Merge neurun kernel and backend build (#6313)

- Merge neurun each backend's kernel and backend build
- Fix build error on cpu backend kernel by restrict build option
  - Invalid type comparison
  - Unused variable
  - Unused function

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>

diff --git a/runtimes/neurun/backend/acl_cl/CMakeLists.txt b/runtimes/neurun/backend/acl_cl/CMakeLists.txt
index c2f1fd6..2dd7959 100644 (file)
--- a/runtimes/neurun/backend/acl_cl/CMakeLists.txt
+++ b/runtimes/neurun/backend/acl_cl/CMakeLists.txt
@@ -5,13 +5,8 @@ if(NOT ARMCompute_FOUND)
 endif(NOT ARMCompute_FOUND)
 
 set(LIB_NEURUN_BACKEND_ACL_CL neurun_backend_acl_cl)
-set(LIB_NEURUN_KERNEL_ACL_CL neurun_kernel_acl_cl)
-
-add_subdirectory(kernel)
 
 file(GLOB_RECURSE SOURCES "*.cc")
-file(GLOB_RECURSE REMOVE_KERNEL_SOURCES "kernel/*.cc")
-list(REMOVE_ITEM SOURCES ${REMOVE_KERNEL_SOURCES})
 
 add_library(${LIB_NEURUN_BACKEND_ACL_CL} SHARED ${SOURCES})
 
@@ -20,7 +15,6 @@ target_include_directories(${LIB_NEURUN_BACKEND_ACL_CL} PUBLIC ${CMAKE_CURRENT_S
 
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_CL} arm_compute)
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_CL} arm_compute_ex)
-target_link_libraries(${LIB_NEURUN_BACKEND_ACL_CL} ${LIB_NEURUN_KERNEL_ACL_CL})
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_CL} neurun-core)
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_CL} ${LIB_NEURUN_BACKEND_ACL_COMMON})
 

diff --git a/runtimes/neurun/backend/acl_cl/kernel/CMakeLists.txt b/runtimes/neurun/backend/acl_cl/kernel/CMakeLists.txt
deleted file mode 100644 (file)
index 1dbeba3..0000000
--- a/runtimes/neurun/backend/acl_cl/kernel/CMakeLists.txt
+++ /dev/null
@@ -1,19 +0,0 @@
-# Unsupported architecture
-nnfw_find_package(ARMCompute QUIET)
-if(NOT ARMCompute_FOUND)
-  return()
-endif(NOT ARMCompute_FOUND)
-
-file(GLOB SOURCES "*.cc")
-
-add_library(${LIB_NEURUN_KERNEL_ACL_CL} STATIC ${SOURCES})
-
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_CL} PUBLIC ${NEURUN_INCLUDE_DIR})
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_CL} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/..)
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_CL} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../..)
-
-target_link_libraries(${LIB_NEURUN_KERNEL_ACL_CL} arm_compute)
-target_link_libraries(${LIB_NEURUN_KERNEL_ACL_CL} ${LIB_NEURUN_BACKEND_ACL_COMMON})
-
-set_target_properties(${LIB_NEURUN_KERNEL_ACL_CL} PROPERTIES POSITION_INDEPENDENT_CODE ON)
-set_target_properties(${LIB_NEURUN_KERNEL_ACL_CL} PROPERTIES OUTPUT_NAME kernel_acl_cl)

diff --git a/runtimes/neurun/backend/acl_cl/kernel/ConcatLayer.cc b/runtimes/neurun/backend/acl_cl/kernel/ConcatLayer.cc
index fee3272..04388ce 100644 (file)
--- a/runtimes/neurun/backend/acl_cl/kernel/ConcatLayer.cc
+++ b/runtimes/neurun/backend/acl_cl/kernel/ConcatLayer.cc
@@ -25,8 +25,9 @@
 namespace
 {
 
