Restore -march flag for Android builds

fixes #2419 - renewed discussion in #2112 suggests removal of the option was primarily aimed at non-Android builds

.drone.yml

@@ -92,7 +92,7 @@ steps:
     - mkdir build && cd build
     - cmake $CMAKE_FLAGS ..
     - make -j
-    - ctest
+    - ctest -V
 
 ---
 kind: pipeline
@@ -116,7 +116,7 @@ steps:
     - mkdir build && cd build
     - cmake $CMAKE_FLAGS ..
     - make -j
-    - ctest
+    - ctest -V
 
 ---
 kind: pipeline
@@ -140,4 +140,4 @@ steps:
     - mkdir build && cd build
     - cmake $CMAKE_FLAGS ..
     - make -j
-    - ctest
+    - ctest -V

.gitignore

@@ -87,4 +87,5 @@ build.*
 *.swp
 benchmark/*.goto
 benchmark/smallscaling
-
+CMakeCache.txt
+CMakeFiles/*

CMakeLists.txt

@@ -6,7 +6,7 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 8)
+set(OpenBLAS_PATCH_VERSION 9.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions

Makefile.arm

@@ -1,7 +1,7 @@
 ifeq ($(CORE), $(filter $(CORE),ARMV7 CORTEXA9 CORTEXA15))
 ifeq ($(OSNAME), Android)
-CCOMMON_OPT += -mfpu=neon
-FCOMMON_OPT += -mfpu=neon
+CCOMMON_OPT += -mfpu=neon -march=armv7-a
+FCOMMON_OPT += -mfpu=neon -march=armv7-a
 else
 CCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
 FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.8
+VERSION = 0.3.9.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -97,6 +97,15 @@ VERSION = 0.3.8
 # 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 "${TARGET_CORE}" STREQUAL "LONGSOON3B" OR "${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")
+    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) )
+      set(USE_TRMM true)
+    endif ()
+    if (ZARCH OR (TARGET_CORE MATCHES POWER8) OR (TARGET_CORE MATCHES POWER9)) 
       set(USE_TRMM true)
     endif ()
 

kernel/power/KERNEL.POWER8

@@ -89,30 +89,52 @@ ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
 #SMINKERNEL   = ../arm/min.c
 #DMINKERNEL   = ../arm/min.c
 #
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
+ifneq ($(GCCVERSIONGTEQ9),1)
 ISAMAXKERNEL = isamax_power8.S
 else
 ISAMAXKERNEL = isamax.c
 endif
+else
+ISAMAXKERNEL = isamax.c
+endif
+#
 IDAMAXKERNEL = idamax.c
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+#
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
+ifneq ($(GCCVERSIONGTEQ9),1)
 ICAMAXKERNEL = icamax_power8.S
 else
 ICAMAXKERNEL = icamax.c
 endif
+else
+ICAMAXKERNEL = icamax.c
+endif
+#
 IZAMAXKERNEL = izamax.c
 #
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
+ifneq ($(GCCVERSIONGTEQ9),1)
 ISAMINKERNEL = isamin_power8.S
 else
 ISAMINKERNEL = isamin.c
 endif
+else
+ISAMINKERNEL = isamin.c
+endif
+#
 IDAMINKERNEL = idamin.c
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+#
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
+ifneq ($(GCCVERSIONGTEQ9),1)
 ICAMINKERNEL = icamin_power8.S
 else
 ICAMINKERNEL = icamin.c
 endif
+else
+ICAMINKERNEL = icamin.c
+endif
+#
 IZAMINKERNEL = izamin.c
 #
 #ISMAXKERNEL  = ../arm/imax.c
@@ -128,11 +150,16 @@ ZASUMKERNEL  = zasum.c
 #
 SAXPYKERNEL  = saxpy.c
 DAXPYKERNEL  = daxpy.c
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
+ifneq ($(GCCVERSIONGTEQ9),1)
 CAXPYKERNEL  = caxpy_power8.S
 else
 CAXPYKERNEL  = caxpy.c
 endif
+else
+CAXPYKERNEL  = caxpy.c
+endif
+#
 ZAXPYKERNEL  = zaxpy.c
 #
 SCOPYKERNEL  = scopy.c

kernel/power/KERNEL.PPC440

@@ -15,7 +15,7 @@ ZASUMKERNEL	= zasum_ppc440.S
 
 SAXPYKERNEL	=  axpy_ppc440.S
 DAXPYKERNEL	=  axpy_ppc440.S
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 CAXPYKERNEL	= ../arm/zaxpy.c
 ZAXPYKERNEL	= ../arm/zaxpy.c
 else
@@ -25,7 +25,7 @@ endif
 
