[Bf-blender-cvs] [d86092c] soc-2016-cycles_images: Don't pass flat_slot to tex_free(), handle this inside the device.
Thomas Dinges
noreply at git.blender.org
Thu May 19 11:42:14 CEST 2016
Commit: d86092cecd893d6c6a84bf686b71b67cea508969
Author: Thomas Dinges
Date: Thu May 19 11:41:45 2016 +0200
Branches: soc-2016-cycles_images
https://developer.blender.org/rBd86092cecd893d6c6a84bf686b71b67cea508969
Don't pass flat_slot to tex_free(), handle this inside the device.
===================================================================
M intern/cycles/device/device.h
M intern/cycles/device/device_cuda.cpp
M intern/cycles/device/device_multi.cpp
M intern/cycles/device/device_network.cpp
M intern/cycles/device/device_opencl.cpp
M intern/cycles/render/image.cpp
===================================================================
diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index 1383a17..2bfcf67 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -233,11 +233,7 @@ public:
(void)flat_slot; /* Ignored. */
};
- virtual void tex_free(device_memory& /*mem*/,
- int flat_slot = -1)
- {
- (void)flat_slot; /* Ignored. */
- };
+ virtual void tex_free(device_memory& /*mem*/) {};
/* pixel memory */
virtual void pixels_alloc(device_memory& mem);
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 145ab97..741dcd5 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -85,6 +85,7 @@ public:
CUcontext cuContext;
CUmodule cuModule;
map<device_ptr, bool> tex_interp_map;
+ map<device_ptr, uint> tex_bindless_map;
int cuDevId;
int cuDevArchitecture;
bool first_error;
@@ -223,7 +224,7 @@ public:
task_pool.stop();
if(info.has_bindless_textures)
- tex_free(bindless_mapping, -1);
+ tex_free(bindless_mapping);
cuda_assert(cuCtxDestroy(cuContext));
}
@@ -405,7 +406,7 @@ public:
void load_bindless_mapping()
{
if(info.has_bindless_textures && sync_bindless_mapping) {
- tex_free(bindless_mapping, -1);
+ tex_free(bindless_mapping);
tex_alloc("__bindless_mapping", bindless_mapping, INTERPOLATION_NONE, EXTENSION_REPEAT, 0);
sync_bindless_mapping = false;
}
@@ -661,6 +662,12 @@ public:
else
cuda_assert(cuMemcpyHtoA(handle, 0, (void*)mem.data_pointer, size));
+ /* Fermi and Kepler */
+ mem.device_pointer = (device_ptr)handle;
+ mem.device_size = size;
+
+ stats.mem_alloc(size);
+
/* Bindless Textures - Kepler */
if(has_bindless_textures) {
CUDA_RESOURCE_DESC resDesc;
@@ -684,6 +691,7 @@ public:
bindless_mapping.get_data()[flat_slot] = (uint)tex;
sync_bindless_mapping = true;
+ tex_bindless_map[mem.device_pointer] = (uint)tex;
}
/* Regular Textures - Fermi */
else {
@@ -693,12 +701,6 @@ public:
}
cuda_pop_context();
-
- /* Fermi and Kepler */
- mem.device_pointer = (device_ptr)handle;
- mem.device_size = size;
-
- stats.mem_alloc(size);
}
/* Fermi, Data and Image Textures */
@@ -720,7 +722,7 @@ public:
tex_interp_map[mem.device_pointer] = (interpolation != INTERPOLATION_NONE);
}
- void tex_free(device_memory& mem, int flat_slot)
+ void tex_free(device_memory& mem)
{
if(mem.device_pointer) {
if(tex_interp_map[mem.device_pointer]) {
@@ -728,6 +730,12 @@ public:
cuArrayDestroy((CUarray)mem.device_pointer);
cuda_pop_context();
+ /* Free CUtexObject (Bindless Textures) */
+ if(info.has_bindless_textures && tex_bindless_map[mem.device_pointer]) {
+ uint flat_slot = tex_bindless_map[mem.device_pointer];
+ cuTexObjectDestroy(flat_slot);
