描述

新增fusedSgd算子

关联的Issue

https://gitcode.com/cann/ops-nn/issues/3454

测试

已完成泛化测试

测试用例

  std::vector<float> paramsRefHostData1 = {1, 2, 3, 4, 5, 6, 7, 8};
  std::vector<float> gradsRefHostData1 = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8};
  std::vector<float> momentumHostData1 = {0, 0, 0, 0, 0, 0, 0, 0};
  std::vector<float> paramsRefHostData2 = {9, 10, 11, 12};
  std::vector<float> gradsRefHostData2 = {0.9, 1.0, 1.1, 1.2};
  std::vector<float> momentumHostData2 = {0, 0, 0, 0};
  std::vector<float> gradScaleOptionalHostData = {1.0};
  std::vector<int64_t> inputShape1 = {2, 2, 2};
  std::vector<int64_t> inputShape2 = {2, 2};
  std::vector<int64_t> scalarShape = {1};

执行结果

====== Tensor 2 gradsRef results ======
result[0] is: 0.900000
result[1] is: 1.000000
result[2] is: 1.100000
result[3] is: 1.200000
------ Momentum buffer 2 ------
result[0] is: 0.990000
result[1] is: 1.100000
result[2] is: 1.210000
result[3] is: 1.320000

文档更新

新增README.md、aclnnFusedSgd.md文件

类型标签

• Bug修复
• 新特性
• 性能优化
• 文档更新
• 其他，请描述：