7 files changed, 39 insertions, 35 deletions
diff --git a/indra/cmake/00-Common.cmake b/indra/cmake/00-Common.cmake
index 113e21a715..2cc8fa5e5f 100644
--- a/indra/cmake/00-Common.cmake
+++ b/indra/cmake/00-Common.cmake
@@ -169,6 +169,7 @@ if (LINUX)
     add_definitions(-fvisibility=hidden)
     # don't catch SIGCHLD in our base application class for the viewer - some of our 3rd party libs may need their *own* SIGCHLD handler to work.  Sigh!  The viewer doesn't need to catch SIGCHLD anyway.
     add_definitions(-DLL_IGNORE_SIGCHLD)
+    add_definitions(-march=pentium3 -mfpmath=sse -ftree-vectorize)
     if (NOT STANDALONE)
       # this stops us requiring a really recent glibc at runtime
       add_definitions(-fno-stack-protector)
diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp
index 2a8015e26d..5c6ca30cdd 100644
--- a/indra/llcommon/llmemory.cpp
+++ b/indra/llcommon/llmemory.cpp
@@ -73,25 +73,6 @@ void LLMemory::freeReserve()
 	reserveMem = NULL;
 }
 
-void* ll_allocate (size_t size)
-{
-	if (size == 0)
-	{
-		llwarns << "Null allocation" << llendl;
-	}
-	void *p = malloc(size);
-	if (p == NULL)
-	{
-		LLMemory::freeReserve();
-		llerrs << "Out of memory Error" << llendl;
-	}
-	return p;
-}
-
-void ll_release (void *p)
-{
-	free(p);
-}
 
 //----------------------------------------------------------------------------
 
diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h
index 1c6f64dd8b..f0813fb4ee 100644
--- a/indra/llcommon/llmemory.h
+++ b/indra/llcommon/llmemory.h
@@ -32,14 +32,33 @@
 #ifndef LLMEMORY_H
 #define LLMEMORY_H
 
+#include <stdlib.h>
 
+inline void* ll_aligned_malloc(size_t size, size_t alignment = 16) // alignment MUST be power-of-two multiple of sizeof(void*).   returned hunk MUST be freed with ll_aligned_free().
+{
+#if defined(LL_WINDOWS)
+	return _mm_malloc(size, alignment);
+#else
+	void *rtn;
+	if (LL_LIKELY(0 == posix_memalign(&rtn, alignment, size)))
+	{
+		return rtn;
+	}
+	else // bad alignment requested, or out of memory
+	{
+		return NULL;
+	}
+#endif
+}
 
-extern S32 gTotalDAlloc;
-extern S32 gTotalDAUse;
-extern S32 gDACount;
-
-extern void* ll_allocate (size_t size);
-extern void ll_release (void *p);
+inline void ll_aligned_free(void *p)
+{
+#if defined(LL_WINDOWS)
+	_mm_free(p);
+#else
+	free(p); // posix_memalign() is compatible with heap deallocator
+#endif
+}
 
 class LL_COMMON_API LLMemory
 {
diff --git a/indra/newview/lldrawpoolbump.cpp b/indra/newview/lldrawpoolbump.cpp
index 1fae8e518c..b7092f32a2 100644
--- a/indra/newview/lldrawpoolbump.cpp
+++ b/indra/newview/lldrawpoolbump.cpp
@@ -82,7 +82,7 @@ static S32 bump_channel = -1;
 // static 
 void LLStandardBumpmap::init()
 {
-	LLStandardBumpmap::restoreGL();
+	// do nothing
 }
 
