Merge branch 'develop' into relapack

CMakeLists.txt

@@ -236,7 +236,11 @@ install(TARGETS ${OpenBLAS_LIBNAME}
 	DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/config.h
 	COMMAND ${GENCONFIG_BIN} ${CMAKE_CURRENT_SOURCE_DIR}/config.h ${CMAKE_CURRENT_SOURCE_DIR}/openblas_config_template.h > ${CMAKE_BINARY_DIR}/openblas_config.h
 	)
-	ADD_CUSTOM_TARGET(genconfig DEPENDS openblas_config.h)	
+
+	ADD_CUSTOM_TARGET(genconfig
+	ALL
+	DEPENDS openblas_config.h
+	)
 	add_dependencies(genconfig ${OpenBLAS_LIBNAME})
 
 	install (FILES ${CMAKE_BINARY_DIR}/openblas_config.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
@@ -244,6 +248,7 @@ install(TARGETS ${OpenBLAS_LIBNAME}
 	message(STATUS "Generating f77blas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
 
 	ADD_CUSTOM_TARGET(genf77blas
+	ALL
 	COMMAND ${AWK} 'BEGIN{print \"\#ifndef OPENBLAS_F77BLAS_H\" \; print \"\#define OPENBLAS_F77BLAS_H\" \; print \"\#include \\"openblas_config.h\\" \"}; NF {print}; END{print \"\#endif\"}' ${CMAKE_CURRENT_SOURCE_DIR}/common_interface.h > ${CMAKE_BINARY_DIR}/f77blas.h
 	DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/config.h
 	)
@@ -255,11 +260,11 @@ if(NOT NO_CBLAS)
 	message (STATUS "Generating cblas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
 
 	ADD_CUSTOM_TARGET(gencblas
+	ALL
 	COMMAND ${SED} 's/common/openblas_config/g' ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h > "${CMAKE_BINARY_DIR}/cblas.tmp"
 	COMMAND cp "${CMAKE_BINARY_DIR}/cblas.tmp" "${CMAKE_BINARY_DIR}/cblas.h"
 	DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h
 	)
-
 	add_dependencies(gencblas ${OpenBLAS_LIBNAME})
 
 	install (FILES ${CMAKE_BINARY_DIR}/cblas.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})

Makefile.arm

@@ -1,5 +1,4 @@
-#ifeq logical or
-ifeq ($(CORE), $(filter $(CORE),CORTEXA9 CORTEXA15))
+ifeq ($(CORE), $(filter $(CORE),ARMV7 CORTEXA9 CORTEXA15))
 ifeq ($(OSNAME), Android)
 CCOMMON_OPT += -mfpu=neon  -march=armv7-a
 FCOMMON_OPT += -mfpu=neon  -march=armv7-a
@@ -9,28 +8,12 @@ FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
 endif
 endif
 
-ifeq ($(CORE), ARMV7)
-ifeq ($(OSNAME), Android)
-ifeq ($(ARM_SOFTFP_ABI), 1)
-CCOMMON_OPT += -mfpu=neon -march=armv7-a
-FCOMMON_OPT += -mfpu=neon -march=armv7-a
-else
-CCOMMON_OPT += -mfpu=neon -march=armv7-a -Wl,--no-warn-mismatch
-FCOMMON_OPT += -mfpu=neon -march=armv7-a -Wl,--no-warn-mismatch
-endif
-else
-CCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
-FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
-endif
-endif
-
 ifeq ($(CORE), ARMV6)
 CCOMMON_OPT += -mfpu=vfp -march=armv6
 FCOMMON_OPT += -mfpu=vfp -march=armv6
 endif
 
-
 ifeq ($(CORE), ARMV5)
-CCOMMON_OPT += -marm -march=armv5
-FCOMMON_OPT += -marm -march=armv5
+CCOMMON_OPT += -march=armv5
+FCOMMON_OPT += -march=armv5
 endif

Makefile.arm64

@@ -20,6 +20,6 @@ FCOMMON_OPT += -mtune=thunderx -mcpu=thunderx
 endif
 
 ifeq ($(CORE), THUNDERX2T99)
-CCOMMON_OPT += -mtune=vulcan -mcpu=vulcan
-FCOMMON_OPT += -mtune=vulcan -mcpu=vulcan
+CCOMMON_OPT += -mtune=thunderx2t99 -mcpu=thunderx2t99
+FCOMMON_OPT += -mtune=thunderx2t99 -mcpu=thunderx2t99
 endif

cmake/c_check.cmake

@@ -91,3 +91,8 @@ file(WRITE ${TARGET_CONF}
   "#define __${BINARY}BIT__\t1\n"
   "#define FUNDERSCORE\t${FU}\n")
 
+if (${HOST_OS} STREQUAL "WINDOWSSTORE")
+  file(APPEND ${TARGET_CONF}
+    "#define OS_WINNT\t1\n")
+endif ()
+

cmake/os.cmake

@@ -77,7 +77,7 @@ if (CYGWIN)
   set(NO_EXPRECISION 1)
 endif ()
 
-if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Windows" AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Interix")
+if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Windows" AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Interix" AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Android")
   if (SMP)
     set(EXTRALIB "${EXTRALIB} -lpthread")
   endif ()

cmake/prebuild.cmake

@@ -72,20 +72,26 @@ if (MSVC)
 set(GETARCH_FLAGS ${GETARCH_FLAGS} -DFORCE_GENERIC)
 endif()
 
+if ("${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  # disable WindowsStore strict CRT checks
+  set(GETARCH_FLAGS ${GETARCH_FLAGS} -D_CRT_SECURE_NO_WARNINGS)
+endif ()
+
 set(GETARCH_DIR "${PROJECT_BINARY_DIR}/getarch_build")
 set(GETARCH_BIN "getarch${CMAKE_EXECUTABLE_SUFFIX}")
 file(MAKE_DIRECTORY ${GETARCH_DIR})
-try_compile(GETARCH_RESULT ${GETARCH_DIR}
-  SOURCES ${GETARCH_SRC}
-  COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${PROJECT_SOURCE_DIR}
-  OUTPUT_VARIABLE GETARCH_LOG
-  COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
-)
-
-if (NOT ${GETARCH_RESULT})
-  MESSAGE(FATAL_ERROR "Compiling getarch failed ${GETARCH_LOG}")
+if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  try_compile(GETARCH_RESULT ${GETARCH_DIR}
+    SOURCES ${GETARCH_SRC}
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${PROJECT_SOURCE_DIR}
+    OUTPUT_VARIABLE GETARCH_LOG
+    COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
+  )
+
+  if (NOT ${GETARCH_RESULT})
+    MESSAGE(FATAL_ERROR "Compiling getarch failed ${GETARCH_LOG}")
+  endif ()
 endif ()
-
 message(STATUS "Running getarch")
 
 # use the cmake binary w/ the -E param to run a shell command in a cross-platform way
@@ -101,15 +107,17 @@ ParseGetArchVars(${GETARCH_MAKE_OUT})
 set(GETARCH2_DIR "${PROJECT_BINARY_DIR}/getarch2_build")
 set(GETARCH2_BIN "getarch_2nd${CMAKE_EXECUTABLE_SUFFIX}")
 file(MAKE_DIRECTORY ${GETARCH2_DIR})
-try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
-  SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
-  COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${PROJECT_SOURCE_DIR}
-  OUTPUT_VARIABLE GETARCH2_LOG
-  COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
-)
-
-if (NOT ${GETARCH2_RESULT})
-  MESSAGE(FATAL_ERROR "Compiling getarch_2nd failed ${GETARCH2_LOG}")
+if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
+    SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${PROJECT_SOURCE_DIR}
+    OUTPUT_VARIABLE GETARCH2_LOG
+    COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
+  )
+
+  if (NOT ${GETARCH2_RESULT})
+    MESSAGE(FATAL_ERROR "Compiling getarch_2nd failed ${GETARCH2_LOG}")
+  endif ()
 endif ()
 
 # use the cmake binary w/ the -E param to run a shell command in a cross-platform way
@@ -126,13 +134,15 @@ set(GEN_CONFIG_H_BIN "gen_config_h${CMAKE_EXECUTABLE_SUFFIX}")
 set(GEN_CONFIG_H_FLAGS "-DVERSION=\"${OpenBLAS_VERSION}\"")
 file(MAKE_DIRECTORY ${GEN_CONFIG_H_DIR})
 
-try_compile(GEN_CONFIG_H_RESULT ${GEN_CONFIG_H_DIR}
-  SOURCES ${PROJECT_SOURCE_DIR}/gen_config_h.c
-  COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GEN_CONFIG_H_FLAGS} -I${PROJECT_SOURCE_DIR}
-  OUTPUT_VARIABLE GEN_CONFIG_H_LOG
-  COPY_FILE ${PROJECT_BINARY_DIR}/${GEN_CONFIG_H_BIN}
-)
-
-if (NOT ${GEN_CONFIG_H_RESULT})
-  MESSAGE(FATAL_ERROR "Compiling gen_config_h failed ${GEN_CONFIG_H_LOG}")
-endif ()
+if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  try_compile(GEN_CONFIG_H_RESULT ${GEN_CONFIG_H_DIR}
+    SOURCES ${PROJECT_SOURCE_DIR}/gen_config_h.c
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GEN_CONFIG_H_FLAGS} -I${PROJECT_SOURCE_DIR}
+    OUTPUT_VARIABLE GEN_CONFIG_H_LOG
+    COPY_FILE ${PROJECT_BINARY_DIR}/${GEN_CONFIG_H_BIN}
+  )
+
+  if (NOT ${GEN_CONFIG_H_RESULT})
+    MESSAGE(FATAL_ERROR "Compiling gen_config_h failed ${GEN_CONFIG_H_LOG}")
+  endif ()
+endif ()

common.h

@@ -425,6 +425,10 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #endif
 
 #ifndef ASSEMBLER
+#ifdef OS_WINDOWSSTORE
+typedef char env_var_t[MAX_PATH];
+#define readenv(p, n) 0
+#else
 #ifdef OS_WINDOWS
 typedef char env_var_t[MAX_PATH];
 #define readenv(p, n) GetEnvironmentVariable((LPCTSTR)(n), (LPTSTR)(p), sizeof(p))
@@ -432,6 +436,7 @@ typedef char env_var_t[MAX_PATH];
 typedef char* env_var_t;
 #define readenv(p, n) ((p)=getenv(n))
 #endif
+#endif
 
 #if !defined(RPCC_DEFINED) && !defined(OS_WINDOWS)
 #ifdef _POSIX_MONOTONIC_CLOCK
@@ -654,7 +659,11 @@ static __inline void blas_unlock(volatile BLASULONG *address){
   *address = 0;
 }
 
-
+#ifdef OS_WINDOWSSTORE
+static __inline int readenv_atoi(char *env) {
+	return 0;
+}
+#else
 #ifdef OS_WINDOWS
 static __inline int readenv_atoi(char *env) {
   env_var_t p;
@@ -669,7 +678,7 @@ static __inline int readenv_atoi(char *env) {
 	return(0);
 }
 #endif
-
+#endif
 
 #if !defined(XDOUBLE) || !defined(QUAD_PRECISION)
 

common_arm.h

@@ -111,11 +111,6 @@ REALNAME:
 
 #define PROFCODE
 
-#ifdef __ARM_PCS
-//-mfloat-abi=softfp
-#define SOFT_FLOAT_ABI
-#endif
-
 #endif
 
 

driver/level2/gbmv_thread.c

@@ -177,7 +177,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT *alpha, FLOAT
 
   blas_arg_t args;
   blas_queue_t queue[MAX_CPU_NUMBER];
-  BLASLONG range_m[MAX_CPU_NUMBER];
+  BLASLONG range_m[MAX_CPU_NUMBER + 1];
   BLASLONG range_n[MAX_CPU_NUMBER + 1];
 
   BLASLONG width, i, num_cpu;

driver/level2/sbmv_thread.c

@@ -177,7 +177,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
 #endif
 
   blas_arg_t args;
-  blas_queue_t queue[MAX_CPU_NUMBER];
+  blas_queue_t queue[MAX_CPU_NUMBER + 1];
   BLASLONG range_m[MAX_CPU_NUMBER + 1];
   BLASLONG range_n[MAX_CPU_NUMBER];
 

driver/level2/spmv_thread.c

@@ -182,7 +182,7 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *y,
   blas_arg_t args;
   blas_queue_t queue[MAX_CPU_NUMBER];
   BLASLONG range_m[MAX_CPU_NUMBER + 1];
-  BLASLONG range_n[MAX_CPU_NUMBER];
+  BLASLONG range_n[MAX_CPU_NUMBER + 1];
 
   BLASLONG width, i, num_cpu;
 

driver/level2/tbmv_thread.c

@@ -221,7 +221,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
   blas_arg_t args;
   blas_queue_t queue[MAX_CPU_NUMBER];
   BLASLONG range_m[MAX_CPU_NUMBER + 1];
-  BLASLONG range_n[MAX_CPU_NUMBER];
+  BLASLONG range_n[MAX_CPU_NUMBER + 1];
 
   BLASLONG width, i, num_cpu;
 

driver/level2/tpmv_thread.c

@@ -243,7 +243,7 @@ int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthr
   blas_arg_t args;
   blas_queue_t queue[MAX_CPU_NUMBER];
   BLASLONG range_m[MAX_CPU_NUMBER + 1];
-  BLASLONG range_n[MAX_CPU_NUMBER];
+  BLASLONG range_n[MAX_CPU_NUMBER + 1];
 
   BLASLONG width, i, num_cpu;
 

driver/level2/trmv_thread.c

@@ -281,7 +281,7 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu
   blas_arg_t args;
   blas_queue_t queue[MAX_CPU_NUMBER];
   BLASLONG range_m[MAX_CPU_NUMBER + 1];
-  BLASLONG range_n[MAX_CPU_NUMBER];
+  BLASLONG range_n[MAX_CPU_NUMBER + 1];
 
   BLASLONG width, i, num_cpu;
 

driver/level3/syrk_thread.c

@@ -109,7 +109,7 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
       if (nthreads - num_cpu > 1) {
 
 	di = (double)i;
-	width = ((BLASLONG)( sqrt(di * di + dnum) - di) + mask) & ~mask;
+	width = (BLASLONG)(( sqrt(di * di + dnum) - di + mask)/(mask+1)) * (mask+1);
 
 	if ((width <= 0) || (width > n_to - i)) width = n_to - i;
 
@@ -149,7 +149,7 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
       if (nthreads - num_cpu > 1) {
 
 	di = (double)(arg -> n - i);
-	width = ((BLASLONG)(-sqrt(di * di + dnum) + di) + mask) & ~mask;
+	width = ((BLASLONG)((-sqrt(di * di + dnum) + di) + mask)/(mask+1)) * (mask+1);
 
 	if ((width <= 0) || (width > n_to - i)) width = n_to - i;
 

driver/others/CMakeLists.txt

@@ -12,6 +12,8 @@ if (SMP)
     set(BLAS_SERVER blas_server_omp.c)
   elseif (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
     set(BLAS_SERVER blas_server_win32.c)
+  elseif (${CMAKE_SYSTEM_NAME} STREQUAL "WindowsStore")
+    set(BLAS_SERVER blas_server_win32.c)
   endif ()
 
   if (NOT DEFINED BLAS_SERVER)

driver/others/blas_server_win32.c

@@ -443,8 +443,11 @@ int BLASFUNC(blas_thread_shutdown)(void){
     SetEvent(pool.killed);
 
     for(i = 0; i < blas_num_threads - 1; i++){
-     WaitForSingleObject(blas_threads[i], 5);  //INFINITE);
-	 TerminateThread(blas_threads[i],0);
+      WaitForSingleObject(blas_threads[i], 5);  //INFINITE);
+#ifndef OS_WINDOWSSTORE
+// TerminateThread is only available with WINAPI_DESKTOP and WINAPI_SYSTEM not WINAPI_APP in UWP
+      TerminateThread(blas_threads[i],0);
+#endif
     }
 
     blas_server_avail = 0;

driver/others/init.c

@@ -354,6 +354,24 @@ static int numa_check(void) {
   return common -> num_nodes;
 }
 
