[Bf-blender-cvs] [2ba7c3aa650] master: Cleanup: refactor to make number of channels for shader evaluation variable
Brecht Van Lommel
noreply at git.blender.org
Fri Oct 15 15:59:42 CEST 2021
Commit: 2ba7c3aa650c3c795d903a24998204f67c75b017
Author: Brecht Van Lommel
Date: Wed Oct 13 19:13:35 2021 +0200
Branches: master
https://developer.blender.org/rB2ba7c3aa650c3c795d903a24998204f67c75b017
Cleanup: refactor to make number of channels for shader evaluation variable
===================================================================
M intern/cycles/device/cpu/kernel.h
M intern/cycles/integrator/shader_eval.cpp
M intern/cycles/integrator/shader_eval.h
M intern/cycles/kernel/device/cpu/kernel_arch.h
M intern/cycles/kernel/device/cpu/kernel_arch_impl.h
M intern/cycles/kernel/device/gpu/kernel.h
M intern/cycles/kernel/integrator/integrator_intersect_shadow.h
M intern/cycles/kernel/kernel_bake.h
M intern/cycles/render/light.cpp
M intern/cycles/render/mesh_displace.cpp
===================================================================
diff --git a/intern/cycles/device/cpu/kernel.h b/intern/cycles/device/cpu/kernel.h
index 54b18308544..b5f0d873f30 100644
--- a/intern/cycles/device/cpu/kernel.h
+++ b/intern/cycles/device/cpu/kernel.h
@@ -54,7 +54,7 @@ class CPUKernels {
/* Shader evaluation. */
using ShaderEvalFunction = CPUKernelFunction<void (*)(
- const KernelGlobals *kg, const KernelShaderEvalInput *, float4 *, const int)>;
+ const KernelGlobals *kg, const KernelShaderEvalInput *, float *, const int)>;
ShaderEvalFunction shader_eval_displace;
ShaderEvalFunction shader_eval_background;
diff --git a/intern/cycles/integrator/shader_eval.cpp b/intern/cycles/integrator/shader_eval.cpp
index a14e41ec5be..53546c03872 100644
--- a/intern/cycles/integrator/shader_eval.cpp
+++ b/intern/cycles/integrator/shader_eval.cpp
@@ -34,9 +34,10 @@ ShaderEval::ShaderEval(Device *device, Progress &progress) : device_(device), pr
}
bool ShaderEval::eval(const ShaderEvalType type,
- const int max_num_points,
+ const int max_num_inputs,
+ const int num_channels,
const function<int(device_vector<KernelShaderEvalInput> &)> &fill_input,
- const function<void(device_vector<float4> &)> &read_output)
+ const function<void(device_vector<float> &)> &read_output)
{
bool first_device = true;
bool success = true;
@@ -50,26 +51,27 @@ bool ShaderEval::eval(const ShaderEvalType type,
first_device = false;
device_vector<KernelShaderEvalInput> input(device, "ShaderEval input", MEM_READ_ONLY);
- device_vector<float4> output(device, "ShaderEval output", MEM_READ_WRITE);
+ device_vector<float> output(device, "ShaderEval output", MEM_READ_WRITE);
/* Allocate and copy device buffers. */
DCHECK_EQ(input.device, device);
DCHECK_EQ(output.device, device);
DCHECK_LE(output.size(), input.size());
- input.alloc(max_num_points);
+ input.alloc(max_num_inputs);
int num_points = fill_input(input);
if (num_points == 0) {
return;
}
input.copy_to_device();
- output.alloc(num_points);
+ output.alloc(num_points * num_channels);
output.zero_to_device();
/* Evaluate on CPU or GPU. */
- success = (device->info.type == DEVICE_CPU) ? eval_cpu(device, type, input, output) :
- eval_gpu(device, type, input, output);
+ success = (device->info.type == DEVICE_CPU) ?
+ eval_cpu(device, type, input, output, num_points) :
+ eval_gpu(device, type, input, output, num_points);
/* Copy data back from device if not canceled. */
if (success) {
@@ -87,7 +89,8 @@ bool ShaderEval::eval(const ShaderEvalType type,
bool ShaderEval::eval_cpu(Device *device,
const ShaderEvalType type,
device_vector<KernelShaderEvalInput> &input,
- device_vector<float4> &output)
+ device_vector<float> &output,
+ const int64_t work_size)
{
vector<CPUKernelThreadGlobals> kernel_thread_globals;
device->get_cpu_kernel_thread_globals(kernel_thread_globals);
@@ -96,9 +99,8 @@ bool ShaderEval::eval_cpu(Device *device,
const CPUKernels &kernels = *(device->get_cpu_kernels());
/* Simple parallel_for over all work items. */
- const int64_t work_size = output.size();
KernelShaderEvalInput *input_data = input.data();
- float4 *output_data = output.data();
+ float *output_data = output.data();
bool success = true;
tbb::task_arena local_arena(device->info.cpu_threads);
@@ -130,7 +132,8 @@ bool ShaderEval::eval_cpu(Device *device,
bool ShaderEval::eval_gpu(Device *device,
const ShaderEvalType type,
device_vector<KernelShaderEvalInput> &input,
- device_vector<float4> &output)
+ device_vector<float> &output,
+ const int64_t work_size)
{
/* Find required kernel function. */
DeviceKernel kernel;
@@ -151,7 +154,6 @@ bool ShaderEval::eval_gpu(Device *device,
