Merge remote-tracking branch 'upstream/develop' into dyn

.travis.yml

@@ -1,4 +1,119 @@
+# XXX: Precise is already deprecated, new default is Trusty.
+# https://blog.travis-ci.com/2017-07-11-trusty-as-default-linux-is-coming
+dist: precise
+sudo: false
 language: c
+compiler: gcc
+
+jobs:
+  include:
+    - &test-ubuntu
+      stage: test
+      addons:
+        apt:
+          packages:
+            - gfortran
+      before_script: &common-before
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+      script:
+        - set -e
+        - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+        - make -C test $COMMON_FLAGS $BTYPE
+        - make -C ctest $COMMON_FLAGS $BTYPE
+        - make -C utest $COMMON_FLAGS $BTYPE
+      env:
+        - TARGET_BOX=LINUX64
+        - BTYPE="BINARY=64"
+
+    - <<: *test-ubuntu
+      env:
+        - TARGET_BOX=LINUX64
+        - BTYPE="BINARY=64 USE_OPENMP=1"
+
+    - <<: *test-ubuntu
+      env:
+        - TARGET_BOX=LINUX64
+        - BTYPE="BINARY=64 INTERFACE64=1"
+
+    - <<: *test-ubuntu
+      addons:
+        apt:
+          packages:
+            - gcc-multilib
+            - gfortran-multilib
+      env:
+        - TARGET_BOX=LINUX32
+        - BTYPE="BINARY=32"
+
+    - stage: test
+      addons:
+        apt:
+          packages:
+            - binutils-mingw-w64-x86-64
+            - gcc-mingw-w64-x86-64
+            - gfortran-mingw-w64-x86-64
+      before_script: *common-before
+      script:
+        - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+      env:
+        - TARGET_BOX=WIN64
+        - BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
+
+    # Build & test on Alpine Linux inside chroot, i.e. on system with musl libc.
+    # These jobs needs sudo, so Travis runs them on VM-based infrastructure
+    # which is slower than container-based infrastructure used for jobs
+    # that don't require sudo.
+    - &test-alpine
+      stage: test
+      dist: trusty
+      sudo: true
+      language: minimal
+      before_install:
+        - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.6.0/alpine-chroot-install' \
+              && echo 'a827a4ba3d0817e7c88bae17fe34e50204983d1e  alpine-chroot-install' | sha1sum -c || exit 1"
+        - alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
+      install:
+        - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
+      before_script: *common-before
+      script:
+        - set -e
+        # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
+        - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+              CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
+        - alpine make -C test $COMMON_FLAGS $BTYPE
+        - alpine make -C ctest $COMMON_FLAGS $BTYPE
+        - alpine make -C utest $COMMON_FLAGS $BTYPE
+      env:
+        - TARGET_BOX=LINUX64_MUSL
+        - BTYPE="BINARY=64"
+
+    # XXX: This job segfaults in TESTS OF THE COMPLEX LEVEL 3 BLAS,
+    # so it's "allowed to fail" for now (see allow_failures).
+    - &test-alpine-openmp
+      <<: *test-alpine
+      env:
+        - TARGET_BOX=LINUX64_MUSL
+        - BTYPE="BINARY=64 USE_OPENMP=1"
+
+    - <<: *test-alpine
+      env:
+        - TARGET_BOX=LINUX64_MUSL
+        - BTYPE="BINARY=64 INTERFACE64=1"
+
+    # Build with the same flags as Alpine do in OpenBLAS package.
+    - <<: *test-alpine
+      env:
+        - TARGET_BOX=LINUX64_MUSL
+        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=core2"
+
+  allow_failures:
+    - <<: *test-alpine-openmp
+
+# whitelist
+branches:
+  only:
+    - master
+    - develop
 
 notifications:
   webhooks:
@@ -7,32 +122,3 @@ notifications:
     on_success: change  # options: [always|never|change] default: always
     on_failure: always  # options: [always|never|change] default: always
     on_start: never     # options: [always|never|change] default: always
-
-compiler:
-  - gcc
-
-env:
-  - TARGET_BOX=LINUX64 BTYPE="BINARY=64"
-  - TARGET_BOX=LINUX64 BTYPE="BINARY=64 USE_OPENMP=1"
-  - TARGET_BOX=LINUX64 BTYPE="BINARY=64 INTERFACE64=1"
-  - TARGET_BOX=LINUX32 BTYPE="BINARY=32"
-  - TARGET_BOX=WIN64 BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
-
-before_install:
- - sudo apt-get update -qq
- - sudo apt-get install -qq gfortran
- - if [[ "$TARGET_BOX" == "WIN64" ]]; then sudo apt-get install -qq binutils-mingw-w64-x86-64 gcc-mingw-w64-x86-64 gfortran-mingw-w64-x86-64; fi
- - if [[ "$TARGET_BOX" == "LINUX32" ]]; then sudo apt-get install -qq  gcc-multilib gfortran-multilib; fi
-
-script: 
- - set -e
- - make QUIET_MAKE=1 DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE
- - if [ "$TARGET_BOX" == "LINUX32" ] || [ "$TARGET_BOX" == "LINUX64" ]; then make -C test DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE; fi
- - if [ "$TARGET_BOX" == "LINUX32" ] || [ "$TARGET_BOX" == "LINUX64" ]; then make -C ctest DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE; fi
- - if [ "$TARGET_BOX" == "LINUX32" ] || [ "$TARGET_BOX" == "LINUX64" ]; then make -C utest DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE; fi
-
-# whitelist
-branches:
-  only:
-    - master
-    - develop

appveyor.yml

@@ -12,31 +12,36 @@ clone_folder: c:\projects\OpenBLAS
 init:
   - git config --global core.autocrlf input
 
-build:
-  project: OpenBLAS.sln
-
 clone_depth: 5
 
-#branches to build
-branches:
-  only:
-    - master
-    - develop
-    - cmake
-
 skip_tags: true
 
 matrix:
-  fast_finish: true
+  fast_finish: false
 
 skip_commits:
 # Add [av skip] to commit messages
   message: /\[av skip\]/
 
+environment:
+  matrix:
+    - COMPILER: clang-cl
+    - COMPILER: cl
+
+install:
+  - if [%COMPILER%]==[clang-cl] call C:\Miniconda36-x64\Scripts\activate.bat
+  - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
+  - if [%COMPILER%]==[clang-cl] conda install --yes clangdev ninja cmake
+  - if [%COMPILER%]==[clang-cl] call "C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\vcvarsall.bat" amd64
+
 before_build:
   - echo Running cmake...
   - cd c:\projects\OpenBLAS
-  - cmake -G "Visual Studio 12 Win64" .
+  - if [%COMPILER%]==[cl] cmake -G "Visual Studio 12 Win64" .
+  - if [%COMPILER%]==[clang-cl] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl .
+
+build_script:
+  - cmake --build .
 
