Commit Graph

71 Commits

Author SHA1 Message Date
Jonathan Hamilton 8ae76a6680 Fix arm64 MOVI2R for addresses between 2gb and 4gb offset from PC
The PC offset ADRP() path takes a s32 value, but the input offset was
being tested as abs(ptr) < 0xFFFFFFFF. This caused values between
0x80000000 and 0xFFFFFFFF to incorrectly use this path, despite the
offsets not being representable in an s32.

This caused a crash in the VertexLoader on android 8.1 immediate in wind
waker (and possibly all other apps on android 8.1) as the jit and data
sections happened to be loaded 4gb apart in virtual memory, causing some
pointers to hit this
2017-12-09 13:21:58 -08:00
degasus 304e601ad3 JitArm64: Reimplement aarch64 cycle counters.
CNTVCT_EL0 is force-enabled on all linux plattforms.
Windows is untested, but as this is the best way to get *any* low
overhead performance counters, they likely use it as well.
2017-09-02 13:24:37 +02:00
degasus b00c60618b JitArm64: Fix rlwinmx.
Seems like I was wrong that ANDI2R doesn't require a temporary register here.
There is *one* case when the mask won't fit in the ARM AND instruction:
mask = 0xFFFFFFFF
But let's just use MOV instead of AND here for this case...
2017-08-22 08:47:43 +02:00
Markus Wick d78009877b JitArm64: Fix LSL/LSR/ROR/ASR wrappers.
The other method has a latency of 2 cycles. This also improves the
throughput a lot.
2017-08-12 00:00:41 +02:00
Tillmann Karras c54c49714d Arm64Emitter: add FRECPE 2017-05-03 08:02:35 +01:00
Michael Maltese 3d7bace9da Arm64Emitter: extract lambda to AddImmediate()
Fixes warning:

```
Source/Core/Common/Arm64Emitter.cpp:4108:31: error: declaration shadows a local variable [-Werror,-Wshadow]
    auto addi = [this](ARM64Reg Rd, ARM64Reg Rn, u64 imm, bool shift, bool negative, bool flags) {
                                ^
  /var/lib/buildbot/slave/pr-android/build/Source/Core/Common/Arm64Emitter.cpp:4105:46: note: previous declaration is here
  void ARM64XEmitter::ADDI2R_internal(ARM64Reg Rd, ARM64Reg Rn, u64 imm, bool negative, bool flags,
                                               ^
```
2017-03-25 14:21:19 -07:00
Michael Maltese c58ba93503 Arm64: Use PRIi64/PRIx64 for printf 2017-03-25 14:20:44 -07:00
degasus 6aa54a029e JitArm64: Optimize GPR register push/pop. 2017-02-11 00:59:12 +01:00
BhaaL 23d99f2f2c specify custom brace style to fix unions
BreakBeforeBraces: Allman apparently includes all styles,
except for AfterUnion (which is false) when using clang-format -dump-config
2017-01-05 12:55:13 +01:00
Léo Lam 31ccfffd38 Common: Add alignment header
Gets rid of duplicated alignment code.
2016-12-06 20:33:53 +01:00
degasus 8ad98d0046 ArmEmitter: Merge AddI2R helpers. 2016-10-27 19:19:06 +02:00
degasus 694e9b4132 JitArm64: ADDI2R optimizations 2. 2016-10-27 19:19:06 +02:00
degasus 1df694626d JitArm64: Optimize addi2r & subi2r. 2016-10-26 21:54:13 +02:00
degasus df250b84cc JitArm64: Avoid MOVI2R is possible.
Just use all kind of ADDI2R, SUBI2R, ...
They have some optimizations internally.
2016-10-26 21:54:09 +02:00
degasus 7c9bba2213 Arm64Emitter: Fix std::array initializer. 2016-09-26 22:17:25 +02:00
Bernhard Urban 976da3707a arm64: add comment about data cache flushing 2016-09-10 08:05:16 +02:00
Bernhard Urban fff8221b63 arm64: fixes around icache flushing 2016-09-10 02:31:07 +02:00
Pierre Bourdon 3570c7f03a Reformat all the things. Have fun with merge conflicts. 2016-06-24 10:43:46 +02:00
degasus 9ed465f4ac JitArm64: Implement mulhwx 2016-03-04 22:51:46 +01:00
mathieui f15ffda5a7 Correct ampersands as well 2016-01-21 21:27:56 +01:00
mathieui 3e283ea9f1 More asterisks 2016-01-21 21:16:51 +01:00
mathieui 78aa398e7c Common: asterisks go against the type name
not the variable name
2016-01-21 20:46:25 +01:00
Lioncash 2630752ffe Arm64Emitter: Get rid of a pointer cast 2015-10-22 15:32:11 -04:00
Lioncash 018c85c248 Arm64Emitter: Mark trivial functions as constexpr 2015-10-22 15:22:38 -04:00
Lioncash 19ac565e0d Common: Move asserts to their own header 2015-09-26 18:51:27 -04:00
