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
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
use crate::error::{CudaResult, ToResult};
use crate::memory::device::AsyncCopyDestination;
use crate::memory::device::{CopyDestination, DeviceBuffer};
use crate::memory::DeviceCopy;
use crate::memory::DevicePointer;
use crate::stream::Stream;
use cuda_sys::cuda;
use std::iter::{ExactSizeIterator, FusedIterator};
use std::mem;
use std::ops::{
    Index, IndexMut, Range, RangeFrom, RangeFull, RangeInclusive, RangeTo, RangeToInclusive,
};

use std::os::raw::c_void;
use std::slice::{self, Chunks, ChunksMut};

/// Fixed-size device-side slice.
#[derive(Debug)]
#[repr(C)]
pub struct DeviceSlice<T>([T]);
// This works by faking a regular slice out of the device raw-pointer and the length and transmuting
// I have no idea if this is safe or not. Probably not, though I can't imagine how the compiler
// could possibly know that the pointer is not de-referenceable. I'm banking that we get proper
// Dynamicaly-sized Types before the compiler authors break this assumption.
impl<T> DeviceSlice<T> {
    /// Returns the number of elements in the slice.
    ///
    /// # Examples
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let a = DeviceBuffer::from_slice(&[1, 2, 3]).unwrap();
    /// assert_eq!(a.len(), 3);
    /// ```
    pub fn len(&self) -> usize {
        self.0.len()
    }

    /// Returns `true` if the slice has a length of 0.
    ///
    /// # Examples
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let a : DeviceBuffer<u64> = unsafe { DeviceBuffer::uninitialized(0).unwrap() };
    /// assert!(a.is_empty());
    /// ```
    pub fn is_empty(&self) -> bool {
        self.0.is_empty()
    }

    /// Return a raw device-pointer to the slice's buffer.
    ///
    /// The caller must ensure that the slice outlives the pointer this function returns, or else
    /// it will end up pointing to garbage. The caller must also ensure that the pointer is not
    /// dereferenced by the CPU.
    ///
    /// Examples:
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let a = DeviceBuffer::from_slice(&[1, 2, 3]).unwrap();
    /// println!("{:p}", a.as_ptr());
    /// ```
    pub fn as_ptr(&self) -> *const T {
        self.0.as_ptr()
    }

    /// Returns an unsafe mutable device-pointer to the slice's buffer.
    ///
    /// The caller must ensure that the slice outlives the pointer this function returns, or else
    /// it will end up pointing to garbage. The caller must also ensure that the pointer is not
    /// dereferenced by the CPU.
    ///
    /// Examples:
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let mut a = DeviceBuffer::from_slice(&[1, 2, 3]).unwrap();
    /// println!("{:p}", a.as_mut_ptr());
    /// ```
    pub fn as_mut_ptr(&mut self) -> *mut T {
        self.0.as_mut_ptr()
    }

    /// Divides one DeviceSlice into two at a given index.
    ///
    /// The first will contain all indices from `[0, mid)` (excluding the index `mid` itself) and
    /// the second will contain all indices from `[mid, len)` (excluding the index `len` itself).
    ///
    /// # Panics
    ///
    /// Panics if `min > len`.
    ///
    /// Examples:
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let buf = DeviceBuffer::from_slice(&[0u64, 1, 2, 3, 4, 5]).unwrap();
    /// let (left, right) = buf.split_at(3);
    /// let mut left_host = [0u64, 0, 0];
    /// let mut right_host = [0u64, 0, 0];
    /// left.copy_to(&mut left_host).unwrap();
    /// right.copy_to(&mut right_host).unwrap();
    /// assert_eq!([0u64, 1, 2], left_host);
    /// assert_eq!([3u64, 4, 5], right_host);
    /// ```
    pub fn split_at(&self, mid: usize) -> (&DeviceSlice<T>, &DeviceSlice<T>) {
        let (left, right) = self.0.split_at(mid);
        unsafe {
            (
                DeviceSlice::from_slice(left),
                DeviceSlice::from_slice(right),
            )
        }
    }

