diff --git a/dust3d.pro b/dust3d.pro
index 700144fc..330a5644 100644
--- a/dust3d.pro
+++ b/dust3d.pro
@@ -274,6 +274,8 @@ HEADERS += thirdparty/thekla_atlas/extern/poshlib/posh.h
 SOURCES += thirdparty/thekla_atlas/src/thekla/thekla_atlas.cpp
 HEADERS += thirdparty/thekla_atlas/src/thekla/thekla_atlas.h
 
+HEADERS += thirdparty/thekla_atlas/src/nvcore/nvcore.h
+
 HEADERS += thirdparty/thekla_atlas/src/nvcore/Stream.h
 
 SOURCES += thirdparty/thekla_atlas/src/nvcore/Debug.cpp
diff --git a/thirdparty/thekla_atlas/CMakeLists.txt b/thirdparty/thekla_atlas/CMakeLists.txt
index b74d29b5..3f0bd99b 100755
--- a/thirdparty/thekla_atlas/CMakeLists.txt
+++ b/thirdparty/thekla_atlas/CMakeLists.txt
@@ -23,3 +23,32 @@ add_executable(
 target_link_libraries(
     thekla_atlas_test
     nvmesh)
+
+set(CMAKE_DEBUG_POSTFIX d)
+
+set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
+set(CMAKE_SHARED_LINKER_FLAGS "-Wl,--export-all-symbols")
+
+add_library(libthekla_atlas
+	src/thekla/thekla_atlas.cpp
+)
+
+install(TARGETS libthekla_atlas DESTINATION lib/static)
+
+get_property(HEADER_DIRS DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY INCLUDE_DIRECTORIES)
+
+configure_file(theklaAtlasConfig.cmake.in ${CMAKE_CURRENT_BINARY_DIR}/theklaAtlasConfig.cmake @ONLY)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/theklaAtlasConfig.cmake DESTINATION .)
+
+file(GLOB HEADERS src/*.h)
+install(FILES ${HEADERS} DESTINATION include)
+file(GLOB HEADERS src/thekla/*.h)
+install(FILES ${HEADERS} DESTINATION include/thekla)
+file(GLOB HEADERS src/nvcore/*.h)
+install(FILES ${HEADERS} DESTINATION include/nvcore)
+file(GLOB HEADERS src/nvimage/*.h)
+install(FILES ${HEADERS} DESTINATION include/nvimage)
+file(GLOB HEADERS src/nvmesh/*.h)
+install(FILES ${HEADERS} DESTINATION include/nvmesh)
+file(GLOB HEADERS src/nvmath/*.h)
+install(FILES ${HEADERS} DESTINATION include/nvmath)
diff --git a/thirdparty/thekla_atlas/src/nvcore/Array.h b/thirdparty/thekla_atlas/src/nvcore/Array.h
index d3fab26a..b295cb2b 100755
--- a/thirdparty/thekla_atlas/src/nvcore/Array.h
+++ b/thirdparty/thekla_atlas/src/nvcore/Array.h
@@ -179,6 +179,4 @@ namespace nv
 
 } // nv namespace
 
-#include "Array.inl"
-
 #endif // NV_CORE_ARRAY_H
diff --git a/thirdparty/thekla_atlas/src/nvcore/Array.inl b/thirdparty/thekla_atlas/src/nvcore/Array.inl
index 0b4de28b..67c00296 100755
--- a/thirdparty/thekla_atlas/src/nvcore/Array.inl
+++ b/thirdparty/thekla_atlas/src/nvcore/Array.inl
@@ -12,6 +12,9 @@
 #include <string.h>	// memmove
 #include <new> // for placement new
 
+/////// Changes in Dust3D Begin /////////////
+#include "Vector.inl"
+/////// Changes in Dust3D End /////////////
 
 
 namespace nv 
diff --git a/thirdparty/thekla_atlas/src/nvcore/BitArray.h b/thirdparty/thekla_atlas/src/nvcore/BitArray.h
index 23cf8806..80626973 100755
--- a/thirdparty/thekla_atlas/src/nvcore/BitArray.h
+++ b/thirdparty/thekla_atlas/src/nvcore/BitArray.h
@@ -40,7 +40,7 @@ namespace nv
     {
     public:
 
-        BitArray() {}
+        BitArray() : m_size(0){}
         BitArray(uint sz) {
             resize(sz);
         }
@@ -237,7 +237,7 @@ namespace nv
     private:
 
         // Number of bits stored.
-        uint m_size;
+        uint m_size = 0;
 
         // Array of bits.
         Array<uint> m_wordArray;
diff --git a/thirdparty/thekla_atlas/src/nvcore/CMakeLists.txt b/thirdparty/thekla_atlas/src/nvcore/CMakeLists.txt
index a84a9e99..879be2d9 100755
--- a/thirdparty/thekla_atlas/src/nvcore/CMakeLists.txt
+++ b/thirdparty/thekla_atlas/src/nvcore/CMakeLists.txt
@@ -1,4 +1,5 @@
 PROJECT(nvcore)
+set(CMAKE_DEBUG_POSTFIX d)
 
 SET(CORE_SRCS
 	nvcore.h
diff --git a/thirdparty/thekla_atlas/src/nvcore/Debug.cpp b/thirdparty/thekla_atlas/src/nvcore/Debug.cpp
index 75ac6beb..8edb9ec7 100755
--- a/thirdparty/thekla_atlas/src/nvcore/Debug.cpp
+++ b/thirdparty/thekla_atlas/src/nvcore/Debug.cpp
@@ -106,7 +106,8 @@ namespace
     struct sigaction s_old_sigtrap;
     struct sigaction s_old_sigfpe;
     struct sigaction s_old_sigbus;
-
+    struct sigaction s_old_sigill;
+    
 #endif
 
 
@@ -431,9 +432,9 @@ namespace
             pSymbol->SizeOfStruct = sizeof(IMAGEHLP_SYMBOL64);
             pSymbol->MaxNameLength = MAX_STRING_LEN;
 
-            DWORD64 dwDisplacement;
+            DWORD64 dwDisplacement64;
             
-            if (SymGetSymFromAddr64(hProcess, ip, &dwDisplacement, pSymbol))
+            if (SymGetSymFromAddr64(hProcess, ip, &dwDisplacement64, pSymbol))
             {
                 pSymbol->Name[MAX_STRING_LEN-1] = 0;
                 
@@ -455,8 +456,8 @@ namespace
                 IMAGEHLP_LINE64 theLine = { 0 };
                 theLine.SizeOfStruct = sizeof(theLine);
 
-                DWORD dwDisplacement;
-                if (!SymGetLineFromAddr64(hProcess, ip, &dwDisplacement, &theLine))
+                DWORD dwDisplacement32;
+                if (!SymGetLineFromAddr64(hProcess, ip, &dwDisplacement32, &theLine))
                 {
                     // Do not print unknown symbols anymore.
                     //break;
@@ -679,6 +680,9 @@ namespace
 #    elif NV_CPU_ARM
         ucontext_t * ucp = (ucontext_t *)secret;
         return (void *) ucp->uc_mcontext->__ss.__pc;
+#    elif NV_CPU_ARM_64
+        ucontext_t * ucp = (ucontext_t *)secret;
+        return (void *) ucp->uc_mcontext->__ss.__pc;
 #    else
 #      error "Unknown CPU"
 #    endif
@@ -784,8 +788,8 @@ namespace
             printStackTrace(trace, size, 1);
         }
 #endif // defined(NV_HAVE_EXECINFO_H)
-
-        exit(0);
+    
+        exit(EXIT_FAILURE + 2);
     }
 
 #endif // defined(NV_HAVE_SIGNAL_H)
@@ -933,15 +937,24 @@ namespace
         // Assert handler method.
         virtual int assertion(const char * exp, const char * file, int line, const char * func, const char * msg, va_list arg)
         {
-            int ret = NV_ABORT_EXIT;            
+            int ret = NV_ABORT_IGNORE;
             
-            if( func != NULL ) {
-                nvDebug( "*** Assertion failed: %s\n    On file: %s\n    On function: %s\n    On line: %d\n ", exp, file, func, line );
+            StringBuilder error_string;
+            error_string.format("*** Assertion failed: %s\n    On file: %s\n    On line: %d\n", exp, file, line );
+            if (func != NULL) {
+                error_string.appendFormat("    On function: %s\n", func);
             }
-            else {
-                nvDebug( "*** Assertion failed: %s\n    On file: %s\n    On line: %d\n ", exp, file, line );
+            if (msg != NULL) {
+                error_string.append("    Message: ");
+                va_list tmp;
+                va_copy(tmp, arg);
+                error_string.appendFormatList(msg, tmp);
+                va_end(tmp);
+                error_string.append("\n");
             }
-
+            nvDebug("%s", error_string.str());
+            
+            
 #if _DEBUG
             if (debug::isDebuggerPresent()) {
                 return NV_ABORT_DEBUG;
@@ -1173,7 +1186,7 @@ void debug::enableSigHandler(bool interactive)
     s_interactive = interactive;
 
 #if (NV_OS_WIN32 && NV_CC_MSVC) || NV_OS_DURANGO
-    if (interactive) {
+    if (!interactive) {
 #if NV_OS_WIN32
         // Do not display message boxes on error.
         // http://msdn.microsoft.com/en-us/library/windows/desktop/ms680621(v=vs.85).aspx
@@ -1224,6 +1237,8 @@ void debug::enableSigHandler(bool interactive)
     sigaction(SIGTRAP, &sa, &s_old_sigtrap);
     sigaction(SIGFPE, &sa, &s_old_sigfpe);
     sigaction(SIGBUS, &sa, &s_old_sigbus);
+    sigaction(SIGILL, &sa, &s_old_sigill);
+    
 
 #endif
 }
@@ -1249,7 +1264,8 @@ void debug::disableSigHandler()
     sigaction(SIGTRAP, &s_old_sigtrap, NULL);
     sigaction(SIGFPE, &s_old_sigfpe, NULL);
     sigaction(SIGBUS, &s_old_sigbus, NULL);
-
+    sigaction(SIGILL, &s_old_sigill, NULL);
+    
 #endif
 }
 
diff --git a/thirdparty/thekla_atlas/src/nvcore/Debug.h b/thirdparty/thekla_atlas/src/nvcore/Debug.h
index 3804ed47..98a32fd4 100755
--- a/thirdparty/thekla_atlas/src/nvcore/Debug.h
+++ b/thirdparty/thekla_atlas/src/nvcore/Debug.h
@@ -12,6 +12,8 @@
 #   ifdef __APPLE__
 #       include <TargetConditionals.h>
 #       include <signal.h>
+#       include <sys/types.h>   // getpid
+#       include <unistd.h>
 #   endif
 #endif
 
@@ -43,10 +45,22 @@
 //#       define nvDebugBreak()        __asm { int 3 }
 #   elif NV_OS_ORBIS
 #       define nvDebugBreak()       __debugbreak()
-#   elif NV_OS_IOS && TARGET_OS_IPHONE
-#       define nvDebugBreak()       raise(SIGINT)
 #   elif NV_CC_CLANG
-#       define nvDebugBreak()       __builtin_debugtrap()
+#       if NV_OS_IOS
+#           if NV_CPU_ARM_64
+#               if __clang_major__ >= 9
+#                   define nvDebugBreak()     asm volatile("svc #0x80") // IC: "svc 0" doesn't seem to work anymore.
+#               else
+#                   define nvDebugBreak()     asm volatile("svc 0")
+#               endif
+#           elif NV_CPU_ARM
+#               define nvDebugBreak()     asm("trap")
+#           else // simulator?
+#               define nvDebugBreak()     __builtin_debugtrap()
+#           endif
+#       else
+#           define nvDebugBreak()       __builtin_debugtrap()
+#       endif
 #   elif NV_CC_GNUC
 //#       define nvDebugBreak()       __builtin_debugtrap()     // Does GCC have debugtrap?
 #       define nvDebugBreak()		__builtin_trap()
@@ -98,12 +112,14 @@
     } \
     NV_MULTI_LINE_MACRO_END
 
+// IC: When using this macro in llvm/ios you can ignore future hits typing 'expr ignoreAll=true' in the console.
 // GCC, LLVM need "##" before the __VA_ARGS__, MSVC doesn't care
 #define nvAssertMacroWithIgnoreAll(exp,...) \
     NV_MULTI_LINE_MACRO_BEGIN \
         static bool ignoreAll = false; \
         if (!ignoreAll && !nvExpect(exp)) { \
             int _result = nvAbort(#exp, __FILE__, __LINE__, __FUNC__, ##__VA_ARGS__); \
+            /*nvDebug("Type 'expr ignoreAll=true' to ignore future hits\n");*/ \
             if (_result == NV_ABORT_DEBUG) { \
                 nvDebugBreak(); \
             } else if (_result == NV_ABORT_IGNORE) { \
@@ -124,7 +140,7 @@
 #define nvCheckMacro(exp) \
     (\
         (exp) ? true : ( \
-            (nvAbort(#exp, __FILE__, __LINE__, __FUNC__) == NV_ABORT_DEBUG) ? (nvDebugBreak(), true) : ( false ) \
+                        (nvAbort(#exp, __FILE__, __LINE__, __FUNC__) == NV_ABORT_DEBUG) ? ([](){nvDebugBreak();}(), true) : ( false ) \
         ) \
     )
 
@@ -199,11 +215,15 @@ namespace nv
         if (ptr == NULL) return true;
         if (reinterpret_cast<uint64>(ptr) < 0x10000ULL) return false;
         if (reinterpret_cast<uint64>(ptr) >= 0x000007FFFFFEFFFFULL) return false;
+    #elif NV_CPU_ARM_64
+        if (ptr == NULL) return true;
+        if (reinterpret_cast<uint64>(ptr) < 0x10000ULL) return false;
+        if (reinterpret_cast<uint64>(ptr) >= 0x000007FFFFFEFFFFULL) return false;
     #else
-	    if (reinterpret_cast<uintptr_t>(ptr) == 0xcccccccc) return false;
-	    if (reinterpret_cast<uintptr_t>(ptr) == 0xcdcdcdcd) return false;
-	    if (reinterpret_cast<uintptr_t>(ptr) == 0xdddddddd) return false;
-	    if (reinterpret_cast<uintptr_t>(ptr) == 0xffffffff) return false;
+        if (reinterpret_cast<uint32>(ptr) == 0xcccccccc) return false;
+	    if (reinterpret_cast<uint32>(ptr) == 0xcdcdcdcd) return false;
+	    if (reinterpret_cast<uint32>(ptr) == 0xdddddddd) return false;
+	    if (reinterpret_cast<uint32>(ptr) == 0xffffffff) return false;
     #endif
         return true;
     }
diff --git a/thirdparty/thekla_atlas/src/nvcore/DefsGnucDarwin.h b/thirdparty/thekla_atlas/src/nvcore/DefsGnucDarwin.h
index afb21c3d..14974878 100755
--- a/thirdparty/thekla_atlas/src/nvcore/DefsGnucDarwin.h
+++ b/thirdparty/thekla_atlas/src/nvcore/DefsGnucDarwin.h
@@ -27,8 +27,8 @@
 #define NV_FASTCALL		__attribute__((fastcall))
 #define NV_FORCEINLINE	__attribute__((always_inline)) inline
 #define NV_DEPRECATED   __attribute__((deprecated))
-#if NV_OS_IOS
-#define NV_THREAD_LOCAL // @@ IC: Looks like iOS does not have support for TLS declarations.
+#if 0 //Apple finally added TLS support to iOS!// NV_OS_IOS
+#define NV_THREAD_LOCAL
 #else
 #define NV_THREAD_LOCAL __thread
 #endif
diff --git a/thirdparty/thekla_atlas/src/nvcore/FileSystem.cpp b/thirdparty/thekla_atlas/src/nvcore/FileSystem.cpp
index 5ed0ca07..881ade25 100755
--- a/thirdparty/thekla_atlas/src/nvcore/FileSystem.cpp
+++ b/thirdparty/thekla_atlas/src/nvcore/FileSystem.cpp
@@ -73,3 +73,31 @@ bool FileSystem::removeFile(const char * path)
     // @@ Use unlink or remove?
     return remove(path) == 0;
 }
+
+
+#include "StdStream.h" // for fileOpen
+
+bool FileSystem::copyFile(const char * src, const char * dst) {
+
+    FILE * fsrc = fileOpen(src, "rb");
+    if (fsrc == NULL) return false;
+    NV_ON_RETURN(fclose(fsrc));
+
+    FILE * fdst = fileOpen(dst, "wb");
+    if (fdst == NULL) return false;
+    NV_ON_RETURN(fclose(fdst));
+    
+    char buffer[1024];
+    size_t n;
+
+    while ((n = fread(buffer, sizeof(char), sizeof(buffer), fsrc)) > 0) {
+        if (fwrite(buffer, sizeof(char), n, fdst) != n) {
+            return false;
+        }
+    }
+    
+    return true;
+}
+
+
+
diff --git a/thirdparty/thekla_atlas/src/nvcore/FileSystem.h b/thirdparty/thekla_atlas/src/nvcore/FileSystem.h
index afd0f449..30661c2b 100755
--- a/thirdparty/thekla_atlas/src/nvcore/FileSystem.h
+++ b/thirdparty/thekla_atlas/src/nvcore/FileSystem.h
@@ -15,7 +15,7 @@ namespace nv
         NVCORE_API bool createDirectory(const char * path);
         NVCORE_API bool changeDirectory(const char * path);
         NVCORE_API bool removeFile(const char * path);
-
+        NVCORE_API bool copyFile(const char * src, const char * dst);
     } // FileSystem namespace
 
 } // nv namespace
diff --git a/thirdparty/thekla_atlas/src/nvcore/Memory.cpp b/thirdparty/thekla_atlas/src/nvcore/Memory.cpp
index 03b05dfc..fb4a04f5 100755
--- a/thirdparty/thekla_atlas/src/nvcore/Memory.cpp
+++ b/thirdparty/thekla_atlas/src/nvcore/Memory.cpp
@@ -5,10 +5,12 @@
 #include "Utils.h"
 
 #include <stdlib.h>
+/////// Changes in Dust3D Begin /////////////
 #if defined(__APPLE__)
 #else
 #include <malloc.h>
 #endif
+/////// Changes in Dust3D End /////////////
 
 #define USE_EFENCE 0
 
diff --git a/thirdparty/thekla_atlas/src/nvcore/StrLib.cpp b/thirdparty/thekla_atlas/src/nvcore/StrLib.cpp
index 7ec6c701..8896a543 100755
--- a/thirdparty/thekla_atlas/src/nvcore/StrLib.cpp
+++ b/thirdparty/thekla_atlas/src/nvcore/StrLib.cpp
@@ -526,6 +526,28 @@ StringBuilder & StringBuilder::copy( const StringBuilder & s )
     return *this;
 }
 
+void StringBuilder::removeChar(char c)
+{
+    char * src = strchr(m_str, c);
+    if (src) {
+        char * dst = src;
+        src++;
+        while (*src) {
+            *dst++ = *src++;
+        }
+        *dst = '\0';
+    }
+}
+
+void StringBuilder::replaceChars(char c, char x)
+{
+    char * src = m_str;
+    while (*src) {
+        if (*src == c) *src = x;
+        src++;
+    }
+}
+
 bool StringBuilder::endsWith(const char * str) const
 {
     uint l = uint(strlen(str));
@@ -540,7 +562,7 @@ bool StringBuilder::beginsWith(const char * str) const
     return strncmp(m_str, str, l) == 0;
 }
 
-// Find given char starting from the end.
+// Find given char starting from the end. Why not use strrchr!?
 char * StringBuilder::reverseFind(char c)
 {
     int length = (int)strlen(m_str) - 1;
diff --git a/thirdparty/thekla_atlas/src/nvcore/StrLib.h b/thirdparty/thekla_atlas/src/nvcore/StrLib.h
index ae4b5d12..9634d575 100755
--- a/thirdparty/thekla_atlas/src/nvcore/StrLib.h
+++ b/thirdparty/thekla_atlas/src/nvcore/StrLib.h
@@ -124,6 +124,9 @@ namespace nv
 
         StringBuilder & toLower();
         StringBuilder & toUpper();
+        
+        void removeChar(char c);
+        void replaceChars(char c, char x);
 
         bool endsWith(const char * str) const;
         bool beginsWith(const char * str) const;
@@ -131,13 +134,13 @@ namespace nv
         char * reverseFind(char c);
 
         void reset();
-        bool isNull() const { return m_size == 0; }
+        NV_FORCEINLINE bool isNull() const { return m_size == 0; }
 
         // const char * accessors
         //operator const char * () const { return m_str; }
         //operator char * () { return m_str; }
-        const char * str() const { return m_str; }
-        char * str() { return m_str; }
+        NV_FORCEINLINE const char * str() const { return m_str; }
+        NV_FORCEINLINE char * str() { return m_str; }
 
         char * release();       // Release ownership of string.
         void acquire(char *);   // Take ownership of string.
@@ -283,25 +286,25 @@ namespace nv
         /// Equal operator.
         bool operator==( const String & str ) const
         {
-            return strMatch(str.data, data);
+            return strEqual(str.data, data);
         }
 
         /// Equal operator.
         bool operator==( const char * str ) const
         {
-            return strMatch(str, data);
+            return strEqual(str, data);
         }
 
         /// Not equal operator.
         bool operator!=( const String & str ) const
         {
-            return !strMatch(str.data, data);
+            return !strEqual(str.data, data);
         }
 
         /// Not equal operator.
         bool operator!=( const char * str ) const
         {
-            return !strMatch(str, data);
+            return !strEqual(str, data);
         }
 
         /// Returns true if this string is the null string.
diff --git a/thirdparty/thekla_atlas/src/nvcore/Utils.h b/thirdparty/thekla_atlas/src/nvcore/Utils.h
index f20e42cd..44e17e63 100755
--- a/thirdparty/thekla_atlas/src/nvcore/Utils.h
+++ b/thirdparty/thekla_atlas/src/nvcore/Utils.h
@@ -72,6 +72,11 @@ namespace nv
     //template <> inline uint32 U32<uint8>(uint8 x) { return x; }
     template <> inline uint32 U32<int8>(int8 x) { nvDebugCheck(x >= 0); return (uint32)x; }
 
+#if NV_OS_DARWIN
+    template <> inline uint32 U32<unsigned long>(unsigned long x) { nvDebugCheck(x <= NV_UINT32_MAX); return (uint32)x; }
+    template <> inline uint32 U32<long>(long x) { nvDebugCheck(x >= 0 && x <= NV_UINT32_MAX); return (uint32)x; }
+#endif
+
     // int32 casts:
     template <typename T> inline int32 I32(T x) { return x; }
     template <> inline int32 I32<uint64>(uint64 x) { nvDebugCheck(x <= NV_INT32_MAX); return (int32)x; }
@@ -120,11 +125,11 @@ namespace nv
     // int8 casts:
     template <typename T> inline int8 I8(T x) { return x; }
     template <> inline int8 I8<uint64>(uint64 x) { nvDebugCheck(x <= NV_INT8_MAX); return (int8)x; }
-    template <> inline int8 I8<int64>(int64 x) { nvDebugCheck(x >= NV_INT8_MIN && x <= NV_UINT8_MAX); return (int8)x; }
+    template <> inline int8 I8<int64>(int64 x) { nvDebugCheck(x >= NV_INT8_MIN && x <= NV_INT8_MAX); return (int8)x; }
     template <> inline int8 I8<uint32>(uint32 x) { nvDebugCheck(x <= NV_INT8_MAX); return (int8)x; }
-    template <> inline int8 I8<int32>(int32 x) { nvDebugCheck(x >= NV_INT8_MIN && x <= NV_UINT8_MAX); return (int8)x; }
+    template <> inline int8 I8<int32>(int32 x) { nvDebugCheck(x >= NV_INT8_MIN && x <= NV_INT8_MAX); return (int8)x; }
     template <> inline int8 I8<uint16>(uint16 x) { nvDebugCheck(x <= NV_INT8_MAX); return (int8)x; }
-    template <> inline int8 I8<int16>(int16 x) { nvDebugCheck(x >= NV_INT8_MIN && x <= NV_UINT8_MAX); return (int8)x; }
+    template <> inline int8 I8<int16>(int16 x) { nvDebugCheck(x >= NV_INT8_MIN && x <= NV_INT8_MAX); return (int8)x; }
     template <> inline int8 I8<uint8>(uint8 x) { nvDebugCheck(x <= NV_INT8_MAX); return (int8)x; }
     //template <> inline int8 I8<int8>(int8 x) { return x; }
 
diff --git a/thirdparty/thekla_atlas/src/nvcore/nvcore.h b/thirdparty/thekla_atlas/src/nvcore/nvcore.h
index a3deb66b..9af7ddd6 100755
--- a/thirdparty/thekla_atlas/src/nvcore/nvcore.h
+++ b/thirdparty/thekla_atlas/src/nvcore/nvcore.h
@@ -84,9 +84,9 @@
 
