Commit Graph

87 Commits

Author SHA1 Message Date
Sacha
637d75f47d Unlock the other 16 regs that are available for NEON. 2013-06-09 23:15:59 +10:00
Sacha
29d1e40c79 Buildfix for Blackberry, Linux, Symbian, non-ffmpeg platforms, Meego, ... 2013-06-09 22:53:22 +10:00
Sacha
583927988c Armjit Update: Improvements
Implement vsat0 and vsat1
Introduce VMOV for immediates.  Use in MOVI2F where possible by detecting if float val can be encoded.
Combine some vneg ops in ApplyPrefixST.
2013-06-09 21:27:16 +10:00
Sacha
f21218c3f9 Armjit: Combine mul.s + neg.s to VNMUL. Implement VNMUL, VNMLA, VNMLS.
I had implemented mul.s + add/sub.s + add/sub.s -> VADD/VSUB + V(N)ML(A/S). Turns out it doesn't happen enough though (once or twice per game).
2013-06-08 16:39:59 +10:00
Sacha
10c976b2af Armjit: Improve ApplyPrefixD. Add VABD to emitter. 2013-06-06 03:08:58 +10:00
Sacha
6712de1136 Armjit: Implement wsbh and wsbw (rev16 and rev). Fix encoding for rev16. 2013-06-05 11:32:44 +10:00
Sacha
a14a2fafa9 ARMJIT: Fix and use DIV, DIVU, INS and EXT (ARMv7 and VFPv4 implementations). 2013-05-24 08:26:19 +10:00
Sonicadvance1
cd9b49acfd Updates to ArmEmitter from Dolphin. 2013-04-12 10:14:18 -07:00
Unknown W. Brackets
268f30f522 Buildfix. 2013-03-29 01:27:24 -07:00
Sacha
2f63b56c19 If MOVI2R can be done in 2 ops for ARMv6, prefer that over litpool. Litpool can take 1-3 cycles depending on when the value is used. So, usually 3 cycles :\. 2013-03-28 21:26:13 +10:00
Sacha
471ddd6380 Simplify armjit. 2013-03-26 02:41:15 +10:00
Unknown W. Brackets
de3713fc50 armjit: improve mem speed without fastmem. 2013-03-16 14:37:35 -07:00
Unknown W. Brackets
663b481033 armjit: Fix encoding of LDRH/STRH/LDRSH/LDRSB.
Fixes fastmem issues recently introduced.
2013-03-15 00:12:28 -07:00
Sacha
5091c7aae2 Fix Sonic's ArmEmitter changes. FastMem should work now. 2013-03-15 11:26:56 +10:00
Unknown W. Brackets
ead6983ad2 Don't lock armjit memory before aligning code. 2013-03-14 01:16:28 -07:00
Sacha
f88bf8bbff Attempt at fixing JIT on iOS. Can only have PROT_WRITE or PROT_EXEC enabled. So toggle between them as needed. 2013-03-14 14:40:01 +10:00
Sacha
529803e429 Sonic's ArmEmitter changes (cross-project merge from Dolphin) 2013-03-14 12:47:29 +10:00
Sacha
7782be9a3e Try software rounding mode instead. Use conditionals for VFP functions. Note: Round to Zero not working for some reason? 2013-03-08 22:39:34 +10:00
Sacha
fbef590bed Add emitters for VMRS/VMSR. Undisable ceil, which works. Add stub for rounding mode handling. 2013-03-08 22:39:34 +10:00
Sacha
b9ec281c55 Disable rounding methods that aren't working in hardware. Add some notes. Fix compare fall-through and choose right round method. 2013-03-07 21:33:30 +10:00
Sacha
9f0bb8cf52 armv6: Reuse constants in literal pool.
May reduce memory loads?
2013-03-07 14:44:55 +10:00
Sacha
ae3b881a7f Use correct args for Operand2(..) through armjit. Fix STR(..). 2013-03-07 00:59:07 +10:00
Sacha
268d16bd24 Use correct args for STR(..) throughout armjit. 2013-03-07 00:59:07 +10:00
Sacha
5a134243a7 Armjit: Fix lwl, lwr and enable again. Thanks Sonic. 2013-03-06 03:28:28 +10:00
Sacha
7e67de3334 Armjit: Implement lwl, lwr, swl, swr in ARM JIT. lwr is currently disabled as it isn't working. 2013-03-06 02:11:36 +10:00
Sacha
9152d2f2bb Armjit: Optimise swl+swr and lwl+lwr cases that can be combined to a single sw or lw. Add shift flags to STR/LDR. Add EatInstruction to ArmJit. 2013-03-06 02:11:36 +10:00
Sacha
4641cf376f Armjit: Implement CLZ instruction. Disable reg shifts for now (breaks Wipeout Pure). 2013-03-05 14:16:35 +10:00
Sacha
10ad797c6d Armjit stubs.
Add a double encoding for VCVT. Implement integer divide (but not working yet). Stubs for msub/msubu. Don't detect vfpv3 on Symbian.
2013-03-05 13:16:08 +10:00
Sacha
1089a31a45 Armjit: add reverse bit instruction. 2013-03-05 02:58:51 +10:00
Sacha
bce3295950 Fix graphical issues. DISABLE INS instruction for now. Fix OR (it was doing AND). 2013-03-04 22:09:45 +10:00
Sacha
0fc6b60874 Fix Lit Pools for cases where offset goes out of range. For example: Zero no Kiseki has a block that is 7K large. 2013-03-04 14:36:23 +10:00
Sacha
9633239f18 Update for lit pools. Works in some games now thanks to [Unknown]. 2013-03-04 14:29:17 +10:00
Sacha
bdfe24a86b ARMv6 literal pool method. 2013-03-04 14:26:45 +10:00
Henrik Rydgard
7dc75d87b5 armjit: Re-enable ext/ins, safer implementation. arm7 path disabled for now. 2013-03-03 23:17:21 +01:00
Henrik Rydgard
1cddc86e05 armjit: Temporarily disable ext/ins as they appear to have broken Persona 3 somehow. 2013-03-03 22:26:20 +01:00
Henrik Rydgard
1e3a00ee9d armjit: implement vzero, vone. Use vmla for dot product. 2013-03-03 20:56:22 +01:00
Henrik Rydgard
650c02c3a5 Some more armjit work (ext, ins) and VFPU prefix clamps (not enabled) 2013-03-03 17:36:22 +01:00
Henrik Rydgard
bc15617392 Make un-buffered rendering much smarter, removing flicker.
This turns it into a very viable option for many games. You do lose some FX
but it can as a result even be used as a workaround for the massive glow
in Wipeout...
2013-03-03 13:00:21 +01:00
Henrik Rydgard
b781c8deb5 Delete unused ARM emitter function. 2013-03-03 10:55:49 +01:00
Henrik Rydgard
f5581caccc Some work on ARMJIT FPU compares, still not quite working. 2013-03-02 19:09:24 +01:00
Sacha
6c23e1b6d5 Use flags instead of bools for VCVT. Fix up some spacing. Only Android has ArmEmitterTest. 2013-03-02 11:34:03 +10:00
Henrik Rydgard
253396666c Merge branch 'armjit-fpu' of github.com:hrydgard/ppsspp into armjit-fpu 2013-03-01 18:26:36 +01:00
Sacha
26ebdb4f11 Improve VCMP instruction with option for E.
Add comment to le JIT about how the VCMP crashes on ARM11, with commented code.
2013-03-01 15:41:45 +10:00
Sacha
6d3c89e354 Fix up VCVT function to recognise the difference in encoding for to_int and to_float. There is no 'round to zero' option for to_float. cvt.s.w and cvt.w.s should be unsigned. 2013-03-01 13:45:22 +10:00
Henrik Rydgard
516ca8a0c4 Merge branch 'master' into armjit-fpu
Conflicts:
	Core/MIPS/ARM/ArmJit.h
	Core/MIPS/x86/CompVFPU.cpp
	GPU/GLES/Framebuffer.cpp
2013-02-28 23:56:28 +01:00
Sacha
35a57be115 ARMJIT: Implement MADD, MADDU. Do bitrev if it takes an immediate. Fix a bug where MULTU was being passed through to the interpreter. 2013-02-28 23:45:46 +01:00
Sacha
059abc0d69 ARMJIT: Add floor, ceil, round. Introducing a rounding mode for VCVT.
The cvt and trunc are tested heavily. Floor, ceil, round aren't tested as much as there are very few games that use it.
2013-03-01 01:10:07 +10:00
Sacha
61f5fb35bd ARMJIT: Implement cvt.w.s, cvt.s.w and trunc.w.s that are used heavily in Dragonball.
May need to keep note of FCR to get correct rounding mode? Interpreter doesn't do this either.
2013-02-28 19:46:07 +10:00
Sacha
fe90d5cd06 Add VNEG and VABS implementations and use in FPU2op. 2013-02-27 23:33:59 +10:00
Sacha
ff14815fda ARMJIT: Combine to one instruction for load/stores. 2013-02-27 19:45:01 +10:00