Merge remote branch 'origin/develop' into armv7

19 files changed, 87 insertions, 316 deletions

diff --git a/Makefile b/Makefile
index 67d425359..294502f81 100644
--- a/Makefile
+++ b/Makefile
@@ -219,10 +219,10 @@ prof_lapack : lapack_prebuild
 lapack_prebuild :
 ifndef NOFORTRAN
 	-@echo "FORTRAN     = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "OPTS        = $(FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "POPTS       = $(FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "NOOPT       = $(FFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "PNOOPT      = $(FPFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "OPTS        = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "NOOPT       = $(LAPACK_FFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "PNOOPT      = $(LAPACK_FPFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "LOADOPTS    = $(FFLAGS) $(EXTRALIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "CC          = $(CC)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "override CFLAGS      = $(LAPACK_CFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
diff --git a/Makefile.install b/Makefile.install
index cbe98bc5b..8319b46db 100644
--- a/Makefile.install
+++ b/Makefile.install
@@ -23,8 +23,8 @@ install : 	lib.grd
 #for inc 
 	@echo \#ifndef OPENBLAS_CONFIG_H > $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 	@echo \#define OPENBLAS_CONFIG_H >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
-	@cat config_last.h >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
-	@echo \#define VERSION \" OpenBLAS $(VERSION) \" >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
+	@awk '{print $$1, "OPENBLAS_"$$2, $$3}' config_last.h >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
+	@echo \#define OPENBLAS_VERSION \" OpenBLAS $(VERSION) \" >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 	@cat openblas_config_template.h >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 	@echo \#endif  \/\* OPENBLAS_CONFIG_H \*\/ >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 
diff --git a/Makefile.system b/Makefile.system
index 5545de1b1..d6c172f3d 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -222,6 +222,11 @@ endif
 endif
 endif
 
+# ifeq logical or
+ifeq ($(OSNAME), $(filter $(OSNAME),WINNT CYGWIN_NT Interix))
+OS_WINDOWS=1
+endif
+
 ifdef QUAD_PRECISION
 CCOMMON_OPT	+= -DQUAD_PRECISION
 NO_EXPRECISION = 1
@@ -477,10 +482,8 @@ CCOMMON_OPT += -DF_INTERFACE_GFORT
 FCOMMON_OPT += -Wall
 #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
 ifneq ($(NO_LAPACK), 1)
-ifneq ($(C_COMPILER), LSB)
 EXTRALIB += -lgfortran 
 endif
-endif
 ifdef NO_BINARY_MODE
 ifeq ($(ARCH), mips64)
 ifdef BINARY64
@@ -861,11 +864,18 @@ override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)
 override FPFLAGS    += $(COMMON_OPT) $(FCOMMON_OPT) $(COMMON_PROF)
 #MAKEOVERRIDES =
 
+#For LAPACK Fortran codes.
+LAPACK_FFLAGS := $(filter-out -fopenmp -mp -openmp -xopenmp=parallel,$(FFLAGS))
+LAPACK_FPFLAGS := $(filter-out -fopenmp -mp -openmp -xopenmp=parallel,$(FPFLAGS))
+
 LAPACK_CFLAGS = $(CFLAGS)
 LAPACK_CFLAGS += -DHAVE_LAPACK_CONFIG_H 
 ifdef INTERFACE64
 LAPACK_CFLAGS +=  -DLAPACK_ILP64
 endif
+ifdef OS_WINDOWS
+LAPACK_CFLAGS +=  -DOPENBLAS_OS_WINDOWS
+endif
 ifeq ($(C_COMPILER), LSB)
 LAPACK_CFLAGS +=  -DLAPACK_COMPLEX_STRUCTURE
 endif
diff --git a/Makefile.tail b/Makefile.tail
index 53dd0caad..56f8d820c 100644
--- a/Makefile.tail
+++ b/Makefile.tail
@@ -606,7 +606,8 @@ clean ::
 	@if test -d $(ARCH); then \
 	(cd $(ARCH) && $(MAKE) clean) \
 	fi
-	@rm -rf *.a *.s *.o *.po *.obj *.i *.so core core.* gmon.out *.cso \
+	@find . -name '*.o' | xargs rm -rf
+	@rm -rf *.a *.s *.po *.obj *.i *.so core core.* gmon.out *.cso \
 	*.csx *.is *~ *.exe *.flame *.pdb *.dwf \
 	gen_insn_flash.c gen_insn_flash *.stackdump *.dll *.exp *.lib \
 	*.pc *.pcl *.def *.i *.prof linktest.c \
diff --git a/driver/others/blas_server_win32.c b/driver/others/blas_server_win32.c
index bd1069c5e..8723a6fa7 100644
--- a/driver/others/blas_server_win32.c
+++ b/driver/others/blas_server_win32.c
@@ -441,9 +441,10 @@ int BLASFUNC(blas_thread_shutdown)(void){
   if (blas_server_avail){
 