 // Threading:
 // some platforms don't implement __thread or similar for thread-local-storage
-#if NV_OS_UNIX || NV_OS_ORBIS || NV_OS_IOS //ACStodoIOS darwin instead of ios?
+#if NV_OS_UNIX || NV_OS_ORBIS || NV_OS_IOS
 #   define NV_OS_USE_PTHREAD 1
-#   if NV_OS_IOS
+#   if 0  //Apple finally added TLS support to iOS!// NV_OS_IOS
 #       define NV_OS_HAS_TLS_QUALIFIER 0
 #   else
 #       define NV_OS_HAS_TLS_QUALIFIER 1
@@ -114,6 +114,8 @@
 #   define NV_CPU_PPC 1
 #elif defined POSH_CPU_STRONGARM
 #   define NV_CPU_ARM 1
+#elif defined POSH_CPU_AARCH64
+#   define NV_CPU_ARM_64 1
 #else
 #   error "Unsupported CPU"
 #endif
@@ -151,10 +153,16 @@
 #endif
 
 // Endiannes:
-#define NV_LITTLE_ENDIAN    POSH_LITTLE_ENDIAN
-#define NV_BIG_ENDIAN       POSH_BIG_ENDIAN
-#define NV_ENDIAN_STRING    POSH_ENDIAN_STRING
-
+// @@ POSH endian detection is broken for arm64 on iOS. They are bi-endian and iOS sets all their processors to little endian by default.
+#if NV_OS_IOS
+#   define NV_LITTLE_ENDIAN    1
+#   define NV_BIG_ENDIAN       0
+#   define NV_ENDIAN_STRING    "little"
+#else
+#   define NV_LITTLE_ENDIAN    POSH_LITTLE_ENDIAN
+#   define NV_BIG_ENDIAN       POSH_BIG_ENDIAN
+#   define NV_ENDIAN_STRING    POSH_ENDIAN_STRING
+#endif
 
 // Define the right printf prefix for size_t arguments:
 #if POSH_64BIT_POINTER
@@ -311,6 +319,24 @@ template <typename T, size_t N> char (&ArraySizeHelper(T (&array)[N]))[N];
         NV_STRING_JOIN3(AtStartup_, __LINE__, Instance); \
     }
 
+namespace nv {
+    template <typename F>
+    struct ScopeExit {
+        ScopeExit(F f) : f(f) {}
+        ~ScopeExit() { f(); }
+        F f;
+    };
+
+    template <typename F>
+    ScopeExit<F> MakeScopeExit(F f) {
+        return ScopeExit<F>(f);
+    };
+}
+
+#define NV_ON_RETURN(code) \
+    auto NV_STRING_JOIN2(scope_exit_, __LINE__) = nv::MakeScopeExit([=](){code;})
+
+
 // Indicate the compiler that the parameter is not used to suppress compier warnings.
 #if NV_CC_MSVC
 #define NV_UNUSED(a) ((a)=(a))
@@ -318,6 +344,14 @@ template <typename T, size_t N> char (&ArraySizeHelper(T (&array)[N]))[N];
 #define NV_UNUSED(a) _Pragma(NV_STRING(unused(a)))
 #endif
 
+#if NV_CC_GNUC || NV_CC_CLANG
+#define NV_LIKELY(x) __builtin_expect(!!(x), 1)
+#define NV_UNLIKELY(x) __builtin_expect(!!(x), 0)
+#else
+#define NV_LIKELY(x) x
+#define NV_UNLIKELY(x) x
+#endif
+
 // Null index. @@ Move this somewhere else... it's only used by nvmesh.
 //const unsigned int NIL = unsigned int(~0);
 #define NIL uint(~0)
diff --git a/thirdparty/thekla_atlas/src/nvimage/CMakeLists.txt b/thirdparty/thekla_atlas/src/nvimage/CMakeLists.txt
index 5bc3cf43..c368af24 100755
--- a/thirdparty/thekla_atlas/src/nvimage/CMakeLists.txt
+++ b/thirdparty/thekla_atlas/src/nvimage/CMakeLists.txt
@@ -1,4 +1,5 @@
 PROJECT(nvimage)
+set(CMAKE_DEBUG_POSTFIX d)
 
 SET(IMAGE_SRCS	
     nvimage.h
diff --git a/thirdparty/thekla_atlas/src/nvimage/Image.h b/thirdparty/thekla_atlas/src/nvimage/Image.h
index 4c5748cb..bf54940b 100755
--- a/thirdparty/thekla_atlas/src/nvimage/Image.h
+++ b/thirdparty/thekla_atlas/src/nvimage/Image.h
@@ -7,6 +7,10 @@
 #include "nvimage.h"
 #include "nvcore/Debug.h"
 
+#if NV_USE_ALTIVEC
+#undef pixel
+#endif
+
 namespace nv
 {
     class Color32;
diff --git a/thirdparty/thekla_atlas/src/nvmath/Box.h b/thirdparty/thekla_atlas/src/nvmath/Box.h
index bec04784..19b5f2a3 100755
--- a/thirdparty/thekla_atlas/src/nvmath/Box.h
+++ b/thirdparty/thekla_atlas/src/nvmath/Box.h
@@ -99,7 +99,5 @@ namespace nv
 
 } // nv namespace
 
-#include "Box.inl"
-
 
 #endif // NV_MATH_BOX_H
diff --git a/thirdparty/thekla_atlas/src/nvmath/CMakeLists.txt b/thirdparty/thekla_atlas/src/nvmath/CMakeLists.txt
index 49f50496..45e508e4 100755
--- a/thirdparty/thekla_atlas/src/nvmath/CMakeLists.txt
+++ b/thirdparty/thekla_atlas/src/nvmath/CMakeLists.txt
@@ -1,4 +1,5 @@
 PROJECT(nvmath)
+set(CMAKE_DEBUG_POSTFIX d)
 
 SET(MATH_SRCS
 	nvmath.h
diff --git a/thirdparty/thekla_atlas/src/nvmath/ConvexHull.cpp b/thirdparty/thekla_atlas/src/nvmath/ConvexHull.cpp
index a4a95dac..52cfdff5 100755
--- a/thirdparty/thekla_atlas/src/nvmath/ConvexHull.cpp
+++ b/thirdparty/thekla_atlas/src/nvmath/ConvexHull.cpp
@@ -7,12 +7,9 @@
 #include "nvcore/RadixSort.h"
 #include "nvcore/Array.inl"
 
-using namespace nv;
+#include <float.h>
 
-inline static float triangleArea(Vector2::Arg v1, Vector2::Arg v2, Vector2::Arg v3)
-{
-    return 0.5f * (v3.x * v1.y + v1.x * v2.y + v2.x * v3.y - v2.x * v1.y - v3.x * v2.y - v1.x * v3.y);
-}
+using namespace nv;
 
 
 // Compute the convex hull using Graham Scan.
@@ -61,7 +58,7 @@ void nv::convexHull(const Array<Vector2> & input, Array<Vector2> & output, float
         Vector2 b = output[output.count()-1];
         Vector2 c = top[i];
 
-        float area = triangleArea(a, b, c);
+        float area = triangleArea(a, b, c); // * 0.5
 
         if (area >= -epsilon) {
             output.popBack();
@@ -82,7 +79,7 @@ void nv::convexHull(const Array<Vector2> & input, Array<Vector2> & output, float
         Vector2 b = output[output.count()-1];
         Vector2 c = bottom[i];
 
-        float area = triangleArea(a, b, c);
+        float area = triangleArea(a, b, c); // * 0.5
 
         if (area >= -epsilon) {
             output.popBack();
@@ -99,6 +96,63 @@ void nv::convexHull(const Array<Vector2> & input, Array<Vector2> & output, float
     output.popBack();
 }
 
+
+void nv::reduceConvexHullToNSides(Array<Vector2> *input, uint num_sides) {
+
+    while (input->size() > num_sides) {
+        // find the shortest side and merge it with its shorter neighbor.
+        // we'll see how this goes... if it's not satisfactory, we could also merge
+        // based partially on the "flatness" of adjacent edges.
+        int shortest_index = 0;
+        float shortest_length = nv::length((*input)[1] - (*input)[0]);
+
+        for (uint i = 1; i < input->size(); i++) {
+            nv::Vector2 p1 = (*input)[i];
+            nv::Vector2 p2 = (*input)[(i + 1) % input->size()];
+            float length = nv::length(p2 - p1);
+
+            if (length < shortest_length) {
+                shortest_index = i;
+                shortest_length = length;
+            }
+        }
+
+        int prev_index = (shortest_index + input->size() - 1) % input->size();
+        int next_index = (shortest_index + 1) % input->size();
+        int after_index = (next_index + 1) % input->size();
+        float prev_len = nv::length((*input)[prev_index] - (*input)[shortest_index]);
+        float next_len = nv::length((*input)[next_index] - (*input)[after_index]);
+
+        if (prev_len < next_len) {
+            input->removeAt(shortest_index);
+        } else {
+            input->removeAt(next_index);
+        }
+    }
+}
+
+bool nv::isClockwise(const Array<Vector2> &input) {
+    nv::Vector2 d1 = input[1] - input[0];
+    nv::Vector2 d2 = input[2] - input[0];
+    nv::Vector3 normal = cross(Vector3(d1.x, d1.y, 0), Vector3(d2.x, d2.y, 0));
+    return (normal.z < 0.0f);
+}
+
+void nv::flipWinding(Array<Vector2> *input) {
+    int start = 0;
+    int end = input->size() - 1;
+
+    while (end > start) {
+        nv::Vector2 swap = (*input)[start];
+        (*input)[start] = (*input)[end];
+        (*input)[end] = swap;
+        start++;
+        end--;
+    }
+}
+
+
+
 /*
 void testConvexHull() {
 
