mirror of
https://gitlab.com/libeigen/eigen.git
synced 2025-07-28 15:52:01 +08:00
Added missing NEON pcasts, update packetmath tests.
The NEON `pcast` operators are all implemented and tested for existing
packets. This requires adding a `pcast(a,b,c,d,e,f,g,h)` for casting
between `int64_t` and `int8_t` in `GenericPacketMath.h`.
Removed incorrect `HasHalfPacket` definition for NEON's
`Packet2l`/`Packet2ul`.
Adjustments were also made to the `packetmath` tests. These include
- minor bug fixes for cast tests (i.e. 4:1 casts, only casting for
packets that are vectorizable)
- added 8:1 cast tests
- random number generation
- original had uninteresting 0 to 0 casts for many casts between
floating-point and integers, and exhibited signed overflow
undefined behavior
Tested:
```
$ aarch64-linux-gnu-g++ -static -I./ '-DEIGEN_TEST_PART_ALL=1' test/packetmath.cpp -o packetmath
$ adb push packetmath /data/local/tmp/
$ adb shell "/data/local/tmp/packetmath"
```
This commit is contained in:
Antonio Sanchez
parent
386d809bde
commit
03ebdf6acb
@ -162,12 +162,17 @@ EIGEN_DEVICE_FUNC inline TgtPacket
|
||||
pcast(const SrcPacket& a, const SrcPacket& /*b*/) {
|
||||
return static_cast<TgtPacket>(a);
|
||||
}
|
||||
|
||||
template <typename SrcPacket, typename TgtPacket>
|
||||
EIGEN_DEVICE_FUNC inline TgtPacket
|
||||
pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) {
|
||||
return static_cast<TgtPacket>(a);
|
||||
}
|
||||
template <typename SrcPacket, typename TgtPacket>
|
||||
EIGEN_DEVICE_FUNC inline TgtPacket
|
||||
pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/,
|
||||
const SrcPacket& /*e*/, const SrcPacket& /*f*/, const SrcPacket& /*g*/, const SrcPacket& /*h*/) {
|
||||
return static_cast<TgtPacket>(a);
|
||||
}
|
||||
|
||||
/** \internal \returns reinterpret_cast<Target>(a) */
|
||||
template <typename Target, typename Packet>
|
||||
|
||||
@ -338,7 +338,7 @@ struct packet_traits<int64_t> : default_packet_traits
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 1,
|
||||
size = 2,
|
||||
HasHalfPacket = 1,
|
||||
HasHalfPacket = 0,
|
||||
|
||||
HasCmp = 1,
|
||||
HasAdd = 1,
|
||||
@ -368,7 +368,7 @@ struct packet_traits<uint64_t> : default_packet_traits
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 1,
|
||||
size = 2,
|
||||
HasHalfPacket = 1,
|
||||
HasHalfPacket = 0,
|
||||
|
||||
HasCmp = 1,
|
||||
HasAdd = 1,
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@ -11,38 +11,66 @@
|
||||
#include "packetmath_test_shared.h"
|
||||
|
||||
template <typename T>
|
||||
inline T REF_ADD(const T& a, const T& b) { return a + b;}
|
||||
inline T REF_ADD(const T& a, const T& b) {
|
||||
return a + b;
|
||||
}
|
||||
template <typename T>
|
||||
inline T REF_SUB(const T& a, const T& b) { return a - b;}
|
||||
inline T REF_SUB(const T& a, const T& b) {
|
||||
return a - b;
|
||||
}
|
||||
template <typename T>
|
||||
inline T REF_MUL(const T& a, const T& b) { return a * b;}
|
||||
inline T REF_MUL(const T& a, const T& b) {
|
||||
return a * b;
|
||||
}
|
||||
template <typename T>
|
||||
inline T REF_DIV(const T& a, const T& b) { return a / b;}
|
||||
inline T REF_DIV(const T& a, const T& b) {
|
||||
return a / b;
|
||||
}
|
||||
template <typename T>
|
||||
inline T REF_ABS_DIFF(const T& a, const T& b) { return a>b ? a - b : b-a;}
|
||||
inline T REF_ABS_DIFF(const T& a, const T& b) {
|
||||
return a > b ? a - b : b - a;
|
||||
}
|
||||
|
||||
// Specializations for bool.
|
||||
template <>
|
||||
inline bool REF_ADD(const bool& a, const bool& b) { return a || b;}
|
||||
inline bool REF_ADD(const bool& a, const bool& b) {
|
||||
return a || b;
|
||||
}
|
||||
template <>
|
||||
inline bool REF_SUB(const bool& a, const bool& b) { return a ^ b;}
|
||||
inline bool REF_SUB(const bool& a, const bool& b) {
|
||||
return a ^ b;
|
||||
}
|
||||
template <>
|
||||
inline bool REF_MUL(const bool& a, const bool& b) { return a && b;}
|
||||
inline bool REF_MUL(const bool& a, const bool& b) {
|
||||
return a && b;
|
||||
}
|
||||
|
||||
// Uses pcast to cast from one array to another.
|
||||
template <typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio>
|
||||
struct pcast_array {
|
||||
static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size, typename internal::unpacket_traits<TgtPacket>::type* dst) {
|
||||
struct pcast_array;
|
||||
|
||||
template <typename SrcPacket, typename TgtPacket, int TgtCoeffRatio>
|
||||
struct pcast_array<SrcPacket, TgtPacket, 1, TgtCoeffRatio> {
|
||||
typedef typename internal::unpacket_traits<SrcPacket>::type SrcScalar;
|
||||
typedef typename internal::unpacket_traits<TgtPacket>::type TgtScalar;
|
||||
static void cast(const SrcScalar* src, size_t size, TgtScalar* dst) {
|
||||
static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
|
||||
static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
|
||||
for (size_t i=0; i<size; i+=TgtPacketSize) {
|
||||
size_t i;
|
||||
for (i = 0; i < size && i + SrcPacketSize <= size; i += TgtPacketSize) {
|
||||
internal::pstoreu(dst + i, internal::pcast<SrcPacket, TgtPacket>(internal::ploadu<SrcPacket>(src + i)));
|
||||
}
|
||||
// Leftovers that cannot be loaded into a packet.
