Windows support and version bump (0.4.1)

TensorFlow support WIP. This should addresses
#7 and
#93

CMakeLists.txt

@@ -5,7 +5,7 @@ project(redner VERSION 0.0.2 DESCRIPTION "Differentiable Ray Tracer")
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake/")
 
 if(WIN32)
-    find_package(PythonLibs 3.6 COMPONENTS Development REQUIRED)
+    find_package(Python 3.6 COMPONENTS Development REQUIRED)
     add_subdirectory(pybind11)
 else()
     find_package(Python 3.6 COMPONENTS Development REQUIRED)
@@ -82,7 +82,6 @@ set(SRCS src/aabb.h
          src/primary_contribution.h
          src/primary_intersection.h
          src/ptr.h
-         src/py_utils.h
          src/ray.h
          src/rebuild_topology.h
          src/redner.h
@@ -104,7 +103,6 @@ set(SRCS src/aabb.h
          src/channels.cpp
          src/edge.cpp
          src/edge_tree.cpp
-         src/envmap.cpp
          src/load_serialized.cpp
          src/material.cpp
          src/miniz.c
@@ -131,7 +129,6 @@ if(REDNER_CUDA)
         src/channels.cpp
         src/edge.cpp
         src/edge_tree.cpp
-        src/envmap.cpp
         src/material.cpp
         src/parallel.cpp
         src/path_contribution.cpp
@@ -191,9 +188,15 @@ string( REPLACE "-DNDEBUG" "" CMAKE_C_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}"
 string( REPLACE "/DNDEBUG" "" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
 string( REPLACE "-DNDEBUG" "" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
 
-find_package(TensorFlow)
-if(TensorFlow_FOUND)
-    add_subdirectory(pyredner_tensorflow/custom_ops)
-else()
-    message(INFO " Building without TensorFlow support (not found)")
+if(MSVC)
+    target_compile_options(redner PRIVATE /W3 /wd4244 /wd4146 /wd4244 /wd4305 /wd4334 /wd4996)
 endif()
+
+if(NOT WIN32)
+    find_package(TensorFlow)
+    if(TensorFlow_FOUND)
+        add_subdirectory(pyredner_tensorflow/custom_ops)
+    else()
+        message(INFO " Building without TensorFlow support (not found)")
+    endif()
+endif()

cmake/FindTensorFlow.cmake

@@ -12,11 +12,16 @@ if("${retcode}" STREQUAL "0")
     list(GET TF_INFORMATION_LIST 1 TF_DETECTED_ABI)
     list(GET TF_INFORMATION_LIST 2 TF_DETECTED_INCLUDE_DIR)
     list(GET TF_INFORMATION_LIST 3 TF_DETECTED_LIBRARY_DIR)
-    # For some reason my tensorflow doesn't have a .so file
-    list(APPEND CMAKE_FIND_LIBRARY_SUFFIXES .so.1)
-    list(APPEND CMAKE_FIND_LIBRARY_SUFFIXES .so.2)
-    find_library(TF_DETECTED_LIBRARY NAMES tensorflow_framework PATHS 
-        ${TF_DETECTED_LIBRARY_DIR})
+    if(WIN32)
+        find_library(TF_DETECTED_LIBRARY NAMES _pywrap_tensorflow_internal PATHS 
+            ${TF_DETECTED_LIBRARY_DIR}/python)        
+    else()
+        # For some reason my tensorflow doesn't have a .so file
+        list(APPEND CMAKE_FIND_LIBRARY_SUFFIXES .so.1)
+        list(APPEND CMAKE_FIND_LIBRARY_SUFFIXES .so.2)
+        find_library(TF_DETECTED_LIBRARY NAMES tensorflow_framework PATHS 
+            ${TF_DETECTED_LIBRARY_DIR})
+    endif()
     set(TensorFlow_VERSION ${TF_DETECTED_VERSION})
     set(TensorFlow_ABI ${TF_DETECTED_ABI})
     set(TensorFlow_INCLUDE_DIR ${TF_DETECTED_INCLUDE_DIR})

pyredner_tensorflow/custom_ops/CMakeLists.txt

@@ -5,19 +5,17 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
 project(rednerTFCustomOp)
 
-# TODO: windows support
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
-set(CMAKE_EXE_LINKER_FLAGS  "${CMAKE_EXE_LINKER_FLAGS} -fPIC --shared")
+set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 include_directories(SYSTEM ${TensorFlow_INCLUDE_DIR})
 
 # Compile two versions of the library
-add_library(redner_tf_data_ptr_cxx11_abi MODULE data_ptr.cc)
+add_library(redner_tf_data_ptr_cxx11_abi SHARED data_ptr.cc)
 set_target_properties(redner_tf_data_ptr_cxx11_abi PROPERTIES COMPILE_FLAGS -D_GLIBCXX_USE_CXX11_ABI=1)
 set_target_properties(redner_tf_data_ptr_cxx11_abi PROPERTIES LINK_FLAGS -D_GLIBCXX_USE_CXX11_ABI=1)
 target_link_libraries(redner_tf_data_ptr_cxx11_abi ${TensorFlow_LIBRARY})
 
