Merge pull request #79 from xianyi/develop

rebase

Makefile.install

@@ -13,6 +13,14 @@ OPENBLAS_CMAKE_DIR := $(OPENBLAS_LIBRARY_DIR)/cmake/openblas
 OPENBLAS_CMAKE_CONFIG := OpenBLASConfig.cmake
 OPENBLAS_CMAKE_CONFIG_VERSION := OpenBLASConfigVersion.cmake
 OPENBLAS_PKGCONFIG_DIR := $(OPENBLAS_LIBRARY_DIR)/pkgconfig
+PKG_EXTRALIB := $(EXTRALIB)
+ifeq ($(USE_OPENMP), 1)
+  ifeq ($(C_COMPILER), PGI)
+          PKG_EXTRALIB += -lomp
+  else	  
+          PKG_EXTRALIB += -lgomp
+  endif
+endif	  
 
 .PHONY : install
 .NOTPARALLEL : install
@@ -147,7 +155,7 @@ endif
 	@echo 'includedir='$(OPENBLAS_INCLUDE_DIR) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'openblas_config= USE_64BITINT='$(USE_64BITINT) 'DYNAMIC_ARCH='$(DYNAMIC_ARCH) 'DYNAMIC_OLDER='$(DYNAMIC_OLDER) 'NO_CBLAS='$(NO_CBLAS) 'NO_LAPACK='$(NO_LAPACK) 'NO_LAPACKE='$(NO_LAPACKE) 'NO_AFFINITY='$(NO_AFFINITY) 'USE_OPENMP='$(USE_OPENMP) $(CORE) 'MAX_THREADS='$(NUM_THREADS)>> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'version='$(VERSION) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
-	@echo 'extralib='$(EXTRALIB) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
+	@echo 'extralib='$(PKG_EXTRALIB) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@cat openblas.pc.in >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 
 

Makefile.system

@@ -25,6 +25,10 @@ else ifeq ($(ARCH), powerpc)
 override ARCH=power
 else ifeq ($(ARCH), i386)
 override ARCH=x86
+else ifeq ($(ARCH), armv6)
+override ARCH=arm
+else ifeq ($(ARCH), armv7)
+override ARCH=arm
 else ifeq ($(ARCH), aarch64)
 override ARCH=arm64
 else ifeq ($(ARCH), zarch)

cmake/openblas.pc.in

@@ -7,5 +7,5 @@ Name: OpenBLAS
 Description: OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version
 Version: @OPENBLAS_VERSION@
 URL: https://github.com/xianyi/OpenBLAS
-Libs: -L${libdir} -lopenblas${libsuffix}
+Libs: @OpenMP_C_FLAGS@ -L${libdir} -lopenblas${libsuffix} 
 Cflags: -I${includedir}

cmake/system_check.cmake

@@ -54,14 +54,14 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
   endif()
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "i686.*|i386.*|x86.*|amd64.*|AMD64.*")
   set(X86 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm.*|ARM.*)")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*|arm64.*|ARM64.*)")
-  set(ARM 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*)")
   if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
     set(ARM64 1)
   else()
     set(ARM 1)
   endif()
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm.*|ARM.*)")
+  set(ARM 1)
 elseif (${CMAKE_CROSSCOMPILING})
    if (${TARGET} STREQUAL "CORE2")
     if (NOT BINARY)

cpuid_arm64.c

@@ -197,6 +197,8 @@ int detect(void)
 
 
 	}
+#else
+	return CPU_ARMV8;	
 #endif
 
 	return CPU_UNKNOWN;

cpuid_power.c

@@ -145,7 +145,7 @@ int detect(void){
   if (implementation >= 0x40000u) return CPUTYPE_POWER10;
   else if (implementation & 0x20000) return CPUTYPE_POWER9;
   else if (implementation & 0x10000) return CPUTYPE_POWER8;
-  else if (implementation & 0x08000) return CPUTYPE_POWER7; // POWER 7
+  else if (implementation & 0x08000) return CPUTYPE_POWER6; // POWER 7
   else if (implementation & 0x04000) return CPUTYPE_POWER6;
   else if (implementation & 0x02000) return CPUTYPE_POWER5;
   else if (implementation & 0x01000) return CPUTYPE_POWER4; // MPC7450

driver/others/blas_server_omp.c

@@ -335,7 +335,7 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
       break;
   }
 
-#pragma omp parallel for schedule(OMP_SCHED)
+#pragma omp parallel for num_threads(num) schedule(OMP_SCHED)
   for (i = 0; i < num; i ++) {
 
 #ifndef USE_SIMPLE_THREADED_LEVEL3

kernel/power/dgemm_kernel_power10.c

@@ -87,22 +87,6 @@ typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
           rowC[0] += result[1] * alpha;
 #endif
 
-#define SET_ACC_ZERO4() \
-          __builtin_mma_xxsetaccz (&acc0); \
-          __builtin_mma_xxsetaccz (&acc1); \
-          __builtin_mma_xxsetaccz (&acc2); \
-          __builtin_mma_xxsetaccz (&acc3);
-
-#define SET_ACC_ZERO8() \
-          __builtin_mma_xxsetaccz (&acc0); \
-          __builtin_mma_xxsetaccz (&acc1); \
-          __builtin_mma_xxsetaccz (&acc2); \
-          __builtin_mma_xxsetaccz (&acc3); \
-          __builtin_mma_xxsetaccz (&acc4); \
-          __builtin_mma_xxsetaccz (&acc5); \
-          __builtin_mma_xxsetaccz (&acc6); \
-          __builtin_mma_xxsetaccz (&acc7);
-
 #define PREFETCH1(x, y) asm volatile ("dcbt %0, %1" : : "r" (x), "b" (y) : "memory");
 
 #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
@@ -210,12 +194,22 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  PREFETCH1 (CO + ldc + ldc, 128);
 	  PREFETCH1 (CO + ldc + ldc + ldc, 128);
 	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
-	  SET_ACC_ZERO8 ();
+	  vec_t *rowA = (vec_t *) & AO[0];
-	  for (l = 0; l < temp; l++)
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
+	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[2]);
+	  __builtin_mma_xvf64ger (&acc3, rowB, rowA[3]);
+	  __builtin_mma_xvf64ger (&acc4, rowB, rowA[4]);
+	  __builtin_mma_xvf64ger (&acc5, rowB, rowA[5]);
+	  __builtin_mma_xvf64ger (&acc6, rowB, rowA[6]);
+	  __builtin_mma_xvf64ger (&acc7, rowB, rowA[7]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      rowA = (vec_t *) & AO[l << 4];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & BO[l << 2];
-	      vec_t *rb = (vec_t *) & BO[l << 2];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
@@ -254,13 +248,19 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3;
-	  SET_ACC_ZERO4 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
+	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[2]);
+	  __builtin_mma_xvf64ger (&acc3, rowB, rowA[3]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      rowA = (vec_t *) & AO[l << 3];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & BO[l << 2];
-	      vec_t *rb = (vec_t *) & BO[l << 2];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
@@ -291,14 +291,17 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1;
-	  __builtin_mma_xxsetaccz (&acc0);
-	  __builtin_mma_xxsetaccz (&acc1);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      rowA = (vec_t *) & AO[l << 2];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & BO[l << 2];
-	      vec_t *rb = (vec_t *) & BO[l << 2];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
@@ -325,13 +328,16 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0;
-	  __builtin_mma_xxsetaccz (&acc0);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 1];
+	      rowA = (vec_t *) & AO[l << 1];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & BO[l << 2];
-	      vec_t *rb = (vec_t *) & BO[l << 2];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	    }
@@ -414,16 +420,27 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
-	  SET_ACC_ZERO8 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0, 0, 0, 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & t[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
+	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[2]);
+	  __builtin_mma_xvf64ger (&acc3, rowB, rowA[3]);
+	  __builtin_mma_xvf64ger (&acc4, rowB, rowA[4]);
+	  __builtin_mma_xvf64ger (&acc5, rowB, rowA[5]);
+	  __builtin_mma_xvf64ger (&acc6, rowB, rowA[6]);
+	  __builtin_mma_xvf64ger (&acc7, rowB, rowA[7]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0, 0, 0, 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & t[0];
-	      vec_t *rb = (vec_t *) & t[0];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      rowA = (vec_t *) & AO[l << 4];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
 	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[2]);
@@ -461,16 +478,23 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3;
-	  SET_ACC_ZERO4 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0, 0, 0, 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & t[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
+	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[2]);
+	  __builtin_mma_xvf64ger (&acc3, rowB, rowA[3]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0, 0, 0, 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & t[0];
-	      vec_t *rb = (vec_t *) & t[0];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      rowA = (vec_t *) & AO[l << 3];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
 	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[2]);
@@ -500,17 +524,21 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1;
-	  __builtin_mma_xxsetaccz (&acc0);
-	  __builtin_mma_xxsetaccz (&acc1);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0, 0, 0, 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & t[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0, 0, 0, 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & t[0];
-	      vec_t *rb = (vec_t *) & t[0];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      rowA = (vec_t *) & AO[l << 2];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
 	    }
@@ -536,16 +564,20 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0;
-	  __builtin_mma_xxsetaccz (&acc0);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0, 0, 0, 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  __vector_pair rowB;
+	  vec_t *rb = (vec_t *) & t[0];
+	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0, 0, 0, 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      __vector_pair rowB;
+	      rb = (vec_t *) & t[0];
-	      vec_t *rb = (vec_t *) & t[0];
 	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      vec_t *rowA = (vec_t *) & AO[l << 1];
+	      rowA = (vec_t *) & AO[l << 1];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	    }
 	  SAVE2x4_ACC (&acc0, 0);

