Browse Source

Change a1b0 gemm to b0 gemm.

tags/v0.3.18
Xianyi Zhang Wangyang Guo 5 years ago
parent
commit
17d32a4a82
10 changed files with 70 additions and 70 deletions
  1. +4
    -4
      common_d.h
  2. +9
    -9
      common_level3.h
  3. +12
    -12
      common_macro.h
  4. +4
    -4
      common_s.h
  5. +5
    -5
      interface/gemm.c
  6. +28
    -28
      kernel/Makefile.L3
  7. +2
    -2
      kernel/generic/gemm_small_matrix_kernel_b0_nn.c
  8. +2
    -2
      kernel/generic/gemm_small_matrix_kernel_b0_nt.c
  9. +2
    -2
      kernel/generic/gemm_small_matrix_kernel_b0_tn.c
  10. +2
    -2
      kernel/generic/gemm_small_matrix_kernel_b0_tt.c

+ 4
- 4
common_d.h View File

@@ -163,10 +163,10 @@
#define DGEMM_SMALL_KERNEL_TN dgemm_small_kernel_tn
#define DGEMM_SMALL_KERNEL_TT dgemm_small_kernel_tt

#define DGEMM_SMALL_KERNEL_A1B0_NN dgemm_small_kernel_a1b0_nn
#define DGEMM_SMALL_KERNEL_A1B0_NT dgemm_small_kernel_a1b0_nt
#define DGEMM_SMALL_KERNEL_A1B0_TN dgemm_small_kernel_a1b0_tn
#define DGEMM_SMALL_KERNEL_A1B0_TT dgemm_small_kernel_a1b0_tt
#define DGEMM_SMALL_KERNEL_B0_NN dgemm_small_kernel_b0_nn
#define DGEMM_SMALL_KERNEL_B0_NT dgemm_small_kernel_b0_nt
#define DGEMM_SMALL_KERNEL_B0_TN dgemm_small_kernel_b0_tn
#define DGEMM_SMALL_KERNEL_B0_TT dgemm_small_kernel_b0_tt

#else



+ 9
- 9
common_level3.h View File

@@ -526,15 +526,15 @@ int dgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLO
int dgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
int dgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);

int sgemm_small_kernel_a1b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int sgemm_small_kernel_a1b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int sgemm_small_kernel_a1b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int sgemm_small_kernel_a1b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int dgemm_small_kernel_a1b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int dgemm_small_kernel_a1b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int dgemm_small_kernel_a1b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int dgemm_small_kernel_a1b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int sgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int sgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int sgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int sgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
int dgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int dgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int dgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
int dgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);

#endif



+ 12
- 12
common_macro.h View File

@@ -648,10 +648,10 @@
#define GEMM_SMALL_KERNEL_NT DGEMM_SMALL_KERNEL_NT
#define GEMM_SMALL_KERNEL_TN DGEMM_SMALL_KERNEL_TN
#define GEMM_SMALL_KERNEL_TT DGEMM_SMALL_KERNEL_TT
#define GEMM_SMALL_KERNEL_A1B0_NN DGEMM_SMALL_KERNEL_A1B0_NN
#define GEMM_SMALL_KERNEL_A1B0_NT DGEMM_SMALL_KERNEL_A1B0_NT
#define GEMM_SMALL_KERNEL_A1B0_TN DGEMM_SMALL_KERNEL_A1B0_TN
#define GEMM_SMALL_KERNEL_A1B0_TT DGEMM_SMALL_KERNEL_A1B0_TT
#define GEMM_SMALL_KERNEL_B0_NN DGEMM_SMALL_KERNEL_B0_NN
#define GEMM_SMALL_KERNEL_B0_NT DGEMM_SMALL_KERNEL_B0_NT
#define GEMM_SMALL_KERNEL_B0_TN DGEMM_SMALL_KERNEL_B0_TN
#define GEMM_SMALL_KERNEL_B0_TT DGEMM_SMALL_KERNEL_B0_TT

#elif defined(BFLOAT16)

@@ -945,10 +945,10 @@
#define GEMM_SMALL_KERNEL_TN SGEMM_SMALL_KERNEL_TN
#define GEMM_SMALL_KERNEL_TT SGEMM_SMALL_KERNEL_TT

