[cmake] BTAS fixes

INSTALL.md

@@ -40,7 +40,7 @@ Both methods are supported. However, for most users we _strongly_ recommend to b
   - Boost.Container: header-only
   - Boost.Test: header-only or (optionally) as a compiled library, *only used for unit testing*
   - Boost.Range: header-only, *only used for unit testing*
-- [BTAS](https://github.com/ValeevGroup/BTAS), tag 0dc805fd7f2dea5e56eff94ab1c44b2b2397edd7 . If usable BTAS installation is not found, TiledArray will download and compile
+- [BTAS](https://github.com/ValeevGroup/BTAS), tag 8ac131460e05e9470779880e15acc5642a451e7a . If usable BTAS installation is not found, TiledArray will download and compile
   BTAS from source. *This is the recommended way to compile BTAS for all users*.
 - [MADNESS](https://github.com/m-a-d-n-e-s-s/madness), tag 51f0615f094bca5110ea149100e39c0edb25a4f2 .
   Only the MADworld runtime and BLAS/LAPACK C API component of MADNESS is used by TiledArray.
@@ -275,13 +275,22 @@ algebra in TA:
   BLAS++/LAPACK++ during the TA configuration. There are 2 mechanisms by which BLAS++/LAPACK++
   discover BLAS/LAPACK:
   - _the built-in custom discovery kit_; no options exist to provide any control
-  - standard CMake BLAS/LAPACK modules.
+  - standard CMake [BLAS](https://cmake.org/cmake/help/latest/module/FindBLAS.html)/[LAPACK](https://cmake.org/cmake/help/latest/module/FindLAPACK.html) modules.
 
   The latter is used if CMake cache variable `BLA_VENDOR` is specified:
   - `BLA_VENDOR` -- controls which vendor BLAS/LAPACK library will be sought
-  (see [CMake docs](https://cmake.org/cmake/help/latest/module/FindLAPACK.html));
-  by default all possible vendor libraries will be considered. E.g., to force the use of the Accelerate
-  framework on MacOS use `-DBLA_VENDOR=Apple`.
+    (see [CMake docs](https://cmake.org/cmake/help/latest/module/FindLAPACK.html));
+    by default all possible vendor libraries will be considered. E.g., to force the use of the Accelerate
+    framework on MacOS use `-DBLA_VENDOR=Apple`.
+
+  Unfortunately, if the standard CMake modules discover BLAS/LAPACK,
+  BLAS++/LAPACK++ will not attempt to discover their name mangling convention.
+  To specify the name mangling to be assumed by BLAS++/LAPACK++ specify CMake cache variable `LINALG_MANGLING`:
+  - `LINALG_MANGLING` -- specifies the name mangling assumed by BLAS++/LAPACK++
+     when using BLAS/LAPACK. Valid values are:
+    - `lower`: function/variable `dgemm` will be mangled to `dgemm`,
+    - `UPPER`: function/variable `dgemm` will be mangled to `DGEMM`,
+    - `lower_`: function/variable `dgemm` will be mangled to `dgemm_` (default).
 
   More information can be found in the installation instructions for
   [BLAS++](https://icl.bitbucket.io/blaspp/md__i_n_s_t_a_l_l.html) and

cmake/modules/FindOrFetchBTAS.cmake

@@ -10,9 +10,16 @@ endif (NOT TARGET BTAS::BTAS)
 # if not found, build via FetchContent
 if (NOT TARGET BTAS::BTAS)
 
+  # BTAS will load BLAS++/LAPACK++ ... if those use CMake's FindBLAS/FindLAPACK (as indicated by defined BLA_VENDOR)
+  # will need to specify Fortran linkage convention ... manually for now, switching to NWX's linear algebra discovery
+  # is necessary to handle all the corner cases for automatic discovery
+  if (DEFINED BLA_VENDOR)
+    set(_linalgpp_use_standard_linalg_kits TRUE)
+  endif(DEFINED BLA_VENDOR)
+
   if (NOT TILEDARRAY_HAS_CUDA)
     # tell BLAS++/LAPACK++ to ignore CUDA
-    set(use_cuda FALSE CACHE BOOL "Whether to look for CUDA-enabled libs in BLAS++/LAPACK++")
+    set(gpu_backend none CACHE STRING "The device backend to use for Linalg++")
   endif()
 
   include(FetchContent)
@@ -35,6 +42,26 @@ if (NOT TARGET BTAS::BTAS)
   # set BTAS_CONFIG to the install location so that we know where to find it
   set(BTAS_CONFIG ${CMAKE_INSTALL_PREFIX}/${BTAS_INSTALL_CMAKEDIR}/btas-config.cmake)
 
+  # define macros specifying Fortran mangling convention, if necessary
+  if (_linalgpp_use_standard_linalg_kits)
+    if (NOT TARGET blaspp AND NOT TARGET lapackpp)
+      message(FATAL_ERROR "blaspp or lapackpp targets missing")
+    endif(NOT TARGET blaspp AND NOT TARGET lapackpp)
+    if (LINALG_MANGLING STREQUAL lower)
+      target_compile_definitions(blaspp PUBLIC -DBLAS_FORTRAN_LOWER=1)
+      target_compile_definitions(lapackpp PUBLIC -DLAPACK_FORTRAN_LOWER=1)
+    elseif(LINALG_MANGLING STREQUAL UPPER OR LINALG_MANGLING STREQUAL upper)
+      target_compile_definitions(blaspp PUBLIC -DBLAS_FORTRAN_UPPER=1)
+      target_compile_definitions(lapackpp PUBLIC -DLAPACK_FORTRAN_UPPER=1)
+    else()
+      if (NOT LINALG_MANGLING STREQUAL lower_)
+        message(WARNING "Linear algebra libraries' mangling convention not specified; specify -DLINALG_MANGLING={lower,lower_,UPPER}, if needed; will assume lower_")
+      endif(NOT LINALG_MANGLING STREQUAL lower_)
+      target_compile_definitions(blaspp PUBLIC -DBLAS_FORTRAN_ADD_=1)
+      target_compile_definitions(lapackpp PUBLIC -DLAPACK_FORTRAN_ADD_=1)
+    endif()
+  endif (_linalgpp_use_standard_linalg_kits)
+
 endif(NOT TARGET BTAS::BTAS)
 
 # postcond check

external/versions.cmake

@@ -24,8 +24,8 @@ set(TA_TRACKED_MADNESS_PREVIOUS_TAG 34ef8e6d6c61098d44ca559a530b4096e40b9a01)
 set(TA_TRACKED_MADNESS_VERSION 0.10.1)
 set(TA_TRACKED_MADNESS_PREVIOUS_VERSION 0.10.1)
 
-set(TA_TRACKED_BTAS_TAG 0dc805fd7f2dea5e56eff94ab1c44b2b2397edd7)
-set(TA_TRACKED_BTAS_PREVIOUS_TAG b7bddfcef89612652878854a447bacf4d3f6a5b0)
+set(TA_TRACKED_BTAS_TAG 8ac131460e05e9470779880e15acc5642a451e7a )
+set(TA_TRACKED_BTAS_PREVIOUS_TAG 0dc805fd7f2dea5e56eff94ab1c44b2b2397edd7)
 
 set(TA_TRACKED_CUTT_TAG 0e8685bf82910bc7435835f846e88f1b39f47f09)
 set(TA_TRACKED_CUTT_PREVIOUS_TAG 592198b93c93b7ca79e7900b9a9f2e79f9dafec3)