- Sign extend values read using LH
- Add conversion functions to represent Signed/Unsigned 16-bit ranges
- Add overflow and truncation functions for 16-bit ranges
- Sign extend imm value in ADD(U)
- Add component overflow and truncation to ADD/SUB functions
- Construct new value in logic operators where result using inputs is undefined
- Return a valid W component from logic operators (if either input has one)
- Compare against high value (y), then low value (x) in less than operators
- Use doubles and implement overflow for Multiply operations to try to increase accuracy
- Use unsigned values in both MUL and DIV operations to make output as accurate as possible
- Implement several variants of shift operators. Trying both arithmetically correct and more analytical approaches with varying success.
Debug updates
- Added ability to force all values to be equal to low precision values before operating on them
- Added feature to test output of operations against a tolerance and print only those which fail
GPU updates
- Colour vertices with valid XY coordinates but no W as cyan to make them easier to spot
- Remove cyan colouring for stale vertices (wasn't useful)
- Added ability to skip debug rendering when needed (like seeing the output of offscreen rendering applied to a sprite).
- Added new mode which shows primitive type
Validity of a pgxp value can now be set for any of four components using individual bit flags. This also allows the potential expansion of more flag data for each component.
- Change processor transfer functions to use take both source and destination values
- Add enum values to all debug functions to double check instructions are correct
- Temporarily store instruction codes in interpreter wrappers as these are sometimes changed by branching
- Change recompiler wrapper functions to store register values in global variables in case they're changed by the original function
- Reset CPU on switching between interpret and recompiler
- Call wrapped functions even when tests fail
- Handle retrieved data overwriting address in load operations
- Change PGXP integration to use wrapper functions
- Use modified function lookup tables to call wrappers or default funnctions based on mode
- Implement wrappers for interpreter too, recompiler sometimes calls interpreter functions
- Add UI elements for selecting CPU modes
- Only call precision version of NCLIP in modes where there are not going to be stale vertices in GTE registers
- Added support for CP0 co-processor functions
- Improve support for CPU functions (still broken in many games)
- Improved debug output to show values for inputs before they're modified
GPU plugin
- Ignore stale vertices (occurs when CPU mode no longer writes to memory)
- Add Cyan vertex colouring for stale vertices
- SH: Use appropriate masking when validating 16-bit values before storing
- Shift operations: on shifts larger than 16bits, first shift elements along, then apply arithmetic for remainder
- Restructured project to base interface on PSX instructions
- Support for all relevant CPU arithmetic and logic instructions
- Debug output available via deferred PGXP calls
- Remove most dependencies on PCSXR
- Still very much a work in progress (lots of errors)
- Add extra debug information to GPU plugin (w values)
a37e3f092f
"peopsgl/xvideo: fixed interlace/frame bit31 toggling. Now X-Men Acade…
…mies are fine. Obsoletes use of odd/even hack."
This fixes a hang on going to the title screen in many games.
Fix stray triangles
- Switch to full floating point RTPS/RTPT
- Clamp w values to near plane
Fix texture corruption in 1.78, copy 2.4 Tweak
- Remove matrix replacement
- Add W component to vertices
- Wrap glVertex3fv calls
- Use counter to call glVertex4fv for PGXP vertices only
- replace calls to glOrtho with new matrix (z value becomes w)
- store w value for each vertex
- if any vertex does not have a w value set whole polygon to 1.0
- Reset vertex between draw calls to prevent stale w values persisting
- validate PGXP values using stored copy of original (allow greater variance)
- properly convert addresses before passing to plugin
- rework memory to use a single pool with offsets
- Implement floating point RTPS/RTPT transform, currently disabled.
In some games PGSX memory pointer was not being set after saved states were reloaded, causing the GPU plugin to fall back on fixed point values.
Ensure that PGSX memory pointer in GPU plugin is always set upon initialisation.
- Catch 8bit reads/writes and invalidate registers and memory as needed
- Prevent reading over the end of stride and count arrays when decoding primitive commands
iCatButler