXTURE_COORD_ARRAY);
+    glDisableClientState(GL_VERTEX_ARRAY);
+
+    // draw with VAO
+
+    glBindVertexArray_(vao);
+
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)12);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*) 0);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)24);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)36);
+
+    // END
+
+    SDL_GL_SwapBuffers();
+
+    verify();
+  
+#if !EMSCRIPTEN
+    SDL_Delay(1500);
+#endif
+
+ //   SDL_Quit();
+    
+    return 0;
+}
+
diff --git a/tests/cubegeom_pre2_vao2.c b/tests/cubegeom_pre2_vao2.c
new file mode 100644
index 00000000..e3ab4d62
--- /dev/null
+++ b/tests/cubegeom_pre2_vao2.c
@@ -0,0 +1,381 @@
+/*
+THIS WORK, INCLUDING THE SOURCE CODE, DOCUMENTATION
+AND RELATED MEDIA AND DATA, IS PLACED INTO THE PUBLIC DOMAIN.
+
+THE ORIGINAL AUTHOR IS KYLE FOLEY.
+
+THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
+OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
+MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
+ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
+RESULTING FROM THE USE, MODIFICATION, OR
+REDISTRIBUTION OF THIS SOFTWARE.
+*/
+
+#if !EMSCRIPTEN
+#define USE_GLEW 1
+#endif
+
+#if USE_GLEW
+#include "GL/glew.h"
+#endif
+
+#include "SDL/SDL.h"
+#if !USE_GLEW
+#include "SDL/SDL_opengl.h"
+#endif
+
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+PFNGLGENVERTEXARRAYSPROC glGenVertexArrays_ = NULL;
+PFNGLBINDVERTEXARRAYPROC glBindVertexArray_ = NULL;
+PFNGLDELETEVERTEXARRAYSPROC glDeleteVertexArrays_ = NULL;
+
+void verify() {
+  int width = 640, height = 480;
+  unsigned char *data = (unsigned char*)malloc(width*height*4);
+  glReadPixels(0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, data);
+  int sum = 0;
+  for (int x = 0; x < width*height*4; x++) {
+    if (x % 4 != 3) sum += x * data[x];
+  }
+#if EMSCRIPTEN
+  int result = sum;
+  REPORT_RESULT();
+#endif
+}
+
+int main(int argc, char *argv[])
+{
+    SDL_Surface *screen;
+    if ( SDL_Init(SDL_INIT_VIDEO) != 0 ) {
+        printf("Unable to initialize SDL: %s\n", SDL_GetError());
+        return 1;
+    }
+
+    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 );
+    screen = SDL_SetVideoMode( 640, 480, 24, SDL_OPENGL );
+    if ( !screen ) {
+        printf("Unable to set video mode: %s\n", SDL_GetError());
+        return 1;
+    }
+    
+    glClearColor( 0, 0, 0, 0 );
+    glClear( GL_COLOR_BUFFER_BIT );
+
+    glGenVertexArrays_ =        (PFNGLGENVERTEXARRAYSPROC)     SDL_GL_GetProcAddress("glGenVertexArrays");
+    assert(glGenVertexArrays_);
+    glBindVertexArray_ =        (PFNGLBINDVERTEXARRAYPROC)     SDL_GL_GetProcAddress("glBindVertexArray");
+    assert(glBindVertexArray_);
+    glDeleteVertexArrays_ =     (PFNGLDELETEVERTEXARRAYSPROC)     SDL_GL_GetProcAddress("glDeleteVertexArrays");
+    assert(glDeleteVertexArrays_);
+
+    glEnableClientState(GL_TEXTURE_COORD_ARRAY); // enabling it *before* the vao does nothing, the vao should wipe it out!
+
+    // Generate a VAO
+    GLuint vao;
+    glGenVertexArrays_(1, &vao);
+    glBindVertexArray_(vao);
+
+    // Create a texture
+
+    GLuint texture;
+    glGenTextures( 1, &texture );
+    glBindTexture( GL_TEXTURE_2D, texture );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+    GLubyte textureData[16*16*4];
+    for (int x = 0; x < 16; x++) {
+      for (int y = 0; y < 16; y++) {
+        *((int*)&textureData[(x*16 + y) * 4]) = x*16 + ((y*16) << 8);
+      }
+    }
+    glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, 16, 16, 0,
+                  GL_RGBA, GL_UNSIGNED_BYTE, textureData );
+
+    // Create a second texture
+
+    GLuint texture2;
+    glGenTextures( 1, &texture2 );
+    glBindTexture( GL_TEXTURE_2D, texture2 );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+    GLubyte texture2Data[] = { 0xff,    0,    0, 0xff,
+                                  0, 0xff,    0, 0xaa,
+                                  0,    0, 0xff, 0x55,
+                               0x80, 0x90, 0x70,    0 };
+    glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, 2, 2, 0,
+                  GL_RGBA, GL_UNSIGNED_BYTE, texture2Data );
+    
+    // BEGIN
+
+#if USE_GLEW
+    glewInit();
+#endif
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    // original: glFrustum(-0.6435469817188064, 0.6435469817188064 ,-0.48266022190470925, 0.48266022190470925 ,0.5400000214576721, 2048);
+    //glFrustum(-0.6435469817188064, 0.1435469817188064 ,-0.48266022190470925, 0.88266022190470925 ,0.5400000214576721, 2048);
+    GLfloat pm[] = { 1.372136116027832, 0, 0, 0, 0, 0.7910231351852417, 0, 0, -0.6352481842041016, 0.29297152161598206, -1.0005275011062622, -1, 0, 0, -1.080284833908081, 0 };
+    glLoadMatrixf(pm);
+
+    glMatrixMode(GL_MODELVIEW);
+    GLfloat matrixData[] = { -1, 0, 0, 0,
+                              0, 0,-1, 0,
+                              0, 1, 0, 0,
+                              0, 0, 0, 1 };
+    glLoadMatrixf(matrixData);
+    //glTranslated(-512,-512,-527); // XXX this should be uncommented, but if it is then nothing is shown
+
+//    glEnable(GL_CULL_FACE);
+  //  glEnable(GL_DEPTH_TEST);
+
+    //glClear(GL_DEPTH_BUFFER_BIT);
+
+//    glEnableClientState(GL_NORMAL_ARRAY);
+  //  glEnableClientState(GL_COLOR_ARRAY);
+
+    glActiveTexture(GL_TEXTURE0);
+
+    glEnableClientState(GL_VERTEX_ARRAY);
+
+    GLuint arrayBuffer, elementBuffer;
+    glGenBuffers(1, &arrayBuffer);
+    glGenBuffers(1, &elementBuffer);
+
+    GLubyte arrayData[] = {
+/*
+[0, 0,   0, 67] ==>  128 float
+[0, 0, 128, 67] ==>  256 float
+[0, 0,   0, 68] ==>  512 float
+[0, 0, 128, 68] ==> 1024 float
+
+[vertex x        ] [vertex y         ] [vertex z         ] [nr]                [texture u        ] [texture v        ] [lm u   ] [lm v   ] [color r,g,b,a    ] */
+  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  0
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  1
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  2
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  3
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  4
+  0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  5
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  6
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  7
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  8
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  9
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 10
+  0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 11
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 12
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 13
+  0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 14
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 15
+
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128
+    };
+    assert(sizeof(arrayData) == 1408);
+    glBindBuffer(GL_ARRAY_BUFFER, arrayBuffer);
+    glBufferData(GL_ARRAY_BUFFER, sizeof(arrayData), arrayData, GL_STATIC_DRAW);
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+
+    GLushort elementData[] = { 1, 2, 0, 2, 3, 0, 5, 6, 4, 6, 7, 4, 9, 10, 8, 10, 11, 8, 13, 14, 12, 14, 15, 12 };
+    assert(sizeof(elementData) == 48);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, elementBuffer);
+    glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(elementData), elementData, GL_STATIC_DRAW);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+
+    glBindBuffer(GL_ARRAY_BUFFER, arrayBuffer);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, elementBuffer);
+
+    // sauer vertex data is apparently 0-12: V3F, 12: N1B, 16-24: T2F, 24-28: T2S, 28-32: C4B
+    glVertexPointer(3, GL_FLOAT, 32, (void*)0); // all these apply to the ARRAY_BUFFER that is bound
+    glTexCoordPointer(2, GL_FLOAT, 32, (void*)16);
+//    glClientActiveTexture(GL_TEXTURE1); // XXX seems to be ignored in native build
+//    glTexCoordPointer(2, GL_SHORT, 32, (void*)24);
+//    glClientActiveTexture(GL_TEXTURE0); // likely not needed, it is a cleanup
+//    glNormalPointer(GL_BYTE, 32, (void*)12);
+//    glColorPointer(4, GL_UNSIGNED_BYTE, 32, (void*)28);
+
+    glBindTexture(GL_TEXTURE_2D, texture); // diffuse?
+    glActiveTexture(GL_TEXTURE0);
+    glActiveTexture(GL_TEXTURE1);
+    glBindTexture(GL_TEXTURE_2D, texture2); // lightmap?
+    glActiveTexture(GL_TEXTURE0);
+
+    GLint ok;
+
+    const char *vertexShader = "uniform mat4 u_modelView;\n"
+                               "uniform mat4 u_projection;\n"
+                               "varying vec4 v_texCoord0;\n"
+                               "void main(void)\n"
+                               "{\n" // (gl_ProjectionMatrix * gl_ModelViewMatrix * gl_Vertex)
+                                     // (u_projection * u_modelView * a_position)
+                               "    gl_Position = (u_projection * u_modelView * gl_Vertex) + vec4(200, 0, 0, 0);\n"
+                               "    v_texCoord0.xy = gl_MultiTexCoord0.xy/20.0;\n" // added /100 here
+                               "}\n";
+    const char *fragmentShader = "uniform sampler2D diffusemap;\n"
+                                 "varying vec4 v_texCoord0;\n"
+                                 "void main(void)\n"
+                                 "{\n"
+                                 "    vec4 diffuse = texture2D(diffusemap, v_texCoord0.xy);\n"
+                                 "    gl_FragColor = diffuse;\n"
+                                 "}\n";
+
+    GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+    glShaderSource(vs, 1, &vertexShader, NULL);
+    glCompileShader(vs);
+    glGetShaderiv(vs, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+      printf("Shader compilation error with vertex\n");
+      GLint infoLen = 0;
+      glGetShaderiv (vs, GL_INFO_LOG_LENGTH, &infoLen);
+      if (infoLen > 1)
+      {
+         char* infoLog = (char *)malloc(sizeof(char) * infoLen+1);
+         glGetShaderInfoLog(vs, infoLen, NULL, infoLog);
+         printf("Error compiling shader:\n%s\n", infoLog);            
+      }
+    }
+
+    GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+    glShaderSource(fs, 1, &fragmentShader, NULL);
+    glCompileShader(fs);
+    glGetShaderiv(fs, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+      printf("Shader compilation error with fragment\n");
+      GLint infoLen = 0;
+      glGetShaderiv (vs, GL_INFO_LOG_LENGTH, &infoLen);
+      if (infoLen > 1)
+      {
+         char* infoLog = (char *)malloc(sizeof(char) * infoLen+1);
+         glGetShaderInfoLog(vs, infoLen, NULL, infoLog);
+         printf("Error compiling shader:\n%s\n", infoLog);            
+      }
+    }
+
+    GLuint program = glCreateProgram();
+
+    glAttachShader(program, vs);
+    glAttachShader(program, fs);
+    glLinkProgram(program);
+    glGetProgramiv(program, GL_LINK_STATUS, &ok);
+    assert(ok);
+
+    glUseProgram(program);
+
+    //GLint lightmapLocation = glGetUniformLocation(program, "lightmap");
+    //assert(lightmapLocation >= 0);
+    //glUniform1i(lightmapLocation, 1); // sampler2D? Is it the texture unit?
+
+    GLint diffusemapLocation = glGetUniformLocation(program, "diffusemap");
+    assert(diffusemapLocation >= 0);
+    glUniform1i(diffusemapLocation, 0);
+
+    //GLint texgenscrollLocation = glGetUniformLocation(program, "texgenscroll");
+    //assert(texgenscrollLocation >= 0);
+
+    //GLint colorparamsLocation = glGetUniformLocation(program, "colorparams");
+    //assert(colorparamsLocation >= 0);
+
+    //GLfloat texgenscrollData[] = { 0, 0, 0, 0 };
+    //glUniform4fv(texgenscrollLocation, 1, texgenscrollData);
+
+    //GLfloat colorparamsData[] = { 2, 2, 2, 1 };
+    //glUniform4fv(colorparamsLocation, 1, colorparamsData);
+
+    {
+      GLfloat data[16];
+      glGetFloatv(GL_MODELVIEW_MATRIX, data);
+      printf("Modelview: ");
+      for (int i = 0; i < 16; i++) printf("%.3f, ", data[i]);
+      printf("\n");
+      //memset(data, 0, 16*4);
+      GLint modelViewLocation = glGetUniformLocation(program, "u_modelView");
+      assert(modelViewLocation >= 0);
+      glUniformMatrix4fv(modelViewLocation, 1, GL_FALSE, data);
+    }
+    {
+      GLfloat data[16];
+      glGetFloatv(GL_PROJECTION_MATRIX, data);
+      printf("Projection: ");
+      for (int i = 0; i < 16; i++) printf("%.3f, ", data[i]);
+      printf("\n");
+      //memset(data, 0, 16*4);
+      GLint projectionLocation = glGetUniformLocation(program, "u_projection");
+      assert(projectionLocation >= 0);
+      glUniformMatrix4fv(projectionLocation, 1, GL_FALSE, data);
+    }
+
+/*
+    glBindAttribLocation(program, 0, "a_position");
+    glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 32, (void*)0);
+    glEnableVertexAttribArray(0);
+
+    glBindAttribLocation(program, 1, "v_texCoord0");
+    glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 32, (void*)16);
+    glEnableVertexAttribArray(1);
+*/
+
+    // stop recording in the VAO
+
+    glBindVertexArray_(0);
+
+    // unbind all the stuff the VAO would save for us, so this is a valid test
+
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+
+    glDisableClientState(GL_TEXTURE_COORD_ARRAY);
+    glDisableClientState(GL_VERTEX_ARRAY);
+
+    // draw with VAO
+
+    glBindVertexArray_(vao);
+
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)12);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*) 0);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)24);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)36);
+
+    // END
+
+    SDL_GL_SwapBuffers();
+
+    verify();
+  
+#if !EMSCRIPTEN
+    SDL_Delay(1500);
+#endif
+
+ //   SDL_Quit();
+    
+    return 0;
+}
+
diff --git a/tests/cubegeom_pre_vao.c b/tests/cubegeom_pre_vao.c
new file mode 100644
index 00000000..f1d35fb5
--- /dev/null
+++ b/tests/cubegeom_pre_vao.c
@@ -0,0 +1,333 @@
+/*
+THIS WORK, INCLUDING THE SOURCE CODE, DOCUMENTATION
+AND RELATED MEDIA AND DATA, IS PLACED INTO THE PUBLIC DOMAIN.
+
+THE ORIGINAL AUTHOR IS KYLE FOLEY.
+
+THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
+OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
+MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
+ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
+RESULTING FROM THE USE, MODIFICATION, OR
+REDISTRIBUTION OF THIS SOFTWARE.
+*/
+
+#if !EMSCRIPTEN
+#define USE_GLEW 1
+#endif
+
+#if USE_GLEW
+#include "GL/glew.h"
+#endif
+
+#include "SDL/SDL.h"
+#if !USE_GLEW
+#include "SDL/SDL_opengl.h"
+#endif
+
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+void verify() {
+  int width = 640, height = 480;
+  unsigned char *data = (unsigned char*)malloc(width*height*4);
+  glReadPixels(0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, data);
+  int sum = 0;
+  for (int x = 0; x < width*height*4; x++) {
+    if (x % 4 != 3) sum += x * data[x];
+  }
+#if EMSCRIPTEN
+  int result = sum;
+  REPORT_RESULT();
+#endif
+}
+
+int main(int argc, char *argv[])
+{
+    SDL_Surface *screen;
+    if ( SDL_Init(SDL_INIT_VIDEO) != 0 ) {
+        printf("Unable to initialize SDL: %s\n", SDL_GetError());
+        return 1;
+    }
+
+    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 );
+    screen = SDL_SetVideoMode( 640, 480, 24, SDL_OPENGL );
+    if ( !screen ) {
+        printf("Unable to set video mode: %s\n", SDL_GetError());
+        return 1;
+    }
+    
+    glClearColor( 0, 0, 0, 0 );
+    glClear( GL_COLOR_BUFFER_BIT );
+
+    // Create a texture
+
+    GLuint texture;
+    glGenTextures( 1, &texture );
+    glBindTexture( GL_TEXTURE_2D, texture );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+    GLubyte textureData[16*16*4];
+    for (int x = 0; x < 16; x++) {
+      for (int y = 0; y < 16; y++) {
+        *((int*)&textureData[(x*16 + y) * 4]) = x*16 + ((y*16) << 8);
+      }
+    }
+    glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, 16, 16, 0, 
+                  GL_RGBA, GL_UNSIGNED_BYTE, textureData );
+
+    // Create a second texture
+
+    GLuint texture2;
+    glGenTextures( 1, &texture2 );
+    glBindTexture( GL_TEXTURE_2D, texture2 );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+    glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+    GLubyte texture2Data[] = { 0xff,    0,    0, 0xff,
+                                  0, 0xff,    0, 0xaa,
+                                  0,    0, 0xff, 0x55,
+                               0x80, 0x90, 0x70,    0 };
+    glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, 2, 2, 0,
+                  GL_RGBA, GL_UNSIGNED_BYTE, texture2Data );
+    
+    // BEGIN
+
+#if USE_GLEW
+    glewInit();
+#endif
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    // original: glFrustum(-0.6435469817188064, 0.6435469817188064 ,-0.48266022190470925, 0.48266022190470925 ,0.5400000214576721, 2048);
+    //glFrustum(-0.6435469817188064, 0.1435469817188064 ,-0.48266022190470925, 0.88266022190470925 ,0.5400000214576721, 2048);
+    GLfloat pm[] = { 1.372136116027832, 0, 0, 0, 0, 0.7910231351852417, 0, 0, -0.6352481842041016, 0.29297152161598206, -1.0005275011062622, -1, 0, 0, -1.080284833908081, 0 };
+    glLoadMatrixf(pm);
+
+    glMatrixMode(GL_MODELVIEW);
+    GLfloat matrixData[] = { -1, 0, 0, 0,
+                              0, 0,-1, 0,
+                              0, 1, 0, 0,
+                              0, 0, 0, 1 };
+    glLoadMatrixf(matrixData);
+
+    glActiveTexture(GL_TEXTURE0);
+
+    GLuint arrayBuffer, elementBuffer;
+    glGenBuffers(1, &arrayBuffer);
+    glGenBuffers(1, &elementBuffer);
+
+    GLubyte arrayData[] = {
+/*
+[0, 0,   0, 67] ==>  128 float
+[0, 0, 128, 67] ==>  256 float
+[0, 0,   0, 68] ==>  512 float
+[0, 0, 128, 68] ==> 1024 float
+
+[vertex x        ] [vertex y         ] [vertex z         ] [nr]                [texture u        ] [texture v        ] [lm u   ] [lm v   ] [color r,g,b,a    ] */
+  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  0
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  1
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  2
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  3
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  4
+  0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128, //  5
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  6
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  7
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  8
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, //  9
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 10
+  0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 11
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 12
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0,   0,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 13
+  0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 14
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  67,   0,   0, 128,  67,   0,   0,   0,   0, 128, 128, 128, 128, // 15
+
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,   0,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0,   0,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0, 128,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128,
+  0,   0, 128,  68,   0,   0, 128,  68,   0,   0,   0,  68,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 128, 128, 128, 128
+    };
+
+    // Generate a VAO
+    GLuint vao;
+    glGenVertexArrays(1, &vao);
+    glBindVertexArray(vao);
+
+    assert(sizeof(arrayData) == 1408);
+    glBindBuffer(GL_ARRAY_BUFFER, arrayBuffer);
+    glBufferData(GL_ARRAY_BUFFER, sizeof(arrayData), arrayData, GL_STATIC_DRAW);
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+
+    GLushort elementData[] = { 1, 2, 0, 2, 3, 0, 5, 6, 4, 6, 7, 4, 9, 10, 8, 10, 11, 8, 13, 14, 12, 14, 15, 12 };
+    assert(sizeof(elementData) == 48);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, elementBuffer);
+    glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(elementData), elementData, GL_STATIC_DRAW);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+
+    glBindBuffer(GL_ARRAY_BUFFER, arrayBuffer);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, elementBuffer);
+
+    glBindTexture(GL_TEXTURE_2D, texture); // diffuse?
