xref: /freebsd/share/man/man3/qmath.3 (revision 38effe887ee979f91ad5abf42a2291558e7ff8d1)
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.\" $FreeBSD$
27.\"
28.Dd July 4, 2019
29.Dt QMATH 3
30.Os
31.Sh NAME
32.Nm qmath
33.Nd fixed-point math library based on the
34.Dq Q
35number format
36.Sh SYNOPSIS
37.In sys/qmath.h
38.Sh DESCRIPTION
39The
40.Nm
41data types and APIs support fixed-point math based on the
42.Dq Q
43number format.
44The APIs have been built around the following data types:
45.Vt s8q_t ,
46.Vt u8q_t ,
47.Vt s16q_t ,
48.Vt u16q_t ,
49.Vt s32q_t ,
50.Vt u32q_t ,
51.Vt s64q_t ,
52and
53.Vt u64q_t ,
54which are referred to generically in the earlier API definitions as
55.Fa QTYPE .
56The
57.Fa ITYPE
58refers to the
59.Xr stdint 7
60integer types.
61.Fa NTYPE
62is used to refer to any numeric type and is therefore a superset of
63.Fa QTYPE
64and
65.Fa ITYPE .
66.Pp
67This scheme can represent Q numbers with
68.Bq 2, 4, 6, 8, 16, 32, 48
69bits of precision after the binary radix point,
70depending on the
71.Fa rpshft
72argument to
73.Fn Q_INI .
74The number of bits available for the integral component is not explicitly
75specified, and implicitly consumes the remaining available bits of the chosen Q
76data type.
77.Pp
78Operations on Q numbers maintain the precision of their arguments.
79The fractional component is truncated to fit into the destination,
80with no rounding.
81None of the operations is affected by the floating-point environment.
82.Pp
83For more details, see the
84.Sx IMPLEMENTATION DETAILS
85below.
86.Sh LIST OF FUNCTIONS
87.de Cl
88.Bl -column "isgreaterequal" "bessel function of the second kind of the order 0"
89.Em "Name	Description"
90..
91.Ss Functions which create/initialise a Q number
92.Cl
93.Xr Q_INI 3	initialise a Q number
94.El
95.Ss Numeric functions which operate on two Q numbers
96.Cl
97.Xr Q_QADDQ 3	addition
98.Xr Q_QDIVQ 3	division
99.Xr Q_QMULQ 3	multiplication
100.Xr Q_QSUBQ 3	subtraction
101.Xr Q_NORMPREC 3	normalisation
102.Xr Q_QMAXQ 3	maximum function
103.Xr Q_QMINQ 3	minimum function
104.Xr Q_QCLONEQ 3	identical copy
105.Xr Q_QCPYVALQ 3	representational copy
106.El
107.Ss Numeric functions which apply integers to a Q number
108.Cl
109.Xr Q_QADDI 3	addition
110.Xr Q_QDIVI 3	division
111.Xr Q_QMULI 3	multiplication
112.Xr Q_QSUBI 3	subtraction
113.Xr Q_QFRACI 3	fraction
114.Xr Q_QCPYVALI 3	overwrite
115.El
116.Ss Numeric functions which operate on a single Q number
117.Cl
118.Xr Q_QABS 3	absolute value
119.Xr Q_Q2D 3	double representation
120.Xr Q_Q2F 3	float representation
121.El
122.Ss Comparison and logic functions
123.Cl
124.Xr Q_SIGNED 3	determine sign
125.Xr Q_LTZ 3	less than zero
126.Xr Q_PRECEQ 3	compare bits
127.Xr Q_QLTQ 3	less than
128.Xr Q_QLEQ 3	less or equal
129.Xr Q_QGTQ 3	greater than
130.Xr Q_QGEQ 3	greater or equal
131.Xr Q_QEQ 3	equal
132.Xr Q_QNEQ 3	not equal
133.Xr Q_OFLOW 3	would overflow
134.Xr Q_RELPREC 3	relative precision
135.El
136.Ss Functions which manipulate the control/sign data bits
137.Cl
138.Xr Q_SIGNSHFT 3	sign bit position
139.Xr Q_SSIGN 3	sign bit
140.Xr Q_CRAWMASK 3	control bitmask
141.Xr Q_SRAWMASK 3	sign bitmask
142.Xr Q_GCRAW 3	raw control bits
143.Xr Q_GCVAL 3	value of control bits
144.Xr Q_SCVAL 3	set control bits
145.El
146.Ss Functions which manipulate the combined integer/fractional data bits
147.Cl
148.Xr Q_IFRAWMASK 3	integer/fractional bitmask
149.Xr Q_IFVALIMASK 3	value of integer bits
150.Xr Q_IFVALFMASK 3	value of fractional bits
151.Xr Q_GIFRAW 3	raw integer/fractional bits
152.Xr Q_GIFABSVAL 3	absolute value of fractional bits
153.Xr Q_GIFVAL 3	real value of fractional bits
154.Xr Q_SIFVAL 3	set integer/fractional bits
155.Xr Q_SIFVALS 3	set separate integer/fractional values
