Migrate backends/cadence away from deprecated namespaces (#5905)

Summary:
Pull Request resolved: #5905

Stop using the `torch::` namespace where possible.

For now, ops still live under `torch::executor::`.

Reviewed By: Gasoonjia, zonglinpeng

Differential Revision: D63924099

fbshipit-source-id: e1132f889bfdeccb56e55a5bad6be937cce366e3

Author: dbort

backends/cadence/executor_runner.cpp

@@ -37,7 +37,6 @@
 
 static uint8_t method_allocator_pool[18 * 1024U]; // 4 MB
 
-using namespace torch::executor;
 #include <xtensa/config/core.h>
 
 #define APP_MU MUB
@@ -48,8 +47,8 @@ using namespace torch::executor;
 /* How many message is used to test message sending */
 #define MSG_LENGTH 32U
 
-using torch::executor::Error;
-using torch::executor::Result;
+using executorch::runtime::Error;
+using executorch::runtime::Result;
 
 void LED_INIT();
 void LED_TOGGLE();
@@ -106,13 +105,13 @@ int main(int argc, char** argv) {
   BOARD_InitDebugConsole();
   ET_LOG(Info, "Booted up in DSP.");
 
-  torch::executor::runtime_init();
+  executorch::runtime::runtime_init();
 
   auto loader =
-      torch::executor::util::BufferDataLoader(model_pte, sizeof(model_pte));
+      executorch::extension::BufferDataLoader(model_pte, sizeof(model_pte));
 
-  Result<torch::executor::Program> program =
-      torch::executor::Program::load(&loader);
+  Result<executorch::runtime::Program> program =
+      executorch::runtime::Program::load(&loader);
   if (!program.ok()) {
     ET_LOG(
         Error,
@@ -132,7 +131,7 @@ int main(int argc, char** argv) {
   }
   ET_LOG(Info, "ET: Running method %s", method_name);
 
-  Result<torch::executor::MethodMeta> method_meta =
+  Result<executorch::runtime::MethodMeta> method_meta =
       program->method_meta(method_name);
   if (!method_meta.ok()) {
     ET_LOG(
@@ -142,12 +141,12 @@ int main(int argc, char** argv) {
         (unsigned int)method_meta.error());
   }
 
-  torch::executor::MemoryAllocator method_allocator{
-      torch::executor::MemoryAllocator(
+  executorch::runtime::MemoryAllocator method_allocator{
+      executorch::runtime::MemoryAllocator(
           sizeof(method_allocator_pool), method_allocator_pool)};
 
   std::vector<std::unique_ptr<uint8_t[]>> planned_buffers; // Owns the memory
-  std::vector<torch::executor::Span<uint8_t>>
+  std::vector<executorch::runtime::Span<uint8_t>>
       planned_spans; // Passed to the allocator
   size_t num_memory_planned_buffers = method_meta->num_memory_planned_buffers();
 
@@ -161,13 +160,13 @@ int main(int argc, char** argv) {
     planned_spans.push_back({planned_buffers.back().get(), buffer_size});
   }
 
-  torch::executor::HierarchicalAllocator planned_memory(
+  executorch::runtime::HierarchicalAllocator planned_memory(
       {planned_spans.data(), planned_spans.size()});
 
-  torch::executor::MemoryManager memory_manager(
+  executorch::runtime::MemoryManager memory_manager(
       &method_allocator, &planned_memory);
 
-  Result<torch::executor::Method> method =
+  Result<executorch::runtime::Method> method =
       program->load_method(method_name, &memory_manager);
   if (!method.ok()) {
     ET_LOG(
@@ -178,7 +177,7 @@ int main(int argc, char** argv) {
   }
 
   ET_LOG(Info, "Method loaded.");
-  torch::executor::util::prepare_input_tensors(*method);
+  executorch::extension::prepare_input_tensors(*method);
   ET_LOG(Info, "Starting the model execution...");
 
   Error status = method->execute();

backends/cadence/hifi/operators/dequantize_per_tensor.cpp

@@ -14,9 +14,9 @@ namespace impl {
 namespace HiFi {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::ScalarType;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
-using ScalarType = exec_aten::ScalarType;
 
 void dequantize_per_tensor_out(
     KernelRuntimeContext& context,

backends/cadence/hifi/operators/quantize_per_tensor.cpp

@@ -14,9 +14,9 @@ namespace impl {
 namespace HiFi {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::ScalarType;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
-using ScalarType = exec_aten::ScalarType;
 
