It is really dangerous to fallback to Loongson core on other MIPS64 processors. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>tags/v0.3.22^2
| @@ -677,7 +677,7 @@ endif | |||||
| endif | endif | ||||
| ifeq ($(ARCH), mips64) | ifeq ($(ARCH), mips64) | ||||
| DYNAMIC_CORE = LOONGSON3R3 LOONGSON3R4 | |||||
| DYNAMIC_CORE = LOONGSON3R3 LOONGSON3R4 MIPS64_GENERIC | |||||
| endif | endif | ||||
| ifeq ($(ARCH), loongarch64) | ifeq ($(ARCH), loongarch64) | ||||
| @@ -856,6 +856,11 @@ CCOMMON_OPT += -mabi=32 | |||||
| BINARY_DEFINED = 1 | BINARY_DEFINED = 1 | ||||
| endif | endif | ||||
| ifneq (, $(filter $(CORE), MIPS64_GENERIC)) | |||||
| CCOMMON_OPT += -DNO_MSA | |||||
| FCOMMON_OPT += -DNO_MSA | |||||
| endif | |||||
| ifneq (, $(filter $(CORE),LOONGSON3R3 LOONGSON3R4)) | ifneq (, $(filter $(CORE),LOONGSON3R3 LOONGSON3R4)) | ||||
| CCOMMON_OPT += -march=loongson3a | CCOMMON_OPT += -march=loongson3a | ||||
| FCOMMON_OPT += -march=loongson3a | FCOMMON_OPT += -march=loongson3a | ||||
| @@ -65,6 +65,7 @@ MIPS1004K | |||||
| MIPS24K | MIPS24K | ||||
| 4.MIPS64 CPU: | 4.MIPS64 CPU: | ||||
| MIPS64_GENERIC | |||||
| SICORTEX | SICORTEX | ||||
| LOONGSON3A | LOONGSON3A | ||||
| LOONGSON3B | LOONGSON3B | ||||
| @@ -70,16 +70,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| /* or implied, of The University of Texas at Austin. */ | /* or implied, of The University of Texas at Austin. */ | ||||
| /*********************************************************************/ | /*********************************************************************/ | ||||
| #define CPU_UNKNOWN 0 | |||||
| #define CPU_SICORTEX 1 | |||||
| #define CPU_LOONGSON3R3 2 | |||||
| #define CPU_LOONGSON3R4 3 | |||||
| #define CPU_I6400 4 | |||||
| #define CPU_P6600 5 | |||||
| #define CPU_I6500 6 | |||||
| #define CPU_UNKNOWN 0 | |||||
| #define CPU_MIPS64_GENERIC 1 | |||||
| #define CPU_SICORTEX 2 | |||||
| #define CPU_LOONGSON3R3 3 | |||||
| #define CPU_LOONGSON3R4 4 | |||||
| #define CPU_I6400 5 | |||||
| #define CPU_P6600 6 | |||||
| #define CPU_I6500 7 | |||||
| static char *cpuname[] = { | static char *cpuname[] = { | ||||
| "UNKNOWN", | "UNKNOWN", | ||||
| "MIPS64_GENERIC" | |||||
| "SICORTEX", | "SICORTEX", | ||||
| "LOONGSON3R3", | "LOONGSON3R3", | ||||
| "LOONGSON3R4", | "LOONGSON3R4", | ||||
| @@ -113,8 +115,11 @@ int detect(void){ | |||||
| return CPU_SICORTEX; | return CPU_SICORTEX; | ||||
| } | } | ||||
| } | } | ||||
| return CPU_MIPS64_GENERIC; | |||||
| #else | |||||
| return CPU_UNKNOWN; | |||||
| #endif | #endif | ||||
| return CPU_UNKNOWN; | |||||
| } | } | ||||
| char *get_corename(void){ | char *get_corename(void){ | ||||
| @@ -136,9 +141,11 @@ void get_subarchitecture(void){ | |||||
| printf("P6600"); | printf("P6600"); | ||||
| }else if(detect()==CPU_I6500){ | }else if(detect()==CPU_I6500){ | ||||
| printf("I6500"); | printf("I6500"); | ||||
| }else{ | |||||
| }else if(detect()==CPU_SICORTEX){ | |||||
| printf("SICORTEX"); | printf("SICORTEX"); | ||||
