Merge pull request #108 from xianyi/develop

rebase

Makefile.x86_64

@@ -90,7 +90,7 @@ ifeq ($(F_COMPILER), GFORTRAN)
 GCCVERSIONGTEQ4 := $(shell expr `$(FC) -dumpversion | cut -f1 -d.` \>= 4)
 GCCVERSIONGTEQ5 := $(shell expr `$(FC) -dumpversion | cut -f1 -d.` \>= 5)
 GCCMINORVERSIONGTEQ7 := $(shell expr `$(FC) -dumpversion | cut -f2 -d.` \>= 7)
-GCCVERSIONCHECK := $(GCCVERSIONGTEQ5)$(GCCVERSIONGTEQ4)$(GCCVERSIONMINORGTEQ7)
+GCCVERSIONCHECK := $(GCCVERSIONGTEQ5)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7)
 ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
 FCOMMON_OPT += -mavx2
 endif

common_power.h

@@ -844,8 +844,8 @@ Lmcount$lazy_ptr:
 #define BUFFER_SIZE     (  2 << 20)
 #elif defined(PPC440FP2)
 #define BUFFER_SIZE     ( 16 << 20)
-#elif defined(POWER8) || defined(POWER9) || defined(POWER10)
-#define BUFFER_SIZE     ( 64 << 20)
+#elif defined(POWER6) || defined(POWER8) || defined(POWER9) || defined(POWER10)
+#define BUFFER_SIZE     ( 64 << 22)
 #else
 #define BUFFER_SIZE     ( 16 << 20)
 #endif

kernel/Makefile

@@ -22,20 +22,25 @@ ifeq ($(C_COMPILER), CLANG)
  override CFLAGS += -fno-integrated-as
 endif
 endif
+
 AVX2OPT = 
 ifeq ($(C_COMPILER), GCC)
 # AVX2 support was added in 4.7.0
-  GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
-  GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
-  ifeq ($(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7), 11)
+GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+GCCVERSIONGTEQ5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 5)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+GCCVERSIONCHECK := $(GCCVERSIONGTEQ5)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7)
+ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
    AVX2OPT = -mavx2
   endif
 endif
 ifeq ($(C_COMPILER), CLANG)
 # Any clang posing as gcc 4.2 should be new enough (3.4 or later)
   GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+  GCCVERSIONGTEQ5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 5)
   GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 2)
-  ifeq ($(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ2), 11)
+  GCCVERSIONCHECK := $(GCCVERSIONGTEQ5)$(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7)
+  ifeq ($(GCCVERSIONCHECK), $(filter $(GCCVERSIONCHECK), 011 110 111))
    AVX2OPT = -mavx2
   endif
 endif

kernel/power/zgemv_t_4.c

@@ -513,7 +513,7 @@ static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT al
 
 #endif
 
-static __attribute__((always_inline)) void copy_x(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_src) {
+static __attribute__((always_inline)) inline void copy_x(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_src) {
     BLASLONG i;
     for (i = 0; i < n; i++) {
         *dest = *src;

lapack-netlib/LAPACKE/src/lapacke_zgesvdq.c

@@ -71,7 +71,7 @@ lapack_int LAPACKE_zgesvdq( int matrix_layout, char joba, char jobp,
         goto exit_level_0;
     }
     liwork = iwork_query;
-    lcwork = LAPACK_C2INT(cwork_query);
+    lcwork = LAPACK_Z2INT(cwork_query);
     lrwork = (lapack_int)rwork_query;
     /* Allocate memory for work arrays */
     iwork = (lapack_int*)LAPACKE_malloc( sizeof(lapack_int) * liwork );