|
||||
for (; i < size; ++i) {
|
||||
dst[i] = static_cast<TgtScalar>(src[i]);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename SrcPacket, typename TgtPacket>
|
||||
struct pcast_array<SrcPacket, TgtPacket, 2, 1> {
|
||||
static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size, typename internal::unpacket_traits<TgtPacket>::type* dst) {
|
||||
static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size,
|
||||
typename internal::unpacket_traits<TgtPacket>::type* dst) {
|
||||
static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
|
||||
static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
|
||||
for (size_t i = 0; i < size; i += TgtPacketSize) {
|
||||
@ -55,7 +83,8 @@ struct pcast_array<SrcPacket, TgtPacket, 2, 1>{
|
||||
|
||||
template <typename SrcPacket, typename TgtPacket>
|
||||
struct pcast_array<SrcPacket, TgtPacket, 4, 1> {
|
||||
static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size, typename internal::unpacket_traits<TgtPacket>::type* dst) {
|
||||
static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size,
|
||||
typename internal::unpacket_traits<TgtPacket>::type* dst) {
|
||||
static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
|
||||
static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
|
||||
for (size_t i = 0; i < size; i += TgtPacketSize) {
|
||||
@ -63,7 +92,27 @@ struct pcast_array<SrcPacket, TgtPacket, 4, 1>{
|
||||
SrcPacket b = internal::ploadu<SrcPacket>(src + i + SrcPacketSize);
|
||||
SrcPacket c = internal::ploadu<SrcPacket>(src + i + 2 * SrcPacketSize);
|
||||
SrcPacket d = internal::ploadu<SrcPacket>(src + i + 3 * SrcPacketSize);
|
||||
internal::pstoreu(dst+i, internal::pcast<SrcPacket,TgtPacket>(a, b));
|
||||
internal::pstoreu(dst + i, internal::pcast<SrcPacket, TgtPacket>(a, b, c, d));
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename SrcPacket, typename TgtPacket>
|
||||
struct pcast_array<SrcPacket, TgtPacket, 8, 1> {
|
||||
static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size,
|
||||
typename internal::unpacket_traits<TgtPacket>::type* dst) {
|
||||
static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
|
||||
static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
|
||||
for (size_t i = 0; i < size; i += TgtPacketSize) {
|
||||
SrcPacket a = internal::ploadu<SrcPacket>(src + i);
|
||||
SrcPacket b = internal::ploadu<SrcPacket>(src + i + SrcPacketSize);
|
||||
SrcPacket c = internal::ploadu<SrcPacket>(src + i + 2 * SrcPacketSize);
|
||||
SrcPacket d = internal::ploadu<SrcPacket>(src + i + 3 * SrcPacketSize);
|
||||
SrcPacket e = internal::ploadu<SrcPacket>(src + i + 4 * SrcPacketSize);
|
||||
SrcPacket f = internal::ploadu<SrcPacket>(src + i + 5 * SrcPacketSize);
|
||||
SrcPacket g = internal::ploadu<SrcPacket>(src + i + 6 * SrcPacketSize);
|
||||
SrcPacket h = internal::ploadu<SrcPacket>(src + i + 7 * SrcPacketSize);
|
||||
internal::pstoreu(dst + i, internal::pcast<SrcPacket, TgtPacket>(a, b, c, d, e, f, g, h));
|
||||
}
|
||||
}
|
||||
};
|
||||
@ -76,6 +125,129 @@ struct test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, fals
|
||||
static void run() {}
|
||||
};
|
||||
|
||||
// Generates random values that fit in both SrcScalar and TgtScalar without
|
||||
// overflowing when cast.
|
||||
template <typename SrcScalar, typename TgtScalar, typename EnableIf = void>
|
||||
struct random_without_cast_overflow {
|
||||
static SrcScalar value() { return internal::random<SrcScalar>(); }
|
||||
};
|
||||
|
||||
// Widening integer cast signed to unsigned.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<NumTraits<SrcScalar>::IsInteger && NumTraits<TgtScalar>::IsInteger &&
|
||||
!NumTraits<TgtScalar>::IsSigned &&
|
||||
(std::numeric_limits<SrcScalar>::digits < std::numeric_limits<TgtScalar>::digits ||
|
||||
(std::numeric_limits<SrcScalar>::digits == std::numeric_limits<TgtScalar>::digits &&
|
||||
NumTraits<SrcScalar>::IsSigned))>::type> {
|
||||
static SrcScalar value() {
|
||||
SrcScalar a = internal::random<SrcScalar>();
|
||||
return a < SrcScalar(0) ? -(a + 1) : a;
|
||||
}
|
||||
};
|
||||
|
||||
// Narrowing integer cast to unsigned.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<
|
||||
NumTraits<SrcScalar>::IsInteger && NumTraits<TgtScalar>::IsInteger && !NumTraits<SrcScalar>::IsSigned &&
|
||||
(std::numeric_limits<SrcScalar>::digits > std::numeric_limits<TgtScalar>::digits)>::type> {
|
||||
static SrcScalar value() {
|
||||
TgtScalar b = internal::random<TgtScalar>();
|
||||
return static_cast<SrcScalar>(b < TgtScalar(0) ? -(b + 1) : b);
|
||||
}
|
||||
};
|
||||
|
||||
// Narrowing integer cast to signed.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<
|
||||
NumTraits<SrcScalar>::IsInteger && NumTraits<TgtScalar>::IsInteger && NumTraits<SrcScalar>::IsSigned &&
|
||||
(std::numeric_limits<SrcScalar>::digits > std::numeric_limits<TgtScalar>::digits)>::type> {
|
||||
static SrcScalar value() {
|
||||
TgtScalar b = internal::random<TgtScalar>();
|
||||
return static_cast<SrcScalar>(b);
|
||||
}
|
||||
};
|
||||
|
||||
// Unsigned to signed narrowing cast.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<NumTraits<SrcScalar>::IsInteger && NumTraits<TgtScalar>::IsInteger &&
