paper(ipcq): add latency-model-style flow + stacked architecture diagrams

Adds two new IPCQ architecture views in the latency_model.png aesthetic:
a 2x2 flow figure (one per case) and a 4x1 stacked figure for taller
horizontal panels. Refactors per-case panel drawing into shared helpers
and disambiguates the flit labels (HMQ desc -> Hd so it does not
collide with data D). Two-row legend keys.

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

scripts/paper/paper_plot_ipcq_alternatives_comparison.py

@@ -238,6 +238,267 @@ def _plot_architecture() -> Path:
     return out
 
 
+# Module-level flow-diagram helpers and constants.
+_FLOW_DATA = "#6699CC"   # blue flit (data)
+_FLOW_CTRL = "#ED7D65"   # orange flit (control)
+_FLOW_CRED = "#7BB661"   # green (credit / ack)
+
+
+def _flow_swim(ax, x, y, w, h, label, sub=None):
+    ax.add_patch(mpatches.FancyBboxPatch(
+        (x, y), w, h,
+        boxstyle="round,pad=0.02,rounding_size=0.08",
+        facecolor="white", edgecolor="#3a3a3a", linewidth=1.6))
+    ax.text(x + w / 2, y + h / 2 + (0.20 if sub else 0),
+            label, ha="center", va="center",
+            fontsize=10.5, fontweight="bold", color="#111")
+    if sub:
+        ax.text(x + w / 2, y + h / 2 - 0.35, sub,
+                ha="center", va="center",
+                fontsize=8, color="#555", fontstyle="italic")
+
+
+def _flow_flit(ax, x, y, color, letter="", size=0.32):
+    ax.add_patch(mpatches.FancyBboxPatch(
+        (x - size / 2, y - size / 2), size, size,
+        boxstyle="round,pad=0.01,rounding_size=0.04",
+        facecolor=color, edgecolor="#222", linewidth=0.8))
+    if letter:
+        ax.text(x, y, letter, ha="center", va="center",
+                fontsize=8.5, fontweight="bold", color="white")
+
+
+def _flow_wire(ax, x1, x2, y, color="#888"):
+    ax.plot([x1, x2], [y, y], color=color, lw=1.0, zorder=0)
+
+
+def _flow_dot(ax, x, y, color):
+    ax.add_patch(mpatches.Circle((x, y), 0.13,
+                                 facecolor=color, edgecolor="#333",
+                                 linewidth=0.8, zorder=4))
+
+
+def _flow_label(ax, x, y, text, color="#222", fontsize=9):
+    ax.text(x, y, text, ha="center", va="center",
+            fontsize=fontsize, color=color, fontweight="bold")
+
+
+def _flow_lead(ax, x_tail, y_tail, x_head, y_head, color="#777"):
+    ax.plot([x_tail, x_head], [y_tail, y_head],
+            color=color, lw=0.8, ls=":", zorder=2)
+
+
+def _flow_setup(ax, kind):
+    ax.set_xlim(0, 20); ax.set_ylim(0, 10); ax.axis("off")
+    ax.add_patch(mpatches.FancyBboxPatch(
+        (0.15, 0.15), 19.7, 9.7,
+        boxstyle="round,pad=0.05,rounding_size=0.12",
+        facecolor="white", edgecolor=_COLOR[kind],
+        linewidth=2.2, zorder=0))
+    ax.text(10, 0.65, _TITLE[kind],
+            ha="center", va="center", fontsize=11.5,
+            fontweight="bold",
+            color=("#1B5E20" if kind == "ipcq" else "white"),
+            bbox=dict(
+                facecolor=("#E8F5E9" if kind == "ipcq" else _COLOR[kind]),
+                edgecolor=("#2E7D32" if kind == "ipcq" else _COLOR[kind]),
+                boxstyle="round,pad=0.40", linewidth=1.5))
+
+
+def _draw_flow_case(ax, kind):
+    """Draw one Sender → NoC → Receiver flow panel for `kind`."""