-bool matchSizeExceptAxis(const ::neurun::backend::acl_cl::operand::ICLTensor *t1,
-                         const ::neurun::backend::acl_cl::operand::ICLTensor *t2, uint32_t axis)
+inline bool matchSizeExceptAxis(const ::neurun::backend::acl_cl::operand::ICLTensor *t1,
+                                const ::neurun::backend::acl_cl::operand::ICLTensor *t2,
+                                uint32_t axis)
 {
   assert(t1->num_dimensions() <= 4);
   assert(t2->num_dimensions() <= 4);
@@ -82,8 +83,6 @@ template <typename T> bool ConcatLayer::concatenate()
   {
     uint32_t axis_offset = 0;
 
-    auto &queue = ::arm_compute::CLScheduler::get().queue();
-
     auto outout_fn = [&](::neurun::backend::operand::ITensor &out_tensor) {
       for (auto input : _input_allocs)
       {
@@ -92,13 +91,13 @@ template <typename T> bool ConcatLayer::concatenate()
         auto input_fn = [&](::neurun::backend::operand::ITensor &in_tensor) {
           auto &in_cl_tensor =
               static_cast<::neurun::backend::acl_cl::operand::ICLTensor &>(in_tensor);
-          for (int32_t i = 0; i < in_cl_tensor.info()->dimension(0); i++)
+          for (uint32_t i = 0; i < in_cl_tensor.info()->dimension(0); i++)
           {
-            for (int32_t j = 0; j < in_cl_tensor.info()->dimension(1); j++)
+            for (uint32_t j = 0; j < in_cl_tensor.info()->dimension(1); j++)
             {
-              for (int32_t k = 0; k < in_cl_tensor.info()->dimension(2); k++)
+              for (uint32_t k = 0; k < in_cl_tensor.info()->dimension(2); k++)
               {
-                for (int32_t l = 0; l < in_cl_tensor.info()->dimension(3); l++)
+                for (uint32_t l = 0; l < in_cl_tensor.info()->dimension(3); l++)
                 {
                   int32_t io = (_axis == 0) ? axis_offset : 0;
                   int32_t jo = (_axis == 1) ? axis_offset : 0;

diff --git a/runtimes/neurun/backend/acl_neon/CMakeLists.txt b/runtimes/neurun/backend/acl_neon/CMakeLists.txt
index 8b1ae5d..0ffcc6b 100644 (file)
--- a/runtimes/neurun/backend/acl_neon/CMakeLists.txt
+++ b/runtimes/neurun/backend/acl_neon/CMakeLists.txt
@@ -5,24 +5,18 @@ if(NOT ARMCompute_FOUND)
 endif(NOT ARMCompute_FOUND)
 
 set(LIB_NEURUN_BACKEND_ACL_NEON neurun_backend_acl_neon)
-set(LIB_NEURUN_KERNEL_ACL_NEON neurun_kernel_acl_neon)
-add_subdirectory(kernel)
 
 file(GLOB_RECURSE SOURCES "*.cc")
 
-file(GLOB_RECURSE REMOVE_KERNEL_SOURCES "kernel/*.cc")
-list(REMOVE_ITEM SOURCES ${REMOVE_KERNEL_SOURCES})
-
 add_library(${LIB_NEURUN_BACKEND_ACL_NEON} SHARED ${SOURCES})
 
 target_include_directories(${LIB_NEURUN_BACKEND_ACL_NEON} PUBLIC ${NEURUN_INCLUDE_DIR})
 target_include_directories(${LIB_NEURUN_BACKEND_ACL_NEON} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
 #following line it to reuse codes of other backends
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_NEON} PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/..)
+target_include_directories(${LIB_NEURUN_BACKEND_ACL_NEON} PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/..)
 
