Merge pull request #74 from xianyi/develop

rebase

Makefile.arm64

@@ -56,6 +56,16 @@ CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
 
+ifeq ($(CORE), THUNDERX3T110)
+ifeq ($(GCCVERSIONGTEQ10), 1)
+CCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+FCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+else
+CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
+FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
+endif
+endif
+
 ifeq ($(GCCVERSIONGTEQ9), 1)
 ifeq ($(CORE), TSV110)
 CCOMMON_OPT += -march=armv8.2-a -mtune=tsv110

Makefile.power

@@ -11,34 +11,34 @@ endif
 
 ifeq ($(CORE), POWER10)
 ifeq ($(USE_OPENMP), 1)
-COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
-FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10 -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx  -DUSE_OPENMP -fno-fast-math -fopenmp
+FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -DUSE_OPENMP -fno-fast-math -fopenmp
 else
-COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -malign-power -fno-fast-math
-FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10 -malign-power -fno-fast-math
+COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx  -fno-fast-math
+FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -fno-fast-math
 endif
 endif
 
 ifeq ($(CORE), POWER9)
 ifeq ($(USE_OPENMP), 1)
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx  -DUSE_OPENMP -fno-fast-math -fopenmp
 else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp
 endif
 ifneq ($(F_COMPILER), PGI)
-FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9 -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9  -DUSE_OPENMP -fno-fast-math -fopenmp
 else
 FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp
 endif
 else
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -malign-power -fno-fast-math
+CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx  -fno-fast-math
 else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
-FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9 -malign-power -fno-fast-math
+FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9  -fno-fast-math
 else
 FCOMMON_OPT += -O2 -Mrecursive
 endif
@@ -48,26 +48,26 @@ endif
 ifeq ($(CORE), POWER8)
 ifeq ($(USE_OPENMP), 1)
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx  -DUSE_OPENMP -fno-fast-math -fopenmp
 else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp
 endif
 ifneq ($(F_COMPILER), PGI)
-FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8  -DUSE_OPENMP -fno-fast-math -fopenmp
 else
 FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp
 endif
 else
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -malign-power -fno-fast-math
+CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx  -fno-fast-math
 else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
 ifeq ($(OSNAME), AIX)
-FCOMMON_OPT += -O1 -frecursive -mcpu=power8 -mtune=power8 -malign-power -fno-fast-math 
+FCOMMON_OPT += -O1 -frecursive -mcpu=power8 -mtune=power8  -fno-fast-math 
 else
-FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -malign-power -fno-fast-math 
+FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8  -fno-fast-math 
 endif
 else
 FCOMMON_OPT += -O2 -Mrecursive

Makefile.system

@@ -578,6 +578,7 @@ DYNAMIC_CORE += THUNDERX
 DYNAMIC_CORE += THUNDERX2T99
 DYNAMIC_CORE += TSV110
 DYNAMIC_CORE += EMAG8180
+DYNAMIC_CORE += THUNDERX3T110
 endif
 
 ifeq ($(ARCH), zarch)
@@ -617,7 +618,6 @@ DYNAMIC_CORE += POWER8
 ifneq ($(C_COMPILER), GCC)
 DYNAMIC_CORE += POWER9
 DYNAMIC_CORE += POWER10
-override LDFLAGS += -Wl,-no-power10-stubs
 endif
 ifeq ($(C_COMPILER), GCC)
 ifeq ($(GCCVERSIONGT5), 1)
@@ -627,11 +627,9 @@ $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
 endif
 ifeq ($(GCCVERSIONGTEQ11), 1)
 DYNAMIC_CORE += POWER10
-override LDFLAGS += -Wl,-no-power10-stubs
 else ifeq ($(GCCVERSIONGTEQ10), 1)
 ifeq ($(GCCMINORVERSIONGTEQ2), 1)
 DYNAMIC_CORE += POWER10
-override LDFLAGS += -Wl,-no-power10-stubs
 endif
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)
@@ -1241,7 +1239,9 @@ KERNELDIR	= $(TOPDIR)/kernel/$(ARCH)
 
 include $(TOPDIR)/Makefile.$(ARCH)
 
+ifneq ($(C_COMPILER), PGI)
 CCOMMON_OPT     += -UASMNAME -UASMFNAME -UNAME -UCNAME -UCHAR_NAME -UCHAR_CNAME
+endif
 CCOMMON_OPT	+= -DASMNAME=$(FU)$(*F) -DASMFNAME=$(FU)$(*F)$(BU) -DNAME=$(*F)$(BU) -DCNAME=$(*F) -DCHAR_NAME=\"$(*F)$(BU)\" -DCHAR_CNAME=\"$(*F)\"
 
 ifeq ($(CORE), PPC440)

README.md

@@ -28,7 +28,8 @@ You can download them from [file hosting on sourceforge.net](https://sourceforge
 ## Installation from Source
 
 Download from project homepage, https://xianyi.github.com/OpenBLAS/, or check out the code
-using Git from https://github.com/xianyi/OpenBLAS.git.
+using Git from https://github.com/xianyi/OpenBLAS.git. (If you want the most up to date version, be
+sure to use the develop branch - master is several years out of date due to a change of maintainership.)
 Buildtime parameters can be chosen in Makefile.rule, see there for a short description of each option.
 Most can also be given directly on the make or cmake command line.
 

TargetList.txt

@@ -96,6 +96,7 @@ FALKOR
 THUNDERX
 THUNDERX2T99
 TSV110
+THUNDERX3T110
 
 9.System Z:
 ZARCH_GENERIC

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 EMAG8180 NEOVERSEN1)
+    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
   endif ()
   
   if (POWER)

cmake/prebuild.cmake

@@ -195,8 +195,13 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
       "#define HAVE_VFP\n"
       "#define HAVE_NEON\n"
       "#define ARMV8\n")
+if ("${TCORE}" STREQUAL "CORTEXA57")     
     set(SGEMM_UNROLL_M 16)
     set(SGEMM_UNROLL_N 4)
+else ()
+    set(SGEMM_UNROLL_M 8)
+    set(SGEMM_UNROLL_N 8) 
+endif ()
     set(DGEMM_UNROLL_M 8)
     set(DGEMM_UNROLL_N 4)
     set(CGEMM_UNROLL_M 8)
@@ -338,6 +343,33 @@ 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 "THUNDERX3T110")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define THUNDERX3T110\n"
+      "#define L1_CODE_SIZE\t65536\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t8\n"
+      "#define L1_DATA_SIZE\t65536\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t8\n"
+      "#define L2_SIZE\t524288\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t8\n"
+      "#define L3_SIZE\t94371840\n"
+      "#define L3_LINESIZE\t64\n"
+      "#define L3_ASSOCIATIVE\t32\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
   elseif ("${TCORE}" STREQUAL "TSV110")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define ARMV8\n"

cpuid_arm64.c

@@ -40,6 +40,7 @@
 // Cavium
 #define CPU_THUNDERX      7
 #define CPU_THUNDERX2T99  8
+#define CPU_THUNDERX3T110 12
 //Hisilicon
 #define CPU_TSV110        9
 // Ampere
@@ -57,7 +58,8 @@ static char *cpuname[] = {
   "THUNDERX2T99",
   "TSV110",
   "EMAG8180",
-  "NEOVERSEN1"
+  "NEOVERSEN1",
+  "THUNDERX3T110"
 };
 
 static char *cpuname_lower[] = {
@@ -72,7 +74,8 @@ static char *cpuname_lower[] = {
   "thunderx2t99",
   "tsv110",
   "emag8180",
-  "neoversen1"
+  "neoversen1",
+  "thunderx3t110"
 };
 
 int get_feature(char *search)
@@ -158,6 +161,8 @@ int detect(void)
 			return CPU_THUNDERX;
     else if (strstr(cpu_implementer, "0x43") && strstr(cpu_part, "0x0af"))
 			return CPU_THUNDERX2T99;
+    else if (strstr(cpu_implementer, "0x43") && strstr(cpu_part, "0x0b8"))
+			return CPU_THUNDERX3T110;
     // HiSilicon
     else if (strstr(cpu_implementer, "0x48") && strstr(cpu_part, "0xd01"))
                         return CPU_TSV110;
@@ -372,7 +377,25 @@ void get_cpuconfig(void)
     	printf("#define L2_LINESIZE 64\n");
     	printf("#define DTB_DEFAULT_ENTRIES 64\n");
     	printf("#define DTB_SIZE 4096\n");
+			break;
 
