Implement ADR-0021: PE pipeline refactor with token self-routing

Step 1-2: Backup existing code
- builtin/ → builtin_legacy/ (unchanged backup)
- custom/pe_accel/ → custom/pe_accel_legacy/ (unchanged backup)

Step 3-4: New pipeline types and tiling
- pe_types.py: StageType, Stage, TilePlan, PipelinePlan, PipelineContext, TileToken
- tiling.py: generate_gemm_plan, generate_math_plan (ported from pe_accel)

Step 5: Component implementations (ADR-0021 D4-D6)
- PE_SCHEDULER: _feed_loop (singleton FIFO feeder) + plan generation
- PE_FETCH_STORE: new component — TCM ↔ Register File
- PE_GEMM: TileToken pipeline + legacy PeInternalTxn dual-mode
- PE_MATH: TileToken pipeline + legacy dual-mode
- PE_DMA: TileToken pipeline + legacy + fabric Transaction triple-mode
- PE_TCM: TcmRequest handler with dual-channel BW serialization

Step 6: Infrastructure
- topology.yaml: pe_fetch_store component + chaining edges
- components.yaml: pe_fetch_store_v1 registration
- builder.py: PE_COMP_OFFSETS, _add_pe_internal_edges, PE view positions
- Tests: node/edge counts, PE component sets updated

All components handle both TileToken (pipeline) and PeInternalTxn (legacy).
Token self-routing: components read next stage from token.plan, chain via out_port.
366 tests passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/kernbench/topology/builder.py

@@ -20,6 +20,7 @@ _PE_COMP_OFFSETS = {
     "pe_cpu": (-0.3, 0.0),
     "pe_scheduler": (-0.15, 0.0),
     "pe_dma": (0.0, -0.15),
+    "pe_fetch_store": (0.15, 0.0),
     "pe_gemm": (0.0, 0.0),
     "pe_math": (0.0, 0.15),
     "pe_mmu": (0.15, -0.15),
@@ -637,12 +638,13 @@ def _instantiate_cube(
 
 
 def _add_pe_internal_edges(edges: list[Edge], pp: str, pe_links: dict) -> None:
-    """Add PE-internal edges for a single PE instance."""
+    """Add PE-internal edges for a single PE instance (ADR-0021)."""
     edges.append(Edge(
         src=f"{pp}.pe_cpu", dst=f"{pp}.pe_scheduler",
         distance_mm=pe_links["pe_cpu_to_scheduler_mm"],
         kind="pe_internal",
     ))
+    # Scheduler → engines (initial dispatch)
     for eng, key in [("pe_dma", "scheduler_to_dma_mm"),
                      ("pe_gemm", "scheduler_to_gemm_mm"),
                      ("pe_math", "scheduler_to_math_mm")]:
@@ -651,6 +653,15 @@ def _add_pe_internal_edges(edges: list[Edge], pp: str, pe_links: dict) -> None:
             distance_mm=pe_links[key],
             kind="pe_internal",
         ))
+    # Scheduler → fetch_store (initial dispatch)
+    if "scheduler_to_fetch_store_mm" in pe_links:
+        edges.append(Edge(
+            src=f"{pp}.pe_scheduler", dst=f"{pp}.pe_fetch_store",
+            distance_mm=pe_links["scheduler_to_fetch_store_mm"],
+            kind="pe_internal",
+        ))
+
+    # Engine → TCM (legacy BW edges)
     for eng, mm_key, bw_key in [("pe_dma", "dma_to_tcm_mm", "dma_to_tcm_bw_gbs"),
                                  ("pe_gemm", "gemm_to_tcm_mm", "gemm_to_tcm_bw_gbs"),
                                  ("pe_math", "math_to_tcm_mm", "math_to_tcm_bw_gbs")]:
@@ -661,6 +672,32 @@ def _add_pe_internal_edges(edges: list[Edge], pp: str, pe_links: dict) -> None:
             kind="pe_internal",
         ))
 
+    # Fetch/Store → TCM (ADR-0021 D5)
+    if "fetch_store_to_tcm_mm" in pe_links:
+        edges.append(Edge(
+            src=f"{pp}.pe_fetch_store", dst=f"{pp}.pe_tcm",
+            distance_mm=pe_links["fetch_store_to_tcm_mm"],
+            bw_gbs=pe_links.get("fetch_store_to_tcm_bw_gbs", 512.0),
+            kind="pe_internal",
+        ))
+
+    # Chaining edges (ADR-0021 D4 — token self-routing)
+    chaining = [
+        ("pe_dma", "pe_fetch_store", "dma_to_fetch_store_mm"),
+        ("pe_fetch_store", "pe_gemm", "fetch_store_to_gemm_mm"),
+        ("pe_fetch_store", "pe_math", "fetch_store_to_math_mm"),
+        ("pe_gemm", "pe_fetch_store", "gemm_to_fetch_store_mm"),
+        ("pe_math", "pe_fetch_store", "math_to_fetch_store_mm"),
+        ("pe_fetch_store", "pe_dma", "fetch_store_to_dma_mm"),
+    ]
+    for src_eng, dst_eng, mm_key in chaining:
+        if mm_key in pe_links:
+            edges.append(Edge(
+                src=f"{pp}.{src_eng}", dst=f"{pp}.{dst_eng}",
+                distance_mm=pe_links[mm_key],
+                kind="pe_internal",
+            ))
+
 
 # ── Inter-cube / IO / system edges ──────────────────────────────────
 
@@ -1071,6 +1108,7 @@ def _build_pe_view(spec: dict) -> ViewGraph:
         "pe_cpu": (1.5, 4.0),
         "pe_scheduler": (4.0, 4.0),
         "pe_dma": (7.0, 1.5),
+        "pe_fetch_store": (8.5, 4.0),
         "pe_gemm": (7.0, 4.0),
         "pe_math": (7.0, 6.5),
         "pe_mmu": (4.0, 1.5),
@@ -1101,6 +1139,12 @@ def _build_pe_view(spec: dict) -> ViewGraph:
             distance_mm=pe_links[key],
             kind="pe_internal",
         ))
+    if "scheduler_to_fetch_store_mm" in pe_links:
+        view_edges.append(Edge(
+            src="pe_scheduler", dst="pe_fetch_store",
+            distance_mm=pe_links["scheduler_to_fetch_store_mm"],
+            kind="pe_internal",
+        ))
     for eng, mm_key, bw_key in [("pe_dma", "dma_to_tcm_mm", "dma_to_tcm_bw_gbs"),
                                  ("pe_gemm", "gemm_to_tcm_mm", "gemm_to_tcm_bw_gbs"),
                                  ("pe_math", "math_to_tcm_mm", "math_to_tcm_bw_gbs")]:
@@ -1110,6 +1154,13 @@ def _build_pe_view(spec: dict) -> ViewGraph:
             bw_gbs=pe_links[bw_key],
             kind="pe_internal",
         ))
+    if "fetch_store_to_tcm_mm" in pe_links:
+        view_edges.append(Edge(
+            src="pe_fetch_store", dst="pe_tcm",
+            distance_mm=pe_links["fetch_store_to_tcm_mm"],
+            bw_gbs=pe_links.get("fetch_store_to_tcm_bw_gbs", 512.0),
+            kind="pe_internal",
+        ))
 
     return ViewGraph(
         name="pe", nodes=nodes, edges=view_edges,