1 | //! `x86_64` Streaming SIMD Extensions (SSE) |
2 | |
3 | use crate::core_arch::x86::*; |
4 | |
5 | #[cfg (test)] |
6 | use stdarch_test::assert_instr; |
7 | |
8 | #[allow (improper_ctypes)] |
9 | extern "C" { |
10 | #[link_name = "llvm.x86.sse.cvtss2si64" ] |
11 | fn cvtss2si64(a: __m128) -> i64; |
12 | #[link_name = "llvm.x86.sse.cvttss2si64" ] |
13 | fn cvttss2si64(a: __m128) -> i64; |
14 | #[link_name = "llvm.x86.sse.cvtsi642ss" ] |
15 | fn cvtsi642ss(a: __m128, b: i64) -> __m128; |
16 | } |
17 | |
18 | /// Converts the lowest 32 bit float in the input vector to a 64 bit integer. |
19 | /// |
20 | /// The result is rounded according to the current rounding mode. If the result |
21 | /// cannot be represented as a 64 bit integer the result will be |
22 | /// `0x8000_0000_0000_0000` (`i64::MIN`) or trigger an invalid operation |
23 | /// floating point exception if unmasked (see |
24 | /// [`_mm_setcsr`](fn._mm_setcsr.html)). |
25 | /// |
26 | /// This corresponds to the `CVTSS2SI` instruction (with 64 bit output). |
27 | /// |
28 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtss_si64) |
29 | #[inline ] |
30 | #[target_feature (enable = "sse" )] |
31 | #[cfg_attr (test, assert_instr(cvtss2si))] |
32 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
33 | pub unsafe fn _mm_cvtss_si64(a: __m128) -> i64 { |
34 | cvtss2si64(a) |
35 | } |
36 | |
37 | /// Converts the lowest 32 bit float in the input vector to a 64 bit integer |
38 | /// with truncation. |
39 | /// |
40 | /// The result is rounded always using truncation (round towards zero). If the |
41 | /// result cannot be represented as a 64 bit integer the result will be |
42 | /// `0x8000_0000_0000_0000` (`i64::MIN`) or an invalid operation floating |
43 | /// point exception if unmasked (see [`_mm_setcsr`](fn._mm_setcsr.html)). |
44 | /// |
45 | /// This corresponds to the `CVTTSS2SI` instruction (with 64 bit output). |
46 | /// |
47 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvttss_si64) |
48 | #[inline ] |
49 | #[target_feature (enable = "sse" )] |
50 | #[cfg_attr (test, assert_instr(cvttss2si))] |
51 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
52 | pub unsafe fn _mm_cvttss_si64(a: __m128) -> i64 { |
53 | cvttss2si64(a) |
54 | } |
55 | |
56 | /// Converts a 64 bit integer to a 32 bit float. The result vector is the input |
57 | /// vector `a` with the lowest 32 bit float replaced by the converted integer. |
58 | /// |
59 | /// This intrinsic corresponds to the `CVTSI2SS` instruction (with 64 bit |
60 | /// input). |
61 | /// |
62 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi64_ss) |
63 | #[inline ] |
64 | #[target_feature (enable = "sse" )] |
65 | #[cfg_attr (test, assert_instr(cvtsi2ss))] |
66 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
67 | pub unsafe fn _mm_cvtsi64_ss(a: __m128, b: i64) -> __m128 { |
68 | cvtsi642ss(a, b) |
69 | } |
70 | |
71 | #[cfg (test)] |
72 | mod tests { |
73 | use crate::core_arch::arch::x86_64::*; |
74 | use stdarch_test::simd_test; |
75 | |
76 | #[simd_test(enable = "sse" )] |
77 | unsafe fn test_mm_cvtss_si64() { |
78 | let inputs = &[ |
79 | (42.0f32, 42i64), |
80 | (-31.4, -31), |
81 | (-33.5, -34), |
82 | (-34.5, -34), |
83 | (4.0e10, 40_000_000_000), |
84 | (4.0e-10, 0), |
85 | (f32::NAN, i64::MIN), |
86 | (2147483500.1, 2147483520), |
87 | (9.223371e18, 9223370937343148032), |
88 | ]; |
89 | for (i, &(xi, e)) in inputs.iter().enumerate() { |
90 | let x = _mm_setr_ps(xi, 1.0, 3.0, 4.0); |
91 | let r = _mm_cvtss_si64(x); |
92 | assert_eq!( |
93 | e, r, |
94 | "TestCase # {} _mm_cvtss_si64( {:?}) = {}, expected: {}" , |
95 | i, x, r, e |
96 | ); |
97 | } |
98 | } |
99 | |
100 | #[simd_test(enable = "sse" )] |
101 | unsafe fn test_mm_cvttss_si64() { |
102 | let inputs = &[ |
103 | (42.0f32, 42i64), |
104 | (-31.4, -31), |
105 | (-33.5, -33), |
106 | (-34.5, -34), |
107 | (10.999, 10), |
108 | (-5.99, -5), |
109 | (4.0e10, 40_000_000_000), |
110 | (4.0e-10, 0), |
111 | (f32::NAN, i64::MIN), |
112 | (2147483500.1, 2147483520), |
113 | (9.223371e18, 9223370937343148032), |
114 | (9.223372e18, i64::MIN), |
115 | ]; |
116 | for (i, &(xi, e)) in inputs.iter().enumerate() { |
117 | let x = _mm_setr_ps(xi, 1.0, 3.0, 4.0); |
118 | let r = _mm_cvttss_si64(x); |
119 | assert_eq!( |
120 | e, r, |
121 | "TestCase # {} _mm_cvttss_si64( {:?}) = {}, expected: {}" , |
122 | i, x, r, e |
123 | ); |
124 | } |
125 | } |
126 | |
127 | #[simd_test(enable = "sse" )] |
128 | unsafe fn test_mm_cvtsi64_ss() { |
129 | let inputs = &[ |
130 | (4555i64, 4555.0f32), |
131 | (322223333, 322223330.0), |
132 | (-432, -432.0), |
133 | (-322223333, -322223330.0), |
134 | (9223372036854775807, 9.223372e18), |
135 | (-9223372036854775808, -9.223372e18), |
136 | ]; |
137 | |
138 | for &(x, f) in inputs { |
139 | let a = _mm_setr_ps(5.0, 6.0, 7.0, 8.0); |
140 | let r = _mm_cvtsi64_ss(a, x); |
141 | let e = _mm_setr_ps(f, 6.0, 7.0, 8.0); |
142 | assert_eq_m128(e, r); |
143 | } |
144 | } |
145 | } |
146 | |