Mercury/N64Recomp
1
0
Fork 0
Code Issues Pull requests Packages Projects Releases Wiki Activity

Integrated SDL2 and switch to a test rom, fixed some issues in recompilation

Browse source
This commit is contained in:
Mr-Wiseguy 2022-11-20 12:51:08 -05:00
parent d0c3eb73ec
commit aad1bac933
15 changed files with 5590 additions and 7515 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
.gitignore vendored
View file

@ -13,6 +13,7 @@ build/
*.o
# Windows build output
*.exe
# User-specific files
*.rsuser
@ -40,5 +41,12 @@ bld/
# Visual Studio 2015/2017 cache/options directory
.vs/
# Libraries (binaries that aren't in the repo)
test/Lib
# RT64 (since it's not public yet)
test/RT64
# Runtime files
imgui.ini
rt64.log

50
src/main.cpp
View file

@ -211,7 +211,7 @@ std::unordered_set<std::string> ignored_funcs {
"__osGetTLBPageMask",
"__osGetTLBASID",
"__osProbeTLB",
// Coprocessor 0 functions
// Coprocessor 0/1 functions
"__osSetCount",
"osGetCount",
"__osSetSR",
@ -224,15 +224,35 @@ std::unordered_set<std::string> ignored_funcs {
"__osGetConfig",
"__osSetWatchLo",
"__osGetWatchLo",
"__osSetFpcCsr",
// Cache funcs
"osInvalDCache",
"osInvalICache",
"osWritebackDCache",
"osWritebackDCacheAll"
"osWritebackDCacheAll",
// Microcodes
"rspbootTextStart",
"gspF3DEX2_fifoTextStart",
"gspS2DEX2_fifoTextStart",
"gspL3DEX2_fifoTextStart",
};
std::unordered_set<std::string> renamed_funcs{
"sincosf"
"sincosf",
"sqrtf",
"memcpy",
"memset",
"strchr",
};
// Functions that weren't declared properly and thus have no size in the elf
std::unordered_map<std::string, size_t> unsized_funcs{
{ "guMtxF2L", 0x64 },
{ "guScaleF", 0x48 },
{ "guTranslateF", 0x48 },
{ "guMtxIdentF", 0x48 },
{ "sqrtf", 0x8 },
{ "guMtxIdent", 0x4C },
};
int main(int argc, char** argv) {
@ -308,21 +328,32 @@ int main(int argc, char** argv) {
unsigned char type;
ELFIO::Elf_Half section_index;
unsigned char other;
bool ignored = false;
// Read symbol properties
symbols.get_symbol(sym_index, name, value, size, bind, type,
section_index, other);
// Check if this symbol is unsized and if so populate its size from the unsized_funcs map
if (size == 0) {
auto size_find = unsized_funcs.find(name);
if (size_find != unsized_funcs.end()) {
size = size_find->second;
type = ELFIO::STT_FUNC;
}
}
if (ignored_funcs.contains(name)) {
name = name + "_recomp";
ignored = true;
}
// Check if this symbol is a function or has no type (like a regular glabel would)
// Symbols with no type have a dummy entry created so that their symbol can be looked up for function calls
if (type == ELFIO::STT_FUNC || type == ELFIO::STT_NOTYPE) {
bool ignored = false;
if (ignored || type == ELFIO::STT_FUNC || type == ELFIO::STT_NOTYPE || type == ELFIO::STT_OBJECT) {
if (renamed_funcs.contains(name)) {
name = "_" + name;
}
if (ignored_funcs.contains(name)) {
name = name + "_recomp";
ignored = true;
ignored = false;
}
if (section_index < section_rom_addrs.size()) {
auto section_rom_addr = section_rom_addrs[section_index];
@ -377,6 +408,7 @@ int main(int argc, char** argv) {
//#pragma omp parallel for
for (size_t i = 0; i < context.functions.size(); i++) {
const auto& func = context.functions[i];
if (!func.ignored && func.words.size() != 0) {
fmt::print(func_header_file,
"void {}(uint8_t* restrict rdram, recomp_context* restrict ctx);\n", func.name);

5
src/recompilation.cpp
View file

@ -128,6 +128,7 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::F
case InstrId::cpu_subu:
print_line("{}{} = SUB32({}{}, {}{})", ctx_gpr_prefix(rd), rd, ctx_gpr_prefix(rs), rs, ctx_gpr_prefix(rt), rt);
break;
case InstrId::cpu_addi:
case InstrId::cpu_addiu:
print_line("{}{} = ADD32({}{}, {:#X})", ctx_gpr_prefix(rt), rt, ctx_gpr_prefix(rs), rs, (int16_t)imm);
break;
@ -464,7 +465,7 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::F
// Cop1 compares
case InstrId::cpu_c_lt_s:
if ((fs & 1) == 0 && (ft & 1) == 0) {
print_line("c1cs = ctx->f{}.fl <= ctx->f{}.fl", fs, ft);
print_line("c1cs = ctx->f{}.fl < ctx->f{}.fl", fs, ft);
} else {
fmt::print(stderr, "Invalid operand for c.lt.s: f{} f{}\n", fs, ft);
return false;
@ -472,7 +473,7 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::F
break;
case InstrId::cpu_c_lt_d:
if ((fs & 1) == 0 && (ft & 1) == 0) {
print_line("c1cs = ctx->f{}.d <= ctx->f{}.d", fs, ft);
print_line("c1cs = ctx->f{}.d < ctx->f{}.d", fs, ft);
} else {
fmt::print(stderr, "Invalid operand for c.lt.d: f{} f{}\n", fs, ft);
return false;

