FD.io VPP  v16.09
Vector Packet Processing
vector_altivec.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2015 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 /*
16  Copyright (c) 2009 Eliot Dresselhaus
17 
18  Permission is hereby granted, free of charge, to any person obtaining
19  a copy of this software and associated documentation files (the
20  "Software"), to deal in the Software without restriction, including
21  without limitation the rights to use, copy, modify, merge, publish,
22  distribute, sublicense, and/or sell copies of the Software, and to
23  permit persons to whom the Software is furnished to do so, subject to
24  the following conditions:
25 
26  The above copyright notice and this permission notice shall be
27  included in all copies or substantial portions of the Software.
28 
29  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
33  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
34  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
35  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
36 */
37 
38 #ifndef included_vector_altivec_h
39 #define included_vector_altivec_h
40 
41 /* Splats. */
42 #define _(t,n,ti,fi,tr,fr) \
43  always_inline t##x##n t##x##n##_splat (t v) \
44  { return (t##x##n) __builtin_altivec_##fi ((ti) v); } \
45  \
46  always_inline t##x##n t##x##n##_splat_word (t##x##n x, int word_index) \
47  { return (t##x##n) __builtin_altivec_##fr ((tr) x, word_index); }
48 
49 #define u16x8_splat(i) ((u16x8) __builtin_altivec_vspltish (i))
50 #define i16x8_splat(i) ((i16x8) __builtin_altivec_vspltish (i))
51 #define u32x4_splat(i) ((u32x4) __builtin_altivec_vspltisw (i))
52 #define i32x4_splat(i) ((i32x4) __builtin_altivec_vspltisw (i))
53 
54 #define u16x8_splat_word(x,i) ((u16x8) __builtin_altivec_vsplth ((i16x8) (x), (i)))
55 #define i16x8_splat_word(x,i) ((i16x8) __builtin_altivec_vsplth ((i16x8) (x), (i)))
56 #define u32x4_splat_word(x,i) ((u32x4) __builtin_altivec_vspltw ((i32x4) (x), (i)))
57 #define i32x4_splat_word(x,i) ((i32x4) __builtin_altivec_vspltw ((i32x4) (x), (i)))
58 
59 #undef _
60 
61 /* 128 bit shifts. */
62 #define _(t,ti,lr,f) \
63  always_inline t t##_##lr (t x, t y) \
64  { return (t) __builtin_altivec_##f ((ti) x, (ti) y); } \
65  \
66  always_inline t t##_i##lr (t x, int i) \
67  { \
68  t j = {i,i,i,i}; \
69  return t##_##lr (x, j); \
70  }
71 
72 _(u16x8, i16x8, shift_left, vslh);
73 _(u32x4, i32x4, shift_left, vslw);
74 _(u16x8, i16x8, shift_right, vsrh);
75 _(u32x4, i32x4, shift_right, vsrw);
76 _(i16x8, i16x8, shift_right, vsrah);
77 _(i32x4, i32x4, shift_right, vsraw);
78 _(u16x8, i16x8, rotate_left, vrlh);
79 _(i16x8, i16x8, rotate_left, vrlh);
80 _(u32x4, i32x4, rotate_left, vrlw);
81 _(i32x4, i32x4, rotate_left, vrlw);
82 
83 #undef _
84 
85 #define _(t,it,lr,f) \
86  always_inline t t##_word_shift_##lr (t x, int n_words) \
87  { \
88  i32x4 n_bits = {0,0,0,n_words * BITS (it)}; \
89  return (t) __builtin_altivec_##f ((i32x4) x, n_bits); \
90  }
91 
92 _(u32x4, u32, left, vslo)
93 _(i32x4, i32, left, vslo)