+    glActiveTexture(GL_TEXTURE0);
+    glActiveTexture(GL_TEXTURE1);
+    glBindTexture(GL_TEXTURE_2D, texture2); // lightmap?
+    glActiveTexture(GL_TEXTURE0);
+
+    GLint ok;
+
+    const char *vertexShader = "attribute vec4 a_position;\n"
+                               "attribute vec4 a_texCoord0;\n"
+                               "uniform mat4 u_modelView;\n"
+                               "uniform mat4 u_projection;\n"
+                               "varying vec4 v_texCoord0;\n"
+                               "void main(void)\n"
+                               "{\n"
+                               "    gl_Position = (u_projection * u_modelView * a_position) + vec4(200, 0, 0, 0);\n"
+                               "    v_texCoord0.xy = a_texCoord0.xy/20.0;\n" // added /20 here
+                               "}\n";
+    const char *fragmentShader = "uniform sampler2D diffusemap;\n"
+                                 "varying vec4 v_texCoord0;\n"
+                                 "void main(void)\n"
+                                 "{\n"
+                                 "    vec4 diffuse = texture2D(diffusemap, v_texCoord0.xy);\n"
+                                 "    gl_FragColor = diffuse;\n"
+                                 "}\n";
+
+    GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+    glShaderSource(vs, 1, &vertexShader, NULL);
+    glCompileShader(vs);
+    glGetShaderiv(vs, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+      printf("Shader compilation error with vertex\n");
+      GLint infoLen = 0;
+      glGetShaderiv (vs, GL_INFO_LOG_LENGTH, &infoLen);
+      if (infoLen > 1)
+      {
+         char* infoLog = (char *)malloc(sizeof(char) * infoLen+1);
+         glGetShaderInfoLog(vs, infoLen, NULL, infoLog);
+         printf("Error compiling shader:\n%s\n", infoLog);            
+      }
+    }
+
+    GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+    glShaderSource(fs, 1, &fragmentShader, NULL);
+    glCompileShader(fs);
+    glGetShaderiv(fs, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+      printf("Shader compilation error with fragment\n");
+      GLint infoLen = 0;
+      glGetShaderiv (vs, GL_INFO_LOG_LENGTH, &infoLen);
+      if (infoLen > 1)
+      {
+         char* infoLog = (char *)malloc(sizeof(char) * infoLen+1);
+         glGetShaderInfoLog(vs, infoLen, NULL, infoLog);
+         printf("Error compiling shader:\n%s\n", infoLog);            
+      }
+    }
+
+    GLuint program = glCreateProgram();
+
+    glAttachShader(program, vs);
+    glAttachShader(program, fs);
+    glLinkProgram(program);
+    glGetProgramiv(program, GL_LINK_STATUS, &ok);
+    assert(ok);
+
+    glUseProgram(program);
+
+    GLint diffusemapLocation = glGetUniformLocation(program, "diffusemap");
+    assert(diffusemapLocation >= 0);
+    glUniform1i(diffusemapLocation, 0);
+
+    {
+      GLfloat data[16];
+      glGetFloatv(GL_MODELVIEW_MATRIX, data);
+      printf("Modelview: ");
+      for (int i = 0; i < 16; i++) printf("%.3f, ", data[i]);
+      printf("\n");
+      GLint modelViewLocation = glGetUniformLocation(program, "u_modelView");
+      assert(modelViewLocation >= 0);
+      glUniformMatrix4fv(modelViewLocation, 1, GL_FALSE, data);
+    }
+    {
+      GLfloat data[16];
+      glGetFloatv(GL_PROJECTION_MATRIX, data);
+      printf("Projection: ");
+      for (int i = 0; i < 16; i++) printf("%.3f, ", data[i]);
+      printf("\n");
+      GLint projectionLocation = glGetUniformLocation(program, "u_projection");
+      assert(projectionLocation >= 0);
+      glUniformMatrix4fv(projectionLocation, 1, GL_FALSE, data);
+    }
+
+    glBindAttribLocation(program, 0, "a_position");
+    glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 32, (void*)0);
+    glEnableVertexAttribArray(0);
+
+    glBindAttribLocation(program, 1, "a_texCoord0");
+    glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 32, (void*)16);
+    glEnableVertexAttribArray(1);
+
+    // stop recording in the VAO
+
+    glBindVertexArray(0);
+
+    // unbind all the stuff the VAO would save for us, so this is a valid test
+
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+
+    glDisableVertexAttribArray(0);
+    glDisableVertexAttribArray(1);
+
+    // draw with VAO
+
+    glBindVertexArray(vao);
+
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)12);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*) 0);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)24);
+    glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, (void*)36);
+
+    glBindVertexArray(0);
+
+    glDeleteVertexArrays(1, &vao);
+
+    // END
+
+    SDL_GL_SwapBuffers();
+
+    verify();
+    
+#if !EMSCRIPTEN
+    SDL_Delay(1500);
+#endif
+
+    SDL_Quit();
+    
+    return 0;
+}
diff --git a/tests/float_tex.png b/tests/float_tex.png
index 8c3c6502..8fc005ff 100644
--- a/tests/float_tex.png
+++ b/tests/float_tex.png
diff --git a/tests/fuzz/1.c b/tests/fuzz/1.c
new file mode 100644
index 00000000..2017fb76
--- /dev/null
+++ b/tests/fuzz/1.c
@@ -0,0 +1,117 @@
+/*
+ * This is a RANDOMLY GENERATED PROGRAM.
+ *
+ * Generator: csmith 2.2.0
+ * Git version: a8697aa
+ * Options:   --no-volatiles --no-math64 --max-block-depth 2 --max-block-size 2 --max-expr-complexity 2 --max-funcs 2
+ * Seed:      1880513882
+ */
+
+#include "csmith.h"
+
+
+static long __undefined;
+
+/* --- Struct/Union Declarations --- */
+union U0 {
+   int32_t  f0;
+   int8_t  f1;
+   const int8_t  f2;
+};
+
+/* --- GLOBAL VARIABLES --- */
+static uint32_t g_4 = 9UL;
+static int32_t g_6 = 0xB9DD952EL;
+static const int32_t *g_5 = &g_6;
+static union U0 g_7[3][9][6] = {{{{-8L},{5L},{0x901516EAL},{7L},{7L},{0x901516EAL}},{{0x520EA0C8L},{0x520EA0C8L},{0x0920A6FFL},{0x0021FBB9L},{0x888C5540L},{-8L}},{{0xDC87A9B6L},{0x42B48371L},{8L},{7L},{-8L},{0x0920A6FFL}},{{-1L},{0xDC87A9B6L},{8L},{5L},{0x520EA0C8L},{-8L}},{{-3L},{5L},{0x0920A6FFL},{0x865B49D5L},{0xDC87A9B6L},{0x901516EAL}},{{0x865B49D5L},{0xDC87A9B6L},{0x901516EAL},{0x0021FBB9L},{-1L},{-1L}},{{0x865B49D5L},{0x42B48371L},{0x42B48371L},{0x865B49D5L},{-3L},{0x0920A6FFL}},{{-3L},{0x520EA0C8L},{0x0021FBB9L},{5L},{0x865B49D5L},{-1L}},{{-1L},{5L},{0L},{7L},{0x865B49D5L},{0x901516EAL}}},{{{0xDC87A9B6L},{0x520EA0C8L},{7L},{0x0021FBB9L},{-3L},{-8L}},{{0x520EA0C8L},{0x42B48371L},{5L},{7L},{-1L},{0x0920A6FFL}},{{-8L},{0xDC87A9B6L},{5L},{5L},{0xDC87A9B6L},{-8L}},{{0x888C5540L},{5L},{7L},{0x865B49D5L},{0x520EA0C8L},{0x901516EAL}},{{7L},{0xDC87A9B6L},{0L},{0x0021FBB9L},{-8L},{-1L}},{{7L},{0x42B48371L},{0x0021FBB9L},{0x865B49D5L},{0x888C5540L},{0x0920A6FFL}},{{0x888C5540L},{0x520EA0C8L},{0x42B48371L},{5L},{7L},{-1L}},{{-8L},{5L},{0x901516EAL},{7L},{7L},{0x901516EAL}},{{0x520EA0C8L},{0x520EA0C8L},{0x0920A6FFL},{0x0021FBB9L},{0x888C5540L},{-8L}}},{{{0xDC87A9B6L},{0x42B48371L},{8L},{7L},{-8L},{0x0920A6FFL}},{{-1L},{0xDC87A9B6L},{8L},{5L},{0x520EA0C8L},{-8L}},{{-3L},{5L},{0x0920A6FFL},{0x865B49D5L},{0xDC87A9B6L},{0x901516EAL}},{{0x865B49D5L},{0xDC87A9B6L},{0x901516EAL},{0x0021FBB9L},{-1L},{-1L}},{{0x865B49D5L},{0x42B48371L},{0x42B48371L},{0x865B49D5L},{-3L},{0x0920A6FFL}},{{-3L},{0x520EA0C8L},{0x0021FBB9L},{5L},{0x865B49D5L},{-1L}},{{-1L},{5L},{0L},{7L},{0x865B49D5L},{0x901516EAL}},{{0xDC87A9B6L},{0x520EA0C8L},{7L},{0x0021FBB9L},{-3L},{-8L}},{{0x520EA0C8L},{0x42B48371L},{5L},{7L},{-1L},{0x0920A6FFL}}}};
+static int32_t g_31 = 1L;
+static uint32_t g_32 = 4294967289UL;
+
+
+/* --- FORWARD DECLARATIONS --- */
+static uint16_t  func_1(void);
+static int32_t * func_9(int32_t * p_10, uint16_t  p_11);
+
+
+/* --- FUNCTIONS --- */
+/* ------------------------------------------ */
+/* 
+ * reads : g_4 g_5 g_7 g_32
+ * writes: g_5 g_32
+ */
+static uint16_t  func_1(void)
+{ /* block id: 0 */
+    int32_t l_3 = (-1L);
+    if (((safe_unary_minus_func_int16_t_s(l_3)) >= g_4))
+    { /* block id: 1 */
+        int32_t *l_8 = &l_3;
+        g_5 = g_5;
+        (*l_8) ^= (g_7[1][3][2] , 0x44688D23L);
+    }
+    else
+    { /* block id: 4 */
+        int32_t *l_12 = &l_3;
+        int32_t **l_13 = &l_12;
+        (*l_13) = func_9(((*l_13) = l_12), l_3);
+    }
+    return l_3;
+}
+
+
+/* ------------------------------------------ */
+/* 
+ * reads : g_32
+ * writes: g_32
+ */
+static int32_t * func_9(int32_t * p_10, uint16_t  p_11)
+{ /* block id: 6 */
+    int32_t *l_14 = &g_6;
+    int32_t *l_15 = &g_7[1][3][2].f0;
+    int32_t *l_16 = &g_6;
+    int32_t *l_17 = &g_6;
+    int32_t *l_18 = &g_6;
+    int32_t *l_19 = &g_6;
+    int32_t *l_20 = &g_6;
+    int32_t *l_21 = &g_6;
+    int32_t *l_22 = &g_6;
+    int32_t *l_23 = &g_7[1][3][2].f0;
+    int32_t l_24[8] = {0xF9F11119L,0xF9F11119L,0xF9F11119L,0xF9F11119L,0xF9F11119L,0xF9F11119L,0xF9F11119L,0xF9F11119L};
+    int32_t *l_25 = &l_24[7];
+    int32_t *l_26 = &l_24[2];
+    int32_t *l_27 = &l_24[2];
+    int32_t *l_28 = &l_24[0];
+    int32_t *l_29 = &g_7[1][3][2].f0;
+    int32_t *l_30[10] = {(void*)0,(void*)0,(void*)0,(void*)0,(void*)0,(void*)0,(void*)0,(void*)0,(void*)0,(void*)0};
+    int i;
+    ++g_32;
+    return p_10;
+}
+
+
+
+
+/* ---------------------------------------- */
+int main (int argc, char* argv[])
+{
+    int i, j, k;
+    int print_hash_value = 0;
+    if (argc == 2 && strcmp(argv[1], "1") == 0) print_hash_value = 1;
+    platform_main_begin();
+    crc32_gentab();
+    func_1();
+    for (i = 0; i < 3; i++)
+    {
+        for (j = 0; j < 1; j++)
+        {
+            for (k = 0; k < 1; k++)
+            {
+                transparent_crc(g_7[i][j][k].f0, "g_7[i][j][k].f0", print_hash_value);
+                transparent_crc(g_7[i][j][k].f1, "g_7[i][j][k].f1", print_hash_value);
+                if (print_hash_value) printf("index = [%d][%d][%d]\n", i, j, k);
+            }
+        }
+    }
+    platform_main_end(crc32_context ^ 0xFFFFFFFFUL, print_hash_value);
+    return 0;
+}
+
diff --git a/tests/fuzz/1.c.txt b/tests/fuzz/1.c.txt
new file mode 100644
index 00000000..701369f3
--- /dev/null
+++ b/tests/fuzz/1.c.txt
@@ -0,0 +1 @@
+checksum = 504CCDC
diff --git a/tests/fuzz/creduce_tester.py b/tests/fuzz/creduce_tester.py
new file mode 100755
index 00000000..c3460e9d
--- /dev/null
+++ b/tests/fuzz/creduce_tester.py
@@ -0,0 +1,53 @@
+#!/usr/bin/python
+
+'''
+Runs csmith, a C fuzzer, and looks for bugs
+'''
+
+import os, sys, difflib
+from subprocess import Popen, PIPE, STDOUT
+
+sys.path += [os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(__file__))), 'tools')]
+import shared
+
+filename = sys.argv[1]
+print 'testing file', filename
+
+print '2) Compile natively'
+shared.try_delete(filename)
+shared.execute([shared.CLANG_CC, '-O2', filename + '.c', '-o', filename] + CSMITH_CFLAGS, stderr=PIPE)
+assert os.path.exists(filename)
+print '3) Run natively'
+try:
+  correct = shared.timeout_run(Popen([filename], stdout=PIPE, stderr=PIPE), 3)
+except Exception, e:
+  print 'Failed or infinite looping in native, skipping', e
+  notes['invalid'] += 1
+  os.exit(0) # boring 
+
+print '4) Compile JS-ly and compare'
+
+def try_js(args):
+  shared.try_delete(filename + '.js')
+  shared.execute([shared.EMCC, '-O2', '-s', 'ASM_JS=1', '-s', 'PRECISE_I64_MATH=1', '-s', 'PRECISE_I32_MUL=1', filename + '.c', '-o', filename + '.js'] + CSMITH_CFLAGS + args, stderr=PIPE)
+  assert os.path.exists(filename + '.js')
+  js = shared.run_js(filename + '.js', stderr=PIPE, engine=engine1)
+  assert correct == js, ''.join([a.rstrip()+'\n' for a in difflib.unified_diff(correct.split('\n'), js.split('\n'), fromfile='expected', tofile='actual')])
+
+# Try normally, then try unaligned because csmith does generate nonportable code that requires x86 alignment
+ok = False
+normal = True
+for args, note in [([], None), (['-s', 'UNALIGNED_MEMORY=1'], 'unaligned')]:
+  try:
+    try_js(args)
+    ok = True
+    if note:
+      notes[note] += 1
+    break
+  except Exception, e:
+    print e
+    normal = False
+if not ok: sys.exit(1)
+
+sys.exit(0) # boring
+
diff --git a/tests/fuzz/csmith.h b/tests/fuzz/csmith.h
new file mode 100644
index 00000000..f1334a37
--- /dev/null
+++ b/tests/fuzz/csmith.h
@@ -0,0 +1,130 @@
+/* -*- mode: C -*-
+ *
+ * Copyright (c) 2007-2010 The University of Utah
+ * All rights reserved.
+ *
+ * This file is part of `csmith', a random generator of C programs.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ *   * Redistributions of source code must retain the above copyright notice,
+ *     this list of conditions and the following disclaimer.
+ *
+ *   * Redistributions in binary form must reproduce the above copyright
+ *     notice, this list of conditions and the following disclaimer in the
+ *     documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef RANDOM_RUNTIME_H
+#define RANDOM_RUNTIME_H
+
+#ifdef CSMITH_MINIMAL
+#include "csmith_minimal.h"
+#else
+
+/*****************************************************************************/
+
+#include <string.h>
+
+#define __STDC_LIMIT_MACROS
+#include "random_inc.h"
+
+static uint32_t crc32_tab[256];
+static uint32_t crc32_context = 0xFFFFFFFFUL;
+
+static void 
+crc32_gentab (void)
+{
+	uint32_t crc;
+	const uint32_t poly = 0xEDB88320UL;
+	int i, j;
+	
+	for (i = 0; i < 256; i++) {
+		crc = i;
+		for (j = 8; j > 0; j--) {
+			if (crc & 1) {
+				crc = (crc >> 1) ^ poly;
+			} else {
+				crc >>= 1;
+			}
+		}
+		crc32_tab[i] = crc;
+	}
+}
+
+static void 
+crc32_byte (uint8_t b) {
+	crc32_context = 
+		((crc32_context >> 8) & 0x00FFFFFF) ^ 
+		crc32_tab[(crc32_context ^ b) & 0xFF];
+}
+
+#if defined(__SPLAT__) || defined (__COMPCERT__) || defined(NO_LONGLONG)
+static void 
+crc32_8bytes (uint32_t val)
+{
+	crc32_byte ((val>>0) & 0xff);
+	crc32_byte ((val>>8) & 0xff);
+	crc32_byte ((val>>16) & 0xff);
+	crc32_byte ((val>>24) & 0xff);
+}
+
+static void 
+transparent_crc (uint32_t val, char* vname, int flag)
+{
+	crc32_8bytes(val);
+	if (flag) {
+  		printf("...checksum after hashing %s : %X\n", vname, crc32_context ^ 0xFFFFFFFFU);
+	}
+}
+#else
+static void 
+crc32_8bytes (uint64_t val)
+{
+	crc32_byte ((val>>0) & 0xff);
+	crc32_byte ((val>>8) & 0xff);
+	crc32_byte ((val>>16) & 0xff);
+	crc32_byte ((val>>24) & 0xff);
+	crc32_byte ((val>>32) & 0xff);
+	crc32_byte ((val>>40) & 0xff);
+	crc32_byte ((val>>48) & 0xff);
+	crc32_byte ((val>>56) & 0xff);
+}
+
+static void 
+transparent_crc (uint64_t val, char* vname, int flag)
+{
+	crc32_8bytes(val);
+	if (flag) {
+  		printf("...checksum after hashing %s : %lX\n", vname, crc32_context ^ 0xFFFFFFFFUL);
+	}
+}
+#endif
+
+/*****************************************************************************/
+
+#endif
+
+#endif /* RANDOM_RUNTIME_H */
+
+/*
+ * Local Variables:
+ * c-basic-offset: 4
+ * tab-width: 4
+ * End:
+ */
+
+/* End of file. */
diff --git a/tests/fuzz/csmith_driver.py b/tests/fuzz/csmith_driver.py
new file mode 100755
index 00000000..1cb85451
--- /dev/null
+++ b/tests/fuzz/csmith_driver.py
@@ -0,0 +1,100 @@
+#!/usr/bin/python
+
+'''
+Runs csmith, a C fuzzer, and looks for bugs
+'''
+
+import os, sys, difflib
+from subprocess import Popen, PIPE, STDOUT
+
+sys.path += [os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(__file__))), 'tools')]
+import shared
+
+engine1 = eval('shared.' + sys.argv[1]) if len(sys.argv) > 1 else shared.JS_ENGINES[0]
+engine2 = eval('shared.' + sys.argv[2]) if len(sys.argv) > 2 else None
+
+print 'testing js engines', engine1, engine2
+
+CSMITH = os.path.expanduser('~/Dev/csmith/src/csmith')
+CSMITH_CFLAGS = ['-I' + os.path.expanduser('~/Dev/csmith/runtime/')]
+
+filename = os.path.join(shared.CANONICAL_TEMP_DIR, 'fuzzcode')
+
+shared.DEFAULT_TIMEOUT = 1
+
+tried = 0
+
+notes = { 'invalid': 0, 'unaligned': 0, 'embug': 0 }
+
+while 1:
+  print 'Tried %d, notes: %s' % (tried, notes)
+  tried += 1
+  print '1) Generate C'
+  shared.execute([CSMITH, '--no-volatiles', '--no-math64'], stdout=open(filename + '.c', 'w'))
+
+  print '2) Compile natively'
+  shared.try_delete(filename)
+  shared.execute([shared.CLANG_CC, '-O2', filename + '.c', '-o', filename] + CSMITH_CFLAGS, stderr=PIPE)
+  assert os.path.exists(filename)
+  print '3) Run natively'
+  try:
+    correct = shared.timeout_run(Popen([filename], stdout=PIPE, stderr=PIPE), 3)
+  except Exception, e:
+    print 'Failed or infinite looping in native, skipping', e
+    notes['invalid'] += 1
+    continue
+
+  print '4) Compile JS-ly and compare'
+
+  def try_js(args):
+    shared.try_delete(filename + '.js')
+    shared.execute([shared.EMCC, '-O2', '-s', 'ASM_JS=1', '-s', 'PRECISE_I64_MATH=1', '-s', 'PRECISE_I32_MUL=1', filename + '.c', '-o', filename + '.js'] + CSMITH_CFLAGS + args, stderr=PIPE)
+    assert os.path.exists(filename + '.js')
+    js = shared.run_js(filename + '.js', stderr=PIPE, engine=engine1, check_timeout=True)
+    assert correct == js, ''.join([a.rstrip()+'\n' for a in difflib.unified_diff(correct.split('\n'), js.split('\n'), fromfile='expected', tofile='actual')])
+
+  # Try normally, then try unaligned because csmith does generate nonportable code that requires x86 alignment
+  ok = False
+  normal = True
+  for args, note in [([], None), (['-s', 'UNALIGNED_MEMORY=1'], 'unaligned')]:
+    try:
+      try_js(args)
+      ok = True
+      if note:
+        notes[note] += 1
+      break
+    except Exception, e:
+      print e
+      normal = False
+  if not ok:
+    print "EMSCRIPTEN BUG"
+    notes['embug'] += 1
+    continue #break
+  #if not ok:
+  #  try: # finally, try with safe heap. if that is triggered, this is nonportable code almost certainly
+  #    try_js(['-s', 'SAFE_HEAP=1'])
+  #  except Exception, e:
+  #    print e
+  #    js = shared.run_js(filename + '.js', stderr=PIPE, full_output=True)
+  #  print js
+  #  if 'SAFE_HEAP' in js:
+  #    notes['safeheap'] += 1
+  #  else:
+  #    break
+
+  # This is ok. Try in secondary JS engine too
+  if engine2 and normal:
+    try:
+      js2 = shared.run_js(filename + '.js', stderr=PIPE, engine=engine2, full_output=True, check_timeout=True)
+    except:
+      print 'failed to run in secondary', js2
+      break
+
+    # asm.js testing
+    assert 'warning: Successfully compiled asm.js code' in js2, 'must validate'
+    js2 = js2.replace('\nwarning: Successfully compiled asm.js code\n', '')
+
+    assert js2 == correct, ''.join([a.rstrip()+'\n' for a in difflib.unified_diff(correct.split('\n'), js2.split('\n'), fromfile='expected', tofile='actual')]) + 'ODIN FAIL'
+    print 'odin ok'
+
+
diff --git a/tests/fuzz/platform_generic.h b/tests/fuzz/platform_generic.h
new file mode 100644
index 00000000..b2ef33a3
--- /dev/null
+++ b/tests/fuzz/platform_generic.h
@@ -0,0 +1,132 @@
+/* -*- mode: C -*-
+ *
+ *
+ * Copyright (c) 2007, 2008 The University of Utah
+ * All rights reserved.
+ *
+ * This file is part of `csmith', a random generator of C programs.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ *   * Redistributions of source code must retain the above copyright notice,
+ *     this list of conditions and the following disclaimer.
+ *
+ *   * Redistributions in binary form must reproduce the above copyright
+ *     notice, this list of conditions and the following disclaimer in the
+ *     documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef PLATFORM_GENERIC_H
+#define PLATFORM_GENERIC_H
+
+/*****************************************************************************/
+
+#ifdef STANDALONE
+extern int printf (const char *, ...);
+#else
+#include <stdio.h>
+#endif
+
+static void
+platform_main_begin(void)
+{
+	/* Nothing to do. */
+}
+
+static void
+platform_main_end(uint32_t crc, int flag)
+{
+#if defined (__FRAMAC)
+    Frama_C_dump_assert_each();
+#endif
+	printf ("checksum = %X\n", crc);
+#if defined (LOG_WRAPPERS)
+	{
+		int i, first;
+
+		printf ("executed wrappers: ");
+		first = 1;
+		for (i=1; i<N_WRAP+1; i++) {
+			if (__executed_checks[i]) {
+				if (!first) {
+					printf (",");
+				} else {
+					first = 0;
+				}
+				printf ("%d", i);
+			}
+		}
+		printf ("\n");
+
+		printf ("dead wrappers: ");
+		first = 1;
+		for (i=1; i<N_WRAP+1; i++) {
+			if (!__executed_checks[i]) {
+				if (!first) {
+					printf (",");
+				} else {
+					first = 0;
+				}
+				printf ("%d", i);
+			}
+		}
+		printf ("\n");
+
+		printf ("wrappers that failed at least once: ");