 SDOTKERNEL	=  dot_ppc440.S
 DDOTKERNEL	=  dot_ppc440.S
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 CDOTKERNEL	= zdot_ppc440.S
 ZDOTKERNEL	= zdot_ppc440.S
 else
@@ -62,7 +62,7 @@ ZNRM2KERNEL	= znrm2_ppc440.S
 
 SROTKERNEL	=  rot_ppc440.S
 DROTKERNEL	=  rot_ppc440.S
-ifneq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifneq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 CROTKERNEL	= zrot_ppc440.S
 ZROTKERNEL	= zrot_ppc440.S
 else
@@ -132,7 +132,7 @@ ZTRSMKERNEL_LT	=  ztrsm_kernel_ppc440_LT.S
 ZTRSMKERNEL_RN	=  ztrsm_kernel_ppc440_LT.S
 ZTRSMKERNEL_RT	=  ztrsm_kernel_ppc440_RT.S
 
-ifeq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifeq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 SGEMVNKERNEL = ../arm/gemv_n.c
 DGEMVNKERNEL = ../arm/gemv_n.c
 SGEMVTKERNEL = ../arm/gemv_t.c

kernel/power/KERNEL.PPC970

@@ -1,4 +1,4 @@
-ifeq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifeq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 SGEMMKERNEL    =  gemm_kernel.S
 SGEMMINCOPY    =  
 SGEMMITCOPY    =  
@@ -30,7 +30,7 @@ DGEMMITCOPYOBJ =
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-ifeq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifeq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 CGEMMKERNEL    =  zgemm_kernel.S
 CGEMMINCOPY    =
 CGEMMITCOPY    =
@@ -72,7 +72,7 @@ ZTRSMKERNEL_LT	=  ztrsm_kernel_LT.S
 ZTRSMKERNEL_RN	=  ztrsm_kernel_LT.S
 ZTRSMKERNEL_RT	=  ztrsm_kernel_RT.S
 
-ifeq ($(__BYTE_ORDER__),$(__ORDER_BIG_ENDIAN__))
+ifeq ($(__BYTE_ORDER__),"__ORDER_BIG_ENDIAN__")
 STRSMKERNEL_LN	=  trsm_kernel_LN.S
 STRSMKERNEL_LT	=  trsm_kernel_LT.S
 STRSMKERNEL_RN	=  trsm_kernel_LT.S

kernel/power/caxpy_power8.S

@@ -12,6 +12,13 @@
 
 	PROLOGUE
 
+#if _CALL_ELF ==2
+#ifdef CONJ
+caxpyc_k:
+#else
+caxpy_k:
+#endif
+#endif
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l

kernel/power/icamin_power8.S

@@ -10,7 +10,9 @@
 #include "common.h"
 
 	PROLOGUE
-
+#if _CALL_ELF ==2
+icamin_k:
+#endif
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l

kernel/power/isamax_power8.S

@@ -11,7 +11,10 @@
 #include "common.h"
 
 	PROLOGUE
-
+	
+#if _CALL_ELF == 2	
+isamax_k:
+#endif
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l

kernel/power/isamin_power8.S

@@ -11,6 +11,9 @@
 
 	PROLOGUE
 
+#if _CALL_ELF ==2
+isamin_k:
+#endif
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l

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

lapack/CMakeLists.txt

@@ -4,7 +4,6 @@ include_directories(${PROJECT_BINARY_DIR})
 
 
 set(LAPACK_SOURCES
-  getrf/getrf_single.c
   potrf/potrf_U_single.c
   potrf/potrf_L_single.c
   lauum/lauum_U_single.c
@@ -45,6 +44,10 @@ GenerateNamedObjects("${LAPACK_MANGLED_SOURCES}" "" "" false "" "" false 3)
 GenerateNamedObjects("laswp/generic/laswp_k_4.c" "" "laswp_plus" false "" ""  false 3)
 GenerateNamedObjects("laswp/generic/laswp_k_4.c" "MINUS" "laswp_minus" false "" ""  false 3)
 
+foreach (float_type ${FLOAT_TYPES})
+GenerateNamedObjects("getrf/getrf_single.c" "UNIT" "getrf_single" false "" "" false ${float_type})
+endforeach ()
+
 # dynamic_arch laswp needs arch specific code ?
 #foreach(TARGET_CORE ${DYNAMIC_CORE})
 #      set(TSUFFIX "_${TARGET_CORE}")
@@ -81,7 +84,7 @@ if (USE_THREAD)
   )
 
   foreach (float_type ${FLOAT_TYPES})
-    GenerateNamedObjects("${GETRF_SRC}" "" "getrf_parallel" false "" "" false ${float_type})  
+    GenerateNamedObjects("${GETRF_SRC}" "UNIT" "getrf_parallel" false "" "" false ${float_type})
   endforeach()
 
   GenerateNamedObjects("${PARALLEL_SOURCES}")