Meshoptimizer update library

6 files changed, 220 insertions, 356 deletions

diff --git a/thirdparty/meshoptimizer/indexcodec.cpp b/thirdparty/meshoptimizer/indexcodec.cpp
index eeb541e5be..5c35eb43ae 100644
--- a/thirdparty/meshoptimizer/indexcodec.cpp
+++ b/thirdparty/meshoptimizer/indexcodec.cpp
@@ -4,14 +4,6 @@
 #include <assert.h>
 #include <string.h>
 
-#ifndef TRACE
-#define TRACE 0
-#endif
-
-#if TRACE
-#include <stdio.h>
-#endif
-
 // This work is based on:
 // Fabian Giesen. Simple lossless index buffer compression & follow-up. 2013
 // Conor Stokes. Vertex Cache Optimised Index Buffer Compression. 2014
@@ -167,38 +159,6 @@ static void writeTriangle(void* destination, size_t offset, size_t index_size, u
 	}
 }
 
-#if TRACE
-static size_t sortTop16(unsigned char dest[16], size_t stats[256])
-{
-	size_t destsize = 0;
-
-	for (size_t i = 0; i < 256; ++i)
-	{
-		size_t j = 0;
-		for (; j < destsize; ++j)
-		{
-			if (stats[i] >= stats[dest[j]])
-			{
-				if (destsize < 16)
-					destsize++;
-
-				memmove(&dest[j + 1], &dest[j], destsize - 1 - j);
-				dest[j] = (unsigned char)i;
-				break;
-			}
-		}
-
-		if (j == destsize && destsize < 16)
-		{
-			dest[destsize] = (unsigned char)i;
-			destsize++;
-		}
-	}
-
-	return destsize;
-}
-#endif
-
 } // namespace meshopt
 
 size_t meshopt_encodeIndexBuffer(unsigned char* buffer, size_t buffer_size, const unsigned int* indices, size_t index_count)
@@ -207,11 +167,6 @@ size_t meshopt_encodeIndexBuffer(unsigned char* buffer, size_t buffer_size, cons
 
 	assert(index_count % 3 == 0);
 
-#if TRACE
-	size_t codestats[256] = {};
-	size_t codeauxstats[256] = {};
-#endif
-
 	// the minimum valid encoding is header, 1 byte per triangle and a 16-byte codeaux table
 	if (buffer_size < 1 + index_count / 3 + 16)
 		return 0;
@@ -275,10 +230,6 @@ size_t meshopt_encodeIndexBuffer(unsigned char* buffer, size_t buffer_size, cons
 
 			*code++ = (unsigned char)((fe << 4) | fec);
 
-#if TRACE
-			codestats[code[-1]]++;
-#endif
-
 			// note that we need to update the last index since free indices are delta-encoded
 			if (fec == 15)
 				encodeIndex(data, c, last), last = c;
@@ -334,11 +285,6 @@ size_t meshopt_encodeIndexBuffer(unsigned char* buffer, size_t buffer_size, cons
 				*data++ = codeaux;
 			}
 
-#if TRACE
-			codestats[code[-1]]++;
-			codeauxstats[codeaux]++;
-#endif
-
 			// note that we need to update the last index since free indices are delta-encoded
 			if (fea == 15)
 				encodeIndex(data, a, last), last = a;
@@ -387,30 +333,6 @@ size_t meshopt_encodeIndexBuffer(unsigned char* buffer, size_t buffer_size, cons
 	assert(data >= buffer + index_count / 3 + 16);
 	assert(data <= buffer + buffer_size);
 
-#if TRACE
-	unsigned char codetop[16], codeauxtop[16];
-	size_t codetopsize = sortTop16(codetop, codestats);
-	size_t codeauxtopsize = sortTop16(codeauxtop, codeauxstats);
-
-	size_t sumcode = 0, sumcodeaux = 0;
-	for (size_t i = 0; i < 256; ++i)
-		sumcode += codestats[i], sumcodeaux += codeauxstats[i];
-
-	size_t acccode = 0, acccodeaux = 0;
-
-	printf("code\t\t\t\t\tcodeaux\n");
-
-	for (size_t i = 0; i < codetopsize && i < codeauxtopsize; ++i)
-	{
-		acccode += codestats[codetop[i]];
-		acccodeaux += codeauxstats[codeauxtop[i]];
-
-		printf("%2d: %02x = %d (%.1f%% ..%.1f%%)\t\t%2d: %02x = %d (%.1f%% ..%.1f%%)\n",
-		       int(i), codetop[i], int(codestats[codetop[i]]), double(codestats[codetop[i]]) / double(sumcode) * 100, double(acccode) / double(sumcode) * 100,
-		       int(i), codeauxtop[i], int(codeauxstats[codeauxtop[i]]), double(codeauxstats[codeauxtop[i]]) / double(sumcodeaux) * 100, double(acccodeaux) / double(sumcodeaux) * 100);
-	}
-#endif
-
 	return data - buffer;
 }
 
