Cycles / CUDA: Increase sm_2x registers to 40.

This fixes the ptaxs "ACCESS_VIOLATION" error and should allow our Linux and Windows build bots to compile again.
Unfortunately this comes with a performance penalty on sm_2x cards, so this is only a workaround for now. Branched Path is still globally disabled on GPU.

intern/cycles/device/device_cuda.cpp

@@ -320,7 +320,7 @@ public:
 		/* CUDA 5.x build flags for different archs */
 		if(major == 2) {
 			/* sm_2x */
-			arch_flags = "--maxrregcount=32 --use_fast_math";
+			arch_flags = "--maxrregcount=40 --use_fast_math";
 		}
 		else if(major == 3) {
 			/* sm_3x */

intern/cycles/kernel/CMakeLists.txt

@@ -165,7 +165,7 @@ if(WITH_CYCLES_CUDA_BINARIES)
 		# CUDA 5.x build flags for different archs
 		if(${arch} MATCHES "sm_2[0-9]")
 			# sm_2x
-			set(cuda_arch_flags "--maxrregcount=32")
+			set(cuda_arch_flags "--maxrregcount=40")
 		elseif(${arch} MATCHES "sm_3[0-9]")
 			# sm_3x
 			set(cuda_arch_flags "--maxrregcount=32")

intern/cycles/kernel/SConscript

@@ -90,7 +90,7 @@ if env['WITH_BF_CYCLES_CUDA_BINARIES']:
         # CUDA 5.x build flags for different archs
         if arch.startswith("sm_2"):
             # sm_2x
-            cuda_arch_flags = "--maxrregcount=32 --use_fast_math"
+            cuda_arch_flags = "--maxrregcount=40 --use_fast_math"
         elif arch.startswith("sm_3"):
             # sm_3x
             cuda_arch_flags = "--maxrregcount=32 --use_fast_math"