compilation fails on main branch

[hatemhelal]

I'm trying out this project and ran into a few initial hiccups and wondered if there is a known build recipe that I can follow?

Here is what I tried to use:

 sudo apt install cmake ccache clang ninja-build libopenblas-dev liblapacke-dev
 cmake .. -DLIBINTX_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native -DCMAKE_C_COMPILER_LAUNCHER=ccache -G Ninja
 
 # compile the tests too...
 ninja all.tests

which fails pretty fast due to this:

/root/libintx/src/libintx/ao/md/md3.cc: In member function ‘void libintx::md::IntegralEngine<3>::compute(const Params&, const std::vector<int>&, const std::vector<libintx::pair<int, int> >&, libintx::BraKet<const double*>, const Visitor&)’:
/root/libintx/src/libintx/ao/md/md3.cc:35:52: error: expected ‘#pragma omp’ clause before ‘;’ token
   35 | #pragma omp parallel num_threads(this->num_threads);
      |                                                    ^

removing the ; and the build gets further but eventually fails with the longer error copied below (failed override?). Looking at the source code I see variants of the compute interface, some of which are commented out so here I am asking for help - thanks in advance!!!

FAILED: src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o
/usr/local/cuda/bin/nvcc -forward-unknown-to-host-compiler -DLIBINTX_GPU_MAX_SHMEM=49152 -DLIBINTX_GPU_MD_MD3_KERNEL_KET=0 -DLIBINTX_GPU_MD_MD3_KERNEL_X=1 -I/root/libintx/src -I/root/libintx/include -I/root/libintx/include/eigen3 -I/root/libintx/build -expt-relaxed-constexpr --resource-usage -O3 -DNDEBUG -std=c++20 -arch=native -Xcompiler=-fPIC -MD -MT src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o -MF src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o.d -x cu -c /root/libintx/src/libintx/gpu/md/md3.kernel.cu -o src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o
/root/libintx/src/libintx/gpu/md/engine.h(29): error: member function declared with "override" does not override a base class member
      void compute(
           ^

/root/libintx/src/libintx/gpu/md/engine.h(29): warning #997-D: function "libintx::ao::IntegralEngine<3>::compute(libintx::Operator, const std::vector<libintx::Index1, std::allocator<libintx::Index1>> &, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, libintx::BraKet<const double *, const double *>, double *, const std::array<size_t, 2UL> &)" is hidden by "libintx::gpu::md::IntegralEngine<3>::compute" -- virtual function override intended?
      void compute(
           ^

Remark: The warnings can be suppressed with "-diag-suppress <warning-number>"

/root/libintx/src/libintx/gpu/md/engine.h(82): error: member function declared with "override" does not override a base class member
      void compute(
           ^

/root/libintx/src/libintx/gpu/md/engine.h(82): warning #997-D: function "libintx::ao::IntegralEngine<4>::compute(libintx::Operator, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, libintx::BraKet<const double *, const double *>, double *, const std::array<size_t, 2UL> &)" is hidden by "libintx::gpu::md::IntegralEngine<4>::compute" -- virtual function override intended?
      void compute(
           ^

2 errors detected in the compilation of "/root/libintx/src/libintx/gpu/md/md3.kernel.cu".

[asadchev]

Yes, GPU libraries been broken since SIMD work. I will update branch in next few days where GPU is functional.

