drop __ACCELERATE, use vecLibFort instead

cp2k · May 12, 2023 · e46a8c8 · e46a8c8
1 parent 19e08a4
commit e46a8c8
Show file tree

Hide file tree

Showing 12 changed files with 49 additions and 33 deletions.
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -25,10 +25,18 @@ repos:
   - id: check-yaml
   - id: check-symlinks
   - id: trailing-whitespace
+    exclude: >-
+      (?x)^(
+        tools/vecLibFort/.*|
+      )$
 - repo: https://github.com/pseewald/fprettify
   rev: v0.3.7
   hooks:
   - id: fprettify
+    exclude: >-
+      (?x)^(
+        tools/vecLibFort/.*|
+      )$
 - repo: https://github.com/cheshirekow/cmake-format-precommit
   rev: v0.6.13
   hooks:
@@ -64,3 +72,8 @@ repos:
     files: \.(c|cc|cxx|cpp|cl|frag|glsl|h|hpp|hxx|ih|ispc|ipp|java|js|m|mm|proto|textproto|vert)$
     args: ['-i', '-fallback-style=none', '--style=file']
     additional_dependencies: ['clang-format']
+    exclude: >-
+      (?x)^(
+        tools/vecLibFort/.*|
+      )$
+
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -163,6 +163,23 @@ endif ()
 find_package(LAPACK REQUIRED) # needed for some of the integrated test routines,
                               # also calls find_package(BLAS)
 
+if (APPLE
+    AND (BLAS_LIBRARIES MATCHES "Accelerate"
+         OR BLAS_LIBRARIES MATCHES "vecLib" # automated search
+         OR BLA_VENDOR STREQUAL "Accelerate"
+         OR BLA_VENDOR STREQUAL "NAS" # user override
+        ))
+  message(CHECK_START "Looking for vecLibFort library")
+  find_library(VECLIBFORT_LIBRARY vecLibFort)
+  if (NOT VECLIBFORT_LIBRARY)
+    message(CHECK_FAIL "not found, building it")
+    add_subdirectory(tools/vecLibFort)
+    set(VECLIBFORT_LIBRARY vecLibFort)
+  else ()
+    message(CHECK_PASS "found at " ${VECLIBFORT_LIBRARY})
+  endif ()
+endif ()
+
 # =================================== Python this module looks preferably for
 # version 3 of Python. If not found, version 2 is searched. In CMake 3.15, if a
 # python virtual environment is activated, it will search the virtual

diff --git a/docs/guide/2-user-guide/1-installation/index.md b/docs/guide/2-user-guide/1-installation/index.md
@@ -9,8 +9,12 @@ You need:
 * [CMake](https://cmake.org/) (3.22+)
 * GNU make or Ninja
 * Fortran compiler which supports at least Fortran 2008 (including the TS 29113 when using the C-bindings)
-* BLAS+LAPACK implementation (reference, OpenBLAS and MKL have been tested. Note: DBCSR linked to OpenBLAS 0.3.6 gives wrong results on Power9 architectures.)
-* Python version installed (2.7 or 3.6+ have been tested)
+* BLAS+LAPACK implementation
+    * Reference BLAS/LAPACK, OpenBLAS and MKL have been tested and can be considered supported.
+    * On macOS [vecLibFort](https://github.com/mcg1969/vecLibFort) is required to use Accelerate and/or vecLib.
+      The build system will automatically build a bundled version if not found on the system.
+    * DBCSR linked to OpenBLAS 0.3.6 gives wrong results on Power9 architectures.
+* Python version installed (3.6+ have been tested)
 
