Home

"I'm out of order? You're out of order!"

Pacino is an open source 8-issue out-of-order RISC-V processor targeting the RVA23S64 profile. RTL is written in SystemVerilog, simulated with Verilator, and verified with directed tests, line coverage and mutation testing, with spike as an independent decode oracle. Formal analysis and physical design are in progress; the physical design toolchain has not been chosen. Design decisions, AI-assisted co-design records, and architectural rationale are committed alongside the RTL.

●  Frontend RTL complete: BPU, FTQ, IFU, I-cache, ITLB, PTW, decode
●  Frontend demo: 24 scenarios, waveforms and write-ups, see the Showcase
●  BP-139/BP-140: every FTB branch predicted at any slot count, see IA Tasks tab
●  BP-138: two prediction defects found by running the demo, see IA Tasks tab
●  BPC testbench: 973 checks, full line coverage, mutation-tested, see Blog

architecture
RISC-V RVA23S64
issue width
8-issue OOO
rtl language
SystemVerilog
simulator
Verilator 5.x
compiler
GCC 16.x, LLVM 22.x RISC-V
status
active
frontend
icache · ftq · ifu · ibuf · decode · bpu
dispatch
rename · prf (i/f/v/mm) · rob · reservation stations
execute
int · float · vector · matrix · atomics
lsu
agu · lq · sq · fwd · l1d · vlsu · mmlsu
memory
l2 · l3 · prefetch
mmu
itlb · dtlb · l2tlb · ptw
uncore
tilelink · csr · pmu · trace
protection
pma · pmp

Pacino is designed using a structured AI co-design methodology — a domain expert directing a dual-agent workflow across architectural planning and RTL implementation. Prompts, iteration history, and design rationale are committed alongside the RTL as first-class artifacts. Full methodology documentation is in the FAQ and the repo.