-add_library(redner_tf_data_ptr_no_cxx11_abi MODULE data_ptr.cc)
+add_library(redner_tf_data_ptr_no_cxx11_abi SHARED data_ptr.cc)
 set_target_properties(redner_tf_data_ptr_no_cxx11_abi PROPERTIES COMPILE_FLAGS -D_GLIBCXX_USE_CXX11_ABI=0)
 set_target_properties(redner_tf_data_ptr_no_cxx11_abi PROPERTIES LINK_FLAGS -D_GLIBCXX_USE_CXX11_ABI=0)
 target_link_libraries(redner_tf_data_ptr_no_cxx11_abi ${TensorFlow_LIBRARY})

pyredner_tensorflow/custom_ops/data_ptr.cc

@@ -2,6 +2,11 @@
 
 ==============================================================================*/
 
+#pragma warning(disable : 4003 4061 4100 4127 4242 4244 4267 4355 4365 4388 4464 4514 4574 4623 4625 4626 4647 4668 4710 4820 4946 5026 5027 5031 5039)
+
+// For windows
+#define NOMINMAX
+
 #include "tensorflow/core/framework/op.h"
 #include "tensorflow/core/framework/shape_inference.h"
 #include "tensorflow/core/framework/op_kernel.h"

setup.py

@@ -100,14 +100,14 @@ def build_extension(self, ext):
     import torch
     if torch.cuda.is_available():
         build_with_cuda = True
-if tf_spec is not None:
+if tf_spec is not None and sys.platform != 'win32':
     packages.append('pyredner_tensorflow')
     if not build_with_cuda:
         import tensorflow as tf
         if tf.test.is_gpu_available(cuda_only=True, min_cuda_compute_capability=None):
             build_with_cuda = True
 if len(packages) == 0:
-    print('Error: PyTorch or Tensorflow must be installed.')
+    print('Error: PyTorch or Tensorflow must be installed. For Windows platform only PyTorch is supported.')
     exit()
 # Override build_with_cuda with environment variable
 if 'REDNER_CUDA' in os.environ:
@@ -204,7 +204,7 @@ def build_extension(self, ext):
 if 'PROJECT_NAME' in os.environ:
     project_name = os.environ['PROJECT_NAME']
 setup(name = project_name,
-      version = '0.3.15',
+      version = '0.4.1',
       description = 'Differentiable rendering without approximation.',
       long_description = """redner is a differentiable renderer that can take the
                             derivatives of rendering output with respect to arbitrary

src/active_pixels.cpp

@@ -24,7 +24,7 @@ void init_active_pixels(const BufferView<Ray> &rays,
     auto new_end = DISPATCH_CACHED(use_gpu, thrust_alloc, thrust::remove_if,
         active_pixels.begin(), active_pixels.end(),
         active_pixels.begin(), op);
-    active_pixels.count = new_end - active_pixels.begin();
+    active_pixels.count = int(new_end - active_pixels.begin());
 }
 
 struct is_valid_intersection {
@@ -45,7 +45,7 @@ void update_active_pixels(const BufferView<int> &active_pixels,
     auto new_end = DISPATCH(use_gpu, thrust::copy_if,
         active_pixels.begin(), active_pixels.end(),
         new_active_pixels.begin(), op);
-    new_active_pixels.count = new_end - new_active_pixels.begin();
+    new_active_pixels.count = int(new_end - new_active_pixels.begin());
 }
 
 void test_active_pixels(bool use_gpu) {

src/atomic.h

@@ -37,7 +37,7 @@ inline T0 atomic_add(T0 &target, T1 source) {
     T0 new_val;
     do {
         old_val = target;
-        new_val = old_val + source;
+        new_val = old_val + (T0)source;
   #if defined(USE_GCC_INTRINSICS)
     } while (!__atomic_compare_exchange(&target, &old_val, &new_val, true,
         std::memory_order::memory_order_seq_cst,

src/automatic_uv_map.cpp

@@ -1,4 +1,5 @@
 #include "automatic_uv_map.h"
+
 #include <ctime>
 #include <cstdio>
 #include <cstdarg>
@@ -13,7 +14,7 @@ class Stopwatch {
     clock_t m_start;
 };
 
-static int Print(const char *format, ...) {
+int Print(const char *format, ...) {
     va_list arg;
     va_start(arg, format);
     printf("\r"); // Clear progress text (PrintProgress).
@@ -22,7 +23,7 @@ static int Print(const char *format, ...) {
     return result;
 }
 
-static void PrintProgress(const char *name, const char *indent1, const char *indent2, int progress, Stopwatch *stopwatch) {
+void PrintProgress(const char *name, const char *indent1, const char *indent2, int progress, Stopwatch *stopwatch) {
     if (progress == 0)
         stopwatch->reset();
     printf("\r%s%s [", indent1, name);
@@ -34,7 +35,7 @@ static void PrintProgress(const char *name, const char *indent1, const char *ind
         printf("\n%s%.2f seconds (%g ms) elapsed\n", indent2, stopwatch->elapsed() / 1000.0, stopwatch->elapsed());
 }
 
-static bool ProgressCallback(xatlas::ProgressCategory::Enum category, int progress, void *userData) {
+bool ProgressCallback(xatlas::ProgressCategory::Enum category, int progress, void *userData) {
     Stopwatch *stopwatch = (Stopwatch *)userData;
     PrintProgress(xatlas::StringForEnum(category), "   ", "      ", progress, stopwatch);
     return true;
@@ -50,17 +51,17 @@ std::vector<int> automatic_uv_map(const std::vector<UVTriMesh> &meshes, TextureA
     for (int i = 0; i < (int)meshes.size(); i++) {
         const UVTriMesh &mesh = meshes[i];
         xatlas::MeshDecl meshDecl;
-        meshDecl.vertexCount = mesh.num_vertices;
+        meshDecl.vertexCount = (uint32_t)mesh.num_vertices;
         meshDecl.vertexPositionData = mesh.vertices.get();
         meshDecl.vertexPositionStride = sizeof(float) * 3;
         if (mesh.uvs.get()) {
             meshDecl.vertexUvData = mesh.uvs.get();
             meshDecl.vertexUvStride = sizeof(float) * 2;
         }
-        meshDecl.indexCount = mesh.num_triangles * 3;
+        meshDecl.indexCount = (uint32_t)mesh.num_triangles * 3;
         meshDecl.indexData = mesh.indices.get();
         meshDecl.indexFormat = xatlas::IndexFormat::UInt32;
-        xatlas::AddMeshError::Enum error = xatlas::AddMesh(atlas.atlas, meshDecl, meshes.size());
+        xatlas::AddMeshError::Enum error = xatlas::AddMesh(atlas.atlas, meshDecl, (uint32_t)meshes.size());
         if (error != xatlas::AddMeshError::Success) {
             char buf[256];
             sprintf(buf, "\rError adding mesh %d: %s\n", i, xatlas::StringForEnum(error));
@@ -78,7 +79,7 @@ std::vector<int> automatic_uv_map(const std::vector<UVTriMesh> &meshes, TextureA
     std::vector<int> num_uv_vertices(meshes.size());
     for (uint32_t i = 0; i < atlas.atlas->meshCount; i++) {
         const xatlas::Mesh &mesh = atlas.atlas->meshes[i];
-        num_uv_vertices[i] = mesh.vertexCount;
+        num_uv_vertices[i] = (int)mesh.vertexCount;
     }
     return num_uv_vertices;
 }
@@ -94,7 +95,7 @@ void copy_texture_atlas(const TextureAtlas &atlas, std::vector<UVTriMesh> &meshe
         }
         int *ind_target = meshes[i].uv_indices.get();
         for (uint32_t f = 0; f < mesh.indexCount; f++) {
-            ind_target[f] = mesh.indexArray[f];
+            ind_target[f] = (int)mesh.indexArray[f];
         }
     }
 }

src/buffer.h

@@ -93,8 +93,8 @@ struct Buffer {
         }
     }
 
-    int size() const { return count; }
-    int bytes() const { return count * sizeof(T); }
+    int size() const { return (int)count; }
+    int bytes() const { return (int)count * sizeof(T); }
 
     T* begin() {
         return data;