2011
test/RecompTest.vcxproj
View file

File diff suppressed because it is too large Load diff

6134
test/RecompTest.vcxproj.filters
View file

File diff suppressed because it is too large Load diff

54
test/portultra/events.cpp
View file

@ -3,8 +3,10 @@
#include <chrono>
#include <cinttypes>
#include <variant>
#include <unordered_map>
#include <Windows.h>
#include "SDL.h"
#include "blockingconcurrentqueue.h"
#include "ultra64.h"
@ -85,7 +87,7 @@ extern "C" void osViSetEvent(RDRAM_ARG PTR(OSMesgQueue) mq_, OSMesg msg, u32 ret
events_context.vi.retrace_count = retrace_count;
}
constexpr uint32_t speed_multiplier = 10;
constexpr uint32_t speed_multiplier = 1;
// N64 CPU counter ticks per millisecond
constexpr uint32_t counter_per_ms = 46'875 * speed_multiplier;
@ -176,11 +178,57 @@ void dp_complete() {
void RT64Init(uint8_t* rom, uint8_t* rdram);
void RT64SendDL(uint8_t* rdram, const OSTask* task);
void RT64UpdateScreen(uint32_t vi_origin);
void RT64PumpEvents();
std::unordered_map<SDL_Scancode, int> button_map{
{ SDL_Scancode::SDL_SCANCODE_LEFT, 0x0002 },
{ SDL_Scancode::SDL_SCANCODE_RIGHT, 0x0001 },
{ SDL_Scancode::SDL_SCANCODE_UP, 0x0008 },
{ SDL_Scancode::SDL_SCANCODE_DOWN, 0x0004 }
};
extern int button;
extern int stick_x;
extern int stick_y;
int sdl_event_filter(void* userdata, SDL_Event* event) {
switch (event->type) {
case SDL_EventType::SDL_KEYUP:
case SDL_EventType::SDL_KEYDOWN:
{
const Uint8* key_states = SDL_GetKeyboardState(nullptr);
int new_button = 0;
for (const auto& mapping : button_map) {
if (key_states[mapping.first]) {
new_button |= mapping.second;
}
}
button = new_button;
stick_x = 127 * (key_states[SDL_Scancode::SDL_SCANCODE_D] - key_states[SDL_Scancode::SDL_SCANCODE_A]);
stick_y = 127 * (key_states[SDL_Scancode::SDL_SCANCODE_W] - key_states[SDL_Scancode::SDL_SCANCODE_S]);
}
break;
case SDL_EventType::SDL_QUIT:
std::quick_exit(ERROR_SUCCESS);
break;
}
return 1;
}
void gfx_thread_func(uint8_t* rdram, uint8_t* rom) {
using namespace std::chrono_literals;
if (SDL_Init(SDL_INIT_VIDEO | SDL_INIT_JOYSTICK) < 0) {
fprintf(stderr, "Failed to initialize SDL2: %s\n", SDL_GetError());
std::quick_exit(EXIT_FAILURE);
}
RT64Init(rom, rdram);
SDL_Window* window = SDL_GetWindowFromID(1);
// TODO set this window title in RT64, create the window here and send it to RT64, or something else entirely
// as the current window name visibly changes as RT64 is initialized
SDL_SetWindowTitle(window, "Recomp");
SDL_SetEventFilter(sdl_event_filter, nullptr);
while (true) {
// Try to pull an action from the queue
@ -205,7 +253,7 @@ void gfx_thread_func(uint8_t* rdram, uint8_t* rom) {
}
// Handle events
RT64PumpEvents();
SDL_PumpEvents();
}
}

0
test/portultra/init.cpp → test/portultra/ultrainit.cpp
View file

18
test/src/cont.cpp
View file

@ -16,15 +16,27 @@ struct OSContPad {
u8 errno_;
};
int button = 0;
int stick_x = 0;
int stick_y = 0;
void press_button(int button) {
}
void release_button(int button) {
}
extern "C" void osContGetReadData_recomp(uint8_t* restrict rdram, recomp_context* restrict ctx) {
int32_t pad = (uint32_t)ctx->r4;
// button
MEM_H(0, pad) = 0;
MEM_H(0, pad) = button;
// stick_x
MEM_B(2, pad) = 0;
MEM_B(2, pad) = stick_x;
// stick_y
MEM_B(3, pad) = 0;
MEM_B(3, pad) = stick_y;
// errno
MEM_B(4, pad) = 0;
}

4
test/src/misc_ultra.cpp
View file

@ -41,3 +41,7 @@ extern "C" void __osDisableInt_recomp(uint8_t* restrict rdram, recomp_context* r
extern "C" void __osRestoreInt_recomp(uint8_t* restrict rdram, recomp_context* restrict ctx) {
;
}
extern "C" void __osSetFpcCsr_recomp(uint8_t * restrict rdram, recomp_context * restrict ctx) {
ctx->r2 = 0;
}

43
test/src/pi.cpp
View file

@ -16,7 +16,7 @@ extern "C" void osCreatePiManager_recomp(uint8_t* restrict rdram, recomp_context
constexpr uint32_t rom_base = 0xB0000000;
void do_rom_read(uint8_t* rdram, int32_t ram_address, uint32_t dev_address, size_t num_bytes) {
void do_rom_read(uint8_t* rdram, gpr ram_address, uint32_t dev_address, size_t num_bytes) {
// TODO use word copies when possible
uint8_t* rom_addr = rom.get() + (dev_address | rom_base) - rom_base;
for (size_t i = 0; i < num_bytes; i++) {
@ -30,10 +30,9 @@ extern "C" void osPiStartDma_recomp(uint8_t* restrict rdram, recomp_context* res
uint32_t pri = ctx->r5;
uint32_t direction = ctx->r6;
uint32_t devAddr = ctx->r7;
int32_t dramAddr = MEM_W(0x10, ctx->r29);
gpr dramAddr = MEM_W(0x10, ctx->r29);
uint32_t size = MEM_W(0x14, ctx->r29);
uint32_t mq_ = MEM_W(0x18, ctx->r29);
OSMesgQueue* mq = TO_PTR(OSMesgQueue, mq_);
PTR(OSMesgQueue) mq = MEM_W(0x18, ctx->r29);
debug_printf("[pi] DMA from 0x%08X into 0x%08X of size 0x%08X\n", devAddr, dramAddr, size);
@ -44,11 +43,43 @@ extern "C" void osPiStartDma_recomp(uint8_t* restrict rdram, recomp_context* res
//memcpy(rdram + (dramAddr & 0x3FFFFFF), rom.get() + (devAddr | rom_base) - rom_base, num_bytes);
// Send a message to the mq to indicate that the transfer completed
osSendMesg(rdram, mq_, 0, OS_MESG_NOBLOCK);
osSendMesg(rdram, mq, 0, OS_MESG_NOBLOCK);
}
struct OSIoMesgHdr {
// These 3 reversed due to endianness
u8 status; /* Return status */
u8 pri; /* Message priority (High or Normal) */
u16 type; /* Message type */
PTR(OSMesgQueue) retQueue; /* Return message queue to notify I/O completion */
};
struct OSIoMesg {
OSIoMesgHdr hdr; /* Message header */
PTR(void) dramAddr; /* RDRAM buffer address (DMA) */
u32 devAddr; /* Device buffer address (DMA) */
u32 size; /* DMA transfer size in bytes */
u32 piHandle; /* PI device handle */
};
extern "C" void osEPiStartDma_recomp(uint8_t* restrict rdram, recomp_context* restrict ctx) {
;
OSIoMesg* mb = TO_PTR(OSIoMesg, ctx->r5);
uint32_t direction = ctx->r6;
uint32_t devAddr = mb->devAddr;
gpr dramAddr = mb->dramAddr;
uint32_t size = mb->size;
PTR(OSMesgQueue) mq = mb->hdr.retQueue;
debug_printf("[pi] DMA from 0x%08X into 0x%08X of size 0x%08X\n", devAddr, dramAddr, size);
// TODO asynchronous transfer (will require preemption in the scheduler)
// TODO this won't handle unaligned DMA
do_rom_read(rdram, dramAddr, devAddr, size);
//memcpy(rdram + (dramAddr & 0x3FFFFFF), rom.get() + (devAddr | rom_base) - rom_base, num_bytes);
// Send a message to the mq to indicate that the transfer completed
osSendMesg(rdram, mq, 0, OS_MESG_NOBLOCK);
}
extern "C" void osPiGetStatus_recomp(uint8_t * restrict rdram, recomp_context * restrict ctx) {

