libretroadmin 827dff7998 IOP: the memory map, the MDEC, the PGIF and the PS1 TTY calls in C89 ...

IopMem.cpp, Mdec.cpp, ps2/pgif.cpp and ps2/Iop/PsxBios.cpp become C89
units. iopMemReadString, which nothing called, is gone with the two IopBios
macros that named it. pgif.h keeps the PGIF's register layouts as C
typedefs, read through macros over the EE and IOP register blocks where
references were; its entry points join the GPU ones in pgifio.h, which
Hw.cpp and R5900.cpp now include, and hwIntcIrq and hwDmacIrq get C
linkage. The PS1 kernel's write and puts calls still read the guest
memory they name, but no longer collect text that nothing printed.

Three things guest data could break are fixed. A PGIF DMA with a block
size of 0 divides by 0x10000 rather than by zero, and words waiting for a
DMA to the IOP drain in a loop where each took a call. The MDEC clamps a
DMA to the buffers it copies through, and a pixel past its clamp table's
range to the table's ends, which is the saturation the table encodes.
tests/iopdev drives the memory map, the TTY calls, MDEC decodes at each
depth and PGIF and GPU DMA traffic with seeded sequences and compares the
hash of every value, call and byte with the transcript the C++ made
(expected.txt, written by the same harness with IOPDEV_TREE=); the C build
then checks the three fixes.

2026-11-09 00:35:16 +01:00

..

build.sh

IOP: the memory map, the MDEC, the PGIF and the PS1 TTY calls in C89

2026-11-09 00:35:16 +01:00

expected.txt

IOP: the memory map, the MDEC, the PGIF and the PS1 TTY calls in C89

2026-11-09 00:35:16 +01:00

main.cpp

IOP: the memory map, the MDEC, the PGIF and the PS1 TTY calls in C89

2026-11-09 00:35:16 +01:00