gqa(prefill-4cases): add long-context prefill 4-cases comparative study

Mirrors the existing decode 4-cases study for prefill on the LLaMA-3.1-70B
single-KV-head group (h_q=8, h_kv=1, d_head=128) across 8 cubes × 8 PEs:

  Case 1  Cube-SP × PE-TP  → KV S_kv-Ring across cubes; Q/O T_q-split 64-way
  Case 2  Cube-Repl × PE-TP → full KV per cube; only CUBE 0's P PEs split T_q
  Case 3  Cube-Repl × PE-SP → full KV per cube; PEs SP on S_kv (intra-cube AR)
  Case 4  Cube-SP × PE-SP  → KV split 64-way + Q T_q-split across cubes
                              (Ring KV + per-cube intra-CUBE AR)  ★ optimal

Cases 2 and 3 use Q-axis tiling (TILE_Q) so the per-PE scratch is bounded
by TILE_Q × TILE_S_KV regardless of T_q (the full Q tensor would be
G·T_q·d_head·2 bytes which scales linearly with T_q and blew the 1 MB
budget at T_q≥128 without tiling).

Per-panel try/except in run_sweep tolerates per-panel failures so
sweep.json always lands with whichever cases succeed plus a failures
list. Case 4 uses configure_sfr_intercube_ring with the snake submesh
(2,4) which installs both the Ring E/W lanes and the intra_* lanes
needed for the intra-CUBE reduce — single SFR config for all 4 cases.

The plot script generates 4 PNGs (latency, traffic, memory, parallelism)
into 1H_milestone_output/gqa/long_ctx/. The parallelism chart shows
total compute work (active_PE × T_q_per_PE × S_kv_processed) — exposes
Case 3's 8× redundancy vs. Cases 1/2/4 which all do unique work.

gqa_prefill_long_ctx_4cases.py exposes run_sweep() rather than a @bench
decorator — the umbrella bench in the next commit invokes it.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

scripts/paper/paper_plot_gqa_prefill_long_ctx_4cases.py

@@ -0,0 +1,253 @@
+"""Comparative figures for milestone-gqa-prefill-long-ctx-4cases.
+
+Mirror of paper_plot_gqa_decode_long_ctx_4cases but for the prefill
+variant. Reads sweep.json (emitted by ``kernbench run --bench
+milestone-gqa-prefill-long-ctx-4cases``) and writes four PNGs into
+``docs/report/1H-codesign-paper/figures/``:
+
+  gqa_prefill_long_ctx_4cases_latency.png       end-to-end latency per case
+  gqa_prefill_long_ctx_4cases_traffic.png       ipcq/dma op-count breakdown
+  gqa_prefill_long_ctx_4cases_memory.png        per-PE KV bytes per case
+  gqa_prefill_long_ctx_4cases_parallelism.png   active-PE × S_local work load
+
+Run (after the bench):
+    GQA_PREFILL_LONG_CTX_4CASES_RUN=1 python -m kernbench.cli.main run \\
+        --bench milestone-gqa-prefill-long-ctx-4cases --topology topology.yaml
+    python scripts/paper/paper_plot_gqa_prefill_long_ctx_4cases.py
+"""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import matplotlib
+
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt  # noqa: E402
+
+_REPO_ROOT = Path(__file__).resolve().parents[2]
+# Sweep JSON + PNGs live together under the bench output dir.
+_FIG_DIR = (
+    _REPO_ROOT / "src" / "kernbench" / "benches"
+    / "1H_milestone_output" / "gqa" / "long_ctx"
+)
+_SWEEP_JSON = _FIG_DIR / "sweep_prefill.json"
+
+# Panel name → (short label, case ordinal for left-to-right plot order).
+_CASE_INFO = {
+    "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_tp": (
+        "Case 1\nCube-SP × PE-TP", 1),
+    "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_tp": (
+        "Case 2\nCube-Repl × PE-TP", 2),
+    "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_sp": (
+        "Case 3\nCube-Repl × PE-SP", 3),
+    "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_sp": (
+        "Case 4 ★\nCube-SP × PE-SP", 4),
+}
+
+
+def _load() -> list[dict]:
+    return json.loads(_SWEEP_JSON.read_text())["rows"]
+
+
+def _sorted_by_case(rows: list[dict]) -> list[dict]:
+    return sorted(rows, key=lambda r: _CASE_INFO[r["panel"]][1])
+
+
+def _plot_latency(rows: list[dict]) -> Path:
+    rows = _sorted_by_case(rows)
+    labels = [_CASE_INFO[r["panel"]][0] for r in rows]
+    lat_us = [r["latency_ns"] / 1e3 for r in rows]
+    colors = ["#888", "#888", "#888", "#3b6ea5"]  # Case 4 highlighted
+    fig, ax = plt.subplots(figsize=(8.0, 4.5))
+    bars = ax.bar(labels, lat_us, color=colors, width=0.6)
+    ax.set_ylabel("end-to-end latency (µs)")
+    ax.set_title(
+        "Long-context prefill 4-cases — end-to-end latency per case\n"
+        "LLaMA-3.1-70B single-KV-head group (8 cubes × 8 PEs)"
+    )
+    ax.bar_label(bars, fmt="%.1f", padding=3, fontsize=9)
+    ax.grid(axis="y", ls=":", alpha=0.5)
+    ax.set_ylim(0, max(lat_us) * 1.15)
+    fig.tight_layout()
+    out = _FIG_DIR / "gqa_prefill_long_ctx_4cases_latency.png"
+    fig.savefig(out, dpi=150)
+    plt.close(fig)
+    return out
+
+
+def _plot_traffic(rows: list[dict]) -> Path:
+    rows = _sorted_by_case(rows)
+    labels = [_CASE_INFO[r["panel"]][0] for r in rows]
+    x = list(range(len(rows)))
+    keys = ["ipcq_copy_count", "dma_read_count", "dma_write_count"]
+    disp = ["IPCQ copy", "DMA read", "DMA write"]
+    colors = ["#c0504d", "#9bbb59", "#8064a2"]
+    w = 0.25
+    fig, ax = plt.subplots(figsize=(9.0, 4.5))
+    for i, (k, d, c) in enumerate(zip(keys, disp, colors)):
+        vals = [r["op_log_summary"][k] for r in rows]
+        ax.bar([xi + (i - 1) * w for xi in x], vals, width=w, label=d, color=c)
+    ax.set_xticks(list(x))
+    ax.set_xticklabels(labels, fontsize=9)
+    ax.set_ylabel("op count")
+    ax.set_title("Long-context prefill 4-cases — op-count breakdown per case")
+    ax.legend(fontsize=9)
+    ax.grid(axis="y", ls=":", alpha=0.5)
+    fig.tight_layout()
+    out = _FIG_DIR / "gqa_prefill_long_ctx_4cases_traffic.png"
+    fig.savefig(out, dpi=150)
+    plt.close(fig)
+    return out
+
+
+def _s_local_per_pe(panel: str, *, S_kv: int, C: int, P: int) -> int:
+    """S_local (token count) each PE attends over locally.
+
+    Encodes the cube/pe sharding axes from the panel name:
+      cube_sp_pe_tp (Case 1): S_kv / C       (pe=replicate within cube)
+      cube_repl_pe_tp (Case 2): S_kv         (full S_kv per active PE)
+      cube_repl_pe_sp (Case 3): S_kv / P     (pe=row_wise within cube)
+      cube_sp_pe_sp (Case 4): S_kv / (C·P)   (★ 64-way split)
+    """
+    S_per_cube = S_kv if "cube_repl" in panel else S_kv // C
+    return S_per_cube // P if "pe_sp" in panel else S_per_cube
+
+
+def _active_pe_count(panel: str, *, C: int, P: int) -> int:
+    """Number of PEs doing non-idle attention work.
+
+    Prefill T_q≫1 means PE-TP is *useful* (not wasted as in decode):
+      cube_sp_pe_tp (Case 1): C·P   (T_q sharded across all 64 ranks)
+      cube_repl_pe_tp (Case 2): P   (CUBE 0 only; T_q sharded across its P PEs)
+      cube_repl_pe_sp (Case 3): C·P (all PEs busy but cubes redundant)
+      cube_sp_pe_sp (Case 4): C·P   (all 64 PEs doing unique work)
+    """
+    if "cube_repl" in panel and "pe_tp" in panel:
+        return P
+    return C * P
+
+
+def _kv_bytes_per_pe(panel: str, *, S_kv: int, h_kv: int,
+                     d_head: int, C: int, P: int) -> int:
+    """KV bytes a single PE references (K + V, f16, 2 B/elem)."""
+    s_local = _s_local_per_pe(panel, S_kv=S_kv, C=C, P=P)
+    return 2 * s_local * h_kv * d_head * 2
+
+
+def _plot_memory(rows: list[dict]) -> Path:
+    """Per-PE KV bytes — Case 4 wins (64-way split)."""
+    rows = _sorted_by_case(rows)
+    labels = [_CASE_INFO[r["panel"]][0] for r in rows]
+    mib_per_pe = [
+        _kv_bytes_per_pe(
+            r["panel"], S_kv=r["S_kv"], h_kv=r["h_kv"],
+            d_head=r["d_head"], C=r["C"], P=r["P"],
+        ) / (1024 * 1024)
+        for r in rows
+    ]
+    colors = ["#888", "#c0504d", "#888", "#3b6ea5"]
+    fig, ax = plt.subplots(figsize=(8.0, 4.5))
+    bars = ax.bar(labels, mib_per_pe, color=colors, width=0.6)
+    ax.set_ylabel("KV bytes per PE (MiB, K + V, f16)")
+    ax.set_title(
+        "Long-context prefill 4-cases — KV memory per PE\n"
+        "(one KV-head group; per-layer, full S_kv state)"
+    )
+    ax.bar_label(bars, fmt="%.3f", padding=3, fontsize=9)
+    ax.grid(axis="y", ls=":", alpha=0.5)
+    ax.set_ylim(0, max(mib_per_pe) * 1.15)
+    fig.tight_layout()
+    out = _FIG_DIR / "gqa_prefill_long_ctx_4cases_memory.png"
+    fig.savefig(out, dpi=150)
+    plt.close(fig)
+    return out
+
+
+def _t_q_per_pe(panel: str, *, T_q: int, C: int, P: int) -> int:
+    """T_q row count each active PE computes attention for.
+
+      cube_sp_pe_tp (Case 1): T_q / (C·P)  (T_q sharded across all 64 ranks)
+      cube_repl_pe_tp (Case 2): T_q / P    (T_q sharded across CUBE 0's P PEs)
+      cube_repl_pe_sp (Case 3): T_q        (Q replicated on every PE)
+      cube_sp_pe_sp (Case 4): T_q / C      (Q sharded by cube, replicated within)
+    """
+    if "cube_sp" in panel and "pe_tp" in panel:
+        return T_q // (C * P)
+    if "cube_repl" in panel and "pe_tp" in panel:
+        return T_q // P
+    if "cube_repl" in panel and "pe_sp" in panel:
+        return T_q
+    return T_q // C  # cube_sp_pe_sp
+
+
+def _s_kv_processed_per_pe(panel: str, *, S_kv: int, C: int, P: int) -> int:
+    """S_kv tokens each PE actually processes attention over.
+
+    Differs from ``_s_local_per_pe`` (OWNED KV bytes): for cases with
+    a Ring (Case 1, 4) each PE sees C ring steps so processes more
+    tokens than it locally owns.
+
+      cube_sp_pe_tp (Case 1): S_kv          (Ring + pe=replicate within cube)
+      cube_repl_pe_tp (Case 2): S_kv        (full KV per PE)
+      cube_repl_pe_sp (Case 3): S_kv / P    (pe=row_wise; no Ring)
+      cube_sp_pe_sp (Case 4): S_kv / P      (Ring restores full S_kv/P per PE)
+    """
+    if "cube_sp" in panel and "pe_tp" in panel:
+        return S_kv
+    if "cube_repl" in panel and "pe_tp" in panel:
+        return S_kv
+    return S_kv // P  # both pe_sp cases
+
+
+def _plot_parallelism(rows: list[dict]) -> Path:
+    """Total compute work (PE × T_q × S_kv token-pairs) — exposes Case 3's
+    redundancy. Cases 1, 2, 4 all do the same total work (correct
+    attention over T_q × S_kv); Case 3 does C× more (cubes redundantly
+    repeat the same compute because K/V is cube-replicated).
+    """
+    rows = _sorted_by_case(rows)
+    labels = [_CASE_INFO[r["panel"]][0] for r in rows]
+    total_work = [
+        _active_pe_count(r["panel"], C=r["C"], P=r["P"])
+        * _t_q_per_pe(r["panel"], T_q=r["T_q"], C=r["C"], P=r["P"])
+        * _s_kv_processed_per_pe(
+            r["panel"], S_kv=r["S_kv"], C=r["C"], P=r["P"],
+        )
+        for r in rows
+    ]
+    colors = ["#888", "#888", "#c0504d", "#3b6ea5"]  # Case 3 red, Case 4 highlighted
+    fig, ax = plt.subplots(figsize=(8.0, 4.5))
+    bars = ax.bar(labels, total_work, color=colors, width=0.6)
+    ax.set_ylabel(
+        "total compute (PE × T_q × S_kv token-pairs; lower ⇒ less wasted work)"
+    )
+    ax.set_title(
+        "Long-context prefill 4-cases — total compute work across active PEs\n"
+        "(Case 3 replicates K/V across 8 cubes ⇒ 8× redundant compute)"
+    )
+    ax.bar_label(bars, fmt="%d", padding=3, fontsize=9)
+    ax.grid(axis="y", ls=":", alpha=0.5)
+    ax.set_ylim(0, max(total_work) * 1.15)
+    fig.tight_layout()
+    out = _FIG_DIR / "gqa_prefill_long_ctx_4cases_parallelism.png"
+    fig.savefig(out, dpi=150)
+    plt.close(fig)
+    return out
+
+
+def main() -> None:
+    rows = _load()
+    _FIG_DIR.mkdir(parents=True, exist_ok=True)
+    p1 = _plot_latency(rows)
+    p2 = _plot_traffic(rows)
+    p3 = _plot_memory(rows)
+    p4 = _plot_parallelism(rows)
+    print(f"wrote {p1}")
+    print(f"wrote {p2}")
+    print(f"wrote {p3}")
+    print(f"wrote {p4}")
+
+
+if __name__ == "__main__":
+    main()

