chore: import upstream snapshot with attribution

test/custom_kernel/CMakeLists.txt

@@ -0,0 +1,14 @@
+file(
+  GLOB TEST_OPS
+  RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}"
+  "test_*.py")
+string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}")
+
+set(CUSTOM_ENVS
+    PADDLE_SOURCE_DIR=${PADDLE_SOURCE_DIR}
+    PADDLE_BINARY_DIR=${PADDLE_BINARY_DIR}
+    CUSTOM_DEVICE_ROOT=${CMAKE_BINARY_DIR}/test)
+
+foreach(TEST_OP ${TEST_OPS})
+  py_test(${TEST_OP} SRCS ${TEST_OP}.py ENVS ${CUSTOM_ENVS})
+endforeach()

test/custom_kernel/custom_kernel_dot.cc

@@ -0,0 +1,51 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "paddle/phi/core/kernel_registry.h"
+
+namespace paddle {
+
+namespace custom_kernel {
+
+// Here we use dot <CPU, ANY, INT8> for test
+// This test will fail when this kernel is supported in framework
+template <typename T, typename Context>
+void DotKernel(const Context& dev_ctx,
+               const phi::DenseTensor& x,
+               const phi::DenseTensor& y,
+               phi::DenseTensor* out) {
+  auto const *x_ptr = x.data<T>(), *x_ptr_ = &x_ptr[0];
+  auto const *y_ptr = y.data<T>(), *y_ptr_ = &y_ptr[0];
+  T* z = dev_ctx.template Alloc<T>(out);
+
+  // Loop over the total N elements of both operands while sum-reducing every
+  // B pairs along the way where B is the dimension of the least ordered axis
+  auto&& d = x.dims();
+  auto const N = x.numel();
+  auto const B = d[d.size() - 1];
+
+  for (int j = 0; j < N / B; j++) {
+    T ss = 0;
+    for (int i = 0; i < B; i++) ss += (*x_ptr_++) * (*y_ptr_++);
+    z[j] = ss;
+  }
+}
+
+}  // namespace custom_kernel
+}  // namespace paddle
+
+PD_REGISTER_BUILTIN_KERNEL(
+    dot, CPU, ALL_LAYOUT, paddle::custom_kernel::DotKernel, int8_t) {
+  kernel->OutputAt(0).SetDataType(phi::DataType::INT8);
+}

test/custom_kernel/custom_kernel_dot_c.cc

@@ -0,0 +1,49 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "paddle/phi/core/custom_phi_kernel.h"
+
+namespace paddle {
+
+namespace custom_kernel {
+
+// Here we use dot <CPU, ANY, INT8> for test
+// This test will fail when this kernel is supported in framework
+template <typename T>
+void DotKernel(const phi::Context& dev_ctx,
+               const phi::DenseTensor& x,
+               const phi::DenseTensor& y,
+               phi::DenseTensor* out) {
+  auto const *x_ptr = x.data<T>(), *x_ptr_ = &x_ptr[0];
+  auto const *y_ptr = y.data<T>(), *y_ptr_ = &y_ptr[0];
+  T* z = dev_ctx.template Alloc<T>(out);
+
+  // Loop over the total N elements of both operands while sum-reducing every
+  // B pairs along the way where B is the dimension of the least ordered axis
+  auto&& d = x.dims();
+  auto const N = x.numel();
+  auto const B = d[d.size() - 1];
+
+  for (int j = 0; j < N / B; j++) {
+    T ss = 0;
+    for (int i = 0; i < B; i++) ss += (*x_ptr_++) * (*y_ptr_++);
+    z[j] = ss;
+  }
+}
+
+}  // namespace custom_kernel
+}  // namespace paddle
+
+PD_BUILD_PHI_KERNEL(
+    dot, CPU, ALL_LAYOUT, paddle::custom_kernel::DotKernel, int8_t) {}