 Optional:
 

diff --git a/docs/guide/3-developer-guide/3-programming/1-overview/index.md b/docs/guide/3-developer-guide/3-programming/1-overview/index.md
@@ -46,7 +46,6 @@ Assumed square matrix with 20x20 matrix with 5x5 blocks and a 2x2 processor grid
 | `__NO_STATM_ACCESS`, `__STATM_RESIDENT` or `__STATM_TOTAL` | Toggle memory usage reporting between resident memory and total memory. In particular, macOS users must use `-D__NO_STATM_ACCESS` | Fortran |
 | `__NO_ABORT` | Avoid calling abort, but STOP instead (useful for coverage testing, and to avoid core dumps on some systems) | Fortran |
 | `__LIBXSMM` | Enable [LIBXSMM](https://github.com/hfp/libxsmm/) link for optimized small matrix multiplications on CPU | Fortran |
-| `__ACCELERATE` | Must be defined on macOS when Apple's Accelerate framework is used for BLAS and LAPACK (this is due to some interface incompatibilities between Accelerate and reference BLAS/LAPACK) | Fortran |
 | `NDEBUG`       | Assertions are stripped ("compiled out"), `NDEBUG` is the ANSI-conforming symbol name (not `__NDEBUG`). Regular release builds may carry assertions for safety | Fortran, C, C++ |
 | `__CRAY_PM_ACCEL_ENERGY` or `__CRAY_PM_ENERGY` | Switch on collectin energy profiling on Cray systems | Fortran |
 | `__DBCSR_ACC` | Enable Accelerator compilation | Fortran, C, C++ |

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -176,8 +176,8 @@ if (APPLE)
   # fix /proc/self/statm can not be opened on macOS
   target_compile_definitions(dbcsr PRIVATE __NO_STATM_ACCESS)
 
-  if (BLAS_LIBRARIES MATCHES "Accelerate")
-    target_compile_definitions(dbcsr PRIVATE __ACCELERATE)
+  if (VECLIBFORT_LIBRARY)
+    target_link_libraries(dbcsr PRIVATE ${VECLIBFORT_LIBRARY})
   endif ()
 endif ()
 

diff --git a/src/mm/dbcsr_mm_common.F b/src/mm/dbcsr_mm_common.F
@@ -579,11 +579,7 @@ SUBROUTINE calc_norms_${nametype1}$ (norms, nblks, &
          INTEGER                                  :: blk, bp, bpe, row, col
 
          REAL(KIND=real_8), EXTERNAL              :: DDOT
-#if defined (__ACCELERATE)
-         REAL(KIND=real_8), EXTERNAL              :: SDOT
-#else
          REAL(KIND=real_4), EXTERNAL              :: SDOT
-#endif
 
 !   ---------------------------------------------------------------------------
 

diff --git a/src/mm/dbcsr_mm_multrec.F b/src/mm/dbcsr_mm_multrec.F
@@ -707,11 +707,7 @@ SUBROUTINE multrec_filtering_${nametype1}$ (filter_eps, nblks, rowi, coli, blkp,
          REAL(kind=real_8)                          :: nrm
 
          REAL(KIND=real_8), EXTERNAL                :: DZNRM2, DDOT
-#if defined (__ACCELERATE)
-         REAL(KIND=real_8), EXTERNAL                :: SCNRM2, SDOT
-#else
          REAL(KIND=real_4), EXTERNAL                :: SCNRM2, SDOT
-#endif
 
          REAL(kind=real_8)                          :: filter_eps_opt
 

diff --git a/src/ops/dbcsr_operations.F b/src/ops/dbcsr_operations.F
@@ -1910,11 +1910,7 @@ SUBROUTINE dbcsr_filter_anytype(matrix, eps, method, &
       TYPE(dbcsr_iterator)                               :: iter
 
       REAL(KIND=real_8), EXTERNAL                        :: DZNRM2
-#if defined (__ACCELERATE)
-      REAL(KIND=real_8), EXTERNAL                        :: SCNRM2
-#else
       REAL(KIND=real_4), EXTERNAL                        :: SCNRM2
-#endif
 