…

On Mon, Sep 15, 2025, 12:56 PM Hatem Helal ***@***.***> wrote: *hatemhelal* created an issue (ValeevGroup/libintx#6) <#6> I'm trying out this project and ran into a few initial hiccups and wondered if there is a known build recipe that I can follow? Here is what I tried to use: sudo apt install cmake ccache clang ninja-build libopenblas-dev liblabpe-dev cmake .. -DLIBINTX_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native -DCMAKE_C_COMPILER_LAUNCHER=ccache -G Ninja # compile the tests too... ninja all.tests which fails pretty fast due to this: /root/libintx/src/libintx/ao/md/md3.cc: In member function ‘void libintx::md::IntegralEngine<3>::compute(const Params&, const std::vector<int>&, const std::vector<libintx::pair<int, int> >&, libintx::BraKet<const double*>, const Visitor&)’: /root/libintx/src/libintx/ao/md/md3.cc:35:52: error: expected ‘#pragma omp’ clause before ‘;’ token 35 | #pragma omp parallel num_threads(this->num_threads); | ^ removing the ; and the build gets further but eventually fails with the longer error copied below (failed override?). Looking at the source code I see variants of the compute interface, some of which are commented out so here I am asking for help - thanks in advance!!! FAILED: src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o /usr/local/cuda/bin/nvcc -forward-unknown-to-host-compiler -DLIBINTX_GPU_MAX_SHMEM=49152 -DLIBINTX_GPU_MD_MD3_KERNEL_KET=0 -DLIBINTX_GPU_MD_MD3_KERNEL_X=1 -I/root/libintx/src -I/root/libintx/include -I/root/libintx/include/eigen3 -I/root/libintx/build -expt-relaxed-constexpr --resource-usage -O3 -DNDEBUG -std=c++20 -arch=native -Xcompiler=-fPIC -MD -MT src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o -MF src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o.d -x cu -c /root/libintx/src/libintx/gpu/md/md3.kernel.cu -o src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o /root/libintx/src/libintx/gpu/md/engine.h(29): error: member function declared with "override" does not override a base class member void compute( ^ /root/libintx/src/libintx/gpu/md/engine.h(29): warning #997-D: function "libintx::ao::IntegralEngine<3>::compute(libintx::Operator, const std::vector<libintx::Index1, std::allocator<libintx::Index1>> &, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, libintx::BraKet<const double *, const double *>, double *, const std::array<size_t, 2UL> &)" is hidden by "libintx::gpu::md::IntegralEngine<3>::compute" -- virtual function override intended? void compute( ^ Remark: The warnings can be suppressed with "-diag-suppress <warning-number>" /root/libintx/src/libintx/gpu/md/engine.h(82): error: member function declared with "override" does not override a base class member void compute( ^ /root/libintx/src/libintx/gpu/md/engine.h(82): warning #997-D: function "libintx::ao::IntegralEngine<4>::compute(libintx::Operator, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, libintx::BraKet<const double *, const double *>, double *, const std::array<size_t, 2UL> &)" is hidden by "libintx::gpu::md::IntegralEngine<4>::compute" -- virtual function override intended? void compute( ^ 2 errors detected in the compilation of "/root/libintx/src/libintx/gpu/md/md3.kernel.cu". — Reply to this email directly, view it on GitHub <#6>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABH62KCPHQZ7X4TQYHSMS233S3OR3AVCNFSM6AAAAACGR4CGDSVHI2DSMVQWIX3LMV43ASLTON2WKOZTGQYTQNJWGE2TIOA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[asadchev]

Can you build with just CPU libraries for now? Once fix is pushed the interfaces will be identical save for constructors. PS, apologies for late response, didn't see notification.

