npsimd 0.3.0

An ergonomic library for architecture-specific vectorization.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278

//! Intrinsics for SSSE3.

use super::*;

/// SSSE3 (Supplementary Streaming SIMD Extensions 3).
///
/// SSSE3 is the integer counterpart to SSE3: it adds useful functionality for
/// some integer operations, common and specialized, as well as a very crucial
/// data shuffling instruction, [`pshufb`].  It was introduced in 2006 with the
/// Intel Core microarchitecture.  If you're writing a vectorized algorithm by
/// hand, `pshufb` makes SSSE3 quite essential, even if you are just mapping on
/// array elements.
///
/// [`pshufb`]: Use::shuffle_u8x16
///
/// # Instructions
///
/// These instructions are implemented by [`Use`].
///
/// Common integer operations:
///
/// - [x] `pabsb`
///   - [`Use::abs_s8x16`]
/// - [x] `pabsd`
///   - [`Use::abs_s16x8`]
/// - [x] `pabsw`
///   - [`Use::abs_s32x4`]
///
/// - [x] `phaddw`
///   - [`Use::concat_and_reduce_add_u16x2x4`]
///   - [`Use::concat_and_reduce_add_s16x2x4`]
/// - [x] `phaddd`
///   - [`Use::concat_and_reduce_add_u32x2x2`]
///   - [`Use::concat_and_reduce_add_s32x2x2`]
///
/// - [x] `phaddsw`
///   - [`Use::concat_and_reduce_saturating_add_s16x2x4`]
///
/// - [x] `phsubw`
///   - [`Use::concat_and_reduce_sub_u16x2x4`]
///   - [`Use::concat_and_reduce_sub_s16x2x4`]
/// - [x] `phsubd`
///   - [`Use::concat_and_reduce_sub_u32x2x2`]
///   - [`Use::concat_and_reduce_sub_s32x2x2`]
///
/// - [x] `phsubsw`
///   - [`Use::concat_and_reduce_saturating_sub_s16x2x4`]
///
/// - [x] `psignb`
///   - [`Use::mul_sign_s8x16`]
/// - [x] `psignw`
///   - [`Use::mul_sign_s16x8`]
/// - [x] `psignd`
///   - [`Use::mul_sign_s32x4`]
///
/// Specialized integer operations:
///
/// - [x] `pmaddubsw`
///   - [`Use::sum_of_prod_u8x2x8`]
/// - [x] `pmulhrsw`
///   - [`Use::mul_and_halve_s16x8`]
///
/// Data shuffling:
///
/// - [x] `palignr`
///   - [`Use::align_elems_by_u8x16`]
///   - [`Use::align_elems_by_s8x16`]
///
/// - [x] `pshufb`
///   - [`Use::shuffle_u8x16`]
///   - [`Use::shuffle_s8x16`]
pub struct SSSE3(());

impl Feature<FeatureGroup> for SSSE3 {
    fn get_support(runtime: &RuntimeSupport) -> Option<Self> {
        runtime.ssse3().then_some(Self(()))
    }
}

/// SSSE3 intrinsics.
///
/// See [`SSSE3`] for information about this extension and which instructions
/// are supported here.
impl<FS> Use<FS>
where FS: HasFeature<FeatureGroup, SSSE3> {
    // Take the absolute value of integer elements in a 16-byte vector.
    defn_simd_shared!("ssse3", { simd_abs(x) }, {
        /// Absolute value of signed 8-bit integers.
        #[intrinsic_for("pabsb")]
        #[intel_equivalents("_mm_abs_epi8")]
        pub fn abs_s8x16(x: s8x16) -> s8x16;

        /// Absolute value of signed 16-bit integers.
        #[intrinsic_for("pabsw")]
        #[intel_equivalents("_mm_abs_epi16")]
        pub fn abs_s16x8(x: s16x8) -> s16x8;

        /// Absolute value of signed 32-bit integers.
        #[intrinsic_for("pabsd")]
        #[intel_equivalents("_mm_abs_epi32")]
        pub fn abs_s32x4(x: s32x4) -> s32x4;
    });

