[Bf-blender-cvs] [e6f5632eb11] blender2.7: T61513: Refactored Cycles Attribute Retrieval
Jeroen Bakker
noreply at git.blender.org
Tue Feb 19 16:30:03 CET 2019
Commit: e6f5632eb11b37a2398f80841a77674656243dcf
Author: Jeroen Bakker
Date: Tue Feb 19 15:41:22 2019 +0100
Branches: blender2.7
https://developer.blender.org/rBe6f5632eb11b37a2398f80841a77674656243dcf
T61513: Refactored Cycles Attribute Retrieval
There is a generic function to retrieve float and float3 attributes
`primitive_attribute_float` and primitive_attribute_float3`. Inside
these functions an prioritised if-else construction checked where
the attribute is stored and then retrieved from that location.
Actually the calling function most of the time already knows where
the data is stored. So we could simplify this by splitting these
functions and remove the check logic.
This patch splits the `primitive_attribute_float?` functions into
`primitive_surface_attribute_float?` and `primitive_volume_attribute_float?`.
What leads to less branching and more optimum kernels.
The original function is still being used by OSL and `svm_node_attr`.
This will reduce the compilation time and render time for kernels.
Especially in production scenes there is a lot of benefit.
Impact in compilation times
job | scene_name | previous | new | percentage
-------+-----------------+----------+-------+------------
t61513 | empty | 10.63 | 10.66 | 0%
t61513 | bmw | 17.91 | 17.65 | 1%
t61513 | fishycat | 19.57 | 17.68 | 10%
t61513 | barbershop | 54.10 | 24.41 | 55%
t61513 | classroom | 17.55 | 16.29 | 7%
t61513 | koro | 18.92 | 18.05 | 5%
t61513 | pavillion | 17.43 | 16.52 | 5%
t61513 | splash279 | 16.48 | 14.91 | 10%
t61513 | volume_emission | 36.22 | 21.60 | 40%
Impact in render times
job | scene_name | previous | new | percentage
-------+-----------------+----------+--------+------------
61513 | empty | 21.06 | 20.35 | 3%
61513 | bmw | 198.44 | 190.05 | 4%
61513 | fishycat | 394.20 | 401.25 | -2%
61513 | barbershop | 1188.16 | 912.39 | 23%
61513 | classroom | 341.08 | 340.38 | 0%
61513 | koro | 472.43 | 471.80 | 0%
61513 | pavillion | 905.77 | 899.80 | 1%
61513 | splash279 | 55.26 | 54.86 | 1%
61513 | volume_emission | 62.59 | 61.70 | 1%
There is also a possitive impact when using CPU and CUDA, but they are small.
I didn't split the hair logic from the surface logic due to:
* Hair and surface use same attribute types. It was not clear if it could be
splitted when looking at the code only.
* Hair and surface are quick to compile and to read. So the benefit is quite
small.
Differential Revision: https://developer.blender.org/D4375
===================================================================
M intern/cycles/kernel/geom/geom_primitive.h
M intern/cycles/kernel/geom/geom_volume.h
M intern/cycles/kernel/osl/osl_services.cpp
M intern/cycles/kernel/svm/svm_attribute.h
M intern/cycles/kernel/svm/svm_bump.h
M intern/cycles/kernel/svm/svm_closure.h
M intern/cycles/kernel/svm/svm_displace.h
M intern/cycles/kernel/svm/svm_tex_coord.h
===================================================================
diff --git a/intern/cycles/kernel/geom/geom_primitive.h b/intern/cycles/kernel/geom/geom_primitive.h
index 6db8475d196..e3e2648e9ec 100644
--- a/intern/cycles/kernel/geom/geom_primitive.h
+++ b/intern/cycles/kernel/geom/geom_primitive.h
@@ -22,7 +22,6 @@
CCL_NAMESPACE_BEGIN
/* Generic primitive attribute reading functions */
-
ccl_device_inline float primitive_attribute_float(KernelGlobals *kg,
const ShaderData *sd,
const AttributeDescriptor desc,
@@ -41,7 +40,9 @@ ccl_device_inline float primitive_attribute_float(KernelGlobals *kg,
#endif
#ifdef __VOLUME__
else if(sd->object != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
- return volume_attribute_float(kg, sd, desc, dx, dy);
+ if(dx) *dx = 0.0f;
+ if(dy) *dy = 0.0f;
+ return volume_attribute_float(kg, sd, desc);
}
#endif
else {
@@ -51,6 +52,43 @@ ccl_device_inline float primitive_attribute_float(KernelGlobals *kg,
}
}
+ccl_device_inline float primitive_surface_attribute_float(KernelGlobals *kg,
+ const ShaderData *sd,
+ const AttributeDescriptor desc,
