GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-14 12:46:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refined the #ifdef __CUDACC__ guard to ensure that when trying to compile gpu code with a non cuda compiler results in a linking error instead of bogus code.

Browse Source
This commit is contained in:
Benoit Steiner 2015-10-23 09:15:34 -07:00
parent ac99b49249
commit 9ea39ce13c
3 changed files with 54 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
unsupported/Eigen/CXX11/src/Tensor/TensorDevice.h
View File

@ -106,7 +106,7 @@ template <typename ExpressionType> class TensorDevice<ExpressionType, ThreadPool
#endif #endif
#if defined(EIGEN_USE_GPU) && defined(__CUDACC__) #if defined(EIGEN_USE_GPU)
template <typename ExpressionType> class TensorDevice<ExpressionType, GpuDevice> template <typename ExpressionType> class TensorDevice<ExpressionType, GpuDevice>
{ {
public: public:

11
unsupported/Eigen/CXX11/src/Tensor/TensorDeviceType.h
View File

@ -287,6 +287,7 @@ class StreamInterface {
virtual void deallocate(void* buffer) const = 0; virtual void deallocate(void* buffer) const = 0;
}; };
#if defined(__CUDACC__)
static cudaDeviceProp* m_deviceProperties; static cudaDeviceProp* m_deviceProperties;
static bool m_devicePropInitialized = false; static bool m_devicePropInitialized = false;
@ -362,7 +363,7 @@ class CudaStreamDevice : public StreamInterface {
const cudaStream_t* stream_; const cudaStream_t* stream_;
int device_; int device_;
}; };
#endif // __CUDACC__
struct GpuDevice { struct GpuDevice {
// The StreamInterface is not owned: the caller is // The StreamInterface is not owned: the caller is
@ -450,7 +451,7 @@ struct GpuDevice {
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void synchronize() const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void synchronize() const {
#ifndef __CUDA_ARCH__ #if defined(__CUDACC__) && !defined(__CUDA_ARCH__)
cudaError_t err = cudaStreamSynchronize(stream_->stream()); cudaError_t err = cudaStreamSynchronize(stream_->stream());
assert(err == cudaSuccess); assert(err == cudaSuccess);
#else #else
@ -477,8 +478,12 @@ struct GpuDevice {
// This function checks if the CUDA runtime recorded an error for the // This function checks if the CUDA runtime recorded an error for the
// underlying stream device. // underlying stream device.
inline bool ok() const { inline bool ok() const {
#ifdef __CUDACC__
cudaError_t error = cudaStreamQuery(stream_->stream()); cudaError_t error = cudaStreamQuery(stream_->stream());
return (error == cudaSuccess) || (error == cudaErrorNotReady); return (error == cudaSuccess) || (error == cudaErrorNotReady);
#else
return false;
#endif
} }
private: private:
@ -493,10 +498,12 @@ struct GpuDevice {
// FIXME: Should be device and kernel specific. // FIXME: Should be device and kernel specific.
#ifdef __CUDACC__
static inline void setCudaSharedMemConfig(cudaSharedMemConfig config) { static inline void setCudaSharedMemConfig(cudaSharedMemConfig config) {
cudaError_t status = cudaDeviceSetSharedMemConfig(config); cudaError_t status = cudaDeviceSetSharedMemConfig(config);
assert(status == cudaSuccess); assert(status == cudaSuccess);
} }
#endif
#endif #endif

40
unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h
View File

@ -149,7 +149,24 @@ class TensorExecutor<Expression, ThreadPoolDevice, Vectorizable>
// GPU: the evaluation of the expression is offloaded to a GPU. // GPU: the evaluation of the expression is offloaded to a GPU.
#if defined(EIGEN_USE_GPU) && defined(__CUDACC__) #if defined(EIGEN_USE_GPU)
template <typename Expression>
class TensorExecutor<Expression, GpuDevice, false> {
public:
typedef typename Expression::Index Index;
static void run(const Expression& expr, const GpuDevice& device);
};
template <typename Expression>
class TensorExecutor<Expression, GpuDevice, true> {
public:
typedef typename Expression::Index Index;
static void run(const Expression& expr, const GpuDevice& device);
};
#if defined(__CUDACC__)
template <typename Evaluator, typename Index> template <typename Evaluator, typename Index>
__global__ void __global__ void
__launch_bounds__(1024) __launch_bounds__(1024)
@ -193,13 +210,9 @@ EigenMetaKernel_Vectorizable(Evaluator memcopied_eval, Index size) {
} }
} }
/*static*/
template <typename Expression> template <typename Expression>
class TensorExecutor<Expression, GpuDevice, false> inline void TensorExecutor<Expression, GpuDevice, false>::run(const Expression& expr, const GpuDevice& device)
{
public:
typedef typename Expression::Index Index;
static inline void run(const Expression& expr, const GpuDevice& device)
{ {
TensorEvaluator<Expression, GpuDevice> evaluator(expr, device); TensorEvaluator<Expression, GpuDevice> evaluator(expr, device);
const bool needs_assign = evaluator.evalSubExprsIfNeeded(NULL); const bool needs_assign = evaluator.evalSubExprsIfNeeded(NULL);
@ -212,14 +225,11 @@ class TensorExecutor<Expression, GpuDevice, false>
} }
evaluator.cleanup(); evaluator.cleanup();
} }
};
/*static*/
template<typename Expression> template<typename Expression>
class TensorExecutor<Expression, GpuDevice, true> inline void TensorExecutor<Expression, GpuDevice, false>::run(const Expression& expr, const GpuDevice& device)
{
public:
typedef typename Expression::Index Index;
static inline void run(const Expression& expr, const GpuDevice& device)
{ {
TensorEvaluator<Expression, GpuDevice> evaluator(expr, device); TensorEvaluator<Expression, GpuDevice> evaluator(expr, device);
const bool needs_assign = evaluator.evalSubExprsIfNeeded(NULL); const bool needs_assign = evaluator.evalSubExprsIfNeeded(NULL);
@ -232,9 +242,9 @@ class TensorExecutor<Expression, GpuDevice, true>
} }
evaluator.cleanup(); evaluator.cleanup();
} }
};
#endif #endif // __CUDACC__
#endif // EIGEN_USE_GPU
} // end namespace internal } // end namespace internal