 test_script:
   - echo Running Test

cmake/c_check.cmake

@@ -28,6 +28,8 @@
 set(FU "")
 if(APPLE)
 set(FU "_")
+elseif(MSVC AND ${CMAKE_C_COMPILER_ID} MATCHES "Clang")
+set(FU "")
 elseif(MSVC)
 set(FU "_")
 elseif(UNIX)
@@ -59,7 +61,8 @@ endif ()
 # CMAKE_HOST_SYSTEM_PROCESSOR - The name of the CPU CMake is running on.
 #
 # TODO: CMAKE_SYSTEM_PROCESSOR doesn't seem to be correct - instead get it from the compiler a la c_check
-set(ARCH ${CMAKE_SYSTEM_PROCESSOR})
+set(ARCH ${CMAKE_SYSTEM_PROCESSOR} CACHE STRING "Target Architecture")
+
 if (${ARCH} STREQUAL "AMD64")
   set(ARCH "x86_64")
 endif ()

cmake/export.cmake

@@ -51,7 +51,8 @@ else()
 endif()
 
 add_custom_command(
-  TARGET ${OpenBLAS_LIBNAME} PRE_LINK 
+  OUTPUT ${PROJECT_BINARY_DIR}/openblas.def
+  #TARGET ${OpenBLAS_LIBNAME} PRE_LINK
   COMMAND perl 
   ARGS "${PROJECT_SOURCE_DIR}/exports/gensymbol" "win2k" "${ARCH_IN}" "dummy" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" "${SYMBOLPREFIX}" "${SYMBOLSUFFIX}" > "${PROJECT_BINARY_DIR}/openblas.def"
   COMMENT "Create openblas.def file"

cmake/prebuild.cmake

@@ -66,15 +66,14 @@ set(GETARCH_SRC
   ${CPUIDEMO}
 )
 
-if (NOT MSVC)
+if ("${CMAKE_C_COMPILER_ID}" STREQUAL "MSVC")
+  #Use generic for MSVC now
+  message("MSVC")
+  set(GETARCH_FLAGS ${GETARCH_FLAGS} -DFORCE_GENERIC)
+else()
   list(APPEND GETARCH_SRC ${PROJECT_SOURCE_DIR}/cpuid.S)
 endif ()
 
-if (MSVC)
-#Use generic for MSVC now
-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)

common.h

@@ -495,6 +495,33 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #define MMAP_POLICY (MAP_PRIVATE | MAP_ANONYMOUS)
 #endif
 
+#ifndef ASSEMBLER
+/* C99 supports complex floating numbers natively, which GCC also offers as an
+   extension since version 3.0.  If neither are available, use a compatible
+   structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
+#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
+      (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT))) && !defined(_MSC_VER)
+  #define OPENBLAS_COMPLEX_C99
+  #ifndef __cplusplus
+    #include <complex.h>
+  #endif
+  typedef float _Complex openblas_complex_float;
+  typedef double _Complex openblas_complex_double;
+  typedef xdouble _Complex openblas_complex_xdouble;
+  #define openblas_make_complex_float(real, imag)    ((real) + ((imag) * _Complex_I))
+  #define openblas_make_complex_double(real, imag)   ((real) + ((imag) * _Complex_I))
+  #define openblas_make_complex_xdouble(real, imag)  ((real) + ((imag) * _Complex_I))
+#else
+  #define OPENBLAS_COMPLEX_STRUCT
+  typedef struct { float real, imag; } openblas_complex_float;
+  typedef struct { double real, imag; } openblas_complex_double;
+  typedef struct { xdouble real, imag; } openblas_complex_xdouble;
+  #define openblas_make_complex_float(real, imag)    {(real), (imag)}
+  #define openblas_make_complex_double(real, imag)   {(real), (imag)}
+  #define openblas_make_complex_xdouble(real, imag)  {(real), (imag)}
+#endif
+#endif
+
 #include "param.h"
 #include "common_param.h"
 
@@ -524,31 +551,6 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #include <stdio.h>
 #endif  // NOINCLUDE
 
-/* C99 supports complex floating numbers natively, which GCC also offers as an
-   extension since version 3.0.  If neither are available, use a compatible
-   structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
-#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
-      (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT)))
-  #define OPENBLAS_COMPLEX_C99
-  #ifndef __cplusplus
-    #include <complex.h>
-  #endif
-  typedef float _Complex openblas_complex_float;
-  typedef double _Complex openblas_complex_double;
-  typedef xdouble _Complex openblas_complex_xdouble;
-  #define openblas_make_complex_float(real, imag)    ((real) + ((imag) * _Complex_I))
-  #define openblas_make_complex_double(real, imag)   ((real) + ((imag) * _Complex_I))
-  #define openblas_make_complex_xdouble(real, imag)  ((real) + ((imag) * _Complex_I))
-#else
-  #define OPENBLAS_COMPLEX_STRUCT
-  typedef struct { float real, imag; } openblas_complex_float;
-  typedef struct { double real, imag; } openblas_complex_double;
-  typedef struct { xdouble real, imag; } openblas_complex_xdouble;
-  #define openblas_make_complex_float(real, imag)    {(real), (imag)}
-  #define openblas_make_complex_double(real, imag)   {(real), (imag)}
-  #define openblas_make_complex_xdouble(real, imag)  {(real), (imag)}
-#endif
-
 #ifdef XDOUBLE
 #define OPENBLAS_COMPLEX_FLOAT openblas_complex_xdouble
 #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_xdouble(r,i)

common_param.h

@@ -333,8 +333,8 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   float (*cnrm2_k) (BLASLONG, float *, BLASLONG);
   float (*casum_k) (BLASLONG, float *, BLASLONG);
   int    (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  float _Complex (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  float _Complex (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  openblas_complex_float (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  openblas_complex_float (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   int    (*csrot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
 
   int    (*caxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
@@ -496,8 +496,8 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
   double (*znrm2_k) (BLASLONG, double *, BLASLONG);
   double (*zasum_k) (BLASLONG, double *, BLASLONG);
   int    (*zcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  double _Complex (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  double _Complex (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+  openblas_complex_double (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+  openblas_complex_double (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   int    (*zdrot_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
 
   int    (*zaxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
@@ -661,8 +661,8 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   xdouble (*xnrm2_k) (BLASLONG, xdouble *, BLASLONG);
   xdouble (*xasum_k) (BLASLONG, xdouble *, BLASLONG);
   int    (*xcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  xdouble _Complex (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  xdouble _Complex (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
+  openblas_complex_xdouble (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
+  openblas_complex_xdouble (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
   int    (*xqrot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
 
   int    (*xaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

driver/level2/gbmv_thread.c

@@ -230,8 +230,10 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT *alpha, FLOAT
 
 #ifndef TRANSA
     range_m[num_cpu] = num_cpu * ((m + 15) & ~15);
+    if (range_m[num_cpu] > m) range_m[num_cpu] = m;
 #else
     range_m[num_cpu] = num_cpu * ((n + 15) & ~15);
+    if (range_m[num_cpu] > n) range_m[num_cpu] = n;
 #endif
 
     queue[num_cpu].mode    = mode;

driver/level2/sbmv_thread.c

@@ -246,6 +246,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
 
       range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
+      if (range_n[num_cpu] > n) range_n[num_cpu] = n;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = sbmv_kernel;
@@ -285,6 +286,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
 
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
+      if (range_n[num_cpu] > n) range_n[num_cpu] = n;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = sbmv_kernel;
@@ -316,6 +318,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
 
       range_n[num_cpu] = num_cpu * ((n + 15) & ~15);
+      if (range_n[num_cpu] > n) range_n[num_cpu] = n;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = sbmv_kernel;

driver/level2/spmv_thread.c

@@ -246,6 +246,7 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *y,
 
     range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = spmv_kernel;
@@ -285,6 +286,7 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *y,
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = spmv_kernel;

driver/level2/symv_thread.c

@@ -177,7 +177,8 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG i
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
+    
     queue[MAX_CPU_NUMBER - num_cpu - 1].mode    = mode;
     queue[MAX_CPU_NUMBER - num_cpu - 1].routine = symv_kernel;
     queue[MAX_CPU_NUMBER - num_cpu - 1].args    = &args;
@@ -225,6 +226,7 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG i
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = symv_kernel;

driver/level2/tbmv_thread.c

@@ -288,6 +288,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
 
       range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
+      if (range_n[num_cpu] > n) range_n[num_cpu] = n;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = trmv_kernel;
@@ -327,6 +328,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
 
