[XPU] add increment op. (#51487)

* [XPU] add increment op.

* fix ci

cmake/external/xpu.cmake

@@ -8,8 +8,8 @@ set(XPU_API_LIB_NAME "libxpuapi.so")
 set(XPU_RT_LIB_NAME "libxpurt.so")
 set(XPU_XFT_LIB_NAME "libxft.so")
 
-set(XPU_BASE_DATE "20230308")
-set(XPU_XCCL_BASE_VERSION "1.0.10")
+set(XPU_BASE_DATE "20230310")
+set(XPU_XCCL_BASE_VERSION "1.0.12")
 set(XPU_XFT_BASE_VERSION "latest")
 
 if(NOT DEFINED XPU_BASE_URL)

paddle/phi/backends/xpu/xpu2_op_list.cc

@@ -385,7 +385,10 @@ XPUOpMap& get_kl2_ops() {
       {"huber_loss", XPUKernelSet({phi::DataType::FLOAT32})},
       {"kldiv_loss", XPUKernelSet({phi::DataType::FLOAT32})},
       {"kldiv_loss_grad", XPUKernelSet({phi::DataType::FLOAT32})},
-      {"iou_similarity", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"increment",
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::INT32,
+                     phi::DataType::INT64})},
       {"index_sample",
        XPUKernelSet({phi::DataType::INT8,
                      phi::DataType::INT16,
@@ -400,6 +403,7 @@ XPUOpMap& get_kl2_ops() {
                      phi::DataType::INT64})},
       {"instance_norm", XPUKernelSet({phi::DataType::FLOAT32})},
       {"instance_norm_grad", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"iou_similarity", XPUKernelSet({phi::DataType::FLOAT32})},
       {"label_smooth", XPUKernelSet({phi::DataType::FLOAT32})},
       {"lamb", XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
       {"lars_momentum",

paddle/phi/kernels/xpu/increment_kernel.cc

+// Copyright (c) 2023 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/kernels/increment_kernel.h"
+
+#include "paddle/phi/backends/xpu/enforce_xpu.h"
+#include "paddle/phi/common/memory_utils.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+namespace phi {
+
+template <typename T, typename Context>
+void IncrementKernel(const Context& ctx,
+                     const DenseTensor& x,
+                     float value,
+                     DenseTensor* out) {
+  // check input
+  PADDLE_ENFORCE_EQ(x.numel(),
+                    1,
+                    phi::errors::InvalidArgument(
+                        "input tensor x's numel should be EXACTLY 1."));
+
+  const T* x_data = x.data<T>();
+  T* out_data = ctx.template Alloc<T>(out);
+
+  // allocation for "value" on xpu
+  T value_as_t = static_cast<T>(value);
+  xpu::ctx_guard RAII_GUARD(ctx.x_context());
+  T* value_xpu = RAII_GUARD.alloc_l3_or_gm<T>(1);
+  memory_utils::Copy(ctx.GetPlace(),
+                     value_xpu,
+                     phi::CPUPlace(),
+                     reinterpret_cast<void*>(&value_as_t),
+                     sizeof(T));
+
+  // int add(Context* ctx, const T* x, const T* y, T* z, int64_t len);
+  int ret = xpu::add(ctx.x_context(), x_data, value_xpu, out_data, 1);
+  PADDLE_ENFORCE_XDNN_SUCCESS(ret, "add");
+}
+
+}  // namespace phi
+
+PD_REGISTER_KERNEL(
+    increment, XPU, ALL_LAYOUT, phi::IncrementKernel, float, int, int64_t) {}

python/paddle/fluid/tests/unittests/xpu/test_increment_op_xpu.py

+#   Copyright (c) 2023 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 sys
+import unittest
+
+import numpy as np
+
+sys.path.append("..")
+
+from op_test_xpu import XPUOpTest
+from xpu.get_test_cover_info import (
+    XPUOpTestWrapper,
+    create_test_class,
+    get_xpu_op_support_types,
+)
+
+import paddle
+
+paddle.enable_static()
+
+
+class XPUTestIncrementOP(XPUOpTestWrapper):
+    def __init__(self):
+        self.op_name = 'increment'
+        self.use_dynamic_create_class = False
+
+    class TestXPUIncrementOp(XPUOpTest):
+        def setUp(self):
+            self.place = paddle.XPUPlace(0)
+            self.init_dtype()
+            self.op_type = 'increment'
+
+            self.initTestCase()
+
+            x = np.random.uniform(-100, 100, [1]).astype(self.dtype)
+            output = x + np.cast[self.dtype](self.step)
+            output = output.astype(self.dtype)
+
+            self.inputs = {'X': x}
+            self.attrs = {'step': self.step}
+            self.outputs = {'Out': output}
+
+        def initTestCase(self):
+            self.step = -1.5
+
+        def init_dtype(self):
+            self.dtype = self.in_type
+
+        def test_check_output(self):
+            self.check_output_with_place(self.place)
+
+    class TestIncrement1(TestXPUIncrementOp):
+        def initTestCase(self):
+            self.step = 6.0
+
+    class TestIncrement2(TestXPUIncrementOp):
+        def initTestCase(self):
+            self.step = 2.1
+
+    class TestIncrement3(TestXPUIncrementOp):
+        def initTestCase(self):
+            self.step = -1.5
+
+    class TestIncrement4(TestXPUIncrementOp):
+        def initTestCase(self):
+            self.step = 0.5
+
+    class TestIncrement5(TestXPUIncrementOp):
+        def initTestCase(self):
+            self.step = 3
+
+
+support_types = get_xpu_op_support_types('increment')
+for stype in support_types:
+    create_test_class(globals(), XPUTestIncrementOP, stype)
+
+if __name__ == '__main__':
+    unittest.main()

python/paddle/fluid/tests/unittests/xpu/test_set_value_op_xpu.py

@@ -943,6 +943,8 @@ class XPUTestSetValueOp(XPUOpTestWrapper):
             with paddle.static.program_guard(main_program, startup_program):
                 x = paddle.static.data(name="x", shape=[4, 4], dtype='float32')
                 y = paddle.static.data(name="y", shape=[4, 4], dtype='float32')
+                x.stop_gradient = False
+                y.stop_gradient = False
 
                 label = paddle.static.data(
                     name="label", shape=[4, 1], dtype='int64'

python/paddle/fluid/tests/unittests/xpu/test_where_op_xpu.py

@@ -116,6 +116,7 @@ class TestXPUWhereAPI(unittest.TestCase):
                     y.stop_gradient = y_stop_gradient
 
                     result = paddle.where(cond, x, y)
+                    result.stop_gradient = False
                     append_backward(paddle.mean(result))
 
                     exe = fluid.Executor(self.place)