 !   ---------------------------------------------------------------------------
 

diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
@@ -95,23 +95,20 @@ set(dbcsr_unittest_common_SRCS dbcsr_test_add.F dbcsr_test_multiply.F)
 # instead of building a full-blown lib, it would be better to simply build an
 # OBJECT lib, but we would need cmake 3.12 to be able to specify
 # target_link_libraries on those to get the proper compile flags
-add_library(dbcsr_unittest_common STATIC ${dbcsr_unittest_common_SRCS})
+add_library(dbcsr_unittest_common OBJECT ${dbcsr_unittest_common_SRCS})
+target_link_libraries(dbcsr_unittest_common PUBLIC dbcsr)
 target_link_libraries(dbcsr_unittest_common PUBLIC ${BLAS_LIBRARIES}
                                                    ${LAPACK_LIBRARIES})
 if (OpenMP_FOUND)
   target_link_libraries(dbcsr_unittest_common PUBLIC OpenMP::OpenMP_Fortran)
 endif ()
 
-if (APPLE AND BLAS_LIBRARIES MATCHES "Accelerate")
-  target_compile_definitions(dbcsr_unittest_common PRIVATE __ACCELERATE)
-endif ()
-target_link_libraries(dbcsr_unittest_common PUBLIC dbcsr)
-
 # Compile Fortran tests
 foreach (dbcsr_test ${DBCSR_TESTS_FTN})
   add_executable(${dbcsr_test} ${${dbcsr_test}_SRCS})
-  target_link_libraries(${dbcsr_test} dbcsr_unittest_common)
+  target_link_libraries(${dbcsr_test} PUBLIC dbcsr_unittest_common)
   set_target_properties(${dbcsr_test} PROPERTIES LINKER_LANGUAGE Fortran)
+
   # register unittest executable with CMake
   if (USE_MPI)
     separate_arguments(MPIEXEC_PREFLAGS)
@@ -124,7 +121,6 @@ foreach (dbcsr_test ${DBCSR_TESTS_FTN})
     add_test(NAME ${dbcsr_test} COMMAND ${dbcsr_test})
   endif ()
   if (OpenMP_FOUND)
-    target_link_libraries(${dbcsr_test} OpenMP::OpenMP_Fortran)
     set_tests_properties(
       ${dbcsr_test} PROPERTIES ENVIRONMENT OMP_NUM_THREADS=${TEST_OMP_THREADS})
   endif ()

diff --git a/tests/dbcsr_test_add.F b/tests/dbcsr_test_add.F
@@ -377,11 +377,7 @@ SUBROUTINE dbcsr_check_add(test_name, matrix_a, dense_a_dbcsr, dense_a, dense_b,
 
       LOGICAL                                            :: valid
       REAL(real_4), ALLOCATABLE, DIMENSION(:)            :: work_sp
-#if defined (__ACCELERATE)
-      REAL(real_8), EXTERNAL                             :: clange, slamch, slange
-#else
       REAL(real_4), EXTERNAL                             :: clange, slamch, slange
-#endif
       REAL(real_8)                                       :: a_norm_dbcsr, a_norm_in, a_norm_out, &
                                                             b_norm, eps, residual
       REAL(real_8), ALLOCATABLE, DIMENSION(:)            :: work

diff --git a/tests/dbcsr_test_multiply.F b/tests/dbcsr_test_multiply.F
@@ -553,11 +553,7 @@ SUBROUTINE dbcsr_check_multiply(test_name, matrix_c, dense_c_dbcsr, dense_a, den
 
       LOGICAL                                            :: valid
       REAL(real_4), ALLOCATABLE, DIMENSION(:)            :: work_sp
-#if defined (__ACCELERATE)
-      REAL(real_8), EXTERNAL                             :: clange, slamch, slange
-#else
       REAL(real_4), EXTERNAL                             :: clange, slamch, slange
-#endif
       REAL(real_8)                                       :: a_norm, b_norm, c_norm_dbcsr, c_norm_in, &
                                                             c_norm_out, eps, eps_norm, residual
       REAL(real_8), ALLOCATABLE, DIMENSION(:)            :: work

diff --git a/tools/vecLibFort/CMakeLists.txt b/tools/vecLibFort/CMakeLists.txt
@@ -0,0 +1,7 @@
+add_library(vecLibFort STATIC vecLibFort.c)
+
+install(
+  TARGETS vecLibFort
+  EXPORT DBCSRTargets
+  LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}"
+  ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}")