blender
/
blender
129
397
Fork 571
Code Issues 6.2k Pull Requests 752 Projects 19 Wiki Activity

Cycles OpenCL: Remove OpenCL MegaKernel 

Using OpenCL MegaKernel has been slow and therefore not usefull.
This patch will remove the mega kernel from the OpenCL codebase
and the OpenCLDeviceBase class.

T61736: removal of mega kernel
T61703: baking does not work with mega kernel

Tags: #cycles

Differential Revision: https://developer.blender.org/D4383
This commit is contained in:
Jeroen Bakker 2019-02-20 14:41:56 +01:00
parent 667033e89e
commit 949ab753bb
Notes: blender-bot 2023-02-14 00:06:52 +01:00 
Referenced by commit 44c6b6615b, OpenCL: Bring back CYCLES_OPENCL_TEST override
Referenced by issue #73734, Can't force the detection of OpenCL devices
15 changed files with 1676 additions and 2116 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
intern/cycles/blender/addon/ui.py
View File

@ -1634,7 +1634,6 @@ class CYCLES_RENDER_PT_debug(CyclesButtonsPanel, Panel):
col = layout.column()
col.label('OpenCL Flags:')
col.prop(cscene, "debug_opencl_kernel_type", text="Kernel")
col.prop(cscene, "debug_opencl_device_type", text="Device")
col.prop(cscene, "debug_opencl_kernel_single_program", text="Single Program")
col.prop(cscene, "debug_use_opencl_debug", text="Debug")

20
intern/cycles/blender/blender_python.cpp
View File

@ -67,7 +67,6 @@ bool debug_flags_sync_from_scene(BL::Scene b_scene)
PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
/* Backup some settings for comparison. */
DebugFlags::OpenCL::DeviceType opencl_device_type = flags.opencl.device_type;
DebugFlags::OpenCL::KernelType opencl_kernel_type = flags.opencl.kernel_type;
/* Synchronize shared flags. */
flags.viewport_static_bvh = get_enum(cscene, "debug_bvh_type");
/* Synchronize CPU flags. */
@ -81,18 +80,6 @@ bool debug_flags_sync_from_scene(BL::Scene b_scene)
/* Synchronize CUDA flags. */
flags.cuda.adaptive_compile = get_boolean(cscene, "debug_use_cuda_adaptive_compile");
flags.cuda.split_kernel = get_boolean(cscene, "debug_use_cuda_split_kernel");
/* Synchronize OpenCL kernel type. */
switch(get_enum(cscene, "debug_opencl_kernel_type")) {
case 0:
flags.opencl.kernel_type = DebugFlags::OpenCL::KERNEL_DEFAULT;
break;
case 1:
flags.opencl.kernel_type = DebugFlags::OpenCL::KERNEL_MEGA;
break;
case 2:
flags.opencl.kernel_type = DebugFlags::OpenCL::KERNEL_SPLIT;
break;
}
/* Synchronize OpenCL device type. */
switch(get_enum(cscene, "debug_opencl_device_type")) {
case 0:
@ -118,8 +105,7 @@ bool debug_flags_sync_from_scene(BL::Scene b_scene)
flags.opencl.debug = get_boolean(cscene, "debug_use_opencl_debug");
flags.opencl.mem_limit = ((size_t)get_int(cscene, "debug_opencl_mem_limit"))*1024*1024;
flags.opencl.single_program = get_boolean(cscene, "debug_opencl_kernel_single_program");
return flags.opencl.device_type != opencl_device_type ||
flags.opencl.kernel_type != opencl_kernel_type;
return flags.opencl.device_type != opencl_device_type;
}
/* Reset debug flags to default values.
@ -130,10 +116,8 @@ bool debug_flags_reset()
DebugFlagsRef flags = DebugFlags();
/* Backup some settings for comparison. */
DebugFlags::OpenCL::DeviceType opencl_device_type = flags.opencl.device_type;
DebugFlags::OpenCL::KernelType opencl_kernel_type = flags.opencl.kernel_type;
flags.reset();
return flags.opencl.device_type != opencl_device_type ||
flags.opencl.kernel_type != opencl_kernel_type;
return flags.opencl.device_type != opencl_device_type;
}
} /* namespace */

2
intern/cycles/device/CMakeLists.txt
View File

@ -37,8 +37,6 @@ set(SRC_OPENCL
opencl/opencl.h
opencl/memory_manager.h
opencl/opencl_base.cpp
opencl/opencl_mega.cpp
opencl/opencl_split.cpp
opencl/opencl_util.cpp
opencl/memory_manager.cpp

