You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

opr_impl.h 2.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354
  1. /**
  2. * \file dnn/src/armv7/matrix_mul/opr_impl.h
  3. * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
  4. *
  5. * Copyright (c) 2014-2020 Megvii Inc. All rights reserved.
  6. *
  7. * Unless required by applicable law or agreed to in writing,
  8. * software distributed under the License is distributed on an
  9. * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
  10. * implied.
  11. */
  12. #pragma once
  13. #include "src/arm_common/matrix_mul/opr_impl.h"
  14. namespace megdnn {
  15. namespace armv7 {
  16. class MatrixMulImpl : public arm_common::MatrixMulImpl {
  17. public:
  18. using arm_common::MatrixMulImpl::MatrixMulImpl;
  19. SmallVector<AlgoBase*> algo_pack() override;
  20. private:
  21. class AlgoF32; // Armv7 F32
  22. class AlgoF32MK4Pack4x12; // Armv7 F32 Kernel 4x12 with pack
  23. class AlgoF32MK4_4x8; // Armv7 F32 Kernel 4x8 nopack
  24. class AlgoF32Gemv; // Armv7 F32 Gemv
  25. class AlgoInt8x8x32K4x8x8; // Armv7 Int8x8x32 Kernel 4x8x8
  26. class AlgoInt8x8x32K4x2x16; // Armv7 Int8x8x32 Kernel 4x2x16
  27. class AlgoInt8x8x32MK4_4x2x16; // Armv7 Int8x8x32 Kernel MK4 4x2x16
  28. class AlgoQuint8K4x8x8; // Armv7 Quint8 Kernel 4x8x8
  29. class AlgoInt8x8x16K4x2x16; // Armv7 Int8x8x16 Kernel 4x2x16
  30. class AlgoInt8x8x16K4x8x8; // Armv7 Int8x8x16 Kernel 4x8x8
  31. class AlgoInt8x8x16MK4_8x8x4; // Armv7 Int8x8x16 Kernel 8x8x8
  32. class AlgoInt16x16x32K12x4x1; // Armv7 Int16x16x32 Kernel 12x4x1
  33. class AlgoInt16x16x32MK8_4x8; // Armv7 Int16x16x32 MK8 Format block 4x8
  34. #if __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
  35. class AlgoF16K4x16x1; // Armv7 F16 Kernel 4x16x1
  36. class AlgoF16MK8_4x8; // Armv7 F16 MK8 Format block 4x8
  37. #endif
  38. #if __ARM_FEATURE_DOTPROD
  39. class AlgoInt8x8x32K6x8x4; // Armv7 Int8 Kernel 6x8x4
  40. class AlgoQuint8DotK4x8x4; // Armv7 Quint8 Kernel 6x8x4
  41. class AlgoInt8x8x32MK4_8x4x4DotProd; // Armv7 nchw44 Int8x8x32 Kernel 8x4x4
  42. // DotProduct
  43. #endif
  44. class AlgoPack;
  45. };
  46. } // namespace armv7
  47. } // namespace megdnn
  48. // vim: syntax=cpp.doxygen

MegEngine 安装包中集成了使用 GPU 运行代码所需的 CUDA 环境,不用区分 CPU 和 GPU 版。 如果想要运行 GPU 程序,请确保机器本身配有 GPU 硬件设备并安装好驱动。 如果你想体验在云端 GPU 算力平台进行深度学习开发的感觉,欢迎访问 MegStudio 平台