Cycles: add single program debug option for split kerneltemp_cycles_split_kernel

Single program generally compiles kernels faster (2-3 times), loads faster,
takes less drive space (2-3 times), and reduces the number of cached kernels.

9 files changed, 35 insertions, 12 deletions

diff --git a/intern/cycles/blender/addon/properties.py b/intern/cycles/blender/addon/properties.py
index ca109734314..6c5bcf09305 100644
--- a/intern/cycles/blender/addon/properties.py
+++ b/intern/cycles/blender/addon/properties.py
@@ -695,6 +695,8 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
             update=devices_update_callback
             )
 
+        cls.debug_opencl_kernel_single_program = BoolProperty(name="Single Program", default=False, update=devices_update_callback);
+
         cls.debug_use_opencl_debug = BoolProperty(name="Debug OpenCL", default=False)
 
     @classmethod
diff --git a/intern/cycles/blender/addon/ui.py b/intern/cycles/blender/addon/ui.py
index 7c1e3e270fb..3a689913230 100644
--- a/intern/cycles/blender/addon/ui.py
+++ b/intern/cycles/blender/addon/ui.py
@@ -1529,6 +1529,7 @@ class CyclesRender_PT_debug(CyclesButtonsPanel, Panel):
         col.label('OpenCL Flags:')
         col.prop(cscene, "debug_opencl_kernel_type", text="Kernel")
         col.prop(cscene, "debug_opencl_device_type", text="Device")
+        col.prop(cscene, "debug_opencl_kernel_single_program", text="Single Program")
         col.prop(cscene, "debug_use_opencl_debug", text="Debug")
 
 
diff --git a/intern/cycles/blender/blender_python.cpp b/intern/cycles/blender/blender_python.cpp
index 75118c43747..fc0df3410c6 100644
--- a/intern/cycles/blender/blender_python.cpp
+++ b/intern/cycles/blender/blender_python.cpp
@@ -106,6 +106,7 @@ bool debug_flags_sync_from_scene(BL::Scene b_scene)
 	}
 	/* Synchronize other OpenCL flags. */
 	flags.opencl.debug = get_boolean(cscene, "debug_use_opencl_debug");
+	flags.opencl.single_program = get_boolean(cscene, "debug_opencl_kernel_single_program");
 	return flags.opencl.device_type != opencl_device_type ||
 	       flags.opencl.kernel_type != opencl_kernel_type;
 }
diff --git a/intern/cycles/device/opencl/opencl.h b/intern/cycles/device/opencl/opencl.h
index 14e1db7a7aa..1c47dd18c2d 100644
--- a/intern/cycles/device/opencl/opencl.h
+++ b/intern/cycles/device/opencl/opencl.h
@@ -90,6 +90,7 @@ public:
 	                              cl_device_id device_id);
 	static void get_usable_devices(vector<OpenCLPlatformDevice> *usable_devices,
 	                               bool force_all = false);
+	static bool use_single_program();
 };
 
 /* Thread safe cache for contexts and programs.
diff --git a/intern/cycles/device/opencl/opencl_split.cpp b/intern/cycles/device/opencl/opencl_split.cpp
index be931599e39..4c6c9387864 100644
--- a/intern/cycles/device/opencl/opencl_split.cpp
+++ b/intern/cycles/device/opencl/opencl_split.cpp
@@ -77,16 +77,18 @@ public:
 	virtual bool load_kernels(const DeviceRequestedFeatures& requested_features,
 	                          vector<OpenCLDeviceBase::OpenCLProgram*> &programs)
 	{
+		bool single_program = OpenCLInfo::use_single_program();
 		program_data_init = OpenCLDeviceBase::OpenCLProgram(this,
-		                                  "split_data_init",
-		                                  "kernel_data_init.cl",
+		                                  single_program ? "split" : "split_data_init",
+		                                  single_program ? "kernel_split.cl" : "kernel_data_init.cl",
 		                                  get_build_options(this, requested_features));
+
 		program_data_init.add_kernel(ustring("path_trace_data_init"));
 		programs.push_back(&program_data_init);
 
 		program_state_buffer_size = OpenCLDeviceBase::OpenCLProgram(this,
-		                                  "split_state_buffer_size",
-		                                  "kernel_state_buffer_size.cl",
+		                                  single_program ? "split" : "split_state_buffer_size",
+		                                  single_program ? "kernel_split.cl" : "kernel_state_buffer_size.cl",
 		                                  get_build_options(this, requested_features));
 		program_state_buffer_size.add_kernel(ustring("path_trace_state_buffer_size"));
 		programs.push_back(&program_state_buffer_size);
@@ -209,10 +211,13 @@ public:
 	{
 		OpenCLSplitKernelFunction* kernel = new OpenCLSplitKernelFunction(device);
 
