PVM: Signed division should always round to zero

[koute]

As reported in: w3f/jamtestvectors#3 (comment)

For division instructions we need to round towards zero instead of flooring (the fractional part of the result should always be discarded).

[emielsebastiaan]

This is likely an incomplete solution. We implemented these changes and now test vectors for all shar* (Shift Arithmetic Right) testvectors fail. I.e., instructions: 140, 146, 153, 157, 199 & 209
Interestingly enough before this PR our implementation for these instructions were passing.

Example:

• shar_r_64 (209)
• Testvector: shift_arithmic_right_64.json

Input

ω_A (register: 7) = 9223372036854775925
ω_B (register: 8) = 3
ω_D (register: 9) = Destination register for outcome

Calculations

Z_8(ω_A) = -9223372036854775691
2**(ω_B % 64) = 8
Division: -9223372036854775691 / 8 = -1152921504606846961,375
Floor/Ceil: rtz(-9223372036854775691 / 8) = rtz(-1152921504606846961,375) = -1152921504606846961
Calculated output: ω_D (register: 9) = Z_inv_8(-1152921504606846961) = 17293822569102704655

Expected Output Test Vectors

ω_D= 17293822569102704654 (!= Calculated output: 17293822569102704655)

Suggested follow-up

• Either GP definition issue with current shar* instructions: 140, 146, 153, 157, 199 & 209.
• Or test vectors are incorrect.

[koute]

Hm, at least with arithmetic shifts you're correct. I'll double check this. Thanks.

[koute]

Okay, I double checked, and I indeed had a brain fart here. Apologies. Only divisions should have this; thanks @emielsebastiaan.

[emielsebastiaan]

Yes confirmed... in all those other modifications you made ealier (140, 146, 153, 157, 199, 209, 213 & 215) the division only protected against overflow. All good now! thanks