birkeh/o3de
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[development] removal of unused and low stakes code related to Cry-threading (#2896)

Browse Source 
Removal highlights include:
 - File indexer (used CryThread<>) linked to long gone asset browser
 - Producer/consumer queues from CryMT
 - set/vector/CLocklessPointerQueue containers also from CryMT
 - Cry interlocked linked list and _InterlockedCompareExchange128
 - CryThread type
 - SAtomicVar types
 - CryAutoSet type
 - Various unused lock types
 -- AutoLockModify
 -- AutoLockRead
 -- CryOptionalAutoLock
 -- CryReadModifyLock
 -- CryRWLock
 -- ReadLock
 -- ReadLockCond
 -- WriteAfterReadLock
 - Misc. unused functions
 -- CryInterLockedAdd (not to be confused with CryInterlockedAdd, using a lower case "locked")
 -- CryInterlockedExchange64 (which was only defined for unix platforms)
 -- SpinLock
 -- JobSpinLock
 -- AtomicAdd
 -- JobAtomicAdd

Signed-off-by: AMZN-ScottR <24445312+AMZN-ScottR@users.noreply.github.com>
This commit is contained in:
Scott Romero
2021-08-06 13:23:14 -07:00
committed by GitHub
parent 4f9382e8c6
commit 9a8a411a0b
16 changed files with 1 additions and 2460 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Code/Legacy/CryCommon/AndroidSpecific.h
-11
View File
@@ -30,17 +30,6 @@
#define MOBILE
#endif
#if (defined(__clang__) && NDK_REV_MAJOR >= 14) || (defined(_CPU_ARM) && defined(PLATFORM_64BIT))
// The version of clang that NDK r14+ ships with is seemingly generating different (for better or worse) code for the atomic operations
// used in the LocklessLinkedList. In either case, this is causing deadlocks in the job system and crashes from memory stomps in
// the bucket allocator. By defining INTERLOCKED_COMPARE_EXCHANGE_128_NOT_SUPPORTED it will disable the Cry job system as well as
// change the implementation of the LocklessLinkedList to use a mutex in it's operations instead, essentially use the same behaviour
// as iOS. While not ideal to use this as a band-aid on the problem, it does fix it with a negligible performance impact.
//
// Additionally, arm64 processors do not provide a cmpxchg16b (or equivalent) instruction required for _InterlockedCompareExchange128
#define INTERLOCKED_COMPARE_EXCHANGE_128_NOT_SUPPORTED
#endif
// Force all allocations to be aligned to TARGET_DEFAULT_ALIGN.
// This is because malloc on Android 32 bit returns memory that is not aligned
// to what some structs/classes need.