156.El
157.Ss Functions which manipulate the integer data bits
158.Cl
159.Xr Q_IRAWMASK 3	integer bitmask
160.Xr Q_GIRAW 3	raw integer bits
161.Xr Q_GIABSVAL 3	absolute value of integer bits
162.Xr Q_GIVAL 3	real value of integer bits
163.Xr Q_SIVAL 3	set integer bits
164.El
165.Ss Functions which manipulate the fractional data bits
166.Cl
167.Xr Q_FRAWMASK 3	fractional bitmask
168.Xr Q_GFRAW 3	raw fractional bits
169.Xr Q_GFABSVAL 3	absolute value of fractional bits
170.Xr Q_GFVAL 3	real value of fractional bits
171.Xr Q_SFVAL 3	set fractional bits
172.El
173.Ss Miscellaneous functions/variables
174.Cl
175.Xr Q_NCBITS 3	number of reserved control bits
176.Xr Q_BT 3	C data type
177.Xr Q_TC 3	casted data type
178.Xr Q_NTBITS 3	number of total bits
179.Xr Q_NFCBITS 3	number of control-encoded fractional bits
180.Xr Q_MAXNFBITS 3	number of maximum fractional bits
181.Xr Q_NFBITS 3	number of effective fractional bits
182.Xr Q_NIBITS 3	number of integer bits
183.Xr Q_RPSHFT 3	bit position of radix point
184.Xr Q_ABS 3	absolute value
185.Xr Q_MAXSTRLEN 3	number of characters to render string
186.Xr Q_TOSTR 3	render string
187.Xr Q_SHL 3	left-shifted value
188.Xr Q_SHR 3	right-shifted value
189.Xr Q_DEBUG 3	render debugging information
190.Xr Q_DFV2BFV 3	convert decimal fractional value
191.El
192.Sh IMPLEMENTATION DETAILS
193The
194.Nm
195data types and APIs support fixed-point math based on the
196.Dq Q
197number format.
198This implementation uses the Q notation
199.Em Qm.n ,
200where
201.Em m
202specifies the number of bits for integral data
203.Pq excluding the sign bit for signed types ,
204and
205.Em n
206specifies the number of bits for fractional data.
207.Pp
208The APIs have been built around the following q_t derived data types:
209.Bd -literal -offset indent
210typedef int8_t		s8q_t;
211typedef uint8_t		u8q_t;
212typedef int16_t		s16q_t;
213typedef uint16_t	u16q_t;
214typedef int32_t		s32q_t;
215typedef uint32_t	u32q_t;
216typedef int64_t		s64q_t;
217typedef uint64_t	u64q_t;
218.Ed
219.Pp
220These types are referred to generically in the earlier API definitions as
221.Fa QTYPE ,
222while
223.Fa ITYPE
224refers to the
225.Xr stdint 7
226integer types the Q data types are derived from.
227.Fa NTYPE
228is used to refer to any numeric type and is therefore a superset of
229.Fa QTYPE
230and
231.Fa ITYPE .
232.Pp
233The 3 least significant bits
234.Pq LSBs
235of all q_t data types are reserved for embedded control data:
236.Bl -dash
237.It
238bits 1-2 specify the binary radix point shift index operand, with 00,01,10,11 ==
2391,2,3,4.
240.It
241bit 3 specifies the radix point shift index operand multiplier as 2
242.Pq 0
243or 16
244.Pq 1 .
245.El
246.Pp
247This scheme can therefore represent Q numbers with
248.Bq 2,4,6,8,16,32,48,64
249bits of precision after the binary radix point.
250The number of bits available for the integral component is not explicitly
251specified, and implicitly consumes the remaining available bits of the chosen Q
252data type.
253.Pp
254Additionally, the most significant bit
255.Pq MSB
256of signed Q types stores the sign bit, with bit value 0 representing a positive
257number and bit value 1 representing a negative number.
258Negative numbers are stored as absolute values with the sign bit set, rather
259than the more typical two's complement representation.
260This avoids having to bit shift negative numbers, which can result in undefined
261behaviour from some compilers.
262.Pp
263This binary representation used for Q numbers therefore comprises a set of
264distinct data bit types and associated bit counts.
265Data bit types/labels, listed in LSB to MSB order, are: control
266.Sq C ,
267fractional
268.Sq F ,
269integer
270.Sq I
271and sign
272.Sq S .
