2 .\" Copyright (c) 2018 Netflix, Inc.
3 .\" All rights reserved.
5 .\" Redistribution and use in source and binary forms, with or without
6 .\" modification, are permitted provided that the following conditions
8 .\" 1. Redistributions of source code must retain the above copyright
9 .\" notice, this list of conditions, and the following disclaimer,
10 .\" without modification, immediately at the beginning of the file.
11 .\" 2. The name of the author may not be used to endorse or promote products
12 .\" derived from this software without specific prior written permission.
14 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
18 .\" ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 .Nd fixed-point math library based on the
39 data types and APIs support fixed-point math based on the
42 The APIs have been built around the following data types:
52 which are referred to generically in the earlier API definitions as
60 is used to refer to any numeric type and is therefore a superset of
65 This scheme can represent Q numbers with
66 .Bq 2, 4, 6, 8, 16, 32, 48
67 bits of precision after the binary radix point,
72 The number of bits available for the integral component is not explicitly
73 specified, and implicitly consumes the remaining available bits of the chosen Q
76 Operations on Q numbers maintain the precision of their arguments.
77 The fractional component is truncated to fit into the destination,
79 None of the operations is affected by the floating-point environment.
81 For more details, see the
82 .Sx IMPLEMENTATION DETAILS
86 .Bl -column "isgreaterequal" "bessel function of the second kind of the order 0"
87 .Em "Name Description"
89 .Ss Functions which create/initialise a Q number
91 .Xr Q_INI 3 initialise a Q number
93 .Ss Numeric functions which operate on two Q numbers
95 .Xr Q_QADDQ 3 addition
96 .Xr Q_QDIVQ 3 division
97 .Xr Q_QMULQ 3 multiplication
98 .Xr Q_QSUBQ 3 subtraction
99 .Xr Q_NORMPREC 3 normalisation
100 .Xr Q_QMAXQ 3 maximum function
101 .Xr Q_QMINQ 3 minimum function
102 .Xr Q_QCLONEQ 3 identical copy
103 .Xr Q_QCPYVALQ 3 representational copy
105 .Ss Numeric functions which apply integers to a Q number
107 .Xr Q_QADDI 3 addition
108 .Xr Q_QDIVI 3 division
109 .Xr Q_QMULI 3 multiplication
110 .Xr Q_QSUBI 3 subtraction
111 .Xr Q_QFRACI 3 fraction
112 .Xr Q_QCPYVALI 3 overwrite
114 .Ss Numeric functions which operate on a single Q number
116 .Xr Q_QABS 3 absolute value
117 .Xr Q_Q2D 3 double representation
118 .Xr Q_Q2F 3 float representation
120 .Ss Comparison and logic functions
122 .Xr Q_SIGNED 3 determine sign
123 .Xr Q_LTZ 3 less than zero
124 .Xr Q_PRECEQ 3 compare bits
125 .Xr Q_QLTQ 3 less than
126 .Xr Q_QLEQ 3 less or equal
127 .Xr Q_QGTQ 3 greater than
128 .Xr Q_QGEQ 3 greater or equal
130 .Xr Q_QNEQ 3 not equal
131 .Xr Q_OFLOW 3 would overflow
132 .Xr Q_RELPREC 3 relative precision
134 .Ss Functions which manipulate the control/sign data bits
136 .Xr Q_SIGNSHFT 3 sign bit position
137 .Xr Q_SSIGN 3 sign bit
138 .Xr Q_CRAWMASK 3 control bitmask
139 .Xr Q_SRAWMASK 3 sign bitmask
140 .Xr Q_GCRAW 3 raw control bits
141 .Xr Q_GCVAL 3 value of control bits
142 .Xr Q_SCVAL 3 set control bits
144 .Ss Functions which manipulate the combined integer/fractional data bits
146 .Xr Q_IFRAWMASK 3 integer/fractional bitmask
147 .Xr Q_IFVALIMASK 3 value of integer bits
148 .Xr Q_IFVALFMASK 3 value of fractional bits
149 .Xr Q_GIFRAW 3 raw integer/fractional bits
150 .Xr Q_GIFABSVAL 3 absolute value of fractional bits
151 .Xr Q_GIFVAL 3 real value of fractional bits
152 .Xr Q_SIFVAL 3 set integer/fractional bits
153 .Xr Q_SIFVALS 3 set separate integer/fractional values
155 .Ss Functions which manipulate the integer data bits
157 .Xr Q_IRAWMASK 3 integer bitmask
158 .Xr Q_GIRAW 3 raw integer bits
159 .Xr Q_GIABSVAL 3 absolute value of integer bits
160 .Xr Q_GIVAL 3 real value of integer bits
161 .Xr Q_SIVAL 3 set integer bits
163 .Ss Functions which manipulate the fractional data bits
165 .Xr Q_FRAWMASK 3 fractional bitmask
166 .Xr Q_GFRAW 3 raw fractional bits
167 .Xr Q_GFABSVAL 3 absolute value of fractional bits
168 .Xr Q_GFVAL 3 real value of fractional bits
169 .Xr Q_SFVAL 3 set fractional bits
171 .Ss Miscellaneous functions/variables
173 .Xr Q_NCBITS 3 number of reserved control bits
174 .Xr Q_BT 3 C data type
175 .Xr Q_TC 3 casted data type
176 .Xr Q_NTBITS 3 number of total bits
177 .Xr Q_NFCBITS 3 number of control-encoded fractional bits
178 .Xr Q_MAXNFBITS 3 number of maximum fractional bits
179 .Xr Q_NFBITS 3 number of effective fractional bits
180 .Xr Q_NIBITS 3 number of integer bits
181 .Xr Q_RPSHFT 3 bit position of radix point
182 .Xr Q_ABS 3 absolute value
183 .Xr Q_MAXSTRLEN 3 number of characters to render string
184 .Xr Q_TOSTR 3 render string
185 .Xr Q_SHL 3 left-shifted value
186 .Xr Q_SHR 3 right-shifted value
187 .Xr Q_DEBUG 3 render debugging information
188 .Xr Q_DFV2BFV 3 convert decimal fractional value
190 .Sh IMPLEMENTATION DETAILS
193 data types and APIs support fixed-point math based on the
196 This implementation uses the Q notation
200 specifies the number of bits for integral data
201 .Pq excluding the sign bit for signed types ,
204 specifies the number of bits for fractional data.
