diff options
author | Brecht Van Lommel <brechtvanlommel@gmail.com> | 2017-10-06 22:47:41 +0300 |
---|---|---|
committer | Brecht Van Lommel <brechtvanlommel@gmail.com> | 2017-10-07 15:53:14 +0300 |
commit | 23098cda9936d785988b689ee69e58e900f17cb2 (patch) | |
tree | ed49843e81afbe9c38707324f37bf7e14b234a9b /intern/cycles/device/device_cpu.cpp | |
parent | d013b56dde47580d1907e3a994bc49cfaaa9f90c (diff) |
Code refactor: make texture code more consistent between devices.
* Use common TextureInfo struct for all devices, except CUDA fermi.
* Move image sampling code to kernels/*/kernel_*_image.h files.
* Use arrays for data textures on Fermi too, so device_vector<Struct> works.
Diffstat (limited to 'intern/cycles/device/device_cpu.cpp')
-rw-r--r-- | intern/cycles/device/device_cpu.cpp | 68 |
1 files changed, 60 insertions, 8 deletions
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp index 19e3c0a9075..ac6d3246d38 100644 --- a/intern/cycles/device/device_cpu.cpp +++ b/intern/cycles/device/device_cpu.cpp @@ -163,6 +163,9 @@ public: TaskPool task_pool; KernelGlobals kernel_globals; + device_vector<TextureInfo> texture_info; + bool need_texture_info; + #ifdef WITH_OSL OSLGlobals osl_globals; #endif @@ -235,6 +238,8 @@ public: VLOG(1) << "Will be using split kernel."; } + need_texture_info = false; + #define REGISTER_SPLIT_KERNEL(name) split_kernels[#name] = KernelFunctions<void(*)(KernelGlobals*, KernelData*)>(KERNEL_FUNCTIONS(name)) REGISTER_SPLIT_KERNEL(path_init); REGISTER_SPLIT_KERNEL(scene_intersect); @@ -261,6 +266,7 @@ public: ~CPUDevice() { task_pool.stop(); + tex_free(texture_info); } virtual bool show_samples() const @@ -268,6 +274,15 @@ public: return (TaskScheduler::num_threads() == 1); } + void load_texture_info() + { + if(need_texture_info) { + tex_free(texture_info); + tex_alloc("__texture_info", texture_info, INTERPOLATION_NONE, EXTENSION_REPEAT); + need_texture_info = false; + } + } + void mem_alloc(const char *name, device_memory& mem, MemoryType /*type*/) { if(name) { @@ -333,14 +348,47 @@ public: VLOG(1) << "Texture allocate: " << name << ", " << string_human_readable_number(mem.memory_size()) << " bytes. (" << string_human_readable_size(mem.memory_size()) << ")"; - kernel_tex_copy(&kernel_globals, - name, - mem.data_pointer, - mem.data_width, - mem.data_height, - mem.data_depth, - interpolation, - extension); + + if(interpolation == INTERPOLATION_NONE) { + /* Data texture. */ + kernel_tex_copy(&kernel_globals, + name, + mem.data_pointer, + mem.data_width, + mem.data_height, + mem.data_depth, + interpolation, + extension); + } + else { + /* Image Texture. */ + int flat_slot = 0; + if(string_startswith(name, "__tex_image")) { + int pos = string(name).rfind("_"); + flat_slot = atoi(name + pos + 1); + } + else { + assert(0); + } + + if(flat_slot >= texture_info.size()) { + /* Allocate some slots in advance, to reduce amount + * of re-allocations. */ + texture_info.resize(flat_slot + 128); + } + + TextureInfo& info = texture_info.get_data()[flat_slot]; + info.data = (uint64_t)mem.data_pointer; + info.cl_buffer = 0; + info.interpolation = interpolation; + info.extension = extension; + info.width = mem.data_width; + info.height = mem.data_height; + info.depth = mem.data_depth; + + need_texture_info = true; + } + mem.device_pointer = mem.data_pointer; mem.device_size = mem.memory_size(); stats.mem_alloc(mem.device_size); @@ -352,6 +400,7 @@ public: mem.device_pointer = 0; stats.mem_free(mem.device_size); mem.device_size = 0; + need_texture_info = true; } } @@ -784,6 +833,9 @@ public: void task_add(DeviceTask& task) { + /* Load texture info. */ + load_texture_info(); + /* split task into smaller ones */ list<DeviceTask> tasks; |