SSE2: Remove dead code

This commit is contained in:
kaetemi 2014-06-13 22:21:07 +02:00
parent 56c59d114d
commit f51843a721
3 changed files with 0 additions and 510 deletions

View file

@ -108,281 +108,6 @@ public:
}; };
// ***************************************************************************
// ***************************************************************************
// SSE Matrix
// ***************************************************************************
// ***************************************************************************
// ***************************************************************************
#if defined(NL_OS_WINDOWS) && !defined(NL_NO_ASM)
/** For fast vector/point multiplication. Special usage for Skinning.
* NB: SSE is no more used (no speed gain, some memory problem), but keep it for possible future usage.
*/
class CMatrix3x4SSE
{
public:
// Order them in memory column first, for SSE column multiplication.
float a11, a21, a31, a41;
float a12, a22, a32, a42;
float a13, a23, a33, a43;
float a14, a24, a34, a44;
// Copy from a matrix.
void set(const CMatrix &mat)
{
const float *m =mat.get();
a11= m[0]; a12= m[4]; a13= m[8] ; a14= m[12];
a21= m[1]; a22= m[5]; a23= m[9] ; a24= m[13];
a31= m[2]; a32= m[6]; a33= m[10]; a34= m[14];
// not used.
a41= 0 ; a42= 0 ; a43= 0 ; a44= 1;
}
// mulSetvector. NB: in should be different as v!! (else don't work).
void mulSetVector(const CVector &vin, CVector &vout)
{
__asm
{
mov eax, vin
mov ebx, this
mov edi, vout
// Load in vector in op[0]
movss xmm0, [eax]vin.x
movss xmm1, [eax]vin.y
movss xmm2, [eax]vin.z
// Expand op[0] to op[1], op[2], op[3]
shufps xmm0, xmm0, 0
shufps xmm1, xmm1, 0
shufps xmm2, xmm2, 0
// Mul each vector with 3 Matrix column
mulps xmm0, [ebx]this.a11
mulps xmm1, [ebx]this.a12
mulps xmm2, [ebx]this.a13
// Add each column vector.
addps xmm0, xmm1
addps xmm0, xmm2
// write the result.
movss [edi]vout.x, xmm0
shufps xmm0, xmm0, 33
movss [edi]vout.y, xmm0
movhlps xmm0, xmm0
movss [edi]vout.z, xmm0
}
}
// mulSetpoint. NB: in should be different as v!! (else don't work).
void mulSetPoint(const CVector &vin, CVector &vout)
{
__asm
{
mov eax, vin
mov ebx, this
mov edi, vout
// Load in vector in op[0]
movss xmm0, [eax]vin.x
movss xmm1, [eax]vin.y
movss xmm2, [eax]vin.z
// Expand op[0] to op[1], op[2], op[3]
shufps xmm0, xmm0, 0
shufps xmm1, xmm1, 0
shufps xmm2, xmm2, 0
// Mul each vector with 3 Matrix column
mulps xmm0, [ebx]this.a11
mulps xmm1, [ebx]this.a12
mulps xmm2, [ebx]this.a13
// Add each column vector.
addps xmm0, xmm1
addps xmm0, xmm2
// Add Matrix translate column vector
addps xmm0, [ebx]this.a14
// write the result.
movss [edi]vout.x, xmm0
shufps xmm0, xmm0, 33
movss [edi]vout.y, xmm0
movhlps xmm0, xmm0
movss [edi]vout.z, xmm0
}
}
// mulSetvector. NB: vin should be different as v!! (else don't work).
void mulSetVector(const CVector &vin, float scale, CVector &vout)
{
__asm
{
mov eax, vin
mov ebx, this
mov edi, vout
// Load in vector in op[0]
movss xmm0, [eax]vin.x
movss xmm1, [eax]vin.y
movss xmm2, [eax]vin.z
// Load scale in op[0]
movss xmm3, scale
// Expand op[0] to op[1], op[2], op[3]
shufps xmm0, xmm0, 0
shufps xmm1, xmm1, 0
shufps xmm2, xmm2, 0
