Add ifx support

[Leonard-Reuter]

This is a draft pull request, I simply copied Intel.cmake to IntelLLVM.cmake. With ifx (IFORT) 2021.4.0 Beta 20210924 pFUnit does however not compile due to an compilation error in gFTL-shared:

/home/reuter/GitProjects/pFUnit_github/extern/fArgParse/extern/gFTL-shared/src/v1/demo.F90(299): error #6197: An assignment of different structure types is invalid.   [INTEGER32INTEGER32_MAP]
   Integer32Integer_map = Integer32Integer32_map
--------------------------^
compilation aborted for /home/reuter/GitProjects/pFUnit_github/extern/fArgParse/extern/gFTL-shared/src/v1/demo.F90 (code 1)
make[2]: *** [extern/fArgParse/extern/gFTL-shared/src/v1/CMakeFiles/demo.x.dir/build.make:75: extern/fArgParse/extern/gFTL-shared/src/v1/CMakeFiles/demo.x.dir/demo.F90.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:980: extern/fArgParse/extern/gFTL-shared/src/v1/CMakeFiles/demo.x.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....

[mathomp4]

@Leonard-Reuter You'd probably need to add an IntelLLVM.cmake file to gFTL, gFTL-shared, fArgParse, etc. Essentially all the GFE libraries. I can work with @tclune to start spreading that through GFE.

But as I say in https://github.com/Goddard-Fortran-Ecosystem/pFUnit/pull/403#issuecomment-1331091155, at least with ifx 2022.2 the compiler is still incomplete.

If you are willing to try it out, you can do:

gh repo clone Goddard-Fortran-Ecosystem/GFE
cd GFE
git submodule update --init

to get the latest GFE. Then you can go into each of the submodules and do the cp Intel.cmake IntelLLVM.cmake to prepare it, then try doing a build of GFE. If ifx 2022.4 beta works, maybe that's a good sign ifx is ready soon.

[scivision]

CC=icx FC=ifort does work:

-- The Fortran compiler identification is Intel 2021.7.0.20220726
-- The C compiler identification is IntelLLVM 2022.2.0

However as noted ifx 2022.2 is not yet ready for pFUnit.

[scivision]

With this patch, ifx 2022.2.0 does build. The only other change I did from pfUnit main branch was copying cmake/Intel.cmake => cmake/IntelLLVM.cmake as above.

diff --git a/src/funit/core/NameFilter.F90 b/src/funit/core/NameFilter.F90
index 8483df4..1f3f3da 100644
--- a/src/funit/core/NameFilter.F90
+++ b/src/funit/core/NameFilter.F90
@@ -30,8 +30,12 @@ contains
   logical function filter(this, a_test)
     class(NameFilter), intent(in) :: this
     class(Test), intent(in) :: a_test
+
+    ! workaround for oneAPI ifx 2022.2.0
+    character(:), allocatable :: name
+    name = a_test%getName()

-    filter = index(a_test%getName(), this%pattern) > 0
+    filter = index(name, this%pattern) > 0
   end function filter

diff --git a/src/funit/fhamcrest/StringContains.F90 b/src/funit/fhamcrest/StringContains.F90
index 127d757..ad23b83 100644
--- a/src/funit/fhamcrest/StringContains.F90
+++ b/src/funit/fhamcrest/StringContains.F90
@@ -35,7 +35,12 @@ contains
     class(StringContains), intent(in) :: this
     character(*), intent(in) :: item

-    eval_substring_of = index(this%converted(item), this%converted(this%get_substring())) > 0
+    ! workaround for Intel oneAPI ifx 2022.2.0
+    character(:), allocatable :: converted, convsub
+    converted = this%converted(item)
+    convsub = this%converted(this%get_substring())
+
+    eval_substring_of = index(converted, convsub) > 0
   end function eval_substring_of

 end module pf_StringContains
diff --git a/src/funit/fhamcrest/StringEndsWith.F90 b/src/funit/fhamcrest/StringEndsWith.F90
index e8f7a31..ce548d1 100644
--- a/src/funit/fhamcrest/StringEndsWith.F90
+++ b/src/funit/fhamcrest/StringEndsWith.F90
@@ -35,8 +35,13 @@ contains
     character(*), intent(in) :: item

     integer :: idx_substring
+
+    ! workaround for Intel oneAPI ifx 2022.2.0
+    character(:), allocatable :: buf1, buf2
+    buf1 = this%converted(item)
+    buf2 = this%converted(this%get_substring())

-    idx_substring = index(this%converted(item), this%converted(this%get_substring()),back=.true.)
+    idx_substring = index(buf1, buf2, back=.true.)

