diff --git a/trunk/src/nvtt/ClusterFit.cpp b/trunk/src/nvtt/ClusterFit.cpp
index 7a072c5..a8e08a4 100644
--- a/trunk/src/nvtt/ClusterFit.cpp
+++ b/trunk/src/nvtt/ClusterFit.cpp
@@ -92,8 +92,8 @@ void ClusterFit::setColourSet(const ColorSet * set)
     {
         int p = order[i];
 #if NVTT_USE_SIMD
-        Vector4 tmp(values[p] * set->weights[p], set->weights[p]);
-        m_weighted[i] = SimdVector(tmp);
+        NV_ALIGN_16 Vector4 tmp(values[p] * set->weights[p], set->weights[p]);
+        m_weighted[i] = SimdVector(tmp.component);
         m_xxsum += m_weighted[i] * m_weighted[i];
         m_xsum += m_weighted[i];
 #else
@@ -110,8 +110,8 @@ void ClusterFit::setColourSet(const ColorSet * set)
 void ClusterFit::setMetric(Vector4::Arg w)
 {
 #if NVTT_USE_SIMD
-    Vector4 tmp(w.xyz(), 1);
-    m_metric = SimdVector(tmp);
+    NV_ALIGN_16 Vector4 tmp(w.xyz(), 1);
+    m_metric = SimdVector(tmp.component);
 #else
     m_metric = w.xyz();
 #endif
@@ -134,13 +134,13 @@ float ClusterFit::bestError() const
 
 bool ClusterFit::compress3( Vector3 * start, Vector3 * end )
 {
-    int const count = m_count;
-    SimdVector const one = SimdVector(1.0f);
-    SimdVector const zero = SimdVector(0.0f);
-    SimdVector const half(0.5f, 0.5f, 0.5f, 0.25f);
-    SimdVector const two = SimdVector(2.0);
-    SimdVector const grid( 31.0f, 63.0f, 31.0f, 0.0f );
-    SimdVector const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f, 0.0f );
+    const int count = m_count;
+    const SimdVector one = SimdVector(1.0f);
+    const SimdVector zero = SimdVector(0.0f);
+    const SimdVector half(0.5f, 0.5f, 0.5f, 0.25f);
+    const SimdVector two = SimdVector(2.0);
+    const SimdVector grid( 31.0f, 63.0f, 31.0f, 0.0f );
+    const SimdVector gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f, 0.0f );
 
     // declare variables
     SimdVector beststart = SimdVector( 0.0f );
@@ -158,23 +158,23 @@ bool ClusterFit::compress3( Vector3 * start, Vector3 * end )
 
         for( int c1 = 0; c1 <= count-c0; c1++)
         {
-            SimdVector const x2 = m_xsum - x1 - x0;
+            const SimdVector x2 = m_xsum - x1 - x0;
 
-            //Vector3 const alphax_sum = x0 + x1 * 0.5f;
-            //float const alpha2_sum = w0 + w1 * 0.25f;
-            SimdVector const alphax_sum = multiplyAdd(x1, half, x0); // alphax_sum, alpha2_sum
-            SimdVector const alpha2_sum = alphax_sum.splatW();
+            //Vector3 alphax_sum = x0 + x1 * 0.5f;
+            //float alpha2_sum = w0 + w1 * 0.25f;
+            const SimdVector alphax_sum = multiplyAdd(x1, half, x0); // alphax_sum, alpha2_sum
+            const SimdVector alpha2_sum = alphax_sum.splatW();
 
-            //Vector3 const betax_sum = x2 + x1 * 0.5f;
-            //float const beta2_sum = w2 + w1 * 0.25f;
-            SimdVector const betax_sum = multiplyAdd(x1, half, x2); // betax_sum, beta2_sum
-            SimdVector const beta2_sum = betax_sum.splatW();
+            //const Vector3 betax_sum = x2 + x1 * 0.5f;
+            //const float beta2_sum = w2 + w1 * 0.25f;
+            const SimdVector betax_sum = multiplyAdd(x1, half, x2); // betax_sum, beta2_sum
+            const SimdVector beta2_sum = betax_sum.splatW();
 