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
+      if (range_n[num_cpu] > n) range_n[num_cpu] = n;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = trmv_kernel;
@@ -356,6 +358,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
 
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
+      if (range_n[num_cpu] > n) range_n[num_cpu] = n;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = trmv_kernel;

driver/level2/tpmv_thread.c

@@ -307,7 +307,8 @@ int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthr
 
     range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
+    
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = tpmv_kernel;
     queue[num_cpu].args    = &args;
@@ -346,6 +347,7 @@ int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthr
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = tpmv_kernel;

driver/level2/trmv_thread.c

@@ -346,6 +346,7 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu
 
     range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = trmv_kernel;
@@ -385,6 +386,7 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
+    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = trmv_kernel;

driver/others/memory.c

@@ -155,7 +155,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef DYNAMIC_ARCH
 gotoblas_t *gotoblas = NULL;
 #endif
-
 extern void openblas_warning(int verbose, const char * msg);
 
 #ifndef SMP
@@ -187,25 +186,24 @@ int i,n;
 
 #if !defined(__GLIBC_PREREQ)
    return nums;
-#endif
-#if !__GLIBC_PREREQ(2, 3)
+#else
+ #if !__GLIBC_PREREQ(2, 3)
    return nums;
-#endif
+ #endif
 
-#if !__GLIBC_PREREQ(2, 7)
+ #if !__GLIBC_PREREQ(2, 7)
   ret = sched_getaffinity(0,sizeof(cpu_set_t), cpusetp);
   if (ret!=0) return nums;
   n=0;
-#if !__GLIBC_PREREQ(2, 6)
+  #if !__GLIBC_PREREQ(2, 6)
   for (i=0;i<nums;i++)
      if (CPU_ISSET(i,cpusetp)) n++;
   nums=n;
-#else
+  #else
   nums = CPU_COUNT(sizeof(cpu_set_t),cpusetp);
-#endif
+  #endif
   return nums;
-#endif
-
+ #else
   cpusetp = CPU_ALLOC(nums);
   if (cpusetp == NULL) return nums;
   size = CPU_ALLOC_SIZE(nums);
@@ -214,6 +212,8 @@ int i,n;
   nums = CPU_COUNT_S(size,cpusetp);
   CPU_FREE(cpusetp);
   return nums;
+ #endif
+#endif
 }
 #endif
 #endif

kernel/CMakeLists.txt

@@ -1,7 +1,6 @@
 
 include_directories(${PROJECT_SOURCE_DIR})
 
-
 # Makefile
 function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     set (OPENBLAS_SRC "")
@@ -21,7 +20,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     endif ()
 
     if (${ARCH} STREQUAL "x86")
-    if (NOT MSVC)
+    if (NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
       GenerateNamedObjects("${KERNELDIR}/cpuid.S" "" "" false "" "" true)
     else()
       GenerateNamedObjects("${KERNELDIR}/cpuid_win.c" "" "" false "" "" true)

kernel/arm64/casum_thunderx2t99.c

@@ -147,57 +147,57 @@ static FLOAT casum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	fmov	s6, "REG0"			\n"
 	"	fmov	s7, "REG0"			\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Lasum_kernel_S_BEGIN		\n"
+	"	bne	5f //asum_kernel_S_BEGIN	\n"
 
-	".Lasum_kernel_F_BEGIN:				\n"
+	"1: //asum_kernel_F_BEGIN:			\n"
 	"	asr	"J", "N", #5			\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Lasum_kernel_F1		\n"
+	"	beq	3f //asum_kernel_F1		\n"
 
-	".Lasum_kernel_F32:				\n"
+	"2: //asum_kernel_F32:				\n"
 	"	"KERNEL_F32"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F32		\n"
+	"	bne	2b //asum_kernel_F32		\n"
 	"	"KERNEL_F32_FINALIZE"			\n"
 
-	".Lasum_kernel_F1:				\n"
+	"3: //asum_kernel_F1:				\n"
 	"	ands	"J", "N", #31			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_F10:				\n"
+	"4: //asum_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs    "J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F10		\n"
-	"	b	.Lasum_kernel_L999		\n"
+	"	bne	4b //asum_kernel_F10		\n"
+	"	b	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S_BEGIN:				\n"
+	"5: //asum_kernel_S_BEGIN:			\n"
 	"	"INIT_S"				\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Lasum_kernel_S1		\n"
+	"	ble	7f //asum_kernel_S1		\n"
 
-	".Lasum_kernel_S4:				\n"
+	"6: //asum_kernel_S4:				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S4		\n"
+	"	bne	6b //asum_kernel_S4		\n"
 
-	".Lasum_kernel_S1:				\n"
+	"7: //asum_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S10:				\n"
+	"8: //asum_kernel_S10:				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S10		\n"
+	"	bne	8b //asum_kernel_S10		\n"
 
-	".Lasum_kernel_L999:				\n"
+	"9: //asum_kernel_L999:				\n"
 	"	fmov	%[ASUM_], "SUMFD"		\n"
 
 	: [ASUM_] "=r" (asum)		//%0

kernel/arm64/copy_thunderx2t99.c

@@ -90,62 +90,62 @@ static int do_copy(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_
 	"	mov	"Y", %[Y_]			\n"
 	"	mov	"INC_Y", %[INCY_]		\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Lcopy_kernel_L999		\n"
+	"	ble	8f //copy_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Lcopy_kernel_S_BEGIN		\n"
+	"	bne	4f //copy_kernel_S_BEGIN	\n"
 	"	cmp	"INC_Y", #1			\n"
-	"	bne	.Lcopy_kernel_S_BEGIN		\n"
+	"	bne	4f //copy_kernel_S_BEGIN	\n"
 
-	".Lcopy_kernel_F_BEGIN:				\n"
+	"// .Lcopy_kernel_F_BEGIN:			\n"
 	"	"INIT"					\n"
 	"	asr	"J", "N", #"N_DIV_SHIFT"	\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Lcopy_kernel_F1		\n"
+	"	beq	2f //copy_kernel_F1		\n"
 	"	.align 5				\n"
 
-	".Lcopy_kernel_F:				\n"
+	"1: //copy_kernel_F:				\n"
 	"	"KERNEL_F"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lcopy_kernel_F			\n"
+	"	bne	1b //copy_kernel_F		\n"
 
-	".Lcopy_kernel_F1:				\n"
+	"2: //copy_kernel_F1:				\n"
 #if defined(COMPLEX) && defined(DOUBLE)
-	"	b	.Lcopy_kernel_L999		\n"
+	"	b	8f //copy_kernel_L999		\n"
 #else
 	"	ands	"J", "N", #"N_REM_MASK"		\n"
-	"	ble	.Lcopy_kernel_L999		\n"
+	"	ble	8f //copy_kernel_L999		\n"
 #endif
 
-	".Lcopy_kernel_F10:				\n"
+	"3: //copy_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lcopy_kernel_F10		\n"
-	"	b	.Lcopy_kernel_L999		\n"
+	"	bne	3b //copy_kernel_F10		\n"
+	"	b	8f //copy_kernel_L999		\n"
 