+		first = 1;
+		for (i=1; i<N_WRAP+1; i++) {
+			if (__failed_checks[i]) {
+				if (!first) {
+					printf (",");
+				} else {
+					first = 0;
+				}
+				printf ("%d", i);
+			}
+		}
+		printf ("\n");
+
+		printf ("wrappers that never failed (or never executed): ");
+		first = 1;
+		for (i=1; i<N_WRAP+1; i++) {
+			if (!__failed_checks[i]) {
+				if (!first) {
+					printf (",");
+				} else {
+					first = 0;
+				}
+				printf ("%d", i);
+			}
+		}
+		printf ("\n");
+	}
+#endif
+}
+
+#define MB (1<<20)
+
+/*****************************************************************************/
+
+#endif /* PLATFORM_GENERIC_H */
+
+/*
+ * Local Variables:
+ * c-basic-offset: 4
+ * tab-width: 4
+ * End:
+ */
+
+/* End of file. */
diff --git a/tests/fuzz/random_inc.h b/tests/fuzz/random_inc.h
new file mode 100644
index 00000000..7559cbd7
--- /dev/null
+++ b/tests/fuzz/random_inc.h
@@ -0,0 +1,129 @@
+/* -*- mode: C -*-
+ *
+ * Copyright (c) 2007-2010 The University of Utah
+ * All rights reserved.
+ *
+ * This file is part of `csmith', a random generator of C programs.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ *   * Redistributions of source code must retain the above copyright notice,
+ *     this list of conditions and the following disclaimer.
+ *
+ *   * Redistributions in binary form must reproduce the above copyright
+ *     notice, this list of conditions and the following disclaimer in the
+ *     documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef RANDOM_INC_H
+#define RANDOM_INC_H
+
+
+#if defined(STANDALONE)
+  #if defined(_MSC_VER)
+    #include <limits.h>
+    #include "windows/stdint.h"
+  #elif defined (IA32_ARCH)
+    #include "stdint_ia32.h"
+  #elif defined (IA64_ARCH)
+    #include "stdint_ia64.h"
+  #elif defined (MSP430)
+    #include "stdint_msp430.h"
+  #elif defined (AVR_ARCH)
+    #include "stdint_avr.h"
+  #else
+    #include "stdint_ia32.h"
+  #endif
+#else
+  #include <limits.h>
+  #if defined(_MSC_VER)
+    #include "windows/stdint.h"
+  #else
+    #include <stdint.h>
+  #endif
+#endif
+
+#include <assert.h>
+
+/*****************************************************************************/
+
+#ifndef DEPUTY
+#define COUNT(n)
+#define TC
+#define SAFE
+#endif
+
+/*****************************************************************************/
+
+#ifdef LOG_WRAPPERS
+#include "wrapper.h"
+char __failed_checks[N_WRAP+1];
+char __executed_checks[N_WRAP+1];
+#define UNDEFINED(__val) (__failed_checks[index]=1,(__val))
+#define LOG_INDEX , int index
+#define LOG_EXEC __executed_checks[index]=1;
+#else
+#define UNDEFINED(__val) (__val)
+#define LOG_INDEX
+#define LOG_EXEC
+#endif
+
+#if defined(AVR_ARCH)
+#include "platform_avr.h"
+#elif defined (MSP430)
+#include "platform_msp430.h"
+#else
+#include "platform_generic.h"
+#endif
+
+#define STATIC static
+
+#if defined (USE_MATH_MACROS_NOTMP)
+#include "safe_math_macros_notmp.h"
+#elif defined (USE_MATH_MACROS)
+#include "safe_math_macros.h"
+#else
+#define FUNC_NAME(x) (safe_##x)
+#include "safe_math.h"
+#undef FUNC_NAME
+#endif
+
+#define INT_BIT (sizeof(int)*CHAR_BIT)
+#define _CSMITH_BITFIELD(x) (((x)>INT_BIT)?((x)%INT_BIT):(x))
+
+#ifdef TCC
+
+void* memcpy(void* dest, const void* src, size_t count) {
+  char* dst8 = (char*)dest;
+  char* src8 = (char*)src;
+  
+  while (count--) {
+    *dst8++ = *src8++;
+  }
+  return dest;
+}
+
+void *memset(void *s, int c, size_t n)
+{
+    unsigned char* p=s;
+    while(n--)
+        *p++ = (unsigned char)c;
+    return s;
+}
+
+#endif
+
+#endif // RANDOM_INC_H
diff --git a/tests/fuzz/safe_math.h b/tests/fuzz/safe_math.h
new file mode 100644
index 00000000..393ebba1
--- /dev/null
+++ b/tests/fuzz/safe_math.h
@@ -0,0 +1,947 @@
+
+#ifndef SAFE_MATH_H
+#define SAFE_MATH_H
+
+
+
+
+
+
+
+
+
+STATIC int8_t
+FUNC_NAME(unary_minus_func_int8_t_s)(int8_t si LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT8_MAX>=INT_MAX)
+    (si==INT8_MIN) ?
+    (UNDEFINED(si)) :
+#endif
+#endif
+    -si;
+}
+
+STATIC int8_t
+FUNC_NAME(add_func_int8_t_s_s)(int8_t si1, int8_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT8_MAX>=INT_MAX)
+    (((si1>0) && (si2>0) && (si1 > (INT8_MAX-si2))) || ((si1<0) && (si2<0) && (si1 < (INT8_MIN-si2)))) ?
+    (UNDEFINED(si1)) :
+#endif
+#endif
+    (si1 + si2);
+}
+
+STATIC int8_t
+FUNC_NAME(sub_func_int8_t_s_s)(int8_t si1, int8_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT8_MAX>=INT_MAX)
+    (((si1^si2) & (((si1 ^ ((si1^si2) & (~INT8_MAX)))-si2)^si2)) < 0) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    (si1 - si2);
+}
+
+STATIC int8_t
+FUNC_NAME(mul_func_int8_t_s_s)(int8_t si1, int8_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT8_MAX>=INT_MAX)
+    (((si1 > 0) && (si2 > 0) && (si1 > (INT8_MAX / si2))) || ((si1 > 0) && (si2 <= 0) && (si2 < (INT8_MIN / si1))) || ((si1 <= 0) && (si2 > 0) && (si1 < (INT8_MIN / si2))) || ((si1 <= 0) && (si2 <= 0) && (si1 != 0) && (si2 < (INT8_MAX / si1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    si1 * si2;
+}
+
+STATIC int8_t
+FUNC_NAME(mod_func_int8_t_s_s)(int8_t si1, int8_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT8_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 % si2);
+}
+
+STATIC int8_t
+FUNC_NAME(div_func_int8_t_s_s)(int8_t si1, int8_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT8_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 / si2);
+}
+
+STATIC int8_t
+FUNC_NAME(lshift_func_int8_t_s_s)(int8_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32) || (left > (INT8_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC int8_t
+FUNC_NAME(lshift_func_int8_t_s_u)(int8_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32) || (left > (INT8_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC int8_t
+FUNC_NAME(rshift_func_int8_t_s_s)(int8_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32))? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC int8_t
+FUNC_NAME(rshift_func_int8_t_s_u)(int8_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+
+
+STATIC int16_t
+FUNC_NAME(unary_minus_func_int16_t_s)(int16_t si LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT16_MAX>=INT_MAX)
+    (si==INT16_MIN) ?
+    (UNDEFINED(si)) :
+#endif
+#endif
+    -si;
+}
+
+STATIC int16_t
+FUNC_NAME(add_func_int16_t_s_s)(int16_t si1, int16_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT16_MAX>=INT_MAX)
+    (((si1>0) && (si2>0) && (si1 > (INT16_MAX-si2))) || ((si1<0) && (si2<0) && (si1 < (INT16_MIN-si2)))) ?
+    (UNDEFINED(si1)) :
+#endif
+#endif
+    (si1 + si2);
+}
+
+STATIC int16_t
+FUNC_NAME(sub_func_int16_t_s_s)(int16_t si1, int16_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT16_MAX>=INT_MAX)
+    (((si1^si2) & (((si1 ^ ((si1^si2) & (~INT16_MAX)))-si2)^si2)) < 0) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    (si1 - si2);
+}
+
+STATIC int16_t
+FUNC_NAME(mul_func_int16_t_s_s)(int16_t si1, int16_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT16_MAX>=INT_MAX)
+    (((si1 > 0) && (si2 > 0) && (si1 > (INT16_MAX / si2))) || ((si1 > 0) && (si2 <= 0) && (si2 < (INT16_MIN / si1))) || ((si1 <= 0) && (si2 > 0) && (si1 < (INT16_MIN / si2))) || ((si1 <= 0) && (si2 <= 0) && (si1 != 0) && (si2 < (INT16_MAX / si1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    si1 * si2;
+}
+
+STATIC int16_t
+FUNC_NAME(mod_func_int16_t_s_s)(int16_t si1, int16_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT16_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 % si2);
+}
+
+STATIC int16_t
+FUNC_NAME(div_func_int16_t_s_s)(int16_t si1, int16_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT16_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 / si2);
+}
+
+STATIC int16_t
+FUNC_NAME(lshift_func_int16_t_s_s)(int16_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32) || (left > (INT16_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC int16_t
+FUNC_NAME(lshift_func_int16_t_s_u)(int16_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32) || (left > (INT16_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC int16_t
+FUNC_NAME(rshift_func_int16_t_s_s)(int16_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32))? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC int16_t
+FUNC_NAME(rshift_func_int16_t_s_u)(int16_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+
+
+STATIC int32_t
+FUNC_NAME(unary_minus_func_int32_t_s)(int32_t si LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT32_MAX>=INT_MAX)
+    (si==INT32_MIN) ?
+    (UNDEFINED(si)) :
+#endif
+#endif
+    -si;
+}
+
+STATIC int32_t
+FUNC_NAME(add_func_int32_t_s_s)(int32_t si1, int32_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT32_MAX>=INT_MAX)
+    (((si1>0) && (si2>0) && (si1 > (INT32_MAX-si2))) || ((si1<0) && (si2<0) && (si1 < (INT32_MIN-si2)))) ?
+    (UNDEFINED(si1)) :
+#endif
+#endif
+    (si1 + si2);
+}
+
+STATIC int32_t
+FUNC_NAME(sub_func_int32_t_s_s)(int32_t si1, int32_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT32_MAX>=INT_MAX)
+    (((si1^si2) & (((si1 ^ ((si1^si2) & (~INT32_MAX)))-si2)^si2)) < 0) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    (si1 - si2);
+}
+
+STATIC int32_t
+FUNC_NAME(mul_func_int32_t_s_s)(int32_t si1, int32_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT32_MAX>=INT_MAX)
+    (((si1 > 0) && (si2 > 0) && (si1 > (INT32_MAX / si2))) || ((si1 > 0) && (si2 <= 0) && (si2 < (INT32_MIN / si1))) || ((si1 <= 0) && (si2 > 0) && (si1 < (INT32_MIN / si2))) || ((si1 <= 0) && (si2 <= 0) && (si1 != 0) && (si2 < (INT32_MAX / si1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    si1 * si2;
+}
+
+STATIC int32_t
+FUNC_NAME(mod_func_int32_t_s_s)(int32_t si1, int32_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT32_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 % si2);
+}
+
+STATIC int32_t
+FUNC_NAME(div_func_int32_t_s_s)(int32_t si1, int32_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT32_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 / si2);
+}
+
+STATIC int32_t
+FUNC_NAME(lshift_func_int32_t_s_s)(int32_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32) || (left > (INT32_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC int32_t
+FUNC_NAME(lshift_func_int32_t_s_u)(int32_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32) || (left > (INT32_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC int32_t
+FUNC_NAME(rshift_func_int32_t_s_s)(int32_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32))? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC int32_t
+FUNC_NAME(rshift_func_int32_t_s_u)(int32_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+#ifndef NO_LONGLONG
+
+
+STATIC int64_t
+FUNC_NAME(unary_minus_func_int64_t_s)(int64_t si LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT64_MAX>=INT_MAX)
+    (si==INT64_MIN) ?
+    (UNDEFINED(si)) :
+#endif
+#endif
+    -si;
+}
+
+STATIC int64_t
+FUNC_NAME(add_func_int64_t_s_s)(int64_t si1, int64_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT64_MAX>=INT_MAX)
+    (((si1>0) && (si2>0) && (si1 > (INT64_MAX-si2))) || ((si1<0) && (si2<0) && (si1 < (INT64_MIN-si2)))) ?
+    (UNDEFINED(si1)) :
+#endif
+#endif
+    (si1 + si2);
+}
+
+STATIC int64_t
+FUNC_NAME(sub_func_int64_t_s_s)(int64_t si1, int64_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT64_MAX>=INT_MAX)
+    (((si1^si2) & (((si1 ^ ((si1^si2) & (~INT64_MAX)))-si2)^si2)) < 0) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    (si1 - si2);
+}
+
+STATIC int64_t
+FUNC_NAME(mul_func_int64_t_s_s)(int64_t si1, int64_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+#if (INT64_MAX>=INT_MAX)
+    (((si1 > 0) && (si2 > 0) && (si1 > (INT64_MAX / si2))) || ((si1 > 0) && (si2 <= 0) && (si2 < (INT64_MIN / si1))) || ((si1 <= 0) && (si2 > 0) && (si1 < (INT64_MIN / si2))) || ((si1 <= 0) && (si2 <= 0) && (si1 != 0) && (si2 < (INT64_MAX / si1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+#endif
+    si1 * si2;
+}
+
+STATIC int64_t
+FUNC_NAME(mod_func_int64_t_s_s)(int64_t si1, int64_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT64_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 % si2);
+}
+
+STATIC int64_t
+FUNC_NAME(div_func_int64_t_s_s)(int64_t si1, int64_t si2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((si2 == 0) || ((si1 == INT64_MIN) && (si2 == (-1)))) ? 
+    (UNDEFINED(si1)) : 
+#endif
+    (si1 / si2);
+}
+
+STATIC int64_t
+FUNC_NAME(lshift_func_int64_t_s_s)(int64_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32) || (left > (INT64_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC int64_t
+FUNC_NAME(lshift_func_int64_t_s_u)(int64_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32) || (left > (INT64_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC int64_t
+FUNC_NAME(rshift_func_int64_t_s_s)(int64_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((int)right) < 0) || (((int)right) >= 32))? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC int64_t
+FUNC_NAME(rshift_func_int64_t_s_u)(int64_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((left < 0) || (((unsigned int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+#endif
+
+
+
+
+
+STATIC uint8_t
+FUNC_NAME(unary_minus_func_uint8_t_u)(uint8_t ui LOG_INDEX)
+{
+  LOG_EXEC
+  return -ui;
+}
+
+STATIC uint8_t
+FUNC_NAME(add_func_uint8_t_u_u)(uint8_t ui1, uint8_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 + ui2;
+}
+
+STATIC uint8_t
+FUNC_NAME(sub_func_uint8_t_u_u)(uint8_t ui1, uint8_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 - ui2;
+}
+
+STATIC uint8_t
+FUNC_NAME(mul_func_uint8_t_u_u)(uint8_t ui1, uint8_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ((unsigned int)ui1) * ((unsigned int)ui2);
+}
+
+STATIC uint8_t
+FUNC_NAME(mod_func_uint8_t_u_u)(uint8_t ui1, uint8_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 % ui2);
+}
+
+STATIC uint8_t
+FUNC_NAME(div_func_uint8_t_u_u)(uint8_t ui1, uint8_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 / ui2);
+}
+
+STATIC uint8_t
+FUNC_NAME(lshift_func_uint8_t_u_s)(uint8_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32) || (left > (UINT8_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC uint8_t
+FUNC_NAME(lshift_func_uint8_t_u_u)(uint8_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((unsigned int)right) >= 32) || (left > (UINT8_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC uint8_t
+FUNC_NAME(rshift_func_uint8_t_u_s)(uint8_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC uint8_t
+FUNC_NAME(rshift_func_uint8_t_u_u)(uint8_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (((unsigned int)right) >= 32) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+
+
+STATIC uint16_t
+FUNC_NAME(unary_minus_func_uint16_t_u)(uint16_t ui LOG_INDEX)
+{
+  LOG_EXEC
+  return -ui;
+}
+
+STATIC uint16_t
+FUNC_NAME(add_func_uint16_t_u_u)(uint16_t ui1, uint16_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 + ui2;
+}
+
+STATIC uint16_t
+FUNC_NAME(sub_func_uint16_t_u_u)(uint16_t ui1, uint16_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 - ui2;
+}
+
+STATIC uint16_t
+FUNC_NAME(mul_func_uint16_t_u_u)(uint16_t ui1, uint16_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ((unsigned int)ui1) * ((unsigned int)ui2);
+}
+
+STATIC uint16_t
+FUNC_NAME(mod_func_uint16_t_u_u)(uint16_t ui1, uint16_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 % ui2);
+}
+
+STATIC uint16_t
+FUNC_NAME(div_func_uint16_t_u_u)(uint16_t ui1, uint16_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 / ui2);
+}
+
+STATIC uint16_t
+FUNC_NAME(lshift_func_uint16_t_u_s)(uint16_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32) || (left > (UINT16_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC uint16_t
+FUNC_NAME(lshift_func_uint16_t_u_u)(uint16_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((unsigned int)right) >= 32) || (left > (UINT16_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC uint16_t
+FUNC_NAME(rshift_func_uint16_t_u_s)(uint16_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC uint16_t
+FUNC_NAME(rshift_func_uint16_t_u_u)(uint16_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (((unsigned int)right) >= 32) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+
+
+STATIC uint32_t
+FUNC_NAME(unary_minus_func_uint32_t_u)(uint32_t ui LOG_INDEX)
+{
+  LOG_EXEC
+  return -ui;
+}
+
+STATIC uint32_t
+FUNC_NAME(add_func_uint32_t_u_u)(uint32_t ui1, uint32_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 + ui2;
+}
+
+STATIC uint32_t
+FUNC_NAME(sub_func_uint32_t_u_u)(uint32_t ui1, uint32_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 - ui2;
+}
+
+STATIC uint32_t
+FUNC_NAME(mul_func_uint32_t_u_u)(uint32_t ui1, uint32_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ((unsigned int)ui1) * ((unsigned int)ui2);
+}
+
+STATIC uint32_t
+FUNC_NAME(mod_func_uint32_t_u_u)(uint32_t ui1, uint32_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 % ui2);
+}
+
+STATIC uint32_t
+FUNC_NAME(div_func_uint32_t_u_u)(uint32_t ui1, uint32_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 / ui2);
+}
+
+STATIC uint32_t
+FUNC_NAME(lshift_func_uint32_t_u_s)(uint32_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32) || (left > (UINT32_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC uint32_t
+FUNC_NAME(lshift_func_uint32_t_u_u)(uint32_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((unsigned int)right) >= 32) || (left > (UINT32_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC uint32_t
+FUNC_NAME(rshift_func_uint32_t_u_s)(uint32_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC uint32_t
+FUNC_NAME(rshift_func_uint32_t_u_u)(uint32_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (((unsigned int)right) >= 32) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+#ifndef NO_LONGLONG
+
+
+STATIC uint64_t
+FUNC_NAME(unary_minus_func_uint64_t_u)(uint64_t ui LOG_INDEX)
+{
+  LOG_EXEC
+  return -ui;
+}
+
+STATIC uint64_t
+FUNC_NAME(add_func_uint64_t_u_u)(uint64_t ui1, uint64_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 + ui2;
+}
+
+STATIC uint64_t
+FUNC_NAME(sub_func_uint64_t_u_u)(uint64_t ui1, uint64_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ui1 - ui2;
+}
+
+STATIC uint64_t
+FUNC_NAME(mul_func_uint64_t_u_u)(uint64_t ui1, uint64_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return ((unsigned long long int)ui1) * ((unsigned long long int)ui2);
+}
+
+STATIC uint64_t
+FUNC_NAME(mod_func_uint64_t_u_u)(uint64_t ui1, uint64_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 % ui2);
+}
+
+STATIC uint64_t
+FUNC_NAME(div_func_uint64_t_u_u)(uint64_t ui1, uint64_t ui2 LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (ui2 == 0) ? 
+    (UNDEFINED(ui1)) : 
+#endif
+    (ui1 / ui2);
+}
+
+STATIC uint64_t
+FUNC_NAME(lshift_func_uint64_t_u_s)(uint64_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32) || (left > (UINT64_MAX >> ((int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((int)right));
+}
+
+STATIC uint64_t
+FUNC_NAME(lshift_func_uint64_t_u_u)(uint64_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((unsigned int)right) >= 32) || (left > (UINT64_MAX >> ((unsigned int)right)))) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left << ((unsigned int)right));
+}
+
+STATIC uint64_t
+FUNC_NAME(rshift_func_uint64_t_u_s)(uint64_t left, int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    ((((int)right) < 0) || (((int)right) >= 32)) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((int)right));
+}
+
+STATIC uint64_t
+FUNC_NAME(rshift_func_uint64_t_u_u)(uint64_t left, unsigned int right LOG_INDEX)
+{
+  LOG_EXEC
+  return 
+#ifndef UNSAFE
+    (((unsigned int)right) >= 32) ? 
+    (UNDEFINED(left)) : 
+#endif
+    (left >> ((unsigned int)right));
+}
+
+#endif
+
+#endif
diff --git a/tests/gears.png b/tests/gears.png
index ffb855c4..a35db625 100644
--- a/tests/gears.png
+++ b/tests/gears.png
diff --git a/tests/gl_ps.c b/tests/gl_ps.c
index 81579c1d..6ea0e3db 100644
--- a/tests/gl_ps.c
+++ b/tests/gl_ps.c
@@ -78,6 +78,9 @@ void shaders() {
   glLinkProgram(program);
   glGetProgramiv(program, GL_LINK_STATUS, &ok);
   assert(ok);
+  assert(glIsProgram(program));
+  assert(!glIsProgram(0));
+  assert(!glIsProgram(program+1)); // a number that can't be a real shader
 
