summaryrefslogtreecommitdiff
path: root/indra/llmath
diff options
context:
space:
mode:
Diffstat (limited to 'indra/llmath')
-rw-r--r--indra/llmath/llsimdmath.h28
-rw-r--r--indra/llmath/llvector4a.cpp50
-rw-r--r--indra/llmath/llvector4a.h4
-rw-r--r--indra/llmath/llvolume.cpp104
-rw-r--r--indra/llmath/llvolume.h21
5 files changed, 14 insertions, 193 deletions
diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h
index 01458521ec..cebd2ace7d 100644
--- a/indra/llmath/llsimdmath.h
+++ b/indra/llmath/llsimdmath.h
@@ -39,34 +39,6 @@
#include <stdint.h>
#endif
-template <typename T> T* LL_NEXT_ALIGNED_ADDRESS(T* address)
-{
- return reinterpret_cast<T*>(
- (reinterpret_cast<uintptr_t>(address) + 0xF) & ~0xF);
-}
-
-template <typename T> T* LL_NEXT_ALIGNED_ADDRESS_64(T* address)
-{
- return reinterpret_cast<T*>(
- (reinterpret_cast<uintptr_t>(address) + 0x3F) & ~0x3F);
-}
-
-#if LL_LINUX || LL_DARWIN
-
-#define LL_ALIGN_PREFIX(x)
-#define LL_ALIGN_POSTFIX(x) __attribute__((aligned(x)))
-
-#elif LL_WINDOWS
-
-#define LL_ALIGN_PREFIX(x) __declspec(align(x))
-#define LL_ALIGN_POSTFIX(x)
-
-#else
-#error "LL_ALIGN_PREFIX and LL_ALIGN_POSTFIX undefined"
-#endif
-
-#define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16)
-
#include <xmmintrin.h>
#include <emmintrin.h>
diff --git a/indra/llmath/llvector4a.cpp b/indra/llmath/llvector4a.cpp
index 6edeb0fefe..570fa41a43 100644
--- a/indra/llmath/llvector4a.cpp
+++ b/indra/llmath/llvector4a.cpp
@@ -41,55 +41,7 @@ extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F
/*static */void LLVector4a::memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes)
{
- assert(src != NULL);
- assert(dst != NULL);
- assert(bytes > 0);
- assert((bytes % sizeof(F32))== 0);
- ll_assert_aligned(src,16);
- ll_assert_aligned(dst,16);
- assert(bytes%16==0);
-
- F32* end = dst + (bytes / sizeof(F32) );
-
- if (bytes > 64)
- {
- F32* begin_64 = LL_NEXT_ALIGNED_ADDRESS_64(dst);
-
- //at least 64 (16*4) bytes before the end of the destination, switch to 16 byte copies
- F32* end_64 = end-16;
-
- _mm_prefetch((char*)begin_64, _MM_HINT_NTA);
- _mm_prefetch((char*)begin_64 + 64, _MM_HINT_NTA);
- _mm_prefetch((char*)begin_64 + 128, _MM_HINT_NTA);
- _mm_prefetch((char*)begin_64 + 192, _MM_HINT_NTA);
-
- while (dst < begin_64)
- {
- copy4a(dst, src);
- dst += 4;
- src += 4;
- }
-
- while (dst < end_64)
- {
- _mm_prefetch((char*)src + 512, _MM_HINT_NTA);
- _mm_prefetch((char*)dst + 512, _MM_HINT_NTA);
- copy4a(dst, src);
- copy4a(dst+4, src+4);
- copy4a(dst+8, src+8);
- copy4a(dst+12, src+12);
-
- dst += 16;
- src += 16;
- }
- }
-
- while (dst < end)
- {
- copy4a(dst, src);
- dst += 4;
- src += 4;
- }
+ ll_memcpy_nonaliased_aligned_16((char*)dst, (char*)src, bytes);
}
void LLVector4a::setRotated( const LLRotation& rot, const LLVector4a& vec )
diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h
index 0526793d3a..1a478bc8de 100644
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -93,7 +93,11 @@ public:
LLVector4a()
{ //DO NOT INITIALIZE -- The overhead is completely unnecessary
+// This assert is causing spurious referenced before set warnings on GCC 4.3.4
+//
+#if !LL_LINUX
ll_assert_aligned(this,16);
+#endif
}
LLVector4a(F32 x, F32 y, F32 z, F32 w = 0.f)
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index 9fc72fd801..edd16b5688 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -96,93 +96,6 @@ extern BOOL gDebugGL;
bool less_than_max_mag(const LLVector4a& vec);
-template <class T, U32 alignment>
-LLAlignedArray<T, alignment>::LLAlignedArray()
-{
- mArray = NULL;
- mElementCount = 0;
- mCapacity = 0;
-}
-