+#if defined(__GLIBC_PREREQ)
+#if !__GLIBC_PREREQ(2, 6)
+int sched_getcpu(void)
+{
+int cpu;
+FILE *fp = NULL;
+if ( (fp = fopen("/proc/self/stat", "r")) == NULL)
+   return -1;
+if ( fscanf( fp, "%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%d", &cpu) != 1) {
+  fclose (fp);
+  return -1;
+  }
+  fclose (fp);
+  return(cpu);
+}
+#endif
+#endif
+
 static void numa_mapping(void) {
 
   int node, cpu, core;
@@ -808,7 +826,6 @@ void gotoblas_affinity_init(void) {
   common -> shmid = pshmid;
 
   if (common -> magic != SH_MAGIC) {
-
 #ifdef DEBUG
     fprintf(stderr, "Shared Memory Initialization.\n");
 #endif
@@ -830,7 +847,7 @@ void gotoblas_affinity_init(void) {
     if (common -> num_nodes > 1) numa_mapping();
 
     common -> final_num_procs = 0;
-    for(i = 0; i < common -> avail_count; i++) common -> final_num_procs += rcount(common -> avail[i]) + 1;   //Make the max cpu number.
+    for(i = 0; i < common -> avail_count; i++) common -> final_num_procs += rcount(common -> avail[i]) + 1;   //Make the max cpu number. 
 
     for (cpu = 0; cpu < common -> final_num_procs; cpu ++) common -> cpu_use[cpu] =  0;
 

kernel/arm/KERNEL.ARMV6

@@ -1,7 +1,5 @@
+include $(KERNELDIR)/KERNEL.ARMV5
 
-
-
-###############################################################################
 SAMAXKERNEL  = iamax_vfp.S
 DAMAXKERNEL  = iamax_vfp.S
 CAMAXKERNEL  = iamax_vfp.S
@@ -44,10 +42,10 @@ DAXPYKERNEL  = axpy_vfp.S
 CAXPYKERNEL  = axpy_vfp.S
 ZAXPYKERNEL  = axpy_vfp.S
 
-SCOPYKERNEL  = copy.c
-DCOPYKERNEL  = copy.c
-CCOPYKERNEL  = zcopy.c
-ZCOPYKERNEL  = zcopy.c
+SROTKERNEL   = rot_vfp.S
+DROTKERNEL   = rot_vfp.S
+CROTKERNEL   = rot_vfp.S
+ZROTKERNEL   = rot_vfp.S
 
 SDOTKERNEL   = sdot_vfp.S
 DDOTKERNEL   = ddot_vfp.S
@@ -59,16 +57,6 @@ DNRM2KERNEL  = nrm2_vfp.S
 CNRM2KERNEL  = nrm2_vfp.S
 ZNRM2KERNEL  = nrm2_vfp.S
 
-SROTKERNEL   = rot_vfp.S
-DROTKERNEL   = rot_vfp.S
-CROTKERNEL   = rot_vfp.S
-ZROTKERNEL   = rot_vfp.S
-
-SSCALKERNEL  = scal.c
-DSCALKERNEL  = scal.c
-CSCALKERNEL  = zscal.c
-ZSCALKERNEL  = zscal.c
-
 SSWAPKERNEL  = swap_vfp.S
 DSWAPKERNEL  = swap_vfp.S
 CSWAPKERNEL  = swap_vfp.S
@@ -84,26 +72,25 @@ DGEMVTKERNEL = gemv_t_vfp.S
 CGEMVTKERNEL = cgemv_t_vfp.S
 ZGEMVTKERNEL = zgemv_t_vfp.S
 
-STRMMKERNEL	= strmm_kernel_4x2_vfp.S
-DTRMMKERNEL	= dtrmm_kernel_4x2_vfp.S
-CTRMMKERNEL	= ctrmm_kernel_2x2_vfp.S
-ZTRMMKERNEL	= ztrmm_kernel_2x2_vfp.S
-
 SGEMMKERNEL    = sgemm_kernel_4x2_vfp.S
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
 SGEMMINCOPY    = sgemm_ncopy_4_vfp.S
 SGEMMITCOPY    = sgemm_tcopy_4_vfp.S
 SGEMMINCOPYOBJ = sgemm_incopy.o
 SGEMMITCOPYOBJ = sgemm_itcopy.o
+endif
 SGEMMONCOPY    = sgemm_ncopy_2_vfp.S
-SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
-SGEMMONCOPYOBJ =  sgemm_oncopy.o
-SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+SGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+SGEMMONCOPYOBJ = sgemm_oncopy.o
+SGEMMOTCOPYOBJ = sgemm_otcopy.o
 
 DGEMMKERNEL    = dgemm_kernel_4x2_vfp.S
+ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
 DGEMMINCOPY    = dgemm_ncopy_4_vfp.S
 DGEMMITCOPY    = dgemm_tcopy_4_vfp.S
 DGEMMINCOPYOBJ = dgemm_incopy.o
 DGEMMITCOPYOBJ = dgemm_itcopy.o
+endif
 DGEMMONCOPY    = dgemm_ncopy_2_vfp.S
 DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
 DGEMMONCOPYOBJ = dgemm_oncopy.o
@@ -121,26 +108,8 @@ ZGEMMOTCOPY    = zgemm_tcopy_2_vfp.S
 ZGEMMONCOPYOBJ = zgemm_oncopy.o
 ZGEMMOTCOPYOBJ = zgemm_otcopy.o
 
-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
-
-
-
+STRMMKERNEL    = strmm_kernel_4x2_vfp.S
+DTRMMKERNEL    = dtrmm_kernel_4x2_vfp.S
+CTRMMKERNEL    = ctrmm_kernel_2x2_vfp.S
+ZTRMMKERNEL    = ztrmm_kernel_2x2_vfp.S
 

kernel/arm/KERNEL.ARMV7

@@ -1,91 +1,12 @@
-
-#################################################################################
-SAMAXKERNEL  = iamax_vfp.S
-DAMAXKERNEL  = iamax_vfp.S
-CAMAXKERNEL  = iamax_vfp.S
-ZAMAXKERNEL  = iamax_vfp.S
-
-SAMINKERNEL  = iamax_vfp.S
-DAMINKERNEL  = iamax_vfp.S
-CAMINKERNEL  = iamax_vfp.S
-ZAMINKERNEL  = iamax_vfp.S
-
-SMAXKERNEL   = iamax_vfp.S
-DMAXKERNEL   = iamax_vfp.S
-
-SMINKERNEL   = iamax_vfp.S
-DMINKERNEL   = iamax_vfp.S
-
-ISAMAXKERNEL = iamax_vfp.S
-IDAMAXKERNEL = iamax_vfp.S
-ICAMAXKERNEL = iamax_vfp.S
-IZAMAXKERNEL = iamax_vfp.S
-
-ISAMINKERNEL = iamax_vfp.S
-IDAMINKERNEL = iamax_vfp.S
-ICAMINKERNEL = iamax_vfp.S
-IZAMINKERNEL = iamax_vfp.S
-
-ISMAXKERNEL  = iamax_vfp.S
-IDMAXKERNEL  = iamax_vfp.S
-
-ISMINKERNEL  = iamax_vfp.S
-IDMINKERNEL  = iamax_vfp.S
-
-SSWAPKERNEL  = swap_vfp.S
-DSWAPKERNEL  = swap_vfp.S
-CSWAPKERNEL  = swap_vfp.S
-ZSWAPKERNEL  = swap_vfp.S
-
-SASUMKERNEL  = asum_vfp.S
-DASUMKERNEL  = asum_vfp.S
-CASUMKERNEL  = asum_vfp.S
-ZASUMKERNEL  = asum_vfp.S
-
-SAXPYKERNEL  = axpy_vfp.S
-DAXPYKERNEL  = axpy_vfp.S
-CAXPYKERNEL  = axpy_vfp.S
-ZAXPYKERNEL  = axpy_vfp.S
-
-SCOPYKERNEL  = copy.c
-DCOPYKERNEL  = copy.c
-CCOPYKERNEL  = zcopy.c
-ZCOPYKERNEL  = zcopy.c
-
-SDOTKERNEL   = sdot_vfp.S
-DDOTKERNEL   = ddot_vfp.S
-CDOTKERNEL   = cdot_vfp.S
-ZDOTKERNEL   = zdot_vfp.S
+include $(KERNELDIR)/KERNEL.ARMV6
 
 SNRM2KERNEL  = nrm2_vfpv3.S
 DNRM2KERNEL  = nrm2_vfpv3.S
 CNRM2KERNEL  = nrm2_vfpv3.S
 ZNRM2KERNEL  = nrm2_vfpv3.S
 
-SROTKERNEL   = rot_vfp.S
-DROTKERNEL   = rot_vfp.S
-CROTKERNEL   = rot_vfp.S
-ZROTKERNEL   = rot_vfp.S
-
-SSCALKERNEL  = scal.c
-DSCALKERNEL  = scal.c
-CSCALKERNEL  = zscal.c
-ZSCALKERNEL  = zscal.c
-
 SGEMVNKERNEL = gemv_n_vfpv3.S
 DGEMVNKERNEL = gemv_n_vfpv3.S
-CGEMVNKERNEL = cgemv_n_vfp.S
-ZGEMVNKERNEL = zgemv_n_vfp.S
-
-SGEMVTKERNEL = gemv_t_vfp.S
-DGEMVTKERNEL = gemv_t_vfp.S
-CGEMVTKERNEL = cgemv_t_vfp.S
-ZGEMVTKERNEL = zgemv_t_vfp.S
-
-STRMMKERNEL  =  strmm_kernel_4x4_vfpv3.S
-DTRMMKERNEL  =  dtrmm_kernel_4x4_vfpv3.S
-CTRMMKERNEL  =  ctrmm_kernel_2x2_vfpv3.S
-ZTRMMKERNEL  =  ztrmm_kernel_2x2_vfpv3.S
 
 SGEMMKERNEL    =  sgemm_kernel_4x4_vfpv3.S
 SGEMMONCOPY    =  sgemm_ncopy_4_vfp.S
@@ -100,35 +21,10 @@ DGEMMONCOPYOBJ = dgemm_oncopy.o
 DGEMMOTCOPYOBJ = dgemm_otcopy.o
 
 CGEMMKERNEL    = cgemm_kernel_2x2_vfpv3.S
-CGEMMONCOPY    = cgemm_ncopy_2_vfp.S
-CGEMMOTCOPY    = cgemm_tcopy_2_vfp.S
-CGEMMONCOPYOBJ = cgemm_oncopy.o
-CGEMMOTCOPYOBJ = cgemm_otcopy.o
-
 ZGEMMKERNEL    = zgemm_kernel_2x2_vfpv3.S
-ZGEMMONCOPY    = zgemm_ncopy_2_vfp.S
-ZGEMMOTCOPY    = zgemm_tcopy_2_vfp.S
-ZGEMMONCOPYOBJ = zgemm_oncopy.o
-ZGEMMOTCOPYOBJ = zgemm_otcopy.o
-
-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
 
+STRMMKERNEL  =  strmm_kernel_4x4_vfpv3.S
+DTRMMKERNEL  =  dtrmm_kernel_4x4_vfpv3.S
+CTRMMKERNEL  =  ctrmm_kernel_2x2_vfpv3.S
+ZTRMMKERNEL  =  ztrmm_kernel_2x2_vfpv3.S
 

kernel/arm/asum_vfp.S

@@ -475,6 +475,14 @@ asum_kernel_L999:
 	vadd.f32	s0 , s0, s1				// set return value
 #endif
 
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	r0, s0
+#else
+	vmov	r0, r1, d0
+#endif
+#endif
+
 	bx	lr
 
 	EPILOGUE

kernel/arm/axpy_vfp.S

@@ -38,18 +38,52 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#ifndef ARM_SOFTFP_ABI
-//hard	
-#define	OLD_INC_X	[fp, #0 ]
-#define	OLD_Y		[fp, #4 ]
-#define	OLD_INC_Y	[fp, #8 ]
-#else
+#if !defined(__ARM_PCS_VFP)
+
+#if !defined(COMPLEX)
+
+#if !defined(DOUBLE)
+#define	OLD_ALPHA	r3
 #define	OLD_X		[fp, #0 ]
 #define	OLD_INC_X	[fp, #4 ]
 #define	OLD_Y		[fp, #8 ]
 #define	OLD_INC_Y	[fp, #12 ]
+#else
+#define	OLD_ALPHA	[fp, #0]
+#define	OLD_X		[fp, #8 ]
+#define	OLD_INC_X	[fp, #12 ]
+#define	OLD_Y		[fp, #16 ]
+#define	OLD_INC_Y	[fp, #20 ]
 #endif
-	
+
+#else //COMPLEX
+
+#if !defined(DOUBLE)
+#define	OLD_ALPHAR	r3
+#define	OLD_ALPHAI	[fp, #0 ]
+#define	OLD_X		[fp, #4 ]
+#define	OLD_INC_X	[fp, #8 ]
+#define	OLD_Y		[fp, #12 ]
+#define	OLD_INC_Y	[fp, #16 ]
+#else
+#define	OLD_ALPHAR	[fp, #0]
+#define	OLD_ALPHAI	[fp, #8]
+#define	OLD_X		[fp, #16 ]
+#define	OLD_INC_X	[fp, #20 ]
+#define	OLD_Y		[fp, #24 ]
+#define	OLD_INC_Y	[fp, #28 ]
+#endif
+
+#endif //!defined(COMPLEX)
+
+#else //__ARM_PCS_VFP
+
+#define	OLD_INC_X	[fp, #0 ]
+#define	OLD_Y		[fp, #4 ]
+#define	OLD_INC_Y	[fp, #8 ]
+
+#endif //!defined(__ARM_PCS_VFP)
+
 #define	N	r0
 #define Y	r1
 #define	INC_X	r2
@@ -71,14 +105,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(DOUBLE)
 
 #define	FMAC_R1	fmacd
-#define FMAC_R2 fnmacd
+#define FMAC_R2 vmls.f64
 #define	FMAC_I1	fmacd
 #define FMAC_I2 fmacd
 
 #else
 
 #define	FMAC_R1	fmacs
-#define FMAC_R2 fnmacs
+#define FMAC_R2 vmls.f32
 #define	FMAC_I1	fmacs
 #define FMAC_I2 fmacs
 
@@ -90,14 +124,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define	FMAC_R1	fmacd
 #define FMAC_R2 fmacd
-#define	FMAC_I1	fnmacd
+#define	FMAC_I1	vmls.f64
 #define FMAC_I2 fmacd
 
 #else
 
 #define	FMAC_R1	fmacs
 #define FMAC_R2 fmacs
-#define	FMAC_I1	fnmacs
+#define	FMAC_I1	vmls.f32
 #define FMAC_I2 fmacs
 
 #endif
@@ -370,13 +404,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         add     fp, sp, #8
 	sub     sp, sp, #STACKSIZE                              // reserve stack
 
-#ifdef ARM_SOFTFP_ABI
-#ifndef DOUBLE
-	vmov	s0, r3     //move alpha to s0
+#if !defined(__ARM_PCS_VFP)
+#if !defined(COMPLEX)
+#if !defined(DOUBLE)
+	vmov	s0, OLD_ALPHA
 	ldr	X, OLD_X
+#else
+	vldr	d0, OLD_ALPHA
+	ldr	X, OLD_X
+#endif
+#else //COMPLEX
+#if !defined(DOUBLE)
+	vmov	s0, OLD_ALPHAR
+	vldr	s1, OLD_ALPHAI
+	ldr	X, OLD_X
+#else
+	vldr	d0, OLD_ALPHAR
+	vldr	d1, OLD_ALPHAI
+	ldr     X, OLD_X
+#endif
 #endif
 #endif
-	
+
 	ldr    INC_X , OLD_INC_X
 	ldr         Y, OLD_Y
 	ldr    INC_Y , OLD_INC_Y

kernel/arm/cdot_vfp.S

@@ -41,8 +41,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r0
 #define	X	r1
 #define	INC_X	r2
-#define	OLD_Y	r3
-
 
 /******************************************************
 * [fp, #-128] - [fp, #-64] is reserved
@@ -50,7 +48,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 * registers
 *******************************************************/
 
-#define OLD_INC_Y	[fp, #4 ]
+#if !defined(__ARM_PCS_VFP)
+#define	OLD_RETURN_ADDR	r0
+#define	OLD_N		r1
+#define	OLD_X		r2
+#define	OLD_INC_X	r3
+#define	OLD_Y		[fp, #0 ]
+#define	OLD_INC_Y	[fp, #4 ]
+#define	RETURN_ADDR	r8
+#else
+#define	OLD_Y		r3
+#define	OLD_INC_Y	[fp, #0 ]
+#endif
 
