blender
/
blender
135
403
Fork 586
Code Issues 6.2k Pull Requests 759 Projects 19 Wiki Activity

Cycles: Add single channel texture support for OpenCL. 

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
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
intern/cycles/kernel/kernel_compat_cpu.h
View File

@ -495,6 +495,7 @@ typedef texture<uint> texture_uint;
typedef texture<int> texture_int;
typedef texture<uint4> texture_uint4;
typedef texture<uchar4> texture_uchar4;
typedef texture<uchar> texture_uchar;
typedef texture_image<float> texture_image_float;
typedef texture_image<uchar> texture_image_uchar;
typedef texture_image<half> texture_image_half;

2
intern/cycles/kernel/kernel_textures.h
View File

@ -188,6 +188,8 @@ KERNEL_TEX(uint, texture_uint, __bindless_mapping)
/* packed image (opencl) */
KERNEL_TEX(uchar4, texture_uchar4, __tex_image_byte4_packed)
KERNEL_TEX(float4, texture_float4, __tex_image_float4_packed)
KERNEL_TEX(uchar, texture_uchar, __tex_image_byte_packed)
KERNEL_TEX(float, texture_float, __tex_image_float_packed)
KERNEL_TEX(uint4, texture_uint4, __tex_image_packed_info)
#undef KERNEL_TEX

17
intern/cycles/kernel/svm/svm_image.h
View File

@ -36,13 +36,26 @@ CCL_NAMESPACE_BEGIN
ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int id, int offset)
{
if(id >= TEX_NUM_FLOAT4_IMAGES) {
/* Float4 */
if(id < TEX_START_BYTE4_OPENCL) {
return kernel_tex_fetch(__tex_image_float4_packed, offset);
}
/* Byte4 */
else if(id < TEX_START_FLOAT_OPENCL) {
uchar4 r = kernel_tex_fetch(__tex_image_byte4_packed, offset);
float f = 1.0f/255.0f;
return make_float4(r.x*f, r.y*f, r.z*f, r.w*f);
}
/* Float */
else if(id < TEX_START_BYTE_OPENCL) {
float f = kernel_tex_fetch(__tex_image_float_packed, offset);
return make_float4(f, f, f, 1.0f);
}
/* Byte */
else {
return kernel_tex_fetch(__tex_image_float4_packed, offset);
uchar r = kernel_tex_fetch(__tex_image_byte_packed, offset);
float f = r * (1.0f/255.0f);
return make_float4(f, f, f, 1.0f);
}
}

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();
}

2
intern/cycles/render/scene.h
View File

@ -123,6 +123,8 @@ public:
/* opencl images */
device_vector<uchar4> tex_image_byte4_packed;
device_vector<float4> tex_image_float4_packed;
device_vector<uchar> tex_image_byte_packed;
device_vector<float> tex_image_float_packed;
device_vector<uint4> tex_image_packed_info;
KernelData data;

4
intern/cycles/util/util_texture.h
View File

@ -67,8 +67,8 @@ CCL_NAMESPACE_BEGIN
#define TEX_NUM_FLOAT4_OPENCL 1024
#define TEX_NUM_BYTE4_OPENCL 1024
#define TEX_NUM_HALF4_OPENCL 0
#define TEX_NUM_FLOAT_OPENCL 0
#define TEX_NUM_BYTE_OPENCL 0
#define TEX_NUM_FLOAT_OPENCL 1024
#define TEX_NUM_BYTE_OPENCL 1024
#define TEX_NUM_HALF_OPENCL 0
#define TEX_START_FLOAT4_OPENCL 0
#define TEX_START_BYTE4_OPENCL TEX_NUM_FLOAT4_OPENCL