4
test/src/portultra_translation.cpp
View file

@ -5,6 +5,10 @@ extern "C" void osInitialize_recomp(uint8_t * restrict rdram, recomp_context * r
osInitialize();
}
extern "C" void __osInitialize_common_recomp(uint8_t * restrict rdram, recomp_context * restrict ctx) {
osInitialize();
}
extern "C" void osCreateThread_recomp(uint8_t* restrict rdram, recomp_context* restrict ctx) {
osCreateThread(rdram, (uint32_t)ctx->r4, (OSId)ctx->r5, (uint32_t)ctx->r6, (uint32_t)ctx->r7,
(uint32_t)MEM_W(0x10, ctx->r29), (OSPri)MEM_W(0x14, ctx->r29));

20
test/src/recomp.cpp
View file

@ -61,8 +61,12 @@ void run_thread_function(uint8_t* rdram, uint64_t addr, uint64_t sp, uint64_t ar
func(rdram, &ctx);
}
extern "C" void game_init(uint8_t* restrict rdram, recomp_context* restrict ctx);
void do_rom_read(uint8_t* rdram, int32_t ram_address, uint32_t dev_address, size_t num_bytes);
extern "C" void init(uint8_t * restrict rdram, recomp_context * restrict ctx);
// rocket robot
//extern "C" void game_init(uint8_t* restrict rdram, recomp_context* restrict ctx);
// test rom
void do_rom_read(uint8_t* rdram, gpr ram_address, uint32_t dev_address, size_t num_bytes);
std::unique_ptr<uint8_t[]> rom;
size_t rom_size;
@ -105,7 +109,7 @@ int main(int argc, char **argv) {
// Get entrypoint from ROM
// TODO fix this for other IPL3 versions
int32_t entrypoint = byteswap(*reinterpret_cast<uint32_t*>(rom.get() + 0x8));
gpr entrypoint = (int32_t)byteswap(*reinterpret_cast<uint32_t*>(rom.get() + 0x8));
// Allocate rdram_buffer
std::unique_ptr<uint8_t[]> rdram_buffer = std::make_unique<uint8_t[]>(8 * 1024 * 1024);
@ -153,13 +157,19 @@ int main(int argc, char **argv) {
// Clear bss
// TODO run the entrypoint instead
memset(rdram_buffer.get() + 0XAF860, 0, 0xC00A0u - 0XAF860);
// rocket robot
//memset(rdram_buffer.get() + 0xAF860, 0, 0xC00A0u - 0XAF860);
// test rom
memset(rdram_buffer.get() + 0x18670, 0, 0x20D120);
debug_printf("[Recomp] Starting\n");
Multilibultra::preinit(rdram_buffer.get(), rom.get());
game_init(rdram_buffer.get(), &context);
// rocket robot
// game_init(rdram_buffer.get(), &context);
// test rom
init(rdram_buffer.get(), &context);
debug_printf("[Recomp] Quitting\n");

