*This was actually breaking the fix for New Tetris.. Originally we were doing -1 for the clear screen check in Fillrect, this been updated too. The check in SetScissor did not need to be updated as it was already clamping when the if the N64 resolution was equal or greater..
*Always clear z buffer first in fillrect
*Properly trigger DP interrupt, some games are very sensitive when we ignore it. Blast Corps no longer hangs after the Rare logo. Also ensure to trigger the interrupt when frameskip its enabled to avoid stalls
*Added bound checks for DLinMem and DLCount
*Disabled invalid address bound checks for the PSP. With the recent changes, now its safer to disable it
*Added more bound checks
*Removed compiler workaround in MatrixFromN64FixedPoint and GBI1_DL, I checked and latest PSP toolchain optimizes this correctly now
*Removed GBI0_Tri4 (F3D microcode doesn't support it)
*Moved GBI0_Tri4 implementation to Ucode_GE.h since this is actually part of the custom microcode used by GE and PD.
*Refactored RDP_MoveMemLight to prevent duplicate code and simplified it for adding custom lighting impls. Also added bound checks for invalid light index and addresses
*Added hash for Rogue Squadron (this a custom GBI1 ucode), currently unimplemented
*Do not hardcode GBI commands when patching custom ucode tables
*Fixed DKR's index mtx address, thanks Rinnegatamante for pointing out this error
*Remove SOTE ucode implementation, this was wrong. Now both SOTE and WRUS will use the Beta ucode
*Fix typo from previous commit that caused that changed ucode hash.
*Added WIP to set the vertex stride at compile time to allow the compiler to do further optimizations when the stride its a power of two. Currently its only done for the new F3DBETA ucode as a test.
* Auto defect and cache ucode functions and vertex stride
* Speed up custom ucode detection, avoid string detection and return as soon as there's a match
* Do not attempt to auto defect ucodes that lack a version string
*Alot of code clean up, now custom ucodes should be easier to add (everything related its now in Microcode.cpp)
*Bound checking for InitMicrocode and make sure to set a default ucode version if there's a failure
*Hardcode code_size to avoid issues when hashing ucodes when games set an invalid size, ex Conker
*Simpler way to clear the ucode cache
*Always include gNormalInstructionName even on release builds, the memory we saved wasn't worth the hassle to keep it out of release builds..
*Implemented IsVertexInfoValid to add proper bound checking our vertextinfo
*Implemented IsAddressValid to add proper bound checking when referencing an N64 ram address, currently its only used for DMA_Tri_DKR
*Fixed bug bugged addresses in DKR ucode
*Removed now redundant bound checks, some of these were either wrong pr inconsistent
*Correct bad pointer casting for optimized copy in Yoshi_Memrect and fixed non optimized copy
*Store n64 ram offset rather than the system memory for Fast TMEM (thanks strmnnrmn!). Fixes random crashes when using the non accurate path for TMEM emulation
*Make sure to reset tmem block for fast TMEM even when accurate TMEM its used since we fall back to the non accurate path when games set line to 0
*Ignore load tile for accurate TMEM when line its 0, this was causing a crash in Paper Mario which sets line = 0
*Fixed possible overflow of ucode entries, also spread entries to avoid always overriding the last entry
*Fixed out of bounds assert for microcode data
*Properly check both microcode data and code base for ucode cache
*Removed gLastUcodeBase, potentially caused ucodes not to be loaded correctly. This was legacy code before we had the ucode cache
*Fixed 32bit pointer assumption on DLParser_SetCustom
[!] LightVerts() now returns only R,G & B (V3 and no A)
[!] Lights color and direction now uses V3
[!] Fixed RDP_MoveMemLight() to proper light direction.
[!] PSP VFPU ASM compatible with new light struct as well possible to interface with Conker's new TnL. (Corn)
[!] Slight tidy in audio Alist processing loop. (Corn)
[~] Keep bitfields for texrect, since compiler generates much better asm when it converts int to float (~30 ops)
[!] Clean ups to gu oshle patches, also optimize when reading from stack since is always physical memory
git-svn-id: https://subversion.assembla.com/svn/Daedalusx64/trunk@1674 42e9bfbe-799a-4a2d-bad1-236e862a387a