    /// Divides one mutable DeviceSlice into two at a given index.
    ///
    /// The first will contain all indices from `[0, mid)` (excluding the index `mid` itself) and
    /// the second will contain all indices from `[mid, len)` (excluding the index `len` itself).
    ///
    /// # Panics
    ///
    /// Panics if `min > len`.
    ///
    /// Examples:
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let mut buf = DeviceBuffer::from_slice(&[0u64, 0, 0, 0, 0, 0]).unwrap();
    ///
    /// {
    ///     let (left, right) = buf.split_at_mut(3);
    ///     let left_host = [0u64, 1, 2];
    ///     let right_host = [3u64, 4, 5];
    ///     left.copy_from(&left_host).unwrap();
    ///     right.copy_from(&right_host).unwrap();
    /// }
    ///
    /// let mut host_full = [0u64; 6];
    /// buf.copy_to(&mut host_full).unwrap();
    /// assert_eq!([0u64, 1, 2, 3, 4, 5], host_full);
    /// ```
    pub fn split_at_mut(&mut self, mid: usize) -> (&mut DeviceSlice<T>, &mut DeviceSlice<T>) {
        let (left, right) = self.0.split_at_mut(mid);
        unsafe {
            (
                DeviceSlice::from_slice_mut(left),
                DeviceSlice::from_slice_mut(right),
            )
        }
    }

    /// Returns an iterator over `chunk_size` elements of the slice at a time. The chunks are device
    /// slices and do not overlap. If `chunk_size` does not divide the length of the slice, then the
    /// last chunk will not have length `chunk_size`.
    ///
    /// See `exact_chunks` for a variant of this iterator that returns chunks of always exactly
    /// `chunk_size` elements.
    ///
    /// # Panics
    ///
    /// Panics if `chunk_size` is 0.
    ///
    /// # Examples
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let slice = DeviceBuffer::from_slice(&[1u64, 2, 3, 4, 5]).unwrap();
    /// let mut iter = slice.chunks(2);
    ///
    /// assert_eq!(iter.next().unwrap().len(), 2);
    ///
    /// let mut host_buf = [0u64, 0];
    /// iter.next().unwrap().copy_to(&mut host_buf).unwrap();
    /// assert_eq!([3, 4], host_buf);
    ///
    /// assert_eq!(iter.next().unwrap().len(), 1);
    ///
    /// ```
    pub fn chunks(&self, chunk_size: usize) -> DeviceChunks<T> {
        DeviceChunks(self.0.chunks(chunk_size))
    }

    /// Returns an iterator over `chunk_size` elements of the slice at a time. The chunks are
    /// mutable device slices and do not overlap. If `chunk_size` does not divide the length of the
    /// slice, then the last chunk will not have length `chunk_size`.
    ///
    /// See `exact_chunks` for a variant of this iterator that returns chunks of always exactly
    /// `chunk_size` elements.
    ///
    /// # Panics
    ///
    /// Panics if `chunk_size` is 0.
    ///
    /// # Examples
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let mut slice = DeviceBuffer::from_slice(&[0u64, 0, 0, 0, 0]).unwrap();
    /// {
    ///     let mut iter = slice.chunks_mut(2);
    ///
    ///     assert_eq!(iter.next().unwrap().len(), 2);
    ///
    ///     let host_buf = [2u64, 3];
    ///     iter.next().unwrap().copy_from(&host_buf).unwrap();
    ///
    ///     assert_eq!(iter.next().unwrap().len(), 1);
    /// }
    ///
    /// let mut host_buf = [0u64, 0, 0, 0, 0];
    /// slice.copy_to(&mut host_buf).unwrap();
    /// assert_eq!([0u64, 0, 2, 3, 0], host_buf);
    /// ```
    pub fn chunks_mut(&mut self, chunk_size: usize) -> DeviceChunksMut<T> {
        DeviceChunksMut(self.0.chunks_mut(chunk_size))
    }

    /// Private function used to transmute a CPU slice (which must have the device pointer as it's
    /// buffer pointer) to a DeviceSlice. Completely unsafe.
    pub(super) unsafe fn from_slice(slice: &[T]) -> &DeviceSlice<T> {
        &*(slice as *const [T] as *const DeviceSlice<T>)
    }