diff --git a/thirdparty/meshoptimizer/meshoptimizer.h b/thirdparty/meshoptimizer/meshoptimizer.h
index fde00f9c82..4071f0a371 100644
--- a/thirdparty/meshoptimizer/meshoptimizer.h
+++ b/thirdparty/meshoptimizer/meshoptimizer.h
@@ -239,7 +239,6 @@ MESHOPTIMIZER_API int meshopt_decodeVertexBuffer(void* destination, size_t verte
 /**
  * Vertex buffer filters
  * These functions can be used to filter output of meshopt_decodeVertexBuffer in-place.
- * count must be aligned by 4 and stride is fixed for each function to facilitate SIMD implementation.
  *
  * meshopt_decodeFilterOct decodes octahedral encoding of a unit vector with K-bit (K <= 16) signed X/Y as an input; Z must store 1.0f.
  * Each component is stored as an 8-bit or 16-bit normalized integer; stride must be equal to 4 or 8. W is preserved as is.
@@ -265,11 +264,10 @@ MESHOPTIMIZER_EXPERIMENTAL void meshopt_decodeFilterExp(void* buffer, size_t ver
  *
  * destination must contain enough space for the *source* index buffer (since optimization is iterative, this means index_count elements - *not* target_index_count!)
  * vertex_positions should have float3 position in the first 12 bytes of each vertex - similar to glVertexPointer
+ * target_error represents the error relative to mesh extents that can be tolerated, e.g. 0.01 = 1% deformation
+ * result_error can be NULL; when it's not NULL, it will contain the resulting (relative) error after simplification
  */
-// -- GODOT start --
-//MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error);
-MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices, size_t index_count, const float *vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float *r_resulting_error);
-// -- GODOT end --
+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float* result_error);
 
 /**
  * Experimental: Mesh simplifier (sloppy)
@@ -297,6 +295,14 @@ MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifySloppy(unsigned int* destinati
 MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifyPoints(unsigned int* destination, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_vertex_count);
 
 /**
+ * Experimental: Returns the error scaling factor used by the simplifier to convert between absolute and relative extents
+ * 
+ * Absolute error must be *divided* by the scaling factor before passing it to meshopt_simplify as target_error
+ * Relative error returned by meshopt_simplify via result_error must be *multiplied* by the scaling factor to get absolute error.
+ */
+MESHOPTIMIZER_EXPERIMENTAL float meshopt_simplifyScale(const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
+
+/**
  * Mesh stripifier
  * Converts a previously vertex cache optimized triangle list to triangle strip, stitching strips using restart index or degenerate triangles
  * Returns the number of indices in the resulting strip, with destination containing new index data
@@ -525,7 +531,7 @@ inline size_t meshopt_encodeIndexSequence(unsigned char* buffer, size_t buffer_s
 template <typename T>
 inline int meshopt_decodeIndexSequence(T* destination, size_t index_count, const unsigned char* buffer, size_t buffer_size);
 template <typename T>
-inline size_t meshopt_simplify(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error);
+inline size_t meshopt_simplify(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float* result_error = 0);
 template <typename T>
 inline size_t meshopt_simplifySloppy(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count);
 template <typename T>
@@ -840,12 +846,12 @@ inline int meshopt_decodeIndexSequence(T* destination, size_t index_count, const
 }
 
 template <typename T>
-inline size_t meshopt_simplify(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error)
+inline size_t meshopt_simplify(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float* result_error)
 {
 	meshopt_IndexAdapter<T> in(0, indices, index_count);
 	meshopt_IndexAdapter<T> out(destination, 0, index_count);
 
-	return meshopt_simplify(out.data, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride, target_index_count, target_error);
+	return meshopt_simplify(out.data, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride, target_index_count, target_error, result_error);
 }
 
 template <typename T>
diff --git a/thirdparty/meshoptimizer/patches/simplifier_get_resulting_error.patch b/thirdparty/meshoptimizer/patches/simplifier_get_resulting_error.patch
deleted file mode 100644
index 1be38e45d2..0000000000
--- a/thirdparty/meshoptimizer/patches/simplifier_get_resulting_error.patch
+++ /dev/null
@@ -1,96 +0,0 @@
-diff --git a/thirdparty/meshoptimizer/meshoptimizer.h b/thirdparty/meshoptimizer/meshoptimizer.h
-index a442d103c8..fde00f9c82 100644
---- a/thirdparty/meshoptimizer/meshoptimizer.h
-+++ b/thirdparty/meshoptimizer/meshoptimizer.h
-@@ -266,7 +266,10 @@ MESHOPTIMIZER_EXPERIMENTAL void meshopt_decodeFilterExp(void* buffer, size_t ver
-  * destination must contain enough space for the *source* index buffer (since optimization is iterative, this means index_count elements - *not* target_index_count!)
-  * vertex_positions should have float3 position in the first 12 bytes of each vertex - similar to glVertexPointer
-  */
--MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error);
-+// -- GODOT start --
-+//MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error);
-+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices, size_t index_count, const float *vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float *r_resulting_error);
-+// -- GODOT end --
- 
- /**
-  * Experimental: Mesh simplifier (sloppy)
-diff --git a/thirdparty/meshoptimizer/simplifier.cpp b/thirdparty/meshoptimizer/simplifier.cpp
-index bd523275ce..51cf634186 100644
---- a/thirdparty/meshoptimizer/simplifier.cpp
-+++ b/thirdparty/meshoptimizer/simplifier.cpp
-@@ -1143,7 +1143,10 @@ unsigned int* meshopt_simplifyDebugLoop = 0;
- unsigned int* meshopt_simplifyDebugLoopBack = 0;
- #endif
- 
--size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error)
-+// -- GODOT start --
-+//size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error)
-+size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices, size_t index_count, const float *vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float *r_resulting_error)
-+// -- GODOT end --
- {
- 	using namespace meshopt;
- 
-@@ -1198,10 +1201,13 @@ size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices,
- 	if (result != indices)
- 		memcpy(result, indices, index_count * sizeof(unsigned int));
- 
-+// -- GODOT start --
- #if TRACE
- 	size_t pass_count = 0;
--	float worst_error = 0;
-+	//float worst_error = 0;
- #endif
-+	float worst_error = 0;
-+// -- GODOT end --
- 
- 	Collapse* edge_collapses = allocator.allocate<Collapse>(index_count);
- 	unsigned int* collapse_order = allocator.allocate<unsigned int>(index_count);
-@@ -1213,6 +1219,12 @@ size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices,
- 	// target_error input is linear; we need to adjust it to match quadricError units
- 	float error_limit = target_error * target_error;
- 
-+// -- GODOT start --
-+	if (r_resulting_error) {
-+		*r_resulting_error = 1.0;
-+	}
-+// -- GODOT end --
-+
- 	while (result_count > target_index_count)
- 	{
- 		size_t edge_collapse_count = pickEdgeCollapses(edge_collapses, result, result_count, remap, vertex_kind, loop);
-@@ -1257,7 +1269,8 @@ size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices,
- 		size_t new_count = remapIndexBuffer(result, result_count, collapse_remap);
- 		assert(new_count < result_count);
- 
--#if TRACE
-+// -- GODOT start --
-+//#if TRACE
- 		float pass_error = 0.f;
- 		for (size_t i = 0; i < edge_collapse_count; ++i)
- 		{
-@@ -1267,15 +1280,24 @@ size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices,
- 				pass_error = c.error;
- 		}
- 
--		pass_count++;
-+		//pass_count++;
- 		worst_error = (worst_error < pass_error) ? pass_error : worst_error;
- 
-+#if TRACE
-+		pass_count++;
- 		printf("pass %d: triangles: %d -> %d, collapses: %d/%d (goal: %d), error: %e (limit %e goal %e)\n", int(pass_count), int(result_count / 3), int(new_count / 3), int(collapses), int(edge_collapse_count), int(edge_collapse_goal), pass_error, error_limit, error_goal);
- #endif
-+// -- GODOT end --
- 
- 		result_count = new_count;
- 	}
- 
-+// -- GODOT start --
-+	if (r_resulting_error) {
-+		*r_resulting_error = sqrt(worst_error);
-+	}
-+// -- GODOT end --
-+
- #if TRACE
- 	printf("passes: %d, worst error: %e\n", int(pass_count), worst_error);
- #endif
diff --git a/thirdparty/meshoptimizer/simplifier.cpp b/thirdparty/meshoptimizer/simplifier.cpp
index b195a8cb5d..5205b01172 100644
--- a/thirdparty/meshoptimizer/simplifier.cpp
+++ b/thirdparty/meshoptimizer/simplifier.cpp
@@ -6,7 +6,6 @@
 #include <math.h>
 #include <string.h>
 
