First PR for vector changes to Testgen.py #580
Conversation
|
Comments at start of Python header |
|
Randomize GPR for address load |
…ing and is now saying it can't create a new process (fork: retry: Resource temporarily unavailable). So Was not able to look at the contents of the obj dump yet. Changes made were minimal though.
|
Looks like we now have a merge conflict. Otherwise, @jordancarlin have you had a chance to review? |
| @@ -137,9 +133,16 @@ def writeSIGUPD_F(rd): | |||
| # *** write this | |||
| return "" | |||
|
|
|||
| def writeSIGUPD_V(vd, sew): | |||
| global sigOffset | |||
There was a problem hiding this comment.
I don't think we have a global variable called sigOffset anymore. And it looks like you don't actually use it here anyway.
There was a problem hiding this comment.
seems like this is still used, or at least shouldnt be delt with in this PR
There was a problem hiding this comment.
The only place I see it is in the unused incrementSigOffset function. The rest of the code uses a different approach now.
| if sew == 16: | ||
| precision = 16 | ||
| loadop = "flh" | ||
| storeop = "sw" |
| def loadVecReg(reg, pointer, sew): | ||
| loadop = f"vle{sew}.v" | ||
| lines = "" | ||
| addrGPR = randint(1,31) |
There was a problem hiding this comment.
Does this need some way of avoiding the signature base register?
There was a problem hiding this comment.
shouldnt the signature register move if it overlaps with the randomly generated register?
There was a problem hiding this comment.
writeCovVector calls handleSignaturePointerConflict to work around this. Looks like this uses writeCovVector_V which doesn't call this.
There was a problem hiding this comment.
yes this is causing and issue. Working on fixing it
| loadop = "fld" | ||
| storeop = "sd" | ||
| lines = lines + "la x2, scratch\n" | ||
| lines = lines + f"li x3, {formatstrFP.format(val)} # load x3 with value {formatstrFP.format(val)}\n" |
There was a problem hiding this comment.
I think there is a chance these hardcoded registers could conflict with the signature register.
There was a problem hiding this comment.
same convention is followed in function "loadFloatReg"
There was a problem hiding this comment.
loadFloatReg (and all of the other floating point code) is not updated to work with signatures yet. Correct me if I'm wrong, but I thought the intention was to work signatures in from the start for vector.
Changes made on a new branch- the PR from yesterday had conflicts that I wasn't able to undo.
TODO: Small changes to testvector files for vlmax-dependent coverpoints in different length suites.
TODO: Change formatting of tests for some of the MVV.