kernel/power/sgemm_kernel_power10.c

@@ -134,21 +134,6 @@ typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
           __builtin_mma_xvf32gerpp (&acc5, rowB[i+1], rowA[j+2]); \
           __builtin_mma_xvf32gerpp (&acc6, rowB[i], rowA[j+3]); \
           __builtin_mma_xvf32gerpp (&acc7, rowB[i+1], rowA[j+3]);
-#define SET_ACC_ZERO4() \
-          __builtin_mma_xxsetaccz (&acc0); \
-          __builtin_mma_xxsetaccz (&acc1); \
-          __builtin_mma_xxsetaccz (&acc2); \
-          __builtin_mma_xxsetaccz (&acc3);
-
-#define SET_ACC_ZERO8() \
-          __builtin_mma_xxsetaccz (&acc0); \
-          __builtin_mma_xxsetaccz (&acc1); \
-          __builtin_mma_xxsetaccz (&acc2); \
-          __builtin_mma_xxsetaccz (&acc3); \
-          __builtin_mma_xxsetaccz (&acc4); \
-          __builtin_mma_xxsetaccz (&acc5); \
-          __builtin_mma_xxsetaccz (&acc6); \
-          __builtin_mma_xxsetaccz (&acc7);
 
 #define PREFETCH1(x, y) asm volatile ("dcbt %0, %1" : : "r" (x), "b" (y) : "memory");
 
