progress: new assembler works, instruction set fully implemented with call/ret and push/pop, bf.dsa works which means the dsa assembly language works reliably. still a few bugs to fix. might be able to squeeze out a tiny bit more performance

dsa/emulator/Cargo.toml

@@ -0,0 +1,40 @@
+[package]
+name = "emulator"
+version = "0.1.0"
+edition = "2024"
+
+[[bin]]
+name = "dsa"
+path = "src/main.rs"
+
+[lib]
+name = "dsa"
+path = "src/lib.rs"
+
+[dev-dependencies]
+criterion = { version = "0.5.0", features = ["html_reports"] }
+
+[dependencies]
+arc-swap = "1.8.2"
+clap = { version = "4.5.60", features = ["derive"] }
+fxhash = "0.2.1"
+ron = "0.12.0"
+serde = { version = "1.0.228", features = ["derive"] }
+
+common = { path = "../common" }
+#assembler = { path = "../assembler" }
+
+
+[[bench]]
+name = "bench_mainstore"
+harness = false
+
+[features]
+default = ["mainstore-bulkalloc"]
+
+# Memory Bank Features
+mainstore-bulkalloc = [] # Fastest for Writes
+mainstore-prealloc = [] # Fastest for Reads
+mainstore-stackarray = [] # Slightly outperforms ArrayMap but requires a large stack
+mainstore-arraymap = [] # Simple implementation
+mainstore-hashmap = [] # Old implementation, no raw pointers. Uses a hashmap

dsa/emulator/benches/bench_mainstore.rs

@@ -0,0 +1,208 @@
+use criterion::{
+    BenchmarkGroup, BenchmarkId, Criterion, Throughput, black_box, criterion_group, criterion_main,
+    measurement::WallTime,
+};
+use std::time::Duration;
+
+use dsa::{Page, RandomAccessMemory};
+
+type PhysAddr = u32;
+
+// ── address generators ────────────────────────────────────────────────────────
+
+fn sequential_addrs(n: usize, max_addr: u32) -> Vec<PhysAddr> {
+    (0..n).map(|i| ((i as u32 * 4) & (max_addr - 1))).collect()
+}
+
+fn random_addrs(n: usize, max_addr: u32) -> Vec<PhysAddr> {
+    let mut addrs = Vec::with_capacity(n);
+    let mut x: u32 = 0xdeadbeef;
+    for _ in 0..n {
+        x = x.wrapping_mul(1664525).wrapping_add(1013904223);
+        addrs.push((x & (max_addr - 1)) & !3);
+    }
+    addrs
+}
+
+fn page_stride_addrs(n: usize) -> Vec<PhysAddr> {
+    (0..n).map(|i| (i as u32 * 4096) & 0x00FF_FFFF).collect()
+}
+
+fn random_page_addrs(n: usize, max_addr: u32) -> Vec<PhysAddr> {
+    let mut addrs = Vec::with_capacity(n);
+    let mut x: u32 = 0xc0ffee42;
+    for _ in 0..n {
+        x = x.wrapping_mul(1664525).wrapping_add(1013904223);
+        addrs.push((x & (max_addr - 1)) & !0xFFF);
+    }
+    addrs
+}
+
+// ── runners ───────────────────────────────────────────────────────────────────
+
+fn run_read_byte(
+    group: &mut BenchmarkGroup<WallTime>,
+    addrs: &[PhysAddr],
+    implementations: &mut [(&str, Box<dyn RandomAccessMemory>)],
+) {
+    group.throughput(Throughput::Elements(addrs.len() as u64));
+    for (name, mem) in implementations.iter() {
+        group.bench_with_input(BenchmarkId::new(*name, ""), addrs, |b, addrs| {
+            b.iter(|| {
+                let mut sum: u32 = 0;
+                for &addr in addrs {
+                    sum = sum.wrapping_add(mem.read_byte(black_box(addr)) as u32);
+                }
+                black_box(sum)
+            })
+        });
+    }
+}
+
+fn run_write_byte(
+    group: &mut BenchmarkGroup<WallTime>,
+    addrs: &[PhysAddr],
+    implementations: &mut [(&str, Box<dyn RandomAccessMemory>)],
+) {
+    group.throughput(Throughput::Elements(addrs.len() as u64));
+    for (name, mem) in implementations.iter_mut() {
+        group.bench_with_input(BenchmarkId::new(*name, ""), addrs, |b, addrs| {
+            b.iter(|| {
+                for (i, &addr) in addrs.iter().enumerate() {
+                    mem.write_byte(black_box(addr), black_box(i as u8));
+                }
+            })
+        });
+    }
+}
+
+fn run_read_word(
+    group: &mut BenchmarkGroup<WallTime>,
+    addrs: &[PhysAddr],
+    implementations: &mut [(&str, Box<dyn RandomAccessMemory>)],
+) {
+    group.throughput(Throughput::Elements(addrs.len() as u64));
+    for (name, mem) in implementations.iter() {
+        group.bench_with_input(BenchmarkId::new(*name, ""), addrs, |b, addrs| {
+            b.iter(|| {
+                let mut sum: u32 = 0;
+                for &addr in addrs {
+                    sum = sum.wrapping_add(mem.read_word(black_box(addr)));
+                }
+                black_box(sum)
+            })
+        });
+    }
+}
+
+fn run_write_word(
+    group: &mut BenchmarkGroup<WallTime>,
+    addrs: &[PhysAddr],
+    implementations: &mut [(&str, Box<dyn RandomAccessMemory>)],
+) {
+    group.throughput(Throughput::Elements(addrs.len() as u64));
+    for (name, mem) in implementations.iter_mut() {
+        group.bench_with_input(BenchmarkId::new(*name, ""), addrs, |b, addrs| {
+            b.iter(|| {
+                for (i, &addr) in addrs.iter().enumerate() {
+                    mem.write_word(black_box(addr), black_box(i as u32));
+                }
+            })
+        });
+    }
+}
+
+fn run_read_page(
+    group: &mut BenchmarkGroup<WallTime>,
+    addrs: &[PhysAddr],
+    implementations: &mut [(&str, Box<dyn RandomAccessMemory>)],
+) {
+    group.throughput(Throughput::Bytes(addrs.len() as u64 * 4096));
+    for (name, mem) in implementations.iter() {
+        group.bench_with_input(BenchmarkId::new(*name, ""), addrs, |b, addrs| {
+            b.iter(|| {
+                let mut sum: u64 = 0;
+                for &addr in addrs {
+                    let page = mem.read_page(black_box(addr));
+                    for chunk in page.chunks_exact(8) {
+                        sum = sum.wrapping_add(u64::from_le_bytes(chunk.try_into().unwrap()));
+                    }
+                }
+                black_box(sum)
+            })
+        });
+    }
+}
+
+fn run_write_page(
+    group: &mut BenchmarkGroup<WallTime>,
+    addrs: &[PhysAddr],
+    implementations: &mut [(&str, Box<dyn RandomAccessMemory>)],
+) {
+    let page_data = Page::from([0xABu8; 4096]);
+    group.throughput(Throughput::Bytes(addrs.len() as u64 * 4096));
+    for (name, mem) in implementations.iter_mut() {
+        group.bench_with_input(BenchmarkId::new(*name, ""), addrs, |b, addrs| {
+            b.iter(|| {
+                for &addr in addrs {
+                    mem.write_page(black_box(addr), black_box(&page_data));
+                }
+            })
+        });
+    }
+}
+
+// ── entry point ───────────────────────────────────────────────────────────────
+
+fn benchmarks(c: &mut Criterion) {
+    const N: usize = 4096;
+    const MAX_ADDR: u32 = 0x00FF_FFFF;
+
+    let seq_addrs = sequential_addrs(N, MAX_ADDR);
+    let rand_addrs = random_addrs(N, MAX_ADDR);
+    let page_addrs_seq = page_stride_addrs(N);
+    let page_addrs_rand = random_page_addrs(N, MAX_ADDR);
+
+    let mut implementations: Vec<(&str, Box<dyn RandomAccessMemory>)> = vec![
+        #[cfg(feature = "mainstore-hashmap")]
+        ("HashStore", Box::new(dsa::HashStore::new())),
+        #[cfg(feature = "mainstore-arraymap")]
+        ("ArrayStore", Box::new(dsa::ArrayStore::new())),
+        #[cfg(feature = "mainstore-stackarray")]
+        ("StackArrayStore", Box::new(dsa::StackArrayStore::new())),
+        #[cfg(feature = "mainstore-prealloc")]
+        ("PreAllocStore", Box::new(dsa::PreAllocStore::new())),
+        #[cfg(feature = "mainstore-bulkalloc")]
+        ("BulkAllocStore", Box::new(dsa::BulkAllocStore::new())),
+    ];
+
+    macro_rules! group {
+        ($name:expr, $secs:expr, $runner:ident, $addrs:expr) => {{
+            let mut g = c.benchmark_group($name);
+            g.measurement_time(Duration::from_secs($secs));
+            $runner(&mut g, $addrs, &mut implementations);
+            g.finish();
+        }};
+    }
+
+    group!("write_word/random", 30, run_write_word, &rand_addrs);
+    group!("write_byte/random", 30, run_write_byte, &rand_addrs);
+
+    group!("read_byte/sequential", 30, run_read_byte, &seq_addrs);
+    group!("read_byte/random", 30, run_read_byte, &rand_addrs);
+
+    group!("read_word/sequential", 30, run_read_word, &seq_addrs);
+    group!("read_word/random", 30, run_read_word, &rand_addrs);
+
+    group!("write_byte/sequential", 30, run_write_byte, &seq_addrs);
+    group!("write_word/sequential", 30, run_write_word, &seq_addrs);
+
+    group!("read_page/sequential", 30, run_read_page, &page_addrs_seq);
+    group!("read_page/random", 30, run_read_page, &page_addrs_rand);
+
+    group!("write_page/sequential", 30, run_write_page, &page_addrs_seq);
+    group!("write_page/random", 30, run_write_page, &page_addrs_rand);
+}
+
+criterion_group!(benches, benchmarks);
+criterion_main!(benches);