-	".Lcopy_kernel_S_BEGIN:				\n"
+	"4: //copy_kernel_S_BEGIN:			\n"
 	"	"INIT"					\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Lcopy_kernel_S1		\n"
+	"	ble	6f //copy_kernel_S1		\n"
 
-	".Lcopy_kernel_S4:				\n"
+	"5: //copy_kernel_S4:				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lcopy_kernel_S4		\n"
+	"	bne	5b //copy_kernel_S4		\n"
 
-	".Lcopy_kernel_S1:				\n"
+	"6: //copy_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Lcopy_kernel_L999		\n"
+	"	ble	8f //copy_kernel_L999		\n"
 
-	".Lcopy_kernel_S10:				\n"
+	"7: //copy_kernel_S10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lcopy_kernel_S10		\n"
+	"	bne	7b //copy_kernel_S10		\n"
 
-	".Lcopy_kernel_L999:				\n"
+	"8: //copy_kernel_L999:				\n"
 
 	:
 	: [N_]    "r"  (n),		//%1

kernel/arm64/dasum_thunderx2t99.c

@@ -141,58 +141,58 @@ static FLOAT dasum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	fmov	d6, "REG0"			\n"
 	"	fmov	d7, "REG0"			\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Lasum_kernel_S_BEGIN		\n"
+	"	bne	5f //asum_kernel_S_BEGIN	\n"
 
-	".Lasum_kernel_F_BEGIN:				\n"
+	"1: //asum_kernel_F_BEGIN:			\n"
 	"	asr	"J", "N", #5			\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Lasum_kernel_F1		\n"
+	"	beq	3f //asum_kernel_F1		\n"
 
 	".align 5					\n"
-	".Lasum_kernel_F32:				\n"
+	"2: //asum_kernel_F32:				\n"
 	"	"KERNEL_F32"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F32		\n"
+	"	bne	2b //asum_kernel_F32		\n"
 	"	"KERNEL_F32_FINALIZE"			\n"
 
-	".Lasum_kernel_F1:				\n"
+	"3: //asum_kernel_F1:				\n"
 	"	ands	"J", "N", #31			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_F10:				\n"
+	"4: //asum_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs    "J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F10		\n"
-	"	b	.Lasum_kernel_L999		\n"
+	"	bne	4b //asum_kernel_F10		\n"
+	"	b	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S_BEGIN:				\n"
+	"5: //asum_kernel_S_BEGIN:			\n"
 	"	"INIT_S"				\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Lasum_kernel_S1		\n"
+	"	ble	7f //asum_kernel_S1		\n"
 
-	".Lasum_kernel_S4:				\n"
+	"6: //asum_kernel_S4:				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S4		\n"
+	"	bne	6b //asum_kernel_S4		\n"
 
-	".Lasum_kernel_S1:				\n"
+	"7: //asum_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S10:				\n"
+	"8: //asum_kernel_S10:				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S10		\n"
+	"	bne	8b //asum_kernel_S10		\n"
 
-	".Lasum_kernel_L999:				\n"
+	"9: //asum_kernel_L999:				\n"
 	"	fmov	%[ASUM_], "SUMF"		\n"
 
 	: [ASUM_] "=r" (asum)		//%0

kernel/arm64/dot_thunderx2t99.c

@@ -291,61 +291,61 @@ static RETURN_TYPE dot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, B
 	"	fmov	d6, xzr				\n"
 	"	fmov	d7, xzr				\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Ldot_kernel_L999		\n"
+	"	ble	9f //dot_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Ldot_kernel_S_BEGIN		\n"
+	"	bne	5f //dot_kernel_S_BEGIN		\n"
 	"	cmp	"INC_Y", #1			\n"
-	"	bne	.Ldot_kernel_S_BEGIN		\n"
+	"	bne	5f //dot_kernel_S_BEGIN		\n"
 
-	".Ldot_kernel_F_BEGIN:				\n"
+	"1: //dot_kernel_F_BEGIN:			\n"
 	"	lsl	"INC_X", "INC_X", "INC_SHIFT"	\n"
 	"	lsl	"INC_Y", "INC_Y", "INC_SHIFT"	\n"
 	"	asr	"J", "N", #"N_DIV_SHIFT"	\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Ldot_kernel_F1			\n"
+	"	beq	3f //dot_kernel_F1		\n"
 
 	"	.align 5				\n"
-	".Ldot_kernel_F:				\n"
+	"2: //dot_kernel_F:				\n"
 	"	"KERNEL_F"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_F			\n"
+	"	bne	2b //dot_kernel_F		\n"
 	"	"KERNEL_F_FINALIZE"			\n"
 
-	".Ldot_kernel_F1:				\n"
+	"3: //dot_kernel_F1:				\n"
 	"	ands	"J", "N", #"N_REM_MASK"		\n"
-	"	ble	.Ldot_kernel_L999		\n"
+	"	ble	9f //dot_kernel_L999		\n"
 
-	".Ldot_kernel_F10:				\n"
+	"4: //dot_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_F10		\n"
-	"	b	.Ldot_kernel_L999		\n"
+	"	bne	4b //dot_kernel_F10		\n"
+	"	b	9f //dot_kernel_L999		\n"
 
-	".Ldot_kernel_S_BEGIN:				\n"
+	"5: //dot_kernel_S_BEGIN:			\n"
 	"	lsl	"INC_X", "INC_X", "INC_SHIFT"	\n"
 	"	lsl	"INC_Y", "INC_Y", "INC_SHIFT"	\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Ldot_kernel_S1			\n"
+	"	ble	7f //dot_kernel_S1		\n"
 
-	".Ldot_kernel_S4:				\n"
+	"6: //dot_kernel_S4:				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_S4			\n"
+	"	bne	6b //dot_kernel_S4		\n"
 
-	".Ldot_kernel_S1:				\n"
+	"7: //dot_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Ldot_kernel_L999		\n"
+	"	ble	9f //dot_kernel_L999		\n"
 
-	".Ldot_kernel_S10:				\n"
+	"8: //dot_kernel_S10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_S10		\n"
+	"	bne	8b //dot_kernel_S10		\n"
 
-	".Ldot_kernel_L999:				\n"
+	"9: //dot_kernel_L999:				\n"
 	"	str	"DOTF", [%[DOT_]]		\n"
 
 	:

kernel/arm64/dznrm2_thunderx2t99.c

@@ -74,33 +74,33 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	fmov	"SCALE", xzr				\n"
 	"	fmov	"SSQ", #1.0				\n"
 	"	cmp	"N", xzr				\n"
-	"	ble	.Lnrm2_kernel_L999			\n"
+	"	ble	9f //nrm2_kernel_L999			\n"
 	"	cmp	"INC_X", xzr				\n"
-	"	ble	.Lnrm2_kernel_L999			\n"
+	"	ble	9f //nrm2_kernel_L999			\n"
 
-	".Lnrm2_kernel_F_BEGIN:					\n"
+	"1: //nrm2_kernel_F_BEGIN:				\n"
 	"	fmov	"REGZERO", xzr				\n"
 	"	fmov	"REGONE", #1.0				\n"
 	"	lsl	"INC_X", "INC_X", #"INC_SHIFT"		\n"
 	"	mov	"J", "N"				\n"
 	"	cmp	"J", xzr				\n"
-	"	beq	.Lnrm2_kernel_L999			\n"
+	"	beq	9f //nrm2_kernel_L999			\n"
 