|
||||
!NumTraits<SrcScalar>::IsSigned && NumTraits<TgtScalar>::IsSigned &&
|
||||
(std::numeric_limits<SrcScalar>::digits ==
|
||||
std::numeric_limits<TgtScalar>::digits)>::type> {
|
||||
static SrcScalar value() { return internal::random<SrcScalar>() / 2; }
|
||||
};
|
||||
|
||||
template <typename Scalar>
|
||||
struct is_floating_point {
|
||||
enum { value = 0 };
|
||||
};
|
||||
template <>
|
||||
struct is_floating_point<float> {
|
||||
enum { value = 1 };
|
||||
};
|
||||
template <>
|
||||
struct is_floating_point<double> {
|
||||
enum { value = 1 };
|
||||
};
|
||||
template <>
|
||||
struct is_floating_point<half> {
|
||||
enum { value = 1 };
|
||||
};
|
||||
template <>
|
||||
struct is_floating_point<bfloat16> {
|
||||
enum { value = 1 };
|
||||
};
|
||||
|
||||
// Floating-point to integer, full precision.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<is_floating_point<SrcScalar>::value && NumTraits<TgtScalar>::IsInteger &&
|
||||
(std::numeric_limits<TgtScalar>::digits <=
|
||||
std::numeric_limits<SrcScalar>::digits)>::type> {
|
||||
static SrcScalar value() { return static_cast<SrcScalar>(internal::random<TgtScalar>()); }
|
||||
};
|
||||
|
||||
// Floating-point to integer, narrowing precision.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<is_floating_point<SrcScalar>::value && NumTraits<TgtScalar>::IsInteger &&
|
||||
(std::numeric_limits<TgtScalar>::digits >
|
||||
std::numeric_limits<SrcScalar>::digits)>::type> {
|
||||
static SrcScalar value() {
|
||||
static const int BitShift = std::numeric_limits<TgtScalar>::digits - std::numeric_limits<SrcScalar>::digits;
|
||||
return static_cast<SrcScalar>(internal::random<TgtScalar>() >> BitShift);
|
||||
}
|
||||
};
|
||||
|
||||
// Floating-point target from integer, re-use above logic.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<NumTraits<SrcScalar>::IsInteger && is_floating_point<TgtScalar>::value>::type> {
|
||||
static SrcScalar value() {
|
||||
return static_cast<SrcScalar>(random_without_cast_overflow<TgtScalar, SrcScalar>::value());
|
||||
}
|
||||
};
|
||||
|
||||
// Floating-point narrowing conversion.
|
||||
template <typename SrcScalar, typename TgtScalar>
|
||||
struct random_without_cast_overflow<
|
||||
SrcScalar, TgtScalar,
|
||||
typename internal::enable_if<is_floating_point<SrcScalar>::value && is_floating_point<TgtScalar>::value &&
|
||||
(std::numeric_limits<SrcScalar>::digits >
|
||||
std::numeric_limits<TgtScalar>::digits)>::type> {
|
||||
static SrcScalar value() { return static_cast<SrcScalar>(internal::random<TgtScalar>()); }
|
||||
};
|
||||
|
||||
template <typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio>
|
||||
struct test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, true> {
|
||||
static void run() {
|
||||
@ -83,23 +255,20 @@ struct test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, true
|
||||
typedef typename internal::unpacket_traits<TgtPacket>::type TgtScalar;
|
||||
static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
|
||||
static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
|
||||
static const int DataSize = SrcPacketSize*SrcCoeffRatio;
|
||||
VERIFY(DataSize == TgtPacketSize*TgtCoeffRatio && "Packet sizes and cast ratios are mismatched.");
|
||||
static const int BlockSize = SrcPacketSize * SrcCoeffRatio;
|
||||
eigen_assert(BlockSize == TgtPacketSize * TgtCoeffRatio && "Packet sizes and cast ratios are mismatched.");
|
||||
|
||||
static const int DataSize = 10 * BlockSize;
|
||||
EIGEN_ALIGN_MAX SrcScalar data1[DataSize];
|
||||
EIGEN_ALIGN_MAX TgtScalar data2[DataSize];
|
||||
EIGEN_ALIGN_MAX TgtScalar ref[DataSize];
|
||||
|
||||
// Construct a packet of scalars that will not overflow when casting
|
||||
for (int i = 0; i < DataSize; ++i) {
|
||||
const SrcScalar a = Array<SrcScalar,1,1>::Random().value();
|
||||
const TgtScalar b = Array<TgtScalar,1,1>::Random().value();
|
||||
const SrcScalar c = sizeof(TgtScalar) > sizeof(SrcScalar) ? static_cast<SrcScalar>(b) : a;
|
||||
data1[i] = (NumTraits<SrcScalar>::IsSigned && !NumTraits<TgtScalar>::IsSigned) ? numext::abs(c) : c;
|
||||
data1[i] = random_without_cast_overflow<SrcScalar, TgtScalar>::value();
|
||||
}
|
||||
|
||||
for (int i=0; i<DataSize; ++i)
|
||||
ref[i] = static_cast<const TgtScalar>(data1[i]);
|
||||
for (int i = 0; i < DataSize; ++i) ref[i] = static_cast<const TgtScalar>(data1[i]);
|
||||
|
||||
pcast_array<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio>::cast(data1, DataSize, data2);
|
||||
|
||||
@ -107,48 +276,72 @@ struct test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, true
|
||||
}
|
||||
};
|
||||
|
||||
template<typename SrcScalar, typename TgtScalar>
|
||||
void test_cast() {
|
||||
typedef typename internal::packet_traits<SrcScalar> SrcPacketTraits;
|
||||
typedef typename internal::packet_traits<TgtScalar> TgtPacketTraits;
|
||||
template <typename SrcPacket, typename TgtPacket>
|
||||
struct test_cast {
|
||||
static void run() {
|
||||
typedef typename internal::unpacket_traits<SrcPacket>::type SrcScalar;
|
||||
typedef typename internal::unpacket_traits<TgtPacket>::type TgtScalar;
|
||||
