void Mjoin(Mjoin(PATL,symmL),UploNM)
(const int M, const int N, const void *valpha, const void *A, const int lda,
const void *B, const int ldb, const void *vbeta, void *C, const int ldc)
{
#ifdef TREAL
const SCALAR alpha=*( (const SCALAR *)valpha );
const SCALAR beta =*( (const SCALAR *)vbeta );
const SCALAR one=1.0;
#else
#define alpha valpha
#define beta vbeta
#endif
TYPE *a;
void *va;
if (N > SYMM_Xover)
{
va = malloc(ATL_Cachelen + (ATL_MulBySize(M)*M));
ATL_assert(va);
a = ATL_AlignPtr(va);
#ifdef TREAL
if ( SCALAR_IS_ONE(alpha) )
Mjoin(Mjoin(Mjoin(PATL,sycopy),UploNM),_a1)(M, alpha, A, lda, a);
else Mjoin(Mjoin(Mjoin(PATL,sycopy),UploNM),_aX)(M, alpha, A, lda, a);
CgemmTN(M, N, M, one, a, M, B, ldb, beta, C, ldc);
#else
Mjoin(Mjoin(PATL,sycopy),UploNM)(M, A, lda, a);
CgemmTN(M, N, M, valpha, a, M, B, ldb, vbeta, C, ldc);
#endif
free(va);
}
else Mjoin(PATL,refsymm)(AtlasLeft, Uplo_, M, N, alpha, A, lda, B, ldb,
beta, C, ldc);
}
int Mjoin(PATL,syr2kLT)
#endif
(const int N, const int K, const void *valpha, const void *A, const int lda,
const void *B, const int ldb, const void *vbeta, void *C, const int ldc)
{
int i;
void *vc=NULL;
TYPE *c;
#ifdef TREAL
const SCALAR alpha=*( (const SCALAR *)valpha );
const SCALAR beta =*( (const SCALAR *)vbeta );
const SCALAR one=1.0, zero=0.0;
#else
#define alpha valpha
const TYPE *beta=vbeta;
const TYPE one[2]={1.0,0.0}, zero[2]={0.0,0.0};
#endif
i = ATL_MulBySize(N)*N;
if (i <= ATL_MaxMalloc) vc = malloc(ATL_Cachelen+i);
if (vc == NULL) return(1);
c = ATL_AlignPtr(vc);
CgemmTN(N, N, K, alpha, A, lda, B, ldb, zero, c, N);
if ( SCALAR_IS_ONE(beta) ) Mjoin(syr2k_put,_b1)(N, c, beta, C, ldc);
else if ( SCALAR_IS_ZERO(beta) ) Mjoin(syr2k_put,_b0)(N, c, beta, C, ldc);
#ifdef TCPLX
else if (SCALAR_IS_NONE(beta)) Mjoin(syr2k_put,_bn1)(N, c, beta, C, ldc);
else if (beta[1] == *zero) Mjoin(syr2k_put,_bXi0)(N, c, beta, C, ldc);
#endif
else Mjoin(syr2k_put,_bX)(N, c, beta, C, ldc);
free(vc);
return(0);
}
void Mjoin(Mjoin(Mjoin(PATL,syrk),UploNM),T)
(const int N, const int K, const void *valpha, const void *A, const int lda,
const void *vbeta, void *C, const int ldc)
{
void *vc;
TYPE *c;
#ifdef TREAL
const SCALAR alpha=*( (const SCALAR *)valpha );
const SCALAR beta =*( (const SCALAR *)vbeta );
const SCALAR one=1.0, zero=0.0;
#else
#define alpha valpha
const TYPE *beta=vbeta;
const TYPE one[2]={1.0,0.0}, zero[2]={0.0,0.0};
#endif
if (K > SYRK_Xover)
{
vc = malloc(ATL_Cachelen+ATL_MulBySize(N)*N);
ATL_assert(vc);
c = ATL_AlignPtr(vc);
CgemmTN(N, N, K, alpha, A, lda, A, lda, zero, c, N);
if ( SCALAR_IS_ONE(beta) ) Mjoin(syr_put,_b1)(N, c, beta, C, ldc);
else if ( SCALAR_IS_ZERO(beta) ) Mjoin(syr_put,_b0)(N, c, beta, C, ldc);
#ifdef TCPLX
else if ( SCALAR_IS_NONE(beta) )
Mjoin(syr_put,_bn1)(N, c, beta, C, ldc);
else if (beta[1] == *zero) Mjoin(syr_put,_bXi0)(N, c, beta, C, ldc);
#endif
else Mjoin(syr_put,_bX)(N, c, beta, C, ldc);
free(vc);
}
else Mjoin(PATL,refsyrk)(Uplo_, AtlasTrans, N, K, alpha, A, lda,
beta, C, ldc);
}
void Cgemm(const enum ATLAS_TRANS TA, const enum ATLAS_TRANS TB,
const int M, const int N, const int K, const SCALAR alpha,
const TYPE *A, const int lda, const TYPE *B, const int ldb,
const SCALAR beta, TYPE *C, const int ldc)
/*
* Error checks have been done by interface routine
*/
{
if (!M || !N) return; /* quick return */
if ( SCALAR_IS_ZERO(alpha) || !K)
{
#ifdef TREAL
if (beta == ATL_rzero) Mjoin(PATL,gezero)(M, N, C, ldc);
else if (beta != ATL_rone) Mjoin(PATL,gescal_bX)(M, N, beta, C, ldc);
#else
if (beta[1] == ATL_rzero)
{
if (*beta == ATL_rzero) Mjoin(PATL,gezero)(M, N, C, ldc);
else if (*beta != ATL_rone)
Mjoin(PATL,gescal_bXi0)(M, N, beta, C, ldc);
}
else Mjoin(PATL,gescal_bX)(M, N, beta, C, ldc);
#endif
return;
}
if (TA == AtlasNoTrans)
{
if (TB == AtlasNoTrans)
CgemmNN(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
#ifdef TCPLX
else if (TB == AtlasConjTrans)
CgemmNC(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
#endif
else
CgemmNT(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
}
#ifdef TCPLX
else if (TA == AtlasConjTrans)
{
if (TB == AtlasNoTrans)
CgemmCN(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
else if (TB == AtlasConjTrans)
CgemmCC(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
else
CgemmCT(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
}
#endif
else
{
if (TB == AtlasNoTrans)
CgemmTN(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
#ifdef TCPLX
else if (TB == AtlasConjTrans)
CgemmTC(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
#endif
else
CgemmTT(M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
}
}