src/kernbench/benches/gqa_helpers/long_ctx/_gqa_attention_prefill_long_ctx_cube_repl_pe_sp.py

@@ -0,0 +1,177 @@
+"""GQA prefill kernel — Case 3 (Cube-Repl × PE-SP) at single-KV-head group.
+
+Per GQA_full_deck.pptx slide 11 (prefill counterpart, T_q ≫ 1):
+  - K, V replicated across all 8 cubes (the 8× memory waste).
+  - PEs SP on S_kv inside each cube: each PE attends to its
+    ``S_local = S_kv / P`` slice.
+  - Q replicated on every rank (every cube does full T_q × S_local).
+  - Intra-CUBE 8-way reduce on the partial ``(m, ℓ, O)`` triple
+    per Q-tile (row chain along intra_W + col bridge along intra_N
+    over the 2×4 PE grid).
+  - NO inter-CUBE communication — every cube ends with the full
+    answer (8× redundant compute is the inherent cost of Case 3).
+  - Designated writer: only PE 0 of CUBE 0 stores ``O`` to avoid
+    8 redundant DMA writes.
+
+Q-axis tiling (vs decode Case 3): an outer Q-tile loop wraps the
+entire per-tile pipeline (Q load → bootstrap → S_kv-tile fold →
+intra-CUBE AR → store) in ``tl.scratch_scope``. Peak scratch is
+bounded by ``TILE_Q × TILE_S_KV`` instead of ``T_q × S_kv``, so the
+kernel runs at large T_q without exceeding the ~1 MB per-PE budget.
+Each Q-tile's (m, ℓ, O) is independent of other Q-tiles, so no
+state leaks across iterations.
+
+Tensor layout:
+  Q : (T_q, h_q · d_head) replicated on every rank; loaded per
+      Q-tile as ``(TILE_Q, d_head)`` slices (G is folded into the
+      T_q row axis, total ``G·T_q`` rows).
+  K : (S_kv, h_kv · d_head) with cube=replicate, pe=row_wise — each
+      PE owns its ``(S_local, h_kv·d_head)`` shard within its cube.
+  V : same as K.
+  O : (T_q, h_q · d_head) — only PE 0 of CUBE 0 stores, one
+      ``(TILE_Q, d_head)`` slice per outer Q-tile iteration.
+
+Topology / SFR:
+  - Requires intra_* lanes; ``configure_sfr_intercube_ring`` (with the
+    snake submesh) or ``configure_sfr_intercube_multisip`` both install
+    them.
+"""
+from __future__ import annotations
+
+
+TILE_S_KV = 64  # per-tile S_kv width (small to keep ``scores`` bounded).
+TILE_Q = 64     # per-tile Q row count (outer-loop tile size).
+
+
+def _merge_running(m_local, l_local, O_local, m_other, l_other, O_other, *, tl):
+    """Online-softmax merge of two partial ``(m, ℓ, O)`` triples."""
+    m_new = tl.maximum(m_local, m_other)
+    scale_old = tl.exp(m_local - m_new)
+    scale_new = tl.exp(m_other - m_new)
+    l_new = l_local * scale_old + l_other * scale_new
+    O_new = O_local * scale_old + O_other * scale_new
+    return m_new, l_new, O_new
+
+
+def gqa_attention_prefill_long_ctx_cube_repl_pe_sp_kernel(
+    q_ptr: int,
+    k_ptr: int,
+    v_ptr: int,
+    o_ptr: int,
+    T_q: int,
+    S_kv: int,
+    h_q: int,
+    h_kv: int,
+    d_head: int,
+    C: int,
+    P: int,
+    *,
+    tl,
+) -> None:
+    """Case-3 prefill: PE-SP S_kv split + Q-tiled intra-CUBE AR."""
+    G = h_q // h_kv
+    S_local = S_kv // P  # each PE owns S_kv/P (cube=replicate, pe=row_wise)
+    pe_id = tl.program_id(axis=0)
+    cube_id = tl.program_id(axis=1)
+
+    PE_GRID_COLS = 4
+    pe_col = pe_id % PE_GRID_COLS
+    pe_row = pe_id // PE_GRID_COLS
+    pe_cols_used = min(PE_GRID_COLS, P)
+    pe_rows_used = (P + PE_GRID_COLS - 1) // PE_GRID_COLS
+
+    KV_ROW_BYTES = d_head * 2  # f16
+    Q_ROW_BYTES = d_head * 2   # G is folded into the row axis, not d_head
+    total_q_rows = G * T_q
+    n_q_tiles = (total_q_rows + TILE_Q - 1) // TILE_Q
+    n_kv_tiles = (S_local + TILE_S_KV - 1) // TILE_S_KV
+
+    # ── Outer Q-tile loop ──
+    # Per Q-tile: load Q-slice, run full S_kv attention, intra-CUBE AR,
+    # store. All scratch is recycled at outer-scope exit.
+    for q_idx in range(n_q_tiles):
+        q_start = q_idx * TILE_Q
+        q_rows = min(TILE_Q, total_q_rows - q_start)
+        with tl.scratch_scope():
+            Q = tl.load(q_ptr + q_start * Q_ROW_BYTES,
+                        shape=(q_rows, d_head), dtype="f16")
+
+            # ── Bootstrap (S_kv tile 0) ──
+            # K_T, V, scores, exp_scores live in the OUTER (Q-tile)
+            # scope. They survive the inner S_kv-tile loop (each
+            # iteration's allocations are recycled) and are freed
+            # together when the outer scope exits.
+            tile_s0 = min(TILE_S_KV, S_local)
+            K_T = tl.load(k_ptr, shape=(d_head, tile_s0), dtype="f16")
+            V = tl.load(v_ptr, shape=(tile_s0, d_head), dtype="f16")
+            scores = tl.dot(Q, K_T)
+            m_local = tl.max(scores, axis=-1)
+            exp_scores = tl.exp(scores - m_local)
+            l_local = tl.sum(exp_scores, axis=-1)
+            O_local = tl.dot(exp_scores, V)
+
+            # ── S_kv tiles 1..N: fold via online-softmax merge ──
+            for tile_idx in range(1, n_kv_tiles):
+                tile_start = tile_idx * TILE_S_KV
+                tile_s = min(TILE_S_KV, S_local - tile_start)
+                with tl.scratch_scope():
+                    K_T_t = tl.load(k_ptr + tile_start * KV_ROW_BYTES,
+                                    shape=(d_head, tile_s), dtype="f16")
+                    V_t = tl.load(v_ptr + tile_start * KV_ROW_BYTES,
+                                  shape=(tile_s, d_head), dtype="f16")
+                    scores_t = tl.dot(Q, K_T_t)
+                    m_tile = tl.max(scores_t, axis=-1)
+                    exp_scores_t = tl.exp(scores_t - m_tile)
+                    l_tile = tl.sum(exp_scores_t, axis=-1)
+                    O_tile = tl.dot(exp_scores_t, V_t)
+                    m_new, l_new, O_new = _merge_running(
+                        m_local, l_local, O_local,
+                        m_tile, l_tile, O_tile, tl=tl,
+                    )
+                    tl.copy_to(m_local, m_new)
+                    tl.copy_to(l_local, l_new)
+                    tl.copy_to(O_local, O_new)
+
+            # ── Intra-CUBE 8-way reduce-to-PE0 (row chain + col bridge) ──
+            # (m_local, l_local, O_local) are still alive in the outer
+            # scope; the send/recv operate on them directly.
+            if pe_cols_used > 1:
+                if pe_col < pe_cols_used - 1:
+                    with tl.scratch_scope():
+                        m_other = tl.recv(dir="intra_E", shape=m_local.shape, dtype="f16")
+                        l_other = tl.recv(dir="intra_E", shape=l_local.shape, dtype="f16")
+                        O_other = tl.recv(dir="intra_E", shape=O_local.shape, dtype="f16")
+                        m_new, l_new, O_new = _merge_running(
+                            m_local, l_local, O_local,
+                            m_other, l_other, O_other, tl=tl,
+                        )
+                        tl.copy_to(m_local, m_new)
+                        tl.copy_to(l_local, l_new)
+                        tl.copy_to(O_local, O_new)
+                if pe_col > 0:
+                    tl.send(dir="intra_W", src=m_local)
+                    tl.send(dir="intra_W", src=l_local)
+                    tl.send(dir="intra_W", src=O_local)
+
+            if pe_col == 0 and pe_rows_used > 1:
+                if pe_row < pe_rows_used - 1:
+                    with tl.scratch_scope():
+                        m_other = tl.recv(dir="intra_S", shape=m_local.shape, dtype="f16")
+                        l_other = tl.recv(dir="intra_S", shape=l_local.shape, dtype="f16")
+                        O_other = tl.recv(dir="intra_S", shape=O_local.shape, dtype="f16")
+                        m_new, l_new, O_new = _merge_running(
+                            m_local, l_local, O_local,
+                            m_other, l_other, O_other, tl=tl,
+                        )
+                        tl.copy_to(m_local, m_new)
+                        tl.copy_to(l_local, l_new)
+                        tl.copy_to(O_local, O_new)
+                if pe_row > 0:
+                    tl.send(dir="intra_N", src=m_local)
+                    tl.send(dir="intra_N", src=l_local)
+                    tl.send(dir="intra_N", src=O_local)
+
+            # ── Store this Q-tile (designated writer: cube 0, PE 0) ──
+            if pe_id == 0 and cube_id == 0:
+                O_final = O_local / l_local
+                tl.store(o_ptr + q_start * Q_ROW_BYTES, O_final)