-	".Lnrm2_kernel_F_ZERO_SKIP:				\n"
+	"2: //nrm2_kernel_F_ZERO_SKIP:				\n"
 	"	ldr	d4, ["X"]				\n"
 	"	fcmp	d4, "REGZERO"				\n"
-	"	bne	.Lnrm2_kernel_F_INIT			\n"
+	"	bne	3f //nrm2_kernel_F_INIT			\n"
 #if defined(COMPLEX)
 	"	ldr	d4, ["X", #8]				\n"
 	"	fcmp	d4, "REGZERO"				\n"
-	"	bne	.Lnrm2_kernel_F_INIT_I			\n"
+	"	bne	4f //nrm2_kernel_F_INIT_I		\n"
 #endif
 	"	add	"X", "X", "INC_X"			\n"
 	"	subs	"J", "J", #1				\n"
-	"	beq	.Lnrm2_kernel_L999			\n"
-	"	b	.Lnrm2_kernel_F_ZERO_SKIP		\n"
+	"	beq	9f //nrm2_kernel_L999			\n"
+	"	b	2b //nrm2_kernel_F_ZERO_SKIP		\n"
 
-	".Lnrm2_kernel_F_INIT:					\n"
+	"3: //nrm2_kernel_F_INIT:				\n"
 	"	ldr	d4, ["X"]				\n"
 	"	fabs	d4, d4					\n"
 	"	fmax	"CUR_MAX", "SCALE", d4			\n"
@@ -112,7 +112,7 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	fadd	"SSQ", "SSQ", d4			\n"
 	"	fmov	"SCALE", "CUR_MAX"			\n"
 #if defined(COMPLEX)
-	".Lnrm2_kernel_F_INIT_I:				\n"
+	"4: //nrm2_kernel_F_INIT_I:				\n"
 	"	ldr	d3, ["X", #8]				\n"
 	"	fabs	d3, d3					\n"
 	"	fmax	"CUR_MAX", "SCALE", d3			\n"
@@ -126,16 +126,16 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 #endif
 	"	add	"X", "X", "INC_X"			\n"
 	"	subs	"J", "J", #1				\n"
-	"	beq	.Lnrm2_kernel_L999			\n"
+	"	beq	9f //nrm2_kernel_L999			\n"
 
-	".Lnrm2_kernel_F_START:					\n"
+	"5: //nrm2_kernel_F_START:				\n"
 	"	cmp	"INC_X", #"SZ"				\n"
-	"	bne	.Lnrm2_kernel_F1			\n"
+	"	bne	8f //nrm2_kernel_F1			\n"
 	"	asr	"K", "J", #4				\n"
 	"	cmp	"K", xzr				\n"
-	"	beq	.Lnrm2_kernel_F1			\n"
+	"	beq	8f //nrm2_kernel_F1			\n"
 
-	".Lnrm2_kernel_F:					\n"
+	"6: //nrm2_kernel_F:					\n"
 	"	ldp	q16, q17,  ["X"]			\n"
 	"	ldp	q18, q19,  ["X", #32]			\n"
 	"	ldp	q20, q21,  ["X", #64]			\n"
@@ -255,13 +255,13 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	fmov	"SCALE", "CUR_MAX"			\n"
 #endif
 	"	subs	"K", "K", #1				\n"
-	"	bne	.Lnrm2_kernel_F				\n"
+	"	bne	6b //nrm2_kernel_F			\n"
 
-	".Lnrm2_kernel_F_DONE:					\n"
+	"7: //nrm2_kernel_F_DONE:				\n"
 	"	ands	"J", "J", #15				\n"
-	"	beq	.Lnrm2_kernel_L999			\n"
+	"	beq	9f //nrm2_kernel_L999			\n"
 
-	".Lnrm2_kernel_F1:					\n"
+	"8: //nrm2_kernel_F1:					\n"
 	"	ldr	d4, ["X"]				\n"
 	"	fabs	d4, d4					\n"
 	"	fmax	"CUR_MAX", "SCALE", d4			\n"
@@ -286,9 +286,9 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 #endif
 	"	add	"X", "X", "INC_X"			\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Lnrm2_kernel_F1			\n"
+	"	bne	8b //nrm2_kernel_F1			\n"
 
-	".Lnrm2_kernel_L999:					\n"
+	"9: //nrm2_kernel_L999:					\n"
 	"	str	"SSQ", [%[SSQ_]]			\n"
 	"	str	"SCALE", [%[SCALE_]]			\n"
 

kernel/arm64/iamax_thunderx2t99.c

@@ -208,7 +208,7 @@ extern int blas_level1_thread_with_return_value(int mode, BLASLONG m, BLASLONG n
 #endif
 
 
-static BLASLONG iamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+static BLASLONG __attribute__((noinline)) iamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 {
 	BLASLONG index = 0;
 
@@ -220,72 +220,72 @@ static BLASLONG iamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	mov	"INC_X", %[INCX_]			\n"
 
 	"	cmp	"N", xzr				\n"
-	"	ble	.Liamax_kernel_zero			\n"
+	"	ble	10f //iamax_kernel_zero			\n"
 	"	cmp	"INC_X", xzr				\n"
-	"	ble	.Liamax_kernel_zero			\n"
+	"	ble	10f //iamax_kernel_zero			\n"
 	"	cmp	"INC_X", #1				\n"
-	"	bne	.Liamax_kernel_S_BEGIN			\n"
+	"	bne	5f //iamax_kernel_S_BEGIN		\n"
 	"	mov	x7, "X"					\n"
 
-	".Liamax_kernel_F_BEGIN:				\n"
+	"1: //iamax_kernel_F_BEGIN:				\n"
 	"	"INIT"						\n"
 	"	subs	"N", "N", #1				\n"
-	"	ble	.Liamax_kernel_L999			\n"
+	"	ble	9f //iamax_kernel_L999			\n"
 	"	asr	"J", "N", #"N_DIV_SHIFT"		\n"
 	"	cmp	"J", xzr				\n"
-	"	beq	.Liamax_kernel_F1			\n"
+	"	beq	3f //iamax_kernel_F1			\n"
 	"	add	"Z", "Z", #1				\n"
 
-	".Liamax_kernel_F:					\n"
+	"2: //iamax_kernel_F:					\n"
 	"	"KERNEL_F"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Liamax_kernel_F			\n"
+	"	bne	2b //iamax_kernel_F			\n"
 	"	"KERNEL_F_FINALIZE"				\n"
 	"	sub	"Z", "Z", #1				\n"
 
-	".Liamax_kernel_F1:					\n"
+	"3: //iamax_kernel_F1:					\n"
 	"	ands	"J", "N", #"N_REM_MASK"			\n"
-	"	ble	.Liamax_kernel_L999			\n"
+	"	ble	9f //iamax_kernel_L999			\n"
 
-	".Liamax_kernel_F10:					\n"
+	"4: //iamax_kernel_F10:					\n"
 	"	"KERNEL_S1"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Liamax_kernel_F10			\n"
-	"	b	.Liamax_kernel_L999			\n"
+	"	bne	4b //iamax_kernel_F10			\n"
+	"	b	9f //iamax_kernel_L999			\n"
 
-	".Liamax_kernel_S_BEGIN:				\n"
+	"5: //iamax_kernel_S_BEGIN:				\n"
 	"	"INIT"						\n"
 	"	subs	"N", "N", #1				\n"
-	"	ble	.Liamax_kernel_L999			\n"
+	"	ble	9f //iamax_kernel_L999			\n"
 	"	asr	"J", "N", #2				\n"
 	"	cmp	"J", xzr				\n"
-	"	ble	.Liamax_kernel_S1			\n"
+	"	ble	7f //iamax_kernel_S1			\n"
 
