Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

int2c2e doesn't work #211

Open
n-gao opened this issue Sep 20, 2024 · 1 comment
Open

int2c2e doesn't work #211

n-gao opened this issue Sep 20, 2024 · 1 comment
Assignees
@wxj6000
Labels
bug Something isn't working

Comments

@n-gao
Copy link

n-gao commented Sep 20, 2024

It appears that the int2c2e function is not working properly. The following code doesn't work

import pyscf
from pyscf.df.addons import make_auxmol
from gpu4pyscf.df.int3c2e import get_int2c2e
m = pyscf.gto.M(atom='''
        C   -1.1367    0.0103    0.0000
        C    0.1372   -0.0024    0.0000
        C    1.0258    1.2064    0.0000
        C    2.3997    1.1549    0.0000
        O    2.9930    0.0003    0.0000
        O    3.0209    2.3217    0.0000
        C    1.0136    2.4426    0.0000
        C   -0.3603    2.4961    0.0000
        H   -2.0364   -0.6442    0.0000
        H    1.7805   -0.5956    0.0000
        H   -2.0216    0.6637    0.0000
        H    0.0660   -1.0129    0.0000
        O   -0.5710    3.7755    0.0000
        C    3.9200    0.9435    0.0000
        H   -1.2845    3.9715    0.0000
        H    4.5367    1.2207    0.0000
    ''', basis='6-31G(d)')
aux_mol = make_auxmol(m, 'weigend')
int2c2e = get_int2c2e(m, aux_mol)

Error

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[19], [line 23](vscode-notebook-cell:?execution_count=19&line=23)
      [4](vscode-notebook-cell:?execution_count=19&line=4) m = pyscf.gto.M(atom='''
      [5](vscode-notebook-cell:?execution_count=19&line=5)         C   -1.1367    0.0103    0.0000
      [6](vscode-notebook-cell:?execution_count=19&line=6)         C    0.1372   -0.0024    0.0000
   (...)
     [20](vscode-notebook-cell:?execution_count=19&line=20)         H    4.5367    1.2207    0.0000
     [21](vscode-notebook-cell:?execution_count=19&line=21)     ''', basis='6-31G(d)')
     [22](vscode-notebook-cell:?execution_count=19&line=22) aux_mol = make_auxmol(m, 'weigend')
---> [23](vscode-notebook-cell:?execution_count=19&line=23) int2c2e = get_int2c2e(m, aux_mol)

File /ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1563, in get_int2c2e(mol, auxmol, direct_scf_tol)
   [1561](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1561) intopt = VHFOpt(mol, auxmol, 'int2e')
   [1562](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1562) intopt.build(direct_scf_tol, diag_block_with_triu=True, aosym=True)
-> [1563](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1563) int2c = get_int2c2e_sorted(mol, auxmol, intopt=intopt)
   [1564](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1564) aux_idx = np.argsort(intopt.aux_ao_idx)
   [1565](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1565) int2c = int2c[np.ix_(aux_idx, aux_idx)]

File /ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1478, in get_int2c2e_sorted(mol, auxmol, intopt, direct_scf_tol, aosym, omega, stream)
   [1476](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1476) strides = np.array([1, naux_cart, naux_cart, naux_cart*naux_cart], dtype=np.int32)
   [1477](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1477) for k_id, log_q_k in enumerate(intopt.aux_log_qs):
-> [1478](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1478)     bins_locs_k = _make_s_index_offsets(log_q_k, nbins)
   [1479](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1479)     cp_k_id = k_id + len(intopt.log_qs)
   [1480](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/df/int3c2e.py:1480)     for l_id, log_q_l in enumerate(intopt.aux_log_qs):

File /ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/scf/hf.py:1014, in _make_s_index_offsets(log_q, nbins, cutoff)
   [1012](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/scf/hf.py:1012) scale = nbins / np.log(min(cutoff, .1))
   [1013](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/scf/hf.py:1013) s_index = np.floor(scale * log_q).astype(np.int32)
-> [1014](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/scf/hf.py:1014) bins = np.bincount(s_index)
   [1015](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/scf/hf.py:1015) if bins.size < nbins:
   [1016](https://vscode-remote+ssh-002dremote-002bgpu06.vscode-resource.vscode-cdn.net/ceph/ssd/staff/gaoni/repos/mldft/.venv/lib/python3.12/site-packages/gpu4pyscf/scf/hf.py:1016)     bins = np.append(bins, np.zeros(nbins-bins.size, dtype=np.int32))

ValueError: 'list' argument must have no negative elements

Python 3.12.5
gpu4pyscf-cuda12x 1.0.2
gpu4pyscf-libxc-cuda12x 0.5
pyscf 2.6.2
numpy 1.26.4
cupy-cuda12x 13.3.0
cudatoolkit 12.5.0
@wxj6000 wxj6000 self-assigned this Sep 20, 2024
@wxj6000 wxj6000 added the bug Something isn't working label Sep 20, 2024
@wxj6000
Copy link
Collaborator

wxj6000 commented Sep 21, 2024

@n-gao Thank you for your feedback. The bug is fixed in #212. However, it is not necessary faster than its implementation on CPU. Mostly, we are still using int2c2e on CPU. It probably can be much faster than CPU for large systems with further optimization.

And I moved int2c2e to avoid possible confusions. Here is some test script.

import cupy
import pyscf
from pyscf.df.addons import make_auxmol
from gpu4pyscf.scf.int2c2e import get_int2c2e

m = pyscf.gto.M(atom='''
        C   -1.1367    0.0103    0.0000
        C    0.1372   -0.0024    0.0000
        C    1.0258    1.2064    0.0000
        C    2.3997    1.1549    0.0000
        O    2.9930    0.0003    0.0000
        O    3.0209    2.3217    0.0000
        C    1.0136    2.4426    0.0000
        C   -0.3603    2.4961    0.0000
        H   -2.0364   -0.6442    0.0000
        H    1.7805   -0.5956    0.0000
        H   -2.0216    0.6637    0.0000
        H    0.0660   -1.0129    0.0000
        O   -0.5710    3.7755    0.0000
        C    3.9200    0.9435    0.0000
        H   -1.2845    3.9715    0.0000
        H    4.5367    1.2207    0.0000
    ''', basis='6-31G(d)')
aux_mol = make_auxmol(m, 'weigend')

start_event = cupy.cuda.Event()
end_event = cupy.cuda.Event()
start_event.record()
for i in range(100):
    int2c2e = get_int2c2e(aux_mol)
end_event.record()
end_event.synchronize()
elapsed_time = cupy.cuda.get_elapsed_time(start_event, end_event) 
print(f'{elapsed_time/1000:.3f} s', 'with GPU')

start_event = cupy.cuda.Event()
end_event = cupy.cuda.Event()
start_event.record()
for i in range(100):
    int2c2e = aux_mol.intor('int2c2e_sph')
    int2c2e = cupy.asarray(int2c2e)
end_event.record()
end_event.synchronize()
elapsed_time = cupy.cuda.get_elapsed_time(start_event, end_event) 
print(f'{elapsed_time/1000:.3f} s', 'with CPU')

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@wxj6000 wxj6000

Labels
bug Something isn't working
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@wxj6000 @n-gao