18
intern/cycles/device/device_opencl.cpp
View File

@ -29,19 +29,7 @@ CCL_NAMESPACE_BEGIN
Device *device_opencl_create(DeviceInfo& info, Stats &stats, Profiler &profiler, bool background)
{
vector<OpenCLPlatformDevice> usable_devices;
OpenCLInfo::get_usable_devices(&usable_devices);
assert(info.num < usable_devices.size());
const OpenCLPlatformDevice& platform_device = usable_devices[info.num];
const string& platform_name = platform_device.platform_name;
const cl_device_type device_type = platform_device.device_type;
if(OpenCLInfo::kernel_use_split(platform_name, device_type)) {
VLOG(1) << "Using split kernel.";
return opencl_create_split_device(info, stats, profiler, background);
} else {
VLOG(1) << "Using mega kernel.";
return opencl_create_mega_device(info, stats, profiler, background);
}
return opencl_create_split_device(info, stats, profiler, background);
}
bool device_opencl_init()
@ -111,7 +99,6 @@ void device_opencl_info(vector<DeviceInfo>& devices)
foreach(OpenCLPlatformDevice& platform_device, usable_devices) {
/* Compute unique ID for persistent user preferences. */
const string& platform_name = platform_device.platform_name;
const cl_device_type device_type = platform_device.device_type;
const string& device_name = platform_device.device_name;
string hardware_id = platform_device.hardware_id;
if(hardware_id == "") {
@ -133,8 +120,7 @@ void device_opencl_info(vector<DeviceInfo>& devices)
/* We don't know if it's used for display, but assume it is. */
info.display_device = true;
info.advanced_shading = OpenCLInfo::kernel_use_advanced_shading(platform_name);
info.use_split_kernel = OpenCLInfo::kernel_use_split(platform_name,
device_type);
info.use_split_kernel = true;
info.has_volume_decoupled = false;
info.id = id;

6
intern/cycles/device/opencl/memory_manager.cpp
View File

@ -28,7 +28,7 @@ void MemoryManager::DeviceBuffer::add_allocation(Allocation& allocation)
allocations.push_back(&allocation);
}
void MemoryManager::DeviceBuffer::update_device_memory(OpenCLDeviceBase *device)
void MemoryManager::DeviceBuffer::update_device_memory(OpenCLDevice *device)
{
bool need_realloc = false;
@ -142,7 +142,7 @@ void MemoryManager::DeviceBuffer::update_device_memory(OpenCLDeviceBase *device)
clFinish(device->cqCommandQueue);
}
void MemoryManager::DeviceBuffer::free(OpenCLDeviceBase *)
void MemoryManager::DeviceBuffer::free(OpenCLDevice *)
{
buffer->free();
}
@ -160,7 +160,7 @@ MemoryManager::DeviceBuffer* MemoryManager::smallest_device_buffer()
return smallest;
}
MemoryManager::MemoryManager(OpenCLDeviceBase *device)
MemoryManager::MemoryManager(OpenCLDevice *device)
: device(device), need_update(false)
{
foreach(DeviceBuffer& device_buffer, device_buffers) {

10
intern/cycles/device/opencl/memory_manager.h
View File

@ -26,7 +26,7 @@
CCL_NAMESPACE_BEGIN
class OpenCLDeviceBase;
class OpenCLDevice;
class MemoryManager {
public:
@ -73,12 +73,12 @@ private:
void add_allocation(Allocation& allocation);
void update_device_memory(OpenCLDeviceBase *device);
void update_device_memory(OpenCLDevice *device);
void free(OpenCLDeviceBase *device);
void free(OpenCLDevice *device);
};
OpenCLDeviceBase *device;
OpenCLDevice *device;
DeviceBuffer device_buffers[NUM_DEVICE_BUFFERS];
@ -90,7 +90,7 @@ private:
DeviceBuffer* smallest_device_buffer();
public:
MemoryManager(OpenCLDeviceBase *device);
MemoryManager(OpenCLDevice *device);
void free(); /* Free all memory. */