    /// Private function used to transmute a mutable CPU slice (which must have the device pointer
    /// as it's buffer pointer) to a mutable DeviceSlice. Completely unsafe.
    pub(super) unsafe fn from_slice_mut(slice: &mut [T]) -> &mut DeviceSlice<T> {
        &mut *(slice as *mut [T] as *mut DeviceSlice<T>)
    }

    /// Returns a `DevicePointer<T>` to the buffer.
    ///
    /// The caller must ensure that the buffer outlives the returned pointer, or it will end up
    /// pointing to garbage.
    ///
    /// Modifying `DeviceBuffer` is guaranteed not to cause its buffer to be reallocated, so pointers
    /// cannot be invalidated in that manner, but other types may be added in the future which can
    /// reallocate.
    pub fn as_device_ptr(&mut self) -> DevicePointer<T> {
        unsafe { DevicePointer::wrap(self.0.as_mut_ptr()) }
    }

    /// Forms a slice from a `DevicePointer` and a length.
    ///
    /// The `len` argument is the number of _elements_, not the number of bytes.
    ///
    /// # Safety
    ///
    /// This function is unsafe as there is no guarantee that the given pointer is valid for `len`
    /// elements, nor whether the lifetime inferred is a suitable lifetime for the returned slice.
    ///
    /// # Caveat
    ///
    /// The lifetime for the returned slice is inferred from its usage. To prevent accidental misuse,
    /// it's suggested to tie the lifetime to whatever source lifetime is safe in the context, such
    /// as by providing a helper function taking the lifetime of a host value for the slice or
    /// by explicit annotation.
    ///
    /// # Examples
    ///
    /// ```
    /// # let _context = rustacuda::quick_init().unwrap();
    /// use rustacuda::memory::*;
    /// let mut x = DeviceBuffer::from_slice(&[0u64, 1, 2, 3, 4, 5]).unwrap();
    /// // Manually slice the buffer (this is not recommended!)
    /// let ptr = unsafe { x.as_device_ptr().offset(1) };
    /// let slice = unsafe { DeviceSlice::from_raw_parts(ptr, 2) };
    /// let mut host_buf = [0u64, 0];
    /// slice.copy_to(&mut host_buf).unwrap();
    /// assert_eq!([1u64, 2], host_buf);
    /// ```
    #[allow(clippy::needless_pass_by_value)]
    pub unsafe fn from_raw_parts<'a>(data: DevicePointer<T>, len: usize) -> &'a DeviceSlice<T> {
        DeviceSlice::from_slice(slice::from_raw_parts(data.as_raw(), len))
    }

    /// Performs the same functionality as `from_raw_parts`, except that a
    /// mutable slice is returned.
    ///
    /// # Safety
    ///
    /// This function is unsafe as there is no guarantee that the given pointer is valid for `len`
    /// elements, nor whether the lifetime inferred is a suitable lifetime for the returned slice.
    ///
    /// This function is unsafe as there is no guarantee that the given pointer is valid for `len`
    /// elements, not whether the lifetime inferred is a suitable lifetime for the returned slice,
    /// as well as not being able to provide a non-aliasing guarantee of the returned
    /// mutable slice. `data` must be non-null and aligned even for zero-length
    /// slices as with `from_raw_parts`.
    ///
    /// See the documentation of `from_raw_parts` for more details.
    pub unsafe fn from_raw_parts_mut<'a>(
        mut data: DevicePointer<T>,
        len: usize,
    ) -> &'a mut DeviceSlice<T> {
        DeviceSlice::from_slice_mut(slice::from_raw_parts_mut(data.as_raw_mut(), len))
    }
}

/// An iterator over a [`DeviceSlice`](struct.DeviceSlice.html) in (non-overlapping) chunks
/// (`chunk_size` elements at a time).
///
/// When the slice len is not evenly divided by the chunk size, the last slice of the iteration will
/// be the remainder.
///
/// This struct is created by the `chunks` method on `DeviceSlices`.
#[derive(Debug, Clone)]
pub struct DeviceChunks<'a, T: 'a>(Chunks<'a, T>);
impl<'a, T> Iterator for DeviceChunks<'a, T> {
    type Item = &'a DeviceSlice<T>;

    fn next(&mut self) -> Option<&'a DeviceSlice<T>> {
        self.0
            .next()
            .map(|slice| unsafe { DeviceSlice::from_slice(slice) })
    }

    fn size_hint(&self) -> (usize, Option<usize>) {
        self.0.size_hint()
    }

    fn count(self) -> usize {
        self.0.len()
    }

    fn nth(&mut self, n: usize) -> Option<Self::Item> {
        self.0
            .nth(n)
            .map(|slice| unsafe { DeviceSlice::from_slice(slice) })
    }