test/custom_kernel/custom_kernel_dot_c_setup.py

@@ -0,0 +1,79 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from distutils.sysconfig import get_python_lib
+
+from setuptools import Extension, setup
+from setuptools.command.build_ext import build_ext
+
+
+# refer: https://note.qidong.name/2018/03/setup-warning-strict-prototypes
+# Avoid a gcc warning below:
+# cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid
+# for C/ObjC but not for C++
+class BuildExt(build_ext):
+    def build_extensions(self):
+        if '-Wstrict-prototypes' in self.compiler.compiler_so:
+            self.compiler.compiler_so.remove('-Wstrict-prototypes')
+        super().build_extensions()
+
+
+# cc flags
+paddle_extra_compile_args = [
+    '-std=c++14',
+    '-shared',
+    '-fPIC',
+    '-Wno-parentheses',
+    '-DPADDLE_WITH_CUSTOM_KERNEL',
+]
+
+# include path
+site_packages_path = get_python_lib()
+paddle_custom_kernel_include = [
+    os.path.join(site_packages_path, 'paddle', 'include'),
+]
+# include path third_party
+compile_third_party_path = os.path.join(
+    os.environ['PADDLE_BINARY_DIR'], 'third_party'
+)
+paddle_custom_kernel_include += [
+    os.path.join(compile_third_party_path, 'install/gflags/include'),  # gflags
+    os.path.join(compile_third_party_path, 'install/glog/include'),  # glog
+]
+
+# libs path
+paddle_custom_kernel_library_dir = [
+    os.path.join(site_packages_path, 'paddle', 'base'),
+]
+
+# libs
+libs = [':libpaddle.so']
+
+custom_kernel_dot_module = Extension(
+    'custom_kernel_dot',
+    sources=['custom_kernel_dot_c.cc'],
+    include_dirs=paddle_custom_kernel_include,
+    library_dirs=paddle_custom_kernel_library_dir,
+    libraries=libs,
+    extra_compile_args=paddle_extra_compile_args,
+)
+
+setup(
+    name='custom_kernel_dot_c',
+    version='1.0',
+    description='custom kernel fot compiling',
+    cmdclass={'build_ext': BuildExt},
+    ext_modules=[custom_kernel_dot_module],
+)

test/custom_kernel/custom_kernel_dot_setup.py

@@ -0,0 +1,80 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import site
+
+from setuptools import Extension, setup
+from setuptools.command.build_ext import build_ext
+
+
+# refer: https://note.qidong.name/2018/03/setup-warning-strict-prototypes
+# Avoid a gcc warning below:
+# cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid
+# for C/ObjC but not for C++
+class BuildExt(build_ext):
+    def build_extensions(self):
+        if '-Wstrict-prototypes' in self.compiler.compiler_so:
+            self.compiler.compiler_so.remove('-Wstrict-prototypes')
+        super().build_extensions()
+
+
+# cc flags
+paddle_extra_compile_args = [
+    '-std=c++14',
+    '-shared',
+    '-fPIC',
+    '-Wno-parentheses',
+    '-DPADDLE_WITH_CUSTOM_KERNEL',
+]
+
+# include path
+site_packages_path = site.getsitepackages()
+paddle_custom_kernel_include = [
+    os.path.join(path, 'paddle', 'include') for path in site_packages_path
+]
+
+# include path third_party
+compile_third_party_path = os.path.join(
+    os.environ['PADDLE_BINARY_DIR'], 'third_party'
+)
+paddle_custom_kernel_include += [
+    os.path.join(compile_third_party_path, 'install/gflags/include'),  # gflags
+    os.path.join(compile_third_party_path, 'install/glog/include'),  # glog
+]
+
+# libs path
+paddle_custom_kernel_library_dir = [
+    os.path.join(path, 'paddle', 'base') for path in site_packages_path
+]
+
+# libs
+libs = [':libpaddle.so']
+
+custom_kernel_dot_module = Extension(
+    'custom_kernel_dot',
+    sources=['custom_kernel_dot.cc'],
+    include_dirs=paddle_custom_kernel_include,
+    library_dirs=paddle_custom_kernel_library_dir,
+    libraries=libs,
+    extra_compile_args=paddle_extra_compile_args,
+)
+
+setup(
+    name='custom_kernel_dot',
+    version='1.0',
+    description='custom kernel fot compiling',
+    cmdclass={'build_ext': BuildExt},
+    ext_modules=[custom_kernel_dot_module],
+)