+		case CPU_THUNDERX3T110:
+			printf("#define THUNDERX3T110                 \n");
+			printf("#define L1_CODE_SIZE         65536    \n");
+			printf("#define L1_CODE_LINESIZE     64       \n");
+			printf("#define L1_CODE_ASSOCIATIVE  8        \n");
+			printf("#define L1_DATA_SIZE         32768    \n");
+			printf("#define L1_DATA_LINESIZE     64       \n");
+			printf("#define L1_DATA_ASSOCIATIVE  8        \n");
+			printf("#define L2_SIZE              524288   \n");
+			printf("#define L2_LINESIZE          64       \n");
+			printf("#define L2_ASSOCIATIVE       8        \n");
+			printf("#define L3_SIZE              94371840 \n");
+			printf("#define L3_LINESIZE          64       \n");
+			printf("#define L3_ASSOCIATIVE       32       \n");
+			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
+			printf("#define DTB_SIZE             4096     \n");
+			break;
 	}
 	get_cpucount();
 }

cpuid_x86.c

@@ -1454,10 +1454,11 @@ int get_cpuname(void){
 	return CPUTYPE_OPTERON;
       case  1:
       case  3:
-      case  7:
-      case 10:
+//      case  7:
+//      case 10:
 	return CPUTYPE_BARCELONA;
       case  5:
+      case  7:		      
 	return CPUTYPE_BOBCAT;
       case  6:
 	switch (model) {
@@ -1507,6 +1508,8 @@ int get_cpuname(void){
 	  // AMD Ryzen
 	case 8:
 	  // AMD Ryzen2
+	default:
+	  // Matisse/Renoir and other recent Ryzen2		
 	  if(support_avx())
 #ifndef NO_AVX2
 	    return CPUTYPE_ZEN;
@@ -1516,6 +1519,16 @@ int get_cpuname(void){
 	  else
 	    return CPUTYPE_BARCELONA;
         }
+	break;	      
+      case 10: // Zen3		      
+	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;
     }
@@ -2107,7 +2120,7 @@ int get_coretype(void){
 	    return CORE_PILEDRIVER;
 	  else
 	    return CORE_BARCELONA; //OS don't support AVX.
-    case 5: // New EXCAVATOR
+        case 5: // New EXCAVATOR
 	  if(support_avx())
 	    return CORE_EXCAVATOR;
 	  else
@@ -2135,12 +2148,14 @@ int get_coretype(void){
 	  }
 	  break;
 	}
-      } else if (exfamily == 8) {
+      } else if (exfamily == 8 || exfamily == 10) {
 	switch (model) {
 	case 1:
 	  // AMD Ryzen
 	case 8:
-	  // Ryzen 2		
+	  // Ryzen 2
+	default:
+	  // Matisse,Renoir Ryzen2 models		
 	  if(support_avx())
 #ifndef NO_AVX2
 	    return CORE_ZEN;

driver/others/dynamic.c

@@ -656,7 +656,7 @@ static gotoblas_t *get_coretype(void){
       if ((exfamily == 0) || (exfamily == 2)) {
 	if (ecx & (1 <<  0)) return &gotoblas_OPTERON_SSE3;
 	else return &gotoblas_OPTERON;
-      }  else if (exfamily == 5) {
+      }  else if (exfamily == 5 || exfamily == 7) {
 	return &gotoblas_BOBCAT;
       } else if (exfamily == 6) {
 	if(model == 1){
@@ -710,7 +710,7 @@ static gotoblas_t *get_coretype(void){
 	  }
 	}
       } else if (exfamily == 8) {
-	if (model == 1 || model == 8) {
+	/* if (model == 1 || model == 8) */ {
 	  if(support_avx())
 	    return &gotoblas_ZEN;
 	  else{
@@ -718,16 +718,24 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
-      } else if (exfamily == 9) {
+      } else if (exfamily == 9) {  
 	  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 if (exfamily == 10) {  
+	  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;
       }
+   
     }
   }
 

driver/others/dynamic_arm64.c

@@ -53,10 +53,11 @@ extern gotoblas_t  gotoblas_THUNDERX2T99;
 extern gotoblas_t  gotoblas_TSV110;
 extern gotoblas_t  gotoblas_EMAG8180;
 extern gotoblas_t  gotoblas_NEOVERSEN1;
+extern gotoblas_t  gotoblas_THUNDERX3T110;
 
 extern void openblas_warning(int verbose, const char * msg);
 
-#define NUM_CORETYPES   11
+#define NUM_CORETYPES   12
 
 /*
  * In case asm/hwcap.h is outdated on the build system, make sure
@@ -82,6 +83,7 @@ static char *corename[] = {
   "tsv110",
   "emag8180",
   "neoversen1",
+  "thunderx3t110",
   "unknown"
 };
 
@@ -97,6 +99,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_TSV110)       return corename[ 8];
   if (gotoblas == &gotoblas_EMAG8180)     return corename[ 9];
   if (gotoblas == &gotoblas_NEOVERSEN1)   return corename[10];
+  if (gotoblas == &gotoblas_THUNDERX3T110) return corename[11];
   return corename[NUM_CORETYPES];
 }
 
@@ -127,6 +130,7 @@ static gotoblas_t *force_coretype(char *coretype) {
     case  8: return (&gotoblas_TSV110);
     case  9: return (&gotoblas_EMAG8180);
     case 10: return (&gotoblas_NEOVERSEN1);
+    case 11: return (&gotoblas_THUNDERX3T110);
   }
   snprintf(message, 128, "Core not found: %s\n", coretype);
   openblas_warning(1, message);
@@ -190,6 +194,8 @@ static gotoblas_t *get_coretype(void) {
           return &gotoblas_THUNDERX;
         case 0x0af: // ThunderX2
           return &gotoblas_THUNDERX2T99;
+        case 0x0b8: // ThunderX3
+          return &gotoblas_THUNDERX3T110;
       }
       break;
     case 0x48: // HiSilicon

getarch.c

@@ -1174,6 +1174,24 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "EMAG8180"
 #endif
 
+#ifdef FORCE_THUNDERX3T110
+#define ARMV8
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "THUNDERX3T110"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DTHUNDERX3T110 " \
+       "-DL1_CODE_SIZE=65536 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=8 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=8 " \
+       "-DL2_SIZE=524288 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \
+       "-DL3_SIZE=94371840 -DL3_LINESIZE=64 -DL3_ASSOCIATIVE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "thunderx3t110"
+#define CORENAME  "THUNDERX3T110"
+#else
+#endif
+
 #ifdef FORCE_ZARCH_GENERIC
 #define FORCE
 #define ARCHITECTURE    "ZARCH"

interface/swap.c

@@ -42,7 +42,7 @@
 #include "functable.h"
 #endif
 
-#if defined(THUNDERX2T99) || defined(VULCAN) || defined(ARMV8)
+#if defined(THUNDERX2T99) || defined(VULCAN) || defined(ARMV8) || defined(THUNDERX3T110)
 // Multithreaded swap gives performance benefits in ThunderX2T99
 #else
 // Disable multi-threading as it does not show any performance

interface/zswap.c

@@ -42,7 +42,7 @@
 #include "functable.h"
 #endif
 
-#if defined(THUNDERX2T99) || defined(VULCAN) || defined(ARMV8)
+#if defined(THUNDERX2T99) || defined(VULCAN) || defined(ARMV8) || defined(THUNDERX3T110)
 // Multithreaded swap gives performance benefits in ThunderX2T99
 #else
 // Disable multi-threading as it does not show any performance

kernel/Makefile

@@ -10,6 +10,11 @@ ifeq ($(C_COMPILER), GCC)
 GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
 endif
 
+ifeq ($(ARCH), power)
+ifeq ($(C_COMPILER), CLANG)
+ override CFLAGS += -fno-integrated-as
+endif
+endif
 AVX2OPT = 
 ifeq ($(C_COMPILER), GCC)
 # AVX2 support was added in 4.7.0

kernel/Makefile.L3

@@ -44,8 +44,10 @@ USE_TRMM = 1
 endif
 
 ifeq ($(CORE), POWER8)
+ifeq ($(BINARY64),1)
 USE_TRMM = 1
 endif
+endif
 
 ifeq ($(CORE), POWER9)
 USE_TRMM = 1

kernel/arm/zdot.c

@@ -48,10 +48,12 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 
 	dot[0]=0.0;
 	dot[1]=0.0;
-
+#if !defined(__PPC__)
 	CREAL(result) = 0.0 ;
 	CIMAG(result) = 0.0 ;
-
+#else
+	result = OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
+#endif
 	if ( n < 1 )  return(result);
 
 	inc_x2 = 2 * inc_x ;
@@ -71,8 +73,12 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 		i++ ;
 
 	}
-	CREAL(result) = dot[0];
+#if !defined(__POWER__)	
+        CREAL(result) = dot[0];
 	CIMAG(result) = dot[1];
+#else
+	result = OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0],dot[1]);
+#endif
 	return(result);
 