-
 #ifndef TRACE
 #define TRACE 0
 #endif
@@ -15,6 +14,12 @@
 #include <stdio.h>
 #endif
 
+#if TRACE
+#define TRACESTATS(i) stats[i]++;
+#else
+#define TRACESTATS(i) (void)0
+#endif
+
 // This work is based on:
 // Michael Garland and Paul S. Heckbert. Surface simplification using quadric error metrics. 1997
 // Michael Garland. Quadric-based polygonal surface simplification. 1999
@@ -26,28 +31,37 @@ namespace meshopt
 
 struct EdgeAdjacency
 {
+	struct Edge
+	{
+		unsigned int next;
+		unsigned int prev;
+	};
+
 	unsigned int* counts;
 	unsigned int* offsets;
-	unsigned int* data;
+	Edge* data;
 };
 
-static void buildEdgeAdjacency(EdgeAdjacency& adjacency, const unsigned int* indices, size_t index_count, size_t vertex_count, meshopt_Allocator& allocator)
+static void prepareEdgeAdjacency(EdgeAdjacency& adjacency, size_t index_count, size_t vertex_count, meshopt_Allocator& allocator)
 {
-	size_t face_count = index_count / 3;
-
-	// allocate arrays
 	adjacency.counts = allocator.allocate<unsigned int>(vertex_count);
 	adjacency.offsets = allocator.allocate<unsigned int>(vertex_count);
-	adjacency.data = allocator.allocate<unsigned int>(index_count);
+	adjacency.data = allocator.allocate<EdgeAdjacency::Edge>(index_count);
+}
+
+static void updateEdgeAdjacency(EdgeAdjacency& adjacency, const unsigned int* indices, size_t index_count, size_t vertex_count, const unsigned int* remap)
+{
+	size_t face_count = index_count / 3;
 
 	// fill edge counts
 	memset(adjacency.counts, 0, vertex_count * sizeof(unsigned int));
 
 	for (size_t i = 0; i < index_count; ++i)
 	{
-		assert(indices[i] < vertex_count);
+		unsigned int v = remap ? remap[indices[i]] : indices[i];
+		assert(v < vertex_count);
 
-		adjacency.counts[indices[i]]++;
+		adjacency.counts[v]++;
 	}
 
 	// fill offset table
@@ -66,9 +80,24 @@ static void buildEdgeAdjacency(EdgeAdjacency& adjacency, const unsigned int* ind
 	{
 		unsigned int a = indices[i * 3 + 0], b = indices[i * 3 + 1], c = indices[i * 3 + 2];
 
-		adjacency.data[adjacency.offsets[a]++] = b;
-		adjacency.data[adjacency.offsets[b]++] = c;
-		adjacency.data[adjacency.offsets[c]++] = a;
+		if (remap)
+		{
+			a = remap[a];
+			b = remap[b];
+			c = remap[c];
+		}
+
+		adjacency.data[adjacency.offsets[a]].next = b;
+		adjacency.data[adjacency.offsets[a]].prev = c;
+		adjacency.offsets[a]++;
+
+		adjacency.data[adjacency.offsets[b]].next = c;
+		adjacency.data[adjacency.offsets[b]].prev = a;
+		adjacency.offsets[b]++;
+
+		adjacency.data[adjacency.offsets[c]].next = a;
+		adjacency.data[adjacency.offsets[c]].prev = b;
+		adjacency.offsets[c]++;
 	}
 
