merged locally

.codebase/pipelines/smoke_test.yml

@@ -0,0 +1,13 @@
+name: Python
+trigger: change
+jobs:
+  benchmarks:
+    image: hub.byted.org/codebase/ci_python_39
+    steps:
+      - name: test
+        commands:
+          - pip install pytest
+      - uses: actions/
+        inputs:
+          file: coverage.xml
+

.vscode/settings.json

@@ -1,4 +1,7 @@
 {
     "C_Cpp.errorSquiggles": "disabled",
-    "cmake.sourceDirectory": "/mlx_devbox/users/xiaojie.wu/playground/gpu4pyscf/build/lib.linux-x86_64-cpython-39/gpu4pyscf/lib"
+    "cmake.sourceDirectory": "/mlx_devbox/users/xiaojie.wu/playground/gpu4pyscf/build/lib.linux-x86_64-cpython-39/gpu4pyscf/lib",
+    "[python]": {
+        "editor.defaultFormatter": "ms-python.autopep8"
+    }
 }

README.md

@@ -110,3 +110,24 @@ Speedup with GPU4PySCF v0.6.0 over Q-Chem 6.1 (Desity fitting, SCF, def2-tzvpp,
 
 Find more benchmarks in gpu4pyscf/benchmarks
 
+Benchmarks
+--------
+Speedup with GPU4PySCF v0.6.0 over Q-Chem 6.1 (Desity fitting, SCF, def2-tzvpp, def2-universal-jkfit, (99,590))
+
+| mol               |   natm |    LDA |    PBE |   B3LYP |    M06 |   wB97m-v |
+|:------------------|-------:|-------:|-------:|--------:|-------:|----------:|
+| 020_Vitamin_C     |     20 |   2.86 |   6.09 |   13.11 |  11.58 |     17.46 |
+| 031_Inosine       |     31 |  13.14 |  15.87 |   16.57 |  25.89 |     26.14 |
+| 033_Bisphenol_A   |     33 |  12.31 |  16.88 |   16.54 |  28.45 |     28.82 |
+| 037_Mg_Porphin    |     37 |  13.85 |  19.03 |   20.53 |  28.31 |     30.27 |
+| 042_Penicillin_V  |     42 |  10.34 |  13.35 |   15.34 |  22.01 |     24.2  |
+| 045_Ochratoxin_A  |     45 |  13.34 |  15.3  |   19.66 |  27.08 |     25.41 |
+| 052_Cetirizine    |     52 |  17.79 |  17.44 |   19    |  24.41 |     25.87 |
+| 057_Tamoxifen     |     57 |  14.7  |  16.57 |   18.4  |  24.86 |     25.47 |
+| 066_Raffinose     |     66 |  13.77 |  14.2  |   20.47 |  22.94 |     25.35 |
+| 084_Sphingomyelin |     84 |  14.24 |  12.82 |   15.96 |  22.11 |     24.46 |
+| 095_Azadirachtin  |     95 |   5.58 |   7.72 |   24.18 |  26.84 |     25.21 |
+| 113_Taxol         |    113 |   5.44 |   6.81 |   24.58 |  29.14 |    nan    |
+
+Find more benchmarks in gpu4pyscf/benchmarks
+

gpu4pyscf/__init__.py

@@ -1,2 +1,8 @@
 from . import lib, grad, hessian, solvent, scf, dft
-__version__ = '0.6.5'
+__version__ = '0.6.1'
+
+import pyscf
+from gpu4pyscf.lib.utils import patch_cpu_kernel
+
+# patch tag_array for compatibility
+pyscf.lib.tag_array = patch_cpu_kernel(pyscf.lib.tag_array)(lib.cupy_helper.tag_array)

gpu4pyscf/df/grad/rks.py

@@ -115,7 +115,11 @@ def get_veff(ks_grad, mol=None, dm=None):
     vxc = tag_array(vxc, aux=e1_aux)
     return vxc
 
