rsx: Rewrite FP aliased registers handling #17701
Conversation
| * Calculates the lane mask for a given input | ||
| * This is a temporary workaround until the decompiler is rewritten with some IR to allow granular split/gather passes | ||
| */ | ||
| u32 get_src_vector_mask(u32 opcode, int operand); |
There was a problem hiding this comment.
can it be const ?
| u32 get_src_vector_mask(u32 opcode, int operand); | |
| u32 get_src_vector_mask(u32 opcode, int operand) const; |
kd-11
changed the title
|
Breaks outlines in The Simpsons Game. |
|
Regression with COD4. Tried with clean cache, different shader modes etc. PR: cod.mp4Master: 
Ryzen 7 2700 |
|
COD is fixed. Simpsons requires a compiler overhaul. I'll look at the easiest way to do that while also adding UTs. Otherwise the eventual rewrites will be too painful. |
|
Regression with The BIGS 2. Most text in the game is now blue. PR:
Master:
|
kd-11
changed the title
kd-11
changed the title
|
Regression with RR7 demo with similar blue color affecting tire smoke and nitrous. I don't want to spam this topic with regressions since I assume you are well aware of what the issues are. I will wait for any further commits before making any other reports here. Cheers. RR7 PR: rr7PR.mp4RR7 master: rr7master.mp4 |
| u8 fragment_texture::convolution_filter() const | ||
| { | ||
| return ((registers[NV4097_SET_TEXTURE_FILTER + (m_index * 8)] >> 13) & 0xf); | ||
| return ((registers[NV4097_SET_TEXTURE_FILTER + (m_index * 8)] >> 13) & 0x7); |
There was a problem hiding this comment.
Can the convolution filter fix be done in another pr? it seems like it's an important fix with minimal chance for regressions.
There was a problem hiding this comment.
Yea. I'll move that to a smaller PR
kd-11
changed the title
|
CFG is now done with #17743 |
|
Next task: #17753 |
|
This work has been reimplemented from the ground up. Final PR: #17837 |
5 participants
Fixes most visual bugs reported in "I AM ALIVE". The game has other unrelated issues such as flickering shadows, but the root cause for that is different.
Ref: #6897