You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Please format your code before before opening pull requests using clang-format and the .clang-format file placed in the repository root.
6
+
7
+
### Visual Studio and CLion
8
+
Suport for clang-format is built-in since Visual Studio 2017 15.7 and CLion 2019.1.
9
+
The .clang-format file in the repository will be automatically detected and formatting is done as you type, or triggered when pressing the format hotkey.
10
+
11
+
### Bash
12
+
First install clang-format. Instructions therefore can be found on the web. To format you can run this command in bash:
|**CreationPolicy**| Scatter`<conf1,conf2>`| A scattered allocation to tradeoff fragmentation for allocation time, as proposed in [ScatterAlloc](http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6339604). `conf1` configures the heap layout, `conf2` determines the hashing parameters|
22
+
|**CreationPolicy**| Scatter`<conf1,conf2>`| A scattered allocation to tradeoff fragmentation for allocation time, as proposed in [ScatterAlloc](http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6339604). `conf1` configures the heap layout, `conf2` determines the hashing parameters|
23
23
|| OldMalloc | device-side malloc/new and free/delete syscalls as implemented on NVidia CUDA graphics cards with compute capability sm_20 and higher |
24
-
|**DistributionPolicy**| XMallocSIMD`<conf>`| SIMD optimization for warp-wide allocation on NVIDIA CUDA accelerators, as proposed by [XMalloc](http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=5577907). `conf` is used to determine the pagesize. If used in combination with *Scatter*, the pagesizes must match |
24
+
|**DistributionPolicy**| XMallocSIMD`<conf>`| SIMD optimization for warp-wide allocation on NVIDIA CUDA accelerators, as proposed by [XMalloc](http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=5577907). `conf` is used to determine the pagesize. If used in combination with *Scatter*, the pagesizes must match |
25
25
|| Noop | no workload distribution at all |
26
-
|**OOMPolicy**| ReturnNull | pointers will be *NULL*, if the request could not be fulfilled |
26
+
|**OOMPolicy**| ReturnNull | pointers will be *nullptr*, if the request could not be fulfilled |
27
27
||~~BadAllocException~~| will throw a `std::bad_alloc` exception. The accelerator has to support exceptions |
28
28
|**ReservePoolPolicy**| SimpleCudaMalloc | allocate a fixed heap with `CudaMalloc`|
29
29
|| CudaSetLimits | call to `CudaSetLimits` to increase the available Heap (e.g. when using *OldMalloc*) |
30
-
|**AlignmentPolicy**| Shrink`<conf>`| shrinks the pool so that the starting pointer is well aligned, applies padding to requested memory chunks. `conf` is used to determine the alignment|
30
+
|**AlignmentPolicy**| Shrink`<conf>`| shrinks the pool so that the starting pointer is well aligned, applies padding to requested memory chunks. `conf` is used to determine the alignment|
31
31
|| Noop | no alignment at all |
32
32
33
33
The user has to choose one of each policy that will form a useful allocator
0 commit comments