steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-23 09:00:39 +02:00
Code Issues Packages Projects Releases 67 Wiki Activity
8,965 Commits 11 Branches 71 Tags
002ca360f80bb59bee904ebc2b16404d61398550
Commit Graph
880 Commits
Author SHA1 Message Date
Alyssa Rosenzweig 82ba16c6ed OpcodeDispatcher: optimize LOOP/N/E 
Don't clobber NZCV.

Before/after assembly from the Primary_E1 unit test:

< 4340: [INFO] cset w20, ne
< 4340: [INFO] mrs x21, nzcv
< 4340: [INFO] cmp x5, #0x0 (0)
< 4340: [INFO] cset x22, ne
< 4340: [INFO] and x20, x22, x20
< 4340: [INFO] msr nzcv, x21
< 4340: [INFO] cbnz x20, #+0x8 (addr 0xffff896f8084)
< 4340: [INFO] b #+0x1c (addr 0xffff896f809c)
< 4340: [INFO] ldr x0, pc+8 (addr 0xffff896f808c)
---
> 4340: [INFO] csel x20, x5, xzr, ne
> 4340: [INFO] cbnz x20, #+0x8 (addr 0xfffed7308070)
> 4340: [INFO] b #+0x1c (addr 0xfffed7308088)
> 4340: [INFO] ldr x0, pc+8 (addr 0xfffed7308078)

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-21 12:08:40 -04:00
Billy Laws d490cb1b79 FEXCore: Fallback to the memcpy slow path for overlaps within 32 bytes 
Take e.g a forward rep movsb copy from addr 0 to 1, the expected
behaviour since this is a bytewise copy is:
before: aaabbbb...
after: aaaaaaa...
but by copying in 32-byte chunks we end up with:
after: aaaabbbb...
due to the self overwrites not occuring within a single 32 bit copy.
 2024-03-20 20:54:19 +00:00
Billy Laws 94fecb9dad FEXCore: Remove needless alignment checks for the mem{cpy,set} fastpath 2024-03-20 20:54:09 +00:00
Ryan Houdek 7dcacfe990 Merge pull request #3478 from bylaws/memcpy 
FEXCore: Add non-atomic Memcpy and Memset IR fast paths
 2024-03-18 18:56:44 -07:00
Billy Laws 8d4d8fe3e5 FEXCore: Add non-atomic Memcpy and Memset IR fast paths 
When TSO is disabled, vector LDP/STP can be used for a two
instruction 32 byte memory copy which is significantly faster than the
current byte-by-byte copy. Performing two such copies directly after
oneanother also marginally increases copy speed for all sizes >=64.
 2024-03-18 23:28:50 +00:00
Ryan Houdek ab8ee64352 Merge pull request #3497 from Sonicadvance1/movmaskb_constant 
JIT: Optimize pmovmaskb with a named vector constant
 2024-03-18 16:08:40 -07:00
Alyssa Rosenzweig 2a9fcc6a66 Merge pull request #3492 from Sonicadvance1/implement_prefetch 
OpcodeDispatcher: Implement support for the various prefetch instructions
 2024-03-18 07:49:47 -04:00
Ryan Houdek fd391b1b18 JIT: Optimize pmovmaskb with a named vector constant 
I was looking at some other JIT overheads and this cropped up as some
overhead. Instead of materializing a constant using mov+movk+movk+movk,
load it from the named vector constant array.

