PrusaSlicer-NonPlainar/src/libslic3r/MTUtils.hpp

390 lines
11 KiB
C++
Raw Normal View History

#ifndef MTUTILS_HPP
#define MTUTILS_HPP
#include <atomic> // for std::atomic_flag and memory orders
#include <mutex> // for std::lock_guard
#include <functional> // for std::function
#include <utility> // for std::forward
2019-06-18 09:24:50 +00:00
#include <vector>
#include <algorithm>
2019-06-18 09:24:50 +00:00
#include <cmath>
#include "libslic3r.h"
#include "Point.hpp"
namespace Slic3r {
/// Handy little spin mutex for the cached meshes.
/// Implements the "Lockable" concept
class SpinMutex
{
std::atomic_flag m_flg;
static const /*constexpr*/ auto MO_ACQ = std::memory_order_acquire;
static const /*constexpr*/ auto MO_REL = std::memory_order_release;
public:
inline SpinMutex() { m_flg.clear(MO_REL); }
inline void lock() { while (m_flg.test_and_set(MO_ACQ)) ; }
inline bool try_lock() { return !m_flg.test_and_set(MO_ACQ); }
inline void unlock() { m_flg.clear(MO_REL); }
};
/// A wrapper class around arbitrary object that needs thread safe caching.
template<class T> class CachedObject
{
public:
// Method type which refreshes the object when it has been invalidated
using Setter = std::function<void(T &)>;
private:
T m_obj; // the object itself
bool m_valid; // invalidation flag
SpinMutex m_lck; // to make the caching thread safe
// the setter will be called just before the object's const value is
// about to be retrieved.
std::function<void(T &)> m_setter;
public:
// Forwarded constructor
template<class... Args>
inline CachedObject(Setter fn, Args &&... args)
: m_obj(std::forward<Args>(args)...), m_valid(false), m_setter(fn)
{}
// invalidate the value of the object. The object will be refreshed at
// the next retrieval (Setter will be called). The data that is used in
// the setter function should be guarded as well during modification so
// the modification has to take place in fn.
inline void invalidate(std::function<void()> fn)
{
std::lock_guard<SpinMutex> lck(m_lck);
fn();
m_valid = false;
}
// Get the const object properly updated.
inline const T &get()
{
std::lock_guard<SpinMutex> lck(m_lck);
if (!m_valid) {
m_setter(m_obj);
m_valid = true;
}
return m_obj;
}
};
/// An std compatible random access iterator which uses indices to the
/// source vector thus resistant to invalidation caused by relocations. It
/// also "knows" its container. No comparison is neccesary to the container
/// "end()" iterator. The template can be instantiated with a different
/// value type than that of the container's but the types must be
/// compatible. E.g. a base class of the contained objects is compatible.
///
/// For a constant iterator, one can instantiate this template with a value
/// type preceded with 'const'.
template<class Vector, // The container type, must be random access...
class Value = typename Vector::value_type // The value type
>
class IndexBasedIterator
{
static const size_t NONE = size_t(-1);
std::reference_wrapper<Vector> m_index_ref;
size_t m_idx = NONE;
public:
using value_type = Value;
using pointer = Value *;
using reference = Value &;
using difference_type = long;
using iterator_category = std::random_access_iterator_tag;
inline explicit IndexBasedIterator(Vector &index, size_t idx)
: m_index_ref(index), m_idx(idx)
{}
// Post increment
inline IndexBasedIterator operator++(int)
{
IndexBasedIterator cpy(*this);
++m_idx;
return cpy;
}
inline IndexBasedIterator operator--(int)
{
IndexBasedIterator cpy(*this);
--m_idx;
return cpy;
}
inline IndexBasedIterator &operator++()
{
++m_idx;
return *this;
}
inline IndexBasedIterator &operator--()
{
--m_idx;
return *this;
}
inline IndexBasedIterator &operator+=(difference_type l)
{
m_idx += size_t(l);
return *this;
}
inline IndexBasedIterator operator+(difference_type l)
{
auto cpy = *this;
cpy += l;
return cpy;
}
inline IndexBasedIterator &operator-=(difference_type l)
{
m_idx -= size_t(l);
return *this;
}
inline IndexBasedIterator operator-(difference_type l)
{
auto cpy = *this;
cpy -= l;
return cpy;
}
operator difference_type() { return difference_type(m_idx); }
/// Tesing the end of the container... this is not possible with std
/// iterators.
