e3d95f2258
wince: unify _vmem and vmem32. Use 4GB virtual space on 64-bit arch
...
On 64-bit architectures, _vmem first tries to allocate 4GB then falls
back to 512 MB.
The same virtual space is now used by _vmem and vmem32 (mmu)
2019-05-23 11:40:33 +02:00
c2c0215e1b
Merge remote-tracking branch 'origin/master' into fh/wince-dynarec
2019-05-22 14:13:00 +02:00
fb76efb08d
Standard functions for virtual mem alloc/protect/map/...
2019-05-22 11:41:12 +02:00
3463872fdf
Fix small clang build issue.
2019-05-13 01:35:40 +02:00
65e2218792
Fix ARM64 rewrites with NO_RWX_PAGES
...
Tested on nvidia jetson and Android, so far works great.
2019-05-13 00:23:18 +02:00
aa4fc8dd60
Move cache invalidation to vmem_platform and implement NO_RWX on arm64
...
Still not functional yet, this commit breaks arm64 on android for instance.
Rewrites are not yet figured out, will push them in the next commit.
2019-05-12 23:25:43 +02:00
5ba56627a9
Optimize rec-ARM64 by emitting less instructions on loads.
...
Fast load uses 3 or 4 insts, whereas slow load takes 2+.
On A57 measured ~3% perf, which is expected for an OOO CPU, hopefully
perf will be much better on A53 and other in-order CPUs.
2019-05-09 00:10:01 +02:00
810b8a59da
wince: 32-bit virtual mem space
...
use fast mem read/write for x64 and arm64 dynarecs
2019-04-29 18:23:00 +02:00
011849ea74
wince: arm64 dynarec fixes
2019-04-19 17:20:55 +02:00
553f6054ce
NO_MMU build fixes
2019-04-19 11:56:11 +02:00
dece3fc13e
wince: use setjmp/longjmp instead of try/catch for dynarecs
...
WinCE fast mmu implementation
WIP arm64 dynarec
2019-04-15 18:02:34 +02:00
af6993a819
dynarec: Refactor smc-option a bit
2019-03-30 07:16:44 +01:00
6438a402df
dynarec: add option to control smc code checks: none, fast, full
2019-03-29 19:23:37 +01:00
5242af5879
rec-arm64: better pref fix
2019-03-28 14:26:04 +01:00
20b09cf1b7
arm64: disable partially implemented 0x1337 pref optimization hack
2019-03-28 13:02:08 +01:00
fca6f3cecb
dsp arm64: fix crash when VIXL_DEBUG is off
2019-01-24 21:47:55 +01:00
91cfd4b2f7
Reserve and allocate maximum RAM/VRAM/ARAM in all cases
...
Reserve enough virtual memory space for DC and Naomi
Allocate dynarec entry point tables for max possible ram
Free mem and release vmem on exit
2019-01-24 09:48:58 +01:00
91b76f0c16
arm64 dynarec: missing edits
2019-01-22 18:40:36 +01:00
75a04d6d28
arm64 dynarec: implement tail calling and block linking
2019-01-22 18:37:04 +01:00
a3682e7b22
arm64 dynarec: revert to non-explode spans and minor optimizations
...
Added some profiling
2019-01-21 12:54:29 +01:00
573f285f3b
arm64 dynarec: save x29 and x30 in ngen_mainloop prologue
2019-01-18 00:06:33 +01:00
cb8e81d473
arm64: direct memory access and jit rewrite
...
generates direct vmem read & write accesses
trap sigsegv and rewrite using slow path
add w29 to allocatable registers
get rid of literals and use pc-rel branching
minor optimizations and cleanup
2019-01-16 13:04:16 +01:00
c2a048e8d8
arm64: use explode_spans to allocate regs for V2 and F64 params
2019-01-15 08:47:07 +01:00
523b110412
arm64: dynarec fixes
2019-01-13 11:32:28 +01:00
fc05727538
dynarecs clean-up
...
move GetRegPtr and ngen_FailedToFindBlock to sh4/dyna
2019-01-11 23:52:20 +01:00
e5ee48efa9
arm64: implement swaplb and pref. Various optimizations
2019-01-11 15:58:48 +01:00
1c80207879
arm64: implement ftrv and frswap
2019-01-10 21:22:53 +01:00
18a16f83ac
arm64: more native opcodes implemented
2019-01-10 18:58:29 +01:00
a9a2aad8f6
arm64: use register spans allocation. Implement some opcodes natively
2019-01-09 16:35:23 +01:00
3d8b01c515
arm64: Use x28 as sh4 regs base pointer. Reg cache to avoid fetches.
...
Block check optimization
2019-01-08 13:29:08 +01:00
67a4eb8f1f
arm64 dynarec using vixl
2019-01-07 21:50:46 +01:00
Flyinghead