src/kernbench/benches/gqa_helpers/long_ctx/_gqa_attention_prefill_long_ctx_cube_repl_pe_tp.py

@@ -0,0 +1,128 @@
+"""GQA prefill kernel — Case 2 (Cube-Repl × PE-TP) at single-KV-head group.
+
+Per GQA_full_deck.pptx slide 11 (prefill counterpart, T_q ≫ 1):
+  - K, V replicated across all 8 cubes (the 8× memory waste — this is
+    the inherent cost Case 2 demonstrates). PE-replicate within each
+    cube too: every PE on every cube holds an HBM copy.
+  - PE-TP on T_q: only CUBE 0 has work; within CUBE 0 the P PEs split
+    the T_q axis disjointly. CUBEs 1..7 idle (slide-11 "PE-TP only
+    uses one cube; replicate is pure waste here").
+  - NO inter-rank communication (each active rank has full KV and
+    disjoint T_q rows — slide 11 lists comm cost as "none").
+
+Distinction from decode Case 2: decode T_q=1 makes PE-TP wasteful
+(only PE 0 of CUBE 0 works → 1 of 64 ranks active). Prefill T_q≫1
+makes PE-TP useful — all P PEs of CUBE 0 work on disjoint T_q rows
+(P of 64 ranks active; CUBEs 1..7 are pure memory waste).
+
+Q-axis tiling: an outer Q-tile loop wraps the entire per-tile
+pipeline (Q load → bootstrap → S_kv-tile fold → store) in
+``tl.scratch_scope``. Peak scratch is bounded by
+``TILE_Q × TILE_S_KV`` instead of ``T_q_local × S_kv``, so the
+kernel runs at large T_q without exceeding the ~1 MB per-PE budget.
+Each Q-tile is independent (disjoint output rows; no inter-tile state).
+
+Tensor layout:
+  Q : (T_q, h_q · d_head) cube=replicate, pe=row_wise on T_q. Only
+      CUBE 0's PEs load — each owns ``G · T_q_local`` total rows with
+      ``T_q_local = T_q / P``; loaded per Q-tile.
+  K : (S_kv, h_kv · d_head) replicated on every rank.
+  V : (S_kv, h_kv · d_head) replicated on every rank.
+  O : (T_q, h_q · d_head) cube=replicate, pe=row_wise on T_q —
+      only CUBE 0's PEs write their disjoint ``T_q_local`` rows,
+      one ``(TILE_Q, d_head)`` slice per outer iteration.
+
+Topology / SFR:
+  - SFR setup is benign (no inter-rank sends/recvs); any of the
+    ``configure_sfr_intercube_*`` helpers works.
+
+Requires ``T_q % P == 0``.
+"""
+from __future__ import annotations
+
+
+TILE_S_KV = 64  # per-tile S_kv width (small to keep ``scores`` bounded).
+TILE_Q = 64     # per-tile Q row count (outer-loop tile size).
+
+
+def gqa_attention_prefill_long_ctx_cube_repl_pe_tp_kernel(
+    q_ptr: int,
+    k_ptr: int,
+    v_ptr: int,
+    o_ptr: int,
+    T_q: int,
+    S_kv: int,
+    h_q: int,
+    h_kv: int,
+    d_head: int,
+    C: int,
+    P: int,
+    *,
+    tl,
+) -> None:
+    """Case-2 prefill: CUBE 0 only; Q-tiled PE-TP; full KV per rank; no comm."""
+    if T_q % P != 0:
+        raise ValueError(
+            f"Case 2 prefill requires T_q={T_q} divisible by P={P}"
+        )
+
+    cube_id = tl.program_id(axis=1)
+    # CUBE 0 only — CUBEs 1..7 are pure memory-replication waste, no work.
+    if cube_id != 0:
+        return
+
+    G = h_q // h_kv
+    T_q_local = T_q // P
+    n_kv_tiles = (S_kv + TILE_S_KV - 1) // TILE_S_KV
+    KV_ROW_BYTES = d_head * 2  # f16
+    Q_ROW_BYTES = d_head * 2
+    total_q_rows = G * T_q_local
+    n_q_tiles = (total_q_rows + TILE_Q - 1) // TILE_Q
+
+    # ── Outer Q-tile loop ──
+    # Per Q-tile: load Q-slice, run full S_kv attention, store. All
+    # scratch is recycled at outer-scope exit, so peak usage is bounded
+    # by TILE_Q × TILE_S_KV.
+    for q_idx in range(n_q_tiles):
+        q_start = q_idx * TILE_Q
+        q_rows = min(TILE_Q, total_q_rows - q_start)
+        with tl.scratch_scope():
+            Q = tl.load(q_ptr + q_start * Q_ROW_BYTES,
+                        shape=(q_rows, d_head), dtype="f16")
+
+            # ── Bootstrap (S_kv tile 0) ──
+            tile_s0 = min(TILE_S_KV, S_kv)
+            K_T = tl.load(k_ptr, shape=(d_head, tile_s0), dtype="f16")
+            V = tl.load(v_ptr, shape=(tile_s0, d_head), dtype="f16")
+            scores = tl.dot(Q, K_T)
+            m_local = tl.max(scores, axis=-1)
+            exp_scores = tl.exp(scores - m_local)
+            l_local = tl.sum(exp_scores, axis=-1)
+            O_local = tl.dot(exp_scores, V)
+
+            # ── S_kv tiles 1..N: fold via online-softmax merge ──
+            for tile_idx in range(1, n_kv_tiles):
+                tile_start = tile_idx * TILE_S_KV
+                tile_s = min(TILE_S_KV, S_kv - tile_start)
+                with tl.scratch_scope():
+                    K_T_t = tl.load(k_ptr + tile_start * KV_ROW_BYTES,
+                                    shape=(d_head, tile_s), dtype="f16")
+                    V_t = tl.load(v_ptr + tile_start * KV_ROW_BYTES,
+                                  shape=(tile_s, d_head), dtype="f16")
+                    scores_t = tl.dot(Q, K_T_t)
+                    m_tile = tl.max(scores_t, axis=-1)
+                    exp_scores_t = tl.exp(scores_t - m_tile)
+                    l_tile = tl.sum(exp_scores_t, axis=-1)
+                    O_tile = tl.dot(exp_scores_t, V_t)
+                    m_new = tl.maximum(m_local, m_tile)
+                    scale_old = tl.exp(m_local - m_new)
+                    scale_new = tl.exp(m_tile - m_new)
+                    l_new = l_local * scale_old + l_tile * scale_new
+                    O_new = O_local * scale_old + O_tile * scale_new
+                    tl.copy_to(m_local, m_new)
+                    tl.copy_to(l_local, l_new)
+                    tl.copy_to(O_local, O_new)
+
+            # ── Store this Q-tile (each active PE writes its slice) ──
+            O_final = O_local / l_local
+            tl.store(o_ptr + q_start * Q_ROW_BYTES, O_final)

src/kernbench/benches/gqa_helpers/long_ctx/_gqa_attention_prefill_long_ctx_cube_sp_pe_sp.py