 #define I	r5
 #define Y	r6
@@ -179,7 +188,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	.align 5
 
 	push	{r4 - r9, fp}
-	add	fp, sp, #24
+	add	fp, sp, #28
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
 	sub	r4, fp, #128
@@ -191,8 +200,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmov	s2, s0
 	vmov	s3, s0
 
+#if !defined(__ARM_PCS_VFP)
+	mov	RETURN_ADDR, OLD_RETURN_ADDR
+	mov	N, OLD_N
+	mov	X, OLD_X
+	mov	INC_X, OLD_INC_X
+	ldr	Y, OLD_Y
+	ldr	INC_Y, OLD_INC_Y
+#else
 	mov	Y, OLD_Y
 	ldr	INC_Y, OLD_INC_Y
+#endif
 
 	cmp	N, #0
 	ble	cdot_kernel_L999
@@ -265,7 +283,6 @@ cdot_kernel_S10:
 
 
 cdot_kernel_L999:
-
 	sub	r3, fp, #128
 	vldm	r3, { s8 - s15}					// restore floating point registers
 
@@ -276,8 +293,11 @@ cdot_kernel_L999:
 	vadd.f32	s0 , s0, s2
 	vsub.f32	s1 , s1, s3
 #endif
+#if !defined(__ARM_PCS_VFP)
+	vstm	RETURN_ADDR, {s0 - s1}
+#endif
 
-	sub	sp, fp, #24
+	sub	sp, fp, #28
 	pop	{r4 - r9, fp}
 	bx	lr
 

kernel/arm/cgemm_kernel_2x2_vfp.S

@@ -64,9 +64,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	r3
+#define OLD_ALPHAI_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define B	[fp, #12 ]
+#define C	[fp, #16 ]
+#define OLD_LDC	[fp, #20 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -94,42 +103,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(NN) || defined(NT) || defined(TN) || defined(TT)
 
-	#define	KMAC_R	fnmacs
+	#define	KMAC_R	vmls.f32
 	#define	KMAC_I	fmacs
 
 	#define	FMAC_R1	fmacs
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmacs
 	#define	FMAC_I2	fmacs
 
 #elif defined(CN) || defined(CT)
 
 	#define	KMAC_R	fmacs
-	#define	KMAC_I	fnmacs
+	#define	KMAC_I	vmls.f32
 
 	#define	FMAC_R1	fmacs
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmacs
 	#define	FMAC_I2	fmacs
 
 #elif defined(NC) || defined(TC)
 
 	#define	KMAC_R	fmacs
-	#define	KMAC_I	fnmacs
+	#define	KMAC_I	vmls.f32
 
 	#define	FMAC_R1	fmacs
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmacs
+	#define	FMAC_I1	vmls.f32
 	#define	FMAC_I2	fmacs
 
 #else
 
-	#define	KMAC_R  fnmacs
+	#define	KMAC_R  vmls.f32
 	#define	KMAC_I	fmacs
 
 	#define	FMAC_R1	fmacs
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmacs
+	#define	FMAC_I1	vmls.f32
 	#define	FMAC_I2	fmacs
 
 #endif
@@ -816,6 +825,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/cgemm_kernel_2x2_vfpv3.S

@@ -80,9 +80,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	r3
+#define OLD_ALPHAI_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define B	[fp, #12 ]
+#define C	[fp, #16 ]
+#define OLD_LDC	[fp, #20 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -106,10 +115,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R	fsubs
 	#define	FADD_I	fadds
 
-	#define	FMAC_R1	fnmacs
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R1	vmls.f32
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmacs
-	#define	FMAC_I2	fnmacs
+	#define	FMAC_I2	vmls.f32
 
 #elif defined(CN) || defined(CT)
 
@@ -118,7 +127,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	#define	FMAC_R1	fmacs
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmacs
+	#define	FMAC_I1	vmls.f32
 	#define	FMAC_I2	fmacs
 
 #elif defined(NC) || defined(TC)
@@ -127,7 +136,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_I	fsubs
 
 	#define	FMAC_R1	fmacs
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmacs
 	#define	FMAC_I2	fmacs
 
@@ -136,10 +145,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R  fsubs
 	#define	FADD_I	fadds
 
-	#define	FMAC_R1	fnmacs
+	#define	FMAC_R1	vmls.f32
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmacs
-	#define	FMAC_I2	fnmacs
+	#define	FMAC_I1	vmls.f32
+	#define	FMAC_I2	vmls.f32
 
 #endif
 
@@ -873,6 +882,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/cgemv_n_vfp.S

@@ -38,11 +38,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR	r3
+#define OLD_ALPHAI	[fp, #0 ]
+#define OLD_A_SOFTFP	[fp, #4 ]
+#define OLD_LDA		[fp, #8 ]
+#define X		[fp, #12 ]
+#define OLD_INC_X	[fp, #16 ]
+#define Y		[fp, #20 ]
+#define OLD_INC_Y	[fp, #24 ]
+#else
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+#endif
+
 #define OLD_A		r3
 #define	OLD_M		r0
 
@@ -78,42 +90,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if !defined(CONJ) && !defined(XCONJ)
 
-        #define KMAC_R  fnmacs
+        #define KMAC_R  vmls.f32
         #define KMAC_I  fmacs
 
         #define FMAC_R1 fmacs
-        #define FMAC_R2 fnmacs
+        #define FMAC_R2 vmls.f32
         #define FMAC_I1 fmacs
         #define FMAC_I2 fmacs
 
 #elif defined(CONJ) && !defined(XCONJ)
 
         #define KMAC_R  fmacs
-        #define KMAC_I  fnmacs
+        #define KMAC_I  vmls.f32
 
         #define FMAC_R1 fmacs
-        #define FMAC_R2 fnmacs
+        #define FMAC_R2 vmls.f32
         #define FMAC_I1 fmacs
         #define FMAC_I2 fmacs
 
 #elif !defined(CONJ) && defined(XCONJ)
 
         #define KMAC_R  fmacs
-        #define KMAC_I  fnmacs
+        #define KMAC_I  vmls.f32
 
         #define FMAC_R1 fmacs
         #define FMAC_R2 fmacs
-        #define FMAC_I1 fnmacs
+        #define FMAC_I1 vmls.f32
         #define FMAC_I2 fmacs
 
 #else
 
-        #define KMAC_R  fnmacs
+        #define KMAC_R  vmls.f32
         #define KMAC_I  fmacs
 
         #define FMAC_R1 fmacs
         #define FMAC_R2 fmacs
-        #define FMAC_I1 fnmacs
+        #define FMAC_I1 vmls.f32
         #define FMAC_I2 fmacs
 
 #endif
@@ -462,6 +474,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	N, #0
 	ble	cgemvn_kernel_L999
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	s0, OLD_ALPHAR
+	vldr	s1, OLD_ALPHAI
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_A, A
 	str	OLD_M, M
 	vstr    s0 , ALPHA_R

kernel/arm/cgemv_t_vfp.S

@@ -38,11 +38,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR	r3
+#define OLD_ALPHAI	[fp, #0 ]
+#define OLD_A_SOFTFP	[fp, #4 ]
+#define OLD_LDA		[fp, #8 ]
+#define X		[fp, #12 ]
+#define OLD_INC_X	[fp, #16 ]
+#define Y		[fp, #20 ]
+#define OLD_INC_Y	[fp, #24 ]
+#else
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+#endif
+
 #define OLD_A		r3
 #define	OLD_N		r1
 
@@ -76,42 +88,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if !defined(CONJ) && !defined(XCONJ)
 
-        #define KMAC_R  fnmacs
+        #define KMAC_R  vmls.f32
         #define KMAC_I  fmacs
 
         #define FMAC_R1 fmacs
-        #define FMAC_R2 fnmacs
+        #define FMAC_R2 vmls.f32
         #define FMAC_I1 fmacs
         #define FMAC_I2 fmacs
 
 #elif defined(CONJ) && !defined(XCONJ)
 
         #define KMAC_R  fmacs
-        #define KMAC_I  fnmacs
+        #define KMAC_I  vmls.f32
 
         #define FMAC_R1 fmacs
-        #define FMAC_R2 fnmacs
+        #define FMAC_R2 vmls.f32
         #define FMAC_I1 fmacs
         #define FMAC_I2 fmacs
 
 #elif !defined(CONJ) && defined(XCONJ)
 
         #define KMAC_R  fmacs
-        #define KMAC_I  fnmacs
+        #define KMAC_I  vmls.f32
 
         #define FMAC_R1 fmacs
         #define FMAC_R2 fmacs
-        #define FMAC_I1 fnmacs
+        #define FMAC_I1 vmls.f32
         #define FMAC_I2 fmacs
 
 #else
 
-        #define KMAC_R  fnmacs
+        #define KMAC_R  vmls.f32
         #define KMAC_I  fmacs
 
         #define FMAC_R1 fmacs
         #define FMAC_R2 fmacs
-        #define FMAC_I1 fnmacs
+        #define FMAC_I1 vmls.f32
         #define FMAC_I2 fmacs
 
 #endif
@@ -359,6 +371,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	OLD_N, #0
 	ble	cgemvt_kernel_L999
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	s0, OLD_ALPHAR
+	vldr	s1, OLD_ALPHAI
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_A, A
 	str	OLD_N, N
 

kernel/arm/ctrmm_kernel_2x2_vfp.S

@@ -67,10 +67,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	r3
+#define OLD_ALPHAI_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define B	[fp, #12 ]
+#define C	[fp, #16 ]
+#define OLD_LDC	[fp, #20 ]
+#define OFFSET  [fp, #24 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -98,42 +108,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(NN) || defined(NT) || defined(TN) || defined(TT)
 
-	#define	KMAC_R	fnmacs
+	#define	KMAC_R	vmls.f32
 	#define	KMAC_I	fmacs
 
 	#define	FMAC_R1	fmacs
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmacs
 	#define	FMAC_I2	fmacs
 
 #elif defined(CN) || defined(CT)
 
 	#define	KMAC_R	fmacs
-	#define	KMAC_I	fnmacs
+	#define	KMAC_I	vmls.f32
 
 	#define	FMAC_R1	fmacs
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmacs
 	#define	FMAC_I2	fmacs
 
 #elif defined(NC) || defined(TC)
 
 	#define	KMAC_R	fmacs
-	#define	KMAC_I	fnmacs
+	#define	KMAC_I	vmls.f32
 
 	#define	FMAC_R1	fmacs
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmacs
+	#define	FMAC_I1	vmls.f32
 	#define	FMAC_I2	fmacs
 
 #else
 
-	#define	KMAC_R  fnmacs
+	#define	KMAC_R  vmls.f32
 	#define	KMAC_I	fmacs
 
 	#define	FMAC_R1	fmacs
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmacs
+	#define	FMAC_I1	vmls.f32
 	#define	FMAC_I2	fmacs
 
 #endif
@@ -826,6 +836,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/ctrmm_kernel_2x2_vfpv3.S

@@ -66,10 +66,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	r3
+#define OLD_ALPHAI_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define B	[fp, #12 ]
+#define C	[fp, #16 ]
+#define OLD_LDC	[fp, #20 ]
+#define OFFSET  [fp, #24 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -93,10 +103,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R	fsubs
 	#define	FADD_I	fadds
 
-	#define	FMAC_R1	fnmuls
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R1	vnmul.f32
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmuls
-	#define	FMAC_I2	fnmacs
+	#define	FMAC_I2	vmls.f32
 
 #elif defined(CN) || defined(CT)
 
@@ -105,7 +115,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	#define	FMAC_R1	fmuls
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmuls
+	#define	FMAC_I1	vnmul.f32
 	#define	FMAC_I2	fmacs
 
 #elif defined(NC) || defined(TC)
@@ -114,7 +124,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_I	fsubs
 
 	#define	FMAC_R1	fmuls
-	#define	FMAC_R2	fnmacs
+	#define	FMAC_R2	vmls.f32
 	#define	FMAC_I1	fmuls
 	#define	FMAC_I2	fmacs
 
@@ -123,10 +133,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R  fsubs
 	#define	FADD_I	fadds
 
-	#define	FMAC_R1	fnmuls
+	#define	FMAC_R1	vnmul.f32
 	#define	FMAC_R2	fmacs
-	#define	FMAC_I1	fnmuls
-	#define	FMAC_I2	fnmacs
+	#define	FMAC_I1	vnmul.f32
+	#define	FMAC_I2	vmls.f32
 
 #endif
 
@@ -846,6 +856,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/ddot_vfp.S

@@ -246,6 +246,9 @@ ddot_kernel_L999:
 	vldm	r3, { d8 - d15}					// restore floating point registers
 
 	vadd.f64	d0 , d0, d1				// set return value
+#if !defined(__ARM_PCS_VFP)
+	vmov	r0, r1, d0
+#endif
 	sub	sp, fp, #24
 	pop	{r4 - r9, fp}
 	bx	lr

kernel/arm/dgemm_kernel_4x2_vfp.S

@@ -62,10 +62,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-280]
 
-
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #12 ]
+#define B	[fp, #16 ]
+#define C	[fp, #20 ]
+#define OLD_LDC	[fp, #24 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -429,6 +436,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/dgemm_kernel_4x4_vfpv3.S

@@ -79,9 +79,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #12 ]
+#define B	[fp, #16 ]
+#define C	[fp, #20 ]
+#define OLD_LDC	[fp, #24 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -878,6 +886,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/dtrmm_kernel_4x2_vfp.S

@@ -65,10 +65,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-276 ]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #12 ]
+#define B	[fp, #16 ]
+#define OLD_C	[fp, #20 ]
+#define OLD_LDC	[fp, #24 ]
+#define OFFSET  [fp, #28 ]
+#else
 #define B	[fp, #4 ]
 #define OLD_C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -404,6 +413,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/dtrmm_kernel_4x4_vfpv3.S

@@ -66,10 +66,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-276 ]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	[fp, #4]
+#define OLD_A_SOFTFP	[fp, #12 ]
+#define B	[fp, #16 ]
+#define OLD_C	[fp, #20 ]
+#define OLD_LDC	[fp, #24 ]
+#define OFFSET  [fp, #28 ]
+#else
 #define B	[fp, #4 ]
 #define OLD_C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -846,6 +855,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/gemv_n_vfp.S

@@ -38,11 +38,36 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
+#if !defined(__ARM_PCS_VFP)
+
+#if !defined(DOUBLE)
+#define OLD_ALPHA	r3
+#define OLD_A_SOFTFP	[fp, #0 ]
+#define OLD_LDA		[fp, #4 ]
+#define X		[fp, #8 ]
+#define OLD_INC_X	[fp, #12 ]
+#define Y		[fp, #16 ]
+#define OLD_INC_Y	[fp, #20 ]
+#else
+#define OLD_ALPHA	[fp, #0 ]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define OLD_LDA		[fp, #12]
+#define X		[fp, #16]
+#define OLD_INC_X	[fp, #20]
+#define Y		[fp, #24]
+#define OLD_INC_Y	[fp, #28]
+#endif
+
+#else
+
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+
+#endif
+
 #define OLD_A		r3
 #define	OLD_M		r0
 
@@ -508,6 +533,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	N, #0
 	ble	gemvn_kernel_L999
 
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	s0, OLD_ALPHA
+#else
+	vldr	d0, OLD_ALPHA
+#endif
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_A, A
 	str	OLD_M, M
 

kernel/arm/gemv_n_vfpv3.S

@@ -38,25 +38,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#ifndef ARM_SOFTFP_ABI
-//hard
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
-#define OLD_A		r3
-#else
-#define OLD_A_SOFTFP	[fp, #0 ]
-#define	OLD_LDA		[fp, #4 ]
-#define	X		[fp, #8 ]
-#define	OLD_INC_X	[fp, #12 ]
-#define	Y		[fp, #16 ]
-#define	OLD_INC_Y	[fp, #20 ]
+#if !defined(__ARM_PCS_VFP)
+
+#if !defined(DOUBLE)
 #define OLD_ALPHA	r3
-#define OLD_A		r3
+#define OLD_A_SOFTFP	[fp, #0 ]
+#define OLD_LDA		[fp, #4 ]
+#define X		[fp, #8 ]
+#define OLD_INC_X	[fp, #12 ]
+#define Y		[fp, #16 ]
+#define OLD_INC_Y	[fp, #20 ]
+#else
+#define OLD_ALPHA	[fp, #0 ]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define OLD_LDA		[fp, #12]
+#define X		[fp, #16]
+#define OLD_INC_X	[fp, #20]
+#define Y		[fp, #24]
+#define OLD_INC_Y	[fp, #28]
+#endif
+
+#else
+
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+
 #endif
 
