void Mjoin(PATL,mmJIK2)
(int K, int nMb, int nNb, int nKb, int ib, int jb, int kb,
const SCALAR alpha, const TYPE *pA0, const TYPE *B, int ldb,
TYPE *pB0, int incB, MAT2BLK B2blk, const SCALAR beta,
TYPE *C, int ldc, MATSCAL gescal, NBMM0 NBmm0)
{
const int incK = ATL_MulByNB(K)SHIFT, incC = ATL_MulByNB(ldc-nMb) SHIFT;
const int ZEROC = ((gescal == NULL) && SCALAR_IS_ZERO(beta));
int i, j = nNb;
const TYPE *pA=pA0;
const TYPE rbeta = ( (gescal) ? ATL_rone : *beta );
TYPE *pB=pB0, *stB=pB0+(ATL_MulByNBNB(nKb)SHIFT);
if (nNb)
{
do /* Loop over full column panels of B */
{
if (B)
{
B2blk(K, NB, B, ldb, pB, alpha);
B += incB;
}
if (nMb)
{
i = nMb;
do /* loop over full row panels of A */
{
if (gescal) gescal(NB, NB, beta, C, ldc);
if (nKb) /* loop over full blocks in panels */
{
NBmm0(MB, NB, KB, ATL_rone, pA, KB, pB, KB, rbeta, C, ldc);
pA += NBNB2;
pB += NBNB2;
if (nKb != 1)
{
do
{
NBmm_b1(MB, NB, KB, ATL_rone, pA, KB, pB, KB, ATL_rone,
C, ldc);
pA += NBNB2;
pB += NBNB2;
}
while (pB != stB);
}
if (kb)
{
KBmm(MB, NB, kb, ATL_rone, pA, kb, pB, kb, ATL_rone,
C, ldc);
pA += ATL_MulByNB(kb)<<1;
}
}
else if (kb)
{
if (ZEROC) Mjoin(PATL,gezero)(MB, NB, C, ldc);
KBmm(MB, NB, kb, ATL_rone, pA, kb, pB, kb, rbeta, C, ldc);
pA += ATL_MulByNB(kb)<<1;
}
pB = pB0;
C += NB2;
}
while (--i);
}
if (ib)
{
if (gescal) gescal(ib, NB, beta, C, ldc);
IBNBmm(ib, K, pA, pB, rbeta, C, ldc);
}
if (!B)
{
pB0 += incK;
pB = pB0;
stB += incK;
}
C += incC;
pA = pA0;
}
while (--j);
}
if (jb)
{
if (B) B2blk(K, jb, B, ldb, pB, alpha);
for (i=nMb; i; i--)
{
if (gescal) gescal(NB, jb, beta, C, ldc);
NBJBmm(jb, K, pA, pB, rbeta, C, ldc);
pA += incK;
C += NB2;
}
if (ib)
{
if (gescal) gescal(ib, jb, beta, C, ldc);
IBJBmm(ib, jb, K, pA, pB, rbeta, C, ldc);
}
}
}
void Mjoin(PATL,mmIJK2)(int K, int nMb, int nNb, int nKb, int ib, int jb,
int kb, const SCALAR alpha, const TYPE *A, int lda,
TYPE *pA0, int incA, MAT2BLK A2blk, const TYPE *pB0,
const SCALAR beta, TYPE *C, int ldc, TYPE *pC,
PUTBLK putblk, NBMM0 NBmm0)
/*
* Outer three loops for matmul with outer loop over rows of A
*/
{
int i, j, ldpc;
const int ZEROC = ((putblk == NULL) && SCALAR_IS_ZERO(beta));
const int incK = ATL_MulByNB(K), incC = ATL_MulByNB(ldc);
TYPE *pA=pA0, *stA=pA0+ATL_MulByNBNB(nKb);
const TYPE *pB=pB0;
const TYPE cubeta = ( (putblk) ? ATL_rzero : beta );
TYPE *c;
if (putblk)
{
ldpc = NB;
if (!nKb && kb) Mjoin(PATL,gezero)(MB, NB, pC, MB);
}
else ldpc = ldc;
for (i=nMb; i; i--) /* loop over full row panels of A */
{
if (A)
{
A2blk(K, NB, A, lda, pA, alpha); /* get 1 row panel of A */
A += incA;
}
if (!putblk) pC = C;
c = C;
C += NB;
for (j=nNb; j; j--) /* full column panels of B */
{
if (nKb)
{
NBmm0(MB, NB, KB, ATL_rone, pA, KB, pB, KB, beta, pC, ldpc);
pA += NBNB;
pB += NBNB;
