Add ZEN support

Makefile.system

@@ -68,6 +68,9 @@ endif
 ifeq ($(TARGET), EXCAVATOR)
 GETARCH_FLAGS := -DFORCE_BARCELONA
 endif
+ifeq ($(TARGET), ZEN)
+GETARCH_FLAGS := -DFORCE_BARCELONA
+endif
 endif
 
 
@@ -98,6 +101,9 @@ endif
 ifeq ($(TARGET_CORE), EXCAVATOR)
 GETARCH_FLAGS := -DFORCE_BARCELONA
 endif
+ifeq ($(TARGET_CORE), ZEN)
+GETARCH_FLAGS := -DFORCE_BARCELONA
+endif
 endif
 
 
@@ -443,7 +449,7 @@ ifneq ($(NO_AVX), 1)
 DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR
 endif
 ifneq ($(NO_AVX2), 1)
-DYNAMIC_CORE += HASWELL
+DYNAMIC_CORE += HASWELL ZEN
 endif
 endif
 

TargetList.txt

@@ -34,6 +34,7 @@ BULLDOZER
 PILEDRIVER
 STEAMROLLER
 EXCAVATOR
+ZEN
 
 c)VIA CPU:
 SSE_GENERIC

cmake/arch.cmake

@@ -73,7 +73,7 @@ if (DYNAMIC_ARCH)
       set(DYNAMIC_CORE "${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER")
     endif ()
     if (NOT NO_AVX2)
-      set(DYNAMIC_CORE "${DYNAMIC_CORE} HASWELL")
+      set(DYNAMIC_CORE "${DYNAMIC_CORE} HASWELL ZEN")
     endif ()
   endif ()
 

cmake/system.cmake

@@ -22,7 +22,7 @@ if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
   if (${TARGET} STREQUAL "HASWELL" OR ${TARGET} STREQUAL "SANDYBRIDGE")
     set(TARGET "NEHALEM")
   endif ()
-  if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER")
+  if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "ZEN")
     set(TARGET "BARCELONA")
   endif ()
 endif ()

cpuid.h

@@ -114,6 +114,7 @@
 #define CORE_HASWELL     24
 #define CORE_STEAMROLLER 25
 #define CORE_EXCAVATOR   26
+#define CORE_ZEN         27
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -209,5 +210,6 @@ typedef struct {
 #define CPUTYPE_HASWELL 		48
 #define CPUTYPE_STEAMROLLER 		49
 #define CPUTYPE_EXCAVATOR 		50
+#define CPUTYPE_ZEN 			51
 
 #endif

cpuid_x86.c

@@ -1281,6 +1281,8 @@ int get_cpuname(void){
       case  3:
       case 10:
 	return CPUTYPE_BARCELONA;
+      case  5:
+	return CPUTYPE_BOBCAT;
       case  6:
 	switch (model) {
 	case 1:
@@ -1295,8 +1297,8 @@ int get_cpuname(void){
 	    return CPUTYPE_PILEDRIVER;
 	  else
 	    return CPUTYPE_BARCELONA; //OS don't support AVX.
-    case 5: // New EXCAVATOR CPUS
-      if(support_avx())
+	case 5: // New EXCAVATOR CPUS
+	  if(support_avx())
 	    return CPUTYPE_EXCAVATOR;
 	  else
 	    return CPUTYPE_BARCELONA; //OS don't support AVX.
@@ -1322,8 +1324,19 @@ int get_cpuname(void){
 	  break;
 	}
 	break;
-      case  5:
-	return CPUTYPE_BOBCAT;
+      case 8:
+	switch (model) {
+	case 1:
+	  // AMD Ryzen
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CPUTYPE_ZEN;
+#else
+	    return CPUTYPE_SANDYBRIDGE; // Zen is closer in architecture to Sandy Bridge than to Excavator
+#endif
+	  else
+	    return CPUTYPE_BARCELONA;
+        }
       }
       break;
     }
@@ -1450,6 +1463,7 @@ static char *cpuname[] = {
   "HASWELL",
   "STEAMROLLER",
   "EXCAVATOR",
+  "ZEN",
 };
 
