TOGSim C++ trace-generation pipeline (P0-P3): explicit dataflow producer + barriers#267

Open

YWHyuk wants to merge 26 commits into

developfrom

feature/togsim-cpp-trace

Commits on Jun 25, 2026

[Docs] C++ trace pipeline design (runtime-tag pairing, ABI)

YWHyuk
and
claude
committed
[TOGSim] C++ trace pipeline: front end, runtime, loader, bridge, Core feed

YWHyuk
and
claude
committed
[TOGSim] Per-iteration tag pairing for multi-tile-K and conv

YWHyuk
and
claude
committed
[TOGSim] Work-item outlining and ABI v12 dispatch

YWHyuk
and
claude
committed
[TOGSim] SRAM-capacity and SA weight-buffer throttle for the trace path

YWHyuk
and
claude
committed
[Tooling] TOGSim trace timeline (Perfetto) and the trace emits it needs

YWHyuk
and
claude
committed
[TOGSim] Make the C++ trace path the default and stabilize it

YWHyuk
and
claude
committed
[TOGSim] Make the trace runtime test self-contained

YWHyuk
and
claude
committed
[Frontend] Trace cache-safe replay and compile-race fixes

YWHyuk
and
claude
committed
[TOGSim] Redesign trace-bridge dependency, barrier, SRAM-version, and runtime model

YWHyuk
and
claude
committed
[Frontend] Run the spad-overflow check in timing-only mode, budget at spad/2

YWHyuk
and
claude
committed
[Frontend] Generate the trace.cpp ABI/API banner from togsim_runtime.h

YWHyuk
and
claude
committed
[TOGSim] Pick 1- vs 2-dispatch concurrency by per-dispatch spad footprint

YWHyuk
and
claude
committed
[Frontend] Budget fused-epilogue spad buffers honestly in GEMM tile selection

YWHyuk
and
claude
committed
[Frontend] Stop charging the kernel stack frame against the spad budget

YWHyuk
and
claude
committed
[CI] Bump spike pin to v1.0.3
YWHyuk
committed
[Frontend] Add per-kernel CPU functional verify sub-option

YWHyuk
and
claude
committed
[Docs] Document per-kernel functional verify

YWHyuk
and
claude
committed
[TOGSim] Drop the ABI version changelog comment
YWHyuk
committed
[Frontend] Budget fused-prologue spad buffers in BMM tile selection

YWHyuk
and
claude
committed
[Frontend] Condense BMM prologue spad-budget comments to one line

YWHyuk
and
claude
committed
Merge pull request #278 from PSAL-POSTECH/fix/bmm-prologue-spad-budget
YWHyuk
authored

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

TOGSim C++ trace-generation pipeline (P0-P3): explicit dataflow producer + barriers#267

TOGSim C++ trace-generation pipeline (P0-P3): explicit dataflow producer + barriers#267
YWHyuk wants to merge 26 commits into
developfrom
feature/togsim-cpp-trace

Commits on Jun 25, 2026

[Docs] C++ trace pipeline design (runtime-tag pairing, ABI)

[TOGSim] C++ trace pipeline: front end, runtime, loader, bridge, Core feed

[TOGSim] Per-iteration tag pairing for multi-tile-K and conv

[TOGSim] Work-item outlining and ABI v12 dispatch

[TOGSim] SRAM-capacity and SA weight-buffer throttle for the trace path

[Tooling] TOGSim trace timeline (Perfetto) and the trace emits it needs

[TOGSim] Make the C++ trace path the default and stabilize it

[TOGSim] Make the trace runtime test self-contained

[Frontend] Trace cache-safe replay and compile-race fixes

[TOGSim] Redesign trace-bridge dependency, barrier, SRAM-version, and runtime model

[Frontend] Run the spad-overflow check in timing-only mode, budget at spad/2

[Frontend] Generate the trace.cpp ABI/API banner from togsim_runtime.h

[TOGSim] Pick 1- vs 2-dispatch concurrency by per-dispatch spad footprint

[Frontend] Budget fused-epilogue spad buffers honestly in GEMM tile selection

[Frontend] Stop charging the kernel stack frame against the spad budget

[CI] Bump spike pin to v1.0.3

[Frontend] Add per-kernel CPU functional verify sub-option

[Docs] Document per-kernel functional verify

[TOGSim] Drop the ABI version changelog comment

[Frontend] Budget fused-prologue spad buffers in BMM tile selection

[Frontend] Condense BMM prologue spad-budget comments to one line

Merge pull request #278 from PSAL-POSTECH/fix/bmm-prologue-spad-budget

Commits on Jun 26, 2026

[TOGSim] Skip per-instruction trace string build when trace logging is off

[TOGSim] Skip the per-cycle issue scan when nothing can newly issue

[TOGSim] Fix use-after-erase in the zero-cycle COMP skip path

[Frontend] Make the C++ trace the sole main TOG path; drop legacy ONNX TOG