diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/common/CMakeLists.txt | 3 | ||||
| -rw-r--r-- | src/common/atomic.h | 16 | ||||
| -rw-r--r-- | src/common/atomic_gcc.h | 110 | ||||
| -rw-r--r-- | src/common/atomic_win32.h | 69 | ||||
| -rw-r--r-- | src/core/core_timing.cpp | 8 |
5 files changed, 4 insertions, 202 deletions
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index f8a55c2a7..55a5f9eba 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt | |||
| @@ -23,9 +23,6 @@ set(SRCS | |||
| 23 | ) | 23 | ) |
| 24 | 24 | ||
| 25 | set(HEADERS | 25 | set(HEADERS |
| 26 | atomic.h | ||
| 27 | atomic_gcc.h | ||
| 28 | atomic_win32.h | ||
| 29 | bit_field.h | 26 | bit_field.h |
| 30 | break_points.h | 27 | break_points.h |
| 31 | chunk_file.h | 28 | chunk_file.h |
diff --git a/src/common/atomic.h b/src/common/atomic.h deleted file mode 100644 index 941f5ad5e..000000000 --- a/src/common/atomic.h +++ /dev/null | |||
| @@ -1,16 +0,0 @@ | |||
| 1 | // Copyright 2013 Dolphin Emulator Project | ||
| 2 | // Licensed under GPLv2 | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #pragma once | ||
| 6 | |||
| 7 | #ifdef _WIN32 | ||
| 8 | |||
| 9 | #include "common/atomic_win32.h" | ||
| 10 | |||
| 11 | #else | ||
| 12 | |||
| 13 | // GCC-compatible compiler assumed! | ||
| 14 | #include "common/atomic_gcc.h" | ||
| 15 | |||
| 16 | #endif | ||
diff --git a/src/common/atomic_gcc.h b/src/common/atomic_gcc.h deleted file mode 100644 index 117e342f6..000000000 --- a/src/common/atomic_gcc.h +++ /dev/null | |||
| @@ -1,110 +0,0 @@ | |||
| 1 | // Copyright 2013 Dolphin Emulator Project | ||
| 2 | // Licensed under GPLv2 | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #pragma once | ||
| 6 | |||
| 7 | #include "common/common.h" | ||
| 8 | |||
| 9 | // Atomic operations are performed in a single step by the CPU. It is | ||
| 10 | // impossible for other threads to see the operation "half-done." | ||
| 11 | // | ||
| 12 | // Some atomic operations can be combined with different types of memory | ||
| 13 | // barriers called "Acquire semantics" and "Release semantics", defined below. | ||
| 14 | // | ||
| 15 | // Acquire semantics: Future memory accesses cannot be relocated to before the | ||
| 16 | // operation. | ||
| 17 | // | ||
| 18 | // Release semantics: Past memory accesses cannot be relocated to after the | ||
| 19 | // operation. | ||
| 20 | // | ||
| 21 | // These barriers affect not only the compiler, but also the CPU. | ||
| 22 | |||
| 23 | namespace Common | ||
| 24 | { | ||
| 25 | |||
| 26 | inline void AtomicAdd(volatile u32& target, u32 value) { | ||
| 27 | __sync_add_and_fetch(&target, value); | ||
| 28 | } | ||
| 29 | |||
| 30 | inline void AtomicAnd(volatile u32& target, u32 value) { | ||
| 31 | __sync_and_and_fetch(&target, value); | ||
| 32 | } | ||
| 33 | |||
| 34 | inline void AtomicDecrement(volatile u32& target) { | ||
| 35 | __sync_add_and_fetch(&target, -1); | ||
| 36 | } | ||
| 37 | |||
| 38 | inline void AtomicIncrement(volatile u32& target) { | ||
| 39 | __sync_add_and_fetch(&target, 1); | ||
| 40 | } | ||
| 41 | |||
| 42 | inline u32 AtomicLoad(volatile u32& src) { | ||
| 43 | return src; // 32-bit reads are always atomic. | ||
| 44 | } | ||
| 45 | inline u32 AtomicLoadAcquire(volatile u32& src) { | ||
| 46 | //keep the compiler from caching any memory references | ||
| 47 | u32 result = src; // 32-bit reads are always atomic. | ||
| 48 | //__sync_synchronize(); // TODO: May not be necessary. | ||
| 49 | // Compiler instruction only. x86 loads always have acquire semantics. | ||
| 50 | __asm__ __volatile__ ( "":::"memory" ); | ||
| 51 | return result; | ||
| 52 | } | ||
| 53 | |||
| 54 | inline void AtomicOr(volatile u32& target, u32 value) { | ||
| 55 | __sync_or_and_fetch(&target, value); | ||
| 56 | } | ||
| 57 | |||
| 58 | inline void AtomicStore(volatile u32& dest, u32 value) { | ||