+#define OLD_A		r3
 #define	OLD_M		r0
 
 #define AO1	r0
@@ -565,18 +577,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	N, #0
 	ble	gemvn_kernel_L999
 
-#ifndef DOUBLE
-#ifdef ARM_SOFTFP_ABI
-
-	vmov	s0, 	OLD_ALPHA
-	ldr	OLD_A,	OLD_A_SOFTFP
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	s0, OLD_ALPHA
+#else
+	vldr	d0, OLD_ALPHA
 #endif
+	ldr	OLD_A, OLD_A_SOFTFP
 #endif
 
 	str	OLD_A, A
 	str	OLD_M, M
-	
-	
+
 	ldr    INC_X , OLD_INC_X
 	ldr    INC_Y , OLD_INC_Y
 

kernel/arm/gemv_t_vfp.S

@@ -38,25 +38,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#ifndef ARM_SOFTFP_ABI
-//hard abi	
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
-#define OLD_A		r3
-#else
-#define OLD_A_SOFTFP	[fp, #0 ]
-#define	OLD_LDA		[fp, #4 ]
-#define	X		[fp, #8 ]
-#define	OLD_INC_X	[fp, #12 ]
-#define	Y		[fp, #16 ]
-#define	OLD_INC_Y	[fp, #20 ]
+#if !defined(__ARM_PCS_VFP)
+
+#if !defined(DOUBLE)
 #define OLD_ALPHA	r3
-#define OLD_A		r3
+#define OLD_A_SOFTFP	[fp, #0 ]
+#define OLD_LDA		[fp, #4 ]
+#define X		[fp, #8 ]
+#define OLD_INC_X	[fp, #12 ]
+#define Y		[fp, #16 ]
+#define OLD_INC_Y	[fp, #20 ]
+#else
+#define OLD_ALPHA	[fp, #0 ]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define OLD_LDA		[fp, #12]
+#define X		[fp, #16]
+#define OLD_INC_X	[fp, #20]
+#define Y		[fp, #24]
+#define OLD_INC_Y	[fp, #28]
 #endif
 
+#else
+
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+
+#endif
+
+#define OLD_A		r3
 #define	OLD_N		r1
 
 #define M	r0
@@ -518,11 +530,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	OLD_N, #0
 	ble	gemvt_kernel_L999
 
-#ifndef DOUBLE
-#ifdef ARM_SOFTFP_ABI
-	vmov	s0,	OLD_ALPHA
-	ldr	OLD_A,	OLD_A_SOFTFP
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	s0, OLD_ALPHA
+#else
+	vldr	d0, OLD_ALPHA
 #endif
+	ldr	OLD_A, OLD_A_SOFTFP
 #endif
 
 	str	OLD_A, A

kernel/arm/gemv_t_vfpv3.S

@@ -38,11 +38,36 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
+#if !defined(__ARM_PCS_VFP)
+
+#if !defined(DOUBLE)
+#define OLD_ALPHA	r3
+#define OLD_A_SOFTFP	[fp, #0 ]
+#define OLD_LDA		[fp, #4 ]
+#define X		[fp, #8 ]
+#define OLD_INC_X	[fp, #12 ]
+#define Y		[fp, #16 ]
+#define OLD_INC_Y	[fp, #20 ]
+#else
+#define OLD_ALPHA	[fp, #0 ]
+#define OLD_A_SOFTFP	[fp, #8 ]
+#define OLD_LDA		[fp, #12]
+#define X		[fp, #16]
+#define OLD_INC_X	[fp, #20]
+#define Y		[fp, #24]
+#define OLD_INC_Y	[fp, #28]
+#endif
+
+#else
+
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+
+#endif
+
 #define OLD_A		r3
 #define	OLD_N		r1
 
@@ -476,6 +501,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	OLD_N, #0
 	ble	gemvt_kernel_L999
 
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	s0, OLD_ALPHA
+#else
+	vldr	d0, OLD_ALPHA
+#endif
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_A, A
 	str	OLD_N, N
 

kernel/arm/nrm2_vfp.S

@@ -573,6 +573,13 @@ nrm2_kernel_L999:
 #else
 	vsqrt.f32	s1, s1
 	vmul.f32	s0, s0, s1
+#endif
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	r0, s0
+#else
+	vmov	r0, r1, d0
+#endif
 #endif
 
 	bx	lr

kernel/arm/nrm2_vfpv3.S

@@ -503,8 +503,13 @@ nrm2_kernel_L999:
 #else
 	vsqrt.f32	s1, s1
 	vmul.f32	s0, s0, s1
-#ifdef ARM_SOFTFP_ABI
-	vmov		r0, s0
+#endif
+
+#if !defined(__ARM_PCS_VFP)
+#if defined(DOUBLE)
+	vmov	r0, r1, d0
+#else
+	vmov	r0, s0
 #endif
 #endif
 

kernel/arm/rot_vfp.S

@@ -40,6 +40,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define	OLD_INC_Y	[fp, #0 ]
 
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+#define	OLD_C		[fp, #4]
+#define	OLD_S		[fp, #8]
+#else
+#define	OLD_C		[fp, #8]
+#define	OLD_S		[fp, #16]
+#endif
+#endif
 
 #define	N	r0
 #define X	r1
@@ -73,7 +82,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d5
 	vmul.f64    d3 , d0, d5
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -82,7 +91,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d5
 	vmul.f64    d3 , d0, d5
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -91,7 +100,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d5
 	vmul.f64    d3 , d0, d5
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -100,7 +109,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d5
 	vmul.f64    d3 , d0, d5
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -114,7 +123,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d5
 	vmul.f64    d3 , d0, d5
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -127,7 +136,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d5
 	vmul.f64    d3 , d0, d5
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X, { d2 }
 	fstmiad	Y, { d3 }
 
@@ -145,7 +154,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s5
 	vmul.f32    s3 , s0, s5
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -154,7 +163,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s5
 	vmul.f32    s3 , s0, s5
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -163,7 +172,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s5
 	vmul.f32    s3 , s0, s5
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -172,7 +181,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s5
 	vmul.f32    s3 , s0, s5
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -186,7 +195,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s5
 	vmul.f32    s3 , s0, s5
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -199,7 +208,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s5
 	vmul.f32    s3 , s0, s5
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X, { s2 }
 	fstmias	Y, { s3 }
 
@@ -226,13 +235,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d6
 	vmul.f64    d3 , d0, d6
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 	vmul.f64    d2 , d0, d5
 	fmacd       d2 , d1, d7
 	vmul.f64    d3 , d0, d7
-	fnmacd      d3 , d1, d5
+	vmls.f64    d3 , d1, d5
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -241,13 +250,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d6
 	vmul.f64    d3 , d0, d6
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 	vmul.f64    d2 , d0, d5
 	fmacd       d2 , d1, d7
 	vmul.f64    d3 , d0, d7
-	fnmacd      d3 , d1, d5
+	vmls.f64    d3 , d1, d5
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -259,13 +268,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d6
 	vmul.f64    d3 , d0, d6
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 	vmul.f64    d2 , d0, d5
 	fmacd       d2 , d1, d7
 	vmul.f64    d3 , d0, d7
-	fnmacd      d3 , d1, d5
+	vmls.f64    d3 , d1, d5
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -274,13 +283,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d6
 	vmul.f64    d3 , d0, d6
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 	vmul.f64    d2 , d0, d5
 	fmacd       d2 , d1, d7
 	vmul.f64    d3 , d0, d7
-	fnmacd      d3 , d1, d5
+	vmls.f64    d3 , d1, d5
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -294,13 +303,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d6
 	vmul.f64    d3 , d0, d6
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 	vmul.f64    d2 , d0, d5
 	fmacd       d2 , d1, d7
 	vmul.f64    d3 , d0, d7
-	fnmacd      d3 , d1, d5
+	vmls.f64    d3 , d1, d5
 	fstmiad	X!, { d2 }
 	fstmiad	Y!, { d3 }
 
@@ -314,13 +323,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f64    d2 , d0, d4
 	fmacd       d2 , d1, d6
 	vmul.f64    d3 , d0, d6
-	fnmacd      d3 , d1, d4
+	vmls.f64    d3 , d1, d4
 	vstr	    d2 , [ X, #0 ]
 	vstr	    d3 , [ Y, #0 ]
 	vmul.f64    d2 , d0, d5
 	fmacd       d2 , d1, d7
 	vmul.f64    d3 , d0, d7
-	fnmacd      d3 , d1, d5
+	vmls.f64    d3 , d1, d5
 	vstr	    d2 , [ X, #8 ]
 	vstr	    d3 , [ Y, #8 ]
 
@@ -343,13 +352,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s6
 	vmul.f32    s3 , s0, s6
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 	vmul.f32    s2 , s0, s5
 	fmacs       s2 , s1, s7
 	vmul.f32    s3 , s0, s7
-	fnmacs      s3 , s1, s5
+	vmls.f32    s3 , s1, s5
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -358,13 +367,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s6
 	vmul.f32    s3 , s0, s6
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 	vmul.f32    s2 , s0, s5
 	fmacs       s2 , s1, s7
 	vmul.f32    s3 , s0, s7
-	fnmacs      s3 , s1, s5
+	vmls.f32    s3 , s1, s5
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -376,13 +385,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s6
 	vmul.f32    s3 , s0, s6
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 	vmul.f32    s2 , s0, s5
 	fmacs       s2 , s1, s7
 	vmul.f32    s3 , s0, s7
-	fnmacs      s3 , s1, s5
+	vmls.f32    s3 , s1, s5
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -391,13 +400,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s6
 	vmul.f32    s3 , s0, s6
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 	vmul.f32    s2 , s0, s5
 	fmacs       s2 , s1, s7
 	vmul.f32    s3 , s0, s7
-	fnmacs      s3 , s1, s5
+	vmls.f32    s3 , s1, s5
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -411,13 +420,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s6
 	vmul.f32    s3 , s0, s6
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 	vmul.f32    s2 , s0, s5
 	fmacs       s2 , s1, s7
 	vmul.f32    s3 , s0, s7
-	fnmacs      s3 , s1, s5
+	vmls.f32    s3 , s1, s5
 	fstmias	X!, { s2 }
 	fstmias	Y!, { s3 }
 
@@ -431,13 +440,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmul.f32    s2 , s0, s4
 	fmacs       s2 , s1, s6
 	vmul.f32    s3 , s0, s6
-	fnmacs      s3 , s1, s4
+	vmls.f32    s3 , s1, s4
 	vstr	    s2 , [ X, #0 ]
 	vstr	    s3 , [ Y, #0 ]
 	vmul.f32    s2 , s0, s5
 	fmacs       s2 , s1, s7
 	vmul.f32    s3 , s0, s7
-	fnmacs      s3 , s1, s5
+	vmls.f32    s3 , s1, s5
 	vstr	    s2 , [ X, #4 ]
 	vstr	    s3 , [ Y, #4 ]
 
@@ -462,7 +471,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         add     fp, sp, #8
 
 	ldr    INC_Y , OLD_INC_Y
-
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vldr	s0, OLD_C
+	vldr	s1, OLD_S
+#else
+	vldr	d0, OLD_C
+	vldr	d1, OLD_S
+#endif
+#endif
 
 	cmp	N, #0
 	ble	rot_kernel_L999

kernel/arm/scal_vfp.S

@@ -138,14 +138,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmiad	X,  { d4 - d5 }
 	vmul.f64    d2, d0, d4
-	fnmacd     d2, d1, d5
+	vmls.f64    d2, d1, d5
 	vmul.f64    d3, d0, d5
 	fmacd      d3, d1, d4
 	fstmiad	X!, { d2 - d3 }
 
 	fldmiad	X,  { d4 - d5 }
 	vmul.f64    d2, d0, d4
-	fnmacd     d2, d1, d5
+	vmls.f64    d2, d1, d5
 	vmul.f64    d3, d0, d5
 	fmacd      d3, d1, d4
 	fstmiad	X!, { d2 - d3 }
@@ -154,14 +154,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmiad	X,  { d4 - d5 }
 	vmul.f64    d2, d0, d4
-	fnmacd     d2, d1, d5
+	vmls.f64    d2, d1, d5
 	vmul.f64    d3, d0, d5
 	fmacd      d3, d1, d4
 	fstmiad	X!, { d2 - d3 }
 
 	fldmiad	X,  { d4 - d5 }
 	vmul.f64    d2, d0, d4
-	fnmacd     d2, d1, d5
+	vmls.f64    d2, d1, d5
 	vmul.f64    d3, d0, d5
 	fmacd      d3, d1, d4
 	fstmiad	X!, { d2 - d3 }
@@ -173,7 +173,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmiad	X,  { d4 - d5 }
 	vmul.f64    d2, d0, d4
-	fnmacd     d2, d1, d5
+	vmls.f64    d2, d1, d5
 	vmul.f64    d3, d0, d5
 	fmacd      d3, d1, d4
 	fstmiad	X!, { d2 - d3 }
@@ -184,7 +184,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmiad	X,  { d4 - d5 }
 	vmul.f64    d2, d0, d4
-	fnmacd     d2, d1, d5
+	vmls.f64    d2, d1, d5
 	vmul.f64    d3, d0, d5
 	fmacd      d3, d1, d4
 	fstmiad	X, { d2 - d3 }
@@ -201,28 +201,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmias	X,  { s4 - s5 }
 	vmul.f32    s2, s0, s4
-	fnmacs     s2, s1, s5
+	vmls.f32    s2, s1, s5
 	vmul.f32    s3, s0, s5
 	fmacs      s3, s1, s4
 	fstmias	X!, { s2 - s3 }
 
 	fldmias	X,  { s4 - s5 }
 	vmul.f32    s2, s0, s4
-	fnmacs     s2, s1, s5
+	vmls.f32    s2, s1, s5
 	vmul.f32    s3, s0, s5
 	fmacs      s3, s1, s4
 	fstmias	X!, { s2 - s3 }
 
 	fldmias	X,  { s4 - s5 }
 	vmul.f32    s2, s0, s4
-	fnmacs     s2, s1, s5
+	vmls.f32    s2, s1, s5
 	vmul.f32    s3, s0, s5
 	fmacs      s3, s1, s4
 	fstmias	X!, { s2 - s3 }
 
 	fldmias	X,  { s4 - s5 }
 	vmul.f32    s2, s0, s4
-	fnmacs     s2, s1, s5
+	vmls.f32    s2, s1, s5
 	vmul.f32    s3, s0, s5
 	fmacs      s3, s1, s4
 	fstmias	X!, { s2 - s3 }
@@ -234,7 +234,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmias	X,  { s4 - s5 }
 	vmul.f32    s2, s0, s4
-	fnmacs     s2, s1, s5
+	vmls.f32    s2, s1, s5
 	vmul.f32    s3, s0, s5
 	fmacs      s3, s1, s4
 	fstmias	X!, { s2 - s3 }
@@ -245,7 +245,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	fldmias	X,  { s4 - s5 }
 	vmul.f32    s2, s0, s4
-	fnmacs     s2, s1, s5
+	vmls.f32    s2, s1, s5
 	vmul.f32    s3, s0, s5
 	fmacs      s3, s1, s4
 	fstmias	X, { s2 - s3 }

kernel/arm/sdot_vfp.S

@@ -329,20 +329,19 @@ sdot_kernel_L999:
 	vldm	r3, { s8 - s15}					// restore floating point registers
 
 #if	defined(DSDOT)
-
 	vadd.f64	d0 , d0, d1				// set return value
-
-#ifdef ARM_SOFTFP_ABI
-	vmov	r0, r1, d0
+#else
+	vadd.f32	s0 , s0, s1				// set return value
 #endif
 
+#if !defined(__ARM_PCS_VFP)
+#if defined(DSDOT)
+	vmov	r0, r1, d0
 #else
-
-	vadd.f32	s0 , s0, s1				// set return value
-#ifdef ARM_SOFTFP_ABI
 	vmov	r0, s0
 #endif
 #endif
+
 	sub	sp, fp, #24
 	pop	{r4 - r9, fp}
 	bx	lr

kernel/arm/sgemm_kernel_4x2_vfp.S

@@ -62,9 +62,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	r3
+#define OLD_A_SOFTFP	[fp, #4 ]
+#define B	[fp, #8 ]
+#define C	[fp, #12 ]
+#define OLD_LDC	[fp, #16 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -416,6 +424,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/sgemm_kernel_4x4_vfpv3.S