 }

kernel/arm64/KERNEL.THUNDERX3T110

@@ -0,0 +1,184 @@
+SAMINKERNEL  = ../arm/amin.c
+DAMINKERNEL  = ../arm/amin.c
+CAMINKERNEL  = ../arm/zamin.c
+ZAMINKERNEL  = ../arm/zamin.c
+
+SMAXKERNEL   = ../arm/max.c
+DMAXKERNEL   = ../arm/max.c
+
+SMINKERNEL   = ../arm/min.c
+DMINKERNEL   = ../arm/min.c
+
+ISAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = ../arm/iamin.c
+ICAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = ../arm/izamin.c
+
+ISMAXKERNEL  = ../arm/imax.c
+IDMAXKERNEL  = ../arm/imax.c
+
+ISMINKERNEL  = ../arm/imin.c
+IDMINKERNEL  = ../arm/imin.c
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+SAMAXKERNEL  = amax.S
+DAMAXKERNEL  = amax.S
+CAMAXKERNEL  = zamax.S
+ZAMAXKERNEL  = zamax.S
+
+SAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = daxpy_thunderx2t99.S
+CAXPYKERNEL  = zaxpy.S
+ZAXPYKERNEL  = zaxpy.S
+
+SROTKERNEL   = rot.S
+DROTKERNEL   = rot.S
+CROTKERNEL   = zrot.S
+ZROTKERNEL   = zrot.S
+
+SSCALKERNEL  = scal.S
+DSCALKERNEL  = scal.S
+CSCALKERNEL  = zscal.S
+ZSCALKERNEL  = zscal.S
+
+SGEMVNKERNEL = gemv_n.S
+DGEMVNKERNEL = gemv_n.S
+CGEMVNKERNEL = zgemv_n.S
+ZGEMVNKERNEL = zgemv_n.S
+
+SGEMVTKERNEL = gemv_t.S
+DGEMVTKERNEL = gemv_t.S
+CGEMVTKERNEL = zgemv_t.S
+ZGEMVTKERNEL = zgemv_t.S
+
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+SGEMMONCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+
+ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
+
+ifeq ($(DGEMM_UNROLL_M), 8)
+DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
+DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
+else
+DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
+endif
+
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifeq ($(DGEMM_UNROLL_N), 4)
+DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
+DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
+else
+DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
+endif
+
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+SASUMKERNEL    = sasum_thunderx2t99.c
+DASUMKERNEL    = dasum_thunderx2t99.c
+CASUMKERNEL    = casum_thunderx2t99.c
+ZASUMKERNEL    = zasum_thunderx2t99.c
+
+SCOPYKERNEL    = copy_thunderx2t99.c
+DCOPYKERNEL    = copy_thunderx2t99.c
+CCOPYKERNEL    = copy_thunderx2t99.c
+ZCOPYKERNEL    = copy_thunderx2t99.c
+
+SSWAPKERNEL    = swap_thunderx2t99.S
+DSWAPKERNEL    = swap_thunderx2t99.S
+CSWAPKERNEL    = swap_thunderx2t99.S
+ZSWAPKERNEL    = swap_thunderx2t99.S
+
+ISAMAXKERNEL   = iamax_thunderx2t99.c
+IDAMAXKERNEL   = iamax_thunderx2t99.c
+ICAMAXKERNEL   = izamax_thunderx2t99.c
+IZAMAXKERNEL   = izamax_thunderx2t99.c
+
+SNRM2KERNEL    = scnrm2_thunderx2t99.c
+CNRM2KERNEL    = scnrm2_thunderx2t99.c
+#DNRM2KERNEL    = dznrm2_thunderx2t99_fast.c
+#ZNRM2KERNEL    = dznrm2_thunderx2t99_fast.c
+DNRM2KERNEL    = dznrm2_thunderx2t99.c
+ZNRM2KERNEL    = dznrm2_thunderx2t99.c
+
+
+DDOTKERNEL     = dot_thunderx2t99.c
+SDOTKERNEL     = dot_thunderx2t99.c
+CDOTKERNEL     = zdot_thunderx2t99.c
+ZDOTKERNEL     = zdot_thunderx2t99.c
+DSDOTKERNEL    = dot.S
+
+ifeq ($(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N), 8x4)
+DGEMMKERNEL    = dgemm_kernel_8x4_thunderx2t99.S
+endif
+
+ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 16x4)
+SGEMMKERNEL    =  sgemm_kernel_16x4_thunderx2t99.S
+endif
+
+ifeq ($(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N), 8x4)
+CGEMMKERNEL    =  cgemm_kernel_8x4_thunderx2t99.S
+endif
+
+ifeq ($(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N), 4x4)
+ZGEMMKERNEL    =  zgemm_kernel_4x4_thunderx2t99.S
+endif

kernel/power/KERNEL.POWER8

@@ -1,3 +1,44 @@
+# Big-endian 32bit (AIX) is supported through the POWER6 GEMM kernels, no separate TRMM
+ifeq ($(__BYTE_ORDER__)$(BINARY32),__ORDER_BIG_ENDIAN__1)
+SGEMMKERNEL    =  gemm_kernel_power6.S
+SGEMMINCOPY    =
+SGEMMITCOPY    =
+SGEMMONCOPY    =  gemm_ncopy_4.S
+SGEMMOTCOPY    =  gemm_tcopy_4.S
+SGEMMINCOPYOBJ =
+SGEMMITCOPYOBJ =
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+DGEMMKERNEL    =  gemm_kernel_power6.S
+DGEMMINCOPY    =
+DGEMMITCOPY    =
+DGEMMONCOPY    =  gemm_ncopy_4.S
+DGEMMOTCOPY    =  gemm_tcopy_4.S
+DGEMMINCOPYOBJ =
+DGEMMITCOPYOBJ =
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+CGEMMKERNEL    =  zgemm_kernel_power6.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_2.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_2.c
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMKERNEL    =  zgemm_kernel_power6.S
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_2.c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+else
+
 #SGEMM_BETA = ../generic/gemm_beta.c
 #DGEMM_BETA = ../generic/gemm_beta.c
 #CGEMM_BETA = ../generic/zgemm_beta.c
@@ -47,16 +88,24 @@ ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
 ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
 ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
 
 STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
 STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
 STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
 STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
 
+ifeq ($(__BYTE_ORDER__)$(BINARY32),__ORDER_BIG_ENDIAN__1)
+DTRSMKERNEL_LN  =  trsm_kernel_power6_LN.S
+DTRSMKERNEL_LT  =  trsm_kernel_power6_LT.S
+DTRSMKERNEL_RN  =  trsm_kernel_power6_LT.S
+DTRSMKERNEL_RT  =  trsm_kernel_power6_RT.S
+else
 DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 DTRSMKERNEL_LT	= dtrsm_kernel_LT_16x4_power8.S
 DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
 DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+endif
 
 CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
@@ -153,6 +202,10 @@ ZASUMKERNEL  = zasum.c
 #
 SAXPYKERNEL  = saxpy.c
 DAXPYKERNEL  = daxpy.c
+#
+ifeq ($(__BYTE_ORDER__)$(BINARY32),__ORDER_BIG_ENDIAN__1)
+CAXPYKERNEL  = zaxpy.S
+else
 ifneq ($(__BYTE_ORDER__),__ORDER_BIG_ENDIAN__)
 ifneq ($(GCCVERSIONGTEQ9),1)
 CAXPYKERNEL  = caxpy_power8.S
@@ -162,6 +215,7 @@ endif
 else
 CAXPYKERNEL  = caxpy.c
 endif
+endif
 #
 ZAXPYKERNEL  = zaxpy.c
 #
@@ -239,4 +293,3 @@ IDAMINKERNEL  = ../arm/iamin.c
 IZAMAXKERNEL  = ../arm/izamax.c
 IZAMINKERNEL  = ../arm/izamin.c
 endif
-

kernel/power/casum.c

@@ -47,8 +47,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #if defined(POWER8)  || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "casum_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_16

kernel/power/ccopy.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "ccopy_microk_power8.c"
 #endif
+#endif
 
 #ifndef HAVE_KERNEL_32
 

kernel/power/cdot.c

@@ -23,6 +23,9 @@ CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *****************************************************************************/
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/zdot.c"
+#else
 
 #include "common.h"
 #ifndef HAVE_KERNEL_8
@@ -168,3 +171,4 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
     return (result);
 
 }
+#endif

kernel/power/cgemv_n.c

@@ -23,7 +23,10 @@ SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- *****************************************************************************/
+*****************************************************************************/
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/zgemv_n.c"
+#else
 
 #include <stdlib.h>
 #include <stdio.h>
@@ -591,4 +594,4 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
 
     return (0);
 }
-
+#endif

kernel/power/cgemv_t.c

@@ -23,7 +23,10 @@ SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- *****************************************************************************/
+*****************************************************************************/
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/zgemv_t.c"
+#else
 
 #include "common.h"
 
@@ -595,4 +598,4 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
     return (0);
 
 }
-
+#endif

kernel/power/crot.c

@@ -28,6 +28,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
  
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 
 static void crot_kernel_8 (long n, float *x, float *y, float c, float s)
 {
@@ -169,6 +170,7 @@ static void crot_kernel_8 (long n, float *x, float *y, float c, float s)
 }
  
 #endif
+#endif
 
 
 int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)
@@ -183,7 +185,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
 
     if ( (inc_x == 1) && (inc_y == 1) )
     {
-
+#if defined(__VEC__) || defined(__ALTIVEC__)
         BLASLONG n1 = n & -8; 
         if ( n1 > 0 )
         { 
@@ -191,7 +193,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
             i=n1; 
             ix=2*n1; 
         }
-
+#endif
          while(i < n)
            {
                 temp[0]   = c*x[ix]   + s*y[ix] ;

kernel/power/cswap.c

@@ -37,8 +37,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8)  || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "cswap_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_32

kernel/power/dasum.c

@@ -47,8 +47,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "dasum_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_16

kernel/power/daxpy.c

@@ -37,8 +37,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "daxpy_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_8

kernel/power/dcopy.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "dcopy_microk_power8.c"
 #endif
+#endif
 
 #ifndef HAVE_KERNEL_32
 

kernel/power/ddot.c

@@ -37,8 +37,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8)  || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "ddot_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_8

kernel/power/dgemm_kernel_power10.c

@@ -27,64 +27,64 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 #include <altivec.h>
 
-typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef __vector unsigned char  vec_t;
 typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
 typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
 
 #ifdef TRMMKERNEL
 #define SAVE_ACC(ACC, J)  \
-          __builtin_mma_disassemble_acc (result, ACC); \
+          __builtin_mma_disassemble_acc ((void *)result, ACC); \
           rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] = result[3] * alpha; \
+          rowC[0] = result[0] * alpha; \
           rowC = (v4sf_t *) &CO[1*ldc+J]; \
-          rowC[0] = result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[2*ldc+J]; \
           rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
           rowC = (v4sf_t *) &CO[3*ldc+J]; \
-          rowC[0] = result[0] * alpha;
+          rowC[0] = result[3] * alpha;
 #define SAVE_ACC1(ACC, J)  \
-          __builtin_mma_disassemble_acc (result, ACC); \
+          __builtin_mma_disassemble_acc ((void *)result, ACC); \
           rowC = (v4sf_t *) &CO[4* ldc+J]; \
-          rowC[0] = result[3] * alpha; \
+          rowC[0] = result[0] * alpha; \
           rowC = (v4sf_t *) &CO[5*ldc+J]; \
-          rowC[0] = result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[6*ldc+J]; \
           rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
           rowC = (v4sf_t *) &CO[7*ldc+J]; \
-          rowC[0] = result[0] * alpha;
+          rowC[0] = result[3] * alpha;
 #define  SAVE2x4_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] = result[3] * alpha; \
+          rowC[0] = result[0] * alpha; \
 	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
-          rowC[0] = result[2] * alpha;
+          rowC[0] = result[1] * alpha;
 #else
 #define SAVE_ACC(ACC, J)  \
-          __builtin_mma_disassemble_acc (result, ACC); \
+          __builtin_mma_disassemble_acc ((void *)result, ACC); \
           rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
           rowC = (v4sf_t *) &CO[1*ldc+J]; \
-          rowC[0] += result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[2*ldc+J]; \
           rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
           rowC = (v4sf_t *) &CO[3*ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[3] * alpha;
 #define SAVE_ACC1(ACC, J)  \
-          __builtin_mma_disassemble_acc (result, ACC); \
+          __builtin_mma_disassemble_acc ((void *)result, ACC); \
           rowC = (v4sf_t *) &CO[4* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
           rowC = (v4sf_t *) &CO[5*ldc+J]; \
-          rowC[0] += result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[6*ldc+J]; \
           rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
           rowC = (v4sf_t *) &CO[7*ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[3] * alpha;
 #define  SAVE2x4_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
-          rowC[0] += result[2] * alpha;
+          rowC[0] += result[1] * alpha;
 #endif
 