-	".Liamax_kernel_S4:					\n"
+	"6: //iamax_kernel_S4:					\n"
 	"	"KERNEL_S1"					\n"
 	"	"KERNEL_S1"					\n"
 	"	"KERNEL_S1"					\n"
 	"	"KERNEL_S1"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Liamax_kernel_S4			\n"
+	"	bne	6b //iamax_kernel_S4			\n"
 
-	".Liamax_kernel_S1:					\n"
+	"7: //iamax_kernel_S1:					\n"
 	"	ands	"J", "N", #3				\n"
-	"	ble	.Liamax_kernel_L999			\n"
+	"	ble	9f //iamax_kernel_L999			\n"
 
-	".Liamax_kernel_S10:					\n"
+	"8: //iamax_kernel_S10:					\n"
 	"	"KERNEL_S1"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Liamax_kernel_S10			\n"
+	"	bne	8b //iamax_kernel_S10			\n"
 
-	".Liamax_kernel_L999:					\n"
+	"9: //iamax_kernel_L999:				\n"
 	"	mov	x0, "INDEX"				\n"
-	"	b	.Liamax_kernel_DONE			\n"
+	"	b	11f //iamax_kernel_DONE			\n"
 
-	".Liamax_kernel_zero:					\n"
+	"10: //iamax_kernel_zero:				\n"
 	"	mov	x0, xzr					\n"
 
-	".Liamax_kernel_DONE:					\n"
+	"11: //iamax_kernel_DONE:				\n"
 	"	mov	%[INDEX_], "INDEX"			\n"
 
 	: [INDEX_] "=r" (index)		//%0

kernel/arm64/izamax_thunderx2t99.c

@@ -229,72 +229,72 @@ static BLASLONG izamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	mov	"INC_X", %[INCX_]			\n"
 
 	"	cmp	"N", xzr				\n"
-	"	ble	.Lizamax_kernel_zero			\n"
+	"	ble	10f //izamax_kernel_zero		\n"
 	"	cmp	"INC_X", xzr				\n"
-	"	ble	.Lizamax_kernel_zero			\n"
+	"	ble	10f //izamax_kernel_zero		\n"
 	"	cmp	"INC_X", #1				\n"
-	"	bne	.Lizamax_kernel_S_BEGIN			\n"
+	"	bne	5f //izamax_kernel_S_BEGIN		\n"
 	"	mov	x7, "X"					\n"
 
-	".Lizamax_kernel_F_BEGIN:				\n"
+	"1: //izamax_kernel_F_BEGIN:				\n"
 	"	"INIT"						\n"
 	"	subs	"N", "N", #1				\n"
-	"	ble	.Lizamax_kernel_L999			\n"
+	"	ble	9f //izamax_kernel_L999			\n"
 	"	asr	"J", "N", #"N_DIV_SHIFT"		\n"
 	"	cmp	"J", xzr				\n"
-	"	beq	.Lizamax_kernel_F1			\n"
+	"	beq	3f //izamax_kernel_F1			\n"
 	"	add	"Z", "Z", #1				\n"
 
-	".Lizamax_kernel_F:					\n"
+	"2: //izamax_kernel_F:					\n"
 	"	"KERNEL_F"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Lizamax_kernel_F			\n"
+	"	bne	2b //izamax_kernel_F			\n"
 	"	"KERNEL_F_FINALIZE"				\n"
 	"	sub	"Z", "Z", #1				\n"
 
-	".Lizamax_kernel_F1:					\n"
+	"3: //izamax_kernel_F1:					\n"
 	"	ands	"J", "N", #"N_REM_MASK"			\n"
-	"	ble	.Lizamax_kernel_L999			\n"
+	"	ble	9f //izamax_kernel_L999			\n"
 
-	".Lizamax_kernel_F10:					\n"
+	"4: //izamax_kernel_F10:				\n"
 	"	"KERNEL_S1"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Lizamax_kernel_F10			\n"
-	"	b	.Lizamax_kernel_L999			\n"
+	"	bne	4b //izamax_kernel_F10			\n"
+	"	b	9f //izamax_kernel_L999			\n"
 
-	".Lizamax_kernel_S_BEGIN:				\n"
+	"5: //izamax_kernel_S_BEGIN:				\n"
 	"	"INIT"						\n"
 	"	subs	"N", "N", #1				\n"
-	"	ble	.Lizamax_kernel_L999			\n"
+	"	ble	9f //izamax_kernel_L999			\n"
 	"	asr	"J", "N", #2				\n"
 	"	cmp	"J", xzr				\n"
-	"	ble	.Lizamax_kernel_S1			\n"
+	"	ble	7f //izamax_kernel_S1			\n"
 
-	".Lizamax_kernel_S4:					\n"
+	"6: //izamax_kernel_S4:					\n"
 	"	"KERNEL_S1"					\n"
 	"	"KERNEL_S1"					\n"
 	"	"KERNEL_S1"					\n"
 	"	"KERNEL_S1"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Lizamax_kernel_S4			\n"
+	"	bne	6b //izamax_kernel_S4			\n"
 
-	".Lizamax_kernel_S1:					\n"
+	"7: //izamax_kernel_S1:					\n"
 	"	ands	"J", "N", #3				\n"
-	"	ble	.Lizamax_kernel_L999			\n"
+	"	ble	9f //izamax_kernel_L999			\n"
 
-	".Lizamax_kernel_S10:					\n"
+	"8: //izamax_kernel_S10:				\n"
 	"	"KERNEL_S1"					\n"
 	"	subs	"J", "J", #1				\n"
-	"	bne	.Lizamax_kernel_S10			\n"
+	"	bne	8b //izamax_kernel_S10			\n"
 
-	".Lizamax_kernel_L999:					\n"
+	"9: //izamax_kernel_L999:				\n"
 	"	mov	x0, "INDEX"				\n"
-	"	b	.Lizamax_kernel_DONE			\n"
+	"	b	11f //izamax_kernel_DONE		\n"
 
-	".Lizamax_kernel_zero:					\n"
+	"10: //izamax_kernel_zero:				\n"
 	"	mov	x0, xzr					\n"
 
-	".Lizamax_kernel_DONE:					\n"
+	"11: //izamax_kernel_DONE:				\n"
 	"	mov	%[INDEX_], "INDEX"			\n"
 
 	: [INDEX_] "=r" (index)		//%0

kernel/arm64/sasum_thunderx2t99.c

@@ -143,58 +143,58 @@ static FLOAT sasum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	fmov	s6, "REG0"			\n"
 	"	fmov	s7, "REG0"			\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Lasum_kernel_S_BEGIN		\n"
+	"	bne	5f //asum_kernel_S_BEGIN	\n"
 
-	".Lasum_kernel_F_BEGIN:				\n"
+	"1: //asum_kernel_F_BEGIN:			\n"
 	"	asr	"J", "N", #6			\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Lasum_kernel_F1		\n"
+	"	beq	3f //asum_kernel_F1		\n"
 
 	".align 5					\n"
-	".Lasum_kernel_F64:				\n"
+	"2: //asum_kernel_F64:				\n"
 	"	"KERNEL_F64"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F64		\n"
+	"	bne	2b //asum_kernel_F64		\n"
 	"	"KERNEL_F64_FINALIZE"			\n"
 