     SetEvent(pool.killed);
-    
+    printf("blas_num_threads=%d\n", blas_num_threads);
     for(i = 0; i < blas_num_threads - 1; i++){
-      WaitForSingleObject(blas_threads[i], INFINITE);
+     WaitForSingleObject(blas_threads[i], 5);  //INFINITE);
+	 TerminateThread(blas_threads[i],0);
     }
     
     blas_server_avail = 0;
diff --git a/driver/others/memory.c b/driver/others/memory.c
index 4f35691ff..35758d13c 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -363,7 +363,7 @@ static void *alloc_mmap(void *address){
 #define BENCH_ITERATION 4
 #define SCALING		2
 
-static inline BLASULONG run_bench(BLASULONG address, long size) {
+static inline BLASULONG run_bench(BLASULONG address, BLASULONG size) {
 
   BLASULONG original, *p;
   BLASULONG start, stop, min;
@@ -450,12 +450,12 @@ static void *alloc_mmap(void *address){
 	current = (SCALING - 1) * BUFFER_SIZE;
 	
 	while(current > 0) {
-	  *(long *)start = (long)start + PAGESIZE;
+	  *(BLASLONG *)start = (BLASLONG)start + PAGESIZE;
 	  start += PAGESIZE;
 	  current -= PAGESIZE;
 	}
 	
-	*(long *)(start - PAGESIZE) = (BLASULONG)map_address;
+	*(BLASLONG *)(start - PAGESIZE) = (BLASULONG)map_address;
 	
 	start = (BLASULONG)map_address;
 	
@@ -1170,7 +1170,7 @@ static void _touch_memory(blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n,
 
 #if !defined(ARCH_POWER) && !defined(ARCH_SPARC)
 
-  long size;
+  size_t size;
   BLASULONG buffer;
 
   size   = BUFFER_SIZE - PAGESIZE;
diff --git a/exports/Makefile b/exports/Makefile
index 0bc9ec6e0..8e50a9809 100644
--- a/exports/Makefile
+++ b/exports/Makefile
@@ -111,7 +111,7 @@ libgoto_hpl.def : gensymbol
 	perl ./gensymbol win2khpl $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) > $(@F)
 
 $(LIBDYNNAME) : ../$(LIBNAME) osx.def
-	$(CC) $(CFLAGS) -all_load -headerpad_max_install_names -install_name $(CURDIR)/../$(LIBDYNNAME) -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(FC) $(FFLAGS) -all_load -headerpad_max_install_names -install_name $(CURDIR)/../$(LIBDYNNAME) -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 
 symbol.$(SUFFIX) : symbol.S
 	$(CC) $(CFLAGS) -c -o $(@F) $^
@@ -124,14 +124,17 @@ ifeq ($(OSNAME), Linux)
 so : ../$(LIBSONAME)
 
 ../$(LIBSONAME) : ../$(LIBNAME) linux.def linktest.c
+ifneq ($(C_COMPILER), LSB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive ../$(LIBNAME) -Wl,--no-whole-archive \
 	-Wl,-soname,$(LIBPREFIX).so.$(MAJOR_VERSION) $(EXTRALIB)
-ifneq ($(C_COMPILER), LSB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 else
-#Use FC on LSB
-	$(FC) $(FFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
+#for LSB
+	env LSBCC_SHAREDLIBS=gfortran $(CC) $(CFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
+	-Wl,--whole-archive ../$(LIBNAME) -Wl,--no-whole-archive \
+	-Wl,-soname,$(LIBPREFIX).so.$(MAJOR_VERSION) $(EXTRALIB)
+	$(FC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 endif
 	rm -f linktest
 
diff --git a/interface/trtri.c b/interface/trtri.c
index 007dbd7fa..5aa5e9b9b 100644
--- a/interface/trtri.c
+++ b/interface/trtri.c
@@ -60,7 +60,6 @@ static blasint (*trtri_parallel[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *
 };
 #endif
 
