Commit graph

316 commits

Author SHA1 Message Date
Nekotekina
487b572aa6 Linux: remove assertions on MADV_HUGEPAGE
It's a hint and is allowed to fail.
2020-11-25 10:41:17 +03:00
Nekotekina
600c0ea5ba atomic.cpp: fix windows/std path
Always check condition after rearming internal semaphore.
2020-11-25 10:41:17 +03:00
Nekotekina
00e64920c8 Implement "slow mutex" without spin-waiting.
Benefits: only 1 byte.
Testing atomic wait.
2020-11-25 10:41:17 +03:00
Nekotekina
d8e9382573 asm.hpp: constexpr refactoring
Removed asm (what an irony), made many functions constexpr.
Removed some unused function, readding them should be trivial.
2020-11-24 13:20:40 +03:00
Nekotekina
ab248e170c vm_native.cpp: add MADV_HUGEPAGE
It enables Transparend Huge Pages for some regions on Linux.
Although it can't be actively useful, it seems to do something.
Maybe it's even harmful for recompilers.
2020-11-24 12:31:11 +03:00
Nekotekina
27097cbcfe atomic.hpp: alter atomic_storage<>::store a bit
Always fallback to exchange. May improve codegen on gcc.
2020-11-24 12:31:11 +03:00
Nekotekina
43952e18e2 Implement prefetch_write() and prefetch_exec() wrappers
Do some refactoring to prefetch_read() in util/asm.hpp as well.
Make all these function constexpr because they are no-ops.
2020-11-24 12:31:11 +03:00
Nekotekina
14fa96fdef atomic.cpp: don't use empty notify callback
Instead, allow setting it to null and check for null.
Similar could be done for wait cb, but it's set for all named threads.
2020-11-24 01:09:04 +03:00
Nekotekina
16cc88dd0e atomic.hpp: fix atomic_wait::list regression
Typo-regressions. Also improve static asserts.
2020-11-23 09:26:25 +03:00
Nekotekina
280f5f0b54 atomic.hpp: redefine op::ne to actually mean "not equal"
WTF is "native endianness", it's native by default.
2020-11-23 09:24:16 +03:00
Nekotekina
d70ab11477 atomic.cpp: remove redundant constexpr
Static initialization is all-zeros anyway.
But constexpr was killing my intellisense.
And probably also affected compile time.

Also make some internal structures hidden ("static").
2020-11-22 12:07:49 +03:00
Nekotekina
e89a568765 Attempt to fix Emu.Quit()
There was some kind of "initialization order fiasco" of g_fxo.
Not only should it be used after main, but also destroyed before exit.
2020-11-21 18:27:16 +03:00
Nekotekina
60cff6f3d4 Attempt to fix #9271 2020-11-21 07:09:48 +03:00
Nekotekina
5f010c2f4c atomic.cpp: optimize "cond" allocation with TLS
Billions of events can reduce to thousands, saving CPU time a few s.
Only 4 slots are available (arbitrarily), and only 1 is usually used.
Other slots are used only for waiting on multiple atomics.
2020-11-20 19:13:22 +03:00
Nekotekina
6c18e00977 atomic.cpp: use deterministic "cond" allocation
Takes ~80 ns instead of ~40 ns, same about deallocation.
Loops don't exist here, only 4-level semaphore tree.
Worst case only happens with concurrence, not from looping.
This optimization is not really necessary at current state of RPCS3.
This is more like to test C++ compilers and MSVC u128 implementation.
2020-11-20 15:04:17 +03:00
Nekotekina
72ac1aa537 atomic.cpp: use some actual hash for hashtable (std::minstd_rand)
Should only take about 5ns for initialization.
Should also make collisions almost impossible.