* TODO : query appropriate size from device.*/
const int64_t chunk_size = 65536;
- const int64_t work_size = output.size();
void *d_input = (void *)input.device_pointer;
void *d_output = (void *)output.device_pointer;
diff --git a/intern/cycles/integrator/shader_eval.h b/intern/cycles/integrator/shader_eval.h
index 7dbf334b8d7..013fad17d4f 100644
--- a/intern/cycles/integrator/shader_eval.h
+++ b/intern/cycles/integrator/shader_eval.h
@@ -40,19 +40,22 @@ class ShaderEval {
/* Evaluate shader at points specified by KernelShaderEvalInput and write out
* RGBA colors to output. */
bool eval(const ShaderEvalType type,
- const int max_num_points,
+ const int max_num_inputs,
+ const int num_channels,
const function<int(device_vector<KernelShaderEvalInput> &)> &fill_input,
- const function<void(device_vector<float4> &)> &read_output);
+ const function<void(device_vector<float> &)> &read_output);
protected:
bool eval_cpu(Device *device,
const ShaderEvalType type,
device_vector<KernelShaderEvalInput> &input,
- device_vector<float4> &output);
+ device_vector<float> &output,
+ const int64_t work_size);
bool eval_gpu(Device *device,
const ShaderEvalType type,
device_vector<KernelShaderEvalInput> &input,
- device_vector<float4> &output);
+ device_vector<float> &output,
+ const int64_t work_size);
Device *device_;
Progress &progress_;
diff --git a/intern/cycles/kernel/device/cpu/kernel_arch.h b/intern/cycles/kernel/device/cpu/kernel_arch.h
index 81f328c710b..8b7b0ec0548 100644
--- a/intern/cycles/kernel/device/cpu/kernel_arch.h
+++ b/intern/cycles/kernel/device/cpu/kernel_arch.h
@@ -58,11 +58,11 @@ KERNEL_INTEGRATOR_SHADE_FUNCTION(megakernel);
void KERNEL_FUNCTION_FULL_NAME(shader_eval_background)(const KernelGlobals *kg,
const KernelShaderEvalInput *input,
- float4 *output,
+ float *output,
const int offset);
void KERNEL_FUNCTION_FULL_NAME(shader_eval_displace)(const KernelGlobals *kg,
const KernelShaderEvalInput *input,
- float4 *output,
+ float *output,
const int offset);
/* --------------------------------------------------------------------
diff --git a/intern/cycles/kernel/device/cpu/kernel_arch_impl.h b/intern/cycles/kernel/device/cpu/kernel_arch_impl.h
index 1432abfd330..23e371f165f 100644
--- a/intern/cycles/kernel/device/cpu/kernel_arch_impl.h
+++ b/intern/cycles/kernel/device/cpu/kernel_arch_impl.h
@@ -114,7 +114,7 @@ DEFINE_INTEGRATOR_SHADE_KERNEL(megakernel)
void KERNEL_FUNCTION_FULL_NAME(shader_eval_displace)(const KernelGlobals *kg,
const KernelShaderEvalInput *input,
- float4 *output,
+ float *output,
const int offset)
{
#ifdef KERNEL_STUB
@@ -126,7 +126,7 @@ void KERNEL_FUNCTION_FULL_NAME(shader_eval_displace)(const KernelGlobals *kg,
void KERNEL_FUNCTION_FULL_NAME(shader_eval_background)(const KernelGlobals *kg,
const KernelShaderEvalInput *input,
- float4 *output,
+ float *output,
const int offset)
{
#ifdef KERNEL_STUB
diff --git a/intern/cycles/kernel/device/gpu/kernel.h b/intern/cycles/kernel/device/gpu/kernel.h
index 3379114fc62..21901215757 100644
--- a/intern/cycles/kernel/device/gpu/kernel.h
+++ b/intern/cycles/kernel/device/gpu/kernel.h
@@ -615,7 +615,7 @@ KERNEL_FILM_CONVERT_DEFINE(float4, rgba)
ccl_gpu_kernel(GPU_KERNEL_BLOCK_NUM_THREADS, GPU_KERNEL_MAX_REGISTERS)
kernel_gpu_shader_eval_displace(KernelShaderEvalInput *input,
- float4 *output,
+ float *output,
const int offset,
const int work_size)
{
@@ -629,7 +629,7 @@ ccl_gpu_kernel(GPU_KERNEL_BLOCK_NUM_THREADS, GPU_KERNEL_MAX_REGISTERS)
ccl_gpu_kernel(GPU_KERNEL_BLOCK_NUM_THREADS, GPU_KERNEL_MAX_REGISTERS)
kernel_gpu_shader_eval_background(KernelShaderEvalInput *input,
- float4 *output,
+ float *output,
const int offset,
const int work_size)
{
diff --git a/intern/cycles/kernel/integrator/integrator_intersect_shadow.h b/intern/cycles/kernel/integrator/integrator_intersect_shadow.h
index 00d44f0e5ed..3ebd21e4651 100644
--- a/intern/cycles/kernel/integrator/integrator_intersect_shadow.h
+++ b/intern/cycles/kernel/integrator/integrator_intersect_shadow.h
@@ -85,7 +85,8 @@ ccl_device bool integrate_intersect_shadow_transparent(INTEGRATOR_STATE_ARGS,
if (num_recorded_hits > 0) {
sort_intersections(isect, num_recorded_hits);
- /* Write intersection result into global integrator state memory. */
+ /* Write intersection result into global integrator state memory.
+ * More efficient may be to do this directly from the intersection kernel. */
for (int hit = 0; hit < num_recorded_hits; hit++) {
integrator_state_write_shadow_isect(INTEGRATOR_STATE_PASS, &isect[hit]
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list