Skip to content

Fix complex conjugate case in GEMMT, add its prototypes; remove overoptimization in zimatcopy_cnc (from #4452) #4460

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 4 commits into from
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions cblas.h
Original file line number Diff line number Diff line change
Expand Up @@ -290,6 +290,14 @@ void cblas_zgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLA
void cblas_zgemm3m(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);

void cblas_sgemmt(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint K,
OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
void cblas_dgemmt(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint K,
OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
void cblas_cgemmt(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint K,
OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
void cblas_zgemmt(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint K,
OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);

void cblas_ssymm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
Expand Down
9 changes: 9 additions & 0 deletions common_interface.h
Original file line number Diff line number Diff line change
Expand Up @@ -498,6 +498,15 @@ void BLASFUNC(zgemm3m)(char *, char *, blasint *, blasint *, blasint *, double *
void BLASFUNC(xgemm3m)(char *, char *, blasint *, blasint *, blasint *, xdouble *,
xdouble *, blasint *, xdouble *, blasint *, xdouble *, xdouble *, blasint *);

void BLASFUNC(sgemmt)(char*, char *, char *, blasint *, blasint *, float *,
float *, blasint *, float *, blasint *, float *, float *, blasint *);
void BLASFUNC(dgemmt)(char*, char *, char *, blasint *, blasint *, double *,
double *, blasint *, double *, blasint *, double *, double *, blasint *);
void BLASFUNC(cgemmt)(char*, char *, char *, blasint *, blasint *, float *,
float *, blasint *, float *, blasint *, float *, float *, blasint *);
void BLASFUNC(zgemmt)(char*, char *, char *, blasint *, blasint *, double *,
double *, blasint *, double *, blasint *, double *, double *, blasint *);

int BLASFUNC(sge2mm)(char *, char *, char *, blasint *, blasint *,
float *, float *, blasint *, float *, blasint *,
float *, float *, blasint *);
Expand Down
114 changes: 82 additions & 32 deletions interface/gemmt.c
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,9 @@ void NAME(char *UPLO, char *TRANSA, char *TRANSB,

char transA, transB, Uplo;
blasint nrowa, nrowb;
#if defined(COMPLEX)
blasint ncolb;
#endif
IFLOAT *buffer;
IFLOAT *aa, *bb;
FLOAT *cc;
Expand Down Expand Up @@ -157,17 +160,25 @@ void NAME(char *UPLO, char *TRANSA, char *TRANSB,
uplo = 1;

nrowa = m;
if (transa) nrowa = k;
if (transa & 1) nrowa = k;
nrowb = k;
if (transb) nrowb = m;
#if defined(COMPLEX)
ncolb = m;
#endif
if (transb & 1) {
nrowb = m;
#if defined(COMPLEX)
ncolb = k;
#endif
}

info = 0;

if (ldc < MAX(1, m))
info = 13;
if (ldb < MAX(1, nrowa))
if (ldb < MAX(1, nrowb))
info = 10;
if (lda < MAX(1, nrowb))
if (lda < MAX(1, nrowa))
info = 8;
if (k < 0)
info = 5;
Expand Down Expand Up @@ -211,6 +222,9 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
blasint info;
blasint lda, ldb;
FLOAT *a, *b;
#if defined(COMPLEX)
blasint nrowb, ncolb;
#endif
XFLOAT *buffer;

PRINT_DEBUG_CNAME;
Expand Down Expand Up @@ -262,11 +276,22 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,

info = -1;

blasint nrowa, nrowb;
blasint nrowa;
#if !defined(COMPLEX)
blasint nrowb;
#endif
nrowa = m;
if (transa) nrowa = k;
if (transa & 1) nrowa = k;
nrowb = k;
if (transb) nrowb = m;
#if defined(COMPLEX)
ncolb = m;
#endif
if (transb & 1) {
nrowb = m;
#if defined(COMPLEX)
ncolb = k;
#endif
}

if (ldc < MAX(1, m))
info = 13;
Expand Down Expand Up @@ -330,26 +355,38 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,

info = -1;

blasint ncola, ncolb;
ncola = k;
if (transa) ncola = m;
ncolb = m;
if (transb) ncolb = k;
blasint ncola;
#if !defined(COMPLEX)
blasint ncolb;
#endif
ncola = m;
if (transa & 1) ncola = k;
ncolb = k;
#if defined(COMPLEX)
nrowb = m;
#endif

if (transb & 1) {
#if defined(COMPLEX)
nrowb = k;
#endif
ncolb = m;
}

if (ldc < MAX(1,m))
info = 13;
if (ldb < MAX(1, ncolb))
info = 10;
if (lda < MAX(1, ncola))
info = 8;
if (lda < MAX(1, ncola))
info = 10;
if (k < 0)
info = 5;
if (m < 0)
info = 4;
if (transb < 0)
info = 3;
if (transa < 0)
info = 2;
if (transa < 0)
info = 3;
if (uplo < 0)
info = 1;
}
Expand Down Expand Up @@ -428,7 +465,20 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,