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_NEON} arm_compute)
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_NEON} arm_compute_ex)
-target_link_libraries(${LIB_NEURUN_BACKEND_ACL_NEON} ${LIB_NEURUN_KERNEL_ACL_NEON})
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_NEON} neurun-core)
 
 target_link_libraries(${LIB_NEURUN_BACKEND_ACL_NEON} ${LIB_NEURUN_BACKEND_ACL_COMMON})

diff --git a/runtimes/neurun/backend/acl_neon/kernel/CMakeLists.txt b/runtimes/neurun/backend/acl_neon/kernel/CMakeLists.txt
deleted file mode 100644 (file)
index 3944436..0000000
--- a/runtimes/neurun/backend/acl_neon/kernel/CMakeLists.txt
+++ /dev/null
@@ -1,20 +0,0 @@
-# Unsupported architecture
-nnfw_find_package(ARMCompute QUIET)
-if(NOT ARMCompute_FOUND)
-  return()
-endif(NOT ARMCompute_FOUND)
-
-file(GLOB SOURCES "*.cc")
-
-add_library(${LIB_NEURUN_KERNEL_ACL_NEON} STATIC ${SOURCES})
-
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_NEON} PUBLIC ${NEURUN_INCLUDE_DIR})
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_NEON} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/..)
-#following line it to reuse codes of other backends
-target_include_directories(${LIB_NEURUN_KERNEL_ACL_NEON} PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/../..)
-
-target_link_libraries(${LIB_NEURUN_KERNEL_ACL_NEON} arm_compute)
-target_link_libraries(${LIB_NEURUN_KERNEL_ACL_NEON} ${LIB_NEURUN_BACKEND_ACL_COMMON})
-
-set_target_properties(${LIB_NEURUN_KERNEL_ACL_NEON} PROPERTIES POSITION_INDEPENDENT_CODE ON)
-set_target_properties(${LIB_NEURUN_KERNEL_ACL_NEON} PROPERTIES OUTPUT_NAME kernel_acl_neon)

diff --git a/runtimes/neurun/backend/acl_neon/kernel/ConcatLayer.cc b/runtimes/neurun/backend/acl_neon/kernel/ConcatLayer.cc
index 8d73980..49ffa7b 100644 (file)
--- a/runtimes/neurun/backend/acl_neon/kernel/ConcatLayer.cc
+++ b/runtimes/neurun/backend/acl_neon/kernel/ConcatLayer.cc
@@ -23,8 +23,9 @@
 namespace
 {
 
-bool matchSizeExceptAxis(const ::neurun::backend::acl_neon::operand::INETensor *t1,
-                         const ::neurun::backend::acl_neon::operand::INETensor *t2, uint32_t axis)
+inline bool matchSizeExceptAxis(const ::neurun::backend::acl_neon::operand::INETensor *t1,
+                                const ::neurun::backend::acl_neon::operand::INETensor *t2,
+                                uint32_t axis)
 {
   assert(t1->num_dimensions() <= 4);
   assert(t2->num_dimensions() <= 4);
@@ -84,13 +85,13 @@ template <typename T> bool ConcatLayer::concatenate()
 
     for (auto input : _input_allocs)
     {
-      for (int32_t i = 0; i < input->info()->dimension(0); i++)
+      for (uint32_t i = 0; i < input->info()->dimension(0); i++)
       {
-        for (int32_t j = 0; j < input->info()->dimension(1); j++)
+        for (uint32_t j = 0; j < input->info()->dimension(1); j++)
         {
-          for (int32_t k = 0; k < input->info()->dimension(2); k++)
+          for (uint32_t k = 0; k < input->info()->dimension(2); k++)
           {
-            for (int32_t l = 0; l < input->info()->dimension(3); l++)
+            for (uint32_t l = 0; l < input->info()->dimension(3); l++)
             {
               uint32_t io = (_axis == 0) ? axis_offset : 0;
               uint32_t jo = (_axis == 1) ? axis_offset : 0;

diff --git a/runtimes/neurun/backend/cpu/CMakeLists.txt b/runtimes/neurun/backend/cpu/CMakeLists.txt
index fb75d6e..029fd03 100644 (file)
--- a/runtimes/neurun/backend/cpu/CMakeLists.txt
+++ b/runtimes/neurun/backend/cpu/CMakeLists.txt
@@ -1,19 +1,12 @@
 set(LIB_NEURUN_BACKEND_CPU neurun_backend_cpu)
-set(LIB_NEURUN_KERNEL_CPU neurun_kernel_cpu)
-
-add_subdirectory(kernel)
 
 file(GLOB_RECURSE SOURCES "*.cc")
-file(GLOB_RECURSE REMOVE_KERNEL_SOURCES "kernel/*.cc")
 file(GLOB_RECURSE TESTS "*.test.cc")
-list(REMOVE_ITEM SOURCES ${REMOVE_KERNEL_SOURCES})
 list(REMOVE_ITEM SOURCES ${TESTS})
 
 add_library(${LIB_NEURUN_BACKEND_CPU} SHARED ${SOURCES})
 