-extern void BLASFUNC(dtrtrilapack)(char *UPLO, char *DIAG, int *N, double *a, int *ldA, int *Info);
 
 int NAME(char *UPLO, char *DIAG, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
 
@@ -133,18 +132,6 @@ int NAME(char *UPLO, char *DIAG, blasint *N, FLOAT *a, blasint *ldA, blasint *In
   if (args.nthreads == 1) {
 #endif
 
-#if DOUBLE
-    // double trtri_U single thread error
-    // call dtrtri from lapack for a walk around.
-    if(uplo==0){
-      BLASFUNC(dtrtrilapack)(UPLO, DIAG, N, a, ldA, Info);
-#ifndef PPC440
-      blas_memory_free(buffer);
-#endif
-      return 0;
-    }
-#endif
-
     *Info = (trtri_single[(uplo << 1) | diag])(&args, NULL, NULL, sa, sb, 0);
     
 #ifdef SMP
diff --git a/kernel/x86_64/dtrsm_kernel_LT_8x2_bulldozer.S b/kernel/x86_64/dtrsm_kernel_LT_8x2_bulldozer.S
index 374f45096..9e15fa240 100644
--- a/kernel/x86_64/dtrsm_kernel_LT_8x2_bulldozer.S
+++ b/kernel/x86_64/dtrsm_kernel_LT_8x2_bulldozer.S
@@ -103,7 +103,7 @@
 	vmovups		-10*SIZE(AO,%rax,8), %xmm6
 	vfmaddpd	%xmm14, %xmm6 , %xmm1 , %xmm14
 	vfmaddpd	%xmm15, %xmm6 , %xmm2 , %xmm15
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_8x2
@@ -265,7 +265,7 @@
 	vmovups		-14*SIZE(AO,%rax,4), %xmm0
 	vfmaddpd	%xmm10, %xmm0 , %xmm1 , %xmm10
 	vfmaddpd	%xmm11, %xmm0 , %xmm2 , %xmm11
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -338,7 +338,7 @@
 	vmovups		-16*SIZE(AO,%rax,2), %xmm0
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
 	vfmaddpd	%xmm9 , %xmm0 , %xmm2 , %xmm9
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -378,7 +378,7 @@
 	vmovups 	-16*SIZE(BO,%rax,2), %xmm1
 	vmovddup	-16*SIZE(AO,%rax,1), %xmm0
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_1x2
@@ -411,7 +411,7 @@
 	vfmaddpd	%xmm10, %xmm0 , %xmm1 , %xmm10
 	vmovups		-10*SIZE(AO,%rax,8), %xmm0
 	vfmaddpd	%xmm11, %xmm0 , %xmm1 , %xmm11
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_8x1
@@ -510,7 +510,7 @@
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
 	vmovups		-14*SIZE(AO,%rax,4), %xmm0
 	vfmaddpd	%xmm9 , %xmm0 , %xmm1 , %xmm9
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -560,7 +560,7 @@
 	vmovddup	-16*SIZE(BO,%rax,1), %xmm1
 	vmovups		-16*SIZE(AO,%rax,2), %xmm0
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -592,7 +592,7 @@
 	vmovsd  	-16*SIZE(BO,%rax,1), %xmm1
 	vmovsd 		-16*SIZE(AO,%rax,1), %xmm0
 	vfmaddsd	%xmm8 , %xmm0 , %xmm1 , %xmm8
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_1x1
diff --git a/kernel/x86_64/dtrsm_kernel_RN_8x2_bulldozer.S b/kernel/x86_64/dtrsm_kernel_RN_8x2_bulldozer.S
index 8fa53efa7..8d3964aee 100644
--- a/kernel/x86_64/dtrsm_kernel_RN_8x2_bulldozer.S
+++ b/kernel/x86_64/dtrsm_kernel_RN_8x2_bulldozer.S
@@ -103,7 +103,7 @@
 	vmovups		-10*SIZE(AO,%rax,8), %xmm6
 	vfmaddpd	%xmm14, %xmm6 , %xmm1 , %xmm14
 	vfmaddpd	%xmm15, %xmm6 , %xmm2 , %xmm15
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_8x2
@@ -177,7 +177,7 @@
 	vmovups		-14*SIZE(AO,%rax,4), %xmm0
 	vfmaddpd	%xmm10, %xmm0 , %xmm1 , %xmm10
 	vfmaddpd	%xmm11, %xmm0 , %xmm2 , %xmm11
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -226,7 +226,7 @@
 	vmovups		-16*SIZE(AO,%rax,2), %xmm0
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
 	vfmaddpd	%xmm9 , %xmm0 , %xmm2 , %xmm9
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -262,7 +262,7 @@
 	vmovups 	-16*SIZE(BO,%rax,2), %xmm1
 	vmovddup	-16*SIZE(AO,%rax,1), %xmm0
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_1x2
@@ -306,7 +306,7 @@
 	vfmaddpd	%xmm10, %xmm0 , %xmm1 , %xmm10
 	vmovups		-10*SIZE(AO,%rax,8), %xmm0
 	vfmaddpd	%xmm11, %xmm0 , %xmm1 , %xmm11
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_8x1
@@ -347,7 +347,7 @@
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
 	vmovups		-14*SIZE(AO,%rax,4), %xmm0
 	vfmaddpd	%xmm9 , %xmm0 , %xmm1 , %xmm9
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -377,7 +377,7 @@
 	vmovddup	-16*SIZE(BO,%rax,1), %xmm1
 	vmovups		-16*SIZE(AO,%rax,2), %xmm0
 	vfmaddpd	%xmm8 , %xmm0 , %xmm1 , %xmm8
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 
@@ -402,7 +402,7 @@
 	vmovsd  	-16*SIZE(BO,%rax,1), %xmm1
 	vmovsd 		-16*SIZE(AO,%rax,1), %xmm0
 	vfmaddsd	%xmm8 , %xmm0 , %xmm1 , %xmm8
-	addq    $SIZE, %rax
+	addq    $ SIZE, %rax
 .endm
 