src/camera.cpp

@@ -126,7 +126,7 @@ void test_d_sample_primary_rays() {
     auto finite_delta = Real(1e-6);
     for (int i = 0; i < 3; i++) {
         auto delta_pos = pos;
-        delta_pos[i] += finite_delta;
+        delta_pos[i] += float(finite_delta);
         auto delta_camera = Camera{
             1, 1,
             &delta_pos[0],
@@ -140,7 +140,7 @@ void test_d_sample_primary_rays() {
             CameraType::Perspective};
         auto positive_ray =
             sample_primary(delta_camera, Vector2{0.5, 0.5});
-        delta_pos[i] -= 2 * finite_delta;
+        delta_pos[i] -= float(2 * finite_delta);
         delta_camera = Camera{
             1, 1,
             &delta_pos[0],
@@ -161,7 +161,7 @@ void test_d_sample_primary_rays() {
     }
     for (int i = 0; i < 3; i++) {
         auto delta_look = look;
-        delta_look[i] += finite_delta;
+        delta_look[i] += float(finite_delta);
         auto delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -175,7 +175,7 @@ void test_d_sample_primary_rays() {
             CameraType::Perspective};
         auto positive_ray =
             sample_primary(delta_camera, Vector2{0.5, 0.5});
-        delta_look[i] -= 2 * finite_delta;
+        delta_look[i] -= float(2 * finite_delta);
         delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -196,7 +196,7 @@ void test_d_sample_primary_rays() {
     }
     for (int i = 0; i < 3; i++) {
         auto delta_up = up;
-        delta_up[i] += finite_delta;
+        delta_up[i] += float(finite_delta);
         auto delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -210,7 +210,7 @@ void test_d_sample_primary_rays() {
             CameraType::Perspective};
         auto positive_ray =
             sample_primary(delta_camera, Vector2{0.5, 0.5});
-        delta_up[i] -= 2 * finite_delta;
+        delta_up[i] -= float(2 * finite_delta);
         delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -283,7 +283,7 @@ void test_d_camera_to_screen() {
     auto finite_delta = Real(1e-6);
     for (int i = 0; i < 3; i++) {
         auto delta_pos = pos;
-        delta_pos[i] += finite_delta;
+        delta_pos[i] += float(finite_delta);
         auto delta_camera = Camera{
             1, 1,
             &delta_pos[0],
@@ -296,7 +296,7 @@ void test_d_camera_to_screen() {
             1e-2f,
             CameraType::Perspective};
         auto pxy = camera_to_screen(delta_camera, pt);
-        delta_pos[i] -= 2 * finite_delta;
+        delta_pos[i] -= float(2 * finite_delta);
         delta_camera = Camera{
             1, 1,
             &delta_pos[0],
@@ -314,7 +314,7 @@ void test_d_camera_to_screen() {
     }
     for (int i = 0; i < 3; i++) {
         auto delta_look = look;
-        delta_look[i] += finite_delta;
+        delta_look[i] += float(finite_delta);
         auto delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -327,7 +327,7 @@ void test_d_camera_to_screen() {
             1e-2f,
             CameraType::Perspective};
         auto pxy = camera_to_screen(delta_camera, pt);
-        delta_look[i] -= 2 * finite_delta;
+        delta_look[i] -= float(2 * finite_delta);
         delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -345,7 +345,7 @@ void test_d_camera_to_screen() {
     }
     for (int i = 0; i < 3; i++) {
         auto delta_up = up;
-        delta_up[i] += finite_delta;
+        delta_up[i] += float(finite_delta);
         auto delta_camera = Camera{
             1, 1,
             &pos[0],
@@ -358,7 +358,7 @@ void test_d_camera_to_screen() {
             1e-2f,
             CameraType::Perspective};
         auto pxy = camera_to_screen(delta_camera, pt);
-        delta_up[i] -= 2 * finite_delta;
+        delta_up[i] -= float(2 * finite_delta);
         delta_camera = Camera{
             1, 1,
             &pos[0],

src/edge.cpp

@@ -236,7 +236,7 @@ EdgeSampler::EdgeSampler(const std::vector<const Shape*> &shapes,
         // No need to collect edges
         return;
     }
-    auto shapes_buffer = scene.shapes.view(0, shapes.size());
+    auto shapes_buffer = scene.shapes.view(0, (int)shapes.size());
     // Conservatively allocate a big buffer for all edges
     auto num_total_triangles = 0;
     for (int shape_id = 0; shape_id < (int)shapes.size(); shape_id++) {
@@ -288,7 +288,7 @@ EdgeSampler::EdgeSampler(const std::vector<const Shape*> &shapes,
         }, num_edges, scene.use_gpu);
 
         DISPATCH(scene.use_gpu, thrust::copy, edges_buffer_begin, new_end, edges_begin);
-        current_num_edges += num_edges;
+        current_num_edges += (int)num_edges;
     }
     // Remove edges with 180 degree dihedral angles
     auto edges_end = DISPATCH(scene.use_gpu, thrust::remove_if, edges.begin(),
@@ -1421,7 +1421,7 @@ struct secondary_edge_sampler {
         }
         auto diffuse_pmf = diffuse_weight / weight_sum;
         auto specular_pmf = specular_weight / weight_sum;
-        auto m_pmf = 0.f;
+        auto m_pmf = Real(0);
         auto n = shading_point.shading_frame.n;
         if (material.two_sided) {
             if (dot(wi, n) < 0.f) {

src/envmap.h

@@ -8,8 +8,12 @@
 #include "transform.h"
 #include "buffer.h"
 
+#pragma warning(push)
+#pragma warning(disable : 4324 4668)
 #include <thrust/execution_policy.h>
 #include <thrust/binary_search.h>
+#pragma warning(pop)
+
 struct Scene;
 
 struct EnvironmentMap {