floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add small marix optimization kernel interface.

Browse Source 
make SMALL_MATRIX_OPT=1
This commit is contained in:
Xianyi Zhang
2020-04-28 19:01:36 +08:00
committed by Wangyang Guo
parent e6d6d3ee43
commit 0a2077901c
11 changed files with 340 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
kernel/Makefile.L3
+73
View File
@@ -447,6 +447,19 @@ XBLASOBJS += \
endif
###### BLAS small matrix optimization #####
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).$(SUFFIX)
DBLASOBJS += \
dgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
dgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX)
endif
###### BLAS extensions #####
ifeq ($(BUILD_SINGLE),1)
@@ -4237,3 +4250,63 @@ endif
$(KDIR)zgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEADD_K)
$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM $< -o $@
###### BLAS small matrix optimization #####
ifndef DGEMM_SAMLL_K_NN
DGEMM_SAMLL_K_NN = ../generic/gemm_small_matrix_kernel_nn.c
endif
ifndef DGEMM_SAMLL_K_NT
DGEMM_SAMLL_K_NT = ../generic/gemm_small_matrix_kernel_nt.c
endif
ifndef DGEMM_SAMLL_K_TN
DGEMM_SAMLL_K_TN = ../generic/gemm_small_matrix_kernel_tn.c
endif
ifndef DGEMM_SAMLL_K_TT
DGEMM_SAMLL_K_TT = ../generic/gemm_small_matrix_kernel_tt.c
endif
$(KDIR)dgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SAMLL_K_NN)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
$(KDIR)dgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SAMLL_K_NT)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
$(KDIR)dgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SAMLL_K_TN)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
$(KDIR)dgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SAMLL_K_TT)
$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
ifndef SGEMM_SAMLL_K_NN
SGEMM_SAMLL_K_NN = ../generic/gemm_small_matrix_kernel_nn.c
endif
ifndef SGEMM_SAMLL_K_NT
SGEMM_SAMLL_K_NT = ../generic/gemm_small_matrix_kernel_nt.c
endif
ifndef SGEMM_SAMLL_K_TN
SGEMM_SAMLL_K_TN = ../generic/gemm_small_matrix_kernel_tn.c
endif
ifndef SGEMM_SAMLL_K_TT
SGEMM_SAMLL_K_TT = ../generic/gemm_small_matrix_kernel_tt.c
endif
$(KDIR)sgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SAMLL_K_NN)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
$(KDIR)sgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SAMLL_K_NT)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
$(KDIR)sgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SAMLL_K_TN)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
$(KDIR)sgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SAMLL_K_TT)
$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@