add in runtime cpu detection for zarch (#2349)

add in runtime cpu detection for zarch

Makefile.system

@@ -25,6 +25,8 @@ else ifeq ($(ARCH), i386)
 override ARCH=x86
 else ifeq ($(ARCH), aarch64)
 override ARCH=arm64
+else ifeq ($(ARCH), zarch)
+override ARCH=zarch
 endif
 
 NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-netlib
@@ -558,6 +560,11 @@ DYNAMIC_CORE += THUNDERX2T99
 DYNAMIC_CORE += TSV110
 endif
 
+ifeq ($(ARCH), zarch)
+DYNAMIC_CORE = Z13
+DYNAMIC_CORE += Z14
+endif
+
 ifeq ($(ARCH), power)
 DYNAMIC_CORE = POWER6
 DYNAMIC_CORE += POWER8

driver/others/Makefile

@@ -21,9 +21,13 @@ else
 ifeq ($(ARCH),power)
 COMMONOBJS	+=  dynamic_power.$(SUFFIX)
 else
+ifeq ($(ARCH),zarch)
+COMMONOBJS += dynamic_zarch.$(SUFFIX)
+else
 COMMONOBJS	+=  dynamic.$(SUFFIX)
 endif
 endif
+endif
 else
 COMMONOBJS	+=  parameter.$(SUFFIX)
 endif
@@ -85,9 +89,13 @@ else
 ifeq ($(ARCH),power)
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic_power.$(SUFFIX)
 else
+ifeq ($(ARCH),zarch)
+HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic_zarch.$(SUFFIX)
+else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic.$(SUFFIX)
 endif
 endif
+endif
 else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) parameter.$(SUFFIX)
 endif

driver/others/dynamic_zarch.c

@@ -0,0 +1,131 @@
+
+#include "common.h"
+
+extern gotoblas_t gotoblas_Z13;
+extern gotoblas_t gotoblas_Z14;
+extern gotoblas_t gotoblas_Z15;
+//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+//extern gotoblas_t gotoblas_Z14;
+//#endif
+
+#define NUM_CORETYPES 5
+
+extern void openblas_warning(int verbose, const char* msg);
+
+static char* corename[] = {
+	"unknown",
+	"Z13",
+	"Z14",
+	"Z15",
+	"ZARCH_GENERIC",
+};
+
+char* gotoblas_corename(void) {
+	if (gotoblas == &gotoblas_Z13)	return corename[1];
+	if (gotoblas == &gotoblas_Z14)	return corename[2];
+	if (gotoblas == &gotoblas_Z15)	return corename[3];
+//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+//	if (gotoblas == &gotoblas_POWER9)	return corename[3];
+//#endif
+	return corename[0]; // try generic?
+}
+
+// __builtin_cpu_is is not supported by zarch
+static gotolabs_t* get_coretype(void) {
+	FILE* infile;
+	char buffer[512], * p;
+
+	p = (char*)NULL;
+	infile = fopen("/proc/sysinfo", "r");
+	while (fgets(buffer, sizeof(buffer), infile)) {
+		if (!strncmp("Type", buffer, 4)) {
+			p = strchr(buffer, ':') + 2;
+#if 0
+			fprintf(stderr, "%s\n", p);
+#endif
+			break;
+		}
+	}
+
+	fclose(infile);
+
+	if (strstr(p, "2964")) return &gotoblas_Z13;
+	if (strstr(p, "2965")) return &gotoblas_Z13;
+	if (strstr(p, "3906")) return &gotoblas_Z14;
+	if (strstr(p, "3907")) return &gotoblas_Z14;
+	if (strstr(p, "8561")) return &gotoblas_Z14;        // fallback z15 to z14
+	if (strstr(p, "8562")) return &gotoblas_Z14;        // fallback z15 to z14
+
+	return NULL; // should be ZARCH_GENERIC
+}
+
+static gotoblas_t* force_coretype(char* coretype) {
+
+	int i;
+	int found = -1;
+	char message[128];
+
+	for (i = 0; i < NUM_CORETYPES; i++)
+	{
+		if (!strncasecmp(coretype, corename[i], 20))
+		{
+			found = i;
+			break;
+		}
+	}
+
+	switch (found)
+	{
+	case  1: return (&gotoblas_Z13);
+	case  2: return (&gotoblas_Z14);
+	case  3: return (&gotoblas_Z15);
+//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+//	case  3: return (&gotoblas_POWER9);
+//#endif
+	default: return NULL;
+	}
+	snprintf(message, 128, "Core not found: %s\n", coretype);
+	openblas_warning(1, message);
+}
+
+void gotoblas_dynamic_init(void) {
+
+	char coremsg[128];
+	char coren[22];
+	char* p;
+
+
+	if (gotoblas) return;
+
+	p = getenv("OPENBLAS_CORETYPE");
+	if (p)
+	{
+		gotoblas = force_coretype(p);
+	}
+	else
+	{
+		gotoblas = get_coretype();
+	}
+
+	if (gotoblas == NULL)
+	{
+		snprintf(coremsg, 128, "Falling back to Z14 core\n");
+		openblas_warning(1, coremsg);
+		gotoblas = &gotoblas_Z14;
+	}
+
+	if (gotoblas && gotoblas->init) {
+		strncpy(coren, gotoblas_corename(), 20);
+		sprintf(coremsg, "Core: %s\n", coren);
+		openblas_warning(2, coremsg);
+		gotoblas->init();
+	}
+	else {
+		openblas_warning(0, "OpenBLAS : Architecture Initialization failed. No initialization function found.\n");
+		exit(1);
+	}
+}
+
+void gotoblas_dynamic_quit(void) {
+	gotoblas = NULL;
+}

