import span to speed up & fix memory leak

Author: ivanallen

examples/eager_op_multithread/CMakeLists.txt

@@ -6,9 +6,11 @@ find_package(Threads REQUIRED)
 
 set(CMAKE_CXX_STANDARD 17)
 
+
 #set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer -fsanitize=thread")
 #set(CMAKE_LINKER_FLAGS "${CMAKE_LINKER_FLAGS} -fno-omit-frame-pointer -fsanitize=thread")
 
 add_executable(example main.cpp)
 target_include_directories(example PRIVATE ../../include $ENV{HOME}/libtensorflow2/include)
+target_link_libraries (example "${TENSORFLOW_LIB}")
 target_link_libraries(example "${TENSORFLOW_LIB}" Threads::Threads)

examples/tensor_span/CMakeLists.txt

@@ -0,0 +1,16 @@
+cmake_minimum_required(VERSION 3.10)
+project(example)
+
+find_library(TENSORFLOW_LIB tensorflow HINT $ENV{HOME}/libtensorflow2/lib)
+find_package(Threads REQUIRED)
+
+#set(CMAKE_CXX_STANDARD 20)
+set(CMAKE_CXX_STANDARD 17)
+
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer -fsanitize=address")
+set(CMAKE_LINKER_FLAGS "${CMAKE_LINKER_FLAGS} -lasan")
+
+add_executable(example main.cpp)
+target_include_directories(example PRIVATE ../../include $ENV{HOME}/libtensorflow2/include)
+target_link_libraries (example "${TENSORFLOW_LIB}")

examples/tensor_span/main.cpp

@@ -0,0 +1,23 @@
+#include <cmath>
+#include <vector>
+#include <cassert>
+
+#include "cppflow/cppflow.h"
+
+template<typename T>
+void test(const std::initializer_list<T>& input) {
+    std::vector<T> values = input;
+    cppflow::tensor t(input);
+    auto span_result = t.get_data<T>();
+
+    assert(span_result.size() == values.size());
+    for(size_t i = 0; i < span_result.size(); i++) {
+        assert(std::abs(values[i]/span_result[i]-1.0f) < 1e-6);
+    }
+}
+
+int main() {
+    test({10, 20, 30});
+    test({10.0, 20.1, 30.3});
+    return 0;
+}

include/cppflow/context.h

@@ -6,6 +6,7 @@
 #define CPPFLOW2_CONTEXT_H
 
 #include <memory>
+#include <stdexcept>
 #include <utility>
 
 #include <tensorflow/c/c_api.h>

include/cppflow/model.h

@@ -13,6 +13,7 @@
 
 #include "context.h"
 #include "tensor.h"
+#include "defer.h"
 
 namespace cppflow {
 
@@ -84,6 +85,12 @@ namespace cppflow {
 
         std::vector<TF_Output> inp_ops(inputs.size());
         std::vector<TF_Tensor*> inp_val(inputs.size());
+
+        defer d([&inp_val]{
+            for (auto* tf_tensor : inp_val) {
+                TF_DeleteTensor(tf_tensor);
+            }
+        });
         for (int i=0; i<inputs.size(); i++) {
 
             // Operations