 .macro SOLVE_1x1
diff --git a/lapack-netlib/lapacke/include/lapacke_config.h b/lapack-netlib/lapacke/include/lapacke_config.h
index 1e2509bf0..561b2736b 100644
--- a/lapack-netlib/lapacke/include/lapacke_config.h
+++ b/lapack-netlib/lapacke/include/lapacke_config.h
@@ -45,7 +45,11 @@ extern "C" {
 
 #ifndef lapack_int
 #if defined(LAPACK_ILP64)
+#if defined(OPENBLAS_OS_WINDOWS)
+#define lapack_int              long long
+#else
 #define lapack_int              long
+#endif
 #else
 #define lapack_int              int
 #endif
diff --git a/lapack/getrf/getrf_parallel.c b/lapack/getrf/getrf_parallel.c
index 21ea9d5f5..3dbc70e9d 100644
--- a/lapack/getrf/getrf_parallel.c
+++ b/lapack/getrf/getrf_parallel.c
@@ -67,14 +67,14 @@ double sqrt(double);
 #undef  GETRF_FACTOR
 #define GETRF_FACTOR 1.00
 
-static inline long FORMULA1(long M, long N, long IS, long BK, long T) {
+static inline BLASLONG FORMULA1(BLASLONG M, BLASLONG N, BLASLONG IS, BLASLONG BK, BLASLONG T) {
 
   double m = (double)(M - IS - BK);
   double n = (double)(N - IS - BK);
   double b = (double)BK;
   double a = (double)T;
 
-  return (long)((n + GETRF_FACTOR * m * b * (1. - a) / (b + m)) / a);
+  return (BLASLONG)((n + GETRF_FACTOR * m * b * (1. - a) / (b + m)) / a);
 
 }
 
@@ -111,7 +111,7 @@ static void inner_basic_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *ra
 
   if (args -> a == NULL) {
     TRSM_ILTCOPY(k, k, (FLOAT *)args -> b, lda, 0, sb);
-    sbb = (FLOAT *)((((long)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+    sbb = (FLOAT *)((((BLASULONG)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
   } else {
     sb  = (FLOAT *)args -> a;
   }
@@ -221,7 +221,7 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
 
   if (args -> a == NULL) {
     TRSM_ILTCOPY(k, k, (FLOAT *)args -> b, lda, 0, sb);
-    sbb = (FLOAT *)((((long)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+    sbb = (FLOAT *)((((BLASULONG)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
   } else {
     sb  = (FLOAT *)args -> a;
   }
@@ -448,7 +448,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 
   TRSM_ILTCOPY(bk, bk, a, lda, 0, sb);
 
