[Bf-blender-cvs] [087331c495b] master: Cycles: Replace __MAX_CLOSURE__ build option with runtime integrator variable
Mai Lavelle
noreply at git.blender.org
Thu Nov 9 07:05:55 CET 2017
Commit: 087331c495b04ebd37903c0dc0e46262354cf026
Author: Mai Lavelle
Date: Thu Nov 9 00:49:15 2017 -0500
Branches: master
https://developer.blender.org/rB087331c495b04ebd37903c0dc0e46262354cf026
Cycles: Replace __MAX_CLOSURE__ build option with runtime integrator variable
Goal is to reduce OpenCL kernel recompilations.
Currently viewport renders are still set to use 64 closures as this seems to
be faster and we don't want to cause a performance regression there. Needs
to be investigated.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2775
===================================================================
M intern/cycles/device/device.cpp
M intern/cycles/device/device.h
M intern/cycles/device/device_cpu.cpp
M intern/cycles/device/device_cuda.cpp
M intern/cycles/device/device_split_kernel.cpp
M intern/cycles/device/device_split_kernel.h
M intern/cycles/kernel/kernel_bake.h
M intern/cycles/kernel/kernel_path.h
M intern/cycles/kernel/kernel_path_branched.h
M intern/cycles/kernel/kernel_subsurface.h
M intern/cycles/kernel/kernel_types.h
M intern/cycles/kernel/kernel_volume.h
M intern/cycles/kernel/split/kernel_branched.h
M intern/cycles/kernel/split/kernel_direct_lighting.h
M intern/cycles/kernel/split/kernel_do_volume.h
M intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
M intern/cycles/kernel/split/kernel_indirect_background.h
M intern/cycles/kernel/split/kernel_lamp_emission.h
M intern/cycles/kernel/split/kernel_next_iteration_setup.h
M intern/cycles/kernel/split/kernel_shader_eval.h
M intern/cycles/kernel/split/kernel_shader_setup.h
M intern/cycles/kernel/split/kernel_shader_sort.h
M intern/cycles/kernel/split/kernel_shadow_blocked_ao.h
M intern/cycles/kernel/split/kernel_shadow_blocked_dl.h
M intern/cycles/kernel/split/kernel_split_data.h
M intern/cycles/kernel/split/kernel_split_data_types.h
M intern/cycles/kernel/split/kernel_subsurface_scatter.h
M intern/cycles/render/session.cpp
===================================================================
diff --git a/intern/cycles/device/device.cpp b/intern/cycles/device/device.cpp
index 641e3fde140..a3ff5481cef 100644
--- a/intern/cycles/device/device.cpp
+++ b/intern/cycles/device/device.cpp
@@ -47,7 +47,6 @@ std::ostream& operator <<(std::ostream &os,
{
os << "Experimental features: "
<< (requested_features.experimental ? "On" : "Off") << std::endl;
- os << "Max closure count: " << requested_features.max_closure << std::endl;
os << "Max nodes group: " << requested_features.max_nodes_group << std::endl;
/* TODO(sergey): Decode bitflag into list of names. */
os << "Nodes features: " << requested_features.nodes_features << std::endl;
diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index 6bf3bbe6c25..35b545388f2 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -91,9 +91,6 @@ public:
/* Use experimental feature set. */
bool experimental;
- /* Maximum number of closures in shader trees. */
- int max_closure;
-
/* Selective nodes compilation. */
/* Identifier of a node group up to which all the nodes needs to be
@@ -146,7 +143,6 @@ public:
{
/* TODO(sergey): Find more meaningful defaults. */
experimental = false;
- max_closure = 0;
max_nodes_group = 0;
nodes_features = 0;
use_hair = false;
@@ -167,7 +163,6 @@ public:
bool modified(const DeviceRequestedFeatures& requested_features)
{
return !(experimental == requested_features.experimental &&
- max_closure == requested_features.max_closure &&
max_nodes_group == requested_features.max_nodes_group &&
nodes_features == requested_features.nodes_features &&
use_hair == requested_features.use_hair &&
@@ -198,7 +193,6 @@ public:
string_printf("%d", max_nodes_group);
build_options += " -D__NODES_FEATURES__=" +
string_printf("%d", nodes_features);
