Cycles

* Compile all of cycles with -ffast-math again
* Add scons compilation of cuda binaries, tested on mac/linux.
* Add UI option for supported/experimental features, to make it
  more clear what is supported, opencl/subdivision is experimental.
* Remove cycles xml exporter, was just for testing.

SConstruct

@@ -581,6 +581,14 @@ if env['OURPLATFORM']!='darwin':
             source.remove('CMakeLists.txt')
             source=['intern/cycles/doc/license/'+s for s in source]
             scriptinstall.append(env.Install(dir=dir,source=source))
+
+            # cuda binaries
+            if env['WITH_BF_CYCLES_CUDA_BINARIES']:
+                dir=os.path.join(env['BF_INSTALLDIR'], VERSION, 'scripts', 'addons','cycles', 'lib')
+                for arch in env['BF_CYCLES_CUDA_BINARIES_ARCH']:
+                    kernel_build_dir = os.path.join(B.root_build_dir, 'intern/cycles/kernel')
+                    cubin_file = os.path.join(kernel_build_dir, "kernel_%s.cubin" % arch)
+                    scriptinstall.append(env.Install(dir=dir,source=cubin_file))
     
     if env['WITH_BF_INTERNATIONAL']:
         internationalpaths=['release' + os.sep + 'datafiles']

build_files/scons/config/darwin-config.py

@@ -283,7 +283,7 @@ BF_PCRE_LIBPATH = '${BF_PCRE}/lib'
 #BF_EXPAT_LIB = 'expat'
 #BF_EXPAT_LIBPATH = '/usr/lib'
 
-#Cycles
+# Cycles
 WITH_BF_CYCLES = True
 
 WITH_BF_OIIO = True
@@ -298,6 +298,10 @@ BF_BOOST_INC = BF_BOOST + '/include'
 BF_BOOST_LIB = 'boost_date_time-mt boost_filesystem-mt boost_regex-mt boost_system-mt boost_thread-mt'
 BF_BOOST_LIBPATH = BF_BOOST + '/lib'
 
+WITH_BF_CYCLES_CUDA_BINARIES = False
+BF_CYCLES_CUDA_NVCC = '/usr/local/cuda/bin/nvcc'
+BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_13', 'sm_20', 'sm_21']
+
 #Ray trace optimization
 if MACOSX_ARCHITECTURE == 'x86_64' or MACOSX_ARCHITECTURE == 'i386':
     WITH_BF_RAYOPTIMIZATION = True

build_files/scons/config/linux-config.py

@@ -219,6 +219,10 @@ BF_BOOST_LIBPATH = BF_BOOST + '/lib'
 
 WITH_BF_CYCLES = WITH_BF_OIIO and WITH_BF_BOOST
 
+WITH_BF_CYCLES_CUDA_BINARIES = False
+BF_CYCLES_CUDA_NVCC = '/usr/local/cuda/bin/nvcc'
+BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_13', 'sm_20', 'sm_21']
+
 WITH_BF_OPENMP = True
 
 #Ray trace optimization

build_files/scons/tools/Blender.py

@@ -291,7 +291,7 @@ def setup_syslibs(lenv):
 
     syslibs += Split(lenv['BF_JPEG_LIB'])
     syslibs += Split(lenv['BF_PNG_LIB'])
-	
+
     syslibs += lenv['LLIBS']
 
     return syslibs
@@ -588,6 +588,8 @@ def AppIt(target=None, source=None, env=None):
             commands.getoutput(cmd)
             cmd = 'mkdir %s/kernel' % (cinstalldir)
             commands.getoutput(cmd)
+            cmd = 'mkdir %s/lib' % (cinstalldir)
+            commands.getoutput(cmd)
             cmd = 'cp -R %s/blender/addon/*.py %s/' % (croot, cinstalldir)
             commands.getoutput(cmd)
             cmd = 'cp -R %s/doc/license %s/license' % (croot, cinstalldir)
@@ -596,6 +598,8 @@ def AppIt(target=None, source=None, env=None):
             commands.getoutput(cmd)
             cmd = 'cp -R %s/kernel/svm %s/util/util_color.h %s/util/util_math.h %s/util/util_transform.h %s/util/util_types.h %s/kernel/' % (croot, croot, croot, croot, croot, cinstalldir)
             commands.getoutput(cmd)
+            cmd = 'cp -R %s/../intern/cycles/kernel/*.cubin %s/lib/' % (builddir, cinstalldir)
+            commands.getoutput(cmd)
 
     if env['WITH_OSX_STATICPYTHON']:
         cmd = 'mkdir %s/%s.app/Contents/MacOS/%s/python/'%(installdir,binary, VERSION)

