PC is consistently wrong.

[redsuperbat]

These tests are great for sanity checking the CPU of my emulator.
I have noticed though that for some tests the PC is consistently 1-3 off mark.

I have checked the length of my instructions, they are conforming to the spec. I have also written tests to assure that.
If I disregard the PC check all tests pass with flying colors.

I assume that these tests are correct, but I can't figure out why it's consistently 1-3 off. Perhaps there is some setup I'm missing?

Thanks for the help!

[eeeyrie]

I think I'm having this same issue but I could be wrong
If I remove the check for PC then I go from 52 tests passed to 96 tests passed, and I also notice the same 1-3 off mark issue with at least 1 test (but to be fair I didn't look further than that)

[untrioctium]

I'm passing these and Blargg's CPU instruction tests for my Game Boy emulator. I think the issue is that the fetch behavior of the tests isn't entirely clear. Pseudo-code for my test harness basically looks like:

cpu.load_registers(test.registers)
cpu.fetch_cycle()

for length(test.cycles):
    cpu.cycle()

cpu.register.pc--
check_results()

Notice that I have to prefetch with fetch_cycle. This fetches at the current PC and increments PC, consuming one cycle. The readme states that the memory info in the cycles list is sampled between cycles, meaning you should always have that length plus one cycles. NOP for example implicitly takes two cycles (the length of the list is one), and the first cycle is the fetch of that NOP.

I also have to decrement PC when done because my instructions perform the fetch as part of their last cycle, and the tests don't simulate that overlap. You'll also need to do this if you overlap in the execution.

I can't speak to the accuracy of the actual memory cycle data since I am not emulating on that level.

Hope this helps and was clear.