+ float *dx, float *dy)
+{
+ if(sd->type & PRIMITIVE_ALL_TRIANGLE) {
+ if(subd_triangle_patch(kg, sd) == ~0)
+ return triangle_attribute_float(kg, sd, desc, dx, dy);
+ else
+ return subd_triangle_attribute_float(kg, sd, desc, dx, dy);
+ }
+#ifdef __HAIR__
+ else if(sd->type & PRIMITIVE_ALL_CURVE) {
+ return curve_attribute_float(kg, sd, desc, dx, dy);
+ }
+#endif
+ else {
+ if(dx) *dx = 0.0f;
+ if(dy) *dy = 0.0f;
+ return 0.0f;
+ }
+}
+
+#ifdef __VOLUME__
+ccl_device_inline float primitive_volume_attribute_float(KernelGlobals *kg,
+ const ShaderData *sd,
+ const AttributeDescriptor desc)
+{
+ if(sd->object != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
+ return volume_attribute_float(kg, sd, desc);
+ }
+ else {
+ return 0.0f;
+ }
+}
+#endif
+
ccl_device_inline float3 primitive_attribute_float3(KernelGlobals *kg,
const ShaderData *sd,
const AttributeDescriptor desc,
@@ -69,7 +107,32 @@ ccl_device_inline float3 primitive_attribute_float3(KernelGlobals *kg,
#endif
#ifdef __VOLUME__
else if(sd->object != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
- return volume_attribute_float3(kg, sd, desc, dx, dy);
+ if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
+ if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
+ return volume_attribute_float3(kg, sd, desc);
+ }
+#endif
+ else {
+ if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
+ if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
+ return make_float3(0.0f, 0.0f, 0.0f);
+ }
+}
+
+ccl_device_inline float3 primitive_surface_attribute_float3(KernelGlobals *kg,
+ const ShaderData *sd,
+ const AttributeDescriptor desc,
+ float3 *dx, float3 *dy)
+{
+ if(sd->type & PRIMITIVE_ALL_TRIANGLE) {
+ if(subd_triangle_patch(kg, sd) == ~0)
+ return triangle_attribute_float3(kg, sd, desc, dx, dy);
+ else
+ return subd_triangle_attribute_float3(kg, sd, desc, dx, dy);
+ }
+#ifdef __HAIR__
+ else if(sd->type & PRIMITIVE_ALL_CURVE) {
+ return curve_attribute_float3(kg, sd, desc, dx, dy);
}
#endif
else {
@@ -79,6 +142,20 @@ ccl_device_inline float3 primitive_attribute_float3(KernelGlobals *kg,
}
}
+#ifdef __VOLUME__
+ccl_device_inline float3 primitive_volume_attribute_float3(KernelGlobals *kg,
+ const ShaderData *sd,
+ const AttributeDescriptor desc)
+{
+ if(sd->object != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
+ return volume_attribute_float3(kg, sd, desc);
+ }
+ else {
+ return make_float3(0.0f, 0.0f, 0.0f);
+ }
+}
+#endif
+
/* Default UV coordinate */
ccl_device_inline float3 primitive_uv(KernelGlobals *kg, ShaderData *sd)
@@ -88,7 +165,7 @@ ccl_device_inline float3 primitive_uv(KernelGlobals *kg, ShaderData *sd)
if(desc.offset == ATTR_STD_NOT_FOUND)
return make_float3(0.0f, 0.0f, 0.0f);
- float3 uv = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
+ float3 uv = primitive_surface_attribute_float3(kg, sd, desc, NULL, NULL);
uv.z = 1.0f;
return uv;
}
@@ -104,8 +181,8 @@ ccl_device bool primitive_ptex(KernelGlobals *kg, ShaderData *sd, float2 *uv, in
if(desc_face_id.offset == ATTR_STD_NOT_FOUND || desc_uv.offset == ATTR_STD_NOT_FOUND)
return false;
- float3 uv3 = primitive_attribute_float3(kg, sd, desc_uv, NULL, NULL);
- float face_id_f = primitive_attribute_float(kg, sd, desc_face_id, NULL, NULL);
+ float3 uv3 = primitive_surface_attribute_float3(kg, sd, desc_uv, NULL, NULL);
+ float face_id_f = primitive_surface_attribute_float(kg, sd, desc_face_id, NULL, NULL);
*uv = make_float2(uv3.x, uv3.y);
*face_id = (int)face_id_f;
@@ -130,7 +207,7 @@ ccl_device float3 primitive_tangent(KernelGlobals *kg, ShaderData *sd)
const AttributeDescriptor desc = find_attribute(kg, sd, ATTR_STD_GENERATED);
if(desc.offset != ATTR_STD_NOT_FOUND) {
- float3 data = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
+ float3 data = primitive_surface_attribute_float3(kg, sd, desc, NULL, NULL);
data = make_float3(-(data.y - 0.5f), (data.x - 0.5f), 0.0f);
object_normal_transform(kg, sd, &data);
return cross(sd->N, normalize(cross(data, sd->N)));
@@ -176,10 +253,10 @@ ccl_device_inline float4 primitive_motion_vector(KernelGlobals *kg, ShaderData *