dsa/emulator/src/args.rs

@@ -0,0 +1,73 @@
+use std::{fs, path::PathBuf};
+
+use clap::{Parser, ValueEnum};
+use serde::{Deserialize, Serialize};
+
+use crate::{MemoryMap, RandomAccessMemory, config::IoMapping, io::DeviceId};
+
+#[derive(Parser, Debug)]
+#[command(version, about, long_about = None)]
+pub struct DsaArgs {
+    /// Memory map to use on boot.
+    /// Overrides default value.
+    #[arg(long = "mmap")]
+    memory_map: Option<PathBuf>,
+
+    #[arg(long = "bin")]
+    binary: Option<PathBuf>,
+
+    #[arg(long = "iomap")]
+    io_map: Option<PathBuf>,
+
+    #[arg(value_enum, long = "mem", default_value = "bulk-alloc")]
+    pub memory_bank: MemoryBank,
+}
+
+impl DsaArgs {
+    pub fn get_memory_map(&self) -> MemoryMap {
+        self.memory_map
+            .as_ref()
+            .and_then(|m| {
+                fs::read_to_string(m)
+                    .ok()
+                    .and_then(|map| ron::from_str(&map).ok())
+            })
+            .unwrap_or_else(|| {
+                ron::from_str(include_str!("./config/default/dsa.mmap.ron")).unwrap()
+            })
+    }
+
+    pub fn get_binary(&self) -> Option<Vec<u8>> {
+        self.binary
+            .clone()
+            .and_then(|path| Some(fs::read(path).unwrap()))
+    }
+
+    pub fn get_io_map(&self) -> Vec<IoMapping> {
+        self.io_map
+            .as_ref()
+            .and_then(|m| {
+                fs::read_to_string(m)
+                    .ok()
+                    .and_then(|map| ron::from_str(&map).ok())
+            })
+            .unwrap_or_else(|| {
+                ron::from_str(include_str!("./config/default/dsa.iomap.ron")).unwrap()
+            })
+    }
+}
+
+#[derive(Debug, Clone, ValueEnum, Default)]
+pub enum MemoryBank {
+    #[cfg(feature = "mainstore-bulkalloc")]
+    #[default]
+    BulkAlloc,
+    #[cfg(feature = "mainstore-arraymap")]
+    ArrayMap,
+    #[cfg(feature = "mainstore-hashmap")]
+    HashMap,
+    #[cfg(feature = "mainstore-prealloc")]
+    PreAlloc,
+    #[cfg(feature = "mainstore-stackarray")]
+    StackArray,
+}

dsa/emulator/src/bin/bench_emu.rs

@@ -0,0 +1,330 @@
+use common::instructions::Instruction;
+use dsa::{BulkAllocStore, Emulator, Page, RandomAccessMemory};
+/// DSA Emulator Benchmark
+///
+/// Place this in `dsa/src/bin/bench.rs` and run with:
+///   cargo run --bin bench --release
+///
+/// Three workloads are tested:
+///   1. ALU-heavy   — tight arithmetic loop, no memory, no branches
+///   2. Memory-heavy — repeated load/store to a working set of addresses
+///   3. Branch-heavy — the bf.dsa dispatch pattern (ieq + jnz chains)
+///
+/// Each workload is a hand-assembled flat binary loaded directly into
+/// the emulator, bypassing all the args/display/IO plumbing.
+use std::{
+    thread,
+    time::{Duration, Instant},
+    u16,
+};
+
+// ---------------------------------------------------------------------------
+// Minimal no-op memory map + IO so the emulator boots without a config file.
+// ---------------------------------------------------------------------------
+
+fn make_emulator() -> Emulator<BulkAllocStore> {
+    use dsa::{
+        config::{MemoryMap, Region, RegionType},
+        io::display::DisplayDevice,
+    };
+
+    let mmap = MemoryMap {
+        table_addr: 0x40000,
+        regions: vec![
+            Region {
+                base: 0x0000_0000,
+                size: 0x0010_0000, // 1 MiB RAM
+                region_type: RegionType::Usable,
+                identity_map_on_boot: true,
+            },
+            Region {
+                base: 0x0020_0000,
+                size: 0x0002_0000, // 128 KiB MMIO (display)
+                region_type: RegionType::MMIO,
+                identity_map_on_boot: false,
+            },
+        ],
+    };
+
+    let iomap = vec![dsa::config::IoMapping {
+        device: dsa::io::DeviceId::Display,
+        base: 0x0020_0000,
+        size: 0x0000_07D0,
+    }];
+
+    let (display, _handle) = DisplayDevice::new(80, 25);
+    Emulator::new(BulkAllocStore::new())
+        .with_device(display)
+        .apply_memory_map(mmap)
+        .apply_io_map(iomap)
+        .unwrap()
+}
+
+// ---------------------------------------------------------------------------
+// Instruction encoding helpers (little-endian words)
+// ---------------------------------------------------------------------------
+
+/// Encode an I-type instruction: [op:6][src:5][dst:5][imm:16]
+fn enc_i(op: u8, src: u8, dst: u8, imm: u16) -> u32 {
+    ((op as u32 & 0x3F) << 26)
+        | ((src as u32 & 0x1F) << 21)
+        | ((dst as u32 & 0x1F) << 16)
+        | imm as u32
+}
+
+/// Encode an R-type instruction: [op:6][r1:5][r2:5][r3:5][r4:5][u6:6]
+fn enc_r(op: u8, r1: u8, r2: u8, r3: u8, r4: u8, u6: u8) -> u32 {
+    ((op as u32 & 0x3F) << 26)
+        | ((r1 as u32 & 0x1F) << 21)
+        | ((r2 as u32 & 0x1F) << 16)
+        | ((r3 as u32 & 0x1F) << 11)
+        | ((r4 as u32 & 0x1F) << 6)
+        | (u6 as u32 & 0x3F)
+}
+
+// Register indices — must match your Register enum discriminants
+const RG0: u8 = 0;
+const RG1: u8 = 1;
+const RG2: u8 = 2;
+const RG3: u8 = 3;
+const RG4: u8 = 4;
+const RG5: u8 = 5;
+const ZERO: u8 = 16;
+
+// Opcodes — must match your Opcode enum discriminants.
+// Fill these in from your ISA.md / encode.rs.
+const OP_NOP: u8 = 0x0D; // adjust to match your actual opcode values
+const OP_HLT: u8 = 0x2B;
+const OP_ADD: u8 = 0x22;
+const OP_ADDI: u8 = 0x23;
+const OP_SUBI: u8 = 0x24;
+const OP_LDW: u8 = 0x07;
+const OP_STW: u8 = 0x0A;
+const OP_LLI: u8 = 0x0B;
+const OP_LUI: u8 = 0x0C;
+const OP_IEQ: u8 = 0x0E;
+const OP_JNZ: u8 = 0x15;
+const OP_JMP: u8 = 0x12;
+
+// ---------------------------------------------------------------------------
+// Workload 1: ALU-heavy
+//
+// Counts down from 10_000_000 using subi, jumps back to top.
+// Pure register arithmetic, no memory accesses after init.
+// Instruction mix: lli, subi, ieq, jnz, hlt
+//
+// Layout (each word at address = index * 4):
+//   0: lli rg0, 0 (upper half of count)        [lui follows]
+//   4: lui rg0, <high>
+//   8: loop: subi rg0, 1, rg0
+//  12: ieq rg0, zero, rg1
+//  16: jnz rg1, zero, 8                         [jump to loop]
+//  20: hlt
+// ---------------------------------------------------------------------------
+fn alu_workload() -> Vec<u8> {
+    let count: u32 = 10_000_000;
+    let lo = (count & 0xFFFF) as u16;
+    let hi = ((count >> 16) & 0xFFFF) as u16;
+
+    let words: &[u32] = &[
+        enc_i(OP_LLI, 0, RG0, lo), // 0x00: lli rg0, lo(count)
+        enc_i(OP_LUI, 0, RG0, hi), // 0x04: lui rg0, hi(count)
+        // loop @ 0x08:
+        enc_i(OP_SUBI, RG0, RG0, 1),         // 0x08: subi rg0, rg0, 1
+        enc_r(OP_IEQ, RG0, ZERO, RG1, 0, 0), // 0x0C: ieq rg0, zero, rg1
+        enc_i(OP_JNZ, RG1, ZERO, 0x0008), // 0x10: jnz rg1, zero, 0x08  <- BUG: jnz jumps when rg1!=0
+        //   rg1=1 when rg0==0, so we want jez here to keep looping
+        //   fix: use jez (jump when rg1==0, i.e. rg0!=0)
+        enc_i(OP_HLT, 0, 0, 0), // 0x14: hlt
+    ];
+    // Note: replace OP_JNZ above with your JEZ opcode so the loop continues
+    // while rg0 != 0. JNZ exits when rg1 != 0, i.e. when rg0 == 0 (done).
+    // The ieq sets rg1=1 only when rg0==0, so jnz rg1 exits the loop correctly.
+    // Actually this IS correct: loop while ieq returns 0 (rg0 != 0),
+    // exit when ieq returns 1 (rg0 == 0). jnz rg1 = jump when rg1 != 0 = jump when done.
+    // We want to jump BACK while not done, so we need jez:
+    //   jez rg1, zero, 0x08  = jump back when rg1==0 (rg0 != 0, not done yet)
+    // Swap OP_JNZ for your JEZ opcode.
+    words_to_bytes(words)
+}
+
+// ---------------------------------------------------------------------------
+// Workload 2: Memory-heavy
+//
+// Repeatedly writes and reads a 64-word working set.
+// Tests page table lookup performance under repeated memory access.
+//
+// rg0 = base address (0x1000, well within RAM, page-aligned)
+// rg1 = loop counter (1000 outer iterations)
+// rg2 = inner counter (64 words per pass)
+// rg3 = scratch for store value
+// ---------------------------------------------------------------------------
+fn memory_workload() -> Vec<u8> {
+    let base: u32 = 0x1000;
+    let outer: u16 = 1000;
+    let inner: u16 = 64;
+
+    let words: &[u32] = &[
+        // init
+        enc_i(OP_LLI, 0, RG0, (base & 0xFFFF) as u16), // rg0 = base
+        enc_i(OP_LLI, 0, RG1, outer),                  // rg1 = outer count
+        // outer_loop @ 0x08:
+        enc_i(OP_LLI, 0, RG2, inner),  // rg2 = inner count
+        enc_i(OP_LLI, 0, RG3, 0xABCD), // rg3 = value to write
+        // inner_loop @ 0x10:
+        enc_i(OP_STW, RG3, RG0, 0),          // stw rg3, rg0, 0
+        enc_i(OP_LDW, RG0, RG3, 0),          // ldw rg0, rg3, 0  (read back)
+        enc_i(OP_ADDI, RG0, RG0, 4),         // rg0 += 4
+        enc_i(OP_SUBI, RG2, RG2, 1),         // rg2 -= 1
+        enc_r(OP_IEQ, RG2, ZERO, RG4, 0, 0), // ieq rg2, zero, rg4
+        enc_i(OP_JNZ, RG4, ZERO, 0x0010),    // jez rg4 -> inner_loop (swap opcode)
+        // restore base, dec outer
+        enc_i(OP_LLI, 0, RG0, (base & 0xFFFF) as u16),
+        enc_i(OP_SUBI, RG1, RG1, 1),
+        enc_r(OP_IEQ, RG1, ZERO, RG4, 0, 0),
+        enc_i(OP_JNZ, RG4, ZERO, 0x0008), // jez rg4 -> outer_loop (swap opcode)
+        enc_i(OP_HLT, 0, 0, 0),
+    ];
+    words_to_bytes(words)
+}
+
+// ---------------------------------------------------------------------------
+// Workload 3: Branch-heavy
+//
+// Mimics the bf.dsa dispatch pattern: compare a value against 8 constants,
+// branch on each. Worst case for branch predictor.
+// rg0 = current "instruction" (cycles through 8 values)
+// rg8-rgf = the 8 bf opcode constants
+// ---------------------------------------------------------------------------
+fn branch_workload() -> Vec<u8> {
+    // 8 bf opcodes: + - > < . , [ ]
+    let opcodes: [u16; 8] = [43, 45, 62, 60, 46, 44, 91, 93];
+    // We'll cycle rg0 through these values and do the full dispatch chain.
+    // To keep it self-contained: load opcodes, set outer loop counter,
+    // inner loop cycles rg0 through all 8 values doing ieq+jnz for each.
+    let outer: u16 = u16::MAX;
+
+    let mut w: Vec<u32> = Vec::new();
+
+    // Load 8 opcode constants into rg8-rgf (regs 8-15)
+    for (i, &op) in opcodes.iter().enumerate() {
+        w.push(enc_i(OP_LLI, 0, 8 + i as u8, op));
+    }
+    // rg1 = outer counter
+    w.push(enc_i(OP_LLI, 0, RG1, outer));
+
+    // outer_loop:
+    let outer_loop_addr = (w.len() * 4) as u16;
+
+    // rg5 = inner counter (8 opcodes)
+    w.push(enc_i(OP_LLI, 0, RG5, 8));
+    // rg2 = pointer into opcode table (we'll use addi to step rg0 through values)
+    // For simplicity: just set rg0 to each opcode in sequence via lli each iteration
+    // This is branch-heavy, not arithmetic-heavy, so the lli overhead is fine.
+    for (i, &op) in opcodes.iter().enumerate() {
+        let next_offset = ((w.len() + 4) * 4) as u16; // addr after the jnz
+        w.push(enc_i(OP_LLI, 0, RG0, op)); // rg0 = opcode
+        w.push(enc_r(OP_IEQ, RG0, 8 + i as u8, RG4, 0, 0)); // ieq rg0, rgX, rg4
+        w.push(enc_i(OP_JNZ, RG4, ZERO, next_offset)); // jnz rg4 (jez to skip)
+        w.push(enc_i(OP_NOP, 0, 0, 0)); // "handler" nop
+    }
+    // dec outer, loop
+    let after_dispatch = (w.len() * 4) as u16;
+    w.push(enc_i(OP_SUBI, RG1, RG1, 1));
+    w.push(enc_r(OP_IEQ, RG1, ZERO, RG4, 0, 0));
+    w.push(enc_i(OP_JNZ, RG4, ZERO, outer_loop_addr)); // jez -> outer_loop
+    w.push(enc_i(OP_HLT, 0, 0, 0));
+
+    words_to_bytes(&w)
+}
+
+fn words_to_bytes(words: &[u32]) -> Vec<u8> {
+    words.iter().flat_map(|w| w.to_le_bytes()).collect()
+}
+
+// ---------------------------------------------------------------------------
+// Runner
+// ---------------------------------------------------------------------------
+
+struct BenchResult {
+    name: &'static str,
+    instructions: u64,
+    elapsed: Duration,
+    mips: f64,
+}
+
+fn run_bench(name: &'static str, binary: Vec<u8>) -> BenchResult {
+    let mut emu = make_emulator();
+    let handle = emu.state_handle();
+
+    let pages: Vec<Page> = Page::paginate(&binary).collect();
+    for (i, page) in pages.iter().enumerate() {
+        emu.memory_mut().write_page(i as u32 * 4096, page);
+    }
+
+    let start = Instant::now();
+    // run() blocks until HLT
+    emu.run().expect("emulator run failed");
+    let elapsed = start.elapsed();
+
+    // read clock from emulator — we need to expose it.
+    // For now, estimate from elapsed + target MIPS.
+    // TODO: expose emulator.clock() -> u64 and use that directly.
+    let instructions = handle.proc.load().clock as u64;
+    let mips = instructions as f64 / elapsed.as_micros() as f64;
+
+    BenchResult {
+        name,
+        instructions,
+        elapsed,
+        mips,
+    }
+}
+
+fn main() {
+    println!("DSA Emulator Benchmark");
+    println!("======================");
+    println!();
+
+    // NOTE: you need to expose `pub fn clock(&self) -> usize` on Emulator
+    // and ensure `run()` returns normally on HLT without blocking in idle_wait.
+    // The idle_wait loop currently spins forever after HLT — for benchmarking,
+    // you want run() to return when the program halts. Add a flag or check:
+    //   if !self.internal_state.running { break 'emu; }
+    // after the HLT handler sets running=false.
+
+    let workloads: &[(&'static str, fn() -> Vec<u8>)] = &[
+        ("ALU-heavy (10M countdown)", alu_workload),
+        ("Memory-heavy (64K rw ops)", memory_workload),
+        ("Branch-heavy (bf dispatch)", branch_workload),
+    ];
+
+    let mut results = Vec::new();
+    for &(name, build) in workloads {
+        print!("Running {}... ", name);
+        let binary = build();
+        let result = run_bench(name, binary);
+        println!("done ({:.1}ms)", result.elapsed.as_secs_f64() * 1000.0);
+        results.push(result);
+    }
+
+    println!();
+    println!(
+        "{:<35} {:>12} {:>12} {:>10}",
+        "Workload", "Instructions", "Time", "MIPS"
+    );
+    println!("{}", "-".repeat(72));
+    for r in &results {
+        println!(
+            "{:<35} {:>12} {:>10.1}ms {:>10.1}",
+            r.name,
+            r.instructions,
+            r.elapsed.as_secs_f64() * 1000.0,
+            r.mips,
+        );
+    }
+
+    let avg_mips: f64 = results.iter().map(|r| r.mips).sum::<f64>() / results.len() as f64;
+    println!("{}", "-".repeat(72));
+    println!("{:<35} {:>35.1} MIPS avg", "Overall", avg_mips);
+}

