portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-03 05:24:32 +00:00
Code Releases Activity

Partial implementation of the vector types SIMD

Browse Source 
Implementing SSE4 now
A systematic series of tests must be written.
This commit is contained in:
neo
2015-05-19 17:21:17 +09:00
parent 4cadf11d1d
commit b29caead32
12 changed files with 516 additions and 145 deletions
Download Patch File Download Diff File Expand all files Collapse all files

127
lib/Grid_config.h Normal file
View File

@@ -0,0 +1,127 @@
/* lib/Grid_config.h. Generated from Grid_config.h.in by configure. */
/* lib/Grid_config.h.in. Generated from configure.ac by autoheader. */
/* AVX */
/* #undef AVX1 */
/* AVX2 */
/* #undef AVX2 */
/* AVX512 */
/* #undef AVX512 */
/* GRID_COMMS_MPI */
/* #undef GRID_COMMS_MPI */
/* GRID_COMMS_NONE */
#define GRID_COMMS_NONE 1
/* Define to 1 if you have the declaration of `be64toh', and to 0 if you
don't. */
#define HAVE_DECL_BE64TOH 1
/* Define to 1 if you have the declaration of `ntohll', and to 0 if you don't.
*/
#define HAVE_DECL_NTOHLL 0
/* Define to 1 if you have the <endian.h> header file. */
#define HAVE_ENDIAN_H 1
/* Define to 1 if you have the `gettimeofday' function. */
#define HAVE_GETTIMEOFDAY 1
/* Define to 1 if you have the <gmp.h> header file. */
#define HAVE_GMP_H 1
/* Define to 1 if you have the <inttypes.h> header file. */
#define HAVE_INTTYPES_H 1
/* Define to 1 if you have the `gmp' library (-lgmp). */
#define HAVE_LIBGMP 1
/* Define to 1 if you have the `mpfr' library (-lmpfr). */
#define HAVE_LIBMPFR 1
/* Define to 1 if you have the <malloc.h> header file. */
#define HAVE_MALLOC_H 1
/* Define to 1 if you have the <malloc/malloc.h> header file. */
/* #undef HAVE_MALLOC_MALLOC_H */
/* Define to 1 if you have the <memory.h> header file. */
#define HAVE_MEMORY_H 1
/* Define to 1 if you have the <mm_malloc.h> header file. */
#define HAVE_MM_MALLOC_H 1
/* Define to 1 if you have the <stdint.h> header file. */
#define HAVE_STDINT_H 1
/* Define to 1 if you have the <stdlib.h> header file. */
#define HAVE_STDLIB_H 1
/* Define to 1 if you have the <strings.h> header file. */
#define HAVE_STRINGS_H 1
/* Define to 1 if you have the <string.h> header file. */
#define HAVE_STRING_H 1
/* Define to 1 if you have the <sys/stat.h> header file. */
#define HAVE_SYS_STAT_H 1
/* Define to 1 if you have the <sys/types.h> header file. */
#define HAVE_SYS_TYPES_H 1
/* Define to 1 if you have the <unistd.h> header file. */
#define HAVE_UNISTD_H 1
/* Name of package */
#define PACKAGE "grid"
/* Define to the address where bug reports for this package should be sent. */
#define PACKAGE_BUGREPORT "paboyle@ph.ed.ac.uk"
/* Define to the full name of this package. */
#define PACKAGE_NAME "Grid"
/* Define to the full name and version of this package. */
#define PACKAGE_STRING "Grid 1.0"
/* Define to the one symbol short name of this package. */
#define PACKAGE_TARNAME "grid"
/* Define to the home page for this package. */
#define PACKAGE_URL ""
/* Define to the version of this package. */
#define PACKAGE_VERSION "1.0"
/* SSE4 */
#define SSE4 1
/* Define to 1 if you have the ANSI C header files. */
#define STDC_HEADERS 1
/* Version number of package */
#define VERSION "1.0"
/* Define for Solaris 2.5.1 so the uint32_t typedef from <sys/synch.h>,
<pthread.h>, or <semaphore.h> is not used. If the typedef were allowed, the
#define below would cause a syntax error. */
/* #undef _UINT32_T */
/* Define for Solaris 2.5.1 so the uint64_t typedef from <sys/synch.h>,
<pthread.h>, or <semaphore.h> is not used. If the typedef were allowed, the
#define below would cause a syntax error. */
/* #undef _UINT64_T */
/* Define to `unsigned int' if <sys/types.h> does not define. */
/* #undef size_t */
/* Define to the type of an unsigned integer type of width exactly 32 bits if
such a type exists and the standard includes do not define it. */
/* #undef uint32_t */
/* Define to the type of an unsigned integer type of width exactly 64 bits if
such a type exists and the standard includes do not define it. */
/* #undef uint64_t */

158
lib/Makefile.am
View File

@@ -14,87 +14,89 @@ endif
# Libraries
#
lib_LIBRARIES = libGrid.a
libGrid_a_SOURCES =\
Grid_init.cc\
stencil/Grid_stencil_common.cc\
qcd/Grid_qcd_dirac.cc\
qcd/Grid_qcd_wilson_dop.cc\
algorithms/approx/Zolotarev.cc\
algorithms/approx/Remez.cc\
libGrid_a_SOURCES = \
Grid_init.cc \
stencil/Grid_stencil_common.cc \
qcd/Grid_qcd_dirac.cc \
qcd/Grid_qcd_wilson_dop.cc \
algorithms/approx/Zolotarev.cc \
algorithms/approx/Remez.cc \
$(extra_sources)
#
# Include files
#
nobase_include_HEADERS = algorithms/approx/bigfloat.h\
algorithms/approx/Chebyshev.h\
algorithms/approx/Remez.h\
algorithms/approx/Zolotarev.h\
algorithms/iterative/ConjugateGradient.h\
algorithms/iterative/NormalEquations.h\
algorithms/iterative/SchurRedBlack.h\
algorithms/LinearOperator.h\
algorithms/SparseMatrix.h\
cartesian/Grid_cartesian_base.h\
cartesian/Grid_cartesian_full.h\
cartesian/Grid_cartesian_red_black.h\
communicator/Grid_communicator_base.h\
cshift/Grid_cshift_common.h\
cshift/Grid_cshift_mpi.h\
cshift/Grid_cshift_none.h\
Grid.h\
Grid_algorithms.h\
Grid_aligned_allocator.h\
Grid_cartesian.h\
Grid_communicator.h\
Grid_comparison.h\
Grid_cshift.h\
Grid_extract.h\
Grid_lattice.h\
Grid_math.h\
Grid_simd.h\
Grid_stencil.h\
Grid_threads.h\
lattice/Grid_lattice_arith.h\
lattice/Grid_lattice_base.h\
lattice/Grid_lattice_comparison.h\
lattice/Grid_lattice_conformable.h\
lattice/Grid_lattice_coordinate.h\
lattice/Grid_lattice_ET.h\
lattice/Grid_lattice_local.h\
lattice/Grid_lattice_overload.h\
lattice/Grid_lattice_peekpoke.h\
lattice/Grid_lattice_reality.h\
lattice/Grid_lattice_reduction.h\
lattice/Grid_lattice_rng.h\
lattice/Grid_lattice_trace.h\
lattice/Grid_lattice_transfer.h\
lattice/Grid_lattice_transpose.h\
lattice/Grid_lattice_where.h\
math/Grid_math_arith.h\
math/Grid_math_arith_add.h\
math/Grid_math_arith_mac.h\
math/Grid_math_arith_mul.h\
math/Grid_math_arith_scalar.h\
math/Grid_math_arith_sub.h\
math/Grid_math_inner.h\
math/Grid_math_outer.h\
math/Grid_math_peek.h\
math/Grid_math_poke.h\
math/Grid_math_reality.h\
math/Grid_math_tensors.h\
math/Grid_math_trace.h\
math/Grid_math_traits.h\
math/Grid_math_transpose.h\
parallelIO/GridNerscIO.h\
qcd/Grid_qcd.h\
qcd/Grid_qcd_2spinor.h\
qcd/Grid_qcd_dirac.h\
qcd/Grid_qcd_wilson_dop.h\
simd/Grid_vComplexD.h\
simd/Grid_vComplexF.h\
simd/Grid_vInteger.h\
simd/Grid_vRealD.h\
simd/Grid_vRealF.h\
simd/Grid_vector_types.h
nobase_include_HEADERS = algorithms/approx/bigfloat.h \
algorithms/approx/Chebyshev.h \
algorithms/approx/Remez.h \
algorithms/approx/Zolotarev.h \
algorithms/iterative/ConjugateGradient.h \
algorithms/iterative/NormalEquations.h \
algorithms/iterative/SchurRedBlack.h \
algorithms/LinearOperator.h \
algorithms/SparseMatrix.h \
cartesian/Grid_cartesian_base.h \
cartesian/Grid_cartesian_full.h \
cartesian/Grid_cartesian_red_black.h \
communicator/Grid_communicator_base.h \
cshift/Grid_cshift_common.h \
cshift/Grid_cshift_mpi.h \
cshift/Grid_cshift_none.h \
Grid.h \
Grid_algorithms.h \
Grid_aligned_allocator.h \
Grid_cartesian.h \
Grid_communicator.h \
Grid_comparison.h \
Grid_cshift.h \
Grid_extract.h \
Grid_lattice.h \
Grid_math.h \
Grid_simd.h \
Grid_stencil.h \
Grid_threads.h \
lattice/Grid_lattice_arith.h \
lattice/Grid_lattice_base.h \
lattice/Grid_lattice_comparison.h \
lattice/Grid_lattice_conformable.h \
lattice/Grid_lattice_coordinate.h \
lattice/Grid_lattice_ET.h \
lattice/Grid_lattice_local.h \
lattice/Grid_lattice_overload.h \
lattice/Grid_lattice_peekpoke.h \
lattice/Grid_lattice_reality.h \
lattice/Grid_lattice_reduction.h \
lattice/Grid_lattice_rng.h \
lattice/Grid_lattice_trace.h \
lattice/Grid_lattice_transfer.h \
lattice/Grid_lattice_transpose.h \
lattice/Grid_lattice_where.h \
math/Grid_math_arith.h \
math/Grid_math_arith_add.h \
math/Grid_math_arith_mac.h \
math/Grid_math_arith_mul.h \
math/Grid_math_arith_scalar.h \
math/Grid_math_arith_sub.h \
math/Grid_math_inner.h \
math/Grid_math_outer.h \
math/Grid_math_peek.h \
math/Grid_math_poke.h \
math/Grid_math_reality.h \
math/Grid_math_tensors.h \
math/Grid_math_trace.h \
math/Grid_math_traits.h \
math/Grid_math_transpose.h \
parallelIO/GridNerscIO.h \
qcd/Grid_qcd.h \
qcd/Grid_qcd_2spinor.h \
qcd/Grid_qcd_dirac.h \
qcd/Grid_qcd_wilson_dop.h \
simd/Grid_vComplexD.h \
simd/Grid_vComplexF.h \
simd/Grid_vInteger.h \
simd/Grid_vRealD.h \
simd/Grid_vRealF.h \
simd/Grid_vector_types.h \
simd/Grid_sse4.h

0
lib/algorithms/approx/.dirstamp Normal file
View File

0
lib/communicator/.dirstamp Normal file
View File

0
lib/qcd/.dirstamp Normal file
View File

0
lib/simd/.dirstamp Normal file
View File

19
lib/simd/Grid_sse4.cpp
View File

@@ -1,19 +0,0 @@
//----------------------------------------------------------------------
/*! @file Grid_vector_types.h
@brief Defines templated class to deal with inner vector types
*/
// Time-stamp: <2015-05-19 13:53:47 neo>
//----------------------------------------------------------------------
namespace Optimization {
}
// Here assign types
namespace Grid {
}

194
lib/simd/Grid_sse4.h Normal file
View File

@@ -0,0 +1,194 @@
//----------------------------------------------------------------------
/*! @file Grid_sse4.h
@brief Optimization libraries
*/
// Time-stamp: <2015-05-19 17:06:51 neo>
//----------------------------------------------------------------------
#include <pmmintrin.h>
namespace Optimization {
struct Vsplat{
//Complex float
inline __m128 operator()(float a, float b){
return _mm_set_ps(b,a,b,a);
}
// Real float
inline __m128 operator()(float a){
return _mm_set_ps(a,a,a,a);
}
//Complex double
inline __m128d operator()(double a, double b){
return _mm_set_pd(b,a);
}
//Real double
inline __m128d operator()(double a){
return _mm_set_pd(a,a);
}
//Integer
inline __m128i operator()(Integer a){
return _mm_set1_epi32(a);
}
};
struct Vstore{
//Float
inline void operator()(__m128 a, float* F){
_mm_store_ps(F,a);
}
//Double
inline void operator()(__m128d a, double* D){
_mm_store_pd(D,a);
}
//Integer
inline void operator()(__m128i a, Integer* I){
_mm_store_si128((__m128i *)I,a);
}
};
struct Vset{
// Complex float
inline __m128 operator()(Grid::ComplexF *a){
return _mm_set_ps(a[1].imag(), a[1].real(),a[0].imag(),a[0].real());
}
// Complex double
inline __m128d operator()(Grid::ComplexD *a){
return _mm_set_pd(a[0].imag(),a[0].real());
}
// Real float
inline __m128 operator()(float *a){
return _mm_set_ps(a[3],a[2],a[1],a[0]);
}
// Real double
inline __m128d operator()(double *a){
return _mm_set_pd(a[1],a[0]);
}
// Integer
inline __m128i operator()(Integer *a){
return _mm_set_epi32(a[0],a[1],a[2],a[3]);
}
};
struct Reduce{
//Complex float
inline Grid::ComplexF operator()(__m128 in){
union {
__m128 v1;
float f[4];
} u128;
u128.v1 = _mm_add_ps(in, _mm_shuffle_ps(in,in, 0b01001110)); // FIXME Prefer to use _MM_SHUFFLE macros
return Grid::ComplexF(u128.f[0], u128.f[1]);
}
//Complex double
inline Grid::ComplexD operator()(__m128d in){
printf("Missing complex double implementation -> FIX\n");
return Grid::ComplexD(0,0); // FIXME wrong
}
};
/////////////////////////////////////////////////////
// Arithmetic operations
/////////////////////////////////////////////////////
struct Sum{
//Complex/Real float
inline __m128 operator()(__m128 a, __m128 b){
return _mm_add_ps(a,b);
}
//Complex/Real double
inline __m128d operator()(__m128d a, __m128d b){
return _mm_add_pd(a,b);
}
//Integer
inline __m128i operator()(__m128i a, __m128i b){
return _mm_add_epi32(a,b);
}
};
struct Sub{
//Complex/Real float
inline __m128 operator()(__m128 a, __m128 b){
return _mm_sub_ps(a,b);
}
//Complex/Real double
inline __m128d operator()(__m128d a, __m128d b){
return _mm_sub_pd(a,b);
}
//Integer
inline __m128i operator()(__m128i a, __m128i b){
return _mm_sub_epi32(a,b);
}
};
struct MultComplex{
// Complex float
inline __m128 operator()(__m128 a, __m128 b){
__m128 ymm0,ymm1,ymm2;
ymm0 = _mm_shuffle_ps(a,a,_MM_SHUFFLE(2,2,0,0)); // ymm0 <- ar ar,
ymm0 = _mm_mul_ps(ymm0,b); // ymm0 <- ar bi, ar br
ymm1 = _mm_shuffle_ps(b,b,_MM_SHUFFLE(2,3,0,1)); // ymm1 <- br,bi
ymm2 = _mm_shuffle_ps(a,a,_MM_SHUFFLE(3,3,1,1)); // ymm2 <- ai,ai
ymm1 = _mm_mul_ps(ymm1,ymm2); // ymm1 <- br ai, ai bi
return _mm_addsub_ps(ymm0,ymm1);
}
// Complex double
inline __m128d operator()(__m128d a, __m128d b){
__m128d ymm0,ymm1,ymm2;
ymm0 = _mm_shuffle_pd(a,a,0x0); // ymm0 <- ar ar,
ymm0 = _mm_mul_pd(ymm0,b); // ymm0 <- ar bi, ar br
ymm1 = _mm_shuffle_pd(b,b,0x1); // ymm1 <- br,bi b01
ymm2 = _mm_shuffle_pd(a,a,0x3); // ymm2 <- ai,ai b11
ymm1 = _mm_mul_pd(ymm1,ymm2); // ymm1 <- br ai, ai bi
return _mm_addsub_pd(ymm0,ymm1);
}
};
struct Mult{
// Real float
inline __m128 operator()(__m128 a, __m128 b){
return _mm_mul_ps(a,b);
}
// Real double
inline __m128d operator()(__m128d a, __m128d b){
return _mm_mul_pd(a,b);
}
// Integer
inline __m128i operator()(__m128i a, __m128i b){
return _mm_mul_epi32(a,b);
}
};
}
// Here assign types
namespace Grid {
typedef __m128 SIMD_Ftype; // Single precision type
typedef __m128d SIMD_Dtype; // Double precision type
typedef __m128i SIMD_Itype; // Integer type
// Function names
typedef Optimization::Vsplat VsplatSIMD;
typedef Optimization::Vstore VstoreSIMD;
// Arithmetic operations
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Vset VsetSIMD;
}

125
lib/simd/Grid_vector_types.h
View File

@@ -1,12 +1,14 @@
//----------------------------------------------------------------------
//---------------------------------------------------------------------------
/*! @file Grid_vector_types.h
@brief Defines templated class to deal with inner vector types
@brief Defines templated class Grid_simd to deal with inner vector types
*/
// Time-stamp: <2015-05-19 13:41:47 neo>
//----------------------------------------------------------------------
// Time-stamp: <2015-05-19 17:20:36 neo>
//---------------------------------------------------------------------------
#ifndef GRID_VECTOR_TYPES
#define GRID_VECTOR_TYPES
#include "Grid_sse4.h"
namespace Grid {
@@ -27,18 +29,20 @@ namespace Grid {
template < typename T >
struct is_complex< std::complex<T> >: std::true_type {};
////////////////////////////////////////////////////////
// Define the operation templates functors
template < class SIMD, class Operation >
SIMD binary(SIMD src_1, SIMD src_2, Operation op){
// general forms to allow for vsplat syntax
// need explicit declaration of types when used since
// clang cannot automatically determine the output type sometimes
template < class Out, class Input1, class Input2, class Operation >
Out binary(Input1 src_1, Input2 src_2, Operation op){
return op(src_1, src_2);
}
}
template < class SIMD, class Operation >
SIMD unary(SIMD src, Operation op){
template < class SIMDout, class Input, class Operation >
SIMDout unary(Input src, Operation op){
return op(src);
}
}
///////////////////////////////////////////////
/*
@@ -74,36 +78,42 @@ namespace Grid {
};
///////////////////////////////////////////////
// mac, mult, sub, add, adj
// Should do an AVX2 version with mac.
///////////////////////////////////////////////
friend inline void mac (Grid_simd * __restrict__ y,const Grid_simd * __restrict__ a,const Grid_simd *__restrict__ x){ *y = (*a)*(*x)+(*y); };
friend inline void mult(Grid_simd * __restrict__ y,const Grid_simd * __restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) * (*r); }
friend inline void sub (Grid_simd * __restrict__ y,const Grid_simd * __restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) - (*r); }
friend inline void add (Grid_simd * __restrict__ y,const Grid_simd * __restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) + (*r); }
//not for integer types... FIXME
friend inline Grid_simd adj(const Grid_simd &in){ return conj(in); }
//////////////////////////////////
// Initialise to 1,0,i
//////////////////////////////////
///////////////////////////////////////////////
// Initialise to 1,0,i for the correct types
///////////////////////////////////////////////
// if not complex overload here
friend inline void vone(Grid_simd &ret) { vsplat(ret,1.0); }
friend inline void vzero(Grid_simd &ret) { vsplat(ret,0.0); }
template < class S = Scalar_type,typename std::enable_if < !is_complex < S >::value, int >::type = 0 >
friend inline void vone(Grid_simd &ret) { vsplat(ret,1.0); }
template < class S = Scalar_type,typename std::enable_if < !is_complex < S >::value, int >::type = 0 >
friend inline void vzero(Grid_simd &ret) { vsplat(ret,0.0); }
// overload for complex type
template < class S = Scalar_type,typename std::enable_if < is_complex < S >::value, int >::type = 0 >
friend inline void vone(Grid_simd &ret) { vsplat(ret,1.0,0.0); }
friend inline void vone(Grid_simd &ret) { vsplat(ret,1.0,0.0); }
template < class S = Scalar_type,typename std::enable_if < is_complex < S >::value, int >::type = 0 >
friend inline void vzero(Grid_simd &ret) { vsplat(ret,0.0,0.0); }
friend inline void vzero(Grid_simd &ret) { vsplat(ret,0.0,0.0); }// use xor?
// For integral type
template < class S = Scalar_type,typename std::enable_if < std::is_integral < S >::value, int >::type = 0 >
friend inline void vone(Grid_simd &ret) { vsplat(ret,1); }
friend inline void vone(Grid_simd &ret) { vsplat(ret,1); }
template < class S = Scalar_type,typename std::enable_if < std::is_integral < S >::value, int >::type = 0 >
friend inline void vzero(Grid_simd &ret) { vsplat(ret,0); }
friend inline void vzero(Grid_simd &ret) { vsplat(ret,0); }
template < class S = Scalar_type,typename std::enable_if < std::is_integral < S >::value, int >::type = 0 >
friend inline void vtrue (Grid_simd &ret){vsplat(ret,0xFFFFFFFF);}
template < class S = Scalar_type,typename std::enable_if < std::is_integral < S >::value, int >::type = 0 >
friend inline void vfalse(vInteger &ret){vsplat(ret,0);}
// do not compile if real or integer, send an error message from the compiler
template < class S = Scalar_type,typename std::enable_if < is_complex < S >::value, int >::type = 0 >
@@ -114,31 +124,44 @@ namespace Grid {
////////////////////////////////////
friend inline Grid_simd operator + (Grid_simd a, Grid_simd b)
{
vComplexF ret;
// FIXME call the binary op
Grid_simd ret;
ret.v = binary<Vector_type>(a.v, b.v, SumSIMD());
return ret;
};
friend inline Grid_simd operator - (Grid_simd a, Grid_simd b)
{
vComplexF ret;
// FIXME call the binary op
Grid_simd ret;
ret.v = binary<Vector_type>(a.v, b.v, SubSIMD());
return ret;
};
friend inline Grid_simd operator * (Grid_simd a, Grid_simd b)
// Distinguish between complex types and others
template < class S = Scalar_type, typename std::enable_if < is_complex < S >::value, int >::type = 0 >
friend inline Grid_simd operator * (Grid_simd a, Grid_simd b)
{
vComplexF ret;
// FIXME call the binary op
Grid_simd ret;
ret.v = binary<Vector_type>(a.v,b.v, MultComplexSIMD());
return ret;
};
// Real/Integer types
template < class S = Scalar_type,typename std::enable_if < !is_complex < S >::value, int >::type = 0 >
friend inline Grid_simd operator * (Grid_simd a, Grid_simd b)
{
Grid_simd ret;
