Skip to content
Merged
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
3 changes: 3 additions & 0 deletions cmake/kernel.cmake
Original file line number Diff line number Diff line change
Expand Up @@ -79,6 +79,9 @@ macro(SetDefaultL1)
SetFallback(CROTKERNEL zrot.S)
SetFallback(ZROTKERNEL zrot.S)
SetFallback(XROTKERNEL zrot.S)
SetFallback(SROTMKERNEL rotm.S)
SetFallback(DROTMKERNEL rotm.S)
SetFallback(QROTMKERNEL rotm.S)
SetFallback(SSCALKERNEL scal.S)
SetFallback(DSCALKERNEL scal.S)
SetFallback(CSCALKERNEL zscal.S)
Expand Down
2 changes: 2 additions & 0 deletions common_d.h
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
#define DSUM_K dsum_k
#define DSWAP_K dswap_k
#define DROT_K drot_k
#define DROTM_K drotm_k

#define DGEMV_N dgemv_n
#define DGEMV_T dgemv_t
Expand Down Expand Up @@ -180,6 +181,7 @@
#define DSUM_K gotoblas -> dsum_k
#define DSWAP_K gotoblas -> dswap_k
#define DROT_K gotoblas -> drot_k
#define DROTM_K gotoblas -> drotm_k

#define DGEMV_N gotoblas -> dgemv_n
#define DGEMV_T gotoblas -> dgemv_t
Expand Down
6 changes: 3 additions & 3 deletions common_level1.h
Original file line number Diff line number Diff line change
Expand Up @@ -213,9 +213,9 @@ int srotmg_k(float *, float *, float *, float *, float *);
int drotmg_k(double *, double *, double *, double *, double *);
int qrotmg_k(xdouble *, xdouble *, xdouble *, xdouble *, xdouble *);

int srotm_k (BLASLONG, float, BLASLONG, float, BLASLONG, float);
int drotm_k (BLASLONG, double, BLASLONG, double, BLASLONG, double);
int qrotm_k (BLASLONG, xdouble, BLASLONG, xdouble, BLASLONG, xdouble);
int srotm_k (BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
int drotm_k (BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
int qrotm_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);


int saxpby_k (BLASLONG, float, float *, BLASLONG, float, float *, BLASLONG);
Expand Down
3 changes: 3 additions & 0 deletions common_macro.h
Original file line number Diff line number Diff line change
Expand Up @@ -70,6 +70,7 @@
#define SUM_K QSUM_K
#define SWAP_K QSWAP_K
#define ROT_K QROT_K
#define ROTM_K QROTM_K

#define GEMV_N QGEMV_N
#define GEMV_T QGEMV_T
Expand Down Expand Up @@ -361,6 +362,7 @@
#define SUM_K DSUM_K
#define SWAP_K DSWAP_K
#define ROT_K DROT_K
#define ROTM_K DROTM_K

#define GEMV_N DGEMV_N
#define GEMV_T DGEMV_T
Expand Down Expand Up @@ -977,6 +979,7 @@
#define SUM_K SSUM_K
#define SWAP_K SSWAP_K
#define ROT_K SROT_K
#define ROTM_K SROTM_K

#define GEMV_N SGEMV_N
#define GEMV_T SGEMV_T
Expand Down
3 changes: 3 additions & 0 deletions common_param.h
Original file line number Diff line number Diff line change
Expand Up @@ -197,6 +197,7 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
//double (*dsdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);

int (*srot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
int (*srotm_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
#endif
#if (BUILD_SINGLE==1) || (BUILD_DOUBLE==1) || (BUILD_COMPLEX==1)
int (*saxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
Expand Down Expand Up @@ -330,6 +331,7 @@ BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
#endif
#if (BUILD_DOUBLE==1) || (BUILD_COMPLEX16==1)
int (*drot_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
int (*drotm_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
int (*daxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
int (*dscal_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
int (*dswap_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
Expand Down Expand Up @@ -439,6 +441,7 @@ BLASLONG (*iqmin_k) (BLASLONG, xdouble *, BLASLONG);
int (*qcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
xdouble (*qdot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
int (*qrot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
int (*qrotm_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);

int (*qaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
int (*qscal_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
Expand Down
2 changes: 2 additions & 0 deletions common_q.h
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
#define QSUM_K qsum_k
#define QSWAP_K qswap_k
#define QROT_K qrot_k
#define QROTM_K qrotm_k