 	// fix offsets that have been disturbed by the previous pass
@@ -209,10 +238,10 @@ const unsigned char kHasOpposite[Kind_Count][Kind_Count] = {
 static bool hasEdge(const EdgeAdjacency& adjacency, unsigned int a, unsigned int b)
 {
 	unsigned int count = adjacency.counts[a];
-	const unsigned int* data = adjacency.data + adjacency.offsets[a];
+	const EdgeAdjacency::Edge* edges = adjacency.data + adjacency.offsets[a];
 
 	for (size_t i = 0; i < count; ++i)
-		if (data[i] == b)
+		if (edges[i].next == b)
 			return true;
 
 	return false;
@@ -234,11 +263,11 @@ static void classifyVertices(unsigned char* result, unsigned int* loop, unsigned
 		unsigned int vertex = unsigned(i);
 
 		unsigned int count = adjacency.counts[vertex];
-		const unsigned int* data = adjacency.data + adjacency.offsets[vertex];
+		const EdgeAdjacency::Edge* edges = adjacency.data + adjacency.offsets[vertex];
 
 		for (size_t j = 0; j < count; ++j)
 		{
-			unsigned int target = data[j];
+			unsigned int target = edges[j].next;
 
 			if (!hasEdge(adjacency, target, vertex))
 			{
@@ -249,10 +278,7 @@ static void classifyVertices(unsigned char* result, unsigned int* loop, unsigned
 	}
 
 #if TRACE
-	size_t lockedstats[4] = {};
-#define TRACELOCKED(i) lockedstats[i]++;
-#else
-#define TRACELOCKED(i) (void)0
+	size_t stats[4] = {};
 #endif
 
 	for (size_t i = 0; i < vertex_count; ++i)
@@ -278,7 +304,7 @@ static void classifyVertices(unsigned char* result, unsigned int* loop, unsigned
 				else
 				{
 					result[i] = Kind_Locked;
-					TRACELOCKED(0);
+					TRACESTATS(0);
 				}
 			}
 			else if (wedge[wedge[i]] == i)
@@ -299,20 +325,20 @@ static void classifyVertices(unsigned char* result, unsigned int* loop, unsigned
 					else
 					{
 						result[i] = Kind_Locked;
-						TRACELOCKED(1);
+						TRACESTATS(1);
 					}
 				}
 				else
 				{
 					result[i] = Kind_Locked;
-					TRACELOCKED(2);
+					TRACESTATS(2);
 				}
 			}
 			else
 			{
 				// more than one vertex maps to this one; we don't have classification available
 				result[i] = Kind_Locked;
-				TRACELOCKED(3);
+				TRACESTATS(3);
 			}
 		}
 		else
@@ -325,7 +351,7 @@ static void classifyVertices(unsigned char* result, unsigned int* loop, unsigned
 
 #if TRACE
 	printf("locked: many open edges %d, disconnected seam %d, many seam edges %d, many wedges %d\n",
-	       int(lockedstats[0]), int(lockedstats[1]), int(lockedstats[2]), int(lockedstats[3]));
+	       int(stats[0]), int(stats[1]), int(stats[2]), int(stats[3]));
 #endif
 }
 
@@ -333,11 +359,8 @@ struct Vector3
 {
 	float x, y, z;
 };
-// -- GODOT start --
-//static void rescalePositions(Vector3* result, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride)
-static float rescalePositions(Vector3* result, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride)
-// -- GODOT end --
 
+static float rescalePositions(Vector3* result, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride)
 {
 	size_t vertex_stride_float = vertex_positions_stride / sizeof(float);
 
@@ -348,9 +371,12 @@ static float rescalePositions(Vector3* result, const float* vertex_positions_dat
 	{
 		const float* v = vertex_positions_data + i * vertex_stride_float;
 
-		result[i].x = v[0];
-		result[i].y = v[1];
-		result[i].z = v[2];
+		if (result)
+		{
+			result[i].x = v[0];
+			result[i].y = v[1];
+			result[i].z = v[2];
+		}
 
 		for (int j = 0; j < 3; ++j)
 		{
@@ -367,18 +393,19 @@ static float rescalePositions(Vector3* result, const float* vertex_positions_dat
 	extent = (maxv[1] - minv[1]) < extent ? extent : (maxv[1] - minv[1]);
 	extent = (maxv[2] - minv[2]) < extent ? extent : (maxv[2] - minv[2]);
 
-	float scale = extent == 0 ? 0.f : 1.f / extent;
-
-	for (size_t i = 0; i < vertex_count; ++i)
+	if (result)
 	{
-		result[i].x = (result[i].x - minv[0]) * scale;
-		result[i].y = (result[i].y - minv[1]) * scale;
-		result[i].z = (result[i].z - minv[2]) * scale;
+		float scale = extent == 0 ? 0.f : 1.f / extent;
+
+		for (size_t i = 0; i < vertex_count; ++i)
+		{
+			result[i].x = (result[i].x - minv[0]) * scale;
+			result[i].y = (result[i].y - minv[1]) * scale;
+			result[i].z = (result[i].z - minv[2]) * scale;
+		}
 	}
-// -- GODOT start --	
-	return extent;
-// -- GODOT end --
 
+	return extent;
 }
 
 struct Quadric
@@ -594,6 +621,48 @@ static void fillEdgeQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
 	}
 }
 
