Skip to content

GitLab

Commit 3ea343b7 authored by incardon's avatar incardon
Browse files

Compilable cuda on cpu

parent 84e4f62c
Pipeline #2508 failed with stages
in 7 seconds
Hide whitespace changes
Inline Side-by-side
CMakeLists.txt
View file @ 3ea343b7
......@@ -6,12 +6,18 @@ set(BOOST_INCLUDE ${Boost_INCLUDE_DIR} CACHE PATH "Include directory for BOOST")
set (CMAKE_CXX_STANDARD 14)
set (CMAKE_CUDA_STANDARD 14)
find_package(Boost 1.68.0 REQUIRED COMPONENTS unit_test_framework iostreams program_options)
find_package(Boost 1.68.0 REQUIRED COMPONENTS unit_test_framework iostreams program_options fiber)
set(CUDA_ON_CPU CACHE BOOL "Make Cuda work on heap")
# Save Boost_LIBRARIES before alpaka fuck-up Boost
set(Boost_LIBRARIES_BCK ${Boost_LIBRARIES})
if(ENABLE_GPU)
enable_language(CUDA)
find_package(CUDA)
else()
find_package(alpaka)
endif()
###### CONFIG.h FILE ######
......
src/CMakeLists.txt
View file @ 3ea343b7
if(CUDA_FOUND)
if(CUDA_FOUND OR CUDA_ON_CPU)
set(CUDA_SOURCES memory/CudaMemory.cu)
else()
set(CUDA_SOURCES )
endif()
if (CUDA_ON_CPU)
add_definitions(-DCUDA_ON_CPU)
set_source_files_properties(${CUDA_SOURCES} PROPERTIES LANGUAGE CXX)
if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
add_definitions("-x c++")
endif()
endif()
add_executable(mem main.cpp memory/HeapMemory.cpp ${CUDA_SOURCES})
add_executable(mem main.cpp memory/HeapMemory.cpp util/cudify_vars.cpp util/cudify_unit_test.cpp ${CUDA_SOURCES})
add_library(ofpmmemory STATIC memory/HeapMemory.cpp memory/PtrMemory.cpp ${CUDA_SOURCES})
add_library(ofpmmemory_dl SHARED memory/HeapMemory.cpp memory/PtrMemory.cpp ${CUDA_SOURCES})
add_library(ofpmmemory STATIC memory/HeapMemory.cpp util/cudify_vars.cpp memory/PtrMemory.cpp ${CUDA_SOURCES})
add_library(ofpmmemory_dl SHARED memory/HeapMemory.cpp util/cudify_vars.cpp memory/PtrMemory.cpp ${CUDA_SOURCES})
if ( CMAKE_COMPILER_IS_GNUCC )
target_compile_options(mem PRIVATE "-Wno-deprecated-declarations")
......@@ -30,17 +37,19 @@ target_include_directories (mem PUBLIC ${CUDA_INCLUDE_DIRS})
target_include_directories (mem PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
target_include_directories (mem PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/config)
target_include_directories (mem PUBLIC ${Boost_INCLUDE_DIRS})
target_include_directories (mem PUBLIC ${ALPAKA_ROOT}/include)
if (CUDA_FOUND)
target_include_directories (mem PUBLIC ${CUDA_INCLUDE_DIRS})
endif()
target_include_directories (ofpmmemory PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
target_include_directories (ofpmmemory PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/config)
target_include_directories (ofpmmemory PUBLIC ${ALPAKA_ROOT}/include)
target_include_directories (ofpmmemory_dl PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
target_include_directories (ofpmmemory_dl PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/config)
target_link_libraries(mem ${Boost_LIBRARIES})
target_include_directories (ofpmmemory_dl PUBLIC ${ALPAKA_ROOT}/include)
target_link_libraries(mem ${Boost_LIBRARIES_BCK})
if (TEST_COVERAGE)
target_link_libraries(mem -lgcov)
......
