1 | use crate::core_arch::x86::*; |
2 | |
3 | #[cfg (test)] |
4 | use stdarch_test::assert_instr; |
5 | |
6 | /// Multiply packed unsigned 52-bit integers in each 64-bit element of |
7 | /// `b` and `c` to form a 104-bit intermediate result. Add the high 52-bit |
8 | /// unsigned integer from the intermediate result with the |
9 | /// corresponding unsigned 64-bit integer in `a`, and store the |
10 | /// results in `dst`. |
11 | /// |
12 | /// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#avx512techs=AVX512IFMA52&expand=3488) |
13 | #[inline ] |
14 | #[target_feature (enable = "avx512ifma" )] |
15 | #[cfg_attr (test, assert_instr(vpmadd52huq))] |
16 | pub unsafe fn _mm512_madd52hi_epu64(a: __m512i, b: __m512i, c: __m512i) -> __m512i { |
17 | vpmadd52huq_512(z:a, x:b, y:c) |
18 | } |
19 | |
20 | /// Multiply packed unsigned 52-bit integers in each 64-bit element of |
21 | /// `b` and `c` to form a 104-bit intermediate result. Add the low 52-bit |
22 | /// unsigned integer from the intermediate result with the |
23 | /// corresponding unsigned 64-bit integer in `a`, and store the |
24 | /// results in `dst`. |
25 | /// |
26 | /// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=3497&avx512techs=AVX512IFMA52) |
27 | #[inline ] |
28 | #[target_feature (enable = "avx512ifma" )] |
29 | #[cfg_attr (test, assert_instr(vpmadd52luq))] |
30 | pub unsafe fn _mm512_madd52lo_epu64(a: __m512i, b: __m512i, c: __m512i) -> __m512i { |
31 | vpmadd52luq_512(z:a, x:b, y:c) |
32 | } |
33 | |
34 | /// Multiply packed unsigned 52-bit integers in each 64-bit element of |
35 | /// `b` and `c` to form a 104-bit intermediate result. Add the high 52-bit |
36 | /// unsigned integer from the intermediate result with the |
37 | /// corresponding unsigned 64-bit integer in `a`, and store the |
38 | /// results in `dst`. |
39 | /// |
40 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=vpmadd52&avx512techs=AVX512IFMA52,AVX512VL&expand=3485) |
41 | #[inline ] |
42 | #[target_feature (enable = "avx512ifma,avx512vl" )] |
43 | #[cfg_attr (test, assert_instr(vpmadd52huq))] |
44 | pub unsafe fn _mm256_madd52hi_epu64(a: __m256i, b: __m256i, c: __m256i) -> __m256i { |
45 | vpmadd52huq_256(z:a, x:b, y:c) |
46 | } |
47 | |
48 | /// Multiply packed unsigned 52-bit integers in each 64-bit element of |
49 | /// `b` and `c` to form a 104-bit intermediate result. Add the low 52-bit |
50 | /// unsigned integer from the intermediate result with the |
51 | /// corresponding unsigned 64-bit integer in `a`, and store the |
52 | /// results in `dst`. |
53 | /// |
54 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=vpmadd52&avx512techs=AVX512IFMA52,AVX512VL&expand=3494) |
55 | #[inline ] |
56 | #[target_feature (enable = "avx512ifma,avx512vl" )] |
57 | #[cfg_attr (test, assert_instr(vpmadd52luq))] |
58 | pub unsafe fn _mm256_madd52lo_epu64(a: __m256i, b: __m256i, c: __m256i) -> __m256i { |
59 | vpmadd52luq_256(z:a, x:b, y:c) |
60 | } |
61 | |
62 | /// Multiply packed unsigned 52-bit integers in each 64-bit element of |
63 | /// `b` and `c` to form a 104-bit intermediate result. Add the high 52-bit |
64 | /// unsigned integer from the intermediate result with the |
65 | /// corresponding unsigned 64-bit integer in `a`, and store the |
66 | /// results in `dst`. |
67 | /// |
68 | /// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=3488,3482&text=vpmadd52&avx512techs=AVX512IFMA52,AVX512VL) |
69 | #[inline ] |
70 | #[target_feature (enable = "avx512ifma,avx512vl" )] |
71 | #[cfg_attr (test, assert_instr(vpmadd52huq))] |
72 | pub unsafe fn _mm_madd52hi_epu64(a: __m128i, b: __m128i, c: __m128i) -> __m128i { |
73 | vpmadd52huq_128(z:a, x:b, y:c) |
74 | } |
75 | |
76 | /// Multiply packed unsigned 52-bit integers in each 64-bit element of |
77 | /// `b` and `c` to form a 104-bit intermediate result. Add the low 52-bit |
78 | /// unsigned integer from the intermediate result with the |
79 | /// corresponding unsigned 64-bit integer in `a`, and store the |
80 | /// results in `dst`. |
81 | /// |
82 | /// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=3488,3491&text=vpmadd52&avx512techs=AVX512IFMA52,AVX512VL) |
83 | #[inline ] |
84 | #[target_feature (enable = "avx512ifma,avx512vl" )] |