+// does triangle ABC flip when C is replaced with D?
+static bool hasTriangleFlip(const Vector3& a, const Vector3& b, const Vector3& c, const Vector3& d)
+{
+	Vector3 eb = {b.x - a.x, b.y - a.y, b.z - a.z};
+	Vector3 ec = {c.x - a.x, c.y - a.y, c.z - a.z};
+	Vector3 ed = {d.x - a.x, d.y - a.y, d.z - a.z};
+
+	Vector3 nbc = {eb.y * ec.z - eb.z * ec.y, eb.z * ec.x - eb.x * ec.z, eb.x * ec.y - eb.y * ec.x};
+	Vector3 nbd = {eb.y * ed.z - eb.z * ed.y, eb.z * ed.x - eb.x * ed.z, eb.x * ed.y - eb.y * ed.x};
+
+	return nbc.x * nbd.x + nbc.y * nbd.y + nbc.z * nbd.z < 0;
+}
+
+static bool hasTriangleFlips(const EdgeAdjacency& adjacency, const Vector3* vertex_positions, const unsigned int* collapse_remap, unsigned int i0, unsigned int i1)
+{
+	assert(collapse_remap[i0] == i0);
+	assert(collapse_remap[i1] == i1);
+
+	const Vector3& v0 = vertex_positions[i0];
+	const Vector3& v1 = vertex_positions[i1];
+
+	const EdgeAdjacency::Edge* edges = &adjacency.data[adjacency.offsets[i0]];
+	size_t count = adjacency.counts[i0];
+
+	for (size_t i = 0; i < count; ++i)
+	{
+		unsigned int a = collapse_remap[edges[i].next];
+		unsigned int b = collapse_remap[edges[i].prev];
+
+		// skip triangles that get collapsed
+		// note: this is mathematically redundant as if either of these is true, the dot product in hasTriangleFlip should be 0
+		if (a == i1 || b == i1)
+			continue;
+
+		// early-out when at least one triangle flips due to a collapse
+		if (hasTriangleFlip(vertex_positions[a], vertex_positions[b], v0, v1))
+			return true;
+	}
+
+	return false;
+}
+
 static size_t pickEdgeCollapses(Collapse* collapses, const unsigned int* indices, size_t index_count, const unsigned int* remap, const unsigned char* vertex_kind, const unsigned int* loop)
 {
 	size_t collapse_count = 0;
@@ -704,7 +773,7 @@ static void dumpEdgeCollapses(const Collapse* collapses, size_t collapse_count,
 	for (int k0 = 0; k0 < Kind_Count; ++k0)
 		for (int k1 = 0; k1 < Kind_Count; ++k1)
 			if (ckinds[k0][k1])
-				printf("collapses %d -> %d: %d, min error %e\n", k0, k1, int(ckinds[k0][k1]), cerrors[k0][k1]);
+				printf("collapses %d -> %d: %d, min error %e\n", k0, k1, int(ckinds[k0][k1]), ckinds[k0][k1] ? sqrtf(cerrors[k0][k1]) : 0.f);
 }
 
 static void dumpLockedCollapses(const unsigned int* indices, size_t index_count, const unsigned char* vertex_kind)
@@ -772,22 +841,38 @@ static void sortEdgeCollapses(unsigned int* sort_order, const Collapse* collapse
 	}
 }
 
-static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char* collapse_locked, Quadric* vertex_quadrics, const Collapse* collapses, size_t collapse_count, const unsigned int* collapse_order, const unsigned int* remap, const unsigned int* wedge, const unsigned char* vertex_kind, size_t triangle_collapse_goal, float error_goal, float error_limit)
+static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char* collapse_locked, Quadric* vertex_quadrics, const Collapse* collapses, size_t collapse_count, const unsigned int* collapse_order, const unsigned int* remap, const unsigned int* wedge, const unsigned char* vertex_kind, const Vector3* vertex_positions, const EdgeAdjacency& adjacency, size_t triangle_collapse_goal, float error_limit, float& result_error)
 {
 	size_t edge_collapses = 0;
 	size_t triangle_collapses = 0;
 
+	// most collapses remove 2 triangles; use this to establish a bound on the pass in terms of error limit
+	// note that edge_collapse_goal is an estimate; triangle_collapse_goal will be used to actually limit collapses
+	size_t edge_collapse_goal = triangle_collapse_goal / 2;
+
+#if TRACE
+	size_t stats[4] = {};
+#endif
+
 	for (size_t i = 0; i < collapse_count; ++i)
 	{
 		const Collapse& c = collapses[collapse_order[i]];
 
+		TRACESTATS(0);
+
 		if (c.error > error_limit)
 			break;
 
-		if (c.error > error_goal && triangle_collapses > triangle_collapse_goal / 10)
+		if (triangle_collapses >= triangle_collapse_goal)
 			break;
 
-		if (triangle_collapses >= triangle_collapse_goal)
+		// we limit the error in each pass based on the error of optimal last collapse; since many collapses will be locked
+		// as they will share vertices with other successfull collapses, we need to increase the acceptable error by some factor
+		float error_goal = edge_collapse_goal < collapse_count ? 1.5f * collapses[collapse_order[edge_collapse_goal]].error : FLT_MAX;
+
+		// on average, each collapse is expected to lock 6 other collapses; to avoid degenerate passes on meshes with odd
+		// topology, we only abort if we got over 1/6 collapses accordingly.
+		if (c.error > error_goal && triangle_collapses > triangle_collapse_goal / 6)
 			break;
 
 		unsigned int i0 = c.v0;
@@ -800,7 +885,19 @@ static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char*
 		// it's important to not move the vertices twice since it complicates the tracking/remapping logic
 		// it's important to not move other vertices towards a moved vertex to preserve error since we don't re-rank collapses mid-pass
 		if (collapse_locked[r0] | collapse_locked[r1])
+		{
+			TRACESTATS(1);
+			continue;
+		}
+
+		if (hasTriangleFlips(adjacency, vertex_positions, collapse_remap, r0, r1))
+		{
+			// adjust collapse goal since this collapse is invalid and shouldn't factor into error goal
+			edge_collapse_goal++;
+
+			TRACESTATS(2);
 			continue;
+		}
 
 		assert(collapse_remap[r0] == r0);
 		assert(collapse_remap[r1] == r1);
@@ -842,8 +939,18 @@ static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char*
 		// border edges collapse 1 triangle, other edges collapse 2 or more
 		triangle_collapses += (vertex_kind[i0] == Kind_Border) ? 1 : 2;
 		edge_collapses++;
+
+		result_error = result_error < c.error ? c.error : result_error;
 	}
 
