Update base for Update on "make etrecord support export program"

Differential Revision: [D77965102](https://our.internmc.facebook.com/intern/diff/D77965102/)

[ghstack-poisoned]

Author: Gasoonjia

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-9b498d3bb28b8e3411ce464dd2755c5b96d92c8f
+7cda4017ddda554752e89069ae205be5e8388f59

.ci/scripts/check_c10_sync.sh

@@ -12,4 +12,4 @@ pushd pytorch
 git checkout "$pytorch_pin"
 popd
 "$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/c10 pytorch/c10
-"$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/torch/standalone pytorch/torch/standalone
+"$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/torch/headeronly pytorch/torch/headeronly

backends/arm/_passes/TARGETS

@@ -6,6 +6,7 @@ python_library(
     deps = [
         "//executorch/backends/arm:tosa_quant_utils",
         "//executorch/backends/arm:tosa_utils",
+        "//executorch/backends/arm/tosa/dialect:lib",
         "//executorch/backends/transforms:fuse_view_copy",
         "//executorch/backends/transforms:remove_getitem_op",
         "//executorch/backends/transforms:replace_scalar_with_tensor",

backends/arm/tosa/dialect/TARGETS

@@ -0,0 +1,6 @@
+load("@fbcode_macros//build_defs:python_library.bzl", "python_library")
+
+python_library(
+    name = "lib",
+    srcs = glob(["*.py"]),
+)

backends/cadence/aot/compiler.py

@@ -66,9 +66,6 @@ def trace(
     # Make the model inference mode by calling model.eval()
     model.eval()
 
-    # Prevent mkldnn decompositions
-    torch._C._set_mkldnn_enabled(False)
-
     # Get default decompositions
     decomp_table = torch.export.default_decompositions()
 

backends/cadence/aot/functions_hifi.yaml

@@ -75,7 +75,7 @@
 - op: clamp.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::clamp_tensor_out
+      kernel_name: cadence::impl::HiFi::clamp_Tensor_out
 
 - op: clone.out
   kernels:
@@ -100,7 +100,7 @@
 - op: eq.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::eq_tensor_out
+      kernel_name: cadence::impl::HiFi::eq_Tensor_out
 
 - op: fmod.Tensor_out
   kernels:
@@ -120,12 +120,12 @@
 - op: ge.Scalar_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::ge_scalar_out
+      kernel_name: cadence::impl::HiFi::ge_Scalar_out
 
 - op: ge.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::ge_tensor_out
+      kernel_name: cadence::impl::HiFi::ge_Tensor_out
 
 - op: gelu.out
   kernels:
@@ -135,12 +135,12 @@
 - op: gt.Scalar_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::gt_scalar_out
+      kernel_name: cadence::impl::HiFi::gt_Scalar_out
 
 - op: gt.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::gt_tensor_out
+      kernel_name: cadence::impl::HiFi::gt_Tensor_out
 
 - op: hardtanh.out
   kernels:
@@ -150,27 +150,27 @@
 - op: le.Scalar_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::le_scalar_out
+      kernel_name: cadence::impl::HiFi::le_Scalar_out
 
 - op: le.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::le_tensor_out
+      kernel_name: cadence::impl::HiFi::le_Tensor_out
 
 - op: lt.Scalar_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::lt_scalar_out
+      kernel_name: cadence::impl::HiFi::lt_Scalar_out
 
 - op: lt.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::lt_tensor_out
+      kernel_name: cadence::impl::HiFi::lt_Tensor_out
 
 - op: masked_fill.Scalar_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::masked_fill_scalar_out
+      kernel_name: cadence::impl::HiFi::masked_fill_Scalar_out
 
 - op: max_pool2d_with_indices.out
   kernels:
@@ -185,7 +185,7 @@
 - op: mean.out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::mean_out   
+      kernel_name: cadence::impl::HiFi::mean_out
 
 - op: minimum.out
   kernels:
@@ -205,7 +205,7 @@
 - op: ne.Tensor_out
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::ne_tensor_out
+      kernel_name: cadence::impl::HiFi::ne_Tensor_out
 