    #[inline]
    fn last(self) -> Option<Self::Item> {
        self.0
            .last()
            .map(|slice| unsafe { DeviceSlice::from_slice(slice) })
    }
}
impl<'a, T> DoubleEndedIterator for DeviceChunks<'a, T> {
    #[inline]
    fn next_back(&mut self) -> Option<&'a DeviceSlice<T>> {
        self.0
            .next_back()
            .map(|slice| unsafe { DeviceSlice::from_slice(slice) })
    }
}
impl<'a, T> ExactSizeIterator for DeviceChunks<'a, T> {}
impl<'a, T> FusedIterator for DeviceChunks<'a, T> {}

/// An iterator over a [`DeviceSlice`](struct.DeviceSlice.html) in (non-overlapping) mutable chunks
/// (`chunk_size` elements at a time).
///
/// When the slice len is not evenly divided by the chunk size, the last slice of the iteration will
/// be the remainder.
///
/// This struct is created by the `chunks` method on `DeviceSlices`.
#[derive(Debug)]
pub struct DeviceChunksMut<'a, T: 'a>(ChunksMut<'a, T>);
impl<'a, T> Iterator for DeviceChunksMut<'a, T> {
    type Item = &'a mut DeviceSlice<T>;

    fn next(&mut self) -> Option<&'a mut DeviceSlice<T>> {
        self.0
            .next()
            .map(|slice| unsafe { DeviceSlice::from_slice_mut(slice) })
    }

    fn size_hint(&self) -> (usize, Option<usize>) {
        self.0.size_hint()
    }

    fn count(self) -> usize {
        self.0.len()
    }

    fn nth(&mut self, n: usize) -> Option<Self::Item> {
        self.0
            .nth(n)
            .map(|slice| unsafe { DeviceSlice::from_slice_mut(slice) })
    }

    #[inline]
    fn last(self) -> Option<Self::Item> {
        self.0
            .last()
            .map(|slice| unsafe { DeviceSlice::from_slice_mut(slice) })
    }
}
impl<'a, T> DoubleEndedIterator for DeviceChunksMut<'a, T> {
    #[inline]
    fn next_back(&mut self) -> Option<&'a mut DeviceSlice<T>> {
        self.0
            .next_back()
            .map(|slice| unsafe { DeviceSlice::from_slice_mut(slice) })
    }
}
impl<'a, T> ExactSizeIterator for DeviceChunksMut<'a, T> {}
impl<'a, T> FusedIterator for DeviceChunksMut<'a, T> {}

macro_rules! impl_index {
    ($($t:ty)*) => {
        $(
            impl<T> Index<$t> for DeviceSlice<T>
            {
                type Output = DeviceSlice<T>;

                fn index(&self, index: $t) -> &Self {
                    unsafe { DeviceSlice::from_slice(self.0.index(index)) }
                }
            }

            impl<T> IndexMut<$t> for DeviceSlice<T>
            {
                fn index_mut(&mut self, index: $t) -> &mut Self {
                    unsafe { DeviceSlice::from_slice_mut( self.0.index_mut(index)) }
                }
            }
        )*
    }
}
impl_index! {
    Range<usize>
    RangeFull
    RangeFrom<usize>
    RangeInclusive<usize>
    RangeTo<usize>
    RangeToInclusive<usize>
}
impl<T> crate::private::Sealed for DeviceSlice<T> {}
impl<T: DeviceCopy, I: AsRef<[T]> + AsMut<[T]> + ?Sized> CopyDestination<I> for DeviceSlice<T> {
    fn copy_from(&mut self, val: &I) -> CudaResult<()> {
        let val = val.as_ref();
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            unsafe {
                cuda::cuMemcpyHtoD_v2(
                    self.0.as_mut_ptr() as u64,
                    val.as_ptr() as *const c_void,
                    size,
                )
                .to_result()?
            }
        }
        Ok(())
    }

