steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-14 23:00:44 +02:00
Code Issues Packages Projects Releases 67 Wiki Activity
445 Commits 11 Branches 71 Tags
ea167cf7d97f8aaf5750909fc133fa21db1f0d4e
Commit Graph
100 Commits
Author SHA1 Message Date
Ryan Houdek 55a2c3ca99 Merge pull request #41 from Sonicadvance1/fix_llvm9 
LLVM 9 fixes
 2020-03-18 11:05:40 -07:00
Ryan Houdek fc452de973 Some more fixes for LLVM picked up from the AArch64 side 
AArch64 is a bit more picky when mixing vector and scalar ops
 2020-03-17 19:28:17 -07:00
Ryan Houdek 2e0851975f Makes sure LLVM on AArch64 can read the cycle counter 2020-03-17 19:28:08 -07:00
Ryan Houdek d6fb1f2d17 Make sure OpDispatcher is using explicit scalar operations 
LLVM specifically gets really picky if you mix vector and scalar ops.
Additionally AArch64 seems to be even pickier than x86 for this
 2020-03-17 19:27:58 -07:00
Ryan Houdek 94cf11068c Makes sure Interpreter zeros the upper 64bits when doing Double->Float 2020-03-17 19:27:50 -07:00
Ryan Houdek 90c4b3dae7 Merge pull request #33 from Sonicadvance1/minor_x86_vsqxtun 
Minor optimization in vsqxtun for x86
 2020-03-17 17:01:18 -07:00
Ryan Houdek 388e723c78 Fixes shift ops in LLVM9 
These don't need to operate at 128bit, just one size larger than
incoming.
This is because of 8bit and 16bit shifts rely on 32bit size masking
rather than 8bit and 32bit, so they have to shift at a larger size due
to ensure correct behaviour(that is not undefined).
LLVM-8 worked with the previous behaviour but broke with LLVM-9.
So changing it to this slightly smarter code gen.
 2020-03-17 13:52:23 -07:00
Ryan Houdek 16382c4202 Fixes Scalar FCMP in LLVM9 
LLVM9 is a bit more strict when mixing scalar and vector ops.
Was causing crashes inside of LLVM when mixing a vector source with a
scalar source
 2020-03-17 13:51:07 -07:00
Ryan Houdek 8ea266e10f Implements the new IR ops in the CPU backends 2020-03-17 13:50:37 -07:00
Ryan Houdek 4692a761b6 Implements two new IR Op codes 
Extracts a vector element to a scalar value
Inserts a scalar element in to a vector
 2020-03-17 13:49:47 -07:00
Stefanos Kornilios Mitsis Poiitidis 2993e4dab6 Merge pull request #37 from Sonicadvance1/unittests/bt_reg 
Unit tests and fixes for BT ops with GPR destination
 2020-03-17 19:43:44 +02:00
Ryan Houdek 7ba9aef304 Fixes AArch64 LLVM intrinsic typo 
arch64 != aarch64
 2020-03-17 09:35:39 -07:00
Ryan Houdek 469935e5b8 Fixes x86-64 JIT using invalid movzx 
Found with BT test ops
 2020-03-17 09:26:52 -07:00
Ryan Houdek 21e9ec5f8a Fixes BT ops masking bit selection on GPR destination 
When the destination is a GPR instead of a memory location then the
source bitselect register is masked by the size of the operation
 2020-03-17 09:25:21 -07:00
Ryan Houdek a16638844f Merge pull request #31 from Sonicadvance1/fix_llvm_jit 
Implements all the ops missing in the LLVM JIT
 2020-03-17 09:22:44 -07:00
Ryan Houdek 0a87cbf9d9 Merge pull request #30 from Sonicadvance1/aarch64_jit_fixes 
Get AArch64 JIT to feature parity of the x86 JIT
 2020-03-17 09:22:34 -07:00
Ryan Houdek 4718493d6c Merge pull request #27 from Sonicadvance1/fixes_jit_x86_#2 
x86-64 JIT unit test fixes
 2020-03-17 09:22:17 -07:00
Ryan Houdek 9a54578fd8 Minor optimization in vsqxtun for x86 
No need to zero the register when we are going to logical shift the
upper 64bits to the lower 64bits anyway
 2020-03-15 22:32:23 -07:00
Ryan Houdek 568af31578 Implements all the ops missing in the LLVM JIT 
Lets the LLVM pass the unit tests as well as the interpreter and other
JITs
 2020-03-15 22:30:11 -07:00
Ryan Houdek 1cf60f43cb [AArch64] Implements vector integer length and narrow ops 2020-03-15 00:54:33 -07:00
Ryan Houdek 99d95d5bca [AArch64] Implements float conversion ops 2020-03-15 00:53:49 -07:00
Ryan Houdek deac8d5a75 [AArch64] Implements vector float compare ops 2020-03-15 00:52:43 -07:00
Ryan Houdek 823c803142 [AArch64] Fixes GPR<->FPR conversion moves 2020-03-15 00:45:50 -07:00
Ryan Houdek 702473d3a5 [AArch64] Implements vector shift by immediate ops 2020-03-15 00:44:35 -07:00
Ryan Houdek 4212d177a3 [AArch64] Fixes VSLI/VSRI for byte shifts larger than the register. Needs to zero the reg 2020-03-15 00:44:15 -07:00
Ryan Houdek 35317d0e8a [AArch64] Fixes and implements vector shift ops 2020-03-15 00:43:06 -07:00
Ryan Houdek a670d05274 [AArch64] Fixes VEXTR op 2020-03-15 00:42:11 -07:00
Ryan Houdek 8f1d5e9a4c [AArch64] Fixes FRSQRT Op 2020-03-15 00:40:38 -07:00
Ryan Houdek 803bbe817e [AArch64] Fixes FSQRT Op 2020-03-15 00:39:40 -07:00
Ryan Houdek 214728e19e [AArch64] Implements vector saturating add/sub ops 2020-03-15 00:38:43 -07:00
Ryan Houdek 5530e9eb0e [AArch64] Implements VUMULL IR op 2020-03-15 00:38:17 -07:00
Ryan Houdek 746f399e00 [AArch64] Implements VNOT op 2020-03-15 00:37:30 -07:00
Ryan Houdek c79ee8314e [AArch64] Implements SplatVector Ops 2020-03-15 00:37:14 -07:00
Ryan Houdek cc68d1f88e [AArch64] Implements VInsGPR Op 2020-03-15 00:36:51 -07:00
Ryan Houdek 180944a194 [AArch64] Fixes integer division and remainder ops 2020-03-15 00:35:09 -07:00
Ryan Houdek 9ae0930515 [AArch64] Implements FindTrailingZeros IR Op 2020-03-15 00:33:46 -07:00
Ryan Houdek 5f295ffc5e [AArch64] Fixes FindMSB for 16bit and 32bit register sizes 2020-03-15 00:33:23 -07:00
Ryan Houdek d75fed7691 [AArch64] Fixes ROR for 8bit and 16bit register sizes 2020-03-15 00:32:51 -07:00
Ryan Houdek 8686c0f269 [AArch64] Fixes Break IR op. No immediate offset on atomics. 2020-03-15 00:32:20 -07:00
Ryan Houdek 8c374142e6 [AArch64] Implements Load/Store context indexed 2020-03-15 00:31:25 -07:00
Ryan Houdek 1f112bccbb [AArch64] Fixes VExtractToGPR for 8bit and 16bit sizes 2020-03-15 00:30:53 -07:00
Ryan Houdek 2a37f5fea6 Fixes FindTrailingZeros IR op for <64bit sizes in x86 JIT 
32bit and 16bit were returning wrong values for values of zero
 2020-03-15 00:26:05 -07:00
Ryan Houdek c4b40b2d8f Implements vector shift by immediate ops 2020-03-13 17:55:28 -07:00
Ryan Houdek 664466b60f Fixes 64bit Vector SToF 
Was just using the 32bit instruction on accident
 2020-03-13 17:54:43 -07:00
Ryan Houdek 26aa13e7a9 Replaces x86 rsqrtss usage with sqrtss+divss 
x86 documentation only claims 1.5 * 2^-12 precision on rsqrtss.
Which was enough to make rsqrt of 1.0 be off by a significant amount
This made unit tests fail, so let's just make it more accurate
 2020-03-13 17:52:26 -07:00
Ryan Houdek de444dc62b Fixes SplatVector IR Op 2020-03-13 17:52:07 -07:00
Ryan Houdek 39bac4efea Fixes 32bit memory load op 
xbyak was silently treating this dword load as a 64bit load because the
destination was a 64bit register
 2020-03-13 17:51:07 -07:00
Ryan Houdek 7fa3a318db Division op fixes 
Some division ops weren't sign extending properly and implements 16bit
ops
 2020-03-13 17:49:38 -07:00
Ryan Houdek 23d516d7f3 Fixes SHL/SHR ops 
Makes sure they are operating at the correct sizes.
Also removes the BMI2 SHRX/SHLX ops
 2020-03-13 17:47:53 -07:00
Ryan Houdek e426a622bb Fixes PRINT op in IR JIT. Needs correct stack alignment 2020-03-13 17:47:25 -07:00