#define QGEMV_N qgemv_n
#define QGEMV_T qgemv_t
Expand Down Expand Up @@ -165,6 +166,7 @@
#define QSUM_K gotoblas -> qsum_k
#define QSWAP_K gotoblas -> qswap_k
#define QROT_K gotoblas -> qrot_k
#define QROTM_K gotoblas -> qrotm_k

#define QGEMV_N gotoblas -> qgemv_n
#define QGEMV_T gotoblas -> qgemv_t
Expand Down
2 changes: 2 additions & 0 deletions common_s.h
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@
#define SSCAL_K sscal_k
#define SSWAP_K sswap_k
#define SROT_K srot_k
#define SROTM_K srotm_k

#define SGEMV_N sgemv_n
#define SGEMV_T sgemv_t
Expand Down Expand Up @@ -189,6 +190,7 @@
#define SSCAL_K gotoblas -> sscal_k
#define SSWAP_K gotoblas -> sswap_k
#define SROT_K gotoblas -> srot_k
#define SROTM_K gotoblas -> srotm_k

#define SGEMV_N gotoblas -> sgemv_n
#define SGEMV_T gotoblas -> sgemv_t
Expand Down
140 changes: 6 additions & 134 deletions interface/rotm.c
Original file line number Diff line number Diff line change
Expand Up @@ -7,149 +7,21 @@

void NAME(blasint *N, FLOAT *dx, blasint *INCX, FLOAT *dy, blasint *INCY, FLOAT *dparam){

blasint n = *N;
blasint incx = *INCX;
blasint incy = *INCY;
blasint n = *N;
blasint incx = *INCX;
blasint incy = *INCY;

PRINT_DEBUG_NAME
#else

void CNAME(blasint n, FLOAT *dx, blasint incx, FLOAT *dy, blasint incy, FLOAT *dparam){

#endif

blasint i__1, i__2;
PRINT_DEBUG_CNAME;

blasint i__;
FLOAT w, z__;
blasint kx, ky;
FLOAT dh11, dh12, dh22, dh21, dflag;
blasint nsteps;

#ifndef CBLAS
PRINT_DEBUG_CNAME;
#else
PRINT_DEBUG_CNAME;
#endif

--dparam;
--dy;
--dx;

dflag = dparam[1];
if (n <= 0 || dflag == - 2.0) goto L140;

if (! (incx == incy && incx > 0)) goto L70;

nsteps = n * incx;
if (dflag < 0.) {
goto L50;
} else if (dflag == 0) {
goto L10;
} else {
goto L30;
}
L10:
dh12 = dparam[4];
dh21 = dparam[3];
i__1 = nsteps;
i__2 = incx;
for (i__ = 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ += i__2) {
w = dx[i__];
z__ = dy[i__];
dx[i__] = w + z__ * dh12;
dy[i__] = w * dh21 + z__;
/* L20: */
}
goto L140;
L30:
dh11 = dparam[2];
dh22 = dparam[5];
i__2 = nsteps;
i__1 = incx;
for (i__ = 1; i__1 < 0 ? i__ >= i__2 : i__ <= i__2; i__ += i__1) {
w = dx[i__];
z__ = dy[i__];
dx[i__] = w * dh11 + z__;
dy[i__] = -w + dh22 * z__;
/* L40: */
}
goto L140;
L50:
dh11 = dparam[2];
dh12 = dparam[4];
dh21 = dparam[3];
dh22 = dparam[5];
i__1 = nsteps;
i__2 = incx;
for (i__ = 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ += i__2) {
w = dx[i__];
z__ = dy[i__];
dx[i__] = w * dh11 + z__ * dh12;
dy[i__] = w * dh21 + z__ * dh22;
/* L60: */
}
goto L140;
L70:
kx = 1;
ky = 1;
if (incx < 0) {
kx = (1 - n) * incx + 1;
}
if (incy < 0) {
ky = (1 - n) * incy + 1;
}
ROTM_K(n, dx, incx, dy, incy, dparam);

