zircon/kernel/lib/fixed_point/include/lib/fixed_point.h

// Copyright 2016 The Fuchsia Authors
// Copyright (c) 2013, Google Inc. All rights reserved.
//
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file or at
// https://opensource.org/licenses/MIT

#ifndef ZIRCON_KERNEL_LIB_FIXED_POINT_INCLUDE_LIB_FIXED_POINT_H_
#define ZIRCON_KERNEL_LIB_FIXED_POINT_INCLUDE_LIB_FIXED_POINT_H_

#include <stdint.h>

#ifndef DEBUG_FIXED_POINT
#define DEBUG_FIXED_POINT 0
#endif

struct fp_32_64 {
  uint32_t l0;  /* unshifted value */
  uint32_t l32; /* value shifted left 32 bits (or bit -1 to -32) */
  uint32_t l64; /* value shifted left 64 bits (or bit -33 to -64) */
};

#include "fixed_point_debug.h"

static inline void fp_32_64_div_32_32(struct fp_32_64* result, uint32_t dividend, uint32_t divisor) {
  uint64_t tmp;
  uint32_t rem;

  tmp = ((uint64_t)dividend << 32) / divisor;
  rem = (uint32_t)(((uint64_t)dividend << 32) % divisor);
  result->l0 = (uint32_t)(tmp >> 32);
  result->l32 = (uint32_t)tmp;
  tmp = ((uint64_t)rem << 32) / divisor;
  result->l64 = (uint32_t)tmp;
}

static inline uint64_t mul_u32_u32(uint32_t a, uint32_t b, int a_shift, int b_shift) {
  uint64_t ret = (uint64_t)a * b;
  debug_mul_u32_u32(a, b, a_shift, b_shift, ret);
  return ret;
}

static inline uint64_t u64_mul_u32_fp32_64(uint32_t a, struct fp_32_64 b) {
  uint64_t tmp;
  uint64_t res_0;
  uint64_t res_l32;
  uint32_t res_l32_32;
  uint64_t ret;

  res_0 = mul_u32_u32(a, b.l0, 0, 0);
  tmp = mul_u32_u32(a, b.l32, 0, -32);
  res_0 += tmp >> 32;
  res_l32 = (uint32_t)tmp;
  res_l32 += mul_u32_u32(a, b.l64, 0, -64) >> 32; /* Improve rounding accuracy */
  res_0 += res_l32 >> 32;
  res_l32_32 = (uint32_t)res_l32;
  ret = res_0 + (res_l32_32 >> 31); /* Round to nearest integer */

  debug_u64_mul_u32_fp32_64(a, b, res_0, res_l32_32, ret);

  return ret;
}

static inline uint32_t u32_mul_u64_fp32_64(uint64_t a, struct fp_32_64 b) {
  uint32_t a_r32 = (uint32_t)(a >> 32);
  uint32_t a_0 = (uint32_t)a;
  uint64_t res_l32;
  uint32_t ret;

  /* mul_u32_u32(a_r32, b.l0, 32, 0) does not affect result */
  res_l32 = mul_u32_u32(a_0, b.l0, 0, 0) << 32;
  res_l32 += mul_u32_u32(a_r32, b.l32, 32, -32) << 32;
  res_l32 += mul_u32_u32(a_0, b.l32, 0, -32);
  res_l32 += mul_u32_u32(a_r32, b.l64, 32, -64);
  res_l32 += mul_u32_u32(a_0, b.l64, 0, -64) >> 32;              /* Improve rounding accuracy */
  ret = (uint32_t)((res_l32 >> 32) + ((uint32_t)res_l32 >> 31)); /* Round to nearest integer */

  debug_u32_mul_u64_fp32_64(a, b, res_l32, ret);

  return ret;
}

static inline uint64_t u64_mul_u64_fp32_64(uint64_t a, struct fp_32_64 b) {
  uint32_t a_r32 = (uint32_t)(a >> 32);
  uint32_t a_0 = (uint32_t)a;
  uint64_t res_0;
  uint64_t res_l32;
  uint32_t res_l32_32;
  uint64_t tmp;
  uint64_t ret;

  tmp = mul_u32_u32(a_r32, b.l0, 32, 0);
  res_0 = tmp << 32;
  tmp = mul_u32_u32(a_0, b.l0, 0, 0);
  res_0 += tmp;
  tmp = mul_u32_u32(a_r32, b.l32, 32, -32);
  res_0 += tmp;
  tmp = mul_u32_u32(a_0, b.l32, 0, -32);
  res_0 += tmp >> 32;
  res_l32 = (uint32_t)tmp;
  tmp = mul_u32_u32(a_r32, b.l64, 32, -64);
  res_0 += tmp >> 32;
  res_l32 += (uint32_t)tmp;
  tmp = mul_u32_u32(a_0, b.l64, 0, -64); /* Improve rounding accuracy */
  res_l32 += tmp >> 32;
  res_0 += res_l32 >> 32;
  res_l32_32 = (uint32_t)(res_l32);
  ret = res_0 + (res_l32_32 >> 31); /* Round to nearest integer */

  debug_u64_mul_u64_fp32_64(a, b, res_0, res_l32_32, ret);

  return ret;
}

#endif  // ZIRCON_KERNEL_LIB_FIXED_POINT_INCLUDE_LIB_FIXED_POINT_H_