[KUNLUNXIN] speed up rsqrt op (#1266)

Author: dongjibin1996

src/flag_gems/runtime/backend/_kunlunxin/ops/rsqrt.py

@@ -2,6 +2,7 @@
 
 import triton
 import triton.language as tl
+import triton.language.extra.xpu.libdevice as xpu
 
 from ..utils.pointwise_dynamic import pointwise_dynamic
 
@@ -11,7 +12,7 @@
 @pointwise_dynamic(promotion_methods=[(0, "INT_TO_FLOAT")])
 @triton.jit
 def rsqrt_func(x):
-    return 1.0 / tl.sqrt(x.to(tl.float32))
+    return xpu.rsqrt(x.to(tl.float32))
 
 
 def rsqrt(A):