…

On Sun, Sep 21, 2025, 1:53 PM Andrey Asadchev ***@***.***> wrote: Yes, GPU libraries been broken since SIMD work. I will update branch in next few days where GPU is functional. On Mon, Sep 15, 2025, 12:56 PM Hatem Helal ***@***.***> wrote: > *hatemhelal* created an issue (ValeevGroup/libintx#6) > <#6> > > I'm trying out this project and ran into a few initial hiccups and > wondered if there is a known build recipe that I can follow? > > Here is what I tried to use: > > sudo apt install cmake ccache clang ninja-build libopenblas-dev liblabpe-dev > cmake .. -DLIBINTX_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native -DCMAKE_C_COMPILER_LAUNCHER=ccache -G Ninja > > # compile the tests too... > ninja all.tests > > which fails pretty fast due to this: > > /root/libintx/src/libintx/ao/md/md3.cc: In member function ‘void libintx::md::IntegralEngine<3>::compute(const Params&, const std::vector<int>&, const std::vector<libintx::pair<int, int> >&, libintx::BraKet<const double*>, const Visitor&)’: > /root/libintx/src/libintx/ao/md/md3.cc:35:52: error: expected ‘#pragma omp’ clause before ‘;’ token > 35 | #pragma omp parallel num_threads(this->num_threads); > | ^ > > removing the ; and the build gets further but eventually fails with the > longer error copied below (failed override?). Looking at the source code I > see variants of the compute interface, some of which are commented out so > here I am asking for help - thanks in advance!!! > > FAILED: src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o > /usr/local/cuda/bin/nvcc -forward-unknown-to-host-compiler -DLIBINTX_GPU_MAX_SHMEM=49152 -DLIBINTX_GPU_MD_MD3_KERNEL_KET=0 -DLIBINTX_GPU_MD_MD3_KERNEL_X=1 -I/root/libintx/src -I/root/libintx/include -I/root/libintx/include/eigen3 -I/root/libintx/build -expt-relaxed-constexpr --resource-usage -O3 -DNDEBUG -std=c++20 -arch=native -Xcompiler=-fPIC -MD -MT src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o -MF src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o.d -x cu -c /root/libintx/src/libintx/gpu/md/md3.kernel.cu -o src/libintx/gpu/md/CMakeFiles/libintx.gpu.md3.kernel.1_0.dir/md3.kernel.cu.o > /root/libintx/src/libintx/gpu/md/engine.h(29): error: member function declared with "override" does not override a base class member > void compute( > ^ > > /root/libintx/src/libintx/gpu/md/engine.h(29): warning #997-D: function "libintx::ao::IntegralEngine<3>::compute(libintx::Operator, const std::vector<libintx::Index1, std::allocator<libintx::Index1>> &, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, libintx::BraKet<const double *, const double *>, double *, const std::array<size_t, 2UL> &)" is hidden by "libintx::gpu::md::IntegralEngine<3>::compute" -- virtual function override intended? > void compute( > ^ > > Remark: The warnings can be suppressed with "-diag-suppress <warning-number>" > > /root/libintx/src/libintx/gpu/md/engine.h(82): error: member function declared with "override" does not override a base class member > void compute( > ^ > > /root/libintx/src/libintx/gpu/md/engine.h(82): warning #997-D: function "libintx::ao::IntegralEngine<4>::compute(libintx::Operator, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, const std::vector<libintx::Index2, std::allocator<libintx::Index2>> &, libintx::BraKet<const double *, const double *>, double *, const std::array<size_t, 2UL> &)" is hidden by "libintx::gpu::md::IntegralEngine<4>::compute" -- virtual function override intended? > void compute( > ^ > > 2 errors detected in the compilation of "/root/libintx/src/libintx/gpu/md/md3.kernel.cu". > > — > Reply to this email directly, view it on GitHub > <#6>, or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ABH62KCPHQZ7X4TQYHSMS233S3OR3AVCNFSM6AAAAACGR4CGDSVHI2DSMVQWIX3LMV43ASLTON2WKOZTGQYTQNJWGE2TIOA> > . > You are receiving this because you are subscribed to this thread.Message > ID: ***@***.***> >

[hatemhelal]

Thanks @asadchev and happy to share that I've built the cpu-lib after disabling the simd feature. Running the tests outputs several errors...Would it be possible to share a last known good commit hash that I can give a try?

[hatemhelal]

Here is the output from the test run:

    Start 1: pure.test
1/6 Test #1: pure.test ........................   Passed    0.13 sec
    Start 2: boys.test
2/6 Test #2: boys.test ........................***Failed    0.22 sec
    Start 3: libintx.md.test
3/6 Test #3: libintx.md.test ..................   Passed    0.18 sec
    Start 4: libintx.md2.test
4/6 Test #4: libintx.md2.test .................   Passed  256.68 sec
    Start 5: libintx.md3.test
5/6 Test #5: libintx.md3.test .................***Failed  353.35 sec
    Start 6: libintx.md4.test
6/6 Test #6: libintx.md4.test .................***Timeout 1500.02 sec

50% tests passed, 3 tests failed out of 6

Total Test time (real) = 2110.65 sec

The following tests FAILED:
	  2 - boys.test (Failed)
	  5 - libintx.md3.test (Failed)
	  6 - libintx.md4.test (Timeout)

[asadchev]

That should be correct branch, sha should be 54f47 Tell me your OS, compiler, maximum L, etc. For high L there is some noise in results and likely to see few errors. Reference is computed via completely different methods (including Boys) and default epsilon may be too optimistic. Paste few error entries too, will help me narrow it down.