In a micro-benchmark this improved performance by 34%.
In bytemark this improved on subbench by 0.82%
 2024-03-17 18:40:46 -07:00
Ryan Houdek f79991a9d8 OpcodeDispatcher: Implement rdpid 
Missed this instruction when implementing rdtscp. Returns the same ID
result in a register just like rdtscp, but without the cycle counter
results. Doesn't touch any flags just like rdtscp.
 2024-03-14 20:07:58 -07:00
Ryan Houdek ca6b2e43e6 Merge pull request #3491 from alyssarosenzweig/rclse/waw 
RCLSE: Optimize store-after-store
 2024-03-14 03:23:05 -07:00
Ryan Houdek 8056bee82b OpcodeDispatcher: Implement support for the various prefetch instructions 
x86 has a few prefetch instructions.
- prefetch - One of two classic 3DNow! instructions
   - Prefetch in to L1 data cache
- prefetchw - One of two classic 3DNow! instructions
   - Implies prefetch in to L1 data cache
   - Prefetch cacheline with intent to write and exclusive ownership

- prefetchnta
   - Prefetch non-temporal data in respect to /all/ cache levels
   - Assumes inclusive caches?
- prefetch{t0,t1,t2}
   - Prefetch data with respect to each cache level
   - T0 = L1 and higher
   - T1 = L2 and higher
   - T2 = L3 and higher

**Some silly duplicates**
- prefetchwt1
   - Duplicate of prefetchw but explicitly L1 data cache
- prefetch_exclusive
   - Duplicate of prefetch

God Of War 2018 uses prefetchw as a hint for exclusive ownership of the
cacheline in some very aggressive spin-loops. Let's implement the
operations to help it along.
 2024-03-12 21:37:31 -07:00
Ryan Houdek cc635a54f8 IR: Implements support for prefetch operation 2024-03-12 21:19:50 -07:00
Ryan Houdek 217d9d8c50 ARMEmitter: Fixes prfm with negative or unaligned offsets 2024-03-12 21:18:23 -07:00
Alyssa Rosenzweig 7629007cfa OpcodeDispatcher: allow upper garbage on STOS 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:50:31 -04:00
Alyssa Rosenzweig 03c6abdad4 OpcodeDispatcher: optimize DF add 
fuse the shift the right way

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:50:31 -04:00
Alyssa Rosenzweig c99cbe6d0a JIT: switch DF representation 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:50:31 -04:00
Alyssa Rosenzweig e3ee65e491 OpcodeDispatcher: use transformed DF for memset/memcpy 
Use the 1/-1 representation instead of 0/1. This will be better by the end of
the series.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:50:31 -04:00
Alyssa Rosenzweig aee00f524c OpcodeDispatcher: use DF retrieval helpers 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:50:31 -04:00
Alyssa Rosenzweig a76321c6c1 OpcodeDispatcher: add DF retrieval helpers 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:50:31 -04:00
Alyssa Rosenzweig 85f8ad3842 JIT: fix sha256msg1 encoding 
botched move in the !tied reg case.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-11 18:41:23 -04:00
Alyssa Rosenzweig 11880459a5 OpcodeDispatcher: use SETF for DEC 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-01 19:40:53 -04:00
Alyssa Rosenzweig 0ef0bb2c97 OpcodeDispatcher: use SETF for INC 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-01 19:40:53 -04:00
Alyssa Rosenzweig 72edee7c6f IR: add SETF8/SETF16 ir ops 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-03-01 19:40:53 -04:00
Ryan Houdek 009ae55ff0 Merge pull request #3475 from alyssarosenzweig/opt/lock-dec 
Optimize lock dec
 2024-02-29 08:44:24 -08:00
Ryan Houdek 98572b9e23 Merge pull request #3473 from Sonicadvance1/remove_mov_swap 
Arm64: Stop moving source in atomic swap
 2024-02-29 08:44:16 -08:00
Alyssa Rosenzweig fed5e6d546 OpcodeDispatcher: use fetchadd for atomic DEC 
Avoids a NEG.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-29 09:28:21 -04:00
Ryan Houdek c318947695 Arm64: Stop moving source in atomic swap 
ldswpal doesn't overwrite the source register and only reads the bits
required for the sized operation.
Not sure exactly why we were doing a copy here.

