[MI35X] Fix meta mha 950 error to make guard return bool func itself (ROCm#1170)

* Make bool guard return func itself

* make all return func

* write querysol fake

Author: ZhangLirong-amd

aiter/jit/utils/torch_guard.py

@@ -106,10 +106,10 @@ def fake_impl(dummy_tensor, *args, **kwargs):
                 if gen_fake is not None:
                     return gen_fake(*args, **kwargs)
                 return func(*args, **kwargs)
-            default_values = {int: 0, bool: True, float: 0.0}
 
-            default_value = default_values.get(return_annotation, None)
-            return out, default_value
+            if gen_fake is not None:
+                return out, gen_fake(*args, **kwargs)
+            return out, func(*args, **kwargs)
 
         if is_torch_equal_or_newer("2.8.0"):
             tags = ()

aiter/tuned_gemm.py

@@ -96,7 +96,28 @@ def query_sol_core(
     return solution_idx
 
 
-@torch_compile_guard()
+def query_sol_fake(
+    m: int, n: int, k: int, bias: bool, dtype: str, otype: str, scaleAB: bool = False
+) -> int:
+    global solids, solMap, soltype
+    # soltype = None
+    solution_idx = 0
+    cu_count = get_cu_num()
+    if dtype in [dtypes.fp16, dtypes.bf16] and k % 8 == 0:
+        if (
+            ((m == 1 and n <= 2 * cu_count) or (m > 1 and m <= 4 and n <= cu_count))
+            and k <= 9216
+            or (m > 4 and m <= 8 and n <= cu_count)
+            and k <= 5120
+            or (m > 8 and m <= 16 and n <= cu_count)
+            and k <= 256
+        ):
+            soltype, solution_idx = 3, 2
+
+    return solution_idx
+
+
+@torch_compile_guard(gen_fake=query_sol_fake)
 def query_sol(
     m: int, n: int, k: int, bias: bool, dtype: str, otype: str, scaleAB: bool = False
 ) -> int: