Welcome to mirror list, hosted at ThFree Co, Russian Federation.

git.blender.org/blender.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
path: root/intern/cycles/device
diff options
context:
space:
mode:
Diffstat (limited to 'intern/cycles/device')
-rw-r--r--intern/cycles/device/cuda/device_impl.cpp6
-rw-r--r--intern/cycles/device/cuda/queue.cpp6
-rw-r--r--intern/cycles/device/cuda/queue.h4
-rw-r--r--intern/cycles/device/hip/device_impl.cpp6
-rw-r--r--intern/cycles/device/hip/queue.cpp6
-rw-r--r--intern/cycles/device/hip/queue.h4
-rw-r--r--intern/cycles/device/optix/device_impl.cpp88
-rw-r--r--intern/cycles/device/optix/queue.cpp8
-rw-r--r--intern/cycles/device/optix/queue.h4
-rw-r--r--intern/cycles/device/queue.h70
10 files changed, 141 insertions, 61 deletions
diff --git a/intern/cycles/device/cuda/device_impl.cpp b/intern/cycles/device/cuda/device_impl.cpp
index e05fef3897c..ee55e6dc632 100644
--- a/intern/cycles/device/cuda/device_impl.cpp
+++ b/intern/cycles/device/cuda/device_impl.cpp
@@ -477,10 +477,10 @@ void CUDADevice::reserve_local_memory(const uint kernel_features)
* still to make it faster. */
CUDADeviceQueue queue(this);
- void *d_path_index = nullptr;
- void *d_render_buffer = nullptr;
+ device_ptr d_path_index = 0;
+ device_ptr d_render_buffer = 0;
int d_work_size = 0;
- void *args[] = {&d_path_index, &d_render_buffer, &d_work_size};
+ DeviceKernelArguments args(&d_path_index, &d_render_buffer, &d_work_size);
queue.init_execution();
queue.enqueue(test_kernel, 1, args);
diff --git a/intern/cycles/device/cuda/queue.cpp b/intern/cycles/device/cuda/queue.cpp
index 09352a84181..880d7ca4cf2 100644
--- a/intern/cycles/device/cuda/queue.cpp
+++ b/intern/cycles/device/cuda/queue.cpp
@@ -89,7 +89,9 @@ bool CUDADeviceQueue::kernel_available(DeviceKernel kernel) const
return cuda_device_->kernels.available(kernel);
}
-bool CUDADeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *args[])
+bool CUDADeviceQueue::enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args)
{
if (cuda_device_->have_error()) {
return false;
@@ -133,7 +135,7 @@ bool CUDADeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *ar
1,
shared_mem_bytes,
cuda_stream_,
- args,
+ const_cast<void**>(args.values),
0),
"enqueue");
diff --git a/intern/cycles/device/cuda/queue.h b/intern/cycles/device/cuda/queue.h
index 28613cda071..0836af12098 100644
--- a/intern/cycles/device/cuda/queue.h
+++ b/intern/cycles/device/cuda/queue.h
@@ -42,7 +42,9 @@ class CUDADeviceQueue : public DeviceQueue {
virtual bool kernel_available(DeviceKernel kernel) const override;
- virtual bool enqueue(DeviceKernel kernel, const int work_size, void *args[]) override;
+ virtual bool enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args) override;
virtual bool synchronize() override;
diff --git a/intern/cycles/device/hip/device_impl.cpp b/intern/cycles/device/hip/device_impl.cpp
index 53c4f3f0b3f..4f1cbabc89b 100644
--- a/intern/cycles/device/hip/device_impl.cpp
+++ b/intern/cycles/device/hip/device_impl.cpp
@@ -440,10 +440,10 @@ void HIPDevice::reserve_local_memory(const uint kernel_features)
* still to make it faster. */
HIPDeviceQueue queue(this);
- void *d_path_index = nullptr;
- void *d_render_buffer = nullptr;
+ device_ptr d_path_index = 0;
+ device_ptr d_render_buffer = 0;
int d_work_size = 0;
- void *args[] = {&d_path_index, &d_render_buffer, &d_work_size};
+ DeviceKernelArguments args(&d_path_index, &d_render_buffer, &d_work_size);
queue.init_execution();
queue.enqueue(test_kernel, 1, args);
diff --git a/intern/cycles/device/hip/queue.cpp b/intern/cycles/device/hip/queue.cpp
index 0f053ccbeb5..42841324ed6 100644
--- a/intern/cycles/device/hip/queue.cpp
+++ b/intern/cycles/device/hip/queue.cpp
@@ -89,7 +89,9 @@ bool HIPDeviceQueue::kernel_available(DeviceKernel kernel) const
return hip_device_->kernels.available(kernel);
}
-bool HIPDeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *args[])
+bool HIPDeviceQueue::enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args)
{
if (hip_device_->have_error()) {
return false;
@@ -132,7 +134,7 @@ bool HIPDeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *arg
1,
shared_mem_bytes,
hip_stream_,
- args,
+ const_cast<void**>(args.values),
0),
"enqueue");
diff --git a/intern/cycles/device/hip/queue.h b/intern/cycles/device/hip/queue.h
index 95d1afaff0f..8040d367798 100644
--- a/intern/cycles/device/hip/queue.h
+++ b/intern/cycles/device/hip/queue.h
@@ -42,7 +42,9 @@ class HIPDeviceQueue : public DeviceQueue {
virtual bool kernel_available(DeviceKernel kernel) const override;
- virtual bool enqueue(DeviceKernel kernel, const int work_size, void *args[]) override;
+ virtual bool enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args) override;
virtual bool synchronize() override;
diff --git a/intern/cycles/device/optix/device_impl.cpp b/intern/cycles/device/optix/device_impl.cpp
index b82b1281eb8..1d893d9c65b 100644
--- a/intern/cycles/device/optix/device_impl.cpp
+++ b/intern/cycles/device/optix/device_impl.cpp
@@ -667,22 +667,22 @@ bool OptiXDevice::denoise_filter_guiding_preprocess(DenoiseContext &context)
const int work_size = buffer_params.width * buffer_params.height;
- void *args[] = {const_cast<device_ptr *>(&context.guiding_params.device_pointer),
- const_cast<int *>(&context.guiding_params.pass_stride),
- const_cast<int *>(&context.guiding_params.pass_albedo),
- const_cast<int *>(&context.guiding_params.pass_normal),
- &context.render_buffers->buffer.device_pointer,
- const_cast<int *>(&buffer_params.offset),
- const_cast<int *>(&buffer_params.stride),
- const_cast<int *>(&buffer_params.pass_stride),
- const_cast<int *>(&context.pass_sample_count),
- const_cast<int *>(&context.pass_denoising_albedo),
- const_cast<int *>(&context.pass_denoising_normal),
- const_cast<int *>(&buffer_params.full_x),
- const_cast<int *>(&buffer_params.full_y),
- const_cast<int *>(&buffer_params.width),
- const_cast<int *>(&buffer_params.height),
- const_cast<int *>(&context.num_samples)};
+ DeviceKernelArguments args(&context.guiding_params.device_pointer,
+ &context.guiding_params.pass_stride,
+ &context.guiding_params.pass_albedo,
+ &context.guiding_params.pass_normal,
+ &context.render_buffers->buffer.device_pointer,
+ &buffer_params.offset,
+ &buffer_params.stride,
+ &buffer_params.pass_stride,
+ &context.pass_sample_count,
+ &context.pass_denoising_albedo,
+ &context.pass_denoising_normal,
+ &buffer_params.full_x,
+ &buffer_params.full_y,
+ &buffer_params.width,
+ &buffer_params.height,
+ &context.num_samples);
return denoiser_.queue.enqueue(DEVICE_KERNEL_FILTER_GUIDING_PREPROCESS, work_size, args);
}
@@ -693,11 +693,11 @@ bool OptiXDevice::denoise_filter_guiding_set_fake_albedo(DenoiseContext &context
const int work_size = buffer_params.width * buffer_params.height;
