griefith/test2
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: work around OpenCL performance regression after AOVs and vector rotate

Browse Source 
We appear to be hitting some limit where adding any amount of code causes a
significant performance regression, no matter what it does. To work around
that a new node level was added.

Ref T71479
This commit is contained in:
Brecht Van Lommel
2020-03-24 01:27:38 +01:00
parent f8a4fb43fb
commit 2bec6f1f06
3 changed files with 27 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
intern/cycles/kernel/svm/svm.h
View File

@@ -459,17 +459,6 @@ ccl_device_noinline void svm_eval_nodes(KernelGlobals *kg,
case NODE_IES:
svm_node_ies(kg, sd, stack, node, &offset);
break;
case NODE_AOV_START:
if (!svm_node_aov_check(state, buffer)) {
return;
}
break;
case NODE_AOV_COLOR:
svm_node_aov_color(kg, sd, stack, node, buffer);
break;
case NODE_AOV_VALUE:
svm_node_aov_value(kg, sd, stack, node, buffer);
break;
#endif /* NODES_GROUP(NODE_GROUP_LEVEL_2) */
#if NODES_GROUP(NODE_GROUP_LEVEL_3)
@@ -522,11 +511,6 @@ ccl_device_noinline void svm_eval_nodes(KernelGlobals *kg,
case NODE_CLAMP:
svm_node_clamp(kg, sd, stack, node.y, node.z, node.w, &offset);
break;
# if NODES_FEATURE(NODE_FEATURE_VOLUME)
case NODE_TEX_VOXEL:
svm_node_tex_voxel(kg, sd, stack, node, &offset);
break;
# endif /* NODES_FEATURE(NODE_FEATURE_VOLUME) */
# ifdef __SHADER_RAYTRACE__
case NODE_BEVEL:
svm_node_bevel(kg, sd, state, stack, node);
@@ -536,6 +520,25 @@ ccl_device_noinline void svm_eval_nodes(KernelGlobals *kg,
break;
# endif /* __SHADER_RAYTRACE__ */
#endif /* NODES_GROUP(NODE_GROUP_LEVEL_3) */
#if NODES_GROUP(NODE_GROUP_LEVEL_4)
# if NODES_FEATURE(NODE_FEATURE_VOLUME)
case NODE_TEX_VOXEL:
svm_node_tex_voxel(kg, sd, stack, node, &offset);
break;
# endif /* NODES_FEATURE(NODE_FEATURE_VOLUME) */
case NODE_AOV_START:
if (!svm_node_aov_check(state, buffer)) {
return;
}
break;
case NODE_AOV_COLOR:
svm_node_aov_color(kg, sd, stack, node, buffer);
break;
case NODE_AOV_VALUE:
svm_node_aov_value(kg, sd, stack, node, buffer);
break;
#endif /* NODES_GROUP(NODE_GROUP_LEVEL_4) */
case NODE_END:
return;
default:

3
intern/cycles/kernel/svm/svm_types.h
View File

@@ -42,7 +42,8 @@ CCL_NAMESPACE_BEGIN
#define NODE_GROUP_LEVEL_1 1
#define NODE_GROUP_LEVEL_2 2
#define NODE_GROUP_LEVEL_3 3
#define NODE_GROUP_LEVEL_MAX NODE_GROUP_LEVEL_3
#define NODE_GROUP_LEVEL_4 4
#define NODE_GROUP_LEVEL_MAX NODE_GROUP_LEVEL_4
#define NODE_FEATURE_VOLUME (1 << 0)
#define NODE_FEATURE_HAIR (1 << 1)

7
intern/cycles/render/nodes.h
View File

@@ -197,6 +197,11 @@ class OutputAOVNode : public ShaderNode {
ustring name;
virtual int get_group()
{
return NODE_GROUP_LEVEL_4;
}
/* Don't allow output node de-duplication. */
virtual bool equals(const ShaderNode & /*other*/)
{
@@ -339,7 +344,7 @@ class PointDensityTextureNode : public ShaderNode {
SHADER_NODE_NO_CLONE_CLASS(PointDensityTextureNode)
virtual int get_group()
{
return NODE_GROUP_LEVEL_3;
return NODE_GROUP_LEVEL_4;
}
~PointDensityTextureNode();