-	".Lasum_kernel_F1:				\n"
+	"3: //asum_kernel_F1:				\n"
 	"	ands	"J", "N", #63			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_F10:				\n"
+	"4: //asum_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs    "J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F10		\n"
-	"	b	.Lasum_kernel_L999		\n"
+	"	bne	4b //asum_kernel_F10		\n"
+	"	b	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S_BEGIN:				\n"
+	"5: //asum_kernel_S_BEGIN:			\n"
 	"	"INIT_S"				\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Lasum_kernel_S1		\n"
+	"	ble	7f //asum_kernel_S1		\n"
 
-	".Lasum_kernel_S4:				\n"
+	"6: //asum_kernel_S4:				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S4		\n"
+	"	bne	6b //asum_kernel_S4		\n"
 
-	".Lasum_kernel_S1:				\n"
+	"7: //asum_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S10:				\n"
+	"8: //asum_kernel_S10:				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S10		\n"
+	"	bne	8b //asum_kernel_S10		\n"
 
-	".Lasum_kernel_L999:				\n"
+	"9: //asum_kernel_L999:				\n"
 	"	fmov	%[ASUM_], "SUMFD"		\n"
 
 	: [ASUM_] "=r" (asum)		//%0

kernel/arm64/scnrm2_thunderx2t99.c

@@ -227,58 +227,58 @@ static double nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	fmov	d6, xzr				\n"
 	"	fmov	d7, xzr				\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Lnrm2_kernel_L999		\n"
+	"	ble	9f //nrm2_kernel_L999		\n"
 	"	cmp	"INC_X", xzr			\n"
-	"	ble	.Lnrm2_kernel_L999		\n"
+	"	ble	9f //nrm2_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Lnrm2_kernel_S_BEGIN		\n"
+	"	bne	5f //nrm2_kernel_S_BEGIN	\n"
 
-	".Lnrm2_kernel_F_BEGIN:				\n"
+	"1: //nrm2_kernel_F_BEGIN:			\n"
 	"	asr	"J", "N", #"N_DIV_SHIFT"	\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Lnrm2_kernel_S_BEGIN		\n"
+	"	beq	5f //nrm2_kernel_S_BEGIN	\n"
 
 	"	.align 5				\n"
-	".Lnrm2_kernel_F:				\n"
+	"2: //nrm2_kernel_F:				\n"
 	"	"KERNEL_F"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lnrm2_kernel_F			\n"
+	"	bne	2b //nrm2_kernel_F		\n"
 	"	"KERNEL_F_FINALIZE"			\n"
 
-	".Lnrm2_kernel_F1:				\n"
+	"3: //nrm2_kernel_F1:				\n"
 	"	ands	"J", "N", #"N_REM_MASK"		\n"
-	"	ble	.Lnrm2_kernel_L999		\n"
+	"	ble	9f //nrm2_kernel_L999		\n"
 
-	".Lnrm2_kernel_F10:				\n"
+	"4: //nrm2_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lnrm2_kernel_F10		\n"
-	"	b	.Lnrm2_kernel_L999		\n"
+	"	bne	4b //nrm2_kernel_F10		\n"
+	"	b	9f //nrm2_kernel_L999		\n"
 
-	".Lnrm2_kernel_S_BEGIN:				\n"
+	"5: //nrm2_kernel_S_BEGIN:			\n"
 	"	lsl	"INC_X", "INC_X", #"INC_SHIFT"	\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Lnrm2_kernel_S1		\n"
+	"	ble	7f //nrm2_kernel_S1		\n"
 
-	".Lnrm2_kernel_S4:				\n"
+	"6: //nrm2_kernel_S4:				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lnrm2_kernel_S4		\n"
+	"	bne	6b //nrm2_kernel_S4		\n"
 
-	".Lnrm2_kernel_S1:				\n"
+	"7: //nrm2_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Lnrm2_kernel_L999		\n"
+	"	ble	9f //nrm2_kernel_L999		\n"
 
-	".Lnrm2_kernel_S10:				\n"
+	"8: //nrm2_kernel_S10:				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lnrm2_kernel_S10		\n"
+	"	bne	8b //nrm2_kernel_S10		\n"
 
-	".Lnrm2_kernel_L999:				\n"
+	"9: //nrm2_kernel_L999:				\n"
 	"	"KERNEL_FINALIZE"			\n"
 	"	fmov	%[RET_], "SSQD"			\n"
 

kernel/arm64/zasum_thunderx2t99.c

@@ -143,58 +143,58 @@ static FLOAT zasum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	"	fmov	d6, "REG0"			\n"
 	"	fmov	d7, "REG0"			\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", xzr			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Lasum_kernel_S_BEGIN		\n"
+	"	bne	5f //asum_kernel_S_BEGIN	\n"
 
-	".Lasum_kernel_F_BEGIN:				\n"
+	"1: //asum_kernel_F_BEGIN:			\n"
 	"	asr	"J", "N", #4			\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Lasum_kernel_F1		\n"
+	"	beq	3f //asum_kernel_F1		\n"
 
 	".align 5					\n"
-	".Lasum_kernel_F16:				\n"
+	"2: //asum_kernel_F16:				\n"
 	"	"KERNEL_F16"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F16		\n"
+	"	bne	2b //asum_kernel_F16		\n"
 	"	"KERNEL_F16_FINALIZE"			\n"
 
-	".Lasum_kernel_F1:				\n"
+	"3: //asum_kernel_F1:				\n"
 	"	ands	"J", "N", #15			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_F10:				\n"
+	"4: //asum_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs    "J", "J", #1			\n"
-	"	bne	.Lasum_kernel_F10		\n"
-	"	b	.Lasum_kernel_L999		\n"
+	"	bne	4b //asum_kernel_F10		\n"
+	"	b	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S_BEGIN:				\n"
+	"5: //asum_kernel_S_BEGIN:			\n"
 	"	"INIT_S"				\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Lasum_kernel_S1		\n"
+	"	ble	7f //asum_kernel_S1		\n"
 
-	".Lasum_kernel_S4:				\n"
+	"6: //asum_kernel_S4:				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S4		\n"
+	"	bne	6b //asum_kernel_S4		\n"
 
-	".Lasum_kernel_S1:				\n"
+	"7: //asum_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Lasum_kernel_L999		\n"
+	"	ble	9f //asum_kernel_L999		\n"
 
-	".Lasum_kernel_S10:				\n"
+	"8: //asum_kernel_S10:				\n"
 	"	"KERNEL_S1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Lasum_kernel_S10		\n"
+	"	bne	8b //asum_kernel_S10		\n"
 
-	".Lasum_kernel_L999:				\n"
+	"9: //asum_kernel_L999:				\n"
 	"	fmov	%[ASUM_], "SUMF"		\n"
 
 	: [ASUM_] "=r" (asum)		//%0

kernel/arm64/zdot_thunderx2t99.c

@@ -218,61 +218,61 @@ static void zdot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLON
 	"	fmov	d6, xzr				\n"
 	"	fmov	d7, xzr				\n"
 	"	cmp	"N", xzr			\n"
-	"	ble	.Ldot_kernel_L999		\n"
+	"	ble	9f //dot_kernel_L999		\n"
 	"	cmp	"INC_X", #1			\n"
-	"	bne	.Ldot_kernel_S_BEGIN		\n"
+	"	bne	5f //dot_kernel_S_BEGIN		\n"
 	"	cmp	"INC_Y", #1			\n"
-	"	bne	.Ldot_kernel_S_BEGIN		\n"
+	"	bne	5f //dot_kernel_S_BEGIN		\n"
 