…

On Sun, Sep 21, 2025, 5:40 PM Hatem Helal ***@***.***> wrote: *hatemhelal* left a comment (ValeevGroup/libintx#6) <#6 (comment)> Here is the output from the test run: Start 1: pure.test 1/6 Test #1: pure.test ........................ Passed 0.13 sec Start 2: boys.test 2/6 Test #2: boys.test ........................***Failed 0.22 sec Start 3: libintx.md.test 3/6 Test #3: libintx.md.test .................. Passed 0.18 sec Start 4: libintx.md2.test 4/6 Test #4: libintx.md2.test ................. Passed 256.68 sec Start 5: libintx.md3.test 5/6 Test #5: libintx.md3.test .................***Failed 353.35 sec Start 6: libintx.md4.test 6/6 Test #6: libintx.md4.test .................***Timeout 1500.02 sec 50% tests passed, 3 tests failed out of 6 Total Test time (real) = 2110.65 sec The following tests FAILED: 2 - boys.test (Failed) 5 - libintx.md3.test (Failed) 6 - libintx.md4.test (Timeout) — Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABH62KDRIYO2SIZ3GX5O5CT3T4EMPAVCNFSM6AAAAACGR4CGDSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMJWGI2DSOBSGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[hatemhelal]

Thank you, appreciate the help narrowing this down. My env:

• macOS on an M4 Apple silicon
• Apple clang v17.0.0
• system BLAS (apple accelerate framework), wonder if this is worth trying openblas instead.
• LIBINTX_MAX_L=3
• LIBINTX_SIMD=OFF

Looks like the boys test is failing due to small eps level differences (here is a representative example):

/Users/hatem/code/libintx/tests/boys.test.cc:14:
TEST CASE:  chebyshev interpolation

/Users/hatem/code/libintx/tests/boys.test.cc:43: ERROR: CHECK( px == ref.epsilon(1e-15) ) is NOT correct!
  values: CHECK( 0.0116948957 == 0.0116948957 @ [ 7 2.000000 0 ] )
  logged: Validating F(7)

both the md3 and md4 test failures looks more concerning - there are many errors where the numerical differences are far off, here is one example:

/Users/hatem/code/libintx/tests/libintx.md3.test.cc:96:
TEST CASE:  libintx.md3.Coulomb

/Users/hatem/code/libintx/tests/libintx.md3.test.cc:60: ERROR: CHECK( result(i,idx...,(int)kl) == ref.epsilon(1e-9) ) is NOT correct!
  values: CHECK( 16.3089062585 == 6514.4833019485 @ [ 0 0 0 ] )

The full test log is ~75MB but I can share a link to that if that is at all useful. Thank you and appreciate the help!

[asadchev]

This does not seem like an obvious fix. Are you ok taking conversation to email?

…

On Mon, Sep 22, 2025, 7:56 AM Hatem Helal ***@***.***> wrote: *hatemhelal* left a comment (ValeevGroup/libintx#6) <#6 (comment)> Thank you, appreciate the help narrowing this down. My env: - macOS on an M4 Apple silicon - Apple clang v17.0.0 - system BLAS (apple accelerate framework), wonder if this is worth trying openblas instead. - LIBINTX_MAX_L=3 - LIBINTX_SIMD=OFF Looks like the boys test is failing due to small eps level differences (here is a representative example): /Users/hatem/code/libintx/tests/boys.test.cc:14: TEST CASE: chebyshev interpolation /Users/hatem/code/libintx/tests/boys.test.cc:43: ERROR: CHECK( px == ref.epsilon(1e-15) ) is NOT correct! values: CHECK( 0.0116948957 == 0.0116948957 @ [ 7 2.000000 0 ] ) logged: Validating F(7) both the md3 and md4 test failures looks more concerning - there are many errors where the numerical differences are far off, here is one example: /Users/hatem/code/libintx/tests/libintx.md3.test.cc:96: TEST CASE: libintx.md3.Coulomb /Users/hatem/code/libintx/tests/libintx.md3.test.cc:60: ERROR: CHECK( result(i,idx...,(int)kl) == ref.epsilon(1e-9) ) is NOT correct! values: CHECK( 16.3089062585 == 6514.4833019485 @ [ 0 0 0 ] ) The full test log is ~75MB but I can share a link to that if that is at all useful. Thank you and appreciate the help! — Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABH62KG35YA5QT7TJ3J6SU33T7IVNAVCNFSM6AAAAACGR4CGDSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMJYGIZDSNBWGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>