if (dflag < 0.) {
goto L120;
} else if (dflag == 0) {
goto L80;
} else {
goto L100;
}
L80:
dh12 = dparam[4];
dh21 = dparam[3];
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
w = dx[kx];
z__ = dy[ky];
dx[kx] = w + z__ * dh12;
dy[ky] = w * dh21 + z__;
kx += incx;
ky += incy;
/* L90: */
}
goto L140;
L100:
dh11 = dparam[2];
dh22 = dparam[5];
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
w = dx[kx];
z__ = dy[ky];
dx[kx] = w * dh11 + z__;
dy[ky] = -w + dh22 * z__;
kx += incx;
ky += incy;
/* L110: */
}
goto L140;
L120:
dh11 = dparam[2];
dh12 = dparam[4];
dh21 = dparam[3];
dh22 = dparam[5];
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
w = dx[kx];
z__ = dy[ky];
dx[kx] = w * dh11 + z__ * dh12;
dy[ky] = w * dh21 + z__ * dh22;
kx += incx;
ky += incy;
/* L130: */
}
L140:
return;
}

4 changes: 4 additions & 0 deletions kernel/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
GenerateNamedObjects("${KERNELDIR}/${${float_char}COPYKERNEL}" "C_INTERFACE" "copy_k" false "" "" false ${float_type})
GenerateNamedObjects("${KERNELDIR}/${${float_char}NRM2KERNEL}" "" "nrm2_k" false "" "" false ${float_type})
GenerateNamedObjects("${KERNELDIR}/${${float_char}ROTKERNEL}" "" "rot_k" false "" "" false ${float_type})
GenerateNamedObjects("${KERNELDIR}/${${float_char}ROTMKERNEL}" "" "rotm_k" false "" "" false ${float_type})
GenerateNamedObjects("${KERNELDIR}/${${float_char}SCALKERNEL}" "" "scal_k" false "" "" false ${float_type})
GenerateNamedObjects("${KERNELDIR}/${${float_char}SWAPKERNEL}" "" "swap_k" false "" "" false ${float_type})
GenerateNamedObjects("${KERNELDIR}/${${float_char}AXPBYKERNEL}" "" "axpby_k" false "" "" false ${float_type})
Expand Down Expand Up @@ -125,6 +126,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
GenerateNamedObjects("${KERNELDIR}/${SNRM2KERNEL}" "" "nrm2_k" false "" "" false "SINGLE")
GenerateNamedObjects("${KERNELDIR}/${SDOTKERNEL}" "" "dot_k" false "" "" false "SINGLE")
GenerateNamedObjects("${KERNELDIR}/${SROTKERNEL}" "" "rot_k" false "" "" false "SINGLE")
GenerateNamedObjects("${KERNELDIR}/${SROTMKERNEL}" "" "rotm_k" false "" "" false "SINGLE")
endif ()
if (BUILD_COMPLEX16 AND NOT BUILD_DOUBLE)
GenerateNamedObjects("${KERNELDIR}/${DAMAXKERNEL}" "USE_ABS" "amax_k" false "" "" false "DOUBLE")
Expand All @@ -148,6 +150,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
GenerateNamedObjects("${KERNELDIR}/${DCOPYKERNEL}" "C_INTERFACE" "copy_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DNRM2KERNEL}" "" "nrm2_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DROTKERNEL}" "" "rot_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DROTMKERNEL}" "" "rotm_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DDOTKERNEL}" "" "dot_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DSWAPKERNEL}" "" "swap_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DAXPYKERNEL}" "" "axpy_k" false "" "" false "DOUBLE")
Expand Down Expand Up @@ -1105,6 +1108,7 @@ endif ()
GenerateNamedObjects("${KERNELDIR}/${DCOPYKERNEL}" "C_INTERFACE" "copy_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DNRM2KERNEL}" "" "nrm2_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DROTKERNEL}" "" "rot_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DROTMKERNEL}" "" "rotm_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DDOTKERNEL}" "" "dot_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DSWAPKERNEL}" "" "swap_k" false "" "" false "DOUBLE")
GenerateNamedObjects("${KERNELDIR}/${DAXPYKERNEL}" "" "axpy_k" false "" "" false "DOUBLE")
Expand Down
29 changes: 25 additions & 4 deletions kernel/Makefile.L1
Original file line number Diff line number Diff line change
Expand Up @@ -336,6 +336,18 @@ ifndef XROTKERNEL
XROTKERNEL = zrot.S
endif