582
test/thirdparty/blockingconcurrentqueue.h vendored Normal file
View file

@ -0,0 +1,582 @@
// Provides an efficient blocking version of moodycamel::ConcurrentQueue.
// ©2015-2020 Cameron Desrochers. Distributed under the terms of the simplified
// BSD license, available at the top of concurrentqueue.h.
// Also dual-licensed under the Boost Software License (see LICENSE.md)
// Uses Jeff Preshing's semaphore implementation (under the terms of its
// separate zlib license, see lightweightsemaphore.h).
#pragma once
#include "concurrentqueue.h"
#include "lightweightsemaphore.h"
#include <type_traits>
#include <cerrno>
#include <memory>
#include <chrono>
#include <ctime>
namespace moodycamel
{
// This is a blocking version of the queue. It has an almost identical interface to
// the normal non-blocking version, with the addition of various wait_dequeue() methods
// and the removal of producer-specific dequeue methods.
template<typename T, typename Traits = ConcurrentQueueDefaultTraits>
class BlockingConcurrentQueue
{
private:
typedef ::moodycamel::ConcurrentQueue<T, Traits> ConcurrentQueue;
typedef ::moodycamel::LightweightSemaphore LightweightSemaphore;
public:
typedef typename ConcurrentQueue::producer_token_t producer_token_t;
typedef typename ConcurrentQueue::consumer_token_t consumer_token_t;
typedef typename ConcurrentQueue::index_t index_t;
typedef typename ConcurrentQueue::size_t size_t;
typedef typename std::make_signed<size_t>::type ssize_t;
static const size_t BLOCK_SIZE = ConcurrentQueue::BLOCK_SIZE;
static const size_t EXPLICIT_BLOCK_EMPTY_COUNTER_THRESHOLD = ConcurrentQueue::EXPLICIT_BLOCK_EMPTY_COUNTER_THRESHOLD;
static const size_t EXPLICIT_INITIAL_INDEX_SIZE = ConcurrentQueue::EXPLICIT_INITIAL_INDEX_SIZE;
static const size_t IMPLICIT_INITIAL_INDEX_SIZE = ConcurrentQueue::IMPLICIT_INITIAL_INDEX_SIZE;
static const size_t INITIAL_IMPLICIT_PRODUCER_HASH_SIZE = ConcurrentQueue::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE;
static const std::uint32_t EXPLICIT_CONSUMER_CONSUMPTION_QUOTA_BEFORE_ROTATE = ConcurrentQueue::EXPLICIT_CONSUMER_CONSUMPTION_QUOTA_BEFORE_ROTATE;
static const size_t MAX_SUBQUEUE_SIZE = ConcurrentQueue::MAX_SUBQUEUE_SIZE;
public:
// Creates a queue with at least `capacity` element slots; note that the
// actual number of elements that can be inserted without additional memory
// allocation depends on the number of producers and the block size (e.g. if
// the block size is equal to `capacity`, only a single block will be allocated
// up-front, which means only a single producer will be able to enqueue elements
// without an extra allocation -- blocks aren't shared between producers).
// This method is not thread safe -- it is up to the user to ensure that the
// queue is fully constructed before it starts being used by other threads (this
// includes making the memory effects of construction visible, possibly with a
// memory barrier).
explicit BlockingConcurrentQueue(size_t capacity = 6 * BLOCK_SIZE)
: inner(capacity), sema(create<LightweightSemaphore, ssize_t, int>(0, (int)Traits::MAX_SEMA_SPINS), &BlockingConcurrentQueue::template destroy<LightweightSemaphore>)
{
assert(reinterpret_cast<ConcurrentQueue*>((BlockingConcurrentQueue*)1) == &((BlockingConcurrentQueue*)1)->inner && "BlockingConcurrentQueue must have ConcurrentQueue as its first member");
if (!sema) {
MOODYCAMEL_THROW(std::bad_alloc());
}
}
BlockingConcurrentQueue(size_t minCapacity, size_t maxExplicitProducers, size_t maxImplicitProducers)
: inner(minCapacity, maxExplicitProducers, maxImplicitProducers), sema(create<LightweightSemaphore, ssize_t, int>(0, (int)Traits::MAX_SEMA_SPINS), &BlockingConcurrentQueue::template destroy<LightweightSemaphore>)
{
assert(reinterpret_cast<ConcurrentQueue*>((BlockingConcurrentQueue*)1) == &((BlockingConcurrentQueue*)1)->inner && "BlockingConcurrentQueue must have ConcurrentQueue as its first member");
if (!sema) {
MOODYCAMEL_THROW(std::bad_alloc());
}
}
// Disable copying and copy assignment
BlockingConcurrentQueue(BlockingConcurrentQueue const&) MOODYCAMEL_DELETE_FUNCTION;
BlockingConcurrentQueue& operator=(BlockingConcurrentQueue const&) MOODYCAMEL_DELETE_FUNCTION;
// Moving is supported, but note that it is *not* a thread-safe operation.
// Nobody can use the queue while it's being moved, and the memory effects
// of that move must be propagated to other threads before they can use it.
// Note: When a queue is moved, its tokens are still valid but can only be
// used with the destination queue (i.e. semantically they are moved along
// with the queue itself).
BlockingConcurrentQueue(BlockingConcurrentQueue&& other) MOODYCAMEL_NOEXCEPT
: inner(std::move(other.inner)), sema(std::move(other.sema))
{ }
inline BlockingConcurrentQueue& operator=(BlockingConcurrentQueue&& other) MOODYCAMEL_NOEXCEPT
{
return swap_internal(other);
}
// Swaps this queue's state with the other's. Not thread-safe.
// Swapping two queues does not invalidate their tokens, however
// the tokens that were created for one queue must be used with
// only the swapped queue (i.e. the tokens are tied to the
// queue's movable state, not the object itself).
inline void swap(BlockingConcurrentQueue& other) MOODYCAMEL_NOEXCEPT
{
swap_internal(other);
}
private:
BlockingConcurrentQueue& swap_internal(BlockingConcurrentQueue& other)
{
if (this == &other) {
return *this;
}
inner.swap(other.inner);
sema.swap(other.sema);
return *this;
}
public:
// Enqueues a single item (by copying it).
// Allocates memory if required. Only fails if memory allocation fails (or implicit
// production is disabled because Traits::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE is 0,
// or Traits::MAX_SUBQUEUE_SIZE has been defined and would be surpassed).
// Thread-safe.
inline bool enqueue(T const& item)
{
if ((details::likely)(inner.enqueue(item))) {
sema->signal();
return true;
}
return false;
}
// Enqueues a single item (by moving it, if possible).
// Allocates memory if required. Only fails if memory allocation fails (or implicit
// production is disabled because Traits::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE is 0,
// or Traits::MAX_SUBQUEUE_SIZE has been defined and would be surpassed).
// Thread-safe.
inline bool enqueue(T&& item)
{
if ((details::likely)(inner.enqueue(std::move(item)))) {
sema->signal();
return true;
}
return false;
}
// Enqueues a single item (by copying it) using an explicit producer token.
// Allocates memory if required. Only fails if memory allocation fails (or
// Traits::MAX_SUBQUEUE_SIZE has been defined and would be surpassed).
// Thread-safe.
inline bool enqueue(producer_token_t const& token, T const& item)
{
if ((details::likely)(inner.enqueue(token, item))) {
sema->signal();
return true;
}
return false;
}
// Enqueues a single item (by moving it, if possible) using an explicit producer token.
// Allocates memory if required. Only fails if memory allocation fails (or
