Cr8-xyz/blender
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: Add single channel texture support for OpenCL.

Browse Source 
This way OpenCL devices can also benefit from a smaller memory footprint, when using e.g. bumpmaps (greyscale, 1 channel).

Additional target for my GSoC 2016.
This commit is contained in:
Thomas Dinges
2016-08-14 20:21:08 +02:00
parent da77d9873f
commit 5c0a67b325
6 changed files with 105 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

90
intern/cycles/render/image.cpp
View File

@@ -284,7 +284,7 @@ int ImageManager::add_image(const string& filename,
if(type == IMAGE_DATA_TYPE_FLOAT || type == IMAGE_DATA_TYPE_FLOAT4)
is_float = true;
/* No single channel and half textures on CUDA (Fermi) and OpenCL, use available slots */
/* No single channel and half textures on CUDA (Fermi) and no half on OpenCL, use available slots */
if((type == IMAGE_DATA_TYPE_FLOAT ||
type == IMAGE_DATA_TYPE_HALF4 ||
type == IMAGE_DATA_TYPE_HALF) &&
@@ -1105,10 +1105,11 @@ void ImageManager::device_pack_images(Device *device,
size_t size = 0, offset = 0;
ImageDataType type;
int info_size = tex_num_images[IMAGE_DATA_TYPE_FLOAT4] + tex_num_images[IMAGE_DATA_TYPE_BYTE4];
int info_size = tex_num_images[IMAGE_DATA_TYPE_FLOAT4] + tex_num_images[IMAGE_DATA_TYPE_BYTE4]
+ tex_num_images[IMAGE_DATA_TYPE_FLOAT] + tex_num_images[IMAGE_DATA_TYPE_BYTE];
uint4 *info = dscene->tex_image_packed_info.resize(info_size);
/* Byte Textures*/
/* Byte4 Textures*/
type = IMAGE_DATA_TYPE_BYTE4;
for(size_t slot = 0; slot < images[type].size(); slot++) {
@@ -1119,7 +1120,7 @@ void ImageManager::device_pack_images(Device *device,
size += tex_img.size();
}
uchar4 *pixels_byte = dscene->tex_image_byte4_packed.resize(size);
uchar4 *pixels_byte4 = dscene->tex_image_byte4_packed.resize(size);
for(size_t slot = 0; slot < images[type].size(); slot++) {
if(!images[type][slot])
@@ -1131,11 +1132,11 @@ void ImageManager::device_pack_images(Device *device,
info[type_index_to_flattened_slot(slot, type)] = make_uint4(tex_img.data_width, tex_img.data_height, offset, options);
memcpy(pixels_byte+offset, (void*)tex_img.data_pointer, tex_img.memory_size());
memcpy(pixels_byte4+offset, (void*)tex_img.data_pointer, tex_img.memory_size());
offset += tex_img.size();
}
/* Float Textures*/
/* Float4 Textures*/
type = IMAGE_DATA_TYPE_FLOAT4;
size = 0, offset = 0;
@@ -1147,7 +1148,7 @@ void ImageManager::device_pack_images(Device *device,
size += tex_img.size();
}
float4 *pixels_float = dscene->tex_image_float4_packed.resize(size);
float4 *pixels_float4 = dscene->tex_image_float4_packed.resize(size);
for(size_t slot = 0; slot < images[type].size(); slot++) {
if(!images[type][slot])
@@ -1160,6 +1161,63 @@ void ImageManager::device_pack_images(Device *device,
uint8_t options = pack_image_options(type, slot);
info[type_index_to_flattened_slot(slot, type)] = make_uint4(tex_img.data_width, tex_img.data_height, offset, options);
memcpy(pixels_float4+offset, (void*)tex_img.data_pointer, tex_img.memory_size());
offset += tex_img.size();
}
/* Byte Textures*/
type = IMAGE_DATA_TYPE_BYTE;
size = 0, offset = 0;
for(size_t slot = 0; slot < images[type].size(); slot++) {
if(!images[type][slot])
continue;
device_vector<uchar>& tex_img = dscene->tex_byte_image[slot];
size += tex_img.size();
}
uchar *pixels_byte = dscene->tex_image_byte_packed.resize(size);
for(size_t slot = 0; slot < images[type].size(); slot++) {
if(!images[type][slot])
continue;
device_vector<uchar>& tex_img = dscene->tex_byte_image[slot];
uint8_t options = pack_image_options(type, slot);
info[type_index_to_flattened_slot(slot, type)] = make_uint4(tex_img.data_width, tex_img.data_height, offset, options);
memcpy(pixels_byte+offset, (void*)tex_img.data_pointer, tex_img.memory_size());
offset += tex_img.size();
}
/* Float Textures*/
type = IMAGE_DATA_TYPE_FLOAT;
size = 0, offset = 0;
for(size_t slot = 0; slot < images[type].size(); slot++) {
if(!images[type][slot])
continue;
device_vector<float>& tex_img = dscene->tex_float_image[slot];
size += tex_img.size();
}
float *pixels_float = dscene->tex_image_float_packed.resize(size);
for(size_t slot = 0; slot < images[type].size(); slot++) {
if(!images[type][slot])
continue;
device_vector<float>& tex_img = dscene->tex_float_image[slot];
/* todo: support 3D textures, only CPU for now */
uint8_t options = pack_image_options(type, slot);
info[type_index_to_flattened_slot(slot, type)] = make_uint4(tex_img.data_width, tex_img.data_height, offset, options);
memcpy(pixels_float+offset, (void*)tex_img.data_pointer, tex_img.memory_size());
offset += tex_img.size();
}
@@ -1178,6 +1236,20 @@ void ImageManager::device_pack_images(Device *device,
}
device->tex_alloc("__tex_image_float4_packed", dscene->tex_image_float4_packed);
}
if(dscene->tex_image_byte_packed.size()) {
if(dscene->tex_image_byte_packed.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
device->tex_free(dscene->tex_image_byte_packed);
}
device->tex_alloc("__tex_image_byte_packed", dscene->tex_image_byte_packed);
}
if(dscene->tex_image_float_packed.size()) {
if(dscene->tex_image_float_packed.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
device->tex_free(dscene->tex_image_float_packed);
}
device->tex_alloc("__tex_image_float_packed", dscene->tex_image_float_packed);
}
if(dscene->tex_image_packed_info.size()) {
if(dscene->tex_image_packed_info.device_pointer) {
thread_scoped_lock device_lock(device_mutex);
@@ -1208,10 +1280,14 @@ void ImageManager::device_free(Device *device, DeviceScene *dscene)
device->tex_free(dscene->tex_image_byte4_packed);
device->tex_free(dscene->tex_image_float4_packed);
device->tex_free(dscene->tex_image_byte_packed);
device->tex_free(dscene->tex_image_float_packed);
device->tex_free(dscene->tex_image_packed_info);
dscene->tex_image_byte4_packed.clear();
dscene->tex_image_float4_packed.clear();
dscene->tex_image_byte_packed.clear();
dscene->tex_image_float_packed.clear();
dscene->tex_image_packed_info.clear();
}