Commit Graph

120 Commits

Author SHA1 Message Date
Henrik Rydgård
b346142df8 Merge pull request #11954 from unknownbrackets/vfpu-chunk5
Fix prefix and size handling for vsbx, vsocp, and integer conv ops
2019-04-01 17:12:03 +02:00
Unknown W. Brackets
b24f84d1a2 interp: Handle prefixes on matrix init ops. 2019-03-31 17:11:24 -07:00
Unknown W. Brackets
59905de719 interp: Correct vsgn out of swizzle bounds. 2019-03-31 17:10:51 -07:00
Unknown W. Brackets
b881a689c4 interp: Ignore high bit in vmfvc/vmtvc.
Both 0 and 128 read/write the S prefix, for example.
2019-03-31 17:09:55 -07:00
Unknown W. Brackets
175ceef583 interp: Cleanup vsocp size handling. 2019-03-31 13:52:07 -07:00
Unknown W. Brackets
4a2f8a74dc interp: Correct size handling for vi2x ops. 2019-03-31 13:51:12 -07:00
Unknown W. Brackets
b75690787e interp: Correct swizzle on vx2i ops. 2019-03-31 13:51:12 -07:00
Unknown W. Brackets
68cdcba6c5 interp: Don't write lane 2 on single colorconv.
Not that it's valid to use the op with that size anyway.
2019-03-31 13:51:12 -07:00
Unknown W. Brackets
5749ae09d0 interp: Correct vmfvc register behavior.
The target and source registers were completely wrong.
2019-03-31 13:41:48 -07:00
Unknown W. Brackets
b2e024025f interp: Handle wrong sizes of vf2h/vh2f.
Probably not ever used, but they have consistent behavior.
2019-03-31 13:41:48 -07:00
Unknown W. Brackets
af3ed69144 interp: Mask moves to vfpu ctrl.
These bits of the registers can't be written.
2019-03-31 10:37:07 -07:00
Unknown W. Brackets
d40ac043d4 interp: Handle prefixes for Vmmov/Vmmul/Vmscl.
I doubt any actual code uses this, but we have some tricky VFPU bugs left,
so just trying for maximum accuracy in the interpreter.
2019-03-31 10:37:07 -07:00
Unknown W. Brackets
ec1dae57eb interp: Fix vbfy prefix handling. 2019-03-31 10:09:18 -07:00
Unknown W. Brackets
d5273f589a interp: Mask value in vpfxd.
The actual register ends up with only the lower 12 bits, which makes sense
since those are the only ones that do anything.
2019-03-31 08:23:36 -07:00
Unknown W. Brackets
7dc775e54f IR: Use interp for unhandled prefix cases.
The interpreter is not changed yet, so in theory this shouldn't change
behavior.
2019-03-31 08:17:11 -07:00
Unknown W. Brackets
4c3aa841d3 IR: Correct vmmul optimizations.
It's a bit confusing since it's not D = S*T, but rather D = S'*T.
2019-03-23 15:31:10 -07:00
Unknown W. Brackets
6178a1fb33 Jit: Correct vocp prefix handling.
See #5549.  Matches tests for various prefix settings.
2019-02-23 09:15:26 -08:00
Unknown W. Brackets
d7f40afd9d interp: Correct vocp prefix handling.
Also, guess that vsocp also applies prefixes.  See #5549.
2019-02-21 19:02:16 -08:00
Unknown W. Brackets
419c1fbd73 Jit: Respect flags for jit types and features.
Left some free space for more.
2019-02-03 14:57:08 -08:00
Henrik Rydgård
34f79904fd IR: This optimization is safe when all three regs are consecutive, so avoid disabling it unnecessarily. 2018-01-10 09:19:27 +01:00
Henrik Rydgård
18be23eccc IR: More fixes. Still something wrong with VFPU compares (not caused by this PR). 2018-01-04 19:38:36 +01:00
Henrik Rydgård
1a97f62dc9 Fix running the CPU test from the UI. 2018-01-04 18:10:41 +01:00
Unknown W. Brackets
905d2c2da6 irjit: Cleanup some invalid op handling.
And log blocks the same way as other backends.
2018-01-01 08:38:11 -08:00
Henrik Rydgård
22e65ba80d Get rid of ugly alignment macros and some other cruft, we now have alignas(16) from C++11 2017-08-31 01:14:51 +02:00
Unknown W. Brackets
b483444fab IR: Cleanup some invalid ops. 2017-04-20 21:11:40 -07:00