@@ -58,14 +58,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	OLD_M	r0
 #define	OLD_N	r1
 #define	OLD_K	r2
-
-#ifdef ARM_SOFTFP_ABI
-#define OLD_ALPHA r3
-//#define OLD_A	
-#else //hard
 #define	OLD_A	r3
 #define OLD_ALPHA s0
-#endif
 
 /******************************************************
 * [fp, #-128] - [fp, #-64] is reserved
@@ -77,10 +71,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define M	[fp, #-256 ]
 #define N	[fp, #-260 ]
 #define K	[fp, #-264 ]
-
-#ifndef ARM_SOFTFP_ABI
 #define A	[fp, #-268 ]
-#endif
 
 #define FP_ZERO [fp, #-240]
 #define FP_ZERO_0 [fp, #-240]
@@ -88,17 +79,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA   [fp, #-280]
 
-#ifdef ARM_SOFTFP_ABI
-#define A	[fp, #4 ]
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	r3
+#define OLD_A_SOFTFP	[fp, #4 ]
 #define B	[fp, #8 ]
 #define C	[fp, #12 ]
 #define OLD_LDC	[fp, #16 ]
-#else //hard
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #endif
-	
+
 #define I	r0
 #define J	r1
 #define L	r2
@@ -867,16 +859,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K
-	
-#ifdef ARM_SOFTFP_ABI
-	str	OLD_ALPHA, ALPHA
-#else //hard
 	str	OLD_A, A
 	vstr	OLD_ALPHA, ALPHA
-#endif
+
 	sub	r3, fp, #128
 	vstm	r3, { s8 - s31} 				// store floating point registers
 

kernel/arm/strmm_kernel_4x2_vfp.S

@@ -65,10 +65,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-276 ]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	r3
+#define OLD_A_SOFTFP	[fp, #4 ]
+#define B	[fp, #8 ]
+#define OLD_C	[fp, #12 ]
+#define OLD_LDC	[fp, #16 ]
+#define OFFSET  [fp, #20 ]
+#else
 #define B	[fp, #4 ]
 #define OLD_C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -395,6 +404,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/strmm_kernel_4x4_vfpv3.S

@@ -64,10 +64,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHA_SOFTFP	r3
+#define OLD_A_SOFTFP	[fp, #4 ]
+#define B	[fp, #8 ]
+#define C	[fp, #12 ]
+#define OLD_LDC	[fp, #16 ]
+#define OFFSET  [fp, #20 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -782,6 +791,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vmov	OLD_ALPHA, OLD_ALPHA_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/swap_vfp.S

@@ -38,9 +38,43 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
+#if !defined(__ARM_PCS_VFP)
+
+#if !defined(COMPLEX)
+
+#if !defined(DOUBLE)
+#define OLD_X		[fp, #0 ]
+#define	OLD_INC_X	[fp, #4 ]
+#define	OLD_Y		[fp, #8 ]
+#define	OLD_INC_Y	[fp, #12 ]
+#else
+#define OLD_X		[fp, #8 ]
+#define	OLD_INC_X	[fp, #12]
+#define	OLD_Y		[fp, #16]
+#define	OLD_INC_Y	[fp, #20]
+#endif
+
+#else //COMPLEX
+
+#if !defined(DOUBLE)
+#define OLD_X		[fp, #4 ]
+#define	OLD_INC_X	[fp, #8 ]
+#define	OLD_Y		[fp, #12 ]
+#define	OLD_INC_Y	[fp, #16 ]
+#else
+#define OLD_X		[fp, #16]
+#define	OLD_INC_X	[fp, #20]
+#define	OLD_Y		[fp, #24]
+#define	OLD_INC_Y	[fp, #28]
+#endif
+
+#endif // !defined(__ARM_PCS_VFP)
+
+#else
 #define	OLD_INC_X	[fp, #0 ]
 #define	OLD_Y		[fp, #4 ]
 #define	OLD_INC_Y	[fp, #8 ]
+#endif
 
 
 #define	N	r0
@@ -229,6 +263,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	push    {r4 , fp}
         add     fp, sp, #8
 
+#if !defined(__ARM_PCS_VFP)
+	ldr	X, OLD_X
+#endif
 	ldr    INC_X , OLD_INC_X
 	ldr         Y, OLD_Y
 	ldr    INC_Y , OLD_INC_Y

kernel/arm/zdot_vfp.S

@@ -41,8 +41,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r0
 #define	X	r1
 #define	INC_X	r2
-#define	OLD_Y	r3
-
 
 /******************************************************
 * [fp, #-128] - [fp, #-64] is reserved
@@ -50,7 +48,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 * registers
 *******************************************************/
 
-#define OLD_INC_Y	[fp, #4 ]
+#if !defined(__ARM_PCS_VFP)
+#define	OLD_RETURN_ADDR	r0
+#define	OLD_N		r1
+#define	OLD_X		r2
+#define	OLD_INC_X	r3
+#define	OLD_Y		[fp, #0 ]
+#define	OLD_INC_Y	[fp, #4 ]
+#define	RETURN_ADDR	r8
+#else
+#define	OLD_Y		r3
+#define	OLD_INC_Y	[fp, #0 ]
+#endif
 
 #define I	r5
 #define Y	r6
@@ -181,7 +190,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	.align 5
 
 	push	{r4 - r9, fp}
-	add	fp, sp, #24
+	add	fp, sp, #28
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
 	sub	r4, fp, #128
@@ -194,9 +203,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         vcvt.f64.f32    d2, s0
         vcvt.f64.f32    d3, s0
 
+#if !defined(__ARM_PCS_VFP)
+	mov	RETURN_ADDR, OLD_RETURN_ADDR
+	mov	N, OLD_N
+	mov	X, OLD_X
+	mov	INC_X, OLD_INC_X
+	ldr	Y, OLD_Y
+	ldr	INC_Y, OLD_INC_Y
+#else
 	mov	Y, OLD_Y
 	ldr	INC_Y, OLD_INC_Y
-
+#endif
 
 	cmp	N, #0
 	ble	zdot_kernel_L999
@@ -280,8 +297,11 @@ zdot_kernel_L999:
 	vadd.f64	d0 , d0, d2
 	vsub.f64	d1 , d1, d3
 #endif
+#if !defined(__ARM_PCS_VFP)
+	vstm	RETURN_ADDR, {d0 - d1}
+#endif
 
-	sub	sp, fp, #24
+	sub	sp, fp, #28
 	pop	{r4 - r9, fp}
 	bx	lr
 

kernel/arm/zgemm_kernel_2x2_vfp.S

@@ -64,9 +64,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	[fp, #4]
+#define OLD_ALPHAI_SOFTFP	[fp, #12]
+#define OLD_A_SOFTFP	[fp, #20 ]
+#define B	[fp, #24 ]
+#define C	[fp, #28 ]
+#define OLD_LDC	[fp, #32 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -87,42 +96,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(NN) || defined(NT) || defined(TN) || defined(TT)
 
-	#define	KMAC_R	fnmacd
+	#define	KMAC_R	vmls.f64
 	#define	KMAC_I	fmacd
 
 	#define	FMAC_R1	fmacd
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmacd
 	#define	FMAC_I2	fmacd
 
 #elif defined(CN) || defined(CT)
 
 	#define	KMAC_R	fmacd
-	#define	KMAC_I	fnmacd
+	#define	KMAC_I	vmls.f64
 
 	#define	FMAC_R1	fmacd
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmacd
 	#define	FMAC_I2	fmacd
 
 #elif defined(NC) || defined(TC)
 
 	#define	KMAC_R	fmacd
-	#define	KMAC_I	fnmacd
+	#define	KMAC_I	vmls.f64
 
 	#define	FMAC_R1	fmacd
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmacd
+	#define	FMAC_I1	vmls.f64
 	#define	FMAC_I2	fmacd
 
 #else
 
-	#define	KMAC_R  fnmacd
+	#define	KMAC_R  vmls.f64
 	#define	KMAC_I	fmacd
 
 	#define	FMAC_R1	fmacd
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmacd
+	#define	FMAC_I1	vmls.f64
 	#define	FMAC_I2	fmacd
 
 #endif
@@ -863,6 +872,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/zgemm_kernel_2x2_vfpv3.S

@@ -80,9 +80,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	[fp, #4]
+#define OLD_ALPHAI_SOFTFP	[fp, #12]
+#define OLD_A_SOFTFP	[fp, #20 ]
+#define B	[fp, #24 ]
+#define C	[fp, #28 ]
+#define OLD_LDC	[fp, #32 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -106,10 +115,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R	fsubd
 	#define	FADD_I	faddd
 
-	#define	FMAC_R1	fnmacd
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R1	vmls.f64
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmacd
-	#define	FMAC_I2	fnmacd
+	#define	FMAC_I2	vmls.f64
 
 #elif defined(CN) || defined(CT)
 
@@ -118,7 +127,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	#define	FMAC_R1	fmacd
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmacd
+	#define	FMAC_I1	vmls.f64
 	#define	FMAC_I2	fmacd
 
 #elif defined(NC) || defined(TC)
@@ -127,7 +136,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_I	fsubd
 
 	#define	FMAC_R1	fmacd
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmacd
 	#define	FMAC_I2	fmacd
 
@@ -136,10 +145,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R  fsubd
 	#define	FADD_I	faddd
 
-	#define	FMAC_R1	fnmacd
+	#define	FMAC_R1	vmls.f64
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmacd
-	#define	FMAC_I2	fnmacd
+	#define	FMAC_I1	vmls.f64
+	#define	FMAC_I2	vmls.f64
 
 #endif
 
@@ -909,6 +918,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/zgemv_n_vfp.S

@@ -38,11 +38,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR	[fp, #0 ]
+#define OLD_ALPHAI	[fp, #8 ]
+#define OLD_A_SOFTFP	[fp, #16]
+#define OLD_LDA		[fp, #20]
+#define X		[fp, #24]
+#define OLD_INC_X	[fp, #28]
+#define Y		[fp, #32]
+#define OLD_INC_Y	[fp, #36]
+#else
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+#endif
+
 #define OLD_A		r3
 #define	OLD_M		r0
 
@@ -79,42 +91,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if !defined(CONJ) && !defined(XCONJ)
 
-        #define KMAC_R  fnmacd
+        #define KMAC_R  vmls.f64
         #define KMAC_I  fmacd
 
         #define FMAC_R1 fmacd
-        #define FMAC_R2 fnmacd
+        #define FMAC_R2 vmls.f64
         #define FMAC_I1 fmacd
         #define FMAC_I2 fmacd
 
 #elif defined(CONJ) && !defined(XCONJ)
 
         #define KMAC_R  fmacd
-        #define KMAC_I  fnmacd
+        #define KMAC_I  vmls.f64
 
         #define FMAC_R1 fmacd
-        #define FMAC_R2 fnmacd
+        #define FMAC_R2 vmls.f64
         #define FMAC_I1 fmacd
         #define FMAC_I2 fmacd
 
 #elif !defined(CONJ) && defined(XCONJ)
 
         #define KMAC_R  fmacd
-        #define KMAC_I  fnmacd
+        #define KMAC_I  vmls.f64
 
         #define FMAC_R1 fmacd
         #define FMAC_R2 fmacd
-        #define FMAC_I1 fnmacd
+        #define FMAC_I1 vmls.f64
         #define FMAC_I2 fmacd
 
 #else
 
-        #define KMAC_R  fnmacd
+        #define KMAC_R  vmls.f64
         #define KMAC_I  fmacd
 
         #define FMAC_R1 fmacd
         #define FMAC_R2 fmacd
-        #define FMAC_I1 fnmacd
+        #define FMAC_I1 vmls.f64
         #define FMAC_I2 fmacd
 
 #endif
@@ -465,6 +477,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	N, #0
 	ble	zgemvn_kernel_L999
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	d0, OLD_ALPHAR
+	vldr	d1, OLD_ALPHAI
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_A, A
 	str	OLD_M, M
 	vstr    d0 , ALPHA_R

kernel/arm/zgemv_t_vfp.S

@@ -38,11 +38,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
-#define	OLD_LDA		[fp, #0 ]
-#define	X		[fp, #4 ]
-#define	OLD_INC_X	[fp, #8 ]
-#define	Y		[fp, #12 ]
-#define	OLD_INC_Y	[fp, #16 ]
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR	[fp, #0 ]
+#define OLD_ALPHAI	[fp, #8 ]
+#define OLD_A_SOFTFP	[fp, #16]
+#define OLD_LDA		[fp, #20]
+#define X		[fp, #24]
+#define OLD_INC_X	[fp, #28]
+#define Y		[fp, #32]
+#define OLD_INC_Y	[fp, #36]
+#else
+#define OLD_LDA		[fp, #0 ]
+#define X		[fp, #4 ]
+#define OLD_INC_X	[fp, #8 ]
+#define Y		[fp, #12 ]
+#define OLD_INC_Y	[fp, #16 ]
+#endif
+
 #define OLD_A		r3
 #define	OLD_N		r1
 
@@ -77,42 +89,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if !defined(CONJ) && !defined(XCONJ)
 
-        #define KMAC_R  fnmacd
+        #define KMAC_R  vmls.f64
         #define KMAC_I  fmacd
 
         #define FMAC_R1 fmacd
-        #define FMAC_R2 fnmacd
+        #define FMAC_R2 vmls.f64
         #define FMAC_I1 fmacd
         #define FMAC_I2 fmacd
 
 #elif defined(CONJ) && !defined(XCONJ)
 
         #define KMAC_R  fmacd
-        #define KMAC_I  fnmacd
+        #define KMAC_I  vmls.f64
 
         #define FMAC_R1 fmacd
-        #define FMAC_R2 fnmacd
+        #define FMAC_R2 vmls.f64
         #define FMAC_I1 fmacd
         #define FMAC_I2 fmacd
 
 #elif !defined(CONJ) && defined(XCONJ)
 
         #define KMAC_R  fmacd
-        #define KMAC_I  fnmacd
+        #define KMAC_I  vmls.f64
 
         #define FMAC_R1 fmacd
         #define FMAC_R2 fmacd
-        #define FMAC_I1 fnmacd
+        #define FMAC_I1 vmls.f64
         #define FMAC_I2 fmacd
 
 #else
 
-        #define KMAC_R  fnmacd
+        #define KMAC_R  vmls.f64
         #define KMAC_I  fmacd
 
         #define FMAC_R1 fmacd
         #define FMAC_R2 fmacd
-        #define FMAC_I1 fnmacd
+        #define FMAC_I1 vmls.f64
         #define FMAC_I2 fmacd
 
 #endif
@@ -360,6 +372,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	OLD_N, #0
 	ble	zgemvt_kernel_L999
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	d0, OLD_ALPHAR
+	vldr	d1, OLD_ALPHAI
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
+
 	str	OLD_A, A
 	str	OLD_N, N
 

kernel/arm/ztrmm_kernel_2x2_vfp.S

@@ -66,10 +66,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	[fp, #4]
+#define OLD_ALPHAI_SOFTFP	[fp, #12]
+#define OLD_A_SOFTFP	[fp, #20 ]
+#define B	[fp, #24 ]
+#define C	[fp, #28 ]
+#define OLD_LDC	[fp, #32 ]
+#define OFFSET  [fp, #36 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -96,42 +106,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(NN) || defined(NT) || defined(TN) || defined(TT)
 
-	#define	KMAC_R	fnmacd
+	#define	KMAC_R	vmls.f64
 	#define	KMAC_I	fmacd
 
 	#define	FMAC_R1	fmacd
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmacd
 	#define	FMAC_I2	fmacd
 
 #elif defined(CN) || defined(CT)
 
 	#define	KMAC_R	fmacd
-	#define	KMAC_I	fnmacd
+	#define	KMAC_I	vmls.f64
 
 	#define	FMAC_R1	fmacd
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmacd
 	#define	FMAC_I2	fmacd
 
 #elif defined(NC) || defined(TC)
 
 	#define	KMAC_R	fmacd
-	#define	KMAC_I	fnmacd
+	#define	KMAC_I	vmls.f64
 
 	#define	FMAC_R1	fmacd
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmacd
+	#define	FMAC_I1	vmls.f64
 	#define	FMAC_I2	fmacd
 
 #else
 
-	#define	KMAC_R  fnmacd
+	#define	KMAC_R  vmls.f64
 	#define	KMAC_I	fmacd
 
 	#define	FMAC_R1	fmacd
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmacd
+	#define	FMAC_I1	vmls.f64
 	#define	FMAC_I2	fmacd
 
 #endif
@@ -882,6 +892,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/arm/ztrmm_kernel_2x2_vfpv3.S

@@ -66,10 +66,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ALPHA_I	[fp, #-272]
 #define ALPHA_R	[fp, #-280]
 