78
intern/cycles/device/opencl/opencl.h
View File

@ -18,6 +18,7 @@
#include "device/device.h"
#include "device/device_denoising.h"
#include "device/device_split_kernel.h"
#include "util/util_map.h"
#include "util/util_param.h"
@ -84,8 +85,6 @@ public:
static cl_device_type device_type();
static bool use_debug();
static bool kernel_use_advanced_shading(const string& platform_name);
static bool kernel_use_split(const string& platform_name,
const cl_device_type device_type);
static bool device_supported(const string& platform_name,
const cl_device_id device_id);
static bool platform_version_check(cl_platform_id platform,
@ -259,7 +258,7 @@ public:
} \
} (void) 0
class OpenCLDeviceBase : public Device
class OpenCLDevice : public Device
{
public:
DedicatedTaskPool task_pool;
@ -273,7 +272,7 @@ public:
class OpenCLProgram {
public:
OpenCLProgram() : loaded(false), program(NULL), device(NULL) {}
OpenCLProgram(OpenCLDeviceBase *device,
OpenCLProgram(OpenCLDevice *device,
const string& program_name,
const string& kernel_name,
const string& kernel_build_options,
@ -311,7 +310,7 @@ public:
bool loaded;
cl_program program;
OpenCLDeviceBase *device;
OpenCLDevice *device;
/* Used for the OpenCLCache key. */
string program_name;
@ -325,6 +324,32 @@ public:
map<ustring, cl_kernel> kernels;
};
DeviceSplitKernel *split_kernel;
OpenCLProgram program_data_init;
OpenCLProgram program_state_buffer_size;
OpenCLProgram program_split;
OpenCLProgram program_path_init;
OpenCLProgram program_scene_intersect;
OpenCLProgram program_lamp_emission;
OpenCLProgram program_do_volume;
OpenCLProgram program_queue_enqueue;
OpenCLProgram program_indirect_background;
OpenCLProgram program_shader_setup;
OpenCLProgram program_shader_sort;
OpenCLProgram program_shader_eval;
OpenCLProgram program_holdout_emission_blurring_pathtermination_ao;
OpenCLProgram program_subsurface_scatter;
OpenCLProgram program_direct_lighting;
OpenCLProgram program_shadow_blocked_ao;
OpenCLProgram program_shadow_blocked_dl;
OpenCLProgram program_enqueue_inactive;
OpenCLProgram program_next_iteration_setup;
OpenCLProgram program_indirect_subsurface;
OpenCLProgram program_buffer_update;
OpenCLProgram base_program;
OpenCLProgram bake_program;
OpenCLProgram displace_program;
@ -346,8 +371,8 @@ public:
void opencl_error(const string& message);
void opencl_assert_err(cl_int err, const char* where);
OpenCLDeviceBase(DeviceInfo& info, Stats &stats, Profiler &profiler, bool background_);
~OpenCLDeviceBase();
OpenCLDevice(DeviceInfo& info, Stats &stats, Profiler &profiler, bool background_);
~OpenCLDevice();
static void CL_CALLBACK context_notify_callback(const char *err_info,
const void * /*private_info*/, size_t /*cb*/, void *user_data);
@ -355,17 +380,14 @@ public:
bool opencl_version_check();
string device_md5_hash(string kernel_custom_build_options = "");
virtual bool load_kernels(const DeviceRequestedFeatures& requested_features);
/* Has to be implemented by the real device classes.
* The base device will then load all these programs. */
virtual bool add_kernel_programs(const DeviceRequestedFeatures& requested_features,
vector<OpenCLProgram*> &programs) = 0;
bool load_kernels(const DeviceRequestedFeatures& requested_features);
/* Get the name of the opencl program for the given kernel */
virtual const string get_opencl_program_name(bool single_program, const string& kernel_name) = 0;
const string get_opencl_program_name(bool single_program, const string& kernel_name);
/* Get the program file name to compile (*.cl) for the given kernel */
virtual const string get_opencl_program_filename(bool single_program, const string& kernel_name) = 0;
const string get_opencl_program_filename(bool single_program, const string& kernel_name);
string get_build_options(const DeviceRequestedFeatures& requested_features);
string get_build_options_for_bake(const DeviceRequestedFeatures& requested_features);
void mem_alloc(device_memory& mem);
void mem_copy_to(device_memory& mem);
@ -393,10 +415,10 @@ public:
class OpenCLDeviceTask : public DeviceTask {
public:
OpenCLDeviceTask(OpenCLDeviceBase *device, DeviceTask& task)
OpenCLDeviceTask(OpenCLDevice *device, DeviceTask& task)
: DeviceTask(task)
{
run = function_bind(&OpenCLDeviceBase::thread_run,
run = function_bind(&OpenCLDevice::thread_run,
device,
this);
}
@ -422,9 +444,16 @@ public:
task_pool.cancel();
}
virtual void thread_run(DeviceTask * /*task*/) = 0;
void thread_run(DeviceTask *task);
virtual BVHLayoutMask get_bvh_layout_mask() const {
return BVH_LAYOUT_BVH2;
}
virtual bool show_samples() const {
return true;
}
virtual bool is_split_kernel() = 0;
protected:
string kernel_build_options(const string *debug_src = NULL);
@ -566,18 +595,15 @@ protected:
/* ** Those guys are for workign around some compiler-specific bugs ** */
virtual cl_program load_cached_kernel(
cl_program load_cached_kernel(
ustring key,
thread_scoped_lock& cache_locker);
virtual void store_cached_kernel(
void store_cached_kernel(
cl_program program,
ustring key,
thread_scoped_lock& cache_locker);
virtual string build_options_for_bake_program(
const DeviceRequestedFeatures& /*requested_features*/);
private:
MemoryManager memory_manager;
friend class MemoryManager;
@ -592,9 +618,11 @@ private:
protected:
void flush_texture_buffers();
friend class OpenCLSplitKernel;
friend class OpenCLSplitKernelFunction;
};
Device *opencl_create_mega_device(DeviceInfo& info, Stats& stats, Profiler &profiler, bool background);
Device *opencl_create_split_device(DeviceInfo& info, Stats& stats, Profiler &profiler, bool background);
CCL_NAMESPACE_END