 // Quantize the input tensor (PT2 version). Note that quant_<min,max> are not
 // used in any computation.

backends/cadence/hifi/operators/quantized_layer_norm.cpp

@@ -12,7 +12,7 @@
 #include <cmath>
 #include <tuple>
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
 
 namespace impl {
@@ -119,14 +119,14 @@ void quantized_layer_norm_out(
     const Tensor& input,
     const Tensor& in_scale,
     const Tensor& in_zero_point,
-    const exec_aten::IntArrayRef normalized_shape,
+    const executorch::aten::IntArrayRef normalized_shape,
     const Tensor& weight,
     const Tensor& bias,
     double eps,
     double output_scale,
     int64_t output_zero_point,
     Tensor& out) {
-  if (input.scalar_type() == exec_aten::ScalarType::Byte) {
+  if (input.scalar_type() == executorch::aten::ScalarType::Byte) {
     quantized_layer_norm_<uint8_t>(
         input,
         in_scale,
@@ -137,7 +137,7 @@ void quantized_layer_norm_out(
         output_scale,
         output_zero_point,
         out);
-  } else if (input.scalar_type() == exec_aten::ScalarType::Char) {
+  } else if (input.scalar_type() == executorch::aten::ScalarType::Char) {
     quantized_layer_norm_<int8_t>(
         input,
         in_scale,

backends/cadence/hifi/operators/quantized_linear_out.cpp

@@ -15,7 +15,7 @@ namespace impl {
 namespace HiFi {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
 
 void quantized_linear_out(
@@ -28,7 +28,7 @@ void quantized_linear_out(
     const Tensor& out_multiplier,
     const Tensor& out_shift,
     int64_t out_zero_point,
-    const exec_aten::optional<Tensor>& offset,
+    const executorch::aten::optional<Tensor>& offset,
     Tensor& out) {
   // input comes in shape [leading_dims, in_dim]
   // weight comes in shape [out_dim, in_dim]

backends/cadence/reference/operators/dequantize_per_tensor.cpp

@@ -13,9 +13,9 @@ namespace impl {
 namespace reference {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::ScalarType;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
-using ScalarType = exec_aten::ScalarType;
 
 void dequantize_per_tensor_out(
     KernelRuntimeContext& context,

backends/cadence/reference/operators/op_embedding.cpp

@@ -12,7 +12,7 @@ namespace torch {
 namespace executor {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
 
 void embedding_out(

backends/cadence/reference/operators/op_full.cpp

@@ -13,8 +13,8 @@ namespace torch {
 namespace executor {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
-using ScalarType = exec_aten::ScalarType;
+using executorch::aten::ScalarType;
+using executorch::aten::Tensor;
 
 Tensor& full_out(
     KernelRuntimeContext& ctx,

backends/cadence/reference/operators/op_view_copy.cpp

@@ -12,7 +12,7 @@ namespace torch {
 namespace executor {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
 
 Tensor& view_copy_out(

backends/cadence/reference/operators/quantize_per_tensor.cpp

@@ -13,9 +13,9 @@ namespace impl {
 namespace reference {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::ScalarType;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
-using ScalarType = exec_aten::ScalarType;
 
 // Quantize the input tensor (PT2 version). Note that quant_<min,max> are not
 // used in any computation.

backends/cadence/reference/operators/quantized_conv_out.cpp

@@ -14,7 +14,7 @@ namespace impl {
 namespace reference {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
 