shufps xmm3, xmm3, 0
// Store vertex column in other regs.
movaps xmm5, xmm0
movaps xmm6, xmm1
movaps xmm7, xmm2
// Mul each vector with 3 Matrix column
mulps xmm0, [ebx]this.a11
mulps xmm1, [ebx]this.a12
mulps xmm2, [ebx]this.a13
// Add each column vector.
addps xmm0, xmm1
addps xmm0, xmm2
// mul final result with scale
mulps xmm0, xmm3
// store it in xmm4 for future use.
movaps xmm4, xmm0
}
}
// mulSetpoint. NB: vin should be different as v!! (else don't work).
void mulSetPoint(const CVector &vin, float scale, CVector &vout)
{
__asm
{
mov eax, vin
mov ebx, this
mov edi, vout
// Load in vector in op[0]
movss xmm0, [eax]vin.x
movss xmm1, [eax]vin.y
movss xmm2, [eax]vin.z
// Load scale in op[0]
movss xmm3, scale
// Expand op[0] to op[1], op[2], op[3]
shufps xmm0, xmm0, 0
shufps xmm1, xmm1, 0
shufps xmm2, xmm2, 0
shufps xmm3, xmm3, 0
// Store vertex column in other regs.
movaps xmm5, xmm0
movaps xmm6, xmm1
movaps xmm7, xmm2
// Mul each vector with 3 Matrix column
mulps xmm0, [ebx]this.a11
mulps xmm1, [ebx]this.a12
mulps xmm2, [ebx]this.a13
// Add each column vector.
addps xmm0, xmm1
addps xmm0, xmm2
// Add Matrix translate column vector
addps xmm0, [ebx]this.a14
// mul final result with scale
mulps xmm0, xmm3
// store it in xmm4 for future use.
movaps xmm4, xmm0
}
}
// mulAddvector. NB: vin should be different as v!! (else don't work).
void mulAddVector(const CVector &/* vin */, float scale, CVector &vout)
{
__asm
{
mov ebx, this
mov edi, vout
// Load vin vector loaded in mulSetVector
movaps xmm0, xmm5
movaps xmm1, xmm6
movaps xmm2, xmm7
// Load scale in op[0]
movss xmm3, scale
// Expand op[0] to op[1], op[2], op[3]
shufps xmm3, xmm3, 0
// Mul each vector with 3 Matrix column
mulps xmm0, [ebx]this.a11
mulps xmm1, [ebx]this.a12
mulps xmm2, [ebx]this.a13
// Add each column vector.
addps xmm0, xmm1
addps xmm0, xmm2
// mul final result with scale
mulps xmm0, xmm3
// Add result, with prec sum.
addps xmm0, xmm4
// store it in xmm4 for future use.
movaps xmm4, xmm0
// write the result.
movss [edi]vout.x, xmm0
shufps xmm0, xmm0, 33
movss [edi]vout.y, xmm0
movhlps xmm0, xmm0
movss [edi]vout.z, xmm0
}
}
// mulAddpoint. NB: vin should be different as v!! (else don't work).
void mulAddPoint(const CVector &/* vin */, float scale, CVector &vout)
{
__asm
{
mov ebx, this
mov edi, vout
// Load vin vector loaded in mulSetPoint
movaps xmm0, xmm5
movaps xmm1, xmm6
movaps xmm2, xmm7
// Load scale in op[0]
movss xmm3, scale
// Expand op[0] to op[1], op[2], op[3]
shufps xmm3, xmm3, 0
// Mul each vector with 3 Matrix column
mulps xmm0, [ebx]this.a11
mulps xmm1, [ebx]this.a12
mulps xmm2, [ebx]this.a13
// Add each column vector.
addps xmm0, xmm1
addps xmm0, xmm2
// Add Matrix translate column vector
addps xmm0, [ebx]this.a14
// mul final result with scale
mulps xmm0, xmm3
// Add result, with prec sum.
addps xmm0, xmm4
// store it in xmm4 for future use.
movaps xmm4, xmm0
// write the result.
movss [edi]vout.x, xmm0
shufps xmm0, xmm0, 33
movss [edi]vout.y, xmm0
movhlps xmm0, xmm0
movss [edi]vout.z, xmm0
}
}
};
#else // NL_OS_WINDOWS
/// dummy CMatrix3x4SSE for non windows platform
class CMatrix3x4SSE : public CMatrix3x4 { };
#endif
} // NL3D } // NL3D

View file

@ -39,124 +39,6 @@ namespace NL3D
{ {
// ***************************************************************************
// ***************************************************************************
// CMatrix3x4SSE array correctly aligned
// ***************************************************************************
// ***************************************************************************
// ***************************************************************************
#define NL3D_SSE_ALIGNEMENT 16
/**
* A CMatrix3x4SSE array correctly aligned
* NB: SSE is no more used (no speed gain, some memory problem), but keep it for possible future usage.
*/
class CMatrix3x4SSEArray
{
private:
void *_AllocData;
void *_Data;
uint _Size;
uint _Capacity;
public:
CMatrix3x4SSEArray()
{
_AllocData= NULL;
_Data= NULL;
_Size= 0;
_Capacity= 0;
}
~CMatrix3x4SSEArray()
{
clear();
}
CMatrix3x4SSEArray(const CMatrix3x4SSEArray &other)
{
_AllocData= NULL;
_Data= NULL;
_Size= 0;
_Capacity= 0;
*this= other;
}
CMatrix3x4SSEArray &operator=(const CMatrix3x4SSEArray &other)
{
if( this == &other)
return *this;
resize(other.size());
// copy data from aligned pointers to aligned pointers.
memcpy(_Data, other._Data, size() * sizeof(CMatrix3x4SSE) );
return *this;
}
CMatrix3x4SSE *getPtr()
{
return (CMatrix3x4SSE*)_Data;
}
void clear()
{
delete [] ((uint8 *)_AllocData);
_AllocData= NULL;
_Data= NULL;
_Size= 0;
_Capacity= 0;
}
void resize(uint n)
{
// reserve ??
if(n>_Capacity)
reserve( max(2*_Capacity, n));
_Size= n;
}
void reserve(uint n)
{
if(n==0)
clear();
else if(n>_Capacity)
{
// Alloc new data.
void *newAllocData;
void *newData;
// Alloc for alignement.
newAllocData= new uint8 [n * sizeof(CMatrix3x4SSE) + NL3D_SSE_ALIGNEMENT-1];
if(newAllocData==NULL)
throw Exception("SSE Allocation Failed");
// Align ptr
newData= (void*) ( ((ptrdiff_t)newAllocData+NL3D_SSE_ALIGNEMENT-1) & (~(NL3D_SSE_ALIGNEMENT-1)) );
// copy valid data from old to new.
memcpy(newData, _Data, size() * sizeof(CMatrix3x4SSE) );
// release old.
if(_AllocData)
delete [] ((uint8*)_AllocData);
// change ptrs and capacity.
_Data= newData;
_AllocData= newAllocData;
_Capacity= n;
// TestYoyo
//nlwarning("YOYO Tst SSE P4: %X, %d", _Data, n);
}
}
uint size() const {return _Size;}
CMatrix3x4SSE &operator[](uint i) {return ((CMatrix3x4SSE*)_Data)[i];}
};
// *************************************************************************** // ***************************************************************************
// *************************************************************************** // ***************************************************************************

View file

@ -2247,123 +2247,6 @@ void CMeshMRMSkinnedGeom::getSkinWeights (std::vector<CMesh::CSkinWeight> &skinW
} }
} }
// ***************************************************************************
// ***************************************************************************
// CMatrix3x4SSE array correctly aligned
// ***************************************************************************
// ***************************************************************************
// ***************************************************************************
#define NL3D_SSE_ALIGNEMENT 16
/**
* A CMatrix3x4SSEArray array correctly aligned
* NB: SSE is no more used (no speed gain, some memory problem), but keep it for possible future usage.
*/
class CMatrix3x4SSEArray
{
private:
void *_AllocData;
void *_Data;
uint _Size;
uint _Capacity;
public:
CMatrix3x4SSEArray()
{
_AllocData= NULL;
_Data= NULL;
_Size= 0;
_Capacity= 0;
}
~CMatrix3x4SSEArray()
{
clear();
}
CMatrix3x4SSEArray(const CMatrix3x4SSEArray &other)
{
_AllocData= NULL;
_Data= NULL;
_Size= 0;
_Capacity= 0;
*this= other;
}
CMatrix3x4SSEArray &operator=(const CMatrix3x4SSEArray &other)
{
if( this == &other)
return *this;
resize(other.size());
// copy data from aligned pointers to aligned pointers.
memcpy(_Data, other._Data, size() * sizeof(CMatrix3x4SSE) );
return *this;
}
CMatrix3x4SSE *getPtr()
{
return (CMatrix3x4SSE*)_Data;
}
void clear()
{
delete [] ((uint8 *) _AllocData);
_AllocData= NULL;
_Data= NULL;
_Size= 0;
_Capacity= 0;
}
void resize(uint n)
{
// reserve ??
if(n>_Capacity)
reserve( max(2*_Capacity, n));
_Size= n;
}
void reserve(uint n)
{
if(n==0)
clear();
else if(n>_Capacity)
{
// Alloc new data.
void *newAllocData;
void *newData;
// Alloc for alignement.
newAllocData= new uint8 [n * sizeof(CMatrix3x4SSE) + NL3D_SSE_ALIGNEMENT-1];
if(newAllocData==NULL)
throw Exception("SSE Allocation Failed");
// Align ptr
newData= (void*) ( ((ptrdiff_t)newAllocData+NL3D_SSE_ALIGNEMENT-1) & (~(NL3D_SSE_ALIGNEMENT-1)) );
// copy valid data from old to new.
memcpy(newData, _Data, size() * sizeof(CMatrix3x4SSE) );
// release old.
if(_AllocData)
delete [] ((uint8*)_AllocData);
// change ptrs and capacity.
_Data= newData;
_AllocData= newAllocData;
_Capacity= n;
// TestYoyo
//nlwarning("YOYO Tst SSE P4: %X, %d", _Data, n);
}
}
uint size() const {return _Size;}
CMatrix3x4SSE &operator[](uint i) {return ((CMatrix3x4SSE*)_Data)[i];}
};
// *************************************************************************** // ***************************************************************************
// *************************************************************************** // ***************************************************************************