-target_link_libraries(${LIB_NEURUN_BACKEND_CPU} nnfw_lib_misc)
-target_link_libraries(${LIB_NEURUN_BACKEND_CPU} nnfw_lib_cpp14)
-target_link_libraries(${LIB_NEURUN_BACKEND_CPU} ${LIB_NEURUN_KERNEL_CPU})
+target_link_libraries(${LIB_NEURUN_BACKEND_CPU} nnfw_lib_misc nnfw_lib_cpp14 nnfw_lib_cker)
 target_link_libraries(${LIB_NEURUN_BACKEND_CPU} neurun-core)
 
 target_compile_options(${LIB_NEURUN_BACKEND_CPU} PRIVATE -Wall -Wextra -Werror)

diff --git a/runtimes/neurun/backend/cpu/kernel/AddLayer.cc b/runtimes/neurun/backend/cpu/kernel/AddLayer.cc
index 56ad6b6..14e2afe 100644 (file)
--- a/runtimes/neurun/backend/cpu/kernel/AddLayer.cc
+++ b/runtimes/neurun/backend/cpu/kernel/AddLayer.cc
@@ -47,9 +47,9 @@ void AddLayer::addQuant8()
   int32_t output_activation_min, output_activation_max;
   CalculateActivationRangeUint8(_activation, _outputShape, &output_activation_min,
                                 &output_activation_max);
-  nnfw::cker::AddParam op_params;
-  op_params.quantized_activation_max = output_activation_max;
-  op_params.quantized_activation_min = output_activation_min;
+  // nnfw::cker::AddParam op_params;
+  // op_params.quantized_activation_max = output_activation_max;
+  // op_params.quantized_activation_min = output_activation_min;
 
   // cker quant8 add is not implemented yet
   throw std::runtime_error{"NYI"};

diff --git a/runtimes/neurun/backend/cpu/kernel/CMakeLists.txt b/runtimes/neurun/backend/cpu/kernel/CMakeLists.txt
deleted file mode 100644 (file)
index fa95817..0000000
--- a/runtimes/neurun/backend/cpu/kernel/CMakeLists.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-file(GLOB SOURCES "*.cc")
-
-add_library(${LIB_NEURUN_KERNEL_CPU} STATIC ${SOURCES})
-
-target_include_directories(${LIB_NEURUN_KERNEL_CPU} PUBLIC ${NEURUN_INCLUDE_DIR})
-
-target_link_libraries(${LIB_NEURUN_KERNEL_CPU} nnfw_lib_misc nnfw_lib_cpp14 nnfw_lib_cker)
-
-set_target_properties(${LIB_NEURUN_KERNEL_CPU} PROPERTIES POSITION_INDEPENDENT_CODE ON)
-set_target_properties(${LIB_NEURUN_KERNEL_CPU} PROPERTIES OUTPUT_NAME kernel_cpu)

diff --git a/runtimes/neurun/backend/cpu/kernel/ConcatLayer.cc b/runtimes/neurun/backend/cpu/kernel/ConcatLayer.cc
index 1dac452..c390436 100644 (file)
--- a/runtimes/neurun/backend/cpu/kernel/ConcatLayer.cc
+++ b/runtimes/neurun/backend/cpu/kernel/ConcatLayer.cc
@@ -68,7 +68,7 @@ void ConcatLayer::concatenationFloat32()
 }
 void ConcatLayer::concatenationQuant8()
 {
-  int num_inputs = _inputShapes.size();
+  uint32_t num_inputs = _inputShapes.size();
 
   std::vector<int32_t> input_zeropoints(num_inputs);
   std::vector<float> input_scales(num_inputs);