inline bool is_end() const
{
return m_idx >= m_index_ref.get().size();
}
inline Value &operator*() const
{
assert(m_idx < m_index_ref.get().size());
return m_index_ref.get().operator[](m_idx);
}
inline Value *operator->() const
{
assert(m_idx < m_index_ref.get().size());
return &m_index_ref.get().operator[](m_idx);
}
/// If both iterators point past the container, they are equal...
inline bool operator==(const IndexBasedIterator &other)
{
size_t e = m_index_ref.get().size();
return m_idx == other.m_idx || (m_idx >= e && other.m_idx >= e);
}
inline bool operator!=(const IndexBasedIterator &other)
{
return !(*this == other);
}
inline bool operator<=(const IndexBasedIterator &other)
{
return (m_idx < other.m_idx) || (*this == other);
}
inline bool operator<(const IndexBasedIterator &other)
{
return m_idx < other.m_idx && (*this != other);
}
inline bool operator>=(const IndexBasedIterator &other)
{
return m_idx > other.m_idx || *this == other;
}
inline bool operator>(const IndexBasedIterator &other)
{
return m_idx > other.m_idx && *this != other;
}
};
/// A very simple range concept implementation with iterator-like objects.
template<class It> class Range
{
It from, to;
public:
// The class is ready for range based for loops.
It begin() const { return from; }
It end() const { return to; }
// The iterator type can be obtained this way.
using Type = It;
Range() = default;
Range(It &&b, It &&e)
: from(std::forward<It>(b)), to(std::forward<It>(e))
{}
// Some useful container-like methods...
inline size_t size() const { return end() - begin(); }
inline bool empty() const { return size() == 0; }
};
template<class C> bool all_of(const C &container)
{
return std::all_of(container.begin(),
container.end(),
[](const typename C::value_type &v) {
return static_cast<bool>(v);
});
}
2019-06-28 13:42:59 +00:00
template<class T> struct remove_cvref
2019-06-18 09:24:50 +00:00
{
using type =
typename std::remove_cv<typename std::remove_reference<T>::type>::type;
};
2019-06-28 13:42:59 +00:00
template<class T> using remove_cvref_t = typename remove_cvref<T>::type;
2019-06-18 09:24:50 +00:00
template<template<class> class C, class T>
2019-06-28 13:42:59 +00:00
class Container : public C<remove_cvref_t<T>>
{
2019-06-18 09:24:50 +00:00
public:
2019-06-28 13:42:59 +00:00
explicit Container(size_t count, T &&initval)
: C<remove_cvref_t<T>>(count, initval)
{}
2019-06-18 09:24:50 +00:00
};
template<class T> using DefaultContainer = std::vector<T>;
/// Exactly like Matlab https://www.mathworks.com/help/matlab/ref/linspace.html
template<class T, class I, template<class> class C = DefaultContainer>
inline C<remove_cvref_t<T>> linspace(const T &start, const T &stop, const I &n)
{
Container<C, T> vals(n, T());
2019-06-28 13:42:59 +00:00
T stride = (stop - start) / n;
size_t i = 0;
2019-06-18 09:24:50 +00:00
std::generate(vals.begin(), vals.end(), [&i, start, stride] {
2019-06-28 13:42:59 +00:00
return start + i++ * stride;
2019-06-18 09:24:50 +00:00
});
2019-06-28 13:42:59 +00:00
2019-06-18 09:24:50 +00:00
return vals;
}
/// A set of equidistant values starting from 'start' (inclusive), ending
/// in the closest multiple of 'stride' less than or equal to 'end' and
/// leaving 'stride' space between each value.