typedef typename internal::type_casting_traits<SrcScalar, TgtScalar> TypeCastingTraits;
|
||||
static const int SrcCoeffRatio = TypeCastingTraits::SrcCoeffRatio;
|
||||
static const int TgtCoeffRatio = TypeCastingTraits::TgtCoeffRatio;
|
||||
|
||||
static const bool HasFullCast = TypeCastingTraits::VectorizedCast;
|
||||
static const bool HasHalfCast = HasFullCast && internal::packet_traits<SrcScalar>::HasHalfPacket && internal::packet_traits<TgtScalar>::HasHalfPacket;
|
||||
|
||||
test_cast_helper<typename SrcPacketTraits::type, typename TgtPacketTraits::type, SrcCoeffRatio, TgtCoeffRatio, HasFullCast>::run();
|
||||
test_cast_helper<typename SrcPacketTraits::half, typename TgtPacketTraits::half, SrcCoeffRatio, TgtCoeffRatio, HasHalfCast>::run();
|
||||
static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
|
||||
static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
|
||||
static const bool HasCast =
|
||||
internal::unpacket_traits<SrcPacket>::vectorizable && internal::unpacket_traits<TgtPacket>::vectorizable &&
|
||||
TypeCastingTraits::VectorizedCast && (SrcPacketSize * SrcCoeffRatio == TgtPacketSize * TgtCoeffRatio);
|
||||
test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, HasCast>::run();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Scalar, typename Packet> void packetmath_pcast_ops() {
|
||||
const static bool IsFullPacket = internal::is_same<typename internal::packet_traits<Scalar>::type,Packet>::value;
|
||||
if (IsFullPacket) {
|
||||
test_cast<Scalar, float>();
|
||||
test_cast<Scalar, double>();
|
||||
test_cast<Scalar, int8_t>();
|
||||
test_cast<Scalar, uint8_t>();
|
||||
test_cast<Scalar, int16_t>();
|
||||
test_cast<Scalar, uint16_t>();
|
||||
test_cast<Scalar, int32_t>();
|
||||
test_cast<Scalar, uint32_t>();
|
||||
test_cast<Scalar, int64_t>();
|
||||
test_cast<Scalar, uint64_t>();
|
||||
template <typename SrcPacket, typename TgtScalar,
|
||||
typename TgtPacket = typename internal::packet_traits<TgtScalar>::type,
|
||||
bool Vectorized = internal::packet_traits<TgtScalar>::Vectorizable,
|
||||
bool HasHalf = !internal::is_same<typename internal::unpacket_traits<TgtPacket>::half, TgtPacket>::value>
|
||||
struct test_cast_runner;
|
||||
|
||||
template <typename SrcPacket, typename TgtScalar, typename TgtPacket>
|
||||
struct test_cast_runner<SrcPacket, TgtScalar, TgtPacket, true, false> {
|
||||
static void run() { test_cast<SrcPacket, TgtPacket>::run(); }
|
||||
};
|
||||
|
||||
template <typename SrcPacket, typename TgtScalar, typename TgtPacket>
|
||||
struct test_cast_runner<SrcPacket, TgtScalar, TgtPacket, true, true> {
|
||||
static void run() {
|
||||
test_cast<SrcPacket, TgtPacket>::run();
|
||||
test_cast_runner<SrcPacket, TgtScalar, typename internal::unpacket_traits<TgtPacket>::half>::run();
|
||||
}
|
||||
};
|
||||
|
||||
template <typename SrcPacket, typename TgtScalar, typename TgtPacket>
|
||||
struct test_cast_runner<SrcPacket, TgtScalar, TgtPacket, false, false> {
|
||||
static void run() {}
|
||||
};
|
||||
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath_pcast_ops() {
|
||||
test_cast_runner<Packet, float>::run();
|
||||
test_cast_runner<Packet, double>::run();
|
||||
test_cast_runner<Packet, int8_t>::run();
|
||||
test_cast_runner<Packet, uint8_t>::run();
|
||||
test_cast_runner<Packet, int16_t>::run();
|
||||
test_cast_runner<Packet, uint16_t>::run();
|
||||
test_cast_runner<Packet, int32_t>::run();
|
||||
test_cast_runner<Packet, uint32_t>::run();
|
||||
test_cast_runner<Packet, int64_t>::run();
|
||||
test_cast_runner<Packet, uint64_t>::run();
|
||||
test_cast_runner<Packet, bool>::run();
|
||||
test_cast_runner<Packet, half>::run();
|
||||
}
|
||||
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath_boolean_mask_ops()
|
||||
{
|
||||
void packetmath_boolean_mask_ops() {
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
const int size = 2 * PacketSize;
|
||||
EIGEN_ALIGN_MAX Scalar data1[size];
|
||||
EIGEN_ALIGN_MAX Scalar data2[size];
|
||||
EIGEN_ALIGN_MAX Scalar ref[size];
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>();
|
||||
}
|
||||
CHECK_CWISE1(internal::ptrue, internal::ptrue);
|
||||
@ -164,21 +357,18 @@ void packetmath_boolean_mask_ops()
|
||||
// (for some compilers) compute the bitwise and with 0x1 of the results to keep the value in [0,1].
|
||||
#ifdef EIGEN_PACKET_MATH_SSE_H
|
||||
template <>
|
||||
void packetmath_boolean_mask_ops<bool, internal::Packet16b>()
|
||||
{
|
||||
}
|
||||
void packetmath_boolean_mask_ops<bool, internal::Packet16b>() {}
|
||||
#endif
|
||||
|
||||
template<typename Scalar,typename Packet> void packetmath()
|
||||
{
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath() {
|
||||
typedef internal::packet_traits<Scalar> PacketTraits;
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
if (g_first_pass)
|
||||
std::cerr << "=== Testing packet of type '" << typeid(Packet).name()
|
||||
<< "' and scalar type '" << typeid(Scalar).name()
|
||||
<< "' and size '" << PacketSize << "' ===\n" ;
|
||||
std::cerr << "=== Testing packet of type '" << typeid(Packet).name() << "' and scalar type '"
|
||||
<< typeid(Scalar).name() << "' and size '" << PacketSize << "' ===\n";
|
||||
|
||||
const int max_size = PacketSize > 4 ? PacketSize : 4;
|
||||
const int size = PacketSize * max_size;
|
||||
@ -187,8 +377,7 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
EIGEN_ALIGN_MAX Scalar data3[size];
|
||||
EIGEN_ALIGN_MAX Scalar ref[size];
|
||||
RealScalar refvalue = RealScalar(0);
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>() / RealScalar(PacketSize);
|
||||
data2[i] = internal::random<Scalar>() / RealScalar(PacketSize);
|
||||
refvalue = (std::max)(refvalue, numext::abs(data1[i]));
|
||||
@ -197,48 +386,38 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
internal::pstore(data2, internal::pload<Packet>(data1));
|
||||
VERIFY(test::areApprox(data1, data2, PacketSize) && "aligned load/store");
|
||||
|
||||
for (int offset=0; offset<PacketSize; ++offset)
|
||||
{
|
||||
for (int offset = 0; offset < PacketSize; ++offset) {
|
||||
internal::pstore(data2, internal::ploadu<Packet>(data1 + offset));
|
||||
VERIFY(test::areApprox(data1 + offset, data2, PacketSize) && "internal::ploadu");
|
||||
}
|
||||
|
||||
for (int offset=0; offset<PacketSize; ++offset)
|
||||
{
|
||||
for (int offset = 0; offset < PacketSize; ++offset) {
|
||||
internal::pstoreu(data2 + offset, internal::pload<Packet>(data1));
|
||||
VERIFY(test::areApprox(data1, data2 + offset, PacketSize) && "internal::pstoreu");
|
||||
}
|
||||
|
||||
if (internal::unpacket_traits<Packet>::masked_load_available)
|
||||
{
|
||||
if (internal::unpacket_traits<Packet>::masked_load_available) {
|
||||
test::packet_helper<internal::unpacket_traits<Packet>::masked_load_available, Packet> h;
|
||||
unsigned long long max_umask = (0x1ull << PacketSize);
|
||||
|
||||
for (int offset=0; offset<PacketSize; ++offset)
|
||||
{
|
||||
for (unsigned long long umask=0; umask<max_umask; ++umask)
|
||||
{
|
||||
for (int offset = 0; offset < PacketSize; ++offset) {
|
||||
for (unsigned long long umask = 0; umask < max_umask; ++umask) {
|
||||
h.store(data2, h.load(data1 + offset, umask));
|
||||
for (int k=0; k<PacketSize; ++k)
|
||||
data3[k] = ((umask & ( 0x1ull << k )) >> k) ? data1[k+offset] : Scalar(0);
|
||||
for (int k = 0; k < PacketSize; ++k) data3[k] = ((umask & (0x1ull << k)) >> k) ? data1[k + offset] : Scalar(0);
|
||||
VERIFY(test::areApprox(data3, data2, PacketSize) && "internal::ploadu masked");
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (internal::unpacket_traits<Packet>::masked_store_available)
|
||||
{
|
||||
if (internal::unpacket_traits<Packet>::masked_store_available) {
|
||||
test::packet_helper<internal::unpacket_traits<Packet>::masked_store_available, Packet> h;
|
||||
unsigned long long max_umask = (0x1ull << PacketSize);
|
||||
|
||||
for (int offset=0; offset<PacketSize; ++offset)
|
||||
{
|
||||
for (unsigned long long umask=0; umask<max_umask; ++umask)
|
||||
{
|
||||
for (int offset = 0; offset < PacketSize; ++offset) {
|
||||
for (unsigned long long umask = 0; umask < max_umask; ++umask) {
|
||||
internal::pstore(data2, internal::pset1<Packet>(Scalar(0)));
|
||||
h.store(data2, h.loadu(data1 + offset), umask);
|
||||
for (int k=0; k<PacketSize; ++k)
|
||||
data3[k] = ((umask & ( 0x1ull << k )) >> k) ? data1[k+offset] : Scalar(0);
|
||||
for (int k = 0; k < PacketSize; ++k) data3[k] = ((umask & (0x1ull << k)) >> k) ? data1[k + offset] : Scalar(0);
|
||||
VERIFY(test::areApprox(data3, data2, PacketSize) && "internal::pstoreu masked");
|
||||
}
|
||||
}
|
||||
@ -253,21 +432,17 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
CHECK_CWISE2_IF(PacketTraits::HasMul, REF_MUL, internal::pmul);
|
||||
CHECK_CWISE2_IF(PacketTraits::HasDiv, REF_DIV, internal::pdiv);
|
||||
|
||||
if (PacketTraits::HasNegate)
|
||||
CHECK_CWISE1(internal::negate, internal::pnegate);
|
||||
if (PacketTraits::HasNegate) CHECK_CWISE1(internal::negate, internal::pnegate);
|
||||
CHECK_CWISE1(numext::conj, internal::pconj);
|
||||
|
||||
for(int offset=0;offset<3;++offset)
|
||||
{
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[i] = data1[offset];
|
||||
for (int offset = 0; offset < 3; ++offset) {
|
||||
for (int i = 0; i < PacketSize; ++i) ref[i] = data1[offset];
|
||||
internal::pstore(data2, internal::pset1<Packet>(data1[offset]));
|
||||
VERIFY(test::areApprox(ref, data2, PacketSize) && "internal::pset1");
|
||||
}
|
||||
|
||||
{
|
||||
for (int i=0; i<PacketSize*4; ++i)
|
||||
ref[i] = data1[i/PacketSize];
|
||||
for (int i = 0; i < PacketSize * 4; ++i) ref[i] = data1[i / PacketSize];
|
||||
Packet A0, A1, A2, A3;
|
||||
internal::pbroadcast4<Packet>(data1, A0, A1, A2, A3);
|
||||
internal::pstore(data2 + 0 * PacketSize, A0);
|
||||
@ -278,8 +453,7 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
}
|
||||
|
||||
{
|
||||
for (int i=0; i<PacketSize*2; ++i)
|
||||
ref[i] = data1[i/PacketSize];
|
||||
for (int i = 0; i < PacketSize * 2; ++i) ref[i] = data1[i / PacketSize];
|
||||
Packet A0, A1;
|
||||
internal::pbroadcast2<Packet>(data1, A0, A1);
|
||||
internal::pstore(data2 + 0 * PacketSize, A0);
|
||||
@ -289,23 +463,18 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
|
||||
VERIFY(internal::isApprox(data1[0], internal::pfirst(internal::pload<Packet>(data1))) && "internal::pfirst");
|
||||
|
||||