   glUseProgram(program);
 
diff --git a/tests/gl_ps.png b/tests/gl_ps.png
index 185f7166..ac13ec0b 100644
--- a/tests/gl_ps.png
+++ b/tests/gl_ps.png
diff --git a/tests/gl_ps_packed.c b/tests/gl_ps_packed.c
new file mode 100644
index 00000000..9ab99cb8
--- /dev/null
+++ b/tests/gl_ps_packed.c
@@ -0,0 +1,230 @@
+/*******************************************************************
+ *                                                                 *
+ *                        Using SDL With OpenGL                    *
+ *                                                                 *
+ *                    Tutorial by Kyle Foley (sdw)                 *
+ *                                                                 *
+ * http://gpwiki.org/index.php/SDL:Tutorials:Using_SDL_with_OpenGL *
+ *                                                                 *
+ *******************************************************************/
+
+/*
+THIS WORK, INCLUDING THE SOURCE CODE, DOCUMENTATION
+AND RELATED MEDIA AND DATA, IS PLACED INTO THE PUBLIC DOMAIN.
+
+THE ORIGINAL AUTHOR IS KYLE FOLEY.
+
+THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
+OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
+MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
+ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
+RESULTING FROM THE USE, MODIFICATION, OR
+REDISTRIBUTION OF THIS SOFTWARE.
+*/
+
+#if !EMSCRIPTEN
+#define USE_GLEW 1
+#endif
+
+#if USE_GLEW
+#include "GL/glew.h"
+#endif
+
+#include "SDL/SDL.h"
+#include "SDL/SDL_image.h"
+#if !USE_GLEW
+#include "SDL/SDL_opengl.h"
+#endif
+
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+void shaders() {
+#if USE_GLEW
+  glewInit();
+#endif
+
+  GLint ok;
+
+  const char *vertexShader = "void main(void) \n"
+                             "{ \n"
+                             "    gl_Position = ftransform(); \n"
+                             "    gl_TexCoord[0] = gl_MultiTexCoord0; \n"
+                             "    gl_FrontColor = gl_Color; \n"
+                             "} \n";
+  const char *fragmentShader = "uniform sampler2D tex0; \n"
+                               "void main(void) \n"
+                               "{ \n"
+                               "    gl_FragColor = gl_Color * texture2D(tex0, gl_TexCoord[0].xy); \n"
+                               "} \n";
+
+  GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+  glShaderSource(vs, 1, &vertexShader, NULL);
+  glCompileShader(vs);
+  glGetShaderiv(vs, GL_COMPILE_STATUS, &ok);
+  assert(ok);
+
+  GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+  glShaderSource(fs, 1, &fragmentShader, NULL);
+  glCompileShader(fs);
+  glGetShaderiv(fs, GL_COMPILE_STATUS, &ok);
+  assert(ok);
+
+  GLuint program = glCreateProgram();
+
+  glAttachShader(program, vs);
+  glAttachShader(program, fs);
+  glLinkProgram(program);
+  glGetProgramiv(program, GL_LINK_STATUS, &ok);
+  assert(ok);
+
+  glUseProgram(program);
+}
+
+int main(int argc, char *argv[])
+{
+    SDL_Surface *screen;
+
+    // Slightly different SDL initialization
+    if ( SDL_Init(SDL_INIT_VIDEO) != 0 ) {
+        printf("Unable to initialize SDL: %s\n", SDL_GetError());
+        return 1;
+    }
+
+    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 ); // *new*
+
+    screen = SDL_SetVideoMode( 640, 480, 16, SDL_OPENGL ); // *changed*
+    if ( !screen ) {
+        printf("Unable to set video mode: %s\n", SDL_GetError());
+        return 1;
+    }
+    
+    // Set the OpenGL state after creating the context with SDL_SetVideoMode
+
+    glClearColor( 0, 0, 0, 0 );
+    
+#if !EMSCRIPTEN
+    glEnable( GL_TEXTURE_2D ); // Need this to display a texture XXX unnecessary in OpenGL ES 2.0/WebGL
+#endif
+
+    glViewport( 0, 0, 640, 480 );
+
+    glMatrixMode( GL_PROJECTION );
+    GLfloat matrixData[] = { 2.0/640,        0,  0,  0,
+                                   0, -2.0/480,  0,  0,
+                                   0,        0, -1,  0,
+                                  -1,        1,  0,  1 };
+    glLoadMatrixf(matrixData); // test loadmatrix
+
+    glMatrixMode( GL_MODELVIEW );
+    glLoadIdentity();
+    
+    // Load the OpenGL texture
+
+    GLuint texture; // Texture object handle
+    SDL_Surface *surface; // Gives us the information to make the texture
+    
+    if ( (surface = IMG_Load("screenshot.png")) ) { 
+    
+        // Check that the image's width is a power of 2
+        if ( (surface->w & (surface->w - 1)) != 0 ) {
+            printf("warning: image.bmp's width is not a power of 2\n");
+        }
+    
+        // Also check if the height is a power of 2
+        if ( (surface->h & (surface->h - 1)) != 0 ) {
+            printf("warning: image.bmp's height is not a power of 2\n");
+        }
+    
+        // Have OpenGL generate a texture object handle for us
+        glGenTextures( 1, &texture );
+    
+        // Bind the texture object
+        glBindTexture( GL_TEXTURE_2D, texture );
+        
+        // Set the texture's stretching properties
+        glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+        glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+
+        //SDL_LockSurface(surface);
+
+        // Add some greyness
+        memset(surface->pixels, 0x66, surface->w*surface->h);
+
+        // Edit the texture object's image data using the information SDL_Surface gives us
+        glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, surface->w, surface->h, 0, 
+                      GL_RGBA, GL_UNSIGNED_BYTE, surface->pixels );
+
+        //SDL_UnlockSurface(surface);
+    } 
+    else {
+        printf("SDL could not load image.bmp: %s\n", SDL_GetError());
+        SDL_Quit();
+        return 1;
+    }    
+    
+    // Free the SDL_Surface only if it was successfully created
+    if ( surface ) { 
+        SDL_FreeSurface( surface );
+    }
+    
+    // Clear the screen before drawing
+    glClear( GL_COLOR_BUFFER_BIT );
+    
+    shaders();
+
+    // Bind the texture to which subsequent calls refer to
+    glBindTexture( GL_TEXTURE_2D, texture );
+
+    // Use clientside vertex pointers to render two items. In this test we have each
+    // attribute in a separate buffer, packed (i.e. stride == 0)
+    GLfloat vertexData[] = {  10,  10,
+                             300,  10,
+                             300, 128,
+                              10, 128,
+                             410,  10,
+                             600,  10,
+                             630, 200,
+                             310, 250,
+                             100, 300,
+                             300, 300,
+                             300, 400,
+                             100, 400 };
+    GLfloat textureData[] = { 0, 0,
+                              1, 0,
+                              1, 1,
+                              0, 1,
+                              0, 0.5,
+                              1, 0.5,
+                              1,   1,
+                              0.5, 1,
+                              0,   0,
+                              1,   0,
+                              1,   1,
+                              0,   1, };
+
+    glEnableClientState(GL_TEXTURE_COORD_ARRAY);
+    glTexCoordPointer(2, GL_FLOAT, 0, textureData);
+    glEnableClientState(GL_VERTEX_ARRAY);
+    glVertexPointer(2, GL_FLOAT, 0, vertexData);
+
+    glDrawArrays(GL_QUADS, 0, 12);
+
+    glDisableClientState(GL_TEXTURE_COORD_ARRAY);
+    glDisableClientState(GL_VERTEX_ARRAY);
+
+    SDL_GL_SwapBuffers();
+    
+#if !EMSCRIPTEN
+    // Wait for 3 seconds to give us a chance to see the image
+    SDL_Delay(3000);
+#endif
+
+    // Now we can delete the OpenGL texture and close down SDL
+    glDeleteTextures( 1, &texture );
+    
+    SDL_Quit();
+    
+    return 0;
+}
diff --git a/tests/gl_ps_workaround.c b/tests/gl_ps_workaround.c
new file mode 100644
index 00000000..1e2a5c41
--- /dev/null
+++ b/tests/gl_ps_workaround.c
@@ -0,0 +1,230 @@
+/*******************************************************************
+ *                                                                 *
+ *                        Using SDL With OpenGL                    *
+ *                                                                 *
+ *                    Tutorial by Kyle Foley (sdw)                 *
+ *                                                                 *
+ * http://gpwiki.org/index.php/SDL:Tutorials:Using_SDL_with_OpenGL *
+ *                                                                 *
+ *******************************************************************/
+
+/*
+THIS WORK, INCLUDING THE SOURCE CODE, DOCUMENTATION
+AND RELATED MEDIA AND DATA, IS PLACED INTO THE PUBLIC DOMAIN.
+
+THE ORIGINAL AUTHOR IS KYLE FOLEY.
+
+THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
+OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
+MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
+ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
+RESULTING FROM THE USE, MODIFICATION, OR
+REDISTRIBUTION OF THIS SOFTWARE.
+*/
+
+#if !EMSCRIPTEN
+#define USE_GLEW 1
+#endif
+
+#if USE_GLEW
+#include "GL/glew.h"
+#endif
+
+#include "SDL/SDL.h"
+#include "SDL/SDL_image.h"
+#if !USE_GLEW
+#include "SDL/SDL_opengl.h"
+#endif
+
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+void shaders() {
+#if USE_GLEW
+  glewInit();
+#endif
+
+  GLint ok;
+
+  const char *vertexShader = "void main(void) \n"
+                             "{ \n"
+                             "    gl_Position = ftransform(); \n"
+                             "    gl_TexCoord[0] = gl_MultiTexCoord0; \n"
+                             "    gl_FrontColor = gl_Color; \n"
+                             "} \n";
+  const char *fragmentShader = "uniform sampler2D tex0; \n"
+                               "void main(void) \n"
+                               "{ \n"
+                               "    gl_FragColor = gl_Color * texture2D(tex0, gl_TexCoord[0].xy); \n"
+                               "} \n";
+
+  GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+  glShaderSource(vs, 1, &vertexShader, NULL);
+  glCompileShader(vs);
+  glGetShaderiv(vs, GL_COMPILE_STATUS, &ok);
+  assert(ok);
+
+  GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+  glShaderSource(fs, 1, &fragmentShader, NULL);
+  glCompileShader(fs);
+  glGetShaderiv(fs, GL_COMPILE_STATUS, &ok);
+  assert(ok);
+
+  GLuint program = glCreateProgram();
+
+  glAttachShader(program, vs);
+  glAttachShader(program, fs);
+  glLinkProgram(program);
+  glGetProgramiv(program, GL_LINK_STATUS, &ok);
+  assert(ok);
+
+  glUseProgram(program);
+}
+
+int main(int argc, char *argv[])
+{
+    SDL_Surface *screen;
+
+    // Slightly different SDL initialization
+    if ( SDL_Init(SDL_INIT_VIDEO) != 0 ) {
+        printf("Unable to initialize SDL: %s\n", SDL_GetError());
+        return 1;
+    }
+
+    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 ); // *new*
+
+    screen = SDL_SetVideoMode( 640, 480, 16, SDL_OPENGL ); // *changed*
+    if ( !screen ) {
+        printf("Unable to set video mode: %s\n", SDL_GetError());
+        return 1;
+    }
+    
+    // Set the OpenGL state after creating the context with SDL_SetVideoMode
+
+    glClearColor( 0, 0, 0, 0 );
+    
+#if !EMSCRIPTEN
+    glEnable( GL_TEXTURE_2D ); // Need this to display a texture XXX unnecessary in OpenGL ES 2.0/WebGL
+#endif
+
+    glViewport( 0, 0, 640, 480 );
+
+    glMatrixMode( GL_PROJECTION );
+    GLfloat matrixData[] = { 2.0/640,        0,  0,  0,
+                                   0, -2.0/480,  0,  0,
+                                   0,        0, -1,  0,
+                                  -1,        1,  0,  1 };
+    glLoadMatrixf(matrixData); // test loadmatrix
+
+    glMatrixMode( GL_MODELVIEW );
+    glLoadIdentity();
+    
+    // Load the OpenGL texture
+
+    GLuint texture; // Texture object handle
+    SDL_Surface *surface; // Gives us the information to make the texture
+    
+    if ( (surface = IMG_Load("screenshot.png")) ) { 
+    
+        // Check that the image's width is a power of 2
+        if ( (surface->w & (surface->w - 1)) != 0 ) {
+            printf("warning: image.bmp's width is not a power of 2\n");
+        }
+    
+        // Also check if the height is a power of 2
+        if ( (surface->h & (surface->h - 1)) != 0 ) {
+            printf("warning: image.bmp's height is not a power of 2\n");
+        }
+    
+        // Have OpenGL generate a texture object handle for us
+        glGenTextures( 1, &texture );
+    
+        // Bind the texture object
+        glBindTexture( GL_TEXTURE_2D, texture );
+        
+        // Set the texture's stretching properties
+        glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+        glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+
+        //SDL_LockSurface(surface);
+
+        // Add some greyness
+        memset(surface->pixels, 0x66, surface->w*surface->h);
+
+        // Edit the texture object's image data using the information SDL_Surface gives us
+        glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, surface->w, surface->h, 0, 
+                      GL_RGBA, GL_UNSIGNED_BYTE, surface->pixels );
+
+        //SDL_UnlockSurface(surface);
+    } 
+    else {
+        printf("SDL could not load image.bmp: %s\n", SDL_GetError());
+        SDL_Quit();
+        return 1;
+    }    
+    
+    // Free the SDL_Surface only if it was successfully created
+    if ( surface ) { 
+        SDL_FreeSurface( surface );
+    }
+    
+    // Clear the screen before drawing
+    glClear( GL_COLOR_BUFFER_BIT );
+    
+    shaders();
+
+    // Bind the texture to which subsequent calls refer to
+    glBindTexture( GL_TEXTURE_2D, texture );
+
+    // Use clientside vertex pointers to render two items. In this test we have each
+    // attribute in a separate buffer, packed (i.e. stride == 0)
+    GLfloat vertexData[] = {  10,  10,
+                             300,  10,
+                             300, 128,
+                              10, 128,
+                             410,  10,
+                             600,  10,
+                             630, 200,
+                             310, 250,
+                             100, 300,
+                             300, 300,
+                             300, 400,
+                             100, 400 };
+    GLfloat textureData[] = { 0, 0,
+                              1, 0,
+                              1, 1,
+                              0, 1,
+                              0, 0.5,
+                              1, 0.5,
+                              1,   1,
+                              0.5, 1,
+                              0,   0,
+                              1,   0,
+                              1,   1,
+                              0,   1, };
+
+    glEnableClientState(GL_TEXTURE_2D); // XXX should be GL_TEXTURE_COORD_ARRAY); // XXX 
+    glTexCoordPointer(2, GL_FLOAT, 0, textureData);
+    glEnableClientState(GL_VERTEX_ARRAY);
+    glVertexPointer(2, GL_FLOAT, 0, vertexData);
+
+    glDrawArrays(GL_QUADS, 0, 12);
+
+    glDisableClientState(GL_TEXTURE_COORD_ARRAY);
+    glDisableClientState(GL_VERTEX_ARRAY);
+
+    SDL_GL_SwapBuffers();
+    
+#if !EMSCRIPTEN
+    // Wait for 3 seconds to give us a chance to see the image
+    SDL_Delay(3000);
+#endif
+
+    // Now we can delete the OpenGL texture and close down SDL
+    glDeleteTextures( 1, &texture );
+    
+    SDL_Quit();
+    
+    return 0;
+}
diff --git a/tests/gl_ps_workaround2.c b/tests/gl_ps_workaround2.c
new file mode 100644
index 00000000..e5bd2fd1
--- /dev/null
+++ b/tests/gl_ps_workaround2.c
@@ -0,0 +1,230 @@
+/*******************************************************************
+ *                                                                 *
+ *                        Using SDL With OpenGL                    *
+ *                                                                 *
+ *                    Tutorial by Kyle Foley (sdw)                 *
+ *                                                                 *
+ * http://gpwiki.org/index.php/SDL:Tutorials:Using_SDL_with_OpenGL *
+ *                                                                 *
+ *******************************************************************/
+
+/*
+THIS WORK, INCLUDING THE SOURCE CODE, DOCUMENTATION
+AND RELATED MEDIA AND DATA, IS PLACED INTO THE PUBLIC DOMAIN.
+
+THE ORIGINAL AUTHOR IS KYLE FOLEY.
+
+THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
+OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
+MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
+ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
+RESULTING FROM THE USE, MODIFICATION, OR
+REDISTRIBUTION OF THIS SOFTWARE.
+*/
+
+#if !EMSCRIPTEN
+#define USE_GLEW 1
+#endif
+
+#if USE_GLEW
+#include "GL/glew.h"
+#endif
+
+#include "SDL/SDL.h"
+#include "SDL/SDL_image.h"
+#if !USE_GLEW
+#include "SDL/SDL_opengl.h"
+#endif
+
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+void shaders() {
+#if USE_GLEW
+  glewInit();
+#endif
+
+  GLint ok;
+
+  const char *vertexShader = "void main(void) \n"
+                             "{ \n"
+                             "    gl_Position = ftransform(); \n"
+                             "    gl_TexCoord[0] = gl_MultiTexCoord0; \n"
+                             "    gl_FrontColor = gl_Color; \n"
+                             "} \n";
+  const char *fragmentShader = "uniform sampler2D tex0; \n"
+                               "void main(void) \n"
+                               "{ \n"
+                               "    gl_FragColor = gl_Color * texture2D(tex0, gl_TexCoord[0].xy); \n"