 // static 
@@ -876,6 +876,8 @@ void LLBumpImageList::clear()
 	// these will be re-populated on-demand
 	mBrightnessEntries.clear();
 	mDarknessEntries.clear();
+
+	LLStandardBumpmap::clear();
 }
 
 void LLBumpImageList::shutdown()
@@ -892,8 +894,8 @@ void LLBumpImageList::destroyGL()
 
 void LLBumpImageList::restoreGL()
 {
-	// Images will be recreated as they are needed.
 	LLStandardBumpmap::restoreGL();
+	// Images will be recreated as they are needed.
 }
 
 
diff --git a/indra/newview/llpolymesh.cpp b/indra/newview/llpolymesh.cpp
index 98c0191397..d0164c1033 100644
--- a/indra/newview/llpolymesh.cpp
+++ b/indra/newview/llpolymesh.cpp
@@ -35,7 +35,8 @@
 //-----------------------------------------------------------------------------
 #include "llviewerprecompiledheaders.h"
 
-#include "llpolymesh.h"
+#include "llfasttimer.h"
+#include "llmemory.h"
 
 #include "llviewercontrol.h"
 #include "llxmltree.h"
@@ -45,7 +46,7 @@
 #include "llvolume.h"
 #include "llendianswizzle.h"
 
-#include "llfasttimer.h"
+#include "llpolymesh.h"
 
 #define HEADER_ASCII "Linden Mesh 1.0"
 #define HEADER_BINARY "Linden Binary Mesh 1.0"
@@ -715,7 +716,7 @@ LLPolyMesh::LLPolyMesh(LLPolyMeshSharedData *shared_data, LLPolyMesh *reference_
 		int nfloats = nverts * (2*4 + 3*3 + 2 + 4);
 
 		//use aligned vertex data to make LLPolyMesh SSE friendly
-		mVertexData = (F32*) _mm_malloc(nfloats*4, 16);
+		mVertexData = (F32*) ll_aligned_malloc(nfloats*4, 16);
 		int offset = 0;
 
 		//all members must be 16-byte aligned except the last 3
@@ -766,7 +767,7 @@ LLPolyMesh::~LLPolyMesh()
 	delete [] mClothingWeights;
 	delete [] mTexCoords;
 #else
-	_mm_free(mVertexData);
+	ll_aligned_free(mVertexData);
 #endif
 }
 
diff --git a/indra/newview/llviewerjointmesh_sse.cpp b/indra/newview/llviewerjointmesh_sse.cpp
index e586b910cd..4fb5fafad1 100644
--- a/indra/newview/llviewerjointmesh_sse.cpp
+++ b/indra/newview/llviewerjointmesh_sse.cpp
@@ -94,8 +94,8 @@ void LLViewerJointMesh::updateGeometrySSE(LLFace *face, LLPolyMesh *mesh)
 	buffer->getNormalStrider(o_normals,   mesh->mFaceVertexOffset);
 
 	const F32*			weights			= mesh->getWeights();
-	const LLVector3*	coords			= mesh->getCoords();
-	const LLVector3*	normals			= mesh->getNormals();
+	const LLVector3*	coords			= (const LLVector3*)mesh->getCoords();
+	const LLVector3*	normals			= (const LLVector3*)mesh->getNormals();
 	for (U32 index = 0, index_end = mesh->getNumVertices(); index < index_end; ++index)
 	{
 		if( weight != weights[index])
diff --git a/indra/newview/llviewerjointmesh_sse2.cpp b/indra/newview/llviewerjointmesh_sse2.cpp
index 550044c321..58cd75871d 100644
--- a/indra/newview/llviewerjointmesh_sse2.cpp
+++ b/indra/newview/llviewerjointmesh_sse2.cpp
@@ -101,8 +101,8 @@ void LLViewerJointMesh::updateGeometrySSE2(LLFace *face, LLPolyMesh *mesh)
 	buffer->getNormalStrider(o_normals,   mesh->mFaceVertexOffset);
 
 	const F32*			weights			= mesh->getWeights();
-	const LLVector3*	coords			= mesh->getCoords();
-	const LLVector3*	normals			= mesh->getNormals();
+	const LLVector3*	coords			= (const LLVector3*)mesh->getCoords();
+	const LLVector3*	normals			= (const LLVector3*)mesh->getNormals();
 	for (U32 index = 0, index_end = mesh->getNumVertices(); index < index_end; ++index)
 	{
 		if( weight != weights[index])