#5 Changed diffusion to work with uint16 instead of uint8

src/lib/cpp/gpu/diffusion.cc

@@ -230,7 +230,7 @@ namespace gpu {
 
     void diffusion_step(const uint8_t *__restrict__ voxels, float *buf0, float *buf1, const shape_t &N, const float *__restrict__ kernel, const int64_t radius) {
         for (int64_t dim = 0; dim < 3; dim++) {
-            diffusion_core(buf0, kernel, buf1, N, dim, radius, 0);
+            diffusion_core(buf0, kernel, buf1, N, dim, radius);
             std::swap(buf0, buf1);
         }
         illuminate(voxels, buf0, N.z*N.y*N.x);
@@ -243,7 +243,7 @@ namespace gpu {
         }
     }
 
-    void diffusion_in_memory(const uint8_t *__restrict__ voxels, const shape_t &N, const float *__restrict__ kernel, const int64_t kernel_size, const int64_t repititions, uint8_t *__restrict__ output) {
+    void diffusion_in_memory(const uint8_t *__restrict__ voxels, const shape_t &N, const float *__restrict__ kernel, const int64_t kernel_size, const int64_t repititions, uint16_t *__restrict__ output) {
         const int64_t
             total_size = N.z*N.y*N.x,
             radius = kernel_size / 2;
@@ -258,7 +258,7 @@ namespace gpu {
                 diffusion_step(voxels, buf0, buf1, N, kernel, radius);
                 std::swap(buf0, buf1);
             }
-            convert_float_to_uint8(buf0, output, total_size);
+            convert_float_to_uint16(buf0, output, total_size);
         }
 
         delete[] buf0;
@@ -336,8 +336,8 @@ namespace gpu {
             }
         }
 
-        // Convert to uint8
-        convert_float_to_uint8(temp0, output_file, total_flat_size);
+        // Convert to uint16
+        convert_float_to_uint16(temp0, output_file, total_flat_size);
 
         // Free memory
         free(buf0);

src/lib/cpp/include/diffusion.hh

@@ -9,7 +9,7 @@ namespace NS {
     //void convert_uint8_to_float(const std::string &src, const std::string &dst, const int64_t total_flat_size);
     //void diffusion(const std::string &input_file, const std::vector<float>& kernel, const std::string &output_file, const shape_t &total_shape, const shape_t &global_shape, const int64_t repititions, const bool verbose = false);
     //void diffusion_core(const float *__restrict__ input, const float *__restrict__ kernel, float *__restrict__ output, const shape_t &N, const int64_t dim, const int64_t radius, const int64_t padding);
-    void diffusion_in_memory(const uint8_t *__restrict__ voxels, const shape_t &N, const double *__restrict__ kernel, const int64_t kernel_size, const int64_t repititions, uint8_t *__restrict__ output);
+    void diffusion_in_memory(const uint8_t *__restrict__ voxels, const shape_t &N, const double *__restrict__ kernel, const int64_t kernel_size, const int64_t repititions, uint16_t *__restrict__ output);
     //void illuminate(const bool *__restrict__ mask, float *__restrict__ output, const int64_t local_flat_size);
     //void store_mask(const float *__restrict__ input, bool *__restrict__ mask, const int64_t local_flat_size);
     //void stage_to_device(float *__restrict__ stage, const float *__restrict__ src, const idx3drange &range, const shape_t &global_shape, const int64_t kernel_size);

src/processing_steps/0900_generate_gauss_c.py

@@ -14,7 +14,7 @@
 NA = np.newaxis
 
 internal_type = np.float32
-result_type = np.uint8
+result_type = np.uint16
 
 def toint(arr, dtype=np.uint8):
     vmin, vmax = arr.min(), arr.max()

src/pybind/diffusion-pybind.cc

@@ -2,15 +2,15 @@
 
 namespace python_api {
 
-    void diffusion_in_memory(const np_array<uint8_t> &np_voxels, const np_array<float> &np_kernel, np_array<uint8_t> &np_output, const int64_t repititions = 1) {
+    void diffusion_in_memory(const np_array<uint8_t> &np_voxels, const np_array<float> &np_kernel, np_array<uint16_t> &np_output, const int64_t repititions = 1) {
         auto
             voxels_info = np_voxels.request(),
             kernel_info = np_kernel.request(),
             output_info = np_output.request();
 
         const uint8_t *voxels = static_cast<const uint8_t*>(voxels_info.ptr);
         const float *kernel = static_cast<const float*>(kernel_info.ptr);
-        uint8_t *output = static_cast<uint8_t*>(output_info.ptr);
+        uint16_t *output = static_cast<uint16_t*>(output_info.ptr);
 
         const shape_t N = {voxels_info.shape[0], voxels_info.shape[1], voxels_info.shape[2]};
         const int64_t kernel_size = kernel_info.shape[0];