Enhancement: recompiler for PADDSW/PSUBSW

[GoogleCodeExporter]

I've written a recompiler for PADDSW and PSUBSW, which is currently
implemented as an interpreter. I don't know if this improves performance,
though.

Original issue reported on code.google.com by tmkkmac on 18 Jan 2009 at 9:40

Attachments:

• paddsubsw.patch

[GoogleCodeExporter]

Checked Kingdom Hearts2 vids which uses both.
Congratulations: It works, and it adds about 3% more fps :D

Keep those coming, and drop by #pcsx2 on efnet, send me a pm ;)

Original comment by ramapcsx2 on 18 Jan 2009 at 10:49

• Changed state: Accepted

[GoogleCodeExporter]

Reduced instructions a bit, and added a new sse4.1 instruction which helps.

Original comment by tmkkmac on 18 Jan 2009 at 7:22

Attachments:

• paddsubsw_2.patch

[GoogleCodeExporter]

Reduced instruction more, and added some comments.

#instructions for paddsw
rev1: 66 insts
rev2: 47 insts (sse4), 57 insts (sse2)
rev3: 33 insts (sse4), 43 insts (sse2)

Original comment by tmkkmac on 19 Jan 2009 at 11:01

Attachments:

• paddsubsw_3.patch

[GoogleCodeExporter]

Nice work.  We'll run the latest rev through some additional tests and put it 
on SVN
if no problems (don't anticipate any, but no harm in double-checking).

Original comment by Jake.Stine on 20 Jan 2009 at 12:20

[GoogleCodeExporter]

Reduced some redundant insts for sse2.

rev4: 33 insts (sse4), 39 insts (sse2)

Original comment by tmkkmac on 20 Jan 2009 at 6:32

Attachments:

• paddsubsw_4.patch

[GoogleCodeExporter]

Reduced just one more insts. No speed impact on 65-nm era Core2 CPUs.

rev5: 32 insts (sse4), 38 insts (sse2)

Original comment by tmkkmac on 20 Jan 2009 at 2:09

Attachments:

• paddsubsw_5.patch

[GoogleCodeExporter]

Original comment by ramapcsx2 on 21 Jan 2009 at 8:29

• Changed state: Fixed