@@ -118,3 +172,280 @@ void testConvexHull() {
 }
 */
 
+
+
+struct Line {
+    Vector2 v;
+    Vector2 d;
+};
+
+
+bool intersect(const Line &line0, const Line &line1, Vector2 * point)
+{
+	float d = triangleArea(line0.d, line1.d);
+	if (fabsf(d) < NV_EPSILON) // Parallel lines
+		return false;
+
+	float t = triangleArea(line1.d, line0.v - line1.v) / d;
+
+	if (t < 0.5f) // Intersects on the wrong side
+		return false;
+
+	*point = line0.v + t * line0.d;
+	return true;
+}
+
+static inline Line buildLine(const Array<Vector2> & vertices, uint i0, uint i1) {
+    Line line;
+    line.v = vertices[i0];
+    line.d = vertices[i1] - vertices[i0]; // @@ Normalize?
+    return line;
+}
+
+
+static bool buildPolygon(const Array<Vector2> & vertices, const Array<uint> & edges, Array<Vector2> & output) {
+
+    const uint vertexCount = vertices.count();
+    const uint edgeCount = edges.count();
+
+    output.clear();
+    output.reserve(edgeCount);
+
+    // Intersect each edge against the next.
+    Line first, prev, current;
+    Vector2 point;
+    
+    first = prev = buildLine(vertices, 0, 1);
+
+    for (uint i = 1; i < edgeCount; i++) {
+        current = buildLine(vertices, edges[i], (edges[i]+1) % vertexCount);
+
+        if (!intersect(prev, current, &point)) {
+            return false;
+        }
+        output.append(point);
+
+        prev = current;
+    }
+
+    if (!intersect(current, first, &point)) {
+        return false;
+    }
+    output.append(point);
+
+    return true;
+}
+
+// Triangulate polygon and sum triangle areas.
+static float polygonArea(const Array<Vector2> & vertices) {
+
+    float area = 0;
+    Vector2 v0 = vertices[0];
+
+    for (uint i = 2; i < vertices.count(); i++)
+    {
+        Vector2 v1 = vertices[i];
+        Vector2 v2 = vertices[i-1]; 
+
+        area += triangleArea(v0, v1, v2);
+    }
+
+    return area * 0.5f;
+}
+
+// @@ Test this!
+static bool next(Array<uint> & edges, uint vertexCount) {
+    int count = edges.count();
+
+    uint previous = vertexCount;
+
+    for (int i = count-1; i >= 0; i--) {
+        uint e = edges[i];
+        if (e + 1 < previous) {
+            // increment current and reset remainder of the sequence.
+            for (; i < count; i++) {
+                edges[i] = ++e;
+            }
+            return true;
+        }
+        previous = e;
+    }
+
+    return false;
+}
+
+
+
+// Given a convex hull, finds best fit polygon with the given number of edges.
+// Brute force implementation, tries all possible edges, but may not arrive to best result, because edges of best fit polygon are not necessarily aligned to convex hull edges. 
+bool nv::bestFitPolygon(const Array<Vector2> & hullVertices, uint edgeCount, Array<Vector2> * bestPolygon) {
+    uint vertexCount = hullVertices.count();
+
+    // Not a valid polygon.
+    if (edgeCount <= 2) return false;
+
+    // Hull has same or less vertices that desired polygon.
+    if (edgeCount >= vertexCount) return false;
+
+    bestPolygon->reserve(edgeCount);
+    float bestArea = FLT_MAX;
+
+    if (edgeCount == 4) {
+        // Try with axis aligned box first.
+        Vector2 box_min(FLT_MAX);
+        Vector2 box_max(-FLT_MAX);
+
+        for (uint i = 0; i < hullVertices.count(); i++) {
+            box_min.x = min(box_min.x, hullVertices[i].x);
+            box_min.y = min(box_min.y, hullVertices[i].y);
+            box_max.x = max(box_max.x, hullVertices[i].x);
+            box_max.y = max(box_max.y, hullVertices[i].y);
+        }
+
+        bestArea = (box_max.x - box_min.x) * (box_max.y - box_min.y);
+
+        bestPolygon->append(Vector2(box_min.x, box_min.y));
+        bestPolygon->append(Vector2(box_min.x, box_max.y));
+        bestPolygon->append(Vector2(box_max.x, box_max.y));
+        bestPolygon->append(Vector2(box_max.x, box_min.y));
+    }
+
+
+    // Select all possible combinations of consecutive edges.
+
+    // For example, for groups of 3 out of 5 we want:
+
+    // 012
+    // 013
+    // 014
+    // 023
+    // 024
+    // 034
+    // 123
+    // 124
+    // 134
+
+
+    Array<uint> edges(edgeCount);
+
+    // Init edge sequence.
+    for (uint i = 0; i < edgeCount; i++) {
+        edges.append(i);
+    }
+
+    // Traverse sequence testing all 
+    Array<Vector2> polygon(edgeCount);
+    do {
+        if (buildPolygon(hullVertices, edges, polygon)) {
+            float area = polygonArea(polygon);
+            if (area < bestArea) {
+                bestArea = area;
+                *bestPolygon = polygon;
+            }
+        }
+    } while (next(edges, vertexCount));
+
+    return bestArea < FLT_MAX;
+}
+
+
+static bool pointInTriangle(const Vector2 & p, const Vector2 & a, const Vector2 & b, const Vector2 & c)
+{
+    return triangleArea(a, b, p) >= 0.00001f && 
+        triangleArea(b, c, p) >= 0.00001f && 
+        triangleArea(c, a, p) >= 0.00001f; 
+}
+
+void nv::triangulate(const Array<Vector2> & input, Array<uint> * output) {
+
+    const uint edgeCount = input.count();
+    nvDebugCheck(edgeCount >= 3);
+
+    output->clear();    // @@ Do we want to clear?
+    output->reserve(edgeCount);
+
+    if (edgeCount == 3) {
+        // Simple case for triangles.
+        output->append(0);
+        output->append(1);
+        output->append(2);
+    }
+    else {
+        Array<int> polygonVertices;
+        Array<float> polygonAngles;
+
+        polygonVertices.resize(edgeCount);
+        polygonAngles.resize(edgeCount);
+
+        for (uint i = 0; i < edgeCount; ++i) {
+            polygonVertices[i] = i;
+        }
+
+        while (polygonVertices.size() > 2) {
+            uint size = polygonVertices.size();
+
+            // Update polygon angles. @@ Update only those that have changed.
+            float minAngle = 2 * PI;
+            uint bestEar = 0; // Use first one if none of them is valid.
+            bool bestIsValid = false;
+            for (uint i = 0; i < size; i++) {
+                uint i0 = i; 
+                uint i1 = (i+1) % size; // @@ Use Sean's polygon interation trick.
+                uint i2 = (i+2) % size;
+
+                Vector2 p0 = input[polygonVertices[i0]];
+                Vector2 p1 = input[polygonVertices[i1]];
+                Vector2 p2 = input[polygonVertices[i2]];
+
+                float d = clamp(dot(p0-p1, p2-p1) / (length(p0-p1) * length(p2-p1)), -1.0f, 1.0f);
+                float angle = acosf(d);
+                
+                float area = triangleArea(p0, p1, p2);
+                if (area < 0.0f) angle = 2.0f * PI - angle;
+
+                polygonAngles[i1] = angle;
+
+                if (angle < minAngle || !bestIsValid) {
+
+                    // Make sure this is a valid ear, if not, skip this point.
+                    bool valid = true;
+                    for (uint j = 0; j < size; j++) {
+                        if (j == i0 || j == i1 || j == i2) continue;
+                        Vector2 p = input[polygonVertices[j]];
+
+                        if (pointInTriangle(p, p0, p1, p2)) {
+                            valid = false;
+                            break;
+                        }
+                    }
+
+                    if (valid || !bestIsValid) {
+                        minAngle = angle;
+                        bestEar = i1;
+                        bestIsValid = valid;
+                    }
+                }
+            }
+
+            nvDebugCheck(minAngle <= 2 * PI);
+
+            // Clip best ear:
+
+            uint i0 = (bestEar+size-1) % size;
+            uint i1 = (bestEar+0) % size;
+            uint i2 = (bestEar+1) % size;
+
+            int v0 = polygonVertices[i0];
+            int v1 = polygonVertices[i1];
+            int v2 = polygonVertices[i2];
+            
+            output->append(v0);
+            output->append(v1);
+            output->append(v2);
+
+            polygonVertices.removeAt(i1);
+            polygonAngles.removeAt(i1);
+        }
+    }
+
+}
diff --git a/thirdparty/thekla_atlas/src/nvmath/ConvexHull.h b/thirdparty/thekla_atlas/src/nvmath/ConvexHull.h
index 6c2db5d7..ea5a7a23 100755
--- a/thirdparty/thekla_atlas/src/nvmath/ConvexHull.h
+++ b/thirdparty/thekla_atlas/src/nvmath/ConvexHull.h
@@ -12,6 +12,16 @@ namespace nv {
  
     void convexHull(const Array<Vector2> & input, Array<Vector2> & output, float epsilon = 0);
 
+    //ACS: moved these down from collision_volume.cpp
+    bool isClockwise(const Array<Vector2> & input);
+    void reduceConvexHullToNSides(Array<Vector2> *input, uint num_sides);
+    void flipWinding(Array<Vector2> *input);
+
+    bool bestFitPolygon(const Array<Vector2> & input, uint vertexCount, Array<Vector2> * output);
+
+    // Basic ear-clipping algorithm.
+    void triangulate(const Array<Vector2> & input, Array<uint> * output);
+
 } // namespace nv
 
 #endif // NV_MATH_CONVEXHULL_H
diff --git a/thirdparty/thekla_atlas/src/nvmath/Matrix.cpp b/thirdparty/thekla_atlas/src/nvmath/Matrix.cpp
index 29bd19f5..d3d55711 100755
--- a/thirdparty/thekla_atlas/src/nvmath/Matrix.cpp
+++ b/thirdparty/thekla_atlas/src/nvmath/Matrix.cpp
@@ -57,7 +57,7 @@ static bool ludcmp(float **a, int n, int *indx, float *d)
             }
             a[i][j]=sum;
 
-            float dum = vv[i]*fabs(sum);
+            float dum = vv[i]*fabsf(sum);
             if (dum >= big) {
                 // Is the figure of merit for the pivot better than the best so far?
                 big = dum;
@@ -197,6 +197,36 @@ bool nv::solveLU(const Matrix3 & A, const Vector3 & b, Vector3 * x)
     return true;
 }
 
+bool nv::solveLU(const Matrix2 & A, const Vector2 & b, Vector2 * x)
+{
+    nvDebugCheck(x != NULL);
+    
+    float m[2][2];
+    float *a[2] = {m[0], m[1]};
+    int idx[2];
+    float d;
+    
+    for (int y = 0; y < 2; y++) {
+        for (int x = 0; x < 2; x++) {
+            a[x][y] = A(x, y);
+        }
+    }
+    
+    // Create LU decomposition.
+    if (!ludcmp(a, 2, idx, &d)) {
+        // Singular matrix.
+        return false;
+    }
+    
+    // Init solution.
+    *x = b;
+    
+    // Do back substitution.
+    lubksb(a, 2, idx, x->component);
+    
+    return true;
+}
+
 
 bool nv::solveCramer(const Matrix & A, const Vector4 & b, Vector4 * x)
 {
@@ -223,6 +253,22 @@ bool nv::solveCramer(const Matrix3 & A, const Vector3 & b, Vector3 * x)
     return true;
 }
 
+bool nv::solveCramer(const Matrix2 & A, const Vector2 & b, Vector2 * x)
+{
+    nvDebugCheck(x != NULL);
+    
+    const float det = A.determinant();
+    if (equal(det, 0.0f)) {   // @@ Use input epsilon.
+        return false;
+    }
+    
+    Matrix2 Ai = inverseCramer(A);
+    
+    *x = transform(Ai, b);
+    
+    return true;
+}
+
 
 
 // Inverse using gaussian elimination. From Jon's code.
@@ -238,8 +284,8 @@ Matrix nv::inverse(const Matrix & m) {
     for (i=0; i<4; i++) {               /* eliminate in column i, below diag */
         max = -1.;
         for (k=i; k<4; k++)             /* find pivot for column i */
-            if (fabs(A(k, i)) > max) {
-                max = fabs(A(k, i));
+            if (fabsf(A(k, i)) > max) {
+                max = fabsf(A(k, i));
                 j = k;
             }
         if (max<=0.) return B;         /* if no nonzero pivot, PUNT */
@@ -293,8 +339,8 @@ Matrix3 nv::inverse(const Matrix3 & m) {
     for (i=0; i<3; i++) {               /* eliminate in column i, below diag */
         max = -1.;
         for (k=i; k<3; k++)             /* find pivot for column i */
-            if (fabs(A(k, i)) > max) {
-                max = fabs(A(k, i));
+            if (fabsf(A(k, i)) > max) {
+                max = fabsf(A(k, i));
                 j = k;
             }
         if (max<=0.) return B;         /* if no nonzero pivot, PUNT */
diff --git a/thirdparty/thekla_atlas/src/nvmath/Matrix.h b/thirdparty/thekla_atlas/src/nvmath/Matrix.h
index 506bdad1..e48f1992 100755
--- a/thirdparty/thekla_atlas/src/nvmath/Matrix.h
+++ b/thirdparty/thekla_atlas/src/nvmath/Matrix.h
@@ -14,6 +14,46 @@ namespace nv
 {
     enum identity_t { identity };
 
+    // 2x2 matrix.
+    class NVMATH_CLASS Matrix2
+    {
+    public:
+        Matrix2();
+        explicit Matrix2(float f);
+        explicit Matrix2(identity_t);
+        Matrix2(const Matrix2 & m);
+        Matrix2(Vector2::Arg v0, Vector2::Arg v1);
+        Matrix2(float a, float b, float c, float d);
+        
+        float data(uint idx) const;
+        float & data(uint idx);
+        float get(uint row, uint col) const;
+        float operator()(uint row, uint col) const;
+        float & operator()(uint row, uint col);
+        
+        Vector2 row(uint i) const;
+        Vector2 column(uint i) const;
+        
+        void operator*=(float s);
+        void operator/=(float s);
+        void operator+=(const Matrix2 & m);
+        void operator-=(const Matrix2 & m);
+        
+        void scale(float s);
+        void scale(Vector2::Arg s);
+        float determinant() const;
+        
+    private:
+        float m_data[4];
+    };
+    
+    // Solve equation system using LU decomposition and back-substitution.
+    extern bool solveLU(const Matrix2 & m, const Vector2 & b, Vector2 * x);
+    
+    // Solve equation system using Cramer's inverse.
+    extern bool solveCramer(const Matrix2 & A, const Vector2 & b, Vector2 * x);
+    
+    
     // 3x3 matrix.
     class NVMATH_CLASS Matrix3
     {
@@ -52,6 +92,8 @@ namespace nv
     // Solve equation system using Cramer's inverse.
     extern bool solveCramer(const Matrix3 & A, const Vector3 & b, Vector3 * x);
 
+    extern Matrix3 inverse(const Matrix3 & m);
+    
 
     // 4x4 matrix.
     class NVMATH_CLASS Matrix
@@ -106,7 +148,6 @@ namespace nv
 
     // Compute inverse using Gaussian elimination and partial pivoting.
     extern Matrix inverse(const Matrix & m);
-    extern Matrix3 inverse(const Matrix3 & m);
 
 } // nv namespace
 
diff --git a/thirdparty/thekla_atlas/src/nvmath/Matrix.inl b/thirdparty/thekla_atlas/src/nvmath/Matrix.inl
index c0d99d9f..b7319848 100755
--- a/thirdparty/thekla_atlas/src/nvmath/Matrix.inl
+++ b/thirdparty/thekla_atlas/src/nvmath/Matrix.inl
@@ -8,6 +8,199 @@
 
 namespace nv
 {
+    inline Matrix2::Matrix2() {}
+    
+    inline Matrix2::Matrix2(float f)
+    {
+        for(int i = 0; i < 4; i++) {
+            m_data[i] = f;
+        }
+    }
+    
+    inline Matrix2::Matrix2(identity_t)
+    {
+        for(int i = 0; i < 2; i++) {
+            for(int j = 0; j < 2; j++) {
+                m_data[2*j+i] = (i == j) ? 1.0f : 0.0f;
+            }
+        }
+    }
+    
+    inline Matrix2::Matrix2(const Matrix2 & m)
+    {
+        for(int i = 0; i < 4; i++) {
+            m_data[i] = m.m_data[i];
+        }
+    }
+    
+    inline Matrix2::Matrix2(Vector2::Arg v0, Vector2::Arg v1)
+    {
+        m_data[0] = v0.x; m_data[1] = v0.y;
+        m_data[2] = v1.x; m_data[3] = v1.y;
+    }
+    
+    inline Matrix2::Matrix2(float a, float b, float c, float d)
+    {
+        m_data[0] = a; m_data[1] = b;
+        m_data[2] = c; m_data[3] = d;
+    }
+    
+    inline float Matrix2::data(uint idx) const
+    {
+        nvDebugCheck(idx < 4);
+        return m_data[idx];
+    }
+    inline float & Matrix2::data(uint idx)
+    {
+        nvDebugCheck(idx < 4);
+        return m_data[idx];
+    }
+    inline float Matrix2::get(uint row, uint col) const
+    {
+        nvDebugCheck(row < 2 && col < 2);
+        return m_data[col * 2 + row];
+    }
+    inline float Matrix2::operator()(uint row, uint col) const
+    {
+        nvDebugCheck(row < 2 && col < 2);
+        return m_data[col * 2 + row];
+    }
+    inline float & Matrix2::operator()(uint row, uint col)
+    {
+        nvDebugCheck(row < 2 && col < 2);
+        return m_data[col * 2 + row];
+    }
+    
+    inline Vector2 Matrix2::row(uint i) const
+    {
+        nvDebugCheck(i < 2);
+        return Vector2(get(i, 0), get(i, 1));
+    }
+    inline Vector2 Matrix2::column(uint i) const
+    {
+        nvDebugCheck(i < 2);
+        return Vector2(get(0, i), get(1, i));
+    }
+    
+    inline void Matrix2::operator*=(float s)
+    {
+        for(int i = 0; i < 4; i++) {
+            m_data[i] *= s;
+        }
+    }
+    
+    inline void Matrix2::operator/=(float s)
+    {
+        float is = 1.0f /s;
+        for(int i = 0; i < 4; i++) {
+            m_data[i] *= is;
+        }
+    }
+    
+    inline void Matrix2::operator+=(const Matrix2 & m)
+    {
+        for(int i = 0; i < 4; i++) {
+            m_data[i] += m.m_data[i];
+        }
+    }
+    
+    inline void Matrix2::operator-=(const Matrix2 & m)
+    {
+        for(int i = 0; i < 4; i++) {
+            m_data[i] -= m.m_data[i];
+        }
+    }
+    
+    inline Matrix2 operator+(const Matrix2 & a, const Matrix2 & b)
+    {
+        Matrix2 m = a;
+        m += b;
+        return m;
+    }
+    
+    inline Matrix2 operator-(const Matrix2 & a, const Matrix2 & b)
+    {
+        Matrix2 m = a;
+        m -= b;
+        return m;
+    }
+    
+    inline Matrix2 operator*(const Matrix2 & a, float s)
+    {
+        Matrix2 m = a;
+        m *= s;
+        return m;
+    }
+    
+    inline Matrix2 operator*(float s, const Matrix2 & a)
+    {
+        Matrix2 m = a;
+        m *= s;
+        return m;
+    }
+    
+    inline Matrix2 operator/(const Matrix2 & a, float s)
+    {
+        Matrix2 m = a;
+        m /= s;
+        return m;
+    }
+    
+    inline Matrix2 mul(const Matrix2 & a, const Matrix2 & b)
+    {
+        Matrix2 m;
+        
+        for(int i = 0; i < 2; i++) {
+            const float ai0 = a(i,0), ai1 = a(i,1);
+            m(i, 0) = ai0 * b(0,0) + ai1 * b(1,0);
+            m(i, 1) = ai0 * b(0,1) + ai1 * b(1,1);
+        }
+        
+        return m;
+    }
+    
+    inline Matrix2 operator*(const Matrix2 & a, const Matrix2 & b)
+    {
+        return mul(a, b);
+    }
+    
+    // Transform the given 3d vector with the given matrix.
+    inline Vector2 transform(const Matrix2 & m, const Vector2 & p)
+    {
+        return Vector2(p.x * m(0,0) + p.y * m(0,1),
+                       p.x * m(1,0) + p.y * m(1,1));
+    }
+    
+    inline void Matrix2::scale(float s)
+    {
+        for (int i = 0; i < 4; i++) {
+            m_data[i] *= s;
+        }
+    }
+    
+    inline void Matrix2::scale(Vector2::Arg s)
+    {
+        m_data[0] *= s.x; m_data[1] *= s.x;
+        m_data[2] *= s.y; m_data[3] *= s.y;
+    }
+    
+    inline float Matrix2::determinant() const
+    {
+        return get(0,0) * get(1,1) - get(0,1) * get(1,0);
+    }
+    
+    // Inverse using Cramer's rule.
+    inline Matrix2 inverseCramer(const Matrix2 & m)
+    {
+        const float det = m.determinant();
+        if (equal(det, 0.0f, 0.0f)) {
+            return Matrix2(0);
+        }
+        
+        return m * (1/det);
+    }
+    
+    
     inline Matrix3::Matrix3() {}
     
     inline Matrix3::Matrix3(float f)
@@ -16,7 +209,7 @@ namespace nv
             m_data[i] = f;
         }
     }
-
+    
     inline Matrix3::Matrix3(identity_t)
     {
         for(int i = 0; i < 3; i++) {
@@ -536,7 +729,7 @@ namespace nv
     // Get perspective matrix.
     inline Matrix perspective(float fovy, float aspect, float zNear, float zFar)
     {
-        float xmax = zNear * tan(fovy / 2);
+        float xmax = zNear * tanf(fovy / 2);
         float xmin = -xmax;
 
         float ymax = xmax / aspect;
@@ -548,7 +741,7 @@ namespace nv
     // Get inverse perspective matrix.
     inline Matrix perspectiveInverse(float fovy, float aspect, float zNear, float zFar)
     {
-        float xmax = zNear * tan(fovy / 2);
+        float xmax = zNear * tanf(fovy / 2);
         float xmin = -xmax;
 
         float ymax = xmax / aspect;
@@ -560,7 +753,7 @@ namespace nv
     // Get infinite perspective matrix.
     inline Matrix perspective(float fovy, float aspect, float zNear)
     {
-        float x = zNear * tan(fovy / 2);
+        float x = zNear * tanf(fovy / 2);
         float y = x / aspect;
         return frustum( -x, x, -y, y, zNear );	
     }
diff --git a/thirdparty/thekla_atlas/src/nvmath/ProximityGrid.cpp b/thirdparty/thekla_atlas/src/nvmath/ProximityGrid.cpp
index 3553e48f..d5ff0b9f 100755
--- a/thirdparty/thekla_atlas/src/nvmath/ProximityGrid.cpp
+++ b/thirdparty/thekla_atlas/src/nvmath/ProximityGrid.cpp
@@ -65,7 +65,7 @@ void ProximityGrid::init(const Box & box, uint count) {
     else {
         // Ideally we want one cell per point.
         float cellVolume = volume / count;
-        cellWidth = pow(cellVolume, 1.0f / 3.0f);
+        cellWidth = powf(cellVolume, 1.0f / 3.0f);
     }
 
     nvDebugCheck(cellWidth != 0);
diff --git a/thirdparty/thekla_atlas/src/nvmath/Quaternion.h b/thirdparty/thekla_atlas/src/nvmath/Quaternion.h
index dc5219e5..c2890a51 100755
--- a/thirdparty/thekla_atlas/src/nvmath/Quaternion.h
+++ b/thirdparty/thekla_atlas/src/nvmath/Quaternion.h
@@ -22,6 +22,7 @@ namespace nv
         Quaternion(Vector4::Arg v);
 
         const Quaternion & operator=(Quaternion::Arg v);
+        const Quaternion & operator*=(float s);
 
         Vector4 asVector() const;
 
@@ -47,6 +48,13 @@ namespace nv
         w = v.w;
         return *this;
     }
+    inline const Quaternion & Quaternion::operator*=(float s) {
+        x *= s;
+        y *= s;
+        z *= s;
+        w *= s;
+        return *this;
+    }
 
     inline Vector4 Quaternion::asVector() const { return Vector4(x, y, z, w); }
 
@@ -105,7 +113,7 @@ namespace nv
     {
         return scale(q, s);
     }
-
+    
     inline Quaternion scale(Quaternion::Arg q, Vector4::Arg s)
     {
         return scale(q.asVector(), s);
@@ -170,15 +178,15 @@ namespace nv
     /// Transform vector.
     inline Vector3 transform(Quaternion::Arg q, Vector3::Arg v)
     {
-        //Quaternion t = q * v * conjugate(q);
-        //return imag(t);
+        Quaternion t = q * v * conjugate(q);
+        return imag(t);
 
         // Faster method by Fabian Giesen and others:
         // http://molecularmusings.wordpress.com/2013/05/24/a-faster-quaternion-vector-multiplication/
         // http://mollyrocket.com/forums/viewtopic.php?t=833&sid=3a84e00a70ccb046cfc87ac39881a3d0
         
-        Vector3 t = 2 * cross(imag(q), v);
-        return v + q.w * t + cross(imag(q), t);
+        //Vector3 t = 2 * cross(imag(q), v);
+        //return v + q.w * t + cross(imag(q), t);
     }
 
     // @@ Not tested.
@@ -207,7 +215,73 @@ namespace nv
         }
     }
 
+    inline Quaternion fromMatrix(const Matrix3 & m) {
+    #if 0 // IC: There must be a bug in this code:
+        if (m(2, 2) < 0) {
+            if (m(0, 0) < m(1,1)) {
+                float t = 1 - m(0, 0) - m(1, 1) - m(2, 2);
+                return Quaternion(t, m(0,1)+m(1,0), m(2,0)+m(0,2), m(1,2)-m(2,1));
+            }
+            else {
+                float t = 1 - m(0, 0) + m(1, 1) - m(2, 2);
+                return Quaternion(t, m(0,1) + m(1,0), m(1,2) + m(2,1), m(2,0) - m(0,2));
+            }
+        }
+        else {
+            if (m(0, 0) < -m(1, 1)) {
+                float t = 1 - m(0, 0) - m(1, 1) + m(2, 2);
+                return Quaternion(t, m(2,0) + m(0,2), m(1,2) + m(2,1), m(0,1) - m(1,0));
+            }
+            else {
+                float t = 1 + m(0, 0) + m(1, 1) + m(2, 2);
+                return Quaternion(t, m(1,2) - m(2,1), m(2,0) - m(0,2), m(0,1) - m(1,0));
+            }
+        }
+    #else
+        Quaternion q;
+        float tr = m(0,0) + m(1,1) + m(2,2);
+        if (tr > 0) {
+            float s = sqrtf(tr + 1.0f);
+            float p = 0.5f / s;
+            q.x = (m(2,1) - m(1,2)) * p;
+            q.y = (m(0,2) - m(2,0)) * p;
+            q.z = (m(1,0) - m(0,1)) * p;
+            q.w = s * 0.5f;
+        }
+        else if ((m(0,0) >= m(1,1)) && (m(0,0) >= m(2,2))) {
+            float s = sqrtf(m(0,0) - m(1,1) - m(2,2) + 1.0f);
+            float p = 0.5f / s;
+            q.x = s * 0.5f;
+            q.y = (m(1,0) + m(0,1)) * p;
+            q.z = (m(2,0) + m(0,2)) * p;
+            q.w = (m(2,1) - m(1,2)) * p;
+        }
+        else if ((m(1,1) >= m(0,0)) && (m(1,1) >= m(2,2))) {
+            float s = sqrtf(m(1,1) - m(2,2) - m(0,0) + 1.0f);
+            float p = 0.5f / s;
+            q.x = (m(0,1) + m(1,0)) * p; 
+            q.y = s * 0.5f;
+            q.z = (m(2,1) + m(1,2)) * p;
+            q.w = (m(0,2) - m(2,0)) * p;
+        }
+        else if ((m(2,2) >= m(0,0)) && (m(2,2) >= m(1,1))) {
+            float s = sqrtf(m(2,2) - m(0,0) - m(1,1) + 1.0f);
+            float p = 0.5f / s;
+            q.x = (m(0,2) + m(2,0)) * p;
+            q.y = (m(1,2) + m(2,1)) * p;
+            q.z = s * 0.5f;
+            q.w = (m(1,0) - m(0,1)) * p;
+        }
+        return q;
+    #endif
+    }
+    
 
+    
+    
+    
+    
+    
 } // nv namespace
 
 #endif // NV_MATH_QUATERNION_H
diff --git a/thirdparty/thekla_atlas/src/nvmath/Vector.h b/thirdparty/thekla_atlas/src/nvmath/Vector.h
index 9236f2a5..ad18672a 100755
--- a/thirdparty/thekla_atlas/src/nvmath/Vector.h
+++ b/thirdparty/thekla_atlas/src/nvmath/Vector.h
@@ -146,6 +146,4 @@ template <typename T> T to(const nv::Vector2 & v) { NV_COMPILER_CHECK(sizeof(T)
 template <typename T> T to(const nv::Vector3 & v) { NV_COMPILER_CHECK(sizeof(T) == sizeof(nv::Vector3)); return T(v.x, v.y, v.z); }
 template <typename T> T to(const nv::Vector4 & v) { NV_COMPILER_CHECK(sizeof(T) == sizeof(nv::Vector4)); return T(v.x, v.y, v.z, v.w); }
 
-#include "Vector.inl"
-
 #endif // NV_MATH_VECTOR_H
diff --git a/thirdparty/thekla_atlas/src/nvmath/ftoi.h b/thirdparty/thekla_atlas/src/nvmath/ftoi.h
index 5a1c8c58..bee15c09 100755
--- a/thirdparty/thekla_atlas/src/nvmath/ftoi.h
+++ b/thirdparty/thekla_atlas/src/nvmath/ftoi.h
@@ -53,8 +53,7 @@ namespace nv
         return (val<0) ? ftoi_ceil_xs(val) : ftoi_floor_xs(val);
     }
 
-
-#if NV_USE_SSE
+#if NV_CPU_X86 || NV_CPU_X86_64
 
     NV_FORCEINLINE int ftoi_round_sse(float f) {
         return _mm_cvt_ss2si(_mm_set_ss(f));
@@ -64,6 +63,12 @@ namespace nv
       return _mm_cvtt_ss2si(_mm_set_ss(f));
     }
 
+#endif
+
+
+
+#if NV_USE_SSE
+
     NV_FORCEINLINE int ftoi_round(float val) {
         return ftoi_round_sse(val);
     }
diff --git a/thirdparty/thekla_atlas/src/nvmath/nvmath.h b/thirdparty/thekla_atlas/src/nvmath/nvmath.h
index d07300ee..eaf85f42 100755
--- a/thirdparty/thekla_atlas/src/nvmath/nvmath.h
+++ b/thirdparty/thekla_atlas/src/nvmath/nvmath.h
@@ -14,6 +14,12 @@
 #include <float.h>  // finite, isnan
 #endif
 
+#if NV_CPU_X86 || NV_CPU_X86_64
+    //#include <intrin.h>
+    #include <xmmintrin.h>
+#endif
+
+
 
 // Function linkage
 #if NVMATH_SHARED
@@ -36,33 +42,36 @@
 #endif
 
 #ifndef NV_USE_SSE
-    // 1=SSE, 2=SSE2
 #   if NV_CPU_X86_64
         // x64 always supports at least SSE2
 #       define NV_USE_SSE 2
 #   elif NV_CC_MSVC && defined(_M_IX86_FP)
         // Also on x86 with the /arch:SSE flag in MSVC.
-#       define NV_USE_SSE _M_IX86_FP
-#   elif defined(__SSE2__)
-#       define NV_USE_SSE 2
+#       define NV_USE_SSE _M_IX86_FP       // 1=SSE, 2=SS2
 #   elif defined(__SSE__)
 #       define NV_USE_SSE 1
+#   elif defined(__SSE2__)
+#       define NV_USE_SSE 2
 #   else
         // Otherwise we assume no SSE.
 #       define NV_USE_SSE 0
 #   endif
 #endif
 
-#if NV_USE_SSE
-    #include <xmmintrin.h>
+// Temporarily disable SSE to make things compile for now
+#if TARGET_OS_TV
+#undef NV_USE_SSE
+#define NV_USE_SSE 0
 #endif
 
+
 // Internally set NV_USE_SIMD when either altivec or sse is available.
 #if NV_USE_ALTIVEC && NV_USE_SSE
 #	error "Cannot enable both altivec and sse!"
 #endif
 
 
+
 #ifndef PI
 #define PI                  float(3.1415926535897932384626433833)
 #endif
@@ -169,10 +178,8 @@ namespace nv
     {
 #if NV_OS_WIN32 || NV_OS_XBOX || NV_OS_DURANGO
         return _finite(f) != 0;
-#elif NV_OS_DARWIN || NV_OS_FREEBSD || NV_OS_OPENBSD || NV_OS_ORBIS
+#elif NV_OS_DARWIN || NV_OS_FREEBSD || NV_OS_OPENBSD || NV_OS_ORBIS || NV_OS_LINUX
         return isfinite(f);
-#elif NV_OS_LINUX
-        return finitef(f);
 #else
 #   error "isFinite not supported"
 #endif
@@ -184,10 +191,8 @@ namespace nv
     {
 #if NV_OS_WIN32 || NV_OS_XBOX || NV_OS_DURANGO
         return _isnan(f) != 0;
-#elif NV_OS_DARWIN || NV_OS_FREEBSD || NV_OS_OPENBSD || NV_OS_ORBIS
+#elif NV_OS_DARWIN || NV_OS_FREEBSD || NV_OS_OPENBSD || NV_OS_ORBIS || NV_OS_LINUX
         return isnan(f);
-#elif NV_OS_LINUX
-        return isnanf(f);
 #else
 #   error "isNan not supported"
 #endif
@@ -221,7 +226,7 @@ namespace nv
 
     inline float frac(float f)
     {
-        return f - floor(f);
+        return f - floorf(f);
     }
 
     inline float floatRound(float f)
diff --git a/thirdparty/thekla_atlas/src/nvmesh/BaseMesh.cpp b/thirdparty/thekla_atlas/src/nvmesh/BaseMesh.cpp
index f17d3b46..d6cdef77 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/BaseMesh.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/BaseMesh.cpp
@@ -1,9 +1,11 @@
 // This code is in the public domain -- Ignacio Castaņo <castano@gmail.com>
 
 #include "BaseMesh.h"
-#include "Stream.h"
+#include "nvcore/Stream.h"
 #include "nvmath/TypeSerialization.h"
-
+/////// Changes in Dust3D Begin /////////////
+#include "Array.inl"
+/////// Changes in Dust3D End /////////////
 
 namespace nv
 {
diff --git a/thirdparty/thekla_atlas/src/nvmesh/CMakeLists.txt b/thirdparty/thekla_atlas/src/nvmesh/CMakeLists.txt
index b2b1fc06..ae40805f 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/CMakeLists.txt
+++ b/thirdparty/thekla_atlas/src/nvmesh/CMakeLists.txt
@@ -1,4 +1,5 @@
 PROJECT(nvmesh)
+set(CMAKE_DEBUG_POSTFIX d)
 
 SET(MESH_SRCS
 	nvmesh.h
diff --git a/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.cpp b/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.cpp
index 24d8ddff..cfae58f3 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.cpp
@@ -77,6 +77,7 @@ namespace
 
         String name;
         uint faceCount;
+        bool doubleSided = false;
     };
 
     struct Vertex
@@ -141,6 +142,7 @@ struct MeshBuilder::PrivateData
     Array<Face> faceArray;
 
     uint maxFaceIndexCount;
+	bool detect_duplicate_faces = true;
 };
 
 
@@ -221,7 +223,7 @@ void MeshBuilder::endGroup()
 }
 
 // Add named material, check for uniquenes.
-uint MeshBuilder::addMaterial(const char * name)
+uint MeshBuilder::addMaterial(const char * name, bool doubleSided)
 {
     uint index;
     if (d->materialMap.get(name, &index)) {
@@ -232,6 +234,7 @@ uint MeshBuilder::addMaterial(const char * name)
         d->materialMap.add(name, index);
         
         Material material(name);
+        material.doubleSided = doubleSided;
         d->materialArray.append(material);
     }
     return index;
@@ -297,7 +300,7 @@ uint MeshBuilder::addVertex(const Vector3 & pos, const Vector3 & nor, const Vect
 #endif
 
 // Return true if the face is valid and was added to the mesh.
-bool MeshBuilder::endPolygon()
+bool MeshBuilder::endPolygon(bool check_duplicates/*=true*/)
 {
     const Face & face = d->faceArray.back();
     const uint count = face.indexCount;
@@ -353,6 +356,61 @@ bool MeshBuilder::endPolygon()
 
     }
 
+    if (!invalid) {
+
+        Array<Vector3> faceVertices;
+        faceVertices.reserve(4);
+
+        for (uint i = 0; i < face.indexCount; i++) {
+            uint v = d->indexArray[face.firstIndex + i];
+            uint p = d->vertexArray[v].pos;
+            faceVertices.append(d->posArray[p]);
+        }
+
+		if (check_duplicates) {
+
+			uint faceCount = d->faceArray.count() - 1;
+			for (uint f = 0; f < faceCount; f++) {
+				const Face & other = d->faceArray[f];
+
+				// Skip faces with different vertex count.
+				if (other.indexCount != face.indexCount) {
+					continue;
+				}
+
+				const uint orientationCount = (face.material != NIL && d->materialArray[face.material].doubleSided) ? 2 : 1;
+
+				// Compare face vertices in all valid permutations.
+				bool duplicateFace;
+				for (uint orientation = 0; orientation < orientationCount; orientation++) {
+					for (uint order = 0; order < face.indexCount; order++) {
+						duplicateFace = true;
+						for (uint i = 0; i < other.indexCount; i++) {
+							uint v = d->indexArray[other.firstIndex + i];
+							uint p = d->vertexArray[v].pos;
+
+							uint idx = orientation ? (i + order) % face.indexCount : (face.indexCount - 1 - i + order) % face.indexCount;
+
+							if (!equal(faceVertices[idx], d->posArray[p], 0.0f)) {
+								duplicateFace = false;
+								break;
+							}
+						}
+						if (duplicateFace) {
+							goto duplicate;
+						}
+					}
+				}
+
+			duplicate:
+				if (duplicateFace) {
+					invalid = true;
+					break;
+				}
+			}
+		}
+    }
+
     if (invalid)
     {
         d->indexArray.resize(d->indexArray.size() - count);
diff --git a/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.h b/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.h
index 5b3af3fc..196a1933 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.h
+++ b/thirdparty/thekla_atlas/src/nvmesh/MeshBuilder.h
@@ -36,7 +36,7 @@ namespace nv
         void beginGroup(uint id);
         void endGroup();
 
-        uint addMaterial(const char * name);
+        uint addMaterial(const char * name, bool deleteDuplicateFaces);
         void beginMaterial(uint id);
         void endMaterial();
 
@@ -44,7 +44,7 @@ namespace nv
         uint addVertex(uint p, uint n = NIL, uint t0 = NIL, uint t1 = NIL, uint c0 = NIL, uint c1 = NIL, uint c2 = NIL);
         uint addVertex(const Vector3 & p);
         //uint addVertex(const Vector3 & p, const Vector3 & n, const Vector2 & t0 = Vector2(0), const Vector2 & t1 = Vector2(0), const Vector4 & c0 = Vector4(0), const Vector4 & c1 = Vector4(0));
-        bool endPolygon();
+        bool endPolygon(bool check_duplicates = true);
 
         uint weldPositions();
         uint weldNormals();
diff --git a/thirdparty/thekla_atlas/src/nvmesh/QuadTriMesh.cpp b/thirdparty/thekla_atlas/src/nvmesh/QuadTriMesh.cpp
index 64a071ab..48ea50a5 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/QuadTriMesh.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/QuadTriMesh.cpp
@@ -1,7 +1,10 @@
 // This code is in the public domain -- Ignacio Castaņo <castano@gmail.com>
 
 #include "QuadTriMesh.h"
-#include "Stream.h"
+#include "nvcore/Stream.h"
+/////// Changes in Dust3D Begin /////////////
+#include "Array.inl"
+/////// Changes in Dust3D End /////////////
 
 using namespace nv;
 
diff --git a/thirdparty/thekla_atlas/src/nvmesh/param/Atlas.h b/thirdparty/thekla_atlas/src/nvmesh/param/Atlas.h
index 8b478207..856ae60c 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/param/Atlas.h
+++ b/thirdparty/thekla_atlas/src/nvmesh/param/Atlas.h
@@ -157,6 +157,7 @@ namespace nv
         float scale = 1.0f;
         uint vertexMapWidth;
         uint vertexMapHeight;
+        bool blockAligned = true;
 
     private:
 
diff --git a/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.cpp b/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.cpp
index f2156899..fd074bd3 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.cpp
@@ -114,9 +114,12 @@ namespace
 		for (int j = 0; j < h; j++) {
             for (int i = 0; i < w; i++) {
                 Color32 color = bitmap.pixel(i, j);
-                fwrite(&color.r, 1, 1, fp);
-                fwrite(&color.g, 1, 1, fp);
-                fwrite(&color.b, 1, 1, fp);
+                uint8 r = color.r;  // current version of apple's llvm compiling for arm64 doesn't like taking the address of a bit-field. Workaround by using the stack
+                uint8 g = color.g;
+                uint8 b = color.b;
+                fwrite(&r, 1, 1, fp);
+                fwrite(&g, 1, 1, fp);
+                fwrite(&b, 1, 1, fp);
             }
 		}
 
@@ -378,6 +381,8 @@ static void computeBoundingBox(Chart * chart, Vector2 * majorAxis, Vector2 * min
 
 void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned, bool conservative)
 {
+    nvDebugCheck(texelsPerUnit > 0.0f);
+
     const uint chartCount = m_atlas->chartCount();
     if (chartCount == 0) return;
 
@@ -416,7 +421,7 @@ void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned
             //chartOrderArray[c] = chartArea;
 
             // Compute chart scale
-            float parametricArea = fabs(chart->computeParametricArea());    // @@ There doesn't seem to be anything preventing parametric area to be negative.
+            float parametricArea = fabsf(chart->computeParametricArea());    // @@ There doesn't seem to be anything preventing parametric area to be negative.
             if (parametricArea < NV_EPSILON) {
                 // When the parametric area is too small we use a rough approximation to prevent divisions by very small numbers.
                 Vector2 bounds = chart->computeParametricBounds();
@@ -499,7 +504,7 @@ void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned
             if (extents.x > 0) {
                 int cw = ftoi_ceil(extents.x);
 
-                if (blockAligned) {
+                if (blockAligned && chart->blockAligned) {
                     // Align all chart extents to 4x4 blocks, but taking padding into account.
                     if (conservative) {
                         cw = align(cw + 2, 4) - 2;
@@ -517,7 +522,7 @@ void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned
             if (extents.y > 0) {
                 int ch = ftoi_ceil(extents.y);
 
-                if (blockAligned) {
+                if (blockAligned && chart->blockAligned) {
                     // Align all chart extents to 4x4 blocks, but taking padding into account.
                     if (conservative) {
                         ch = align(ch + 2, 4) - 2;
@@ -613,6 +618,8 @@ void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned
         BitMap chart_bitmap;
 
         if (chart->isVertexMapped()) {
+            chart->blockAligned = false;
+
             // Init all bits to 1.
             chart_bitmap.resize(ftoi_ceil(chartExtents[c].x), ftoi_ceil(chartExtents[c].y), /*initValue=*/true);
 
@@ -650,10 +657,14 @@ void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned
             }
         }
 
+        if (!chart->blockAligned) {
+            int k = 1;
+        }
+
         int best_x, best_y;
         int best_cw, best_ch;   // Includes padding now.
         int best_r;
-        findChartLocation(quality, &chart_bitmap, chartExtents[c], w, h, &best_x, &best_y, &best_cw, &best_ch, &best_r);
+        findChartLocation(quality, &chart_bitmap, chartExtents[c], w, h, &best_x, &best_y, &best_cw, &best_ch, &best_r, chart->blockAligned);
         
         /*if (w < best_x + best_cw || h < best_y + best_ch)
         {
@@ -849,7 +860,7 @@ void AtlasPacker::packCharts(int quality, float texelsPerUnit, bool blockAligned
 // is occupied at this point. At the end we have many small charts and a large atlas with sparse holes. Finding those holes randomly is slow. A better approach would be to 
 // start stacking large charts as if they were tetris pieces. Once charts get small try to place them randomly. It may be interesting to try a intermediate strategy, first try 
 // along one axis and then try exhaustively along that axis.
-void AtlasPacker::findChartLocation(int quality, const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r)
+void AtlasPacker::findChartLocation(int quality, const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, bool blockAligned)
 {
     int attempts = 256;
     if (quality == 1) attempts = 4096;
@@ -859,20 +870,23 @@ void AtlasPacker::findChartLocation(int quality, const BitMap * bitmap, Vector2:
 
     if (quality == 0 || w*h < attempts)
     {
-        findChartLocation_bruteForce(bitmap, extents, w, h, best_x, best_y, best_w, best_h, best_r);
+        findChartLocation_bruteForce(bitmap, extents, w, h, best_x, best_y, best_w, best_h, best_r, blockAligned);
     }
     else
     {
-        findChartLocation_random(bitmap, extents, w, h, best_x, best_y, best_w, best_h, best_r, attempts);
+        findChartLocation_random(bitmap, extents, w, h, best_x, best_y, best_w, best_h, best_r, attempts, blockAligned);
     }
 }
 
 #define BLOCK_SIZE 4
 
-void AtlasPacker::findChartLocation_bruteForce(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r)
+void AtlasPacker::findChartLocation_bruteForce(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, bool blockAligned)
 {
     int best_metric = INT_MAX;
 
+    int step_size = blockAligned ? BLOCK_SIZE : 1;
+    
+
     // Try two different orientations.
     for (int r = 0; r < 2; r++)
     {
@@ -880,9 +894,9 @@ void AtlasPacker::findChartLocation_bruteForce(const BitMap * bitmap, Vector2::A
         int ch = bitmap->height();
         if (r & 1) swap(cw, ch);
 
-        for (int y = 0; y <= h + 1; y += BLOCK_SIZE) // + 1 to extend atlas in case atlas full.
+        for (int y = 0; y <= h + 1; y += step_size) // + 1 to extend atlas in case atlas full.
         {
-            for (int x = 0; x <= w + 1; x += BLOCK_SIZE) // + 1 not really necessary here.
+            for (int x = 0; x <= w + 1; x += step_size) // + 1 not really necessary here.
             {
                 // Early out.
                 int area = max(w, x+cw) * max(h, y+ch);
@@ -923,7 +937,7 @@ done:
 }
 
 
-void AtlasPacker::findChartLocation_random(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, int minTrialCount)
+void AtlasPacker::findChartLocation_random(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, int minTrialCount, bool blockAligned)
 {
     int best_metric = INT_MAX;
 
@@ -933,8 +947,10 @@ void AtlasPacker::findChartLocation_random(const BitMap * bitmap, Vector2::Arg e
         int x = m_rand.getRange(w + 1); // + 1 to extend atlas in case atlas full. We may want to use a higher number to increase probability of extending atlas.
         int y = m_rand.getRange(h + 1); // + 1 to extend atlas in case atlas full.
 
-        x = align(x, BLOCK_SIZE);
-        y = align(y, BLOCK_SIZE);
+        if (blockAligned) {
+            x = align(x, BLOCK_SIZE);
+            y = align(y, BLOCK_SIZE);
+        }
 
         int cw = bitmap->width();
         int ch = bitmap->height();
diff --git a/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.h b/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.h
index 2d305f38..08aede94 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.h
+++ b/thirdparty/thekla_atlas/src/nvmesh/param/AtlasPacker.h
@@ -28,9 +28,9 @@ namespace nv
 
     private:
 
-        void findChartLocation(int quality, const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r);
-        void findChartLocation_bruteForce(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r);
-        void findChartLocation_random(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, int minTrialCount);
+        void findChartLocation(int quality, const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, bool blockAligned);
+        void findChartLocation_bruteForce(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, bool blockAligned);
+        void findChartLocation_random(const BitMap * bitmap, Vector2::Arg extents, int w, int h, int * best_x, int * best_y, int * best_w, int * best_h, int * best_r, int minTrialCount, bool blockAligned);
 
         void drawChartBitmapDilate(const Chart * chart, BitMap * bitmap, int padding);
         void drawChartBitmap(const Chart * chart, BitMap * bitmap, const Vector2 & scale, const Vector2 & offset);
diff --git a/thirdparty/thekla_atlas/src/nvmesh/param/ConformalMap.cpp b/thirdparty/thekla_atlas/src/nvmesh/param/ConformalMap.cpp
index 81ab6455..15431f53 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/param/ConformalMap.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/param/ConformalMap.cpp
@@ -2,7 +2,9 @@
 
 #include "ConformalMap.h"
 #include "Util.h"
-
+/////// Changes in Dust3D Begin /////////////
+#include "Array.inl"
+/////// Changes in Dust3D End /////////////
 #include <nvmath/Sparse.h>
 #include <nvmath/Vector.inl>
 #include <nvmath/Solver.h>
diff --git a/thirdparty/thekla_atlas/src/nvmesh/raster/ClippedTriangle.h b/thirdparty/thekla_atlas/src/nvmesh/raster/ClippedTriangle.h
index 0947d485..ecb40c97 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/raster/ClippedTriangle.h
+++ b/thirdparty/thekla_atlas/src/nvmesh/raster/ClippedTriangle.h
@@ -116,7 +116,7 @@ namespace nv
                 centroidx += f * (v[k].x + v[k+1].x);
                 centroidy += f * (v[k].y + v[k+1].y);
             }
-            m_area = 0.5f * fabs(m_area);
+            m_area = 0.5f * fabsf(m_area);
             if (m_area==0) {
                 m_centroid = Vector2(0.0f);
             } else {
diff --git a/thirdparty/thekla_atlas/src/nvmesh/weld/Snap.cpp b/thirdparty/thekla_atlas/src/nvmesh/weld/Snap.cpp
index b6bff4d8..4ef83992 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/weld/Snap.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/weld/Snap.cpp
@@ -1,7 +1,10 @@
 // This code is in the public domain -- castanyo@yahoo.es
 