@@ -249,8 +234,20 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
-	  SET_ACC_ZERO8 ();
 	  BLASLONG l = 0;
+	  vec_t *rowA1 = (vec_t *) & AO[0];
+	  vec_t *rowB1 = (vec_t *) & BO[0];
+          __builtin_mma_xvf32ger (&acc0, rowB1[0], rowA1[0]);
+          __builtin_mma_xvf32ger (&acc1, rowB1[1], rowA1[0]);
+          __builtin_mma_xvf32ger (&acc2, rowB1[0], rowA1[1]);
+          __builtin_mma_xvf32ger (&acc3, rowB1[1], rowA1[1]);
+          __builtin_mma_xvf32ger (&acc4, rowB1[0], rowA1[2]);
+          __builtin_mma_xvf32ger (&acc5, rowB1[1], rowA1[2]);
+          __builtin_mma_xvf32ger (&acc6, rowB1[0], rowA1[3]);
+          __builtin_mma_xvf32ger (&acc7, rowB1[1], rowA1[3]);
+	  AO += 16;
+	  BO += 8;
+	  temp--;
 	  BLASLONG K = temp / 64;
 	  for (l = 0; l < K; l++)
 	    {
@@ -454,12 +451,17 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3;
-	  SET_ACC_ZERO4 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[1], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc2, rowB[0], rowA[1]);
+	  __builtin_mma_xvf32ger (&acc3, rowB[1], rowA[1]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      rowA = (vec_t *) & AO[l << 3];
-	      vec_t *rowB = (vec_t *) & BO[l << 3];
+	      rowB = (vec_t *) & BO[l << 3];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[1], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[1]);
@@ -489,13 +491,15 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1;
-	  __builtin_mma_xxsetaccz (&acc0);
-	  __builtin_mma_xxsetaccz (&acc1);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[1], rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      rowA = (vec_t *) & AO[l << 2];
-	      vec_t *rowB = (vec_t *) & BO[l << 3];
+	      rowB = (vec_t *) & BO[l << 3];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[1], rowA[0]);
 	    }
@@ -522,15 +526,18 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v2sf_t *rowC;
 	  v2sf_t result[8];
 	  __vector_quad acc0, acc1;
-	  __builtin_mma_xxsetaccz (&acc0);
-	  __builtin_mma_xxsetaccz (&acc1);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0 };
+	  t[0] = AO[0], t[1] = AO[1];
+	  vec_t *rowA = (vec_t *) & t[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[1], rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0 };
 	      t[0] = AO[l << 1], t[1] = AO[(l << 1) + 1];
-	      vec_t *rowA = (vec_t *) & t[0];
+	      rowA = (vec_t *) & t[0];
-	      vec_t *rowB = (vec_t *) & BO[l << 3];
+	      rowB = (vec_t *) & BO[l << 3];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[1], rowA[0]);
 	    }
@@ -625,13 +632,23 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  FLOAT *A1;
 	  A1 = AO + (16 * k);
 	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
-	  SET_ACC_ZERO8 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < k; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowA1 = (vec_t *) & A1[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[0], rowA[1]);
+	  __builtin_mma_xvf32ger (&acc2, rowB[0], rowA[2]);
+	  __builtin_mma_xvf32ger (&acc3, rowB[0], rowA[3]);
+	  __builtin_mma_xvf32ger (&acc4, rowB[0], rowA1[0]);
+	  __builtin_mma_xvf32ger (&acc5, rowB[0], rowA1[1]);
+	  __builtin_mma_xvf32ger (&acc6, rowB[0], rowA1[2]);
+	  __builtin_mma_xvf32ger (&acc7, rowB[0], rowA1[3]);
+	  for (l = 1; l < k; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      rowA = (vec_t *) & AO[l << 4];
-	      vec_t *rowA1 = (vec_t *) & A1[l << 4];
+	      rowA1 = (vec_t *) & A1[l << 4];
-	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      rowB = (vec_t *) & BO[l << 2];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
 	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
@@ -673,12 +690,17 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3;
-	  SET_ACC_ZERO4 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[0], rowA[1]);
+	  __builtin_mma_xvf32ger (&acc2, rowB[0], rowA[2]);
+	  __builtin_mma_xvf32ger (&acc3, rowB[0], rowA[3]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      rowA = (vec_t *) & AO[l << 4];
-	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      rowB = (vec_t *) & BO[l << 2];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
 	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
@@ -710,13 +732,15 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1;
-	  __builtin_mma_xxsetaccz (&acc0);
-	  __builtin_mma_xxsetaccz (&acc1);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[0], rowA[1]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      rowA = (vec_t *) & AO[l << 3];
-	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      rowB = (vec_t *) & BO[l << 2];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
 	    }
@@ -742,12 +766,14 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  __vector_quad acc0;
 	  v4sf_t result[4];
-	  __builtin_mma_xxsetaccz (&acc0);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      rowA = (vec_t *) & AO[l << 2];
