Rebasing with latest MR on dev

.gitlab-ci.yml

@@ -20,11 +20,11 @@ include:
 
       - '.gitlab/ci/ubuntu_python.gitlab-ci.yml'
       - '.gitlab/ci/release/cibuildwheel_ubuntu.gitlab-ci.yml'   
+ 
+      # - '.gitlab/ci/windows_cpp.gitlab-ci.yml'
 
-      - '.gitlab/ci/windows_cpp.gitlab-ci.yml'
-
-      - '.gitlab/ci/windows_python.gitlab-ci.yml'   
-      - '.gitlab/ci/release/cibuildwheel_windows.gitlab-ci.yml'   
+      # - '.gitlab/ci/windows_python.gitlab-ci.yml'   
+      # - '.gitlab/ci/release/cibuildwheel_windows.gitlab-ci.yml'   
 
 
 test:ubuntu_python:
@@ -49,28 +49,28 @@ coverage:ubuntu_python:
 release:pip:ubuntu:
   before_script:
     - !reference [.retrieve_deps:apt, script]
-    - DEPS_NAMES=("aidge_core" "aidge_backend_cpu" "aidge_onnx")
+    - DEPS_NAMES=("aidge_core" "aidge_backend_cpu" "aidge_backend_cuda" "aidge_onnx")
     - DEPENDENCY_JOB="build:ubuntu_python"
     - !reference [.ubuntu:download:repositories, script] # located in common.gitlab-ci.yml
     - curl -sSL https://get.docker.com/ | sh
 
 
-release:pip:windows:
-  before_script:
-    - (Get-WmiObject -Class Win32_OperatingSystem).OSArchitecture
-    - (Get-WmiObject -Class Win32_OperatingSystem).Caption
-    - $PSVersionTable.PSVersion
-    # Install Chocolatey
-    - Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; iex ((New-Object System.Net.WebClient).DownloadString('https://community.chocolatey.org/install.ps1'))
-    # Install dependencies
-    - choco install cmake.install --installargs '"ADD_CMAKE_TO_PATH=System"' -Y
-    - choco install git -Y
-    - choco install python --version=$python_version -Y 
-    # Update PATH
-    - $env:Path = [System.Environment]::GetEnvironmentVariable("Path","Machine") + ";" + [System.Environment]::GetEnvironmentVariable("Path","User")
-    - python -m pip install cibuildwheel==2.17.0
-    # Download repositories
-    - $DEPS_NAMES = "aidge_core","aidge_backend_cpu","aidge_onnx"
-    - $DEPENDENCY_JOB="build:windows_python"
-    - !reference [.windows:download:repositories, script] 
+# release:pip:windows:
+#   before_script:
+#     - (Get-WmiObject -Class Win32_OperatingSystem).OSArchitecture
+#     - (Get-WmiObject -Class Win32_OperatingSystem).Caption
+#     - $PSVersionTable.PSVersion
+#     # Install Chocolatey
+#     - Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; iex ((New-Object System.Net.WebClient).DownloadString('https://community.chocolatey.org/install.ps1'))
+#     # Install dependencies
+#     - choco install cmake.install --installargs '"ADD_CMAKE_TO_PATH=System"' -Y
+#     - choco install git -Y
+#     - choco install python --version=$python_version -Y 
+#     # Update PATH
+#     - $env:Path = [System.Environment]::GetEnvironmentVariable("Path","Machine") + ";" + [System.Environment]::GetEnvironmentVariable("Path","User")
+#     - python -m pip install cibuildwheel==2.17.0
+#     # Download repositories
+#     - $DEPS_NAMES = "aidge_core","aidge_backend_cpu","aidge_backend_cuda","aidge_onnx"
+#     - $DEPENDENCY_JOB="build:windows_python"
+#     - !reference [.windows:download:repositories, script] 
 

CMakeLists.txt

 # CMake >= 3.18 is required for good support of FindCUDAToolkit
 cmake_minimum_required(VERSION 3.18) # XXX 3.18
+set(CXX_STANDARD 14)
 
 file(READ "${CMAKE_SOURCE_DIR}/version.txt" version)
 file(READ "${CMAKE_SOURCE_DIR}/project_name.txt" project)
@@ -25,6 +26,7 @@ option(WERROR "Warning as error" OFF)
 option(TEST "Enable tests" ON)
 option(COVERAGE "Enable coverage" OFF)
 option(CUDA "Enable CUDA backend" ON) # XXX OFF
+option(ENABLE_ASAN "Enable ASan (AddressSanitizer) for runtime analysis of memory use (over/underflow, memory leak, ...)" OFF)
 
 ##############################################
 # Import utils CMakeLists
@@ -43,6 +45,14 @@ if(CUDA)
     add_definitions(-DCUDA_COMPILER_VERSION="${CMAKE_CUDA_COMPILER_VERSION}")
 endif()
 