#define GEMM_SMALL_KERNEL_A1B0_NN SGEMM_SMALL_KERNEL_A1B0_NN
#define GEMM_SMALL_KERNEL_A1B0_NT SGEMM_SMALL_KERNEL_A1B0_NT
#define GEMM_SMALL_KERNEL_A1B0_TN SGEMM_SMALL_KERNEL_A1B0_TN
#define GEMM_SMALL_KERNEL_A1B0_TT SGEMM_SMALL_KERNEL_A1B0_TT
#define GEMM_SMALL_KERNEL_B0_NN SGEMM_SMALL_KERNEL_B0_NN
#define GEMM_SMALL_KERNEL_B0_NT SGEMM_SMALL_KERNEL_B0_NT
#define GEMM_SMALL_KERNEL_B0_TN SGEMM_SMALL_KERNEL_B0_TN
#define GEMM_SMALL_KERNEL_B0_TT SGEMM_SMALL_KERNEL_B0_TT

#endif

@@ -1261,10 +1261,10 @@
#define GEMM_SMALL_KERNEL_TN SGEMM_SMALL_KERNEL_TN
#define GEMM_SMALL_KERNEL_TT SGEMM_SMALL_KERNEL_TT

#define GEMM_SMALL_KERNEL_A1B0_NN SGEMM_SMALL_KERNEL_A1B0_NN
#define GEMM_SMALL_KERNEL_A1B0_NT SGEMM_SMALL_KERNEL_A1B0_NT
#define GEMM_SMALL_KERNEL_A1B0_TN SGEMM_SMALL_KERNEL_A1B0_TN
#define GEMM_SMALL_KERNEL_A1B0_TT SGEMM_SMALL_KERNEL_A1B0_TT
#define GEMM_SMALL_KERNEL_B0_NN SGEMM_SMALL_KERNEL_B0_NN
#define GEMM_SMALL_KERNEL_B0_NT SGEMM_SMALL_KERNEL_B0_NT
#define GEMM_SMALL_KERNEL_B0_TN SGEMM_SMALL_KERNEL_B0_TN
#define GEMM_SMALL_KERNEL_B0_TT SGEMM_SMALL_KERNEL_B0_TT

#endif
#else


+ 4
- 4
common_s.h View File

@@ -169,10 +169,10 @@
#define SGEMM_SMALL_KERNEL_TN sgemm_small_kernel_tn
#define SGEMM_SMALL_KERNEL_TT sgemm_small_kernel_tt

#define SGEMM_SMALL_KERNEL_A1B0_NN sgemm_small_kernel_a1b0_nn
#define SGEMM_SMALL_KERNEL_A1B0_NT sgemm_small_kernel_a1b0_nt
#define SGEMM_SMALL_KERNEL_A1B0_TN sgemm_small_kernel_a1b0_tn
#define SGEMM_SMALL_KERNEL_A1B0_TT sgemm_small_kernel_a1b0_tt
#define SGEMM_SMALL_KERNEL_B0_NN sgemm_small_kernel_b0_nn
#define SGEMM_SMALL_KERNEL_B0_NT sgemm_small_kernel_b0_nt
#define SGEMM_SMALL_KERNEL_B0_TN sgemm_small_kernel_b0_tn
#define SGEMM_SMALL_KERNEL_B0_TT sgemm_small_kernel_b0_tt

#else



+ 5
- 5
interface/gemm.c View File

@@ -116,11 +116,11 @@ static int (*gemm_small_kernel[])(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLON
#endif
};

static int (*gemm_small_kernel_a1b0[])(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG) = {
static int (*gemm_small_kernel_b0[])(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT, FLOAT *, BLASLONG, FLOAT *, BLASLONG) = {
#ifndef GEMM3M
#ifndef COMPLEX
GEMM_SMALL_KERNEL_A1B0_NN, GEMM_SMALL_KERNEL_A1B0_TN, NULL, NULL,
GEMM_SMALL_KERNEL_A1B0_NT, GEMM_SMALL_KERNEL_A1B0_TT, NULL, NULL,
GEMM_SMALL_KERNEL_B0_NN, GEMM_SMALL_KERNEL_B0_TN, NULL, NULL,
GEMM_SMALL_KERNEL_B0_NT, GEMM_SMALL_KERNEL_B0_TT, NULL, NULL,
#endif
#endif
};
@@ -450,8 +450,8 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
//need to tune small matrices cases.
if(MNK <= 100.0*100.0*100.0){