85 | #[cfg_attr (test, assert_instr(vpmadd52luq))] |
86 | pub unsafe fn _mm_madd52lo_epu64(a: __m128i, b: __m128i, c: __m128i) -> __m128i { |
87 | vpmadd52luq_128(z:a, x:b, y:c) |
88 | } |
89 | |
90 | #[allow (improper_ctypes)] |
91 | extern "C" { |
92 | #[link_name = "llvm.x86.avx512.vpmadd52l.uq.128" ] |
93 | fn vpmadd52luq_128(z: __m128i, x: __m128i, y: __m128i) -> __m128i; |
94 | #[link_name = "llvm.x86.avx512.vpmadd52h.uq.128" ] |
95 | fn vpmadd52huq_128(z: __m128i, x: __m128i, y: __m128i) -> __m128i; |
96 | #[link_name = "llvm.x86.avx512.vpmadd52l.uq.256" ] |
97 | fn vpmadd52luq_256(z: __m256i, x: __m256i, y: __m256i) -> __m256i; |
98 | #[link_name = "llvm.x86.avx512.vpmadd52h.uq.256" ] |
99 | fn vpmadd52huq_256(z: __m256i, x: __m256i, y: __m256i) -> __m256i; |
100 | #[link_name = "llvm.x86.avx512.vpmadd52l.uq.512" ] |
101 | fn vpmadd52luq_512(z: __m512i, x: __m512i, y: __m512i) -> __m512i; |
102 | #[link_name = "llvm.x86.avx512.vpmadd52h.uq.512" ] |
103 | fn vpmadd52huq_512(z: __m512i, x: __m512i, y: __m512i) -> __m512i; |
104 | } |
105 | |
106 | #[cfg (test)] |
107 | mod tests { |
108 | |
109 | use stdarch_test::simd_test; |
110 | |
111 | use crate::core_arch::x86::*; |
112 | |
113 | #[simd_test(enable = "avx512ifma" )] |
114 | unsafe fn test_mm512_madd52hi_epu64() { |
115 | let mut a = _mm512_set1_epi64(10 << 40); |
116 | let b = _mm512_set1_epi64((11 << 40) + 4); |
117 | let c = _mm512_set1_epi64((12 << 40) + 3); |
118 | |
119 | a = _mm512_madd52hi_epu64(a, b, c); |
120 | |
121 | // (10 << 40) + ((((11 << 40) + 4) * ((12 << 40) + 3)) >> 52) |
122 | let expected = _mm512_set1_epi64(11030549757952); |
123 | |
124 | assert_eq_m512i(a, expected); |
125 | } |
126 | |
127 | #[simd_test(enable = "avx512ifma" )] |
128 | unsafe fn test_mm512_madd52lo_epu64() { |
129 | let mut a = _mm512_set1_epi64(10 << 40); |
130 | let b = _mm512_set1_epi64((11 << 40) + 4); |
131 | let c = _mm512_set1_epi64((12 << 40) + 3); |
132 | |
133 | a = _mm512_madd52lo_epu64(a, b, c); |
134 | |
135 | // (10 << 40) + ((((11 << 40) + 4) * ((12 << 40) + 3)) % (1 << 52)) |
136 | let expected = _mm512_set1_epi64(100055558127628); |
137 | |
138 | assert_eq_m512i(a, expected); |
139 | } |
140 | |
141 | #[simd_test(enable = "avx512ifma,avx512vl" )] |
142 | unsafe fn test_mm256_madd52hi_epu64() { |
143 | let mut a = _mm256_set1_epi64x(10 << 40); |
144 | let b = _mm256_set1_epi64x((11 << 40) + 4); |
145 | let c = _mm256_set1_epi64x((12 << 40) + 3); |
146 | |
147 | a = _mm256_madd52hi_epu64(a, b, c); |
148 | |
149 | // (10 << 40) + ((((11 << 40) + 4) * ((12 << 40) + 3)) >> 52) |
150 | let expected = _mm256_set1_epi64x(11030549757952); |
151 | |
152 | assert_eq_m256i(a, expected); |
153 | } |
154 | |
155 | #[simd_test(enable = "avx512ifma,avx512vl" )] |
156 | unsafe fn test_mm256_madd52lo_epu64() { |
157 | let mut a = _mm256_set1_epi64x(10 << 40); |
158 | let b = _mm256_set1_epi64x((11 << 40) + 4); |
159 | let c = _mm256_set1_epi64x((12 << 40) + 3); |
160 | |
161 | a = _mm256_madd52lo_epu64(a, b, c); |
162 | |
163 | // (10 << 40) + ((((11 << 40) + 4) * ((12 << 40) + 3)) % (1 << 52)) |
164 | let expected = _mm256_set1_epi64x(100055558127628); |
165 | |
166 | assert_eq_m256i(a, expected); |
167 | } |
168 | |
169 | #[simd_test(enable = "avx512ifma,avx512vl" )] |
170 | unsafe fn test_mm_madd52hi_epu64() { |
171 | let mut a = _mm_set1_epi64x(10 << 40); |
172 | let b = _mm_set1_epi64x((11 << 40) + 4); |
173 | let c = _mm_set1_epi64x((12 << 40) + 3); |
174 | |
175 | a = _mm_madd52hi_epu64(a, b, c); |
176 | |
177 | // (10 << 40) + ((((11 << 40) + 4) * ((12 << 40) + 3)) >> 52) |
178 | let expected = _mm_set1_epi64x(11030549757952); |
179 | |
180 | assert_eq_m128i(a, expected); |
181 | } |
182 | |
183 | #[simd_test(enable = "avx512ifma,avx512vl" )] |
184 | unsafe fn test_mm_madd52lo_epu64() { |
185 | let mut a = _mm_set1_epi64x(10 << 40); |
186 | let b = _mm_set1_epi64x((11 << 40) + 4); |
187 | let c = _mm_set1_epi64x((12 << 40) + 3); |
188 | |
189 | a = _mm_madd52hi_epu64(a, b, c); |
190 | |
191 | // (10 << 40) + ((((11 << 40) + 4) * ((12 << 40) + 3)) >> 52) |
192 | let expected = _mm_set1_epi64x(11030549757952); |
193 | |
194 | assert_eq_m128i(a, expected); |
195 | } |
196 | } |
197 | |