// Traits::MAX_SUBQUEUE_SIZE has been defined and would be surpassed).
// Thread-safe.
inline bool enqueue(producer_token_t const& token, T&& item)
{
if ((details::likely)(inner.enqueue(token, std::move(item)))) {
sema->signal();
return true;
}
return false;
}
// Enqueues several items.
// Allocates memory if required. Only fails if memory allocation fails (or
// implicit production is disabled because Traits::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE
// is 0, or Traits::MAX_SUBQUEUE_SIZE has been defined and would be surpassed).
// Note: Use std::make_move_iterator if the elements should be moved instead of copied.
// Thread-safe.
template<typename It>
inline bool enqueue_bulk(It itemFirst, size_t count)
{
if ((details::likely)(inner.enqueue_bulk(std::forward<It>(itemFirst), count))) {
sema->signal((LightweightSemaphore::ssize_t)(ssize_t)count);
return true;
}
return false;
}
// Enqueues several items using an explicit producer token.
// Allocates memory if required. Only fails if memory allocation fails
// (or Traits::MAX_SUBQUEUE_SIZE has been defined and would be surpassed).
// Note: Use std::make_move_iterator if the elements should be moved
// instead of copied.
// Thread-safe.
template<typename It>
inline bool enqueue_bulk(producer_token_t const& token, It itemFirst, size_t count)
{
if ((details::likely)(inner.enqueue_bulk(token, std::forward<It>(itemFirst), count))) {
sema->signal((LightweightSemaphore::ssize_t)(ssize_t)count);
return true;
}
return false;
}
// Enqueues a single item (by copying it).
// Does not allocate memory. Fails if not enough room to enqueue (or implicit
// production is disabled because Traits::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE
// is 0).
// Thread-safe.
inline bool try_enqueue(T const& item)
{
if (inner.try_enqueue(item)) {
sema->signal();
return true;
}
return false;
}
// Enqueues a single item (by moving it, if possible).
// Does not allocate memory (except for one-time implicit producer).
// Fails if not enough room to enqueue (or implicit production is
// disabled because Traits::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE is 0).
// Thread-safe.
inline bool try_enqueue(T&& item)
{
if (inner.try_enqueue(std::move(item))) {
sema->signal();
return true;
}
return false;
}
// Enqueues a single item (by copying it) using an explicit producer token.
// Does not allocate memory. Fails if not enough room to enqueue.
// Thread-safe.
inline bool try_enqueue(producer_token_t const& token, T const& item)
{
if (inner.try_enqueue(token, item)) {
sema->signal();
return true;
}
return false;
}
// Enqueues a single item (by moving it, if possible) using an explicit producer token.
// Does not allocate memory. Fails if not enough room to enqueue.
// Thread-safe.
inline bool try_enqueue(producer_token_t const& token, T&& item)
{
if (inner.try_enqueue(token, std::move(item))) {
sema->signal();
return true;
}
return false;
}
// Enqueues several items.
// Does not allocate memory (except for one-time implicit producer).
// Fails if not enough room to enqueue (or implicit production is
// disabled because Traits::INITIAL_IMPLICIT_PRODUCER_HASH_SIZE is 0).
// Note: Use std::make_move_iterator if the elements should be moved
// instead of copied.
// Thread-safe.
template<typename It>
inline bool try_enqueue_bulk(It itemFirst, size_t count)
{
if (inner.try_enqueue_bulk(std::forward<It>(itemFirst), count)) {
sema->signal((LightweightSemaphore::ssize_t)(ssize_t)count);
return true;
}
return false;
}
// Enqueues several items using an explicit producer token.
// Does not allocate memory. Fails if not enough room to enqueue.
// Note: Use std::make_move_iterator if the elements should be moved
// instead of copied.
// Thread-safe.
template<typename It>
inline bool try_enqueue_bulk(producer_token_t const& token, It itemFirst, size_t count)
{
if (inner.try_enqueue_bulk(token, std::forward<It>(itemFirst), count)) {
sema->signal((LightweightSemaphore::ssize_t)(ssize_t)count);
return true;
}
return false;
}
// Attempts to dequeue from the queue.
// Returns false if all producer streams appeared empty at the time they
// were checked (so, the queue is likely but not guaranteed to be empty).
// Never allocates. Thread-safe.
template<typename U>
inline bool try_dequeue(U& item)
{
if (sema->tryWait()) {
while (!inner.try_dequeue(item)) {
continue;
}
return true;
}
return false;
}
// Attempts to dequeue from the queue using an explicit consumer token.
// Returns false if all producer streams appeared empty at the time they
// were checked (so, the queue is likely but not guaranteed to be empty).
// Never allocates. Thread-safe.
template<typename U>
inline bool try_dequeue(consumer_token_t& token, U& item)
{
if (sema->tryWait()) {
while (!inner.try_dequeue(token, item)) {
continue;
}
return true;
}
return false;
}
// Attempts to dequeue several elements from the queue.
// Returns the number of items actually dequeued.
// Returns 0 if all producer streams appeared empty at the time they
// were checked (so, the queue is likely but not guaranteed to be empty).
// Never allocates. Thread-safe.
template<typename It>
inline size_t try_dequeue_bulk(It itemFirst, size_t max)
{
size_t count = 0;
max = (size_t)sema->tryWaitMany((LightweightSemaphore::ssize_t)(ssize_t)max);
while (count != max) {
count += inner.template try_dequeue_bulk<It&>(itemFirst, max - count);
}
return count;
}
// Attempts to dequeue several elements from the queue using an explicit consumer token.
// Returns the number of items actually dequeued.
// Returns 0 if all producer streams appeared empty at the time they
// were checked (so, the queue is likely but not guaranteed to be empty).
// Never allocates. Thread-safe.
template<typename It>
inline size_t try_dequeue_bulk(consumer_token_t& token, It itemFirst, size_t max)
{
size_t count = 0;
max = (size_t)sema->tryWaitMany((LightweightSemaphore::ssize_t)(ssize_t)max);
while (count != max) {
count += inner.template try_dequeue_bulk<It&>(token, itemFirst, max - count);
}
return count;
}
// Blocks the current thread until there's something to dequeue, then
// dequeues it.
// Never allocates. Thread-safe.
template<typename U>
inline void wait_dequeue(U& item)
{
while (!sema->wait()) {
continue;
}
while (!inner.try_dequeue(item)) {
continue;
}
}
// Blocks the current thread until either there's something to dequeue
// or the timeout (specified in microseconds) expires. Returns false
// without setting `item` if the timeout expires, otherwise assigns
// to `item` and returns true.
// Using a negative timeout indicates an indefinite timeout,
// and is thus functionally equivalent to calling wait_dequeue.
// Never allocates. Thread-safe.
template<typename U>
inline bool wait_dequeue_timed(U& item, std::int64_t timeout_usecs)
{
if (!sema->wait(timeout_usecs)) {
return false;
}
while (!inner.try_dequeue(item)) {
continue;
}
return true;
}
// Blocks the current thread until either there's something to dequeue