-	".Ldot_kernel_F_BEGIN:				\n"
+	"1: //dot_kernel_F_BEGIN:			\n"
 	"	lsl	"INC_X", "INC_X", "INC_SHIFT"	\n"
 	"	lsl	"INC_Y", "INC_Y", "INC_SHIFT"	\n"
 	"	asr	"J", "N", #"N_DIV_SHIFT"	\n"
 	"	cmp	"J", xzr			\n"
-	"	beq	.Ldot_kernel_F1			\n"
+	"	beq	3f //dot_kernel_F1		\n"
 
 	"	.align 5				\n"
-	".Ldot_kernel_F:				\n"
+	"2: //dot_kernel_F:				\n"
 	"	"KERNEL_F"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_F			\n"
+	"	bne	2b //dot_kernel_F		\n"
 	"	"KERNEL_F_FINALIZE"			\n"
 
-	".Ldot_kernel_F1:				\n"
+	"3: //dot_kernel_F1:				\n"
 	"	ands	"J", "N", #"N_REM_MASK"		\n"
-	"	ble	.Ldot_kernel_L999		\n"
+	"	ble	9f //dot_kernel_L999		\n"
 
-	".Ldot_kernel_F10:				\n"
+	"4: //dot_kernel_F10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_F10		\n"
-	"	b	.Ldot_kernel_L999		\n"
+	"	bne	4b //dot_kernel_F10		\n"
+	"	b	9f //dot_kernel_L999		\n"
 
-	".Ldot_kernel_S_BEGIN:				\n"
+	"5: //dot_kernel_S_BEGIN:			\n"
 	"	lsl	"INC_X", "INC_X", "INC_SHIFT"	\n"
 	"	lsl	"INC_Y", "INC_Y", "INC_SHIFT"	\n"
 	"	asr	"J", "N", #2			\n"
 	"	cmp	"J", xzr			\n"
-	"	ble	.Ldot_kernel_S1			\n"
+	"	ble	7f //dot_kernel_S1		\n"
 
-	".Ldot_kernel_S4:				\n"
+	"6: //dot_kernel_S4:				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_S4			\n"
+	"	bne	6b //dot_kernel_S4		\n"
 
-	".Ldot_kernel_S1:				\n"
+	"7: //dot_kernel_S1:				\n"
 	"	ands	"J", "N", #3			\n"
-	"	ble	.Ldot_kernel_L999		\n"
+	"	ble	9f //dot_kernel_L999		\n"
 
-	".Ldot_kernel_S10:				\n"
+	"8: //dot_kernel_S10:				\n"
 	"	"KERNEL_F1"				\n"
 	"	subs	"J", "J", #1			\n"
-	"	bne	.Ldot_kernel_S10		\n"
+	"	bne	8b //dot_kernel_S10		\n"
 
-	".Ldot_kernel_L999:				\n"
+	"9: //dot_kernel_L999:				\n"
 	"	str	"DOTF", [%[DOTR_]]		\n"
 	"	str	"DOTI", [%[DOTI_]]		\n"
 

kernel/x86_64/cdot.c

@@ -91,16 +91,15 @@ static void cdot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
 
 #endif
 
-FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
 {
 	BLASLONG i;
 	BLASLONG ix,iy;
-	FLOAT _Complex result;
 	FLOAT  dot[8] = { 0.0, 0.0, 0.0 , 0.0, 0.0, 0.0, 0.0, 0.0 } ; 
 
 	if ( n <= 0 ) 
 	{
-	        result = OPENBLAS_MAKE_COMPLEX_FLOAT (0.0, 0.0) ;
+	    OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (0.0, 0.0) ;
 		return(result);
 
 	}
@@ -160,11 +159,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
 	}
 
 #if !defined(CONJ)
-	        result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]-dot[1], dot[4]+dot[5]) ;
+	OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]-dot[1], dot[4]+dot[5]) ;
 //	CREAL(result) = dot[0] - dot[1];
 //	CIMAG(result) = dot[4] + dot[5];
 #else
-	        result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]+dot[1], dot[4]-dot[5]) ;
+	OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]+dot[1], dot[4]-dot[5]) ;
 //	CREAL(result) = dot[0] + dot[1];
 //	CIMAG(result) = dot[4] - dot[5];
 

kernel/x86_64/zdot.c

@@ -86,18 +86,17 @@ static void zdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
 
 #endif
 
-FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
 {
 	BLASLONG i;
 	BLASLONG ix,iy;
-	FLOAT _Complex result;
 	FLOAT  dot[4] = { 0.0, 0.0, 0.0 , 0.0 } ; 
 
 	if ( n <= 0 ) 
 	{
 //	        CREAL(result) = 0.0 ;
 //        	CIMAG(result) = 0.0 ;
-		result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
+		OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
 		return(result);
 
 	}
@@ -151,11 +150,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
 	}
 
 #if !defined(CONJ)
-	result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
+	OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
 //	CREAL(result) = dot[0] - dot[1];
 //	CIMAG(result) = dot[2] + dot[3];
 #else
-	result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
+	OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
 //	CREAL(result) = dot[0] + dot[1];
 //	CIMAG(result) = dot[2] - dot[3];
 

openblas_config_template.h

@@ -59,7 +59,7 @@ typedef int blasint;
    extension since version 3.0.  If neither are available, use a compatible
    structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
 #if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
-      (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT)))
+      (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT))) && !defined(_MSC_VER)
   #define OPENBLAS_COMPLEX_C99
 #ifndef __cplusplus
   #include <complex.h>

utest/CMakeLists.txt

@@ -1,10 +1,14 @@
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_BINARY_DIR})
 
-set(OpenBLAS_utest_src
-  utest_main.c
-  test_amax.c
+if (MSVC AND "${CMAKE_C_COMPILER_ID}" MATCHES Clang)
+  set(OpenBLAS_utest_src utest_main2.c)
+else ()
+  set(OpenBLAS_utest_src
+    utest_main.c
+    test_amax.c
   )
+endif ()
 
 if (NOT NO_LAPACK)
 set(OpenBLAS_utest_src
@@ -36,7 +40,7 @@ endforeach()
 if (MSVC)
 add_custom_command(TARGET ${OpenBLAS_utest_bin}
           POST_BUILD
-          COMMAND ${CMAKE_COMMAND} -E copy ${PROJECT_BINARY_DIR}/lib/$<CONFIG>/${OpenBLAS_LIBNAME}.dll ${CMAKE_CURRENT_BINARY_DIR}/.
+          COMMAND ${CMAKE_COMMAND} -E copy ${PROJECT_BINARY_DIR}/lib/${CMAKE_CFG_INTDIR}/${OpenBLAS_LIBNAME}.dll ${CMAKE_CURRENT_BINARY_DIR}/.
 )
 endif()
 

utest/utest_main2.c

@@ -0,0 +1,61 @@
+/*****************************************************************************
+Copyright (c) 2011-2016, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of 
+      its contributors may be used to endorse or promote products 
+      derived from this software without specific prior written 
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+**********************************************************************************/
+
+#include <stdio.h>
+
+#define CTEST_MAIN
+#define CTEST_SEGFAULT
+#define CTEST_ADD_TESTS_MANUALLY
+
+#include "openblas_utest.h"
+
+CTEST(amax, samax){
+  blasint N=3, inc=1;
+  float te_max=0.0, tr_max=0.0;
+  float x[]={-1.1, 2.2, -3.3};
+  te_max=BLASFUNC(samax)(&N, x, &inc);
+  tr_max=3.3;
+  
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
+}
+
+int main(int argc, const char ** argv){
+
+  CTEST_ADD(amax, samax);
+  int num_fail=0;
+
+  num_fail=ctest_main(argc, argv);
+
+  return num_fail;
+}
+