tl.composite: fused epilogue ops with per-op scope

Extend tl.composite() with an ordered epilogue list. Each op carries
a scope flag - output_tile (default, runs once per (m,n) before
STORE), k_tile (every K-tile right after GEMM), or kernel. Plan
generator slots MATH stages by scope; pe_math reuses pe_dma's
local-loop pattern so chained epilogues (bias->relu) skip the port
hop. op_log captures per-stage params for telemetry. Topology
gains a gemm->math edge (snapshot test updated).

API stays backward-compatible - `epilogue=` is opt-in.

Example:
    h = tl.composite(
        op="gemm", a=a, b=b, out_ptr=int(out),
        epilogue=[
            {"op": "dequant", "scale": s_per_k, "scope": "k_tile"},
            {"op": "bias",    "bias":  bias_vec},
            {"op": "relu"},
            {"op": "scale",   "factor": 0.5},
        ],
    )
    tl.wait(h)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-05-15 10:16:47 -07:00
parent a76487ca48
commit a7fe785e5f
12 changed files with 382 additions and 20 deletions
+3 -1
View File
@@ -31,7 +31,9 @@ def test_full_graph_edge_count():
# ADR-0023: +3 IPCQ edges per PE
# ADR-0019 D1 (restored): HBM↔router edges drop from 32 routers × 2
# to 8 PE-routers × 2 per cube. 32 cubes × (16-64) = -1536 edges.
assert len(g.edges) == 12156
# Multi-op composite (ADR-0021): +1 gemm→math edge per PE for
# epilogue chaining = 2 SIPs × 16 cubes × 8 PEs = +256 edges.
assert len(g.edges) == 12412
# -- Full graph: specific nodes exist -----------------------------------------