+#if TRACE
+	float error_goal_perfect = edge_collapse_goal < collapse_count ? collapses[collapse_order[edge_collapse_goal]].error : 0.f;
+
+	printf("removed %d triangles, error %e (goal %e); evaluated %d/%d collapses (done %d, skipped %d, invalid %d)\n",
+		int(triangle_collapses), sqrtf(result_error), sqrtf(error_goal_perfect),
+		int(stats[0]), int(collapse_count), int(edge_collapses), int(stats[1]), int(stats[2]));
+#endif
+
 	return edge_collapses;
 }
 
@@ -1151,10 +1258,7 @@ unsigned int* meshopt_simplifyDebugLoop = 0;
 unsigned int* meshopt_simplifyDebugLoopBack = 0;
 #endif
 
-// -- GODOT start --
-//size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error)
-size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices, size_t index_count, const float *vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float *r_resulting_error)
-// -- GODOT end --
+size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions_data, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float* out_result_error)
 {
 	using namespace meshopt;
 
@@ -1169,7 +1273,8 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 
 	// build adjacency information
 	EdgeAdjacency adjacency = {};
-	buildEdgeAdjacency(adjacency, indices, index_count, vertex_count, allocator);
+	prepareEdgeAdjacency(adjacency, index_count, vertex_count, allocator);
+	updateEdgeAdjacency(adjacency, indices, index_count, vertex_count, NULL);
 
 	// build position remap that maps each vertex to the one with identical position
 	unsigned int* remap = allocator.allocate<unsigned int>(vertex_count);
@@ -1198,10 +1303,7 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 #endif
 
 	Vector3* vertex_positions = allocator.allocate<Vector3>(vertex_count);
-// -- GODOT start --
-	//rescalePositions(vertex_positions, vertex_positions_data, vertex_count, vertex_positions_stride);	
-	float extent = rescalePositions(vertex_positions, vertex_positions_data, vertex_count, vertex_positions_stride);
-// -- GODOT end --
+	rescalePositions(vertex_positions, vertex_positions_data, vertex_count, vertex_positions_stride);
 
 	Quadric* vertex_quadrics = allocator.allocate<Quadric>(vertex_count);
 	memset(vertex_quadrics, 0, vertex_count * sizeof(Quadric));
@@ -1212,13 +1314,9 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 	if (result != indices)
 		memcpy(result, indices, index_count * sizeof(unsigned int));
 
-// -- GODOT start --
 #if TRACE
 	size_t pass_count = 0;
-	//float worst_error = 0;
 #endif
-	float worst_error = 0;
-// -- GODOT end --
 
 	Collapse* edge_collapses = allocator.allocate<Collapse>(index_count);
 	unsigned int* collapse_order = allocator.allocate<unsigned int>(index_count);
@@ -1226,18 +1324,16 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 	unsigned char* collapse_locked = allocator.allocate<unsigned char>(vertex_count);
 
 	size_t result_count = index_count;
+	float result_error = 0;
 
 	// target_error input is linear; we need to adjust it to match quadricError units
 	float error_limit = target_error * target_error;
 
-// -- GODOT start --
-	if (r_resulting_error) {
-		*r_resulting_error = 1.0;
-	}
-// -- GODOT end --
-
 	while (result_count > target_index_count)
 	{
+		// note: throughout the simplification process adjacency structure reflects welded topology for result-in-progress
+		updateEdgeAdjacency(adjacency, result, result_count, vertex_count, remap);
+
 		size_t edge_collapse_count = pickEdgeCollapses(edge_collapses, result, result_count, remap, vertex_kind, loop);
 
 		// no edges can be collapsed any more due to topology restrictions
@@ -1252,23 +1348,18 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 
 		sortEdgeCollapses(collapse_order, edge_collapses, edge_collapse_count);
 
-		// most collapses remove 2 triangles; use this to establish a bound on the pass in terms of error limit
-		// note that edge_collapse_goal is an estimate; triangle_collapse_goal will be used to actually limit collapses
 		size_t triangle_collapse_goal = (result_count - target_index_count) / 3;
-		size_t edge_collapse_goal = triangle_collapse_goal / 2;
-
-		// we limit the error in each pass based on the error of optimal last collapse; since many collapses will be locked
-		// as they will share vertices with other successfull collapses, we need to increase the acceptable error by this factor
-		const float kPassErrorBound = 1.5f;
-
-		float error_goal = edge_collapse_goal < edge_collapse_count ? edge_collapses[collapse_order[edge_collapse_goal]].error * kPassErrorBound : FLT_MAX;
 
 		for (size_t i = 0; i < vertex_count; ++i)
 			collapse_remap[i] = unsigned(i);
 
 		memset(collapse_locked, 0, vertex_count);
 