Ryan Houdek 2ad26ab3e9 [AArch64] Fix Test&Branch to relative location instructions.
Wasn't masking by the size of the offset encoding so negative values were killing the instruction
Missed commiting this in my integer gatherpipe PR.
Fixes crashing on AArch64.
2015-09-07 13:38:58 -05:00
Ryan Houdek d003934b8a Merge pull request #2929 from Sonicadvance1/aarch64_optimize_gpr_flush
Aarch64 optimize gpr flush
2015-08-31 10:55:45 -05:00
Ryan Houdek f2c17436ab [AArch64] Fix issue in emitter.
Loadstore pairs support only signed offsets, not unsigned.
2015-08-30 23:05:59 -05:00
Ryan Houdek b907576510 [AArch64] Support profiling by cycle counters if they are available to EL0 2015-08-30 10:25:16 -05:00
Ryan Houdek 4fa23abbe1 [AArch64] Implement MOVI and ORR(imm) in the NEON emitter. 2015-08-23 15:34:53 -05:00
degasus 9bfff0d461 JitArm64: Fix jit clearing
We have to reset m_lastCacheFlushEnd on clearing.
2015-08-15 11:41:01 +02:00
Lioncash 144ea9f4aa Arm64Emitter: Fix encoding of '2-reg misc' variant of FCMEQ 2015-08-10 19:48:36 -04:00
Ryan Houdek 922d476dab [AArch64] Fix FCMGE instruction encoding.
Fixes a crash when ps_sel is used (PSO 1&2 intro movies).
2015-08-09 14:54:55 -05:00
degasus b8dd68beef JitArm64: Far Code Cache 2015-07-12 09:41:32 +02:00
Lioncash d09d59007a Arm64Emitter: Add a missing const specifier for an array table 2015-07-02 11:09:44 -04:00
Ryan Houdek afc3d30f5c [AArch64] Implement BFI & UBFIZ in the emitter.
Also fixes a bug in the UBFX instruction emitter. Naughty Naughty PPSSPP, not testing emitter functions you add.
2015-06-29 19:00:22 -05:00
Ryan Houdek 5dc148159f [AArch64] Implement {U, S}QXTN{,2}
Also split out XTN to XTN and XTN2.
2015-06-13 23:16:17 -05:00
Lioncash 74b359e390 Arm64Emitter: Remove unused variable from EncodeLoadStoreRegisterOffset 2015-06-13 14:27:15 -04:00
Ryan Houdek 3d2b116323 [AArch64] Implement a couple instructions in the emitter.
Implements LD2R.
Implements LD1R/LD2R with post-indexing support.
Implements vector min/max instructions.
2015-06-09 18:10:56 -05:00
Ryan Houdek 8ae12d8005 [AArch64] Add ASIMD LDR/STR with register offset 2015-06-07 19:53:05 -05:00
Ryan Houdek 05b72c5d31 [AArch64] Upstream PPSSPP's emitter changes.
Requires a minor change to in the JIT to make sure everything still works.
2015-06-07 19:50:21 -05:00
Tillmann Karras 30ebb2459e Set copyright year to when a file was created 2015-05-25 13:22:31 +02:00
Tillmann Karras cefcb0ace9 Update license headers to GPLv2+ 2015-05-25 13:22:31 +02:00
Ryan Houdek f6511c3ba5 [AArch64] Add an assert to SMOV in the emitter.
SMOV doesn't have an encoding for moving a 32bit element to a 32bit GPR.
One should use UMOV if they want that.
2015-03-08 12:29:45 -05:00
Ryan Houdek fbdee7b15f [AArch64] Handle FPR island registers in a less dumb way. 2015-03-03 00:30:05 -06:00
Ryan Houdek f1a9db9bdc [AArch64] Stop violating the AAPCS64 so much. 2015-03-02 11:21:15 -06:00
Ryan Houdek fad46729b0 [AArch64] Implemented paired pushing/popping for the VFP.
A bit more efficient if we are only pushing two VFP registers.
We can probably be a bit more efficient in the future by mixing paired loadstores in to the other paths as well.
2015-03-02 06:27:47 -06:00
Ryan Houdek 39e357d62d [AArch64] Implement VFP loadstore paired in the emitter. 2015-03-02 06:27:17 -06:00
Ryan Houdek 8b8310d28c [AArch64] Optimize FPR pushing and popping.
Previously on FPR pushing and popping we would do a single STR/LDR per quad FPR we wanted to push/pop.
In most of our cases when we are pushing and popping VFP registers they will be consecutive registers that will save more efficiently using the NEON
loadstores that can do up to four quad registers.
So this can potentially cutting instructions down to ~1/4th the amount of instructions if the registers are all consecutive.

On the Cortex-A57 this is basically just an icache improvement, but on the Nvidia Denver this may be optimized to be more efficient. Either way it's a
win.
2015-03-02 06:27:13 -06:00
Ryan Houdek 120df4c688 [AArch64] Implement loadstore unscaled. 2015-02-16 22:00:43 -06:00