dsa/emulator/src/config/default/dsa.iomap.ron

@@ -0,0 +1,7 @@
+[
+    IoMapping(
+        device: Display,
+        base: 0x0002_0000,
+        size: 0x0000_07D0, // 2000 bytes (80x25)
+    ),
+]

dsa/emulator/src/config/default/dsa.mmap.ron

@@ -0,0 +1,41 @@
+MemoryMap(
+    table_addr: 0x1000,
+    regions: [
+        // 0000-1000 = reserved
+        // Region(
+        //     base: 0x0,
+        //     size: 0x1000,
+        //     region_type: Reserved
+        // ),
+        // // 1000-4000 = bootloader
+        // Region(
+        //     base: 0x1000,
+        //     size: 0x3000,
+        //     region_type: Bootloader
+        // ),
+        // 5000-0FFFFFFF = MMIO
+        Region(
+            base: 0x20000,
+            size: 0x0FFF_FFFF,
+            region_type: MMIO
+        ),
+        // 10000000-BFFFFFFF = userspace
+        Region(
+            base: 0x1000_0000,
+            size: 0xBFFF_FFFF,
+            region_type: Usable
+        ),
+        // C0000000-C0FFFFFF = kernel page tables
+        Region(
+            base: 0xC000_0000,
+            size: 0x100_0000,
+            region_type: KernelTables
+        ),
+        // C1000000-FFFFFFFF = kernel
+        Region(
+            base: 0xC100_0000,
+            size: 0x3EFF_FFFF,
+            region_type: Kernel
+        )
+    ]
+)

dsa/emulator/src/config/mod.rs

@@ -0,0 +1,80 @@
+use serde::{Deserialize, Serialize};
+
+use crate::{
+    Emulator, RandomAccessMemory,
+    io::DeviceId,
+    memory::{PhysAddr, mmu::MMU},
+};
+
+#[repr(u32)]
+#[derive(Serialize, Deserialize, Clone, Copy, Debug)]
+pub enum RegionType {
+    Reserved,
+    Bootloader,
+    Kernel,
+    KernelTables,
+    Usable,
+    MMIO,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, Copy)]
+pub struct Region {
+    pub base: PhysAddr,
+    pub size: u32,
+    pub region_type: RegionType,
+
+    // flags for the emulator
+    #[serde(default)]
+    pub identity_map_on_boot: bool,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone)]
+pub struct MemoryMap {
+    pub table_addr: PhysAddr,
+    pub regions: Vec<Region>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct IoMapping {
+    pub device: DeviceId,
+    pub base: u32,
+    pub size: u32,
+}
+
+impl MemoryMap {
+    pub const DEFAULT_MEMORY_MAP_ADDR: PhysAddr = 0x1000;
+
+    pub fn new() -> Self {
+        MemoryMap {
+            table_addr: Self::DEFAULT_MEMORY_MAP_ADDR,
+            regions: Vec::new(),
+        }
+    }
+
+    pub fn apply(&self, cpu: &mut Emulator<impl RandomAccessMemory>) {
+        for reg in &self.regions {
+            if reg.identity_map_on_boot {
+                Self::identity_map(cpu.mmu_mut(), reg);
+            }
+        }
+
+        let mem = cpu.memory_mut();
+        let mut offset = self.table_addr;
+        for region in &self.regions {
+            println!("Wrote entry: {region:?} to page table");
+            mem.write_word(offset, region.base);
+            mem.write_word(offset + 4, region.size);
+            mem.write_word(offset + 8, region.region_type as u32);
+            offset += 12;
+        }
+    }
+
+    pub fn identity_map(mmu: &mut MMU, region: &Region) {
+        let first_page = region.base >> 12;
+        let page_count = region.size >> 12;
+
+        for page in (first_page..first_page + page_count).map(|p| p << 12) {
+            mmu.create_mapping(page, page);
+        }
+    }
+}

dsa/emulator/src/io/display.rs

@@ -0,0 +1,80 @@
+use std::sync::{
+    Arc,
+    atomic::{AtomicBool, AtomicU8, Ordering},
+};
+
+use crate::{
+    io::{IoAccess, IoDevice},
+    processor::interrupts::Interrupt,
+};
+
+pub struct DisplayDevice {
+    buffer: Arc<Box<[AtomicU8]>>,
+    dirty: Arc<AtomicBool>,
+}
+
+impl DisplayDevice {
+    pub fn new(width: usize, height: usize) -> (Self, DisplayHandle) {
+        let buffer = Arc::new(
+            (0..width * height)
+                .map(|_| AtomicU8::new(0))
+                .collect::<Vec<_>>()
+                .into_boxed_slice(),
+        );
+        let dirty = Arc::new(AtomicBool::new(false));
+
+        let handle = DisplayHandle {
+            buffer: Arc::clone(&buffer),
+            dirty: Arc::clone(&dirty),
+        };
+
+        (Self { buffer, dirty }, handle)
+    }
+}
+
+impl IoDevice for DisplayDevice {
+    fn size(&self) -> u32 {
+        (self.buffer.len()) as u32
+    }
+    fn access(&self) -> IoAccess {
+        IoAccess::WRITE
+    }
+
+    fn id(&self) -> super::DeviceId {
+        super::DeviceId::Display
+    }
+
+    fn write_word(&mut self, offset: u32, val: u32) -> Result<(), Interrupt> {
+        let bytes = val.to_le_bytes();
+        self.buffer[offset as usize].store(bytes[0], Ordering::Relaxed);
+        self.buffer[offset as usize + 1].store(bytes[1], Ordering::Relaxed);
+        self.buffer[offset as usize + 2].store(bytes[2], Ordering::Relaxed);
+        self.buffer[offset as usize + 3].store(bytes[3], Ordering::Relaxed);
+        self.dirty.store(true, Ordering::Relaxed);
+        Ok(())
+    }
+
+    fn write_byte(&mut self, offset: u32, val: u8) -> Result<(), Interrupt> {
+        self.buffer[offset as usize].store(val, Ordering::Relaxed);
+        self.dirty.store(true, Ordering::Relaxed);
+        Ok(())
+    }
+}
+
+pub struct DisplayHandle {
+    pub buffer: Arc<Box<[AtomicU8]>>,
+    pub dirty: Arc<AtomicBool>,
+}
+
+impl DisplayHandle {
+    pub fn is_dirty(&self) -> bool {
+        self.dirty.swap(false, Ordering::Relaxed)
+    }
+
+    pub fn read_all(&self) -> Vec<u8> {
+        self.buffer
+            .iter()
+            .map(|b| b.load(Ordering::Relaxed))
+            .collect()
+    }
+}

dsa/emulator/src/io/mod.rs

@@ -0,0 +1,124 @@
+use serde::{Deserialize, Serialize};
+
+use crate::{Emulator, RandomAccessMemory, processor::interrupts::Interrupt};
+
+pub mod display;
+
+pub struct MappedDevice {
+    pub base: u32,
+    pub device: Box<dyn IoDevice>,
+}
+
+#[derive(Clone, Copy, Debug, PartialEq)]
+pub struct IoAccess(u8);
+
+impl IoAccess {
+    pub const READ: Self = Self(0b01);
+    pub const WRITE: Self = Self(0b10);
+
+    /// Convenience alias for Read + Write.
+    pub const READWRITE: Self = Self(Self::READ.0 | Self::WRITE.0);
+    pub const NONE: Self = Self(0);
+
+    /// Bitwise OR – returns a new `IoAccess` that contains all flags from both operands.
+    #[inline]
+    pub fn or(self, other: Self) -> Self {
+        Self(self.0 | other.0)
+    }
+
+    /// Check if a flag is present.
+    #[inline]
+    pub fn contains(&self, flag: Self) -> bool {
+        self.0 & flag.0 != 0
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub enum DeviceId {
+    Display,
+    Serial,
+    Random,
+    Timer,
+}
+
+/// Trait representing an input/output device that can be mapped into a memory space.
+///
+/// - Default methods return dummy data.
+pub trait IoDevice: Send {
+    /// Return the size (in bytes) of the device's addressable region.
+    fn size(&self) -> u32;
+
+    /// Return the access permissions for the device.
+    fn access(&self) -> IoAccess;
+
+    /// Return the DeviceId
+    fn id(&self) -> DeviceId;
+
+    #[inline]
+    /// Read a single byte from the specified address.
+    ///
+    /// By default, write‑only devices return `None`.  All other devices
+    /// return `Some(0)` as placeholder data.  Implementations should override
+    /// this method to provide actual read logic.
+    fn read_byte(&self, offset: u32) -> Result<u8, Interrupt> {
+        if self.access().contains(IoAccess::READ) {
+            Ok(0)
+        } else {
+            Err(Interrupt::ReadFromWriteOnly)
+        }
+    }
+
+    #[inline]
+    /// Write a single byte to the specified address.
+    ///
+    /// By default, read‑only devices return `false` indicating failure.
+    /// All other devices return `true`.  Implementations should override
+    /// this method to perform real write operations and return whether
+    /// the operation succeeded.
+    fn write_byte(&mut self, offset: u32, val: u8) -> Result<(), Interrupt> {
+        if self.access().contains(IoAccess::WRITE) {
+            Ok(())
+        } else {
+            Err(Interrupt::WriteToReadOnly)
+        }
+    }
+
+    /// Read a 32‑bit word from the specified address in little‑endian order.
+    ///
+    /// The default implementation composes four consecutive byte reads using
+    /// `read_byte`.  If any byte read fails, the whole operation returns `None`.
+    /// Write‑only devices return `None` immediately.  Override this method for
+    /// more efficient word reads.
+    #[inline]
+    fn read_word(&self, offset: u32) -> Result<u32, Interrupt> {
+        if self.access().contains(IoAccess::READ) {
+            // default: compose from bytes
+            let b0 = self.read_byte(offset)? as u32;
+            let b1 = self.read_byte(offset + 1)? as u32;
+            let b2 = self.read_byte(offset + 2)? as u32;
+            let b3 = self.read_byte(offset + 3)? as u32;
+            return Ok(b0 | b1 << 8 | b2 << 16 | b3 << 24);
+        }
+
+        Err(Interrupt::ReadFromWriteOnly)
+    }
+
+    /// Write a 32‑bit word to the specified address in little‑endian order.
+    ///
+    /// The default implementation splits the value into bytes and writes each
+    /// using `write_byte`.  Read‑only devices return `false`; otherwise returns
+    /// `true` after attempting all byte writes.  Override for more efficient
+    /// word writes or to handle partial failures.
+    fn write_word(&mut self, offset: u32, val: u32) -> Result<(), Interrupt> {
+        if self.access().contains(IoAccess::WRITE) {
+            let bytes = val.to_le_bytes();
+            self.write_byte(offset, bytes[0]);
+            self.write_byte(offset + 1, bytes[1]);
+            self.write_byte(offset + 2, bytes[2]);
+            self.write_byte(offset + 3, bytes[3]);
+            return Ok(());
+        }
+
+        Err(Interrupt::WriteToReadOnly)
+    }
+}

dsa/emulator/src/lib.rs

@@ -0,0 +1,11 @@
+#![feature(likely_unlikely)]
+
+pub mod args;
+pub mod config;
+pub mod io;
+mod memory;
+mod processor;
+
+pub use {config::MemoryMap, processor::processor::Emulator, processor::state::SharedState};
+
+pub use memory::ram::*;

dsa/emulator/src/main.rs

@@ -0,0 +1,96 @@
+use clap::Parser;
+use common::{instructions::Instruction, register::Register};
+use dsa::{
+    BulkAllocStore, Emulator, Page, RandomAccessMemory, SharedState,
+    args::DsaArgs,
+    io::display::{DisplayDevice, DisplayHandle},
+};
+use std::{
+    os::unix::thread::JoinHandleExt,
+    process::exit,
+    sync::{Arc, atomic::Ordering},
+    thread,
+    time::Duration,
+};
+
+fn main() {
+    let args = DsaArgs::parse();
+
+    let mmap = args.get_memory_map();
+    let iomap = args.get_io_map();
+
+    let store = BulkAllocStore::new();
+    let (display, handle) = DisplayDevice::new(80, 25);
+
+    let mut emulator = Emulator::new(store)
+        .with_device(display)
+        .apply_memory_map(mmap)
+        .apply_io_map(iomap)
+        .unwrap();
+
+    if let Some(bin) = args.get_binary() {
+        for (i, ch) in bin.chunks(4).enumerate() {
+            let instruction = Instruction(u32::from_le_bytes(ch.try_into().unwrap()));
+            let hex = ch
+                .iter()
+                .map(|b| format!("{:02X}", b))
+                .collect::<Vec<_>>()
+                .join(" ");
+            let chars = ch
+                .iter()
+                .map(|b| {
+                    if b.is_ascii_graphic() {
+                        *b as char
+                    } else {
+                        '.'
+                    }
+                })
+                .collect::<String>();
+            println!("{:04X}: {:<12} {:4} {:?}", i * 4, hex, chars, instruction);
+        }
+
+        let program: Vec<Page> = Page::paginate(&bin).collect();
+
+        for (i, page) in program.iter().enumerate() {
+            emulator.memory_mut().write_page(i as u32 * 4096, &page);
+        }
+    }
+
+    let state = emulator.state_handle();
+
+    const STACK_SIZE: usize = 1024 * 1024 * 16;
+    let runner = thread::Builder::new()
+        .stack_size(STACK_SIZE)
+        .spawn(move || emulator.run().unwrap())
+        .unwrap();
+
+    let observer = thread::spawn(|| observe(state, handle));
+
+    runner.join().unwrap();
+    thread::sleep(Duration::from_millis(100));
+
+    // yeah we don't need this tbh.
+    // observer.join().unwrap();
+}
+
+/// todo: remove this!
+fn observe(state: Arc<SharedState>, handle: DisplayHandle) {
+    loop {
+        state.update_req.store(true, Ordering::Relaxed);
+        thread::sleep(std::time::Duration::from_millis(20));
+        let state = state.proc.load();
+        // println!(
+        //     "clock: {}, GP Registers: {:?}",
+        //     state.clock, state.registers
+        // );
+
+        if handle.is_dirty() {
+            for line in handle.read_all().chunks(80) {
+                for (i, byte) in line.iter().enumerate() {
+                    print!("{}", *byte as char);
+                }
+                println!();
+            }
+        }
+    }
+}

dsa/emulator/src/memory/mmu.rs

@@ -0,0 +1,109 @@
+use crate::memory::{FaultInfo, PhysAddr, VirtAddr, tlb::TLB};
+
+pub struct MMU {
+    buffer: TLB,
+    enable_paging: bool,
+
+    // Memory Management Register (MMR) Tells MMU where to find page tables
+    mmr_value: u32,
+
+    page_table: Box<[Option<PageTable>; 1024]>,
+}
+
+impl MMU {
+    pub fn new() -> Self {
+        MMU {
+            buffer: TLB::new(),
+            enable_paging: false,
+            mmr_value: 0,
+            page_table: Box::new([const { None }; 1024]),
+        }
+    }
+
+    pub fn paging_enabled(&mut self, enable: bool) {
+        self.enable_paging = enable;
+    }
+
+    pub fn tlb_insert(&mut self, virtual_address: VirtAddr, physical_address: PhysAddr) {
+        self.buffer.insert(virtual_address, physical_address);
+    }
+
+    pub fn create_mapping(&mut self, virtual_address: VirtAddr, physical_address: PhysAddr) {
+        let table = (virtual_address >> 22) as usize;
+        let page = ((virtual_address >> 12) & 0x3FF) as usize;
+
+        self.page_table[table].get_or_insert_default().entries[page] =
+            Some(PageTableEntry::new(physical_address, true, true, false));
+    }
+
+    #[inline(always)]
+    pub fn lookup(&mut self, virtual_address: VirtAddr) -> Result<PhysAddr, FaultInfo> {
+        if !self.enable_paging {
+            return Ok(virtual_address);
+        }
+
+        if let Some(addr) = self.buffer.lookup(virtual_address) {
+            return Ok(addr);
+        }
+
+        self.walk(virtual_address)
+    }
+
+    #[cold]
+    #[inline(never)]
+    fn walk(&mut self, virtual_address: VirtAddr) -> Result<PhysAddr, FaultInfo> {
+        let phys_addr = self.walk_page_table(virtual_address)?;
+        self.buffer.insert(virtual_address, phys_addr);
+        Ok(phys_addr)
+    }
+
+    fn walk_page_table(&mut self, virtual_address: VirtAddr) -> Result<PhysAddr, FaultInfo> {
+        let table = (virtual_address >> 22) as usize;
+        let page = ((virtual_address >> 12) & 0x3FF) as usize;
+
+        if let Some(table) = &self.page_table[table] {
+            return table.entries[page]
+                .map(|entry| entry.physical_addr)
+                .ok_or(FaultInfo::PageFault);
+        } else {
+            Err(FaultInfo::PageFault)
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+struct PageDir {
+    entries: Box<[PageTable; 1024]>,
+}
+
+#[derive(Debug, Clone)]
+struct PageTable {
+    entries: Box<[Option<PageTableEntry>; 1024]>,
+}
+
+impl Default for PageTable {
+    fn default() -> Self {
+        Self {
+            entries: Box::new([None; 1024]),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+struct PageTableEntry {
+    physical_addr: PhysAddr,
+    writable: bool,
+    present: bool,
+    user: bool,
+}
+
+impl PageTableEntry {
+    fn new(physical_addr: PhysAddr, writable: bool, present: bool, user: bool) -> Self {
+        Self {
+            physical_addr,
+            writable,
+            present,
+            user,
+        }
+    }
+}

dsa/emulator/src/memory/mod.rs

@@ -0,0 +1,15 @@
+use serde::{Deserialize, Serialize};
+
+use crate::{RandomAccessMemory, memory::mmu::MMU, processor::processor::Emulator};
+
+mod cache;
+pub mod mmu;
+pub mod ram;
+mod tlb;
+
+pub type VirtAddr = u32;
+pub type PhysAddr = u32;
+
+pub enum FaultInfo {
+    PageFault,
+}

dsa/emulator/src/memory/ram/arraymap.rs

@@ -0,0 +1,107 @@
+use std::{
+    alloc::{Layout, alloc_zeroed},
+    hint::{likely, unlikely},
+};
+
+use crate::memory::{
+    PhysAddr, RandomAccessMemory,
+    ram::{NUM_PAGES, Page, idx},
+};
+
+pub struct ArrayStore {
+    pages: Box<[*mut Page; NUM_PAGES]>,
+}
+
+unsafe impl Send for ArrayStore {}
+impl ArrayStore {
+    pub fn new() -> Self {
+        let pages = vec![0 as *mut Page; 2 << 20]
+            .into_boxed_slice()
+            .try_into()
+            .unwrap();
+        Self { pages }
+    }
+
+    fn alloc(&mut self) -> *mut Page {
+        let layout = Layout::from_size_align(4096, 4096).unwrap();
+        unsafe { alloc_zeroed(layout) as *mut Page }
+    }
+}
+
+impl RandomAccessMemory for ArrayStore {
+    #[inline(always)]
+    fn read_word(&self, addr: PhysAddr) -> u32 {
+        debug_assert_eq!(addr % 4, 0);
+        let (idx, offset) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { (self.pages[idx] as *const u32).byte_add(offset).read() };
+        }
+
+        // Slow path: MMIO, fault, etc — never inlined
+        // Since we're only reading, we can safely return 0
+        return 0;
+    }
+
+    #[inline(always)]
+    fn read_byte(&self, addr: PhysAddr) -> u8 {
+        let (idx, offset) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { (self.pages[idx] as *const u8).byte_add(offset).read() };
+        }
+
+        return 0;
+    }
+
+    #[inline(always)]
+    fn read_page(&self, addr: PhysAddr) -> &Page {
+        debug_assert_eq!(addr % 4096, 0);
+        let (idx, _) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { &*(self.pages[idx] as *const Page) };
+        }
+
+        return &Page([0; 4096]);
+    }
+
+    #[inline(always)]
+    fn write_byte(&mut self, addr: PhysAddr, value: u8) {
+        let (idx, offset) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe {
+            (self.pages[idx] as *mut u8).byte_add(offset).write(value);
+        }
+    }
+
+    #[inline(always)]
+    fn write_word(&mut self, addr: PhysAddr, value: u32) {
+        debug_assert_eq!(addr % 4, 0);
+        let (idx, offset) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe { (self.pages[idx] as *mut u32).byte_add(offset).write(value) }
+    }
+
+    #[inline(always)]
+    fn write_page(&mut self, addr: PhysAddr, value: &Page) {
+        debug_assert_eq!(addr % 4096, 0);
+        let (idx, _) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe {
+            (self.pages[idx] as *mut Page).copy_from(value, 1);
+        }
+    }
+}

dsa/emulator/src/memory/ram/bulkalloc.rs

@@ -0,0 +1,134 @@
+use std::{
+    alloc::{Layout, alloc_zeroed},
+    hint::{likely, unlikely},
+};
+
+use crate::memory::{
+    PhysAddr, RandomAccessMemory,
+    ram::{Page, idx},
+};
+
+pub struct BulkAllocStore {
+    pages: Box<[*mut Page; Page::COUNT]>,
+    pre_allocated: [*mut Page; Self::ALLOC_AT_ONCE],
+    idx: u8,
+}
+
+unsafe impl Send for BulkAllocStore {}
+impl BulkAllocStore {
+    const ALLOC_AT_ONCE: usize = 256 as usize;
+
+    pub fn new() -> Self {
+        let pages = vec![0 as *mut Page; Page::COUNT]
+            .into_boxed_slice()
+            .try_into()
+            .unwrap();
+
+        Self {
+            pages,
+            pre_allocated: Self::alloc_set(),
+            idx: 0,
+        }
+    }
+
+    fn alloc_set() -> [*mut Page; Self::ALLOC_AT_ONCE] {
+        let layout = Layout::from_size_align(Page::SIZE * Self::ALLOC_AT_ONCE, Page::SIZE).unwrap();
+        let mut region = unsafe { alloc_zeroed(layout) as *mut Page };
+        let mut set = [0 as *mut Page; Self::ALLOC_AT_ONCE];
+        for i in 0..Self::ALLOC_AT_ONCE {
+            set[i] = region;
+            region = unsafe { region.byte_add(4096) };
+        }
+        set
+    }
+
+    fn alloc(&mut self) -> *mut Page {
+        if self.idx < Self::ALLOC_AT_ONCE as u8 {
+            let page = self.pre_allocated[self.idx as usize];
+            self.idx += 1;
+            page
+        } else {
+            self.idx = 0;
+            self.pre_allocated = Self::alloc_set();
+            self.pre_allocated[0]
+        }
+    }
+}
+
+impl RandomAccessMemory for BulkAllocStore {
+    #[inline(always)]
+    fn read_word(&self, addr: PhysAddr) -> u32 {
+        debug_assert_eq!(addr % 4, 0);
+        let (idx, offset) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { (self.pages[idx] as *const u32).byte_add(offset).read() };
+        }
+
+        // Slow path: MMIO, fault, etc — never inlined
+        // Since we're only reading, we can safely return 0
+        return 0;
+    }
+
+    #[inline(always)]
+    fn read_byte(&self, addr: PhysAddr) -> u8 {
+        let (idx, offset) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { (self.pages[idx] as *const u8).byte_add(offset).read() };
+        }
+
+        return 0;
+    }
+
+    #[inline(always)]
+    fn read_page(&self, addr: PhysAddr) -> &Page {
+        debug_assert_eq!(addr % Page::SIZE as u32, 0);
+        let (idx, _) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { &*(self.pages[idx] as *const Page) };
+        }
+
+        return &Page::ZERO;
+    }
+
+    #[inline(always)]
+    fn write_byte(&mut self, addr: PhysAddr, value: u8) {
+        let (idx, offset) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe {
+            (self.pages[idx] as *mut u8).byte_add(offset).write(value);
+        }
+    }
+
+    #[inline(always)]
+    fn write_word(&mut self, addr: PhysAddr, value: u32) {
+        debug_assert_eq!(addr % 4, 0);
+        let (idx, offset) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe { (self.pages[idx] as *mut u32).byte_add(offset).write(value) }
+    }
+
+    #[inline(always)]
+    fn write_page(&mut self, addr: PhysAddr, value: &Page) {
+        debug_assert_eq!(addr % Page::SIZE as u32, 0);
+        let (idx, _) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe {
+            (self.pages[idx] as *mut Page).copy_from(value, 1);
+        }
+    }
+}

dsa/emulator/src/memory/ram/hashmap.rs

@@ -0,0 +1,66 @@
+use fxhash::FxHashMap;
+
+use crate::memory::{PhysAddr, RandomAccessMemory, ram::Page};
+
+pub struct HashStore {
+    pages: FxHashMap<PhysAddr, Page>,
+}
+
+impl HashStore {
+    pub fn new() -> Self {
+        Self {
+            pages: FxHashMap::default(),
+        }
+    }
+
+    #[inline(always)]
+    const fn segment_addr(addr: PhysAddr) -> (PhysAddr, usize) {
+        (addr & !(0xFFF), (addr & 0xFFF) as usize)
+    }
+}
+
+impl RandomAccessMemory for HashStore {
+    #[inline(always)]
+    fn read_byte(&self, addr: PhysAddr) -> u8 {
+        let (page, offset) = Self::segment_addr(addr);
+        self.pages
+            .get(&page)
+            .map(|p| p.as_ref()[offset])
+            .unwrap_or(0)
+    }
+
+    #[inline(always)]
+    fn read_word(&self, addr: PhysAddr) -> u32 {
+        let (page, offset) = Self::segment_addr(addr);
+        debug_assert_eq!(offset % 4, 0);
+        let page = self.pages.get(&page).unwrap_or(&Page::ZERO);
+        u32::from_be_bytes(page[offset..=offset + 3].try_into().unwrap())
+    }
+
+    #[inline(always)]
+    fn read_page(&self, addr: PhysAddr) -> &Page {
+        debug_assert_eq!(addr % 0x1000, 0);
+        self.pages.get(&addr).unwrap_or(&Page::ZERO)
+    }
+
+    #[inline(always)]
+    fn write_byte(&mut self, addr: PhysAddr, value: u8) {
+        let (page, offset) = Self::segment_addr(addr);
+        self.pages.entry(page).or_insert_with(|| Page::zeroed())[offset] = value;
+    }
+
+    #[inline(always)]
+    fn write_word(&mut self, addr: PhysAddr, value: u32) {
+        let (page, offset) = Self::segment_addr(addr);
+        debug_assert_eq!(offset % 4, 0);
+        let page = self.pages.entry(page).or_insert_with(|| Page::zeroed());
+        page[offset..=offset + 3].copy_from_slice(&value.to_be_bytes());
+    }
+
+    #[inline(always)]
+    fn write_page(&mut self, addr: PhysAddr, value: &Page) {
+        debug_assert_eq!(addr % 0x1000, 0);
+        let page = self.pages.entry(addr).or_insert_with(|| Page::zeroed());
+        page.0 = value.0;
+    }
+}

dsa/emulator/src/memory/ram/mod.rs

@@ -0,0 +1,108 @@
+use std::ops::{Deref, DerefMut};
+
+use crate::memory::PhysAddr;
+
+#[cfg(feature = "mainstore-arraymap")]
+pub mod arraymap;
+#[cfg(feature = "mainstore-arraymap")]
+pub use arraymap::ArrayStore;
+
+#[cfg(feature = "mainstore-hashmap")]
+pub mod hashmap;
+#[cfg(feature = "mainstore-hashmap")]
+pub use hashmap::HashStore;
+
+#[cfg(feature = "mainstore-stackarray")]
+pub mod stackarray;
+#[cfg(feature = "mainstore-stackarray")]
+pub use stackarray::StackArrayStore;
+
+#[cfg(feature = "mainstore-bulkalloc")]
+pub mod bulkalloc;
+#[cfg(feature = "mainstore-bulkalloc")]
+pub use bulkalloc::BulkAllocStore;
+
+#[cfg(feature = "mainstore-prealloc")]
+pub mod prealloc;
+#[cfg(feature = "mainstore-prealloc")]
+pub use prealloc::PreAllocStore;
+
+#[repr(transparent)]
+#[derive(Clone)]
+pub struct Page([u8; 4096]);
+
+impl Page {
+    pub const SIZE: usize = 4096;
+    pub const COUNT: usize = u32::MAX as usize / Self::SIZE + 1;
+
+    pub const MASK: u32 = 0xFFF;
+    pub const ZERO: Self = Page::zeroed();
+
+    pub const fn zeroed() -> Self {
+        Self([0; 4096])
+    }
+
+    pub const fn from(data: [u8; 4096]) -> Self {
+        Self(data)
+    }
+
+    pub fn read_word(&self, offset: usize) -> u32 {
+        u32::from_le_bytes(self.0[offset..offset + 4].try_into().unwrap())
+    }
+
+    pub fn write_word(&mut self, offset: usize, value: u32) {
+        self.0[offset..offset + 4].copy_from_slice(&value.to_le_bytes());
+    }
+
+    pub fn paginate(data: &[u8]) -> impl Iterator<Item = Page> + '_ {
+        data.chunks(4096).map(|chunk| {
+            let mut arr = [0u8; 4096];
+            arr[..chunk.len()].copy_from_slice(chunk);
+            Page(arr)
+        })
+    }
+}
+
+impl Deref for Page {
+    type Target = [u8];
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl DerefMut for Page {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        &mut self.0
+    }
+}
+
+impl TryFrom<Vec<u8>> for Page {
+    type Error = String;
+
+    fn try_from(data: Vec<u8>) -> Result<Self, Self::Error> {
+        let arr: [u8; 4096] = data
+            .try_into()
+            .map_err(|v: Vec<u8>| format!("Expected 4096 bytes, got {}", v.len()))?;
+        Ok(Page(arr))
+    }
+}
+
+#[inline(always)]
+const fn idx(addr: PhysAddr) -> (usize, usize) {
+    ((addr >> 12) as usize, (addr & 0xFFF) as usize)
+}
+
+pub trait RandomAccessMemory {
+    fn read_byte(&self, addr: PhysAddr) -> u8;
+
+    fn write_byte(&mut self, addr: PhysAddr, value: u8);
+
+    fn read_word(&self, addr: PhysAddr) -> u32;
+
+    fn write_word(&mut self, addr: PhysAddr, value: u32);
+
+    fn read_page(&self, addr: PhysAddr) -> &Page;
+
+    fn write_page(&mut self, addr: PhysAddr, value: &Page);
+}

dsa/emulator/src/memory/ram/prealloc.rs

@@ -0,0 +1,50 @@
+use crate::memory::{
+    PhysAddr, RandomAccessMemory,
+    ram::{NUM_PAGES, Page},
+};
+
+pub struct PreAllocStore {
+    heap: Box<[u8; NUM_PAGES * 4096]>,
+}
+
+unsafe impl Send for PreAllocStore {}
+impl PreAllocStore {
+    pub fn new() -> Self {
+        Self {
+            heap: unsafe { Box::new_zeroed().assume_init() },
+        }
+    }
+}
+
+impl RandomAccessMemory for PreAllocStore {
+    #[inline(always)]
+    fn read_word(&self, addr: PhysAddr) -> u32 {
+        unsafe { (self.heap.as_ptr().add(addr as usize) as *const u32).read() }
+    }
+
+    #[inline(always)]
+    fn read_byte(&self, addr: PhysAddr) -> u8 {
+        self.heap[addr as usize]
+    }
+
+    #[inline(always)]
+    fn read_page(&self, addr: PhysAddr) -> &Page {
+        debug_assert_eq!(addr % 4096, 0);
+        unsafe { (self.heap.as_ptr().add(addr as usize) as *const Page).as_ref_unchecked() }
+    }
+
+    #[inline(always)]
+    fn write_byte(&mut self, addr: PhysAddr, value: u8) {
+        self.heap[addr as usize] = value;
+    }
+
+    #[inline(always)]
+    fn write_word(&mut self, addr: PhysAddr, value: u32) {
+        unsafe { (self.heap.as_ptr().add(addr as usize) as *mut u32).write(value) };
+    }
+
+    #[inline(always)]
+    fn write_page(&mut self, addr: PhysAddr, value: &Page) {
+        self.heap[addr as usize..addr as usize + 4096].copy_from_slice(value);
+    }
+}

dsa/emulator/src/memory/ram/stackarray.rs

@@ -0,0 +1,103 @@
+use std::{
+    alloc::{Layout, alloc_zeroed},
+    hint::{likely, unlikely},
+};
+
+use crate::memory::{
+    PhysAddr, RandomAccessMemory,
+    ram::{NUM_PAGES, Page, idx},
+};
+
+pub struct StackArrayStore {
+    pages: [*mut Page; NUM_PAGES],
+}
+
+unsafe impl Send for StackArrayStore {}
+impl StackArrayStore {
+    pub fn new() -> Self {
+        Self {
+            pages: [0 as *mut Page; 2 << 20],
+        }
+    }
+
+    fn alloc(&mut self) -> *mut Page {
+        let layout = Layout::from_size_align(4096, 4096).unwrap();
+        unsafe { alloc_zeroed(layout) as *mut Page }
+    }
+}
+
+impl RandomAccessMemory for StackArrayStore {
+    #[inline(always)]
+    fn read_word(&self, addr: PhysAddr) -> u32 {
+        debug_assert_eq!(addr % 4, 0);
+        let (idx, offset) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { (self.pages[idx] as *const u32).byte_add(offset).read() };
+        }
+
+        // Slow path: MMIO, fault, etc — never inlined
+        // Since we're only reading, we can safely return 0
+        return 0;
+    }
+
+    #[inline(always)]
+    fn read_byte(&self, addr: PhysAddr) -> u8 {
+        let (idx, offset) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { (self.pages[idx] as *const u8).byte_add(offset).read() };
+        }
+
+        return 0;
+    }
+
+    #[inline(always)]
+    fn read_page(&self, addr: PhysAddr) -> &Page {
+        debug_assert_eq!(addr % 4096, 0);
+        let (idx, _) = idx(addr);
+
+        if likely(!self.pages[idx].is_null()) {
+            return unsafe { &*(self.pages[idx] as *const Page) };
+        }
+
+        return &Page::ZERO;
+    }
+
+    #[inline(always)]
+    fn write_byte(&mut self, addr: PhysAddr, value: u8) {
+        let (idx, offset) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe {
+            (self.pages[idx] as *mut u8).byte_add(offset).write(value);
+        }
+    }
+
+    #[inline(always)]
+    fn write_word(&mut self, addr: PhysAddr, value: u32) {
+        debug_assert_eq!(addr % 4, 0);
+        let (idx, offset) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe { (self.pages[idx] as *mut u32).byte_add(offset).write(value) }
+    }
+
+    #[inline(always)]
+    fn write_page(&mut self, addr: PhysAddr, value: &Page) {
+        debug_assert_eq!(addr % 4096, 0);
+        let (idx, _) = idx(addr);
+
+        if unlikely(self.pages[idx].is_null()) {
+            self.pages[idx] = self.alloc();
+        }
+
+        unsafe { (self.pages[idx] as *mut Page).copy_from(value, 1) }
+    }
+}

dsa/emulator/src/memory/tlb.rs

@@ -0,0 +1,54 @@
+use crate::memory::{PhysAddr, VirtAddr};
+
+const TLB_SIZE: usize = 256;
+const TLB_MASK: u32 = (TLB_SIZE - 1) as u32;
+
+pub struct TLB {
+    entries: [TlbEntry; TLB_SIZE],
+}
+
+#[derive(Debug, Clone, Copy)]
+struct TlbEntry {
+    virtual_address: VirtAddr,
+    physical_address: PhysAddr,
+    valid: bool,
+}
+
+impl TLB {
+    pub fn new() -> Self {
+        TLB {
+            entries: [TlbEntry {
+                virtual_address: 0,
+                physical_address: 0,
+                valid: false,
+            }; TLB_SIZE],
+        }
+    }
+
+    #[inline(always)]
+    fn index(page: u32) -> usize {
+        let hash = page ^ (page >> 12);
+        (hash & TLB_MASK) as usize
+    }
+
+    #[inline(always)]
+    pub fn lookup(&self, virtual_address: VirtAddr) -> Option<PhysAddr> {
+        let page = virtual_address >> 12; // for 4KB pages
+        let entry = self.entries[Self::index(page)];
+        if entry.valid && entry.virtual_address == virtual_address {
+            Some(entry.physical_address)
+        } else {
+            None
+        }
+    }
+
+    #[inline(always)]
+    pub fn insert(&mut self, virt: VirtAddr, phys: PhysAddr) {
+        let idx = Self::index(virt) as usize;
+        self.entries[idx] = TlbEntry {
+            virtual_address: virt,
+            physical_address: phys,
+            valid: true,
+        };
+    }
+}

dsa/emulator/src/processor/interrupts.rs

@@ -0,0 +1,52 @@
+#[repr(u8)]
+#[derive(Clone, Copy, Debug)]
+pub enum Interrupt {
+    // CPU exceptions 0-31
+    InvalidInterrupt = 0,
+    PageFault = 1,
+    ProtectionFault = 2,
+    InvalidOpcode = 3,
+    DivideByZero = 4,
+    StackOverflow = 5,
+    WriteToReadOnly = 6,
+    ReadFromWriteOnly = 7,
+    UnmappedIo = 8,
+    // 8-31 reserved for future CPU exceptions
+
+    // IRQ's (32-63)
+    Hardware(u8),
+
+    // Syscalls (64-255)
+    // OS defined, program uses INT 64-127
+    Software(u8) = 128,
+}
+
+impl Interrupt {
+    pub fn code(&self) -> u8 {
+        match self {
+            Interrupt::InvalidInterrupt => 0,
+            Interrupt::PageFault => 1,
+            Interrupt::ProtectionFault => 2,
+            Interrupt::InvalidOpcode => 3,
+            Interrupt::DivideByZero => 4,
+            Interrupt::StackOverflow => 5,
+            Interrupt::WriteToReadOnly => 6,
+            Interrupt::ReadFromWriteOnly => 7,
+            Interrupt::UnmappedIo => 8,
+            Interrupt::Hardware(x) => {
+                if *x < 32 || *x > 63 {
+                    0
+                } else {
+                    *x
+                }
+            }
+            Interrupt::Software(x) => {
+                if *x < 64 {
+                    0
+                } else {
+                    *x
+                }
+            }
+        }
+    }
+}

