Opencl datahandling

Since the target_dh_refactoring was never merged, we decided to do the refactoring at smaller steps and recover the original OpenCL datahandling PR.

Refactoring to merge target and backend can be done later.

added 1 commit

• a136fa22 - Simplify boundary handling with OpenCL

Compare with previous version

Can we merge this? It's the previous OpenCL datahandling PR rebased on current master. @kuron @bauer

Test pass apart from documentation with a inter-sphinx problem with sympy.

If the integration tests pass as well this looks fine to me

Could you please apply

--- a/pystencils_tests/test_staggered_kernel.py
+++ b/pystencils_tests/test_staggered_kernel.py
@@ -4,13 +4,13 @@ import sympy as sp
 
 
 class TestStaggeredDiffusion:
-    def _run(self, num_neighbors):
+    def _run(self, num_neighbors, target='cpu'):
         L = (40, 40)
         D = 0.066
         dt = 1
         T = 100
 
-        dh = ps.create_data_handling(L, periodicity=True, default_target='cpu')
+        dh = ps.create_data_handling(L, periodicity=True, default_target=target)
 
         c = dh.add_array('c', values_per_cell=1)
         j = dh.add_array('j', values_per_cell=num_neighbors, field_type=ps.FieldType.STAGGERED_FLUX)
@@ -67,6 +67,11 @@ class TestStaggeredDiffusion:
     def test_diffusion_4(self):
         self._run(4)
 
+    @pytest.mark.skipif(not HAS_OPENCL, reason="Test requires pyopencl")
+    def test_diffusion_opencl(self):
+        import pystencils.opencl.autoinit
+        self._run(4, 'opencl')
+
 
 def test_staggered_subexpressions():
     dh = ps.create_data_handling((10, 10), periodicity=True, default_target='cpu')

That was the test I used two months ago and it still works after your rebase.

added 20 commits

• 880a3b6e - Add InterpolatorAccess.__getnewargs__
• 8aa4fe9f - Allow functions for Field.coordinate_transform
• fc24639d - Merge remote-tracking branch 'fork/InterpolatorAccess.__getnewargs__'
• ca134c22 - Also update Field.physical_to_index,index_to_physical for coordinate_transform as a function
• b508fbc3 - Lint
• 97ed02cf - Improve OpenCL integration via create_kernel
• c1afdc38 - Extend SerialDataHandling to handle alternative GPU libraries
• 190aafcd - Allow creation of indexed kernels with opencl
• f6ba722b - Let SerialDataHandling.synchronization_function work with target OpenCL
• bf7e314e - Teach SympyPrinters how to print Type
• b56d69d0 - Make SerialDataHandling.serialization_function work with OpenCL
• 5f7b9de4 - Add two helper functions to manage global cl ctx
• a08762bd - Add pystencils.opencl.autoinit
• 2e74f600 - Protect pystencils.opencl.autoinit with try..except
• 9221a1a0 - Add pycuda.autoinit to GPU test
• 66c37ce5 - Allow default_target=='opencl' in SerialDataHandling
• 87323f2b - Add friendly hint when calling OpenCL kernels with PyCUDA/Numpy arrays
• f917588e - Add weird construction to handle OpenCL boundary handling (don't show
• 38501e48 - Conftest ignore opencl.autoinit
• 218e868c - Simplify boundary handling with OpenCL

Compare with previous version

added 15 commits

• 85d72db7 - Improve OpenCL integration via create_kernel
• 3b2487c8 - Extend SerialDataHandling to handle alternative GPU libraries
• 489b8c1a - Allow creation of indexed kernels with opencl
• 96f3383d - Let SerialDataHandling.synchronization_function work with target OpenCL
• df8d438a - Teach SympyPrinters how to print Type
• d20202ed - Make SerialDataHandling.serialization_function work with OpenCL
• d30d1b92 - Add two helper functions to manage global cl ctx
• 60b3f9db - Add pystencils.opencl.autoinit
• f76d4b78 - Protect pystencils.opencl.autoinit with try..except
• 718d505c - Add pycuda.autoinit to GPU test
• 11401ba3 - Allow default_target=='opencl' in SerialDataHandling
• 0871a1e2 - Add friendly hint when calling OpenCL kernels with PyCUDA/Numpy arrays
• 0eb5693f - Add weird construction to handle OpenCL boundary handling (don't show
• 7465b426 - Conftest ignore opencl.autoinit
• a136fa22 - Simplify boundary handling with OpenCL

Compare with previous version

added 1 commit

• 6e3c1582 - test_staggered_kernel also for OpenCL

Compare with previous version

Ok. I added the test. Also work on my machine.

enabled an automatic merge when the pipeline for 6e3c1582 succeeds

added 20 commits

• 6e3c1582...9b57cf87 - 4 commits from branch pycodegen:master
• de753874 - Improve OpenCL integration via create_kernel
• e6a30483 - Extend SerialDataHandling to handle alternative GPU libraries
• 061ab58b - Allow creation of indexed kernels with opencl
• e02acf7d - Let SerialDataHandling.synchronization_function work with target OpenCL
• 585084e7 - Teach SympyPrinters how to print Type
• 0e56c1e3 - Make SerialDataHandling.serialization_function work with OpenCL
• 0b0ba401 - Add two helper functions to manage global cl ctx
• 5437459f - Add pystencils.opencl.autoinit
• 2901a5b5 - Protect pystencils.opencl.autoinit with try..except
• 62803d18 - Add pycuda.autoinit to GPU test
• de710402 - Allow default_target=='opencl' in SerialDataHandling
• 58e62e71 - Add friendly hint when calling OpenCL kernels with PyCUDA/Numpy arrays
• bb21c9b1 - Add weird construction to handle OpenCL boundary handling (don't show
• d08987c1 - Conftest ignore opencl.autoinit
• c989d8ae - Simplify boundary handling with OpenCL
• 226abc3a - test_staggered_kernel also for OpenCL

Compare with previous version

enabled an automatic merge when the pipeline for 226abc3a succeeds

merged

mentioned in commit 171d844a