[AMD] Enable preshuffle paged MQA and page_size=64 for NSA indexer (#23562)

python/sglang/srt/layers/attention/nsa/index_buf_accessor.py

@@ -5,10 +5,14 @@ import triton
 import triton.language as tl
 
 from sglang.srt.layers.quantization.fp8_kernel import is_fp8_fnuz
-from sglang.srt.utils import is_hip
+from sglang.srt.utils import get_bool_env_var, is_hip
 
 _is_hip = is_hip()
 _is_fp8_fnuz = is_fp8_fnuz()
+_use_aiter = get_bool_env_var("SGLANG_USE_AITER") and _is_hip
+
+if _use_aiter:
+    from aiter.ops.cache import cp_gather_indexer_k_quant_cache
 
 if TYPE_CHECKING:
     from sglang.srt.mem_cache.memory_pool import NSATokenToKVPool
@@ -163,8 +167,52 @@ class GetS:
 class GetKAndS:
     @classmethod
     def execute(cls, *args, **kwargs):
+        if _use_aiter:
+            return cls.aiter(*args, **kwargs)
         return cls.triton(*args, **kwargs)
 
+    @classmethod
+    def aiter(
+        cls,
+        pool: "NSATokenToKVPool",
+        buf: torch.Tensor,
+        page_indices: torch.Tensor,
+        seq_len_tensor: torch.Tensor,
+        seq_len_sum: int,
+        max_seq_len: int,
+    ):
+        from sglang.srt.layers.quantization.fp8_kernel import fp8_dtype
+
+        page_size = pool.page_size
+        index_head_dim = pool.index_head_dim
+        quant_block_size = pool.quant_block_size
+        scale_elems = index_head_dim // quant_block_size
+
+        kv_cache = buf.view(-1, page_size, index_head_dim + scale_elems * 4).view(
+            fp8_dtype
+        )
+        dst_k = torch.empty(
+            (seq_len_sum, index_head_dim), dtype=torch.uint8, device=buf.device
+        )
+        dst_scale = torch.empty(
+            (seq_len_sum, scale_elems * 4), dtype=torch.uint8, device=buf.device
+        )
+
+        cu_seq_lens = torch.zeros(
+            seq_len_tensor.shape[0] + 1, dtype=torch.int32, device=buf.device
+        )
+        torch.cumsum(seq_len_tensor.to(torch.int32), dim=0, out=cu_seq_lens[1:])
+
+        cp_gather_indexer_k_quant_cache(
+            kv_cache,
+            dst_k.view(fp8_dtype),
+            dst_scale,
+            page_indices.to(torch.int32),
+            cu_seq_lens,
+            preshuffle=True,
+        )
+        return dst_k, dst_scale
+
     @classmethod
     def triton(
         cls,
@@ -366,15 +414,14 @@ def _set_k_and_s_triton(
         raise ValueError(
             f"index_k_scale must be 1D or 2D, got shape {index_k_scale.shape}"
         )
-    if _is_hip:
+    assert buf_numel_per_page == page_size * (128 + 4)
-        assert buf_numel_per_page == 1 * (128 + 4)
-    else:
-        assert buf_numel_per_page == 64 * (128 + 4)
     assert num_tokens_to_write == num_tokens_to_write_ == num_tokens_to_write__
     assert index_head_dim == 128
     assert scale_dim == 1
     if _is_hip:
-        assert page_size == 1
+        assert (
+            page_size % 16 == 0
+        ), f"HIP preshuffle requires page_size to be a multiple of 16, got {page_size}"
     else:
         assert page_size == 64
 

python/sglang/srt/layers/attention/nsa/nsa_indexer.py

@@ -431,12 +431,13 @@ class Indexer(MultiPlatformOp):
         page_size = forward_batch.token_to_kv_pool.page_size
         # NOTE(dark): blocksize = 64 is hardcoded in deep_gemm
         if _is_hip:
-            assert page_size == 1, "only support page size 1"
+            assert (
-            block_tables = metadata.get_page_table_1()
+                page_size % 16 == 0
+            ), f"HIP preshuffle requires page_size to be a multiple of 16, got {page_size}"
         else:
             assert page_size == 64, "only support page size 64"
-            # NOTE(dark): this support extend/decode/decode+graph
+        # NOTE(dark): this support extend/decode/decode+graph
-            block_tables = metadata.get_page_table_64()
+        block_tables = metadata.get_page_table_64()
 