I also suspect I fixed some bad bug in process...
2020-11-20 15:04:17 +03:00
Nekotekina
3dd48a24ce atomic.cpp: remove tiny redundancy 2020-11-17 15:22:04 +03:00
Nekotekina
984e97cbdc Potential fix for Linux compilation 2020-11-17 15:22:04 +03:00
Nekotekina
333fe9a561 atomic.cpp: fix notification on windows 2020-11-15 04:40:44 +03:00
Nekotekina
badb3dc2dd atomic.cpp/threads: remove old wait callback
Add new wait callback which simply collects statistics.
Shift workarounds towards actual problem detection.
2020-11-14 18:16:27 +03:00
Nekotekina
3ac819ee70 atomic.cpp: fix waiting on multiple variables
Silly bugs are so silly...
2020-11-14 17:25:02 +03:00
Nekotekina
00f877e812 atomic.cpp: fix rare overload scenario
If slot allocation fails, increase reference counter in every slot.
This prevents search routine from "losing" some slots.
2020-11-14 13:56:23 +03:00
Nekotekina
ab365fe494 Fixed thread pool a bit
Use 128-bit allocator instead of queue.
When pool is full (128), threads just terminate as before.
2020-11-13 13:32:44 +03:00
Nekotekina
e48f160a29 atomic.cpp: continuation of fixing all-ones masks
Also added missing const noexcept.
2020-11-13 12:40:02 +03:00
Nekotekina
cfda4d0ade atomic.cpp: optimize raw_notify() for unspecified pointer mode
Remove unnecessary optimization from cond_alloc().
Optimistic case was absolutely dominating anyway.
Although the whole function is a dirty hack.
Now scanning through all threads is faster.
2020-11-13 07:05:33 +03:00
Nekotekina
3888b0429c atomic.cpp: print some stats on stop 2020-11-12 05:52:22 +03:00
Nekotekina
70761a4ef0 atomic.cpp: fix notification bug from few commits before
Make sure forced notification ignores mask comparison.
Also fix masks provided by smaller types to correct ones.
2020-11-12 05:01:44 +03:00
Nekotekina
e37df998e9 atomic.cpp: minor optimization
Silly MSVC emits unnecessary initialize-on-first-use code guards.
Didn't check other compiler. But workaround just works anyway.
2020-11-12 03:41:19 +03:00
Nekotekina
c30cdfb3a7 atomic.cpp: merge more atomic variables for simplicity
Compress 16-bit ref counter and two 48+64 bit slot allocators.
This allowed to remove some weird unnecessary logic paths.
Adjust hashtable size to keep it the same.
2020-11-12 03:39:05 +03:00
Nekotekina
350b704cd7 atomic.cpp: combine two atomics (ref ctr and wait ptr)
Should fit into cache line perfectly.
Also makes ptr check more robust.
2020-11-12 01:06:12 +03:00
Nekotekina
ad013d59f4 atomic.cpp: don't load actual memory on empty mask
It's not an optimization. It just allows some tricks safely.
2020-11-11 23:59:09 +03:00
Nekotekina
d391133524 atomic.cpp: minor notification improvement
Check mask before locking the reference.
Can help to not bother non-eligible waiters.
2020-11-11 23:51:41 +03:00
Nekotekina
4b823469f7 atomic.cpp: restore notify_all() behaviour
Restore batch notification with non-blocking attempts.
2020-11-11 22:43:13 +03:00
Nekotekina
7cd1e767be atomic.cpp: implement some atomic wait operations.
Instead of plain waiting while equal to some value,
it can be something like less, or greater, or even bitcount.
But it's a draft and untested. Hopefully doesn't break anything.
2020-11-11 20:46:37 +03:00
Nekotekina
c047ab6147 atomic.cpp: waitable atomics got brutally rewritten
Hashtable increased and flatten, tree-alike extensions removed.
Some things simplified, so it can actually decrease perf a bit.
But most platforms shouldn't be affected.
Removed limit of 56 waiters per pointer.
Real limit now is about 65535.
2020-11-11 05:35:27 +03:00
Nekotekina
d1ee7c651f Make memory locking optional (mlock, VirtualLock).
Fix desired locking operation (to fix "sudo" memory).
It was discovered that some systems have outdated configuration.
With too tight limit, it's almost impossible to lock anything in memory.
2020-11-10 06:05:37 +03:00
Nekotekina
dcbe8ef5f4 Import VirtualAlloc2 if available (Win10+)
Import MapViewOfFile3 as well.
Unused, this crap is tricky.
2020-11-10 03:04:38 +03:00
Nekotekina
bd5253047b Improve DYNAMIC_IMPORT
Don't call get_proc_address every time if if failed.
Also rename Utilities/dynamic_library.h to util/dyn_lib.hpp
2020-11-10 01:44:31 +03:00
Nekotekina
e98eb4659e Implement utils::memory_lock 2020-11-09 22:57:36 +03:00
Nekotekina
d1988d384c vm: remove unnecessary memset() on stop 2020-11-09 22:57:36 +03:00
Nekotekina
b05d12df0e Hotfix for shm->map_self()
Make sure mmap returns 64K-aligned results, as on Windows.
2020-11-08 22:15:59 +03:00
Nekotekina
1c99a2e7fb vm: add map_self() method to utils::shm
Add complementary unmap_self() method.
Move VirtualMemory to util/vm.hpp
Minor associated include cleanup.
Move asm.h to util/asm.hpp
2020-11-08 16:43:15 +03:00
Nekotekina
fd7ea82965 atomic.cpp: fix 3-arg notify functions
Third arg forces comparison with itself, and possibly forced wakeup.
2020-11-07 19:18:14 +03:00
Nekotekina
0bd989b118 atomic.cpp: allow waiting on multiple variables simultaneously
Arbitrary maximum set to 8, but really we need 2, maybe 3.
Added atomic_wait::list object for multi-waiting.
Added atomic_wait::get_unique_tsc just in case.
2020-11-07 18:18:02 +03:00
Nekotekina
bd0b292b39 atomic.cpp: adjust assumed pointer mask
Increase waiter mask (allow 65535 waiters).
Assume 47-bit pointers instead of 48-bit.
2020-11-07 15:08:46 +03:00
Nekotekina
31f0a0c064 atomic.cpp: add minor safety check 2020-11-07 15:08:46 +03:00
Nekotekina
3e445eff8e atomic.cpp: maintenance of wait() function
Move hashtable allocation into another function.
2020-11-07 15:08:46 +03:00
Nekotekina
0a5742587a atomic.cpp: upgrade raw_notify()
Now it searches all semaphores if data arg is nullptr.
Also it tries to wake up all threads if thread_id is 0.
2020-11-06 17:15:40 +03:00
Nekotekina
ad4df2d946 atomic.cpp: refactor native signaling functions
Add native_alert() and try_native_alert()
2020-11-06 17:15:40 +03:00
Nekotekina
bfe9580551 atomic.cpp: fix cond_handle data structures
Fix a critical bug with possible id out of range.
2020-11-06 16:56:59 +03:00