| 59 | dest = value; // 32-bit writes are always atomic. | ||
| 60 | } | ||
| 61 | inline void AtomicStoreRelease(volatile u32& dest, u32 value) { | ||
| 62 | __sync_lock_test_and_set(&dest, value); // TODO: Wrong! This function is has acquire semantics. | ||
| 63 | } | ||
| 64 | |||
| 65 | } | ||
| 66 | |||
| 67 | // Old code kept here for reference in case we need the parts with __asm__ __volatile__. | ||
| 68 | #if 0 | ||
| 69 | LONG SyncInterlockedIncrement(LONG *Dest) | ||
| 70 | { | ||
| 71 | #if defined(__GNUC__) && defined (__GNUC_MINOR__) && ((4 < __GNUC__) || (4 == __GNUC__ && 1 <= __GNUC_MINOR__)) | ||
| 72 | return __sync_add_and_fetch(Dest, 1); | ||
| 73 | #else | ||
| 74 | register int result; | ||
| 75 | __asm__ __volatile__("lock; xadd %0,%1" | ||
| 76 | : "=r" (result), "=m" (*Dest) | ||
| 77 | : "0" (1), "m" (*Dest) | ||
| 78 | : "memory"); | ||
| 79 | return result; | ||
| 80 | #endif | ||
| 81 | } | ||
| 82 | |||
| 83 | LONG SyncInterlockedExchangeAdd(LONG *Dest, LONG Val) | ||
| 84 | { | ||
| 85 | #if defined(__GNUC__) && defined (__GNUC_MINOR__) && ((4 < __GNUC__) || (4 == __GNUC__ && 1 <= __GNUC_MINOR__)) | ||
| 86 | return __sync_add_and_fetch(Dest, Val); | ||
| 87 | #else | ||
| 88 | register int result; | ||
| 89 | __asm__ __volatile__("lock; xadd %0,%1" | ||
| 90 | : "=r" (result), "=m" (*Dest) | ||
| 91 | : "0" (Val), "m" (*Dest) | ||
| 92 | : "memory"); | ||
| 93 | return result; | ||
| 94 | #endif | ||
| 95 | } | ||
| 96 | |||
| 97 | LONG SyncInterlockedExchange(LONG *Dest, LONG Val) | ||
| 98 | { | ||
| 99 | #if defined(__GNUC__) && defined (__GNUC_MINOR__) && ((4 < __GNUC__) || (4 == __GNUC__ && 1 <= __GNUC_MINOR__)) | ||
| 100 | return __sync_lock_test_and_set(Dest, Val); | ||
| 101 | #else | ||
| 102 | register int result; | ||
| 103 | __asm__ __volatile__("lock; xchg %0,%1" | ||
| 104 | : "=r" (result), "=m" (*Dest) | ||
| 105 | : "0" (Val), "m" (*Dest) | ||
| 106 | : "memory"); | ||
| 107 | return result; | ||
| 108 | #endif | ||
| 109 | } | ||
| 110 | #endif | ||
diff --git a/src/common/atomic_win32.h b/src/common/atomic_win32.h deleted file mode 100644 index 0808905f0..000000000 --- a/src/common/atomic_win32.h +++ /dev/null | |||
| @@ -1,69 +0,0 @@ | |||
| 1 | // Copyright 2013 Dolphin Emulator Project | ||
| 2 | // Licensed under GPLv2 | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #pragma once | ||
| 6 | |||
| 7 | #include "common/common.h" | ||
| 8 | #include <intrin.h> | ||
| 9 | #include <Windows.h> | ||
| 10 | |||
| 11 | // Atomic operations are performed in a single step by the CPU. It is | ||
| 12 | // impossible for other threads to see the operation "half-done." | ||
| 13 | // | ||
| 14 | // Some atomic operations can be combined with different types of memory | ||
| 15 | // barriers called "Acquire semantics" and "Release semantics", defined below. | ||
| 16 | // | ||
| 17 | // Acquire semantics: Future memory accesses cannot be relocated to before the | ||
| 18 | // operation. | ||
| 19 | // | ||
| 20 | // Release semantics: Past memory accesses cannot be relocated to after the | ||
| 21 | // operation. | ||
| 22 | // | ||
| 23 | // These barriers affect not only the compiler, but also the CPU. | ||
| 24 | // | ||
| 25 | // NOTE: Acquire and Release are not differentiated right now. They perform a | ||
| 26 | // full memory barrier instead of a "one-way" memory barrier. The newest | ||
| 27 | // Windows SDK has Acquire and Release versions of some Interlocked* functions. | ||
| 28 | |||
| 29 | namespace Common | ||
| 30 | { | ||
| 31 | |||
| 32 | inline void AtomicAdd(volatile u32& target, u32 value) { | ||
| 33 | InterlockedExchangeAdd((volatile LONG*)&target, (LONG)value); | ||
| 34 | } | ||
| 35 | |||
| 36 | inline void AtomicAnd(volatile u32& target, u32 value) { | ||
| 37 | _InterlockedAnd((volatile LONG*)&target, (LONG)value); | ||
