Merge pull request #27 from xianyi/develop

rebase

Makefile.rule

@@ -97,6 +97,15 @@ VERSION = 0.3.9.dev
 # they need to wait for the preceding API calls to finish or risk data corruption.
 # NUM_PARALLEL = 2
 
+# When multithreading, OpenBLAS needs to use a memory buffer for communicating
+# and collating results for individual subranges of the original matrix. Since
+# the original GotoBLAS of the early 2000s, the default size of this buffer has
+# been set at a value of 32<<20 (which is 32MB) on x86_64 , twice that on PPC.
+# If you expect to handle large problem sizes (beyond about 30000x30000) uncomment
+# this line and adjust the (32<<n) factor if necessary. Usually an insufficient value
+# manifests itself as a crash in the relevant scal kernel (sscal_k, dscal_k etc) 
+# BUFFERSIZE = 25
+
 # If you don't need to install the static library, please comment this in.
 # NO_STATIC = 1
 

c_check

@@ -195,7 +195,7 @@ if (($architecture eq "mips") || ($architecture eq "mips64")) {
 	print $tmpf "void main(void){ __asm__ volatile($code); }\n";
 
 	$args = "$msa_flags -o $tmpf.o $tmpf";
-	my @cmd = ("$compiler_name $args");
+	my @cmd = ("$compiler_name $args >/dev/null 2>/dev/null");
 	system(@cmd) == 0;
 	if ($? != 0) {
 	    $have_msa = 0;

cmake/system.cmake

@@ -289,6 +289,10 @@ set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_CPU_NUMBER=${NUM_THREADS}")
 
 set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_PARALLEL_NUMBER=${NUM_PARALLEL}")
 
+if (BUFFERSIZE)
+set(CCOMMON_OPT "${CCOMMON_OPT} -DBUFFERSIZE=${BUFFERSIZE}")
+endif ()
+
 if (USE_SIMPLE_THREADED_LEVEL3)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_SIMPLE_THREADED_LEVEL3")
 endif ()

common_x86_64.h

@@ -225,7 +225,11 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #endif
 #define HUGE_PAGESIZE	( 2 << 20)
 
+#ifndef BUFFERSIZE
 #define BUFFER_SIZE	(32 << 20)
+#else
+#define BUFFER_SIZE	(32 << BUFFERSIZE)
+#endif
 
 #define SEEK_ADDRESS
 

cpuid_x86.c

@@ -2006,6 +2006,38 @@ int get_coretype(void){
 	    return CORE_NEHALEM;
         }
 	break;
+      case 6:
+        if (model == 6)
+#ifndef NO_AVX512
+	    return CORE_SKYLAKEX;
+#else
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_HASWELL;
+#else
+	    return CORE_SANDYBRIDGE;
+#endif
+	  else
+	    return CORE_NEHALEM;
+#endif			
+        break;    	
+      case 7:
+        if (model == 10) 
+            return CORE_NEHALEM;
+        if (model == 14)
+#ifndef NO_AVX512
+	    return CORE_SKYLAKEX;
+#else
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_HASWELL;
+#else
+	    return CORE_SANDYBRIDGE;
+#endif
+	  else
+	    return CORE_NEHALEM;
+#endif			
+        break;    	
       case 9:
       case 8:
         if (model == 14) { // Kaby Lake 

driver/others/dynamic_zarch.c

@@ -3,12 +3,12 @@
 
 extern gotoblas_t gotoblas_Z13;
 extern gotoblas_t gotoblas_Z14;
-extern gotoblas_t gotoblas_Z15;
+//extern gotoblas_t gotoblas_Z15;
 //#if (!defined C_GCC) || (GCC_VERSION >= 60000)
 //extern gotoblas_t gotoblas_Z14;
 //#endif
 
-#define NUM_CORETYPES 5
+#define NUM_CORETYPES 4
 
 extern void openblas_warning(int verbose, const char* msg);
 
@@ -16,14 +16,14 @@ static char* corename[] = {
 	"unknown",
 	"Z13",
 	"Z14",
-	"Z15",
+//	"Z15",
 	"ZARCH_GENERIC",
 };
 
 char* gotoblas_corename(void) {
 	if (gotoblas == &gotoblas_Z13)	return corename[1];
 	if (gotoblas == &gotoblas_Z14)	return corename[2];
-	if (gotoblas == &gotoblas_Z15)	return corename[3];
+//	if (gotoblas == &gotoblas_Z15)	return corename[3];
 //#if (!defined C_GCC) || (GCC_VERSION >= 60000)
 //	if (gotoblas == &gotoblas_POWER9)	return corename[3];
 //#endif
@@ -31,7 +31,7 @@ char* gotoblas_corename(void) {
 }
 
 // __builtin_cpu_is is not supported by zarch
-static gotolabs_t* get_coretype(void) {
+static gotoblas_t* get_coretype(void) {
 	FILE* infile;
 	char buffer[512], * p;
 
@@ -78,7 +78,7 @@ static gotoblas_t* force_coretype(char* coretype) {
 	{
 	case  1: return (&gotoblas_Z13);
 	case  2: return (&gotoblas_Z14);
-	case  3: return (&gotoblas_Z15);
+//	case  3: return (&gotoblas_Z15);
 //#if (!defined C_GCC) || (GCC_VERSION >= 60000)
 //	case  3: return (&gotoblas_POWER9);
 //#endif

kernel/CMakeLists.txt

@@ -121,8 +121,10 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
 
     # Makefile.L3
     set(USE_TRMM false)
-
+    if (ARM OR ARM64 OR (TARGET_CORE MATCHES LONGSOON3B) OR (TARGET_CORE MATCHES GENERIC) OR (TARGET_CORE MATCHES HASWELL) OR (TARGET_CORE MATCHES ZEN) OR (TARGET_CORE MATCHES SKYLAKEX) )
-    if (ARM OR ARM64 OR "${TARGET_CORE}" STREQUAL "LONGSOON3B" OR "${TARGET_CORE}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET_CORE}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell" OR "${CORE}" STREQUAL "zen" OR "${TARGET_CORE}" STREQUAL "SKYLAKEX" OR "${CORE}" STREQUAL "skylakex")
+      set(USE_TRMM true)
+    endif ()
+    if (ZARCH OR (TARGET_CORE MATCHES POWER8) OR (TARGET_CORE MATCHES POWER9)) 
       set(USE_TRMM true)
     endif ()
 

kernel/x86_64/dscal.c

@@ -136,10 +136,10 @@ static void dscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_
 	"jnz    1b					    \n\t"
 
         :
-          "+r" (n)      // 0
+          "+r" (n),     // 0
+          "+r" (x),     // 1
+          "+r" (x1)     // 2
         :
-          "r" (x),      // 1
-          "r" (x1),     // 2
           "r" (alpha),  // 3
           "r" (inc_x),  // 4
           "r" (inc_x3)  // 5