1422
intern/cycles/device/opencl/opencl_base.cpp
View File

File diff suppressed because it is too large Load Diff

186
intern/cycles/device/opencl/opencl_mega.cpp
View File

@ -1,186 +0,0 @@
/*
* Copyright 2011-2013 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifdef WITH_OPENCL
#include "device/opencl/opencl.h"
#include "render/buffers.h"
#include "kernel/kernel_types.h"
#include "util/util_md5.h"
#include "util/util_path.h"
#include "util/util_time.h"
CCL_NAMESPACE_BEGIN
class OpenCLDeviceMegaKernel : public OpenCLDeviceBase
{
public:
OpenCLProgram path_trace_program;
OpenCLDeviceMegaKernel(DeviceInfo& info, Stats &stats, Profiler &profiler, bool background_)
: OpenCLDeviceBase(info, stats, profiler, background_),
path_trace_program(this,
get_opencl_program_name(false, "megakernel"),
get_opencl_program_filename(false, "megakernel"),
"-D__COMPILE_ONLY_MEGAKERNEL__ ")
{
}
virtual bool show_samples() const
{
return true;
}
virtual BVHLayoutMask get_bvh_layout_mask() const
{
return BVH_LAYOUT_BVH2;
}
const string get_opencl_program_name(bool /*single_program*/, const string& kernel_name)
{
return kernel_name;
}
const string get_opencl_program_filename(bool /*single_program*/, const string& /*kernel_name*/)
{
return "kernel.cl";
}
virtual bool add_kernel_programs(const DeviceRequestedFeatures& /*requested_features*/,
vector<OpenCLProgram*> &programs)
{
path_trace_program.add_kernel(ustring("path_trace"));
programs.push_back(&path_trace_program);
return true;
}
~OpenCLDeviceMegaKernel()
{
task_pool.stop();
path_trace_program.release();
}
void path_trace(RenderTile& rtile, int sample)
{
scoped_timer timer(&rtile.buffers->render_time);
/* Cast arguments to cl types. */
cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
cl_mem d_buffer = CL_MEM_PTR(rtile.buffer);
cl_int d_x = rtile.x;
cl_int d_y = rtile.y;
cl_int d_w = rtile.w;
cl_int d_h = rtile.h;
cl_int d_offset = rtile.offset;
cl_int d_stride = rtile.stride;
/* Sample arguments. */
cl_int d_sample = sample;
cl_kernel ckPathTraceKernel = path_trace_program(ustring("path_trace"));
cl_uint start_arg_index =
kernel_set_args(ckPathTraceKernel,
0,
d_data,
d_buffer);
set_kernel_arg_buffers(ckPathTraceKernel, &start_arg_index);
start_arg_index += kernel_set_args(ckPathTraceKernel,
start_arg_index,
d_sample,
d_x,
d_y,
d_w,
d_h,
d_offset,
d_stride);
enqueue_kernel(ckPathTraceKernel, d_w, d_h);
}
void thread_run(DeviceTask *task)
{
if(task->type == DeviceTask::FILM_CONVERT) {
film_convert(*task, task->buffer, task->rgba_byte, task->rgba_half);
}
else if(task->type == DeviceTask::SHADER) {
shader(*task);
}
else if(task->type == DeviceTask::RENDER) {
RenderTile tile;
DenoisingTask denoising(this, *task);
/* Keep rendering tiles until done. */
while(task->acquire_tile(this, tile)) {
if(tile.task == RenderTile::PATH_TRACE) {
int start_sample = tile.start_sample;
int end_sample = tile.start_sample + tile.num_samples;
for(int sample = start_sample; sample < end_sample; sample++) {
if(task->get_cancel()) {
if(task->need_finish_queue == false)
break;
}
path_trace(tile, sample);
tile.sample = sample + 1;
task->update_progress(&tile, tile.w*tile.h);
}
/* Complete kernel execution before release tile */
/* This helps in multi-device render;
* The device that reaches the critical-section function
* release_tile waits (stalling other devices from entering
* release_tile) for all kernels to complete. If device1 (a
* slow-render device) reaches release_tile first then it would
* stall device2 (a fast-render device) from proceeding to render
* next tile.
*/
clFinish(cqCommandQueue);
}
else if(tile.task == RenderTile::DENOISE) {
tile.sample = tile.start_sample + tile.num_samples;
denoise(tile, denoising);
task->update_progress(&tile, tile.w*tile.h);
}
task->release_tile(tile);
}
}
}
bool is_split_kernel()
{
return false;
}
};
Device *opencl_create_mega_device(DeviceInfo& info, Stats& stats, Profiler &profiler, bool background)
{
return new OpenCLDeviceMegaKernel(info, stats, profiler, background);
}
CCL_NAMESPACE_END
#endif