build_files/scons/tools/btools.py

@@ -157,7 +157,7 @@ def validate_arguments(args, bc):
             'WITH_BF_JEMALLOC', 'WITH_BF_STATICJEMALLOC', 'BF_JEMALLOC', 'BF_JEMALLOC_INC', 'BF_JEMALLOC_LIBPATH', 'BF_JEMALLOC_LIB', 'BF_JEMALLOC_LIB_STATIC',
             'BUILDBOT_BRANCH',
             'WITH_BF_3DMOUSE', 'WITH_BF_STATIC3DMOUSE', 'BF_3DMOUSE', 'BF_3DMOUSE_INC', 'BF_3DMOUSE_LIB', 'BF_3DMOUSE_LIBPATH', 'BF_3DMOUSE_LIB_STATIC',
-            'WITH_BF_CYCLES', 'WITH_BF_CYCLES_BINARIES' 'BF_CYCLES_BINARIES_ARCH',
+            'WITH_BF_CYCLES', 'WITH_BF_CYCLES_CUDA_BINARIES' 'BF_CYCLES_CUDA_NVCC', 'BF_CYCLES_CUDA_NVCC',
             'WITH_BF_OIIO', 'WITH_BF_STATICOIIO', 'BF_OIIO', 'BF_OIIO_INC', 'BF_OIIO_LIB', 'BF_OIIO_LIB_STATIC', 'BF_OIIO_LIBPATH',
             'WITH_BF_BOOST', 'WITH_BF_STATICBOOST', 'BF_BOOST', 'BF_BOOST_INC', 'BF_BOOST_LIB', 'BF_BOOST_LIB_STATIC', 'BF_BOOST_LIBPATH'
             ]
@@ -544,7 +544,9 @@ def read_opts(env, cfg, args):
 
     localopts.AddVariables(
         (BoolVariable('WITH_BF_CYCLES', 'Build with the Cycles engine', True)),
-        (BoolVariable('WITH_BF_CYCLES_BINARIES', 'Build with precompiled CUDA binaries', False)),
+        (BoolVariable('WITH_BF_CYCLES_CUDA_BINARIES', 'Build with precompiled CUDA binaries', False)),
+        ('BF_CYCLES_CUDA_NVCC', 'CUDA nvcc compiler path', ''),
+        ('BF_CYCLES_CUDA_BINARIES_ARCH', 'CUDA architectures to compile binaries for', []),
 
         (BoolVariable('WITH_BF_OIIO', 'Build with OpenImageIO', False)),
         (BoolVariable('WITH_BF_STATICOIIO', 'Staticly link to OpenImageIO', False)),

intern/cycles/CMakeLists.txt

@@ -10,12 +10,14 @@ include(cmake/external_libs.cmake)
 
 if(WITH_RAYOPTIMIZATION AND SUPPORT_SSE_BUILD)
 	set(WITH_CYCLES_OPTIMIZED_KERNEL ON)
+endif()
 
-	if(WIN32 AND MSVC)
+if(WIN32 AND MSVC)
-		set(CYCLES_OPTIMIZED_KERNEL_FLAGS "/arch:SSE2 -D_CRT_SECURE_NO_WARNINGS /fp:fast")
+	set(CYCLES_OPTIMIZED_KERNEL_FLAGS "/arch:SSE2 -D_CRT_SECURE_NO_WARNINGS /fp:fast /EHsc")
-	elseif(CMAKE_COMPILER_IS_GNUCC)
+	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /fp:fast -D_CRT_SECURE_NO_WARNINGS /EHsc")
-		set(CYCLES_OPTIMIZED_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3")
+elseif(CMAKE_COMPILER_IS_GNUCC)
-	endif()
+	set(CYCLES_OPTIMIZED_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3")
+	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math")
 endif()
 
 # for OSL, not needed yet

intern/cycles/SConscript

@@ -14,6 +14,7 @@ sources.remove(path.join('kernel', 'kernel_optimized.cpp'))
 
 incs = [] 
 defs = []
+cxxflags = []
 
 defs.append('CCL_NAMESPACE_BEGIN=namespace ccl {')
 defs.append('CCL_NAMESPACE_END=}')
@@ -22,7 +23,7 @@ defs.append('WITH_OPENCL')
 defs.append('WITH_MULTI')
 defs.append('WITH_CUDA')
 
-if env['WITH_BF_CYCLES_BINARIES']:
+if env['WITH_BF_CYCLES_CUDA_BINARIES']:
     defs.append('WITH_CUDA_BINARIES')
 
 incs.extend('. bvh render device kernel kernel/osl kernel/svm util subd'.split())
@@ -33,12 +34,17 @@ incs.append(cycles['BF_OIIO_INC'])
 incs.append(cycles['BF_BOOST_INC'])
 incs.append(cycles['BF_PYTHON_INC'])
 
+if env['OURPLATFORM'] in ('win32-vc', 'win64-vc'):
+    cxxflags.append('-D_CRT_SECURE_NO_WARNINGS /fp:fast /EHsc'.split())
+else:
+    cxxflags.append('-ffast-math'.split())
+
 # optimized kernel
 if env['WITH_BF_RAYOPTIMIZATION']:
     optim_cxxflags = []
 
     if env['OURPLATFORM'] in ('win32-vc', 'win64-vc'):
-        optim_cxxflags.append('/Ox /Ot /arch:SSE2 -D_CRT_SECURE_NO_WARNINGS /EHsc /fp:fast'.split())
+        optim_cxxflags.append('/arch:SSE2 -D_CRT_SECURE_NO_WARNINGS /fp:fast /EHsc'.split())
     else:
         optim_cxxflags.append('-ffast-math -msse -msse2 -msse3'.split())
     
@@ -48,5 +54,10 @@ if env['WITH_BF_RAYOPTIMIZATION']:
     cycles_optim = cycles.Clone()
     cycles_optim.BlenderLib('bf_intern_cycles_optimized', optim_sources, incs, optim_defs, libtype=['intern'], priority=[0], compileflags=[None], cxx_compileflags=optim_cxxflags)
 
-cycles.BlenderLib('bf_intern_cycles', sources, incs, defs, libtype=['intern'], priority=[0], compileflags=[None])
+cycles.BlenderLib('bf_intern_cycles', sources, incs, defs, libtype=['intern'], priority=[0], compileflags=[None], cxx_compileflags=cxxflags)
+
+# cuda kernel binaries
+if env['WITH_BF_CYCLES_CUDA_BINARIES']:
+    kernel_binaries = SConscript(['kernel/SConscript'])
+    cycles.Depends("device/device_cuda.o", kernel_binaries)
 

intern/cycles/blender/CMakeLists.txt

@@ -39,7 +39,6 @@ set(ADDON_FILES
 	addon/presets.py
 	addon/properties.py
 	addon/ui.py
-	addon/xml.py
 )
 
 blender_add_lib(bf_intern_cycles "${SRC}" "${INC}" "${INC_SYS}")

intern/cycles/blender/addon/__init__.py

@@ -33,7 +33,7 @@ bl_info = {
     "category": "Render"}
 
 import bpy
-from . import ui, properties, xml, engine, presets
+from . import ui, properties, engine, presets
 
 
 class CyclesRender(bpy.types.RenderEngine):
@@ -77,13 +77,11 @@ class CyclesRender(bpy.types.RenderEngine):
 def register():
     properties.register()
     ui.register()
-    xml.register()
     presets.register()
     bpy.utils.register_module(__name__)
 
 
 def unregister():
-    xml.unregister()
     ui.unregister()
     properties.unregister()
     presets.unregister()

intern/cycles/blender/addon/enums.py

@@ -41,7 +41,12 @@ devices = (
 
 gpu_type = (
     ("CUDA", "CUDA", "NVidia only"),
-    ("OPENCL", "OpenCL (incomplete)", ""),
+    ("OPENCL", "OpenCL", ""),
+    )
+
+feature_set = (
+    ("SUPPORTED", "Supported", "Only use finished and supported features"),
+    ("EXPERIMENTAL", "Experimental", "Use experimental and incomplete features that might be broken or change in the future"),
     )
 
 shading_systems = (

intern/cycles/blender/addon/properties.py

@@ -41,6 +41,9 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         cls.gpu_type = EnumProperty(name="GPU Type", description="Processing system to use on the GPU",
             items=enums.gpu_type, default="CUDA")
 
+        cls.feature_set = EnumProperty(name="Feature Set", description="Feature set to use for rendering",
+            items=enums.feature_set, default="SUPPORTED")
+
         cls.shading_system = EnumProperty(name="Shading System", description="Shading system to use for rendering",
             items=enums.shading_systems, default="GPU_COMPATIBLE")
 

intern/cycles/blender/addon/ui.py

@@ -294,7 +294,12 @@ class Cycles_PT_mesh_displacement(CyclesButtonsPanel, Panel):
 
     @classmethod
     def poll(cls, context):
-        return CyclesButtonsPanel.poll(context) and (context.mesh or context.curve or context.meta_ball)
+        if CyclesButtonsPanel.poll(context):
+            if context.mesh or context.curve or context.meta_ball:
+                if context.scene.cycles.feature_set == 'EXPERIMENTAL':
+                    return True
+
+        return False
 
     def draw(self, context):
         layout = self.layout
@@ -706,18 +711,20 @@ def draw_device(self, context):
     if scene.render.engine == "CYCLES":
         cscene = scene.cycles
 
+        layout.prop(cscene, "feature_set")
+        experimental = cscene.feature_set == 'EXPERIMENTAL'
+
         available_devices = engine.available_devices()
         available_cuda = 'cuda' in available_devices
-        available_opencl = 'opencl' in available_devices
+        available_opencl = experimental and 'opencl' in available_devices
 
         if available_cuda or available_opencl:
             layout.prop(cscene, "device")
             if cscene.device == 'GPU' and available_cuda and available_opencl:
                 layout.prop(cscene, "gpu_type")
-        if cscene.device == 'CPU' and engine.with_osl():
+        if experimental and cscene.device == 'CPU' and engine.with_osl():
             layout.prop(cscene, "shading_system")
 
-
 def draw_pause(self, context):
     layout = self.layout
     scene = context.scene

intern/cycles/blender/addon/xml.py

@@ -1,105 +0,0 @@
-#
-# Copyright 2011, Blender Foundation.
-#
-# This program is free software; you can redistribute it and/or
-# modify it under the terms of the GNU General Public License
-# as published by the Free Software Foundation; either version 2
-# of the License, or (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program; if not, write to the Free Software Foundation,
-# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
-#
-
-# <pep8 compliant>
-
-# XML exporter for generating test files, not intended for end users
-
-import os
-import bpy
-from bpy_extras.io_utils import ExportHelper
-import xml.etree.ElementTree as etree
-import xml.dom.minidom as dom
-
-
-def strip(root):
-    root.text = None
-    root.tail = None
-
-    for elem in root:
-        strip(elem)
-
-
-def write(node, fname):
-    strip(node)
-
-    s = etree.tostring(node)
-    s = dom.parseString(s).toprettyxml()
-
-    f = open(fname, "w")
-    f.write(s)
-
-
-class ExportCyclesXML(bpy.types.Operator, ExportHelper):
-    ''''''
-    bl_idname = "export_mesh.cycles_xml"
-    bl_label = "Export Cycles XML"
-
-    filename_ext = ".xml"
-
-    @classmethod
-    def poll(cls, context):
-        return context.active_object != None
-
-    def execute(self, context):
-        filepath = bpy.path.ensure_ext(self.filepath, ".xml")
-
-        # get mesh
-        scene = context.scene
-        obj = context.object
-
-        if not obj:
-            raise Exception("No active object")
-
-        mesh = obj.to_mesh(scene, True, 'PREVIEW')
-
-        if not mesh:
-            raise Exception("No mesh data in active object")
-
-        # generate mesh node
-        nverts = ""
-        verts = ""
-        P = ""
-
-        for v in mesh.vertices:
-            P += "%f %f %f  " % (v.co[0], v.co[1], v.co[2])
-
-        for i, f in enumerate(mesh.faces):
-            nverts += str(len(f.vertices)) + " "
-
-            for v in f.vertices:
-                verts += str(v) + " "
-            verts += " "
-
-        node = etree.Element('mesh', attrib={'nverts': nverts, 'verts': verts, 'P': P})
-
-        # write to file
-        write(node, filepath)
-
-        return {'FINISHED'}
-
-
-def register():
-    pass
-
-
-def unregister():
-    pass
-
-if __name__ == "__main__":
-    register()

intern/cycles/blender/blender_mesh.cpp

@@ -281,7 +281,7 @@ Mesh *BlenderSync::sync_mesh(BL::Object b_ob, bool object_updated)
 	mesh->name = ustring(b_ob_data.name().c_str());
 
 	if(b_mesh) {
-		if(cmesh.data && RNA_boolean_get(&cmesh, "use_subdivision"))
+		if(cmesh.data && experimental && RNA_boolean_get(&cmesh, "use_subdivision"))
 			create_subd_mesh(mesh, b_mesh, &cmesh, used_shaders);
 		else
 			create_mesh(scene, mesh, b_mesh, used_shaders);
@@ -294,7 +294,7 @@ Mesh *BlenderSync::sync_mesh(BL::Object b_ob, bool object_updated)
 	if(cmesh.data) {
 		int method = RNA_enum_get(&cmesh, "displacement_method");
 