if(*(FLOAT *)(args.alpha) == 1.0 && *(FLOAT *)(args.beta) == 0.0){
(gemm_small_kernel_a1b0[(transb << 2) | transa])(args.m, args.n, args.k, args.a, args.lda,args.b, args.ldb, args.c, args.ldc);
if(*(FLOAT *)(args.beta) == 0.0){
(gemm_small_kernel_b0[(transb << 2) | transa])(args.m, args.n, args.k, args.a, args.lda, *(FLOAT *)(args.alpha), args.b, args.ldb, args.c, args.ldc);
}else{
(gemm_small_kernel[(transb << 2) | transa])(args.m, args.n, args.k, args.a, args.lda, *(FLOAT *)(args.alpha), args.b, args.ldb, *(FLOAT *)(args.beta), args.c, args.ldc);
}


+ 28
- 28
kernel/Makefile.L3 View File

@@ -453,14 +453,14 @@ ifeq ($(SMALL_MATRIX_OPT), 1)
SBLASOBJS += \
sgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
sgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
sgemm_small_kernel_a1b0_nn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_a1b0_nt$(TSUFFIX).$(SUFFIX) \
sgemm_small_kernel_a1b0_tn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_a1b0_tt$(TSUFFIX).$(SUFFIX)
sgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
sgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX)

DBLASOBJS += \
dgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
dgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
dgemm_small_kernel_a1b0_nn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_a1b0_nt$(TSUFFIX).$(SUFFIX) \
dgemm_small_kernel_a1b0_tn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_a1b0_tt$(TSUFFIX).$(SUFFIX)
dgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
dgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX)

endif

@@ -4286,32 +4286,32 @@ $(KDIR)dgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_
$(KDIR)dgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_TT)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@

ifndef DGEMM_SMALL_K_A1B0_NN
DGEMM_SMALL_K_A1B0_NN = ../generic/gemm_small_matrix_kernel_a1b0_nn.c
ifndef DGEMM_SMALL_K_B0_NN
DGEMM_SMALL_K_B0_NN = ../generic/gemm_small_matrix_kernel_b0_nn.c
endif

ifndef DGEMM_SMALL_K_A1B0_NT
DGEMM_SMALL_K_A1B0_NT = ../generic/gemm_small_matrix_kernel_a1b0_nt.c
ifndef DGEMM_SMALL_K_B0_NT
DGEMM_SMALL_K_B0_NT = ../generic/gemm_small_matrix_kernel_b0_nt.c
endif

ifndef DGEMM_SMALL_K_A1B0_TN
DGEMM_SMALL_K_A1B0_TN = ../generic/gemm_small_matrix_kernel_a1b0_tn.c
ifndef DGEMM_SMALL_K_B0_TN
DGEMM_SMALL_K_B0_TN = ../generic/gemm_small_matrix_kernel_b0_tn.c
endif

ifndef DGEMM_SMALL_K_A1B0_TT
DGEMM_SMALL_K_A1B0_TT = ../generic/gemm_small_matrix_kernel_a1b0_tt.c
ifndef DGEMM_SMALL_K_B0_TT
DGEMM_SMALL_K_B0_TT = ../generic/gemm_small_matrix_kernel_b0_tt.c
endif

$(KDIR)dgemm_small_kernel_a1b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_A1B0_NN)
$(KDIR)dgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_NN)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@

$(KDIR)dgemm_small_kernel_a1b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_A1B0_NT)
$(KDIR)dgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_NT)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@

$(KDIR)dgemm_small_kernel_a1b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_A1B0_TN)
$(KDIR)dgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_TN)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@

$(KDIR)dgemm_small_kernel_a1b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_A1B0_TT)
$(KDIR)dgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_TT)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@


@@ -4343,30 +4343,30 @@ $(KDIR)sgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_
$(KDIR)sgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_TT)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@

ifndef SGEMM_SMALL_K_A1B0_NN
SGEMM_SMALL_K_A1B0_NN = ../generic/gemm_small_matrix_kernel_a1b0_nn.c
ifndef SGEMM_SMALL_K_B0_NN
SGEMM_SMALL_K_B0_NN = ../generic/gemm_small_matrix_kernel_b0_nn.c
endif

ifndef SGEMM_SMALL_K_A1B0_NT
SGEMM_SMALL_K_A1B0_NT = ../generic/gemm_small_matrix_kernel_a1b0_nt.c
ifndef SGEMM_SMALL_K_B0_NT
SGEMM_SMALL_K_B0_NT = ../generic/gemm_small_matrix_kernel_b0_nt.c
endif

ifndef SGEMM_SMALL_K_A1B0_TN
SGEMM_SMALL_K_A1B0_TN = ../generic/gemm_small_matrix_kernel_a1b0_tn.c
ifndef SGEMM_SMALL_K_B0_TN
SGEMM_SMALL_K_B0_TN = ../generic/gemm_small_matrix_kernel_b0_tn.c
endif

ifndef SGEMM_SMALL_K_A1B0_TT
SGEMM_SMALL_K_A1B0_TT = ../generic/gemm_small_matrix_kernel_a1b0_tt.c
ifndef SGEMM_SMALL_K_B0_TT
SGEMM_SMALL_K_B0_TT = ../generic/gemm_small_matrix_kernel_b0_tt.c
endif

$(KDIR)sgemm_small_kernel_a1b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_A1B0_NN)
$(KDIR)sgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_NN)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@

$(KDIR)sgemm_small_kernel_a1b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_A1B0_NT)
$(KDIR)sgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_NT)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@

$(KDIR)sgemm_small_kernel_a1b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_A1B0_TN)
$(KDIR)sgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_TN)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@

$(KDIR)sgemm_small_kernel_a1b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_A1B0_TT)
$(KDIR)sgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_TT)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@

kernel/generic/gemm_small_matrix_kernel_a1b0_nn.c → kernel/generic/gemm_small_matrix_kernel_b0_nn.c View File

@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#include "common.h"

int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT * B, BLASLONG ldb,FLOAT * C, BLASLONG ldc)
int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha, FLOAT * B, BLASLONG ldb,FLOAT * C, BLASLONG ldc)
{
//naive implemtation
//Column major
@@ -41,7 +41,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT * B
for(k=0; k<K; k++){
result += A[i+k*lda] * B[k+j*ldb];
}
C[i+j*ldc]=result;
C[i+j*ldc]=alpha * result;
}
}

kernel/generic/gemm_small_matrix_kernel_a1b0_nt.c → kernel/generic/gemm_small_matrix_kernel_b0_nt.c View File

@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#include "common.h"

int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
{
//naive implemtation
//Column major
@@ -41,7 +41,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT * B
for(k=0; k<K; k++){
result += A[i+k*lda] * B[k*ldb+j];
}
C[i+j*ldc]=result;
C[i+j*ldc]=alpha * result;
}
}

kernel/generic/gemm_small_matrix_kernel_a1b0_tn.c → kernel/generic/gemm_small_matrix_kernel_b0_tn.c View File

@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#include "common.h"

int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda,FLOAT * B, BLASLONG ldb,FLOAT * C, BLASLONG ldc)
int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha, FLOAT * B, BLASLONG ldb,FLOAT * C, BLASLONG ldc)
{
//naive implemtation
//Column major
@@ -41,7 +41,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda,FLOAT * B,
for(k=0; k<K; k++){
result += A[i*lda+k] * B[k+j*ldb];
}
C[i+j*ldc]=result;
C[i+j*ldc]=alpha * result;
}
}


kernel/generic/gemm_small_matrix_kernel_a1b0_tt.c → kernel/generic/gemm_small_matrix_kernel_b0_tt.c View File

@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#include "common.h"

int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
{
//naive implemtation
//Column major
@@ -41,7 +41,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT * B
for(k=0; k<K; k++){
result += A[i*lda+k] * B[k*ldb+j];
}
C[i+j*ldc]=result;
C[i+j*ldc]=alpha * result;
}
}


Loading…
Cancel
Save