Mercurial > vec

comparison src/impl/x86/avx512bw.c @ 28:c6c99ab1088a

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
*: add min/max functions and a big big refactor (again) agh, this time I added a few more implementations (and generally made the code just a little faster...)
author Paper <paper@tflc.us>
date Thu, 24 Apr 2025 00:54:02 -0400
parents
children
comparison
equal deleted inserted replaced
27:d00b95f95dd1 28:c6c99ab1088a
1 /**
2 * vec - a tiny SIMD vector library in C99
3 *
4 * Copyright (c) 2024 Paper
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 **/
24
25 #include "vec/impl/x86/avx512bw.h"
26
27 #include <immintrin.h>
28
29 /* ------------------------------------------------------------------------ */
30
31 #define VEC_AVX512BW_MINMAX_TEMPLATE(SIGN, BITS, SIZE, INTLSIGN, OP) \
32 VEC_FUNC_IMPL v##SIGN##int##BITS##x##SIZE v##SIGN##int##BITS##x##SIZE##_avx512bw_##OP(v##SIGN##int##BITS##x##SIZE vec1, v##SIGN##int##BITS##x##SIZE vec2) \
33 { \
34 union v##SIGN##int##BITS##x##SIZE##_impl_data *vec1d = (union v##SIGN##int##BITS##x##SIZE##_impl_data *)&vec1; \
35 union v##SIGN##int##BITS##x##SIZE##_impl_data *vec2d = (union v##SIGN##int##BITS##x##SIZE##_impl_data *)&vec2; \
36 \
37 vec1d->avx512bw = _mm512_##OP##_ep##INTLSIGN##BITS(vec1d->avx512bw, vec2d->avx512bw); \
38 \
39 return vec1d->vec; \
40 }
41
42 #define VEC_AVX512BW_MINMAX_8x64(OP) VEC_AVX512BW_MINMAX_TEMPLATE( , 8, 64, i, OP)
43 #define VEC_AVX512BW_MINMAX_u8x64(OP) VEC_AVX512BW_MINMAX_TEMPLATE(u, 8, 64, u, OP)
44 #define VEC_AVX512BW_MINMAX_16x32(OP) VEC_AVX512BW_MINMAX_TEMPLATE( , 16, 32, i, OP)
45 #define VEC_AVX512BW_MINMAX_u16x32(OP) VEC_AVX512BW_MINMAX_TEMPLATE(u, 16, 32, u, OP)
46
47 #define VEC_AVX512BW_STRUCT_MINMAX_8x64(OP, SIGN) v##SIGN##int8x64_avx512bw_##OP
48 #define VEC_AVX512BW_STRUCT_MINMAX_16x32(OP, SIGN) v##SIGN##int16x32_avx512bw_##OP
49
50 /* ------------------------------------------------------------------------ */
51
52 #define VEC_AVX512BW_OPERATION_EX(name, op, sign, bits, size, secondsign) \
53 VEC_FUNC_IMPL v##sign##int##bits##x##size v##sign##int##bits##x##size##_avx512bw_##name(v##sign##int##bits##x##size vec1, v##secondsign##int##bits##x##size vec2) \
54 { \
55 union v##sign##int##bits##x##size##_impl_data *vec1d = (union v##sign##int##bits##x##size##_impl_data *)&vec1; \
56 union v##secondsign##int##bits##x##size##_impl_data *vec2d = (union v##secondsign##int##bits##x##size##_impl_data *)&vec2; \
57 \
58 vec1d->avx512bw = _mm512_##op##_epi##bits(vec1d->avx512bw, vec2d->avx512bw); \
59 \
60 return vec1d->vec; \
61 }
62
63 #define VEC_AVX512BW_OPERATION(name, op, sign, bits, size) \
64 VEC_AVX512BW_OPERATION_EX(name, op, sign, bits, size, sign)
65
66 #define VEC_AVX512BW_OPERATION_SHIFT(name, op, sign, bits, size) \
67 VEC_AVX512BW_OPERATION_EX(name, op, sign, bits, size, u)
68
69 #define VEC_AVX512BW_ADD_8x64(sign) VEC_AVX512BW_OPERATION(add, add, sign, 8, 64)
70 #define VEC_AVX512BW_ADD_16x32(sign) VEC_AVX512BW_OPERATION(add, add, sign, 16, 32)
71
72 #define VEC_AVX512BW_SUB_8x64(sign) VEC_AVX512BW_OPERATION(sub, sub, sign, 8, 64)
73 #define VEC_AVX512BW_SUB_16x32(sign) VEC_AVX512BW_OPERATION(sub, sub, sign, 16, 32)
74
75 #define VEC_AVX512BW_MUL_8x64(sign) /* nothing */