- void *args[] = {const_cast<device_ptr *>(&context.guiding_params.device_pointer),
- const_cast<int *>(&context.guiding_params.pass_stride),
- const_cast<int *>(&context.guiding_params.pass_albedo),
- const_cast<int *>(&buffer_params.width),
- const_cast<int *>(&buffer_params.height)};
+ DeviceKernelArguments args(&context.guiding_params.device_pointer,
+ &context.guiding_params.pass_stride,
+ &context.guiding_params.pass_albedo,
+ &buffer_params.width,
+ &buffer_params.height);
return denoiser_.queue.enqueue(DEVICE_KERNEL_FILTER_GUIDING_SET_FAKE_ALBEDO, work_size, args);
}
@@ -793,15 +793,15 @@ bool OptiXDevice::denoise_filter_color_preprocess(DenoiseContext &context, const
const int work_size = buffer_params.width * buffer_params.height;
- void *args[] = {&context.render_buffers->buffer.device_pointer,
- const_cast<int *>(&buffer_params.full_x),
- const_cast<int *>(&buffer_params.full_y),
- const_cast<int *>(&buffer_params.width),
- const_cast<int *>(&buffer_params.height),
- const_cast<int *>(&buffer_params.offset),
- const_cast<int *>(&buffer_params.stride),
- const_cast<int *>(&buffer_params.pass_stride),
- const_cast<int *>(&pass.denoised_offset)};
+ DeviceKernelArguments args(&context.render_buffers->buffer.device_pointer,
+ &buffer_params.full_x,
+ &buffer_params.full_y,
+ &buffer_params.width,
+ &buffer_params.height,
+ &buffer_params.offset,
+ &buffer_params.stride,
+ &buffer_params.pass_stride,
+ &pass.denoised_offset);
return denoiser_.queue.enqueue(DEVICE_KERNEL_FILTER_COLOR_PREPROCESS, work_size, args);
}
@@ -813,20 +813,20 @@ bool OptiXDevice::denoise_filter_color_postprocess(DenoiseContext &context,
const int work_size = buffer_params.width * buffer_params.height;
- void *args[] = {&context.render_buffers->buffer.device_pointer,
- const_cast<int *>(&buffer_params.full_x),
- const_cast<int *>(&buffer_params.full_y),
- const_cast<int *>(&buffer_params.width),
- const_cast<int *>(&buffer_params.height),
- const_cast<int *>(&buffer_params.offset),
- const_cast<int *>(&buffer_params.stride),
- const_cast<int *>(&buffer_params.pass_stride),
- const_cast<int *>(&context.num_samples),
- const_cast<int *>(&pass.noisy_offset),
- const_cast<int *>(&pass.denoised_offset),
- const_cast<int *>(&context.pass_sample_count),
- const_cast<int *>(&pass.num_components),
- const_cast<bool *>(&pass.use_compositing)};
+ DeviceKernelArguments args(&context.render_buffers->buffer.device_pointer,
+ &buffer_params.full_x,
+ &buffer_params.full_y,
+ &buffer_params.width,
+ &buffer_params.height,
+ &buffer_params.offset,
+ &buffer_params.stride,
+ &buffer_params.pass_stride,
+ &context.num_samples,
+ &pass.noisy_offset,
+ &pass.denoised_offset,
+ &context.pass_sample_count,
+ &pass.num_components,
+ &pass.use_compositing);
return denoiser_.queue.enqueue(DEVICE_KERNEL_FILTER_COLOR_POSTPROCESS, work_size, args);
}
diff --git a/intern/cycles/device/optix/queue.cpp b/intern/cycles/device/optix/queue.cpp
index e3946d94f5d..1a437878b5f 100644
--- a/intern/cycles/device/optix/queue.cpp
+++ b/intern/cycles/device/optix/queue.cpp
@@ -47,7 +47,9 @@ static bool is_optix_specific_kernel(DeviceKernel kernel)
kernel == DEVICE_KERNEL_INTEGRATOR_INTERSECT_VOLUME_STACK);
}
-bool OptiXDeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *args[])
+bool OptiXDeviceQueue::enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args)
{
if (!is_optix_specific_kernel(kernel)) {
return CUDADeviceQueue::enqueue(kernel, work_size, args);