-            //float const alphabeta_sum = w1 * 0.25f;
-            SimdVector const alphabeta_sum = (x1 * half).splatW(); // alphabeta_sum
+            //const float alphabeta_sum = w1 * 0.25f;
+            const SimdVector alphabeta_sum = (x1 * half).splatW(); // alphabeta_sum
 
-            // float const factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
-            SimdVector const factor = reciprocal( negativeMultiplySubtract(alphabeta_sum, alphabeta_sum, alpha2_sum*beta2_sum) );
+            // const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
+            const SimdVector factor = reciprocal( negativeMultiplySubtract(alphabeta_sum, alphabeta_sum, alpha2_sum*beta2_sum) );
 
             SimdVector a = negativeMultiplySubtract(betax_sum, alphabeta_sum, alphax_sum*beta2_sum) * factor;
             SimdVector b = negativeMultiplySubtract(alphax_sum, alphabeta_sum, betax_sum*alpha2_sum) * factor;
@@ -229,16 +229,16 @@ bool ClusterFit::compress3( Vector3 * start, Vector3 * end )
 
 bool ClusterFit::compress4( Vector3 * start, Vector3 * end )
 {
-    int const count = m_count;
-    SimdVector const one = SimdVector(1.0f);
-    SimdVector const zero = SimdVector(0.0f);
-    SimdVector const half = SimdVector(0.5f);
-    SimdVector const two = SimdVector(2.0);
-    SimdVector const onethird( 1.0f/3.0f, 1.0f/3.0f, 1.0f/3.0f, 1.0f/9.0f );
-    SimdVector const twothirds( 2.0f/3.0f, 2.0f/3.0f, 2.0f/3.0f, 4.0f/9.0f );
-    SimdVector const twonineths = SimdVector( 2.0f/9.0f );
-    SimdVector const grid( 31.0f, 63.0f, 31.0f, 0.0f );
-    SimdVector const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f, 0.0f );
+    const int count = m_count;
+    const SimdVector one = SimdVector(1.0f);
+    const SimdVector zero = SimdVector(0.0f);
+    const SimdVector half = SimdVector(0.5f);
+    const SimdVector two = SimdVector(2.0);
+    const SimdVector onethird( 1.0f/3.0f, 1.0f/3.0f, 1.0f/3.0f, 1.0f/9.0f );
+    const SimdVector twothirds( 2.0f/3.0f, 2.0f/3.0f, 2.0f/3.0f, 4.0f/9.0f );
+    const SimdVector twonineths = SimdVector( 2.0f/9.0f );
+    const SimdVector grid( 31.0f, 63.0f, 31.0f, 0.0f );
+    const SimdVector gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f, 0.0f );
 
     // declare variables
     SimdVector beststart = SimdVector( 0.0f );
@@ -259,23 +259,23 @@ bool ClusterFit::compress4( Vector3 * start, Vector3 * end )
 
             for( int c2 = 0; c2 <= count-c0-c1; c2++)
             {
-                SimdVector const x3 = m_xsum - x2 - x1 - x0;
+                const SimdVector x3 = m_xsum - x2 - x1 - x0;
 
-                //Vector3 const alphax_sum = x0 + x1 * (2.0f / 3.0f) + x2 * (1.0f / 3.0f);
-                //float const alpha2_sum = w0 + w1 * (4.0f/9.0f) + w2 * (1.0f/9.0f);
-                SimdVector const alphax_sum = multiplyAdd(x2, onethird, multiplyAdd(x1, twothirds, x0)); // alphax_sum, alpha2_sum
-                SimdVector const alpha2_sum = alphax_sum.splatW();
+                //const Vector3 alphax_sum = x0 + x1 * (2.0f / 3.0f) + x2 * (1.0f / 3.0f);
+                //const float alpha2_sum = w0 + w1 * (4.0f/9.0f) + w2 * (1.0f/9.0f);
+                const SimdVector alphax_sum = multiplyAdd(x2, onethird, multiplyAdd(x1, twothirds, x0)); // alphax_sum, alpha2_sum
+                const SimdVector alpha2_sum = alphax_sum.splatW();
 
