#include <limits.h>
#include <riscv_vector.h>
#include <springbok.h>
#include <stdio.h>
#include <stdlib.h>

#include <bit>
#include <tuple>

#include "pw_unit_test/framework.h"
#include "test_v_helpers.h"

namespace vadd_vv_test {
namespace {

const int LMUL_MAX = 8;
const int VLEN = 512;
const int MAXVL_BYTES = VLEN / sizeof(uint8_t) * LMUL_MAX;

using namespace test_v_helpers;

uint8_t test_vector_1[MAXVL_BYTES];
uint8_t test_vector_2[MAXVL_BYTES];

static void zero_vector_registers() {
  // Clear all vector registers
  int vlmax = get_vsetvlmax_intrinsic(VSEW::SEW_E32, VLMUL::LMUL_M8);
  set_vsetvl_intrinsic(VSEW::SEW_E32, VLMUL::LMUL_M8, vlmax);
  __asm__ volatile("vmv.v.i v0, 0");
  __asm__ volatile("vmv.v.i v8, 0");
  __asm__ volatile("vmv.v.i v16, 0");
  __asm__ volatile("vmv.v.i v24, 0");
}

template <typename T>
static std::tuple<int, int> vadd_vv_test_setup(VLMUL lmul, int32_t avl) {
  // Clear all vector registers
  zero_vector_registers();

  // Initialize test_vector_1 and determine vl, vlmax
  uint32_t bw = std::__bit_width(sizeof(T));
  VSEW sew = static_cast<VSEW>(bw - 1);
  int vlmax = get_vsetvlmax_intrinsic(sew, lmul);
  if (avl > vlmax) {
    avl = vlmax;
  }
  memset(test_vector_1, 0, MAXVL_BYTES);
  memset(test_vector_2, 0, MAXVL_BYTES);
  int vl = set_vsetvl_intrinsic(sew, lmul, avl);
  EXPECT_EQ(avl, vl);

  return std::make_tuple(vlmax, vl);
}

template <typename T>
static void vadd_vv_check(int avl) {
  T *ptr_vec_1 = reinterpret_cast<T *>(test_vector_1);
  T *ptr_vec_2 = reinterpret_cast<T *>(test_vector_2);
  for (int idx = 0; idx < avl; idx++) {
    ASSERT_EQ(ptr_vec_1[idx], ptr_vec_2[idx]);
  }
}

class VaddVxTest : public ::testing::Test {
 protected:
  void SetUp() override { zero_vector_registers(); }
  void TearDown() override { zero_vector_registers(); }
};

TEST_F(VaddVxTest, vadd_vv_demo) {
  for (int i = 0; i < AVL_COUNT; i++) {
    int32_t avl = AVLS[i];
    int vlmax;
    int vl;
    std::tie(vlmax, vl) = vadd_vv_test_setup<uint8_t>(VLMUL::LMUL_M1, avl);
    if (avl > vlmax) {
      continue;
    }
    int8_t *ptr_vec_1 = reinterpret_cast<int8_t *>(test_vector_1);
    int8_t *ptr_vec_2 = reinterpret_cast<int8_t *>(test_vector_2);

    // set up values to test up to index of the AVL
    for (int idx = 0; idx < avl; idx++) {
      // restrict values to valid int8_t range
      ptr_vec_1[idx] = idx % (INT8_MAX) + (INT8_MIN);
    }
    __asm__ volatile("vle8.v v8, (%0)" : : "r"(ptr_vec_1));
    __asm__ volatile("vadd.vv v8, v8, v8");
    for (int idx = 0; idx < vl; idx++) {
      ptr_vec_1[idx] =
          ptr_vec_1[idx] + ptr_vec_1[idx];  // restrict values to valid int8
    }
    __asm__ volatile("vse8.v v8, (%0)" : : "r"(ptr_vec_2));
    vadd_vv_check<int8_t>(vlmax);
  }
}

#define DEFINE_TEST_VADD_VV(_SEW_, _LMUL_)                                \
  TEST_F(VaddVxTest, vadd_vv##_SEW_##m##_LMUL_) {                         \
    for (int i = 0; i < AVL_COUNT; i++) {                                 \
      int32_t avl = AVLS[i];                                              \
      int vlmax;                                                          \
      int vl;                                                             \
      std::tie(vlmax, vl) =                                               \
          vadd_vv_test_setup<int##_SEW_##_t>(VLMUL::LMUL_M##_LMUL_, avl); \
      if (avl > vlmax) {                                                  \
        continue;                                                         \
      }                                                                   \
      int##_SEW_##_t *ptr_vec_1 =                                         \
          reinterpret_cast<int##_SEW_##_t *>(test_vector_1);              \
      int##_SEW_##_t *ptr_vec_2 =                                         \
          reinterpret_cast<int##_SEW_##_t *>(test_vector_2);              \
                                                                          \
      for (long long idx = 0; idx < avl; idx++) {                         \
        ptr_vec_1[idx] = idx % INT##_SEW_##_MAX + INT##_SEW_##_MIN;       \
      }                                                                   \
                                                                          \
      __asm__ volatile("vle" #_SEW_ ".v v8, (%0)" : : "r"(ptr_vec_1));    \
      __asm__ volatile("vadd.vv v8, v8, v8");                             \
      for (int idx = 0; idx < vl; idx++) {                                \
        ptr_vec_1[idx] = ptr_vec_1[idx] + ptr_vec_1[idx];                 \
      }                                                                   \
      __asm__ volatile("vse" #_SEW_ ".v v8, (%0)" : : "r"(ptr_vec_2));    \
      vadd_vv_check<int##_SEW_##_t>(vlmax);                               \
    }                                                                     \
  }

// TODO(gkielian): modify macro to permit more than one test per sew/lmul pair
DEFINE_TEST_VADD_VV(8, 1)
DEFINE_TEST_VADD_VV(8, 2)
DEFINE_TEST_VADD_VV(8, 4)
DEFINE_TEST_VADD_VV(8, 8)

DEFINE_TEST_VADD_VV(16, 1)
DEFINE_TEST_VADD_VV(16, 2)
DEFINE_TEST_VADD_VV(16, 4)
DEFINE_TEST_VADD_VV(16, 8)

DEFINE_TEST_VADD_VV(32, 1)
DEFINE_TEST_VADD_VV(32, 2)
DEFINE_TEST_VADD_VV(32, 4)
DEFINE_TEST_VADD_VV(32, 8)

}  // namespace
}  // namespace vadd_vv_test