+                               "} \n";
+
+  GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+  glShaderSource(vs, 1, &vertexShader, NULL);
+  glCompileShader(vs);
+  glGetShaderiv(vs, GL_COMPILE_STATUS, &ok);
+  assert(ok);
+
+  GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+  glShaderSource(fs, 1, &fragmentShader, NULL);
+  glCompileShader(fs);
+  glGetShaderiv(fs, GL_COMPILE_STATUS, &ok);
+  assert(ok);
+
+  GLuint program = glCreateProgram();
+
+  glAttachShader(program, vs);
+  glAttachShader(program, fs);
+  glLinkProgram(program);
+  glGetProgramiv(program, GL_LINK_STATUS, &ok);
+  assert(ok);
+
+  glUseProgram(program);
+}
+
+int main(int argc, char *argv[])
+{
+    SDL_Surface *screen;
+
+    // Slightly different SDL initialization
+    if ( SDL_Init(SDL_INIT_VIDEO) != 0 ) {
+        printf("Unable to initialize SDL: %s\n", SDL_GetError());
+        return 1;
+    }
+
+    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 ); // *new*
+
+    screen = SDL_SetVideoMode( 640, 480, 16, SDL_OPENGL ); // *changed*
+    if ( !screen ) {
+        printf("Unable to set video mode: %s\n", SDL_GetError());
+        return 1;
+    }
+    
+    // Set the OpenGL state after creating the context with SDL_SetVideoMode
+
+    glClearColor( 0, 0, 0, 0 );
+    
+#if !EMSCRIPTEN
+    glEnable( GL_TEXTURE_2D ); // Need this to display a texture XXX unnecessary in OpenGL ES 2.0/WebGL
+#endif
+
+    glViewport( 0, 0, 640, 480 );
+
+    glMatrixMode( GL_PROJECTION );
+    GLfloat matrixData[] = { 2.0/640,        0,  0,  0,
+                                   0, -2.0/480,  0,  0,
+                                   0,        0, -1,  0,
+                                  -1,        1,  0,  1 };
+    glLoadMatrixf(matrixData); // test loadmatrix
+
+    glMatrixMode( GL_MODELVIEW );
+    glLoadIdentity();
+    
+    // Load the OpenGL texture
+
+    GLuint texture; // Texture object handle
+    SDL_Surface *surface; // Gives us the information to make the texture
+    
+    if ( (surface = IMG_Load("screenshot.png")) ) { 
+    
+        // Check that the image's width is a power of 2
+        if ( (surface->w & (surface->w - 1)) != 0 ) {
+            printf("warning: image.bmp's width is not a power of 2\n");
+        }
+    
+        // Also check if the height is a power of 2
+        if ( (surface->h & (surface->h - 1)) != 0 ) {
+            printf("warning: image.bmp's height is not a power of 2\n");
+        }
+    
+        // Have OpenGL generate a texture object handle for us
+        glGenTextures( 1, &texture );
+    
+        // Bind the texture object
+        glBindTexture( GL_TEXTURE_2D, texture );
+        
+        // Set the texture's stretching properties
+        glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
+        glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
+
+        //SDL_LockSurface(surface);
+
+        // Add some greyness
+        memset(surface->pixels, 0x66, surface->w*surface->h);
+
+        // Edit the texture object's image data using the information SDL_Surface gives us
+        glTexImage2D( GL_TEXTURE_2D, 0, GL_RGBA, surface->w, surface->h, 0, 
+                      GL_RGBA, GL_UNSIGNED_BYTE, surface->pixels );
+
+        //SDL_UnlockSurface(surface);
+    } 
+    else {
+        printf("SDL could not load image.bmp: %s\n", SDL_GetError());
+        SDL_Quit();
+        return 1;
+    }    
+    
+    // Free the SDL_Surface only if it was successfully created
+    if ( surface ) { 
+        SDL_FreeSurface( surface );
+    }
+    
+    // Clear the screen before drawing
+    glClear( GL_COLOR_BUFFER_BIT );
+    
+    shaders();
+
+    // Bind the texture to which subsequent calls refer to
+    glBindTexture( GL_TEXTURE_2D, texture );
+
+    // Use clientside vertex pointers to render two items. In this test we have each
+    // attribute in a separate buffer, packed (i.e. stride == 0)
+    GLfloat vertexData[] = {  10,  10,
+                             300,  10,
+                             300, 128,
+                              10, 128,
+                             410,  10,
+                             600,  10,
+                             630, 200,
+                             310, 250,
+                             100, 300,
+                             300, 300,
+                             300, 400,
+                             100, 400 };
+    GLfloat textureData[] = { 0, 0,
+                              1, 0,
+                              1, 1,
+                              0, 1,
+                              0, 0.5,
+                              1, 0.5,
+                              1,   1,
+                              0.5, 1,
+                              0,   0,
+                              1,   0,
+                              1,   1,
+                              0,   1, };
+
+    glEnable(GL_TEXTURE_2D); // XXX should be GL_TEXTURE_COORD_ARRAY); and also glEnableClientState! XXX two workarounds here
+    glTexCoordPointer(2, GL_FLOAT, 0, textureData);
+    glEnableClientState(GL_VERTEX_ARRAY);
+    glVertexPointer(2, GL_FLOAT, 0, vertexData);
+
+    glDrawArrays(GL_QUADS, 0, 12);
+
+    glDisableClientState(GL_TEXTURE_COORD_ARRAY);
+    glDisableClientState(GL_VERTEX_ARRAY);
+
+    SDL_GL_SwapBuffers();
+    
+#if !EMSCRIPTEN
+    // Wait for 3 seconds to give us a chance to see the image
+    SDL_Delay(3000);
+#endif
+
+    // Now we can delete the OpenGL texture and close down SDL
+    glDeleteTextures( 1, &texture );
+    
+    SDL_Quit();
+    
+    return 0;
+}
diff --git a/tests/glbook/CH13_ParticleSystem.png b/tests/glbook/CH13_ParticleSystem.png
index ff9c3496..4b69414f 100644
--- a/tests/glbook/CH13_ParticleSystem.png
+++ b/tests/glbook/CH13_ParticleSystem.png
diff --git a/tests/perspective.c b/tests/perspective.c
new file mode 100644
index 00000000..72f4c50f
--- /dev/null
+++ b/tests/perspective.c
@@ -0,0 +1,408 @@
+/*
+ * SDL OpenGL Tutorial.
+ * (c) Michael Vance, 2000
+ * briareos@lokigames.com
+ *
+ * Distributed under terms of the LGPL.
+ */
+
+#include <SDL/SDL.h>
+
+#ifdef EMSCRIPTEN
+#include <GL/gl.h>
+#include <GL/glu.h>
+#include "emscripten.h"
+#else
+#include <GL/gl.h>
+#include <GL/glu.h>
+#endif
+#include <stdio.h>
+#include <stdlib.h>
+
+#ifdef EMSCRIPTEN
+  #define emColor4ubv(x)
+#else
+#define emColor4ubv(x) glColor4ubv(x)
+#endif
+
+static GLboolean should_rotate = GL_TRUE;
+
+static void quit_tutorial( int code )
+{
+    /*
+     * Quit SDL so we can release the fullscreen
+     * mode and restore the previous video settings,
+     * etc.
+     */
+    SDL_Quit( );
+    
+    /* Exit program. */
+    exit( code );
+}
+
+static void handle_key_down( SDL_keysym* keysym )
+{
+    
+    /*
+     * We're only interested if 'Esc' has
+     * been presssed.
+     *
+     * EXERCISE:
+     * Handle the arrow keys and have that change the
+     * viewing position/angle.
+     */
+    switch( keysym->sym ) {
+        case SDLK_ESCAPE:
+            quit_tutorial( 0 );
+            break;
+        case SDLK_SPACE:
+            should_rotate = !should_rotate;
+            break;
+        default:
+            break;
+    }
+    
+}
+
+static void process_events( void )
+{
+    /* Our SDL event placeholder. */
+    SDL_Event event;
+    
+    /* Grab all the events off the queue. */
+    while( SDL_PollEvent( &event ) ) {
+        
+        switch( event.type ) {
+            case SDL_KEYDOWN:
+                /* Handle key presses. */
+                handle_key_down( &event.key.keysym );
+                break;
+            case SDL_QUIT:
+                /* Handle quit requests (like Ctrl-c). */
+                quit_tutorial( 0 );
+                break;
+        }
+        
+    }
+    
+}
+
+static void draw_screen( void )
+{
+    /* Our angle of rotation. */
+    static float angle = 0.0f;
+    
+    /*
+     * EXERCISE:
+     * Replace this awful mess with vertex
+     * arrays and a call to glDrawElements.
+     *
+     * EXERCISE:
+     * After completing the above, change
+     * it to use compiled vertex arrays.
+     *
+     * EXERCISE:
+     * Verify my windings are correct here ;).
+     */
+    static GLfloat v0[] = { -1.0f, -1.0f,  1.0f };
+    static GLfloat v1[] = {  1.0f, -1.0f,  1.0f };
+    static GLfloat v2[] = {  1.0f,  1.0f,  1.0f };
+    static GLfloat v3[] = { -1.0f,  1.0f,  1.0f };
+    static GLfloat v4[] = { -1.0f, -1.0f, -1.0f };
+    static GLfloat v5[] = {  1.0f, -1.0f, -1.0f };
+    static GLfloat v6[] = {  1.0f,  1.0f, -1.0f };
+    static GLfloat v7[] = { -1.0f,  1.0f, -1.0f };
+    static GLubyte red[]    = { 255,   0,   0, 255 };
+    static GLubyte green[]  = {   0, 255,   0, 255 };
+    static GLubyte blue[]   = {   0,   0, 255, 255 };
+    static GLubyte white[]  = { 255, 255, 255, 255 };
+    static GLubyte yellow[] = {   0, 255, 255, 255 };
+    static GLubyte black[]  = {   0,   0,   0, 255 };
+    static GLubyte orange[] = { 255, 255,   0, 255 };
+    static GLubyte purple[] = { 255,   0, 255,   0 };
+    
+    /* Clear the color and depth buffers. */
+    glClear( GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT );
+    
+    /* We don't want to modify the projection matrix. */
+    glMatrixMode( GL_MODELVIEW );
+    glLoadIdentity( );
+    
+    /* Move down the z-axis. */
+    glTranslatef( 0.0, 0.0, -5.0 );
+    
+    /* Rotate. */
+    glRotatef( angle, 0.0, 1.0, 0.0 );
+    
+    if( should_rotate ) {
+        
+        if( ++angle > 360.0f ) {
+            angle = 0.0f;
+        }
+        
+    }
+    
+    /* Send our triangle data to the pipeline. */
+    glBegin( GL_TRIANGLES );
+    
+    emColor4ubv( red );
+    glVertex3fv( v0 );
+    emColor4ubv( green );
+    glVertex3fv( v1 );
+    emColor4ubv( blue );
+    glVertex3fv( v2 );
+    
+    emColor4ubv( red );
+    glVertex3fv( v0 );
+    emColor4ubv( blue );
+    glVertex3fv( v2 );
+    emColor4ubv( white );
+    glVertex3fv( v3 );
+    
+    emColor4ubv( green );
+    glVertex3fv( v1 );
+    emColor4ubv( black );
+    glVertex3fv( v5 );
+    emColor4ubv( orange );
+    glVertex3fv( v6 );
+    
+    emColor4ubv( green );
+    glVertex3fv( v1 );
+    emColor4ubv( orange );
+    glVertex3fv( v6 );
+    emColor4ubv( blue );
+    glVertex3fv( v2 );
+    
+    emColor4ubv( black );
+    glVertex3fv( v5 );
+    emColor4ubv( yellow );
+    glVertex3fv( v4 );
+    emColor4ubv( purple );
+    glVertex3fv( v7 );
+    
+    emColor4ubv( black );
+    glVertex3fv( v5 );
+    emColor4ubv( purple );
+    glVertex3fv( v7 );
+    emColor4ubv( orange );
+    glVertex3fv( v6 );
+    
+    emColor4ubv( yellow );
+    glVertex3fv( v4 );
+    emColor4ubv( red );
+    glVertex3fv( v0 );
+    emColor4ubv( white );
+    glVertex3fv( v3 );
+    
+    emColor4ubv( yellow );
+    glVertex3fv( v4 );
+    emColor4ubv( white );
+    glVertex3fv( v3 );
+    emColor4ubv( purple );
+    glVertex3fv( v7 );
+    
+    emColor4ubv( white );
+    glVertex3fv( v3 );
+    emColor4ubv( blue );
+    glVertex3fv( v2 );
+    emColor4ubv( orange );
+    glVertex3fv( v6 );
+    
+    emColor4ubv( white );
+    glVertex3fv( v3 );
+    emColor4ubv( orange );
+    glVertex3fv( v6 );
+    emColor4ubv( purple );
+    glVertex3fv( v7 );
+    
+    emColor4ubv( green );
+    glVertex3fv( v1 );
+    emColor4ubv( red );
+    glVertex3fv( v0 );
+    emColor4ubv( yellow );
+    glVertex3fv( v4 );
+    
+    emColor4ubv( green );
+    glVertex3fv( v1 );
+    emColor4ubv( yellow );
+    glVertex3fv( v4 );
+    emColor4ubv( black );
+    glVertex3fv( v5 );
+    
+    glEnd( );
+    
+    /*
+     * EXERCISE:
+     * Draw text telling the user that 'Spc'
+     * pauses the rotation and 'Esc' quits.
+     * Do it using vetors and textured quads.
+     */
+    
+    /*
+     * Swap the buffers. This this tells the driver to
+     * render the next frame from the contents of the
+     * back-buffer, and to set all rendering operations
+     * to occur on what was the front-buffer.
+     *
+     * Double buffering prevents nasty visual tearing
+     * from the application drawing on areas of the
+     * screen that are being updated at the same time.
+     */
+    SDL_GL_SwapBuffers( );
+}
+
+static void setup_opengl( int width, int height )
+{
+    float ratio = (float) width / (float) height;
+    
+    /* Our shading model--Gouraud (smooth). */
+    glShadeModel( GL_SMOOTH );
+    
+    /* Culling. */
+    glCullFace( GL_BACK );
+    glFrontFace( GL_CCW );
+    glEnable( GL_CULL_FACE );
+    
+    /* Set the clear color. */
+    glClearColor( 0, 0, 0, 0 );
+    
+    /* Setup our viewport. */
+    glViewport( 0, 0, width, height );
+    
+    /*
+     * Change to the projection matrix and set
+     * our viewing volume.
+     */
+    glMatrixMode( GL_PROJECTION );
+    glLoadIdentity( );
+    /*
+     * EXERCISE:
+     * Replace this with a call to glFrustum.
+     */
+    gluPerspective( 60.0, ratio, 1.0, 1024.0 );
+}
+
+void one_iter();
+void one_iter() {
+    process_events( );
+    /* Draw the screen. */
+    draw_screen( );
+}
+
+int main( int argc, char* argv[] )
+{
+    /* Information about the current video settings. */
+    const SDL_VideoInfo* info = NULL;
+    /* Dimensions of our window. */
+    int width = 0;
+    int height = 0;
+    /* Color depth in bits of our window. */
+    int bpp = 0;
+    /* Flags we will pass into SDL_SetVideoMode. */
+    int flags = 0;
+    
+    /* First, initialize SDL's video subsystem. */
+    if( SDL_Init( SDL_INIT_VIDEO ) < 0 ) {
+        /* Failed, exit. */
+        fprintf( stderr, "Video initialization failed: %s\n",
+                SDL_GetError( ) );
+        quit_tutorial( 1 );
+    }
+    
+    /* Let's get some video information. */
+    info = SDL_GetVideoInfo( );
+    
+    if( !info ) {
+        /* This should probably never happen. */
+        fprintf( stderr, "Video query failed: %s\n",
+                SDL_GetError( ) );
+        quit_tutorial( 1 );
+    }
+    
+    /*
+     * Set our width/height to 640/480 (you would
+     * of course let the user decide this in a normal
+     * app). We get the bpp we will request from
+     * the display. On X11, VidMode can't change
+     * resolution, so this is probably being overly
+     * safe. Under Win32, ChangeDisplaySettings
+     * can change the bpp.
+     */
+    width = 640;
+    height = 480;
+    bpp = info->vfmt->BitsPerPixel;
+    
+    /*
+     * Now, we want to setup our requested
+     * window attributes for our OpenGL window.
+     * We want *at least* 5 bits of red, green
+     * and blue. We also want at least a 16-bit
+     * depth buffer.
+     *
+     * The last thing we do is request a double
+     * buffered window. '1' turns on double
+     * buffering, '0' turns it off.
+     *
+     * Note that we do not use SDL_DOUBLEBUF in
+     * the flags to SDL_SetVideoMode. That does
+     * not affect the GL attribute state, only
+     * the standard 2D blitting setup.
+     */
+//    SDL_GL_SetAttribute( SDL_GL_RED_SIZE, 5 );
+//    SDL_GL_SetAttribute( SDL_GL_GREEN_SIZE, 5 );
+//    SDL_GL_SetAttribute( SDL_GL_BLUE_SIZE, 5 );
+//    SDL_GL_SetAttribute( SDL_GL_DEPTH_SIZE, 16 );
+//    SDL_GL_SetAttribute( SDL_GL_DOUBLEBUFFER, 1 );
+    
+    /*
+     * We want to request that SDL provide us
+     * with an OpenGL window, in a fullscreen
+     * video mode.
+     *
+     * EXERCISE:
+     * Make starting windowed an option, and
+     * handle the resize events properly with
+     * glViewport.
+     */
+    flags = SDL_OPENGL;// | SDL_FULLSCREEN;
+    
+    /*
+     * Set the video mode
+     */
+    if( SDL_SetVideoMode( width, height, bpp, flags ) == 0 ) {
+        /*
+         * This could happen for a variety of reasons,
+         * including DISPLAY not being set, the specified
+         * resolution not being available, etc.
+         */
+        fprintf( stderr, "Video mode set failed: %s\n",
+                SDL_GetError( ) );
+        quit_tutorial( 1 );
+    }
+    
+    /*
+     * At this point, we should have a properly setup
+     * double-buffered window for use with OpenGL.
+     */
+    setup_opengl( width, height );
+    
+    /*
+     * Now we want to begin our normal app process--
+     * an event loop with a lot of redrawing.
+     */
+    one_iter(); // just one for testing purposes
+
+#ifndef EMSCRIPTEN
+    SDL_Delay(2000);
+#endif    
+    
+    /*
+     * EXERCISE:
+     * Record timings using SDL_GetTicks() and
+     * and print out frames per second at program
+     * end.
+     */
+    
+    /* Never reached. */
+    return 0;
+}
diff --git a/tests/perspective.png b/tests/perspective.png
new file mode 100644
index 00000000..04cedc71
--- /dev/null
+++ b/tests/perspective.png
diff --git a/tests/runner.py b/tests/runner.py
index b6980933..96a2dabd 100755
--- a/tests/runner.py
+++ b/tests/runner.py
@@ -100,14 +100,8 @@ class RunnerCore(unittest.TestCase):
       for temp_file in os.listdir(TEMP_DIR):
         if temp_file.endswith('.ll'):
           self.has_prev_ll = True
-    
+
   def tearDown(self):
-    if self.save_JS:
-      for name in os.listdir(self.get_dir()):
-        if name.endswith(('.o.js', '.cc.js')):
-          suff = '.'.join(name.split('.')[-2:])
-          shutil.copy(os.path.join(self.get_dir(), name),
-                      os.path.join(TEMP_DIR, self.id().replace('__main__.', '').replace('.test_', '.')+'.'+suff))
     if not self.save_dir:
       # rmtree() fails on Windows if the current working directory is inside the tree.
       os.chdir(os.path.join(self.get_dir(), '..'))
@@ -140,6 +134,12 @@ class RunnerCore(unittest.TestCase):
   def get_stdout_path(self):
     return os.path.join(self.get_dir(), 'stdout')
 
