devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

merge

Browse Source
This commit is contained in:
Gael Guennebaud
2016-06-14 15:33:47 +02:00
parent 1004c4df99 4c61f00838
commit 76236cdea4
28 changed files with 376 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/Array.h
View File

@@ -149,7 +149,7 @@ class Array
#if EIGEN_HAS_RVALUE_REFERENCES
EIGEN_DEVICE_FUNC
Array(Array&& other)
Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
: Base(std::move(other))
{
Base::_check_template_params();
@@ -157,7 +157,7 @@ class Array
Base::_set_noalias(other);
}
EIGEN_DEVICE_FUNC
Array& operator=(Array&& other)
Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
{
other.swap(*this);
return *this;

2
Eigen/src/Core/arch/CUDA/PacketMathHalf.h
View File

@@ -28,6 +28,8 @@ template<> struct packet_traits<Eigen::half> : default_packet_traits
AlignedOnScalar = 1,
size=2,
HasHalfPacket = 0,
HasAdd = 1,
HasMul = 1,
HasDiv = 1,
HasSqrt = 1,
HasRsqrt = 1,

8
Eigen/src/Core/arch/NEON/Complex.h
View File

@@ -14,8 +14,9 @@ namespace Eigen {
namespace internal {
static uint32x4_t p4ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET4(0x00000000, 0x80000000, 0x00000000, 0x80000000);
static uint32x2_t p2ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET2(0x00000000, 0x80000000);
const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
static uint32x4_t p4ui_CONJ_XOR = vld1q_u32( conj_XOR_DATA );
static uint32x2_t p2ui_CONJ_XOR = vld1_u32( conj_XOR_DATA );
//---------- float ----------
struct Packet2cf
@@ -274,7 +275,8 @@ ptranspose(PacketBlock<Packet2cf,2>& kernel) {
//---------- double ----------
#if EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
static uint64x2_t p2ul_CONJ_XOR = EIGEN_INIT_NEON_PACKET2(0x0, 0x8000000000000000);
const uint64_t p2ul_conj_XOR_DATA[] = { 0x0, 0x8000000000000000 };
static uint64x2_t p2ul_CONJ_XOR = vld1q_u64( p2ul_conj_XOR_DATA );
struct Packet1cd
{

20
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@@ -49,17 +49,6 @@ typedef uint32x4_t Packet4ui;
#define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
const Packet4i p4i_##NAME = pset1<Packet4i>(X)
#if EIGEN_COMP_LLVM && !EIGEN_COMP_CLANG
//Special treatment for Apple's llvm-gcc, its NEON packet types are unions
#define EIGEN_INIT_NEON_PACKET2(X, Y) {{X, Y}}
#define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {{X, Y, Z, W}}
#else
//Default initializer for packets
#define EIGEN_INIT_NEON_PACKET2(X, Y) {X, Y}
#define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {X, Y, Z, W}
#endif
// arm64 does have the pld instruction. If available, let's trust the __builtin_prefetch built-in function
// which available on LLVM and GCC (at least)
#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
@@ -122,12 +111,14 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int& from) {
template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)
{
Packet4f countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
const float32_t f[] = {0, 1, 2, 3};
Packet4f countdown = vld1q_f32(f);
return vaddq_f32(pset1<Packet4f>(a), countdown);
}
template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)
{
Packet4i countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
const int32_t i[] = {0, 1, 2, 3};
Packet4i countdown = vld1q_s32(i);
return vaddq_s32(pset1<Packet4i>(a), countdown);
}
@@ -585,7 +576,8 @@ template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { r
template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a)
{
Packet2d countdown = EIGEN_INIT_NEON_PACKET2(0, 1);
const double countdown_raw[] = {0.0,1.0};
const Packet2d countdown = vld1q_f64(countdown_raw);
return vaddq_f64(pset1<Packet2d>(a), countdown);
}
template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return vaddq_f64(a,b); }

24
Eigen/src/Core/util/Meta.h
View File

@@ -328,6 +328,30 @@ struct result_of<Func(ArgType0,ArgType1)> {
enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
typedef typename binary_result_of_select<Func, ArgType0, ArgType1, FunctorType>::type type;
};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2, int SizeOf=sizeof(has_none)>
struct ternary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_std_result_type)>
{typedef typename Func::result_type type;};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_tr1_result)>
{typedef typename Func::template result<Func(ArgType0,ArgType1,ArgType2)>::type type;};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
struct result_of<Func(ArgType0,ArgType1,ArgType2)> {
template<typename T>
static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
template<typename T>
static has_tr1_result testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1,ArgType2)>::type const * = 0);
static has_none testFunctor(...);
// note that the following indirection is needed for gcc-3.3
enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
typedef typename ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, FunctorType>::type type;
};
#endif
/** \internal In short, it computes int(sqrt(\a Y)) with \a Y an integer.