ret.v = binary<Vector_type>(a.v,b.v, MultSIMD());
return ret;
};
////////////////////////////////////////////////////////////////////////
// FIXME: gonna remove these load/store, get, set, prefetch
////////////////////////////////////////////////////////////////////////
friend inline void vset(Grid_simd &ret, Scalar_type *a){
// FIXME set
ret.v = unary<Vector_type>(a, VsetSIMD());
}
///////////////////////
@@ -147,34 +170,33 @@ namespace Grid {
// overload if complex
template < class S = Scalar_type >
friend inline void vsplat(Grid_simd &ret, typename std::enable_if< is_complex < S >::value, S>::type c){
Real a= real(c);
Real b= imag(c);
Real a = real(c);
Real b = imag(c);
vsplat(ret,a,b);
}
// this only for the complex version
template < class S = Scalar_type, typename std::enable_if < is_complex < S >::value, int >::type = 0 >
friend inline void vsplat(Grid_simd &ret,Real a, Real b){
// FIXME add operator
ret.v = binary<Vector_type>(a, b, VsplatSIMD());
}
//if real fill with a, if complex fill with a in the real part
//if real fill with a, if complex fill with a in the real part (first function above)
friend inline void vsplat(Grid_simd &ret,Real a){
// FIXME add operator
ret.v = unary<Vector_type>(a, VsplatSIMD());
}
friend inline void vstore(const Grid_simd &ret, Scalar_type *a){
//FIXME
binary<void>(ret.v, (Real*)a, VstoreSIMD());
}
friend inline void vprefetch(const Grid_simd &v)
{
_mm_prefetch((const char*)&v.v,_MM_HINT_T0);
}
friend inline Scalar_type Reduce(const Grid_simd & in)
{
// FIXME add operator
@@ -221,6 +243,7 @@ namespace Grid {
inline Grid_simd &operator *=(const Grid_simd &r) {
*this = (*this)*r;
return *this;
// return (*this)*r; ?
}
inline Grid_simd &operator +=(const Grid_simd &r) {
*this = *this+r;
@@ -233,6 +256,12 @@ namespace Grid {
friend inline void permute(Grid_simd &y,Grid_simd b,int perm)
{
Gpermute<Grid_simd>(y,b,perm);
}
/*
friend inline void permute(Grid_simd &y,Grid_simd b,int perm)
{
Gpermute<Grid_simd>(y,b,perm);
@@ -253,7 +282,7 @@ namespace Grid {
{
Gextract<Grid_simd,Scalar_type>(y,extracted);
}
*/
};// end of Grid_simd class definition
@@ -286,11 +315,11 @@ namespace Grid {
// Define available types (now change names to avoid clashing)
typedef __m128 SIMD_type;// decided at compilation time
typedef Grid_simd< float , SIMD_type > MyRealF;
typedef Grid_simd< double , SIMD_type > MyRealD;
typedef Grid_simd< std::complex< float > , SIMD_type > MyComplexF;
typedef Grid_simd< std::complex< double >, SIMD_type > MyComplexD;
typedef Grid_simd< float , SIMD_Ftype > MyRealF;
typedef Grid_simd< double , SIMD_Dtype > MyRealD;
typedef Grid_simd< std::complex< float > , SIMD_Ftype > MyComplexF;
typedef Grid_simd< std::complex< double >, SIMD_Dtype > MyComplexD;

1
lib/stamp-h1 Normal file
View File

@@ -0,0 +1 @@
timestamp for lib/Grid_config.h

0
lib/stencil/.dirstamp Normal file
View File