-  sbb = (FLOAT *)((((long)(sb + bk * bk * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+  sbb = (FLOAT *)((((BLASULONG)(sb + bk * bk * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
   
   is = 0;
   num_cpu = 0;
@@ -685,7 +685,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
   if (width > n - init_bk) width = n - init_bk;
 
   if (width < init_bk) {
-    long temp;
+    BLASLONG temp;
 
     temp = FORMULA2(m, n, 0, init_bk, args -> nthreads);
     temp = (temp + GEMM_UNROLL_N - 1) & ~(GEMM_UNROLL_N - 1);
@@ -708,7 +708,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
   is = 0;
   num_cpu = 0;
 
-  sbb = (FLOAT *)((((long)(sb + GEMM_PQ * GEMM_PQ * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+  sbb = (FLOAT *)((((BLASULONG)(sb + GEMM_PQ * GEMM_PQ * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
 
   while (is < mn) {
 
diff --git a/lapack/getrf/getrf_parallel_omp.c b/lapack/getrf/getrf_parallel_omp.c
index 4922b9b52..6eda30a52 100644
--- a/lapack/getrf/getrf_parallel_omp.c
+++ b/lapack/getrf/getrf_parallel_omp.c
@@ -178,7 +178,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
     return info;
   }
   
-  sbb = (FLOAT *)((((long)(sb + blocking * blocking * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+  sbb = (FLOAT *)((((BLASULONG)(sb + blocking * blocking * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
 
   info = 0;
 
diff --git a/lapack/getrf/getrf_single.c b/lapack/getrf/getrf_single.c
index fcea0ae89..f1818ea97 100644
--- a/lapack/getrf/getrf_single.c
+++ b/lapack/getrf/getrf_single.c
@@ -82,7 +82,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
     return info;
   }
   
-  sbb = (FLOAT *)((((long)(sb + blocking * blocking * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+  sbb = (FLOAT *)((((BLASULONG)(sb + blocking * blocking * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
 
   info = 0;
 
diff --git a/lapack/potrf/potrf_parallel.c b/lapack/potrf/potrf_parallel.c
index eec9b6e05..11f7f533c 100644
--- a/lapack/potrf/potrf_parallel.c
+++ b/lapack/potrf/potrf_parallel.c
@@ -185,7 +185,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   div_n = ((m_to - m_from + DIVIDE_RATE - 1) / DIVIDE_RATE + GEMM_UNROLL_MN - 1) & ~(GEMM_UNROLL_MN - 1);
 
-  buffer[0] = (FLOAT *)((((long)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
+  buffer[0] = (FLOAT *)((((BLASULONG)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
   for (i = 1; i < DIVIDE_RATE; i++) {
     buffer[i] = buffer[i - 1] + GEMM_Q * div_n * COMPSIZE;
   }
diff --git a/lapack/trtri/Makefile b/lapack/trtri/Makefile
index 10d3cb7fd..626c47bbf 100644
--- a/lapack/trtri/Makefile
+++ b/lapack/trtri/Makefile
@@ -13,7 +13,6 @@ ZBLASOBJS = ztrtri_UU_single.$(SUFFIX) ztrtri_UN_single.$(SUFFIX) ztrtri_LU_sing
 
 XBLASOBJS = xtrtri_UU_single.$(SUFFIX) xtrtri_UN_single.$(SUFFIX) xtrtri_LU_single.$(SUFFIX) xtrtri_LN_single.$(SUFFIX)
 
-DBLASOBJS += dtrtri_lapack.$(SUFFIX)
 
 ifdef SMP
 SBLASOBJS += strtri_UU_parallel.$(SUFFIX) strtri_UN_parallel.$(SUFFIX) strtri_LU_parallel.$(SUFFIX) strtri_LN_parallel.$(SUFFIX)
@@ -54,9 +53,6 @@ dtrtri_UU_single.$(SUFFIX) : trtri_U_single.c
 dtrtri_UN_single.$(SUFFIX) : trtri_U_single.c
 	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -UUNIT $< -o $(@F)
 
