Optimization for AscendPagedTokenToKVPoolAllocator (#8293)

Co-authored-by: ronnie_zheng <zl19940307@163.com>
Co-authored-by: VDV1985 <vladdv85@mail.ru>

python/sglang/srt/mem_cache/allocator.py

@@ -632,27 +632,6 @@ def alloc_extend_kernel_ascend(
             out_indices[end_pos[i] - num3 : end_pos[i]] = (
                 free_pages[end_new_pages[i] - 1] * page_size + pos_in_page[:num3]
             ).view(-1)
-    return num_new_pages
-
-
-def alloc_decode_kernel_ascend(
-    seq_lens,
-    last_loc,
-    free_pages,
-    out_indices,
-    page_size,
-):
-    num_new_pages = (seq_lens + page_size - 1) // page_size - (
-        seq_lens - 1 + page_size - 1
-    ) // page_size
-    end_new_pages = torch.cumsum(num_new_pages, 0)
-    start_new_pages = end_new_pages - num_new_pages
-    for i in range(len(seq_lens)):
-        if num_new_pages[i]:
-            out_indices[i] = free_pages[start_new_pages[i]] * page_size
-        else:
-            out_indices[i] = last_loc[i] + 1
-    return num_new_pages
 
 
 class AscendPagedTokenToKVPoolAllocator(PagedTokenToKVPoolAllocator):
@@ -667,7 +646,6 @@ class AscendPagedTokenToKVPoolAllocator(PagedTokenToKVPoolAllocator):
         need_sort: bool,
     ):
         super().__init__(size, page_size, dtype, device, kvcache, need_sort)
-        self.ret_values = torch.empty((), dtype=torch.int32, device=self.device)
 
     def alloc_extend(
         self,
@@ -681,17 +659,25 @@ class AscendPagedTokenToKVPoolAllocator(PagedTokenToKVPoolAllocator):
                 (last_loc + 1) % self.page_size == prefix_lens % self.page_size
             )
 
-        bs = len(prefix_lens)
+        estimated_num_new_pages = (
-        if self.need_sort and self.estimated_num_new_pages(bs, extend_num_tokens) > len(
+            (
-            self.free_pages
+                (seq_lens + self.page_size - 1) // self.page_size
-        ):
+                - (prefix_lens + self.page_size - 1) // self.page_size
+            )
+            .sum()
+            .item()
+        )
+        if self.need_sort and estimated_num_new_pages > len(self.free_pages):
             self.merge_and_sort_free()
 
+        if estimated_num_new_pages > len(self.free_pages):
+            return None
+
         out_indices = torch.empty(
             (extend_num_tokens,), dtype=torch.int32, device=self.device
         )
 
-        self.ret_values = alloc_extend_kernel_ascend(
+        alloc_extend_kernel_ascend(
             prefix_lens,
             seq_lens,
             last_loc,
@@ -704,11 +690,7 @@ class AscendPagedTokenToKVPoolAllocator(PagedTokenToKVPoolAllocator):
         if self.debug_mode:
             assert len(torch.unique(out_indices)) == len(out_indices)
 
-        num_new_pages = self.ret_values.sum()
+        self.free_pages = self.free_pages[estimated_num_new_pages:]
-        if num_new_pages > len(self.free_pages):
-            return None
-
-        self.free_pages = self.free_pages[num_new_pages:]
         return out_indices
 
     def alloc_decode(
@@ -721,33 +703,26 @@ class AscendPagedTokenToKVPoolAllocator(PagedTokenToKVPoolAllocator):
                 (last_loc + 2) % self.page_size == seq_lens % self.page_size
             )
 
-        bs = len(seq_lens)
+        need_new_pages = (seq_lens % self.page_size == 1).int()
-        if self.need_sort and self.estimated_num_new_pages(bs, 1) > len(
+        num_new_pages = need_new_pages.sum().item()
-            self.free_pages
+
-        ):
+        if num_new_pages > len(self.free_pages):
             self.merge_and_sort_free()
 
-        out_indices = torch.empty((bs,), dtype=torch.int32, device=self.device)
+        if num_new_pages > len(self.free_pages):
+            return None
 
-        self.ret_values = alloc_decode_kernel_ascend(
+        end_new_pages = torch.cumsum(need_new_pages, 0)
-            seq_lens,
+        start_new_pages = end_new_pages - need_new_pages
-            last_loc,
+        if num_new_pages == 0:
-            self.free_pages,
+            out_indices = last_loc + 1
-            out_indices,
+        else:
-            self.page_size,
+            out_indices = (last_loc + 1) * (1 - need_new_pages) + self.free_pages[
-        )
+                start_new_pages
+            ] * self.page_size * need_new_pages
 
         if self.debug_mode:
             assert len(torch.unique(out_indices)) == len(out_indices)
 
-        num_new_pages = self.ret_values.sum()
-        if num_new_pages > len(self.free_pages):
-            return None
-
         self.free_pages = self.free_pages[num_new_pages:]
-        return out_indices
+        return out_indices.int()
-
-    def clear(self):
-        super().clear()
-        self.free_pages = self.free_pages.to(torch.int32)
-        self.release_pages = self.release_pages.to(torch.int32)