 #include <nvcore/RadixSort.h>
-
+/////// Changes in Dust3D Begin /////////////
+#include "Array.inl"
+#include "nvmath/Box.inl"
+/////// Changes in Dust3D End /////////////
 #include <nvmesh/weld/Snap.h>
 #include <nvmesh/TriMesh.h>
 #include <nvmesh/geometry/Bounds.h>
diff --git a/thirdparty/thekla_atlas/src/nvmesh/weld/VertexWeld.cpp b/thirdparty/thekla_atlas/src/nvmesh/weld/VertexWeld.cpp
index 2ba4dcae..917fee29 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/weld/VertexWeld.cpp
+++ b/thirdparty/thekla_atlas/src/nvmesh/weld/VertexWeld.cpp
@@ -2,7 +2,9 @@
 
 #include <nvmesh/TriMesh.h>
 #include <nvmesh/QuadTriMesh.h>
-
+/////// Changes in Dust3D Begin /////////////
+#include "Array.inl"
+/////// Changes in Dust3D End /////////////
 #include <nvmesh/weld/VertexWeld.h>
 #include <nvmesh/weld/Weld.h>
 
diff --git a/thirdparty/thekla_atlas/src/nvmesh/weld/Weld.h b/thirdparty/thekla_atlas/src/nvmesh/weld/Weld.h
index e6155394..d9815bf8 100755
--- a/thirdparty/thekla_atlas/src/nvmesh/weld/Weld.h
+++ b/thirdparty/thekla_atlas/src/nvmesh/weld/Weld.h
@@ -111,7 +111,7 @@ inline void reverseXRefs(Array<uint> & xrefs, uint count)
 //
 struct WeldN
 {
-    uint vertexSize;
+    const uint vertexSize;
 
     WeldN(uint n) : vertexSize(n) {}
 
diff --git a/thirdparty/thekla_atlas/src/thekla/thekla_atlas.h b/thirdparty/thekla_atlas/src/thekla/thekla_atlas.h
index 1d0716e7..239dfcf8 100755
--- a/thirdparty/thekla_atlas/src/thekla/thekla_atlas.h
+++ b/thirdparty/thekla_atlas/src/thekla/thekla_atlas.h
@@ -1,21 +1,39 @@
 
 // Thekla Atlas Generator
 
+#if THEKLA_ATLAS_SHARED
+#if _WIN32
+#define DLL_EXPORT __declspec(dllexport)
+#define DLL_IMPORT __declspec(dllimport)
+#else
+#define DLL_EXPORT
+#define DLL_IMPORT
+#endif
+#ifdef THEKLA_ATLAS_EXPORTS
+#define THEKLA_ATLAS_API DLL_EXPORT
+#else
+#define THEKLA_ATLAS_API DLL_IMPORT
+#endif
+#else
+#define THEKLA_ATLAS_API
+#endif
+
+
 namespace Thekla {
 
 enum Atlas_Charter {
-    Atlas_Charter_Witness,  // Options: threshold
-    Atlas_Charter_Extract,  // Options: ---
+    Atlas_Charter_Witness,
+    Atlas_Charter_Extract,
     Atlas_Charter_Default = Atlas_Charter_Witness
 };
 
 enum Atlas_Mapper {
-    Atlas_Mapper_LSCM,      // Options: ---
+    Atlas_Mapper_LSCM,
     Atlas_Mapper_Default = Atlas_Mapper_LSCM
 };
 
 enum Atlas_Packer {
-    Atlas_Packer_Witness,   // Options: texel_area
+    Atlas_Packer_Witness,
     Atlas_Packer_Default = Atlas_Packer_Witness
 };
 
@@ -92,11 +110,11 @@ enum Atlas_Error {
     Atlas_Error_Not_Implemented,
 };
 
-void atlas_set_default_options(Atlas_Options * options);
+THEKLA_ATLAS_API void atlas_set_default_options(Atlas_Options * options);
 
-Atlas_Output_Mesh * atlas_generate(const Atlas_Input_Mesh * input, const Atlas_Options * options, Atlas_Error * error);
+THEKLA_ATLAS_API Atlas_Output_Mesh * atlas_generate(const Atlas_Input_Mesh * input, const Atlas_Options * options, Atlas_Error * error);
 
-void atlas_free(Atlas_Output_Mesh * output);
+THEKLA_ATLAS_API void atlas_free(Atlas_Output_Mesh * output);
 
 
 /*
diff --git a/thirdparty/thekla_atlas/src/thekla/thekla_atlas_test.cpp b/thirdparty/thekla_atlas/src/thekla/thekla_atlas_test.cpp
index 02fbfe6e..296fc390 100755
--- a/thirdparty/thekla_atlas/src/thekla/thekla_atlas_test.cpp
+++ b/thirdparty/thekla_atlas/src/thekla/thekla_atlas_test.cpp
@@ -54,5 +54,4 @@ int main(int argc, char * argv[]) {
     atlas_free(output_mesh);
  
     return 0;
-
 }
diff --git a/thirdparty/thekla_atlas/src/thekla/thekla_mesh_load.h b/thirdparty/thekla_atlas/src/thekla/thekla_mesh_load.h
index 6dcf769d..4b721152 100755
--- a/thirdparty/thekla_atlas/src/thekla/thekla_mesh_load.h
+++ b/thirdparty/thekla_atlas/src/thekla/thekla_mesh_load.h
@@ -58,36 +58,63 @@ Obj_Mesh * obj_mesh_load(const char * filename, const Obj_Load_Options * options
         return NULL;
     }
 
-    printf("%lu shapes\n", shapes.size());
-    printf("%lu materials\n", materials.size());
+    printf("%zu shapes\n", shapes.size());
+    printf("%zu materials\n", materials.size());
 
     assert(shapes.size() > 0);
 
     Obj_Mesh* mesh = new Obj_Mesh();
 
-    mesh->vertex_count = shapes[0].mesh.positions.size() / 3;
-    mesh->vertex_array = new Obj_Vertex[mesh->vertex_count];
-    for (int nvert = 0; nvert < mesh->vertex_count; nvert++) {
-        mesh->vertex_array[nvert].position[0] = shapes[0].mesh.positions[nvert * 3];
-        mesh->vertex_array[nvert].position[1] = shapes[0].mesh.positions[nvert * 3 + 1];
-        mesh->vertex_array[nvert].position[2] = shapes[0].mesh.positions[nvert * 3 + 2];
-        mesh->vertex_array[nvert].normal[0] = shapes[0].mesh.normals[nvert * 3];
-        mesh->vertex_array[nvert].normal[1] = shapes[0].mesh.normals[nvert * 3 + 1];
-        mesh->vertex_array[nvert].normal[2] = shapes[0].mesh.normals[nvert * 3 + 2];
-        mesh->vertex_array[nvert].uv[0] = 0;
-        mesh->vertex_array[nvert].uv[1] = 0;
-        mesh->vertex_array[nvert].first_colocal = nvert;
+    // Count vertices and faces in all shapes.
+    mesh->vertex_count = 0;
+    mesh->face_count = 0;
+    for (int s = 0; s < shapes.size(); s++) {
+        mesh->vertex_count += (int)shapes[s].mesh.positions.size() / 3;
+        mesh->face_count += (int)shapes[s].mesh.indices.size() / 3;
     }
 
-    mesh->face_count = shapes[0].mesh.indices.size() / 3;
-    mesh->face_array = new Obj_Face[mesh->face_count];
-    for (int nface = 0; nface < mesh->face_count; nface++) {
-        mesh->face_array[nface].material_index = 0;
-        mesh->face_array[nface].vertex_index[0] = shapes[0].mesh.indices[nface * 3];
-        mesh->face_array[nface].vertex_index[1] = shapes[0].mesh.indices[nface * 3 + 1];
-        mesh->face_array[nface].vertex_index[2] = shapes[0].mesh.indices[nface * 3 + 2];
+    mesh->vertex_array = new Obj_Vertex[mesh->vertex_count];
+    for (int s = 0, nv = 0; s < shapes.size(); s++) {
+        const int shape_vertex_count = (int)shapes[s].mesh.positions.size() / 3;
+
+        for (int v = 0; v < shape_vertex_count; v++) {
+            mesh->vertex_array[nv+v].position[0] = shapes[s].mesh.positions[v * 3 + 0];
+            mesh->vertex_array[nv+v].position[1] = shapes[s].mesh.positions[v * 3 + 1];
+            mesh->vertex_array[nv+v].position[2] = shapes[s].mesh.positions[v * 3 + 2];
+            mesh->vertex_array[nv+v].normal[0] = shapes[s].mesh.normals[v * 3 + 0];
+            mesh->vertex_array[nv+v].normal[1] = shapes[s].mesh.normals[v * 3 + 1];
+            mesh->vertex_array[nv+v].normal[2] = shapes[s].mesh.normals[v * 3 + 2];
+            mesh->vertex_array[nv+v].uv[0] = shapes[s].mesh.texcoords[v * 3 + 0];      // The input UVs are provided as a hint to the chart generator.
+            mesh->vertex_array[nv+v].uv[1] = shapes[s].mesh.texcoords[v * 3 + 1];
+            mesh->vertex_array[nv+v].first_colocal = nv+v;
+
+            // Link colocals. You probably want to do this more efficiently! Sort by one axis or use a hash or grid.
+            for (int vv = 0; vv < v; vv++) {
+                if (mesh->vertex_array[nv+v].position[0] == mesh->vertex_array[nv+vv].position[0] &&
+                    mesh->vertex_array[nv+v].position[1] == mesh->vertex_array[nv+vv].position[1] &&
+                    mesh->vertex_array[nv+v].position[2] == mesh->vertex_array[nv+vv].position[2])
+                {
+                    mesh->vertex_array[nv+v].first_colocal = nv+vv;
+                }
+            }
+        }
+        nv += shape_vertex_count;
     }
 
+    mesh->face_array = new Obj_Face[mesh->face_count];
+    for (int s = 0, nf = 0; s < shapes.size(); s++) {
+        const int shape_face_count = (int)shapes[s].mesh.indices.size() / 3;
+        for (int f = 0; f < shape_face_count; f++) {
+            mesh->face_array[nf+f].material_index = 0;
+            mesh->face_array[nf+f].vertex_index[0] = shapes[s].mesh.indices[f * 3 + 0];
+            mesh->face_array[nf+f].vertex_index[1] = shapes[s].mesh.indices[f * 3 + 1];
+            mesh->face_array[nf+f].vertex_index[2] = shapes[s].mesh.indices[f * 3 + 2];
+        }
+        nf += shape_face_count;
+    }
+
+    // @@ Add support for obj materials! Charter also uses material boundaries as a hint to cut charts.
+
     printf("Reading %d verts\n", mesh->vertex_count);
     printf("Reading %d triangles\n", mesh->face_count);
 
diff --git a/thirdparty/thekla_atlas/theklaAtlasConfig.cmake.in b/thirdparty/thekla_atlas/theklaAtlasConfig.cmake.in
new file mode 100755
index 00000000..3c4b7653
--- /dev/null
+++ b/thirdparty/thekla_atlas/theklaAtlasConfig.cmake.in
@@ -0,0 +1,21 @@
+set(THEKLA_ATLAS_INCLUDE_DIRS "@CMAKE_INSTALL_PREFIX@/include")
+set(THEKLA_ATLAS_LIBRARIES @CMAKE_INSTALL_PREFIX@/lib/static/libthekla_atlas.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvcore.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvimage.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvmath.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvmesh.lib
+)
+set(THEKLA_ATLAS_LIBRARIES_DEBUG @CMAKE_INSTALL_PREFIX@/lib/static/libTHEKLA_ATLAS@CMAKE_DEBUG_POSTFIX@.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvcore@CMAKE_DEBUG_POSTFIX@.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvimage@CMAKE_DEBUG_POSTFIX@.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvmath@CMAKE_DEBUG_POSTFIX@.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvmesh@CMAKE_DEBUG_POSTFIX@.lib
+)
+set(THEKLA_ATLAS_LIBRARIES_RELEASE @CMAKE_INSTALL_PREFIX@/lib/static/libthekla_atlas.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvcore.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvimage.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvmath.lib
+	@CMAKE_INSTALL_PREFIX@/lib/static/nvmesh.lib
+)
+set(THEKLA_ATLAS_COMPILE_DEFINITIONS @COMPILE_DEFINITIONS@)
+set(THEKLA_ATLAS_FOUND 1)