dsa/emulator/src/processor/mod.rs

@@ -0,0 +1,3 @@
+pub mod interrupts;
+pub mod processor;
+pub mod state;

dsa/emulator/src/processor/processor.rs

@@ -0,0 +1,653 @@
+use std::{
+    hint::unlikely,
+    ops::{Add, AddAssign},
+    sync::{
+        Arc,
+        atomic::Ordering,
+        mpsc::{self, TryRecvError},
+    },
+    thread,
+    time::{Duration, Instant},
+};
+
+use common::{
+    instructions::{Instruction, Opcode},
+    register::Register,
+};
+
+use crate::{
+    Page,
+    config::{IoMapping, MemoryMap, RegionType},
+    io::{IoAccess, IoDevice, MappedDevice},
+    memory::{mmu::MMU, ram::RandomAccessMemory},
+    processor::{interrupts::Interrupt, state::SharedState},
+};
+
+pub struct Emulator<Mem: RandomAccessMemory> {
+    internal_state: ProcessorSnapshot,
+    shared_state: Arc<SharedState>,
+
+    // Interrupts
+    interrupts: mpsc::Receiver<u8>,
+    pending_fault: Option<Interrupt>,
+
+    // memory
+    mmu: MMU,
+    mainstore: Mem,
+
+    // IO
+    mmio: Vec<MappedDevice>,
+    mmio_region: Option<(u32, u32)>, // start end end of segment
+
+    // optionals - not necessarily set on boot.
+    memory_map: Option<MemoryMap>,
+
+    // Config params
+    __mmap_configured: bool,
+    __io_configured: bool,
+}
+
+unsafe impl<Mem: RandomAccessMemory> Send for Emulator<Mem> {}
+impl<Mem: RandomAccessMemory> Emulator<Mem> {
+    pub fn new(mem: Mem) -> Self {
+        let (sender, receiver) = mpsc::channel::<u8>();
+
+        Self {
+            interrupts: receiver,
+            pending_fault: None,
+
+            internal_state: ProcessorSnapshot::default(),
+            shared_state: Arc::new(SharedState::new(sender)),
+
+            memory_map: None,
+            mmu: MMU::new(),
+            mainstore: mem,
+
+            // mmio
+            mmio: Vec::new(),
+            mmio_region: None,
+
+            // Config params
+            __io_configured: false,
+            __mmap_configured: false,
+        }
+    }
+
+    pub fn with_device(mut self, device: impl IoDevice + 'static) -> Self {
+        self.mmio.push(MappedDevice {
+            base: 0,
+            device: Box::new(device),
+        });
+        self
+    }
+
+    pub fn state_handle(&self) -> Arc<SharedState> {
+        self.shared_state.clone()
+    }
+
+    #[inline]
+    pub fn mmu_mut(&mut self) -> &mut MMU {
+        &mut self.mmu
+    }
+
+    #[inline]
+    pub fn memory_mut(&mut self) -> &mut impl RandomAccessMemory {
+        &mut self.mainstore
+    }
+
+    #[must_use]
+    #[inline]
+    pub fn reg(&self, reg: Register) -> u32 {
+        if reg as u8 == Register::Zero as u8 {
+            return 0;
+        }
+        debug_assert!((reg as usize) < ProcessorSnapshot::REG_COUNT);
+
+        unsafe { *self.internal_state.registers.get_unchecked(reg as usize) }
+    }
+
+    #[inline]
+    pub fn mut_reg(&mut self, reg: Register) -> &mut u32 {
+        debug_assert!((reg as usize) < ProcessorSnapshot::REG_COUNT);
+
+        unsafe {
+            self.internal_state
+                .registers
+                .get_unchecked_mut(reg as usize)
+        }
+    }
+
+    #[cold]
+    pub fn apply_memory_map(mut self, map: MemoryMap) -> Self {
+        self.mmio_region = map
+            .regions
+            .iter()
+            .find(|r| matches!(r.region_type, RegionType::MMIO))
+            .map(|r| (r.base, r.base + r.size));
+
+        map.apply(&mut self);
+        self.memory_map = Some(map);
+        self.__mmap_configured = true;
+        self
+    }
+
+    pub fn apply_io_map(mut self, map: Vec<IoMapping>) -> Result<Self, String> {
+        if !self.__mmap_configured {
+            return Err("You must map memory before applying I/O mappings".to_string());
+        }
+
+        for entry in map {
+            if let Some(idx) = self.mmio.iter().position(|d| d.device.id() == entry.device) {
+                self.mmio[idx].base = entry.base;
+            } else {
+                eprintln!("WARN: no device registgered for {:?}", entry.device);
+            }
+        }
+
+        self.__io_configured = true;
+        Ok(self)
+    }
+
+    #[cold]
+    fn update(&mut self) {
+        self.shared_state
+            .proc
+            .store(Arc::new(self.internal_state.clone()));
+    }
+
+    #[cold]
+    fn boot(&mut self) -> Result<(), String> {
+        if !(self.__io_configured && self.__mmap_configured) {
+            return Err("Processor not configured".to_string());
+        }
+
+        if let Some(map) = &self.memory_map {
+            MemoryMap::identity_map(&mut self.mmu, map.regions.get(0).unwrap());
+        }
+
+        self.internal_state.running = true;
+        Ok(())
+    }
+
+    #[cold]
+    fn shutdown(&mut self) {
+        self.internal_state.running = false;
+    }
+
+    #[cold]
+    pub fn idle_wait(&mut self) {
+        self.internal_state.running = false;
+
+        // Wait for an interrupt or state update to continue.
+        loop {
+            // Check for interrupts.
+            if let Ok(code) = self.interrupts.recv_timeout(Duration::from_millis(100)) {
+                self.interrupt(Interrupt::Software(code));
+                break;
+            }
+
+            // // If we've received a request to continue running. DEPRECATED (we can run through interrupts)
+            // if self.shared_state.running.load(Ordering::Relaxed) {
+            //     panic!("3");
+            //     break;
+            // }
+
+            // UI requested a state update.
+            if self.shared_state.update_req.load(Ordering::Relaxed) {
+                self.update();
+            }
+        }
+
+        self.internal_state.running = true;
+    }
+
+    pub fn run(&mut self) -> Result<(), String> {
+        self.boot()?;
+
+        let mut time = Instant::now();
+
+        'emu: loop {
+            // Update UI thread (roughly every 512k cycles targeting 60 UPS)
+            if unlikely(self.internal_state.clock & 0x7FFFF == 0) {
+                if self.shared_state.update_req.load(Ordering::Relaxed) {
+                    self.update();
+                }
+            }
+
+            // Check for commands or hardware Interrupts every 32k cycles
+            if self.internal_state.clock % 0x7FFF == 0 {
+                if self.shared_state.running.load(Ordering::Relaxed) == false {
+                    self.idle_wait();
+                }
+
+                match self.interrupts.try_recv() {
+                    Ok(code) => self.interrupt(Interrupt::Software(code)),
+                    Err(TryRecvError::Disconnected) => break 'emu,
+                    Err(TryRecvError::Empty) => {}
+                }
+            }
+
+            // if we got a halt instruction, wait
+            if unlikely(!self.internal_state.running) {
+                let mips = (self.internal_state.clock as u128 / time.elapsed().as_micros());
+                println!(
+                    "TIME TAKEN: {:?}, clock: {}, {}MIPS",
+                    time.elapsed(),
+                    self.internal_state.clock,
+                    mips
+                );
+                time = Instant::now();
+
+                // temporary while i figure out a better solution
+                // exit when we hit halt (not useful for a real os ofc)
+                break 'emu;
+
+                self.idle_wait();
+            }
+
+            let pc = self.reg(Register::Pcx);
+            let instruction = self.mem_read_word(pc);
+
+            if cfg!(debug_assertions) {
+                println!(
+                    "Clock: {} Executing {:?} PCX: {}, reg: {:?}",
+                    self.internal_state.clock,
+                    Instruction(instruction),
+                    pc,
+                    self.internal_state.registers
+                );
+                thread::sleep(Duration::from_micros(10));
+            }
+
+            self.mut_reg(Register::Pcx).add_assign(4);
+            self.execute(Instruction(instruction));
+
+            // Check if executing the interrupt caused a fault.
+            if let Some(fault) = self.pending_fault {
+                self.pending_fault = None;
+                println!("WARN fault: {:?}", fault);
+                self.interrupt(fault);
+            }
+
+            // always increment clock.
+            self.internal_state.clock += 1;
+        }
+
+        self.shutdown();
+
+        Ok(())
+    }
+
+    #[inline]
+    fn interrupt(&mut self, int: Interrupt) {
+        let idt = self.reg(Register::Idr);
+
+        *self.mut_reg(Register::Spr) -= 4;
+        let spr = self.reg(Register::Spr);
+        let pcx = self.reg(Register::Pcx);
+        self.mem_write_word(spr, pcx);
+
+        *self.mut_reg(Register::Pcx) = self.mem_read_word(idt + int.code() as u32 * 4)
+    }
+
+    #[inline]
+    fn execute(&mut self, word: Instruction) {
+        // This needs to be unsafe as we're using word.xxxx_uc() functions for decoding which are unsafe
+        // as they perform unchecked transmute operations (performance critical)
+        unsafe {
+            match Opcode::from_u8(word.opcode()) {
+                // Nothing
+                Some(Opcode::Nop) => {}
+
+                // move
+                Some(Opcode::Mov) => {
+                    *self.mut_reg(word.dest_uc()) = self.reg(word.src1_uc());
+                }
+                Some(Opcode::CMov) => {
+                    if self.reg(word.misc_uc()) != 0 {
+                        *self.mut_reg(word.dest_uc()) = self.reg(word.src1_uc());
+                    }
+                }
+
+                // load
+                Some(Opcode::Ldbs) => todo!(),
+                Some(Opcode::Ldhs) => todo!(),
+                Some(Opcode::Ldb) => {
+                    *self.mut_reg(word.dest_uc()) = u32::from(
+                        self.mem_read_byte(self.reg(word.src1_uc()) + word.imm16() as u32),
+                    )
+                }
+                Some(Opcode::Ldh) => {
+                    *self.mut_reg(word.dest_uc()) = u32::from(
+                        self.mem_read_word(self.reg(word.src1_uc()) + word.imm16() as u32) >> 16,
+                    )
+                }
+                Some(Opcode::Ldw) => {
+                    *self.mut_reg(word.dest_uc()) = u32::from(
+                        self.mem_read_word(self.reg(word.src1_uc()) + word.imm16() as u32),
+                    )
+                }
+
+                // store
+                Some(Opcode::Stb) => {
+                    self.mem_write_byte(
+                        self.reg(word.dest_uc()) + word.imm16() as u32,
+                        self.reg(word.src1_uc()) as u8,
+                    );
+                }
+                Some(Opcode::Sth) => {
+                    self.mem_write_byte(
+                        self.reg(word.dest_uc()) + word.imm16() as u32,
+                        (self.reg(word.src1_uc()) as u16 >> 8) as u8,
+                    );
+                    self.mem_write_byte(
+                        self.reg(word.dest_uc()) + word.imm16() as u32 + 1,