+#if !defined(__ARM_PCS_VFP)
+#define OLD_ALPHAR_SOFTFP	[fp, #4]
+#define OLD_ALPHAI_SOFTFP	[fp, #12]
+#define OLD_A_SOFTFP	[fp, #20 ]
+#define B	[fp, #24 ]
+#define C	[fp, #28 ]
+#define OLD_LDC	[fp, #32 ]
+#define OFFSET  [fp, #36 ]
+#else
 #define B	[fp, #4 ]
 #define C	[fp, #8 ]
 #define OLD_LDC	[fp, #12 ]
 #define OFFSET  [fp, #16 ]
+#endif
 
 #define I	r0
 #define J	r1
@@ -93,10 +103,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R	fsubd
 	#define	FADD_I	faddd
 
-	#define	FMAC_R1	fnmuld
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R1	vnmul.f64
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmuld
-	#define	FMAC_I2	fnmacd
+	#define	FMAC_I2	vmls.f64
 
 #elif defined(CN) || defined(CT)
 
@@ -105,7 +115,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	#define	FMAC_R1	fmuld
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmuld
+	#define	FMAC_I1	vnmul.f64
 	#define	FMAC_I2	fmacd
 
 #elif defined(NC) || defined(TC)
@@ -114,7 +124,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_I	fsubd
 
 	#define	FMAC_R1	fmuld
-	#define	FMAC_R2	fnmacd
+	#define	FMAC_R2	vmls.f64
 	#define	FMAC_I1	fmuld
 	#define	FMAC_I2	fmacd
 
@@ -123,10 +133,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	#define	FADD_R  fsubd
 	#define	FADD_I	faddd
 
-	#define	FMAC_R1	fnmuld
+	#define	FMAC_R1	vnmul.f64
 	#define	FMAC_R2	fmacd
-	#define	FMAC_I1	fnmuld
-	#define	FMAC_I2	fnmacd
+	#define	FMAC_I1	vnmul.f64
+	#define	FMAC_I2	vmls.f64
 
 #endif
 
@@ -883,6 +893,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	fp, sp, #24
 	sub	sp, sp, #STACKSIZE				// reserve stack
 
+#if !defined(__ARM_PCS_VFP)
+	vldr	OLD_ALPHA_R, OLD_ALPHAR_SOFTFP
+	vldr	OLD_ALPHA_I, OLD_ALPHAI_SOFTFP
+	ldr	OLD_A, OLD_A_SOFTFP
+#endif
 	str	OLD_M, M
 	str	OLD_N, N
 	str	OLD_K, K

kernel/power/casum_microk_power8.c

@@ -56,14 +56,14 @@ static float casum_kernel_16 (long n, float *x)
        "xxlxor		38, 38,	38	\n\t"
        "xxlxor		39, 39,	39	\n\t"
 
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %8, %2	\n\t"
-       "lxvw4x		42, %9, %2	\n\t"
-       "lxvw4x		43, %10, %2	\n\t"
-       "lxvw4x		44, %11, %2	\n\t"
-       "lxvw4x		45, %12, %2	\n\t"
-       "lxvw4x		46, %13, %2	\n\t"
-       "lxvw4x		47, %14, %2	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %8, %2	\n\t"
+       "lxvd2x		42, %9, %2	\n\t"
+       "lxvd2x		43, %10, %2	\n\t"
+       "lxvd2x		44, %11, %2	\n\t"
+       "lxvd2x		45, %12, %2	\n\t"
+       "lxvd2x		46, %13, %2	\n\t"
+       "lxvd2x		47, %14, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 
@@ -78,26 +78,26 @@ static float casum_kernel_16 (long n, float *x)
        "xvabssp		50, 42		\n\t"
        "xvabssp		51, 43		\n\t"
 
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %8, %2	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %8, %2	\n\t"
 
        "xvabssp		%x3, 44		\n\t"
        "xvabssp		%x4, 45		\n\t"
 
-       "lxvw4x		42, %9, %2	\n\t"
-       "lxvw4x		43, %10, %2	\n\t"
+       "lxvd2x		42, %9, %2	\n\t"
+       "lxvd2x		43, %10, %2	\n\t"
 
        "xvabssp		%x5, 46		\n\t"
        "xvabssp		%x6, 47		\n\t"
 
-       "lxvw4x		44, %11, %2	\n\t"
-       "lxvw4x		45, %12, %2	\n\t"
+       "lxvd2x		44, %11, %2	\n\t"
+       "lxvd2x		45, %12, %2	\n\t"
 
        "xvaddsp		32, 32, 48	\n\t"
        "xvaddsp		33, 33, 49	\n\t"
 
-       "lxvw4x		46, %13, %2	\n\t"
-       "lxvw4x		47, %14, %2	\n\t"
+       "lxvd2x		46, %13, %2	\n\t"
+       "lxvd2x		47, %14, %2	\n\t"
 
        "xvaddsp		34, 34, 50	\n\t"
        "xvaddsp		35, 35, 51	\n\t"

kernel/power/ccopy_microk_power8.c

@@ -39,25 +39,25 @@ static void ccopy_kernel_32 (long n, float *x, float *y)
 {
   __asm__
     (
-       "lxvw4x		32, 0, %2	\n\t"
-       "lxvw4x		33, %5, %2	\n\t"
-       "lxvw4x		34, %6, %2	\n\t"
-       "lxvw4x		35, %7, %2	\n\t"
-       "lxvw4x		36, %8, %2	\n\t"
-       "lxvw4x		37, %9, %2	\n\t"
-       "lxvw4x		38, %10, %2	\n\t"
-       "lxvw4x		39, %11, %2	\n\t"
+       "lxvd2x		32, 0, %2	\n\t"
+       "lxvd2x		33, %5, %2	\n\t"
+       "lxvd2x		34, %6, %2	\n\t"
+       "lxvd2x		35, %7, %2	\n\t"
+       "lxvd2x		36, %8, %2	\n\t"
+       "lxvd2x		37, %9, %2	\n\t"
+       "lxvd2x		38, %10, %2	\n\t"
+       "lxvd2x		39, %11, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %5, %2	\n\t"
-       "lxvw4x		42, %6, %2	\n\t"
-       "lxvw4x		43, %7, %2	\n\t"
-       "lxvw4x		44, %8, %2	\n\t"
-       "lxvw4x		45, %9, %2	\n\t"
-       "lxvw4x		46, %10, %2	\n\t"
-       "lxvw4x		47, %11, %2	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %5, %2	\n\t"
+       "lxvd2x		42, %6, %2	\n\t"
+       "lxvd2x		43, %7, %2	\n\t"
+       "lxvd2x		44, %8, %2	\n\t"
+       "lxvd2x		45, %9, %2	\n\t"
+       "lxvd2x		46, %10, %2	\n\t"
+       "lxvd2x		47, %11, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 
@@ -67,42 +67,42 @@ static void ccopy_kernel_32 (long n, float *x, float *y)
        ".p2align	5		\n"
      "1:				\n\t"
 
-       "stxvw4x		32, 0, %3	\n\t"
-       "stxvw4x		33, %5, %3	\n\t"
-       "lxvw4x		32, 0, %2	\n\t"
-       "lxvw4x		33, %5, %2	\n\t"
-       "stxvw4x		34, %6, %3	\n\t"
-       "stxvw4x		35, %7, %3	\n\t"
-       "lxvw4x		34, %6, %2	\n\t"
-       "lxvw4x		35, %7, %2	\n\t"
-       "stxvw4x		36, %8, %3	\n\t"
-       "stxvw4x		37, %9, %3	\n\t"
-       "lxvw4x		36, %8, %2	\n\t"
-       "lxvw4x		37, %9, %2	\n\t"
-       "stxvw4x		38, %10, %3	\n\t"
-       "stxvw4x		39, %11, %3	\n\t"
-       "lxvw4x		38, %10, %2	\n\t"
-       "lxvw4x		39, %11, %2	\n\t"
+       "stxvd2x		32, 0, %3	\n\t"
+       "stxvd2x		33, %5, %3	\n\t"
+       "lxvd2x		32, 0, %2	\n\t"
+       "lxvd2x		33, %5, %2	\n\t"
+       "stxvd2x		34, %6, %3	\n\t"
+       "stxvd2x		35, %7, %3	\n\t"
+       "lxvd2x		34, %6, %2	\n\t"
+       "lxvd2x		35, %7, %2	\n\t"
+       "stxvd2x		36, %8, %3	\n\t"
+       "stxvd2x		37, %9, %3	\n\t"
+       "lxvd2x		36, %8, %2	\n\t"
+       "lxvd2x		37, %9, %2	\n\t"
+       "stxvd2x		38, %10, %3	\n\t"
+       "stxvd2x		39, %11, %3	\n\t"
+       "lxvd2x		38, %10, %2	\n\t"
+       "lxvd2x		39, %11, %2	\n\t"
 
        "addi		%3, %3, 128	\n\t"
        "addi		%2, %2, 128	\n\t"
 
-       "stxvw4x		40, 0, %3	\n\t"
-       "stxvw4x		41, %5, %3	\n\t"
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %5, %2	\n\t"
-       "stxvw4x		42, %6, %3	\n\t"
-       "stxvw4x		43, %7, %3	\n\t"
-       "lxvw4x		42, %6, %2	\n\t"
-       "lxvw4x		43, %7, %2	\n\t"
-       "stxvw4x		44, %8, %3	\n\t"
-       "stxvw4x		45, %9, %3	\n\t"
-       "lxvw4x		44, %8, %2	\n\t"
-       "lxvw4x		45, %9, %2	\n\t"
-       "stxvw4x		46, %10, %3	\n\t"
-       "stxvw4x		47, %11, %3	\n\t"
-       "lxvw4x		46, %10, %2	\n\t"
-       "lxvw4x		47, %11, %2	\n\t"
+       "stxvd2x		40, 0, %3	\n\t"
+       "stxvd2x		41, %5, %3	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %5, %2	\n\t"
+       "stxvd2x		42, %6, %3	\n\t"
+       "stxvd2x		43, %7, %3	\n\t"
+       "lxvd2x		42, %6, %2	\n\t"
+       "lxvd2x		43, %7, %2	\n\t"
+       "stxvd2x		44, %8, %3	\n\t"
+       "stxvd2x		45, %9, %3	\n\t"
+       "lxvd2x		44, %8, %2	\n\t"
+       "lxvd2x		45, %9, %2	\n\t"
+       "stxvd2x		46, %10, %3	\n\t"
+       "stxvd2x		47, %11, %3	\n\t"
+       "lxvd2x		46, %10, %2	\n\t"
+       "lxvd2x		47, %11, %2	\n\t"
 
        "addi		%3, %3, 128	\n\t"
        "addi		%2, %2, 128	\n\t"
@@ -112,25 +112,25 @@ static void ccopy_kernel_32 (long n, float *x, float *y)
 
      "2:				\n\t"
 
-       "stxvw4x		32, 0, %3	\n\t"
-       "stxvw4x		33, %5, %3	\n\t"
-       "stxvw4x		34, %6, %3	\n\t"
-       "stxvw4x		35, %7, %3	\n\t"
-       "stxvw4x		36, %8, %3	\n\t"
-       "stxvw4x		37, %9, %3	\n\t"
-       "stxvw4x		38, %10, %3	\n\t"
-       "stxvw4x		39, %11, %3	\n\t"
+       "stxvd2x		32, 0, %3	\n\t"
+       "stxvd2x		33, %5, %3	\n\t"
+       "stxvd2x		34, %6, %3	\n\t"
+       "stxvd2x		35, %7, %3	\n\t"
+       "stxvd2x		36, %8, %3	\n\t"
+       "stxvd2x		37, %9, %3	\n\t"
+       "stxvd2x		38, %10, %3	\n\t"
+       "stxvd2x		39, %11, %3	\n\t"
 
        "addi		%3, %3, 128	\n\t"
 
-       "stxvw4x		40, 0, %3	\n\t"
-       "stxvw4x		41, %5, %3	\n\t"
-       "stxvw4x		42, %6, %3	\n\t"
-       "stxvw4x		43, %7, %3	\n\t"
-       "stxvw4x		44, %8, %3	\n\t"
-       "stxvw4x		45, %9, %3	\n\t"
-       "stxvw4x		46, %10, %3	\n\t"
-       "stxvw4x		47, %11, %3	\n"
+       "stxvd2x		40, 0, %3	\n\t"
+       "stxvd2x		41, %5, %3	\n\t"
+       "stxvd2x		42, %6, %3	\n\t"
+       "stxvd2x		43, %7, %3	\n\t"
+       "stxvd2x		44, %8, %3	\n\t"
+       "stxvd2x		45, %9, %3	\n\t"
+       "stxvd2x		46, %10, %3	\n\t"
+       "stxvd2x		47, %11, %3	\n"
 
      "#n=%1 x=%4=%2 y=%0=%3 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :

kernel/power/cswap_microk_power8.c

@@ -42,91 +42,91 @@ static void cswap_kernel_32 (long n, float *x, float *y)
        ".p2align	5		\n"
      "1:				\n\t"
 
-       "lxvw4x		32, 0, %4	\n\t"
-       "lxvw4x		33, %5, %4	\n\t"
-       "lxvw4x		34, %6, %4	\n\t"
-       "lxvw4x		35, %7, %4	\n\t"
-       "lxvw4x		36, %8, %4	\n\t"
-       "lxvw4x		37, %9, %4	\n\t"
-       "lxvw4x		38, %10, %4	\n\t"
-       "lxvw4x		39, %11, %4	\n\t"
+       "lxvd2x		32, 0, %4	\n\t"
+       "lxvd2x		33, %5, %4	\n\t"
+       "lxvd2x		34, %6, %4	\n\t"
+       "lxvd2x		35, %7, %4	\n\t"
+       "lxvd2x		36, %8, %4	\n\t"
+       "lxvd2x		37, %9, %4	\n\t"
+       "lxvd2x		38, %10, %4	\n\t"
+       "lxvd2x		39, %11, %4	\n\t"
 
        "addi		%4, %4, 128	\n\t"
 
-       "lxvw4x		40, 0, %4	\n\t"
-       "lxvw4x		41, %5, %4	\n\t"
-       "lxvw4x		42, %6, %4	\n\t"
-       "lxvw4x		43, %7, %4	\n\t"
-       "lxvw4x		44, %8, %4	\n\t"
-       "lxvw4x		45, %9, %4	\n\t"
-       "lxvw4x		46, %10, %4	\n\t"
-       "lxvw4x		47, %11, %4	\n\t"
+       "lxvd2x		40, 0, %4	\n\t"
+       "lxvd2x		41, %5, %4	\n\t"
+       "lxvd2x		42, %6, %4	\n\t"
+       "lxvd2x		43, %7, %4	\n\t"
+       "lxvd2x		44, %8, %4	\n\t"
+       "lxvd2x		45, %9, %4	\n\t"
+       "lxvd2x		46, %10, %4	\n\t"
+       "lxvd2x		47, %11, %4	\n\t"
 
        "addi		%4, %4, -128	\n\t"
 
-       "lxvw4x		48, 0, %3	\n\t"
-       "lxvw4x		49, %5, %3	\n\t"
-       "lxvw4x		50, %6, %3	\n\t"
-       "lxvw4x		51, %7, %3	\n\t"
-       "lxvw4x		0, %8, %3	\n\t"
-       "lxvw4x		1, %9, %3	\n\t"
-       "lxvw4x		2, %10, %3	\n\t"
-       "lxvw4x		3, %11, %3	\n\t"
+       "lxvd2x		48, 0, %3	\n\t"
+       "lxvd2x		49, %5, %3	\n\t"
+       "lxvd2x		50, %6, %3	\n\t"
+       "lxvd2x		51, %7, %3	\n\t"
+       "lxvd2x		0, %8, %3	\n\t"
+       "lxvd2x		1, %9, %3	\n\t"
+       "lxvd2x		2, %10, %3	\n\t"
+       "lxvd2x		3, %11, %3	\n\t"
 
        "addi		%3, %3, 128	\n\t"
 
