| 1 | // SPDX-License-Identifier: MIT |
| 2 | /* |
| 3 | * Copyright 2023 Advanced Micro Devices, Inc. |
| 4 | * |
| 5 | * Permission is hereby granted, free of charge, to any person obtaining a |
| 6 | * copy of this software and associated documentation files (the "Software"), |
| 7 | * to deal in the Software without restriction, including without limitation |
| 8 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 9 | * and/or sell copies of the Software, and to permit persons to whom the |
| 10 | * Software is furnished to do so, subject to the following conditions: |
| 11 | * |
| 12 | * The above copyright notice and this permission notice shall be included in |
| 13 | * all copies or substantial portions of the Software. |
| 14 | * |
| 15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 18 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 19 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 20 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 21 | * OTHER DEALINGS IN THE SOFTWARE. |
| 22 | * |
| 23 | * Authors: AMD |
| 24 | * |
| 25 | */ |
| 26 | #include "dm_services.h" |
| 27 | #include "custom_float.h" |
| 28 | |
| 29 | static bool build_custom_float(struct fixed31_32 value, |
| 30 | const struct custom_float_format *format, |
| 31 | bool *negative, |
| 32 | uint32_t *mantissa, |
| 33 | uint32_t *exponenta) |
| 34 | { |
| 35 | uint32_t exp_offset = (1 << (format->exponenta_bits - 1)) - 1; |
| 36 | |
| 37 | const struct fixed31_32 mantissa_constant_plus_max_fraction = |
| 38 | dc_fixpt_from_fraction(numerator: (1LL << (format->mantissa_bits + 1)) - 1, |
| 39 | denominator: 1LL << format->mantissa_bits); |
| 40 | |
| 41 | struct fixed31_32 mantiss; |
| 42 | |
| 43 | if (dc_fixpt_eq(arg1: value, arg2: dc_fixpt_zero)) { |
| 44 | *negative = false; |
| 45 | *mantissa = 0; |
| 46 | *exponenta = 0; |
| 47 | return true; |
| 48 | } |
| 49 | |
| 50 | if (dc_fixpt_lt(arg1: value, arg2: dc_fixpt_zero)) { |
| 51 | *negative = format->sign; |
| 52 | value = dc_fixpt_neg(arg: value); |
| 53 | } else { |
| 54 | *negative = false; |
| 55 | } |
| 56 | |
| 57 | if (dc_fixpt_lt(arg1: value, arg2: dc_fixpt_one)) { |
| 58 | uint32_t i = 1; |
| 59 | |
| 60 | do { |
| 61 | value = dc_fixpt_shl(arg: value, shift: 1); |
| 62 | ++i; |
| 63 | } while (dc_fixpt_lt(arg1: value, arg2: dc_fixpt_one)); |
| 64 | |
| 65 | --i; |
| 66 | |
| 67 | if (exp_offset <= i) { |
| 68 | *mantissa = 0; |
| 69 | *exponenta = 0; |
| 70 | return true; |
| 71 | } |
| 72 | |
| 73 | *exponenta = exp_offset - i; |
| 74 | } else if (dc_fixpt_le(arg1: mantissa_constant_plus_max_fraction, arg2: value)) { |
| 75 | uint32_t i = 1; |
| 76 | |
| 77 | do { |
| 78 | value = dc_fixpt_shr(arg: value, shift: 1); |
| 79 | ++i; |
| 80 | } while (dc_fixpt_lt(arg1: mantissa_constant_plus_max_fraction, arg2: value)); |
| 81 | |
| 82 | *exponenta = exp_offset + i - 1; |
| 83 | } else { |
| 84 | *exponenta = exp_offset; |
| 85 | } |
| 86 | |
| 87 | mantiss = dc_fixpt_sub(arg1: value, arg2: dc_fixpt_one); |
| 88 | |
| 89 | if (dc_fixpt_lt(arg1: mantiss, arg2: dc_fixpt_zero) || |
| 90 | dc_fixpt_lt(arg1: dc_fixpt_one, arg2: mantiss)) |
| 91 | mantiss = dc_fixpt_zero; |
| 92 | else |
| 93 | mantiss = dc_fixpt_shl(arg: mantiss, shift: format->mantissa_bits); |
| 94 | |
| 95 | *mantissa = dc_fixpt_floor(arg: mantiss); |
| 96 | |
| 97 | return true; |
| 98 | } |
| 99 | |
| 100 | static bool setup_custom_float(const struct custom_float_format *format, |
| 101 | bool negative, |
| 102 | uint32_t mantissa, |
| 103 | uint32_t exponenta, |
| 104 | uint32_t *result) |
| 105 | { |
| 106 | uint32_t i = 0; |
| 107 | uint32_t j = 0; |
| 108 | uint32_t value = 0; |
| 109 | |
| 110 | /* verification code: |
| 111 | * once calculation is ok we can remove it |
| 112 | */ |
| 113 | |
| 114 | const uint32_t mantissa_mask = |
| 115 | (1 << (format->mantissa_bits + 1)) - 1; |
| 116 | |
| 117 | const uint32_t exponenta_mask = |
| 118 | (1 << (format->exponenta_bits + 1)) - 1; |
| 119 | |
| 120 | if (mantissa & ~mantissa_mask) { |
| 121 | BREAK_TO_DEBUGGER(); |
| 122 | mantissa = mantissa_mask; |
| 123 | } |
| 124 | |
| 125 | if (exponenta & ~exponenta_mask) { |
| 126 | BREAK_TO_DEBUGGER(); |
| 127 | exponenta = exponenta_mask; |
| 128 | } |
| 129 | |
| 130 | /* end of verification code */ |
| 131 | |
| 132 | while (i < format->mantissa_bits) { |
| 133 | uint32_t mask = 1 << i; |
| 134 | |
| 135 | if (mantissa & mask) |
| 136 | value |= mask; |
| 137 | |
| 138 | ++i; |
| 139 | } |
| 140 | |
| 141 | while (j < format->exponenta_bits) { |
| 142 | uint32_t mask = 1 << j; |
| 143 | |
| 144 | if (exponenta & mask) |
| 145 | value |= mask << i; |
| 146 | |
| 147 | ++j; |
| 148 | } |
| 149 | |
| 150 | if (negative && format->sign) |
| 151 | value |= 1 << (i + j); |
| 152 | |
| 153 | *result = value; |
| 154 | |
| 155 | return true; |
| 156 | } |
| 157 | |
| 158 | bool convert_to_custom_float_format(struct fixed31_32 value, |
| 159 | const struct custom_float_format *format, |
| 160 | uint32_t *result) |
| 161 | { |
| 162 | uint32_t mantissa; |
| 163 | uint32_t exponenta; |
| 164 | bool negative; |
| 165 | |
| 166 | return build_custom_float(value, format, negative: &negative, mantissa: &mantissa, exponenta: &exponenta) && |
| 167 | setup_custom_float(format, |
| 168 | negative, |
| 169 | mantissa, |
| 170 | exponenta, |
| 171 | result); |
| 172 | } |
| 173 | |
| 174 | |