JIT64: Fix fmul rounding issues #834
Conversation
|
Verified a bunch of Mario Kart Wii replays, Some Mario Kart Double Dash replays, and many Super Smash Bros. Brawl Replays. Also seemed to fix the fact that Mario keeps sticking to angled surfaces on Super Mario Galaxy 2 when going across certain gravity changes in cylindrical gravity fields. SoftFP did other stuff that fixed F-Zero GX and Dave Mirra Pro BMX's replays. This does not seem to affect them in any way, unfortunately. The only thing I was able to acertain is that it is something in the interpreter, not the JIT. |
|
The super guide of DKCR seems to work fine too. See https://wiki.dolphin-emu.org/index.php?title=SF8E01 |
|
Latest update fixes F-Zero GX and all the other issues I was running into. |
| rPS1(_inst.FD) = ForceSingle( -NI_msub( rPS1(_inst.FA), rPS1(_inst.FC), rPS1(_inst.FB) ) ); | ||
| double c0 = Force25Bit(rPS0(_inst.FC)); | ||
| double c1 = Force25Bit(rPS1(_inst.FC)); | ||
| rPS0(_inst.FD) = ForceSingle( -NI_msub( rPS0(_inst.FA), c0, rPS0(_inst.FB) ) ); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
FioraAeterna
force-pushed
the
fixfmulrounding
branch
from
0b67ed2
to
6c2a516
Compare
| @@ -111,24 +129,25 @@ void Jit64::fmaddXX(UGeckoInstruction inst) | |||
| int d = inst.FD; | |||
|
|
|||
| fpr.Lock(a, b, c, d); | |||
| MOVSD(XMM0, fpr.R(a)); | |||
| MOVSD(XMM0, fpr.R(c)); | |||
| Force25BitPrecision(XMM0, XMM1); | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
FioraAeterna
force-pushed
the
fixfmulrounding
branch
from
6c2a516
to
881cb21
Compare
|
I don't really know anything about proper PPC-ish float rounding, so I'll take the easy way out and say: If it passes more tests than before, it's good. In other words, hwtest pls :) |
|
I checked the Hardware tests, and on Master the CPU test that magumagu gave me for the softFP rounding issues goes off screen and I can't copy all of the errors. On this branch, it's slightly better in that the number of problems seems to fit on the screen. I don't know if I have the latest version of the CPU test for softFP. |
|
Could you rebase and fix the OS X build issues? |
Thanks to magumagu's softfp experiments, we know a lot more about the Wii's strange floating point unit than we used to. In particular, when doing a single-precision floating point multiply (fmulsx), it rounds the right hand side's mantissa so as to lose the low 28 bits (of the 53-bit mantissa). Emulating this behavior in Dolphin fixes a bunch of issues with games that require extremely precise emulation of floating point hardware, especially game replays. Fortunately, we can do this with rather little CPU cost; just ~5 extra instructions per multiply, instead of the vast load of a pure-software float implementation. This doesn't make floating-point behavior at all perfect. I still suspect fmadd rounding might not be quite right, since the Wii uses fused instructions and Dolphin doesn't, and NaN/infinity/exception handling is probably off in various ways... but it's definitely way better than before. This appears to fix replays in Mario Kart Wii, Mario Kart Double Dash, and Super Smash Brothers Brawl. I wouldn't be surprised if it fixes a bunch of other stuff too. The changes to instructions other than fmulsx may not be strictly necessary, but I included them for completeness, since it feels wrong to fix some instructions but not others, since some games we didn't test might rely on them.
FioraAeterna
force-pushed
the
fixfmulrounding
branch
from
881cb21
to
4d7b127
Compare
JIT64: Fix fmul rounding issues
Thanks to magumagu's softfp experiments, we know a lot more about the Wii's
strange floating point unit than we used to. In particular, when doing a
single-precision floating point multiply (fmulsx), it rounds the right hand
side's mantissa so as to lose the low 28 bits (of the 53-bit mantissa).
Emulating this behavior in Dolphin fixes a bunch of issues with games that
require extremely precise emulation of floating point hardware, especially
game replays. Fortunately, we can do this with rather little CPU cost; just ~5
extra instructions per multiply, instead of the vast load of a pure-software
float implementation.
This doesn't make floating-point behavior at all perfect. I still suspect
fmadd rounding might not be quite right, since the Wii uses fused instructions
and Dolphin doesn't, and NaN/infinity/exception handling is probably off in
various ways... but it's definitely way better than before.
This appears to fix replays in Mario Kart Wii, Mario Kart Double Dash, and
Super Smash Brothers Brawl. I wouldn't be surprised if it fixes a bunch of
other stuff too.
The changes to instructions other than fmulsx may not be strictly necessary,
but I included them for completeness, since it feels wrong to fix some
instructions but not others, since some games we didn't test might rely on
them.