     eval_substring_of = idx_substring == (len(item) - len(this%get_substring()) + 1)
   end function eval_substring_of
diff --git a/src/funit/fhamcrest/StringStartsWith.F90 b/src/funit/fhamcrest/StringStartsWith.F90
index 4caf65b..fb90ef5 100644
--- a/src/funit/fhamcrest/StringStartsWith.F90
+++ b/src/funit/fhamcrest/StringStartsWith.F90
@@ -33,7 +33,13 @@ contains
   logical function eval_substring_of(this, item)
     class(StringStartsWith), intent(in) :: this
     character(*), intent(in) :: item
-    eval_substring_of = index(this%converted(item), this%converted(this%get_substring())) == 1
+
+    ! workaround for Intel oneAPI ifx 2022.2.0
+    character(:), allocatable :: converted, convsub
+    converted = this%converted(item)
+    convsub = this%converted(this%get_substring())
+
+    eval_substring_of = index(converted, convsub) == 1
   end function eval_substring_of

 end module pf_StringStartsWith

Running ctest results in:

1/6 Test #2: robust_tests.x ...................   Passed    0.00 sec
2/6 Test #1: old_tests ........................***Failed    0.00 sec
3/6 Test #4: fhamcrest_tests.x ................***Failed    0.01 sec
4/6 Test #3: new_tests.x ......................   Passed    0.01 sec
5/6 Test #5: mpi-tests ........................***Failed    1.99 sec
6/6 Test #6: new_ptests .......................   Passed    1.99 sec

50% tests passed, 3 tests failed out of 6

Total Test time (real) =   1.99 sec

The following tests FAILED:
          1 - old_tests (Failed)
          4 - fhamcrest_tests.x (Failed)
          5 - mpi-tests (Failed)

With the same patched code, icx + ifort does pass all ctest. The failing ifx ctest is segfaults.

[scivision]

rebuilding using

cmake -Bbuildi -DCMAKE_C_COMPILER=icx -DCMAKE_Fortran_COMPILER=ifx  -DSKIP_MPI=yes -DSKIP_OPENMP=yes -DSKIP_FHAMCREST=yes -DSKIP_ROBUST=yes

the ctest output is then

1/2 Test #2: new_tests.x ......................   Passed    0.01 sec
2/2 Test #1: old_tests ........................   Passed    0.01 sec

100% tests passed, 0 tests failed out of 2

Total Test time (real) =   0.01 sec

[scivision]

This patch helps Intel compiler build by not overriding necessary factory options. Especially on windows, where numerous errors occur on any source file before this patch

diff --git a/cmake/Intel.cmake b/cmake/Intel.cmake
index 3dbbea8..7405fcb 100644
--- a/cmake/Intel.cmake
+++ b/cmake/Intel.cmake
@@ -1,22 +1,21 @@
 # Compiler specific flags for Intel Fortran compiler

+set(check_all -check)
+# check without options == "check all". This helps avoid command line processing issues
+
+set(traceback -traceback)
 if(WIN32)
-  set(no_optimize "-Od")
-  set(check_all "-check:all")
+  set(disable_warning_for_long_names /Qdiag-disable:5462)
 else()
-  set(no_optimize "-O0")
-  set(check_all "-check all")
+  set(disable_warning_for_long_names -diag-disable=5462)
 endif()
-
-
-set(disable_warning_for_long_names "-diag-disable 5462")
-set(traceback "-traceback")
-set(cpp "-cpp")

+set(common_flags "${disable_warning_for_long_names}")

-set(common_flags "${cpp} ${disable_warning_for_long_names}")
-set(CMAKE_Fortran_FLAGS_DEBUG  "-O0 -g ${common_flags} ${traceback} ${no_optimize} ${check_all}")
-set(CMAKE_Fortran_FLAGS_RELEASE "-O3 ${common_flags}")
+add_compile_options(
+"$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:${traceback};${check_all}>"
+"$<$<COMPILE_LANGUAGE:Fortran>:${common_flags}>"
+)

-add_definitions(-D_INTEL)
-add_definitions(-D__ifort_18)
+add_compile_definitions(_INTEL)
+add_compile_definitions(__ifort_18)

[Leonard-Reuter]

Just mentioning the related issue: https://github.com/Goddard-Fortran-Ecosystem/pFUnit/issues/336

[scivision]

Build is now successful with Intel 2023.0 and current pfUnit master, just copying Intel.cmake to IntelLLVM.cmake.

However, all tests fail with default configuration.

cmake -Bbuildi
-- The Fortran compiler identification is IntelLLVM 2023.0.0
-- The C compiler identification is IntelLLVM 2023.0.0
cmake --build buildi
ctest --test-dir buildi

1/6 Test #1: old_tests ........................***Failed    0.00 sec
2/6 Test #4: fhamcrest_tests.x ................***Failed    0.00 sec
3/6 Test #3: new_tests.x ......................***Failed    0.01 sec
4/6 Test #2: robust_tests.x ...................Subprocess aborted***Exception:   0.17 sec
5/6 Test #6: new_ptests .......................***Failed    1.97 sec
6/6 Test #5: mpi-tests ........................***Failed    1.98 sec

If I use the following configuration, tests also pass:

cmake  -DSKIP_MPI=yes -DSKIP_OPENMP=yes -DSKIP_FHAMCREST=yes -DSKIP_ROBUST=yes

1/2 Test #2: new_tests.x ...................... Passed 0.01 sec 2/2 Test #1: old_tests ........................ Passed 0.02 sec