| } | |||||
| }else{ | |||||
| printf("MIPS64_GENERIC"); | |||||
| } | |||||
| } | } | ||||
| void get_subdirname(void){ | void get_subdirname(void){ | ||||
| @@ -215,8 +222,8 @@ void get_libname(void){ | |||||
| printf("p6600\n"); | printf("p6600\n"); | ||||
| }else if(detect()==CPU_I6500) { | }else if(detect()==CPU_I6500) { | ||||
| printf("i6500\n"); | printf("i6500\n"); | ||||
| }else{ | |||||
| printf("mips64\n"); | |||||
| }else { | |||||
| printf("mips64_generic\n"); | |||||
| } | } | ||||
| } | } | ||||
| @@ -49,20 +49,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| extern gotoblas_t gotoblas_LOONGSON3R3; | extern gotoblas_t gotoblas_LOONGSON3R3; | ||||
| extern gotoblas_t gotoblas_LOONGSON3R4; | extern gotoblas_t gotoblas_LOONGSON3R4; | ||||
| extern gotoblas_t gotoblas_MIPS64_GENERIC; | |||||
| extern void openblas_warning(int verbose, const char * msg); | extern void openblas_warning(int verbose, const char * msg); | ||||
| #define NUM_CORETYPES 2 | |||||
| #define NUM_CORETYPES 3 | |||||
| static char *corename[] = { | static char *corename[] = { | ||||
| "MIPS64_GENERIC" | |||||
| "loongson3r3", | "loongson3r3", | ||||
| "loongson3r4", | "loongson3r4", | ||||
| "UNKNOWN" | "UNKNOWN" | ||||
| }; | }; | ||||
| char *gotoblas_corename(void) { | char *gotoblas_corename(void) { | ||||
| if (gotoblas == &gotoblas_LOONGSON3R3) return corename[0]; | |||||
| if (gotoblas == &gotoblas_LOONGSON3R4) return corename[1]; | |||||
| if (gotoblas == &gotoblas_MIPS64_GENERIC) return corename[0]; | |||||
| if (gotoblas == &gotoblas_LOONGSON3R3) return corename[1]; | |||||
| if (gotoblas == &gotoblas_LOONGSON3R4) return corename[2]; | |||||
| return corename[NUM_CORETYPES]; | return corename[NUM_CORETYPES]; | ||||
| } | } | ||||
| @@ -82,8 +85,9 @@ static gotoblas_t *force_coretype(char *coretype) { | |||||
| switch (found) | switch (found) | ||||
| { | { | ||||
| case 0: return (&gotoblas_LOONGSON3R3); | |||||
| case 1: return (&gotoblas_LOONGSON3R4); | |||||
| case 0: return (&gotoblas_MIPS64_GENERIC); | |||||
| case 1: return (&gotoblas_LOONGSON3R3); | |||||
| case 2: return (&gotoblas_LOONGSON3R4); | |||||
| } | } | ||||
| snprintf(message, 128, "Core not found: %s\n", coretype); | snprintf(message, 128, "Core not found: %s\n", coretype); | ||||
| openblas_warning(1, message); | openblas_warning(1, message); | ||||
| @@ -173,9 +177,9 @@ void gotoblas_dynamic_init(void) { | |||||
| if (gotoblas == NULL) | if (gotoblas == NULL) | ||||
| { | { | ||||
| snprintf(coremsg, 128, "Falling back to loongson3r3 core\n"); | |||||
| snprintf(coremsg, 128, "Falling back to MIPS64_GENEIRC\n"); | |||||
| openblas_warning(1, coremsg); | openblas_warning(1, coremsg); | ||||
| gotoblas = &gotoblas_LOONGSON3R3; | |||||
| gotoblas = &gotoblas_MIPS64_GENERIC; | |||||
| } | } | ||||
| if (gotoblas && gotoblas->init) { | if (gotoblas && gotoblas->init) { | ||||
