Files
thejayman77 ec82764bef Initial commit: retroDE_ps2 — first-of-its-kind PS2 GS FPGA core (DE25-Nano / Agilex 5)
RTL (GS rasterizer, EE core stub, platform bridge, LPDDR4B path), sim regression
(272 TBs), docs, and tooling. Copyrighted PS2 content (BIOS, game code, GS dumps,
and all dump-derived textures/traces) is excluded via .gitignore and stays local.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-29 20:10:50 -04:00

50 lines
2.0 KiB
Markdown

# rtl/intc
Interrupt controller scaffolding. Matches `docs/contracts/intc.md`.
## Current contents
- `intc_stub.sv` — generic PS2-style INTC register shell.
Register-visible INTC_STAT / INTC_MASK (offsets parameterized) plus a
16-source injection port `irq_src[15:0]`. The aggregate output
`cpu_irq` is polarity-neutral: the same module is instantiated both
as the EE INTC and as the IOP INTC (with appropriate offsets and a
different set of wired sources in each case).
## Register semantics
- `INTC_STAT` (offset is a parameter; default 0x00): W1C on writes;
sticky until cleared. `irq_src` sets bits on each cycle they're
observed; same-cycle inject-over-W1C collisions keep the pending bit
— interrupts are never silently swallowed.
- `INTC_MASK` (offset is a parameter; default 0x10): plain write-to-set.
Real PS2 uses XOR/toggle semantics on mask writes; stub uses plain
write for simplicity. Escalate if a BIOS trace demands it.
## Instantiation conventions
- **EE INTC**: default offsets (STAT=0x00, MASK=0x10). Instantiated
stand-alone in most benches; the EE memory map does not route INTC
addresses yet (deferred).
- **IOP INTC**: parameterized to STAT=0x70, MASK=0x74 to match real
PS2 IOP INTC placement. Reached through `iop_memory_map_stub` at
physical address 0x1F80_1070+ (region id = 5).
## Wired sources (current)
- EE INTC bit 0 = EE DMAC completion (`dmac_reg_stub.irq_completion_o`).
- IOP INTC bit 0 = IOP DMAC ch9 completion
(`iop_dmac_reg_stub.irq_completion_o`).
Both are one-cycle pulses driven from the respective DMAC's `S_DONE`
state. The INTC latches them into its own pending bit; software (the
TB, for now) reads STAT through the architectural register port and
acks with a W1C write.
## Scope boundary
Module is side-neutral by design. Source-routing from other real
subsystems (timers, GIF/GS, IPU, SPU2, bridge `last_seen_o`) is the
next natural expansion. Re-arm / re-assertion ordering is already
proven in the integration benches.