Use xt macros to capture errors returned by nnlib calls (pytorch#15259)

hsharma35 · meta-codesync[bot] · commit 097ec6409f4f · 2025-10-20T20:00:41.000-07:00
Summary: Pull Request resolved: pytorch#15259 This diff implements the use of xt macros to capture errors returned by nnlib calls in the Cadence backend operators. Differential Revision: D83355891
diff --git a/backends/cadence/common/xt_macros.h b/backends/cadence/common/xt_macros.h
diff --git a/backends/cadence/fusion_g3/operators/op_add.cpp b/backends/cadence/fusion_g3/operators/op_add.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/kernels/portable/cpu/util/kernel_ops_util.h>
diff --git a/backends/cadence/fusion_g3/operators/op_cat.cpp b/backends/cadence/fusion_g3/operators/op_cat.cpp
@@ -13,7 +13,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/copy_ops_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_clamp.cpp b/backends/cadence/fusion_g3/operators/op_clamp.cpp
@@ -15,7 +15,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/kernels/portable/cpu/util/math_util.h>
diff --git a/backends/cadence/fusion_g3/operators/op_dequantize.cpp b/backends/cadence/fusion_g3/operators/op_dequantize.cpp
@@ -14,7 +14,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/reduce_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_div.cpp b/backends/cadence/fusion_g3/operators/op_div.cpp
@@ -12,7 +12,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/kernels/portable/cpu/util/math_util.h>
diff --git a/backends/cadence/fusion_g3/operators/op_exp.cpp b/backends/cadence/fusion_g3/operators/op_exp.cpp
@@ -12,7 +12,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/pattern/pattern.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_hardtanh.cpp b/backends/cadence/fusion_g3/operators/op_hardtanh.cpp
@@ -11,7 +11,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/functional_util.h>
 #include <executorch/kernels/portable/cpu/util/math_util.h>
diff --git a/backends/cadence/fusion_g3/operators/op_lt.cpp b/backends/cadence/fusion_g3/operators/op_lt.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/pattern/comparison_op.h>
 
 using ::executorch::aten::Scalar;
diff --git a/backends/cadence/fusion_g3/operators/op_mean.cpp b/backends/cadence/fusion_g3/operators/op_mean.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/kernel_ops_util.h>
 #include <executorch/kernels/portable/cpu/util/reduce_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_mul.cpp b/backends/cadence/fusion_g3/operators/op_mul.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp b/backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp
@@ -13,7 +13,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/normalization_ops_util.h>
 #include <executorch/kernels/portable/cpu/vec_ops.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_permute_copy.cpp b/backends/cadence/fusion_g3/operators/op_permute_copy.cpp
@@ -11,7 +11,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/copy_ops_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_quantize.cpp b/backends/cadence/fusion_g3/operators/op_quantize.cpp
@@ -14,7 +14,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/reduce_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_rsqrt.cpp b/backends/cadence/fusion_g3/operators/op_rsqrt.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/pattern/pattern.h>
 #include <executorch/kernels/portable/cpu/util/functional_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_sigmoid.cpp b/backends/cadence/fusion_g3/operators/op_sigmoid.cpp
@@ -12,7 +12,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/kernels/portable/cpu/util/functional_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_slice_copy.cpp b/backends/cadence/fusion_g3/operators/op_slice_copy.cpp
@@ -13,7 +13,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/slice_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_softmax.cpp b/backends/cadence/fusion_g3/operators/op_softmax.cpp
@@ -12,7 +12,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/activation_ops_util.h>
 #include <executorch/kernels/portable/cpu/util/functional_util.h>
 #include <executorch/kernels/portable/cpu/util/reduce_util.h>
diff --git a/backends/cadence/fusion_g3/operators/op_sqrt.cpp b/backends/cadence/fusion_g3/operators/op_sqrt.cpp
@@ -12,7 +12,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/pattern/pattern.h>
 #include <executorch/kernels/portable/cpu/util/functional_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_sub.cpp b/backends/cadence/fusion_g3/operators/op_sub.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_tanh.cpp b/backends/cadence/fusion_g3/operators/op_tanh.cpp
@@ -12,7 +12,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/pattern/pattern.h>
 #include <executorch/kernels/portable/cpu/util/functional_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
diff --git a/backends/cadence/fusion_g3/operators/op_transpose_copy.cpp b/backends/cadence/fusion_g3/operators/op_transpose_copy.cpp
@@ -11,7 +11,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/transpose_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/op_where.cpp b/backends/cadence/fusion_g3/operators/op_where.cpp
@@ -10,7 +10,7 @@
 
