fix conflicts

docs/developer-guide/operators.md

@@ -6,6 +6,7 @@
 * [BinaryOp](#binaryop)
 * [BNLL](#bnll)
 * [Cast](#cast)
+* [CELU](#celu)
 * [Clip](#clip)
 * [Concat](#concat)
 * [Convolution](#convolution)
@@ -197,6 +198,19 @@ Element type:
 - 3 = int8
 - 4 = bfloat16
 
+# CELU
+```
+if x < 0    y = (exp(x / alpha) - 1.f) * alpha
+else        y = x
+```
+
+* one_blob_only
+* support_inplace
+
+| param id  | name          | type  | default   | description       |
+| --------- | ------------- | ----- | --------- | ----------------- |
+| 0         | alpha         | float | 1.f       |                   |
+
 # Clip
 ```
 y = clamp(x, min, max)

src/CMakeLists.txt

@@ -164,6 +164,7 @@ ncnn_add_layer(CumulativeSum)
 ncnn_add_layer(CopyTo)
 ncnn_add_layer(Erf)
 ncnn_add_layer(Diag)
+ncnn_add_layer(CELU)
 ncnn_add_layer(Size)
 
 if(NCNN_VULKAN)

src/layer/celu.cpp

@@ -0,0 +1,57 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2023 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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 "celu.h"
+
+#include <math.h>
+
+namespace ncnn {
+
+CELU::CELU()
+{
+    one_blob_only = true;
+    support_inplace = true;
+}
+
+int CELU::load_param(const ParamDict& pd)
+{
+    alpha = pd.get(0, 1.f);
+
+    return 0;
+}
+
+int CELU::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
+{
+    int w = bottom_top_blob.w;
+    int h = bottom_top_blob.h;
+    int d = bottom_top_blob.d;
+    int channels = bottom_top_blob.c;
+    int size = w * h * d;
+
+    #pragma omp parallel for num_threads(opt.num_threads)
+    for (int q = 0; q < channels; q++)
+    {
+        float* ptr = bottom_top_blob.channel(q);
+
+        for (int i = 0; i < size; i++)
+        {
+            if (ptr[i] < 0.f)
+                ptr[i] = (expf(ptr[i] / alpha) - 1.f) * alpha;
+        }
+    }
+
+    return 0;
+}
+
+} // namespace ncnn

src/layer/celu.h

@@ -0,0 +1,37 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2023 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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.
+
+#ifndef LAYER_CELU_H
+#define LAYER_CELU_H
+
+#include "layer.h"
+
+namespace ncnn {
+
+class CELU : public Layer
+{
+public:
+    CELU();
+
+    virtual int load_param(const ParamDict& pd);
+
+    virtual int forward_inplace(Mat& bottom_top_blob, const Option& opt) const;
+
+public:
+    float alpha;
+};
+
+} // namespace ncnn
+
+#endif // LAYER_CELU_H

src/layer/vulkan/celu_vulkan.cpp

@@ -0,0 +1,182 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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 "celu_vulkan.h"
+
+#include "layer_shader_type.h"
+
+namespace ncnn {
+
+CELU_vulkan::CELU_vulkan()
+{
+    support_vulkan = true;
+    support_image_storage = true;
+
+    pipeline_celu = 0;
+    pipeline_celu_pack4 = 0;
+    pipeline_celu_pack8 = 0;
+}
+
+int CELU_vulkan::create_pipeline(const Option& opt)
+{
+    const Mat& shape = top_shapes.empty() ? Mat() : top_shapes[0];
+
+    int elempack = 1;
+    if (shape.dims == 1) elempack = opt.use_shader_pack8 && shape.w % 8 == 0 ? 8 : shape.w % 4 == 0 ? 4 : 1;
+    if (shape.dims == 2) elempack = opt.use_shader_pack8 && shape.h % 8 == 0 ? 8 : shape.h % 4 == 0 ? 4 : 1;
+    if (shape.dims == 3 || shape.dims == 4) elempack = opt.use_shader_pack8 && shape.c % 8 == 0 ? 8 : shape.c % 4 == 0 ? 4 : 1;
+
+    size_t elemsize;
+    if (opt.use_fp16_storage)
+    {
+        elemsize = elempack * 2u;
+    }
+    else if (opt.use_fp16_packed)
+    {
+        elemsize = elempack == 1 ? 4u : elempack * 2u;
+    }
+    else
+    {
+        elemsize = elempack * 4u;
+    }
+
+    Mat shape_packed;
+    if (shape.dims == 1) shape_packed = Mat(shape.w / elempack, (void*)0, elemsize, elempack);
+    if (shape.dims == 2) shape_packed = Mat(shape.w, shape.h / elempack, (void*)0, elemsize, elempack);
+    if (shape.dims == 3) shape_packed = Mat(shape.w, shape.h, shape.c / elempack, (void*)0, elemsize, elempack);
+    if (shape.dims == 4) shape_packed = Mat(shape.w, shape.h, shape.d, shape.c / elempack, (void*)0, elemsize, elempack);
+
+    std::vector<vk_specialization_type> specializations(1 + 5);
+    specializations[0].f = alpha;
+    specializations[1 + 0].i = shape_packed.dims;
+    specializations[1 + 1].i = shape_packed.w;
+    specializations[1 + 2].i = shape_packed.h * shape_packed.d;
+    specializations[1 + 3].i = shape_packed.c;
+    specializations[1 + 4].i = shape_packed.cstep;
