/* Copyright (c) 2018 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. */

#pragma once

#include <string>
#include "paddle/fluid/operators/math/jit_gen.h"
namespace paddle {
namespace operators {
namespace math {
namespace jitkernel {
namespace gen {

using reg64_t = const Xbyak::Reg64;
using reg32_t = const Xbyak::Reg32;
using xmm_t = const Xbyak::Xmm;
using ymm_t = const Xbyak::Ymm;
using zmm_t = const Xbyak::Zmm;
using Label = Xbyak::Label;

typedef enum {
  mul = 0,
  add,
  sub,
  relu,
  exp,
  sigmoid,
  tanh,
  identity
} operand_type;

// function: vec = Operand(vec(or scalar), vec(or scalar)) (maybe with relu)
class VXXJitCode : public JitCode {
 public:
  const char* name() const override {
    std::string base = "VXXJitCode";
    if (scalar_index_ == 1) {
      base += "_Scalar";
    } else {
      base += "_Vec";
    }
    if (type_ == operand_type::mul) {
      base += "_Mul";
    } else if (type_ == operand_type::add) {
      base += "_Add";
    }
    if (scalar_index_ == 2) {
      base += "_Scalar";
    } else {
      base += "_Vec";
    }
    base += (with_relu_ ? "_Relu" : "");
    return base.c_str();
  }
  explicit VXXJitCode(int d, operand_type type, int scalar_index,
                      bool with_relu, size_t code_size = 256 * 1024,
                      void* code_ptr = nullptr)
      : JitCode(code_size, code_ptr),
        num_(d),
        type_(type),
        scalar_index_(scalar_index),
        with_relu_(with_relu) {}
  static bool init(int d, int scalar_index = 0);
  void generate() override;

 private:
  int num_;
  operand_type type_;
  int scalar_index_;
  bool with_relu_;
  reg64_t param1{abi_param1};
  reg64_t param2{abi_param2};
  reg64_t param3{abi_param3};

  xmm_t xmm_src1 = xmm_t(0);
  xmm_t xmm_src2 = xmm_t(1);
  xmm_t xmm_dst = xmm_t(2);
  xmm_t xmm_zero = xmm_t(3);

  ymm_t ymm_src1 = ymm_t(0);
  ymm_t ymm_src2 = ymm_t(1);
  ymm_t ymm_dst = ymm_t(2);
  ymm_t ymm_zero = ymm_t(3);
};

class VActJitCode : public JitCode {
 public:
  const char* name() const override {
    std::string base = "VActJitCode";
    switch (type_) {
      case operand_type::relu:
        base += "_Relu";
        break;
      case operand_type::exp:
        base += "_Exp";
        break;
      case operand_type::sigmoid:
        base += "_Sigmoid";
        break;
      case operand_type::tanh:
        base += "_Tanh";
        break;
      case operand_type::identity:
        base += "_Identity";
        break;
      default:
        break;
    }
    return base.c_str();
  }

  explicit VActJitCode(int d, operand_type type, size_t code_size = 256 * 1024,
                       void* code_ptr = nullptr)
      : JitCode(code_size, code_ptr), num_(d), type_(type) {}
  static bool init(int d, operand_type type);
  void generate() override;

 protected:
  // compute relu with ymm
  void relu_ymm(const Xbyak::Ymm& dst, const Xbyak::Ymm& src,
                const Xbyak::Ymm& zero);

  // compute exp with ymm
  void exp_ymm(const Xbyak::Ymm& dst, const Xbyak::Ymm& src, int fx_idx = 2,
               int fy_idx = 3, int mask_idx = 4, int tmp_idx = 5);

  // compute sigmoid with ymm
  void sigmoid_ymm(const Xbyak::Ymm& dst, const Xbyak::Ymm& src, int fx_idx = 2,
                   int fy_idx = 3, int mask_idx = 4, int tmp_idx = 5);

  // compute tanh with ymm
  void tanh_ymm(const Xbyak::Ymm& dst, const Xbyak::Ymm& src, int fx_idx = 2,
                int fy_idx = 3, int mask_idx = 4, int tmp_idx = 5);

 protected:
  int num_;
  operand_type type_;
  reg64_t param1{abi_param1};
  reg64_t param2{abi_param2};

  xmm_t xmm_src = xmm_t(0);
  ymm_t ymm_src = ymm_t(0);

  xmm_t xmm_dst = xmm_t(1);
  ymm_t ymm_dst = ymm_t(1);
};

#ifdef PADDLE_WITH_MKLDNN
struct EltwiseMulnChw16cNC : public Xbyak::CodeGenerator {
  explicit EltwiseMulnChw16cNC(size_t code_size = 256 * 1024)
      : Xbyak::CodeGenerator(code_size) {
    // RDI is ptr x_input
    // RSI is ptr y_input
    // RDX is ptr output
    // RCX is height
    // r8 is width

    push(rbx);

    xor_(rax, rax);
    xor_(r10, r10);
    vmovups(zmm3, ptr[rsi]);

    L("h_loop");
    xor_(rbx, rbx);
    L("w_loop");
    vmovups(zmm2, ptr[rdi + rax]);
    vmulps(zmm1, zmm2, zmm3);
    vmovups(ptr[rdx + rax], zmm1);
    add(rax, 64);
    inc(rbx);
    cmp(r8, rbx);
    jnz("w_loop");
    inc(r10);
    cmp(r10, rcx);
    jnz("h_loop");

    pop(rbx);
    ret();
  }
};
#endif

}  // namespace gen
}  // namespace jitkernel
}  // namespace math
}  // namespace operators
}  // namespace paddle