 #include <xa_nnlib_kernels_api.h>
 
-#include <executorch/backends/cadence/fusion_g3/operators/xt_macros.h>
+#include <executorch/backends/cadence/common/xt_macros.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
diff --git a/backends/cadence/fusion_g3/operators/targets.bzl b/backends/cadence/fusion_g3/operators/targets.bzl
@@ -10,8 +10,11 @@ def define_operator(name: str, deps: list[str] | None = None) -> None:
         "//executorch/kernels/portable/cpu/pattern:all_deps",
         "//executorch/runtime/kernel:kernel_includes",
         "//executorch/kernels/portable/cpu:scalar_utils",
+        "//executorch/backends/cadence/common:xt_macros",
         "fbsource//third-party/nnlib-FusionG3/xa_nnlib:libxa_nnlib_common",
         "fbsource//third-party/nnlib-FusionG3/xa_nnlib:libxa_nnlib",
+        ":operators_header",
+        ":xt_utils",
     ]
     if deps == None:
         deps = []
@@ -26,11 +29,6 @@ def define_operator(name: str, deps: list[str] | None = None) -> None:
         ],
         compatible_with = ["ovr_config//cpu:xtensa"],
         deps = deps + common_deps,
-        exported_deps = [
-            ":operators_header",
-            ":xt_macros",
-            ":xt_utils",
-        ],
     )
 
 OPERATORS = [
diff --git a/backends/cadence/hifi/operators/op_add.cpp b/backends/cadence/hifi/operators/op_add.cpp
@@ -16,6 +16,8 @@
 #include <executorch/runtime/kernel/kernel_includes.h>
 #include <executorch/runtime/platform/assert.h>
 
+#include <executorch/backends/cadence/common/xt_macros.h>
+
 using executorch::aten::Scalar;
 using executorch::aten::ScalarType;
 using executorch::aten::Tensor;
@@ -184,10 +186,25 @@ Tensor& add_out(
       for (int i = 0; i < b.dim(); i++)
         inp2_shape[i + off_b] = b.size(i);
 
-      xa_nn_elm_add_broadcast_4D_f32xf32_f32(
-          out_data, out_shape, a_data, inp1_shape, b_data, inp2_shape);
+      XT_KERNEL_CHECK(
+          ctx,
+          out,
+          xa_nn_elm_add_broadcast_4D_f32xf32_f32,
+          out_data,
+          out_shape,
+          a_data,
+          inp1_shape,
+          b_data,
+          inp2_shape);
     } else {
-      xa_nn_elm_add_f32xf32_f32(out_data, a_data, b_data, out.numel());
+      XT_KERNEL_CHECK(
+          ctx,
+          out,
+          xa_nn_elm_add_f32xf32_f32,
+          out_data,
+          a_data,
+          b_data,
+          out.numel());
     }
 
     return out;
diff --git a/backends/cadence/hifi/operators/op_dequantize_per_tensor.cpp b/backends/cadence/hifi/operators/op_dequantize_per_tensor.cpp
@@ -10,6 +10,8 @@
 #include <executorch/runtime/kernel/kernel_includes.h>
 #include <xa_nnlib_kernels_api.h>
 
+#include <executorch/backends/cadence/common/xt_macros.h>
+
 namespace impl {
 namespace HiFi {
 namespace native {
diff --git a/backends/cadence/hifi/operators/targets.bzl b/backends/cadence/hifi/operators/targets.bzl
@@ -16,7 +16,8 @@ def define_operator(name: str, deps: list[str] | None = None) -> None:
         "//executorch/kernels/portable/cpu/util:elementwise_util",
         "//executorch/kernels/portable/cpu/pattern:bitwise_op",
         "//executorch/backends/cadence/hifi/third-party/nnlib:nnlib-extensions",
-        "//executorch/kernels/portable/cpu/pattern:comparison_op"
+        "//executorch/kernels/portable/cpu/pattern:comparison_op",
+        "//executorch/backends/cadence/common:xt_macros"
     ]
     if deps == None:
         deps = []
