paper: §2 platform deep-edit + §1 GQA framing + figure regen

§2 (KernBench Platform):
- Fix Table 2 HBM aggregate BW (1024 → 2048 GB/s); drop stale
  hbm_total_bw_gbs from topology.yaml (never read by sim_engine)
- Split PE_CPU / PE_SCHED fixed-cost row; disambiguate from the
  40-cycle command-dispatch FIXED term
- §2.2 two-pass: expand to describe Pass 1 timing and Pass 2 data
  data-correctness path
- §2.3 dispatch model: add motivation sentence for the
  descriptor-size linear form
- §2.4 Accuracy: reorder GEMM → All-reduce → Probe →
  Simplifications; drop FSIM aside (already covered by §4 Fig 5
  caption); soften 'every ns' → 'every modeled latency
  contribution'
- §2.5 HW config: add 64 TFLOP/s vs 2048 GB/s (~31 FLOP/byte)
  balance-point intuition and forward pointer to §5 GQA decode
- Fig 1 caption: separate illustrative topology from experimental
  configuration

§1: tighten GQA-as-primary-bandwidth-bottleneck framing.

Figures: regenerate SIP / CUBE architecture (SVG sources + PDF +
generator scripts).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

scripts/paper/paper_sip_architecture.py

@@ -0,0 +1,91 @@
+"""Re-emit sip_view.svg in an academic palette (white background, black
+strokes) and convert it to PDF for the 1H-codesign-paper Figure 1.
+
+Source of truth: docs/diagrams/sip_view.svg (generated by
+src/kernbench/topology/visualizer.py, dark-ish theme).
+
+Treatment per user request: keep the layout intact, just force a white
+canvas and turn every stroke black; promote faint label text to black so
+all annotations stay legible on white.
+"""
+from __future__ import annotations
+
+import re
+import shutil
+import subprocess
+from pathlib import Path
+
+REPO = Path(__file__).resolve().parents[2]
+SRC_SVG = REPO / "docs" / "diagrams" / "sip_view.svg"
+OUT_DIR = REPO / "docs" / "report" / "1H-codesign-paper" / "figures"
+OUT_SVG = OUT_DIR / "sip_architecture.svg"
+OUT_PDF = OUT_DIR / "sip_architecture.pdf"
+
+# Drop the "SIP VIEW" title; the figure caption already names the level.
+TITLE_REMOVE: list[tuple[str, str]] = [
+    ('  <text x="324" y="18" text-anchor="middle" font-family="monospace" '
+     'font-size="14" font-weight="bold" fill="#1e293b">SIP VIEW</text>\n',
+     ''),
+]
+
+COLOR_MAP: list[tuple[str, str]] = [
+    # canvas background slate-50 -> pure white
+    ('fill="#f8fafc"', 'fill="#ffffff"'),
+    # all strokes -> black
+    ('stroke="#3b82f6"', 'stroke="#000000"'),  # UCIe mesh blue lines
+    ('stroke="#475569"', 'stroke="#000000"'),  # cube block borders
+    ('stroke="#0ea5e9"', 'stroke="#000000"'),  # I/O sky-blue lines
+    # link annotation text slate-500 -> black for readability
+    ('fill="#64748b"', 'fill="#000000"'),
+]
+
+# Font-size bumps so labels survive LaTeX \linewidth scaling at the
+# half-text-width subfigure. CUBE block labels overflow the 48px block
+# rects, which is acceptable here.
+FONT_MAP: dict[str, str] = {
+    "7":  "10",
+    "14": "17",
+}
+
+# Tighten whitespace: cube grid occupies x=[84,564], y=[128,520]; IO
+# chiplet sits around y~50. With the title removed, crop top to y=40 so
+# the IO chiplet keeps a small headroom. Crop 70px on each side and 113
+# px from bottom.
+LAYOUT_FIXUP: list[tuple[str, str]] = [
+    ('<svg xmlns="http://www.w3.org/2000/svg" width="648" height="648" '
+     'viewBox="0 0 648 648">',
+     '<svg xmlns="http://www.w3.org/2000/svg" width="508" height="495" '
+     'viewBox="70 40 508 495">'),
+]
+
+
+def _bump_font(m: re.Match) -> str:
+    return f'font-size="{FONT_MAP.get(m.group(1), m.group(1))}"'
+
+
+def main() -> None:
+    if not SRC_SVG.exists():
+        raise SystemExit(f"source SVG missing: {SRC_SVG}")
+    rsvg = shutil.which("rsvg-convert")
+    if rsvg is None:
+        raise SystemExit("rsvg-convert not found (brew install librsvg)")
+
+    svg = SRC_SVG.read_text(encoding="utf-8")
+    for old, new in TITLE_REMOVE + COLOR_MAP + LAYOUT_FIXUP:
+        if old not in svg:
+            print(f"warn: pattern not present in source SVG: {old}")
+        svg = svg.replace(old, new)
+    svg = re.sub(r'font-size="(\d+)"', _bump_font, svg)
+
+    OUT_DIR.mkdir(parents=True, exist_ok=True)
+    OUT_SVG.write_text(svg, encoding="utf-8")
+    subprocess.run(
+        [rsvg, "-f", "pdf", "-o", str(OUT_PDF), str(OUT_SVG)],
+        check=True,
+    )
+    print(f"wrote {OUT_SVG.relative_to(REPO)}")
+    print(f"wrote {OUT_PDF.relative_to(REPO)}")
+
+
+if __name__ == "__main__":
+    main()