/// Very similar to Matlab [start:stride:end] notation.
template<class T, template<class> class C = DefaultContainer>
inline C<remove_cvref_t<T>> grid(const T &start, const T &stop, const T &stride)
{
Container<C, T> vals(size_t(std::ceil((stop - start) / stride)), T());
int i = 0;
std::generate(vals.begin(), vals.end(), [&i, start, stride] {
return start + i++ * stride;
});
return vals;
}
// A shorter C++14 style form of the enable_if metafunction
template<bool B, class T>
using enable_if_t = typename std::enable_if<B, T>::type;
// /////////////////////////////////////////////////////////////////////////////
// Type safe conversions to and from scaled and unscaled coordinates
// /////////////////////////////////////////////////////////////////////////////
// A meta-predicate which is true for integers wider than or equal to coord_t
template<class I> struct is_scaled_coord
{
static const SLIC3R_CONSTEXPR bool value =
std::is_integral<I>::value &&
std::numeric_limits<I>::digits >=
std::numeric_limits<coord_t>::digits;
};
// Meta predicates for floating, 'scaled coord' and generic arithmetic types
2019-07-17 14:47:09 +00:00
template<class T, class O = T>
using FloatingOnly = enable_if_t<std::is_floating_point<T>::value, O>;
2019-07-17 14:47:09 +00:00
template<class T, class O = T>
using ScaledCoordOnly = enable_if_t<is_scaled_coord<T>::value, O>;
2019-07-17 14:47:09 +00:00
template<class T, class O = T>
using ArithmeticOnly = enable_if_t<std::is_arithmetic<T>::value, O>;
// Semantics are the following:
// Upscaling (scaled()): only from floating point types (or Vec) to either
// floating point or integer 'scaled coord' coordinates.
2019-06-27 19:13:44 +00:00
// Downscaling (unscaled()): from arithmetic (or Vec) to floating point only
// Conversion definition from unscaled to floating point scaled
template<class Tout,
class Tin,
2019-07-17 12:34:28 +00:00
class = FloatingOnly<Tin>>
2019-07-17 14:47:09 +00:00
inline constexpr FloatingOnly<Tout> scaled(const Tin &v) noexcept
{
2019-06-27 19:13:44 +00:00
return Tout(v / Tin(SCALING_FACTOR));
}
// Conversion definition from unscaled to integer 'scaled coord'.
2019-06-27 19:13:44 +00:00
// TODO: is the rounding necessary? Here it is commented out to show that
// it can be different for integers but it does not have to be. Using
// std::round means loosing noexcept and constexpr modifiers
template<class Tout = coord_t, class Tin, class = FloatingOnly<Tin>>
2019-07-17 14:47:09 +00:00
inline constexpr ScaledCoordOnly<Tout> scaled(const Tin &v) noexcept
{
//return static_cast<Tout>(std::round(v / SCALING_FACTOR));
2019-06-27 19:13:44 +00:00
return Tout(v / Tin(SCALING_FACTOR));
}
// Conversion for Eigen vectors (N dimensional points)
2019-07-17 14:47:09 +00:00
template<class Tout = coord_t,
class Tin,
int N,
class = FloatingOnly<Tin>,
int...EigenArgs>
inline Eigen::Matrix<ArithmeticOnly<Tout>, N, EigenArgs...>
scaled(const Eigen::Matrix<Tin, N, EigenArgs...> &v)
{
2019-06-28 15:03:50 +00:00
return (v / SCALING_FACTOR).template cast<Tout>();
}
// Conversion from arithmetic scaled type to floating point unscaled
template<class Tout = double,
class Tin,
class = ArithmeticOnly<Tin>,
class = FloatingOnly<Tout>>
2019-07-17 14:47:09 +00:00
inline constexpr Tout unscaled(const Tin &v) noexcept
{
2019-06-27 19:13:44 +00:00
return Tout(v * Tout(SCALING_FACTOR));
}
// Unscaling for Eigen vectors. Input base type can be arithmetic, output base
// type can only be floating point.
template<class Tout = double,
class Tin,
int N,
class = ArithmeticOnly<Tin>,
2019-07-17 14:47:09 +00:00
class = FloatingOnly<Tout>,
int...EigenArgs>
inline constexpr Eigen::Matrix<Tout, N, EigenArgs...>
unscaled(const Eigen::Matrix<Tin, N, EigenArgs...> &v) noexcept
{
return v.template cast<Tout>() * SCALING_FACTOR;
}
} // namespace Slic3r
#endif // MTUTILS_HPP