+  def hardcode_arguments(self, filename, args):
+    # Hardcode in the arguments, so js is portable without manual commandlinearguments
+    if not args: return
+    js = open(filename).read()
+    open(filename, 'w').write(js.replace('var ret = run();', 'var ret = run(%s);' % str(args)))
+
   def prep_ll_run(self, filename, ll_file, force_recompile=False, build_ll_hook=None):
     if ll_file.endswith(('.bc', '.o')):
       if ll_file != filename + '.o':
@@ -435,6 +435,8 @@ process(sys.argv[1])
 
 sys.argv = map(lambda arg: arg if not arg.startswith('test_') else 'default.' + arg, sys.argv)
 
+test_index = 0
+
 if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'browser' not in str(sys.argv):
   # Tests
 
@@ -448,29 +450,35 @@ if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'brows
   class T(RunnerCore): # Short name, to make it more fun to use manually on the commandline
     ## Does a complete test - builds, runs, checks output, etc.
     def do_run(self, src, expected_output, args=[], output_nicerizer=None, output_processor=None, no_build=False, main_file=None, additional_files=[], js_engines=None, post_build=None, basename='src.cpp', libraries=[], includes=[], force_c=False, build_ll_hook=None, extra_emscripten_args=[]):
-        if force_c or (main_file is not None and main_file[-2:]) == '.c':
-          basename = 'src.c'
-          Building.COMPILER = to_cc(Building.COMPILER)
-
-        dirname = self.get_dir()
-        filename = os.path.join(dirname, basename)
-        if not no_build:
-          self.build(src, dirname, filename, main_file=main_file, additional_files=additional_files, libraries=libraries, includes=includes,
-                     build_ll_hook=build_ll_hook, extra_emscripten_args=extra_emscripten_args, post_build=post_build)
-
-        # Run in both JavaScript engines, if optimizing - significant differences there (typed arrays)
-        if js_engines is None:
-          js_engines = JS_ENGINES
-        if Settings.USE_TYPED_ARRAYS:
-          js_engines = filter(lambda engine: engine != V8_ENGINE, js_engines) # V8 issue 1822
-        js_engines = filter(lambda engine: engine not in self.banned_js_engines, js_engines)
-        if len(js_engines) == 0: return self.skip('No JS engine present to run this test with. Check %s and the paths therein.' % EM_CONFIG)
-        for engine in js_engines:
-          js_output = self.run_generated_code(engine, filename + '.o.js', args, output_nicerizer=output_nicerizer)
-          self.assertContained(expected_output, js_output.replace('\r\n', '\n'))
-          self.assertNotContained('ERROR', js_output)
-
-        #shutil.rmtree(dirname) # TODO: leave no trace in memory. But for now nice for debugging
+      if force_c or (main_file is not None and main_file[-2:]) == '.c':
+        basename = 'src.c'
+        Building.COMPILER = to_cc(Building.COMPILER)
+
+      dirname = self.get_dir()
+      filename = os.path.join(dirname, basename)
+      if not no_build:
+        self.build(src, dirname, filename, main_file=main_file, additional_files=additional_files, libraries=libraries, includes=includes,
+                   build_ll_hook=build_ll_hook, extra_emscripten_args=extra_emscripten_args, post_build=post_build)
+
+      # Run in both JavaScript engines, if optimizing - significant differences there (typed arrays)
+      if js_engines is None:
+        js_engines = JS_ENGINES
+      if Settings.USE_TYPED_ARRAYS:
+        js_engines = filter(lambda engine: engine != V8_ENGINE, js_engines) # V8 issue 1822
+      js_engines = filter(lambda engine: engine not in self.banned_js_engines, js_engines)
+      if len(js_engines) == 0: return self.skip('No JS engine present to run this test with. Check %s and the paths therein.' % EM_CONFIG)
+      for engine in js_engines:
+        js_output = self.run_generated_code(engine, filename + '.o.js', args, output_nicerizer=output_nicerizer)
+        self.assertContained(expected_output, js_output.replace('\r\n', '\n'))
+        self.assertNotContained('ERROR', js_output)
+
+      #shutil.rmtree(dirname) # TODO: leave no trace in memory. But for now nice for debugging
+
+      if self.save_JS:
+        global test_index
+        self.hardcode_arguments(filename + '.o.js', args)
+        shutil.copyfile(filename + '.o.js', os.path.join(TEMP_DIR, str(test_index) + '.js'))
+        test_index += 1
 
     # No building - just process an existing .ll file (or .bc, which we turn into .ll)
     def do_ll_run(self, ll_file, expected_output=None, args=[], js_engines=None, output_nicerizer=None, post_build=None, force_recompile=False, build_ll_hook=None, extra_emscripten_args=[]):
@@ -1131,7 +1139,6 @@ m_divisor is 1091269979
     def test_i32_mul_precise(self):
       if self.emcc_args == None: return self.skip('needs ta2')
 
-      self.emcc_args += ['-s', 'PRECISE_I32_MUL=1']
       src = r'''
         #include <stdio.h>
 
@@ -1146,6 +1153,10 @@ m_divisor is 1091269979
       self.do_run(src, '3217489085')
 
     def test_i32_mul_semiprecise(self):
+      if Settings.ASM_JS: return self.skip('asm is always fully precise')
+
+      Settings.PRECISE_I32_MUL = 0 # we want semiprecise here
+
       src = r'''
         #include <stdio.h>
 
@@ -2845,6 +2856,23 @@ Exiting setjmp function, level: 0, prev_jmp: -1
         '''
         self.do_run(src, 'a1: 0\na2: 0\na3: 1\nb1: 0\nb2: 1\nb3: 1\nc1: 1\nc2: 1\nc3: 1\n')
 
+    def test_dynamic_cast_2(self):
+      if self.emcc_args is None: return self.skip('need libcxxabi')
+
+      src = r'''
+        #include <stdio.h>
+        #include <typeinfo>
+
+        class Class {};
+
+        int main() {
+            const Class* dp = dynamic_cast<const Class*>(&typeid(Class));
+            // should return dp == NULL,
+            printf("pointer: %p\n", dp);
+        }
+        '''
+      self.do_run(src, "pointer: (nil)")
+
     def test_funcptr(self):
         src = '''
           #include <stdio.h>
@@ -3966,6 +3994,18 @@ The current type of b is: 9
         '''
       self.do_run(src, '*0\n')
 
+    def test_time_c(self):
+      src = r'''
+        #include <time.h>
+        #include <stdio.h>
+
+        int main() {
+          time_t t = time(0);
+          printf("time: %s\n", ctime(&t));
+        }
+      '''
+      self.do_run(src, 'time: ') # compilation check, mainly
+
     def test_intentional_fault(self):
       if Settings.ASM_JS: return self.skip('no throw support in asm')
       # Some programs intentionally segfault themselves, we should compile that into a throw
@@ -4095,10 +4135,12 @@ The current type of b is: 9
           '''
         self.do_run(src, '*0.00,0.00,0.00*\n*0,77,0*\n*0,77,0*\n*0,77,0*')
 
-    def test_memcpy(self):
+    def test_memcpy_memcmp(self):
         src = '''
           #include <stdio.h>
           #include <string.h>
+          #include <assert.h>
+
           #define MAXX 48
           void reset(unsigned char *buffer) {
             for (int i = 0; i < MAXX; i++) buffer[i] = i+1;
@@ -4119,6 +4161,20 @@ The current type of b is: 9
                   reset(buffer);
                   memcpy(buffer+i, buffer+j, k);
                   dump(buffer);
+                  assert(memcmp(buffer+i, buffer+j, k) == 0);
+                  buffer[i + k/2]++;
+                  if (buffer[i + k/2] != 0) {
+                    assert(memcmp(buffer+i, buffer+j, k) > 0);
+                  } else {
+                    assert(memcmp(buffer+i, buffer+j, k) < 0);
+                  }
+                  buffer[i + k/2]--;
+                  buffer[j + k/2]++;
+                  if (buffer[j + k/2] != 0) {
+                    assert(memcmp(buffer+i, buffer+j, k) < 0);
+                  } else {
+                    assert(memcmp(buffer+i, buffer+j, k) > 0);
+                  }
                 }
               }
             }
@@ -4197,6 +4253,19 @@ The current type of b is: 9
       '''
       self.do_run(src, 'final: -403200.');
 
+    def test_memmove3(self):
+      src = '''
+        #include <stdio.h>
+        #include <string.h>
+        int main() {
+          char str[] = "memmove can be vvery useful....!";
+          memmove(str+15, str+16, 17);
+          puts(str);
+          return 0;
+        }
+      '''
+      self.do_run(src, 'memmove can be very useful....!')
+
     def test_bsearch(self):
       if Settings.QUANTUM_SIZE == 1: return self.skip('Test cannot work with q1')
 
@@ -5299,9 +5368,30 @@ Pass: 0.000012 0.000012''')
           return(0);
         }
       '''
-
       self.do_run(src, '3\nday 19, month Nov, year 2012');
 
+    def test_sscanf_5(self):
+      src = r'''
+        #include "stdio.h"
+
+        static const char *colors[] = {
+          "  c black",
+          ". c #001100",
+          "X c #111100"
+        };
+
+        int main(){
+          unsigned char code;
+          char color[32];
+          int rcode;
+          for(int i = 0; i < 3; i++) {
+            rcode = sscanf(colors[i], "%c c %s", &code, color);
+            printf("%i, %c, %s\n", rcode, code, color);
+          }
+        }
+      '''
+      self.do_run(src, '2,  , black\n2, ., #001100\n2, X, #111100');
+
     def test_langinfo(self):
       src = open(path_from_root('tests', 'langinfo', 'test.c'), 'r').read()
       expected = open(path_from_root('tests', 'langinfo', 'output.txt'), 'r').read()
@@ -6283,6 +6373,53 @@ int main(int argc, char **argv) {
 
       self.do_run(src, '789:123.46\n0:100.1')
 
+    def test_reinterpreted_ptrs(self):
+      if self.emcc_args is None: return self.skip('needs emcc and libc')
+
+      src = r'''
+#include <stdio.h>
+
+class Foo {
+private:
+    float bar;
+public:
+    int baz;
+
+    Foo(): bar(0), baz(4711) {};
+
+    int getBar() const;
+};
+
+int Foo::getBar() const {
+    return this->bar;
+};
+
+const Foo *magic1 = reinterpret_cast<Foo*>(0xDEAD111F);
+const Foo *magic2 = reinterpret_cast<Foo*>(0xDEAD888F);
+
+static void runTest() {
+
+    const Foo *a = new Foo();
+    const Foo *b = a;
+
+    if (a->getBar() == 0) {
+        if (a->baz == 4712)
+            b = magic1;
+        else
+            b = magic2;
+    }
+
+    printf("%s\n", (b == magic1 ? "magic1" : (b == magic2 ? "magic2" : "neither")));
+};
+
+extern "C" {
+    int main(int argc, char **argv) {
+        runTest();
+    }
+}
+'''
+      self.do_run(src, 'magic2')
+
     def test_jansson(self):
         return self.skip('currently broken')
 
@@ -6783,7 +6920,7 @@ def process(filename):
       if self.emcc_args is None: return self.skip('Very slow without ta2, and we would also need to include dlmalloc manually without emcc')
       if Settings.QUANTUM_SIZE == 1: return self.skip('TODO FIXME')
 
-      Settings.CORRECT_SIGNS = 1 # XXX: in default, we fail with 2 here, even though the pgo_data should be correct (and works in s_0_0). Investigate this.
+      Settings.CORRECT_SIGNS = 1
       Settings.CORRECT_OVERFLOWS = 0
       Settings.CORRECT_ROUNDINGS = 0
       if self.emcc_args is None: Settings.SAFE_HEAP = 0 # uses time.h to set random bytes, other stuff
@@ -6995,6 +7132,14 @@ def process(filename):
           self.assertIdentical(clean(open('release.js').read()), clean(open('debug%d.js' % debug).read())) # EMCC_DEBUG=1 mode must not generate different code!
           print >> sys.stderr, 'debug check %d passed too' % debug
 
+        try:
+          os.environ['EMCC_FORCE_STDLIBS'] = '1'
+          print 'EMCC_FORCE_STDLIBS'
+          do_test()
+        finally:
+          del os.environ['EMCC_FORCE_STDLIBS']
+        print >> sys.stderr, 'EMCC_FORCE_STDLIBS ok'
+
         try_delete(CANONICAL_TEMP_DIR)
       else:
         print >> sys.stderr, 'not doing debug check'
@@ -7062,6 +7207,16 @@ def process(filename):
       finally:
         del os.environ['EMCC_LEAVE_INPUTS_RAW']
 
+    def test_fuzz(self):
+      if Settings.USE_TYPED_ARRAYS != 2: return self.skip('needs ta2')
+
+      Building.COMPILER_TEST_OPTS += ['-I' + path_from_root('tests', 'fuzz')]
+
+      for name in glob.glob(path_from_root('tests', 'fuzz', '*.c')):
+        print name
+        self.do_run(open(path_from_root('tests', 'fuzz', name)).read(),
+                    open(path_from_root('tests', 'fuzz', name + '.txt')).read(), force_c=True)
+
     # Autodebug the code
     def do_autodebug(self, filename):
       output = Popen([PYTHON, AUTODEBUGGER, filename+'.o.ll', filename+'.o.ll.ll'], stdout=PIPE, stderr=self.stderr_redirect).communicate()[0]
@@ -7113,40 +7268,101 @@ def process(filename):
         '''
       self.do_run(src, '''AD:-1,1''', build_ll_hook=self.do_autodebug)
 
-    def test_profiling(self):
-      if Settings.ASM_JS: return self.skip('asm does not support profiling')
+    def test_corruption(self):
+      if Settings.ASM_JS: return self.skip('cannot use corruption checks in asm')
+      if Settings.USE_TYPED_ARRAYS != 2: return self.skip('needs ta2 for actual test')
 
-      src = '''
-          #include <emscripten.h>
-          #include <unistd.h>
+      Settings.CORRUPTION_CHECK = 1
 
-          int main()
+      src = r'''
+        #include <stdio.h>
+        #include <stdlib.h>
+        #include <string.h>
+        int main(int argc, char **argv) {
+          int size = 1024*argc;
+          char *buffer = (char*)malloc(size);
+        #if CORRUPT
+          memset(buffer, argc, size+15);
+        #else
+          memset(buffer, argc, size);
+        #endif
+          for (int x = 0; x < size; x += argc*3) buffer[x] = x/3;
+          int ret = 0;
+          for (int x = 0; x < size; x++) ret += buffer[x];
+          free(buffer);
+          printf("All ok, %d\n", ret);
+        }
+      '''
+
+      for corrupt in [1]:
+        self.do_run(src.replace('CORRUPT', str(corrupt)), 'Heap corruption detected!' if corrupt else 'All ok, 4209')
+
+    def test_corruption_2(self):
+      if Settings.ASM_JS: return self.skip('cannot use corruption checks in asm')
+      if Settings.USE_TYPED_ARRAYS != 2: return self.skip('needs ta2 for actual test')
+
+      Settings.SAFE_HEAP = 1
+      Settings.CORRUPTION_CHECK = 1
+
+      # test for free(0), malloc(0), etc.
+      src = r'''
+        #include <iostream>
+        #include <fstream>
+        #include <stdlib.h>
+        #include <stdio.h>
+
+        void bye() {
+          printf("all ok\n");
+        }
+
+        int main() {
+          atexit(bye);
+
+          std::string testPath = "/Script/WA-KA.txt";
+          std::fstream str(testPath.c_str(), std::ios::in | std::ios::binary);
+
+          if (str.is_open())
           {
-            EMSCRIPTEN_PROFILE_INIT(3);
-            EMSCRIPTEN_PROFILE_BEGIN(0);
-            usleep(10 * 1000);
-            EMSCRIPTEN_PROFILE_END(0);
-            EMSCRIPTEN_PROFILE_BEGIN(1);
-            usleep(50 * 1000);
-            EMSCRIPTEN_PROFILE_END(1);
-            EMSCRIPTEN_PROFILE_BEGIN(2);
-            usleep(250 * 1000);
-            EMSCRIPTEN_PROFILE_END(2);
-            return 0;
+            std::cout << "open!" << std::endl;
+          } else {
+            std::cout << "missing!" << std::endl;
           }
+
+          return 1;
+        }
         '''
+      self.do_run(src, 'missing!\nall ok\n')
 
-      post1 = '''
-def process(filename):
-  src = open(filename, 'a')
-  src.write(\'\'\'
-    Profiling.dump();
-  \'\'\')
-  src.close()
-'''
+    def test_corruption_3(self):
+      if Settings.ASM_JS: return self.skip('cannot use corruption checks in asm')
+      if Settings.USE_TYPED_ARRAYS != 2: return self.skip('needs ta2 for actual test')
 
-      self.do_run(src, '''Profiling data:
-Block 0: ''', post_build=post1)
+      Settings.CORRUPTION_CHECK = 1
+
+      # realloc
+      src = r'''
+        #include <stdlib.h>
+        #include <stdio.h>
+        #include <assert.h>
+
+        void bye() {
+          printf("all ok\n");
+        }
+
+        int main(int argc, char **argv) {
+          atexit(bye);
+
+          char *buffer = (char*)malloc(100);
+          for (int i = 0; i < 100; i++) buffer[i] = (i*i)%256;
+          buffer = (char*)realloc(buffer, argc + 50);
+          for (int i = 0; i < argc + 50; i++) {
+            //printf("%d : %d : %d : %d\n", i, (int)(buffer + i), buffer[i], (char)((i*i)%256));
+            assert(buffer[i] == (char)((i*i)%256));
+          }
+          return 1;
+        }
+        '''
+      self.do_run(src, 'all ok\n')
 
     ### Integration tests
 
@@ -7676,6 +7892,8 @@ def process(filename):
 
     def test_debug(self):
       if '-g' not in Building.COMPILER_TEST_OPTS: Building.COMPILER_TEST_OPTS.append('-g')
+      if self.emcc_args is not None:
+        if '-O1' in self.emcc_args or '-O2' in self.emcc_args: return self.skip('optimizations remove LLVM debug info')
 
       src = '''
         #include <stdio.h>
@@ -8051,7 +8269,6 @@ class %s(T):
     Settings.ASSERTIONS = 1-embetter
     Settings.SAFE_HEAP = 1-(embetter and llvm_opts)
     Building.LLVM_OPTS = llvm_opts
-    Settings.PGO = 0
     Settings.CHECK_OVERFLOWS = 1-(embetter or llvm_opts)
     Settings.CORRECT_OVERFLOWS = 1-(embetter and llvm_opts)
     Settings.CORRECT_SIGNS = 0
@@ -8061,7 +8278,6 @@ class %s(T):
     Settings.INIT_STACK = 0
     Settings.RUNTIME_TYPE_INFO = 0
     Settings.DISABLE_EXCEPTION_CATCHING = 0
-    Settings.PROFILE = 0
     Settings.INCLUDE_FULL_LIBRARY = 0
     Settings.BUILD_AS_SHARED_LIB = 0
     Settings.RUNTIME_LINKED_LIBS = []
@@ -8084,11 +8300,11 @@ TT = %s
   exec('o1 = make_run("o1", compiler=CLANG, emcc_args=["-O1", "-s", "SAFE_HEAP=1"])')
 
   # Make one run with -O2, but without closure (we enable closure in specific tests, otherwise on everything it is too slow)
-  exec('o2 = make_run("o2", compiler=CLANG, emcc_args=["-O2", "--closure", "0"])')
+  exec('o2 = make_run("o2", compiler=CLANG, emcc_args=["-O2"])')
 
