[Bf-blender-cvs] [a1aad1f8d15] master: Fix T53134: denoising with CPU + GPU render leaves some tiles noisy.

[Brecht Van Lommel]

Commit: a1aad1f8d15ecdad4a22c34a3708ba6f8c815079
Author: Brecht Van Lommel
Date:   Tue Oct 24 04:07:24 2017 +0200
Branches: master
https://developer.blender.org/rBa1aad1f8d15ecdad4a22c34a3708ba6f8c815079

Fix T53134: denoising with CPU + GPU render leaves some tiles noisy.

===================================================================

M	intern/cycles/device/device_multi.cpp
M	intern/cycles/render/buffers.cpp
M	intern/cycles/render/buffers.h

===================================================================

diff --git a/intern/cycles/device/device_multi.cpp b/intern/cycles/device/device_multi.cpp
index 0a6dd90c86d..ed2e46965ac 100644
--- a/intern/cycles/device/device_multi.cpp
+++ b/intern/cycles/device/device_multi.cpp
@@ -248,26 +248,36 @@ public:
 			if(!tiles[i].buffers) {
 				continue;
 			}
+
 			/* If the tile was rendered on another device, copy its memory to
 			 * to the current device now, for the duration of the denoising task.
 			 * Note that this temporarily modifies the RenderBuffers and calls
 			 * the device, so this function is not thread safe. */
 			device_vector<float> &mem = tiles[i].buffers->buffer;
 			if(mem.device != sub_device) {
-				tiles[i].buffers->copy_from_device();
+				/* Only copy from device to host once. This is faster, but
+				 * also required for the case where a CPU thread is denoising
+				 * a tile rendered on the GPU. In that case we have to avoid
+				 * overwriting the buffer being denoised by the CPU thread. */
+				if(!tiles[i].buffers->map_neighbor_copied) {
+					tiles[i].buffers->map_neighbor_copied = true;
+					mem.copy_from_device(0, mem.data_size, 1);
+				}
 
 				Device *original_device = mem.device;
 				device_ptr original_ptr = mem.device_pointer;
+				size_t original_size = mem.device_size;
 
 				mem.device = sub_device;
 				mem.device_pointer = 0;
+				mem.device_size = 0;
 
-				sub_device->mem_alloc(mem);
-				sub_device->mem_copy_to(mem);
+				mem.copy_to_device();
 				tiles[i].buffer = mem.device_pointer;
 
 				mem.device = original_device;
 				mem.device_pointer = original_ptr;
+				mem.device_size = original_size;
 			}
 		}
 	}
@@ -290,7 +300,7 @@ public:
 
 				/* Copy denoised tile to the host. */
 				if(i == 4) {
-					tiles[i].buffers->copy_from_device();
+					mem.copy_from_device(0, mem.data_size, 1);
 				}
 
 				sub_device->mem_free(mem);
diff --git a/intern/cycles/render/buffers.cpp b/intern/cycles/render/buffers.cpp
index 01f853dda71..23125f66d66 100644
--- a/intern/cycles/render/buffers.cpp
+++ b/intern/cycles/render/buffers.cpp
@@ -115,7 +115,8 @@ RenderTile::RenderTile()
 /* Render Buffers */
 
 RenderBuffers::RenderBuffers(Device *device)
-: buffer(device, "RenderBuffers", MEM_READ_WRITE)
+: buffer(device, "RenderBuffers", MEM_READ_WRITE),
+  map_neighbor_copied(false)
 {
 }
 
diff --git a/intern/cycles/render/buffers.h b/intern/cycles/render/buffers.h
index 8563d6674ec..9fa0cdd4e27 100644
--- a/intern/cycles/render/buffers.h
+++ b/intern/cycles/render/buffers.h
@@ -74,6 +74,7 @@ public:
 
 	/* float buffer */
 	device_vector<float> buffer;
+	bool map_neighbor_copied;
 
 	explicit RenderBuffers(Device *device);
 	~RenderBuffers();