| @@ -131,6 +131,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| /* #define FORCE_PPC440 */ | /* #define FORCE_PPC440 */ | ||||
| /* #define FORCE_PPC440FP2 */ | /* #define FORCE_PPC440FP2 */ | ||||
| /* #define FORCE_CELL */ | /* #define FORCE_CELL */ | ||||
| /* #define FORCE_MIPS64_GENERIC */ | |||||
| /* #define FORCE_SICORTEX */ | /* #define FORCE_SICORTEX */ | ||||
| /* #define FORCE_LOONGSON3R3 */ | /* #define FORCE_LOONGSON3R3 */ | ||||
| /* #define FORCE_LOONGSON3R4 */ | /* #define FORCE_LOONGSON3R4 */ | ||||
| @@ -915,6 +916,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| #define CORENAME "CELL" | #define CORENAME "CELL" | ||||
| #endif | #endif | ||||
| #ifdef FORCE_MIPS64_GENERIC | |||||
| #define FORCE | |||||
| #define ARCHITECTURE "MIPS" | |||||
| #define SUBARCHITECTURE "MIPS64_GENERIC" | |||||
| #define SUBDIRNAME "mips64" | |||||
| #define ARCHCONFIG "-DMIPS64_GENERIC " \ | |||||
| "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=32 " \ | |||||
| "-DL2_SIZE=1048576 -DL2_LINESIZE=32 " \ | |||||
| "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=8 " | |||||
| #define LIBNAME "mips64_generic" | |||||
| #define CORENAME "MIPS64_GENERIC" | |||||
| #else | |||||
| #endif | |||||
| #ifdef FORCE_SICORTEX | #ifdef FORCE_SICORTEX | ||||
| #define FORCE | #define FORCE | ||||
| #define ARCHITECTURE "MIPS" | #define ARCHITECTURE "MIPS" | ||||
| @@ -42,50 +42,58 @@ endif | |||||
| ifndef SGEMMKERNEL | ifndef SGEMMKERNEL | ||||
| SGEMMKERNEL = gemm_kernel.S | SGEMMKERNEL = gemm_kernel.S | ||||
| ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N)) | |||||
| SGEMMINCOPY = ../generic/gemm_ncopy_2.c | SGEMMINCOPY = ../generic/gemm_ncopy_2.c | ||||
| SGEMMITCOPY = ../generic/gemm_tcopy_2.c | SGEMMITCOPY = ../generic/gemm_tcopy_2.c | ||||
| SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX) | |||||
| SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | |||||
| SGEMMONCOPY = ../generic/gemm_ncopy_8.c | SGEMMONCOPY = ../generic/gemm_ncopy_8.c | ||||
| SGEMMOTCOPY = ../generic/gemm_tcopy_8.c | SGEMMOTCOPY = ../generic/gemm_tcopy_8.c | ||||
| SGEMMINCOPYOBJ = sgemm_incopy.o | |||||
| SGEMMITCOPYOBJ = sgemm_itcopy.o | |||||
| SGEMMONCOPYOBJ = sgemm_oncopy.o | |||||
| SGEMMOTCOPYOBJ = sgemm_otcopy.o | |||||
| SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX) | |||||
| SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | endif | ||||
| ifndef DGEMMKERNEL | ifndef DGEMMKERNEL | ||||
| DGEMMKERNEL = gemm_kernel.S | DGEMMKERNEL = gemm_kernel.S | ||||
| ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N)) | |||||
| DGEMMINCOPY = ../generic/gemm_ncopy_2.c | DGEMMINCOPY = ../generic/gemm_ncopy_2.c | ||||
| DGEMMITCOPY = ../generic/gemm_tcopy_2.c | DGEMMITCOPY = ../generic/gemm_tcopy_2.c | ||||
| DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX) | |||||
| DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | |||||
| DGEMMONCOPY = ../generic/gemm_ncopy_8.c | DGEMMONCOPY = ../generic/gemm_ncopy_8.c | ||||
