Merge branch 'develop' of github.com:xianyi/OpenBLAS into arm_soft_fp_abi

Author: xianyi

.gitignore

@@ -68,3 +68,4 @@ test/zblat2
 test/zblat3
 build
 build.*
+*.swp

CMakeLists.txt

@@ -6,7 +6,7 @@ cmake_minimum_required(VERSION 2.8.4)
 project(OpenBLAS)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 2)
-set(OpenBLAS_PATCH_VERSION 14)
+set(OpenBLAS_PATCH_VERSION 16.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 enable_language(ASM)

Makefile.arm64

@@ -4,4 +4,8 @@ CCOMMON_OPT += -march=armv8-a
 FCOMMON_OPT += -march=armv8-a
 endif
 
+ifeq ($(CORE), CORTEXA57)
+CCOMMON_OPT += -march=armv8-a+crc+crypto+fp+simd -mtune=cortex-a57
+FCOMMON_OPT += -march=armv8-a+crc+crypto+fp+simd -mtune=cortex-a57
+endif
 

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.2.15
+VERSION = 0.2.16.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library

TargetList.txt

@@ -74,3 +74,5 @@ ARMV5
 
 7.ARM 64-bit CPU:
 ARMV8
+CORTEXA57
+

benchmark/gemm.c

@@ -172,7 +172,7 @@ int main(int argc, char *argv[]){
   srandom(getpid());
 #endif
 
-	for(j = 0; j < m; j++){
+	for(j = 0; j < to; j++){
       		for(i = 0; i < to * COMPSIZE; i++){
 			a[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
 			b[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;

c_check

@@ -6,6 +6,7 @@ $hostarch = `uname -m | sed -e s/i.86/x86/`;chop($hostarch);
 $hostarch = "x86_64" if ($hostarch eq "amd64");
 $hostarch = "arm" if ($hostarch =~ /^arm.*/);
 $hostarch = "arm64" if ($hostarch eq "aarch64");
+$hostarch = "power" if ($hostarch =~ /^(powerpc|ppc).*/);
 
 $binary = $ENV{"BINARY"};
 

common.h

@@ -86,6 +86,7 @@ extern "C" {
 #if !defined(_MSC_VER)
 #include <unistd.h>
 #endif
+#include <time.h>
 
 #ifdef OS_LINUX
 #include <malloc.h>

common_arm64.h

@@ -89,8 +89,10 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #if defined(ASSEMBLER) && !defined(NEEDPARAM)
 
 #define PROLOGUE \
+	.text ;\
+	.align	4 ;\
 	.global	REALNAME ;\
-	.func	REALNAME  ;\
+	.type	REALNAME, %function ;\
 REALNAME:
 
 #define EPILOGUE
@@ -107,7 +109,11 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
+#if defined(CORTEXA57)
+#define BUFFER_SIZE     (40 << 20)
+#else
 #define BUFFER_SIZE     (16 << 20)
+#endif
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)

cpuid_arm64.c

@@ -29,12 +29,19 @@
 
 #define CPU_UNKNOWN     	0
 #define CPU_ARMV8       	1
+#define CPU_CORTEXA57       	2
 
 static char *cpuname[] = {
-  "UNKOWN",
-  "ARMV8"
+  "UNKNOWN",
+  "ARMV8" ,
+  "CORTEXA57"
 };
 
+static char *cpuname_lower[] = {
+  "unknown",
+  "armv8" ,
+  "cortexa57"
+};
 
 int get_feature(char *search)
 {
@@ -53,13 +60,13 @@ int get_feature(char *search)
 		{
 			p = strchr(buffer, ':') + 2;
 			break;
-      		}
-  	}
+		}
+	}
 
-  	fclose(infile);
+	fclose(infile);
 
 
-	if( p == NULL ) return;
+	if( p == NULL ) return 0;
 
 	t = strtok(p," ");
 	while( t = strtok(NULL," "))
@@ -82,11 +89,30 @@ int detect(void)
   	p = (char *) NULL ;
 
   	infile = fopen("/proc/cpuinfo", "r");
+	while (fgets(buffer, sizeof(buffer), infile))
+	{
 
+		if (!strncmp("CPU part", buffer, 8))
+		{
+			p = strchr(buffer, ':') + 2;
+			break;
+		}
+	}
+
+	fclose(infile);
+	if(p != NULL) {
+	  if (strstr(p, "0xd07")) {
+	    return CPU_CORTEXA57;
+	  }
+	}
+
+	p = (char *) NULL ;
+	infile = fopen("/proc/cpuinfo", "r");
 	while (fgets(buffer, sizeof(buffer), infile))
 	{
 
-		if ((!strncmp("model name", buffer, 10)) || (!strncmp("Processor", buffer, 9)))
+		if ((!strncmp("model name", buffer, 10)) || (!strncmp("Processor", buffer, 9)) ||
+		    (!strncmp("CPU architecture", buffer, 16)))
 		{
 			p = strchr(buffer, ':') + 2;
 			break;
@@ -100,7 +126,7 @@ int detect(void)
 
 		if (strstr(p, "AArch64"))
 		{
-			 	return CPU_ARMV8;
+			return CPU_ARMV8;
 
 		}
 
@@ -118,23 +144,13 @@ char *get_corename(void)
 
 void get_architecture(void)
 {
-	printf("ARM");
+	printf("ARM64");
 }
 
 void get_subarchitecture(void)
 {
 	int d = detect();
-	switch (d)
-	{
-
-		case CPU_ARMV8:
-			printf("ARMV8");
-			break;
-
-		default:
-			printf("UNKNOWN");
-			break;
-	}
+	printf("%s", cpuname[d]);
 }
 
 void get_subdirname(void)
@@ -160,26 +176,32 @@ void get_cpuconfig(void)
     			printf("#define L2_ASSOCIATIVE 4\n");
 			break;
 
-
+		case CPU_CORTEXA57:
+			printf("#define CORTEXA57\n");
+			printf("#define HAVE_VFP\n");
+			printf("#define HAVE_VFPV3\n");
+			printf("#define HAVE_NEON\n");
+			printf("#define HAVE_VFPV4\n");
+			printf("#define L1_CODE_SIZE 49152\n");
+			printf("#define L1_CODE_LINESIZE 64\n");
+			printf("#define L1_CODE_ASSOCIATIVE 3\n");
+			printf("#define L1_DATA_SIZE 32768\n");
+			printf("#define L1_DATA_LINESIZE 64\n");
+			printf("#define L1_DATA_ASSOCIATIVE 2\n");
+			printf("#define L2_SIZE 2097152\n");
+			printf("#define L2_LINESIZE 64\n");
+			printf("#define L2_ASSOCIATIVE 16\n");
+			break;
 	}
 }
 
 
 void get_libname(void)
 {
-
 	int d = detect();
-	switch (d)
-	{
-
-		case CPU_ARMV8:
-    			printf("armv8\n");
-			break;
-
-	}
+	printf("%s", cpuname_lower[d]);
 }
 
-
 void get_features(void)
 {