    // Horizontally add pairs of integers from two concatenated 16-byte vectors.
    defn_simd_llvm!("ssse3", {
        /// Horizontally add pairs of unsigned 16-bit integers from two vectors.
        #[intrinsic_for("phaddw")]
        #[intel_equivalents("_mm_hadd_epi16")]
        pub fn concat_and_reduce_add_u16x2x4(x: u16x8, y: u16x8) -> u16x8
            = "llvm.x86.ssse3.phadd.w.128";

        /// Horizontally add pairs of unsigned 32-bit integers from two vectors.
        #[intrinsic_for("phaddd")]
        #[intel_equivalents("_mm_hadd_epi32")]
        pub fn concat_and_reduce_add_u32x2x2(x: u32x4, y: u32x4) -> u32x4
            = "llvm.x86.ssse3.phadd.d.128";

        /// Horizontally add pairs of signed 16-bit integers from two vectors.
        #[intrinsic_for("phaddw")]
        #[intel_equivalents("_mm_hadd_epi16")]
        pub fn concat_and_reduce_add_s16x2x4(x: s16x8, y: s16x8) -> s16x8
            = "llvm.x86.ssse3.phadd.w.128";

        /// Horizontally add pairs of signed 32-bit integers from two vectors.
        #[intrinsic_for("phaddd")]
        #[intel_equivalents("_mm_hadd_epi32")]
        pub fn concat_and_reduce_add_s32x2x2(x: s32x4, y: s32x4) -> s32x4
            = "llvm.x86.ssse3.phadd.d.128";
    });

    // Horizontally add pairs of integers, with saturation, from two
    // concatenated 16-byte vectors.
    defn_simd_llvm!("ssse3", {
        /// Horizontally add pairs of signed 16-bit integers with saturation
        /// from two vectors.
        #[intrinsic_for("phaddsw")]
        #[intel_equivalents("_mm_hadds_epi16")]
        pub fn concat_and_reduce_saturating_add_s16x2x4
            (x: s16x8, y: s16x8) -> s16x8
            = "llvm.x86.ssse3.phadd.sw.128";
    });

    // Horizontally subtract pairs of integers from two concatenated 16-byte
    // vectors.
    defn_simd_llvm!("ssse3", {
        /// Horizontally subtract pairs of unsigned 16-bit integers from two
        /// vectors.
        #[intrinsic_for("phsubw")]
        #[intel_equivalents("_mm_hsub_epi16")]
        pub fn concat_and_reduce_sub_u16x2x4(x: u16x8, y: u16x8) -> u16x8
            = "llvm.x86.ssse3.phsub.w.128";

        /// Horizontally subtract pairs of unsigned 32-bit integers from two
        /// vectors.
        #[intrinsic_for("phsubd")]
        #[intel_equivalents("_mm_hsub_epi32")]
        pub fn concat_and_reduce_sub_u32x2x2(x: u32x4, y: u32x4) -> u32x4
            = "llvm.x86.ssse3.phsub.d.128";

        /// Horizontally subtract pairs of signed 16-bit integers from two
        /// vectors.
        #[intrinsic_for("phsubw")]
        #[intel_equivalents("_mm_hsub_epi16")]
        pub fn concat_and_reduce_sub_s16x2x4(x: s16x8, y: s16x8) -> s16x8
            = "llvm.x86.ssse3.phsub.w.128";

        /// Horizontally subtract pairs of signed 32-bit integers from two
        /// vectors.
        #[intrinsic_for("phsubd")]
        #[intel_equivalents("_mm_hsub_epi32")]
        pub fn concat_and_reduce_sub_s32x2x2(x: s32x4, y: s32x4) -> s32x4
            = "llvm.x86.ssse3.phsub.d.128";
    });

    // Horizontally subtract pairs of integers, with saturation, from two
    // concatenated 16-byte vectors.
    defn_simd_llvm!("ssse3", {
        /// Horizontally subtract pairs of signed 16-bit integers with
        /// saturation from two vectors.
        #[intrinsic_for("phsubsw")]
        #[intel_equivalents("_mm_hsubs_epi16")]
        pub fn concat_and_reduce_saturating_sub_s16x2x4
            (x: s16x8, y: s16x8) -> s16x8
            = "llvm.x86.ssse3.phsub.sw.128";
    });