if(PacketSize>1)
|
||||
{
|
||||
if (PacketSize > 1) {
|
||||
// apply different offsets to check that ploaddup is robust to unaligned inputs
|
||||
for(int offset=0;offset<4;++offset)
|
||||
{
|
||||
for(int i=0;i<PacketSize/2;++i)
|
||||
ref[2*i+0] = ref[2*i+1] = data1[offset+i];
|
||||
for (int offset = 0; offset < 4; ++offset) {
|
||||
for (int i = 0; i < PacketSize / 2; ++i) ref[2 * i + 0] = ref[2 * i + 1] = data1[offset + i];
|
||||
internal::pstore(data2, internal::ploaddup<Packet>(data1 + offset));
|
||||
VERIFY(test::areApprox(ref, data2, PacketSize) && "ploaddup");
|
||||
}
|
||||
}
|
||||
|
||||
if(PacketSize>2)
|
||||
{
|
||||
if (PacketSize > 2) {
|
||||
// apply different offsets to check that ploadquad is robust to unaligned inputs
|
||||
for(int offset=0;offset<4;++offset)
|
||||
{
|
||||
for (int offset = 0; offset < 4; ++offset) {
|
||||
for (int i = 0; i < PacketSize / 4; ++i)
|
||||
ref[4 * i + 0] = ref[4 * i + 1] = ref[4 * i + 2] = ref[4 * i + 3] = data1[offset + i];
|
||||
internal::pstore(data2, internal::ploadquad<Packet>(data1 + offset));
|
||||
@ -314,28 +483,24 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
}
|
||||
|
||||
ref[0] = Scalar(0);
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[0] += data1[i];
|
||||
for (int i = 0; i < PacketSize; ++i) ref[0] += data1[i];
|
||||
VERIFY(test::isApproxAbs(ref[0], internal::predux(internal::pload<Packet>(data1)), refvalue) && "internal::predux");
|
||||
|
||||
if(PacketSize==8 && internal::unpacket_traits<typename internal::unpacket_traits<Packet>::half>::size ==4) // so far, predux_half_downto4 is only required in such a case
|
||||
if (PacketSize == 8 && internal::unpacket_traits<typename internal::unpacket_traits<Packet>::half>::size ==
|
||||
4) // so far, predux_half_downto4 is only required in such a case
|
||||
{
|
||||
int HalfPacketSize = PacketSize > 4 ? PacketSize / 2 : PacketSize;
|
||||
for (int i=0; i<HalfPacketSize; ++i)
|
||||
ref[i] = Scalar(0);
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[i%HalfPacketSize] += data1[i];
|
||||
for (int i = 0; i < HalfPacketSize; ++i) ref[i] = Scalar(0);
|
||||
for (int i = 0; i < PacketSize; ++i) ref[i % HalfPacketSize] += data1[i];
|
||||
internal::pstore(data2, internal::predux_half_dowto4(internal::pload<Packet>(data1)));
|
||||
VERIFY(test::areApprox(ref, data2, HalfPacketSize) && "internal::predux_half_dowto4");
|
||||
}
|
||||
|
||||
ref[0] = Scalar(1);
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[0] = REF_MUL(ref[0], data1[i]);
|
||||
for (int i = 0; i < PacketSize; ++i) ref[0] = REF_MUL(ref[0], data1[i]);
|
||||
VERIFY(internal::isApprox(ref[0], internal::predux_mul(internal::pload<Packet>(data1))) && "internal::predux_mul");
|
||||
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[i] = data1[PacketSize-i-1];
|
||||
for (int i = 0; i < PacketSize; ++i) ref[i] = data1[PacketSize - i - 1];
|
||||
internal::pstore(data2, internal::preverse(internal::pload<Packet>(data1)));
|
||||
VERIFY(test::areApprox(ref, data2, PacketSize) && "internal::preverse");
|
||||
|
||||
@ -351,7 +516,6 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
if (PacketTraits::HasBlend) {
|
||||
Packet thenPacket = internal::pload<Packet>(data1);
|
||||
Packet elsePacket = internal::pload<Packet>(data2);
|
||||
@ -386,8 +550,7 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
|
||||
CHECK_CWISE1_IF(PacketTraits::HasSqrt, numext::sqrt, internal::psqrt);
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>();
|
||||
}
|
||||
CHECK_CWISE1(internal::pzero, internal::pzero);
|
||||
@ -399,10 +562,8 @@ template<typename Scalar,typename Packet> void packetmath()
|
||||
packetmath_pcast_ops<Scalar, Packet>();
|
||||
}
|
||||
|
||||
|
||||
|
||||
template<typename Scalar,typename Packet> void packetmath_real()
|
||||
{
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath_real() {
|
||||
typedef internal::packet_traits<Scalar> PacketTraits;
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
|
||||
@ -411,20 +572,17 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
EIGEN_ALIGN_MAX Scalar data2[PacketSize * 4];
|
||||
EIGEN_ALIGN_MAX Scalar ref[PacketSize * 4];
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>(0, 1) * std::pow(Scalar(10), internal::random<Scalar>(-6, 6));
|
||||
data2[i] = internal::random<Scalar>(0, 1) * std::pow(Scalar(10), internal::random<Scalar>(-6, 6));
|
||||
}
|
||||
|
||||
if(internal::random<float>(0,1)<0.1f)
|
||||
data1[internal::random<int>(0, PacketSize)] = 0;
|
||||
if (internal::random<float>(0, 1) < 0.1f) data1[internal::random<int>(0, PacketSize)] = 0;
|
||||
|
||||
CHECK_CWISE1_IF(PacketTraits::HasLog, std::log, internal::plog);
|
||||
CHECK_CWISE1_IF(PacketTraits::HasRsqrt, Scalar(1) / std::sqrt, internal::prsqrt);
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>(-1, 1) * std::pow(Scalar(10), internal::random<Scalar>(-3, 3));
|
||||
data2[i] = internal::random<Scalar>(-1, 1) * std::pow(Scalar(10), internal::random<Scalar>(-3, 3));
|
||||
}
|
||||
@ -438,37 +596,32 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
CHECK_CWISE1_IF(PacketTraits::HasRint, numext::rint, internal::print);
|
||||
|
||||
// See bug 1785.