| 38 | } | ||
| 39 | |||
| 40 | inline void AtomicIncrement(volatile u32& target) { | ||
| 41 | InterlockedIncrement((volatile LONG*)&target); | ||
| 42 | } | ||
| 43 | |||
| 44 | inline void AtomicDecrement(volatile u32& target) { | ||
| 45 | InterlockedDecrement((volatile LONG*)&target); | ||
| 46 | } | ||
| 47 | |||
| 48 | inline u32 AtomicLoad(volatile u32& src) { | ||
| 49 | return src; // 32-bit reads are always atomic. | ||
| 50 | } | ||
| 51 | inline u32 AtomicLoadAcquire(volatile u32& src) { | ||
| 52 | u32 result = src; // 32-bit reads are always atomic. | ||
| 53 | _ReadBarrier(); // Compiler instruction only. x86 loads always have acquire semantics. | ||
| 54 | return result; | ||
| 55 | } | ||
| 56 | |||
| 57 | inline void AtomicOr(volatile u32& target, u32 value) { | ||
| 58 | _InterlockedOr((volatile LONG*)&target, (LONG)value); | ||
| 59 | } | ||
| 60 | |||
| 61 | inline void AtomicStore(volatile u32& dest, u32 value) { | ||
| 62 | dest = value; // 32-bit writes are always atomic. | ||
| 63 | } | ||
| 64 | inline void AtomicStoreRelease(volatile u32& dest, u32 value) { | ||
| 65 | _WriteBarrier(); // Compiler instruction only. x86 stores always have release semantics. | ||
| 66 | dest = value; // 32-bit writes are always atomic. | ||
| 67 | } | ||
| 68 | |||
| 69 | } | ||
diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp index a4fc0aaa4..c30e36732 100644 --- a/src/core/core_timing.cpp +++ b/src/core/core_timing.cpp | |||
| @@ -4,10 +4,10 @@ | |||
| 4 | 4 | ||
| 5 | #include <vector> | 5 | #include <vector> |
| 6 | #include <cstdio> | 6 | #include <cstdio> |
| 7 | #include <atomic> | ||
| 7 | 8 | ||
| 8 | #include "common/msg_handler.h" | 9 | #include "common/msg_handler.h" |
| 9 | #include "common/std_mutex.h" | 10 | #include "common/std_mutex.h" |
| 10 | #include "common/atomic.h" | ||
| 11 | #include "common/chunk_file.h" | 11 | #include "common/chunk_file.h" |
| 12 | 12 | ||
| 13 | #include "core/core_timing.h" | 13 | #include "core/core_timing.h" |
| @@ -54,7 +54,7 @@ Event *eventPool = 0; | |||
| 54 | Event *eventTsPool = 0; | 54 | Event *eventTsPool = 0; |
| 55 | int allocatedTsEvents = 0; | 55 | int allocatedTsEvents = 0; |
| 56 | // Optimization to skip MoveEvents when possible. | 56 | // Optimization to skip MoveEvents when possible. |
| 57 | volatile u32 hasTsEvents = false; | 57 | std::atomic<u32> hasTsEvents; |
| 58 | 58 | ||
| 59 | // Downcount has been moved to currentMIPS, to save a couple of clocks in every ARM JIT block | 59 | // Downcount has been moved to currentMIPS, to save a couple of clocks in every ARM JIT block |
| 60 | // as we can already reach that structure through a register. | 60 | // as we can already reach that structure through a register. |
| @@ -202,7 +202,7 @@ void ScheduleEvent_Threadsafe(s64 cyclesIntoFuture, int event_type, u64 userdata | |||
| 202 | tsLast->next = ne; | 202 | tsLast->next = ne; |
| 203 | tsLast = ne; | 203 | tsLast = ne; |
| 204 | 204 | ||
| 205 | Common::AtomicStoreRelease(hasTsEvents, 1); | 205 | hasTsEvents.store(1, std::memory_order_release); |
| 206 | } | 206 | } |
| 207 | 207 | ||
| 208 | // Same as ScheduleEvent_Threadsafe(0, ...) EXCEPT if we are already on the CPU thread | 208 | // Same as ScheduleEvent_Threadsafe(0, ...) EXCEPT if we are already on the CPU thread |
| @@ -484,7 +484,7 @@ void ProcessFifoWaitEvents() | |||
| 484 | 484 | ||
| 485 | void MoveEvents() | 485 | void MoveEvents() |
| 486 | { | 486 | { |
| 487 | Common::AtomicStoreRelease(hasTsEvents, 0); | 487 | hasTsEvents.store(0, std::memory_order_release); |
| 488 | 488 | ||
| 489 | std::lock_guard<std::recursive_mutex> lk(externalEventSection); | 489 | std::lock_guard<std::recursive_mutex> lk(externalEventSection); |
| 490 | // Move events from async queue into main queue | 490 | // Move events from async queue into main queue |