finish conversion to ll_aligned_*() wrappers

6 files changed, 49 insertions, 27 deletions

diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h
index 93dd7a7fe3..1c8c91f57e 100644
--- a/indra/llcommon/llmemory.h
+++ b/indra/llcommon/llmemory.h
@@ -43,13 +43,9 @@ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed wi
 #else
 	void *rtn;
 	if (LL_LIKELY(0 == posix_memalign(&rtn, 16, size)))
-	{
 		return rtn;
-	}
 	else // bad alignment requested, or out of memory
-	{
 		return NULL;
-	}
 #endif
 }
 
@@ -64,6 +60,32 @@ inline void ll_aligned_free_16(void *p)
 #endif
 }
 
+inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed with ll_aligned_free_32().
+{
+#if defined(LL_WINDOWS)
+	return _mm_malloc(size, 32);
+#elif defined(LL_DARWIN)
+# error implement me.
+#else
+	void *rtn;
+	if (LL_LIKELY(0 == posix_memalign(&rtn, 32, size)))
+		return rtn;
+	else // bad alignment requested, or out of memory
+		return NULL;
+#endif
+}
+
+inline void ll_aligned_free_32(void *p)
+{
+#if defined(LL_WINDOWS)
+	_mm_free(p);
+#elif defined(LL_DARWIN)
+# error implement me.
+#else
+	free(p); // posix_memalign() is compatible with heap deallocator
+#endif
+}
+
 class LL_COMMON_API LLMemory
 {
 public:
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 3f70ccacb1..70601663e6 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -777,8 +777,8 @@ LLRender::LLRender()
 	mTexcoordsp = tc.get();
 	mColorsp = color.get();
 
-	mUIOffset = (LLVector4a*) _mm_malloc(LL_MAX_UI_STACK_DEPTH*sizeof(LLVector4a), 16);
-	mUIScale = (LLVector4a*) _mm_malloc(LL_MAX_UI_STACK_DEPTH*sizeof(LLVector4a), 16);
+	mUIOffset = (LLVector4a*) ll_aligned_malloc_16(LL_MAX_UI_STACK_DEPTH*sizeof(LLVector4a));
+	mUIScale = (LLVector4a*) ll_aligned_malloc_16(LL_MAX_UI_STACK_DEPTH*sizeof(LLVector4a));
 		
 	mTexUnits.reserve(LL_NUM_TEXTURE_LAYERS);
 	for (U32 i = 0; i < LL_NUM_TEXTURE_LAYERS; i++)
@@ -816,9 +816,9 @@ void LLRender::shutdown()
 	delete mDummyTexUnit;
 	mDummyTexUnit = NULL;
 
-	_mm_free(mUIOffset);
+	ll_aligned_free_16(mUIOffset);
 	mUIOffset = NULL;
-	_mm_free(mUIScale);
+	ll_aligned_free_16(mUIScale);
 	mUIScale = NULL;
 }
 
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 4f2dae0cdf..ff966f1e95 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -627,7 +627,7 @@ void LLVertexBuffer::createGLBuffer()
 	{
 		static int gl_buffer_idx = 0;
 		mGLBuffer = ++gl_buffer_idx;
-		mMappedData = (U8*) _mm_malloc(size, 16);
+		mMappedData = (U8*) ll_aligned_malloc_16(size);
 		memset(mMappedData, 0, size);
 	}
 }
@@ -662,7 +662,7 @@ void LLVertexBuffer::createGLIndices()
 	}
 	else
 	{
-		mMappedIndexData = (U8*) _mm_malloc(size, 16);
+		mMappedIndexData = (U8*) ll_aligned_malloc_16(size);
 		static int gl_buffer_idx = 0;
 		mGLIndices = ++gl_buffer_idx;
 	}
@@ -683,7 +683,7 @@ void LLVertexBuffer::destroyGLBuffer()
 		}
 		else
 		{
-			_mm_free(mMappedData);
+			ll_aligned_free_16(mMappedData);
 			mMappedData = NULL;
 			mEmpty = TRUE;
 		}
@@ -710,7 +710,7 @@ void LLVertexBuffer::destroyGLIndices()
 		}
 		else
 		{
-			_mm_free(mMappedIndexData);
+			ll_aligned_free_16(mMappedIndexData);
 			mMappedIndexData = NULL;
 			mEmpty = TRUE;
 		}
@@ -846,8 +846,8 @@ void LLVertexBuffer::resizeBuffer(S32 newnverts, S32 newnindices)
 			{
 				if (!useVBOs())
 				{
-					_mm_free(mMappedData);
-					mMappedData = (U8*) _mm_malloc(newsize, 16);
+					ll_aligned_free_16(mMappedData);
+					mMappedData = (U8*) ll_aligned_malloc_16(newsize);
 				}
 				mResized = TRUE;
 			}
@@ -867,8 +867,8 @@ void LLVertexBuffer::resizeBuffer(S32 newnverts, S32 newnindices)
 			{
 				if (!useVBOs())
 				{
-					_mm_free(mMappedIndexData);
-					mMappedIndexData = (U8*) _mm_malloc(new_index_size, 16);
+					ll_aligned_free_16(mMappedIndexData);
+					mMappedIndexData = (U8*) ll_aligned_malloc_16(new_index_size);
 				}
 				mResized = TRUE;
 			}
diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp
index 04e433dcfd..96ec51300f 100644
--- a/indra/newview/lldrawable.cpp
+++ b/indra/newview/lldrawable.cpp
@@ -95,7 +95,7 @@ void LLDrawable::incrementVisible()
 