Removing it means improving Skyrim's hottest code block, as seen in #3472
 2024-02-29 03:07:05 -08:00
Alyssa Rosenzweig 811487ad98 OpcodeDispatcher: use real branch for INT 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 10:35:12 -04:00
Alyssa Rosenzweig 4f4e38ace2 OpcodeDispatcher: use real branch for rep cmps 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 10:35:12 -04:00
Alyssa Rosenzweig edd6becc56 OpcodeDispatcher: use real branch for rep scas 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 10:35:12 -04:00
Alyssa Rosenzweig e47a94cae7 OpcodeDispatcher: skip mask with shld 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 10:18:29 -04:00
Alyssa Rosenzweig cc82dba1ca OpcodeDispatcher: use mvn for AF with constants 
This reduces pointless constant usage. For now, it's no net change to
instcountci, but it should make it easier to get wins later. Hopefully.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 10:18:28 -04:00
Alyssa Rosenzweig 8232669b22 OpcodeDispatcher: simplify 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 10:00:13 -04:00
Alyssa Rosenzweig 28936073c4 OpcodeDispatcher: allow upper garbage on NEG 
like SUB.

due to RA silliness, this is a loss for inst count but a win for cycles.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 09:59:57 -04:00
Alyssa Rosenzweig ef2559d911 OpcodeDispatcher: allow garbage with SCAS 
it's just feeding SUB flags which allow it

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-28 09:16:08 -04:00
Ryan Houdek f346f89678 OpcodeDispatcher: Don't use AddShift with no shift 
This accidentally removed optimizations elsewhere that was only checking
for Add.
 2024-02-27 19:56:12 -08:00
Alyssa Rosenzweig 49e798ab2b Merge pull request #3461 from alyssarosenzweig/opt/sbc 
Optimize SBC
 2024-02-27 11:29:45 -04:00
Ryan Houdek 946c805d84 Merge pull request #3459 from Sonicadvance1/fix_591 
Capture a 64-bit process trying to jump to 32-bit syscall handler
 2024-02-26 21:57:03 -08:00
Alyssa Rosenzweig 12cc980603 OpcodeDispatcher: shuffle adc flag order 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 16:48:07 -04:00
Alyssa Rosenzweig f3d55dd721 OpcodeDispatcher: shuffle SBC flag order 
avoids clobbering nzcv

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 16:48:07 -04:00
Alyssa Rosenzweig 91cef6b76f OpcodeDispatcher: use native ADC even for 8/16-bit 
we mask off the upper bits, and they agree in the lower bits.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 16:48:07 -04:00
Alyssa Rosenzweig 270cbf39b5 OpcodeDispatcher: specialize SALC 
this gets rid of the awkward non-flag SBB case, which streamlines SBB. while
getting better codegen for the demon opcode (-:

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 16:33:41 -04:00
Alyssa Rosenzweig 2e0be0a5e7 OpcodeDispatcher: allow more upper garbage with adc 
missed this last series.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 15:42:13 -04:00
Alyssa Rosenzweig d60c089697 OpcodeDispatcher: allow upper garbage with sbb 
for the usual reasons

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 15:35:14 -04:00
Alyssa Rosenzweig e76ebeab58 OpcodeDispatcher: use 1-op "src + CF" 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 15:35:14 -04:00
Alyssa Rosenzweig 333271d490 OpcodeDispatcher: fuse sbb when flags calculated 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 15:35:14 -04:00
Alyssa Rosenzweig a750870abf OpcodeDispatcher: use fused sbcs calculations 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 15:25:03 -04:00
Alyssa Rosenzweig 15db72ef60 IR: add Sbb, SbbWithFlags ops 
For fusing sbc+sbcs

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-26 15:25:03 -04:00
Ryan Houdek 4f028b8614 Capture a 64-bit process trying to jump to 32-bit syscall handler 
Fixes #591

Adds a simple unittest
 2024-02-26 05:37:29 -08:00
Alyssa Rosenzweig 1e153e0c81 OpcodeDispatcher: allow garbage with adcs 
for the usual reasons

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-25 10:50:02 -04:00