ifndef SROTMKERNEL
SROTMKERNEL = rotm.S
endif

ifndef DROTMKERNEL
DROTMKERNEL = rotm.S
endif

ifndef QROTMKERNEL
QROTMKERNEL = rotm.S
endif

### SCAL ###

ifndef SSCALKERNEL
Expand Down Expand Up @@ -504,21 +516,21 @@ SBLASOBJS += \
sasum_k$(TSUFFIX).$(SUFFIX) ssum_k$(TSUFFIX).$(SUFFIX) saxpy_k$(TSUFFIX).$(SUFFIX) scopy_k$(TSUFFIX).$(SUFFIX) \
sdot_k$(TSUFFIX).$(SUFFIX) sdsdot_k$(TSUFFIX).$(SUFFIX) dsdot_k$(TSUFFIX).$(SUFFIX) \
snrm2_k$(TSUFFIX).$(SUFFIX) srot_k$(TSUFFIX).$(SUFFIX) sscal_k$(TSUFFIX).$(SUFFIX) sswap_k$(TSUFFIX).$(SUFFIX) \
saxpby_k$(TSUFFIX).$(SUFFIX)
saxpby_k$(TSUFFIX).$(SUFFIX) srotm_k$(TSUFFIX).$(SUFFIX)

DBLASOBJS += \
damax_k$(TSUFFIX).$(SUFFIX) damin_k$(TSUFFIX).$(SUFFIX) dmax_k$(TSUFFIX).$(SUFFIX) dmin_k$(TSUFFIX).$(SUFFIX) \
idamax_k$(TSUFFIX).$(SUFFIX) idamin_k$(TSUFFIX).$(SUFFIX) idmax_k$(TSUFFIX).$(SUFFIX) idmin_k$(TSUFFIX).$(SUFFIX) \
dasum_k$(TSUFFIX).$(SUFFIX) daxpy_k$(TSUFFIX).$(SUFFIX) dcopy_k$(TSUFFIX).$(SUFFIX) ddot_k$(TSUFFIX).$(SUFFIX) \
dnrm2_k$(TSUFFIX).$(SUFFIX) drot_k$(TSUFFIX).$(SUFFIX) dscal_k$(TSUFFIX).$(SUFFIX) dswap_k$(TSUFFIX).$(SUFFIX) \
daxpby_k$(TSUFFIX).$(SUFFIX) dsum_k$(TSUFFIX).$(SUFFIX)
daxpby_k$(TSUFFIX).$(SUFFIX) dsum_k$(TSUFFIX).$(SUFFIX) drotm_k$(TSUFFIX).$(SUFFIX)