@@ -0,0 +1,186 @@
+"""GQA prefill kernel — Case 4 (Cube-SP × PE-SP) at single-KV-head group. ★
+
+Per GQA_full_deck.pptx slide 11 (prefill counterpart, T_q ≫ 1):
+  - K, V split 64-way (cube=row_wise, pe=row_wise on S_kv); each rank
+    owns ``S_local_pe = S_kv / (C·P)``.
+  - Q, O split T_q-wise across cubes (cube=row_wise on T_q),
+    pe=replicate within each cube — each cube owns
+    ``T_q_cube = T_q / C`` rows, all P PEs of the cube hold a copy.
+  - Inter-CUBE Ring KV (snake over 4×2 sub-mesh, ADR-0060 §5.5):
+    P parallel rings — each PE drives its own same-lane ring via
+    independent IPCQ channels. Over C ring steps each PE has
+    processed its strided lane of S_kv (= ``S_kv / P`` tokens) for
+    the cube's T_q_cube query rows.
+  - Intra-CUBE 8-way reduce on the partial ``(m, ℓ, O)`` triple
+    (row chain along intra_W + col bridge along intra_N over the
+    2×4 PE grid) — combines the P PE-lanes per cube so cube's PE 0
+    holds the full S_kv attention for its T_q_cube rows.
+  - NO inter-CUBE reduce — each cube writes its own disjoint
+    T_q_cube rows of O (PE 0 of each cube is the writer).
+
+Distinction from decode Case 4: decode T_q=1 forces an inter-CUBE
+lrab reduce-to-root over (m,ℓ,O). Prefill T_q≫1 allows Q to be
+T_q-sharded across cubes, so disjoint output rows avoid the
+inter-cube reduce — the prefill-native pattern (per user's
+"Ring KV" choice).
+
+Tensor layout:
+  Q : (T_q, h_q · d_head) sharded cube=row_wise on T_q, pe=replicate
+      within cube; each rank loads ``(G · T_q_cube, d_head)``.
+  K : (S_kv, h_kv · d_head) cube=row_wise + pe=row_wise on S_kv;
+      each rank owns ``(S_local_pe, h_kv·d_head)``.
+  V : same as K.
+  O : (T_q, h_q · d_head) sharded same as Q — PE 0 of each cube
+      writes its ``(G · T_q_cube, d_head)`` slice; PEs 1..P-1 idle
+      the write.
+
+Topology / SFR:
+  - Requires ``configure_sfr_intercube_ring(submesh_shape=(2, 4))`` —
+    that helper installs both the snake E/W ring across 8 cubes and
+    the intra_* lanes needed for the intra-CUBE reduce.
+
+Requires ``T_q % C == 0`` and ``S_kv % (C · P) == 0``.
+"""
+from __future__ import annotations
+
+
+TILE_S_KV = 1024  # ADR-0063 §A.2 S_kv-axis tile sweep (per-tile width).
+
+
+def _merge_running(m_local, l_local, O_local, m_other, l_other, O_other, *, tl):
+    """Online-softmax merge of two partial ``(m, ℓ, O)`` triples."""
+    m_new = tl.maximum(m_local, m_other)
+    scale_old = tl.exp(m_local - m_new)
+    scale_new = tl.exp(m_other - m_new)
+    l_new = l_local * scale_old + l_other * scale_new
+    O_new = O_local * scale_old + O_other * scale_new
+    return m_new, l_new, O_new
+
+
+def gqa_attention_prefill_long_ctx_cube_sp_pe_sp_kernel(
+    q_ptr: int,
+    k_ptr: int,
+    v_ptr: int,
+    o_ptr: int,
+    T_q: int,
+    S_kv: int,
+    h_q: int,
+    h_kv: int,
+    d_head: int,
+    C: int,
+    P: int,
+    *,
+    tl,
+) -> None:
+    """Case-4 prefill: Ring KV across cubes + intra-CUBE AR; disjoint T_q/C rows per cube."""
+    if T_q % C != 0:
+        raise ValueError(
+            f"Case 4 prefill requires T_q={T_q} divisible by C={C}"
+        )
+    if S_kv % (C * P) != 0:
+        raise ValueError(
+            f"Case 4 prefill requires S_kv={S_kv} divisible by C·P={C * P}"
+        )
+
+    G = h_q // h_kv
+    T_q_cube = T_q // C
+    S_local_pe = S_kv // (C * P)
+    KV_ROW_BYTES = d_head * 2  # f16
+    n_tiles = (S_local_pe + TILE_S_KV - 1) // TILE_S_KV
+
+    pe_id = tl.program_id(axis=0)
+
+    # Q is this cube's T_q slice (pe=replicate within cube ⇒ every PE
+    # loads the same G·T_q_cube rows).
+    Q = tl.load(q_ptr, shape=(G * T_q_cube, d_head), dtype="f16")
+
+    # ── Bootstrap (t=0, k=0): load own rank's tile 0; send W for ring ──
+    # Persistent (m, ℓ, O) must live OUTSIDE tl.scratch_scope (kernbench
+    # scope teardown discards in-scope tensors).
+    tile_s = min(TILE_S_KV, S_local_pe)
+    K_T = tl.load(k_ptr, shape=(d_head, tile_s), dtype="f16")
+    V = tl.load(v_ptr, shape=(tile_s, d_head), dtype="f16")
+    if C > 1:
+        tl.send(dir="W", src=K_T)
+        tl.send(dir="W", src=V)
+    scores = tl.dot(Q, K_T)
+    m_local = tl.max(scores, axis=-1)
+    exp_scores = tl.exp(scores - m_local)
+    l_local = tl.sum(exp_scores, axis=-1)
+    O_local = tl.dot(exp_scores, V)
+
+    # ── Nested loop (outer tile, inner ring step) — P parallel rings ──
+    for t in range(n_tiles):
+        tile_start = t * TILE_S_KV
+        tile_s = min(TILE_S_KV, S_local_pe - tile_start)
+        k_start = 1 if t == 0 else 0
+        for k in range(k_start, C):
+            with tl.scratch_scope():
+                if k == 0:
+                    K_T_t = tl.load(k_ptr + tile_start * KV_ROW_BYTES,
+                                    shape=(d_head, tile_s), dtype="f16")
+                    V_t = tl.load(v_ptr + tile_start * KV_ROW_BYTES,
+                                  shape=(tile_s, d_head), dtype="f16")
+                else:
+                    K_T_t = tl.recv(dir="E", shape=(d_head, tile_s), dtype="f16")
+                    V_t = tl.recv(dir="E", shape=(tile_s, d_head), dtype="f16")
+                if k < C - 1:
+                    tl.send(dir="W", src=K_T_t)
+                    tl.send(dir="W", src=V_t)
+                scores_t = tl.dot(Q, K_T_t)
+                m_step = tl.max(scores_t, axis=-1)
+                exp_scores_t = tl.exp(scores_t - m_step)
+                l_step = tl.sum(exp_scores_t, axis=-1)
+                O_step = tl.dot(exp_scores_t, V_t)
+                m_new, l_new, O_new = _merge_running(
+                    m_local, l_local, O_local, m_step, l_step, O_step, tl=tl,
+                )
+                tl.copy_to(m_local, m_new)
+                tl.copy_to(l_local, l_new)
+                tl.copy_to(O_local, O_new)
+
+    # ── Intra-CUBE 8-way reduce-to-PE0 (row chain + col bridge) ──
+    PE_GRID_COLS = 4
+    pe_col = pe_id % PE_GRID_COLS
+    pe_row = pe_id // PE_GRID_COLS
+    pe_cols_used = min(PE_GRID_COLS, P)
+    pe_rows_used = (P + PE_GRID_COLS - 1) // PE_GRID_COLS
+
+    if pe_cols_used > 1:
+        if pe_col < pe_cols_used - 1:
+            with tl.scratch_scope():
+                m_other = tl.recv(dir="intra_E", shape=m_local.shape, dtype="f16")
+                l_other = tl.recv(dir="intra_E", shape=l_local.shape, dtype="f16")
+                O_other = tl.recv(dir="intra_E", shape=O_local.shape, dtype="f16")
+                m_new, l_new, O_new = _merge_running(
+                    m_local, l_local, O_local, m_other, l_other, O_other, tl=tl,
+                )
+                tl.copy_to(m_local, m_new)
+                tl.copy_to(l_local, l_new)
+                tl.copy_to(O_local, O_new)
+        if pe_col > 0:
+            tl.send(dir="intra_W", src=m_local)
+            tl.send(dir="intra_W", src=l_local)
+            tl.send(dir="intra_W", src=O_local)
+
+    if pe_col == 0 and pe_rows_used > 1:
+        if pe_row < pe_rows_used - 1:
+            with tl.scratch_scope():
+                m_other = tl.recv(dir="intra_S", shape=m_local.shape, dtype="f16")
+                l_other = tl.recv(dir="intra_S", shape=l_local.shape, dtype="f16")
+                O_other = tl.recv(dir="intra_S", shape=O_local.shape, dtype="f16")
+                m_new, l_new, O_new = _merge_running(
+                    m_local, l_local, O_local, m_other, l_other, O_other, tl=tl,
+                )
+                tl.copy_to(m_local, m_new)
+                tl.copy_to(l_local, l_new)
+                tl.copy_to(O_local, O_new)
+        if pe_row > 0:
+            tl.send(dir="intra_N", src=m_local)
+            tl.send(dir="intra_N", src=l_local)
+            tl.send(dir="intra_N", src=O_local)
+
+    # ── Final normalise + store (PE 0 of each cube writes its T_q_cube rows) ──
+    if pe_id == 0:
+        O_final = O_local / l_local
+        tl.store(o_ptr, O_final)

src/kernbench/benches/gqa_helpers/long_ctx/_gqa_attention_prefill_long_ctx_cube_sp_pe_tp.py

