Skip to content

Overview — what this project is

Multicore-OoO-sim is a from-scratch rebuild of three of my course assignments into one cohesive multi-core, out-of-order, cache-coherent computer-architecture simulator. It exists for two reasons:

  1. Personal portfolio. As a single integrated piece, it reads on GitHub as one piece of engineering, not three homework dumps. Anyone landing on the repo sees one binary, one config format, one trace format, one test suite.
  2. Research scaffold. Once the OoO core and coherence layer land, the simulator becomes a usable platform for my own micro-architecture experiments — branch-predictor variants, prefetcher ablations, coherence-protocol scaling studies — without me having to glue together other peoples' simulators every time.

The architecture in one diagram

                 +-----------------------------+
                 |        Coherent Network     |   <- from project3 (RING / XBAR)
                 |  (directory + interconnect) |
                 +--+-------+-------+-------+--+
                    |       |       |       |
                  Core 0  Core 1  Core 2  Core N-1
                  +----+ +----+ +----+ +----+
                  |OoO | |OoO | |OoO | |OoO |   <- from project2 (Tomasulo + branch pred)
                  | L1 | | L1 | | L1 | | L1 |   <- from project1 (L1)
                  | L2 | | L2 | | L2 | | L2 |   <- from project1 (L2)
                  +----+ +----+ +----+ +----+
                            |
                          DRAM

Each core is a full out-of-order pipeline (fetch → branch predict → dispatch → schedule → execute → writeback → retire), each core has a real L1/L2 cache, and the L1s on different cores stay coherent through a directory-based protocol.

What's source projects vs. new

Source project Kept Dropped
project1 (L1+L2 cache) geometry math, LRU/LIP/MIP, prefetchers (+1, Markov, Hybrid), AAT calc global C-style arrays, course driver, course traces
project2 (OoO + branch) branch predictors, ROB / dispatch / schedQ / FU pipeline pre-baked icache_hit/dcache_hit in trace, course driver
project3 (coherence) event loop, Network (RING/XBAR), Node, agents, directory one-op-per-cycle CPU, course scaffolding files
Everywhere — Docker scripts, validate_* targets, GT/course attribution

The course driver code is gone in every case — the simulator has a single new main.cpp that wires everything together via JSON config and CLI flags.

Phased roadmap

The project ships in self-contained phases, each ending with a runnable, testable thing.

Phase What lands Status This report
0 Skeleton, build system, CLI, JSON config, logging, CI ✅ done 01-phase0-skeleton.md
1 ChampSim trace format, reader/writer, synthetic generator, tracer plan partial 02-phase1-traces.md
2 Cache subsystem (--mode cache) ✅ done 03-phase2-cache.md
3 Branch predictors (--mode predictor) ✅ done 04-phase3-predictor.md
4 OoO core (--mode ooo, single-core full) ✅ done 06-phase4-review.md (review + applied fixes)
5 Multi-core + coherence (--mode full / coherence) pending —
6 Polish, results plots, public release pending —

Phase 1 is "partial" because the ChampSim reader/writer and the synthetic gen_trace are done, but the DynamoRIO-based real-workload tracer is deferred — Phase 3 only needs synthetic and project2-derived traces, which we already have.

What's runnable today

# Build
cmake --preset default
cmake --build --preset default

# Run cache mode on a fixture trace
./build/default/sim --mode cache \
    --config configs/baseline.json \
    --trace tests/cache/fixtures/proj1/short_gcc.trace

# Run predictor mode
./build/default/sim --mode predictor \
    --config configs/baseline.json \
    --trace tests/predictor/fixtures/proj2/branchsim.champsimtrace

# Test suite
ctest --preset default

86 tests, all green. Cache numbers match project1 reference output bit-for-bit; predictor numbers match project2 reference output bit-for-bit (see Phase 2 / Phase 3 reports for the regression methodology). Phase-4 fixes are pinned by their own targeted tests (06-phase4-review.md).

How to read the rest of this report

If you want a quick skim:

  • Read 00-overview.md (this file) for the big picture.
  • Read 05-tools-and-libraries.md if you want to understand the toolchain — what CMake / Catch2 / nlohmann/json / CLI11 / ChampSim / DynamoRIO / Pin actually do. Written for someone who knows architecture but hasn't worked with these particular tools.
  • Read the phase-N report (e.g. 04-phase3-predictor.md) for the details of one specific phase.
  • Read code-review.md for the senior-level code review I did on top of Phases 0–3.

If you're picking up where Phase 3 left off and just want to start Phase 4, the relevant section is "Recommended fixes, in priority order" at the bottom of code-review.md.