+
+    Mat local_size_xyz;
+    if (shape_packed.dims == 1)
+    {
+        local_size_xyz.w = std::min(64, shape_packed.w);
+        local_size_xyz.h = 1;
+        local_size_xyz.c = 1;
+    }
+    if (shape_packed.dims == 2)
+    {
+        local_size_xyz.w = std::min(8, shape_packed.w);
+        local_size_xyz.h = std::min(8, shape_packed.h);
+        local_size_xyz.c = 1;
+    }
+    if (shape_packed.dims == 3)
+    {
+        local_size_xyz.w = std::min(4, shape_packed.w);
+        local_size_xyz.h = std::min(4, shape_packed.h);
+        local_size_xyz.c = std::min(4, shape_packed.c);
+    }
+    if (shape_packed.dims == 4)
+    {
+        local_size_xyz.w = std::min(4, shape_packed.w);
+        local_size_xyz.h = std::min(4, shape_packed.h * shape_packed.d);
+        local_size_xyz.c = std::min(4, shape_packed.c);
+    }
+
+    // pack1
+    if (shape.dims == 0 || elempack == 1)
+    {
+        pipeline_celu = new Pipeline(vkdev);
+        pipeline_celu->set_optimal_local_size_xyz(local_size_xyz);
+        pipeline_celu->create(LayerShaderType::celu, opt, specializations);
+    }
+
+    // pack4
+    if (shape.dims == 0 || elempack == 4)
+    {
+        pipeline_celu_pack4 = new Pipeline(vkdev);
+        pipeline_celu_pack4->set_optimal_local_size_xyz(local_size_xyz);
+        pipeline_celu_pack4->create(LayerShaderType::celu_pack4, opt, specializations);
+    }
+
+    // pack8
+    if ((opt.use_shader_pack8 && shape.dims == 0) || elempack == 8)
+    {
+        pipeline_celu_pack8 = new Pipeline(vkdev);
+        pipeline_celu_pack8->set_optimal_local_size_xyz(local_size_xyz);
+        pipeline_celu_pack8->create(LayerShaderType::celu_pack8, opt, specializations);
+    }
+
+    return 0;
+}
+
+int CELU_vulkan::destroy_pipeline(const Option& /*opt*/)
+{
+    delete pipeline_celu;
+    pipeline_celu = 0;
+
+    delete pipeline_celu_pack4;
+    pipeline_celu_pack4 = 0;
+
+    delete pipeline_celu_pack8;
+    pipeline_celu_pack8 = 0;
+
+    return 0;
+}
+
+int CELU_vulkan::forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
+{
+    int elempack = bottom_top_blob.elempack;
+
+    std::vector<VkMat> bindings(1);
+    bindings[0] = bottom_top_blob;
+
+    std::vector<vk_constant_type> constants(5);
+    constants[0].i = bottom_top_blob.dims;
+    constants[1].i = bottom_top_blob.w;
+    constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
+    constants[3].i = bottom_top_blob.c;
+    constants[4].i = bottom_top_blob.cstep;
+
+    const Pipeline* pipeline = elempack == 8 ? pipeline_celu_pack8
+                               : elempack == 4 ? pipeline_celu_pack4
+                               : pipeline_celu;
+
+    cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);
+
+    return 0;
+}
+
+int CELU_vulkan::forward_inplace(VkImageMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
+{
+    int elempack = bottom_top_blob.elempack;
+
+    std::vector<VkImageMat> bindings(2);
+    bindings[0] = bottom_top_blob;
+    bindings[1] = bottom_top_blob;
+
+    std::vector<vk_constant_type> constants(5);
+    constants[0].i = bottom_top_blob.dims;
+    constants[1].i = bottom_top_blob.w;
+    constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
+    constants[3].i = bottom_top_blob.c;
+    constants[4].i = 0; //bottom_top_blob.cstep;
+
+    const Pipeline* pipeline = elempack == 8 ? pipeline_celu_pack8
+                               : elempack == 4 ? pipeline_celu_pack4
+                               : pipeline_celu;
+
+    cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);
+
+    return 0;
+}
+
+} // namespace ncnn

src/layer/vulkan/celu_vulkan.h

@@ -0,0 +1,42 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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.
+
+#ifndef LAYER_CELU_VULKAN_H
+#define LAYER_CELU_VULKAN_H
+
+#include "celu.h"
+
+namespace ncnn {
+
+class CELU_vulkan : virtual public CELU
+{
+public:
+    CELU_vulkan();
+
+    virtual int create_pipeline(const Option& opt);
+    virtual int destroy_pipeline(const Option& opt);
+
+    using CELU::forward_inplace;
+    virtual int forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& opt) const;
+    virtual int forward_inplace(VkImageMat& bottom_top_blob, VkCompute& cmd, const Option& opt) const;
+
+public:
+    Pipeline* pipeline_celu;
+    Pipeline* pipeline_celu_pack4;
+    Pipeline* pipeline_celu_pack8;
+};
+
+} // namespace ncnn
+
+#endif // LAYER_CELU_VULKAN_H