[cadence][hifi] Fixed all hifi ops and cmake

Differential Revision: D69812342

Pull Request resolved: pytorch#8551

Author: zonglinpeng

backends/cadence/CMakeLists.txt

@@ -22,7 +22,10 @@ endif()
 include(${EXECUTORCH_ROOT}/build/Utils.cmake)
 
 # Let files say "include <executorch/path/to/header.h>".
-set(_common_include_directories ${EXECUTORCH_ROOT}/..)
+set(_common_include_directories ${EXECUTORCH_ROOT}/..
+  ${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
+
+add_compile_definitions(C10_USING_CUSTOM_GENERATED_MACROS)
 
 if(EXECUTORCH_CADENCE_CPU_RUNNER)
   include(${EXECUTORCH_ROOT}/build/Codegen.cmake)
@@ -74,10 +77,12 @@ endif()
 
 if(EXECUTORCH_NNLIB_OPT)
   set(TARGET_DIR hifi)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/${TARGET_DIR}/third-party/nnlib)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/${TARGET_DIR}/third-party/nnlib
+  ${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 elseif(EXECUTORCH_FUSION_G3_OPT)
   set(TARGET_DIR fusion_g3)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/${TARGET_DIR}/third-party/nnlib)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/${TARGET_DIR}/third-party/nnlib
+  ${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 else()
   set(TARGET_DIR reference)
 endif()

backends/cadence/aot/functions_hifi.yaml

@@ -219,11 +219,6 @@
     - arg_meta: null
       kernel_name: cadence::impl::HiFi::quantized_relu_per_tensor_out
 
-- func: cadence::quantized_linear.per_tensor_out(Tensor src, Tensor weight, Tensor bias, SymInt src_zero_point, SymInt weight_zero_point, SymInt out_multiplier, SymInt out_shift, SymInt out_zero_point, Tensor? offset, *, Tensor(a!) out) -> Tensor(a!)
-  kernels:
-    - arg_meta: null
-      kernel_name: cadence::impl::HiFi::quantized_linear_per_tensor_out
-
 - func: cadence::quantized_fully_connected.out(Tensor src, Tensor weight, Tensor bias, int src_zero_point, Tensor weight_zero_point, Tensor out_multiplier, Tensor out_shift, int out_zero_point, Tensor? offset, *, Tensor(a!) out) -> Tensor(a!)
   kernels:
     - arg_meta: null

backends/cadence/fusion_g3/operators/CMakeLists.txt

@@ -64,7 +64,8 @@ target_link_libraries(aten_ops_cadence PUBLIC executorch)
 target_link_libraries(aten_ops_cadence PRIVATE xa_nnlib)
 
 # Let files say "include <executorch/path/to/header.h>".
-set(_common_include_directories ${EXECUTORCH_ROOT}/..)
+set(_common_include_directories ${EXECUTORCH_ROOT}/..
+${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 
 target_include_directories(
   aten_ops_cadence PUBLIC ${ROOT_DIR}/.. ${CMAKE_BINARY_DIR}

backends/cadence/hifi/kernels/CMakeLists.txt

@@ -25,7 +25,8 @@ add_library(
   ${EXECUTORCH_ROOT}/backends/cadence/hifi/third-party/nnlib/xa_nn_transpose_32.c
 )
 # Let files say "include <executorch/path/to/header.h>".
-set(_common_include_directories ${EXECUTORCH_ROOT}/..)
+set(_common_include_directories ${EXECUTORCH_ROOT}/..
+${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 
 target_include_directories(
   cadence_kernels

backends/cadence/hifi/operators/CMakeLists.txt

@@ -67,7 +67,8 @@ target_link_libraries(aten_ops_cadence PUBLIC executorch)
 target_link_libraries(aten_ops_cadence PRIVATE cadence_kernels)
 
 # Let files say "include <executorch/path/to/header.h>".
-set(_common_include_directories ${EXECUTORCH_ROOT}/..)
+set(_common_include_directories ${EXECUTORCH_ROOT}/..
+${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 
 target_include_directories(
   aten_ops_cadence PUBLIC ${ROOT_DIR}/.. ${CMAKE_BINARY_DIR}
@@ -77,7 +78,7 @@ 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"
-             "op_quantize_per_tensor.cpp" "op_quantized_relu_out.cpp" "op_dequantize_per_tensor.cpp"
+             "op_quantize_per_tensor.cpp" "op_quantized_relu_out.cpp" "op_dequantize_per_tensor.cpp" "op_quantized_fully_connected_out"
 )
 target_include_directories(
   custom_ops PUBLIC ${ROOT_DIR}/.. ${CMAKE_BINARY_DIR}

backends/cadence/hifi/operators/op_clamp.cpp

@@ -328,7 +328,7 @@ Tensor& clamp_tensor_out(
     const executorch::aten::optional<Tensor>& min_opt,
     const executorch::aten::optional<Tensor>& max_opt,
     Tensor& out) {
-  clamp_Tensor_out(ctx, in, min_opt, max_opt, out);
+  return clamp_Tensor_out(ctx, in, min_opt, max_opt, out);
 }
 
 } // namespace native

backends/cadence/hifi/operators/op_mean.cpp

@@ -175,7 +175,7 @@ Tensor& mean_dim_out(
     bool keepdim,
     optional<ScalarType> dtype,
     Tensor& out) {
-  mean_out(ctx, in, dim_list, keepdim, dtype, out);
+  return mean_out(ctx, in, dim_list, keepdim, dtype, out);
 }
 
 } // namespace native

backends/cadence/hifi/operators/op_quantized_relu_out.cpp

@@ -100,10 +100,10 @@ void quantized_relu_per_tensor_out(
 void quantized_relu_out(
     KernelRuntimeContext& ctx,
     const Tensor& input,
-    const int64_t in_zero_point,
+    const Tensor& in_zero_point,
     const int64_t out_zero_point,
-    const int64_t out_multiplier,
-    const int64_t out_shift,
+    const Tensor& out_multiplier,
+    const Tensor& out_shift,
     Tensor& output) {
   quantized_relu_per_tensor_out(
       ctx,

backends/cadence/hifi/operators/op_softmax.cpp

@@ -200,7 +200,7 @@ Tensor& softmax_out(
     int64_t dim,
     bool half_to_float,
     Tensor& out) {
-  _softmax_out(ctx, in, dim, half_to_float, out);
+  return _softmax_out(ctx, in, dim, half_to_float, out);
 }
 
 } // namespace native

backends/cadence/hifi/operators/op_where.cpp

@@ -183,6 +183,15 @@ Tensor& where_self_out(
   return out;
 }
 
+Tensor& where_out(
+    RuntimeContext& ctx,
+    const Tensor& cond,
+    const Tensor& a,
+    const Tensor& b,
+    Tensor& out) {
+  return where_out(ctx, cond, a, b, out);
+}
+
 } // namespace native
 } // namespace HiFi
 } // namespace impl