simdutf8/implementation/x86/
sse42.rs1#[cfg(target_arch = "x86")]
4use core::arch::x86::{
5 __m128i, _mm_alignr_epi8, _mm_and_si128, _mm_loadu_si128, _mm_movemask_epi8, _mm_or_si128,
6 _mm_prefetch, _mm_set1_epi8, _mm_setr_epi8, _mm_setzero_si128, _mm_shuffle_epi8,
7 _mm_srli_epi16, _mm_subs_epu8, _mm_testz_si128, _mm_xor_si128, _MM_HINT_T0,
8};
9#[cfg(target_arch = "x86_64")]
10use core::arch::x86_64::{
11 __m128i, _mm_alignr_epi8, _mm_and_si128, _mm_loadu_si128, _mm_movemask_epi8, _mm_or_si128,
12 _mm_prefetch, _mm_set1_epi8, _mm_setr_epi8, _mm_setzero_si128, _mm_shuffle_epi8,
13 _mm_srli_epi16, _mm_subs_epu8, _mm_testz_si128, _mm_xor_si128, _MM_HINT_T0,
14};
15
16use crate::implementation::helpers::Utf8CheckAlgorithm;
17
18type SimdU8Value = crate::implementation::helpers::SimdU8Value<__m128i>;
21
22impl SimdU8Value {
23 #[allow(clippy::too_many_arguments)]
24 #[allow(clippy::cast_possible_wrap)]
25 #[target_feature(enable = "sse4.2")]
26 #[inline]
27 unsafe fn from_32_cut_off_leading(
28 _v0: u8,
29 _v1: u8,
30 _v2: u8,
31 _v3: u8,
32 _v4: u8,
33 _v5: u8,
34 _v6: u8,
35 _v7: u8,
36 _v8: u8,
37 _v9: u8,
38 _v10: u8,
39 _v11: u8,
40 _v12: u8,
41 _v13: u8,
42 _v14: u8,
43 _v15: u8,
44 v16: u8,
45 v17: u8,
46 v18: u8,
47 v19: u8,
48 v20: u8,
49 v21: u8,
50 v22: u8,
51 v23: u8,
52 v24: u8,
53 v25: u8,
54 v26: u8,
55 v27: u8,
56 v28: u8,
57 v29: u8,
58 v30: u8,
59 v31: u8,
60 ) -> Self {
61 Self::from(_mm_setr_epi8(
62 v16 as i8, v17 as i8, v18 as i8, v19 as i8, v20 as i8, v21 as i8, v22 as i8, v23 as i8,
63 v24 as i8, v25 as i8, v26 as i8, v27 as i8, v28 as i8, v29 as i8, v30 as i8, v31 as i8,
64 ))
65 }
66
67 #[allow(clippy::too_many_arguments)]
68 #[allow(clippy::cast_possible_wrap)]
69 #[target_feature(enable = "sse4.2")]
70 #[inline]
71 unsafe fn repeat_16(
72 v0: u8,
73 v1: u8,
74 v2: u8,
75 v3: u8,
76 v4: u8,
77 v5: u8,
78 v6: u8,
79 v7: u8,
80 v8: u8,
81 v9: u8,
82 v10: u8,
83 v11: u8,
84 v12: u8,
85 v13: u8,
86 v14: u8,
87 v15: u8,
88 ) -> Self {
89 Self::from(_mm_setr_epi8(
90 v0 as i8, v1 as i8, v2 as i8, v3 as i8, v4 as i8, v5 as i8, v6 as i8, v7 as i8,
91 v8 as i8, v9 as i8, v10 as i8, v11 as i8, v12 as i8, v13 as i8, v14 as i8, v15 as i8,
92 ))
93 }
94
95 #[allow(clippy::cast_ptr_alignment)]
96 #[target_feature(enable = "sse4.2")]
97 #[inline]
98 unsafe fn load_from(ptr: *const u8) -> Self {
99 Self::from(_mm_loadu_si128(ptr.cast::<__m128i>()))
100 }
101
102 #[target_feature(enable = "sse4.2")]
103 #[inline]
104 unsafe fn lookup_16(self, tbl: Self) -> Self {
105 Self::from(_mm_shuffle_epi8(tbl.0, self.0))
106 }
107
108 #[allow(clippy::cast_possible_wrap)]
109 #[target_feature(enable = "sse4.2")]
110 #[inline]
111 unsafe fn splat(val: u8) -> Self {
112 Self::from(_mm_set1_epi8(val as i8))
113 }
114
115 #[target_feature(enable = "sse4.2")]
116 #[inline]
117 unsafe fn splat0() -> Self {
118 Self::from(_mm_setzero_si128())
119 }
120
121 #[target_feature(enable = "sse4.2")]
122 #[inline]
123 unsafe fn or(self, b: Self) -> Self {
124 Self::from(_mm_or_si128(self.0, b.0))
125 }
126
127 #[target_feature(enable = "sse4.2")]
128 #[inline]
129 unsafe fn and(self, b: Self) -> Self {
130 Self::from(_mm_and_si128(self.0, b.0))
131 }
132
133 #[target_feature(enable = "sse4.2")]
134 #[inline]
135 unsafe fn xor(self, b: Self) -> Self {
136 Self::from(_mm_xor_si128(self.0, b.0))
137 }
138
139 #[target_feature(enable = "sse4.2")]
140 #[inline]
141 unsafe fn saturating_sub(self, b: Self) -> Self {
142 Self::from(_mm_subs_epu8(self.0, b.0))
143 }
144
145 #[target_feature(enable = "sse4.2")]
147 #[inline]
148 unsafe fn shr4(self) -> Self {
149 Self::from(_mm_srli_epi16(self.0, 4)).and(Self::splat(0xFF >> 4))
150 }
151
152 #[target_feature(enable = "sse4.2")]
154 #[inline]
155 unsafe fn prev1(self, prev: Self) -> Self {
156 Self::from(_mm_alignr_epi8(self.0, prev.0, 16 - 1))
157 }
158
159 #[target_feature(enable = "sse4.2")]
161 #[inline]
162 unsafe fn prev2(self, prev: Self) -> Self {
163 Self::from(_mm_alignr_epi8(self.0, prev.0, 16 - 2))
164 }
165
166 #[target_feature(enable = "sse4.2")]
168 #[inline]
169 unsafe fn prev3(self, prev: Self) -> Self {
170 Self::from(_mm_alignr_epi8(self.0, prev.0, 16 - 3))
171 }
172
173 #[target_feature(enable = "sse4.2")]
174 #[inline]
175 unsafe fn any_bit_set(self) -> bool {
176 _mm_testz_si128(self.0, self.0) != 1
177 }
178
179 #[target_feature(enable = "sse4.2")]
180 #[inline]
181 unsafe fn is_ascii(self) -> bool {
182 _mm_movemask_epi8(self.0) == 0
183 }
184}
185
186impl From<__m128i> for SimdU8Value {
187 #[inline]
188 fn from(val: __m128i) -> Self {
189 Self(val)
190 }
191}
192
193#[target_feature(enable = "sse4.2")]
194#[inline]
195unsafe fn simd_prefetch(ptr: *const u8) {
196 _mm_prefetch(ptr.cast::<i8>(), _MM_HINT_T0);
197}
198
199const PREFETCH: bool = false;
200use crate::implementation::helpers::TempSimdChunkA16 as TempSimdChunk;
201simd_input_128_bit!(#[target_feature(enable = "sse4.2")]);
202algorithm_simd!(#[target_feature(enable = "sse4.2")]);
203algorithm_simd_default_special_case_fns!(#[target_feature(enable = "sse4.2")]);