Merge pull request #34 from xianyi/develop

rebase

Author: martin-frbg

Makefile.system

@@ -558,6 +558,7 @@ DYNAMIC_CORE += FALKOR
 DYNAMIC_CORE += THUNDERX
 DYNAMIC_CORE += THUNDERX2T99
 DYNAMIC_CORE += TSV110
+DYNAMIC_CORE += EMAG8180
 endif
 
 ifeq ($(ARCH), zarch)

benchmark/copy.c

@@ -129,7 +129,10 @@ int main(int argc, char *argv[]){
   int step =   1;
 
   struct timeval start, stop;
-  double time1,timeg;
+  double time1 = 0.0, timeg = 0.0;
+  long nanos = 0;
+  time_t seconds = 0;
+  struct timespec time_start = { 0, 0 }, time_end = { 0, 0 };
 
   argc--;argv++;
 
@@ -163,35 +166,32 @@ int main(int argc, char *argv[]){
    timeg=0;
 
    fprintf(stderr, " %6d : ", (int)m);
+   for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+       x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   }
 
+   for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+       y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   }
 
    for (l=0; l<loops; l++)
    {
+       clock_gettime(CLOCK_REALTIME, &time_start);
+       COPY (&m, x, &inc_x, y, &inc_y );
+       clock_gettime(CLOCK_REALTIME, &time_end);
 
-   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
-			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-   	}
+       nanos = time_end.tv_nsec - time_start.tv_nsec;
+       seconds = time_end.tv_sec - time_start.tv_sec;
 
-   	for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
-			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-   	}
-    	gettimeofday( &start, (struct timezone *)0);
+       time1 = seconds + nanos / 1.e9;
+       timeg += time1;
+   }
 
-    	COPY (&m, x, &inc_x, y, &inc_y );
+      timeg /= loops;
 
-    	gettimeofday( &stop, (struct timezone *)0);
-
-    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
-
-	timeg += time1;
-
-    }
-
-    timeg /= loops;
-
-    fprintf(stderr,
-	    " %10.2f MBytes %10.6f sec\n",
-	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+      fprintf(stderr,
+	    " %10.2f MBytes %12.9f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg / 1.e6, timeg);
 
   }
 

cmake/arch.cmake

@@ -45,7 +45,7 @@ endif ()
 
 if (DYNAMIC_ARCH)
   if (ARM64)
-    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110)
+    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180)
   endif ()
 
   if (POWER)

cmake/prebuild.cmake

@@ -332,6 +332,29 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
     set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "EMAG8180")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define ARMV8\n"
+      "#define L1_CODE_SIZE\t32768\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t4\n"
+      "#define L1_DATA_SIZE\t32768\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t4\n"
+      "#define L2_SIZE\t5262144\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t8\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
   elseif ("${TCORE}" STREQUAL "POWER6")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE 32768\n"

cpuid_arm64.c

@@ -41,6 +41,8 @@
 #define CPU_THUNDERX2T99  8
 //Hisilicon
 #define CPU_TSV110        9
+// Ampere
+#define CPU_EMAG8180	 10
 
 static char *cpuname[] = {
   "UNKNOWN",
@@ -52,7 +54,8 @@ static char *cpuname[] = {
   "FALKOR",
   "THUNDERX",
   "THUNDERX2T99",
-  "TSV110"
+  "TSV110",
+  "EMAG8180"
 };
 
 static char *cpuname_lower[] = {
@@ -65,7 +68,8 @@ static char *cpuname_lower[] = {
   "falkor",
   "thunderx",
   "thunderx2t99",
-  "tsv110"
+  "tsv110",
+  "emag8180"
 };
 
 int get_feature(char *search)
@@ -152,6 +156,9 @@ int detect(void)
     // HiSilicon
     else if (strstr(cpu_implementer, "0x48") && strstr(cpu_part, "0xd01"))
                         return CPU_TSV110;
+    // Ampere
+    else if (strstr(cpu_implementer, "0x50") && strstr(cpu_part, "0x000"))
+                        return CPU_EMAG8180;
 	}
 
 	p = (char *) NULL ;
@@ -335,6 +342,18 @@ void get_cpuconfig(void)
 			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
 			printf("#define DTB_SIZE             4096     \n");
 			break;	
+
+		case CPU_EMAG8180:
+      // Minimum parameters for ARMv8 (based on A53)
+	printf("#define EMAG8180\n");
+    	printf("#define L1_CODE_SIZE 32768\n");
+    	printf("#define L1_DATA_SIZE 32768\n");
+    	printf("#define L1_DATA_LINESIZE 64\n");
+    	printf("#define L2_SIZE 262144\n");
+    	printf("#define L2_LINESIZE 64\n");
+    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    	printf("#define DTB_SIZE 4096\n");
+
 	}
 	get_cpucount();
 }