 void LLDrawable::init()
 {
-	mExtents = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*3, 32);
+	mExtents = (LLVector4a*) ll_aligned_malloc_32(sizeof(LLVector4a)*3, 32);
 	mPositionGroup = mExtents + 2;
 
 	// mXform
@@ -150,7 +150,7 @@ void LLDrawable::destroy()
 		llinfos << "- Zombie drawables: " << sNumZombieDrawables << llendl;
 	}*/	
 
-	_mm_free(mExtents);
+	ll_aligned_free_32(mExtents);
 	mExtents = mPositionGroup = NULL;
 }
 
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index 0b5cf78261..9b5a1eb07c 100644
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -152,7 +152,7 @@ void cylindricalProjection(LLVector2 &tc, const LLVector4a& normal, const LLVect
 
 void LLFace::init(LLDrawable* drawablep, LLViewerObject* objp)
 {
-	mExtents = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*2, 16);
+	mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*2);
 
 	mLastUpdateTime = gFrameTimeSeconds;
 	mLastMoveTime = 0.f;
@@ -269,7 +269,7 @@ void LLFace::destroy()
 		mVObjp = NULL;
 	}
 
-	_mm_free(mExtents);
+	ll_aligned_free_16(mExtents);
 	mExtents = NULL;
 }
 
diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp
index 05e7bbb6cd..e5d0e064c6 100644
--- a/indra/newview/llspatialpartition.cpp
+++ b/indra/newview/llspatialpartition.cpp
@@ -239,7 +239,7 @@ void LLSpatialGroup::buildOcclusion()
 {
 	if (!mOcclusionVerts)
 	{
-		mOcclusionVerts = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*8, 16);
+		mOcclusionVerts = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*8);
 	}
 
 	LLVector4a fudge;
@@ -342,13 +342,13 @@ LLSpatialGroup::~LLSpatialGroup()
 		sQueryPool.release(mOcclusionQuery[LLViewerCamera::sCurCameraID]);
 	}
 
-	_mm_free(mOcclusionVerts);
+	ll_aligned_free_16(mOcclusionVerts);
 
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 	clearDrawMap();
 	clearAtlasList() ;
 
-	_mm_free(mBounds);
+	ll_aligned_free_16(mBounds);
 }
 
 BOOL LLSpatialGroup::hasAtlas(LLTextureAtlas* atlasp)
@@ -1163,7 +1163,7 @@ LLSpatialGroup::LLSpatialGroup(OctreeNode* node, LLSpatialPartition* part) :
 	sNodeCount++;
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 
-	mBounds = (LLVector4a*) _mm_malloc(sizeof(LLVector4a) * V4_COUNT, 16);
+	mBounds = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a) * V4_COUNT);
 	mExtents = mBounds + EXTENTS;
 	mObjectBounds = mBounds + OBJECT_BOUNDS;
 	mObjectExtents = mBounds + OBJECT_EXTENTS;
@@ -1434,7 +1434,7 @@ void LLSpatialGroup::destroyGL()
 		}
 	}
 
-	_mm_free(mOcclusionVerts);
+	ll_aligned_free_16(mOcclusionVerts);
 	mOcclusionVerts = NULL;
 
 	for (LLSpatialGroup::element_iter i = getData().begin(); i != getData().end(); ++i)
@@ -3502,7 +3502,7 @@ LLDrawInfo::LLDrawInfo(U16 start, U16 end, U32 count, U32 offset,
 	mDrawMode(LLRender::TRIANGLES)
 {
 	mVertexBuffer->validateRange(mStart, mEnd, mCount, mOffset);
-	mExtents = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*2, 16);
+	mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*2);
 
 	mDebugColor = (rand() << 16) + rand();
 }
@@ -3524,7 +3524,7 @@ LLDrawInfo::~LLDrawInfo()
 		gPipeline.checkReferences(this);
 	}
 
-	_mm_free(mExtents);
+	ll_aligned_free_16(mExtents);
 }
 
 void LLDrawInfo::validate()