 - op: permute_copy.out
   kernels:
@@ -289,11 +289,11 @@
   kernels:
     - arg_meta: null
       kernel_name: cadence::impl::HiFi::dequantize_per_tensor_out
-      
+
 - func: cadence::quantized_conv.out(Tensor input, Tensor weight, Tensor bias, int[] stride, SymInt[] padding, int[] dilation, int groups, int input_zero_point, Tensor weight_zero_point, Tensor bias_scale, float out_scale, int out_zero_point, Tensor out_multiplier, Tensor out_shift, bool channel_last=False, *, Tensor(a!) out) -> Tensor(a!)
   kernels:
     - arg_meta: null
-      kernel_name: cadence::impl::HiFi::quantized_conv_out      
+      kernel_name: cadence::impl::HiFi::quantized_conv_out
 
 - func: cadence::quantized_layer_norm.out(Tensor input, Tensor in_scale, Tensor in_zero_point, int[] normalized_shape, Tensor weight, Tensor bias, float eps, float output_scale, int output_zero_point, *, Tensor(a!) out) -> Tensor(a!)
   kernels:

backends/cadence/aot/replace_ops.py

@@ -2329,7 +2329,7 @@ def call(self, graph_module: torch.fx.GraphModule) -> PassResult:
             # Extract an argument to a separate full op.
             with graph_module.graph.inserting_before(mul_node):
                 full_tensor = graph_module.graph.call_function(
-                    exir_ops.edge.aten.full.default, args=([1], full_arg)
+                    torch.ops.aten.full.default, args=([1], full_arg)
                 )
                 new_mul_node = graph_module.graph.call_function(
                     torch.ops.aten.mul.Tensor, args=(x_arg, full_tensor)
@@ -2449,4 +2449,5 @@ class CadenceReplaceOpsInGraph:
         ReplaceAtenApproxGeluWithApproxGeluPass,
         ReplaceSplitWithSlicePass,
         ReplacePowWithMulPass,
+        ReplaceMulTensorWithMulAndFullOpsPass,
     ]

backends/cadence/aot/tests/test_replace_ops_passes.py

@@ -1933,7 +1933,7 @@ def test_extract_mul_argument_to_full(
                 graph_after_passes,
                 expected_op_counts={
                     torch.ops.aten.mul.Tensor: 1,
-                    exir_ops.edge.aten.full.default: 1,
+                    torch.ops.aten.full.default: 1,
                 },
             )
         )

backends/cadence/hifi/operators/CMakeLists.txt

@@ -88,7 +88,7 @@ target_include_directories(
 
 # Custom ops that are needed to run the test model.
 add_library(
-  custom_ops "op_quantized_linear_out.cpp" "op_quantized_layer_norm.cpp" "quantized_matmul_out.cpp"
+  custom_ops "op_quantized_linear_out.cpp" "op_quantized_layer_norm.cpp" "op_quantized_matmul_out.cpp"
              "op_quantize_per_tensor.cpp" "op_quantized_relu_out.cpp" "op_dequantize_per_tensor.cpp"
              "op_quantized_conv_out.cpp" "op_quantized_fully_connected_out"
 )

backends/cadence/hifi/operators/op_bitwise_and.cpp

@@ -169,8 +169,8 @@ Tensor& bitwise_and_Tensor_out(
     return out;
   }
 
-  return torch::executor::native::internal::bitwise_tensor_out<op_name>(
-      ctx, a, b, out);
+  return torch::executor::native::internal::
+      bitwise_tensor_out<std::bit_and, op_name>(ctx, a, b, out);
 }
 
 Tensor& bitwise_and_Scalar_out(
@@ -180,8 +180,8 @@ Tensor& bitwise_and_Scalar_out(
     Tensor& out) {
   // @lint-ignore CLANGTIDY facebook-hte-CArray
   static constexpr const char op_name[] = "bitwise_and.Scalar_out";
-  return torch::executor::native::internal::bitwise_scalar_out<op_name>(
-      ctx, a, b, out);
+  return torch::executor::native::internal::
+      bitwise_scalar_out<std::bit_and, op_name>(ctx, a, b, out);
 }
 
 } // namespace native