IDEBUG_START;

const blasint incb = (transb == 0) ? 1 : ldb;
#if defined(COMPLEX)
if (transb > 1){
#ifndef CBLAS
IMATCOPY_K_CNC(nrowb, ncolb, (FLOAT)(1.0), (FLOAT)(0.0), b, ldb);
#else
if (order == CblasColMajor)
IMATCOPY_K_CNC(nrowb, ncolb, (FLOAT)(1.0), (FLOAT)(0.0), b, ldb);
if (order == CblasRowMajor)
IMATCOPY_K_RNC(nrowb, ncolb, (FLOAT)(1.0), (FLOAT)(0.0), b, ldb);
#endif
}
#endif

const blasint incb = ((transb & 1) == 0) ? 1 : ldb;

if (uplo == 1) {
for (i = 0; i < m; i++) {
Expand All @@ -438,19 +488,19 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
#if defined(COMPLEX)
aa = a + i * 2;
bb = b + i * ldb * 2;
if (transa) {
if (transa & 1) {
aa = a + lda * i * 2;
}
if (transb)
if (transb & 1)
bb = b + i * 2;
cc = c + i * 2 * ldc + i * 2;
#else
aa = a + i;
bb = b + i * ldb;
if (transa) {
if (transa & 1) {
aa = a + lda * i;
}
if (transb)
if (transb & 1)
bb = b + i;
cc = c + i * ldc + i;
#endif
Expand All @@ -461,7 +511,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
NULL, 0);

if (alpha_r == ZERO && alpha_i == ZERO)
return;
continue;
#else
if (beta != ONE)
SCAL_K(l, 0, 0, beta, cc, 1, NULL, 0, NULL, 0);
Expand Down Expand Up @@ -491,7 +541,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
#endif

#if defined(COMPLEX)
if (!transa)
if (!(transa & 1))
(gemv[(int)transa]) (j, k, 0, alpha_r, alpha_i,
aa, lda, bb, incb, cc, 1,
buffer);
Expand All @@ -500,7 +550,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
aa, lda, bb, incb, cc, 1,
buffer);
#else
if (!transa)
if (!(transa & 1))
(gemv[(int)transa]) (j, k, 0, alpha, aa, lda,
bb, incb, cc, 1, buffer);
else
Expand All @@ -509,7 +559,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
#endif
#ifdef SMP
} else {
if (!transa)
if (!(transa & 1))
(gemv_thread[(int)transa]) (j, k, alpha, aa,
lda, bb, incb, cc,
1, buffer,
Expand All @@ -533,13 +583,13 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
l = j;
#if defined COMPLEX
bb = b + i * ldb * 2;
if (transb) {
if (transb & 1) {
bb = b + i * 2;
}
cc = c + i * 2 * ldc;
#else
bb = b + i * ldb;
if (transb) {
if (transb & 1) {
bb = b + i;
}
cc = c + i * ldc;
Expand All @@ -551,7 +601,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
NULL, 0);

if (alpha_r == ZERO && alpha_i == ZERO)
return;
continue;
#else
if (beta != ONE)
SCAL_K(l, 0, 0, beta, cc, 1, NULL, 0, NULL, 0);
Expand Down Expand Up @@ -580,7 +630,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
#endif

#if defined(COMPLEX)
if (!transa)
if (!(transa & 1))
(gemv[(int)transa]) (j, k, 0, alpha_r, alpha_i,
a, lda, bb, incb, cc, 1,
buffer);
Expand All @@ -589,7 +639,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
a, lda, bb, incb, cc, 1,
buffer);
#else
if (!transa)
if (!(transa & 1))
(gemv[(int)transa]) (j, k, 0, alpha, a, lda, bb,
incb, cc, 1, buffer);
else
Expand All @@ -599,7 +649,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,

#ifdef SMP
} else {
if (!transa)
if (!(transa & 1))
(gemv_thread[(int)transa]) (j, k, alpha, a, lda,
bb, incb, cc, 1,
buffer, nthreads);
Expand All @@ -617,4 +667,4 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
IDEBUG_END;

return;
}
}
1 change: 0 additions & 1 deletion kernel/generic/zimatcopy_cnc.c
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,6 @@ int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a,

if ( rows <= 0 ) return(0);
if ( cols <= 0 ) return(0);
if ( alpha_r == 1.0 && alpha_i == 0.0 ) return (0);

aptr = a;
lda *= 2;
Expand Down