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = -1.5 + i;
|
||||
data2[i] = -1.5 + i;
|
||||
}
|
||||
CHECK_CWISE1_IF(PacketTraits::HasRound, numext::round, internal::pround);
|
||||
CHECK_CWISE1_IF(PacketTraits::HasRint, numext::rint, internal::print);
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>(-1, 1);
|
||||
data2[i] = internal::random<Scalar>(-1, 1);
|
||||
}
|
||||
CHECK_CWISE1_IF(PacketTraits::HasASin, std::asin, internal::pasin);
|
||||
CHECK_CWISE1_IF(PacketTraits::HasACos, std::acos, internal::pacos);
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>(-87, 88);
|
||||
data2[i] = internal::random<Scalar>(-87, 88);
|
||||
}
|
||||
CHECK_CWISE1_IF(PacketTraits::HasExp, std::exp, internal::pexp);
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>(-1, 1) * std::pow(Scalar(10), internal::random<Scalar>(-6, 6));
|
||||
data2[i] = internal::random<Scalar>(-1, 1) * std::pow(Scalar(10), internal::random<Scalar>(-6, 6));
|
||||
}
|
||||
data1[0] = 1e-20;
|
||||
CHECK_CWISE1_IF(PacketTraits::HasTanh, std::tanh, internal::ptanh);
|
||||
if(PacketTraits::HasExp && PacketSize>=2)
|
||||
{
|
||||
if (PacketTraits::HasExp && PacketSize >= 2) {
|
||||
data1[0] = std::numeric_limits<Scalar>::quiet_NaN();
|
||||
data1[1] = std::numeric_limits<Scalar>::epsilon();
|
||||
test::packet_helper<PacketTraits::HasExp, Packet> h;
|
||||
@ -505,8 +658,7 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
|
||||
if (PacketTraits::HasExp) {
|
||||
internal::scalar_logistic_op<Scalar> logistic;
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>(-20, 20);
|
||||
}
|
||||
|
||||
@ -529,12 +681,10 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
CHECK_CWISE1_IF(PacketTraits::HasExpm1, std::expm1, internal::pexpm1);
|
||||
#endif
|
||||
|
||||
if(PacketSize>=2)
|
||||
{
|
||||
if (PacketSize >= 2) {
|
||||
data1[0] = std::numeric_limits<Scalar>::quiet_NaN();
|
||||
data1[1] = std::numeric_limits<Scalar>::epsilon();
|
||||
if(PacketTraits::HasLog)
|
||||
{
|
||||
if (PacketTraits::HasLog) {
|
||||
test::packet_helper<PacketTraits::HasLog, Packet> h;
|
||||
h.store(data2, internal::plog(h.load(data1)));
|
||||
VERIFY((numext::isnan)(data2[0]));
|
||||
@ -574,8 +724,7 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
VERIFY((numext::isnan)(data2[0]));
|
||||
VERIFY((numext::isnan)(data2[1]));
|
||||
}
|
||||
if(PacketTraits::HasSqrt)
|
||||
{
|
||||
if (PacketTraits::HasSqrt) {
|
||||
test::packet_helper<PacketTraits::HasSqrt, Packet> h;
|
||||
data1[0] = Scalar(-1.0f);
|
||||
data1[1] = -std::numeric_limits<Scalar>::denorm_min();
|
||||
@ -583,13 +732,10 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
VERIFY((numext::isnan)(data2[0]));
|
||||
VERIFY((numext::isnan)(data2[1]));
|
||||
}
|
||||
if(PacketTraits::HasCos)
|
||||
{
|
||||
if (PacketTraits::HasCos) {
|
||||
test::packet_helper<PacketTraits::HasCos, Packet> h;
|
||||
for(Scalar k = 1; k<Scalar(10000)/std::numeric_limits<Scalar>::epsilon(); k*=2)
|
||||
{
|
||||
for(int k1=0;k1<=1; ++k1)
|
||||
{
|
||||
for (Scalar k = 1; k < Scalar(10000) / std::numeric_limits<Scalar>::epsilon(); k *= 2) {
|
||||
for (int k1 = 0; k1 <= 1; ++k1) {
|
||||
data1[0] = (2 * k + k1) * Scalar(EIGEN_PI) / 2 * internal::random<Scalar>(0.8, 1.2);
|
||||
data1[1] = (2 * k + 2 + k1) * Scalar(EIGEN_PI) / 2 * internal::random<Scalar>(0.8, 1.2);
|
||||
h.store(data2, internal::pcos(h.load(data1)));
|
||||
@ -629,8 +775,8 @@ template<typename Scalar,typename Packet> void packetmath_real()
|
||||
}
|
||||
}
|
||||
|
||||
template<typename Scalar,typename Packet> void packetmath_notcomplex()
|
||||
{
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath_notcomplex() {
|
||||
typedef internal::packet_traits<Scalar> PacketTraits;
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
|
||||
@ -641,8 +787,7 @@ template<typename Scalar,typename Packet> void packetmath_notcomplex()
|
||||
Array<Scalar, Dynamic, 1>::Map(data1, PacketSize * 4).setRandom();
|
||||
|
||||
ref[0] = data1[0];
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[0] = (std::min)(ref[0],data1[i]);
|
||||
for (int i = 0; i < PacketSize; ++i) ref[0] = (std::min)(ref[0], data1[i]);
|
||||
VERIFY(internal::isApprox(ref[0], internal::predux_min(internal::pload<Packet>(data1))) && "internal::predux_min");
|
||||
|
||||
VERIFY((!PacketTraits::Vectorizable) || PacketTraits::HasMin);
|
||||
@ -654,12 +799,10 @@ template<typename Scalar,typename Packet> void packetmath_notcomplex()
|
||||
CHECK_CWISE2_IF(PacketTraits::HasAbsDiff, REF_ABS_DIFF, internal::pabsdiff);
|
||||
|
||||
ref[0] = data1[0];
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[0] = (std::max)(ref[0],data1[i]);
|
||||
for (int i = 0; i < PacketSize; ++i) ref[0] = (std::max)(ref[0], data1[i]);
|
||||
VERIFY(internal::isApprox(ref[0], internal::predux_max(internal::pload<Packet>(data1))) && "internal::predux_max");
|
||||
|
||||
for (int i=0; i<PacketSize; ++i)
|
||||
ref[i] = data1[0]+Scalar(i);
|
||||
for (int i = 0; i < PacketSize; ++i) ref[i] = data1[0] + Scalar(i);
|
||||
internal::pstore(data2, internal::plset<Packet>(data1[0]));
|
||||
VERIFY(test::areApprox(ref, data2, PacketSize) && "internal::plset");
|
||||
|
||||
@ -678,8 +821,7 @@ template<typename Scalar,typename Packet> void packetmath_notcomplex()