@@ -0,0 +1,136 @@
+"""GQA prefill kernel — Case 1 (Cube-SP × PE-TP) at single-KV-head group.
+
+Per GQA_full_deck.pptx slide 11 (prefill counterpart, T_q ≫ 1):
+  - K, V split S_kv-wise across the 8 cubes (cube=row_wise);
+    replicated within each cube (pe=replicate). Each cube owns
+    ``S_local = S_kv / C``.
+  - Q, O split T_q-wise across cubes AND across PEs intra-cube
+    (cube=row_wise, pe=row_wise on T_q). Each rank owns disjoint
+    ``T_q_local = T_q / (C · P)`` query rows.
+  - Inter-CUBE Ring KV (snake over 4×2 sub-mesh, ADR-0060 §5.5):
+    P parallel rings — each PE drives its own same-lane ring via
+    independent IPCQ channels. Over C ring steps the cube-owned
+    K/V blocks rotate through all C cubes, so every rank sees
+    every block.
+  - NO inter-CUBE reduce — each rank writes its own disjoint
+    T_q_local rows of O.
+  - NO intra-CUBE reduce — PEs are disjoint on T_q.
+
+Distinction from decode Case 1: decode T_q=1 makes PE-TP wasteful
+(only PE 0 of each cube works). Prefill T_q≫1 makes PE-TP useful —
+all 64 ranks work on disjoint Q rows.
+
+Tensor layout:
+  Q : (T_q, h_q · d_head) sharded (cube_row_wise, pe_row_wise) on T_q;
+      each rank loads ``(G · T_q_local, d_head)``.
+  K : (S_kv, h_kv · d_head) with cube=row_wise, pe=replicate — each
+      cube's PE 0..P-1 each hold an HBM copy of the cube's
+      ``(S_local, h_kv·d_head)`` shard.
+  V : same as K.
+  O : (T_q, h_q · d_head) sharded same as Q — each rank writes its
+      own ``(G · T_q_local, d_head)`` slice.
+
+Topology / SFR:
+  - Requires ``configure_sfr_intercube_ring(submesh_shape=(2, 4))``
+    for the snake E/W ring across 8 cubes (wrap from cube 3 ↔ cube 7).
+
+Requires ``T_q % (C · P) == 0``.
+"""
+from __future__ import annotations
+
+
+TILE_S_KV = 1024  # ADR-0063 §A.2 S_kv-axis tile sweep (per-tile width).
+
+
+def _merge_running(m_local, l_local, O_local, m_other, l_other, O_other, *, tl):
+    """Online-softmax merge of two partial ``(m, ℓ, O)`` triples."""
+    m_new = tl.maximum(m_local, m_other)
+    scale_old = tl.exp(m_local - m_new)
+    scale_new = tl.exp(m_other - m_new)
+    l_new = l_local * scale_old + l_other * scale_new
+    O_new = O_local * scale_old + O_other * scale_new
+    return m_new, l_new, O_new
+
+
+def gqa_attention_prefill_long_ctx_cube_sp_pe_tp_kernel(
+    q_ptr: int,
+    k_ptr: int,
+    v_ptr: int,
+    o_ptr: int,
+    T_q: int,
+    S_kv: int,
+    h_q: int,
+    h_kv: int,
+    d_head: int,
+    C: int,
+    P: int,
+    *,
+    tl,
+) -> None:
+    """Case-1 prefill: Ring KV across cubes; disjoint T_q rows per rank."""
+    if T_q % (C * P) != 0:
+        raise ValueError(
+            f"Case 1 prefill requires T_q={T_q} divisible by C·P={C * P}"
+        )
+
+    G = h_q // h_kv
+    T_q_local = T_q // (C * P)
+    S_local = S_kv // C
+    KV_ROW_BYTES = d_head * 2  # f16
+    n_tiles = (S_local + TILE_S_KV - 1) // TILE_S_KV
+
+    # Q is this rank's own disjoint T_q slice (G·T_q_local rows for the
+    # G-grouped attention).
+    Q = tl.load(q_ptr, shape=(G * T_q_local, d_head), dtype="f16")
+
+    # ── Bootstrap (t=0, k=0): load own cube's tile 0; send W for ring ──
+    # Persistent (m, ℓ, O) must live OUTSIDE tl.scratch_scope (kernbench
+    # scope teardown discards in-scope tensors). Mirrors decode Cases
+    # 1, 3, 4 and the existing prefill_long bootstrap pattern.
+    tile_s = min(TILE_S_KV, S_local)
+    K_T = tl.load(k_ptr, shape=(d_head, tile_s), dtype="f16")
+    V = tl.load(v_ptr, shape=(tile_s, d_head), dtype="f16")
+    if C > 1:
+        tl.send(dir="W", src=K_T)
+        tl.send(dir="W", src=V)
+    scores = tl.dot(Q, K_T)
+    m_local = tl.max(scores, axis=-1)
+    exp_scores = tl.exp(scores - m_local)
+    l_local = tl.sum(exp_scores, axis=-1)
+    O_local = tl.dot(exp_scores, V)
+
+    # ── Nested loop (outer tile, inner ring step) ──
+    # Each tile propagates around the ring before the next tile starts;
+    # IPCQ in-flight depth stays at 1 per direction per PE-lane.
+    for t in range(n_tiles):
+        tile_start = t * TILE_S_KV
+        tile_s = min(TILE_S_KV, S_local - tile_start)
+        k_start = 1 if t == 0 else 0
+        for k in range(k_start, C):
+            with tl.scratch_scope():
+                if k == 0:
+                    K_T_t = tl.load(k_ptr + tile_start * KV_ROW_BYTES,
+                                    shape=(d_head, tile_s), dtype="f16")
+                    V_t = tl.load(v_ptr + tile_start * KV_ROW_BYTES,
+                                  shape=(tile_s, d_head), dtype="f16")
+                else:
+                    K_T_t = tl.recv(dir="E", shape=(d_head, tile_s), dtype="f16")
+                    V_t = tl.recv(dir="E", shape=(tile_s, d_head), dtype="f16")
+                if k < C - 1:
+                    tl.send(dir="W", src=K_T_t)
+                    tl.send(dir="W", src=V_t)
+                scores_t = tl.dot(Q, K_T_t)
+                m_step = tl.max(scores_t, axis=-1)
+                exp_scores_t = tl.exp(scores_t - m_step)
+                l_step = tl.sum(exp_scores_t, axis=-1)
+                O_step = tl.dot(exp_scores_t, V_t)
+                m_new, l_new, O_new = _merge_running(
+                    m_local, l_local, O_local, m_step, l_step, O_step, tl=tl,
+                )
+                tl.copy_to(m_local, m_new)
+                tl.copy_to(l_local, l_new)
+                tl.copy_to(O_local, O_new)
+
+    # ── Final normalise + store (every rank writes its own T_q_local rows) ──
+    O_final = O_local / l_local
+    tl.store(o_ptr, O_final)

src/kernbench/benches/gqa_helpers/long_ctx/gqa_prefill_long_ctx_4cases.py

@@ -0,0 +1,384 @@
+"""milestone-gqa-prefill-long-ctx-4cases: long-context prefill 4-cases study.
+
+Prefill counterpart to milestone-gqa-decode-long-ctx-4cases. Per
+GQA_full_deck.pptx slides 11-17: 4 KV-cache sharding strategies on
+the LLaMA-3.1-70B single-KV-head group (1 KV head, 8 Q heads,
+d_head=128) on 8 cubes × 8 PEs.
+
+Bench size T_q=S_kv=512 (equal — prompt length matches K/V length
+for prefill). Cases 2 and 3 use Q-axis tiling (``TILE_Q``) so the
+per-PE scratch is bounded by ``TILE_Q × TILE_S_KV`` regardless of
+T_q — much larger sizes are possible at the cost of bench
+wall-clock (Case 2's single-cube serial path dominates).
+
+    Case 1  Cube-SP / PE-TP  → KV split S_kv across cubes (Ring),
+                               Q/O split T_q across all 64 ranks
+    Case 2  Cube-Repl / PE-TP → full KV per cube; only CUBE 0's P PEs
+                               split T_q; CUBEs 1..7 are pure memory waste
+    Case 3  Cube-Repl / PE-SP → full KV per cube; PEs SP on S_kv
+                               (intra-cube AR); 8× redundant compute
+    Case 4  Cube-SP / PE-SP  → KV split 64-way + Q split T_q across cubes;
+                               Ring KV + intra-cube AR per cube  ★ optimal
+
+Each case is a separate panel. The bench drives all panels in one
+invocation and writes per-panel op_log_summary + latency to sweep.json
+so the comparative analysis (latency, comm volume, parallelism,
+memory) can be generated from a single sweep.
+
+Distinction from decode: decode T_q=1 makes PE-TP cases wasteful (1
+active PE). Prefill T_q≫1 makes PE-TP useful (full parallelism). The
+narrative differs — the figures emphasise Case 2/3 memory waste and
+Case 3 compute redundancy, not PE-TP-at-B=1 wasted ranks.
+
+Per-panel try/except in ``run()`` writes whichever cases succeed —
+the bench tolerates per-panel scratch / SFR failures so sweep.json
+always lands with at least the successful rows + a ``failures`` list.
+
+Gated by ``GQA_PREFILL_LONG_CTX_4CASES_RUN=1``.
+"""
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+
+from kernbench.benches.gqa_helpers.long_ctx._gqa_attention_prefill_long_ctx_cube_repl_pe_sp import (
+    gqa_attention_prefill_long_ctx_cube_repl_pe_sp_kernel,
+)
+from kernbench.benches.gqa_helpers.long_ctx._gqa_attention_prefill_long_ctx_cube_repl_pe_tp import (
+    gqa_attention_prefill_long_ctx_cube_repl_pe_tp_kernel,
+)
+from kernbench.benches.gqa_helpers.long_ctx._gqa_attention_prefill_long_ctx_cube_sp_pe_sp import (
+    gqa_attention_prefill_long_ctx_cube_sp_pe_sp_kernel,
+)
+from kernbench.benches.gqa_helpers.long_ctx._gqa_attention_prefill_long_ctx_cube_sp_pe_tp import (
+    gqa_attention_prefill_long_ctx_cube_sp_pe_tp_kernel,
+)
+from kernbench.benches.gqa_helpers.shared._gqa_panel_helpers import (
+    _ccl_cfg,
+    _summarize_op_log,
+)
+from kernbench.ccl.sfr_config import configure_sfr_intercube_ring
+from kernbench.policy.placement.dp import DPPolicy
+
+# File is at benches/gqa_helpers/long_ctx/ — go up 2 parents to reach
+# benches/, then into 1H_milestone_output/gqa/long_ctx/.
+_OUTPUT_DIR = (
+    Path(__file__).resolve().parents[2]
+    / "1H_milestone_output" / "gqa" / "long_ctx"
+)
+_SWEEP_JSON = _OUTPUT_DIR / "sweep_prefill.json"
+
+
+# ── Panel registry ───────────────────────────────────────────────────
+
+
+_PANELS = (
+    "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_sp",      # Case 4 ★ optimal
+    "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_tp",    # Case 2
+    "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_sp",    # Case 3
+    "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_tp",      # Case 1
+)
+
+# Each entry: (kind, panel-specific params).
+# LLaMA-3.1-70B single-KV-head group target:
+#   1 KV head, h_q = 8 (G = 8 group), d_head = 128
+#   8 cubes (head-parallel group), 8 PEs/cube
+#   T_q = S_kv = 4096 (long-context prefill)
+_PANEL_DISPATCH: dict[str, tuple[str, dict]] = {
+    "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_sp": ("prefill_long_ctx_cube_sp_pe_sp", {
+        # Case 4: KV split 64-way + Q split T_q across cubes; Ring KV
+        # + intra-CUBE AR per cube. PE 0 of each cube writes its
+        # T_q/C disjoint output rows.
+        "C": 8, "P": 8,
+        "T_q": 512, "S_kv": 512,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }),
+    "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_tp": ("prefill_long_ctx_cube_repl_pe_tp", {
+        # Case 2: K, V replicated everywhere (8× memory waste); within
+        # CUBE 0 the P PEs split T_q disjointly. CUBEs 1..7 idle.
+        # No inter-rank communication.
+        "C": 8, "P": 8,
+        "T_q": 512, "S_kv": 512,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }),
+    "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_sp": ("prefill_long_ctx_cube_repl_pe_sp", {
+        # Case 3: K, V replicated per cube (8× memory); PEs SP on S_kv
+        # within each cube. Intra-CUBE 8-way reduce; no inter-CUBE comm
+        # (every cube does redundant T_q × S_kv compute; designated
+        # writer = cube 0).
+        "C": 8, "P": 8,
+        "T_q": 512, "S_kv": 512,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }),
+    "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_tp": ("prefill_long_ctx_cube_sp_pe_tp", {
+        # Case 1: K, V split S_kv across cubes (S_local = S_kv/C per
+        # cube); Q/O split T_q across all 64 ranks. Inter-CUBE snake
+        # Ring KV (P parallel rings); no intra-CUBE comm.
+        "C": 8, "P": 8,
+        "T_q": 512, "S_kv": 512,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }),
+}
+
+
+# ── Per-panel runner ─────────────────────────────────────────────────
+
+
+def _ring_sfr(ctx):
+    """Install snake Ring SFR (also provides intra_* lanes for AR cases).
+
+    Snake submesh of shape (2, 4) over the default 4×4 cube mesh
+    yields a Hamiltonian cycle through all 8 cubes
+    (0→1→2→3→7→6→5→4→0), with intra_* lanes available for the
+    Case 3/4 intra-CUBE reduce.
+    """
+    configure_sfr_intercube_ring(
+        ctx.engine, ctx.spec, _ccl_cfg(),
+        submesh_shape=(2, 4),
+    )
+
+
+def _run_prefill_panel_long_ctx_cube_repl_pe_tp(
+    ctx, *, panel: str, C: int, P: int,
+    T_q: int, S_kv: int,
+    d_head: int, h_q: int, h_kv: int,
+) -> None:
+    """Case 2 runner: K, V replicated everywhere; CUBE 0 PE-TP on T_q.
+
+    DPPolicy models the cluster-wide 8× memory waste — every rank
+    holds full K, V in its HBM region. Within CUBE 0, P PEs split
+    T_q (so P of 64 ranks actually compute). CUBEs 1..7 idle.
+    """
+    _ring_sfr(ctx)
+    dp_repl = DPPolicy(cube="replicate", pe="replicate",
+                       num_cubes=C, num_pes=P)
+    dp_qo = DPPolicy(cube="replicate", pe="row_wise",
+                     num_cubes=C, num_pes=P)
+    q = ctx.zeros((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_qo, name=f"{panel}_q")
+    k = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_repl, name=f"{panel}_k")
+    v = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_repl, name=f"{panel}_v")
+    o = ctx.empty((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_qo, name=f"{panel}_o")
+    ctx.launch(
+        panel, gqa_attention_prefill_long_ctx_cube_repl_pe_tp_kernel,
+        q, k, v, o,
+        T_q, S_kv, h_q, h_kv, d_head, C, P,
+        _auto_dim_remap=False,
+    )
+
+
+def _run_prefill_panel_long_ctx_cube_repl_pe_sp(
+    ctx, *, panel: str, C: int, P: int,
+    T_q: int, S_kv: int,
+    d_head: int, h_q: int, h_kv: int,
+) -> None:
+    """Case 3 runner: K, V replicated per cube; PEs SP on S_kv intra-cube.
+
+    DPPolicy models the cluster-wide 8× memory waste — every cube
+    holds full K, V; intra-cube splits the S_kv axis row_wise across
+    8 PEs. Every rank holds full Q (cube=replicate, pe=replicate).
+    The kernel does an intra-CUBE 8-way reduce on (m, ℓ, O); only
+    cube 0's PE 0 writes the output.
+    """
+    _ring_sfr(ctx)
+    dp_full = DPPolicy(cube="replicate", pe="replicate",
+                       num_cubes=C, num_pes=P)
+    dp_kv = DPPolicy(cube="replicate", pe="row_wise",
+                     num_cubes=C, num_pes=P)
+    q = ctx.zeros((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_full, name=f"{panel}_q")
+    k = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_kv, name=f"{panel}_k")
+    v = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_kv, name=f"{panel}_v")
+    o = ctx.empty((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_full, name=f"{panel}_o")
+    ctx.launch(
+        panel, gqa_attention_prefill_long_ctx_cube_repl_pe_sp_kernel,
+        q, k, v, o,
+        T_q, S_kv, h_q, h_kv, d_head, C, P,
+        _auto_dim_remap=False,
+    )
+
+
+def _run_prefill_panel_long_ctx_cube_sp_pe_tp(
+    ctx, *, panel: str, C: int, P: int,
+    T_q: int, S_kv: int,
+    d_head: int, h_q: int, h_kv: int,
+) -> None:
+    """Case 1 runner: K, V split S_kv across cubes (Ring); Q/O split T_q 64-way.
+
+    DPPolicy: K, V are cube=row_wise on S_kv (S_local = S_kv/C per
+    cube), pe=replicate within cube. Q, O are cube=row_wise +
+    pe=row_wise on T_q — every rank owns T_q/(C·P) disjoint Q rows.
+    The kernel runs P parallel Rings through the snake submesh
+    (each PE has its own IPCQ lane); no intra-CUBE reduce.
+    """
+    _ring_sfr(ctx)
+    dp_qo = DPPolicy(cube="row_wise", pe="row_wise",
+                     num_cubes=C, num_pes=P)
+    dp_kv = DPPolicy(cube="row_wise", pe="replicate",
+                     num_cubes=C, num_pes=P)
+    q = ctx.zeros((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_qo, name=f"{panel}_q")
+    k = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_kv, name=f"{panel}_k")
+    v = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_kv, name=f"{panel}_v")
+    o = ctx.empty((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_qo, name=f"{panel}_o")
+    ctx.launch(
+        panel, gqa_attention_prefill_long_ctx_cube_sp_pe_tp_kernel,
+        q, k, v, o,
+        T_q, S_kv, h_q, h_kv, d_head, C, P,
+        _auto_dim_remap=False,
+    )
+
+
+def _run_prefill_panel_long_ctx_cube_sp_pe_sp(
+    ctx, *, panel: str, C: int, P: int,
+    T_q: int, S_kv: int,
+    d_head: int, h_q: int, h_kv: int,
+) -> None:
+    """Case 4 runner: K, V split 64-way; Q split T_q across cubes.
+
+    DPPolicy: K, V are cube=row_wise + pe=row_wise on S_kv — each
+    rank owns S_local_pe = S_kv/(C·P). Q, O are cube=row_wise on
+    T_q (T_q_cube = T_q/C per cube), pe=replicate within cube. Ring
+    KV through the snake submesh (P parallel lanes per cube); after
+    the ring, an intra-CUBE 8-way AR combines PE lanes so cube's
+    PE 0 has the full attention for its T_q_cube rows. PE 0 of each
+    cube writes its disjoint output rows (no inter-cube reduce).
+    """
+    _ring_sfr(ctx)
+    dp_qo = DPPolicy(cube="row_wise", pe="replicate",
+                     num_cubes=C, num_pes=P)
+    dp_kv = DPPolicy(cube="row_wise", pe="row_wise",
+                     num_cubes=C, num_pes=P)
+    q = ctx.zeros((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_qo, name=f"{panel}_q")
+    k = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_kv, name=f"{panel}_k")
+    v = ctx.zeros((S_kv, h_kv * d_head),
+                  dtype="f16", dp=dp_kv, name=f"{panel}_v")
+    o = ctx.empty((T_q, h_q * d_head),
+                  dtype="f16", dp=dp_qo, name=f"{panel}_o")
+    ctx.launch(
+        panel, gqa_attention_prefill_long_ctx_cube_sp_pe_sp_kernel,
+        q, k, v, o,
+        T_q, S_kv, h_q, h_kv, d_head, C, P,
+        _auto_dim_remap=False,
+    )
+
+
+def _make_bench_fn(panel: str):
+    kind, params = _PANEL_DISPATCH[panel]
+
+    def _bench_fn(ctx):
+        if kind == "prefill_long_ctx_cube_sp_pe_sp":
+            _run_prefill_panel_long_ctx_cube_sp_pe_sp(ctx, panel=panel, **params)
+        elif kind == "prefill_long_ctx_cube_repl_pe_tp":
+            _run_prefill_panel_long_ctx_cube_repl_pe_tp(ctx, panel=panel, **params)
+        elif kind == "prefill_long_ctx_cube_repl_pe_sp":
+            _run_prefill_panel_long_ctx_cube_repl_pe_sp(ctx, panel=panel, **params)
+        elif kind == "prefill_long_ctx_cube_sp_pe_tp":
+            _run_prefill_panel_long_ctx_cube_sp_pe_tp(ctx, panel=panel, **params)
+        else:
+            raise RuntimeError(
+                f"milestone-gqa-prefill-long-ctx-4cases panel {panel!r} has "
+                f"unsupported kind={kind!r}."
+            )
+    return _bench_fn
+
+
+# ── Panel metrics (sweep.json carries these for the comparative plot) ─
+
+
+_ENGINE_SUFFIXES = (
+    "pe_gemm", "pe_math", "pe_dma", "pe_fetch_store", "pe_ipcq", "pe_cpu",
+)
+
+
+def _end_to_end_ns(op_log) -> float:
+    """End-to-end window: ``max(t_end) - min(t_start)`` over all records."""
+    if not op_log:
+        return 0.0
+    return max(r.t_end for r in op_log) - min(r.t_start for r in op_log)
+
+
+def _engine_occupancy_ns(op_log) -> dict[str, float]:
+    """Per-engine summed occupancy (component_id suffix match)."""
+    return {
+        eng: sum(
+            r.t_end - r.t_start
+            for r in op_log
+            if r.component_id.endswith("." + eng)
+        )
+        for eng in _ENGINE_SUFFIXES
+    }
+
+
+def _run_panel(panel: str, topology: str) -> dict:
+    from kernbench.runtime_api.bench_runner import run_bench
+    from kernbench.runtime_api.types import resolve_device
+    from kernbench.sim_engine.engine import GraphEngine
+    from kernbench.topology.builder import resolve_topology
+
+    topo = resolve_topology(topology)
+    result = run_bench(
+        topology=topo, bench_fn=_make_bench_fn(panel),
+        device=resolve_device(None),
+        engine_factory=lambda t, d: GraphEngine(
+            getattr(t, "topology_obj", t), enable_data=True,
+        ),
+    )
+    if not result.completion.ok:
+        raise RuntimeError(
+            f"milestone-gqa-prefill-long-ctx-4cases panel {panel!r} failed: "
+            f"{result.completion}"
+        )
+    kind, params = _PANEL_DISPATCH[panel]
+    op_log = result.engine.op_log
+    return {
+        "panel": panel,
+        "kind": kind,
+        **params,
+        "op_log_summary": _summarize_op_log(op_log),
+        "latency_ns": _end_to_end_ns(op_log),
+        "engine_occupancy_ns": _engine_occupancy_ns(op_log),
+    }
+
+
+# ── Sweep entry (called by the umbrella ``milestone_1h_gqa``) ────────
+
+
+def run_sweep(topology: str = "topology.yaml") -> int:
+    """Drive all prefill case panels; write sweep.json. Returns row count.
+
+    Per-panel try/except: even with Q-axis tiling, very large T_q or
+    unusual config combinations can hit the per-PE scratch budget. We
+    keep going so sweep.json carries whichever panels succeed; failures
+    land in a ``failures`` list for follow-up.
+    """
+    _OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
+    rows: list[dict] = []
+    failures: list[dict] = []
+    for panel in _PANELS:
+        try:
+            rows.append(_run_panel(panel, topology))
+        except Exception as e:
+            print(f"  panel {panel!r} FAILED: {e}")
+            failures.append({"panel": panel, "error": str(e)})
+    sweep = {
+        "version": 1,
+        "panels": list(_PANELS),
+        "rows": rows,
+        "failures": failures,
+    }
+    _SWEEP_JSON.write_text(json.dumps(sweep, indent=2))
+    print(f"  gqa-prefill-long-ctx-4cases: {len(rows)} rows -> {_SWEEP_JSON}")
+    return len(rows)