+if(NOT $ENV{AIDGE_INSTALL} STREQUAL "")
+    set(CMAKE_INSTALL_PREFIX $ENV{AIDGE_INSTALL})
+    list(APPEND CMAKE_PREFIX_PATH $ENV{AIDGE_INSTALL})
+    message(WARNING "Env var AIDGE_INSTALL detected : $ENV{AIDGE_INSTALL}. Set CMAKE_INSTALL_PREFIX to AIDGE_INSTALL & added to CMAKE_PREFIX_PATH"
+                    "\n\tCMAKE_INSTALL_PREFIX = ${CMAKE_INSTALL_PREFIX}"
+                    "\n\tCMAKE_PREFIX_PATH = ${CMAKE_PREFIX_PATH}")
+endif()
+
 # ##############################################
 # Find system dependencies
 Include(FetchContent)

python_binding/pybind_PTQ.cpp

@@ -18,7 +18,6 @@
 #include "aidge/quantization/PTQ/CLE.hpp"
 #include "aidge/quantization/PTQ/PTQ.hpp"
 
-#include "aidge/hook/Hook.hpp"
 #include "aidge/graph/GraphView.hpp"
 
 namespace py = pybind11;
@@ -94,7 +93,7 @@ void init_PTQ(py::module &m) {
     :type verbose: bool
     )mydelimiter");
 
-    m.def("quantize_network", &quantizeNetwork ,py::arg("network"), py::arg("nb_bits"), py::arg("input_dataset"), py::arg("clipping_mode") = Clipping::MAX , py::arg("apply_rounding") = true, py::arg("optimize_signs") = false, py::arg("single_shift") = false,  py::arg("use_cuda"), py::arg("verbose") = false,
+    m.def("quantize_network", &quantizeNetwork ,py::arg("network"), py::arg("nb_bits"), py::arg("input_dataset"), py::arg("clipping_mode") = Clipping::MAX , py::arg("no_quantization") = true, py::arg("optimize_signs") = false, py::arg("single_shift") = false,  py::arg("use_cuda") = false, py::arg("verbose") = false,
     R"mydelimiter(   
     Main quantization routine. Performs every step of the quantization pipeline.
     :param network: The GraphView to be quantized.

python_binding/pybind_QAT_FixedQ.cpp

@@ -13,7 +13,6 @@
 #include <pybind11/stl.h>
 
 #include "aidge/quantization/QAT/QAT_FixedQ.hpp"
-#include "aidge/hook/Hook.hpp"
 #include "aidge/graph/GraphView.hpp"
 
 namespace py = pybind11;

python_binding/pybind_QAT_LSQ.cpp

@@ -13,7 +13,6 @@
 #include <pybind11/stl.h>
 
 #include "aidge/quantization/QAT/QAT_LSQ.hpp"
-#include "aidge/hook/Hook.hpp"
 #include "aidge/graph/GraphView.hpp"
 
 namespace py = pybind11;

python_binding/recipes/pybind_QuantRecipes.cpp

@@ -13,7 +13,7 @@
 #include <pybind11/stl.h>
 
 #include "aidge/recipes/QuantRecipes.hpp"
-#include "aidge/hook/Hook.hpp"
+//#include "aidge/hook/Hook.hpp"
 #include "aidge/graph/GraphView.hpp"
 
 namespace py = pybind11;

src/QAT/QAT_LSQ.cpp

@@ -88,11 +88,19 @@ void QuantLSQ::insertQuantizers(std::shared_ptr<GraphView> graphView, size_t nbB
 
 static float getTensorAbsMean(std::shared_ptr<Tensor> tensor)
 {
+    auto backend = tensor->backend();
+    if (backend == "cuda")
+        tensor->setBackend("cpu");
+
     float acc = 0;
     float* castedTensor = static_cast<float *> (tensor->getImpl()->rawPtr());
     for(std::size_t i = 0; i < tensor->size(); i++)
         acc += std::abs(castedTensor[i]);
     acc /= static_cast<float> (tensor->size());
+
+    if (backend == "cuda")
+        tensor->setBackend("cuda");
+
     return acc;
 }
 
@@ -107,7 +115,7 @@ static std::map<std::string, float> collectInputStats(std::shared_ptr<GraphView>
     // Store the input tensor statistics
 
     if (useCuda)
-        graphView->setBackend("cpu");
+        graphView->setBackend("cpu"); 
 
     std::map<std::string, float> inputStats;
     for (auto node : graphView->getNodes())
@@ -192,6 +200,7 @@ void QuantLSQ::insertAndInitQuantizers(std::shared_ptr<GraphView> graphView, siz
 
     // Collect the tensor statisics
     auto inputStats = collectInputStats(graphView, calibrationData, useCuda);
+
     auto paramStats = collectParamStats(graphView, useCuda);
 
     // Insert the quantizers

src/recipes/QuantRecipes.cpp

@@ -62,7 +62,7 @@ void insertBatchNormNodes(std::shared_ptr<GraphView> graphView)
             std::cout << " NB CHANNELS = " << nb_channels << std::endl; // TODO : remove this ...
 
             std::string batchnormNodeName = makeUniqueName(parentNode->name() + "_BN", graphView);
-            std::shared_ptr<Node> batchnormNode = BatchNorm<2>(nb_channels, 1e-5, 0.1, /*0,*/ batchnormNodeName);
+            std::shared_ptr<Node> batchnormNode = BatchNorm<2>(nb_channels, 1e-5, 0.1, batchnormNodeName);
             batchnormNode->getOperator()->setDataType(DataType::Float32);
             batchnormNode->getOperator()->setBackend("cpu");