xref: /freebsd/share/man/man3/qmath.3 (revision f5463265955b829775bbb32e1fd0bc11dafc36ce)
1.\"
2.\" Copyright (c) 2018 Netflix, Inc.
3.\" All rights reserved.
4.\"
5.\" Redistribution and use in source and binary forms, with or without
6.\" modification, are permitted provided that the following conditions
7.\" are met:
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.
13.\"
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
24.\" SUCH DAMAGE.
25.\"
26.Dd July 4, 2019
27.Dt QMATH 3
28.Os
29.Sh NAME
30.Nm qmath
31.Nd fixed-point math library based on the
32.Dq Q
33number format
34.Sh SYNOPSIS
35.In sys/qmath.h
36.Sh DESCRIPTION
37The
38.Nm
39data types and APIs support fixed-point math based on the
40.Dq Q
41number format.
42The APIs have been built around the following data types:
43.Vt s8q_t ,
44.Vt u8q_t ,
45.Vt s16q_t ,
46.Vt u16q_t ,
47.Vt s32q_t ,
48.Vt u32q_t ,
49.Vt s64q_t ,
50and
51.Vt u64q_t ,
52which are referred to generically in the earlier API definitions as
53.Fa QTYPE .
54The
55.Fa ITYPE
56refers to the
57.Xr stdint 7
58integer types.
59.Fa NTYPE
60is used to refer to any numeric type and is therefore a superset of
61.Fa QTYPE
62and
63.Fa ITYPE .
64.Pp
65This scheme can represent Q numbers with
66.Bq 2, 4, 6, 8, 16, 32, 48
67bits of precision after the binary radix point,
68depending on the
69.Fa rpshft
70argument to
71.Fn Q_INI .
72The number of bits available for the integral component is not explicitly
73specified, and implicitly consumes the remaining available bits of the chosen Q
74data type.
75.Pp
76Operations on Q numbers maintain the precision of their arguments.
77The fractional component is truncated to fit into the destination,
78with no rounding.
79None of the operations is affected by the floating-point environment.
80.Pp
81For more details, see the
82.Sx IMPLEMENTATION DETAILS
83below.
84.Sh LIST OF FUNCTIONS
85.de Cl
86.Bl -column "isgreaterequal" "bessel function of the second kind of the order 0"
87.Em "Name	Description"
88..
89.Ss Functions which create/initialise a Q number
90.Cl
91.Xr Q_INI 3	initialise a Q number
92.El
93.Ss Numeric functions which operate on two Q numbers
94.Cl
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
104.El
105.Ss Numeric functions which apply integers to a Q number
106.Cl
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
113.El
114.Ss Numeric functions which operate on a single Q number
115.Cl
116.Xr Q_QABS 3	absolute value
117.Xr Q_Q2D 3	double representation
118.Xr Q_Q2F 3	float representation
119.El
120.Ss Comparison and logic functions
121.Cl
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
129.Xr Q_QEQ 3	equal
130.Xr Q_QNEQ 3	not equal
131.Xr Q_OFLOW 3	would overflow
132.Xr Q_RELPREC 3	relative precision
133.El
134.Ss Functions which manipulate the control/sign data bits
135.Cl
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
143.El
144.Ss Functions which manipulate the combined integer/fractional data bits
145.Cl
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
154.El
155.Ss Functions which manipulate the integer data bits
156.Cl
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
162.El
163.Ss Functions which manipulate the fractional data bits
164.Cl
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
170.El
171.Ss Miscellaneous functions/variables
172.Cl
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
189.El
190.Sh IMPLEMENTATION DETAILS
191The
192.Nm
193data types and APIs support fixed-point math based on the
194.Dq Q
195number format.
196This implementation uses the Q notation
197.Em Qm.n ,
198where
199.Em m
200specifies the number of bits for integral data
201.Pq excluding the sign bit for signed types ,
202and
203.Em n
204specifies the number of bits for fractional data.
205.Pp
206The APIs have been built around the following q_t derived data types:
207.Bd -literal -offset indent
208typedef int8_t		s8q_t;
209typedef uint8_t		u8q_t;
210typedef int16_t		s16q_t;
211typedef uint16_t	u16q_t;
212typedef int32_t		s32q_t;
213typedef uint32_t	u32q_t;
214typedef int64_t		s64q_t;
215typedef uint64_t	u64q_t;
216.Ed
217.Pp
218These types are referred to generically in the earlier API definitions as
219.Fa QTYPE ,
220while
221.Fa ITYPE
222refers to the
223.Xr stdint 7
224integer types the Q data types are derived from.
225.Fa NTYPE
226is used to refer to any numeric type and is therefore a superset of
227.Fa QTYPE
228and
229.Fa ITYPE .
230.Pp
231The 3 least significant bits
232.Pq LSBs
233of all q_t data types are reserved for embedded control data:
234.Bl -dash
235.It
236bits 1-2 specify the binary radix point shift index operand, with 00,01,10,11 ==
2371,2,3,4.
238.It
239bit 3 specifies the radix point shift index operand multiplier as 2
240.Pq 0
241or 16
242.Pq 1 .
243.El
244.Pp
245This scheme can therefore represent Q numbers with
246.Bq 2,4,6,8,16,32,48,64
247bits of precision after the binary radix point.
248The number of bits available for the integral component is not explicitly
249specified, and implicitly consumes the remaining available bits of the chosen Q
250data type.