 #define SET_ACC_ZERO4() \

kernel/power/dgemv_n.c

@@ -39,8 +39,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "dgemv_n_microk_power8.c"
 #endif
+#endif
 
 
 #define NBMAX 4096

kernel/power/dgemv_t.c

@@ -25,15 +25,19 @@ OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *****************************************************************************/
 
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/gemv_t.c"
+#else
+
 #include "common.h"
 
 #define NBMAX 1024
 //#define PREFETCH 1
+
 #include <altivec.h> 
 
 #define HAVE_KERNEL4x8_ASM 1
 
-
 #if defined(HAVE_KERNEL4x8_ASM)
 static void dgemv_kernel_4x8(BLASLONG n, BLASLONG lda, double *ap, double *x, double *y, double alpha) {
 
@@ -355,7 +359,7 @@ static void dgemv_kernel_4x8(BLASLONG n, BLASLONG lda, double *ap, double *x, do
             "stxvd2x 39, %[off], %[y]  \n\t"
             "stxvd2x 40, %[off2], %[y]  \n\t"     
                  
-            : [memy] "+m" (*(const double (*)[8])y),
+            : [memy] "+m" (*(double (*)[8])y),
             [n] "+&r" (n),
             [a0] "=b" (a0),
             [a1] "=&b" (a1),
@@ -369,7 +373,7 @@ static void dgemv_kernel_4x8(BLASLONG n, BLASLONG lda, double *ap, double *x, do
             [off2]"=&b" (off2),
             [temp] "=&b" (tempR)
             : [memx] "m" (*(const double (*)[n])x),
-            [mem_ap] "m" (*(const double (*)[]) ap),
+            [mem_ap] "m" (*(const double (*)[n*8]) ap),
             [alpha] "d" (alpha),
             "[a0]" (ap),
             [x] "b" (x),
@@ -883,4 +887,5 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
     return (0);
 
 }
+#endif
 

kernel/power/drot.c

@@ -40,8 +40,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #pragma GCC optimize "O1"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "drot_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_16

kernel/power/dscal.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "dscal_microk_power8.c"
 #endif
+#endif
 
 #if !defined(HAVE_KERNEL_8)
 

kernel/power/dswap.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "dswap_microk_power8.c"
 #endif
+#endif
 
 #ifndef HAVE_KERNEL_32
 

kernel/power/idamax.c

@@ -26,7 +26,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *****************************************************************************/
 #include "common.h"
 #include <math.h>
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include <altivec.h>
+#endif
+
 #if defined(DOUBLE)
 
 #define ABS fabs
@@ -37,6 +40,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 /**
  * Find  maximum index 
  * Warning: requirements n>0  and n % 32 == 0
@@ -313,6 +318,7 @@ static BLASLONG diamax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *maxf) {
     return index;
 
 }
+#endif
 
 BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
     BLASLONG i = 0;
@@ -326,12 +332,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
 
         BLASLONG n1 = n & -32;
 #if defined(_CALL_ELF) && (_CALL_ELF == 2)
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 	if (n1 > 0) {
 
             max = diamax_kernel_32(n1, x, &maxf);
 
             i = n1;
         }
+#endif	
 #endif	
         while (i < n) {
             if (ABS(x[i]) > maxf) {

kernel/power/idamin.c

@@ -37,6 +37,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 /**
  * Find  minimum index 
  * Warning: requirements n>0  and n % 32 == 0
@@ -313,7 +315,7 @@ static BLASLONG diamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *minf) {
     return index;
 
 }
-
+#endif
 
 
 BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
@@ -327,12 +329,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
     if (inc_x == 1) {
 
 #if defined(_CALL_ELF) && (_CALL_ELF == 2)
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
         BLASLONG n1 = n & -32;
 	if (n1 > 0) {
 
             min = diamin_kernel_32(n1, x, &minf);
             i = n1;
         }
+#endif
 #endif
         while (i < n) {
             if (ABS(x[i]) < minf) {

kernel/power/izamax.c

@@ -34,6 +34,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 
+#if defined(__VEC__) || defined(__ALTIVEC__)
  
 /**
  * Find  maximum index 
@@ -299,7 +300,7 @@ static BLASLONG   ziamax_kernel_16(BLASLONG n, FLOAT *x, FLOAT *maxf) {
 
 }
  
-  
+#endif  
 
  
  
@@ -317,6 +318,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
     if (inc_x == 1) {
 
 #if defined(_CALL_ELF) && (_CALL_ELF == 2)
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
       BLASLONG n1 = n & -16;
       if (n1 > 0) {
 
@@ -324,6 +327,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
             i = n1;
             ix = n1 << 1;
       }
+#endif
 #endif
 
       while(i < n)

kernel/power/izamin.c

@@ -24,7 +24,6 @@ CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
- 
 
 #include "common.h"
 #include <math.h>
@@ -32,6 +31,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ABS fabs 
 #define CABS1(x,i)    ABS(x[i])+ABS(x[i+1])
 
+#if defined(__VEC__) || defined(__ALTIVEC__)
  
 /**
  * Find  minimum index 
@@ -296,6 +296,7 @@ static BLASLONG ziamin_kernel_16_TUNED(BLASLONG n, FLOAT *x, FLOAT *minf) {
     return index; 
 }
 
+#endif 
  
 
  
@@ -316,6 +317,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         minf = CABS1(x,0); //index will not be incremented
 
 #if defined(_CALL_ELF) && (_CALL_ELF == 2)
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 	BLASLONG n1 = n & -16;
         if (n1 > 0) {
 
@@ -323,6 +326,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
             i = n1;
             ix = n1 << 1;
         }
+#endif 
 #endif 
 
         while(i < n)
@@ -359,5 +363,3 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
     }
  
 }
-
-

kernel/power/sasum.c

@@ -47,8 +47,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "sasum_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_32

kernel/power/saxpy.c

@@ -28,8 +28,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
+#define  offset_0 0
+#define  offset_1 16
+#define  offset_2 32
+#define  offset_3 48
+#define  offset_4 64
+#define  offset_5 80
+#define  offset_6 96
+#define  offset_7 112
+#define  offset_8 128
+#define  offset_9 144
+#define  offset_10 160
+#define  offset_11 176
+#define  offset_12 192
+#define  offset_13 208
+#define  offset_14 224
+#define  offset_15 240
  
 
+#if defined(__VEC__) || defined(__ALTIVEC__)
 
 #ifndef HAVE_KERNEL_8
 #include <altivec.h> 
@@ -37,12 +54,85 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 static void saxpy_kernel_64(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT alpha)
 {
     BLASLONG  i = 0;
-    __vector float v_a = {alpha,alpha,alpha,alpha}; 
-    __vector float * v_y=(__vector float *)y;
-    __vector float * v_x=(__vector float *)x;
+    __vector float v_a __attribute((aligned(16))) = {alpha,alpha,alpha,alpha}; 
+    __vector float * vptr_y =(__vector float *)y;
+    __vector float * vptr_x =(__vector float *)x;
         