|
||||
// predux_any
|
||||
for (unsigned int i = 0; i < PacketSize * sizeof(Scalar); ++i) data1_bits[i] = 0x0;
|
||||
VERIFY((!internal::predux_any(internal::pload<Packet>(data1))) && "internal::predux_any(0000)");
|
||||
for(int k=0; k<PacketSize; ++k)
|
||||
{
|
||||
for (int k = 0; k < PacketSize; ++k) {
|
||||
for (unsigned int i = 0; i < sizeof(Scalar); ++i) data1_bits[k * sizeof(Scalar) + i] = 0xff;
|
||||
VERIFY(internal::predux_any(internal::pload<Packet>(data1)) && "internal::predux_any(0101)");
|
||||
for (unsigned int i = 0; i < sizeof(Scalar); ++i) data1_bits[k * sizeof(Scalar) + i] = 0x00;
|
||||
@ -687,8 +829,8 @@ template<typename Scalar,typename Packet> void packetmath_notcomplex()
|
||||
}
|
||||
}
|
||||
|
||||
template<typename Scalar,typename Packet,bool ConjLhs,bool ConjRhs> void test_conj_helper(Scalar* data1, Scalar* data2, Scalar* ref, Scalar* pval)
|
||||
{
|
||||
template <typename Scalar, typename Packet, bool ConjLhs, bool ConjRhs>
|
||||
void test_conj_helper(Scalar* data1, Scalar* data2, Scalar* ref, Scalar* pval) {
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
|
||||
internal::conj_if<ConjLhs> cj0;
|
||||
@ -696,26 +838,25 @@ template<typename Scalar,typename Packet,bool ConjLhs,bool ConjRhs> void test_co
|
||||
internal::conj_helper<Scalar, Scalar, ConjLhs, ConjRhs> cj;
|
||||
internal::conj_helper<Packet, Packet, ConjLhs, ConjRhs> pcj;
|
||||
|
||||
for(int i=0;i<PacketSize;++i)
|
||||
{
|
||||
for (int i = 0; i < PacketSize; ++i) {
|
||||
ref[i] = cj0(data1[i]) * cj1(data2[i]);
|
||||
VERIFY(internal::isApprox(ref[i], cj.pmul(data1[i], data2[i])) && "conj_helper pmul");
|
||||
}
|
||||
internal::pstore(pval, pcj.pmul(internal::pload<Packet>(data1), internal::pload<Packet>(data2)));
|
||||
VERIFY(test::areApprox(ref, pval, PacketSize) && "conj_helper pmul");
|
||||
|
||||
for(int i=0;i<PacketSize;++i)
|
||||
{
|
||||
for (int i = 0; i < PacketSize; ++i) {
|
||||
Scalar tmp = ref[i];
|
||||
ref[i] += cj0(data1[i]) * cj1(data2[i]);
|
||||
VERIFY(internal::isApprox(ref[i], cj.pmadd(data1[i], data2[i], tmp)) && "conj_helper pmadd");
|
||||
}
|
||||
internal::pstore(pval,pcj.pmadd(internal::pload<Packet>(data1),internal::pload<Packet>(data2),internal::pload<Packet>(pval)));
|
||||
internal::pstore(
|
||||
pval, pcj.pmadd(internal::pload<Packet>(data1), internal::pload<Packet>(data2), internal::pload<Packet>(pval)));
|
||||
VERIFY(test::areApprox(ref, pval, PacketSize) && "conj_helper pmadd");
|
||||
}
|
||||
|
||||
template<typename Scalar,typename Packet> void packetmath_complex()
|
||||
{
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath_complex() {
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
|
||||
const int size = PacketSize * 4;
|
||||
@ -724,8 +865,7 @@ template<typename Scalar,typename Packet> void packetmath_complex()
|
||||
EIGEN_ALIGN_MAX Scalar ref[PacketSize * 4];
|
||||
EIGEN_ALIGN_MAX Scalar pval[PacketSize * 4];
|
||||
|
||||
for (int i=0; i<size; ++i)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data1[i] = internal::random<Scalar>() * Scalar(1e2);
|
||||
data2[i] = internal::random<Scalar>() * Scalar(1e2);
|
||||
}
|
||||
@ -736,15 +876,14 @@ template<typename Scalar,typename Packet> void packetmath_complex()
|
||||
test_conj_helper<Scalar, Packet, true, true>(data1, data2, ref, pval);
|
||||
|
||||
{
|
||||
for(int i=0;i<PacketSize;++i)
|
||||
ref[i] = Scalar(std::imag(data1[i]),std::real(data1[i]));
|
||||
for (int i = 0; i < PacketSize; ++i) ref[i] = Scalar(std::imag(data1[i]), std::real(data1[i]));
|
||||
internal::pstore(pval, internal::pcplxflip(internal::pload<Packet>(data1)));
|
||||
VERIFY(test::areApprox(ref, pval, PacketSize) && "pcplxflip");
|
||||
}
|
||||
}
|
||||
|
||||
template<typename Scalar,typename Packet> void packetmath_scatter_gather()
|
||||
{
|
||||
template <typename Scalar, typename Packet>
|
||||
void packetmath_scatter_gather() {
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
const int PacketSize = internal::unpacket_traits<Packet>::size;
|
||||
EIGEN_ALIGN_MAX Scalar data1[PacketSize];
|
||||
@ -762,11 +901,9 @@ template<typename Scalar,typename Packet> void packetmath_scatter_gather()
|
||||
|
||||
for (int i = 0; i < PacketSize * 20; ++i) {
|
||||
if ((i % stride) == 0 && i < stride * PacketSize) {
|
||||
VERIFY(
|
||||
test::isApproxAbs(buffer[i], data1[i/stride], refvalue) && "pscatter");
|
||||
VERIFY(test::isApproxAbs(buffer[i], data1[i / stride], refvalue) && "pscatter");
|
||||
} else {
|
||||
VERIFY(
|
||||
test::isApproxAbs(buffer[i], Scalar(0), refvalue) && "pscatter");
|
||||
VERIFY(test::isApproxAbs(buffer[i], Scalar(0), refvalue) && "pscatter");
|
||||
}
|
||||
}
|
||||
|
||||
@ -811,15 +948,12 @@ struct runall<Scalar,PacketType,true,false> { // i.e. complex
|
||||
}
|
||||
};
|
||||
|
||||
}
|
||||
}
|
||||
} // namespace test
|
||||
} // namespace Eigen
|
||||
|
||||
|
||||
EIGEN_DECLARE_TEST(packetmath)
|
||||
{
|
||||
EIGEN_DECLARE_TEST(packetmath) {
|
||||
g_first_pass = true;
|
||||
for (int i = 0; i < g_repeat; i++) {
|
||||
|
||||
CALL_SUBTEST_1(test::runner<float>::run());
|
||||
CALL_SUBTEST_2(test::runner<double>::run());
|
||||
CALL_SUBTEST_3(test::runner<int8_t>::run());
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user