-dtrtri_lapack.$(SUFFIX) : dtrtri_lapack.f
-	$(FC) -c $(FFLAGS) -UCOMPLEX -DDOUBLE -DUNIT $< -o $(@F)
-
 dtrtri_LU_single.$(SUFFIX) : trtri_L_single.c
 	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -DUNIT $< -o $(@F)
 
diff --git a/lapack/trtri/dtrtri_lapack.f b/lapack/trtri/dtrtri_lapack.f
deleted file mode 100644
index 8e9a08170..000000000
--- a/lapack/trtri/dtrtri_lapack.f
+++ /dev/null
@@ -1,242 +0,0 @@
-*> \brief \b DTRTRI
-*
-*  =========== DOCUMENTATION ===========
-*
-* Online html documentation available at 
-*            http://www.netlib.org/lapack/explore-html/ 
-*
-*> \htmlonly
-*> Download DTRTRI + dependencies 
-*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dtrtri.f"> 
-*> [TGZ]</a> 
-*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dtrtri.f"> 
-*> [ZIP]</a> 
-*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dtrtri.f"> 
-*> [TXT]</a>
-*> \endhtmlonly 
-*
-*  Definition:
-*  ===========
-*
-*       SUBROUTINE DTRTRI( UPLO, DIAG, N, A, LDA, INFO )
-* 
-*       .. Scalar Arguments ..
-*       CHARACTER          DIAG, UPLO
-*       INTEGER            INFO, LDA, N
-*       ..
-*       .. Array Arguments ..
-*       DOUBLE PRECISION   A( LDA, * )
-*       ..
-*  
-*
-*> \par Purpose:
-*  =============
-*>
-*> \verbatim
-*>
-*> DTRTRI computes the inverse of a real upper or lower triangular
-*> matrix A.
-*>
-*> This is the Level 3 BLAS version of the algorithm.
-*> \endverbatim
-*
-*  Arguments:
-*  ==========
-*
-*> \param[in] UPLO
-*> \verbatim
-*>          UPLO is CHARACTER*1
-*>          = 'U':  A is upper triangular;
-*>          = 'L':  A is lower triangular.
-*> \endverbatim
-*>
-*> \param[in] DIAG
-*> \verbatim
-*>          DIAG is CHARACTER*1
-*>          = 'N':  A is non-unit triangular;
-*>          = 'U':  A is unit triangular.
-*> \endverbatim
-*>
-*> \param[in] N
-*> \verbatim
-*>          N is INTEGER
-*>          The order of the matrix A.  N >= 0.
-*> \endverbatim
-*>
-*> \param[in,out] A
-*> \verbatim
-*>          A is DOUBLE PRECISION array, dimension (LDA,N)
-*>          On entry, the triangular matrix A.  If UPLO = 'U', the
-*>          leading N-by-N upper triangular part of the array A contains
-*>          the upper triangular matrix, and the strictly lower
-*>          triangular part of A is not referenced.  If UPLO = 'L', the
-*>          leading N-by-N lower triangular part of the array A contains
-*>          the lower triangular matrix, and the strictly upper
-*>          triangular part of A is not referenced.  If DIAG = 'U', the
-*>          diagonal elements of A are also not referenced and are
-*>          assumed to be 1.
-*>          On exit, the (triangular) inverse of the original matrix, in
-*>          the same storage format.
-*> \endverbatim
-*>
-*> \param[in] LDA
-*> \verbatim
-*>          LDA is INTEGER
-*>          The leading dimension of the array A.  LDA >= max(1,N).
-*> \endverbatim
-*>
-*> \param[out] INFO
-*> \verbatim
-*>          INFO is INTEGER
-*>          = 0: successful exit
-*>          < 0: if INFO = -i, the i-th argument had an illegal value
-*>          > 0: if INFO = i, A(i,i) is exactly zero.  The triangular
-*>               matrix is singular and its inverse can not be computed.
-*> \endverbatim
-*
-*  Authors:
-*  ========
-*
-*> \author Univ. of Tennessee 
-*> \author Univ. of California Berkeley 
-*> \author Univ. of Colorado Denver 
-*> \author NAG Ltd. 
-*
-*> \date November 2011
-*
-*> \ingroup doubleOTHERcomputational
-*
-*  =====================================================================
-      SUBROUTINE DTRTRILAPACK( UPLO, DIAG, N, A, LDA, INFO )
-*
-*  -- LAPACK computational routine (version 3.4.0) --
-*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
-*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-*     November 2011
-*
-*     .. Scalar Arguments ..