         max_seq_len = block_tables.shape[1] * page_size
         kv_cache_fp8 = forward_batch.token_to_kv_pool.get_index_k_with_scale_buffer(
@@ -471,17 +472,12 @@ class Indexer(MultiPlatformOp):
         assert len(q_fp8.shape) == 3
         q_fp8 = q_fp8.unsqueeze(1)  # the next_n dim is 1 now
         assert len(kv_cache_fp8.shape) == 2
-        block_kv = 1 if _is_hip else 64
+        block_kv = page_size
         num_heads_kv = 1
         head_dim_with_sf = 132
-        if _is_hip:
+        kv_cache_fp8 = kv_cache_fp8.view(
-            kv_cache_fp8 = kv_cache_fp8.view(
+            kv_cache_fp8.shape[0], block_kv, num_heads_kv, head_dim_with_sf
-                -1, block_kv, num_heads_kv, head_dim_with_sf
+        )
-            )
-        else:
-            kv_cache_fp8 = kv_cache_fp8.view(
-                kv_cache_fp8.shape[0], block_kv, num_heads_kv, head_dim_with_sf
-            )
         assert len(weights.shape) == 3
         weights = weights.squeeze(2)
 
@@ -492,9 +488,8 @@ class Indexer(MultiPlatformOp):
             from aiter.ops.triton.pa_mqa_logits import deepgemm_fp8_paged_mqa_logits
 
             batch_size, next_n, heads, _ = q_fp8.shape
-            logits = torch.full(
+            logits = torch.empty(
                 (batch_size * next_n, max_seq_len),
-                float("-inf"),
                 device=q_fp8.device,
                 dtype=torch.float32,
             )
@@ -506,7 +501,7 @@ class Indexer(MultiPlatformOp):
                 seqlens_32,
                 block_tables,
                 max_seq_len,
-                Preshuffle=False,
+                Preshuffle=_use_aiter,
                 KVBlockSize=block_kv,
             )
         else:
@@ -570,7 +565,9 @@ class Indexer(MultiPlatformOp):
 
         page_size = forward_batch.token_to_kv_pool.page_size
         if _is_hip:
-            assert page_size == 1, "only support page size 1"
+            assert (
+                page_size % 16 == 0
+            ), f"HIP preshuffle requires page_size to be a multiple of 16, got {page_size}"
         else:
             assert page_size == 64, "only support page size 64"
 
@@ -581,10 +578,7 @@ class Indexer(MultiPlatformOp):
         )
         weights = weights.squeeze(-1)
 
-        if _is_hip:
+        block_tables = metadata.get_page_table_64()
-            block_tables = metadata.get_page_table_1()
-        else:
-            block_tables = metadata.get_page_table_64()
 
         assert (
             forward_batch.seq_lens_cpu is not None
@@ -1040,19 +1034,24 @@ class Indexer(MultiPlatformOp):
             )
             return
 
-        # Fast path: AITER fused quant + cache store (HIP, page_size=1)
+        # Fast path: AITER fused quant + cache store (HIP, preshuffle)
         if _use_aiter:
+            page_size = forward_batch.token_to_kv_pool.page_size
             buf = forward_batch.token_to_kv_pool.get_index_k_with_scale_buffer(
                 layer_id=layer_id
             )
-            # Reshape from (num_pages, 132) uint8 to (num_pages, 1, 132) fp8
+            # Reshape from (num_pages, page_size*(128+4)) uint8 to (num_pages, page_size, 132) fp8
-            # to match kernel's (num_blocks, block_size, head_dim + scale_bytes) layout
+            kv_cache = buf.view(-1, page_size, 132).view(fp8_dtype)
-            kv_cache = buf.unsqueeze(1).view(fp8_dtype)
             out_loc = forward_batch.out_cache_loc
             if not out_loc.is_contiguous():
                 out_loc = out_loc.contiguous()
             indexer_k_quant_and_cache(
-                key, kv_cache, out_loc, self.block_size, self.scale_fmt
+                key,
+                kv_cache,
+                out_loc,
+                self.block_size,
+                self.scale_fmt,
+                preshuffle=True,
             )
             return
 

python/sglang/srt/mem_cache/memory_pool.py

@@ -2023,7 +2023,9 @@ class NSATokenToKVPool(MLATokenToKVPool):
         assert index_head_dim == 128
 
         if _is_hip:
-            assert self.page_size == 1
+            assert (
+                self.page_size % 16 == 0
+            ), f"HIP preshuffle requires page_size to be a multiple of 16, got {self.page_size}"
         else:
             assert self.page_size == 64
         with (

python/sglang/srt/server_args.py

@@ -1836,15 +1836,8 @@ class ServerArgs:
                                 f"attn_tp_size={self.tp_size}, attention weights will be sharded across {self.tp_size} ranks."
                             )
 
-                    if is_hip():
+                    self.page_size = 64
-                        self.page_size = 1
+                    logger.warning("Setting page size to 64 for DeepSeek DSA.")
-                        logger.warning(
-                            "Setting page size to 1 for DeepSeek DSA on ROCm."
-                        )
-                    else:
-                        # For CUDA GPU
-                        self.page_size = 64
-                        logger.warning("Setting page size to 64 for DeepSeek DSA.")
 
                     import torch