 static char *lowercpuname[] = {
@@ -1503,6 +1517,7 @@ static char *lowercpuname[] = {
   "haswell",
   "steamroller",
   "excavator",
+  "zen",
 };
 
 static char *corename[] = {
@@ -1533,6 +1548,7 @@ static char *corename[] = {
   "HASWELL",
   "STEAMROLLER",
   "EXCAVATOR",
+  "ZEN",
 };
 
 static char *corename_lower[] = {
@@ -1563,6 +1579,7 @@ static char *corename_lower[] = {
   "haswell",
   "steamroller",
   "excavator",
+  "zen",
 };
 
 
@@ -1776,15 +1793,16 @@ int get_coretype(void){
 	break;
       case 9:
       case 8:
-        if (model == 14) // Kaby Lake 
+        if (model == 14) { // Kaby Lake 
 	  if(support_avx())
 #ifndef NO_AVX2
-            return CORE_HASWELL;
+	    return CORE_HASWELL;
 #else
-            return CORE_SANDYBRIDGE;
+	    return CORE_SANDYBRIDGE;
 #endif
 	  else
             return CORE_NEHALEM;
+	}
       }
       break;
 
@@ -1841,9 +1859,22 @@ int get_coretype(void){
 	  }
 	  break;
 	}
-
-
-      }else return CORE_BARCELONA;
+      } else if (exfamily == 8) {
+	switch (model) {
+	case 1:
+	  // AMD Ryzen
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_ZEN;
+#else
+	    return CORE_SANDYBRIDGE; // Zen is closer in architecture to Sandy Bridge than to Excavator
+#endif
+	  else
+	    return CORE_BARCELONA;
+	}
+      } else {
+	return CORE_BARCELONA;
+      }
     }
   }
 

driver/others/dynamic.c

@@ -70,8 +70,10 @@ extern gotoblas_t  gotoblas_STEAMROLLER;
 extern gotoblas_t  gotoblas_EXCAVATOR;
 #ifdef NO_AVX2
 #define gotoblas_HASWELL gotoblas_SANDYBRIDGE
+#define gotoblas_ZEN gotoblas_SANDYBRIDGE
 #else
 extern gotoblas_t  gotoblas_HASWELL;
+extern gotoblas_t  gotoblas_ZEN;
 #endif
 #else
 //Use NEHALEM kernels for sandy bridge
@@ -81,6 +83,7 @@ extern gotoblas_t  gotoblas_HASWELL;
 #define gotoblas_PILEDRIVER gotoblas_BARCELONA
 #define gotoblas_STEAMROLLER gotoblas_BARCELONA
 #define gotoblas_EXCAVATOR gotoblas_BARCELONA
+#define gotoblas_ZEN gotoblas_BARCELONA
 #endif
 
 
@@ -355,14 +358,14 @@ static gotoblas_t *get_coretype(void){
 	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
 	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
 	  }
-    }else if(model == 5){
-      if(support_avx())
-        return &gotoblas_EXCAVATOR;
-      else{
-        openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
-        return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
-      }
-    }else if(model == 0){
+	}else if(model == 5){
+	  if(support_avx())
+	    return &gotoblas_EXCAVATOR;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	  }
+	}else if(model == 0){
 	  if (exmodel == 1) {
 	    //AMD Trinity
 	    if(support_avx())
@@ -389,9 +392,16 @@ static gotoblas_t *get_coretype(void){
 
 	  }
 	}
-
-
-      } else {
+      } else if (family == 8) {
+	if (model == 1) {
+	  if(support_avx())
+	    return &gotoblas_ZEN;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+      }else {
 	return &gotoblas_BARCELONA;
       }
     }
@@ -431,6 +441,7 @@ static char *corename[] = {
     "Haswell",
     "Steamroller",
     "Excavator",
+    "Zen"
 };
 
 char *gotoblas_corename(void) {
@@ -457,6 +468,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_HASWELL)      return corename[20];
   if (gotoblas == &gotoblas_STEAMROLLER)  return corename[21];
   if (gotoblas == &gotoblas_EXCAVATOR)    return corename[22];
+  if (gotoblas == &gotoblas_ZEN)          return corename[23];
 
