Enable CK Attention for Navi31

vllm/attention/backends/rocm_flash_attn.py

@@ -18,6 +18,15 @@
 if TYPE_CHECKING:
     from vllm.worker.model_runner import ModelInputForGPUWithSamplingMetadata
 
+try:
+    from flash_attn import flash_attn_func  # noqa: F401
+    from flash_attn import flash_attn_varlen_func  # noqa: F401
+    flash_attn_available = True
+except ModuleNotFoundError:
+    flash_attn_func = None
+    flash_attn_varlen_func = None
+    flash_attn_available = False
+
 logger = init_logger(__name__)
 
 _PARTITION_SIZE_ROCM = 256
@@ -509,16 +518,18 @@ def __init__(
                                "FA backend instead by setting the env var "
                                "`VLLM_USE_TRITON_FLASH_ATTN=0`")
         else:
-            # if not using triton, navi3x/navi21/navi10 do not use flash-attn
+            # if not using triton, navi21/navi10 do not use flash-attn
             # either
             if not current_platform.has_device_capability(90):
                 self.use_naive_attn = True
             else:
-                try:
-                    from flash_attn import flash_attn_varlen_func  # noqa: F401
-                    self.attn_func = flash_attn_varlen_func
+                if flash_attn_available:
+                    if current_platform.has_device_capability(110):
+                        self.attn_func = _ck_attention
+                    else:
+                        self.attn_func = flash_attn_varlen_func
                     logger.debug("Using CK FA in ROCmBackend")
-                except ModuleNotFoundError:
+                else:
                     self.use_naive_attn = True
 
             if self.use_naive_attn:
@@ -722,20 +733,39 @@ def forward(
                         attn_masks,
                     )
                 else:
-                    out = self.attn_func(
-                        q=query,
-                        k=key,
-                        v=value,
-                        cu_seqlens_q=query_seq_start_loc,
-                        cu_seqlens_k=key_seq_start_loc,
-                        max_seqlen_q=prefill_meta.max_prefill_seq_len,
-                        max_seqlen_k=key_max_seq_len,
-                        softmax_scale=self.scale,
-                        causal=True,
-                        window_size=self.sliding_window,
-                        alibi_slopes=self.alibi_slopes,
-                        softcap=self.logits_soft_cap,
-                    )
+                    if _ON_NAVI:
+                        query = query.view(
+                            (num_prefill_tokens, self.num_heads, -1))
+                        key = key.view(
+                            (num_prefill_tokens, self.num_kv_heads, -1))
+                        value = value.view(
+                            (num_prefill_tokens, self.num_kv_heads, -1))
+                        out = self.attn_func(
+                            query,
+                            key,
+                            value,
+                            prefill_meta.seq_lens,
+                            num_prefill_tokens,
+                            self.num_heads,
+                            self.head_size,
+                            self.scale,
+                            attn_masks,
+                        )
+                    else:
+                        out = self.attn_func(
+                            q=query,
+                            k=key,
+                            v=value,
+                            cu_seqlens_q=query_seq_start_loc,
+                            cu_seqlens_k=key_seq_start_loc,
+                            max_seqlen_q=prefill_meta.max_prefill_seq_len,
+                            max_seqlen_k=key_max_seq_len,
+                            softmax_scale=self.scale,
+                            causal=True,
+                            window_size=self.sliding_window,
+                            alibi_slopes=self.alibi_slopes,
+                            softcap=self.logits_soft_cap,
+                        )
 
                 # common code for prefill
                 assert output[:num_prefill_tokens].shape == out.shape
@@ -891,6 +921,36 @@ def _sdpa_attention(
     return output
 
 
+def _ck_attention(
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    seq_lens: List[int],
+    num_tokens: int,
+    num_heads: int,
+    head_size: int,
+    scale: float,
+    attn_masks: Optional[List[torch.Tensor]] = None,
+) -> torch.Tensor:
+    start = 0
+    output = torch.empty((num_tokens, num_heads, head_size),
+                         dtype=query.dtype,
+                         device=query.device)
+
+    for i, seq_len in enumerate(seq_lens):
+        end = start + seq_len
+        sub_out = flash_attn_func(query[start:end, :, :].unsqueeze(0),
+                                  key[start:end, :, :].unsqueeze(0),
+                                  value[start:end, :, :].unsqueeze(0),
+                                  dropout_p=0.0,
+                                  softmax_scale=scale,
+                                  causal=attn_masks is None)
+        output[start:end, :, :] = sub_out.squeeze(0)
+        start = end
+
+    return output
+
+
 def _use_rocm_custom_paged_attention(qtype: torch.dtype, head_size: int,
                                      block_size: int, gqa_ratio: int,
                                      max_seq_len: int) -> bool: