ENH: add() to API standard

* bring the `add()` ufunc up to the API standard and
turn the relevant test on in the CI

* other changes are similar to other recent binary ufunc
PRs; note one drawback related to copies in broadcasting

* the changes are similar to those in kokkosgh-156, though more
workunit type implementation were needed for this operation
to pass the standard test

Author: tylerjereddy

.github/workflows/array_api.yml

@@ -30,7 +30,7 @@ jobs:
           cd /tmp
           git clone https://github.com/kokkos/pykokkos-base.git
           cd pykokkos-base
-          python setup.py install -- -DENABLE_LAYOUTS=ON -DENABLE_MEMORY_TRAITS=OFF
+          python setup.py install -- -DENABLE_LAYOUTS=ON -DENABLE_MEMORY_TRAITS=OFF -DENABLE_VIEW_RANKS=5
       - name: Install pykokkos
         run: |
           python -m pip install .
@@ -49,4 +49,4 @@ jobs:
           # for hypothesis-driven test case generation
           pytest $GITHUB_WORKSPACE/pre_compile_tools/pre_compile_ufuncs.py -s
           # only run a subset of the conformance tests to get started
-          pytest array_api_tests/meta/test_broadcasting.py array_api_tests/meta/test_equality_mapping.py array_api_tests/meta/test_signatures.py array_api_tests/meta/test_special_cases.py array_api_tests/test_constants.py array_api_tests/meta/test_utils.py array_api_tests/test_creation_functions.py::test_ones array_api_tests/test_creation_functions.py::test_ones_like array_api_tests/test_data_type_functions.py::test_result_type array_api_tests/test_operators_and_elementwise_functions.py::test_log10 array_api_tests/test_operators_and_elementwise_functions.py::test_sqrt array_api_tests/test_operators_and_elementwise_functions.py::test_isfinite array_api_tests/test_operators_and_elementwise_functions.py::test_log2 array_api_tests/test_operators_and_elementwise_functions.py::test_log1p array_api_tests/test_operators_and_elementwise_functions.py::test_isinf array_api_tests/test_operators_and_elementwise_functions.py::test_log array_api_tests/test_array_object.py::test_scalar_casting array_api_tests/test_operators_and_elementwise_functions.py::test_sign array_api_tests/test_operators_and_elementwise_functions.py::test_square array_api_tests/test_operators_and_elementwise_functions.py::test_cos array_api_tests/test_operators_and_elementwise_functions.py::test_round array_api_tests/test_operators_and_elementwise_functions.py::test_trunc array_api_tests/test_operators_and_elementwise_functions.py::test_ceil array_api_tests/test_operators_and_elementwise_functions.py::test_floor
+          pytest array_api_tests/meta/test_broadcasting.py array_api_tests/meta/test_equality_mapping.py array_api_tests/meta/test_signatures.py array_api_tests/meta/test_special_cases.py array_api_tests/test_constants.py array_api_tests/meta/test_utils.py array_api_tests/test_creation_functions.py::test_ones array_api_tests/test_creation_functions.py::test_ones_like array_api_tests/test_data_type_functions.py::test_result_type array_api_tests/test_operators_and_elementwise_functions.py::test_log10 array_api_tests/test_operators_and_elementwise_functions.py::test_sqrt array_api_tests/test_operators_and_elementwise_functions.py::test_isfinite array_api_tests/test_operators_and_elementwise_functions.py::test_log2 array_api_tests/test_operators_and_elementwise_functions.py::test_log1p array_api_tests/test_operators_and_elementwise_functions.py::test_isinf array_api_tests/test_operators_and_elementwise_functions.py::test_log array_api_tests/test_array_object.py::test_scalar_casting array_api_tests/test_operators_and_elementwise_functions.py::test_sign array_api_tests/test_operators_and_elementwise_functions.py::test_square array_api_tests/test_operators_and_elementwise_functions.py::test_cos array_api_tests/test_operators_and_elementwise_functions.py::test_round array_api_tests/test_operators_and_elementwise_functions.py::test_trunc array_api_tests/test_operators_and_elementwise_functions.py::test_ceil array_api_tests/test_operators_and_elementwise_functions.py::test_floor "array_api_tests/test_operators_and_elementwise_functions.py::test_add[add(x1, x2)]"

.github/workflows/main_ci.yml

@@ -30,7 +30,7 @@ jobs:
           cd /tmp
           git clone https://github.com/kokkos/pykokkos-base.git
           cd pykokkos-base
-          python setup.py install -- -DENABLE_LAYOUTS=ON -DENABLE_MEMORY_TRAITS=OFF
+          python setup.py install -- -DENABLE_LAYOUTS=ON -DENABLE_MEMORY_TRAITS=OFF -DENABLE_VIEW_RANKS=5
       - name: Install pykokkos
         run: |
           python -m pip install .

pykokkos/lib/ufunc_workunits.py

@@ -1,11 +1,363 @@
 import pykokkos as pk
 
 
+@pk.workunit
+def add_impl_1d_int64(tid: int, view1: pk.View1D[pk.int64], view2: pk.View1D[pk.int64], out: pk.View1D[pk.int64]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_int64(tid: int, view1: pk.View2D[pk.int64], view2: pk.View2D[pk.int64], out: pk.View2D[pk.int64]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_int64(tid: int, view1: pk.View3D[pk.int64], view2: pk.View3D[pk.int64], out: pk.View3D[pk.int64]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_int64(tid: int, view1: pk.View4D[pk.int64], view2: pk.View4D[pk.int64], out: pk.View4D[pk.int64]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_int64(tid: int, view1: pk.View5D[pk.int64], view2: pk.View5D[pk.int64], out: pk.View5D[pk.int64]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_int32(tid: int, view1: pk.View1D[pk.int32], view2: pk.View1D[pk.int32], out: pk.View1D[pk.int32]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_int32(tid: int, view1: pk.View2D[pk.int32], view2: pk.View2D[pk.int32], out: pk.View2D[pk.int32]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_int32(tid: int, view1: pk.View3D[pk.int32], view2: pk.View3D[pk.int32], out: pk.View3D[pk.int32]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_int32(tid: int, view1: pk.View4D[pk.int32], view2: pk.View4D[pk.int32], out: pk.View4D[pk.int32]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_int32(tid: int, view1: pk.View5D[pk.int32], view2: pk.View5D[pk.int32], out: pk.View5D[pk.int32]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_int16(tid: int, view1: pk.View1D[pk.int16], view2: pk.View1D[pk.int16], out: pk.View1D[pk.int16]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_int16(tid: int, view1: pk.View2D[pk.int16], view2: pk.View2D[pk.int16], out: pk.View2D[pk.int16]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_int16(tid: int, view1: pk.View3D[pk.int16], view2: pk.View3D[pk.int16], out: pk.View3D[pk.int16]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_int16(tid: int, view1: pk.View4D[pk.int16], view2: pk.View4D[pk.int16], out: pk.View4D[pk.int16]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_int16(tid: int, view1: pk.View5D[pk.int16], view2: pk.View5D[pk.int16], out: pk.View5D[pk.int16]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_int8(tid: int, view1: pk.View1D[pk.int8], view2: pk.View1D[pk.int8], out: pk.View1D[pk.int8]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_int8(tid: int, view1: pk.View2D[pk.int8], view2: pk.View2D[pk.int8], out: pk.View2D[pk.int8]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_int8(tid: int, view1: pk.View3D[pk.int8], view2: pk.View3D[pk.int8], out: pk.View3D[pk.int8]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_int8(tid: int, view1: pk.View4D[pk.int8], view2: pk.View4D[pk.int8], out: pk.View4D[pk.int8]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_int8(tid: int, view1: pk.View5D[pk.int8], view2: pk.View5D[pk.int8], out: pk.View5D[pk.int8]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_uint64(tid: int, view1: pk.View1D[pk.uint64], view2: pk.View1D[pk.uint64], out: pk.View1D[pk.uint64]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_uint64(tid: int, view1: pk.View2D[pk.uint64], view2: pk.View2D[pk.uint64], out: pk.View2D[pk.uint64]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_uint64(tid: int, view1: pk.View3D[pk.uint64], view2: pk.View3D[pk.uint64], out: pk.View3D[pk.uint64]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_uint64(tid: int, view1: pk.View4D[pk.uint64], view2: pk.View4D[pk.uint64], out: pk.View4D[pk.uint64]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_uint64(tid: int, view1: pk.View5D[pk.uint64], view2: pk.View5D[pk.uint64], out: pk.View5D[pk.uint64]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_uint32(tid: int, view1: pk.View1D[pk.uint32], view2: pk.View1D[pk.uint32], out: pk.View1D[pk.uint32]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_uint32(tid: int, view1: pk.View2D[pk.uint32], view2: pk.View2D[pk.uint32], out: pk.View2D[pk.uint32]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_uint32(tid: int, view1: pk.View3D[pk.uint32], view2: pk.View3D[pk.uint32], out: pk.View3D[pk.uint32]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_uint32(tid: int, view1: pk.View4D[pk.uint32], view2: pk.View4D[pk.uint32], out: pk.View4D[pk.uint32]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_uint32(tid: int, view1: pk.View5D[pk.uint32], view2: pk.View5D[pk.uint32], out: pk.View5D[pk.uint32]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_uint16(tid: int, view1: pk.View1D[pk.uint16], view2: pk.View1D[pk.uint16], out: pk.View1D[pk.uint16]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_uint16(tid: int, view1: pk.View2D[pk.uint16], view2: pk.View2D[pk.uint16], out: pk.View2D[pk.uint16]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_uint16(tid: int, view1: pk.View3D[pk.uint16], view2: pk.View3D[pk.uint16], out: pk.View3D[pk.uint16]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_uint16(tid: int, view1: pk.View4D[pk.uint16], view2: pk.View4D[pk.uint16], out: pk.View4D[pk.uint16]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_uint16(tid: int, view1: pk.View5D[pk.uint16], view2: pk.View5D[pk.uint16], out: pk.View5D[pk.uint16]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_uint8(tid: int, view1: pk.View1D[pk.uint8], view2: pk.View1D[pk.uint8], out: pk.View1D[pk.uint8]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_uint8(tid: int, view1: pk.View2D[pk.uint8], view2: pk.View2D[pk.uint8], out: pk.View2D[pk.uint8]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_uint8(tid: int, view1: pk.View3D[pk.uint8], view2: pk.View3D[pk.uint8], out: pk.View3D[pk.uint8]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_uint8(tid: int, view1: pk.View4D[pk.uint8], view2: pk.View4D[pk.uint8], out: pk.View4D[pk.uint8]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_uint8(tid: int, view1: pk.View5D[pk.uint8], view2: pk.View5D[pk.uint8], out: pk.View5D[pk.uint8]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_float(tid: int, view1: pk.View1D[pk.float], view2: pk.View1D[pk.float], out: pk.View1D[pk.float]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_float(tid: int, view1: pk.View2D[pk.float], view2: pk.View2D[pk.float], out: pk.View2D[pk.float]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_float(tid: int, view1: pk.View3D[pk.float], view2: pk.View3D[pk.float], out: pk.View3D[pk.float]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_float(tid: int, view1: pk.View4D[pk.float], view2: pk.View4D[pk.float], out: pk.View4D[pk.float]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_float(tid: int, view1: pk.View5D[pk.float], view2: pk.View5D[pk.float], out: pk.View5D[pk.float]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
+@pk.workunit
+def add_impl_1d_double(tid: int, view1: pk.View1D[pk.double], view2: pk.View1D[pk.double], out: pk.View1D[pk.double]):
+    out[tid] = view1[tid] + view2[tid]
+
+
+@pk.workunit
+def add_impl_2d_double(tid: int, view1: pk.View2D[pk.double], view2: pk.View2D[pk.double], out: pk.View2D[pk.double]):
+    for i in range(view1.extent(1)):
+        out[tid][i] = view1[tid][i] + view2[tid][i]
+
+
+@pk.workunit
+def add_impl_3d_double(tid: int, view1: pk.View3D[pk.double], view2: pk.View3D[pk.double], out: pk.View3D[pk.double]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            out[tid][i][j] = view1[tid][i][j] + view2[tid][i][j]
+
+
+@pk.workunit
+def add_impl_4d_double(tid: int, view1: pk.View4D[pk.double], view2: pk.View4D[pk.double], out: pk.View4D[pk.double]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                out[tid][i][j][k] = view1[tid][i][j][k] + view2[tid][i][j][k]
+
+
+@pk.workunit
+def add_impl_5d_double(tid: int, view1: pk.View5D[pk.double], view2: pk.View5D[pk.double], out: pk.View5D[pk.double]):
+    for i in range(view1.extent(1)):
+        for j in range(view1.extent(2)):
+            for k in range(view1.extent(3)):
+                for l in range(view1.extent(4)):
+                    out[tid][i][j][k][l] = view1[tid][i][j][k][l] + view2[tid][i][j][k][l]
+
+
 @pk.workunit
 def floor_impl_1d_double(tid: int, view: pk.View1D[pk.double], out: pk.View1D[pk.double]):
     out[tid] = floor(view[tid])
 
 
+
+
 @pk.workunit
 def floor_impl_2d_double(tid: int, view: pk.View2D[pk.double], out: pk.View2D[pk.double]):
     for i in range(view.extent(1)):