   return corename[0];
 }
@@ -487,6 +499,7 @@ static gotoblas_t *force_coretype(char *coretype){
 
 	switch (found)
 	{
+		case 23: return (&gotoblas_ZEN);
 		case 22: return (&gotoblas_EXCAVATOR);
 		case 21: return (&gotoblas_STEAMROLLER);
 		case 20: return (&gotoblas_HASWELL);

driver/others/parameter.c

@@ -167,7 +167,7 @@ int get_L2_size(void){
 #if defined(ATHLON) || defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT) || defined(BULLDOZER) || \
     defined(CORE_PRESCOTT) || defined(CORE_CORE2) || defined(PENRYN) || defined(DUNNINGTON) || \
     defined(CORE_NEHALEM) || defined(CORE_SANDYBRIDGE) || defined(ATOM) || defined(GENERIC) || \
-    defined(PILEDRIVER) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR)
+    defined(PILEDRIVER) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined(ZEN)
 
   cpuid(0x80000006, &eax, &ebx, &ecx, &edx);
 
@@ -251,7 +251,7 @@ int get_L2_size(void){
 void blas_set_parameter(void){
 
   int factor;
-#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(SANDYBRIDGE) || defined(NEHALEM) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(SANDYBRIDGE) || defined(NEHALEM) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined(ZEN)
   int size = 16;
 #else
   int size = get_L2_size();

getarch.c

@@ -473,6 +473,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "EXCAVATOR"
 #endif
 
+#if defined (FORCE_ZEN)
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "ZEN"
+#define ARCHCONFIG   "-DZEN " \
+		     "-DL1_CODE_SIZE=32768 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=8 " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL2_CODE_ASSOCIATIVE=8 " \
+		     "-DL2_SIZE=524288 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \
+		     "-DL3_SIZE=16777216 -DL3_LINESIZE=64 -DL3_ASSOCIATIVE=8 " \
+		     "-DITB_DEFAULT_ENTRIES=64 -DITB_SIZE=4096 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 " \
+		     "-DHAVE_SSE4A -DHAVE_MISALIGNSSE -DHAVE_128BITFPU -DHAVE_FASTMOVU -DHAVE_CFLUSH " \
+		     "-DHAVE_AVX -DHAVE_FMA3 -DFMA3"
+#define LIBNAME   "excavator"
+#define CORENAME  "EXCAVATOR"
+#endif
+
 
 #ifdef FORCE_SSE_GENERIC
 #define FORCE

kernel/CMakeLists.txt

@@ -118,7 +118,7 @@ endforeach ()
 # Makefile.L3
 set(USE_TRMM false)
 
-if (${ARCH} STREQUAL "arm" OR ${ARCH} STREQUAL "arm64" OR "${TARGET}" STREQUAL "LONGSOON3B" OR "${TARGET}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell")
+if (${ARCH} STREQUAL "arm" OR ${ARCH} STREQUAL "arm64" OR "${TARGET}" STREQUAL "LONGSOON3B" OR "${TARGET}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell" OR "{CORE}" STREQUAL "zen")
   set(USE_TRMM true)
 endif ()
 

kernel/Makefile.L3

@@ -32,6 +32,10 @@ ifeq ($(CORE), HASWELL)
 USE_TRMM = 1
 endif
 
+ifeq ($(CORE), ZEN)
+USE_TRMM = 1
+endif
+
 ifeq ($(CORE), POWER8)
 USE_TRMM = 1
 endif

kernel/setparam-ref.c

@@ -982,6 +982,22 @@ static void init_parameter(void) {
 #endif
 #endif
 
+#ifdef ZEN
+
+#ifdef DEBUG
+  fprintf(stderr, "Zen\n");
+#endif
+
+  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;
+#ifdef EXPRECISION
+  TABLE_NAME.qgemm_p = QGEMM_DEFAULT_P;
+  TABLE_NAME.xgemm_p = XGEMM_DEFAULT_P;
+#endif
+#endif
+
 
 #ifdef NANO
 

kernel/x86/KERNEL.ZEN

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