gqa(adr-0065): N4 — opt2 numeric parity vs opt3 in data mode

composite() now awaits its prologue recipe operands (e.g. the score tile from a prior #1 composite) so the cross-composite RAW (#1 writes scores -> #2 reads them) is ordered: #2 waits for #1 before reading. With N1-N3, opt2 (the two-composite recipe kernel) now computes a correct attention output end-to-end in data mode and matches opt3 within fp tolerance.

Scope note: the e2e opt2<->opt3 parity holds in the near-uniform-attention regime (small inputs). kernbench's K reshape-as-transpose (opt3 kernel docstring: 'correct for zero/symmetric inputs') makes opt2's per-sub-tile K interpretation differ from opt3's single-tile one for high-contrast inputs — a pre-existing kernbench limitation shared by both kernels, not the recipe. The EXACT recipe math (online-softmax merge m/l + O = O*corr + P@V) is verified non-trivially and exactly by N2/N3 (test_recipe_data_mode.py). Suite 817 pass / 3 pre-existing fail.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/kernbench/triton_emu/tl_context.py

@@ -802,6 +802,13 @@ class TLContext:
         rw_handles: tuple[TensorHandle, ...] = ()
         if prologue:
             prologue_ops, primary_out, rw_handles = self._expand_prologue(prologue)
+            # Await any prologue operand that is a pending composite output
+            # (e.g. the score tile produced by a prior #1 composite) so it is
+            # computed before this composite reads it (RAW across composites).
+            for _item in prologue:
+                self._await_pending(
+                    *[v for v in _item.values() if isinstance(v, TensorHandle)]
+                )
 
         # Auto-bind (D6.6): the recipe's primary output fills the head GEMM's
         # `a` unless the kernel supplied `a` explicitly (then it's ambiguous).