273The following example illustrates the binary representation of a Q20.8 number
274represented using a s32q_t variable:
275.Bd -literal -offset indent
276M                                                             L
277S                                                             S
278B                                                             B
279
2803 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
2811 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
282
283S 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.Ed
285.Pp
286Important bit counts are: total, control, control-encoded fractional, maximum
287fractional, effective fractional and integer bits.
288.Pp
289The count of total bits is derived from the size of the q_t data type.
290For example, a s32q_t has 32 total bits.
291.Pp
292The count of control-encoded fractional bits is derived from calculating the
293number of fractional bits per the control bit encoding scheme.
294For example, the control bits binary value of 101 encodes a fractional bit
295count of 2 x 16 = 32 fractional bits.
296.Pp
297The count of maximum fractional bits is derived from the difference between the
298counts of total bits and control/sign bits.
299For example, a s32q_t has a maximum of 32 - 3 - 1 = 28 fractional bits.
300.Pp
301The count of effective fractional bits is derived from the minimum of the
302control-encoded fractional bits and the maximum fractional bits.
303For example, a s32q_t with 32 control-encoded fractional bits is effectively
304limited to 28 fractional bits.
305.Pp
306The count of integer bits is derived from the difference between the counts of
307total bits and all other non-integer data bits
308.Pq the sum of control, fractional and sign bits.
309For example, a s32q_t with 8 effective fractional bits has 32 - 3 - 8 - 1 = 20 integer
310bits.
311The count of integer bits can be zero if all available numeric data bits have
312been reserved for fractional data, e.g., when the number of control-encoded
313fractional bits is greater than or equal to the underlying Q data type's maximum
314fractional bits.
315.Sh EXAMPLES
316.Ss Calculating area of a circle with r=4.2 and rpshft=16
317.Bd -literal -offset indent
318u64q_t a, pi, r;
319char buf[32]
320
321Q_INI(&a, 0, 0, 16);
322Q_INI(&pi, 3, 14159, 16);
323Q_INI(&r, 4, 2, 16);
324
325Q_QCLONEQ(&a, r);
326Q_QMULQ(&a, r);
327Q_QMULQ(&a, pi);
328
329Q_TOSTR(a, -1, 10, buf, sizeof(buf));
330printf("%s\\n", buf);
331.Ed
332.Ss Debugging
333Declare a Q20.8 s32q_t number
334.Fa s32 ,
335initialise it with the fixed-point value for 5/3, and render a debugging
336representation of the variable
337.Pq including its full precision decimal C-string representation ,
338to the console:
339.Bd -literal -offset indent
340s32q_t s32;
341Q_INI(&s32, 0, 0, 8);
342Q_QFRACI(&s32, 5, 3);
343char buf[Q_MAXSTRLEN(s32, 10)];
344Q_TOSTR(s32, -1, 10, buf, sizeof(buf));
345printf(Q_DEBUG(s32, "", "\\n\\ttostr=%s\\n\\n", 0), buf);
346.Ed
347.Pp
348The above code outputs the following to the console:
349.Bd -literal -offset indent
350"s32"@0x7fffffffe7d4
351	type=s32q_t, Qm.n=Q20.8, rpshft=11, imin=0xfff00001, \\
352imax=0xfffff
353	qraw=0x00000d53
354	imask=0x7ffff800, fmask=0x000007f8, cmask=0x00000007, \\
355ifmask=0x7ffffff8
356	iraw=0x00000800, iabsval=0x1, ival=0x1
357	fraw=0x00000550, fabsval=0xaa, fval=0xaa
358	tostr=1.664
359.Ed
360.Pp
361Note: The
362.Qq \e
363present in the rendered output above indicates a manual line break inserted to
364keep the man page within 80 columns and is not part of the actual output.
365.Sh SEE ALSO
366.Xr errno 2 ,
367.Xr math 3 ,
368.Xr Q_FRAWMASK 3 ,
369.Xr Q_IFRAWMASK 3 ,
370.Xr Q_INI 3 ,
371.Xr Q_IRAWMASK 3 ,
372.Xr Q_QABS 3 ,
373.Xr Q_QADDI 3 ,
374.Xr Q_QADDQ 3 ,
375.Xr Q_SIGNED 3 ,
376.Xr Q_SIGNSHFT 3 ,
377.Xr stdint 7
378.Sh HISTORY
379The
380.Nm
381functions first appeared in
382.Fx 13.0 .
383.Sh AUTHORS
384.An -nosplit
385The
386.Nm
387functions and this manual page were written by
388.An Lawrence Stewart Aq Mt lstewart@FreeBSD.org
389and sponsored by Netflix, Inc.
390