Merge remote-tracking branch 'upstream/master' into jiwaszki/fuse_mul…

…_fc_one_cmds

.github/workflows/check-format.yml

@@ -18,6 +18,7 @@ jobs:
   check-format:
     name: Check format
     runs-on: ubuntu-20.04
+    if: github.repository_owner == 'Samsung'
 
     steps:
       - name: Checkout
@@ -54,6 +55,7 @@ jobs:
   check-copyright:
     name: Check copyright
     runs-on: ubuntu-20.04
+    if: github.repository_owner == 'Samsung'
 
     steps:
       - name: Checkout

.github/workflows/check-pr-commit.yml

@@ -20,7 +20,7 @@ jobs:
     name: Check commit message
     runs-on: ubuntu-20.04
     # Skip on draft, check on draft -> ready
-    if: github.event.pull_request.draft == false
+    if: github.repository_owner == 'Samsung' && github.event.pull_request.draft == false
 
     steps:
       - name: Checkout

.github/workflows/run-onecc-build.yml

@@ -38,17 +38,23 @@ jobs:
   onecc-test:
     # Tested ubuntu version is decided by docker image, not runner
     runs-on: ubuntu-latest
+    if: github.repository_owner == 'Samsung'
     strategy:
-      # TODO Support various ubuntu version
       matrix:
         type: [ Debug, Release ]
+        ubuntu_code: [ focal, jammy ]
+        include:
+          - ubuntu_code: focal
+            ubuntu_ver: 20.04
+          - ubuntu_code: jammy
+            ubuntu_ver: 22.04
     container:
-      image: nnfw/one-devtools:focal
+      image: nnfw/one-devtools:${{ matrix.ubuntu_code }}
       options: --user root
     env:
       NNCC_WORKSPACE : build
       NNCC_INSTALL_PREFIX : install
-    name: onecc ubuntu 20.04 ${{ matrix.type }} test
+    name: onecc ubuntu ${{ matrix.ubuntu_ver }} ${{ matrix.type }} test
 
     steps:
       - name: Checkout
@@ -67,7 +73,7 @@ jobs:
         uses: actions/cache@v4
         with:
           path: ${{ env.NNCC_WORKSPACE }}/overlay
-          key: overlay-onecc-focal-${{ hashFiles('compiler/common-artifacts/CMakeLists.txt') }}-${{ hashFiles('infra/cmake/packages/**/*.cmake') }}
+          key: overlay-onecc-${{ matrix.ubuntu_code }}-${{ hashFiles('compiler/common-artifacts/CMakeLists.txt') }}-${{ hashFiles('infra/cmake/packages/**/*.cmake') }}
 
       - name: Build
         run: |

.github/workflows/run-onert-android-build.yml

@@ -1,29 +1,38 @@
 name: Run ONERT Android Release Build
 
 on:
+  push:
+    branches:
+      - master
+      - release/*
+    paths:
+      - '.github/workflows/run-onert-android-build.yml'
+      - 'nn*'
+      - 'Makefile.template'
+      - 'compute/**'
+      - 'infra/buildtool/**'
+      - 'infra/cmake/**'
+      - 'infra/nncc/**'
+      - 'infra/nnfw/**'
+      - 'runtime/**'
+      - 'tests/**'
+      - '!**/*.md'
   pull_request:
     branches:
       - master
       - release/*
     paths:
       - '.github/workflows/run-onert-android-build.yml'
-      - 'nnas'
-      - 'nnfw'
+      - 'nn*'
       - 'Makefile.template'
       - 'compute/**'
-      - 'infra/**'
-      - '!infra/debian/**'
-      - '!infra/docker/**'
-      - '!infra/doxygen/**'
-      - '!infra/git-hooks/**'
-      - '!infra/nncc/**'
-      - '!infra/onert-micro/**'
-      - '!infra/packaging/**'
-      - 'nnpackage/**'
-      - '!nnpackage/spec/**'
+      - 'infra/buildtool/**'
+      - 'infra/cmake/**'
+      - 'infra/nncc/**'
+      - 'infra/nnfw/**'
       - 'runtime/**'
       - 'tests/**'
-      - '!**.md'
+      - '!**/*.md'
 
 defaults:
   run:
@@ -32,6 +41,7 @@ defaults:
 jobs:
   build:
     runs-on: ubuntu-22.04
+    if: github.repository_owner == 'Samsung'
     env:
       TARGET_ARCH: aarch64
       TARGET_OS: android
@@ -45,8 +55,9 @@ jobs:
         uses: actions/cache@v4
         with:
           path: externals
-          key: external-onert-ndk26
+          key: external-onert-ndk26-${{ hashFiles('infra/cmake/packages/**/*.cmake') }}-${{ hashFiles('infra/nnfw/cmake/packages/**/*.cmake') }}
           restore-keys: |
+            external-onert-ndk26-
             external-onert-ndk
             external-onert-
             external-

.github/workflows/run-onert-micro-unit-tests.yml

@@ -1,6 +1,13 @@
 name: Run onert-micro Unit tests
 
 on:
+  push:
+    branches:
+      - master
+      - release/*
+    paths:
+      - 'onert-micro/**'
+      - '.github/workflows/run-onert-micro-unit-tests.yml'
   pull_request:
     branches:
       - master
@@ -23,7 +30,7 @@ jobs:
     name: Run onert-micro Unit tests
     runs-on: ubuntu-20.04
     # Skip on draft, check on draft -> ready
-    if: github.event.pull_request.draft == false
+    if: github.repository_owner == 'Samsung' && github.event.pull_request.draft == false
 
     steps:
       - name: Install Arm GNU Toolchain (arm-none-eabi-gcc)

compiler/circledump/driver/Driver.cpp

@@ -50,7 +50,11 @@ int entry(int argc, char **argv)
 
   std::cout << "Dump: " << circle_path << std::endl << std::endl;
 
-  std::cout << circlemodel << std::endl;
+  circledump::ModelEx modelex;
+  modelex.model = circlemodel;
+  modelex.rawdata = &modelData;
+
+  std::cout << modelex << std::endl;
 
   return 0;
 }

compiler/circledump/include/circledump/Dump.h

@@ -24,9 +24,16 @@
 namespace circledump
 {
 
-void dump_model(std::ostream &os, const circle::Model *model);
-}
+struct ModelEx
+{
+  const circle::Model *model;
+  const std::vector<char> *rawdata;
+};
+
+void dump_model(std::ostream &os, const circledump::ModelEx &model);
+
+} // namespace circledump
 
-std::ostream &operator<<(std::ostream &os, const circle::Model *model);
+std::ostream &operator<<(std::ostream &os, const circledump::ModelEx &model);
 
 #endif // __CIRCLEDUMP_DUMP_H__

compiler/circledump/src/Dump.cpp

@@ -341,9 +341,9 @@ void dump_sub_graph(std::ostream &os, mio::circle::Reader &reader)
   os << std::endl;
 }
 
-void dump_model(std::ostream &os, const circle::Model *model)
+void dump_model(std::ostream &os, const circle::Model *model, const std::vector<char> *rawdata)
 {
-  mio::circle::Reader reader(model);
+  mio::circle::Reader reader(model, rawdata);
 
   uint32_t num_subgraph = reader.num_subgraph();
 
@@ -378,13 +378,17 @@ void dump_model(std::ostream &os, const circle::Model *model)
   os << std::endl;
 
   // dump buffer
-  os << "Buffers: B(index) (length) values, if any" << std::endl;
+  os << "Buffers: B(index) (length) values, if any; (length *) for ext_offset" << std::endl;
   for (uint32_t i = 0; i < buffers->size(); ++i)
   {
+    bool ext_offset = false;
     const uint8_t *buff_data;
-    size_t size = reader.buffer_info(i, &buff_data);
+    size_t size = reader.buffer_info(i, &buff_data, ext_offset);
 
-    os << "B(" << i << ") (" << size << ") ";
+    os << "B(" << i << ") (" << size;
+    if (ext_offset)
+      os << " *";
+    os << ") ";
     if (buff_data != nullptr)
     {
       dump_buffer(os, buff_data, size, 16);
@@ -460,8 +464,8 @@ void dump_model(std::ostream &os, const circle::Model *model)
 
 } // namespace circledump
 
-std::ostream &operator<<(std::ostream &os, const circle::Model *model)
+std::ostream &operator<<(std::ostream &os, const circledump::ModelEx &modelex)
 {
-  circledump::dump_model(os, model);
+  circledump::dump_model(os, modelex.model, modelex.rawdata);
   return os;
 }

compiler/luci-interpreter/pal/linux/PALConv2d.h

@@ -30,14 +30,30 @@ static inline void Conv(const tflite::ConvParams &params, const tflite::RuntimeS
                         float *scratchpad_data)
 {
   (void)scratchpad_shape;
-  if (scratchpad_data)
+
+  const int32_t batches = tflite::MatchingDim(input_shape, 0, output_shape, 0);
+  const int32_t input_depth = tflite::MatchingDim(input_shape, 3, filter_shape, 3);
+  const int32_t output_height = output_shape.Dims(1);
+  const int32_t output_width = output_shape.Dims(2);
+  const int32_t filter_height = filter_shape.Dims(1);
+  const int32_t filter_width = filter_shape.Dims(2);
+
+  int64_t im2col_flat_size = 1;
+  im2col_flat_size *= batches;
+  im2col_flat_size *= output_height;
+  im2col_flat_size *= output_width;
+  im2col_flat_size *= input_depth;
+  im2col_flat_size *= filter_height;
+  im2col_flat_size *= filter_width;
+
+  // This condition checks if integer overflow will occur inside the optimized kernel.
+  // https://github.com/tensorflow/tensorflow/blob/v2.12.1/tensorflow/lite/kernels/internal/optimized/im2col_utils.h#L81
+  // If overflow is expected, we fall back to the reference kernel.
+  // NOTE This is just a rough check.
+  bool opt_kernel_overflow = im2col_flat_size > std::numeric_limits<int32_t>::max();
+
+  if (scratchpad_data and not opt_kernel_overflow)
   {
-    const int32_t batches = tflite::MatchingDim(input_shape, 0, output_shape, 0);
-    const int32_t input_depth = tflite::MatchingDim(input_shape, 3, filter_shape, 3);
-    const int32_t output_height = output_shape.Dims(1);
-    const int32_t output_width = output_shape.Dims(2);
-    const int32_t filter_height = filter_shape.Dims(1);
-    const int32_t filter_width = filter_shape.Dims(2);
     tflite::RuntimeShape im2col_shape{batches, output_height, output_width,
                                       input_depth * filter_height * filter_width};