1883
intern/cycles/device/opencl/opencl_split.cpp
View File

File diff suppressed because it is too large Load Diff

75
intern/cycles/device/opencl/opencl_util.cpp
View File

@ -243,7 +243,7 @@ string OpenCLCache::get_kernel_md5()
return self.kernel_md5;
}
OpenCLDeviceBase::OpenCLProgram::OpenCLProgram(OpenCLDeviceBase *device,
OpenCLDevice::OpenCLProgram::OpenCLProgram(OpenCLDevice *device,
const string& program_name,
const string& kernel_file,
const string& kernel_build_options,
@ -258,12 +258,12 @@ OpenCLDeviceBase::OpenCLProgram::OpenCLProgram(OpenCLDeviceBase *device,
program = NULL;
}
OpenCLDeviceBase::OpenCLProgram::~OpenCLProgram()
OpenCLDevice::OpenCLProgram::~OpenCLProgram()
{
release();
}
void OpenCLDeviceBase::OpenCLProgram::release()
void OpenCLDevice::OpenCLProgram::release()
{
for(map<ustring, cl_kernel>::iterator kernel = kernels.begin(); kernel != kernels.end(); ++kernel) {
if(kernel->second) {
@ -277,7 +277,7 @@ void OpenCLDeviceBase::OpenCLProgram::release()
}
}
void OpenCLDeviceBase::OpenCLProgram::add_log(const string& msg, bool debug)
void OpenCLDevice::OpenCLProgram::add_log(const string& msg, bool debug)
{
if(!use_stdout) {
log += msg + "\n";
@ -291,7 +291,7 @@ void OpenCLDeviceBase::OpenCLProgram::add_log(const string& msg, bool debug)
}
}
void OpenCLDeviceBase::OpenCLProgram::add_error(const string& msg)
void OpenCLDevice::OpenCLProgram::add_error(const string& msg)
{
if(use_stdout) {
fprintf(stderr, "%s\n", msg.c_str());
@ -302,14 +302,14 @@ void OpenCLDeviceBase::OpenCLProgram::add_error(const string& msg)
error_msg += msg;
}
void OpenCLDeviceBase::OpenCLProgram::add_kernel(ustring name)
void OpenCLDevice::OpenCLProgram::add_kernel(ustring name)
{
if(!kernels.count(name)) {
kernels[name] = NULL;
}
}
bool OpenCLDeviceBase::OpenCLProgram::build_kernel(const string *debug_src)
bool OpenCLDevice::OpenCLProgram::build_kernel(const string *debug_src)
{
string build_options;
build_options = device->kernel_build_options(debug_src) + kernel_build_options;
@ -341,7 +341,7 @@ bool OpenCLDeviceBase::OpenCLProgram::build_kernel(const string *debug_src)
return (ciErr == CL_SUCCESS);
}
bool OpenCLDeviceBase::OpenCLProgram::compile_kernel(const string *debug_src)
bool OpenCLDevice::OpenCLProgram::compile_kernel(const string *debug_src)
{
string source = "#include \"kernel/kernels/opencl/" + kernel_file + "\"\n";
/* We compile kernels consisting of many files. unfortunately OpenCL
@ -389,14 +389,13 @@ static void escape_python_string(string& str)
string_replace(str, "'", "\'");
}
bool OpenCLDeviceBase::OpenCLProgram::compile_separate(const string& clbin)
bool OpenCLDevice::OpenCLProgram::compile_separate(const string& clbin)
{
vector<string> args;
args.push_back("--background");
args.push_back("--factory-startup");
args.push_back("--python-expr");
const char *force_all_platforms = (DebugFlags().opencl.kernel_type != DebugFlags::OpenCL::KERNEL_DEFAULT)? "true" : "false";
int device_platform_id = device->device_num;
string device_name = device->device_name;
string platform_name = device->platform_name;
@ -412,8 +411,7 @@ bool OpenCLDeviceBase::OpenCLProgram::compile_separate(const string& clbin)
args.push_back(
string_printf(
"import _cycles; _cycles.opencl_compile(r'%s', r'%d', r'%s', r'%s', r'%s', r'%s', r'%s')",
force_all_platforms,
"import _cycles; _cycles.opencl_compile(r'%d', r'%s', r'%s', r'%s', r'%s', r'%s')",
device_platform_id,
device_name.c_str(),
platform_name.c_str(),
@ -438,20 +436,19 @@ bool OpenCLDeviceBase::OpenCLProgram::compile_separate(const string& clbin)
* module compile kernels. Parameters must match function above. */
bool device_opencl_compile_kernel(const vector<string>& parameters)
{
bool force_all_platforms = parameters[0] == "true";
int device_platform_id = std::stoi(parameters[1]);
const string& device_name = parameters[2];