   # asm.js
-  #exec('asm = make_run("asm", compiler=CLANG, emcc_args=["-O0", "--closure", "0", "-s", "ASM_JS=1"])')
-  exec('asm2 = make_run("asm2", compiler=CLANG, emcc_args=["-O2", "--closure", "0", "-s", "ASM_JS=1"])')
+  #exec('asm = make_run("asm", compiler=CLANG, emcc_args=["-O0", "-s", "ASM_JS=1"])')
+  exec('asm2 = make_run("asm2", compiler=CLANG, emcc_args=["-O2", "-s", "ASM_JS=1"])')
 
   # Make custom runs with various options
   for compiler, quantum, embetter, typed_arrays, llvm_opts in [
@@ -8114,11 +8330,12 @@ TT = %s
 
         # --version
         output = Popen([PYTHON, compiler, '--version'], stdout=PIPE, stderr=PIPE).communicate()
-        self.assertContained('''emcc (Emscripten GCC-like replacement) 2.0
-Copyright (C) 2012 the Emscripten authors.
+        output = output[0].replace('\r', '')
+        self.assertContained('''emcc (Emscripten GCC-like replacement)''', output)
+        self.assertContained('''Copyright (C) 2013 the Emscripten authors (see AUTHORS.txt)
 This is free and open source software under the MIT license.
 There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
-''', output[0].replace('\r', ''), output[1].replace('\r', ''))
+''', output)
 
         # -v, without input files
         output = Popen([PYTHON, compiler, '-v'], stdout=PIPE, stderr=PIPE).communicate()
@@ -8224,20 +8441,22 @@ Options that are modified or new in %s include:
           (['-o', 'something.js', '-O0'],                   0, None, 0, 0),
           (['-o', 'something.js', '-O1'],                   1, None, 0, 0),
           (['-o', 'something.js', '-O1', '--closure', '1'], 1, None, 1, 0),
-          (['-o', 'something.js', '-O2'],                   2, None, 1, 1),
+          (['-o', 'something.js', '-O2'],                   2, None, 0, 1),
           (['-o', 'something.js', '-O2', '--closure', '0'], 2, None, 0, 0),
+          (['-o', 'something.js', '-O2', '-g'],             2, None, 0, 0),
           (['-o', 'something.js', '-O3'],                   3, None, 1, 1),
           (['-o', 'something.js', '-O3', '--closure', '0'], 3, None, 0, 0),
           # and, test compiling to bitcode first
           (['-o', 'something.bc'], 0, [],      0, 0),
           (['-o', 'something.bc'], 0, ['-O0'], 0, 0),
           (['-o', 'something.bc'], 1, ['-O1'], 0, 0),
-          (['-o', 'something.bc'], 2, ['-O2'], 1, 0),
+          (['-o', 'something.bc'], 2, ['-O2'], 0, 0),
           (['-o', 'something.bc'], 3, ['-O3'], 1, 0),
           (['-O1', '-o', 'something.bc'], 0, [], 0, 0), # -Ox is ignored and warned about
         ]:
-          #print params, opt_level, bc_params, closure
+          print params, opt_level, bc_params, closure, has_malloc
           self.clear()
+          keep_debug = '-g' in params
           output = Popen([PYTHON, compiler, path_from_root('tests', 'hello_world_loop' + ('_malloc' if has_malloc else '') + '.cpp')] + params,
                          stdout=PIPE, stderr=PIPE).communicate()
           assert len(output[0]) == 0, output[0]
@@ -8261,14 +8480,13 @@ Options that are modified or new in %s include:
           else:
             # closure has not been run, we can do some additional checks. TODO: figure out how to do these even with closure
             assert 'Module._main = ' not in generated, 'closure compiler should not have been run'
-            # XXX find a way to test this: assert ('& 255' in generated or '&255' in generated) == (opt_level <= 2), 'corrections should be in opt <= 2'
-            assert ('(label)' in generated) == (opt_level <= 1), 'relooping should be in opt >= 2'
-            assert ('assert(STACKTOP < STACK_MAX' in generated) == (opt_level == 0), 'assertions should be in opt == 0'
-            assert 'var $i;' in generated or 'var $i_0' in generated or 'var $storemerge3;' in generated or 'var $storemerge4;' in generated or 'var $i_04;' in generated, 'micro opts should always be on'
+            if keep_debug:
+              assert ('(label)' in generated) == (opt_level <= 1), 'relooping should be in opt >= 2'
+              assert ('assert(STACKTOP < STACK_MAX' in generated) == (opt_level == 0), 'assertions should be in opt == 0'
+              assert 'var $i;' in generated or 'var $i_0' in generated or 'var $storemerge3;' in generated or 'var $storemerge4;' in generated or 'var $i_04;' in generated, 'micro opts should always be on'
             if opt_level >= 2:
-              assert re.search('HEAP8\[\$\w+ \+ \(+\$\w+ ', generated) or re.search('HEAP8\[HEAP32\[', generated), 'eliminator should create compound expressions, and fewer one-time vars' # also in -O1, but easier to test in -O2
+              assert re.search('HEAP8\[\$?\w+ \+ \(+\$?\w+ ', generated) or re.search('HEAP8\[HEAP32\[', generated), 'eliminator should create compound expressions, and fewer one-time vars' # also in -O1, but easier to test in -O2
             assert ('_puts(' in generated) == (opt_level >= 1), 'with opt >= 1, llvm opts are run and they should optimize printf to puts'
-            assert ('function _malloc(bytes) {' in generated) == (not has_malloc), 'If malloc is needed, it should be there, if not not'
             assert 'function _main() {' in generated, 'Should be unminified, including whitespace'
             assert ('-O3' in (params+(bc_params or []))) or'function _dump' in generated, 'No inlining by default'
 
@@ -8588,6 +8806,21 @@ f.close()
 
       self.assertContained('result: 62', run_js(os.path.join(self.get_dir(), 'a.out.js')))
 
+    def test_asm_undefined(self):
+      src = r'''
+        #include <stdio.h>
+        extern void doit();
+        int main(int argc, char **argv) {
+          if (argc == 121) doit();
+          printf("done\n");
+          return 1;
+        }
+      '''
+      filename = self.in_dir('src.cpp')
+      open(filename, 'w').write(src)
+      out, err = Popen([PYTHON, EMCC, filename, '-s', 'ASM_JS=1', '-O2'], stderr=PIPE).communicate()
+      assert 'Warning: Unresolved symbol' in err, 'always warn on undefs in asm, since it breaks validation'
+
     def test_redundant_link(self):
       lib = "int mult() { return 1; }"
       lib_name = os.path.join(self.get_dir(), 'libA.c')
@@ -9011,7 +9244,7 @@ f.close()
           return 0;
         }
       ''')
-      Popen([PYTHON, EMCC, '-O2', '--closure', '-0', os.path.join(self.get_dir(), 'main.cpp')]).communicate()
+      Popen([PYTHON, EMCC, '-O2', os.path.join(self.get_dir(), 'main.cpp')]).communicate()
       output = run_js(os.path.join(self.get_dir(), 'a.out.js'), full_output=True, stderr=PIPE)
       self.assertContained('''0:0
 1:1
@@ -9223,6 +9456,26 @@ f.close()
       finally:
         del os.environ['EMCC_DEBUG']
 
+    def test_debuginfo(self):
+      if os.environ.get('EMCC_DEBUG'): return self.skip('cannot run in debug mode')
+      try:
+        os.environ['EMCC_DEBUG'] = '1'
+        # llvm debug info is kept only when we can see it, which is without the js optimize, -O0. js debug info is lost by registerize in -O2, so - g disables it
+        for args, expect_llvm, expect_js in [
+            (['-O0'], True, True),
+            (['-O0', '-g'], True, True),
+            (['-O1'], False, True),
+            (['-O1', '-g'], False, True),
+            (['-O2'], False, False),
+            (['-O2', '-g'], False, True),
+          ]:
+          print args, expect_llvm, expect_js
+          output, err = Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world.cpp')] + args, stdout=PIPE, stderr=PIPE).communicate()
+          assert expect_llvm == ('strip-debug' not in err)
+          assert expect_js == ('registerize' not in err)
+      finally:
+        del os.environ['EMCC_DEBUG']
+
     def test_scons(self): # also incidentally tests c++11 integration in llvm 3.1
       try_delete(os.path.join(self.get_dir(), 'test'))
       shutil.copytree(path_from_root('tests', 'scons'), os.path.join(self.get_dir(), 'test'))
@@ -9958,6 +10211,12 @@ elif 'browser' in str(sys.argv):
       Popen([PYTHON, EMCC, os.path.join(self.get_dir(), 'sdl_image.c'), '-O2', '--preload-file', 'screenshot.jpg', '-o', 'page.html']).communicate()
       self.run_browser('page.html', '', '/report_result?600')
 
+    def test_sdl_image_jpeg(self):
+      shutil.copyfile(path_from_root('tests', 'screenshot.jpg'), os.path.join(self.get_dir(), 'screenshot.jpeg'))
+      open(os.path.join(self.get_dir(), 'sdl_image_jpeg.c'), 'w').write(self.with_report_result(open(path_from_root('tests', 'sdl_image_jpeg.c')).read()))
+      Popen([PYTHON, EMCC, os.path.join(self.get_dir(), 'sdl_image_jpeg.c'), '--preload-file', 'screenshot.jpeg', '-o', 'page.html']).communicate()
+      self.run_browser('page.html', '', '/report_result?600')
+
     def test_sdl_image_compressed(self):
       for image, width in [(path_from_root('tests', 'screenshot2.png'), 300),
                            (path_from_root('tests', 'screenshot.jpg'), 600)]:
@@ -9977,12 +10236,12 @@ elif 'browser' in str(sys.argv):
         self.run_browser('page.html', '', '/report_result?' + str(width))
 
     def test_sdl_image_prepare(self):
-      # load an image file, get pixel data. Also O2 coverage for --preload-file
+      # load an image file, get pixel data.
       shutil.copyfile(path_from_root('tests', 'screenshot.jpg'), os.path.join(self.get_dir(), 'screenshot.not'))
       self.btest('sdl_image_prepare.c', reference='screenshot.jpg', args=['--preload-file', 'screenshot.not'])
 
     def test_sdl_image_prepare_data(self):
-      # load an image file, get pixel data. Also O2 coverage for --preload-file
+      # load an image file, get pixel data.
       shutil.copyfile(path_from_root('tests', 'screenshot.jpg'), os.path.join(self.get_dir(), 'screenshot.not'))
       self.btest('sdl_image_prepare_data.c', reference='screenshot.jpg', args=['--preload-file', 'screenshot.not'])
 
@@ -10080,56 +10339,56 @@ elif 'browser' in str(sys.argv):
       # SDL, OpenGL, textures, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-gray-purple.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_ogl.c'), '-O2', '--minify', '0', '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_ogl.c'), '-O2', '--minify', '0', '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with gray at the top.', '/report_result?0')
 
     def test_sdl_ogl_defaultmatrixmode(self):
       # SDL, OpenGL, textures, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-gray-purple.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_ogl_defaultMatrixMode.c'), '--minify', '0', '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_ogl_defaultMatrixMode.c'), '--minify', '0', '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with gray at the top.', '/report_result?0')
 
     def test_sdl_ogl_p(self):
       # Immediate mode with pointers
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-gray.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_ogl_p.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_ogl_p.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with gray at the top.', '/report_result?0')
 
     def test_sdl_fog_simple(self):
       # SDL, OpenGL, textures, fog, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-fog-simple.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_simple.c'), '-O2', '--minify', '0', '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_simple.c'), '-O2', '--minify', '0', '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with fog.', '/report_result?0')
 
     def test_sdl_fog_negative(self):
       # SDL, OpenGL, textures, fog, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-fog-negative.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_negative.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_negative.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with fog.', '/report_result?0')
 
     def test_sdl_fog_density(self):
       # SDL, OpenGL, textures, fog, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-fog-density.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_density.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_density.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with fog.', '/report_result?0')
 
     def test_sdl_fog_exp2(self):
       # SDL, OpenGL, textures, fog, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-fog-exp2.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_exp2.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_exp2.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with fog.', '/report_result?0')
 
     def test_sdl_fog_linear(self):
       # SDL, OpenGL, textures, fog, immediate mode. Closure for more coverage
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.reftest(path_from_root('tests', 'screenshot-fog-linear.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_linear.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png']).communicate()
+      Popen([PYTHON, EMCC, path_from_root('tests', 'sdl_fog_linear.c'), '-o', 'something.html', '--pre-js', 'reftest.js', '--preload-file', 'screenshot.png', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see an image with fog.', '/report_result?0')
 
     def test_worker(self):
@@ -10269,26 +10528,26 @@ elif 'browser' in str(sys.argv):
     def test_glgears(self):
       self.reftest(path_from_root('tests', 'gears.png'))
       Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_gles.c'), '-o', 'something.html',
-                                           '-DHAVE_BUILTIN_SINCOS', '--pre-js', 'reftest.js']).communicate()
+                                           '-DHAVE_BUILTIN_SINCOS', '--pre-js', 'reftest.js', '-s', 'GL_TESTING=1']).communicate()
       self.run_browser('something.html', 'You should see animating gears.', '/report_result?0')
 
     def test_glgears_animation(self):
       Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_gles.c'), '-o', 'something.html',
-                                           '-DHAVE_BUILTIN_SINCOS',
+                                           '-DHAVE_BUILTIN_SINCOS', '-s', 'GL_TESTING=1',
                                            '--shell-file', path_from_root('tests', 'hello_world_gles_shell.html')]).communicate()
       self.run_browser('something.html', 'You should see animating gears.', '/report_gl_result?true')
 
     def test_glgears_bad(self):
       # Make sure that OpenGL ES is not available if typed arrays are not used
       Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_gles.c'), '-o', 'something.html',
-                                           '-DHAVE_BUILTIN_SINCOS',
+                                           '-DHAVE_BUILTIN_SINCOS', '-s', 'GL_TESTING=1',
                                            '-s', 'USE_TYPED_ARRAYS=0',
                                            '--shell-file', path_from_root('tests', 'hello_world_gles_shell.html')]).communicate()
       self.run_browser('something.html', 'You should not see animating gears.', '/report_gl_result?false')
 
     def test_glgears_deriv(self):
       self.reftest(path_from_root('tests', 'gears.png'))
-      Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_gles_deriv.c'), '-o', 'something.html',
+      Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_gles_deriv.c'), '-o', 'something.html', '-s', 'GL_TESTING=1',
                                            '-DHAVE_BUILTIN_SINCOS', '--pre-js', 'reftest.js']).communicate()
       self.run_browser('something.html', 'You should see animating gears.', '/report_result?0')
       src = open('something.html').read()
@@ -10319,7 +10578,7 @@ elif 'browser' in str(sys.argv):
           args = ['--preload-file', 'smoke.tga', '-O2'] # test optimizations and closure here as well for more coverage
 
         self.reftest(book_path(basename.replace('.bc', '.png')))
-        Popen([PYTHON, EMCC, program, '-o', 'program.html', '--pre-js', 'reftest.js'] + args).communicate()
+        Popen([PYTHON, EMCC, program, '-o', 'program.html', '--pre-js', 'reftest.js', '-s', 'GL_TESTING=1'] + args).communicate()
         self.run_browser('program.html', '', '/report_result?0')
 
     def btest(self, filename, expected=None, reference=None, reference_slack=0, args=[]): # TODO: use in all other tests
@@ -10334,7 +10593,7 @@ elif 'browser' in str(sys.argv):
         expected = [str(i) for i in range(0, reference_slack+1)]
         shutil.copyfile(path_from_root('tests', filename), os.path.join(self.get_dir(), filename))
         self.reftest(path_from_root('tests', reference))
-        args = args + ['--pre-js', 'reftest.js']
+        args = args + ['--pre-js', 'reftest.js', '-s', 'GL_TESTING=1']
       Popen([PYTHON, EMCC, os.path.join(self.get_dir(), filename), '-o', 'test.html'] + args).communicate()
       if type(expected) is str: expected = [expected]
       self.run_browser('test.html', '.', ['/report_result?' + e for e in expected])
@@ -10369,6 +10628,19 @@ elif 'browser' in str(sys.argv):
       shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
       self.btest('gl_ps.c', reference='gl_ps.png', args=['--preload-file', 'screenshot.png'])
 
+    def test_gl_ps_packed(self):
+      # packed data that needs to be strided
+      shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
+      self.btest('gl_ps_packed.c', reference='gl_ps.png', args=['--preload-file', 'screenshot.png'])
+
+    def test_gl_ps_workaround(self):
+      shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
+      self.btest('gl_ps_workaround.c', reference='gl_ps.png', args=['--preload-file', 'screenshot.png'])
+
+    def test_gl_ps_workaround2(self):
+      shutil.copyfile(path_from_root('tests', 'screenshot.png'), os.path.join(self.get_dir(), 'screenshot.png'))
+      self.btest('gl_ps_workaround2.c', reference='gl_ps.png', args=['--preload-file', 'screenshot.png'])
+
     def test_matrix_identity(self):
       self.btest('gl_matrix_identity.c', expected=['-1882984448', '460451840'])
 
@@ -10417,6 +10689,15 @@ elif 'browser' in str(sys.argv):
     def test_cubegeom_fog(self):
       self.btest('cubegeom_fog.c', expected=['1617140399', '-898782526', '-946179526'])
 
+    def test_cubegeom_pre_vao(self):
+      self.btest('cubegeom_pre_vao.c', expected=['-1472804742', '-1626058463', '-2046234971'])
+
+    def test_cubegeom_pre2_vao(self):
+      self.btest('cubegeom_pre2_vao.c', expected=['-1472804742', '-1626058463', '-2046234971'])
+
+    def test_cubegeom_pre2_vao2(self):
+      self.btest('cubegeom_pre2_vao2.c', expected=['-790445118'])
+
     def test_cube_explosion(self):
       self.btest('cube_explosion.c', expected=['667220544', '-1543354600', '-1485258415'])
 
@@ -10499,14 +10780,17 @@ elif 'browser' in str(sys.argv):
     def test_subdata(self):
       self.btest('gl_subdata.cpp', reference='float_tex.png')
 
+    def test_perspective(self):
+      self.btest('perspective.c', reference='perspective.png')
+
     def test_runtimelink(self):
       main, supp = self.setup_runtimelink_test()
 
       open(self.in_dir('supp.cpp'), 'w').write(supp)
-      Popen([PYTHON, EMCC, self.in_dir('supp.cpp'), '-o', 'supp.js', '-s', 'LINKABLE=1', '-s', 'NAMED_GLOBALS=1', '-s', 'BUILD_AS_SHARED_LIB=2', '-O2', '--closure', '0']).communicate()
+      Popen([PYTHON, EMCC, self.in_dir('supp.cpp'), '-o', 'supp.js', '-s', 'LINKABLE=1', '-s', 'NAMED_GLOBALS=1', '-s', 'BUILD_AS_SHARED_LIB=2', '-O2']).communicate()
       shutil.move(self.in_dir('supp.js'), self.in_dir('supp.so'))
 
-      self.btest(main, args=['-s', 'LINKABLE=1', '-s', 'NAMED_GLOBALS=1', '-s', 'RUNTIME_LINKED_LIBS=["supp.so"]', '-DBROWSER=1', '-O2', '--closure', '0'], expected='76')
+      self.btest(main, args=['-s', 'LINKABLE=1', '-s', 'NAMED_GLOBALS=1', '-s', 'RUNTIME_LINKED_LIBS=["supp.so"]', '-DBROWSER=1', '-O2'], expected='76')
 
     def test_pre_run_deps(self):
       # Adding a dependency in preRun will delay run
@@ -10802,12 +11086,15 @@ elif 'benchmark' in str(sys.argv):
 
       try_delete(final_filename)
       output = Popen([PYTHON, EMCC, filename, #'-O3',
-                      '-O2', '-s', 'INLINING_LIMIT=0', '-s', 'DOUBLE_MODE=0', '-s', 'PRECISE_I64_MATH=0',
-                      '-s', 'ASM_JS=1',# '-s', 'USE_MATH_IMUL=1',
+                      '-O2', '-s', 'INLINING_LIMIT=0', '-s', 'DOUBLE_MODE=0', '-s', 'PRECISE_I64_MATH=0',# '-s', 'EXPLICIT_ZEXT=1',
+                      '-s', 'ASM_JS=1', '-s', 'USE_MATH_IMUL=1',
                       '-s', 'TOTAL_MEMORY=128*1024*1024', '-s', 'FAST_MEMORY=10*1024*1024',
                       '-o', final_filename] + shared_args + emcc_args, stdout=PIPE, stderr=self.stderr_redirect).communicate()
       assert os.path.exists(final_filename), 'Failed to compile file: ' + output[0]
 