- build_options += string_printf(" -D__MAX_CLOSURE__=%d", max_closure);
if(!use_hair) {
build_options += " -D__NO_HAIR__";
}
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index 1a54c3380ee..0f4001ab1a6 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -760,7 +760,6 @@ public:
CPUSplitKernel *split_kernel = NULL;
if(use_split_kernel) {
split_kernel = new CPUSplitKernel(this);
- requested_features.max_closure = MAX_CLOSURE;
if(!split_kernel->load_kernels(requested_features)) {
thread_kernel_globals_free((KernelGlobals*)kgbuffer.device_pointer);
kgbuffer.free();
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 14e3ddc8c7b..d230a0c565d 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -1861,10 +1861,6 @@ public:
DeviceRequestedFeatures requested_features;
if(use_split_kernel()) {
- if(!use_adaptive_compilation()) {
- requested_features.max_closure = 64;
- }
-
if(split_kernel == NULL) {
split_kernel = new CUDASplitKernel(this);
split_kernel->load_kernels(requested_features);
diff --git a/intern/cycles/device/device_split_kernel.cpp b/intern/cycles/device/device_split_kernel.cpp
index 115273d9f0a..566d4020b33 100644
--- a/intern/cycles/device/device_split_kernel.cpp
+++ b/intern/cycles/device/device_split_kernel.cpp
@@ -34,7 +34,6 @@ DeviceSplitKernel::DeviceSplitKernel(Device *device)
use_queues_flag(device, "use_queues_flag"),
work_pool_wgs(device, "work_pool_wgs")
{
- current_max_closure = -1;
first_tile = true;
avg_time_per_sample = 0.0;
@@ -116,8 +115,6 @@ bool DeviceSplitKernel::load_kernels(const DeviceRequestedFeatures& requested_fe
#undef LOAD_KERNEL
- current_max_closure = requested_features.max_closure;
-
return true;
}
diff --git a/intern/cycles/device/device_split_kernel.h b/intern/cycles/device/device_split_kernel.h
index 0647c664447..2ec0261e847 100644
--- a/intern/cycles/device/device_split_kernel.h
+++ b/intern/cycles/device/device_split_kernel.h
@@ -92,9 +92,6 @@ private:
/* Work pool with respect to each work group. */
device_only_memory<unsigned int> work_pool_wgs;
- /* clos_max value for which the kernels have been loaded currently. */
- int current_max_closure;
-
/* Marked True in constructor and marked false at the end of path_trace(). */
bool first_tile;
diff --git a/intern/cycles/kernel/kernel_bake.h b/intern/cycles/kernel/kernel_bake.h
index 73cddeb27f7..8788e89c40e 100644
--- a/intern/cycles/kernel/kernel_bake.h
+++ b/intern/cycles/kernel/kernel_bake.h
@@ -51,7 +51,7 @@ ccl_device_inline void compute_light_pass(KernelGlobals *kg,
path_state_init(kg, &emission_sd, &state, rng_hash, sample, NULL);
/* evaluate surface shader */
- shader_eval_surface(kg, sd, &state, state.flag, MAX_CLOSURE);
+ shader_eval_surface(kg, sd, &state, state.flag, kernel_data.integrator.max_closures);
/* TODO, disable more closures we don't need besides transparent */
shader_bsdf_disable_transparency(kg, sd);
@@ -228,12 +228,12 @@ ccl_device float3 kernel_bake_evaluate_direct_indirect(KernelGlobals *kg,
}
else {
/* surface color of the pass only */
- shader_eval_surface(kg, sd, state, 0, MAX_CLOSURE);
+ shader_eval_surface(kg, sd, state, 0, kernel_data.integrator.max_closures);
return kernel_bake_shader_bsdf(kg, sd, type);
}
}
else {
- shader_eval_surface(kg, sd, state, 0, MAX_CLOSURE);
+ shader_eval_surface(kg, sd, state, 0, kernel_data.integrator.max_closures);
color = kernel_bake_shader_bsdf(kg, sd, type);
}
@@ -333,7 +333,7 @@ ccl_device void kernel_bake_evaluate(KernelGlobals *kg, ccl_global uint4 *input,
{
float3 N = sd.N;
if((sd.flag & SD_HAS_BUMP)) {
- shader_eval_surface(kg, &sd, &state, 0, MAX_CLOSURE);
+ shader_eval_surface(kg, &sd, &state, 0, kernel_data.integrator.max_closures);
N = shader_bsdf_average_normal(kg, &sd);
}
diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h
index 8519e0682e1..207ba741e6f 100644
--- a/intern/cycles/kernel/kernel_path.h
+++ b/intern/cycles/kernel/kernel_path.h
