[Bf-blender-cvs] [297a222] master: Cycles / CUDA: Increase sm_2x registers to 40.
Thomas Dinges
noreply at git.blender.org
Tue Apr 8 23:26:26 CEST 2014
Commit: 297a2223b5535cab21c3758ea10ccb4e0f5668e2
Author: Thomas Dinges
Date: Tue Apr 8 23:25:54 2014 +0200
https://developer.blender.org/rB297a2223b5535cab21c3758ea10ccb4e0f5668e2
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.
===================================================================
M intern/cycles/device/device_cuda.cpp
M intern/cycles/kernel/CMakeLists.txt
M intern/cycles/kernel/SConscript
===================================================================
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index a5e4ec6..edee32e 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/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 */
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 45b4c81..7dab65f 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/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")
diff --git a/intern/cycles/kernel/SConscript b/intern/cycles/kernel/SConscript
index 82aaccb..61ddaff 100644
--- a/intern/cycles/kernel/SConscript
+++ b/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"
More information about the Bf-blender-cvs
mailing list