bench(milestone-gqa-headline): scale single_kv_group prefill panel T_q=S_kv=1K (scratch budget)

The headline T_q=S_kv=32K target overflows the 1 MB per-PE scratch
pool: at T_q_local=4K and tile_s=1024 the scores matrix alone is 8 MB.
The prefill kernel's bootstrap section also leaves K_t/V_t/scores/
exp_scores persistent (outside tl.scratch_scope), inflating baseline.

Scale-down to T_q=S_kv=1K (T_q_local=128, fits comfortably) preserves
the C=8 + P=8 architecture demonstration; the true LLaMA 32K headline
awaits a future increment to add Q-axis tiling and tighten bootstrap
scratch discipline.

Verified end-to-end: kernbench run --bench milestone-gqa-headline now
produces sweep.json with all 5 panels. The new panel shows
ipcq_copy=896 (matches (C-1)·n_tiles·2·C·P = 7·1·2·8·8) and
dma_write=64 (one per PE, head-parallel + intra-CUBE PE-SP).

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

src/kernbench/benches/1H_milestone_output/gqa_headline/sweep.json

@@ -3,6 +3,7 @@
   "panels": [
     "single_user_prefill_gqa",
     "multi_user_prefill_gqa",
+    "single_kv_group_prefill_gqa_c8_p8",
     "single_user_decode_gqa",
     "multi_user_decode_gqa"
   ],
@@ -39,6 +40,21 @@
         "dma_write_count": 4
       }
     },
+    {
+      "panel": "single_kv_group_prefill_gqa_c8_p8",
+      "kind": "prefill",
+      "C": 8,
+      "P": 8,
+      "T_q": 1024,
+      "S_kv": 1024,
+      "d_head": 128,
+      "op_log_summary": {
+        "gemm_count": 1024,
+        "ipcq_copy_count": 896,
+        "dma_read_count": 192,
+        "dma_write_count": 64
+      }
+    },
     {
       "panel": "single_user_decode_gqa",
       "kind": "decode",

src/kernbench/benches/milestone_gqa_headline.py

@@ -18,8 +18,10 @@ Panels:
   multi_user_prefill_gqa          : prefill C=4 Ring KV, T_q=4, S_kv=16
   single_kv_group_prefill_gqa_c8_p8: prefill C=8 snake Ring KV +
                                      intra-CUBE PE-SP (all 64 ranks),
-                                     T_q=S_kv=32K, d_head=128 — the
+                                     T_q=S_kv=1K, d_head=128 — the
                                      LLaMA-3.1-70B single-KV-group target
+                                     (scratch-limited; 32K headline awaits
+                                     Q-axis kernel tiling)
   single_user_decode_gqa          : decode C=1, P=8, h_q=8, h_kv=1, S_kv=64
                                     (M-fold + intra-cube row-then-col chain)
   multi_user_decode_gqa           : decode C=4, P=8, h_q=8, h_kv=1, S_kv=128
@@ -70,7 +72,14 @@ _PANEL_DISPATCH: dict[str, tuple[str, dict]] = {
     "multi_user_prefill_gqa":  ("prefill", {"C": 4, "S_kv": _S_KV_PREFILL}),
     "single_kv_group_prefill_gqa_c8_p8": ("prefill", {
         "C": 8, "P": 8,
-        "T_q": 32_768, "S_kv": 32_768,
+        # T_q = S_kv = 1024 (one-shot long-context prefill, scratch-limited).
+        # The bootstrap section of the prefill kernel leaves K_t/V_t/scores/
+        # exp_scores persistent (outside tl.scratch_scope), inflating the
+        # baseline. At T_q=2048 the peak hits ~1.05 MB vs 1.0 MB budget; at
+        # 1024 it fits comfortably. The true LLaMA 32K headline awaits a
+        # future increment to (a) add Q-axis tiling and (b) move the
+        # bootstrap into scratch discipline.
+        "T_q": 1_024, "S_kv": 1_024,
         "d_head": 128,
     }),
     "single_user_decode_gqa":  ("decode",  {"C": 1, "P": 8, "S_kv": 64}),

tests/attention/test_milestone_gqa_single_kv_group_prefill_panel.py

@@ -4,7 +4,8 @@ The single-KV-group (LLaMA-3.1-70B target) prefill panel wires together
 all Increment 1–4 work in a single bench panel:
   - C=8 head-parallel + Ring KV via snake-mapped 2×4 sub-mesh (Inc 1, 3)
   - P=8 intra-CUBE PE-SP (Inc 4)
-  - d_head=128 (LLaMA-3.1-70B), one-shot prefill T_q = S_kv = 32K
+  - d_head=128 (LLaMA-3.1-70B), one-shot prefill T_q = S_kv = 1K
+    (scratch-limited; LLaMA 32K headline awaits Q-axis kernel tiling)
 
 This file verifies the bench-config wiring:
   T1  the new ``single_kv_group_prefill_gqa_c8_p8`` panel is registered
@@ -50,11 +51,13 @@ def test_single_kv_group_prefill_panel_registered():
     """The ``single_kv_group_prefill_gqa_c8_p8`` panel must be in ``_PANELS`` and
     its dispatch entry must have the expected LLaMA-3.1-70B params.
 
-    Headline config:
+    Headline config (scratch-limited; LLaMA 32K headline awaits
+    Q-axis kernel tiling):
       C = 8        (head-parallel, snake sub-mesh)
       P = 8        (intra-CUBE PE-SP)
-      T_q = 32_768 (one-shot long-context prefill)
+      T_q = 1_024  (one-shot long-context prefill)
-      S_kv = 32_768
+      S_kv = 1_024
+      (scratch-limited; LLaMA 32K headline awaits Q-axis kernel tiling)
       d_head = 128 (LLaMA-3.1-70B)
     """
     panel_name = "single_kv_group_prefill_gqa_c8_p8"
@@ -68,11 +71,11 @@ def test_single_kv_group_prefill_panel_registered():
     assert kind == "prefill", f"kind={kind!r}, expected 'prefill'"
     assert params.get("C") == 8, f"C={params.get('C')}, expected 8"
     assert params.get("P") == 8, f"P={params.get('P')}, expected 8"
-    assert params.get("T_q") == 32_768, (
+    assert params.get("T_q") == 1_024, (
-        f"T_q={params.get('T_q')}, expected 32_768"
+        f"T_q={params.get('T_q')}, expected 1_024"
     )
-    assert params.get("S_kv") == 32_768, (
+    assert params.get("S_kv") == 1_024, (
-        f"S_kv={params.get('S_kv')}, expected 32_768"
+        f"S_kv={params.get('S_kv')}, expected 1_024"
     )
     assert params.get("d_head") == 128, (
         f"d_head={params.get('d_head')}, expected 128"