if (nKb != 1)
{
do
{
NBmm(MB, NB, KB, ATL_rone, pA, KB, pB, KB, ATL_rone,
pC, ldpc);
pA += NBNB;
pB += NBNB;
}
while (pA != stA);
}
if (kb)
{
KBmm(MB, NB, kb, ATL_rone, pA, kb, pB, kb, ATL_rone, pC, ldpc);
pB += kb*NB;
}
}
else
{
if (ZEROC) Mjoin(PATL,gezero)(MB, NB, pC, ldpc);
if (kb)
{
KBmm(MB, NB, kb, ATL_rone, pA, kb, pB, kb, cubeta, pC, ldpc);
pB += kb*NB;
}
}
pA = pA0;
if (putblk) putblk(NB, NB, pC, c, ldc, beta);
else pC += incC;
c += incC;
}
if (jb)
{
NBJBmm(jb, K, pA, pB, cubeta, pC, ldpc);
if (putblk) putblk(NB, jb, pC, c, ldc, beta);
}
pB = pB0;
if (!A)
{
pA0 += incK;
pA = pA0;
stA += incK;
}
}
if (ib)
{
c = C;
if (A) A2blk(K, ib, A, lda, pA, alpha); /* get last row panel of A */
for (j=nNb; j; j--) /* full column panels of B */
{
if (putblk)
{
IBNBmm(ib, K, pA, pB, ATL_rzero, pC, ib);
putblk(ib, NB, pC, c, ldc, beta);
}
else IBNBmm(ib, K, pA, pB, beta, c, ldc);
pB += incK;
c += incC;
}
if (jb)
{
if (putblk)
{
IBJBmm(ib, jb, K, pA, pB, ATL_rzero, pC, ib);
putblk(ib, jb, pC, c, ldc, beta);
}
else IBJBmm(ib, jb, K, pA, pB, beta, c, ldc);
}
}
}
void Mjoin(PATL,mmIJK2)
(int K, int nMb, int nNb, int nKb, int ib, int jb, int kb,
const SCALAR alpha, const TYPE *A, const int lda, TYPE *pA0, const int incA,
MAT2BLK A2blk, TYPE *pB0, const SCALAR beta, TYPE *C, int ldc,
MATSCAL gescal, NBMM0 NBmm0)
{
const int incK = ATL_MulByNB(K)<<1;
const int incCn = ATL_MulByNB(ldc)<<1, incCm = (MB<<1) - nNb*incCn;
const int ZEROC = ((gescal == NULL) && SCALAR_IS_ZERO(beta));
int i, j, k;
const TYPE *pB=pB0;
const TYPE rbeta = ( (gescal) ? ATL_rone : *beta );
TYPE *pA=pA0;
for (i=nMb; i; i--)
{
if (A)
{
A2blk(K, NB, A, lda, pA, alpha); /* get 1 row panel of A */
A += incA;
}
for (j=nNb; j; j--)
{
if (gescal) gescal(MB, NB, beta, C, ldc);
if (nKb)
{
NBmm0(MB, NB, KB, ATL_rone, pA, KB, pB, KB, rbeta, C, ldc);
pA += NBNB2;
pB += NBNB2;
if (nKb != 1)
{
for (k=nKb-1; k; k--, pA += NBNB2, pB += NBNB2)
NBmm_b1(MB, NB, KB, ATL_rone, pA, KB, pB, KB,
ATL_rone, C, ldc);
}
if (kb)
{
KBmm(MB, NB, kb, ATL_rone, pA, kb, pB, kb, ATL_rone, C, ldc);
pB += ATL_MulByNB(kb)<<1;
}
}
else
{
if (ZEROC) Mjoin(PATL,gezero)(MB, NB, C, ldc);
if (kb)
{
KBmm(MB, NB, kb, ATL_rone, pA, kb, pB, kb, rbeta, C, ldc);
pB += ATL_MulByNB(kb)<<1;
}
}
pA = pA0;
C += incCn;
}
if (jb)
{
if (gescal) gescal(MB, jb, beta, C, ldc);
MBJBmm(jb, K, pA, pB, rbeta, C, ldc);
}
pB = pB0;
if (!A)
{
pA0 += incK;
pA = pA0;
}
C += incCm;
}
if (ib)
{
if (A) A2blk(K, ib, A, lda, pA, alpha); /* get last row panel of A */
for(j=nNb; j; j--) /* full column panels of B */
{
if (gescal) gescal(ib, NB, beta, C, ldc);
IBNBmm(ib, K, pA, pB, rbeta, C, ldc);
pB += incK;
C += incCn;
}
if (jb)
{
if (gescal) gescal(ib, jb, beta, C, ldc);
IBJBmm(ib, jb, K, pA, pB, rbeta, C, ldc);
}
}
}