    // Multiply integers from two 16-byte vectors and horizontally add pairs of
    // full-width products with saturation.
    defn_simd_llvm!("ssse3", {
        /// Multiply unsigned 8-bit integers, horizontally add pairs of 16-bit
        /// results, then saturate the sums to signed 16-bit integers.
        #[intrinsic_for("pmaddubsw")]
        #[intel_equivalents("_mm_maddubs_epi16")]
        pub fn sum_of_prod_u8x2x8
            (x: u16x8, y: u16x8) -> s16x8
            = "llvm.x86.ssse3.pmadd.ub.sw.128";
    });

    // Multiply integers from two 16-byte vectors and halve the results with
    // rounding.
    defn_simd_llvm!("ssse3", {
        /// Multiply signed 16-bit integers and halve the results with rounding.
        #[intrinsic_for("pmulhrsw")]
        #[intel_equivalents("_mm_mulhrs_epi16")]
        pub fn mul_and_halve_s16x8
            (x: s16x8, y: s16x8) -> s16x8
            = "llvm.x86.ssse3.pmul.hr.sw.128";
    });

    // Multiply integers in a 16-byte vector by corresponding signs.
    defn_simd_llvm!("ssse3", {
        /// Multiply each signed 8-bit integer by the sign of the corresponding
        /// integer.
        #[intrinsic_for("psignb")]
        #[intel_equivalents("_mm_sign_epi8")]
        pub fn mul_sign_s8x16(x: s8x16, y: s8x16) -> s8x16
            = "llvm.x86.ssse3.psign.b.128";

        /// Multiply each signed 16-bit integer by the sign of the corresponding
        /// integer.
        #[intrinsic_for("psignw")]
        #[intel_equivalents("_mm_sign_epi16")]
        pub fn mul_sign_s16x8(x: s16x8, y: s16x8) -> s16x8
            = "llvm.x86.ssse3.psign.w.128";

        /// Multiply each signed 16-bit integer by the sign of the corresponding
        /// integer.
        #[intrinsic_for("psignd")]
        #[intel_equivalents("_mm_sign_epi32")]
        pub fn mul_sign_s32x4(x: s32x4, y: s32x4) -> s32x4
            = "llvm.x86.ssse3.psign.d.128";
    });

    // Slice the concatenation of two 16-byte vectors.
    defn_simd_shared!("ssse3", fn(T, U) -> R {
        const_assert!(SHIFT < 16);
        simd_shuffle(x, y, const {
            simd_slice_indices::<R>(SHIFT as usize)
        })
    }, {
        /// Extract an unaligned vector between two contiguous vectors.
        ///
        /// The given vectors are concatenated in order and the given number of
        /// elements are truncated from its beginning.
        #[intrinsic_for("palignr")]
        #[intel_equivalents("_mm_alignr_epi8")]
        pub fn align_elems_by_u8x16<SHIFT: u8>(x: u8x16, y: u8x16) -> u8x16;

        /// Extract an unaligned vector between two contiguous vectors.
        ///
        /// The given vectors are concatenated in order and the given number of
        /// elements are truncated from its beginning.
        #[intrinsic_for("palignr")]
        #[intel_equivalents("_mm_alignr_epi8")]
        pub fn align_elems_by_s8x16<SHIFT: u8>(x: s8x16, y: s8x16) -> s8x16;
    });

    // Shuffle bytes within 16-byte vectors.
    defn_simd_llvm!("ssse3", {
        /// Index this array of unsigned 8-bit integers with the given vector.
        ///
        /// Each element of the index value is used modulo 16.  If an index is
        /// negative, then that position is filled with zero.
        #[intrinsic_for("pshufb")]
        #[intel_equivalents("_mm_shuffle_epi8")]
        pub fn shuffle_u8x16(x: u8x16, idxs: s8x16) -> u8x16
            = "llvm.x86.ssse3.pshuf.b.128";

        /// Index this array of signed 8-bit integers with the given vector.
        ///
        /// Each element of the index value is used modulo 16.  If an index is
        /// negative, then that position is filled with zero.
        #[intrinsic_for("pshufb")]
        #[intel_equivalents("_mm_shuffle_epi8")]
        pub fn shuffle_s8x16(x: s8x16, idxs: s8x16) -> s8x16
            = "llvm.x86.ssse3.pshuf.b.128";
    });
}