| DGEMMOTCOPY = ../generic/gemm_tcopy_8.c | DGEMMOTCOPY = ../generic/gemm_tcopy_8.c | ||||
| DGEMMINCOPYOBJ = dgemm_incopy.o | |||||
| DGEMMITCOPYOBJ = dgemm_itcopy.o | |||||
| DGEMMONCOPYOBJ = dgemm_oncopy.o | |||||
| DGEMMOTCOPYOBJ = dgemm_otcopy.o | |||||
| DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX) | |||||
| DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | endif | ||||
| ifndef CGEMMKERNEL | ifndef CGEMMKERNEL | ||||
| CGEMMKERNEL = zgemm_kernel.S | CGEMMKERNEL = zgemm_kernel.S | ||||
| ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N)) | |||||
| CGEMMINCOPY = ../generic/zgemm_ncopy_1.c | CGEMMINCOPY = ../generic/zgemm_ncopy_1.c | ||||
| CGEMMITCOPY = ../generic/zgemm_tcopy_1.c | CGEMMITCOPY = ../generic/zgemm_tcopy_1.c | ||||
| CGEMMINCOPYOBJ = cgemm_incopy$(TSUFFIX).$(SUFFIX) | |||||
| CGEMMITCOPYOBJ = cgemm_itcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | |||||
| CGEMMONCOPY = ../generic/zgemm_ncopy_4.c | CGEMMONCOPY = ../generic/zgemm_ncopy_4.c | ||||
| CGEMMOTCOPY = ../generic/zgemm_tcopy_4.c | CGEMMOTCOPY = ../generic/zgemm_tcopy_4.c | ||||
| CGEMMINCOPYOBJ = cgemm_incopy.o | |||||
| CGEMMITCOPYOBJ = cgemm_itcopy.o | |||||
| CGEMMONCOPYOBJ = cgemm_oncopy.o | |||||
| CGEMMOTCOPYOBJ = cgemm_otcopy.o | |||||
| CGEMMONCOPYOBJ = cgemm_oncopy$(TSUFFIX).$(SUFFIX) | |||||
| CGEMMOTCOPYOBJ = cgemm_otcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | endif | ||||
| ifndef ZGEMMKERNEL | ifndef ZGEMMKERNEL | ||||
| ZGEMMKERNEL = zgemm_kernel.S | ZGEMMKERNEL = zgemm_kernel.S | ||||
| ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N)) | |||||
| ZGEMMINCOPY = ../generic/zgemm_ncopy_1.c | ZGEMMINCOPY = ../generic/zgemm_ncopy_1.c | ||||
| ZGEMMITCOPY = ../generic/zgemm_tcopy_1.c | ZGEMMITCOPY = ../generic/zgemm_tcopy_1.c | ||||
| ZGEMMINCOPYOBJ = zgemm_incopy$(TSUFFIX).$(SUFFIX) | |||||
| ZGEMMITCOPYOBJ = zgemm_itcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | |||||
| ZGEMMONCOPY = ../generic/zgemm_ncopy_4.c | ZGEMMONCOPY = ../generic/zgemm_ncopy_4.c | ||||
| ZGEMMOTCOPY = ../generic/zgemm_tcopy_4.c | ZGEMMOTCOPY = ../generic/zgemm_tcopy_4.c | ||||
| ZGEMMINCOPYOBJ = zgemm_incopy.o | |||||
| ZGEMMITCOPYOBJ = zgemm_itcopy.o | |||||
| ZGEMMONCOPYOBJ = zgemm_oncopy.o | |||||
| ZGEMMOTCOPYOBJ = zgemm_otcopy.o | |||||
| ZGEMMONCOPYOBJ = zgemm_oncopy$(TSUFFIX).$(SUFFIX) | |||||
| ZGEMMOTCOPYOBJ = zgemm_otcopy$(TSUFFIX).$(SUFFIX) | |||||
| endif | endif | ||||
| ifndef SGEMM_BETA | ifndef SGEMM_BETA | ||||
| @@ -2945,7 +2945,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| #define SYMV_P 16 | #define SYMV_P 16 | ||||
| #endif | #endif | ||||
| #if defined(P5600) || defined(MIPS1004K) || defined(MIPS24K) || defined(I6400) || defined(P6600) || defined(I6500) | |||||
| #if defined(MIPS64_GENERIC) || defined(P5600) || defined(MIPS1004K) || defined(MIPS24K) || defined(I6400) || defined(P6600) || defined(I6500) | |||||
| #define SNUMOPT 2 | #define SNUMOPT 2 | ||||
| #define DNUMOPT 2 | #define DNUMOPT 2 | ||||