76 #define VEC_AVX512BW_MUL_16x32(sign) VEC_AVX512BW_OPERATION(mul, mullo, sign, 16, 32)
77
78 #define VEC_AVX512BW_LSHIFT_8x64(sign) /* nothing */
79 #define VEC_AVX512BW_LSHIFT_16x32(sign) VEC_AVX512BW_OPERATION_SHIFT(lshift, sllv, sign, 16, 32)
80
81 #define VEC_AVX512BW_XRSHIFT(name, bits, size, sign, aORl) \
82 VEC_AVX512BW_OPERATION_SHIFT(name, sr##aORl##v, sign, bits, size)
83
84 /* always the same, regardless of signedness */
85 #define VEC_AVX512BW_LRSHIFT_8x64(sign) /* nothing */
86 #define VEC_AVX512BW_LRSHIFT_16x32(sign) VEC_AVX512BW_XRSHIFT(lrshift, 16, 32, sign, l)
87
88 #define VEC_AVX512BW_RSHIFT_8x64(sign) /* nothing */
89 #define VEC_AVX512BW_RSHIFT_16x32(sign) VEC_AVX512BW_XRSHIFT(rshift, 16, 32, sign, a)
90
91 #define VEC_AVX512BW_uRSHIFT_8x64(sign) /* nothing */
92 #define VEC_AVX512BW_uRSHIFT_16x32(sign) VEC_AVX512BW_XRSHIFT(rshift, 16, 32, sign, l)
93
94 /* ok */
95 #define VEC_AVX512BW_STRUCT_ADDSUB_8x64(OP, SIGN) v##SIGN##int8x64_avx512bw_##OP
96 #define VEC_AVX512BW_STRUCT_ADDSUB_16x32(OP, SIGN) v##SIGN##int16x32_avx512bw_##OP
97
98 #define VEC_AVX512BW_STRUCT_OPERATION_8x64(OP, SIGN) NULL
99 #define VEC_AVX512BW_STRUCT_OPERATION_16x32(OP, SIGN) v##SIGN##int16x32_avx512bw_##OP
100
101 /* ------------------------------------------------------------------------ */
102
103 #define VEC_AVX512BW_DEFINE_OPERATIONS_SIGN(sign, bits, size) \
104 union v##sign##int##bits##x##size##_impl_data { \
105 v##sign##int##bits##x##size vec; \
106 __m512i avx512bw; \
107 }; \
108 \
109 VEC_STATIC_ASSERT(VEC_ALIGNOF(__m512i) <= VEC_ALIGNOF(v##sign##int##bits##x##size), "vec: v" #sign "int" #bits "x" #size " alignment needs to be expanded to fit intrinsic type size"); \
110 VEC_STATIC_ASSERT(sizeof(__m512i) <= sizeof(v##sign##int##bits##x##size), "vec: v" #sign "int" #bits "x" #size " needs to be expanded to fit intrinsic type size"); \
111 \
112 VEC_AVX512BW_ADD_##bits##x##size(sign) \
113 VEC_AVX512BW_SUB_##bits##x##size(sign) \
114 VEC_AVX512BW_MUL_##bits##x##size(sign) \
115 \
116 VEC_AVX512BW_LSHIFT_##bits##x##size(sign); \
117 VEC_AVX512BW_##sign##RSHIFT_##bits##x##size(sign); \
118 VEC_AVX512BW_LRSHIFT_##bits##x##size(sign); \
119 \
120 VEC_AVX512BW_MINMAX_##sign##bits##x##size(min) \
121 VEC_AVX512BW_MINMAX_##sign##bits##x##size(max) \
122 \
123 const v##sign##int##bits##x##size##_impl v##sign##int##bits##x##size##_impl_avx512bw = { \
124 .add = VEC_AVX512BW_STRUCT_ADDSUB_##bits##x##size(add, sign), \
125 .sub = VEC_AVX512BW_STRUCT_ADDSUB_##bits##x##size(sub, sign), \
126 .mul = VEC_AVX512BW_STRUCT_OPERATION_##bits##x##size(mul, sign), \
127 .lshift = VEC_AVX512BW_STRUCT_OPERATION_##bits##x##size(lshift, sign), \
128 .rshift = VEC_AVX512BW_STRUCT_OPERATION_##bits##x##size(rshift, sign), \
129 .lrshift = VEC_AVX512BW_STRUCT_OPERATION_##bits##x##size(lrshift, sign), \
130 .min = VEC_AVX512BW_STRUCT_MINMAX_##bits##x##size(min, sign), \
131 .max = VEC_AVX512BW_STRUCT_MINMAX_##bits##x##size(max, sign), \
132 };
133
134 #define VEC_AVX512BW_DEFINE_OPERATIONS(bits, size) \
135 VEC_AVX512BW_DEFINE_OPERATIONS_SIGN(u, bits, size) \
136 VEC_AVX512BW_DEFINE_OPERATIONS_SIGN( , bits, size)
137
138 VEC_AVX512BW_DEFINE_OPERATIONS(8, 64)
139 VEC_AVX512BW_DEFINE_OPERATIONS(16, 32)