[ROCM] Optimize ROCM_AITER_FA spec decode eagle performance (#34541)

Signed-off-by: jennyyyyzhen <yzhen@hmc.edu>

vllm/v1/attention/backends/rocm_aiter_fa.py

@@ -396,8 +396,7 @@ class AiterFlashAttentionMetadata:
 class AiterFlashAttentionMetadataBuilder(
     AttentionMetadataBuilder[AiterFlashAttentionMetadata]
 ):
-    _cudagraph_support = AttentionCGSupport.UNIFORM_SINGLE_TOKEN_DECODE
+    _cudagraph_support = AttentionCGSupport.UNIFORM_BATCH
-    reorder_batch_threshold: int = 1
 
     def __init__(
         self,
@@ -422,6 +421,7 @@ class AiterFlashAttentionMetadataBuilder(
         # populated on first build() call.
         self.aot_sliding_window: tuple[int, int] | None = None
         self.total_tokens: int = 0
+        self._init_reorder_batch_threshold(1, supports_spec_as_decode=True)
 
         sliding_window_configs: set[tuple[int, int] | None] = set()
         layers = get_layers_from_vllm_config(self.vllm_config, Attention)
@@ -466,6 +466,7 @@ class AiterFlashAttentionMetadataBuilder(
         common_attn_metadata: CommonAttentionMetadata,
         fast_build: bool = False,
     ) -> "AiterFlashAttentionMetadata":
+        assert self.reorder_batch_threshold is not None
         split_ret = split_decodes_prefills_and_extends(
             common_attn_metadata,
             decode_threshold=self.reorder_batch_threshold,
@@ -677,6 +678,53 @@ class AiterFlashAttentionMetadataBuilder(
         )
         return attn_metadata
 
+    def build_for_drafting(
+        self,
+        common_attn_metadata: CommonAttentionMetadata,
+        draft_index: int,
+    ) -> AiterFlashAttentionMetadata:
+        """
+        Build attention metadata for draft model without CPU-GPU sync.
+
+        During EAGLE drafting all requests are uniform decodes, so we can
+        skip split_decodes_prefills_and_extends() and avoid all .cpu() /
+        .item() calls that would otherwise break CUDA graph capture.
+        """
+        num_reqs = common_attn_metadata.num_reqs
+        num_tokens = common_attn_metadata.num_actual_tokens
+
+        decode_metadata = AiterFlashAttentionDecodeMetadata(
+            max_query_len=common_attn_metadata.max_query_len,
+            min_query_len=common_attn_metadata.max_query_len,  # uniform batch
+            max_seq_len=common_attn_metadata.max_seq_len,
+            query_start_loc=common_attn_metadata.query_start_loc,
+        )
+
+        return AiterFlashAttentionMetadata(
+            num_actual_tokens=num_tokens,
+            num_actual_kv_tokens=0,  # not used in unified_attention path
+            max_query_len=common_attn_metadata.max_query_len,
+            query_start_loc=common_attn_metadata.query_start_loc,
+            max_seq_len=common_attn_metadata.max_seq_len,
+            seq_lens=common_attn_metadata.seq_lens,
+            block_table=common_attn_metadata.block_table_tensor,
+            slot_mapping=common_attn_metadata.slot_mapping,
+            num_decodes=num_reqs,
+            num_decode_tokens=num_tokens,
+            num_prefills=0,
+            num_prefill_tokens=0,
+            num_extends=0,
+            num_extend_tokens=0,
+            decode_metadata=decode_metadata,
+            prefill_metadata=None,
+            extend_metadata=None,
+            use_cascade=False,
+            common_prefix_len=0,
+            total_tokens=self.total_tokens,
+            k_scale=self.scale,
+            v_scale=self.scale,
+        )
+
     def use_cascade_attention(self, *args, **kwargs) -> bool:
         return False