     for(; i<n/4; i+=16){
 
+
+        register __vector float vy_0 = vec_vsx_ld( offset_0 ,vptr_y ) ;
+        register __vector float vy_1 = vec_vsx_ld( offset_1 ,vptr_y ) ;
+        register __vector float vy_2 = vec_vsx_ld( offset_2 ,vptr_y ) ;
+        register __vector float vy_3 = vec_vsx_ld( offset_3 ,vptr_y ) ;
+        register __vector float vy_4 = vec_vsx_ld( offset_4 ,vptr_y ) ;
+        register __vector float vy_5 = vec_vsx_ld( offset_5 ,vptr_y ) ;
+        register __vector float vy_6 = vec_vsx_ld( offset_6 ,vptr_y ) ;
+        register __vector float vy_7 = vec_vsx_ld( offset_7 ,vptr_y ) ;
+        register __vector float vy_8 = vec_vsx_ld( offset_8 ,vptr_y ) ;
+        register __vector float vy_9 = vec_vsx_ld( offset_9 ,vptr_y ) ;
+        register __vector float vy_10 = vec_vsx_ld( offset_10 ,vptr_y ) ;
+        register __vector float vy_11 = vec_vsx_ld( offset_11 ,vptr_y ) ;
+        register __vector float vy_12 = vec_vsx_ld( offset_12 ,vptr_y ) ;
+        register __vector float vy_13 = vec_vsx_ld( offset_13 ,vptr_y ) ;
+        register __vector float vy_14 = vec_vsx_ld( offset_14 ,vptr_y ) ;
+        register __vector float vy_15 = vec_vsx_ld( offset_15 ,vptr_y ) ;
+
+        register __vector float vx_0 = vec_vsx_ld( offset_0 ,vptr_x ) ;
+        register __vector float vx_1 = vec_vsx_ld( offset_1 ,vptr_x ) ;
+        register __vector float vx_2 = vec_vsx_ld( offset_2 ,vptr_x ) ;
+        register __vector float vx_3 = vec_vsx_ld( offset_3 ,vptr_x ) ;
+        register __vector float vx_4 = vec_vsx_ld( offset_4 ,vptr_x ) ;
+        register __vector float vx_5 = vec_vsx_ld( offset_5 ,vptr_x ) ;
+        register __vector float vx_6 = vec_vsx_ld( offset_6 ,vptr_x ) ;
+        register __vector float vx_7 = vec_vsx_ld( offset_7 ,vptr_x ) ;
+        register __vector float vx_8 = vec_vsx_ld( offset_8 ,vptr_x ) ;
+        register __vector float vx_9 = vec_vsx_ld( offset_9 ,vptr_x ) ;
+        register __vector float vx_10 = vec_vsx_ld( offset_10 ,vptr_x ) ;
+        register __vector float vx_11 = vec_vsx_ld( offset_11 ,vptr_x ) ;
+        register __vector float vx_12 = vec_vsx_ld( offset_12 ,vptr_x ) ;
+        register __vector float vx_13 = vec_vsx_ld( offset_13 ,vptr_x ) ;
+        register __vector float vx_14 = vec_vsx_ld( offset_14 ,vptr_x ) ;
+        register __vector float vx_15 = vec_vsx_ld( offset_15 ,vptr_x ) ;
+        vy_0 += vx_0*v_a;
+        vy_1 += vx_1*v_a;
+        vy_2 += vx_2*v_a;
+        vy_3 += vx_3*v_a;
+        vy_4 += vx_4*v_a;
+        vy_5 += vx_5*v_a;
+        vy_6 += vx_6*v_a;
+        vy_7 += vx_7*v_a;
+        vy_8 += vx_8*v_a;
+        vy_9 += vx_9*v_a;
+        vy_10 += vx_10*v_a;
+        vy_11 += vx_11*v_a;
+        vy_12 += vx_12*v_a;
+        vy_13 += vx_13*v_a;
+        vy_14 += vx_14*v_a;
+        vy_15 += vx_15*v_a;
+
+    	vec_vsx_st( vy_0, offset_0 ,vptr_y ) ;
+        vec_vsx_st( vy_1, offset_1 ,vptr_y ) ;
+        vec_vsx_st( vy_2, offset_2 ,vptr_y ) ;
+        vec_vsx_st( vy_3, offset_3 ,vptr_y ) ;
+        vec_vsx_st( vy_4, offset_4 ,vptr_y ) ;
+        vec_vsx_st( vy_5, offset_5 ,vptr_y ) ;
+        vec_vsx_st( vy_6, offset_6 ,vptr_y ) ;
+	vec_vsx_st( vy_7, offset_7 ,vptr_y ) ; 
+    	vec_vsx_st( vy_8, offset_8 ,vptr_y ) ;
+        vec_vsx_st( vy_9, offset_9 ,vptr_y ) ;
+        vec_vsx_st( vy_10, offset_10 ,vptr_y ) ;
+        vec_vsx_st( vy_11, offset_11 ,vptr_y ) ;
+        vec_vsx_st( vy_12, offset_12 ,vptr_y ) ;
+        vec_vsx_st( vy_13, offset_13 ,vptr_y ) ;
+        vec_vsx_st( vy_14, offset_14 ,vptr_y ) ;
+	vec_vsx_st( vy_15, offset_15 ,vptr_y ) ; 
+
+        vptr_x+=16;
+	vptr_y+=16; 
+
+/*
+
         v_y[i]    += v_a * v_x[i];
         v_y[i+1]  += v_a * v_x[i+1];
         v_y[i+2]  += v_a * v_x[i+2];
@@ -59,9 +149,11 @@ static void saxpy_kernel_64(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT alpha)
         v_y[i+13] += v_a * v_x[i+13];
         v_y[i+14] += v_a * v_x[i+14];
         v_y[i+15] += v_a * v_x[i+15];
+*/
     }
 }
 #endif
+#endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
 {
@@ -74,11 +166,13 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 	{
 
 		BLASLONG n1 = n & -64;
+#if defined(__VEC__) || defined(__ALTIVEC__)
 
 		if ( n1 )
 			saxpy_kernel_64(n1, x, y, da);
 
 		i = n1;
+#endif
 		while(i < n)
 		{
 

kernel/power/scopy.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "scopy_microk_power8.c"
 #endif
+#endif
 
 #ifndef HAVE_KERNEL_32
 

kernel/power/sdot.c

@@ -36,8 +36,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8)  || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 #include "sdot_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_16

kernel/power/sgemm_kernel_power10.c

@@ -27,103 +27,103 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 #include <altivec.h>
 
-typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef __vector unsigned char  vec_t;
 typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
 typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
 #if defined(TRMMKERNEL)
 #define SAVE_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] = result[3] * alpha; \
+          rowC[0] = result[0] * alpha; \
           rowC = (v4sf_t *) &CO[1*ldc+J]; \
-          rowC[0] = result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[2*ldc+J]; \
           rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
           rowC = (v4sf_t *) &CO[3*ldc+J]; \
-          rowC[0] = result[0] * alpha;
+          rowC[0] = result[3] * alpha;
 #define SAVE_ACC1(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
-          rowC[0] = result[3] * alpha; \
+          rowC[0] = result[0] * alpha; \
           rowC = (v4sf_t *) &CO[5*ldc+J]; \
-          rowC[0] = result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[6*ldc+J]; \
           rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
           rowC = (v4sf_t *) &CO[7*ldc+J]; \
-          rowC[0] = result[0] * alpha;
+          rowC[0] = result[3] * alpha;
 #define  SAVE4x2_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
-          rowC[0] = result[6] * alpha; \
+          rowC[0] = result[0] * alpha; \
 	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
-          rowC[0] = result[4] * alpha; \
-	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
           rowC[0] = result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] = result[4] * alpha; \
 	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
-          rowC[0] = result[0] * alpha;
+          rowC[0] = result[6] * alpha;
 #define  SAVE4x2_ACC1(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
-          rowC[0] = result[6] * alpha; \
+          rowC[0] = result[0] * alpha; \
 	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
-          rowC[0] = result[4] * alpha; \
-	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
           rowC[0] = result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] = result[4] * alpha; \
 	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
-          rowC[0] = result[0] * alpha;
+          rowC[0] = result[6] * alpha;
 #define  SAVE2x4_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] = result[3] * alpha; \
+          rowC[0] = result[0] * alpha; \
 	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
-          rowC[0] = result[2] * alpha;
+          rowC[0] = result[1] * alpha;
 #else
 #define SAVE_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
           rowC = (v4sf_t *) &CO[1*ldc+J]; \
-          rowC[0] += result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[2*ldc+J]; \
           rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
           rowC = (v4sf_t *) &CO[3*ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[3] * alpha;
 #define SAVE_ACC1(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
           rowC = (v4sf_t *) &CO[5*ldc+J]; \
-          rowC[0] += result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[6*ldc+J]; \
           rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
           rowC = (v4sf_t *) &CO[7*ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[3] * alpha;
 #define  SAVE4x2_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[6] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
-          rowC[0] += result[4] * alpha; \
-	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
           rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
 	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[6] * alpha;
 #define  SAVE4x2_ACC1(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
-          rowC[0] += result[6] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
-          rowC[0] += result[4] * alpha; \
-	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
           rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
 	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[6] * alpha;
 #define  SAVE2x4_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
-          rowC[0] += result[2] * alpha;
+          rowC[0] += result[1] * alpha;
 #endif
 #define KERNEL(i, j) \
           __builtin_mma_xvf32gerpp (&acc0, rowB[i], rowA[j]); \

kernel/power/sgemv_n.c

@@ -24,7 +24,10 @@ CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/gemv_n.c"
 
+#else
 
 #include "common.h"
 
@@ -463,4 +466,5 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 	return(0);
 }
 
+#endif
 

kernel/power/sgemv_t.c

@@ -24,6 +24,10 @@ CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *****************************************************************************/
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/gemv_t.c"
+
+#else
 
 #include "common.h"
 
@@ -477,3 +481,4 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 
 }
 
+#endif

kernel/power/shgemm_kernel_power10.c

@@ -45,7 +45,7 @@ bfloat16tof32 (bfloat16 f16)
 #define BF16TOF32(x) x
 #endif
 
-typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef __vector unsigned char  vec_t;
 typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
 typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
 
@@ -64,54 +64,54 @@ vector char mask =
 #define MERGE_LOW(x, y) (vec_t) vec_mergel ((vector short)x, (vector short)y)
 
 #define SAVE_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
           rowC = (v4sf_t *) &CO[1*ldc+J]; \
-          rowC[0] += result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[2*ldc+J]; \
           rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
           rowC = (v4sf_t *) &CO[3*ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[3] * alpha;
 #define SAVE_ACC1(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
           rowC = (v4sf_t *) &CO[5*ldc+J]; \
-          rowC[0] += result[2] * alpha; \
-          rowC = (v4sf_t *) &CO[6*ldc+J]; \
           rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
           rowC = (v4sf_t *) &CO[7*ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[3] * alpha;
 #define  SAVE4x2_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[6] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
-          rowC[0] += result[4] * alpha; \
-	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
           rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
 	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[6] * alpha;
 #define  SAVE4x2_ACC1(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
-          rowC[0] += result[6] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
-          rowC[0] += result[4] * alpha; \
-	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
           rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
 	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
-          rowC[0] += result[0] * alpha;
+          rowC[0] += result[6] * alpha;
 
