[Bf-blender-cvs] [a613290] master: Cycles / CUDA: Workaround to make sm_52 (Maxwell) cards work.

Thomas Dinges noreply at git.blender.org
Sun Oct 5 04:14:13 CEST 2014


Commit: a613290775c8571cb1d352256c65e95a50321000
Author: Thomas Dinges
Date:   Sun Oct 5 04:13:40 2014 +0200
Branches: master
https://developer.blender.org/rBa613290775c8571cb1d352256c65e95a50321000

Cycles / CUDA: Workaround to make sm_52 (Maxwell) cards work.

* sm_52 can run a sm_50 kernel, so tell runtime detection to use that until we build a dedicated sm_52 kernel.

===================================================================

M	intern/cycles/device/device_cuda.cpp

===================================================================

diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 5de2efa..bd265f0 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -202,6 +202,10 @@ public:
 		/* compute cubin name */
 		int major, minor;
 		cuDeviceComputeCapability(&major, &minor, cuDevId);
+		
+		/* workaround to make sm_52 cards work, until we bundle kernel */
+		if(major == 5 && minor == 2)
+			minor = 0;
 
 		/* attempt to use kernel provided with blender */
 		string cubin;




More information about the Bf-blender-cvs mailing list