Adding LT model for filter module

Author: ErickOF

modules/filter/include/ips_filter_lt_model.hpp

@@ -0,0 +1,164 @@
+#ifndef IPS_FILTER_LT_MODEL_HPP
+#define IPS_FILTER_LT_MODEL_HPP
+#ifdef IPS_DEBUG_EN
+#include <iostream>
+#endif // IPS_DEBUG_ENi
+#ifdef IPS_DUMP_EN
+#include <sstream>
+#endif // IPS_DUMP_EN
+#include <systemc.h>
+
+
+/**
+ * @brief Filter module.
+ * It takes care of filtering a image/kernel using a median filter or an
+ * equivalent convolution like:
+ *                 | 1/N^2 ... 1/N^2 |   | img(row - N/2, col - N/2) ... img(row + N/2, col + N/2) |
+ * img(row, col) = |  ...  ...  .... | * |            ...            ...            ...            |
+ *                 | 1/N^2 ... 1/N^2 |   | img(row + N/2, col - N/2) ... img(row + N/2, col + N/2) |
+ *
+ * @tparam IN - data type of the inputs
+ * @tparam OUT - data type of the outputs
+ * @tparam N - size of the kernel
+ */
+template <typename IN = sc_uint<8>, typename OUT = sc_uint<8>, uint8_t N = 3>
+SC_MODULE(Filter)
+{
+protected:
+  //----------------------------Internal Variables----------------------------
+#ifdef IPS_DUMP_EN
+  sc_trace_file* wf;
+#endif // IPS_DUMP_EN
+  OUT* img_window_tmp;
+  OUT* kernel;
+  OUT* result_ptr;
+
+  // Event to trigger the filter execution
+  sc_event event;
+
+  //-----------------------------Internal Methods-----------------------------
+  void exec_filter();
+  void init();
+
+public:
+  /**
+   * @brief Default constructor for Filter
+   */
+  SC_HAS_PROCESS(Filter);
+#ifdef IPS_DUMP_EN
+  /**
+   * @brief Construct a new Filter object
+   * 
+   * @param name - name of the module
+   * @param wf - waveform file pointer
+   */
+  Filter(sc_core::sc_module_name name, sc_core::sc_trace_file* wf)
+    : sc_core::sc_module(name), wf(wf)
+#else
+  /**
+   * @brief Construct a new Filter object
+   * 
+   * @param name - name of the module
+   */
+  Filter(sc_core::sc_module_name name) : sc_core::sc_module(name)
+#endif // IPS_DUMP_EN
+  {
+    // Calling this method by default since it is no time consumer
+    // It is assumed that this kernel is already loaded in the model
+    // Kernel does not change after synthesis
+    SC_METHOD(init);
+    // Thread waiting for the request
+    SC_THREAD(exec_filter);
+  }
+
+  //---------------------------------Methods---------------------------------
+  void filter(IN* img_window, OUT* result);
+};
+
+/**
+ * @brief Execute the image filtering
+ * 
+ */
+template <typename IN, typename OUT, uint8_t N>
+void Filter<IN, OUT, N>::exec_filter()
+{
+  size_t i;
+  size_t j;
+
+  while (true)
+  {
+    // Wait to peform the convolution
+    wait(this->event);
+
+    // Default value for the result depending on the output datatype
+    *(this->result_ptr) = (OUT) 0;
+
+    // Perform the convolution
+    for (i = 0; i < N; ++i)
+      for (j = 0; j < N; ++j)
+        *(this->result_ptr) += this->kernel[i * N + j] * this->img_window_tmp[i * N + j];
+  }
+}
+
+/**
+ * @brief Filtering image
+ * 
+ * @param img_window - image window to filter
+ * @param result - resultant pixel
+ */
+template <typename IN, typename OUT, uint8_t N>
+void Filter<IN, OUT, N>::filter(IN* img_window, OUT* result)
+{
+  size_t i;
+  size_t j;
+
+  // Default value for the result depending on the output datatype
+  this->result_ptr = result;
+
+  // Perform the convolution
+  for (i = 0; i < N; ++i)
+    for (j = 0; j < N; ++j)
+        this->img_window_tmp[i * N + j] = (OUT) img_window[i * N + j];
+
+  // N * N * copy_pixel_to_mem_time + mult + redux + copy_pixel_to_mem_time
+  // Image is copied pixel by pixel
+  const float DELAY_TIME = (N * N * 1) + 4 + 2 + 1;
+  this->event.notify(DELAY_TIME, SC_NS);
+}
+
+/**
+ * @brief Initializes a kernel of N x N with default value of 1 / (N^2)
+ * 
+ */
+template <typename IN, typename OUT, uint8_t N>
+void Filter<IN, OUT, N>::init()
+{
+  // Init a kernel of N x N with default value of 1 / (N * N)
+  this->kernel = new OUT[N * N];
+  std::fill_n(this->kernel, N * N, ((OUT) 1) / ((OUT) N * N));
+  // Init image window of N x N with default value of 1 / (N * N)
+  this->img_window_tmp = new OUT[N * N];
+#ifdef IPS_DEBUG_EN
+  // Print the initialized kernel
+  SC_REPORT_INFO(this->name(), "init result");
+  size_t i, j;
+
+  for (i = 0; i < N; ++i)
+  {
+    for (j = 0; j < N; ++j)
+    {
+      std::cout << "[" << this->kernel[i * N + j] << "]";
+
+#ifdef IPS_DUMP_EN
+      // Adding the signals to the waveform
+      std::ostringstream var_name;
+      var_name << "kernel_" << i << "_" << j;
+      sc_trace(this->wf, this->kernel[i * N + j], var_name.str());
+#endif // IPS_DUMP_EN
+    }
+
+    std::cout << std::endl;
+  }
+#endif // IPS_DEBUG_EN
+}
+#endif // IPS_FILTER_LT_MODEL_HPP

modules/filter/src/tb_filter.cpp

@@ -86,6 +86,7 @@ void run_one_window()
   // Instantiate filter module and do the connection
 #ifdef IPS_DUMP_EN
   Filter<IPS_IN_TYPE_TB, IPS_OUT_TYPE_TB, IPS_FILTER_KERNEL_SIZE> filter("filter", wf);
+  sc_trace(wf, result, "result");
 #else
   Filter<IPS_IN_TYPE_TB, IPS_OUT_TYPE_TB, IPS_FILTER_KERNEL_SIZE> filter("filter");
 #endif // IPS_DEBUG_EN
@@ -98,14 +99,19 @@ void run_one_window()
 #endif // IPS_DEBUG_EN
 
   // Apply convolution
+#ifdef IPS_FILTER_PV_EN
   filter.filter(img_window, result);
+#elif defined(IPS_FILTER_LT_EN)
+  filter.filter(img_window, &result);
+  sc_start(100, SC_NS);
+#endif // IPS_FILTER_XX_EN
+
 #ifdef IPS_DEBUG_EN
   SC_REPORT_INFO("TEST_MODE_ONE_WINDOW", "filtering");
   std::cout << "Result = " << result << std::endl;
 #endif // IPS_DEBUG_EN
 
 #ifdef IPS_DUMP_EN
-  sc_trace(wf, result, "result");
   sc_start(1, SC_NS);
 #endif // IPS_DUMP_EN