@@ -443,7 +443,7 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
sd,
&isect,
ray);
- shader_eval_surface(kg, sd, state, state->flag, MAX_CLOSURE);
+ shader_eval_surface(kg, sd, state, state->flag, kernel_data.integrator.max_closures);
shader_prepare_closures(sd, state);
/* Apply shadow catcher, holdout, emission. */
@@ -594,7 +594,7 @@ ccl_device_forceinline void kernel_path_integrate(
/* Setup and evaluate shader. */
shader_setup_from_ray(kg, &sd, &isect, ray);
- shader_eval_surface(kg, &sd, state, state->flag, MAX_CLOSURE);
+ shader_eval_surface(kg, &sd, state, state->flag, kernel_data.integrator.max_closures);
shader_prepare_closures(&sd, state);
/* Apply shadow catcher, holdout, emission. */
diff --git a/intern/cycles/kernel/kernel_path_branched.h b/intern/cycles/kernel/kernel_path_branched.h
index b37bc65f4df..9996f52f9a4 100644
--- a/intern/cycles/kernel/kernel_path_branched.h
+++ b/intern/cycles/kernel/kernel_path_branched.h
@@ -474,7 +474,7 @@ ccl_device void kernel_branched_path_integrate(KernelGlobals *kg,
/* Setup and evaluate shader. */
shader_setup_from_ray(kg, &sd, &isect, &ray);
- shader_eval_surface(kg, &sd, &state, state.flag, MAX_CLOSURE);
+ shader_eval_surface(kg, &sd, &state, state.flag, kernel_data.integrator.max_closures);
shader_merge_closures(&sd);
/* Apply shadow catcher, holdout, emission. */
diff --git a/intern/cycles/kernel/kernel_subsurface.h b/intern/cycles/kernel/kernel_subsurface.h
index 87e7d7ff398..616ad71af3c 100644
--- a/intern/cycles/kernel/kernel_subsurface.h
+++ b/intern/cycles/kernel/kernel_subsurface.h
@@ -76,11 +76,11 @@ ccl_device_inline float3 subsurface_scatter_eval(ShaderData *sd,
}
/* replace closures with a single diffuse bsdf closure after scatter step */
-ccl_device void subsurface_scatter_setup_diffuse_bsdf(ShaderData *sd, const ShaderClosure *sc, float3 weight, bool hit, float3 N)
+ccl_device void subsurface_scatter_setup_diffuse_bsdf(KernelGlobals *kg, ShaderData *sd, const ShaderClosure *sc, float3 weight, bool hit, float3 N)
{
sd->flag &= ~SD_CLOSURE_FLAGS;
sd->num_closure = 0;
- sd->num_closure_left = MAX_CLOSURE;
+ sd->num_closure_left = kernel_data.integrator.max_closures;
if(hit) {
Bssrdf *bssrdf = (Bssrdf *)sc;
@@ -154,7 +154,7 @@ ccl_device void subsurface_color_bump_blur(KernelGlobals *kg,
if(bump || texture_blur > 0.0f) {
/* average color and normal at incoming point */
- shader_eval_surface(kg, sd, state, state_flag, MAX_CLOSURE);
+ shader_eval_surface(kg, sd, state, state_flag, kernel_data.integrator.max_closures);
float3 in_color = shader_bssrdf_sum(sd, (bump)? N: NULL, NULL);
/* we simply divide out the average color and multiply with the average
@@ -342,7 +342,7 @@ ccl_device_noinline void subsurface_scatter_multi_setup(
subsurface_color_bump_blur(kg, sd, state, state_flag, &weight, &N);
/* Setup diffuse BSDF. */
- subsurface_scatter_setup_diffuse_bsdf(sd, sc, weight, true, N);
+ subsurface_scatter_setup_diffuse_bsdf(kg, sd, sc, weight, true, N);
}
/* subsurface scattering step, from a point on the surface to another nearby point on the same object */
@@ -439,7 +439,7 @@ ccl_device void subsurface_scatter_step(KernelGlobals *kg, ShaderData *sd, ccl_a
subsurface_color_bump_blur(kg, sd, state, state_flag, &eval, &N);
/* setup diffuse bsdf */
- subsurface_scatter_setup_diffuse_bsdf(sd, sc, eval, (ss_isect.num_hits > 0), N);
+ subsurface_scatter_setup_diffuse_bsdf(kg, sd, sc, eval, (ss_isect.num_hits > 0), N);
}
CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index fc3e7b3da98..919dafbc780 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -782,10 +782,14 @@ typedef struct AttributeDescriptor {
/* Closure data */
#ifdef __MULTI_CLOSURE__
-# ifndef __MAX_CLOSURE__
-# define MAX_CLOSURE 64
+# ifdef __SPLIT_KERNEL__
+# define MAX_CLOSURE 1
# else
-# define MAX_CLOSUR
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list