| @@ -19,6 +19,7 @@ common: | |||||
| * Fixed #25 a wrong result of rotmg. | * Fixed #25 a wrong result of rotmg. | ||||
| * Fixed a bug about detecting underscore prefix in c_check. | * Fixed a bug about detecting underscore prefix in c_check. | ||||
| * Print the wall time (cycles) with enabling FUNCTION_PROFILE | * Print the wall time (cycles) with enabling FUNCTION_PROFILE | ||||
| * Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1 | |||||
| x86/x86_64: | x86/x86_64: | ||||
| * Fixed #28 a wrong result of dsdot on x86_64. | * Fixed #28 a wrong result of dsdot on x86_64. | ||||
| @@ -515,6 +515,10 @@ ifeq ($(DYNAMIC_ARCH), 1) | |||||
| CCOMMON_OPT += -DDYNAMIC_ARCH | CCOMMON_OPT += -DDYNAMIC_ARCH | ||||
| endif | endif | ||||
| ifeq ($(NO_LAPACK), 1) | |||||
| CCOMMON_OPT += -DNO_LAPACK | |||||
| endif | |||||
| ifdef SMP | ifdef SMP | ||||
| CCOMMON_OPT += -DSMP_SERVER | CCOMMON_OPT += -DSMP_SERVER | ||||
| @@ -53,6 +53,11 @@ SBLASOBJS += setparam$(TSUFFIX).$(SUFFIX) | |||||
| CCOMMON_OPT += -DTS=$(TSUFFIX) | CCOMMON_OPT += -DTS=$(TSUFFIX) | ||||
| endif | endif | ||||
| KERNEL_INTERFACE = ../common_level1.h ../common_level2.h ../common_level3.h | |||||
| ifneq ($(NO_LAPACK), 1) | |||||
| KERNEL_INTERFACE += ../common_lapack.h | |||||
| endif | |||||
| ifeq ($(ARCH), x86) | ifeq ($(ARCH), x86) | ||||
| COMMONOBJS += cpuid.$(SUFFIX) | COMMONOBJS += cpuid.$(SUFFIX) | ||||
| endif | endif | ||||
| @@ -88,9 +93,10 @@ setparam$(TSUFFIX).$(SUFFIX): setparam$(TSUFFIX).c kernel$(TSUFFIX).h | |||||
| setparam$(TSUFFIX).c : setparam-ref.c | setparam$(TSUFFIX).c : setparam-ref.c | ||||
| sed 's/TS/$(TSUFFIX)/g' $< > $(@F) | sed 's/TS/$(TSUFFIX)/g' $< > $(@F) | ||||
| kernel$(TSUFFIX).h : ../common_level1.h ../common_level2.h ../common_level3.h ../common_lapack.h | |||||
| kernel$(TSUFFIX).h : $(KERNEL_INTERFACE) | |||||
| sed 's/\ *(/$(TSUFFIX)(/g' $^ > $(@F) | sed 's/\ *(/$(TSUFFIX)(/g' $^ > $(@F) | ||||
| cpuid.$(SUFFIX): $(KERNELDIR)/cpuid.S | cpuid.$(SUFFIX): $(KERNELDIR)/cpuid.S | ||||
| $(CC) -c $(CFLAGS) $< -o $(@F) | $(CC) -c $(CFLAGS) $< -o $(@F) | ||||
| @@ -101,7 +101,11 @@ gotoblas_t TABLE_NAME = { | |||||
| #endif | #endif | ||||
| ssymm_outcopyTS, ssymm_oltcopyTS, | ssymm_outcopyTS, ssymm_oltcopyTS, | ||||
| #ifndef NO_LAPACK | |||||
| sneg_tcopyTS, slaswp_ncopyTS, | sneg_tcopyTS, slaswp_ncopyTS, | ||||
| #else | |||||
| NULL,NULL, | |||||
| #endif | |||||
| 0, 0, 0, | 0, 0, 0, | ||||
| DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N, MAX(DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N), | DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N, MAX(DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N), | ||||
| @@ -147,7 +151,11 @@ gotoblas_t TABLE_NAME = { | |||||
| #endif | #endif | ||||
| dsymm_outcopyTS, dsymm_oltcopyTS, | dsymm_outcopyTS, dsymm_oltcopyTS, | ||||
| #ifndef NO_LAPACK | |||||
| dneg_tcopyTS, dlaswp_ncopyTS, | dneg_tcopyTS, dlaswp_ncopyTS, | ||||
| #else | |||||
| NULL, NULL, | |||||
| #endif | |||||
| #ifdef EXPRECISION | #ifdef EXPRECISION | ||||
| @@ -286,7 +294,11 @@ gotoblas_t TABLE_NAME = { | |||||
| chemm3m_oucopyrTS, chemm3m_olcopyrTS, | chemm3m_oucopyrTS, chemm3m_olcopyrTS, | ||||
| chemm3m_oucopyiTS, chemm3m_olcopyiTS, | chemm3m_oucopyiTS, chemm3m_olcopyiTS, | ||||
| #ifndef NO_LAPACK | |||||
| cneg_tcopyTS, claswp_ncopyTS, | cneg_tcopyTS, claswp_ncopyTS, | ||||
| #else | |||||
| NULL, NULL, | |||||
| #endif | |||||
| 0, 0, 0, | 0, 0, 0, | ||||
| ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N, MAX(ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N), | ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N, MAX(ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N), | ||||
| @@ -375,7 +387,11 @@ gotoblas_t TABLE_NAME = { | |||||
| zhemm3m_oucopyrTS, zhemm3m_olcopyrTS, | zhemm3m_oucopyrTS, zhemm3m_olcopyrTS, | ||||
| zhemm3m_oucopyiTS, zhemm3m_olcopyiTS, | zhemm3m_oucopyiTS, zhemm3m_olcopyiTS, | ||||
| #ifndef NO_LAPACK | |||||
| zneg_tcopyTS, zlaswp_ncopyTS, | zneg_tcopyTS, zlaswp_ncopyTS, | ||||
| #else | |||||
| NULL, NULL, | |||||
| #endif | |||||
| #ifdef EXPRECISION | #ifdef EXPRECISION | ||||