+    _flow_setup(ax, kind)
+    if kind == "doorbell":
+        _draw_case_doorbell(ax)
+    elif kind == "hmq":
+        _draw_case_hmq(ax)
+    elif kind == "rdma":
+        _draw_case_rdma(ax)
+    elif kind == "ipcq":
+        _draw_case_ipcq(ax)
+
+
+def _draw_case_doorbell(ax):
+    _flow_swim(ax, 0.7, 4.4, 2.8, 2.0, "Sender\nkernel")
+    _flow_swim(ax, 7.4, 4.4, 3.0, 2.0, "NoC / UCIe", sub="2 DMA tx")
+    _flow_swim(ax, 14.4, 4.4, 3.0, 2.0, "Doorbell reg\n+ landing TCM")
+    _flow_swim(ax, 17.6, 6.7, 2.0, 1.7, "Receiver\n(poll)", sub="poll loop")
+    _flow_wire(ax, 3.5, 7.4, 5.7); _flow_wire(ax, 10.4, 14.4, 5.7)
+    for x in [4.0, 4.7, 5.4]:
+        _flow_flit(ax, x, 5.7, _FLOW_DATA, "D")
+    for x in [11.0, 11.7]:
+        _flow_flit(ax, x, 5.7, _FLOW_DATA, "D")
+    _flow_wire(ax, 3.5, 7.4, 5.1); _flow_wire(ax, 10.4, 14.4, 5.1)
+    _flow_flit(ax, 4.0, 5.1, _FLOW_CTRL, "F")
+    _flow_flit(ax, 4.7, 5.1, _FLOW_CTRL, "DB")
+    _flow_flit(ax, 11.0, 5.1, _FLOW_CTRL, "DB")
+    for y in (8.0, 7.5, 7.0):
+        ax.annotate("", xy=(15.5, y), xytext=(17.6, y),
+                    arrowprops=dict(arrowstyle="->", color="#aa3a30",
+                                    lw=1.0, ls=":"))
+    _flow_dot(ax, 5.0, 3.5, "#c5a35a")
+    _flow_lead(ax, 5.0, 3.6, 5.0, 4.4)
+    _flow_label(ax, 5.0, 3.2, "fence + 2 DMA writes", color="#5a3e20")
+    _flow_dot(ax, 17.0, 8.8, "#7BB661")
+    _flow_lead(ax, 17.0, 8.7, 17.0, 8.0)
+    _flow_label(ax, 17.0, 9.1, "receiver polling loop", color="#2E7D32")
+
+
+def _draw_case_hmq(ax):
+    _flow_swim(ax, 0.7, 4.4, 2.8, 2.0, "Sender CPU")
+    _flow_swim(ax, 4.6, 6.6, 2.8, 1.8, "HMQ engine", sub="descriptor")
+    _flow_swim(ax, 4.6, 3.6, 2.8, 1.8, "PE_DMA", sub="data path")
+    _flow_swim(ax, 9.6, 4.4, 3.0, 2.0, "NoC / UCIe", sub="2 paths")
+    _flow_swim(ax, 14.4, 6.6, 2.8, 1.8, "Peer HMQ")
+    _flow_swim(ax, 14.4, 3.6, 2.8, 1.8, "Recv TCM")
+    _flow_swim(ax, 17.6, 4.4, 2.0, 2.0, "Receiver\nCPU")
+    _flow_wire(ax, 3.5, 4.6, 7.5); _flow_wire(ax, 7.4, 9.6, 7.5)
+    _flow_wire(ax, 12.6, 14.4, 7.5); _flow_wire(ax, 17.2, 17.6, 5.4)
+    _flow_flit(ax, 4.0, 7.5, _FLOW_CTRL, "Hd")
+    _flow_flit(ax, 8.5, 7.5, _FLOW_CTRL, "Hd")
+    _flow_flit(ax, 13.5, 7.5, _FLOW_CTRL, "Hd")
+    _flow_wire(ax, 3.5, 4.6, 4.5); _flow_wire(ax, 7.4, 9.6, 4.5)
+    _flow_wire(ax, 12.6, 14.4, 4.5)
+    for x in [4.0, 8.0, 8.7, 13.5]:
+        _flow_flit(ax, x, 4.5, _FLOW_DATA, "D")
+    _flow_dot(ax, 11.1, 9.0, "#c5a35a")
+    _flow_lead(ax, 11.1, 8.9, 11.1, 7.6)
+    _flow_label(ax, 11.1, 9.25, "duplicated control + data paths",
+                color="#5a3e20")
+
+
+def _draw_case_rdma(ax):
+    _flow_swim(ax, 0.7, 4.4, 2.8, 2.0, "Sender\nkernel")
+    _flow_swim(ax, 4.8, 4.4, 2.8, 2.0, "WQE +\nPE_DMA")
+    _flow_swim(ax, 8.5, 4.4, 3.0, 2.0, "NoC / UCIe")
+    _flow_swim(ax, 13.0, 4.4, 3.0, 2.0, "CQ + landing\nTCM",
+               sub="auto-CQE")
+    _flow_swim(ax, 17.0, 6.7, 2.5, 1.7, "Receiver\n(poll CQ)")
+    _flow_wire(ax, 3.5, 4.8, 5.7)
+    _flow_flit(ax, 4.0, 5.7, _FLOW_CTRL, "W")
+    _flow_wire(ax, 7.6, 8.5, 5.4); _flow_wire(ax, 11.5, 13.0, 5.4)
+    _flow_flit(ax, 8.1, 5.4, _FLOW_DATA, "D")
+    _flow_flit(ax, 9.2, 5.4, _FLOW_DATA, "D")
+    _flow_flit(ax, 9.9, 5.4, _FLOW_CTRL, "Q")
+    _flow_flit(ax, 11.0, 5.4, _FLOW_DATA, "D")
+    _flow_flit(ax, 12.0, 5.4, _FLOW_CTRL, "Q")
+    for y in (8.0, 7.5, 7.0):
+        ax.annotate("", xy=(14.5, y), xytext=(17.0, y),
+                    arrowprops=dict(arrowstyle="->", color="#7c1b78",
+                                    lw=1.0, ls=":"))
+    _flow_dot(ax, 11.0, 3.5, "#c5a35a")
+    _flow_lead(ax, 11.0, 3.6, 11.0, 4.4)
+    _flow_label(ax, 11.0, 3.2, "CQE rides with DMA payload",
+                color="#5a3e20")
+    _flow_dot(ax, 18.0, 8.8, "#7BB661")
+    _flow_lead(ax, 18.0, 8.7, 18.0, 8.0)
+    _flow_label(ax, 18.0, 9.1, "CQ poll / IRQ", color="#2E7D32")
+
+
+def _draw_case_ipcq(ax):
+    _flow_swim(ax, 0.7, 4.4, 2.8, 2.0, "Sender PE\n(tl.send)")
+    _flow_swim(ax, 4.5, 6.6, 3.0, 1.8, "PE_IPCQ", sub="ring push")
+    _flow_swim(ax, 4.5, 3.6, 3.0, 1.8, "PE_DMA", sub="data push")
+    _flow_swim(ax, 9.4, 4.4, 3.0, 2.0, "NoC / UCIe", sub="dir-addr")
+    _flow_swim(ax, 13.8, 6.6, 3.0, 1.8, "PE_IPCQ",
+               sub="tail++ , credit")
+    _flow_swim(ax, 13.8, 3.6, 3.0, 1.8, "PE_DMA",
+               sub="landing TCM")
+    _flow_swim(ax, 17.6, 4.4, 2.0, 2.0, "Receiver\nPE\n(tl.recv)")
+    _flow_wire(ax, 3.5, 4.5, 7.5); _flow_wire(ax, 7.5, 9.4, 7.5)
+    _flow_wire(ax, 12.4, 13.8, 7.5); _flow_wire(ax, 16.8, 17.6, 5.4)
+    _flow_flit(ax, 4.0, 7.5, _FLOW_CTRL, "M")
+    _flow_flit(ax, 8.4, 7.5, _FLOW_CTRL, "M")
+    _flow_flit(ax, 13.0, 7.5, _FLOW_CTRL, "M")
+    _flow_wire(ax, 3.5, 4.5, 4.5); _flow_wire(ax, 7.5, 9.4, 4.5)
+    _flow_wire(ax, 12.4, 13.8, 4.5)
+    _flow_flit(ax, 4.0, 4.5, _FLOW_DATA, "D")
+    _flow_flit(ax, 8.4, 4.5, _FLOW_DATA, "D")
+    _flow_flit(ax, 11.0, 4.5, _FLOW_DATA, "D")
+    _flow_flit(ax, 13.0, 4.5, _FLOW_DATA, "D")
+    ax.annotate("", xy=(4.5, 9.0), xytext=(16.8, 9.0),
+                arrowprops=dict(arrowstyle="->", color=_FLOW_CRED,
+                                lw=1.6, ls=(0, (5, 3))))
+    _flow_label(ax, 10.5, 9.35, "credit piggyback (no separate ACK)",
+                color=_FLOW_CRED)
+    _flow_dot(ax, 6.0, 2.6, "#c5a35a")
+    _flow_lead(ax, 6.0, 2.7, 6.0, 3.6)
+    _flow_label(ax, 6.0, 2.3, "one in-line ring push + one DMA",
+                color="#5a3e20")
+
+
+def _flow_legend(fig, leg_y=0.02):
+    """Two-row legend so the long control-flit key doesn't overlap
+    the shorter data/credit entries."""