@@ -69,7 +71,7 @@ bool OptiXDeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *a
cuda_device_assert(
cuda_device_,
cuMemcpyHtoDAsync(launch_params_ptr + offsetof(KernelParamsOptiX, path_index_array),
- args[0], // &d_path_index
+ args.values[0], // &d_path_index
sizeof(device_ptr),
cuda_stream_));
@@ -78,7 +80,7 @@ bool OptiXDeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *a
cuda_device_assert(
cuda_device_,
cuMemcpyHtoDAsync(launch_params_ptr + offsetof(KernelParamsOptiX, render_buffer),
- args[1], // &d_render_buffer
+ args.values[1], // &d_render_buffer
sizeof(device_ptr),
cuda_stream_));
}
diff --git a/intern/cycles/device/optix/queue.h b/intern/cycles/device/optix/queue.h
index 0de422ccc71..5f0e09dff2c 100644
--- a/intern/cycles/device/optix/queue.h
+++ b/intern/cycles/device/optix/queue.h
@@ -31,7 +31,9 @@ class OptiXDeviceQueue : public CUDADeviceQueue {
virtual void init_execution() override;
- virtual bool enqueue(DeviceKernel kernel, const int work_size, void *args[]) override;
+ virtual bool enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args) override;
};
CCL_NAMESPACE_END
diff --git a/intern/cycles/device/queue.h b/intern/cycles/device/queue.h
index 188162f4b74..4e9f41f7875 100644
--- a/intern/cycles/device/queue.h
+++ b/intern/cycles/device/queue.h
@@ -31,6 +31,72 @@ class device_memory;
struct KernelWorkTile;
+/* Container for device kernel arguments with type correctness ensured by API. */
+struct DeviceKernelArguments {
+
+ enum Type {
+ POINTER,
+ INT32,
+ FLOAT32,
+ BOOLEAN,
+ KERNEL_FILM_CONVERT,
+ };
+
+ static const int MAX_ARGS = 16;
+ Type types[MAX_ARGS];
+ void *values[MAX_ARGS];
+ size_t sizes[MAX_ARGS];
+ size_t count = 0;
+
+ DeviceKernelArguments()
+ {
+ }
+
+ template<class T> DeviceKernelArguments(const T *arg)
+ {
+ add(arg);
+ }
+
+ template<class T, class... Args> DeviceKernelArguments(const T *first, Args... args)
+ {
+ add(first);
+ add(args...);
+ }
+
+ void add(const KernelFilmConvert *value)
+ {
+ add(KERNEL_FILM_CONVERT, value, sizeof(KernelFilmConvert));
+ }
+ void add(const device_ptr *value)
+ {
+ add(POINTER, value, sizeof(device_ptr));
+ }
+ void add(const int32_t *value)
+ {
+ add(INT32, value, sizeof(int32_t));
+ }
+ void add(const float *value)
+ {
+ add(FLOAT32, value, sizeof(float));
+ }
+ void add(const bool *value)
+ {
+ add(BOOLEAN, value, 4);
+ }
+ void add(const Type type, const void *value, size_t size)
+ {
+ types[count] = type;
+ values[count] = (void *)value;
+ sizes[count] = size;
+ count++;
+ }
+ template<typename T, typename... Args> void add(const T *first, Args... args)
+ {
+ add(first);
+ add(args...);
+ }
+};
+
/* Abstraction of a command queue for a device.
* Provides API to schedule kernel execution in a specific queue with minimal possible overhead
* from driver side.
@@ -66,7 +132,9 @@ class DeviceQueue {
* - int: pass pointer to the int
* - device memory: pass pointer to device_memory.device_pointer
* Return false if there was an error executing this or a previous kernel. */
- virtual bool enqueue(DeviceKernel kernel, const int work_size, void *args[]) = 0;
+ virtual bool enqueue(DeviceKernel kernel,
+ const int work_size,
+ DeviceKernelArguments const &args) = 0;
/* Wait unit all enqueued kernels have finished execution.
* Return false if there was an error executing any of the enqueued kernels. */
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-05 06:34:15 +0300