Mercurial > vec

annotate README @ 20:627d548b23c8

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
impl/generic: fix load and store implementations this caused a segmentation fault under AltiVec, but it went under the radar on x86 because my main PC supports all of the non-generic vector implementations.
author Paper <paper@tflc.us>
date Thu, 21 Nov 2024 21:19:11 +0000
parents e05c257c6a23
children e26874655738
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
0
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
1 vec - a tiny SIMD vector header-only library written in C99
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
2
15
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
3 it comes with an extremely basic API that is similar to other intrinsics
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
4 libraries; each type is in the exact same format:
0
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
5
15
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
6 v[sign][bits]x[size]
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
7 where `sign' is either nothing (for signed) or `u' (for unsigned),
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
8 `bits' is the bit size of the integer format,
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
9 and `size' is the how many integers are in the vector
0
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
10
15
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
11 vec provides types for 64-bit, 128-bit, 256-bit, and 512-bit SIMD intrinsics
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
12 on processors where vec has an implementation and falls back to array-based
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
13 implementations where they are not.
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
14
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
15 all of these have many operations that are prefixed with the name of the
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
16 type and an underscore, for example:
0
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
17
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
18 vint8x16 vint8x16_splat(uint8_t x)
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
19 - creates a vint8x16 where all of the values are filled
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
20 with the value of `x'
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
21
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
22 the current supported operations are:
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
23
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
24 v[u]intAxB splat([u]intA_t x)
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
25 creates a vector with all of the values are filled with
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
26 the value of `x'
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
27
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
28 v[u]intAxB load(const [u]intA_t x[B])
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
29 copies the values from the memory address stored at `x';
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
30 the address is NOT required to be aligned
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
31
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
32 void store(v[u]intAxB vec, [u]intA_t x[B])
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
33 copies the values from the vector into the memory address
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
34 stored at `x'
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
35
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
36 like with load(), this does not require address alignment
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
37
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
38 v[u]intAxB add(v[u]intAxB vec1, v[u]intAxB vec2)
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
39 adds the value of `vec1' and `vec2' and returns it
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
40
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
41 v[u]intAxB sub(v[u]intAxB vec1, v[u]intAxB vec2)
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
42 subtracts the value of `vec2' from `vec1' and returns it
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
43
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
44 v[u]intAxB mul(v[u]intAxB vec1, v[u]intAxB vec2)
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
45 multiplies the values of `vec1' and `vec2' together and
02a517e4c492 *: initial commit
Paper <paper@paper.us.eu.org>
parents:
diff changeset
46 returns it
2
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
47
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
48 v[u]intAxB div(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
49 divides vec1 by the values in vec2. dividing by zero is
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
50 considered defined behavior and should result in a zero;
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
51 if this doesn't happen it's considered a bug
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
52
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
53 v[u]intAxB and(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
54 bitwise AND (&) of the values in both vectors
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
55
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
56 v[u]intAxB or(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
57 bitwise OR (|) of the values in both vectors
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
58
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
59 v[u]intAxB xor(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
60 bitwise XOR (^) of the values in both vectors
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
61
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
62 v[u]intAxB rshift(v[u]intAxB vec1, vuintAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
63 arithmetic right shift of the values in vec1 by
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
64 the corresponding values in vec2
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
65
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
66 v[u]intAxB lshift(v[u]intAxB vec1, vuintAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
67 arithmetic left shift of the values in vec1 by
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
68 the corresponding values in vec2
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
69
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
70 v[u]intAxB lrshift(v[u]intAxB vec1, vuintAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
71 logical right shift of the values in vec1 by
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
72 the corresponding values in vec2
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
73
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
74 v[u]intAxB avg(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
75 returns the average of the values in both vectors
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
76 i.e., div(mul(vec1, vec2), splat(2))
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
77
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
78 there are also a number of comparisons possible:
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
79
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
80 v[u]intAxB cmplt(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
81 turns on all bits of the corresponding value in
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
82 the result vector if the value in `vec1' is less
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
83 than the corresponding value in `vec2', else all
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
84 of the bits are turned off.
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
85
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
86 v[u]intAxB cmpgt(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
87 turns on all bits of the corresponding value in
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
88 the result vector if the value in `vec1' is greater
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
89 than the corresponding value in `vec2', else all
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
90 of the bits are turned off.
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
91
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
92 v[u]intAxB cmpeq(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
93 turns on all bits of the corresponding value in
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
94 the result vector if the value in `vec1' are equal
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
95 to the corresponding value in `vec2', else all
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
96 of the bits are turned off.
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
97
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
98 v[u]intAxB cmple(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
99 turns on all bits of the corresponding value in
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
100 the result vector if the value in `vec1' is less
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
101 than or equal to the corresponding value in `vec2',
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
102 else all of the bits are turned off.
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
103
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
104 v[u]intAxB cmpge(v[u]intAxB vec1, v[u]intAxB vec2)
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
105 turns on all bits of the corresponding value in
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
106 the result vector if the value in `vec1' is greater
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
107 than or equal to the corresponding value in `vec2',
f12b5dd4e18c *: many new operations and a real test suite
Paper <paper@tflc.us>
parents: 0
diff changeset
108 else all of the bits are turned off.
15
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
109
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
110 to initialize vec, you MUST call `vec_init()' when your programs starts up.
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
111
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
112 note that `vec_init()' is NOT thread-safe, and things can and will
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
113 blow up if you call it simultaneously from different threads (i.e. you
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
114 try to only initialize it when you need to... please just initialize
e05c257c6a23 *: huge refactor, add many new x86 intrinsics and the like
Paper <paper@tflc.us>
parents: 2
diff changeset
115 it on startup so you don't have to worry about that!!!)