+<<<<<<< HEAD
+class Gradients(rks_grad.Gradients):
+=======
 class Gradients(rks_grad.Gradients, pyscf.df.grad.rks.Gradients):
+>>>>>>> dev
     from gpu4pyscf.lib.utils import to_cpu, to_gpu, device
 
     get_jk = get_jk

gpu4pyscf/dft/__init__.py

@@ -1,5 +1,9 @@
+<<<<<<< HEAD
+from . import rks
+=======
 from . import rks
 from .rks import RKS
 from .uks import UKS
 from .gks import GKS
 from .roks import ROKS
+>>>>>>> dev

gpu4pyscf/dft/gen_grid.py

@@ -378,11 +378,18 @@ def arg_group_grids(mol, coords, box_size=GROUP_BOX_SIZE):
                 atom_coords.max(axis=0) + GROUP_BOUNDARY_PENALTY]
     # how many boxes inside the boundary
     boxes = ((boundary[1] - boundary[0]) * (1./box_size)).round().astype(int)
-    tot_boxes = numpy.prod(boxes + 2)
+    boxes = cupy.asarray(boxes)
+    boundary = [cupy.asarray(boundary[0]), cupy.asarray(boundary[1])]
+    tot_boxes = cupy.prod(boxes + 2)
     logger.debug(mol, 'tot_boxes %d, boxes in each direction %s', tot_boxes, boxes)
     # box_size is the length of each edge of the box
+<<<<<<< HEAD
+    box_size = (boundary[1] - boundary[0]) / boxes
+    frac_coords = (coords - boundary[0]) * (1./box_size)
+=======
     box_size = cupy.asarray((boundary[1] - boundary[0]) / boxes)
     frac_coords = (coords - cupy.asarray(boundary[0])) * (1./box_size)
+>>>>>>> dev
     box_ids = cupy.floor(frac_coords).astype(int)
     box_ids[box_ids<-1] = -1
     box_ids[box_ids[:,0] > boxes[0], 0] = boxes[0]

gpu4pyscf/grad/rks.py

@@ -496,9 +496,12 @@ def get_du(ia, ib):  # JCP 98, 5612 (1993); (B10)
         yield coords, w0, w1
 
 class Gradients(rhf_grad.Gradients, pyscf.grad.rks.Gradients):
+<<<<<<< HEAD
+    device = 'gpu'
+    get_veff = patch_cpu_kernel(pyscf.grad.rks.Gradients.get_veff)(_get_veff)
+=======
     from gpu4pyscf.lib.utils import to_cpu, to_gpu, device
-
-    get_veff = _get_veff
+>>>>>>> dev
 
     def get_dispersion(self):
         if self.base.disp[:2].upper() == 'D3':

gpu4pyscf/grad/tests/test_rhf_grad.py

@@ -14,10 +14,20 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
 import pyscf
+<<<<<<< HEAD
+import cupy
+import numpy as np
+import unittest
+from pyscf import lib, scf
+import gpu4pyscf
+
+lib.num_threads(8)
+=======
 import numpy as np
 import unittest
 import gpu4pyscf
 from pyscf import scf
+>>>>>>> dev
 
 atom = '''
 O       0.0000000000    -0.0000000000     0.1174000000

gpu4pyscf/grad/tests/test_rks_grad.py

@@ -16,10 +16,19 @@
 import pyscf
 import cupy
 import unittest
+<<<<<<< HEAD:gpu4pyscf/grad/tests/test_rks_grad.py
+from pyscf import lib
 from pyscf.dft import rks
 import gpu4pyscf
 from gpu4pyscf.dft import numint
 
+lib.num_threads(8)
+=======
+from pyscf.dft import rks
+import gpu4pyscf
+from gpu4pyscf.dft import numint
+>>>>>>> dev:gpu4pyscf/grad/tests/test_grad.py
+
 atom = '''
 O       0.0000000000    -0.0000000000     0.1174000000
 H      -0.7570000000    -0.0000000000    -0.4696000000

gpu4pyscf/scf/hf.py

@@ -614,7 +614,11 @@ def nuc_grad_method(self):
     def density_fit(self, auxbasis=None, with_df=None, only_dfj=False):
         import gpu4pyscf.df.df_jk
         return gpu4pyscf.df.df_jk.density_fit(self, auxbasis, with_df, only_dfj)
+<<<<<<< HEAD
+
+=======
 
+>>>>>>> dev
 class _VHFOpt(_vhf.VHFOpt):
     from gpu4pyscf.lib.utils import to_cpu, to_gpu, device