chore: import upstream snapshot with attribution

2026-07-13 13:33:03 +08:00
commit 5b57521aa1
8226 changed files with 3425766 additions and 0 deletions
@@ -0,0 +1,55 @@
+//
+//  NPUPadding.cpp
+//  MNN
+//
+//  Created by MNN on 2019/09/07.
+//  Copyright © 2018, Alibaba Group Holding Limited
+//
+
+#include "NPUPadding.hpp"
+#include "NPUBackend.hpp"
+
+using namespace std;
+
+namespace MNN {
+
+NPUPadding::NPUPadding(Backend *b, const Op *op, const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) : MNN::NPUCommonExecution(b,op) {
+    auto opName = mOp->name()->str();
+    auto input1 = inputs[1];
+    bool isConst1 = TensorUtils::getDescribe(input1)->usage==Tensor::InsideDescribe::Usage::CONSTANT;
+    MNN_ASSERT(isConst1 == true);
+    auto data = input1->host<int>();
+    //MNN_PRINT("Padding input1->buffer().dim[0].extent=%d\n",input1->buffer().dim[0].extent);
+    if (input1->buffer().dim[0].extent == 3) {
+        mPadData = {0, 0, data[4], data[5], data[0], data[1], data[2], data[3]};
+    } else if ((input1->buffer().dim[0].extent == 4) || (input1->buffer().dim[0].extent == 8)) {
+        mPadData = {data[0], data[1], data[2], data[3], data[4], data[5], data[6], data[7]};
+    }
+    // om input weight const op
+    mConst = hiai::op::Const(opName + "_w_const");
+    {
+        ge::TensorPtr filter = std::make_shared<ge::Tensor>();
+        ge::TensorDesc fdesc(ge::Shape(ge::Shape({4, 2})), ge::FORMAT_NCHW, ge::DT_INT32);
+        filter->SetTensorDesc(fdesc);
+        filter->SetData((uint8_t *)mPadData.data(), mPadData.size() * sizeof(int32_t));
+        mConst.set_attr_value(filter);
+    }
+}
+
+ErrorCode NPUPadding::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
+    mNpuBackend->setNetworkInput(inputs, mOp);
+    
+    auto opName = mOp->name()->str();
+    auto xOp = mNpuBackend->getInputOps(mOp);
+    shared_ptr<hiai::op::Pad> padding(new hiai::op::Pad(opName));
+    auto inputIndex = mOp->inputIndexes()->data()[0];
+    auto iops       = mNpuBackend->mGrapMap[inputIndex]; // x
+    xOp        = iops.back().first;
+    (*padding).set_input_x(*xOp.get()).set_input_paddings(mConst);
+    mNpuBackend->setOutputOps(mOp, {padding}, outputs);
+    return NO_ERROR;
+}
+
+NPUCreatorRegister<TypedCreator<NPUPadding>> __padding_op(OpType_Padding);
+
+} // namespace MNN