-       "lxvw4x		4, 0, %3	\n\t"
-       "lxvw4x		5, %5, %3	\n\t"
-       "lxvw4x		6, %6, %3	\n\t"
-       "lxvw4x		7, %7, %3	\n\t"
-       "lxvw4x		8, %8, %3	\n\t"
-       "lxvw4x		9, %9, %3	\n\t"
-       "lxvw4x		10, %10, %3	\n\t"
-       "lxvw4x		11, %11, %3	\n\t"
+       "lxvd2x		4, 0, %3	\n\t"
+       "lxvd2x		5, %5, %3	\n\t"
+       "lxvd2x		6, %6, %3	\n\t"
+       "lxvd2x		7, %7, %3	\n\t"
+       "lxvd2x		8, %8, %3	\n\t"
+       "lxvd2x		9, %9, %3	\n\t"
+       "lxvd2x		10, %10, %3	\n\t"
+       "lxvd2x		11, %11, %3	\n\t"
 
        "addi		%3, %3, -128	\n\t"
 
-       "stxvw4x		32, 0, %3	\n\t"
-       "stxvw4x		33, %5, %3	\n\t"
-       "stxvw4x		34, %6, %3	\n\t"
-       "stxvw4x		35, %7, %3	\n\t"
-       "stxvw4x		36, %8, %3	\n\t"
-       "stxvw4x		37, %9, %3	\n\t"
-       "stxvw4x		38, %10, %3	\n\t"
-       "stxvw4x		39, %11, %3	\n\t"
+       "stxvd2x		32, 0, %3	\n\t"
+       "stxvd2x		33, %5, %3	\n\t"
+       "stxvd2x		34, %6, %3	\n\t"
+       "stxvd2x		35, %7, %3	\n\t"
+       "stxvd2x		36, %8, %3	\n\t"
+       "stxvd2x		37, %9, %3	\n\t"
+       "stxvd2x		38, %10, %3	\n\t"
+       "stxvd2x		39, %11, %3	\n\t"
 
        "addi		%3, %3, 128	\n\t"
 
-       "stxvw4x		40, 0, %3	\n\t"
-       "stxvw4x		41, %5, %3	\n\t"
-       "stxvw4x		42, %6, %3	\n\t"
-       "stxvw4x		43, %7, %3	\n\t"
-       "stxvw4x		44, %8, %3	\n\t"
-       "stxvw4x		45, %9, %3	\n\t"
-       "stxvw4x		46, %10, %3	\n\t"
-       "stxvw4x		47, %11, %3	\n\t"
+       "stxvd2x		40, 0, %3	\n\t"
+       "stxvd2x		41, %5, %3	\n\t"
+       "stxvd2x		42, %6, %3	\n\t"
+       "stxvd2x		43, %7, %3	\n\t"
+       "stxvd2x		44, %8, %3	\n\t"
+       "stxvd2x		45, %9, %3	\n\t"
+       "stxvd2x		46, %10, %3	\n\t"
+       "stxvd2x		47, %11, %3	\n\t"
 
        "addi		%3, %3, 128	\n\t"
 
-       "stxvw4x		48, 0, %4	\n\t"
-       "stxvw4x		49, %5, %4	\n\t"
-       "stxvw4x		50, %6, %4	\n\t"
-       "stxvw4x		51, %7, %4	\n\t"
-       "stxvw4x		0, %8, %4	\n\t"
-       "stxvw4x		1, %9, %4	\n\t"
-       "stxvw4x		2, %10, %4	\n\t"
-       "stxvw4x		3, %11, %4	\n\t"
+       "stxvd2x		48, 0, %4	\n\t"
+       "stxvd2x		49, %5, %4	\n\t"
+       "stxvd2x		50, %6, %4	\n\t"
+       "stxvd2x		51, %7, %4	\n\t"
+       "stxvd2x		0, %8, %4	\n\t"
+       "stxvd2x		1, %9, %4	\n\t"
+       "stxvd2x		2, %10, %4	\n\t"
+       "stxvd2x		3, %11, %4	\n\t"
 
        "addi		%4, %4, 128	\n\t"
 
-       "stxvw4x		4, 0, %4	\n\t"
-       "stxvw4x		5, %5, %4	\n\t"
-       "stxvw4x		6, %6, %4	\n\t"
-       "stxvw4x		7, %7, %4	\n\t"
-       "stxvw4x		8, %8, %4	\n\t"
-       "stxvw4x		9, %9, %4	\n\t"
-       "stxvw4x		10, %10, %4	\n\t"
-       "stxvw4x		11, %11, %4	\n\t"
+       "stxvd2x		4, 0, %4	\n\t"
+       "stxvd2x		5, %5, %4	\n\t"
+       "stxvd2x		6, %6, %4	\n\t"
+       "stxvd2x		7, %7, %4	\n\t"
+       "stxvd2x		8, %8, %4	\n\t"
+       "stxvd2x		9, %9, %4	\n\t"
+       "stxvd2x		10, %10, %4	\n\t"
+       "stxvd2x		11, %11, %4	\n\t"
 
        "addi		%4, %4, 128	\n\t"
 

kernel/power/sasum_microk_power8.c

@@ -56,14 +56,14 @@ static float sasum_kernel_32 (long n, float *x)
        "xxlxor		38, 38,	38	\n\t"
        "xxlxor		39, 39,	39	\n\t"
 
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %8, %2	\n\t"
-       "lxvw4x		42, %9, %2	\n\t"
-       "lxvw4x		43, %10, %2	\n\t"
-       "lxvw4x		44, %11, %2	\n\t"
-       "lxvw4x		45, %12, %2	\n\t"
-       "lxvw4x		46, %13, %2	\n\t"
-       "lxvw4x		47, %14, %2	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %8, %2	\n\t"
+       "lxvd2x		42, %9, %2	\n\t"
+       "lxvd2x		43, %10, %2	\n\t"
+       "lxvd2x		44, %11, %2	\n\t"
+       "lxvd2x		45, %12, %2	\n\t"
+       "lxvd2x		46, %13, %2	\n\t"
+       "lxvd2x		47, %14, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 
@@ -78,26 +78,26 @@ static float sasum_kernel_32 (long n, float *x)
        "xvabssp		50, 42		\n\t"
        "xvabssp		51, 43		\n\t"
 
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %8, %2	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %8, %2	\n\t"
 
        "xvabssp		%x3, 44		\n\t"
        "xvabssp		%x4, 45		\n\t"
 
-       "lxvw4x		42, %9, %2	\n\t"
-       "lxvw4x		43, %10, %2	\n\t"
+       "lxvd2x		42, %9, %2	\n\t"
+       "lxvd2x		43, %10, %2	\n\t"
 
        "xvabssp		%x5, 46		\n\t"
        "xvabssp		%x6, 47		\n\t"
 
-       "lxvw4x		44, %11, %2	\n\t"
-       "lxvw4x		45, %12, %2	\n\t"
+       "lxvd2x		44, %11, %2	\n\t"
+       "lxvd2x		45, %12, %2	\n\t"
 
        "xvaddsp		32, 32, 48	\n\t"
        "xvaddsp		33, 33, 49	\n\t"
 
-       "lxvw4x		46, %13, %2	\n\t"
-       "lxvw4x		47, %14, %2	\n\t"
+       "lxvd2x		46, %13, %2	\n\t"
+       "lxvd2x		47, %14, %2	\n\t"
 
        "xvaddsp		34, 34, 50	\n\t"
        "xvaddsp		35, 35, 51	\n\t"

kernel/power/scopy_microk_power8.c

@@ -39,14 +39,14 @@ static void scopy_kernel_32 (long n, float *x, float *y)
 {
   __asm__
     (
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %5, %2	\n\t"
-       "lxvw4x		42, %6, %2	\n\t"
-       "lxvw4x		43, %7, %2	\n\t"
-       "lxvw4x		44, %8, %2	\n\t"
-       "lxvw4x		45, %9, %2	\n\t"
-       "lxvw4x		46, %10, %2	\n\t"
-       "lxvw4x		47, %11, %2	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %5, %2	\n\t"
+       "lxvd2x		42, %6, %2	\n\t"
+       "lxvd2x		43, %7, %2	\n\t"
+       "lxvd2x		44, %8, %2	\n\t"
+       "lxvd2x		45, %9, %2	\n\t"
+       "lxvd2x		46, %10, %2	\n\t"
+       "lxvd2x		47, %11, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 
@@ -56,22 +56,22 @@ static void scopy_kernel_32 (long n, float *x, float *y)
        ".p2align	5		\n"
      "1:				\n\t"
 
-       "stxvw4x		40, 0, %3	\n\t"
-       "stxvw4x		41, %5, %3	\n\t"
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		41, %5, %2	\n\t"
-       "stxvw4x		42, %6, %3	\n\t"
-       "stxvw4x		43, %7, %3	\n\t"
-       "lxvw4x		42, %6, %2	\n\t"
-       "lxvw4x		43, %7, %2	\n\t"
-       "stxvw4x		44, %8, %3	\n\t"
-       "stxvw4x		45, %9, %3	\n\t"
-       "lxvw4x		44, %8, %2	\n\t"
-       "lxvw4x		45, %9, %2	\n\t"
-       "stxvw4x		46, %10, %3	\n\t"
-       "stxvw4x		47, %11, %3	\n\t"
-       "lxvw4x		46, %10, %2	\n\t"
-       "lxvw4x		47, %11, %2	\n\t"
+       "stxvd2x		40, 0, %3	\n\t"
+       "stxvd2x		41, %5, %3	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		41, %5, %2	\n\t"
+       "stxvd2x		42, %6, %3	\n\t"
+       "stxvd2x		43, %7, %3	\n\t"
+       "lxvd2x		42, %6, %2	\n\t"
+       "lxvd2x		43, %7, %2	\n\t"
+       "stxvd2x		44, %8, %3	\n\t"
+       "stxvd2x		45, %9, %3	\n\t"
+       "lxvd2x		44, %8, %2	\n\t"
+       "lxvd2x		45, %9, %2	\n\t"
+       "stxvd2x		46, %10, %3	\n\t"
+       "stxvd2x		47, %11, %3	\n\t"
+       "lxvd2x		46, %10, %2	\n\t"
+       "lxvd2x		47, %11, %2	\n\t"
 
        "addi		%3, %3, 128	\n\t"
        "addi		%2, %2, 128	\n\t"
@@ -81,14 +81,14 @@ static void scopy_kernel_32 (long n, float *x, float *y)
 
      "2:				\n\t"
 
-       "stxvw4x		40, 0, %3	\n\t"
-       "stxvw4x		41, %5, %3	\n\t"
-       "stxvw4x		42, %6, %3	\n\t"
-       "stxvw4x		43, %7, %3	\n\t"
-       "stxvw4x		44, %8, %3	\n\t"
-       "stxvw4x		45, %9, %3	\n\t"
-       "stxvw4x		46, %10, %3	\n\t"
-       "stxvw4x		47, %11, %3	\n"
+       "stxvd2x		40, 0, %3	\n\t"
+       "stxvd2x		41, %5, %3	\n\t"
+       "stxvd2x		42, %6, %3	\n\t"
+       "stxvd2x		43, %7, %3	\n\t"
+       "stxvd2x		44, %8, %3	\n\t"
+       "stxvd2x		45, %9, %3	\n\t"
+       "stxvd2x		46, %10, %3	\n\t"
+       "stxvd2x		47, %11, %3	\n"
 
      "#n=%1 x=%4=%2 y=%0=%3 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :

kernel/power/sdot_microk_power8.c

@@ -57,22 +57,22 @@ static float sdot_kernel_16 (long n, float *x, float *y)
        "xxlxor		38, 38,	38	\n\t"
        "xxlxor		39, 39,	39	\n\t"
 
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		48, 0, %3	\n\t"
-       "lxvw4x		41, %10, %2	\n\t"
-       "lxvw4x		49, %10, %3	\n\t"
-       "lxvw4x		42, %11, %2	\n\t"
-       "lxvw4x		50, %11, %3	\n\t"
-       "lxvw4x		43, %12, %2	\n\t"
-       "lxvw4x		51, %12, %3	\n\t"
-       "lxvw4x		44, %13, %2	\n\t"
-       "lxvw4x		%x4, %13, %3	\n\t"
-       "lxvw4x		45, %14, %2	\n\t"
-       "lxvw4x		%x5, %14, %3	\n\t"
-       "lxvw4x		46, %15, %2	\n\t"
-       "lxvw4x		%x6, %15, %3	\n\t"
-       "lxvw4x		47, %16, %2	\n\t"
-       "lxvw4x		%x7, %16, %3	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		48, 0, %3	\n\t"
+       "lxvd2x		41, %10, %2	\n\t"
+       "lxvd2x		49, %10, %3	\n\t"
+       "lxvd2x		42, %11, %2	\n\t"
+       "lxvd2x		50, %11, %3	\n\t"
+       "lxvd2x		43, %12, %2	\n\t"
+       "lxvd2x		51, %12, %3	\n\t"
+       "lxvd2x		44, %13, %2	\n\t"
+       "lxvd2x		%x4, %13, %3	\n\t"
+       "lxvd2x		45, %14, %2	\n\t"
+       "lxvd2x		%x5, %14, %3	\n\t"
+       "lxvd2x		46, %15, %2	\n\t"
+       "lxvd2x		%x6, %15, %3	\n\t"
+       "lxvd2x		47, %16, %2	\n\t"
+       "lxvd2x		%x7, %16, %3	\n\t"
 
        "addi		%2, %2, 128	\n\t"
        "addi		%3, %3, 128	\n\t"
@@ -84,29 +84,29 @@ static float sdot_kernel_16 (long n, float *x, float *y)
      "1:				\n\t"
 
        "xvmaddasp	32, 40, 48	\n\t"
-       "lxvw4x		40, 0, %2	\n\t"
-       "lxvw4x		48, 0, %3	\n\t"
+       "lxvd2x		40, 0, %2	\n\t"
+       "lxvd2x		48, 0, %3	\n\t"
        "xvmaddasp	33, 41, 49	\n\t"
-       "lxvw4x		41, %10, %2	\n\t"
-       "lxvw4x		49, %10, %3	\n\t"
+       "lxvd2x		41, %10, %2	\n\t"
+       "lxvd2x		49, %10, %3	\n\t"
        "xvmaddasp	34, 42, 50	\n\t"
-       "lxvw4x		42, %11, %2	\n\t"
-       "lxvw4x		50, %11, %3	\n\t"
+       "lxvd2x		42, %11, %2	\n\t"
+       "lxvd2x		50, %11, %3	\n\t"
        "xvmaddasp	35, 43, 51	\n\t"
-       "lxvw4x		43, %12, %2	\n\t"
-       "lxvw4x		51, %12, %3	\n\t"
+       "lxvd2x		43, %12, %2	\n\t"
+       "lxvd2x		51, %12, %3	\n\t"
        "xvmaddasp	36, 44, %x4	\n\t"
-       "lxvw4x		44, %13, %2	\n\t"
-       "lxvw4x		%x4, %13, %3	\n\t"
+       "lxvd2x		44, %13, %2	\n\t"
+       "lxvd2x		%x4, %13, %3	\n\t"
        "xvmaddasp	37, 45, %x5	\n\t"
-       "lxvw4x		45, %14, %2	\n\t"
-       "lxvw4x		%x5, %14, %3	\n\t"
+       "lxvd2x		45, %14, %2	\n\t"
+       "lxvd2x		%x5, %14, %3	\n\t"
        "xvmaddasp	38, 46, %x6	\n\t"
-       "lxvw4x		46, %15, %2	\n\t"
-       "lxvw4x		%x6, %15, %3	\n\t"
+       "lxvd2x		46, %15, %2	\n\t"
+       "lxvd2x		%x6, %15, %3	\n\t"
        "xvmaddasp	39, 47, %x7	\n\t"
-       "lxvw4x		47, %16, %2	\n\t"
-       "lxvw4x		%x7, %16, %3	\n\t"
+       "lxvd2x		47, %16, %2	\n\t"
+       "lxvd2x		%x7, %16, %3	\n\t"
 
        "addi		%2, %2, 128	\n\t"
        "addi		%3, %3, 128	\n\t"

kernel/power/srot_microk_power8.c

@@ -57,15 +57,15 @@ static void srot_kernel_16 (long n, float *x, float *y, float c, float s)
        "xscvdpspn	37, %x14	\n\t"	// load s to all words
        "xxspltw		37, 37, 0	\n\t"
 
-       "lxvw4x		32, 0, %3	\n\t"	// load x
-       "lxvw4x		33, %15, %3	\n\t"
-       "lxvw4x		34, %16, %3	\n\t"
-       "lxvw4x		35, %17, %3	\n\t"
+       "lxvd2x		32, 0, %3	\n\t"	// load x
+       "lxvd2x		33, %15, %3	\n\t"
+       "lxvd2x		34, %16, %3	\n\t"
+       "lxvd2x		35, %17, %3	\n\t"
 