 #define MMA __builtin_mma_xvbf16ger2pp
 
 #define  SAVE2x4_ACC(ACC, J)  \
-	  __builtin_mma_disassemble_acc (result, ACC); \
+	  __builtin_mma_disassemble_acc ((void *)result, ACC); \
 	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
-          rowC[0] += result[3] * alpha; \
+          rowC[0] += result[0] * alpha; \
 	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
-          rowC[0] += result[2] * alpha;
+          rowC[0] += result[1] * alpha;
 
 #define SET_ACC_ZERO4() \
 	  __builtin_mma_xxsetaccz (&acc0); \

kernel/power/srot.c

@@ -40,8 +40,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #pragma GCC optimize "O1"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "srot_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_16

kernel/power/sscal.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "sscal_microk_power8.c"
 #endif
+#endif
 
 
 #if !defined(HAVE_KERNEL_16)

kernel/power/sswap.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "sswap_microk_power8.c"
 #endif
+#endif
 
 #ifndef HAVE_KERNEL_32
 

kernel/power/zasum.c

@@ -47,8 +47,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "zasum_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_8

kernel/power/zaxpy.c

@@ -37,8 +37,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "zaxpy_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_4

kernel/power/zcopy.c

@@ -36,8 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "zcopy_microk_power8.c"
 #endif
+#endif
 
 #ifndef HAVE_KERNEL_16
 

kernel/power/zdot.c

@@ -37,8 +37,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "zdot_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_8
@@ -93,9 +95,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
 	FLOAT  dot[4] = { 0.0, 0.0, 0.0 , 0.0 } ; 
 
 	if ( n <= 0 ) 
-	{
+	{	/*
 	        __real__ result = 0.0 ;
         	__imag__ result = 0.0 ;
+		*/
+		result = OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
 		return(result);
 
 	}
@@ -149,11 +153,17 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
 	}
 
 #if !defined(CONJ)
+	/*
 	__real__ result = dot[0] - dot[1];
 	__imag__ result = dot[2] + dot[3];
+	*/
+	result = OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
 #else
+	/*
 	__real__ result = dot[0] + dot[1];
 	__imag__ result = dot[2] - dot[3];
+	*/
+	result = OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
 
 #endif
 

kernel/power/zgemv_n_4.c

@@ -29,6 +29,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <stdio.h>
 #include "common.h"
 
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 #define HAVE_KERNEL_4x4_VEC 1
 #define HAVE_KERNEL_4x2_VEC 1
 #define HAVE_KERNEL_4x1_VEC 1
@@ -37,6 +39,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(HAVE_KERNEL_4x4_VEC) || defined(HAVE_KERNEL_4x2_VEC) || defined(HAVE_KERNEL_4x1_VEC)
 #include <altivec.h> 
 #endif
+#endif
 
 // 
 #define NBMAX 4096

kernel/power/zgemv_t_4.c

@@ -28,10 +28,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #define NBMAX 4096
+#if defined(__VEC__) || defined(__ALTIVEC__)
+
 #define HAVE_KERNEL_4x4_VEC 1
 #define HAVE_KERNEL_4x2_VEC 1
 #define HAVE_KERNEL_4x1_VEC 1
 
+#endif
 #if defined(HAVE_KERNEL_4x4_VEC) || defined(HAVE_KERNEL_4x2_VEC) || defined(HAVE_KERNEL_4x1_VEC)
 #include <altivec.h> 
 #endif

kernel/power/zrot.c

@@ -24,6 +24,9 @@ CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
 USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
+#if !defined(__VEC__) || !defined(__ALTIVEC__)
+#include "../arm/zrot.c"
+#else
 
 #include "common.h"
 
@@ -262,4 +265,4 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
  
 }
 
- 
+#endif 

kernel/power/zscal.c

@@ -39,10 +39,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #pragma GCC optimize "O1"
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #if defined(DOUBLE)
 #include "zscal_microk_power8.c"
 #endif
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_8

kernel/power/zswap.c

@@ -37,8 +37,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #if defined(POWER8) || defined(POWER9) || defined(POWER10)
+#if defined(__VEC__) || defined(__ALTIVEC__)
 #include "zswap_microk_power8.c"
 #endif
+#endif
 