driver/others/dynamic_arm64.c

@@ -51,10 +51,11 @@ extern gotoblas_t  gotoblas_FALKOR;
 extern gotoblas_t  gotoblas_THUNDERX;
 extern gotoblas_t  gotoblas_THUNDERX2T99;
 extern gotoblas_t  gotoblas_TSV110;
+extern gotoblas_t  gotoblas_EMAG8180;
 
 extern void openblas_warning(int verbose, const char * msg);
 
-#define NUM_CORETYPES    9
+#define NUM_CORETYPES   10
 
 /*
  * In case asm/hwcap.h is outdated on the build system, make sure
@@ -78,6 +79,7 @@ static char *corename[] = {
   "thunderx",
   "thunderx2t99",
   "tsv110",
+  "emag8180",
   "unknown"
 };
 
@@ -91,6 +93,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_THUNDERX)     return corename[ 6];
   if (gotoblas == &gotoblas_THUNDERX2T99) return corename[ 7];
   if (gotoblas == &gotoblas_TSV110)       return corename[ 8];
+  if (gotoblas == &gotoblas_EMAG8180)     return corename[ 9];
   return corename[NUM_CORETYPES];
 }
 
@@ -119,6 +122,7 @@ static gotoblas_t *force_coretype(char *coretype) {
     case  6: return (&gotoblas_THUNDERX);
     case  7: return (&gotoblas_THUNDERX2T99);
     case  8: return (&gotoblas_TSV110);
+    case  9: return (&gotoblas_EMAG8180);
   }
   snprintf(message, 128, "Core not found: %s\n", coretype);
   openblas_warning(1, message);
@@ -189,6 +193,13 @@ static gotoblas_t *get_coretype(void) {
           return &gotoblas_TSV110;
       }
       break;
+    case 0x50: // Ampere
+      switch (part)
+      {
+        case 0x000: // Skylark/EMAG8180
+          return &gotoblas_EMAG8180;
+      }
+      break;
     case 0x51: // Qualcomm
       switch (part)
       {

getarch.c

@@ -1093,6 +1093,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_EMAG8180
+#define ARMV8
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "EMAG8180"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DEMAG8180 " \
+       "-DL1_CODE_SIZE=32768 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=8 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=8 " \
+       "-DL2_SIZE=262144 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \
+       "-DL3_SIZE=33554432 -DL3_LINESIZE=64 -DL3_ASSOCIATIVE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "emag8180"
+#define CORENAME  "EMAG8180"
+#endif
 
 #ifdef FORCE_ZARCH_GENERIC
 #define FORCE

kernel/arm64/KERNEL.EMAG8180

@@ -0,0 +1,3 @@
+include $(KERNELDIR)/KERNEL.CORTEXA57
+
+

kernel/mips64/KERNEL

@@ -167,3 +167,27 @@ endif
 
 CGEMM3MKERNEL    =  zgemm3m_kernel.S
 ZGEMM3MKERNEL    =  zgemm3m_kernel.S
+
+ifndef ISMINKERNEL
+ISMINKERNEL = imin.S
+endif
+
+ifndef IDMINKERNEL
+IDMINKERNEL = imin.S
+endif
+
+ifndef IQMINKERNEL
+IQMINKERNEL = imin.S
+endif
+
+ifndef ISMAXKERNEL
+ISMAXKERNEL = imax.S
+endif
+
+ifndef IDMAXKERNEL
+IDMAXKERNEL = imax.S
+endif
+
+ifndef IQMAXKERNEL
+IQMAXKERNEL = imax.S
+endif

kernel/power/KERNEL

@@ -50,3 +50,26 @@ ifndef DSDOTKERNEL
 DSDOTKERNEL = ../generic/dot.c
 endif
 
+ifndef ISMINKERNEL
+ISMINKERNEL = imin.S
+endif
+
+ifndef IDMINKERNEL
+IDMINKERNEL = imin.S
+endif
+
+ifndef IQMINKERNEL
+IQMINKERNEL = imin.S
+endif
+
+ifndef ISMAXKERNEL
+ISMAXKERNEL = imax.S
+endif
+
+ifndef IDMAXKERNEL
+IDMAXKERNEL = imax.S
+endif
+
+ifndef IQMAXKERNEL
+IQMAXKERNEL = imax.S
+endif