const string& platform_name = parameters[3];
const string& build_options = parameters[4];
const string& kernel_file = parameters[5];
const string& binary_path = parameters[6];
int device_platform_id = std::stoi(parameters[0]);
const string& device_name = parameters[1];
const string& platform_name = parameters[2];
const string& build_options = parameters[3];
const string& kernel_file = parameters[4];
const string& binary_path = parameters[5];
if(clewInit() != CLEW_SUCCESS) {
return false;
}
vector<OpenCLPlatformDevice> usable_devices;
OpenCLInfo::get_usable_devices(&usable_devices, force_all_platforms);
OpenCLInfo::get_usable_devices(&usable_devices);
if(device_platform_id >= usable_devices.size()) {
return false;
}
@ -504,7 +501,7 @@ bool device_opencl_compile_kernel(const vector<string>& parameters)
return result;
}
bool OpenCLDeviceBase::OpenCLProgram::load_binary(const string& clbin,
bool OpenCLDevice::OpenCLProgram::load_binary(const string& clbin,
const string *debug_src)
{
/* read binary into memory */
@ -535,7 +532,7 @@ bool OpenCLDeviceBase::OpenCLProgram::load_binary(const string& clbin,
return true;
}
bool OpenCLDeviceBase::OpenCLProgram::save_binary(const string& clbin)
bool OpenCLDevice::OpenCLProgram::save_binary(const string& clbin)
{
size_t size = 0;
clGetProgramInfo(program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t), &size, NULL);
@ -551,7 +548,7 @@ bool OpenCLDeviceBase::OpenCLProgram::save_binary(const string& clbin)
return path_write_binary(clbin, binary);
}
void OpenCLDeviceBase::OpenCLProgram::load()
void OpenCLDevice::OpenCLProgram::load()
{
assert(device);
@ -642,7 +639,7 @@ void OpenCLDeviceBase::OpenCLProgram::load()
loaded = true;
}
void OpenCLDeviceBase::OpenCLProgram::report_error()
void OpenCLDevice::OpenCLProgram::report_error()
{
/* If loaded is true, there was no error. */
if(loaded) return;
@ -656,13 +653,13 @@ void OpenCLDeviceBase::OpenCLProgram::report_error()
}
}
cl_kernel OpenCLDeviceBase::OpenCLProgram::operator()()
cl_kernel OpenCLDevice::OpenCLProgram::operator()()
{
assert(kernels.size() == 1);
return kernels.begin()->second;
}
cl_kernel OpenCLDeviceBase::OpenCLProgram::operator()(ustring name)
cl_kernel OpenCLDevice::OpenCLProgram::operator()(ustring name)
{
assert(kernels.count(name));
return kernels[name];
@ -716,28 +713,6 @@ bool OpenCLInfo::kernel_use_advanced_shading(const string& platform)
return false;
}
bool OpenCLInfo::kernel_use_split(const string& platform_name,
const cl_device_type device_type)
{
if(DebugFlags().opencl.kernel_type == DebugFlags::OpenCL::KERNEL_SPLIT) {
VLOG(1) << "Forcing split kernel to use.";
return true;
}
if(DebugFlags().opencl.kernel_type == DebugFlags::OpenCL::KERNEL_MEGA) {
VLOG(1) << "Forcing mega kernel to use.";
return false;
}
/* TODO(sergey): Replace string lookups with more enum-like API,
* similar to device/vendor checks blender's gpu.
*/
if(platform_name == "AMD Accelerated Parallel Processing" &&
device_type == CL_DEVICE_TYPE_GPU)
{
return true;
}
return false;
}
bool OpenCLInfo::device_supported(const string& platform_name,
const cl_device_id device_id)
{
@ -878,8 +853,6 @@ string OpenCLInfo::get_hardware_id(const string& platform_name, cl_device_id dev
void OpenCLInfo::get_usable_devices(vector<OpenCLPlatformDevice> *usable_devices,
bool force_all)
{
const bool force_all_platforms = force_all ||
(DebugFlags().opencl.kernel_type != DebugFlags::OpenCL::KERNEL_DEFAULT);
const cl_device_type device_type = OpenCLInfo::device_type();
static bool first_time = true;
#define FIRST_VLOG(severity) if(first_time) VLOG(severity)
@ -952,7 +925,7 @@ void OpenCLInfo::get_usable_devices(vector<OpenCLPlatformDevice> *usable_devices
<< " due to old compiler version.";
continue;
}
if(force_all_platforms ||
if(force_all ||
device_supported(platform_name, device_id))
{
cl_device_type device_type;