-                //Vector3 const betax_sum = x3 + x2 * (2.0f / 3.0f) + x1 * (1.0f / 3.0f);
-                //float const beta2_sum = w3 + w2 * (4.0f/9.0f) + w1 * (1.0f/9.0f);
-                SimdVector const betax_sum = multiplyAdd(x2, twothirds, multiplyAdd(x1, onethird, x3)); // betax_sum, beta2_sum
-                SimdVector const beta2_sum = betax_sum.splatW();
+                //const Vector3 betax_sum = x3 + x2 * (2.0f / 3.0f) + x1 * (1.0f / 3.0f);
+                //const float beta2_sum = w3 + w2 * (4.0f/9.0f) + w1 * (1.0f/9.0f);
+                const SimdVector betax_sum = multiplyAdd(x2, twothirds, multiplyAdd(x1, onethird, x3)); // betax_sum, beta2_sum
+                const SimdVector beta2_sum = betax_sum.splatW();
 
-                //float const alphabeta_sum = (w1 + w2) * (2.0f/9.0f);
-                SimdVector const alphabeta_sum = twonineths*( x1 + x2 ).splatW(); // alphabeta_sum
+                //const float alphabeta_sum = (w1 + w2) * (2.0f/9.0f);
+                const SimdVector alphabeta_sum = twonineths*( x1 + x2 ).splatW(); // alphabeta_sum
 
-                // float const factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
-                SimdVector const factor = reciprocal( negativeMultiplySubtract(alphabeta_sum, alphabeta_sum, alpha2_sum*beta2_sum) );
+                //const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
+                const SimdVector factor = reciprocal( negativeMultiplySubtract(alphabeta_sum, alphabeta_sum, alpha2_sum*beta2_sum) );
 
                 SimdVector a = negativeMultiplySubtract(betax_sum, alphabeta_sum, alphax_sum*beta2_sum) * factor;
                 SimdVector b = negativeMultiplySubtract(alphax_sum, alphabeta_sum, betax_sum*alpha2_sum) * factor;
diff --git a/trunk/src/nvtt/CompressorDXT.cpp b/trunk/src/nvtt/CompressorDXT.cpp
index c614bd7..ea8c629 100644
--- a/trunk/src/nvtt/CompressorDXT.cpp
+++ b/trunk/src/nvtt/CompressorDXT.cpp
@@ -113,7 +113,7 @@ void FixedBlockCompressor::compress(nvtt::AlphaMode alphaMode, uint w, uint h, c
 */
 
 
-struct CompressorContext
+struct FixedBlockCompressorContext
 {
     nvtt::AlphaMode alphaMode;
     uint w, h;
@@ -125,10 +125,10 @@ struct CompressorContext
     FixedBlockCompressor * compressor;
 };
 
-// Each task compresses one row.
-void CompressorTask(void * data, int i)
+// Each task compresses one block.
+void FixedBlockCompressorTask(void * data, int i)
 {
-    CompressorContext * d = (CompressorContext *) data;
+    FixedBlockCompressorContext * d = (FixedBlockCompressorContext *) data;
 
     uint x = i % d->bw;
     uint y = i / d->bw;
@@ -147,7 +147,7 @@ void FixedBlockCompressor::compress(nvtt::AlphaMode alphaMode, uint w, uint h, u
 {
     nvDebugCheck(d == 1);
 
-    CompressorContext context;
+    FixedBlockCompressorContext context;
     context.alphaMode = alphaMode;
     context.w = w;
     context.h = h;
@@ -169,7 +169,7 @@ void FixedBlockCompressor::compress(nvtt::AlphaMode alphaMode, uint w, uint h, u
     const uint size = context.bs * count;
     context.mem = new uint8[size];
 