-	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      rowB = (vec_t *) & BO[l << 2];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	    }
 	  SAVE_ACC (&acc0, 0);
@@ -771,14 +797,17 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v2sf_t *rowC;
 	  v2sf_t result[8];
 	  __vector_quad acc0;
-	  __builtin_mma_xxsetaccz (&acc0);
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0 };
+	  t[0] = AO[0], t[1] = AO[1];
+	  vec_t *rowA = (vec_t *) & t[0];
+	  vec_t *rowB = (vec_t *) & BO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0 };
 	      t[0] = AO[l << 1], t[1] = AO[(l << 1) + 1];
-	      vec_t *rowA = (vec_t *) & t[0];
+	      rowA = (vec_t *) & t[0];
-	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      rowB = (vec_t *) & BO[l << 2];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	    }
 	  SAVE4x2_ACC (&acc0, 0);
@@ -856,15 +885,26 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  FLOAT *A1;
 	  A1 = AO + (16 * k);
 	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
-	  SET_ACC_ZERO8 ();
 	  BLASLONG l = 0;
-	  for (l = 0; l < k; l++)
+	  FLOAT t[4] = { 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  vec_t *rowB = (vec_t *) & t[0];
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  vec_t *rowA1 = (vec_t *) & A1[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[0], rowA[1]);
+	  __builtin_mma_xvf32ger (&acc2, rowB[0], rowA[2]);
+	  __builtin_mma_xvf32ger (&acc3, rowB[0], rowA[3]);
+	  __builtin_mma_xvf32ger (&acc4, rowB[0], rowA1[0]);
+	  __builtin_mma_xvf32ger (&acc5, rowB[0], rowA1[1]);
+	  __builtin_mma_xvf32ger (&acc6, rowB[0], rowA1[2]);
+	  __builtin_mma_xvf32ger (&acc7, rowB[0], rowA1[3]);
+	  for (l = 1; l < k; l++)
 	    {
-	      FLOAT t[4] = { 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      vec_t *rowB = (vec_t *) & t[0];
+	      rowB = (vec_t *) & t[0];
-	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      rowA = (vec_t *) & AO[l << 4];
-	      vec_t *rowA1 = (vec_t *) & A1[l << 4];
+	      rowA1 = (vec_t *) & A1[l << 4];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
 	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
@@ -897,7 +937,6 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3;
-	  SET_ACC_ZERO4 ();
 	  BLASLONG l = 0;
 #if defined(TRMMKERNEL)
 	  REFRESH_POINTERS (16, 2)
@@ -905,12 +944,19 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  BO = B;
 	  temp = k;
 #endif
-	  for (l = 0; l < temp; l++)
+	 FLOAT t[4] = { 0 };
+	 t[0] = BO[0], t[1] = BO[1];
+	 vec_t *rowB = (vec_t *) & t[0];
+	 vec_t *rowA = (vec_t *) & AO[0];
+	 __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	 __builtin_mma_xvf32ger (&acc1, rowB[0], rowA[1]);
+	 __builtin_mma_xvf32ger (&acc2, rowB[0], rowA[2]);
+	 __builtin_mma_xvf32ger (&acc3, rowB[0], rowA[3]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      vec_t *rowB = (vec_t *) & t[0];
+	      rowB = (vec_t *) & t[0];
-	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      rowA = (vec_t *) & AO[l << 4];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
 	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
@@ -934,8 +980,6 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1;
-	  __builtin_mma_xxsetaccz (&acc0);
-	  __builtin_mma_xxsetaccz (&acc1);
 #if defined(TRMMKERNEL)
 	  REFRESH_POINTERS (8, 2)
 #else
@@ -943,12 +987,17 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  temp = k;
 #endif
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  vec_t *rowB = (vec_t *) & t[0];
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  __builtin_mma_xvf32ger (&acc1, rowB[0], rowA[1]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      vec_t *rowB = (vec_t *) & t[0];
+	      rowB = (vec_t *) & t[0];
-	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      rowA = (vec_t *) & AO[l << 3];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
 	    }
@@ -968,7 +1017,6 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t *rowC;
 	  v4sf_t result[4];
 	  __vector_quad acc0;
-	  __builtin_mma_xxsetaccz (&acc0);
 #if defined(TRMMKERNEL)
 	  REFRESH_POINTERS (4, 2)
 #else
@@ -976,12 +1024,16 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  temp = k;
 #endif
 	  BLASLONG l = 0;
-	  for (l = 0; l < temp; l++)
+	  FLOAT t[4] = { 0 };
+	  t[0] = BO[0], t[1] = BO[1];
+	  vec_t *rowB = (vec_t *) & t[0];
+	  vec_t *rowA = (vec_t *) & AO[0];
+	  __builtin_mma_xvf32ger (&acc0, rowB[0], rowA[0]);
+	  for (l = 1; l < temp; l++)
 	    {
-	      FLOAT t[4] = { 0 };
 	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      vec_t *rowB = (vec_t *) & t[0];
+	      rowB = (vec_t *) & t[0];
-	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      rowA = (vec_t *) & AO[l << 2];
 	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
 	    }
 	  SAVE2x4_ACC (&acc0, 0);