src/memory/CudaMemory.cu
View file @ 3ea343b7
......@@ -17,7 +17,11 @@ bool CudaMemory::flush()
{
//! copy from host to device memory
CUDA_SAFE_CALL(cudaMemcpy(dm,hm,sz,cudaMemcpyHostToDevice));
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(dm,hm,sz,cudaMemcpyHostToDevice));
#else
memcpy(dm,hm,sz);
#endif
}
return true;
......@@ -34,7 +38,13 @@ bool CudaMemory::allocate(size_t sz)
{
//! Allocate the device memory
if (dm == NULL)
{CUDA_SAFE_CALL(cudaMalloc(&dm,sz));}
{
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMalloc(&dm,sz));
#else
dm = new unsigned char[sz];
#endif
}
else
{
if (sz != this->sz)
......@@ -63,14 +73,22 @@ void CudaMemory::destroy()
if (dm != NULL)
{
//! Release the allocated memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaFree(dm));
#else
delete [] (unsigned char *)dm;
#endif
dm = NULL;
}
if (hm != NULL)
{
//! we invalidate hm
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaFreeHost(hm));
#else
delete [] (unsigned char *)hm;
#endif
hm = NULL;
}
......@@ -87,7 +105,11 @@ void CudaMemory::destroy()
*/
void CudaMemory::deviceToDevice(void * ptr, size_t start, size_t stop, size_t offset)
{
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(((unsigned char *)dm)+offset,((unsigned char *)ptr)+start,(stop-start),cudaMemcpyDeviceToDevice));
#else
memcpy(((unsigned char *)dm)+offset,((unsigned char *)ptr)+start,(stop-start));
#endif
}
/*! \brief Allocate the host buffer
......@@ -99,7 +121,11 @@ void CudaMemory::allocate_host(size_t sz) const
{
if (hm == NULL)
{
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaHostAlloc(&hm,sz,cudaHostAllocMapped))
#else
hm = new unsigned char[sz];
#endif
}
}
......@@ -119,11 +145,15 @@ bool CudaMemory::copyFromPointer(const void * ptr)
// get the device pointer
void * dvp;
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaHostGetDevicePointer(&dvp,hm,0));
// memory copy
memcpy(dvp,ptr,sz);
#else
memcpy(hm,ptr,sz);
#endif
return true;
}
......@@ -147,7 +177,11 @@ bool CudaMemory::copyDeviceToDevice(const CudaMemory & m)
}
//! Copy the memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(dm,m.dm,m.sz,cudaMemcpyDeviceToDevice));
#else
memcpy(dm,m.dm,m.sz);
#endif
return true;
}
......@@ -225,26 +259,46 @@ bool CudaMemory::resize(size_t sz)
{
if (this->sz < sz)
{
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMalloc(&tdm,sz));
#else
tdm = new unsigned char [sz];
#endif
#ifdef FILL_CUDA_MEMORY_WITH_MINUS_ONE
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemset(tdm,-1,sz));
#else
memset(tdm,-1,sz);
#endif
#endif
}
//! copy from the old buffer to the new one
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(tdm,dm,CudaMemory::size(),cudaMemcpyDeviceToDevice));
#else
memcpy(tdm,dm,CudaMemory::size());
#endif
}
if (hm != NULL)
{
if (this->sz < sz)
{
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaHostAlloc(&thm,sz,cudaHostAllocMapped));
#else
thm = new unsigned char [sz];
#endif
}
//! copy from the old buffer to the new one
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(thm,hm,CudaMemory::size(),cudaMemcpyHostToHost));
#else
memcpy(thm,hm,CudaMemory::size());
#endif
}
//! free the old buffer
......@@ -289,8 +343,11 @@ void CudaMemory::deviceToHost()
allocate_host(sz);
//! copy from device to host memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(hm,dm,sz,cudaMemcpyDeviceToHost));
#else