-    dispatcher->dispatch(CompressorTask, &context, count);
+    dispatcher->dispatch(FixedBlockCompressorTask, &context, count);
 
     outputOptions.writeData(context.mem, size);
 
@@ -177,35 +177,67 @@ void FixedBlockCompressor::compress(nvtt::AlphaMode alphaMode, uint w, uint h, u
 }
 
 
+struct ColorSetCompressorContext
+{
+    nvtt::AlphaMode alphaMode;
+    uint w, h;
+    const float * data;
+    const nvtt::CompressionOptions::Private * compressionOptions;
+
+    uint bw, bh, bs;
+    uint8 * mem;
+    ColorSetCompressor * compressor;
+};
+
+
+// Each task compresses one block.
+void ColorSetCompressorTask(void * data, int i)
+{
+    ColorSetCompressorContext * d = (ColorSetCompressorContext *) data;
+
+    uint x = i % d->bw;
+    uint y = i / d->bw;
+
+    //for (uint x = 0; x < d->bw; x++)
+    {
+        ColorSet set;
+        set.setColors(d->data, d->w, d->h, x, y);
+
+        uint8 * ptr = d->mem + (y * d->bw + x) * d->bs;
+        d->compressor->compressBlock(set, d->alphaMode, *d->compressionOptions, ptr);
+    }
+}
 
 
 void ColorSetCompressor::compress(AlphaMode alphaMode, uint w, uint h, uint d, const float * data, nvtt::TaskDispatcher * dispatcher, const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
 {
     nvDebugCheck(d == 1);
 
-    const uint bs = blockSize();
-    const uint bw = (w + 3) / 4;
-    const uint bh = (h + 3) / 4;
+    ColorSetCompressorContext context;
+    context.alphaMode = alphaMode;
+    context.w = w;
+    context.h = h;
+    context.data = data;
+    context.compressionOptions = &compressionOptions;
 
-    //bool singleThreaded = true;
-    //if (singleThreaded)
-    {
-        uint8 * mem = malloc<uint8>(bs * bw);
+    context.bs = blockSize();
+    context.bw = (w + 3) / 4;
+    context.bh = (h + 3) / 4;
 
-        ColorSet set;
+    context.compressor = this;
 
-        for (uint y = 0; y < h; y += 4) {
-        	uint8 * ptr = mem;
-            for (uint x = 0; x < w; x += 4, ptr += bs) {
-                set.setColors(data, w, h, x, y);
-                compressBlock(set, alphaMode, compressionOptions, ptr);
-            }
+    SequentialTaskDispatcher sequential;
 
-            if (outputOptions.outputHandler != NULL) {
-                outputOptions.outputHandler->writeData(mem, bs * bw);
-            }
-        }
+    // Use a single thread to compress small textures.
+    if (context.bh < 4) dispatcher = &sequential;
 
-        free(mem);
-    }
+    const uint count = context.bw * context.bh;
+    const uint size = context.bs * count;
+    context.mem = new uint8[size];
+
+    dispatcher->dispatch(ColorSetCompressorTask, &context, count);
+
+    outputOptions.writeData(context.mem, size);
+
+    delete [] context.mem;
 }
diff --git a/trunk/src/nvtt/QuickCompressDXT.cpp b/trunk/src/nvtt/QuickCompressDXT.cpp
index 2833636..084ad52 100644
--- a/trunk/src/nvtt/QuickCompressDXT.cpp
+++ b/trunk/src/nvtt/QuickCompressDXT.cpp
@@ -721,8 +721,8 @@ void QuickCompress::outputBlock4(const ColorSet & set, const Vector3 & start, co
 
     if (color0 < color1)
     {
-            swap(maxColor, minColor);
-            swap(color0, color1);
+        swap(maxColor, minColor);
+        swap(color0, color1);
     }
 