+                        self.reg(word.src1_uc()) as u8,
+                    );
+                }
+                Some(Opcode::Stw) => {
+                    self.mem_write_word(
+                        self.reg(word.dest_uc()) + word.imm16() as u32,
+                        self.reg(word.src1_uc()),
+                    );
+                }
+
+                // load immediate
+                Some(Opcode::Lli) => {
+                    *self.mut_reg(word.dest_uc()) = word.imm16() as u32;
+                }
+                Some(Opcode::Lui) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (word.imm16() as u32) << 16 | self.reg(word.dest_uc());
+                }
+
+                // Comparison
+                Some(Opcode::Ieq) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (self.reg(word.src1_uc()) == self.reg(word.src2_uc())) as u32;
+                }
+                Some(Opcode::Ine) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (self.reg(word.src1_uc()) != self.reg(word.src2_uc())) as u32;
+                }
+                Some(Opcode::Ilt) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (self.reg(word.src1_uc()) < self.reg(word.src2_uc())) as u32;
+                }
+                Some(Opcode::Ile) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (self.reg(word.src1_uc()) <= self.reg(word.src2_uc())) as u32;
+                }
+                Some(Opcode::Igt) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (self.reg(word.src1_uc()) > self.reg(word.src2_uc())) as u32;
+                }
+                Some(Opcode::Ige) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        (self.reg(word.src1_uc()) >= self.reg(word.src2_uc())) as u32;
+                }
+
+                // Jump
+                Some(Opcode::Jmp) => {
+                    *self.mut_reg(Register::Pcx) = self.reg(word.dest_uc()) + word.imm16() as u32
+                }
+
+                Some(Opcode::Jez) => {
+                    if self.reg(word.src1_uc()) == 0 {
+                        *self.mut_reg(Register::Pcx) =
+                            self.reg(word.dest_uc()) + word.imm16() as u32
+                    }
+                }
+                Some(Opcode::Jnz) => {
+                    if self.reg(word.src1_uc()) != 0 {
+                        *self.mut_reg(Register::Pcx) =
+                            self.reg(word.dest_uc()) + word.imm16() as u32
+                    }
+                }
+                Some(Opcode::Jnc) => todo!(),
+                Some(Opcode::Jic) => todo!(),
+
+                // Bitwise
+                Some(Opcode::And) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        self.reg(word.src1_uc()) & self.reg(word.src2_uc());
+                }
+                Some(Opcode::Nand) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        !(self.reg(word.src1_uc()) & self.reg(word.src2_uc()));
+                }
+                Some(Opcode::Or) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        self.reg(word.src1_uc()) | self.reg(word.src2_uc());
+                }
+                Some(Opcode::Nor) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        !(self.reg(word.src1_uc()) | self.reg(word.src2_uc()));
+                }
+                Some(Opcode::Xor) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        self.reg(word.src1_uc()) ^ self.reg(word.src2_uc());
+                }
+                Some(Opcode::Xnor) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        !(self.reg(word.src1_uc()) ^ self.reg(word.src2_uc()));
+                }
+                Some(Opcode::Not) => {
+                    *self.mut_reg(word.dest_uc()) = !self.reg(word.src1_uc());
+                }
+
+                // Arithmetic
+                Some(Opcode::Add) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        self.reg(word.src1_uc()) + self.reg(word.src2_uc());
+                }
+
+                Some(Opcode::Sub) => {
+                    *self.mut_reg(word.dest_uc()) =
+                        self.reg(word.src1_uc()) - self.reg(word.src2_uc());
+                }
+
+                Some(Opcode::Shl) => {
+                    *self.mut_reg(word.dest_uc()) = self.reg(word.src1_uc())
+                        << (self.reg(word.src2_uc()) + word.shamt() as u32);
+                }
+
+                Some(Opcode::Shr) => {
+                    *self.mut_reg(word.dest_uc()) = self.reg(word.src1_uc())
+                        >> (self.reg(word.src2_uc()) + word.shamt() as u32);
+                }
+
+                Some(Opcode::Addi) => {
+                    *self.mut_reg(word.dest_uc()) = self.reg(word.src1_uc()) + word.imm16() as u32;
+                }
+
+                Some(Opcode::Subi) => {
+                    *self.mut_reg(word.dest_uc()) = self.reg(word.src1_uc()) - word.imm16() as u32;
+                }
+
+                // Utility
+                Some(Opcode::Push) => {
+                    self.push(word.src1_uc());
+                }
+                Some(Opcode::Pop) => {
+                    self.pop(word.dest_uc());
+                }
+
+                // Function
+                Some(Opcode::Call) => {
+                    self.push(Register::Pcx);
+                    *self.mut_reg(Register::Pcx) = self.reg(word.dest_uc()) + word.imm16() as u32
+                }
+                Some(Opcode::Ret) => {
+                    self.pop(Register::Ret);
+                    *self.mut_reg(Register::Pcx) = self.reg(Register::Ret);
+                }
+
+                Some(Opcode::Int) => {
+                    self.interrupt(Interrupt::Software(word.imm16() as u8));
+                }
+                Some(Opcode::Hlt) => self.internal_state.running = false,
+                Some(Opcode::IRet) => todo!(),
+                None => {}
+            }
+        }
+    }
+
+    #[inline]
+    fn push(&mut self, reg: Register) {
+        *self.mut_reg(Register::Spr) -= 4;
+        self.mem_write_word(self.reg(Register::Spr), self.reg(reg));
+    }
+
+    #[inline]
+    fn pop(&mut self, reg: Register) {
+        *self.mut_reg(reg) = self.mem_read_word(self.reg(Register::Spr));
+        *self.mut_reg(Register::Spr) += 4;
+    }
+
+    #[inline]
+    fn mem_read_byte(&mut self, addr: u32) -> u8 {
+        if unlikely(self.is_mmio(addr)) {
+            return self.io_read_byte(addr);
+        }
+        self.mainstore.read_byte(addr)
+    }
+
+    #[inline]
+    fn mem_write_byte(&mut self, addr: u32, val: u8) {
+        if unlikely(self.is_mmio(addr)) {
+            self.io_write_byte(addr, val);
+            return;
+        }
+        self.mainstore.write_byte(addr, val);
+    }
+
+    #[inline]
+    fn mem_read_word(&mut self, addr: u32) -> u32 {
+        if unlikely(self.is_mmio(addr)) {
+            return self.io_read_word(addr);
+        }
+        self.mainstore.read_word(addr)
+    }
+
+    #[inline]
+    fn mem_write_word(&mut self, addr: u32, val: u32) {
+        if unlikely(self.is_mmio(addr)) {
+            self.io_write_word(addr, val);
+            return;
+        }
+        self.mainstore.write_word(addr, val);
+    }
+
+    #[inline]
+    fn mem_read_page(&mut self, addr: u32) -> &Page {
+        if unlikely(self.is_mmio(addr)) {
+            // pages spanning MMIO don't really make sense
+            // treat as fault
+            self.pending_fault = Some(Interrupt::ProtectionFault);
+            return &Page::ZERO;
+        }
+        self.mainstore.read_page(addr)
+    }
+
+    #[inline]
+    fn mem_write_page(&mut self, addr: u32, val: &Page) {
+        if unlikely(self.is_mmio(addr)) {
+            self.pending_fault = Some(Interrupt::ProtectionFault);
+            return;
+        }
+        self.mainstore.write_page(addr, val);
+    }
+
+    // single MMIO check reused by all of the above
+    #[inline]
+    fn is_mmio(&self, addr: u32) -> bool {
+        self.mmio_region
+            .map(|(base, end)| addr >= base && addr < end)
+            .unwrap_or(false)
+    }
+
+    fn fault(&mut self, interrupt: Interrupt) {
+        self.pending_fault = Some(interrupt);
+    }
+
+    #[inline]
+    fn get_device(&mut self, addr: u32) -> Option<(&mut Box<dyn IoDevice>, u32)> {
+        // Find the device that covers the address.
+        self.mmio
+            .iter_mut()
+            .find(|d| d.base <= addr && addr < d.base + d.device.size())
+            .map(|d| (&mut d.device, addr - d.base))
+    }
+    // --- cold IO paths ---
+
+    #[cold]
+    fn io_read_byte(&mut self, addr: u32) -> u8 {
+        if let Some((device, offset)) = self.get_device(addr) {
+            device.read_byte(offset).unwrap_or_else(|fault| {
+                self.fault(fault);
+                0
+            })
+        } else {
+            self.fault(Interrupt::UnmappedIo);
+            0
+        }
+    }
+
+    #[cold]
+    fn io_read_word(&mut self, addr: u32) -> u32 {
+        if let Some((device, offset)) = self.get_device(addr) {
+            device.read_word(offset).unwrap_or_else(|fault| {
+                self.fault(fault);
+                0
+            })
+        } else {
+            self.fault(Interrupt::UnmappedIo);
+
+            0
+        }
+    }
+
+    #[cold]
+    fn io_write_byte(&mut self, addr: u32, val: u8) {
+        if let Some((dev, offset)) = self.get_device(addr) {
+            dev.write_byte(offset, val).unwrap_or_else(|fault| {
+                self.fault(fault);
+            });
+        } else {
+            self.fault(Interrupt::UnmappedIo);
+        }
+    }
+
+    #[cold]
+    fn io_write_word(&mut self, addr: u32, val: u32) {
+        if let Some((dev, offset)) = self.get_device(addr) {
+            dev.write_word(offset, val).unwrap_or_else(|fault| {
+                self.fault(fault);
+            });
+        } else {
+            self.fault(Interrupt::UnmappedIo);
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct ProcessorSnapshot {
+    pub running: bool,
+    pub clock: usize,
+    pub registers: [u32; Self::REG_COUNT],
+}
+
+impl Default for ProcessorSnapshot {
+    fn default() -> Self {
+        Self {
+            running: false,
+            clock: 0,
+            registers: [0; Self::REG_COUNT],
+        }
+    }
+}
+
+impl ProcessorSnapshot {
+    const REG_COUNT: usize = 28;
+}

dsa/emulator/src/processor/state.rs

@@ -0,0 +1,25 @@
+use std::sync::{Arc, atomic::AtomicBool, mpsc};
+
+use crate::processor::processor::ProcessorSnapshot;
+use arc_swap::ArcSwap;
+
+pub struct SharedState {
+    pub proc: ArcSwap<ProcessorSnapshot>,
+
+    pub running: AtomicBool,
+    pub update_req: AtomicBool,
+
+    sender: mpsc::Sender<u8>,
+}
+
+impl SharedState {
+    pub fn new(sender: mpsc::Sender<u8>) -> Self {
+        Self {
+            proc: ArcSwap::new(Arc::new(ProcessorSnapshot::default())),
+
+            running: AtomicBool::new(true),
+            sender: sender,
+            update_req: AtomicBool::new(false),
+        }
+    }
+}