 
 #ifndef HAVE_KERNEL_16

lapack/getrf/potrf_parallel.c

@@ -1,667 +0,0 @@
-/*********************************************************************/
-/* Copyright 2009, 2010 The University of Texas at Austin.           */
-/* All rights reserved.                                              */
-/*                                                                   */
-/* Redistribution and use in source and binary forms, with or        */
-/* without modification, are permitted provided that the following   */
-/* conditions are met:                                               */
-/*                                                                   */
-/*   1. Redistributions of source code must retain the above         */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer.                                                  */
-/*                                                                   */
-/*   2. Redistributions in binary form must reproduce the above      */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer in the documentation and/or other materials       */
-/*      provided with the distribution.                              */
-/*                                                                   */
-/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
-/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
-/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
-/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
-/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
-/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
-/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
-/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
-/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
-/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
-/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
-/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
-/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
-/*    POSSIBILITY OF SUCH DAMAGE.                                    */
-/*                                                                   */
-/* The views and conclusions contained in the software and           */
-/* documentation are those of the authors and should not be          */
-/* interpreted as representing official policies, either expressed   */
-/* or implied, of The University of Texas at Austin.                 */
-/*********************************************************************/
-
-#include <stdio.h>
-#include "common.h"
-
-#ifndef USE_SIMPLE_THREADED_LEVEL3
-
-//The array of job_t may overflow the stack.
-//Instead, use malloc to alloc job_t.
-#if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
-#define USE_ALLOC_HEAP
-#endif
-
-
-static FLOAT dm1 = -1.;
-
-#ifndef KERNEL_FUNC
-#ifndef LOWER
-#define KERNEL_FUNC SYRK_KERNEL_U
-#else
-#define KERNEL_FUNC SYRK_KERNEL_L
-#endif
-#endif
-
-#ifndef LOWER
-#ifndef COMPLEX
-#define TRSM_KERNEL   TRSM_KERNEL_LT
-#else
-#define TRSM_KERNEL   TRSM_KERNEL_LC
-#endif
-#else
-#ifndef COMPLEX
-#define TRSM_KERNEL   TRSM_KERNEL_RN
-#else
-#define TRSM_KERNEL   TRSM_KERNEL_RR
-#endif
-#endif
-
-#ifndef CACHE_LINE_SIZE
-#define CACHE_LINE_SIZE 8
-#endif
-
-#ifndef DIVIDE_RATE
-#define DIVIDE_RATE 2
-#endif
-
-#ifndef SWITCH_RATIO
-#define SWITCH_RATIO 2
-#endif
-
-#ifndef LOWER
-#define TRANS
-#endif
-
-#ifndef SYRK_LOCAL
-#if   !defined(LOWER) && !defined(TRANS)
-#define SYRK_LOCAL    SYRK_UN
-#elif !defined(LOWER) &&  defined(TRANS)
-#define SYRK_LOCAL    SYRK_UT
-#elif  defined(LOWER) && !defined(TRANS)
-#define SYRK_LOCAL    SYRK_LN
-#else
-#define SYRK_LOCAL    SYRK_LT
-#endif
-#endif
-
-typedef struct {
-#ifdef HAVE_C11
-  _Atomic 
-#else
-  volatile 
-#endif
-  BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
-} job_t;
-
-
-#ifndef KERNEL_OPERATION
-#ifndef COMPLEX
-#define KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, C, LDC, X, Y) \
-	KERNEL_FUNC(M, N, K, ALPHA[0], SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC, (X) - (Y))
-#else
-#define KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, C, LDC, X, Y) \
-	KERNEL_FUNC(M, N, K, ALPHA[0], ALPHA[1], SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC, (X) - (Y))
-#endif
-#endif
-
-#ifndef ICOPY_OPERATION
-#ifndef TRANS
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
-#else
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
-#endif
-#endif
-
-#ifndef OCOPY_OPERATION
-#ifdef TRANS
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
-#else
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
-#endif
-#endif
-
-#ifndef S
-#define S	args -> a
-#endif
-#ifndef A
-#define A	args -> b
-#endif
-#ifndef C
-#define C	args -> c
-#endif
-#ifndef LDA
-#define LDA	args -> lda
-#endif
-#ifndef N
-#define N	args -> m
-#endif
-#ifndef K
-#define K	args -> k
-#endif
-
-static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
-
-  FLOAT *buffer[DIVIDE_RATE];
-
-  BLASLONG k, lda;
-  BLASLONG m_from, m_to;
-
-  FLOAT *alpha;
-  FLOAT *a, *c;
-  job_t *job = (job_t *)args -> common;
-  BLASLONG xxx, bufferside;
-
-  BLASLONG jjs, min_jj;
-  BLASLONG is, min_i, div_n;
-
-  BLASLONG i, current;
-
-  k = K;
-
-  a = (FLOAT *)A;
-  c = (FLOAT *)C;
-
-  lda = LDA;
-
-  alpha = (FLOAT *)args -> alpha;
-
-  m_from = range_n[mypos + 0];
-  m_to   = range_n[mypos + 1];
-
-#if 0
-  fprintf(stderr, "Thread[%ld]  m_from : %ld m_to : %ld\n",  mypos, m_from, m_to);
-#endif
-
-  div_n = (((m_to - m_from + DIVIDE_RATE - 1) / DIVIDE_RATE + GEMM_UNROLL_MN - 1)/GEMM_UNROLL_MN) * GEMM_UNROLL_MN;
-
-  buffer[0] = (FLOAT *)((((BLASULONG)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
-  for (i = 1; i < DIVIDE_RATE; i++) {
-    buffer[i] = buffer[i - 1] + GEMM_Q * div_n * COMPSIZE;
-  }
-
-#ifndef LOWER
-  TRSM_IUNCOPY(k, k, (FLOAT *)S, lda, 0, sb);
-#else
-  TRSM_OLTCOPY(k, k, (FLOAT *)S, lda, 0, sb);
-#endif
-
-  for (xxx = m_from, bufferside = 0; xxx < m_to; xxx += div_n, bufferside ++) {
-
-    for(jjs = xxx; jjs < MIN(m_to, xxx + div_n); jjs += min_jj){
-
-      min_jj = MIN(m_to, xxx + div_n) - jjs;
-
-#ifndef LOWER
-      if (min_jj > GEMM_UNROLL_MN) min_jj = GEMM_UNROLL_MN;
-#else
-      if (min_jj > GEMM_P)         min_jj = GEMM_P;
-#endif
-
-#ifndef LOWER
-      OCOPY_OPERATION (k, min_jj, a, lda, 0, jjs, buffer[bufferside] + k * (jjs - xxx) * COMPSIZE);
-
-      TRSM_KERNEL     (k, min_jj, k, dm1,
-#ifdef COMPLEX
-		       ZERO,
-#endif
-		       sb,
-		       buffer[bufferside] + k * (jjs - xxx) * COMPSIZE,
-		       a + jjs * lda * COMPSIZE, lda, 0);
-#else
-      ICOPY_OPERATION (k, min_jj, a, lda, 0, jjs, buffer[bufferside] + k * (jjs - xxx) * COMPSIZE);
-
-      TRSM_KERNEL     (min_jj, k, k, dm1,
-#ifdef COMPLEX
-		       ZERO,
-#endif
-		       buffer[bufferside] + k * (jjs - xxx) * COMPSIZE,
-		       sb,
-		       a + jjs       * COMPSIZE, lda, 0);
-#endif
-    }
-
-#ifndef LOWER
-    for (i = 0; i <= mypos; i++)
-      job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
-#else
-    for (i = mypos; i < args -> nthreads; i++)
-      job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
-#endif
-
-    WMB;
-  }
-
-  min_i = m_to - m_from;
-
-  if (min_i >= GEMM_P * 2) {
-    min_i = GEMM_P;
-  } else
-    if (min_i > GEMM_P) {
-      min_i = (((min_i + 1) / 2 + GEMM_UNROLL_MN - 1)/GEMM_UNROLL_MN) * GEMM_UNROLL_MN;
-    }
-
-#ifndef LOWER
-  ICOPY_OPERATION(k, min_i, a, lda, 0, m_from, sa);
-#else
-  OCOPY_OPERATION(k, min_i, a, lda, 0, m_from, sa);
-#endif
-
-  current = mypos;
-
-#ifndef LOWER
-  while (current < args -> nthreads)
-#else
-  while (current >= 0)
-#endif
-    {
-      div_n = (((range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE + GEMM_UNROLL_MN - 1)/GEMM_UNROLL_MN) * GEMM_UNROLL_MN;
-
-      for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {
-
-	/* thread has to wait */
-	if (current != mypos) while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;};
-
-	KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - xxx, div_n), k, alpha,
-			 sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
-			 c, lda, m_from, xxx);
-
-	if (m_from + min_i >= m_to) {
-	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
-	  WMB;
-	}
-      }
-
-#ifndef LOWER
-      current ++;
-#else
-      current --;
-#endif
-    }
-
-  for(is = m_from + min_i; is < m_to; is += min_i){
-    min_i = m_to - is;
-
-    if (min_i >= GEMM_P * 2) {
-      min_i = GEMM_P;
-    } else
-      if (min_i > GEMM_P) {
-	min_i = (((min_i + 1) / 2 + GEMM_UNROLL_MN - 1)/GEMM_UNROLL_MN) * GEMM_UNROLL_MN;
-      }
-
-#ifndef LOWER
-    ICOPY_OPERATION(k, min_i, a, lda, 0, is, sa);
-#else
-    OCOPY_OPERATION(k, min_i, a, lda, 0, is, sa);
-#endif
-
-    current = mypos;
-
-#ifndef LOWER
-    while (current < args -> nthreads)
-#else
-      while (current >= 0)
-#endif
-	{
-	  div_n = (((range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE + GEMM_UNROLL_MN - 1)/GEMM_UNROLL_MN) * GEMM_UNROLL_MN;
-
-	  for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {
-
-	    KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - xxx, div_n), k, alpha,
-			     sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
-			     c, lda, is, xxx);
-
-	    if (is + min_i >= m_to) {
-	      job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
-	      WMB;
-	    }
-	  }
-#ifndef LOWER
-	  current ++;
-#else
-	  current --;
-#endif
-	}
-  }
-
-  for (i = 0; i < args -> nthreads; i++) {
-    if (i != mypos) {
-      for (xxx = 0; xxx < DIVIDE_RATE; xxx++) {
-	while (job[mypos].working[i][CACHE_LINE_SIZE * xxx] ) {YIELDING;};
-      }
-    }
-  }
-
-  return 0;
-  }
-
-static int thread_driver(blas_arg_t *args, FLOAT *sa, FLOAT *sb){
-
-  blas_arg_t newarg;
-
-#ifndef USE_ALLOC_HEAP
-  job_t          job[MAX_CPU_NUMBER];
-#else
-  job_t *        job = NULL;
-#endif
-
-  blas_queue_t queue[MAX_CPU_NUMBER];
-
-  BLASLONG range[MAX_CPU_NUMBER + 100];
-
-  BLASLONG num_cpu;
-
-  BLASLONG nthreads = args -> nthreads;
-
-  BLASLONG width, i, j, k;
-  BLASLONG n, n_from, n_to;
-  int  mode, mask;
-  double dnum;
-
-#ifndef COMPLEX
-#ifdef XDOUBLE
-  mode  =  BLAS_XDOUBLE | BLAS_REAL;
-  mask  = MAX(QGEMM_UNROLL_M, QGEMM_UNROLL_N) - 1;
-#elif defined(DOUBLE)
-  mode  =  BLAS_DOUBLE  | BLAS_REAL;
-  mask  = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
-#elif defined(HALF)
-  mode  =  BLAS_HALF  | BLAS_REAL;
-  mask  = MAX(SHGEMM_UNROLL_M, SHGEMM_UNROLL_N) - 1;
-#else
-  mode  =  BLAS_SINGLE  | BLAS_REAL;
-  mask  = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;
-#endif
-#else
-#ifdef XDOUBLE
-  mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
-  mask  = MAX(XGEMM_UNROLL_M, XGEMM_UNROLL_N) - 1;
-#elif defined(DOUBLE)
-  mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
-  mask  = MAX(ZGEMM_UNROLL_M, ZGEMM_UNROLL_N) - 1;
-#else
-  mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
-  mask  = MAX(CGEMM_UNROLL_M, CGEMM_UNROLL_N) - 1;
-#endif
-#endif
-
-  newarg.m        = args -> m;
-  newarg.k        = args -> k;
-  newarg.a        = args -> a;
-  newarg.b        = args -> b;
-  newarg.c        = args -> c;
-  newarg.lda      = args -> lda;
-  newarg.alpha    = args -> alpha;
-
-#ifdef USE_ALLOC_HEAP
-  job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
-  if(job==NULL){
-    fprintf(stderr, "OpenBLAS: malloc failed in %s\n", __func__);
-    exit(1);
-  }
-#endif
-
-  newarg.common   = (void *)job;
-
-  n_from = 0;
-  n_to   = args -> m;
-
-#ifndef LOWER
-
-  range[MAX_CPU_NUMBER] = n_to - n_from;
-  range[0] = 0;
-  num_cpu  = 0;
-  i        = 0;
-  n        = n_to - n_from;
-
-  dnum = (double)n * (double)n /(double)nthreads;
-
-  while (i < n){
-
-    if (nthreads - num_cpu > 1) {
-
-      double di   = (double)i;
-
-      width = ((((BLASLONG)(sqrt(di * di + dnum) - di) + mask)/(mask+1)) * (mask+1));
-
-      if (num_cpu == 0) width = n - (((n - width)/(mask+1)) * (mask+1));
-
-      if ((width > n - i) || (width < mask)) width = n - i;
-
-    } else {
-      width = n - i;
-    }
-
-    range[MAX_CPU_NUMBER - num_cpu - 1] = range[MAX_CPU_NUMBER - num_cpu] - width;
-
-    queue[num_cpu].mode    = mode;
-    queue[num_cpu].routine = inner_thread;
-    queue[num_cpu].args    = &newarg;
-    queue[num_cpu].range_m = NULL;
-
-    queue[num_cpu].sa      = NULL;
-    queue[num_cpu].sb      = NULL;
-    queue[num_cpu].next    = &queue[num_cpu + 1];
-
-    num_cpu ++;
-    i += width;
-  }
-
-   for (i = 0; i < num_cpu; i ++) queue[i].range_n = &range[MAX_CPU_NUMBER - num_cpu];
-
-#else
-
-  range[0] = 0;
-  num_cpu  = 0;
-  i        = 0;
-  n        = n_to - n_from;
-
-  dnum = (double)n * (double)n /(double)nthreads;
-
-  while (i < n){
-
-    if (nthreads - num_cpu > 1) {
-
-	double di   = (double)i;
-
-	width = ((((BLASLONG)(sqrt(di * di + dnum) - di) + mask)/(mask+1)) * (mask+1));
-
-      if ((width > n - i) || (width < mask)) width = n - i;
-
-    } else {
-      width = n - i;
-    }
-
-    range[num_cpu + 1] = range[num_cpu] + width;
-
-    queue[num_cpu].mode    = mode;
-    queue[num_cpu].routine = inner_thread;
-    queue[num_cpu].args    = &newarg;
-    queue[num_cpu].range_m = NULL;
-    queue[num_cpu].range_n = range;
-    queue[num_cpu].sa      = NULL;
-    queue[num_cpu].sb      = NULL;
-    queue[num_cpu].next    = &queue[num_cpu + 1];
-
-    num_cpu ++;
-    i += width;
-  }
-
-#endif
-
-  newarg.nthreads = num_cpu;
-
-  if (num_cpu) {
-
-    for (j = 0; j < num_cpu; j++) {
-      for (i = 0; i < num_cpu; i++) {
-	for (k = 0; k < DIVIDE_RATE; k++) {
-	  job[j].working[i][CACHE_LINE_SIZE * k] = 0;
-	}
-      }
-    }
-
-    queue[0].sa = sa;
-    queue[0].sb = sb;
-    queue[num_cpu - 1].next = NULL;
-
-    exec_blas(num_cpu, queue);
-  }
-
-#ifdef USE_ALLOC_HEAP
-  free(job);
-#endif
-
-  return 0;
-}
-
-#endif
-
-blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG myid) {
-
-  BLASLONG n, bk, i, blocking, lda;
-  BLASLONG info;
-  int mode;
-  blas_arg_t newarg;
-  FLOAT *a;
-  FLOAT alpha[2] = { -ONE, ZERO};
-
-#ifndef COMPLEX
-#ifdef XDOUBLE
-  mode  =  BLAS_XDOUBLE | BLAS_REAL;
-#elif defined(DOUBLE)
-  mode  =  BLAS_DOUBLE  | BLAS_REAL;
-#else
-  mode  =  BLAS_SINGLE  | BLAS_REAL;
-#endif
-#else
-#ifdef XDOUBLE
-  mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
-#elif defined(DOUBLE)
-  mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
-#else
-  mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
-#endif
-#endif
-
-  if (args -> nthreads  == 1) {
-#ifndef LOWER
-    info = POTRF_U_SINGLE(args, NULL, NULL, sa, sb, 0);
-#else
-    info = POTRF_L_SINGLE(args, NULL, NULL, sa, sb, 0);
-#endif
-    return info;
-  }
-
-  n  = args -> n;
-  a  = (FLOAT *)args -> a;
-  lda = args -> lda;
-
-  if (range_n) n  = range_n[1] - range_n[0];
-
-  if (n <= GEMM_UNROLL_N * 2) {
-#ifndef LOWER
-    info = POTRF_U_SINGLE(args, NULL, range_n, sa, sb, 0);
-#else
-    info = POTRF_L_SINGLE(args, NULL, range_n, sa, sb, 0);
-#endif
-    return info;
-  }
-
-  newarg.lda = lda;
-  newarg.ldb = lda;
-  newarg.ldc = lda;
-  newarg.alpha = alpha;
-  newarg.beta = NULL;
-  newarg.nthreads = args -> nthreads;
-
-  blocking = ((n / 2 + GEMM_UNROLL_N - 1)/GEMM_UNROLL_N) * GEMM_UNROLL_N;
-  if (blocking > GEMM_Q) blocking = GEMM_Q;
-
-  for (i = 0; i < n; i += blocking) {
-    bk = n - i;
-    if (bk > blocking) bk = blocking;
-
-    newarg.m = bk;
-    newarg.n = bk;
-    newarg.a = a + (i + i * lda) * COMPSIZE;
-
-    info = CNAME(&newarg, NULL, NULL, sa, sb, 0);
-    if (info) return info + i;
-
-    if (n - i - bk > 0) {
-#ifndef USE_SIMPLE_THREADED_LEVEL3
-      newarg.m = n - i - bk;
-      newarg.k = bk;
-#ifndef LOWER
-      newarg.b = a + ( i       + (i + bk) * lda) * COMPSIZE;
-#else
-      newarg.b = a + ((i + bk) +  i       * lda) * COMPSIZE;
-#endif
-      newarg.c = a + ((i + bk) + (i + bk) * lda) * COMPSIZE;
-
-      thread_driver(&newarg, sa, sb);
-#else
-
-#ifndef LOWER
-    newarg.m = bk;
-    newarg.n = n - i - bk;
-    newarg.a = a + (i +  i       * lda) * COMPSIZE;
-    newarg.b = a + (i + (i + bk) * lda) * COMPSIZE;
-
-    gemm_thread_n(mode | BLAS_TRANSA_T,
-		  &newarg, NULL, NULL, (void *)TRSM_LCUN, sa, sb, args -> nthreads);
-
-    newarg.n = n - i - bk;
-    newarg.k = bk;
-    newarg.a = a + ( i       + (i + bk) * lda) * COMPSIZE;
-    newarg.c = a + ((i + bk) + (i + bk) * lda) * COMPSIZE;
-
-#if 0
-    HERK_THREAD_UC(&newarg, NULL, NULL, sa, sb, 0);
-#else
-    syrk_thread(mode | BLAS_TRANSA_N | BLAS_TRANSB_T,
-                &newarg, NULL, NULL, (void *)HERK_UC, sa, sb, args -> nthreads);
-#endif
-#else
-    newarg.m = n - i - bk;
-    newarg.n = bk;
-    newarg.a = a + (i      + i * lda) * COMPSIZE;
-    newarg.b = a + (i + bk + i * lda) * COMPSIZE;
-
-    gemm_thread_m(mode | BLAS_RSIDE | BLAS_TRANSA_T | BLAS_UPLO,
-		  &newarg, NULL, NULL, (void *)TRSM_RCLN, sa, sb, args -> nthreads);
-
-    newarg.n = n - i - bk;
-    newarg.k = bk;
-    newarg.a = a + (i + bk +  i       * lda) * COMPSIZE;
-    newarg.c = a + (i + bk + (i + bk) * lda) * COMPSIZE;
-
-#if 0
-    HERK_THREAD_LN(&newarg, NULL, NULL, sa, sb, 0);
-#else
-    syrk_thread(mode | BLAS_TRANSA_N | BLAS_TRANSB_T | BLAS_UPLO,
-                &newarg, NULL, NULL, (void *)HERK_LN, sa, sb, args -> nthreads);
-#endif
-#endif
-
-#endif
-     }
-  }
-  return 0;
-}