test/custom_kernel/test_custom_kernel_dot.py

@@ -0,0 +1,85 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import unittest
+
+import numpy as np
+
+
+# use dot <CPU, ANY, INT8> as test case.
+class TestCustomKernelDot(unittest.TestCase):
+    def setUp(self):
+        # compile so and set to current path
+        cur_dir = os.path.dirname(os.path.abspath(__file__))
+
+        # --inplace to place output so file to current dir
+        cmd = f'cd {cur_dir} && {sys.executable} custom_kernel_dot_setup.py build_ext --inplace'
+        os.system(cmd)
+
+    def test_custom_kernel_dot_run(self):
+        # test dot run
+        x_data = np.random.uniform(1, 5, [2, 10]).astype(np.int8)
+        y_data = np.random.uniform(1, 5, [2, 10]).astype(np.int8)
+        result = np.sum(x_data * y_data, axis=1).reshape([2, 1])
+
+        import paddle
+
+        paddle.set_device('cpu')
+        x = paddle.to_tensor(x_data)
+        y = paddle.to_tensor(y_data)
+        out = paddle.dot(x, y)
+
+        np.testing.assert_array_equal(
+            out.numpy(),
+            result,
+            err_msg=f'custom kernel dot out: {out.numpy()},\n numpy dot out: {result}',
+        )
+
+
+class TestCustomKernelDotC(unittest.TestCase):
+    def setUp(self):
+        # compile so and set to current path
+        cur_dir = os.path.dirname(os.path.abspath(__file__))
+
+        # --inplace to place output so file to current dir
+        cmd = f'cd {cur_dir} && {sys.executable} custom_kernel_dot_c_setup.py build_ext --inplace'
+        os.system(cmd)
+
+    def test_custom_kernel_dot_run(self):
+        # test dot run
+        x_data = np.random.uniform(1, 5, [2, 10]).astype(np.int8)
+        y_data = np.random.uniform(1, 5, [2, 10]).astype(np.int8)
+        result = np.sum(x_data * y_data, axis=1).reshape([2, 1])
+
+        import paddle
+
+        paddle.set_device('cpu')
+        x = paddle.to_tensor(x_data)
+        y = paddle.to_tensor(y_data)
+        out = paddle.dot(x, y)
+
+        np.testing.assert_array_equal(
+            out.numpy(),
+            result,
+            err_msg=f'custom kernel dot out: {out.numpy()},\n numpy dot out: {result}',
+        )
+
+
+if __name__ == '__main__':
+    if os.name == 'nt' or sys.platform.startswith('darwin'):
+        # only support Linux now
+        sys.exit()
+    unittest.main()

test/custom_kernel/test_custom_kernel_load.py

@@ -0,0 +1,80 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import site
+import sys
+import unittest
+
+import numpy as np
+
+
+class TestCustomKernelLoad(unittest.TestCase):
+    def setUp(self):
+        # compile so and set to current path
+        cur_dir = os.path.dirname(os.path.abspath(__file__))
+
+        # --inplace to place output so file to current dir
+        cmd = f'cd {cur_dir} && {sys.executable} custom_kernel_dot_setup.py build_ext --inplace'
+        os.system(cmd)
+
+        # get paddle lib path and place so
+        paddle_lib_path = ''
+        site_dirs = site.getsitepackages()
+        for site_dir in site_dirs:
+            lib_dir = os.path.sep.join([site_dir, 'paddle', 'libs'])
+            if os.path.exists(lib_dir):
+                paddle_lib_path = lib_dir
+                break
+        if paddle_lib_path == '':
+            if hasattr(site, 'USER_SITE'):
+                lib_dir = os.path.sep.join([site.USER_SITE, 'paddle', 'libs'])
+                if os.path.exists(lib_dir):
+                    paddle_lib_path = lib_dir
+        self.default_path = os.path.sep.join(
+            [paddle_lib_path, '..', '..', 'paddle_custom_device']
+        )
+        # copy so to default path
+        cmd = f'mkdir -p {self.default_path} && cp ./*.so {self.default_path}'
+        os.system(cmd)  # wait
+
+    def test_custom_kernel_dot_load(self):
+        # test dot load
+        x_data = np.random.uniform(1, 5, [2, 10]).astype(np.int8)
+        y_data = np.random.uniform(1, 5, [2, 10]).astype(np.int8)
+        result = np.sum(x_data * y_data, axis=1).reshape([2, 1])
+
+        import paddle
+
+        paddle.set_device('cpu')
+        x = paddle.to_tensor(x_data)
+        y = paddle.to_tensor(y_data)
+        out = paddle.dot(x, y)
+
+        np.testing.assert_array_equal(
+            out.numpy(),
+            result,
+            err_msg=f'custom kernel dot out: {out.numpy()},\n numpy dot out: {result}',
+        )
+
+    def tearDown(self):
+        cmd = f'rm -rf {self.default_path}'
+        os.system(cmd)
+
+
+if __name__ == '__main__':
+    if os.name == 'nt' or sys.platform.startswith('darwin'):
+        # only support Linux now
+        sys.exit()
+    unittest.main()