search

nukeykt / Nuked-OPN2

Cycle-accurate Yamaha YM3438(YM2612) emulator
GNU Lesser General Public License v2.1
214 stars 15 forks source link

Eliminate more shifting of signed integers. #21

Closed Clownacy closed 2 years ago

Clownacy commented 2 years ago

To be honest, this might be a vain effort: there are many more cases of shifting signed integers throughout the codebase, and GCC considers them such a non-issue that it doesn't even have an option for producing compiler warnings when they happen. This might just be one of those cases of something that technically is against the C standard but doesn't harm portability to any significant degree (like assuming that negative signed integers are expressed in binary as two's complement).

nukeykt commented 2 years ago

hmm, doesn't this cause off by one error for negative numbers? like -1 >> 10 = -1, -1 / 1024 = 0

Clownacy commented 2 years ago

You're right...

There are some suggestions here on how to implement a portable arithmetic right shift, but none of them seem very efficient: https://stackoverflow.com/questions/31879878/how-can-i-perform-arithmetic-right-shift-in-c-in-a-portable-way

Arithmetic right shifting hinges on the idea of repeating the sign bit, which doesn't make sense for signed number representations such as sign-magnitude, so arithmetic right shifting in general isn't strictly portable either as it only makes sense for ones' complement and two's complement.

The only "perfect" solution I can think of is forgoing signed types entirely, and implementing two's complement manually with unsigned integers.

jotego commented 2 years ago

I'd rather leave the arithmetic shifts too. What computer architecture is around that doesn't use 2's complement encoding and that may be used for OPN emulation?