QBLASOBJS += \
qamax_k$(TSUFFIX).$(SUFFIX) qamin_k$(TSUFFIX).$(SUFFIX) qmax_k$(TSUFFIX).$(SUFFIX) qmin_k$(TSUFFIX).$(SUFFIX) \
iqamax_k$(TSUFFIX).$(SUFFIX) iqamin_k$(TSUFFIX).$(SUFFIX) iqmax_k$(TSUFFIX).$(SUFFIX) iqmin_k$(TSUFFIX).$(SUFFIX) \
qasum_k$(TSUFFIX).$(SUFFIX) qaxpy_k$(TSUFFIX).$(SUFFIX) qcopy_k$(TSUFFIX).$(SUFFIX) qdot_k$(TSUFFIX).$(SUFFIX) \
qnrm2_k$(TSUFFIX).$(SUFFIX) qrot_k$(TSUFFIX).$(SUFFIX) qscal_k$(TSUFFIX).$(SUFFIX) qswap_k$(TSUFFIX).$(SUFFIX) \
qsum_k$(TSUFFIX).$(SUFFIX)
qsum_k$(TSUFFIX).$(SUFFIX) qrotm_k$(TSUFFIX).$(SUFFIX)

CBLASOBJS += \
camax_k$(TSUFFIX).$(SUFFIX) camin_k$(TSUFFIX).$(SUFFIX) icamax_k$(TSUFFIX).$(SUFFIX) icamin_k$(TSUFFIX).$(SUFFIX) \
Expand Down Expand Up @@ -842,7 +854,16 @@ $(KDIR)drot_k$(TSUFFIX).$(SUFFIX) $(KDIR)drot_k$(TPSUFFIX).$(PSUFFIX) : $(KERN
$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DDOUBLE $< -o $@

$(KDIR)qrot_k$(TSUFFIX).$(SUFFIX) $(KDIR)qrot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QROTKERNEL)
$(CC) -c $(CFLAGS) -UCOMPLEX -UCOMPLEX -DXDOUBLE $< -o $@
$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DXDOUBLE $< -o $@

$(KDIR)srotm_k$(TSUFFIX).$(SUFFIX) $(KDIR)srotm_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SROTMKERNEL)
$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -UDOUBLE $< -o $@

$(KDIR)drotm_k$(TSUFFIX).$(SUFFIX) $(KDIR)drotm_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DROTMKERNEL)
$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DDOUBLE $< -o $@

$(KDIR)qrotm_k$(TSUFFIX).$(SUFFIX) $(KDIR)qrotm_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QROTMKERNEL)
$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DXDOUBLE $< -o $@

$(KDIR)csrot_k$(TSUFFIX).$(SUFFIX) $(KDIR)csrot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CROTKERNEL)
$(CC) -c $(CFLAGS) -DCOMPLEX -DCOMPLEX -UDOUBLE $< -o $@
Expand Down
12 changes: 12 additions & 0 deletions kernel/alpha/KERNEL
Original file line number Diff line number Diff line change
Expand Up @@ -122,3 +122,15 @@ ZTRSMKERNEL_LN = ztrsm_kernel_2x2_LN.S
ZTRSMKERNEL_LT = ztrsm_kernel_2x2_LT.S
ZTRSMKERNEL_RN = ztrsm_kernel_2x2_LT.S
ZTRSMKERNEL_RT = ztrsm_kernel_2x2_RT.S

ifndef SROTMKERNEL
SROTMKERNEL = ../generic/rotm.c
endif

ifndef DROTMKERNEL
DROTMKERNEL = ../generic/rotm.c
endif

ifndef QROTMKERNEL
QROTMKERNEL = ../generic/rotm.c
endif
10 changes: 10 additions & 0 deletions kernel/arm/KERNEL
Original file line number Diff line number Diff line change
Expand Up @@ -43,4 +43,14 @@ ifndef ZGEMM_BETA
ZGEMM_BETA = ../generic/zgemm_beta.c
endif

ifndef SROTMKERNEL
SROTMKERNEL = ../generic/rotm.c
endif

ifndef DROTMKERNEL
DROTMKERNEL = ../generic/rotm.c
endif

ifndef QROTMKERNEL
QROTMKERNEL = ../generic/rotm.c
endif
Loading
Loading