blob: 2d2398fb9ffc984906f31c7177cfdc5f37af4ea7 [file] [log] [blame]
// Copyright 2019 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include <lib/affine/assert.h>
#include <stdint.h>
#include <zircon/compiler.h>
#include <limits>
#include <type_traits>
namespace affine {
class Ratio {
enum class Exact { No, Yes };
// Used to indicate overflow/underflow of scaling operations.
static constexpr int64_t kOverflow = std::numeric_limits<int64_t>::max();
static constexpr int64_t kUnderflow = std::numeric_limits<int64_t>::min();
// Reduces the ratio of N/D
// Defined only for uint32_t and uint64_t
template <typename T>
static void Reduce(T* numerator, T* denominator);
// Reduce the ratio instance, in-place.
void Reduce() { Reduce(&numerator_, &denominator_); }
// Produces the product two 32 bit ratios. If exact is true, ASSERTs on loss
// of precision.
static void Product(uint32_t a_numerator, uint32_t a_denominator, uint32_t b_numerator,
uint32_t b_denominator, uint32_t* product_numerator,
uint32_t* product_denominator, Exact exact = Exact::Yes);
// Produces the product of a 32 bit ratio and the int64_t as an int64_t. Returns
// a saturated value (either kOverflow or kUnderflow) on overflow/underflow.
static int64_t Scale(int64_t value, uint32_t numerator, uint32_t denominator);
// Returns the product of the ratios. If exact is true, ASSERTs on loss of
// precision.
static Ratio Product(Ratio a, Ratio b, Exact exact = Exact::Yes) {
uint32_t result_numerator;
uint32_t result_denominator;
Product(a.numerator(), a.denominator(), b.numerator(), b.denominator(), &result_numerator,
&result_denominator, exact);
return Ratio(result_numerator, result_denominator);
Ratio() = default;
// TODO( : Remove these __LOCAL annotations
// So, there is something wrong with GCC when building this library with -O0
// (see the referenced bug). It does not seem to be respecting the
// -fvisibility-hidden or -fvisibility-inlines-hidden flags which are being
// passed to it.
// As a result, when this library is used by a DSO, the symbols become
// exported by the DSO and require some runtime fixup. This generated a GOT
// which is (by definition) a RW segment. This is a problem when we attempt
// to use this library in the VDSO image ( since the VDSO
// _demands_ that there be no read write segments.
// The workaround here is to put the __LOCAL annotation on the two methods
// that the VDSO images uses directly (the non-default constructor, and the
// Scale method). Under the hood, __LOCAL is attribute("hidden"), which the
// compiler does seem to respect. Once the symbols are no longer exported
// from the VDSO image, the GOT goes away and the system is happy.
// Eventually, if the issue with GCC gets resolved, we can come back and
// remove these explicit annotations.
Ratio(uint32_t numerator, uint32_t denominator) __LOCAL : numerator_(numerator),
denominator_(denominator) {
internal::DebugAssert(denominator_ != 0);
uint32_t numerator() const { return numerator_; }
uint32_t denominator() const { return denominator_; }
bool invertible() const { return numerator_ != 0; }
Ratio Inverse() const {
return Ratio{denominator_, numerator_};
int64_t Scale(int64_t value) const __LOCAL { return Scale(value, numerator_, denominator_); }
uint32_t numerator_ = 1;
uint32_t denominator_ = 1;
// Returns the ratio of the two ratios.
inline Ratio operator/(Ratio a, Ratio b) { return Ratio::Product(a, b.Inverse()); }
// Returns the product of the two ratios.
inline Ratio operator*(Ratio a, Ratio b) { return Ratio::Product(a, b); }
// Returns the product of the rate and the int64_t.
inline int64_t operator*(Ratio a, int64_t b) { return a.Scale(b); }
// Returns the product of the rate and the int64_t.
inline int64_t operator*(int64_t a, Ratio b) { return b.Scale(a); }
// Returns the the int64_t divided by the rate.
inline int64_t operator/(int64_t a, Ratio b) { return b.Inverse().Scale(a); }
} // namespace affine