// or the timeout expires. Returns false without setting `item` if the
// timeout expires, otherwise assigns to `item` and returns true.
// Never allocates. Thread-safe.
template<typename U, typename Rep, typename Period>
inline bool wait_dequeue_timed(U& item, std::chrono::duration<Rep, Period> const& timeout)
{
return wait_dequeue_timed(item, std::chrono::duration_cast<std::chrono::microseconds>(timeout).count());
}
// Blocks the current thread until there's something to dequeue, then
// dequeues it using an explicit consumer token.
// Never allocates. Thread-safe.
template<typename U>
inline void wait_dequeue(consumer_token_t& token, U& item)
{
while (!sema->wait()) {
continue;
}
while (!inner.try_dequeue(token, item)) {
continue;
}
}
// Blocks the current thread until either there's something to dequeue
// or the timeout (specified in microseconds) expires. Returns false
// without setting `item` if the timeout expires, otherwise assigns
// to `item` and returns true.
// Using a negative timeout indicates an indefinite timeout,
// and is thus functionally equivalent to calling wait_dequeue.
// Never allocates. Thread-safe.
template<typename U>
inline bool wait_dequeue_timed(consumer_token_t& token, U& item, std::int64_t timeout_usecs)
{
if (!sema->wait(timeout_usecs)) {
return false;
}
while (!inner.try_dequeue(token, item)) {
continue;
}
return true;
}
// Blocks the current thread until either there's something to dequeue
// or the timeout expires. Returns false without setting `item` if the
// timeout expires, otherwise assigns to `item` and returns true.
// Never allocates. Thread-safe.
template<typename U, typename Rep, typename Period>
inline bool wait_dequeue_timed(consumer_token_t& token, U& item, std::chrono::duration<Rep, Period> const& timeout)
{
return wait_dequeue_timed(token, item, std::chrono::duration_cast<std::chrono::microseconds>(timeout).count());
}
// Attempts to dequeue several elements from the queue.
// Returns the number of items actually dequeued, which will
// always be at least one (this method blocks until the queue
// is non-empty) and at most max.
// Never allocates. Thread-safe.
template<typename It>
inline size_t wait_dequeue_bulk(It itemFirst, size_t max)
{
size_t count = 0;
max = (size_t)sema->waitMany((LightweightSemaphore::ssize_t)(ssize_t)max);
while (count != max) {
count += inner.template try_dequeue_bulk<It&>(itemFirst, max - count);
}
return count;
}
// Attempts to dequeue several elements from the queue.
// Returns the number of items actually dequeued, which can
// be 0 if the timeout expires while waiting for elements,
// and at most max.
// Using a negative timeout indicates an indefinite timeout,
// and is thus functionally equivalent to calling wait_dequeue_bulk.
// Never allocates. Thread-safe.
template<typename It>
inline size_t wait_dequeue_bulk_timed(It itemFirst, size_t max, std::int64_t timeout_usecs)
{
size_t count = 0;
max = (size_t)sema->waitMany((LightweightSemaphore::ssize_t)(ssize_t)max, timeout_usecs);
while (count != max) {
count += inner.template try_dequeue_bulk<It&>(itemFirst, max - count);
}
return count;
}
// Attempts to dequeue several elements from the queue.
// Returns the number of items actually dequeued, which can
// be 0 if the timeout expires while waiting for elements,
// and at most max.
// Never allocates. Thread-safe.
template<typename It, typename Rep, typename Period>
inline size_t wait_dequeue_bulk_timed(It itemFirst, size_t max, std::chrono::duration<Rep, Period> const& timeout)
{
return wait_dequeue_bulk_timed<It&>(itemFirst, max, std::chrono::duration_cast<std::chrono::microseconds>(timeout).count());
}
// Attempts to dequeue several elements from the queue using an explicit consumer token.
// Returns the number of items actually dequeued, which will
// always be at least one (this method blocks until the queue
// is non-empty) and at most max.
// Never allocates. Thread-safe.
template<typename It>
inline size_t wait_dequeue_bulk(consumer_token_t& token, It itemFirst, size_t max)
{
size_t count = 0;
max = (size_t)sema->waitMany((LightweightSemaphore::ssize_t)(ssize_t)max);
while (count != max) {
count += inner.template try_dequeue_bulk<It&>(token, itemFirst, max - count);
}
return count;
}
// Attempts to dequeue several elements from the queue using an explicit consumer token.
// Returns the number of items actually dequeued, which can
// be 0 if the timeout expires while waiting for elements,
// and at most max.
// Using a negative timeout indicates an indefinite timeout,
// and is thus functionally equivalent to calling wait_dequeue_bulk.
// Never allocates. Thread-safe.
template<typename It>
inline size_t wait_dequeue_bulk_timed(consumer_token_t& token, It itemFirst, size_t max, std::int64_t timeout_usecs)
{
size_t count = 0;
max = (size_t)sema->waitMany((LightweightSemaphore::ssize_t)(ssize_t)max, timeout_usecs);
while (count != max) {
count += inner.template try_dequeue_bulk<It&>(token, itemFirst, max - count);
}
return count;
}
// Attempts to dequeue several elements from the queue using an explicit consumer token.
// Returns the number of items actually dequeued, which can
// be 0 if the timeout expires while waiting for elements,
// and at most max.
// Never allocates. Thread-safe.
template<typename It, typename Rep, typename Period>
inline size_t wait_dequeue_bulk_timed(consumer_token_t& token, It itemFirst, size_t max, std::chrono::duration<Rep, Period> const& timeout)
{
return wait_dequeue_bulk_timed<It&>(token, itemFirst, max, std::chrono::duration_cast<std::chrono::microseconds>(timeout).count());
}
// Returns an estimate of the total number of elements currently in the queue. This
// estimate is only accurate if the queue has completely stabilized before it is called
// (i.e. all enqueue and dequeue operations have completed and their memory effects are
// visible on the calling thread, and no further operations start while this method is
// being called).
// Thread-safe.
inline size_t size_approx() const
{
return (size_t)sema->availableApprox();
}
// Returns true if the underlying atomic variables used by
// the queue are lock-free (they should be on most platforms).
// Thread-safe.
static constexpr bool is_lock_free()
{
return ConcurrentQueue::is_lock_free();
}
private:
template<typename U, typename A1, typename A2>
static inline U* create(A1&& a1, A2&& a2)
{
void* p = (Traits::malloc)(sizeof(U));
return p != nullptr ? new (p) U(std::forward<A1>(a1), std::forward<A2>(a2)) : nullptr;
}
template<typename U>
static inline void destroy(U* p)
{
if (p != nullptr) {
p->~U();
}
(Traits::free)(p);
}
private:
ConcurrentQueue inner;
std::unique_ptr<LightweightSemaphore, void (*)(LightweightSemaphore*)> sema;
};
template<typename T, typename Traits>
inline void swap(BlockingConcurrentQueue<T, Traits>& a, BlockingConcurrentQueue<T, Traits>& b) MOODYCAMEL_NOEXCEPT
{
a.swap(b);
}
} // end namespace moodycamel