-		if(method == 0)
+		if(method == 0 || !experimental)
 			mesh->displacement_method = Mesh::DISPLACE_BUMP;
 		else if(method == 1)
 			mesh->displacement_method = Mesh::DISPLACE_TRUE;

intern/cycles/blender/blender_sync.cpp

@@ -47,7 +47,8 @@ BlenderSync::BlenderSync(BL::BlendData b_data_, BL::Scene b_scene_, Scene *scene
   mesh_map(&scene_->meshes),
   light_map(&scene_->lights),
   world_map(NULL),
-  world_recalc(false)
+  world_recalc(false),
+  experimental(false)
 {
 	scene = scene_;
 	preview = preview_;
@@ -134,6 +135,8 @@ void BlenderSync::sync_integrator()
 {
 	PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
 
+	experimental = (RNA_enum_get(&cscene, "feature_set") != 0);
+
 	Integrator *integrator = scene->integrator;
 	Integrator previntegrator = *integrator;
 
@@ -253,6 +256,9 @@ SessionParams BlenderSync::get_session_params(BL::Scene b_scene, bool background
 	SessionParams params;
 	PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
 
+	/* feature set */
+	bool experimental = (RNA_enum_get(&cscene, "feature_set") != 0);
+
 	/* device type */
 	params.device_type = DEVICE_CPU;
 
@@ -262,7 +268,7 @@ SessionParams BlenderSync::get_session_params(BL::Scene b_scene, bool background
 
 		if(device_type_available(types, dtype))
 			params.device_type = dtype;
-		else if(device_type_available(types, DEVICE_OPENCL))
+		else if(experimental && device_type_available(types, DEVICE_OPENCL))
 			params.device_type = DEVICE_OPENCL;
 		else if(device_type_available(types, DEVICE_CUDA))
 			params.device_type = DEVICE_CUDA;

intern/cycles/blender/blender_sync.h

@@ -100,6 +100,7 @@ private:
 
 	Scene *scene;
 	bool preview;
+	bool experimental;
 
 	struct RenderLayerInfo {
 		RenderLayerInfo()

intern/cycles/device/device_cuda.cpp

@@ -237,7 +237,10 @@ public:
 			return cubin;
 
 #ifdef WITH_CUDA_BINARIES
-		cuda_error("CUDA binary kernel for this graphics card not found.");
+		if(major <= 1 && minor <= 2)
+			cuda_error(string_printf("CUDA device supported only with shader model 1.3 or up, found %d.%d.", major, minor));
+		else
+			cuda_error("CUDA binary kernel for this graphics card not found.");
 		return "";
 #else
 		/* if not, find CUDA compiler */

intern/cycles/kernel/SConscript

@@ -0,0 +1,50 @@
+#!/usr/bin/python
+import sys
+import os
+
+def normpath(path):
+    return os.path.abspath(os.path.normpath(path))
+
+Import ('env')
+
+kernel_binaries = []
+
+if env['WITH_BF_CYCLES_CUDA_BINARIES']:
+    kernel = env.Clone()
+
+    # cuda info
+    nvcc = env['BF_CYCLES_CUDA_NVCC']
+    cuda_archs = env['BF_CYCLES_CUDA_BINARIES_ARCH']
+
+    # build directory
+    root_build_dir = normpath(env['BF_BUILDDIR'])
+    build_dir = os.path.join(root_build_dir, 'intern/cycles/kernel')
+
+    # source directories and files
+    source_dir = Dir('.').srcnode().path
+    kernel_file = os.path.join(source_dir, "kernel.cu")
+    util_dir = os.path.join(source_dir, "../util")
+    svm_dir = os.path.join(source_dir, "../svm")
+
+    # nvcc flags
+    nvcc_flags = "--cubin -use_fast_math --ptxas-options=\"-v\" --maxrregcount=24"
+    nvcc_flags += " --opencc-options -OPT:Olimit=0"
+    nvcc_flags += " -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END= -DNVCC"
+    nvcc_flags += " -I \"%s\" -I \"%s\"" % (util_dir, svm_dir)
+
+    # dependencies
+    dependencies = ['kernel.cu'] + kernel.Glob('*.h') + kernel.Glob('../util/*.h') + kernel.Glob('svm/*.h')
+
+    # add command for each cuda architecture
+    for arch in cuda_archs:
+        cubin_file = os.path.join(build_dir, "kernel_%s.cubin" % arch)
+
+        command = "\"%s\" -arch=%s %s \"%s\" -o \"%s\"" % (nvcc, arch, nvcc_flags, kernel_file, cubin_file)
+
+        kernel.Command(cubin_file, 'kernel.cu', command)
+        kernel.Depends(cubin_file, dependencies)
+
+        kernel_binaries.append(cubin_file)
+
+Return('kernel_binaries')
+