     block->col0 = Color16(color0);
@@ -741,8 +741,8 @@ void QuickCompress::outputBlock3(const ColorSet & set, const Vector3 & start, co
 
     if (color0 > color1)
     {
-            swap(maxColor, minColor);
-            swap(color0, color1);
+        swap(maxColor, minColor);
+        swap(color0, color1);
     }
 
     block->col0 = Color16(color0);
diff --git a/trunk/src/nvtt/Surface.cpp b/trunk/src/nvtt/Surface.cpp
index 6e9cf73..86e67d4 100644
--- a/trunk/src/nvtt/Surface.cpp
+++ b/trunk/src/nvtt/Surface.cpp
@@ -432,6 +432,30 @@ bool Surface::save(const char * fileName) const
     return false;
 }
 
+#if 0 //NV_OS_WIN32
+
+#include <windows.h>
+#undef min
+#undef max
+
+static int filter(unsigned int code, struct _EXCEPTION_POINTERS *ep) {
+   if (code == EXCEPTION_ACCESS_VIOLATION) {
+      return EXCEPTION_EXECUTE_HANDLER;
+   }
+   else {
+      return EXCEPTION_CONTINUE_SEARCH;
+   };
+}
+
+#define TRY __try
+    
+#define CATCH __except (filter(GetExceptionCode(), GetExceptionInformation()))
+#else
+#define TRY
+#define CATCH
+#endif
+
+
 bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void * data)
 {
     detach();
@@ -453,7 +477,7 @@ bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void
     {
         const Color32 * src = (const Color32 *)data;
 
-        try {
+        TRY {
             for (int i = 0; i < count; i++)
             {
                 rdst[i] = float(src[i].r) / 255.0f;
@@ -462,7 +486,7 @@ bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void
                 adst[i] = float(src[i].a) / 255.0f;
             }
         }
-        catch(...) {
+        CATCH {
             return false;
         }
     }
@@ -470,7 +494,7 @@ bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void
     {
         const uint16 * src = (const uint16 *)data;
 
-        try {
+        TRY {
             for (int i = 0; i < count; i++)
             {
                 ((uint32 *)rdst)[i] = half_to_float(src[4*i+0]);
@@ -479,7 +503,7 @@ bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void
                 ((uint32 *)adst)[i] = half_to_float(src[4*i+3]);
             }
         }
-        catch(...) {
+        CATCH {
             return false;
         }
     }
@@ -487,7 +511,7 @@ bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void
     {
         const float * src = (const float *)data;
 
-        try {
+        TRY {
             for (int i = 0; i < count; i++)
             {
                 rdst[i] = src[4 * i + 0];
@@ -496,7 +520,7 @@ bool Surface::setImage(nvtt::InputFormat format, int w, int h, int d, const void
                 adst[i] = src[4 * i + 3];
             }
         }
-        catch(...) {
+        CATCH {
             return false;
         }
     }
diff --git a/trunk/src/nvtt/tools/compress.cpp b/trunk/src/nvtt/tools/compress.cpp
index fba7a26..8ac5a16 100644
--- a/trunk/src/nvtt/tools/compress.cpp
+++ b/trunk/src/nvtt/tools/compress.cpp
@@ -270,6 +270,11 @@ int main(int argc, char *argv[])
                 i++;
             }
         }
+        else if (strcmp("-pause", argv[i]) == 0)
+        {
+            printf("Press ENTER\n"); fflush(stdout);
+            getchar();
+        }
 
         // Output options
         else if (strcmp("-silent", argv[i]) == 0)
@@ -529,6 +534,11 @@ int main(int argc, char *argv[])
         compressionOptions.setColorWeights(1, 1, 0);
     }
 
+    
+    //compressionOptions.setColorWeights(0.2126, 0.7152, 0.0722);
+    //compressionOptions.setColorWeights(0.299, 0.587, 0.114);
+    //compressionOptions.setColorWeights(3, 4, 2);
+
     if (externalCompressor != NULL)
     {
         compressionOptions.setExternalCompressor(externalCompressor);