memcpy(hm,dm,sz);
#endif
}
/*! \brief It transfer to device memory from the host of another memory
......@@ -308,7 +365,11 @@ void CudaMemory::deviceToHost(CudaMemory & mem)
{resize(mem.sz);}
//! copy from device to host memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(mem.hm,dm,mem.sz,cudaMemcpyDeviceToHost));
#else
memcpy(mem.hm,dm,mem.sz);
#endif
}
/*! \brief It transfer to device memory from the host of another memory
......@@ -326,7 +387,11 @@ void CudaMemory::hostToDevice(CudaMemory & mem)
{resize(mem.sz);}
//! copy from device to host memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(dm,mem.hm,mem.sz,cudaMemcpyHostToDevice));
#else
memcpy(dm,mem.hm,mem.sz);
#endif
}
void CudaMemory::hostToDevice(size_t start, size_t stop)
......@@ -336,8 +401,11 @@ void CudaMemory::hostToDevice(size_t start, size_t stop)
allocate_host(sz);
//! copy from device to host memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(((unsigned char *)dm)+start,((unsigned char *)hm)+start,(stop-start),cudaMemcpyHostToDevice));
#else
memcpy(((unsigned char *)dm)+start,((unsigned char *)hm)+start,(stop-start));
#endif
}
/*! \brief Return a readable pointer with your data
......@@ -352,8 +420,11 @@ void CudaMemory::deviceToHost(size_t start, size_t stop)
allocate_host(sz);
//! copy from device to host memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(((unsigned char *)hm)+start,((unsigned char *)dm)+start,(stop-start),cudaMemcpyDeviceToHost));
#else
memcpy(((unsigned char *)hm)+start,((unsigned char *)dm)+start,(stop-start));
#endif
}
......@@ -379,7 +450,11 @@ const void * CudaMemory::getPointer() const
*/
void CudaMemory::fill(unsigned char c)
{
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemset(dm,c,size()));
#else
memset(dm,c,size());
#endif
if (hm != NULL)
{memset(hm,c,size());}
}
......@@ -407,8 +482,11 @@ void CudaMemory::hostToDevice()
allocate_host(sz);
//! copy from device to host memory
#if defined(CUDA_GPU) && !defined(CUDA_ON_CPU)
CUDA_SAFE_CALL(cudaMemcpy(dm,hm,sz,cudaMemcpyHostToDevice));
#else
memcpy(dm,hm,sz);
#endif
}
......
src/memory/CudaMemory.cuh
View file @ 3ea343b7
......@@ -37,9 +37,11 @@
#include "memory.hpp"
#include <iostream>
#ifdef CUDA_GPU
#if defined(__NVCC__) && !defined(CUDA_ON_CPU)
#include <cuda_runtime.h>
#else
#include "util/cuda_util.hpp"
#endif
extern size_t TotCudaMemoryAllocated;
......@@ -191,8 +193,6 @@ public:
mem.dm = NULL;
mem.hm = NULL;
mem.ref_cnt = 0;
cudaGetLastError();
}
//! Constructor
......@@ -246,6 +246,6 @@ public:
}
};
#endif
#endif
src/util/cuda_util.hpp
View file @ 3ea343b7
......@@ -9,7 +9,7 @@
#define OPENFPM_DATA_SRC_UTIL_CUDA_UTIL_HPP_
#include "config.h"
#ifdef CUDA_GPU
#if defined(__NVCC__) && !defined(CUDA_ON_CPU)
#include <cuda_runtime.h>
#endif
......@@ -17,6 +17,12 @@
#ifndef __NVCC__
#ifndef __host__
#define __host__
#define __device__
#define __shared__
#define __global__
#endif
#else
......@@ -24,8 +30,16 @@
#define __host__
#define __device__
#define __shared__
#define __global__
#endif
#ifdef CUDA_ON_CPU
#define CUDA_SAFE(cuda_call) \
cuda_call;
#else
#define CUDA_SAFE(cuda_call) \
cuda_call; \
{\
......@@ -37,13 +51,16 @@
}\
}
#endif
#endif
#else
#ifndef __host__
#define __host__
#define __device__
#define __shared__
#define __shared__ static
#define __global__
#endif
#endif
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment