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
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
|
#![allow(dead_code)]
use libc::{c_char, c_double, c_float, c_int, c_long, c_ulong};
mod ffi {
use super::*;
extern "C" {
// It would be nice to be able to switch to the NEWLAPACK version of the function but this
// seems to trigger some link error. Available function names can be seen here:
// /Library/Developer/CommandLineTools/SDKs/MacOSX13.3.sdk/System/Library/Frameworks/Accelerate.framework/Versions/A/Accelerate.tbd
#[link_name = "sgemm_"]
pub fn sgemm_ffi(
transa: *const c_char,
transb: *const c_char,
m: *const c_int,
n: *const c_int,
k: *const c_int,
alpha: *const c_float,
a: *const c_float,
lda: *const c_int,
b: *const c_float,
ldb: *const c_int,
beta: *const c_float,
c: *mut c_float,
ldc: *const c_int,
);
#[link_name = "dgemm_"]
pub fn dgemm_ffi(
transa: *const c_char,
transb: *const c_char,
m: *const c_int,
n: *const c_int,
k: *const c_int,
alpha: *const c_double,
a: *const c_double,
lda: *const c_int,
b: *const c_double,
ldb: *const c_int,
beta: *const c_double,
c: *mut c_double,
ldc: *const c_int,
);
pub fn vvexpf(dst: *mut c_float, src: *const c_float, len: *const c_int);
pub fn vvexp(dst: *mut c_double, src: *const c_double, len: *const c_int);
pub fn vvsqrtf(dst: *mut c_float, src: *const c_float, len: *const c_int);
pub fn vvsqrt(dst: *mut c_double, src: *const c_double, len: *const c_int);
pub fn vvsinf(dst: *mut c_float, src: *const c_float, len: *const c_int);
pub fn vvsin(dst: *mut c_double, src: *const c_double, len: *const c_int);
pub fn vvcosf(dst: *mut c_float, src: *const c_float, len: *const c_int);
pub fn vvcos(dst: *mut c_double, src: *const c_double, len: *const c_int);
pub fn vvlogf(dst: *mut c_float, src: *const c_float, len: *const c_int);
pub fn vvlog(dst: *mut c_double, src: *const c_double, len: *const c_int);
pub fn vDSP_vaddD(
_: *const c_double,
_: c_long,
_: *const c_double,
_: c_long,
_: *mut c_double,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vadd(
_: *const c_float,
_: c_long,
_: *const c_float,
_: c_long,
_: *mut c_float,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vsubD(
_: *const c_double,
_: c_long,
_: *const c_double,
_: c_long,
_: *mut c_double,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vsub(
_: *const c_float,
_: c_long,
_: *const c_float,
_: c_long,
_: *mut c_float,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vmulD(
_: *const c_double,
_: c_long,
_: *const c_double,
_: c_long,
_: *mut c_double,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vmul(
_: *const c_float,
_: c_long,
_: *const c_float,
_: c_long,
_: *mut c_float,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vdivD(
_: *const c_double,
_: c_long,
_: *const c_double,
_: c_long,
_: *mut c_double,
_: c_long,
_: c_ulong,
);
pub fn vDSP_vdiv(
_: *const c_float,
_: c_long,
_: *const c_float,
_: c_long,
_: *mut c_float,
_: c_long,
_: c_ulong,
);
}
}
#[allow(clippy::too_many_arguments)]
#[inline]
pub unsafe fn sgemm(
transa: u8,
transb: u8,
m: i32,
n: i32,
k: i32,
alpha: f32,
a: &[f32],
lda: i32,
b: &[f32],
ldb: i32,
beta: f32,
c: &mut [f32],
ldc: i32,
) {
ffi::sgemm_ffi(
&(transa as c_char),
&(transb as c_char),
&m,
&n,
&k,
&alpha,
a.as_ptr(),
&lda,
b.as_ptr(),
&ldb,
&beta,
c.as_mut_ptr(),
&ldc,
)
}
#[allow(clippy::too_many_arguments)]
#[inline]
pub unsafe fn dgemm(
transa: u8,
transb: u8,
m: i32,
n: i32,
k: i32,
alpha: f64,
a: &[f64],
lda: i32,
b: &[f64],
ldb: i32,
beta: f64,
c: &mut [f64],
ldc: i32,
) {
ffi::dgemm_ffi(
&(transa as c_char),
&(transb as c_char),
&m,
&n,
&k,
&alpha,
a.as_ptr(),
&lda,
b.as_ptr(),
&ldb,
&beta,
c.as_mut_ptr(),
&ldc,
)
}
#[inline]
pub fn vs_exp(a: &[f32], y: &mut [f32]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvexpf(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vd_exp(a: &[f64], y: &mut [f64]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvexp(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vs_sqrt(a: &[f32], y: &mut [f32]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvsqrtf(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vd_sqrt(a: &[f64], y: &mut [f64]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvsqrt(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vs_sin(a: &[f32], y: &mut [f32]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvsinf(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vd_sin(a: &[f64], y: &mut [f64]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvsin(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vs_cos(a: &[f32], y: &mut [f32]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvcosf(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vd_cos(a: &[f64], y: &mut [f64]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvcos(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vs_ln(a: &[f32], y: &mut [f32]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvlogf(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vd_ln(a: &[f64], y: &mut [f64]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
unsafe { ffi::vvlog(y.as_mut_ptr(), a.as_ptr(), &(a_len as i32)) }
}
#[inline]
pub fn vs_sqr(a: &[f32], y: &mut [f32]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
y.iter_mut().zip(a.iter()).for_each(|(y, a)| *y = *a * *a)
}
#[inline]
pub fn vd_sqr(a: &[f64], y: &mut [f64]) {
let a_len = a.len();
let y_len = y.len();
if a_len != y_len {
panic!("a and y have different lengths {a_len} <> {y_len}")
}
y.iter_mut().zip(a.iter()).for_each(|(y, a)| *y = *a * *a)
}
macro_rules! binary_op {
($fn_name:ident, $ty:ty, $accelerate_name:ident) => {
#[inline]
pub fn $fn_name(a: &[$ty], b: &[$ty], y: &mut [$ty]) {
let a_len = a.len();
let b_len = b.len();
let y_len = y.len();
if a_len != y_len || b_len != y_len {
panic!(
"{} a,b,y len mismatch {a_len} {b_len} {y_len}",
stringify!($fn_name)
);
}
unsafe {
// Weird quirk of accelerate, the rhs comes before the lhs.
ffi::$accelerate_name(
b.as_ptr(),
1,
a.as_ptr(),
1,
y.as_mut_ptr(),
1,
a_len as u64,
)
}
}
};
}
binary_op!(vs_add, f32, vDSP_vadd);
binary_op!(vd_add, f64, vDSP_vaddD);
binary_op!(vs_sub, f32, vDSP_vsub);
binary_op!(vd_sub, f64, vDSP_vsubD);
binary_op!(vs_mul, f32, vDSP_vmul);
binary_op!(vd_mul, f64, vDSP_vmulD);
binary_op!(vs_div, f32, vDSP_vdiv);
binary_op!(vd_div, f64, vDSP_vdivD);
|