-      CHARACTER          DIAG, UPLO
-      INTEGER            INFO, LDA, N
-*     ..
-*     .. Array Arguments ..
-      DOUBLE PRECISION   A( LDA, * )
-*     ..
-*
-*  =====================================================================
-*
-*     .. Parameters ..
-      DOUBLE PRECISION   ONE, ZERO
-      PARAMETER          ( ONE = 1.0D+0, ZERO = 0.0D+0 )
-*     ..
-*     .. Local Scalars ..
-      LOGICAL            NOUNIT, UPPER
-      INTEGER            J, JB, NB, NN
-*     ..
-*     .. External Functions ..
-      LOGICAL            LSAME
-      INTEGER            ILAENV
-      EXTERNAL           LSAME, ILAENV
-*     ..
-*     .. External Subroutines ..
-      EXTERNAL           DTRMM, DTRSM, DTRTI2, XERBLA
-*     ..
-*     .. Intrinsic Functions ..
-      INTRINSIC          MAX, MIN
-*     ..
-*     .. Executable Statements ..
-*
-*     Test the input parameters.
-*
-      INFO = 0
-      UPPER = LSAME( UPLO, 'U' )
-      NOUNIT = LSAME( DIAG, 'N' )
-      IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN
-         INFO = -1
-      ELSE IF( .NOT.NOUNIT .AND. .NOT.LSAME( DIAG, 'U' ) ) THEN
-         INFO = -2
-      ELSE IF( N.LT.0 ) THEN
-         INFO = -3
-      ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
-         INFO = -5
-      END IF
-      IF( INFO.NE.0 ) THEN
-         CALL XERBLA( 'DTRTRI', -INFO )
-         RETURN
-      END IF
-*
-*     Quick return if possible
-*
-      IF( N.EQ.0 )
-     $   RETURN
-*
-*     Check for singularity if non-unit.
-*
-      IF( NOUNIT ) THEN
-         DO 10 INFO = 1, N
-            IF( A( INFO, INFO ).EQ.ZERO )
-     $         RETURN
-   10    CONTINUE
-         INFO = 0
-      END IF
-*
-*     Determine the block size for this environment.
-*
-      NB = ILAENV( 1, 'DTRTRI', UPLO // DIAG, N, -1, -1, -1 )
-      IF( NB.LE.1 .OR. NB.GE.N ) THEN
-*
-*        Use unblocked code
-*
-         CALL DTRTI2( UPLO, DIAG, N, A, LDA, INFO )
-      ELSE
-*
-*        Use blocked code
-*
-         IF( UPPER ) THEN
-*
-*           Compute inverse of upper triangular matrix
-*
-            DO 20 J = 1, N, NB
-               JB = MIN( NB, N-J+1 )
-*
-*              Compute rows 1:j-1 of current block column
-*
-               CALL DTRMM( 'Left', 'Upper', 'No transpose', DIAG, J-1,
-     $                     JB, ONE, A, LDA, A( 1, J ), LDA )
-               CALL DTRSM( 'Right', 'Upper', 'No transpose', DIAG, J-1,
-     $                     JB, -ONE, A( J, J ), LDA, A( 1, J ), LDA )
-*
-*              Compute inverse of current diagonal block
-*
-               CALL DTRTI2( 'Upper', DIAG, JB, A( J, J ), LDA, INFO )
-   20       CONTINUE
-         ELSE
-*
-*           Compute inverse of lower triangular matrix
-*
-            NN = ( ( N-1 ) / NB )*NB + 1
-            DO 30 J = NN, 1, -NB
-               JB = MIN( NB, N-J+1 )
-               IF( J+JB.LE.N ) THEN
-*
-*                 Compute rows j+jb:n of current block column
-*
-                  CALL DTRMM( 'Left', 'Lower', 'No transpose', DIAG,
-     $                        N-J-JB+1, JB, ONE, A( J+JB, J+JB ), LDA,
-     $                        A( J+JB, J ), LDA )
-                  CALL DTRSM( 'Right', 'Lower', 'No transpose', DIAG,
-     $                        N-J-JB+1, JB, -ONE, A( J, J ), LDA,
-     $                        A( J+JB, J ), LDA )
-               END IF
-*
-*              Compute inverse of current diagonal block
-*
-               CALL DTRTI2( 'Lower', DIAG, JB, A( J, J ), LDA, INFO )
-   30       CONTINUE
-         END IF
-      END IF
-*
-      RETURN
-*
-*     End of DTRTRI
-*
-      END
diff --git a/lapack/trtri/trtri_U_single.c b/lapack/trtri/trtri_U_single.c
index 72133d896..c79281cfb 100644
--- a/lapack/trtri/trtri_U_single.c
+++ b/lapack/trtri/trtri_U_single.c
@@ -127,8 +127,14 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 	    if (min_i > GEMM_P) min_i = GEMM_P;
 	    