-template <class T, U32 alignment>
-LLAlignedArray<T, alignment>::~LLAlignedArray()
-{
- ll_aligned_free(mArray);
- mArray = NULL;
- mElementCount = 0;
- mCapacity = 0;
-}
-
-template <class T, U32 alignment>
-void LLAlignedArray<T, alignment>::push_back(const T& elem)
-{
- T* old_buf = NULL;
- if (mCapacity <= mElementCount)
- {
- mCapacity++;
- mCapacity *= 2;
- T* new_buf = (T*) ll_aligned_malloc(mCapacity*sizeof(T), alignment);
- if (mArray)
- {
- LLVector4a::memcpyNonAliased16((F32*) new_buf, (F32*) mArray, sizeof(T)*mElementCount);
- }
- old_buf = mArray;
- mArray = new_buf;
- }
-
- mArray[mElementCount++] = elem;
-
- //delete old array here to prevent error on a.push_back(a[0])
- ll_aligned_free(old_buf);
-}
-
-template <class T, U32 alignment>
-void LLAlignedArray<T, alignment>::resize(U32 size)
-{
- if (mCapacity < size)
- {
- mCapacity = size+mCapacity*2;
- T* new_buf = mCapacity > 0 ? (T*) ll_aligned_malloc(mCapacity*sizeof(T), alignment) : NULL;
- if (mArray)
- {
- LLVector4a::memcpyNonAliased16((F32*) new_buf, (F32*) mArray, sizeof(T)*mElementCount);
- ll_aligned_free(mArray);
- }
-
- /*for (U32 i = mElementCount; i < mCapacity; ++i)
- {
- new(new_buf+i) T();
- }*/
- mArray = new_buf;
- }
-
- mElementCount = size;
-}
-
-
-template <class T, U32 alignment>
-T& LLAlignedArray<T, alignment>::operator[](int idx)
-{
- llassert(idx < mElementCount);
- return mArray[idx];
-}
-
-template <class T, U32 alignment>
-const T& LLAlignedArray<T, alignment>::operator[](int idx) const
-{
- llassert(idx < mElementCount);
- return mArray[idx];
-}
-
-template <class T, U32 alignment>
-T* LLAlignedArray<T, alignment>::append(S32 N)
-{
- U32 sz = size();
- resize(sz+N);
- return &((*this)[sz]);
-}
-
-
BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm)
{
LLVector3 test = (pt2-pt1)%(pt3-pt2);
@@ -4816,10 +4729,13 @@ void LLVolumeFace::optimize(F32 angle_cutoff)
}
}
- llassert(new_face.mNumIndices == mNumIndices);
- llassert(new_face.mNumVertices <= mNumVertices);
-
- swapData(new_face);
+ // disallow data amplification
+ //
+ if (new_face.mNumVertices <= mNumVertices)
+ {
+ llassert(new_face.mNumIndices == mNumIndices);
+ swapData(new_face);
+ }
}
class LLVCacheTriangleData;
@@ -5400,12 +5316,7 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build)
S32 max_t = volume->getPath().mPath.size();
// S32 i;
- S32 num_vertices = 0, num_indices = 0;
S32 grid_size = (profile.size()-1)/4;
- S32 quad_count = (grid_size * grid_size);
-
- num_vertices = (grid_size+1)*(grid_size+1);
- num_indices = quad_count * 4;
LLVector4a& min = mExtents[0];
LLVector4a& max = mExtents[1];
@@ -6822,3 +6733,4 @@ void calc_binormal_from_triangle(LLVector4a& binormal,
binormal.set( 0, 1 , 0 );
}
}
+
diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h
index 5e43af92ec..6b599a4126 100644
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -57,6 +57,7 @@ class LLVolumeTriangle;
#include "llrefcount.h"
#include "llpointer.h"
#include "llfile.h"
+#include "llalignedarray.h"
//============================================================================
@@ -195,26 +196,6 @@ const U8 LL_SCULPT_FLAG_MIRROR = 128;
const S32 LL_SCULPT_MESH_MAX_FACES = 8;
-template <class T, U32 alignment>
-class LLAlignedArray
-{
-public:
- T* mArray;
- U32 mElementCount;
- U32 mCapacity;
-
- LLAlignedArray();
- ~LLAlignedArray();
-
- void push_back(const T& elem);
- U32 size() const { return mElementCount; }
- void resize(U32 size);
- T* append(S32 N);
- T& operator[](int idx);
- const T& operator[](int idx) const;
-};
-
-
class LLProfileParams
{
public:
generated by cgit v1.2.3 (git 2.25.1) at 2025-01-29 23:43:58 +0000