tests/attention/test_milestone_gqa_prefill_long_ctx_4cases.py

@@ -0,0 +1,471 @@
+"""Tests for the long-context prefill 4-cases comparative-study bench.
+
+Prefill counterpart to test_milestone_gqa_decode_long_ctx_4cases. The
+4 cases differ in how KV cache is sharded across the 8 cubes and 8
+PEs of a single KV-head group on LLaMA-3.1-70B GQA:
+
+    Case 1  Cube-SP / PE-TP  → KV split by S_kv across cubes (Ring);
+                               Q/O split T_q across all 64 ranks
+    Case 2  Cube-Repl / PE-TP → full KV per cube; only CUBE 0's P PEs
+                               split T_q (CUBEs 1..7 = pure memory waste)
+    Case 3  Cube-Repl / PE-SP → full KV per cube; PEs SP on S_kv
+                               (intra-cube AR); 8× redundant compute
+    Case 4  Cube-SP / PE-SP  → KV split 64-way + Q split T_q across cubes;
+                               Ring KV + intra-cube AR per cube  ★ optimal
+
+Smoke sizes (T_q=256, S_kv=2048) keep the per-PE scratch comfortably
+inside budget; the headline T_q=S_kv=4096 runs come from
+``kernbench run --bench milestone-gqa-prefill-long-ctx-4cases``, not
+pytest.
+"""
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+from kernbench.runtime_api.bench_runner import run_bench
+from kernbench.runtime_api.types import resolve_device
+from kernbench.sim_engine.engine import GraphEngine
+from kernbench.topology.builder import resolve_topology
+
+TOPOLOGY_DEFAULT = Path(__file__).resolve().parents[2] / "topology.yaml"
+
+_CASE4_PANEL = "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_sp"
+_CASE3_PANEL = "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_sp"
+_CASE2_PANEL = "single_kv_group_prefill_long_ctx_gqa_cube_repl_pe_tp"
+_CASE1_PANEL = "single_kv_group_prefill_long_ctx_gqa_cube_sp_pe_tp"
+_CUBE_RE = re.compile(r"\bcube(\d+)\b")
+
+# Smoke sizes: T_q=256 (divisible by C·P=64 for Case 1, by P=8 for
+# Case 2, by C=8 for Case 4) and S_kv=2048 (≥ 2·TILE_S_KV so the
+# per-tile fold loop runs; Case 1/4 have S_local sufficient to exercise
+# the Ring). Assertions are size-independent. The headline T_q=S_kv=4096
+# bench runs come from the env-gated bench, not pytest.
+# Smoke sizes — kept small so pytest stays fast:
+#   T_q=64  : minimum that satisfies Case 1's ``T_q % (C·P) == 0``.
+#   S_kv=512: enough to exercise multi-tile fold for Cases 2/3
+#             (TILE_S_KV=64 → 8 tiles for Case 2; 1 tile/PE for Case 3
+#             after intra-cube S_kv split) without burning minutes on
+#             Case 2's single-rank serial path.
+# Cases 2/3 also use TILE_S_KV=64 in their kernels to keep the
+# persistent ``scores`` tensor inside the 1 MB scratch budget at
+# T_q=64. Headline T_q=S_kv=4096 runs come from the env-gated bench,
+# not pytest.
+_SMOKE_T_Q = 64
+_SMOKE_S_KV = 512
+
+_HEADLINE_T_Q = 512
+_HEADLINE_S_KV = 512
+
+
+def _engine_factory(t, d):
+    return GraphEngine(getattr(t, "topology_obj", t), enable_data=True)
+
+
+def _count(op_log, name: str) -> int:
+    return sum(1 for r in op_log if r.op_name == name)
+
+
+def _dma_write_cubes(op_log) -> list[int]:
+    cubes: list[int] = []
+    for r in op_log:
+        if r.op_name != "dma_write":
+            continue
+        m = _CUBE_RE.search(r.component_id)
+        if m is not None:
+            cubes.append(int(m.group(1)))
+    return cubes
+
+
+# ── Case 4 (Cube-SP × PE-SP) — ★ optimal ────────────────────────────
+
+
+def _run_case4_smoke(*, T_q: int, S_kv: int):
+    """Drive the Case 4 prefill panel via the case-specific runner."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _run_prefill_panel_long_ctx_cube_sp_pe_sp,
+    )
+    topo = resolve_topology(str(TOPOLOGY_DEFAULT))
+
+    def _bench_fn(ctx):
+        _run_prefill_panel_long_ctx_cube_sp_pe_sp(
+            ctx, panel=_CASE4_PANEL,
+            C=8, P=8,
+            T_q=T_q, S_kv=S_kv,
+            d_head=128, h_q=8, h_kv=1,
+        )
+
+    return run_bench(
+        topology=topo, bench_fn=_bench_fn,
+        device=resolve_device(None),
+        engine_factory=_engine_factory,
+    )
+
+
+def test_case4_panel_registered():
+    """Case 4 panel must be in ``_PANELS`` + ``_PANEL_DISPATCH`` with
+    the LLaMA-3.1-70B single-KV-group prefill dims (T_q=S_kv=4096).
+    """
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _PANEL_DISPATCH,
+        _PANELS,
+    )
+    assert _CASE4_PANEL in _PANELS, f"{_CASE4_PANEL!r} not in {_PANELS}"
+    assert _CASE4_PANEL in _PANEL_DISPATCH
+    kind, params = _PANEL_DISPATCH[_CASE4_PANEL]
+    assert kind == "prefill_long_ctx_cube_sp_pe_sp"
+    assert params == {
+        "C": 8, "P": 8,
+        "T_q": _HEADLINE_T_Q, "S_kv": _HEADLINE_S_KV,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }
+
+
+def test_case4_runner_smoke():
+    """Case 4 runner drives the new kernel to completion at smoke sizes."""
+    result = _run_case4_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok, (
+        f"Case 4 prefill smoke at C=8 P=8 must complete; "
+        f"got {result.completion}"
+    )
+
+
+def test_case4_writers_one_per_cube():
+    """Case 4 prefill: each cube writes its own disjoint T_q/C rows
+    (PE 0 of each cube). Output cubes must cover all 8.
+    """
+    result = _run_case4_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    cubes = _dma_write_cubes(result.engine.op_log)
+    assert cubes, "expected at least one dma_write for the final O store"
+    distinct = set(cubes)
+    assert distinct == set(range(8)), (
+        f"Case 4 expected PE 0 of every cube to write; got cubes={sorted(distinct)}"
+    )
+
+
+def test_case4_ipcq_copy_positive():
+    """Case 4 prefill has both inter-CUBE Ring and intra-CUBE AR
+    traffic; total ipcq_copy must be strictly positive.
+    """
+    result = _run_case4_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    n_copy = _count(result.engine.op_log, "ipcq_copy")
+    assert n_copy > 0, f"Case 4 must have Ring + intra-CUBE comm; got 0 ipcq_copy"
+
+
+# ── Case 2 (Cube-Repl × PE-TP) ──────────────────────────────────────
+
+
+def _run_case2_smoke(*, T_q: int, S_kv: int):
+    """Drive the Case 2 prefill panel via the case-specific runner."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _run_prefill_panel_long_ctx_cube_repl_pe_tp,
+    )
+    topo = resolve_topology(str(TOPOLOGY_DEFAULT))
+
+    def _bench_fn(ctx):
+        _run_prefill_panel_long_ctx_cube_repl_pe_tp(
+            ctx, panel=_CASE2_PANEL,
+            C=8, P=8,
+            T_q=T_q, S_kv=S_kv,
+            d_head=128, h_q=8, h_kv=1,
+        )
+
+    return run_bench(
+        topology=topo, bench_fn=_bench_fn,
+        device=resolve_device(None),
+        engine_factory=_engine_factory,
+    )
+
+
+def test_case2_panel_registered():
+    """Case 2 panel registered with single-KV-group prefill dims."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _PANEL_DISPATCH,
+        _PANELS,
+    )
+    assert _CASE2_PANEL in _PANELS, f"{_CASE2_PANEL!r} not in {_PANELS}"
+    assert _CASE2_PANEL in _PANEL_DISPATCH
+    kind, params = _PANEL_DISPATCH[_CASE2_PANEL]
+    assert kind == "prefill_long_ctx_cube_repl_pe_tp"
+    assert params == {
+        "C": 8, "P": 8,
+        "T_q": _HEADLINE_T_Q, "S_kv": _HEADLINE_S_KV,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }
+
+
+def test_case2_runner_smoke():
+    """Case 2 runner drives the new kernel to completion at smoke sizes."""
+    result = _run_case2_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok, (
+        f"Case 2 prefill smoke at C=8 P=8 must complete; "
+        f"got {result.completion}"
+    )
+
+
+def test_case2_zero_ipcq_copy_no_comm():
+    """Case 2's defining property: full KV per rank ⇒ NO inter-rank
+    communication. Slide 11 lists comm cost as 'none'.
+    """
+    result = _run_case2_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    n_copy = _count(result.engine.op_log, "ipcq_copy")
+    assert n_copy == 0, (
+        f"Case 2 must have zero inter-rank comm; got ipcq_copy={n_copy}"
+    )
+
+
+def test_case2_writers_only_cube_0():
+    """Case 2: only CUBE 0 active (PEs 0..P-1 each writing their T_q/P
+    rows). All dma_writes come from cube 0.
+    """
+    result = _run_case2_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    cubes = _dma_write_cubes(result.engine.op_log)
+    assert cubes, "expected at least one dma_write for the final O store"
+    distinct = set(cubes)
+    assert distinct == {0}, (
+        f"Case 2 writers must be cube 0 only; got cubes={sorted(distinct)}"
+    )
+
+
+# ── Case 3 (Cube-Repl × PE-SP) ──────────────────────────────────────