+      if self.save_JS:
+        self.hardcode_arguments(final_filename, args)
+
       # Run JS
       global total_times, tests_done
       times = []
@@ -11060,7 +11347,7 @@ elif 'benchmark' in str(sys.argv):
     def test_dlmalloc(self):
       # XXX This seems to have regressed slightly with emcc. Are -g and the signs lines passed properly?
       src = open(path_from_root('system', 'lib', 'dlmalloc.c'), 'r').read() + '\n\n\n' + open(path_from_root('tests', 'dlmalloc_test.c'), 'r').read()
-      self.do_benchmark('dlmalloc', src, ['400', '3000'], '*3000,0*', emcc_args=['-g', '-s', 'CORRECT_SIGNS=2', '-s', 'CORRECT_SIGNS_LINES=[4820, 4195, 4250, 4203, 4209, 4239, 4231]'])
+      self.do_benchmark('dlmalloc', src, ['400', '3000'], '*3000,0*')
 
     def test_zlib(self):
       src = open(path_from_root('tests', 'zlib', 'benchmark.c'), 'r').read()
@@ -11220,12 +11507,12 @@ elif 'sanity' in str(sys.argv):
       f = open(CONFIG_FILE, 'a')
       f.write('CLOSURE_COMPILER = "/tmp/nowhere/nothingtoseehere/kjadsfkjwelkjsdfkqgas/nonexistent.txt"\n')
       f.close()
-      output = self.check_working([EMCC, '-O2', 'tests/hello_world.cpp'], CLOSURE_FATAL)
+      output = self.check_working([EMCC, '-O2', '--closure', '1', 'tests/hello_world.cpp'], CLOSURE_FATAL)
 
       # With a working path, all is well
       restore()
       try_delete('a.out.js')
-      output = self.check_working([EMCC, '-O2', 'tests/hello_world.cpp'], '')
+      output = self.check_working([EMCC, '-O2', '--closure', '1', 'tests/hello_world.cpp'], '')
       assert os.path.exists('a.out.js')
 
     def test_llvm(self):
@@ -11427,7 +11714,7 @@ fi
             try_delete(basebc_name) # we might need to check this file later
             try_delete(dcebc_name) # we might need to check this file later
             for ll_name in ll_names: try_delete(ll_name)
-            output = self.do([EMCC, '-O' + str(i), '--closure', '0', '-s', 'RELOOP=0', '--llvm-lto', '0', path_from_root('tests', filename)])
+            output = self.do([EMCC, '-O' + str(i), '-s', 'RELOOP=0', '--llvm-lto', '0', path_from_root('tests', filename)])
             #print output
             assert INCLUDING_MESSAGE.replace('X', libname) in output
             if libname == 'libc':
@@ -11475,7 +11762,7 @@ fi
           print >> sys.stderr, phase, i
           opt = min(i, 2)
           try_delete('a.out.js')
-          output = Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_loop.cpp'), '-O' + str(opt), '--closure', '0'],
+          output = Popen([PYTHON, EMCC, path_from_root('tests', 'hello_world_loop.cpp'), '-O' + str(opt)],
                          stdout=PIPE, stderr=PIPE).communicate()
           self.assertContained('hello, world!', run_js('a.out.js'))
           output = '\n'.join(output)
@@ -11530,12 +11817,12 @@ fi
           (['--jcache'], 'hello_libcxx.cpp', False, True, False, True, False, True, []),
           ([], 'hello_libcxx.cpp', False, False, False, False, False, False, []),
           # finally, build a file close to the previous, to see that some chunks are found in the cache and some not
-          (['--jcache'], 'hello_libcxx_mod1.cpp', False, True, True, True, True, False, []), # win on pre, mix on funcs, lose on jsfuncs
+          (['--jcache'], 'hello_libcxx_mod1.cpp', False, True, True, True, True, True, []), # win on pre, mix on funcs, mix on jsfuncs
           (['--jcache'], 'hello_libcxx_mod1.cpp', False, True, False, True, False, True, []),
         ]:
           print >> sys.stderr, args, input_file, expect_pre_save, expect_pre_load, expect_funcs_save, expect_funcs_load, expect_jsfuncs_save, expect_jsfuncs_load, expected
           self.clear()
-          out, err = Popen([PYTHON, EMCC, '-O2', '--closure', '0', path_from_root('tests', input_file)] + args, stdout=PIPE, stderr=PIPE).communicate()
+          out, err = Popen([PYTHON, EMCC, '-O2', path_from_root('tests', input_file)] + args, stdout=PIPE, stderr=PIPE).communicate()
           errtail = err.split('emcc invocation')[-1]
           self.assertContained('hello, world!', run_js('a.out.js'), errtail)
           assert (PRE_SAVE_MSG in err) == expect_pre_save, errtail
diff --git a/tests/s3tc_crunch.png b/tests/s3tc_crunch.png
index 3510b83b..067be7aa 100644
--- a/tests/s3tc_crunch.png
+++ b/tests/s3tc_crunch.png
diff --git a/tests/screenshot-fog-density.png b/tests/screenshot-fog-density.png
index cd1f6f1b..60f0c9ef 100644
--- a/tests/screenshot-fog-density.png
+++ b/tests/screenshot-fog-density.png
diff --git a/tests/screenshot-fog-exp2.png b/tests/screenshot-fog-exp2.png
index cd5e6a63..4737a536 100644
--- a/tests/screenshot-fog-exp2.png
+++ b/tests/screenshot-fog-exp2.png
diff --git a/tests/screenshot-fog-linear.png b/tests/screenshot-fog-linear.png
index 57534566..747c0c25 100644
--- a/tests/screenshot-fog-linear.png
+++ b/tests/screenshot-fog-linear.png
diff --git a/tests/screenshot-fog-negative.png b/tests/screenshot-fog-negative.png
index 5b18a201..747063c9 100644
--- a/tests/screenshot-fog-negative.png
+++ b/tests/screenshot-fog-negative.png
diff --git a/tests/screenshot-fog-simple.png b/tests/screenshot-fog-simple.png
index 527768fc..7624076c 100644
--- a/tests/screenshot-fog-simple.png
+++ b/tests/screenshot-fog-simple.png
diff --git a/tests/screenshot-gray-purple.png b/tests/screenshot-gray-purple.png
index 514b29a1..eaf08e6a 100644
--- a/tests/screenshot-gray-purple.png
+++ b/tests/screenshot-gray-purple.png
diff --git a/tests/screenshot-gray.png b/tests/screenshot-gray.png
index 16e45a7a..af06c256 100644
--- a/tests/screenshot-gray.png
+++ b/tests/screenshot-gray.png
diff --git a/tests/sdl_gl_read.c b/tests/sdl_gl_read.c
index 552eb8c0..d752f94d 100644
--- a/tests/sdl_gl_read.c
+++ b/tests/sdl_gl_read.c
@@ -6,6 +6,7 @@
 
 #include <stdio.h>
 #include <stdlib.h>
+#include <assert.h>
 
 GLuint programObject;
 int width = 512;
@@ -118,8 +119,14 @@ void Draw ()
 }
 
 void Verify() {
-  unsigned char *data = malloc(width*height*4);
+  unsigned char *data = malloc(width*height*4 + 16);
+  int *last = (int*)(data + width*height*4 - 4);
+  int *after = (int*)(data + width*height*4);
+  *last = 0xdeadbeef;
+  *after = 0x12345678;
   glReadPixels(0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, data);
+  assert(*last != 0xdeadbeef); // should overwrite the buffer to the end
+  assert(*after == 0x12345678); // nothing should be written afterwards!
   // Should see some blue, and nothing else
   int seen = 0;
   int ok = 1;
diff --git a/tests/sdl_image_jpeg.c b/tests/sdl_image_jpeg.c
new file mode 100644
index 00000000..10619dad
--- /dev/null
+++ b/tests/sdl_image_jpeg.c
@@ -0,0 +1,45 @@
+#include <stdio.h>
+#include <SDL/SDL.h>
+#include <SDL/SDL_image.h>
+#include <assert.h>
+#include <emscripten.h>
+
+int testImage(SDL_Surface* screen, const char* fileName) {
+  SDL_Surface *image = IMG_Load(fileName);
+  if (!image)
+  {
+     printf("IMG_Load: %s\n", IMG_GetError());
+     return 0;
+  }
+  assert(image->format->BitsPerPixel == 32);
+  assert(image->format->BytesPerPixel == 4);
+  assert(image->pitch == 4*image->w);
+  int result = image->w;
+
+  SDL_BlitSurface (image, NULL, screen, NULL);
+  SDL_FreeSurface (image);
+
+  return result;
+}
+
+int main() {
+  SDL_Init(SDL_INIT_VIDEO);
+  SDL_Surface *screen = SDL_SetVideoMode(600, 450, 32, SDL_SWSURFACE);
+
+  int result = 0;
+  result = testImage(screen, "screenshot.jpeg"); // relative path
+  assert(result != 0);
+  result |= testImage(screen, "/screenshot.jpeg"); // absolute path
+  assert(result != 0);
+
+  SDL_Flip(screen);
+
+  printf("you should see an image.\n");
+
+  SDL_Quit();
+
+  REPORT_RESULT();
+
+  return 0;
+}
+
diff --git a/tests/websockets.c b/tests/websockets.c
index 59acbd69..34aa44b4 100644
--- a/tests/websockets.c
+++ b/tests/websockets.c
@@ -82,7 +82,7 @@ void iter(void *arg) {
     printf("sum: %d\n", sum);
 
 #if EMSCRIPTEN
-    assert(not_always_data == 1);
+    //assert(not_always_data == 1);
 
     int result = sum;
     REPORT_RESULT();
diff --git a/tools/file_packager.py b/tools/file_packager.py
index 7e196efd..bfa8e2f0 100644
--- a/tools/file_packager.py
+++ b/tools/file_packager.py
@@ -228,7 +228,7 @@ if has_preloaded:
     curr = open(file_['localname'], 'rb').read()
     file_['data_end'] = start + len(curr)
     if AV_WORKAROUND: curr += '\x00'
-    print >> sys.stderr, 'bundling', file_['name'], file_['localname'], file_['data_start'], file_['data_end']
+    #print >> sys.stderr, 'bundling', file_['name'], file_['localname'], file_['data_start'], file_['data_end']
     start += len(curr)
     data.write(curr)
   data.close()
diff --git a/tools/js-optimizer.js b/tools/js-optimizer.js
index efbfa8aa..f2dc516a 100644
--- a/tools/js-optimizer.js
+++ b/tools/js-optimizer.js
@@ -1598,7 +1598,7 @@ function registerize(ast) {
             fun[2].push(reg);
           }
         }
-        getStatements(fun).unshift(['var', vars]);
+        if (vars.length > 0) getStatements(fun).unshift(['var', vars]);
       }
     } else {
       //printErr('unfake params: \n\n' + astToSrc(fun) + '\n\n');
diff --git a/tools/js_optimizer.py b/tools/js_optimizer.py
index f2e610d0..cbf64486 100644
--- a/tools/js_optimizer.py
+++ b/tools/js_optimizer.py
@@ -88,10 +88,11 @@ def run_on_js(filename, passes, js_engine, jcache):
     if i < len(parts)-1: func += '\n}\n' # last part needs no }
     m = func_sig.search(func)
     if m:
-      ident = m.group(1)
+      ident = m.group(2)
     else:
       if suffix: continue # ignore whitespace
       ident = 'anon_%d' % i
+    assert ident
     funcs.append((ident, func))
   parts = None
   total_size = len(js)
diff --git a/tools/shared.py b/tools/shared.py
index c283c1f5..ce9001fb 100644
--- a/tools/shared.py
+++ b/tools/shared.py
@@ -566,12 +566,10 @@ class Settings:
         if opt_level >= 2:
           Settings.RELOOP = 1
         if opt_level >= 3:
+          # Aside from these, -O3 also runs closure compiler
           Settings.INLINING_LIMIT = 0
           Settings.DOUBLE_MODE = 0
           Settings.PRECISE_I64_MATH = 0
-          Settings.CORRECT_SIGNS = 0
-          Settings.CORRECT_OVERFLOWS = 0
-          Settings.CORRECT_ROUNDINGS = 0
           if noisy: print >> sys.stderr, 'Warning: Applying some potentially unsafe optimizations! (Use -O2 if this fails.)'
 
     global Settings
@@ -665,6 +663,9 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)''' % { 'winfix': '' if not WINDOWS e
   def make(args, stdout=None, stderr=None, env=None):
     if env is None:
       env = Building.get_building_env()
+    if not args:
+      print >> sys.stderr, 'Error: Executable to run not specified.'
+      sys.exit(1)
     #args += ['VERBOSE=1']
     try:
       Popen(args, stdout=stdout, stderr=stderr, env=env).communicate()
@@ -748,12 +749,16 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)''' % { 'winfix': '' if not WINDOWS e
     resolved_symbols = set()
     temp_dirs = []
     files = map(os.path.abspath, files)
+    has_ar = False
+    for f in files:
+      has_ar = has_ar or Building.is_ar(f)
     for f in files:
       if not Building.is_ar(f):
         if Building.is_bitcode(f):
-          new_symbols = Building.llvm_nm(f)
-          resolved_symbols = resolved_symbols.union(new_symbols.defs)
-          unresolved_symbols = unresolved_symbols.union(new_symbols.undefs.difference(resolved_symbols)).difference(new_symbols.defs)
+          if has_ar:
+            new_symbols = Building.llvm_nm(f)
+            resolved_symbols = resolved_symbols.union(new_symbols.defs)
+            unresolved_symbols = unresolved_symbols.union(new_symbols.undefs.difference(resolved_symbols)).difference(new_symbols.defs)
           actual_files.append(f)
       else:
         # Extract object files from ar archives, and link according to gnu ld semantics
@@ -828,6 +833,7 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)''' % { 'winfix': '' if not WINDOWS e
   def llvm_opt(filename, opts):
     if type(opts) is int:
       opts = Building.pick_llvm_opts(opts)
+    #opts += ['-debug-pass=Arguments']
     if DEBUG: print >> sys.stderr, 'emcc: LLVM opts:', opts
     output = Popen([LLVM_OPT, filename] + opts + ['-o=' + filename + '.opt.bc'], stdout=PIPE).communicate()[0]
     assert os.path.exists(filename + '.opt.bc'), 'Failed to run llvm optimizations: ' + output
@@ -864,8 +870,14 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)''' % { 'winfix': '' if not WINDOWS e
     assert os.path.exists(output_filename), 'Could not create bc file: ' + output
     return output_filename
 
+  nm_cache = {} # cache results of nm - it can be slow to run
+
   @staticmethod
   def llvm_nm(filename, stdout=PIPE, stderr=None):
+    if filename in Building.nm_cache:
+      #if DEBUG: print >> sys.stderr, 'loading nm results for %s from cache' % filename
+      return Building.nm_cache[filename]
+
     # LLVM binary ==> list of symbols
     output = Popen([LLVM_NM, filename], stdout=stdout, stderr=stderr).communicate()[0]
     class ret:
@@ -886,6 +898,7 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)''' % { 'winfix': '' if not WINDOWS e
     ret.defs = set(ret.defs)
     ret.undefs = set(ret.undefs)
     ret.commons = set(ret.commons)
+    Building.nm_cache[filename] = ret
     return ret
 
   @staticmethod
@@ -1086,24 +1099,6 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)''' % { 'winfix': '' if not WINDOWS e
 
   @staticmethod
   def is_bitcode(filename):
-    # checks if a file contains LLVM bitcode
-    # if the file doesn't exist or doesn't have valid symbols, it isn't bitcode
-    try:
-      defs = Building.llvm_nm(filename, stderr=PIPE)
-      # If no symbols found, it might just be an empty bitcode file, try to dis it
-      if len(defs.defs) + len(defs.undefs) + len(defs.commons) == 0:
-        # llvm-nm 3.0 has a bug when reading symbols from ar files
-        # so try to see if we're dealing with an ar file, in which
-        # case we should try to dis it.
-        if not Building.is_ar(filename):
-          test_ll = os.path.join(EMSCRIPTEN_TEMP_DIR, 'test.ll')
-          Building.llvm_dis(filename, test_ll)
-          assert os.path.exists(test_ll)
-          try_delete(test_ll)
-    except Exception, e:
-      if DEBUG: print >> sys.stderr, 'shared.Building.is_bitcode failed to test whether file \'%s\' is a llvm bitcode file! Failed on exception: %s' % (filename, e)
-      return False
-
     # look for magic signature
     b = open(filename, 'r').read(4)
     if b[0] == 'B' and b[1] == 'C':
@@ -1267,6 +1262,7 @@ class JCache:
       if os.path.exists(chunking_file):
         try:
           previous_mapping = cPickle.Unpickler(open(chunking_file, 'rb')).load() # maps a function identifier to the chunk number it will be in
+          #if DEBUG: print >> sys.stderr, 'jscache previous mapping', previous_mapping
         except:
           pass
     chunks = []
@@ -1275,20 +1271,25 @@ class JCache:
       news = []
       for func in funcs:
         ident, data = func
+        assert ident, 'need names for jcache chunking'
         if not ident in previous_mapping:
           news.append(func)
         else:
           n = previous_mapping[ident]
           while n >= len(chunks): chunks.append([])
           chunks[n].append(func)
+      if DEBUG: print >> sys.stderr, 'jscache not in previous chunking', len(news)
       # add news and adjust for new sizes
       spilled = news
-      for chunk in chunks:
+      for i in range(len(chunks)):
+        chunk = chunks[i]
         size = sum([len(func[1]) for func in chunk])
-        while size > 1.5*chunk_size and len(chunk) > 0:
+        #if DEBUG: print >> sys.stderr, 'need spilling?', i, size, len(chunk), 'vs', chunk_size, 1.5*chunk_size
+        while size > 1.5*chunk_size and len(chunk) > 1:
           spill = chunk.pop()
           spilled.append(spill)
           size -= len(spill[1])
+      #if DEBUG: print >> sys.stderr, 'jscache new + spilled', len(spilled)
       for chunk in chunks:
         size = sum([len(func[1]) for func in chunk])
         while size < 0.66*chunk_size and len(spilled) > 0:
@@ -1297,6 +1298,7 @@ class JCache:
           size += len(spill[1])
       chunks = filter(lambda chunk: len(chunk) > 0, chunks) # might have empty ones, eliminate them
       funcs = spilled # we will allocate these into chunks as if they were normal inputs
+      #if DEBUG: print >> sys.stderr, 'leftover spills', len(spilled)
     # initialize reasonably, the rest of the funcs we need to split out
     curr = []
     total_size = 0
@@ -1322,15 +1324,18 @@ class JCache:
       for i in range(len(chunks)):
         chunk = chunks[i]
         for ident, data in chunk:
+          assert ident not in new_mapping, 'cannot have duplicate names in jcache chunking'
           new_mapping[ident] = i
       cPickle.Pickler(open(chunking_file, 'wb')).dump(new_mapping)
       #if DEBUG:
+      #  for i in range(len(chunks)):
+      #    chunk = chunks[i]
+      #    print >> sys.stderr, 'final chunk', i, len(chunk)
+      #  print >> sys.stderr, 'new mapping:', new_mapping
       #  if previous_mapping:
       #    for ident in set(previous_mapping.keys() + new_mapping.keys()):
       #      if previous_mapping.get(ident) != new_mapping.get(ident):
       #        print >> sys.stderr, 'mapping inconsistency', ident, previous_mapping.get(ident), new_mapping.get(ident)
-      #  for key, value in new_mapping.iteritems():
-      #    print >> sys.stderr, 'mapping:', key, value
     return [''.join([func[1] for func in chunk]) for chunk in chunks] # remove function names
 
 class JS:
diff --git a/tools/test-js-optimizer-regs-output.js b/tools/test-js-optimizer-regs-output.js
index 36006b7c..fe7de5fb 100644
--- a/tools/test-js-optimizer-regs-output.js
+++ b/tools/test-js-optimizer-regs-output.js
@@ -225,4 +225,8 @@ function switchey(r1) {
   r9 = r1 + 2;
   pp(r9);
 }
+function __ZN14NetworkAddressC1EPKcti(r1) {
+  __ZN14NetworkAddressC2EPKcti(r1);
+  return;
+}
 
diff --git a/tools/test-js-optimizer-regs.js b/tools/test-js-optimizer-regs.js
index 4802afa3..3013e518 100644
--- a/tools/test-js-optimizer-regs.js
+++ b/tools/test-js-optimizer-regs.js
@@ -230,4 +230,8 @@ function switchey(x) {
   var aaa = x+2;
   pp(aaa);
 }
-// EMSCRIPTEN_GENERATED_FUNCTIONS: ["test", "primes", "atomic", "fcntl_open", "ex", "switchey"]
+function __ZN14NetworkAddressC1EPKcti($this) {
+  __ZN14NetworkAddressC2EPKcti($this);
+  return;
+}
+// EMSCRIPTEN_GENERATED_FUNCTIONS: ["test", "primes", "atomic", "fcntl_open", "ex", "switchey", "__ZN14NetworkAddressC1EPKcti"]