Remove split restore/save from mm7 #4107
Conversation
|
@Sonicadvance1 I have a feeling that the split in saving/restoring mm7 is something due to past representations of the registers. I can't find a reason for these now. While trying to understand why, I go back a few years to when you split a bunch of code into several files, one of them x87.cpp. Removing this split improves instcountci and passes all the tests, so it doesn't feel like it's necessary. Can you take a look? If it's indeed necessary, we should probably add a test that fails when the split in save and restore is removed. |
The split is due to the instruction only saving 10-bytes rather than 16-bytes per register. We're taking advantage of overlapping writes to simulate the same behaviour. The final store still needs to be 10-bytes in order to ensure we don't accidentally write past the end of a page, and also ensure we don't overwrite some other data after the the tail. This change will break that edge case, we just don't have ASM unittests in place for it. |
…equired The instruction definition only allows these instructions to load/store 94 or 108 bytes, not affecting any bytes afterwards. This is a bit awkward because 80-bit x87 registers are stored at the end. FEX has an optimization today where it uses overlapping loads and stores for the first seven x87 registers, and a split loadstore for the final register. This ensures that we get the correct data while reducing the number of loadstores. We didn't have a unittest in place to ensure we only ever write the correct amount of data, so changes like in FEX-Emu#4107 which look correct from an initial glance, would have resulted in broken behaviour. This unittest ensures both that the instructions don't try to access beyond the end of the page, and also ensures that they don't overwrite subsequent data. Making sure that potentially broken behaviour doesn't make its way in.
No description provided.