diff --git a/runtimes/neurun/backend/cpu/kernel/OperationUtils.cc b/runtimes/neurun/backend/cpu/kernel/OperationUtils.cc
index 7de047e..d30eb00 100644 (file)
--- a/runtimes/neurun/backend/cpu/kernel/OperationUtils.cc
+++ b/runtimes/neurun/backend/cpu/kernel/OperationUtils.cc
@@ -80,6 +80,7 @@ void GetQuantizedConvolutionMultiplier(const Shape &inputShape, const Shape &fil
   const float bias_scale = biasShape.scale;
   const float output_scale = outputShape.scale;
   // The following conditions must be guaranteed by the training pipeline.
+  UNUSED_RELEASE(bias_scale);
   assert(std::abs(input_product_scale - bias_scale) <=
          1e-6 * std::min(input_product_scale, bias_scale));
   assert(input_product_scale >= 0);

diff --git a/runtimes/neurun/backend/cpu/kernel/PermuteLayer.h b/runtimes/neurun/backend/cpu/kernel/PermuteLayer.h
index e3c3522..a3cca0c 100644 (file)
--- a/runtimes/neurun/backend/cpu/kernel/PermuteLayer.h
+++ b/runtimes/neurun/backend/cpu/kernel/PermuteLayer.h
@@ -61,7 +61,6 @@ private:
     auto rank = _output_shape.rank();
     auto fn = [&](::neurun::backend::operand::ITensor &in_tensor) {
       _output->access([&](::neurun::backend::operand::ITensor &out_tensor) {
-        auto layout = out_tensor.layout();
         auto input_buffer = in_tensor.buffer();
         auto input_size = in_tensor.total_size();
         auto output_buffer = out_tensor.buffer();

diff --git a/runtimes/neurun/backend/hi_perf_cpu/CMakeLists.txt b/runtimes/neurun/backend/hi_perf_cpu/CMakeLists.txt
index f30ef8c..af27d85 100644 (file)
--- a/runtimes/neurun/backend/hi_perf_cpu/CMakeLists.txt
+++ b/runtimes/neurun/backend/hi_perf_cpu/CMakeLists.txt
@@ -13,19 +13,14 @@ if(NOT BUILD_NEURUN_HI_PERF_CPU_BACKEND)
   return()
 endif(NOT BUILD_NEURUN_HI_PERF_CPU_BACKEND)
 
-add_subdirectory(kernel)
-
 file(GLOB_RECURSE SOURCES "*.cc")
-file(GLOB_RECURSE REMOVE_KERNEL_SOURCES "kernel/*.cc")
 file(GLOB_RECURSE TESTS "*.test.cc")
-#list(REMOVE_ITEM SOURCES ${REMOVE_KERNEL_SOURCES})
 list(REMOVE_ITEM SOURCES ${TESTS})
 
 add_library(${LIB_NEURUN_BACKEND_HI_PERF_CPU} SHARED ${SOURCES})
 
 target_link_libraries(${LIB_NEURUN_BACKEND_HI_PERF_CPU} nnfw_lib_misc)
 target_link_libraries(${LIB_NEURUN_BACKEND_HI_PERF_CPU} nnfw_lib_cpp14)
-target_link_libraries(${LIB_NEURUN_BACKEND_HI_PERF_CPU} ${LIB_NEURUN_KERNEL_NNPACK})
 target_link_libraries(${LIB_NEURUN_BACKEND_HI_PERF_CPU} neurun-core)
 
 target_compile_options(${LIB_NEURUN_BACKEND_HI_PERF_CPU} PRIVATE -Wall -Wextra -Werror)

diff --git a/runtimes/neurun/backend/hi_perf_cpu/kernel/CMakeLists.txt b/runtimes/neurun/backend/hi_perf_cpu/kernel/CMakeLists.txt
deleted file mode 100644 (file)
index c643e26..0000000
--- a/runtimes/neurun/backend/hi_perf_cpu/kernel/CMakeLists.txt
+++ /dev/null
@@ -1 +0,0 @@
-return()