kernel/x86_64/zdot.c

@@ -168,7 +168,7 @@ static void zdot_compute (BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLO
 
 #if defined(SMP)
 static int zdot_thread_function(BLASLONG n, BLASLONG dummy0,
-BLASLONG dummy1, FLOAT dummy2, FLOAT *x, BLASLONG inc_x, FLOAT *y,
+BLASLONG dummy1, FLOAT dummy2r, FLOAT dummy2i, FLOAT *x, BLASLONG inc_x, FLOAT *y,
 BLASLONG inc_y, FLOAT *result, BLASLONG dummy3)
 {
         zdot_compute(n, x, inc_x, y, inc_y, (void *)result);

lapack-netlib/LAPACKE/include/lapack.h

@@ -3650,45 +3650,45 @@ void LAPACK_zggrqf(
     lapack_int* info );
 
 #define LAPACK_sggsvd LAPACK_GLOBAL(sggsvd,SGGSVD)
-lapack_int LAPACKE_sggsvd( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_sggsvd( char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* m, lapack_int* n, lapack_int* p,
                            lapack_int* k, lapack_int* l, float* a,
-                           lapack_int lda, float* b, lapack_int ldb,
+                           lapack_int* lda, float* b, lapack_int* ldb,
-                           float* alpha, float* beta, float* u, lapack_int ldu,
+                           float* alpha, float* beta, float* u, lapack_int* ldu,
-                           float* v, lapack_int ldv, float* q, lapack_int ldq,
+                           float* v, lapack_int* ldv, float* q, lapack_int* ldq,
-                           lapack_int* iwork );
+                           float* work, lapack_int* iwork, lapack_int* info );
 
 #define LAPACK_dggsvd LAPACK_GLOBAL(dggsvd,DGGSVD)
-lapack_int LAPACKE_dggsvd( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_dggsvd( char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* m, lapack_int* n, lapack_int* p,
                            lapack_int* k, lapack_int* l, double* a,
-                           lapack_int lda, double* b, lapack_int ldb,
+                           lapack_int* lda, double* b, lapack_int* ldb,
                            double* alpha, double* beta, double* u,
-                           lapack_int ldu, double* v, lapack_int ldv, double* q,
+                           lapack_int* ldu, double* v, lapack_int* ldv, double* q,
-                           lapack_int ldq, lapack_int* iwork );
+                           lapack_int* ldq, float* work, lapack_int* iwork, lapack_int* info );
 
 #define LAPACK_cggsvd LAPACK_GLOBAL(cggsvd,CGGSVD)
-lapack_int LAPACKE_cggsvd( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_cggsvd( char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* m, lapack_int* n, lapack_int* p,
                            lapack_int* k, lapack_int* l,
-                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* a, lapack_int* lda,
-                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* b, lapack_int* ldb,
                            float* alpha, float* beta, lapack_complex_float* u,
-                           lapack_int ldu, lapack_complex_float* v,
+                           lapack_int* ldu, lapack_complex_float* v,
-                           lapack_int ldv, lapack_complex_float* q,
+                           lapack_int* ldv, lapack_complex_float* q,
-                           lapack_int ldq, lapack_int* iwork );
+                           lapack_int* ldq, float* work, lapack_int* rwork, lapack_int* iwork, lapack_int *info );
 
 #define LAPACK_zggsvd LAPACK_GLOBAL(zggsvd,ZGGSVD)
-lapack_int LAPACKE_zggsvd( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_zggsvd( char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* m, lapack_int* n, lapack_int* p,
                            lapack_int* k, lapack_int* l,
-                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* a, lapack_int* lda,
-                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* b, lapack_int* ldb,
                            double* alpha, double* beta,
-                           lapack_complex_double* u, lapack_int ldu,
+                           lapack_complex_double* u, lapack_int* ldu,
-                           lapack_complex_double* v, lapack_int ldv,
+                           lapack_complex_double* v, lapack_int* ldv,
-                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* q, lapack_int* ldq,
-                           lapack_int* iwork );
+                           float* work, lapack_int* rwork, lapack_int* iwork, lapack_int* info );
 
 #define LAPACK_cggsvd3 LAPACK_GLOBAL(cggsvd3,CGGSVD3)
 void LAPACK_cggsvd3(
@@ -3753,41 +3753,49 @@ void LAPACK_zggsvd3(
     lapack_int* info );
 
 #define LAPACK_sggsvp LAPACK_GLOBAL(sggsvp,SGGSVP)
-lapack_int LAPACKE_sggsvp( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_sggsvp(  char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int p, lapack_int n, float* a,
+                           lapack_int* m, lapack_int* p, lapack_int* n, float* a,
-                           lapack_int lda, float* b, lapack_int ldb, float tola,
+                           lapack_int* lda, float* b, lapack_int* ldb, float* tola,
-                           float tolb, lapack_int* k, lapack_int* l, float* u,
+                           float* tolb, lapack_int* k, lapack_int* l, float* u,
-                           lapack_int ldu, float* v, lapack_int ldv, float* q,
+                           lapack_int* ldu, float* v, lapack_int* ldv, float* q,
-                           lapack_int ldq );
+                           lapack_int* ldq, lapack_int* iwork, float* tau, 
+                           float* work, lapack_int* info);
 
 #define LAPACK_dggsvp LAPACK_GLOBAL(dggsvp,DGGSVP)
-lapack_int LAPACKE_dggsvp( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_dggsvp(  char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int p, lapack_int n, double* a,
+                           lapack_int* m, lapack_int* p, lapack_int* n, double* a,
-                           lapack_int lda, double* b, lapack_int ldb,
+                           lapack_int* lda, double* b, lapack_int* ldb,
-                           double tola, double tolb, lapack_int* k,
+                           double* tola, double* tolb, lapack_int* k,
-                           lapack_int* l, double* u, lapack_int ldu, double* v,
+                           lapack_int* l, double* u, lapack_int* ldu, double* v,
-                           lapack_int ldv, double* q, lapack_int ldq );
+                           lapack_int* ldv, double* q, lapack_int* ldq,
+                           lapack_int* iwork, double* tau, double* work,
+                           lapack_int* info);
 
 #define LAPACK_cggsvp LAPACK_GLOBAL(cggsvp,CGGSVP)
-lapack_int LAPACKE_cggsvp( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_cggsvp(  char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int* m, lapack_int* p, lapack_int* n,
-                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* a, lapack_int* lda,
-                           lapack_complex_float* b, lapack_int ldb, float tola,
+                           lapack_complex_float* b, lapack_int* ldb, float* tola,
-                           float tolb, lapack_int* k, lapack_int* l,
+                           float* tolb, lapack_int* k, lapack_int* l,
-                           lapack_complex_float* u, lapack_int ldu,
+                           lapack_complex_float* u, lapack_int* ldu,
-                           lapack_complex_float* v, lapack_int ldv,
+                           lapack_complex_float* v, lapack_int* ldv,
-                           lapack_complex_float* q, lapack_int ldq );
+                           lapack_complex_float* q, lapack_int* ldq,
+                           lapack_int* iwork, lapack_int* rwork,
+                           lapack_complex_float* tau, lapack_complex_float* work,
+                           lapack_int* info);
 
 #define LAPACK_zggsvp LAPACK_GLOBAL(zggsvp,ZGGSVP)
-lapack_int LAPACKE_zggsvp( int matrix_layout, char jobu, char jobv, char jobq,
+lapack_int LAPACK_zggsvp(  char const* jobu, char const* jobv, char const* jobq,
-                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int* m, lapack_int* p, lapack_int* n,
-                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* a, lapack_int* lda,
-                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* b, lapack_int* ldb,
-                           double tola, double tolb, lapack_int* k,
+                           double* tola, double* tolb, lapack_int* k,
                            lapack_int* l, lapack_complex_double* u,
-                           lapack_int ldu, lapack_complex_double* v,
+                           lapack_int* ldu, lapack_complex_double* v,
-                           lapack_int ldv, lapack_complex_double* q,
+                           lapack_int* ldv, lapack_complex_double* q,
-                           lapack_int ldq );
+                           lapack_int* ldq, lapack_int* iwork, lapack_int* rwork,
+                           lapack_complex_double* tau, lapack_complex_double* work,
+                           lapack_int* info);
 
 #define LAPACK_cggsvp3 LAPACK_GLOBAL(cggsvp3,CGGSVP3)
 void LAPACK_cggsvp3(