3747
test/thirdparty/concurrentqueue.h vendored Normal file
View file

File diff suppressed because it is too large Load diff

425
test/thirdparty/lightweightsemaphore.h vendored Normal file
View file

@ -0,0 +1,425 @@
// Provides an efficient implementation of a semaphore (LightweightSemaphore).
// This is an extension of Jeff Preshing's sempahore implementation (licensed
// under the terms of its separate zlib license) that has been adapted and
// extended by Cameron Desrochers.
#pragma once
#include <cstddef> // For std::size_t
#include <atomic>
#include <type_traits> // For std::make_signed<T>
#if defined(_WIN32)
// Avoid including windows.h in a header; we only need a handful of
// items, so we'll redeclare them here (this is relatively safe since
// the API generally has to remain stable between Windows versions).
// I know this is an ugly hack but it still beats polluting the global
// namespace with thousands of generic names or adding a .cpp for nothing.
extern "C" {
struct _SECURITY_ATTRIBUTES;
__declspec(dllimport) void* __stdcall CreateSemaphoreW(_SECURITY_ATTRIBUTES* lpSemaphoreAttributes, long lInitialCount, long lMaximumCount, const wchar_t* lpName);
__declspec(dllimport) int __stdcall CloseHandle(void* hObject);
__declspec(dllimport) unsigned long __stdcall WaitForSingleObject(void* hHandle, unsigned long dwMilliseconds);
__declspec(dllimport) int __stdcall ReleaseSemaphore(void* hSemaphore, long lReleaseCount, long* lpPreviousCount);
}
#elif defined(__MACH__)
#include <mach/mach.h>
#elif defined(__unix__)
#include <semaphore.h>
#if defined(__GLIBC_PREREQ) && defined(_GNU_SOURCE)
#if __GLIBC_PREREQ(2,30)
#define MOODYCAMEL_LIGHTWEIGHTSEMAPHORE_MONOTONIC
#endif
#endif
#endif
namespace moodycamel
{
namespace details
{
// Code in the mpmc_sema namespace below is an adaptation of Jeff Preshing's
// portable + lightweight semaphore implementations, originally from
// https://github.com/preshing/cpp11-on-multicore/blob/master/common/sema.h
// LICENSE:
// Copyright (c) 2015 Jeff Preshing
//
// This software is provided 'as-is', without any express or implied
// warranty. In no event will the authors be held liable for any damages
// arising from the use of this software.
//
// Permission is granted to anyone to use this software for any purpose,
// including commercial applications, and to alter it and redistribute it
// freely, subject to the following restrictions:
//
// 1. The origin of this software must not be misrepresented; you must not
// claim that you wrote the original software. If you use this software
// in a product, an acknowledgement in the product documentation would be
// appreciated but is not required.
// 2. Altered source versions must be plainly marked as such, and must not be
// misrepresented as being the original software.
// 3. This notice may not be removed or altered from any source distribution.
#if defined(_WIN32)
class Semaphore
{
private:
void* m_hSema;
Semaphore(const Semaphore& other) MOODYCAMEL_DELETE_FUNCTION;
Semaphore& operator=(const Semaphore& other) MOODYCAMEL_DELETE_FUNCTION;
public:
Semaphore(int initialCount = 0)
{
assert(initialCount >= 0);
const long maxLong = 0x7fffffff;
m_hSema = CreateSemaphoreW(nullptr, initialCount, maxLong, nullptr);
assert(m_hSema);
}
~Semaphore()
{
CloseHandle(m_hSema);
}
bool wait()
{
const unsigned long infinite = 0xffffffff;
return WaitForSingleObject(m_hSema, infinite) == 0;
}
bool try_wait()
{
return WaitForSingleObject(m_hSema, 0) == 0;
}
bool timed_wait(std::uint64_t usecs)
{
return WaitForSingleObject(m_hSema, (unsigned long)(usecs / 1000)) == 0;
}
void signal(int count = 1)
{
while (!ReleaseSemaphore(m_hSema, count, nullptr));
}
};
#elif defined(__MACH__)
//---------------------------------------------------------
// Semaphore (Apple iOS and OSX)
// Can't use POSIX semaphores due to http://lists.apple.com/archives/darwin-kernel/2009/Apr/msg00010.html
//---------------------------------------------------------
class Semaphore
{
private:
semaphore_t m_sema;
Semaphore(const Semaphore& other) MOODYCAMEL_DELETE_FUNCTION;
Semaphore& operator=(const Semaphore& other) MOODYCAMEL_DELETE_FUNCTION;
public:
Semaphore(int initialCount = 0)
{
assert(initialCount >= 0);
kern_return_t rc = semaphore_create(mach_task_self(), &m_sema, SYNC_POLICY_FIFO, initialCount);
assert(rc == KERN_SUCCESS);
(void)rc;
}
~Semaphore()
{
semaphore_destroy(mach_task_self(), m_sema);
}
bool wait()
{
return semaphore_wait(m_sema) == KERN_SUCCESS;
}
bool try_wait()
{
return timed_wait(0);
}
bool timed_wait(std::uint64_t timeout_usecs)
{
mach_timespec_t ts;
ts.tv_sec = static_cast<unsigned int>(timeout_usecs / 1000000);
ts.tv_nsec = static_cast<int>((timeout_usecs % 1000000) * 1000);
// added in OSX 10.10: https://developer.apple.com/library/prerelease/mac/documentation/General/Reference/APIDiffsMacOSX10_10SeedDiff/modules/Darwin.html
kern_return_t rc = semaphore_timedwait(m_sema, ts);
return rc == KERN_SUCCESS;
}
void signal()
{
while (semaphore_signal(m_sema) != KERN_SUCCESS);
}
void signal(int count)
{
while (count-- > 0)
{
while (semaphore_signal(m_sema) != KERN_SUCCESS);
}
}
};
#elif defined(__unix__)
//---------------------------------------------------------
// Semaphore (POSIX, Linux)
//---------------------------------------------------------
class Semaphore
{
private:
sem_t m_sema;
Semaphore(const Semaphore& other) MOODYCAMEL_DELETE_FUNCTION;
Semaphore& operator=(const Semaphore& other) MOODYCAMEL_DELETE_FUNCTION;
public:
Semaphore(int initialCount = 0)
{
assert(initialCount >= 0);
int rc = sem_init(&m_sema, 0, static_cast<unsigned int>(initialCount));
assert(rc == 0);
(void)rc;
}
~Semaphore()
{
sem_destroy(&m_sema);
}
bool wait()