-		size_t collapses = performEdgeCollapses(collapse_remap, collapse_locked, vertex_quadrics, edge_collapses, edge_collapse_count, collapse_order, remap, wedge, vertex_kind, triangle_collapse_goal, error_goal, error_limit);
+#if TRACE
+		printf("pass %d: ", int(pass_count++));
+#endif
+
+		size_t collapses = performEdgeCollapses(collapse_remap, collapse_locked, vertex_quadrics, edge_collapses, edge_collapse_count, collapse_order, remap, wedge, vertex_kind, vertex_positions, adjacency, triangle_collapse_goal, error_limit, result_error);
 
 		// no edges can be collapsed any more due to hitting the error limit or triangle collapse limit
 		if (collapses == 0)
@@ -1280,37 +1371,11 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 		size_t new_count = remapIndexBuffer(result, result_count, collapse_remap);
 		assert(new_count < result_count);
 
-// -- GODOT start --
-//#if TRACE
-		float pass_error = 0.f;
-		for (size_t i = 0; i < edge_collapse_count; ++i)
-		{
-			Collapse& c = edge_collapses[collapse_order[i]];
-
-			if (collapse_remap[c.v0] == c.v1)
-				pass_error = c.error;
-		}
-
-		//pass_count++;
-		worst_error = (worst_error < pass_error) ? pass_error : worst_error;
-
-#if TRACE
-		pass_count++;
-		printf("pass %d: triangles: %d -> %d, collapses: %d/%d (goal: %d), error: %e (limit %e goal %e)\n", int(pass_count), int(result_count / 3), int(new_count / 3), int(collapses), int(edge_collapse_count), int(edge_collapse_goal), pass_error, error_limit, error_goal);
-#endif
-// -- GODOT end --
-
 		result_count = new_count;
 	}
 
-// -- GODOT start --
-	if (r_resulting_error) {
-		*r_resulting_error = sqrt(worst_error) * extent;
-	}
-// -- GODOT end --
-
 #if TRACE
-	printf("passes: %d, worst error: %e\n", int(pass_count), worst_error);
+	printf("result: %d triangles, error: %e; total %d passes\n", int(result_count), sqrtf(result_error), int(pass_count));
 #endif
 
 #if TRACE > 1
@@ -1328,6 +1393,10 @@ size_t meshopt_simplify(unsigned int *destination, const unsigned int *indices,
 		memcpy(meshopt_simplifyDebugLoopBack, loopback, vertex_count * sizeof(unsigned int));
 #endif
 
+	// result_error is quadratic; we need to remap it back to linear
+	if (out_result_error)
+		*out_result_error = sqrtf(result_error);
+
 	return result_count;
 }
 
@@ -1560,3 +1629,15 @@ size_t meshopt_simplifyPoints(unsigned int* destination, const float* vertex_pos
 
 	return cell_count;
 }
+
+float meshopt_simplifyScale(const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
+{
+	using namespace meshopt;
+
+	assert(vertex_positions_stride > 0 && vertex_positions_stride <= 256);
+	assert(vertex_positions_stride % sizeof(float) == 0);
+
+	float extent = rescalePositions(NULL, vertex_positions, vertex_count, vertex_positions_stride);
+
+	return extent;
+}
diff --git a/thirdparty/meshoptimizer/vertexcodec.cpp b/thirdparty/meshoptimizer/vertexcodec.cpp
index 784c9a13db..2cbfaac367 100644
--- a/thirdparty/meshoptimizer/vertexcodec.cpp
+++ b/thirdparty/meshoptimizer/vertexcodec.cpp
@@ -80,14 +80,6 @@
 #include <wasm_simd128.h>
 #endif
 
-#ifndef TRACE
-#define TRACE 0
-#endif
-
-#if TRACE
-#include <stdio.h>
-#endif
-
 #ifdef SIMD_WASM
 #define wasmx_splat_v32x4(v, i) wasm_v32x4_shuffle(v, v, i, i, i, i)
 #define wasmx_unpacklo_v8x16(a, b) wasm_v8x16_shuffle(a, b, 0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23)
@@ -133,19 +125,6 @@ inline unsigned char unzigzag8(unsigned char v)
 	return -(v & 1) ^ (v >> 1);
 }
 
-#if TRACE
-struct Stats
-{
-	size_t size;
-	size_t header;
-	size_t bitg[4];
-	size_t bitb[4];
-};
-
-Stats* bytestats;
-Stats vertexstats[256];
-#endif
-
 static bool encodeBytesGroupZero(const unsigned char* buffer)
 {
 	for (size_t i = 0; i < kByteGroupSize; ++i)
@@ -267,17 +246,8 @@ static unsigned char* encodeBytes(unsigned char* data, unsigned char* data_end,
 
 		assert(data + best_size == next);
 		data = next;
-
-#if TRACE > 1
-		bytestats->bitg[bitslog2]++;
-		bytestats->bitb[bitslog2] += best_size;
-#endif
 	}
 
-#if TRACE > 1
-	bytestats->header += header_size;
-#endif
-
 	return data;
 }
 
@@ -306,19 +276,9 @@ static unsigned char* encodeVertexBlock(unsigned char* data, unsigned char* data
 			vertex_offset += vertex_size;
 		}
 
-#if TRACE
-		const unsigned char* olddata = data;
-		bytestats = &vertexstats[k];
-#endif
-
 		data = encodeBytes(data, data_end, buffer, (vertex_count + kByteGroupSize - 1) & ~(kByteGroupSize - 1));
 		if (!data)
 			return 0;
-
-#if TRACE
-		bytestats = 0;
-		vertexstats[k].size += data - olddata;
-#endif
 	}
 
 	memcpy(last_vertex, &vertex_data[vertex_size * (vertex_count - 1)], vertex_size);
@@ -1086,10 +1046,6 @@ size_t meshopt_encodeVertexBuffer(unsigned char* buffer, size_t buffer_size, con
 	assert(vertex_size > 0 && vertex_size <= 256);
 	assert(vertex_size % 4 == 0);
 