-       "lxvw4x		48, 0, %4	\n\t"	// load y
-       "lxvw4x		49, %15, %4	\n\t"
-       "lxvw4x		50, %16, %4	\n\t"
-       "lxvw4x		51, %17, %4	\n\t"
+       "lxvd2x		48, 0, %4	\n\t"	// load y
+       "lxvd2x		49, %15, %4	\n\t"
+       "lxvd2x		50, %16, %4	\n\t"
+       "lxvd2x		51, %17, %4	\n\t"
 
        "addi		%3, %3, 64	\n\t"
        "addi		%4, %4, 64	\n\t"
@@ -89,26 +89,26 @@ static void srot_kernel_16 (long n, float *x, float *y, float c, float s)
        "xvmulsp		44, 32, 37	\n\t"	// s * x
        "xvmulsp		45, 33, 37	\n\t"
 
-       "lxvw4x		32, 0, %3	\n\t"	// load x
-       "lxvw4x		33, %15, %3	\n\t"
+       "lxvd2x		32, 0, %3	\n\t"	// load x
+       "lxvd2x		33, %15, %3	\n\t"
 
        "xvmulsp		46, 34, 37	\n\t"
        "xvmulsp		47, 35, 37	\n\t"
 
-       "lxvw4x		34, %16, %3	\n\t"
-       "lxvw4x		35, %17, %3	\n\t"
+       "lxvd2x		34, %16, %3	\n\t"
+       "lxvd2x		35, %17, %3	\n\t"
 
        "xvmulsp		%x9, 48, 37	\n\t"	// s * y
        "xvmulsp		%x10, 49, 37	\n\t"
 
-       "lxvw4x		48, 0, %4	\n\t"	// load y
-       "lxvw4x		49, %15, %4	\n\t"
+       "lxvd2x		48, 0, %4	\n\t"	// load y
+       "lxvd2x		49, %15, %4	\n\t"
 
        "xvmulsp		%x11, 50, 37	\n\t"
        "xvmulsp		%x12, 51, 37	\n\t"
 
-       "lxvw4x		50, %16, %4	\n\t"
-       "lxvw4x		51, %17, %4	\n\t"
+       "lxvd2x		50, %16, %4	\n\t"
+       "lxvd2x		51, %17, %4	\n\t"
 
        "xvaddsp		40, 40, %x9	\n\t"	// c * x + s * y
        "xvaddsp		41, 41, %x10	\n\t"	// c * x + s * y
@@ -124,15 +124,15 @@ static void srot_kernel_16 (long n, float *x, float *y, float c, float s)
        "xvsubsp		%x7, %x7, 46	\n\t"	// c * y - s * x
        "xvsubsp		%x8, %x8, 47	\n\t"	// c * y - s * x
 
-       "stxvw4x		40, 0, %3	\n\t"	// store x
-       "stxvw4x		41, %15, %3	\n\t"
-       "stxvw4x		42, %16, %3	\n\t"
-       "stxvw4x		43, %17, %3	\n\t"
+       "stxvd2x		40, 0, %3	\n\t"	// store x
+       "stxvd2x		41, %15, %3	\n\t"
+       "stxvd2x		42, %16, %3	\n\t"
+       "stxvd2x		43, %17, %3	\n\t"
 
-       "stxvw4x		%x5, 0, %4	\n\t"	// store y
-       "stxvw4x		%x6, %15, %4	\n\t"
-       "stxvw4x		%x7, %16, %4	\n\t"
-       "stxvw4x		%x8, %17, %4	\n\t"
+       "stxvd2x		%x5, 0, %4	\n\t"	// store y
+       "stxvd2x		%x6, %15, %4	\n\t"
+       "stxvd2x		%x7, %16, %4	\n\t"
+       "stxvd2x		%x8, %17, %4	\n\t"
 
        "addi		%3, %3, 128	\n\t"
        "addi		%4, %4, 128	\n\t"
@@ -175,15 +175,15 @@ static void srot_kernel_16 (long n, float *x, float *y, float c, float s)
        "xvsubsp		%x7, %x7, 46	\n\t"	// c * y - s * x
        "xvsubsp		%x8, %x8, 47	\n\t"	// c * y - s * x
 
-       "stxvw4x		40, 0, %3	\n\t"	// store x
-       "stxvw4x		41, %15, %3	\n\t"
-       "stxvw4x		42, %16, %3	\n\t"
-       "stxvw4x		43, %17, %3	\n\t"
+       "stxvd2x		40, 0, %3	\n\t"	// store x
+       "stxvd2x		41, %15, %3	\n\t"
+       "stxvd2x		42, %16, %3	\n\t"
+       "stxvd2x		43, %17, %3	\n\t"
 
-       "stxvw4x		%x5, 0, %4	\n\t"	// store y
-       "stxvw4x		%x6, %15, %4	\n\t"
-       "stxvw4x		%x7, %16, %4	\n\t"
-       "stxvw4x		%x8, %17, %4	\n"
+       "stxvd2x		%x5, 0, %4	\n\t"	// store y
+       "stxvd2x		%x6, %15, %4	\n\t"
+       "stxvd2x		%x7, %16, %4	\n\t"
+       "stxvd2x		%x8, %17, %4	\n"
 
      "#n=%2 x=%0=%3 y=%1=%4 c=%13 s=%14 o16=%15 o32=%16 o48=%17\n"
      "#t0=%x5 t1=%x6 t2=%x7 t3=%x8 t4=%x9 t5=%x10 t6=%x11 t7=%x12"

kernel/power/sscal_microk_power8.c

@@ -44,14 +44,14 @@ static void sscal_kernel_16 (long n, float *x, float alpha)
        "xscvdpspn	%x3, %x3	\n\t"
        "xxspltw		%x3, %x3, 0	\n\t"
 
-       "lxvw4x		32, 0, %2	\n\t"
-       "lxvw4x		33, %4, %2	\n\t"
-       "lxvw4x		34, %5, %2	\n\t"
-       "lxvw4x		35, %6, %2	\n\t"
-       "lxvw4x		36, %7, %2	\n\t"
-       "lxvw4x		37, %8, %2	\n\t"
-       "lxvw4x		38, %9, %2	\n\t"
-       "lxvw4x		39, %10, %2	\n\t"
+       "lxvd2x		32, 0, %2	\n\t"
+       "lxvd2x		33, %4, %2	\n\t"
+       "lxvd2x		34, %5, %2	\n\t"
+       "lxvd2x		35, %6, %2	\n\t"
+       "lxvd2x		36, %7, %2	\n\t"
+       "lxvd2x		37, %8, %2	\n\t"
+       "lxvd2x		38, %9, %2	\n\t"
+       "lxvd2x		39, %10, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 
@@ -63,31 +63,31 @@ static void sscal_kernel_16 (long n, float *x, float alpha)
 
        "xvmulsp		40, 32, %x3	\n\t"
        "xvmulsp		41, 33, %x3	\n\t"
-       "lxvw4x		32, 0, %2	\n\t"
-       "lxvw4x		33, %4, %2	\n\t"
+       "lxvd2x		32, 0, %2	\n\t"
+       "lxvd2x		33, %4, %2	\n\t"
        "xvmulsp		42, 34, %x3	\n\t"
        "xvmulsp		43, 35, %x3	\n\t"
-       "lxvw4x		34, %5, %2	\n\t"
-       "lxvw4x		35, %6, %2	\n\t"
+       "lxvd2x		34, %5, %2	\n\t"
+       "lxvd2x		35, %6, %2	\n\t"
        "xvmulsp		44, 36, %x3	\n\t"
        "xvmulsp		45, 37, %x3	\n\t"
-       "lxvw4x		36, %7, %2	\n\t"
-       "lxvw4x		37, %8, %2	\n\t"
+       "lxvd2x		36, %7, %2	\n\t"
+       "lxvd2x		37, %8, %2	\n\t"
        "xvmulsp		46, 38, %x3	\n\t"
        "xvmulsp		47, 39, %x3	\n\t"
-       "lxvw4x		38, %9, %2	\n\t"
-       "lxvw4x		39, %10, %2	\n\t"
+       "lxvd2x		38, %9, %2	\n\t"
+       "lxvd2x		39, %10, %2	\n\t"
 
        "addi		%2, %2, -128	\n\t"
 
-       "stxvw4x		40, 0, %2	\n\t"
-       "stxvw4x		41, %4, %2	\n\t"
-       "stxvw4x		42, %5, %2	\n\t"
-       "stxvw4x		43, %6, %2	\n\t"
-       "stxvw4x		44, %7, %2	\n\t"
-       "stxvw4x		45, %8, %2	\n\t"
-       "stxvw4x		46, %9, %2	\n\t"
-       "stxvw4x		47, %10, %2	\n\t"
+       "stxvd2x		40, 0, %2	\n\t"
+       "stxvd2x		41, %4, %2	\n\t"
+       "stxvd2x		42, %5, %2	\n\t"
+       "stxvd2x		43, %6, %2	\n\t"
+       "stxvd2x		44, %7, %2	\n\t"
+       "stxvd2x		45, %8, %2	\n\t"
+       "stxvd2x		46, %9, %2	\n\t"
+       "stxvd2x		47, %10, %2	\n\t"
 
        "addi		%2, %2, 256	\n\t"
 
@@ -108,14 +108,14 @@ static void sscal_kernel_16 (long n, float *x, float alpha)
        "xvmulsp		46, 38, %x3	\n\t"
        "xvmulsp		47, 39, %x3	\n\t"
 
-       "stxvw4x		40, 0, %2	\n\t"
-       "stxvw4x		41, %4, %2	\n\t"
-       "stxvw4x		42, %5, %2	\n\t"
-       "stxvw4x		43, %6, %2	\n\t"
-       "stxvw4x		44, %7, %2	\n\t"
-       "stxvw4x		45, %8, %2	\n\t"
-       "stxvw4x		46, %9, %2	\n\t"
-       "stxvw4x		47, %10, %2	\n"
+       "stxvd2x		40, 0, %2	\n\t"
+       "stxvd2x		41, %4, %2	\n\t"
+       "stxvd2x		42, %5, %2	\n\t"
+       "stxvd2x		43, %6, %2	\n\t"
+       "stxvd2x		44, %7, %2	\n\t"
+       "stxvd2x		45, %8, %2	\n\t"
+       "stxvd2x		46, %9, %2	\n\t"
+       "stxvd2x		47, %10, %2	\n"
 
      "#n=%1 alpha=%3 x=%0=%2 o16=%4 o32=%5 o48=%6 o64=%7 o80=%8 o96=%9 o112=%10"
      :
@@ -150,14 +150,14 @@ static void sscal_kernel_16_zero (long n, float *x)
        ".p2align	5		\n"
      "1:				\n\t"
 
-       "stxvw4x		%x3, 0, %2	\n\t"
-       "stxvw4x		%x3, %4, %2	\n\t"
-       "stxvw4x		%x3, %5, %2	\n\t"
-       "stxvw4x		%x3, %6, %2	\n\t"
-       "stxvw4x		%x3, %7, %2	\n\t"
-       "stxvw4x		%x3, %8, %2	\n\t"
-       "stxvw4x		%x3, %9, %2	\n\t"
-       "stxvw4x		%x3, %10, %2	\n\t"
+       "stxvd2x		%x3, 0, %2	\n\t"
+       "stxvd2x		%x3, %4, %2	\n\t"
+       "stxvd2x		%x3, %5, %2	\n\t"
+       "stxvd2x		%x3, %6, %2	\n\t"
+       "stxvd2x		%x3, %7, %2	\n\t"
+       "stxvd2x		%x3, %8, %2	\n\t"
+       "stxvd2x		%x3, %9, %2	\n\t"
+       "stxvd2x		%x3, %10, %2	\n\t"
 
        "addi		%2, %2, 128	\n\t"
 

kernel/power/sswap_microk_power8.c

@@ -42,43 +42,43 @@ static void sswap_kernel_32 (long n, float *x, float *y)
        ".p2align	5		\n"
      "1:				\n\t"
 
-       "lxvw4x		32, 0, %4	\n\t"
-       "lxvw4x		33, %5, %4	\n\t"
-       "lxvw4x		34, %6, %4	\n\t"
-       "lxvw4x		35, %7, %4	\n\t"
-       "lxvw4x		36, %8, %4	\n\t"
-       "lxvw4x		37, %9, %4	\n\t"
-       "lxvw4x		38, %10, %4	\n\t"
-       "lxvw4x		39, %11, %4	\n\t"
+       "lxvd2x		32, 0, %4	\n\t"
+       "lxvd2x		33, %5, %4	\n\t"
+       "lxvd2x		34, %6, %4	\n\t"
+       "lxvd2x		35, %7, %4	\n\t"
+       "lxvd2x		36, %8, %4	\n\t"
+       "lxvd2x		37, %9, %4	\n\t"
+       "lxvd2x		38, %10, %4	\n\t"
+       "lxvd2x		39, %11, %4	\n\t"
 
-       "lxvw4x		40, 0, %3	\n\t"
-       "lxvw4x		41, %5, %3	\n\t"
-       "lxvw4x		42, %6, %3	\n\t"
-       "lxvw4x		43, %7, %3	\n\t"
-       "lxvw4x		44, %8, %3	\n\t"
-       "lxvw4x		45, %9, %3	\n\t"
-       "lxvw4x		46, %10, %3	\n\t"
-       "lxvw4x		47, %11, %3	\n\t"
+       "lxvd2x		40, 0, %3	\n\t"
+       "lxvd2x		41, %5, %3	\n\t"
+       "lxvd2x		42, %6, %3	\n\t"
+       "lxvd2x		43, %7, %3	\n\t"
+       "lxvd2x		44, %8, %3	\n\t"
+       "lxvd2x		45, %9, %3	\n\t"
+       "lxvd2x		46, %10, %3	\n\t"
+       "lxvd2x		47, %11, %3	\n\t"
 
-       "stxvw4x		32, 0, %3	\n\t"
-       "stxvw4x		33, %5, %3	\n\t"
-       "stxvw4x		34, %6, %3	\n\t"
-       "stxvw4x		35, %7, %3	\n\t"
-       "stxvw4x		36, %8, %3	\n\t"
-       "stxvw4x		37, %9, %3	\n\t"
-       "stxvw4x		38, %10, %3	\n\t"
-       "stxvw4x		39, %11, %3	\n\t"
+       "stxvd2x		32, 0, %3	\n\t"
+       "stxvd2x		33, %5, %3	\n\t"
+       "stxvd2x		34, %6, %3	\n\t"
+       "stxvd2x		35, %7, %3	\n\t"
+       "stxvd2x		36, %8, %3	\n\t"
+       "stxvd2x		37, %9, %3	\n\t"
+       "stxvd2x		38, %10, %3	\n\t"
+       "stxvd2x		39, %11, %3	\n\t"
 
        "addi		%3, %3, 128	\n\t"
 
-       "stxvw4x		40, 0, %4	\n\t"
-       "stxvw4x		41, %5, %4	\n\t"
-       "stxvw4x		42, %6, %4	\n\t"
-       "stxvw4x		43, %7, %4	\n\t"
-       "stxvw4x		44, %8, %4	\n\t"
-       "stxvw4x		45, %9, %4	\n\t"
-       "stxvw4x		46, %10, %4	\n\t"
-       "stxvw4x		47, %11, %4	\n\t"
+       "stxvd2x		40, 0, %4	\n\t"
+       "stxvd2x		41, %5, %4	\n\t"
+       "stxvd2x		42, %6, %4	\n\t"
+       "stxvd2x		43, %7, %4	\n\t"
+       "stxvd2x		44, %8, %4	\n\t"
+       "stxvd2x		45, %9, %4	\n\t"
+       "stxvd2x		46, %10, %4	\n\t"
+       "stxvd2x		47, %11, %4	\n\t"
 
        "addi		%4, %4, 128	\n\t"
 

utest/CMakeLists.txt

@@ -21,6 +21,10 @@ if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
 target_link_libraries(${OpenBLAS_utest_bin} m)
 endif()
 
+if (${CMAKE_SYSTEM_NAME} STREQUAL "WindowsStore")
+set_target_properties( ${OpenBLAS_utest_bin} PROPERTIES COMPILE_DEFINITIONS "_CRT_SECURE_NO_WARNINGS")
+endif()
+
 #Set output for utest
 set_target_properties( ${OpenBLAS_utest_bin} PROPERTIES RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
 foreach (OUTPUTCONFIG ${CMAKE_CONFIGURATION_TYPES})