94 _(u32x4, u32, right, vsro)
95 _(i32x4, i32, right, vsro)
96 _(u16x8, u16, left, vslo)
97 _(i16x8, i16, left, vslo)
98 _(u16x8, u16, right, vsro) _(i16x8, i16, right, vsro)
99 #undef _
101  u32
102  u32x4_get0 (u32x4 x)
103 {
104  u32x4_union_t y;
105  y.as_u32x4 = x;
106  return y.as_u32[3];
107 }
108 
109 /* Interleave. */
110 #define _(t,it,lh,f) \
111  always_inline t t##_interleave_##lh (t x, t y) \
112  { return (t) __builtin_altivec_##f ((it) x, (it) y); }
113 
114 _(u32x4, i32x4, lo, vmrglw)
115 _(i32x4, i32x4, lo, vmrglw)
116 _(u16x8, i16x8, lo, vmrglh)
117 _(i16x8, i16x8, lo, vmrglh)
118 _(u32x4, i32x4, hi, vmrghw)
119 _(i32x4, i32x4, hi, vmrghw)
120 _(u16x8, i16x8, hi, vmrghh) _(i16x8, i16x8, hi, vmrghh)
121 #undef _
122 /* Unaligned loads/stores. */
123 #ifndef __cplusplus
124 #define _(t) \
125  always_inline void t##_store_unaligned (t x, t * a) \
126  { clib_mem_unaligned (a, t) = x; } \
127  always_inline t t##_load_unaligned (t * a) \
128  { return clib_mem_unaligned (a, t); }
129  _(u8x16) _(u16x8) _(u32x4) _(u64x2) _(i8x16) _(i16x8) _(i32x4) _(i64x2)
130 #undef _
131 #endif
132 #define _signed_binop(n,m,f,g) \
133  /* Unsigned */ \
134  always_inline u##n##x##m \
135  u##n##x##m##_##f (u##n##x##m x, u##n##x##m y) \
136  { return (u##n##x##m) __builtin_altivec_##g ((i##n##x##m) x, (i##n##x##m) y); } \
137  \
138  /* Signed */ \
139  always_inline i##n##x##m \
140  i##n##x##m##_##f (i##n##x##m x, i##n##x##m y) \
141  { return (i##n##x##m) __builtin_altivec_##g ((i##n##x##m) x, (i##n##x##m) y); }
142 /* Compare operations. */
143  _signed_binop (16, 8, is_equal, vcmpequh)
144 _signed_binop (32, 4, is_equal, vcmpequw)
145 #undef _signed_binop
146  always_inline u16x8 u16x8_is_zero (u16x8 x)
147 {
148  u16x8 zero = { 0 };
149  return u16x8_is_equal (x, zero);
150 }
151 
152 always_inline u32x4
153 u32x4_is_zero (u32x4 x)
154 {
155  u32x4 zero = { 0 };
156  return u32x4_is_equal (x, zero);
157 }
158 
159 always_inline u32
161 {
162  u32x4 cmp = u32x4_is_zero (x);
163  u32x4 tmp = { 0x000f, 0x00f0, 0x0f00, 0xf000, };
164  cmp &= tmp;
165  cmp |= u32x4_word_shift_right (cmp, 2);
166  cmp |= u32x4_word_shift_right (cmp, 1);
167  return u32x4_get0 (cmp);
168 }
169 
170 #endif /* included_vector_altivec_h */
171 
172 /*
173  * fd.io coding-style-patch-verification: ON
174  *
175  * Local Variables:
176  * eval: (c-set-style "gnu")
177  * End:
178  */
static u32x4 u32x4_is_equal(u32x4 x, u32x4 y)
Definition: vector_sse2.h:504
wsllwi shift_right
static u16x8 u16x8_is_zero(u16x8 x)
Definition: vector_sse2.h:541
add_epi add_epi sub_epi sub_epi adds_epu subs_epu i16x8 y
Definition: vector_sse2.h:299
#define always_inline
Definition: clib.h:84
#define u32x4_word_shift_right(a, n)
Definition: vector_sse2.h:410
static u16x8 u16x8_is_equal(u16x8 x, u16x8 y)
Definition: vector_sse2.h:492
vslo vsro vslo static vsro u32 u32x4_get0(u32x4 x)
static u32x4 u32x4_is_zero(u32x4 x)
shift_left
static uword rotate_left(uword x, uword i)
Definition: bitops.h:134
static u32 u32x4_zero_byte_mask(u32x4 x)