-#if TRACE
-	memset(vertexstats, 0, sizeof(vertexstats));
-#endif
-
 	const unsigned char* vertex_data = static_cast<const unsigned char*>(vertices);
 
 	unsigned char* data = buffer;
@@ -1142,28 +1098,6 @@ size_t meshopt_encodeVertexBuffer(unsigned char* buffer, size_t buffer_size, con
 	assert(data >= buffer + tail_size);
 	assert(data <= buffer + buffer_size);
 
-#if TRACE
-	size_t total_size = data - buffer;
-
-	for (size_t k = 0; k < vertex_size; ++k)
-	{
-		const Stats& vsk = vertexstats[k];
-
-		printf("%2d: %d bytes\t%.1f%%\t%.1f bpv", int(k), int(vsk.size), double(vsk.size) / double(total_size) * 100, double(vsk.size) / double(vertex_count) * 8);
-
-#if TRACE > 1
-		printf("\t\thdr %d bytes\tbit0 %d (%d bytes)\tbit1 %d (%d bytes)\tbit2 %d (%d bytes)\tbit3 %d (%d bytes)",
-		       int(vsk.header),
-		       int(vsk.bitg[0]), int(vsk.bitb[0]),
-		       int(vsk.bitg[1]), int(vsk.bitb[1]),
-		       int(vsk.bitg[2]), int(vsk.bitb[2]),
-		       int(vsk.bitg[3]), int(vsk.bitb[3]));
-#endif
-
-		printf("\n");
-	}
-#endif
-
 	return data - buffer;
 }
 
diff --git a/thirdparty/meshoptimizer/vertexfilter.cpp b/thirdparty/meshoptimizer/vertexfilter.cpp
index e7ad2c9d39..39946f46ed 100644
--- a/thirdparty/meshoptimizer/vertexfilter.cpp
+++ b/thirdparty/meshoptimizer/vertexfilter.cpp
@@ -2,6 +2,7 @@
 #include "meshoptimizer.h"
 
 #include <math.h>
+#include <string.h>
 
 // The block below auto-detects SIMD ISA that can be used on the target platform
 #ifndef MESHOPTIMIZER_NO_SIMD
@@ -159,6 +160,25 @@ static void decodeFilterExp(unsigned int* data, size_t count)
 #endif
 
 #if defined(SIMD_SSE) || defined(SIMD_NEON) || defined(SIMD_WASM)
+template <typename T> static void dispatchSimd(void (*process)(T*, size_t), T* data, size_t count, size_t stride)
+{
+	assert(stride <= 4);
+
+	size_t count4 = count & ~size_t(3);
+	process(data, count4);
+
+	if (count4 < count)
+	{
+		T tail[4 * 4] = {}; // max stride 4, max count 4
+		size_t tail_size = (count - count4) * stride * sizeof(T);
+		assert(tail_size <= sizeof(tail));
+
+		memcpy(tail, data + count4 * stride, tail_size);
+		process(tail, count - count4);
+		memcpy(data + count4 * stride, tail, tail_size);
+	}
+}
+
 inline uint64_t rotateleft64(uint64_t v, int x)
 {
 #if defined(_MSC_VER) && !defined(__clang__)
@@ -775,14 +795,13 @@ void meshopt_decodeFilterOct(void* buffer, size_t vertex_count, size_t vertex_si
 {
 	using namespace meshopt;
 
-	assert(vertex_count % 4 == 0);
 	assert(vertex_size == 4 || vertex_size == 8);
 
 #if defined(SIMD_SSE) || defined(SIMD_NEON) || defined(SIMD_WASM)
 	if (vertex_size == 4)
-		decodeFilterOctSimd(static_cast<signed char*>(buffer), vertex_count);
+		dispatchSimd(decodeFilterOctSimd, static_cast<signed char*>(buffer), vertex_count, 4);
 	else
-		decodeFilterOctSimd(static_cast<short*>(buffer), vertex_count);
+		dispatchSimd(decodeFilterOctSimd, static_cast<short*>(buffer), vertex_count, 4);
 #else
 	if (vertex_size == 4)
 		decodeFilterOct(static_cast<signed char*>(buffer), vertex_count);
@@ -795,12 +814,11 @@ void meshopt_decodeFilterQuat(void* buffer, size_t vertex_count, size_t vertex_s
 {
 	using namespace meshopt;
 
-	assert(vertex_count % 4 == 0);
 	assert(vertex_size == 8);
 	(void)vertex_size;
 
 #if defined(SIMD_SSE) || defined(SIMD_NEON) || defined(SIMD_WASM)
-	decodeFilterQuatSimd(static_cast<short*>(buffer), vertex_count);
+	dispatchSimd(decodeFilterQuatSimd, static_cast<short*>(buffer), vertex_count, 4);
 #else
 	decodeFilterQuat(static_cast<short*>(buffer), vertex_count);
 #endif
@@ -810,11 +828,10 @@ void meshopt_decodeFilterExp(void* buffer, size_t vertex_count, size_t vertex_si
 {
 	using namespace meshopt;
 
-	assert(vertex_count % 4 == 0);
 	assert(vertex_size % 4 == 0);
 
 #if defined(SIMD_SSE) || defined(SIMD_NEON) || defined(SIMD_WASM)
-	decodeFilterExpSimd(static_cast<unsigned int*>(buffer), vertex_count * (vertex_size / 4));
+	dispatchSimd(decodeFilterExpSimd, static_cast<unsigned int*>(buffer), vertex_count * (vertex_size / 4), 1);
 #else
 	decodeFilterExp(static_cast<unsigned int*>(buffer), vertex_count * (vertex_size / 4));
 #endif