lapack/potrf/potrf_parallel.c

@@ -101,7 +101,12 @@ static FLOAT dm1 = -1.;
 #endif
 
 typedef struct {
-  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+#ifdef HAVE_C11
+  _Atomic 
+#else
+  volatile 
+#endif
+  BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
 
@@ -375,6 +380,9 @@ static int thread_driver(blas_arg_t *args, FLOAT *sa, FLOAT *sb){
 #elif defined(DOUBLE)
   mode  =  BLAS_DOUBLE  | BLAS_REAL;
   mask  = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
+#elif defined(HALF)
+  mode  =  BLAS_HALF  | BLAS_REAL;
+  mask  = MAX(SHGEMM_UNROLL_M, SHGEMM_UNROLL_N) - 1;
 #else
   mode  =  BLAS_SINGLE  | BLAS_REAL;
   mask  = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;

param.h

@@ -2225,7 +2225,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define GEMM_DEFAULT_OFFSET_A 0 
 #define GEMM_DEFAULT_OFFSET_B 65536
 #define GEMM_DEFAULT_ALIGN 0x0ffffUL
-
+#if defined(__32BIT__)
+#warning using BINARY32==POWER6
+#define SGEMM_DEFAULT_UNROLL_M 4
+#define SGEMM_DEFAULT_UNROLL_N 4
+#define DGEMM_DEFAULT_UNROLL_M 4
+#define DGEMM_DEFAULT_UNROLL_N 4
+#define CGEMM_DEFAULT_UNROLL_M 2
+#define CGEMM_DEFAULT_UNROLL_N 4
+#define ZGEMM_DEFAULT_UNROLL_M 2
+#define ZGEMM_DEFAULT_UNROLL_N 4
+#else
 #define SGEMM_DEFAULT_UNROLL_M 16
 #define SGEMM_DEFAULT_UNROLL_N 8
 #define DGEMM_DEFAULT_UNROLL_M 16
@@ -2234,7 +2244,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_UNROLL_N 4
 #define ZGEMM_DEFAULT_UNROLL_M 8
 #define ZGEMM_DEFAULT_UNROLL_N 2
-
+#endif
 #define SGEMM_DEFAULT_P  1280UL
 #define DGEMM_DEFAULT_P  640UL
 #define CGEMM_DEFAULT_P  640UL
@@ -2769,6 +2779,35 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define CGEMM_DEFAULT_R 4096
 #define ZGEMM_DEFAULT_R 4096
 
+#elif defined(THUNDERX3T110)
+
+#define SGEMM_DEFAULT_UNROLL_M  16
+#define SGEMM_DEFAULT_UNROLL_N  4
+
+#define DGEMM_DEFAULT_UNROLL_M  8
+#define DGEMM_DEFAULT_UNROLL_N  4
+
+#define CGEMM_DEFAULT_UNROLL_M  8
+#define CGEMM_DEFAULT_UNROLL_N  4
+
+#define ZGEMM_DEFAULT_UNROLL_M  4
+#define ZGEMM_DEFAULT_UNROLL_N  4
+
+#define SGEMM_DEFAULT_P	128
+#define DGEMM_DEFAULT_P	320
+#define CGEMM_DEFAULT_P 128
+#define ZGEMM_DEFAULT_P 128
+
+#define SGEMM_DEFAULT_Q 352
+#define DGEMM_DEFAULT_Q 128
+#define CGEMM_DEFAULT_Q 224
+#define ZGEMM_DEFAULT_Q 112
+
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 4096
+
 #elif defined(NEOVERSEN1)
 
 #define SGEMM_DEFAULT_UNROLL_M  16