[Kernel] [V1] Fix performance regression for triton unified attention (vllm-project#18161)

Signed-off-by: Thomas Parnell <[email protected]>
Co-authored-by: Lucas Wilkinson <[email protected]>
Signed-off-by: Yuqi Zhang <[email protected]>

Author: 2 people

vllm/attention/ops/triton_unified_attention.py

@@ -56,11 +56,11 @@ def kernel_unified_attention_2d(
     stride_k_cache_0: tl.int64,  # int
     stride_k_cache_1: tl.int64,  # int
     stride_k_cache_2: tl.int64,  # int
-    stride_k_cache_3: tl.int64,  # int
+    stride_k_cache_3: tl.constexpr,  # int
     stride_v_cache_0: tl.int64,  # int
     stride_v_cache_1: tl.int64,  # int
     stride_v_cache_2: tl.int64,  # int
-    stride_v_cache_3: tl.int64,  # int
+    stride_v_cache_3: tl.constexpr,  # int
     query_start_len_ptr,  # [num_seqs+1]
     BLOCK_Q: tl.constexpr,  # int
     num_seqs: tl.int32,

vllm/v1/attention/backends/triton_attn.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 """Attention layer with PagedAttention and Triton prefix prefill."""
-from typing import Any, Optional
+from typing import TYPE_CHECKING, Any, Optional
 
 import torch
 
@@ -12,10 +12,23 @@
 from vllm.platforms import current_platform
 from vllm.v1.attention.backends.flash_attn import (
     FlashAttentionMetadata, FlashAttentionMetadataBuilder)
+from vllm.v1.kv_cache_interface import AttentionSpec
+from vllm.v1.worker.block_table import BlockTable
+
+if TYPE_CHECKING:
+    from vllm.v1.worker.gpu_model_runner import GPUModelRunner
 
 logger = init_logger(__name__)
 
 
+class TritonAttentionMetadataBuilder(FlashAttentionMetadataBuilder):
+
+    def __init__(self, runner: "GPUModelRunner", kv_cache_spec: AttentionSpec,
+                 block_table: BlockTable):
+        super().__init__(runner, kv_cache_spec, block_table)
+        self.aot_schedule = False
+
+
 class TritonAttentionBackend(AttentionBackend):
 
     accept_output_buffer: bool = True
@@ -52,8 +65,8 @@ def use_cascade_attention(*args, **kwargs) -> bool:
         return False
 
     @staticmethod
-    def get_builder_cls() -> type["FlashAttentionMetadataBuilder"]:
-        return FlashAttentionMetadataBuilder
+    def get_builder_cls() -> type["TritonAttentionMetadataBuilder"]:
+        return TritonAttentionMetadataBuilder
 
 
 class TritonAttentionImpl(AttentionImpl):