2
intern/cycles/kernel/CMakeLists.txt
View File

@ -36,8 +36,8 @@ set(SRC_CUDA_KERNELS
)
set(SRC_OPENCL_KERNELS
kernels/opencl/kernel.cl
kernels/opencl/kernel_bake.cl
kernels/opencl/kernel_base.cl
kernels/opencl/kernel_displace.cl
kernels/opencl/kernel_background.cl
kernels/opencl/kernel_state_buffer_size.cl

62
intern/cycles/kernel/kernels/opencl/kernel.cl → intern/cycles/kernel/kernels/opencl/kernel_base.cl
View File

@ -14,72 +14,14 @@
* limitations under the License.
*/
/* OpenCL kernel entry points - unfinished */
/* OpenCL base kernels entry points */
#include "kernel/kernel_compat_opencl.h"
#include "kernel/kernel_math.h"
#include "kernel/kernel_types.h"
#include "kernel/kernel_globals.h"
#include "kernel/kernel_color.h"
#include "kernel/kernels/opencl/kernel_opencl_image.h"
#include "kernel/kernel_film.h"
#if defined(__COMPILE_ONLY_MEGAKERNEL__) || !defined(__NO_BAKING__)
# include "kernel/kernel_path.h"
# include "kernel/kernel_path_branched.h"
#else /* __COMPILE_ONLY_MEGAKERNEL__ */
/* Include only actually used headers for the case
* when path tracing kernels are not needed.
*/
# include "kernel/kernel_random.h"
# include "kernel/kernel_differential.h"
# include "kernel/kernel_montecarlo.h"
# include "kernel/kernel_projection.h"
# include "kernel/geom/geom.h"
# include "kernel/bvh/bvh.h"
# include "kernel/kernel_accumulate.h"
# include "kernel/kernel_camera.h"
# include "kernel/kernel_shader.h"
#endif /* defined(__COMPILE_ONLY_MEGAKERNEL__) || !defined(__NO_BAKING__) */
#include "kernel/kernel_bake.h"
#ifdef __COMPILE_ONLY_MEGAKERNEL__
__kernel void kernel_ocl_path_trace(
ccl_constant KernelData *data,
ccl_global float *buffer,
KERNEL_BUFFER_PARAMS,
int sample,
int sx, int sy, int sw, int sh, int offset, int stride)
{
KernelGlobals kglobals, *kg = &kglobals;
kg->data = data;
kernel_set_buffer_pointers(kg, KERNEL_BUFFER_ARGS);
kernel_set_buffer_info(kg);
int x = sx + ccl_global_id(0);
int y = sy + ccl_global_id(1);
bool thread_is_active = x < sx + sw && y < sy + sh;
if(thread_is_active) {
kernel_path_trace(kg, buffer, sample, x, y, offset, stride);
}
if(kernel_data.film.cryptomatte_passes) {
/* Make sure no thread is writing to the buffers. */
ccl_barrier(CCL_LOCAL_MEM_FENCE);
if(thread_is_active) {
kernel_cryptomatte_post(kg, buffer, sample, x, y, offset, stride);
}
}
}
#else /* __COMPILE_ONLY_MEGAKERNEL__ */
__kernel void kernel_ocl_convert_to_byte(
ccl_constant KernelData *data,
@ -144,5 +86,3 @@ __kernel void kernel_ocl_zero_buffer(ccl_global float4 *buffer, uint64_t size, u
}
}
}
#endif /* __COMPILE_ONLY_MEGAKERNEL__ */