    fn copy_to(&self, val: &mut I) -> CudaResult<()> {
        let val = val.as_mut();
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            unsafe {
                cuda::cuMemcpyDtoH_v2(val.as_mut_ptr() as *mut c_void, self.as_ptr() as u64, size)
                    .to_result()?
            }
        }
        Ok(())
    }
}
impl<T: DeviceCopy> CopyDestination<DeviceSlice<T>> for DeviceSlice<T> {
    fn copy_from(&mut self, val: &DeviceSlice<T>) -> CudaResult<()> {
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            unsafe {
                cuda::cuMemcpyDtoD_v2(self.0.as_mut_ptr() as u64, val.as_ptr() as u64, size)
                    .to_result()?
            }
        }
        Ok(())
    }

    fn copy_to(&self, val: &mut DeviceSlice<T>) -> CudaResult<()> {
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            unsafe {
                cuda::cuMemcpyDtoD_v2(val.as_mut_ptr() as u64, self.as_ptr() as u64, size)
                    .to_result()?
            }
        }
        Ok(())
    }
}
impl<T: DeviceCopy> CopyDestination<DeviceBuffer<T>> for DeviceSlice<T> {
    fn copy_from(&mut self, val: &DeviceBuffer<T>) -> CudaResult<()> {
        self.copy_from(val as &DeviceSlice<T>)
    }

    fn copy_to(&self, val: &mut DeviceBuffer<T>) -> CudaResult<()> {
        self.copy_to(val as &mut DeviceSlice<T>)
    }
}
impl<T: DeviceCopy, I: AsRef<[T]> + AsMut<[T]> + ?Sized> AsyncCopyDestination<I>
    for DeviceSlice<T>
{
    unsafe fn async_copy_from(&mut self, val: &I, stream: &Stream) -> CudaResult<()> {
        let val = val.as_ref();
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            cuda::cuMemcpyHtoDAsync_v2(
                self.0.as_mut_ptr() as u64,
                val.as_ptr() as *const c_void,
                size,
                stream.as_inner(),
            )
            .to_result()?
        }
        Ok(())
    }

    unsafe fn async_copy_to(&self, val: &mut I, stream: &Stream) -> CudaResult<()> {
        let val = val.as_mut();
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            cuda::cuMemcpyDtoHAsync_v2(
                val.as_mut_ptr() as *mut c_void,
                self.as_ptr() as u64,
                size,
                stream.as_inner(),
            )
            .to_result()?
        }
        Ok(())
    }
}
impl<T: DeviceCopy> AsyncCopyDestination<DeviceSlice<T>> for DeviceSlice<T> {
    unsafe fn async_copy_from(&mut self, val: &DeviceSlice<T>, stream: &Stream) -> CudaResult<()> {
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            cuda::cuMemcpyDtoDAsync_v2(
                self.0.as_mut_ptr() as u64,
                val.as_ptr() as u64,
                size,
                stream.as_inner(),
            )
            .to_result()?
        }
        Ok(())
    }

    unsafe fn async_copy_to(&self, val: &mut DeviceSlice<T>, stream: &Stream) -> CudaResult<()> {
        assert!(
            self.len() == val.len(),
            "destination and source slices have different lengths"
        );
        let size = mem::size_of::<T>() * self.len();
        if size != 0 {
            cuda::cuMemcpyDtoDAsync_v2(
                val.as_mut_ptr() as u64,
                self.as_ptr() as u64,
                size,
                stream.as_inner(),
            )
            .to_result()?
        }
        Ok(())
    }
}
impl<T: DeviceCopy> AsyncCopyDestination<DeviceBuffer<T>> for DeviceSlice<T> {
    unsafe fn async_copy_from(&mut self, val: &DeviceBuffer<T>, stream: &Stream) -> CudaResult<()> {
        self.async_copy_from(val as &DeviceSlice<T>, stream)
    }

    unsafe fn async_copy_to(&self, val: &mut DeviceBuffer<T>, stream: &Stream) -> CudaResult<()> {
        self.async_copy_to(val as &mut DeviceSlice<T>, stream)
    }
}