object_motion_info(kg, sd->object, NULL, &numverts, &numkeys);
/* lookup attributes */
- motion_pre = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
+ motion_pre = primitive_surface_attribute_float3(kg, sd, desc, NULL, NULL);
desc.offset += (sd->type & PRIMITIVE_ALL_TRIANGLE)? numverts: numkeys;
- motion_post = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
+ motion_post = primitive_surface_attribute_float3(kg, sd, desc, NULL, NULL);
#ifdef __HAIR__
if(is_curve_primitive && (sd->object_flag & SD_OBJECT_HAS_VERTEX_MOTION) == 0) {
diff --git a/intern/cycles/kernel/geom/geom_volume.h b/intern/cycles/kernel/geom/geom_volume.h
index 688413b74a1..1977d263ece 100644
--- a/intern/cycles/kernel/geom/geom_volume.h
+++ b/intern/cycles/kernel/geom/geom_volume.h
@@ -47,27 +47,20 @@ ccl_device_inline float3 volume_normalized_position(KernelGlobals *kg,
return P;
}
-ccl_device float volume_attribute_float(KernelGlobals *kg, const ShaderData *sd, const AttributeDescriptor desc, float *dx, float *dy)
+ccl_device float volume_attribute_float(KernelGlobals *kg, const ShaderData *sd, const AttributeDescriptor desc)
{
float3 P = volume_normalized_position(kg, sd, sd->P);
InterpolationType interp = (sd->flag & SD_VOLUME_CUBIC)? INTERPOLATION_CUBIC: INTERPOLATION_NONE;
float4 r = kernel_tex_image_interp_3d(kg, desc.offset, P.x, P.y, P.z, interp);
-
- if(dx) *dx = 0.0f;
- if(dy) *dy = 0.0f;
-
return average(float4_to_float3(r));
}
-ccl_device float3 volume_attribute_float3(KernelGlobals *kg, const ShaderData *sd, const AttributeDescriptor desc, float3 *dx, float3 *dy)
+ccl_device float3 volume_attribute_float3(KernelGlobals *kg, const ShaderData *sd, const AttributeDescriptor desc)
{
float3 P = volume_normalized_position(kg, sd, sd->P);
InterpolationType interp = (sd->flag & SD_VOLUME_CUBIC)? INTERPOLATION_CUBIC: INTERPOLATION_NONE;
float4 r = kernel_tex_image_interp_3d(kg, desc.offset, P.x, P.y, P.z, interp);
- if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
- if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
-
if(r.w > 1e-6f && r.w != 1.0f) {
/* For RGBA colors, unpremultiply after interpolation. */
return float4_to_float3(r) / r.w;
diff --git a/intern/cycles/kernel/osl/osl_services.cpp b/intern/cycles/kernel/osl/osl_services.cpp
index 5b4bddbb6de..5436a66c9d4 100644
--- a/intern/cycles/kernel/osl/osl_services.cpp
+++ b/intern/cycles/kernel/osl/osl_services.cpp
@@ -561,7 +561,7 @@ static bool set_attribute_matrix(const Transform& tfm, TypeDesc type, void *val)
return false;
}
-static bool get_mesh_element_attribute(KernelGlobals *kg, const ShaderData *sd, const OSLGlobals::Attribute& attr,
+static bool get_primitive_attribute(KernelGlobals *kg, const ShaderData *sd, const OSLGlobals::Attribute& attr,
const TypeDesc& type, bool derivatives, void *val)
{
if(attr.type == TypeDesc::TypePoint || attr.type == TypeDesc::TypeVector ||
@@ -849,7 +849,7 @@ bool OSLRenderServices::get_attribute(ShaderData *sd, bool derivatives, ustring
if(attr.desc.element != ATTR_ELEMENT_OBJECT) {
/* triangle and vertex attributes */
- if(get_mesh_element_attribute(kg, sd, attr, type, derivatives, val))
+ if(get_primitive_attribute(kg, sd, attr, type, derivatives, val))
return true;
else
return get_mesh_attribute(kg, sd, attr, type, derivatives, val);
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index d98f538d089..ef6f7d7cbb5 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -52,24 +52,22 @@ ccl_device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, u
AttributeDescriptor desc = svm_node_attr_init(kg, sd, node, &type, &out_offset);
/* fetch and store attribute */
- if(type == NODE_ATTR_FLOAT) {
- if(desc.type == NODE_ATTR_FLOAT) {
- float f = primitive_attribute_float(kg, sd, desc, NULL, NULL);
+ if (desc.type == NODE_ATTR_FLOAT) {
+ float f = primitive_attribute_float(kg, sd, desc, NULL, NULL);
+ if (type == NODE_ATTR_FLOAT) {
stack_store_float(stack, out_offset, f);
}
else {
- float3 f = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
- stack_store_float(stack, out_offset, avera
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list