Home | History | Annotate | Download | only in Tensor
      1 // This file is part of Eigen, a lightweight C++ template library
      2 // for linear algebra.
      3 //
      4 // Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog (at) gmail.com>
      5 //
      6 // This Source Code Form is subject to the terms of the Mozilla
      7 // Public License v. 2.0. If a copy of the MPL was not distributed
      8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
      9 
     10 #ifndef EIGEN_CXX11_TENSOR_TENSOR_DEVICE_DEFAULT_H
     11 #define EIGEN_CXX11_TENSOR_TENSOR_DEVICE_DEFAULT_H
     12 
     13 
     14 namespace Eigen {
     15 
     16 // Default device for the machine (typically a single cpu core)
     17 struct DefaultDevice {
     18   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void* allocate(size_t num_bytes) const {
     19     return internal::aligned_malloc(num_bytes);
     20   }
     21   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void deallocate(void* buffer) const {
     22     internal::aligned_free(buffer);
     23   }
     24   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void memcpy(void* dst, const void* src, size_t n) const {
     25     ::memcpy(dst, src, n);
     26   }
     27   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void memcpyHostToDevice(void* dst, const void* src, size_t n) const {
     28     memcpy(dst, src, n);
     29   }
     30   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void memcpyDeviceToHost(void* dst, const void* src, size_t n) const {
     31     memcpy(dst, src, n);
     32   }
     33   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void memset(void* buffer, int c, size_t n) const {
     34     ::memset(buffer, c, n);
     35   }
     36 
     37   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE size_t numThreads() const {
     38 #ifndef __CUDA_ARCH__
     39     // Running on the host CPU
     40     return 1;
     41 #else
     42     // Running on a CUDA device
     43     return 32;
     44 #endif
     45   }
     46 
     47   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE size_t firstLevelCacheSize() const {
     48 #ifndef __CUDA_ARCH__
     49     // Running on the host CPU
     50     return l1CacheSize();
     51 #else
     52     // Running on a CUDA device, return the amount of shared memory available.
     53     return 48*1024;
     54 #endif
     55   }
     56 
     57   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE size_t lastLevelCacheSize() const {
     58 #ifndef __CUDA_ARCH__
     59     // Running single threaded on the host CPU
     60     return l3CacheSize();
     61 #else
     62     // Running on a CUDA device
     63     return firstLevelCacheSize();
     64 #endif
     65   }
     66 
     67   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE int majorDeviceVersion() const {
     68 #ifndef __CUDA_ARCH__
     69     // Running single threaded on the host CPU
     70     // Should return an enum that encodes the ISA supported by the CPU
     71     return 1;
     72 #else
     73     // Running on a CUDA device
     74     return __CUDA_ARCH__ / 100;
     75 #endif
     76   }
     77 };
     78 
     79 }  // namespace Eigen
     80 
     81 #endif // EIGEN_CXX11_TENSOR_TENSOR_DEVICE_DEFAULT_H
     82