+
+
+def _run_case3_smoke(*, T_q: int, S_kv: int):
+    """Drive the Case 3 prefill panel via the case-specific runner."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _run_prefill_panel_long_ctx_cube_repl_pe_sp,
+    )
+    topo = resolve_topology(str(TOPOLOGY_DEFAULT))
+
+    def _bench_fn(ctx):
+        _run_prefill_panel_long_ctx_cube_repl_pe_sp(
+            ctx, panel=_CASE3_PANEL,
+            C=8, P=8,
+            T_q=T_q, S_kv=S_kv,
+            d_head=128, h_q=8, h_kv=1,
+        )
+
+    return run_bench(
+        topology=topo, bench_fn=_bench_fn,
+        device=resolve_device(None),
+        engine_factory=_engine_factory,
+    )
+
+
+def test_case3_panel_registered():
+    """Case 3 panel registered with single-KV-group prefill dims."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _PANEL_DISPATCH,
+        _PANELS,
+    )
+    assert _CASE3_PANEL in _PANELS, f"{_CASE3_PANEL!r} not in {_PANELS}"
+    assert _CASE3_PANEL in _PANEL_DISPATCH
+    kind, params = _PANEL_DISPATCH[_CASE3_PANEL]
+    assert kind == "prefill_long_ctx_cube_repl_pe_sp"
+    assert params == {
+        "C": 8, "P": 8,
+        "T_q": _HEADLINE_T_Q, "S_kv": _HEADLINE_S_KV,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }
+
+
+def test_case3_runner_smoke():
+    """Case 3 runner drives the new kernel to completion at smoke sizes."""
+    result = _run_case3_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok, (
+        f"Case 3 prefill smoke at C=8 P=8 must complete; "
+        f"got {result.completion}"
+    )
+
+
+def test_case3_intra_cube_ar_only_ipcq():
+    """Case 3 reduce pattern: per-CUBE 8-way PE-SP AR (same structural
+    cost as decode Case 3's intra-CUBE phase = 21 ipcq_copy per cube),
+    and NO inter-CUBE traffic (each cube has a full copy of KV).
+
+    Q-axis tiling tradeoff: prefill Case 3 wraps the intra-CUBE AR
+    INSIDE the outer Q-tile loop (the kernel can't hold full
+    (m, ℓ, O) for the whole T_q in scratch). So the AR runs once
+    per Q-tile → total ipcq = n_q_tiles × per-tile cost.
+
+      per-CUBE intra (2×4 PE grid), PER Q-TILE:
+        row chain along intra_W: cols 1,2,3 each row × 2 rows ×
+          3 tensors (m, ℓ, O)                          = 18
+        col bridge along intra_N: pe4 only × 3 tensors =  3
+        per-CUBE intra total                           = 21
+        × 8 CUBEs                                      = 168
+        × n_q_tiles (= G·T_q / TILE_Q = 8·64/64 = 8)   = 1344
+
+      inter-CUBE: 0 (replicated KV ⇒ no AllReduce needed).
+    """
+    from kernbench.benches.gqa_helpers.long_ctx._gqa_attention_prefill_long_ctx_cube_repl_pe_sp import (
+        TILE_Q,
+    )
+
+    result = _run_case3_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    G = 8  # h_q / h_kv
+    n_q_tiles = (G * _SMOKE_T_Q + TILE_Q - 1) // TILE_Q
+    expected = n_q_tiles * 168
+    n_copy = _count(result.engine.op_log, "ipcq_copy")
+    assert n_copy == expected, (
+        f"Case 3 expected {expected} ipcq_copy "
+        f"({n_q_tiles} Q-tiles × 168 per Q-tile intra-CUBE AR); "
+        f"got {n_copy}"
+    )
+
+
+def test_case3_single_dma_write_at_cube_0():
+    """Every cube ends with the full answer after intra-CUBE AR; only
+    the designated writer (cube 0, PE 0) stores O.
+    """
+    result = _run_case3_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    cubes = _dma_write_cubes(result.engine.op_log)
+    assert cubes, "expected at least one dma_write for the final O store"
+    distinct = set(cubes)
+    assert distinct == {0}, (
+        f"Case 3 designated writer must be cube 0; got cubes={sorted(distinct)}"
+    )
+
+
+# ── Case 1 (Cube-SP × PE-TP) ────────────────────────────────────────
+
+
+def _run_case1_smoke(*, T_q: int, S_kv: int):
+    """Drive the Case 1 prefill panel via the case-specific runner."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _run_prefill_panel_long_ctx_cube_sp_pe_tp,
+    )
+    topo = resolve_topology(str(TOPOLOGY_DEFAULT))
+
+    def _bench_fn(ctx):
+        _run_prefill_panel_long_ctx_cube_sp_pe_tp(
+            ctx, panel=_CASE1_PANEL,
+            C=8, P=8,
+            T_q=T_q, S_kv=S_kv,
+            d_head=128, h_q=8, h_kv=1,
+        )
+
+    return run_bench(
+        topology=topo, bench_fn=_bench_fn,
+        device=resolve_device(None),
+        engine_factory=_engine_factory,
+    )
+
+
+def test_case1_panel_registered():
+    """Case 1 panel registered with single-KV-group prefill dims."""
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _PANEL_DISPATCH,
+        _PANELS,
+    )
+    assert _CASE1_PANEL in _PANELS, f"{_CASE1_PANEL!r} not in {_PANELS}"
+    assert _CASE1_PANEL in _PANEL_DISPATCH
+    kind, params = _PANEL_DISPATCH[_CASE1_PANEL]
+    assert kind == "prefill_long_ctx_cube_sp_pe_tp"
+    assert params == {
+        "C": 8, "P": 8,
+        "T_q": _HEADLINE_T_Q, "S_kv": _HEADLINE_S_KV,
+        "d_head": 128, "h_q": 8, "h_kv": 1,
+    }
+
+
+def test_case1_runner_smoke():
+    """Case 1 runner drives the new kernel to completion at smoke sizes."""
+    result = _run_case1_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok, (
+        f"Case 1 prefill smoke at C=8 P=8 must complete; "
+        f"got {result.completion}"
+    )
+
+
+def test_case1_ipcq_copy_positive_ring_only():
+    """Case 1 has inter-CUBE Ring (P parallel lanes) only; total ipcq
+    must be positive. No intra-CUBE comm (PEs are disjoint on T_q).
+    """
+    result = _run_case1_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    n_copy = _count(result.engine.op_log, "ipcq_copy")
+    assert n_copy > 0, f"Case 1 must have Ring KV comm; got 0 ipcq_copy"
+
+
+def test_case1_writers_all_64_ranks():
+    """Case 1 prefill: every rank owns disjoint T_q/(C·P) rows and
+    writes them. All 8 cubes appear in dma_write component ids.
+    """
+    result = _run_case1_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    cubes = _dma_write_cubes(result.engine.op_log)
+    assert cubes, "expected at least one dma_write for the final O store"
+    distinct = set(cubes)
+    assert distinct == set(range(8)), (
+        f"Case 1 expected every cube to write; got cubes={sorted(distinct)}"
+    )
+
+
+# ── Panel-metrics helpers + _run_panel wiring ───────────────────────
+
+
+_EXPECTED_ENGINES = {
+    "pe_gemm", "pe_math", "pe_dma", "pe_fetch_store", "pe_ipcq", "pe_cpu",
+}
+
+
+def test_panel_metrics_helpers_present_and_correct():
+    """Bench file must expose ``_end_to_end_ns`` and
+    ``_engine_occupancy_ns`` for the comparative plot script.
+    Exercised via the Case 2 smoke runner's op_log (cheapest case).
+    """
+    from kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases import (
+        _end_to_end_ns,
+        _engine_occupancy_ns,
+    )
+    result = _run_case2_smoke(T_q=_SMOKE_T_Q, S_kv=_SMOKE_S_KV)
+    assert result.completion.ok
+    op_log = result.engine.op_log
+
+    lat = _end_to_end_ns(op_log)
+    assert lat > 0, f"expected positive end-to-end latency; got {lat}"
+
+    occ = _engine_occupancy_ns(op_log)
+    assert isinstance(occ, dict)
+    assert set(occ.keys()) >= _EXPECTED_ENGINES, (
+        f"engine_occupancy_ns missing required engines; "
+        f"got {set(occ.keys())}"
+    )
+    assert occ["pe_gemm"] > 0, (
+        f"expected pe_gemm occupancy > 0; got {occ['pe_gemm']}"
+    )
+
+
+def test_run_panel_returns_latency_and_engine_occupancy(monkeypatch):
+    """``_run_panel`` must include ``latency_ns`` + ``engine_occupancy_ns``
+    alongside ``op_log_summary`` so sweep.json carries them for the
+    comparative plot script. Uses monkeypatch to swap the Case 2 panel's
+    T_q/S_kv to smoke sizes for speed.
+    """
+    import kernbench.benches.gqa_helpers.long_ctx.gqa_prefill_long_ctx_4cases as mod
+
+    orig_kind, orig_params = mod._PANEL_DISPATCH[_CASE2_PANEL]
+    fast_params = {**orig_params, "T_q": _SMOKE_T_Q, "S_kv": _SMOKE_S_KV}
+    monkeypatch.setitem(
+        mod._PANEL_DISPATCH, _CASE2_PANEL, (orig_kind, fast_params),
+    )
+
+    row = mod._run_panel(_CASE2_PANEL, str(TOPOLOGY_DEFAULT))
+    assert "op_log_summary" in row
+    assert "latency_ns" in row, (
+        f"_run_panel row missing latency_ns; keys={sorted(row.keys())}"
+    )
+    assert row["latency_ns"] > 0
+    assert "engine_occupancy_ns" in row, (
+        f"_run_panel row missing engine_occupancy_ns; "
+        f"keys={sorted(row.keys())}"
+    )
+    assert isinstance(row["engine_occupancy_ns"], dict)
+    assert set(row["engine_occupancy_ns"].keys()) >= _EXPECTED_ENGINES