 // This implements a generic 2d conv kernel that operates on raw pointers.
@@ -158,9 +158,9 @@ void quantized_conv_out(
     const Tensor& input,
     const Tensor& weight,
     const Tensor& bias,
-    exec_aten::IntArrayRef stride,
-    exec_aten::IntArrayRef padding,
-    exec_aten::IntArrayRef dilation,
+    executorch::aten::IntArrayRef stride,
+    executorch::aten::IntArrayRef padding,
+    executorch::aten::IntArrayRef dilation,
     int64_t groups,
     int64_t in_zero_point,
     const Tensor& weight_zero_point,

backends/cadence/reference/operators/quantized_layer_norm.cpp

@@ -115,14 +115,14 @@ void quantized_layer_norm_out(
     const Tensor& input,
     const Tensor& in_scale,
     const Tensor& in_zero_point,
-    const exec_aten::IntArrayRef normalized_shape,
+    const executorch::aten::IntArrayRef normalized_shape,
     const Tensor& weight,
     const Tensor& bias,
     double eps,
     double output_scale,
     int64_t output_zero_point,
     Tensor& out) {
-  if (input.scalar_type() == exec_aten::ScalarType::Byte) {
+  if (input.scalar_type() == executorch::aten::ScalarType::Byte) {
     quantized_layer_norm_<uint8_t>(
         input,
         in_scale,
@@ -133,7 +133,7 @@ void quantized_layer_norm_out(
         output_scale,
         output_zero_point,
         out);
-  } else if (input.scalar_type() == exec_aten::ScalarType::Char) {
+  } else if (input.scalar_type() == executorch::aten::ScalarType::Char) {
     quantized_layer_norm_<int8_t>(
         input,
         in_scale,

backends/cadence/reference/operators/quantized_linear_out.cpp

@@ -27,7 +27,7 @@ void quantized_linear_out(
     const Tensor& out_multiplier,
     const Tensor& out_shift,
     int64_t out_zero_point,
-    const exec_aten::optional<Tensor>& offset,
+    const executorch::aten::optional<Tensor>& offset,
     Tensor& out) {
   // Assuming uint8_t for now, but needs to be updated for other quantization
   // types

backends/cadence/reference/operators/quantized_matmul_out.cpp

@@ -60,7 +60,7 @@ void inline _typed_quantized_matmul(
     int64_t X_zero_point,
     const Tensor& Y,
     int64_t Y_zero_point,
-    const exec_aten::optional<Tensor>& bias,
+    const executorch::aten::optional<Tensor>& bias,
     int64_t out_multiplier,
     int64_t out_shift,
     int64_t out_zero_point,
@@ -114,13 +114,13 @@ void quantized_matmul_out(
     int64_t X_zero_point,
     const Tensor& Y,
     int64_t Y_zero_point,
-    const exec_aten::optional<Tensor>& bias,
+    const executorch::aten::optional<Tensor>& bias,
     int64_t out_multiplier,
     int64_t out_shift,
     int64_t out_zero_point,
     bool transposed,
     Tensor& out) {
-  if (out.scalar_type() == exec_aten::ScalarType::Byte) {
+  if (out.scalar_type() == executorch::aten::ScalarType::Byte) {
     _typed_quantized_matmul<uint8_t>(
         X,
         X_zero_point,
@@ -132,7 +132,7 @@ void quantized_matmul_out(
         out_zero_point,
         transposed,
         out);
-  } else if (out.scalar_type() == exec_aten::ScalarType::Char) {
+  } else if (out.scalar_type() == executorch::aten::ScalarType::Char) {
     _typed_quantized_matmul<int8_t>(
         X,
         X_zero_point,

backends/cadence/reference/operators/quantized_relu_out.cpp

@@ -13,7 +13,7 @@ namespace impl {
 namespace reference {
 namespace native {
 
-using Tensor = exec_aten::Tensor;
+using executorch::aten::Tensor;
 using executorch::runtime::KernelRuntimeContext;
 
 template <typename T>
@@ -51,15 +51,15 @@ void quantized_relu_out(
     const Tensor& out_multiplier,
     const Tensor& out_shift,
     Tensor& output) {
-  if (input.scalar_type() == exec_aten::ScalarType::Byte) {
+  if (input.scalar_type() == executorch::aten::ScalarType::Byte) {
     quantized_relu_<uint8_t>(
         input,
         in_zero_point,
         out_zero_point,
         out_multiplier,
         out_shift,
         output);
-  } else if (input.scalar_type() == exec_aten::ScalarType::Char) {
+  } else if (input.scalar_type() == executorch::aten::ScalarType::Char) {
     quantized_relu_<int8_t>(
         input,
         in_zero_point,