Add Paired Single Move Tests #56
Conversation
Adds a new file for common float tests -- This will likely be used more in the future as well Modifies two other tests which use thing now use functions in the CommonFloat file (the location of frsqrte_expected very well may change) Adds software implementation of mantissa truncation and rounding given the expected rounding mode Both of these are for tests for float move instructions These instructions include ps_merge, ps_mr, ps_neg, ps_abs, ps_nabs, and ps_rsqrte is included to check for a similar quirk
|
Your CommonFloat.h file is rather similar to Dolphin's Common/FloatUtils.h. I think it would be better to name the file FloatUtils.h, and to follow the approach of Dolphin's Common/FloatUtils.h where relevant (e.g. constexpr for constants instead of enums). Will review the new hwtest in more detail once you've moved this out of draft state. |
Adds instructions I initially forgot - ps_sel, ps_sum0, ps_sum1, ps_res Adds more test cases Adds isync Corrects *everything* to sync on hardware
|
I still haven't been able to get |
`fres` is reimplemented in an unoptimized way but it handles all edge cases The specific magic numbers were binary searched on hardware Someday NI maybe should be an option with the normal fres tests and not only ps_res
In my experience, wiiload over wi-fi can sometimes be unreliable (but I think this does depend on your access point hardware). If you use a wired ethernet/USB connection (for the Wii, though your PC can be wireless still) it's much more stable.
Are you checking different versions of hwtests, or different versions of Dolphin? It'd be useful to make that some of the tests fail before your PR (dolphin-emu/dolphin#13059) and pass with the PR. |
|
I'm sorry I forgot to update on that, I did get it resolved ^^ It was 5g vs 2.4g wifi!! When connecting to the same network it worked """perfectly""" (sometimes it just got stuck for some reason and I'd have to restart my entire Wii U via unplug and plug back in -> VWii, but this is apparently a common occurrence for some reason), which is what I used to verify all tests worked in the end :> |
Adds a new file for common float tests -- This will likely be used more in the future as well
Modifies two other test files which now use functions in the CommonFloat file
The location of
frsqrte_expectedmight change because it's not exactly common? ^^;Adds software implementation of mantissa truncation and rounding given the expected rounding mode
Both of these are for tests for float move instructions
These instructions include:
ps_merge, ps_mr, ps_neg, ps_abs, ps_nabs
ps_rsqrte is included to check for a similar quirk (it returns an f64)
Marked as a draft until I or someone else can test on hardware (so like. a few hours if I can get it working)