 	  if (ls == i + bk) {
-	    NEG_TCOPY (bk, min_i, a + (is + i * lda) * COMPSIZE, lda, sa);
-	    
+	    //NEG_TCOPY (bk, min_i, a + (is + i * lda) * COMPSIZE, lda, sa);
+
+	    GEMM_BETA(min_i, bk, 0, dm1, 
+#ifdef COMPLEX
+		      ZERO, 
+#endif
+		      NULL, 0, NULL, 0, a + (is + i * lda) * COMPSIZE, lda);
+
 	    TRSM_KERNEL_RN(min_i, bk, bk, dm1, 
 #ifdef COMPLEX
 			   ZERO,
@@ -171,8 +177,13 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 	  min_i = i - is;
 	  if (min_i > GEMM_P) min_i = GEMM_P;
 	  
-	  NEG_TCOPY (bk, min_i, a + (is + i * lda) * COMPSIZE, lda, sa);
-	  
+	  //NEG_TCOPY (bk, min_i, a + (is + i * lda) * COMPSIZE, lda, sa);
+	  GEMM_BETA(min_i, bk, 0, dm1, 
+#ifdef COMPLEX
+		    ZERO, 
+#endif
+		    NULL, 0, NULL, 0, a + (is + i * lda) * COMPSIZE, lda);
+
 	  TRSM_KERNEL_RN(min_i, bk, bk, dm1, 
 #ifdef COMPLEX
 			 ZERO,
diff --git a/openblas_config_template.h b/openblas_config_template.h
index 1017caff9..3b3435b0e 100644
--- a/openblas_config_template.h
+++ b/openblas_config_template.h
@@ -1,8 +1,8 @@
 /*This is only for "make install" target.*/
 
-#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
-#define WINDOWS_ABI
-#define OS_WINDOWS
+#if defined(OPENBLAS_OS_WINNT) || defined(OPENBLAS_OS_CYGWIN_NT) || defined(OPENBLAS_OS_INTERIX)
+#define OPENBLAS_WINDOWS_ABI
+#define OPENBLAS_OS_WINDOWS
 
 #ifdef DOUBLE
 #define DOUBLE_DEFINED DOUBLE
@@ -10,23 +10,23 @@
 #endif
 #endif
 
-#ifdef NEEDBUNDERSCORE
+#ifdef OPENBLAS_NEEDBUNDERSCORE
 #define BLASFUNC(FUNC) FUNC##_
 #else
 #define BLASFUNC(FUNC) FUNC
 #endif
 
-#ifdef QUAD_PRECISION
+#ifdef OPENBLAS_QUAD_PRECISION
 typedef struct {
   unsigned long x[2];
 }  xdouble;
-#elif defined EXPRECISION
+#elif defined OPENBLAS_EXPRECISION
 #define xdouble long double
 #else
 #define xdouble double
 #endif
 
-#if defined(OS_WINDOWS) && defined(__64BIT__)
+#if defined(OPENBLAS_OS_WINDOWS) && defined(OPENBLAS___64BIT__)
 typedef long long BLASLONG;
 typedef unsigned long long BLASULONG;
 #else
@@ -34,7 +34,7 @@ typedef long BLASLONG;
 typedef unsigned long BLASULONG;
 #endif
 
-#ifdef USE64BITINT
+#ifdef OPENBLAS_USE64BITINT
 typedef BLASLONG blasint;
 #else
 typedef int blasint;