+    row_top = [
+        (_FLOW_DATA, "Data flit  D = payload"),
+        (_FLOW_CRED, "Credit / piggyback return"),
+    ]
+    row_bottom = [
+        (_FLOW_CTRL, "Control flit:  W=WQE  ·  Q=CQE  ·  M=ring meta"
+                     "  ·  DB=doorbell  ·  F=fence  ·  Hd=HMQ desc"),
+    ]
+
+    def _emit(items, y):
+        lx = 0.04
+        for color, text in items:
+            fig.patches.append(mpatches.FancyBboxPatch(
+                (lx, y), 0.020, 0.022,
+                boxstyle="round,pad=0.005,rounding_size=0.005",
+                facecolor=color, edgecolor="#222", linewidth=0.6,
+                transform=fig.transFigure))
+            fig.text(lx + 0.025, y + 0.011, text,
+                     ha="left", va="center", fontsize=10,
+                     fontweight="bold", color="#222")
+            lx += 0.32
+
+    _emit(row_top,    y=leg_y + 0.030)
+    _emit(row_bottom, y=leg_y)
+
+
+# ─── Diagram 1b : 4-case flow-style architecture (latency_model look)
+def _plot_architecture_flow() -> Path:
+    """4 panels arranged 2×2 — each a horizontal Sender → NoC → Receiver
+    flow with small flit-style packets, dot-marker annotations and a
+    bottom name plate. Aesthetic mirrors latency_model.png."""
+    fig, axes = plt.subplots(2, 2, figsize=(18.0, 11.5))
+    _draw_flow_case(axes[0, 0], "doorbell")
+    _draw_flow_case(axes[0, 1], "hmq")
+    _draw_flow_case(axes[1, 0], "rdma")
+    _draw_flow_case(axes[1, 1], "ipcq")
+    fig.suptitle(
+        "Per-send architecture  =  $\\Sigma$ control overhead  +  "
+        "data-path DMA  +  receiver wake",
+        fontsize=14.5, fontweight="bold", y=0.995)
+    _flow_legend(fig, leg_y=0.015)
+    fig.tight_layout(rect=(0, 0.07, 1, 0.96))
+    out = _OUT_DIR / "ipcq_alternatives_architecture_flow.png"
+    fig.savefig(out, dpi=150, bbox_inches="tight")
+    plt.close(fig)
+    return out
+
+
+# ─── Diagram 1c : 4-case flow-style architecture — stacked 4×1 ──────
+def _plot_architecture_stacked() -> Path:
+    """Same per-case flow content as the 2×2 figure, but stacked one
+    case per row (4 rows × 1 column). Wider panels give each case more
+    horizontal room."""
+    fig, axes = plt.subplots(4, 1, figsize=(16.0, 19.0))
+    for ax, kind in zip(axes, ["doorbell", "hmq", "rdma", "ipcq"]):
+        _draw_flow_case(ax, kind)
+    fig.suptitle(
+        "Per-send architecture  =  $\\Sigma$ control overhead  +  "
+        "data-path DMA  +  receiver wake",
+        fontsize=14.5, fontweight="bold", y=0.997)
+    _flow_legend(fig, leg_y=0.010)
+    fig.tight_layout(rect=(0, 0.05, 1, 0.97))
+    out = _OUT_DIR / "ipcq_alternatives_architecture_stacked.png"
+    fig.savefig(out, dpi=150, bbox_inches="tight")
+    plt.close(fig)
+    return out
+
+
+
 # ─── Diagram 2 : Per-send latency breakdown (vertical stacks) ───────
 def _plot_latency() -> Path:
     # (label, short, illustrative cycles, fill)
@@ -782,7 +1043,9 @@ def _plot_control_data_overlay() -> Path:
 
 def main() -> None:
     _OUT_DIR.mkdir(parents=True, exist_ok=True)
-    for plot in (_plot_architecture, _plot_latency,
+    for plot in (_plot_architecture, _plot_architecture_flow,
+                 _plot_architecture_stacked,
+                 _plot_latency,
                  _plot_opcounts, _plot_sequence,
                  _plot_decision_matrix,
                  _plot_control_data_overlay):