24
intern/cycles/util/util_debug.cpp
View File

@ -90,7 +90,6 @@ void DebugFlags::CUDA::reset()
DebugFlags::OpenCL::OpenCL()
: device_type(DebugFlags::OpenCL::DEVICE_ALL),
kernel_type(DebugFlags::OpenCL::KERNEL_DEFAULT),
debug(false),
single_program(false)
{
@ -122,14 +121,6 @@ void DebugFlags::OpenCL::reset()
device_type = DebugFlags::OpenCL::DEVICE_ACCELERATOR;
}
}
/* Initialize kernel type from environment variables. */
kernel_type = DebugFlags::OpenCL::KERNEL_DEFAULT;
if(getenv("CYCLES_OPENCL_MEGA_KERNEL_TEST") != NULL) {
kernel_type = DebugFlags::OpenCL::KERNEL_MEGA;
}
else if(getenv("CYCLES_OPENCL_SPLIT_KERNEL_TEST") != NULL) {
kernel_type = DebugFlags::OpenCL::KERNEL_SPLIT;
}
/* Initialize other flags from environment variables. */
debug = (getenv("CYCLES_OPENCL_DEBUG") != NULL);
single_program = (getenv("CYCLES_OPENCL_SINGLE_PROGRAM") != NULL);
@ -164,8 +155,7 @@ std::ostream& operator <<(std::ostream &os,
os << "CUDA flags:\n"
<< " Adaptive Compile: " << string_from_bool(debug_flags.cuda.adaptive_compile) << "\n";
const char *opencl_device_type,
*opencl_kernel_type;
const char *opencl_device_type;
switch(debug_flags.opencl.device_type) {
case DebugFlags::OpenCL::DEVICE_NONE:
opencl_device_type = "NONE";
@ -186,20 +176,8 @@ std::ostream& operator <<(std::ostream &os,
opencl_device_type = "ACCELERATOR";
break;
}
switch(debug_flags.opencl.kernel_type) {
case DebugFlags::OpenCL::KERNEL_DEFAULT:
opencl_kernel_type = "DEFAULT";
break;
case DebugFlags::OpenCL::KERNEL_MEGA:
opencl_kernel_type = "MEGA";
break;
case DebugFlags::OpenCL::KERNEL_SPLIT:
opencl_kernel_type = "SPLIT";
break;
}
os << "OpenCL flags:\n"
<< " Device type : " << opencl_device_type << "\n"
<< " Kernel type : " << opencl_kernel_type << "\n"
<< " Debug : " << string_from_bool(debug_flags.opencl.debug) << "\n"
<< " Single program : " << string_from_bool(debug_flags.opencl.single_program) << "\n"
<< " Memory limit : " << string_human_readable_size(debug_flags.opencl.mem_limit) << "\n";

3
intern/cycles/util/util_debug.h
View File

@ -123,9 +123,6 @@ public:
/* Requested device type. */
DeviceType device_type;
/* Requested kernel type. */
KernelType kernel_type;
/* Use debug version of the kernel. */
bool debug;