kernel/setparam-ref.c

@@ -739,6 +739,26 @@ static void init_parameter(void) {
 }
 #else //POWER
 
+#if defined(ARCH_ZARCH)
+static void init_parameter(void) {
+	TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
+	TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
+	TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
+	TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
+
+	TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
+	TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
+	TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
+	TABLE_NAME.zgemm_r = ZGEMM_DEFAULT_R;
+
+
+	TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
+	TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
+	TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
+	TABLE_NAME.zgemm_q = ZGEMM_DEFAULT_Q;
+}
+#else //ZARCH
+
 #ifdef ARCH_X86
 static int get_l2_size_old(void){
   int i, eax, ebx, ecx, edx, cpuid_level;
@@ -1325,4 +1345,5 @@ static void init_parameter(void) {
 
 }
 #endif //POWER
+#endif //ZARCH
 #endif //defined(ARCH_ARM64)

kernel/zarch/KERNEL.Z13

@@ -96,10 +96,10 @@ SGEMMINCOPY    = ../generic/gemm_ncopy_8.c
 SGEMMITCOPY    = ../generic/gemm_tcopy_8.c
 SGEMMONCOPY    = ../generic/gemm_ncopy_4.c
 SGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
-SGEMMINCOPYOBJ = sgemm_incopy.o
+SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
-SGEMMITCOPYOBJ = sgemm_itcopy.o
+SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
-SGEMMONCOPYOBJ = sgemm_oncopy.o
+SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)
-SGEMMOTCOPYOBJ = sgemm_otcopy.o
+SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 
  
@@ -108,16 +108,16 @@ DGEMMINCOPY    = ../generic/gemm_ncopy_8.c
 DGEMMITCOPY    = ../generic/gemm_tcopy_8.c
 DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
 DGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
-DGEMMINCOPYOBJ = dgemm_incopy.o
+DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
-DGEMMITCOPYOBJ = dgemm_itcopy.o
+DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
-DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
-DGEMMOTCOPYOBJ = dgemm_otcopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 CGEMMKERNEL    = ctrmm4x4V.S
 CGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
 CGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
-CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
-CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 ZGEMMKERNEL    = ztrmm4x4V.S
 ZGEMMONCOPY    = ../generic/zgemm_ncopy_4.c

kernel/zarch/KERNEL.Z14

@@ -96,10 +96,10 @@ SGEMMINCOPY    = ../generic/gemm_ncopy_8.c
 SGEMMITCOPY    = ../generic/gemm_tcopy_8.c
 SGEMMONCOPY    = ../generic/gemm_ncopy_4.c
 SGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
-SGEMMINCOPYOBJ = sgemm_incopy.o
+SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
-SGEMMITCOPYOBJ = sgemm_itcopy.o
+SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
-SGEMMONCOPYOBJ = sgemm_oncopy.o
+SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)
-SGEMMOTCOPYOBJ = sgemm_otcopy.o
+SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 
  
@@ -108,16 +108,16 @@ DGEMMINCOPY    = ../generic/gemm_ncopy_8.c
 DGEMMITCOPY    = ../generic/gemm_tcopy_8.c
 DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
 DGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
-DGEMMINCOPYOBJ = dgemm_incopy.o
+DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
-DGEMMITCOPYOBJ = dgemm_itcopy.o
+DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
-DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
-DGEMMOTCOPYOBJ = dgemm_otcopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 CGEMMKERNEL    = ctrmm4x4V.S
 CGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
 CGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
-CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
-CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 ZGEMMKERNEL    = ztrmm4x4V.S
 ZGEMMONCOPY    = ../generic/zgemm_ncopy_4.c

kernel/zarch/KERNEL.ZARCH_GENERIC

@@ -94,26 +94,26 @@ ZTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
 SGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
 SGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
 SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
-SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
-SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 DGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
 DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
 DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
-DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
-DGEMMOTCOPYOBJ = dgemm_otcopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
 CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
 CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
-CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
-CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
 ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
 ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
-ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
-ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
 STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c