1
0
Fork 0
forked from suyu/suyu

common: Implement a ring buffer

This commit is contained in:
MerryMage 2018-09-08 14:28:39 +01:00
parent a6efff8b02
commit 112351d557
4 changed files with 243 additions and 0 deletions

View file

@ -71,6 +71,7 @@ add_library(common STATIC
param_package.cpp
param_package.h
quaternion.h
ring_buffer.h
scm_rev.cpp
scm_rev.h
scope_exit.h

111
src/common/ring_buffer.h Normal file
View file

@ -0,0 +1,111 @@
// Copyright 2018 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <algorithm>
#include <array>
#include <atomic>
#include <cstddef>
#include <cstring>
#include <type_traits>
#include <vector>
#include "common/common_types.h"
namespace Common {
/// SPSC ring buffer
/// @tparam T Element type
/// @tparam capacity Number of slots in ring buffer
/// @tparam granularity Slot size in terms of number of elements
template <typename T, size_t capacity, size_t granularity = 1>
class RingBuffer {
/// A "slot" is made of `granularity` elements of `T`.
static constexpr size_t slot_size = granularity * sizeof(T);
// T must be safely memcpy-able and have a trivial default constructor.
static_assert(std::is_trivial_v<T>);
// Ensure capacity is sensible.
static_assert(capacity < std::numeric_limits<size_t>::max() / 2 / granularity);
static_assert((capacity & (capacity - 1)) == 0, "capacity must be a power of two");
// Ensure lock-free.
static_assert(std::atomic<size_t>::is_always_lock_free);
public:
/// Pushes slots into the ring buffer
/// @param new_slots Pointer to the slots to push
/// @param slot_count Number of slots to push
/// @returns The number of slots actually pushed
size_t Push(const void* new_slots, size_t slot_count) {
const size_t write_index = m_write_index.load();
const size_t slots_free = capacity + m_read_index.load() - write_index;
const size_t push_count = std::min(slot_count, slots_free);
const size_t pos = write_index % capacity;
const size_t first_copy = std::min(capacity - pos, push_count);
const size_t second_copy = push_count - first_copy;
const char* in = static_cast<const char*>(new_slots);
std::memcpy(m_data.data() + pos * granularity, in, first_copy * slot_size);
in += first_copy * slot_size;
std::memcpy(m_data.data(), in, second_copy * slot_size);
m_write_index.store(write_index + push_count);
return push_count;
}
size_t Push(const std::vector<T>& input) {
return Push(input.data(), input.size());
}
/// Pops slots from the ring buffer
/// @param output Where to store the popped slots
/// @param max_slots Maximum number of slots to pop
/// @returns The number of slots actually popped
size_t Pop(void* output, size_t max_slots = ~size_t(0)) {
const size_t read_index = m_read_index.load();
const size_t slots_filled = m_write_index.load() - read_index;
const size_t pop_count = std::min(slots_filled, max_slots);
const size_t pos = read_index % capacity;
const size_t first_copy = std::min(capacity - pos, pop_count);
const size_t second_copy = pop_count - first_copy;
char* out = static_cast<char*>(output);
std::memcpy(out, m_data.data() + pos * granularity, first_copy * slot_size);
out += first_copy * slot_size;
std::memcpy(out, m_data.data(), second_copy * slot_size);
m_read_index.store(read_index + pop_count);
return pop_count;
}
std::vector<T> Pop(size_t max_slots = ~size_t(0)) {
std::vector<T> out(std::min(max_slots, capacity) * granularity);
const size_t count = Pop(out.data(), out.size() / granularity);
out.resize(count * granularity);
return out;
}
/// @returns Number of slots used
size_t Size() const {
return m_write_index.load() - m_read_index.load();
}
/// @returns Maximum size of ring buffer
constexpr size_t Capacity() const {
return capacity;
}
private:
// It is important to align the below variables for performance reasons:
// Having them on the same cache-line would result in false-sharing between them.
alignas(128) std::atomic<size_t> m_read_index{0};
alignas(128) std::atomic<size_t> m_write_index{0};
std::array<T, granularity * capacity> m_data;
};
} // namespace Common

View file

@ -1,5 +1,6 @@
add_executable(tests
common/param_package.cpp
common/ring_buffer.cpp
core/arm/arm_test_common.cpp
core/arm/arm_test_common.h
core/core_timing.cpp

View file

@ -0,0 +1,130 @@
// Copyright 2018 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <cstddef>
#include <numeric>
#include <thread>
#include <vector>
#include <catch2/catch.hpp>
#include "common/ring_buffer.h"
namespace Common {
TEST_CASE("RingBuffer: Basic Tests", "[common]") {
RingBuffer<char, 4, 1> buf;
// Pushing values into a ring buffer with space should succeed.
for (size_t i = 0; i < 4; i++) {
const char elem = static_cast<char>(i);
const size_t count = buf.Push(&elem, 1);
REQUIRE(count == 1);
}
REQUIRE(buf.Size() == 4);
// Pushing values into a full ring buffer should fail.
{
const char elem = static_cast<char>(42);
const size_t count = buf.Push(&elem, 1);
REQUIRE(count == 0);
}
REQUIRE(buf.Size() == 4);
// Popping multiple values from a ring buffer with values should succeed.
{
const std::vector<char> popped = buf.Pop(2);
REQUIRE(popped.size() == 2);
REQUIRE(popped[0] == 0);
REQUIRE(popped[1] == 1);
}
REQUIRE(buf.Size() == 2);
// Popping a single value from a ring buffer with values should succeed.
{
const std::vector<char> popped = buf.Pop(1);
REQUIRE(popped.size() == 1);
REQUIRE(popped[0] == 2);
}
REQUIRE(buf.Size() == 1);
// Pushing more values than space available should partially suceed.
{
std::vector<char> to_push(6);
std::iota(to_push.begin(), to_push.end(), 88);
const size_t count = buf.Push(to_push);
REQUIRE(count == 3);
}
REQUIRE(buf.Size() == 4);
// Doing an unlimited pop should pop all values.
{
const std::vector<char> popped = buf.Pop();
REQUIRE(popped.size() == 4);
REQUIRE(popped[0] == 3);
REQUIRE(popped[1] == 88);
REQUIRE(popped[2] == 89);
REQUIRE(popped[3] == 90);
}
REQUIRE(buf.Size() == 0);
}
TEST_CASE("RingBuffer: Threaded Test", "[common]") {
RingBuffer<char, 4, 2> buf;
const char seed = 42;
const size_t count = 1000000;
size_t full = 0;
size_t empty = 0;
const auto next_value = [](std::array<char, 2>& value) {
value[0] += 1;
value[1] += 2;
};
std::thread producer{[&] {
std::array<char, 2> value = {seed, seed};
size_t i = 0;
while (i < count) {
if (const size_t c = buf.Push(&value[0], 1); c > 0) {
REQUIRE(c == 1);
i++;
next_value(value);
} else {
full++;
std::this_thread::yield();
}
}
}};
std::thread consumer{[&] {
std::array<char, 2> value = {seed, seed};
size_t i = 0;
while (i < count) {
if (const std::vector<char> v = buf.Pop(1); v.size() > 0) {
REQUIRE(v.size() == 2);
REQUIRE(v[0] == value[0]);
REQUIRE(v[1] == value[1]);
i++;
next_value(value);
} else {
empty++;
std::this_thread::yield();
}
}
}};
producer.join();
consumer.join();
REQUIRE(buf.Size() == 0);
printf("RingBuffer: Threaded Test: full: %zu, empty: %zu\n", full, empty);
}
} // namespace Common