Commit: e3bb3ed5ba3dc8911af52bad4f0355f0caff93e4
Author: Mai Lavelle
Date: Thu Oct 27 18:52:04 2016 +0200
Branches: cycles_split_kernel
https://developer.blender.org/rBe3bb3ed5ba3dc8911af52bad4f0355f0caff93e4
Cycles: Add function so each device can specify its ideal local work size
===================================================================
M intern/cycles/device/device.h
M intern/cycles/device/device_cpu.cpp
M intern/cycles/device/device_split_kernel.cpp
M intern/cycles/device/device_split_kernel.h
M intern/cycles/device/opencl/opencl_split.cpp
===================================================================
diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index 71e9a02..c832e19 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -328,6 +328,12 @@ public:
assert(!"not implemented for this device");
}
+ virtual int2 split_kernel_local_size()
+ {
+ assert(!"not implemented for this device");
+ return make_int2(0, 0);
+ }
+
/* tasks */
virtual int get_split_task_count(DeviceTask& task) = 0;
virtual void task_add(DeviceTask& task) = 0;
diff --git a/intern/cycles/device/device_cpu.cpp
b/intern/cycles/device/device_cpu.cpp
index 1955952..50b6ad1 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -713,6 +713,11 @@ protected:
{
thread_kernel_globals_free((KernelGlobals*)mem.device_pointer);
}
+
+ virtual int2 split_kernel_local_size()
+ {
+ return make_int2(1, 1);
+ }
};
unordered_map<string, void*> CPUDevice::kernel_functions;
diff --git a/intern/cycles/device/device_split_kernel.cpp
b/intern/cycles/device/device_split_kernel.cpp
index b9cec54..4bb3c5d 100644
--- a/intern/cycles/device/device_split_kernel.cpp
+++ b/intern/cycles/device/device_split_kernel.cpp
@@ -84,13 +84,22 @@ bool DeviceSplitKernel::path_trace(DeviceTask *task,
tile.buffer_rng_state_stride = tile.stride;
tile.stride = tile.w;
+ size_t global_size[2];
+ size_t local_size[2];
+
+ {
+ int2 lsize = device->split_kernel_local_size();
+ local_size[0] = lsize[0];
+ local_size[1] = lsize[1];
+ }
+
/* Make sure that set render feasible tile size is a multiple of local
* work size dimensions.
*/
int2 max_render_feasible_tile_size;
const int2 tile_size = task->requested_tile_size;
- max_render_feasible_tile_size.x = ROUND_UP(tile_size.x,
SPLIT_KERNEL_LOCAL_SIZE_X);
- max_render_feasible_tile_size.y = ROUND_UP(tile_size.y,
SPLIT_KERNEL_LOCAL_SIZE_Y);
+ max_render_feasible_tile_size.x = ROUND_UP(tile_size.x, local_size[0]);
+ max_render_feasible_tile_size.y = ROUND_UP(tile_size.y, local_size[1]);
/* Calculate per_thread_output_buffer_size. */
size_t per_thread_output_buffer_size;
@@ -116,10 +125,6 @@ bool DeviceSplitKernel::path_trace(DeviceTask *task,
tile.buffers->params.height);
}
- size_t global_size[2];
- size_t local_size[2] = {SPLIT_KERNEL_LOCAL_SIZE_X,
- SPLIT_KERNEL_LOCAL_SIZE_Y};
-
int d_w = tile.w;
int d_h = tile.h;
diff --git a/intern/cycles/device/device_split_kernel.h
b/intern/cycles/device/device_split_kernel.h
index 643a2fd..619702c 100644
--- a/intern/cycles/device/device_split_kernel.h
+++ b/intern/cycles/device/device_split_kernel.h
@@ -23,11 +23,6 @@
CCL_NAMESPACE_BEGIN
-/* Macro declarations used with split kernel */
-
-#define SPLIT_KERNEL_LOCAL_SIZE_X 64
-#define SPLIT_KERNEL_LOCAL_SIZE_Y 1
-
/* This value may be tuned according to the scene we are rendering.
*
* Modifying PATH_ITER_INC_FACTOR value proportional to number of expected
diff --git a/intern/cycles/device/opencl/opencl_split.cpp
b/intern/cycles/device/opencl/opencl_split.cpp
index 7f50968..2ab78f8 100644
--- a/intern/cycles/device/opencl/opencl_split.cpp
+++ b/intern/cycles/device/opencl/opencl_split.cpp
@@ -257,6 +257,11 @@ public:
return true;
}
+ virtual int2 split_kernel_local_size()
+ {
+ return make_int2(64, 1);
+ }
+
void thread_run(DeviceTask *task)
{
if(task->type == DeviceTask::FILM_CONVERT) {
_______________________________________________
Bf-blender-cvs mailing list
[email protected]
https://lists.blender.org/mailman/listinfo/bf-blender-cvs