251.Pp
252Additionally, the most significant bit
253.Pq MSB
254of signed Q types stores the sign bit, with bit value 0 representing a positive
255number and bit value 1 representing a negative number.
256Negative numbers are stored as absolute values with the sign bit set, rather
257than the more typical two's complement representation.
258This avoids having to bit shift negative numbers, which can result in undefined
259behaviour from some compilers.
260.Pp
261This binary representation used for Q numbers therefore comprises a set of
262distinct data bit types and associated bit counts.
263Data bit types/labels, listed in LSB to MSB order, are: control
264.Sq C ,
265fractional
266.Sq F ,
267integer
268.Sq I
269and sign
270.Sq S .
271The following example illustrates the binary representation of a Q20.8 number
272represented using a s32q_t variable:
273.Bd -literal -offset indent
274M                                                             L
275S                                                             S
276B                                                             B
277
2783 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
2791 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
280
281S 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
282.Ed
283.Pp
284Important bit counts are: total, control, control-encoded fractional, maximum
285fractional, effective fractional and integer bits.
286.Pp
287The count of total bits is derived from the size of the q_t data type.
288For example, a s32q_t has 32 total bits.
289.Pp
290The count of control-encoded fractional bits is derived from calculating the
291number of fractional bits per the control bit encoding scheme.
292For example, the control bits binary value of 101 encodes a fractional bit
293count of 2 x 16 = 32 fractional bits.
294.Pp
295The count of maximum fractional bits is derived from the difference between the
296counts of total bits and control/sign bits.
297For example, a s32q_t has a maximum of 32 - 3 - 1 = 28 fractional bits.
298.Pp
299The count of effective fractional bits is derived from the minimum of the
300control-encoded fractional bits and the maximum fractional bits.
301For example, a s32q_t with 32 control-encoded fractional bits is effectively
302limited to 28 fractional bits.
303.Pp
304The count of integer bits is derived from the difference between the counts of
305total bits and all other non-integer data bits
306.Pq the sum of control, fractional and sign bits.
307For example, a s32q_t with 8 effective fractional bits has 32 - 3 - 8 - 1 = 20 integer
308bits.
309The count of integer bits can be zero if all available numeric data bits have
310been reserved for fractional data, e.g., when the number of control-encoded
311fractional bits is greater than or equal to the underlying Q data type's maximum
312fractional bits.
313.Sh EXAMPLES
314.Ss Calculating area of a circle with r=4.2 and rpshft=16
315.Bd -literal -offset indent
316u64q_t a, pi, r;
317char buf[32]
318
319Q_INI(&a, 0, 0, 16);
320Q_INI(&pi, 3, 14159, 16);
321Q_INI(&r, 4, 2, 16);
322
323Q_QCLONEQ(&a, r);
324Q_QMULQ(&a, r);
325Q_QMULQ(&a, pi);
326
327Q_TOSTR(a, -1, 10, buf, sizeof(buf));
328printf("%s\\n", buf);
329.Ed
330.Ss Debugging
331Declare a Q20.8 s32q_t number
332.Fa s32 ,
333initialise it with the fixed-point value for 5/3, and render a debugging
334representation of the variable
335.Pq including its full precision decimal C-string representation ,
336to the console:
337.Bd -literal -offset indent
338s32q_t s32;
339Q_INI(&s32, 0, 0, 8);
340Q_QFRACI(&s32, 5, 3);
341char buf[Q_MAXSTRLEN(s32, 10)];
342Q_TOSTR(s32, -1, 10, buf, sizeof(buf));
343printf(Q_DEBUG(s32, "", "\\n\\ttostr=%s\\n\\n", 0), buf);
344.Ed
345.Pp
346The above code outputs the following to the console:
347.Bd -literal -offset indent
348"s32"@0x7fffffffe7d4
349	type=s32q_t, Qm.n=Q20.8, rpshft=11, imin=0xfff00001, \\
350imax=0xfffff
351	qraw=0x00000d53
352	imask=0x7ffff800, fmask=0x000007f8, cmask=0x00000007, \\
353ifmask=0x7ffffff8
354	iraw=0x00000800, iabsval=0x1, ival=0x1
355	fraw=0x00000550, fabsval=0xaa, fval=0xaa
356	tostr=1.664
357.Ed
358.Pp
359Note: The
360.Qq \e
361present in the rendered output above indicates a manual line break inserted to
362keep the man page within 80 columns and is not part of the actual output.
363.Sh SEE ALSO
364.Xr errno 2 ,
365.Xr math 3 ,
366.Xr Q_FRAWMASK 3 ,
367.Xr Q_IFRAWMASK 3 ,
368.Xr Q_INI 3 ,
369.Xr Q_IRAWMASK 3 ,
370.Xr Q_QABS 3 ,
371.Xr Q_QADDI 3 ,
372.Xr Q_QADDQ 3 ,
373.Xr Q_SIGNED 3 ,
374.Xr Q_SIGNSHFT 3 ,
375.Xr stdint 7
376.Sh HISTORY
377The
378.Nm
379functions first appeared in
380.Fx 13.0 .
381.Sh AUTHORS
382.An -nosplit
383The
384.Nm
385functions and this manual page were written by
386.An Lawrence Stewart Aq Mt lstewart@FreeBSD.org
387and sponsored by Netflix, Inc.
388