diff options
| author | Sergey Sharybin <sergey.vfx@gmail.com> | 2016-08-02 16:04:34 +0300 |
|---|---|---|
| committer | Sergey Sharybin <sergey.vfx@gmail.com> | 2016-08-02 16:13:34 +0300 |
| commit | 500e0e9a3d5a9e7982d4df31a1477b8732c76e71 (patch) | |
| tree | 92a4f5f1559143efdfeb2b087549c587477bba25 /intern/cycles/kernel/svm/svm_attribute.h | |
| parent | b416168d85a7cf5a27370f5c933fd1ad41ee333d (diff) | |
Cycles: Some more inline policy tweaks for CUDA 8
Makes it so toolkit does exactly the same decision about what to inline,
but unfortunately it has really barely visible difference on GTX-980.
Diffstat (limited to 'intern/cycles/kernel/svm/svm_attribute.h')
| -rw-r--r-- | intern/cycles/kernel/svm/svm_attribute.h | 20 |
1 files changed, 15 insertions, 5 deletions
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h index ff92920c610..bd6013e9205 100644 --- a/intern/cycles/kernel/svm/svm_attribute.h +++ b/intern/cycles/kernel/svm/svm_attribute.h @@ -87,7 +87,12 @@ ccl_device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, u } } -ccl_device void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node) +#ifndef __KERNEL_CUDS__ +ccl_device +#else +ccl_device_noinline +#endif +void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node) { NodeAttributeType type, mesh_type; AttributeElement elem; @@ -123,10 +128,15 @@ ccl_device void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float * } } -ccl_device void svm_node_attr_bump_dy(KernelGlobals *kg, - ShaderData *sd, - float *stack, - uint4 node) +#ifndef __KERNEL_CUDS__ +ccl_device +#else +ccl_device_noinline +#endif +void svm_node_attr_bump_dy(KernelGlobals *kg, + ShaderData *sd, + float *stack, + uint4 node) { NodeAttributeType type, mesh_type; AttributeElement elem; |