Linux Devs: Let's get this memcpy thing finalized, if its not already. I'd like to merge the current state of this branch into trunk as soon as possible, since its currently looking very stable and has been, up to this point, a code cleanup and stabilization project. (more invasive changes coming soon)
git-svn-id: http://pcsx2.googlecode.com/svn/branches/ReorderingMTGS@3518 96395faa-99c1-11dd-bbfe-3dabce05a288
* fixes flickering screen in Soul Calibur 3 (caused by VSYNC register bug)
* Optimized upload of queued Path1 transfers; such that all Path1's are uploaded as a single MTGS packet.
git-svn-id: http://pcsx2.googlecode.com/svn/branches/ReorderingMTGS@3515 96395faa-99c1-11dd-bbfe-3dabce05a288
* Added a few assertions to detect when PATH transfers are started that violate other pending PATH transfers.
* Removed a lot of obsolete code from vif1's DIRECT handler (Vif_Codes.cpp)
* Add alignment to Path1buffer to avoid SSE alignment faults.
git-svn-id: http://pcsx2.googlecode.com/svn/branches/ReorderingMTGS@3503 96395faa-99c1-11dd-bbfe-3dabce05a288
* Removed extern "C" and applied static const to s_clut16 vars. Should be fine since the old x86.S files that needed extern "C" have been removed from zzogl.
* Fixed a compilation error in Win32/Debug builds.
* Changed some references of DEVBUILD ZEROGS_DEVBUILD. Not sure if all of them should be changed over or not, so I just stuck to some of the more obviously correct bits.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3500 96395faa-99c1-11dd-bbfe-3dabce05a288
* use volatile keyword to avoid gcc removing the function...
* Use name variable in asm code instead of %n
* Fix constraint on s_clut16mask. There are input, not output...
Arcum can you look at this 2 things thanks.
-> code is still broken in one place. s_clut16mask & s_clut16mask2
re null in the code generated by gcc ! To fix it (do not know why), we can declare them as static. But I'm not sure
on the impact and I can not test windows...
-> s_clut16mask is declared as a 256bits numbers instead of 128 !
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3494 96395faa-99c1-11dd-bbfe-3dabce05a288
* Make PCSX2 bare minimum reqs include SSE as well as MMX.
* Minor bugfix which could have affected MTGS performance.
* Default GIFpath stuff to use SSE opts.
git-svn-id: http://pcsx2.googlecode.com/svn/branches/ReorderingMTGS@3491 96395faa-99c1-11dd-bbfe-3dabce05a288
Note: What I understand some define use the minus symbol and so gcc does a sign extension on 32bits.
Not compatible with u16...
To fix it I kept only the first 16bits. We could also redefine the constant without the minus
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3487 96395faa-99c1-11dd-bbfe-3dabce05a288
* add xmm register to the clobber list.
* Add input contrainst (avoid to depend of the function attribute and easier to understand)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3480 96395faa-99c1-11dd-bbfe-3dabce05a288
* Major rework of the linker flags. Use some globals flags for -s and -m32
Add a USER_CMAKE_LD_FLAGS variable. Easier to play with advanced link flags for future gcc version (>=4.5)
* Remove useless stub file
[debian]
* minor dependency fix
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3475 96395faa-99c1-11dd-bbfe-3dabce05a288
* Disabled the SSE store version of memzero (no performance benefits and it was messy anyway)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3473 96395faa-99c1-11dd-bbfe-3dabce05a288
Seems slower than memcpy_amd_ so didn't enable it.
I tried two different versions, one generates 0x400 different functions, and the other generates 1 function and jumps to the correct entry point.
The later seems faster, but still slower than memcpy_amd_...
(only tested the title-screen of GoW though...)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3465 96395faa-99c1-11dd-bbfe-3dabce05a288
arcum42