{
// http://stackoverflow.com/questions/2013181/gdb-causes-sem-wait-to-fail-with-eintr-error
int rc;
do {
rc = sem_wait(&m_sema);
} while (rc == -1 && errno == EINTR);
return rc == 0;
}
bool try_wait()
{
int rc;
do {
rc = sem_trywait(&m_sema);
} while (rc == -1 && errno == EINTR);
return rc == 0;
}
bool timed_wait(std::uint64_t usecs)
{
struct timespec ts;
const int usecs_in_1_sec = 1000000;
const int nsecs_in_1_sec = 1000000000;
#ifdef MOODYCAMEL_LIGHTWEIGHTSEMAPHORE_MONOTONIC
clock_gettime(CLOCK_MONOTONIC, &ts);
#else
clock_gettime(CLOCK_REALTIME, &ts);
#endif
ts.tv_sec += (time_t)(usecs / usecs_in_1_sec);
ts.tv_nsec += (long)(usecs % usecs_in_1_sec) * 1000;
// sem_timedwait bombs if you have more than 1e9 in tv_nsec
// so we have to clean things up before passing it in
if (ts.tv_nsec >= nsecs_in_1_sec) {
ts.tv_nsec -= nsecs_in_1_sec;
++ts.tv_sec;
}
int rc;
do {
#ifdef MOODYCAMEL_LIGHTWEIGHTSEMAPHORE_MONOTONIC
rc = sem_clockwait(&m_sema, CLOCK_MONOTONIC, &ts);
#else
rc = sem_timedwait(&m_sema, &ts);
#endif
} while (rc == -1 && errno == EINTR);
return rc == 0;
}
void signal()
{
while (sem_post(&m_sema) == -1);
}
void signal(int count)
{
while (count-- > 0)
{
while (sem_post(&m_sema) == -1);
}
}
};
#else
#error Unsupported platform! (No semaphore wrapper available)
#endif
} // end namespace details
//---------------------------------------------------------
// LightweightSemaphore
//---------------------------------------------------------
class LightweightSemaphore
{
public:
typedef std::make_signed<std::size_t>::type ssize_t;
private:
std::atomic<ssize_t> m_count;
details::Semaphore m_sema;
int m_maxSpins;
bool waitWithPartialSpinning(std::int64_t timeout_usecs = -1)
{
ssize_t oldCount;
int spin = m_maxSpins;
while (--spin >= 0)
{
oldCount = m_count.load(std::memory_order_relaxed);
if ((oldCount > 0) && m_count.compare_exchange_strong(oldCount, oldCount - 1, std::memory_order_acquire, std::memory_order_relaxed))
return true;
std::atomic_signal_fence(std::memory_order_acquire); // Prevent the compiler from collapsing the loop.
}
oldCount = m_count.fetch_sub(1, std::memory_order_acquire);
if (oldCount > 0)
return true;
if (timeout_usecs < 0)
{
if (m_sema.wait())
return true;
}
if (timeout_usecs > 0 && m_sema.timed_wait((std::uint64_t)timeout_usecs))
return true;
// At this point, we've timed out waiting for the semaphore, but the
// count is still decremented indicating we may still be waiting on
// it. So we have to re-adjust the count, but only if the semaphore
// wasn't signaled enough times for us too since then. If it was, we
// need to release the semaphore too.
while (true)
{
oldCount = m_count.load(std::memory_order_acquire);
if (oldCount >= 0 && m_sema.try_wait())
return true;
if (oldCount < 0 && m_count.compare_exchange_strong(oldCount, oldCount + 1, std::memory_order_relaxed, std::memory_order_relaxed))
return false;
}
}
ssize_t waitManyWithPartialSpinning(ssize_t max, std::int64_t timeout_usecs = -1)
{
assert(max > 0);
ssize_t oldCount;
int spin = m_maxSpins;
while (--spin >= 0)
{
oldCount = m_count.load(std::memory_order_relaxed);
if (oldCount > 0)
{
ssize_t newCount = oldCount > max ? oldCount - max : 0;
if (m_count.compare_exchange_strong(oldCount, newCount, std::memory_order_acquire, std::memory_order_relaxed))
return oldCount - newCount;
}
std::atomic_signal_fence(std::memory_order_acquire);
}
oldCount = m_count.fetch_sub(1, std::memory_order_acquire);
if (oldCount <= 0)
{
if ((timeout_usecs == 0) || (timeout_usecs < 0 && !m_sema.wait()) || (timeout_usecs > 0 && !m_sema.timed_wait((std::uint64_t)timeout_usecs)))
{
while (true)
{
oldCount = m_count.load(std::memory_order_acquire);
if (oldCount >= 0 && m_sema.try_wait())
break;
if (oldCount < 0 && m_count.compare_exchange_strong(oldCount, oldCount + 1, std::memory_order_relaxed, std::memory_order_relaxed))
return 0;
}
}
}
if (max > 1)
return 1 + tryWaitMany(max - 1);
return 1;
}
public:
LightweightSemaphore(ssize_t initialCount = 0, int maxSpins = 10000) : m_count(initialCount), m_maxSpins(maxSpins)
{
assert(initialCount >= 0);
assert(maxSpins >= 0);
}
bool tryWait()
{
ssize_t oldCount = m_count.load(std::memory_order_relaxed);
while (oldCount > 0)
{
if (m_count.compare_exchange_weak(oldCount, oldCount - 1, std::memory_order_acquire, std::memory_order_relaxed))
return true;
}
return false;
}
bool wait()
{
return tryWait() || waitWithPartialSpinning();
}
bool wait(std::int64_t timeout_usecs)
{
return tryWait() || waitWithPartialSpinning(timeout_usecs);
}
// Acquires between 0 and (greedily) max, inclusive
ssize_t tryWaitMany(ssize_t max)
{
assert(max >= 0);
ssize_t oldCount = m_count.load(std::memory_order_relaxed);
while (oldCount > 0)
{
ssize_t newCount = oldCount > max ? oldCount - max : 0;
if (m_count.compare_exchange_weak(oldCount, newCount, std::memory_order_acquire, std::memory_order_relaxed))
return oldCount - newCount;
}
return 0;
}
// Acquires at least one, and (greedily) at most max
ssize_t waitMany(ssize_t max, std::int64_t timeout_usecs)
{
assert(max >= 0);
ssize_t result = tryWaitMany(max);
if (result == 0 && max > 0)
result = waitManyWithPartialSpinning(max, timeout_usecs);
return result;
}
ssize_t waitMany(ssize_t max)
{
ssize_t result = waitMany(max, -1);
assert(result > 0);
return result;
}
void signal(ssize_t count = 1)
{
assert(count >= 0);
ssize_t oldCount = m_count.fetch_add(count, std::memory_order_release);
ssize_t toRelease = -oldCount < count ? -oldCount : count;
if (toRelease > 0)
{
m_sema.signal((int)toRelease);
}
}
std::size_t availableApprox() const
{
ssize_t count = m_count.load(std::memory_order_relaxed);
return count > 0 ? static_cast<std::size_t>(count) : 0;
}
};
} // end namespace moodycamel