206 The APIs have been built around the following q_t derived data types:
207 .Bd -literal -offset indent
208 typedef int8_t s8q_t;
209 typedef uint8_t u8q_t;
210 typedef int16_t s16q_t;
211 typedef uint16_t u16q_t;
212 typedef int32_t s32q_t;
213 typedef uint32_t u32q_t;
214 typedef int64_t s64q_t;
215 typedef uint64_t u64q_t;
218 These types are referred to generically in the earlier API definitions as
224 integer types the Q data types are derived from.
226 is used to refer to any numeric type and is therefore a superset of
231 The 3 least significant bits
233 of all q_t data types are reserved for embedded control data:
236 bits 1-2 specify the binary radix point shift index operand, with 00,01,10,11 ==
239 bit 3 specifies the radix point shift index operand multiplier as 2
245 This scheme can therefore represent Q numbers with
246 .Bq 2,4,6,8,16,32,48,64
247 bits of precision after the binary radix point.
248 The number of bits available for the integral component is not explicitly
249 specified, and implicitly consumes the remaining available bits of the chosen Q
252 Additionally, the most significant bit
254 of signed Q types stores the sign bit, with bit value 0 representing a positive
255 number and bit value 1 representing a negative number.
256 Negative numbers are stored as absolute values with the sign bit set, rather
257 than the more typical two's complement representation.
258 This avoids having to bit shift negative numbers, which can result in undefined
259 behaviour from some compilers.
261 This binary representation used for Q numbers therefore comprises a set of
262 distinct data bit types and associated bit counts.
263 Data bit types/labels, listed in LSB to MSB order, are: control
271 The following example illustrates the binary representation of a Q20.8 number
272 represented using a s32q_t variable:
273 .Bd -literal -offset indent
278 3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
279 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
281 S I I I I I I I I I I I I I I I I I I I I F F F F F F F F C C C
284 Important bit counts are: total, control, control-encoded fractional, maximum
285 fractional, effective fractional and integer bits.
287 The count of total bits is derived from the size of the q_t data type.
288 For example, a s32q_t has 32 total bits.
290 The count of control-encoded fractional bits is derived from calculating the
291 number of fractional bits per the control bit encoding scheme.
292 For example, the control bits binary value of 101 encodes a fractional bit
293 count of 2 x 16 = 32 fractional bits.
295 The count of maximum fractional bits is derived from the difference between the
296 counts of total bits and control/sign bits.
297 For example, a s32q_t has a maximum of 32 - 3 - 1 = 28 fractional bits.
299 The count of effective fractional bits is derived from the minimum of the
300 control-encoded fractional bits and the maximum fractional bits.
301 For example, a s32q_t with 32 control-encoded fractional bits is effectively
302 limited to 28 fractional bits.
304 The count of integer bits is derived from the difference between the counts of
305 total bits and all other non-integer data bits
306 .Pq the sum of control, fractional and sign bits.
307 For example, a s32q_t with 8 effective fractional bits has 32 - 3 - 8 - 1 = 20 integer
309 The count of integer bits can be zero if all available numeric data bits have
310 been reserved for fractional data, e.g., when the number of control-encoded
311 fractional bits is greater than or equal to the underlying Q data type's maximum
314 .Ss Calculating area of a circle with r=4.2 and rpshft=16
315 .Bd -literal -offset indent
320 Q_INI(&pi, 3, 14159, 16);
327 Q_TOSTR(a, -1, 10, buf, sizeof(buf));
328 printf("%s\\n", buf);
331 Declare a Q20.8 s32q_t number
333 initialise it with the fixed-point value for 5/3, and render a debugging
334 representation of the variable
335 .Pq including its full precision decimal C-string representation ,
337 .Bd -literal -offset indent
339 Q_INI(&s32, 0, 0, 8);
340 Q_QFRACI(&s32, 5, 3);
341 char buf[Q_MAXSTRLEN(s32, 10)];
342 Q_TOSTR(s32, -1, 10, buf, sizeof(buf));
343 printf(Q_DEBUG(s32, "", "\\n\\ttostr=%s\\n\\n", 0), buf);
346 The above code outputs the following to the console:
347 .Bd -literal -offset indent
349 type=s32q_t, Qm.n=Q20.8, rpshft=11, imin=0xfff00001, \\
352 imask=0x7ffff800, fmask=0x000007f8, cmask=0x00000007, \\
354 iraw=0x00000800, iabsval=0x1, ival=0x1
355 fraw=0x00000550, fabsval=0xaa, fval=0xaa
361 present in the rendered output above indicates a manual line break inserted to
362 keep the man page within 80 columns and is not part of the actual output.
379 functions first appeared in
385 functions and this manual page were written by
386 .An Lawrence Stewart Aq Mt lstewart@FreeBSD.org
387 and sponsored by Netflix, Inc.