+ }
+
tex_interp_map.erase(tex_interp_map.find(mem.device_pointer));
mem.device_pointer = 0;
@@ -739,11 +747,6 @@ public:
mem_free(mem);
}
}
-
- /* Free CUtexObject (Bindless Textures) */
- if(info.has_bindless_textures && flat_slot != -1) {
- cuTexObjectDestroy(bindless_mapping.get_data()[flat_slot]);
- }
}
void path_trace(RenderTile& rtile, int sample, bool branched)
diff --git a/intern/cycles/device/device_multi.cpp b/intern/cycles/device/device_multi.cpp
index 472d7fa..34a97db 100644
--- a/intern/cycles/device/device_multi.cpp
+++ b/intern/cycles/device/device_multi.cpp
@@ -187,13 +187,13 @@ public:
mem.device_pointer = unique_ptr++;
}
- void tex_free(device_memory& mem, int flat_slot)
+ void tex_free(device_memory& mem)
{
device_ptr tmp = mem.device_pointer;
foreach(SubDevice& sub, devices) {
mem.device_pointer = sub.ptr_map[tmp];
- sub.device->tex_free(mem, flat_slot);
+ sub.device->tex_free(mem);
sub.ptr_map.erase(sub.ptr_map.find(tmp));
}
diff --git a/intern/cycles/device/device_network.cpp b/intern/cycles/device/device_network.cpp
index 8b7ee7c..449f543 100644
--- a/intern/cycles/device/device_network.cpp
+++ b/intern/cycles/device/device_network.cpp
@@ -188,7 +188,7 @@ public:
snd.write_buffer((void*)mem.data_pointer, mem.memory_size());
}
- void tex_free(device_memory& mem, int flat_slot)
+ void tex_free(device_memory& mem)
{
if(mem.device_pointer) {
thread_scoped_lock lock(rpc_lock);
@@ -196,7 +196,6 @@ public:
RPCSend snd(socket, &error_func, "tex_free");
snd.add(mem);
- snd.add(flat_slot);
snd.write();
mem.device_pointer = 0;
diff --git a/intern/cycles/device/device_opencl.cpp b/intern/cycles/device/device_opencl.cpp
index 6a0399b..61f83f2 100644
--- a/intern/cycles/device/device_opencl.cpp
+++ b/intern/cycles/device/device_opencl.cpp
@@ -1195,7 +1195,7 @@ public:
mem_map.insert(MemMap::value_type(name, mem.device_pointer));
}
- void tex_free(device_memory& mem, int /*flat_slot*/)
+ void tex_free(device_memory& mem)
{
if(mem.device_pointer) {
foreach(const MemMap::value_type& value, mem_map) {
diff --git a/intern/cycles/render/image.cpp b/intern/cycles/render/image.cpp
index 0c60d96..d8fb5ba 100644
--- a/intern/cycles/render/image.cpp
+++ b/intern/cycles/render/image.cpp
@@ -771,7 +771,7 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
if(!file_load_float4_image(img, tex_img)) {
@@ -798,7 +798,7 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
if(!file_load_float_image(img, tex_img)) {
@@ -822,7 +822,7 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
if(!file_load_byte4_image(img, tex_img)) {
@@ -849,7 +849,7 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
if(!file_load_byte_image(img, tex_img)) {
@@ -890,7 +890,7 @@ void ImageManager::device_free_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
tex_img.clear();
@@ -900,7 +900,7 @@ void ImageManager::device_free_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
tex_img.clear();
@@ -910,7 +910,7 @@ void ImageManager::device_free_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
tex_img.clear();
@@ -920,7 +920,7 @@ void ImageManager::device_free_image(Device *device, DeviceScene *dscene, ImageD
if(tex_img.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
- device->tex_free(tex_img, flat_slot);
+ device->tex_free(tex_img);
}
tex_img.clear();
More information about the Bf-blender-cvs
mailing list