File formats
============
tests.in
--------
Each test has the format:
<arbitrary test description>
AF BC DE HL AF' BC' DE' HL' IX IY SP PC
I R IFF1 IFF2 IM <halted> <tstates>
<halted> specifies whether the Z80 is halted.
<tstates> specifies the number of tstates to run the test for, in
decimal; the number actually executed may be higher, as the final
instruction is allowed to complete.
Then followed by lines specifying the initial memory setup. Each has
the format:
<start address> <byte1> <byte2> ... -1
eg
1234 56 78 9a -1
says to put 0x56 at 0x1234, 0x78 at 0x1235 and 0x9a at 0x1236.
Finally, -1 to end the test. Blank lines may follow before the next test.
tests.expected
--------------
Each test output starts with the test description, followed by a list
of 'events': each has the format
<time> <type> <address> <data>
<time> is simply the time at which the event occurs.
<type> is one of MR (memory read), MW (memory write), MC (memory
contend), PR (port read), PW (port write) or PC (port contend).
<address> is the address (or IO port) affected.
<data> is the byte written or read. Missing for contentions.
After that, lines specifying AF, BC etc as for .in files. <tstates>
now specifies the final time.
After that, lines specifying which bits of memory have changed since
the initial setup. Same format as for .in files.
Why some specific tests are here
================================
37_{1,2,3}: check the behaviour of SCF with respect to bits 3 and 5
(bug fixed on 20040225).
cb{4,5,6,7}{7,f}_1: designed to check that bits 3 and 5 are copied to
F only for BIT 3,<arg> and BIT 5,<arg> respectively
(bug fixed on 20040225).
However, later research has revealed the bits 3
and 5 are copied on all BIT instructions, l.
tests are now essentially redundant.
d{3,b}_{1,2,3}: check for correct port contention on IO in the four
relevant states (port high byte in 0x40 to 0x7f or not,
port low bit set or reset).
dd00.in, ddfd00.in: test timings of "extended NOP" opcodes DD 00 and
DD FD 00; the extra 00 at the end is to check the
next opcode executes at the right time (bug fixed
on 20060722).
