us1zcljz
/
mathnet-numerics
forked from tsai/mathnet-numerics
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
Browse Source

Merge pull request #71 from cuda/mkl 

minor cleanup on mkl project
v2
Christoph Ruegg 14 years ago
parent
38fade8a84 3243c2c015
commit
0a54d4b236
6 changed files with 14 additions and 60 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 4
      src/NativeWrappers/Common/wrapper_common.h
  2. 49
      src/NativeWrappers/MKL/lapack.cpp
  3. 8
      src/NativeWrappers/MKL/lapack.h
  4. 9
      src/NativeWrappers/Windows/MKL/MKLWrapper.vcxproj
  5. 3
      src/NativeWrappers/Windows/MKL/MKLWrapper.vcxproj.filters
  6. 1
      src/NativeWrappers/Windows/NativeWrappers.sln

4
src/NativeWrappers/Common/wrapper_common.h
View File

@ -1,5 +1,5 @@
#ifndef COMMON_H
#define COMMON_H
#ifndef WRAPPER_COMMON_H
#define WRAPPER_COMMON_H
#ifdef _WINDOWS
#define DLLEXPORT __declspec( dllexport )

49
src/NativeWrappers/MKL/lapack.cpp
View File

@ -1,31 +1,9 @@
#include "mkl_lapack.h"
#include "mkl_cblas.h"
#include "lapack_common.h"
#include "wrapper_common.h"
#include <algorithm>
template<typename T>
inline void copyBtoX (MKL_INT m, MKL_INT n, MKL_INT bn, T b[], T x[]){
for (MKL_INT i = 0; i < n; ++i)
{
for (MKL_INT j = 0; j < bn; ++j)
{
x[j * n + i] = clone_b[j * m + i];
}
}
};
inline void shift_ipiv_down(MKL_INT m, MKL_INT ipiv[]){
for(MKL_INT i = 0; i < m; ++i ){
ipiv[i] -= 1;
}
}
inline void shift_ipiv_up(MKL_INT m, MKL_INT ipiv[]){
for(MKL_INT i = 0; i < m; ++i ){
ipiv[i] += 1;
}
}
template<typename T>
inline MKL_INT lu_factor(MKL_INT m, T a[], MKL_INT ipiv[],
void (*getrf) (const MKL_INT*, const MKL_INT*, T*, const MKL_INT*, MKL_INT*, MKL_INT*))
@ -83,9 +61,7 @@ inline MKL_INT lu_solve(MKL_INT n, MKL_INT nrhs, T a[], T b[],
void (*getrf) (const MKL_INT*, const MKL_INT*, T*, const MKL_INT*, MKL_INT*, MKL_INT*),
void (*getrs) (const char*, const MKL_INT*, const MKL_INT*, const T*, const MKL_INT*, const MKL_INT*, T*, const MKL_INT*, MKL_INT*))
{
T* clone = new T[n*n];
std::memcpy(clone, a, n*n*sizeof(T));
T* clone = Clone(n, n, a);
MKL_INT* ipiv = new MKL_INT[n];
MKL_INT info = 0;
getrf(&n, &n, clone, &n, ipiv, &info);
@ -128,8 +104,7 @@ inline MKL_INT cholesky_solve(MKL_INT n, MKL_INT nrhs, T a[], T b[],
void (*potrf) (const char*, const MKL_INT*, T*, const MKL_INT*, MKL_INT*),
void (*potrs) (const char*, const MKL_INT*, const MKL_INT*, const T*, const MKL_INT*, T*, const MKL_INT*, MKL_INT*))
{
T* clone = new T[n*n];
std::memcpy(clone, a, n*n*sizeof(T));
T* clone = Clone(n, n, a);
char uplo = 'L';
MKL_INT info = 0;
potrf(&uplo, &n, clone, &n, &info);
@ -214,11 +189,8 @@ inline MKL_INT qr_solve(MKL_INT m, MKL_INT n, MKL_INT bn, T a[], T b[], T x[], T
void (*gels) (const char*, const MKL_INT*, const MKL_INT*, const MKL_INT*, T*,
const MKL_INT*, T* b, const MKL_INT*, T*, const MKL_INT*, MKL_INT*))
{
T* clone_a = new T[m*n];
std::memcpy(clone_a, a, m*n*sizeof(T));
T* clone_b = new T[m*bn];
std::memcpy(clone_b, b, m*bn*sizeof(T));
T* clone_a = Clone(m, n, a);
T* clone_b = Clone(m, bn, b);
char N = 'N';
MKL_INT info = 0;
@ -237,8 +209,6 @@ inline MKL_INT qr_solve(MKL_INT m, MKL_INT n, MKL_INT bn, T a[], T b[], T x[], T
return info;
}
// combine the next two some how
// the problem is that complex trsm takes void* instead of MKL_COMPLEX
template<typename T>
inline MKL_INT qr_solve_factored(MKL_INT m, MKL_INT n, MKL_INT bn, T r[], T b[], T tau[], T x[], T work[], MKL_INT len,
void (*ormqr) (const char*, const char*, const MKL_INT*, const MKL_INT*, const MKL_INT*,
@ -246,9 +216,7 @@ inline MKL_INT qr_solve_factored(MKL_INT m, MKL_INT n, MKL_INT bn, T r[], T b[],
void (*trsm) (const CBLAS_ORDER, const CBLAS_SIDE, const CBLAS_UPLO, const CBLAS_TRANSPOSE, const CBLAS_DIAG,
const MKL_INT, const MKL_INT, const T, const T*, const MKL_INT, T*, const MKL_INT))
{
T* clone_b = new T[m*bn];
std::memcpy(clone_b, b, m*bn*sizeof(T));
T* clone_b = Clone(m, bn, b);
char side ='L';
char tran = 'T';
MKL_INT info = 0;
@ -273,9 +241,7 @@ inline MKL_INT complex_qr_solve_factored(MKL_INT m, MKL_INT n, MKL_INT bn, T r[]
void (*trsm) (const CBLAS_ORDER, const CBLAS_SIDE, const CBLAS_UPLO, const CBLAS_TRANSPOSE, const CBLAS_DIAG,
const MKL_INT, const MKL_INT, const void*, const void*, const MKL_INT, void*, const MKL_INT ldb))
{
T* clone_b = new T[m*bn];
std::memcpy(clone_b, b, m*bn*sizeof(T));
T* clone_b = Clone(m, bn, b);
char side ='L';
char tran = 'C';
MKL_INT info = 0;
@ -306,7 +272,6 @@ inline MKL_INT svd_factor(bool compute_vectors, MKL_INT m, MKL_INT n, T a[], T s
return info;
}
template<typename T, typename R>
inline MKL_INT complex_svd_factor(bool compute_vectors, MKL_INT m, MKL_INT n, T a[], T s[], T u[], T v[], T work[], MKL_INT len,
void (*gesvd) (const char*, const char*, const MKL_INT*, const MKL_INT*, T*, const MKL_INT*,

8
src/NativeWrappers/MKL/lapack.h
View File

@ -1,8 +0,0 @@
#ifndef LAPACK_H
#define LAPACK_H
//#include "blas.h"
#include "mkl_lapack.h"
#endif

9
src/NativeWrappers/Windows/MKL/MKLWrapper.vcxproj
View File

@ -104,7 +104,7 @@
<TargetMachine>MachineX86</TargetMachine>
</Link>
<PostBuildEvent>
<Command>copy "$(CompilerPathForVC)\libiomp5md.dll" $(OutDir)</Command>
<Command>copy "$(INTEL_DEV_REDIST)\redist\$(IntelPlatform)\compiler\libiomp5md.dll" $(OutDir)</Command>
</PostBuildEvent>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
@ -130,7 +130,7 @@
<TargetMachine>MachineX64</TargetMachine>
</Link>
<PostBuildEvent>
<Command>copy "$(CompilerPathForVC)\libiomp5md.dll" $(OutDir)</Command>
<Command>copy "$(INTEL_DEV_REDIST)\redist\$(IntelPlatform)\compiler\libiomp5md.dll" $(OutDir)</Command>
</PostBuildEvent>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
@ -155,7 +155,7 @@
<TargetMachine>MachineX86</TargetMachine>
</Link>
<PostBuildEvent>
<Command>copy "$(CompilerPathForVC)\libiomp5md.dll" $(OutDir)</Command>
<Command>copy "$(INTEL_DEV_REDIST)\redist\$(IntelPlatform)\compiler\libiomp5md.dll" $(OutDir)</Command>
</PostBuildEvent>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
@ -183,7 +183,7 @@
<TargetMachine>MachineX64</TargetMachine>
</Link>
<PostBuildEvent>
<Command>copy "$(CompilerPathForVC)\libiomp5md.dll" $(OutDir)</Command>
<Command>copy "$(INTEL_DEV_REDIST)\redist\$(IntelPlatform)\compiler\libiomp5md.dll" $(OutDir)</Command>
</PostBuildEvent>
</ItemDefinitionGroup>
<ItemGroup>
@ -194,7 +194,6 @@
</ItemGroup>
<ItemGroup>
<ClInclude Include="..\..\Common\wrapper_common.h" />
<ClInclude Include="..\..\MKL\lapack.h" />
</ItemGroup>
<ItemGroup>
<ResourceCompile Include="..\..\Common\resource.rc" />

3
src/NativeWrappers/Windows/MKL/MKLWrapper.vcxproj.filters
View File

@ -29,9 +29,6 @@
</ClCompile>
</ItemGroup>
<ItemGroup>
<ClInclude Include="..\..\MKL\lapack.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\..\Common\wrapper_common.h">
<Filter>Header Files</Filter>
</ClInclude>

1
src/NativeWrappers/Windows/NativeWrappers.sln
View File

@ -3,6 +3,7 @@ Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2012
Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "Common", "Common", "{5A0892FF-82CE-40FC-BCE1-73810C615F52}"
ProjectSection(SolutionItems) = preProject
..\Common\lapack_common.h = ..\Common\lapack_common.h
..\Common\resource.h = ..\Common\resource.h
..\Common\resource.rc = ..\Common\resource.rc
..\Common\WindowsDLL.cpp = ..\Common\WindowsDLL.cpp

Write Preview
Loading…
Cancel
Save