-		kernel->program = OpenCLDeviceBase::OpenCLProgram(device,
-		                                "split_" + kernel_name,
-		                                "kernel_" + kernel_name + ".cl",
-		                                get_build_options(device, requested_features));
+		bool single_program = OpenCLInfo::use_single_program();
+		kernel->program =
+			OpenCLDeviceBase::OpenCLProgram(device,
+			                                single_program ? "split" : "split_" + kernel_name,
+			                                single_program ? "kernel_split.cl" : "kernel_" + kernel_name + ".cl",
+			                                get_build_options(device, requested_features));
+
 		kernel->program.add_kernel(ustring("path_trace_" + kernel_name));
 		kernel->program.load();
 
diff --git a/intern/cycles/device/opencl/opencl_util.cpp b/intern/cycles/device/opencl/opencl_util.cpp
index d5c19bf5386..e2d0ff71786 100644
--- a/intern/cycles/device/opencl/opencl_util.cpp
+++ b/intern/cycles/device/opencl/opencl_util.cpp
@@ -552,6 +552,11 @@ bool OpenCLInfo::use_debug()
 	return DebugFlags().opencl.debug;
 }
 
+bool OpenCLInfo::use_single_program()
+{
+	return DebugFlags().opencl.single_program;
+}
+
 bool OpenCLInfo::kernel_use_advanced_shading(const string& platform)
 {
 	/* keep this in sync with kernel_types.h! */
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index e285c382946..18c352d835c 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -16,6 +16,7 @@ set(SRC
 	kernels/cpu/kernel_split.cpp
 	kernels/opencl/kernel.cl
 	kernels/opencl/kernel_state_buffer_size.cl
+	kernels/opencl/kernel_split.cl
 	kernels/opencl/kernel_data_init.cl
 	kernels/opencl/kernel_path_init.cl
 	kernels/opencl/kernel_queue_enqueue.cl
@@ -412,6 +413,7 @@ endif()
 
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_state_buffer_size.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
+delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_split.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_data_init.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_path_init.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_queue_enqueue.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
diff --git a/intern/cycles/util/util_debug.cpp b/intern/cycles/util/util_debug.cpp
index f12c5e28c80..da3ccae45bf 100644
--- a/intern/cycles/util/util_debug.cpp
+++ b/intern/cycles/util/util_debug.cpp
@@ -77,7 +77,8 @@ void DebugFlags::CUDA::reset()
 DebugFlags::OpenCL::OpenCL()
   : device_type(DebugFlags::OpenCL::DEVICE_ALL),
     kernel_type(DebugFlags::OpenCL::KERNEL_DEFAULT),
-    debug(false)
+    debug(false),
+    single_program(false)
 {
 	reset();
 }
@@ -117,6 +118,7 @@ void DebugFlags::OpenCL::reset()
 	}
 	/* Initialize other flags from environment variables. */
 	debug = (getenv("CYCLES_OPENCL_DEBUG") != NULL);
+	single_program = (getenv("CYCLES_OPENCL_SINGLE_PROGRAM") != NULL);
 }
 
 DebugFlags::DebugFlags()
@@ -179,9 +181,10 @@ std::ostream& operator <<(std::ostream &os,
 			break;
 	}
 	os << "OpenCL flags:\n"
-	   << "  Device type : " << opencl_device_type << "\n"
-	   << "  Kernel type : " << opencl_kernel_type << "\n"
-	   << "  Debug       : " << string_from_bool(debug_flags.opencl.debug)
+	   << "  Device type    : " << opencl_device_type << "\n"
+	   << "  Kernel type    : " << opencl_kernel_type << "\n"
+	   << "  Debug          : " << string_from_bool(debug_flags.opencl.debug) << "\n"
+	   << "  Signle program : " << string_from_bool(debug_flags.opencl.single_program)
 	   << "\n";
 	return os;
 }
diff --git a/intern/cycles/util/util_debug.h b/intern/cycles/util/util_debug.h
index 911c95de4ab..5aa0d911884 100644
--- a/intern/cycles/util/util_debug.h
+++ b/intern/cycles/util/util_debug.h
@@ -112,6 +112,9 @@ public:
 
 		/* Use debug version of the kernel. */
 		bool debug;
+
+		/* Use single program */
+		bool single_program;
 	};
 
 	/* Get instance of debug flags registry. */