add size and rank of nightly facto (#7962)

Summary:

x1.3 scale number of testcases (10-40 cases per op). The hardcoded numbers (size, rank, etc) are defined by statistics from test_aten_ops.py. For most ops the inputs have 1~4 dims, 1~512 of each size, -16.0~16.0 of each value. Dtype is any of float and int expecte 64 bits.

Reviewed By: hsharma35

Differential Revision: D68690963

Author: zonglinpeng

backends/cadence/fusion_g3/operators/op_exp.cpp

@@ -49,9 +49,9 @@ Tensor& exp_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
       out);
 #endif
 
-  if (out.scalar_type() == ScalarType::Float) {
-    float* const out_data = out.mutable_data_ptr<float>();
-    const float* const in_data = in.const_data_ptr<float>();
+  if (in.scalar_type() == ScalarType::Float) {
+    float* __restrict__ out_data = out.mutable_data_ptr<float>();
+    const float* __restrict__ in_data = in.const_data_ptr<float>();
 
     XT_KERNEL_CHECK(
         ctx, out, xa_nn_elm_exp_f32_f32, out_data, in_data, out.numel());
@@ -66,4 +66,4 @@ Tensor& exp_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
 } // namespace native
 } // namespace G3
 } // namespace impl
-} // namespace cadence
+} // namespace cadence

examples/cadence/operators/facto_util.py

@@ -18,58 +18,36 @@
 
 def apply_tensor_contraints(op_name: str, tensor_constraints: list[object]) -> None:
     match op_name:
-        case (
-            "sigmoid.default"
-            | "_softmax.default"
-            | "rsqrt.default"
-            | "exp.default"
-            | "mul.Tensor"
-            | "div.Tensor"
-        ):
+        case "sigmoid.default" | "rsqrt.default":
             tensor_constraints.extend(
                 [
                     cp.Dtype.In(lambda deps: [torch.float]),
-                    cp.Size.Le(lambda deps, r, d: 2),
-                    cp.Rank.Le(lambda deps: 2),
+                    cp.Rank.Le(lambda deps: 2**3),
                 ]
             )
-        case (
-            "add.Tensor"
-            | "sub.Tensor"
-            | "add.Scalar"
-            | "sub.Scalar"
-            | "mul.Scalar"
-            | "div.Scalar"
-        ):
+        case "exp.default":
             tensor_constraints.extend(
                 [
-                    cp.Dtype.In(lambda deps: [torch.float, torch.int32]),
-                    cp.Size.Le(lambda deps, r, d: 2),
-                    cp.Rank.Le(lambda deps: 2),
-                ]
-            )
-        case "native_layer_norm.default":
-            tensor_constraints.extend(
-                [
-                    cp.Dtype.In(lambda deps: [torch.float, torch.int32]),
-                    cp.Size.Le(lambda deps, r, d: 2**4),
-                    cp.Rank.Le(lambda deps: 2**4),
+                    cp.Rank.Le(lambda deps: 2**3),
+                    cp.Value.Ge(lambda deps, dtype, struct: -2**2),
+                    cp.Value.Le(lambda deps, dtype, struct: 2**2),
                 ]
             )
         case _:
             tensor_constraints.extend(
                 [
-                    cp.Dtype.In(lambda deps: [torch.float, torch.int32]),
-                    cp.Size.Le(lambda deps, r, d: 2),
-                    cp.Rank.Le(lambda deps: 2),
+                    cp.Rank.Le(lambda deps: 2**2),
                 ]
             )
     tensor_constraints.extend(
         [
-            cp.Value.Ge(lambda deps, dtype, struct: -(2**8)),
-            cp.Value.Le(lambda deps, dtype, struct: 2**8),
+            cp.Dtype.In(lambda deps: [torch.int, torch.float]),
+            cp.Dtype.NotIn(lambda deps: [torch.int64, torch.float64]),
+            cp.Value.Ge(lambda deps, dtype, struct: -(2**4)),
+            cp.Value.Le(lambda deps, dtype, struct: 2**4),
             cp.Rank.Ge(lambda deps: 1),
             cp.Size.Ge(lambda deps, r, d: 1),
+            cp.Size.Le(lambda deps, r, d: 2**9),
         ]
     )