Cycles: Add an experimental CUDA kernel.

Now we build 2 .cubins per architecture (e.g. kernel_sm_21.cubin, kernel_experimental_sm_21.cubin).
The experimental kernel can be used by switching to the Experimental Feature Set: http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Cycles/Experimental_Features

This enables Subsurface Scattering and Correlated Multi Jitter Sampling on GPU, while keeping the stability and performance of the regular kernel.

Differential Revision: https://developer.blender.org/D762
Patch by Sergey and myself.

Developer / Builder Note:
CUDA Toolkit 6.5 is highly recommended for this, also note that building the experimental kernel requires a lot of system memory (~7-8GB).

1 files changed, 14 insertions, 4 deletions

diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index f2e470c21d5..1ed26717f4b 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -197,14 +197,18 @@ public:
 		return true;
 	}
 
-	string compile_kernel()
+	string compile_kernel(bool experimental)
 	{
 		/* compute cubin name */
 		int major, minor;
 		cuDeviceComputeCapability(&major, &minor, cuDevId);
 
 		/* attempt to use kernel provided with blender */
-		string cubin = path_get(string_printf("lib/kernel_sm_%d%d.cubin", major, minor));
+		string cubin;
+		if(experimental)
+			cubin = path_get(string_printf("lib/kernel_experimental_sm_%d%d.cubin", major, minor));
+		else
+			cubin = path_get(string_printf("lib/kernel_sm_%d%d.cubin", major, minor));
 		if(path_exists(cubin))
 			return cubin;
 
@@ -212,7 +216,10 @@ public:
 		string kernel_path = path_get("kernel");
 		string md5 = path_files_md5_hash(kernel_path);
 
-		cubin = string_printf("cycles_kernel_sm%d%d_%s.cubin", major, minor, md5.c_str());
+		if(experimental)
+			cubin = string_printf("cycles_kernel_experimental_sm%d%d_%s.cubin", major, minor, md5.c_str());
+		else
+			cubin = string_printf("cycles_kernel_sm%d%d_%s.cubin", major, minor, md5.c_str());
 		cubin = path_user_get(path_join("cache", cubin));
 
 		/* if exists already, use it */
@@ -263,6 +270,9 @@ public:
 		string command = string_printf("\"%s\" -arch=sm_%d%d -m%d --cubin \"%s\" "
 			"-o \"%s\" --ptxas-options=\"-v\" -I\"%s\" -DNVCC -D__KERNEL_CUDA_VERSION__=%d",
 			nvcc, major, minor, machine, kernel.c_str(), cubin.c_str(), include.c_str(), cuda_version);
+		
+		if(experimental)
+			command += " -D__KERNEL_CUDA_EXPERIMENTAL__";
 
 		printf("%s\n", command.c_str());
 
@@ -293,7 +303,7 @@ public:
 			return false;
 
 		/* get kernel */
-		string cubin = compile_kernel();
+		string cubin = compile_kernel(experimental);
 
 		if(cubin == "")
 			return false;