7245262de8
The GPU implementation is a bit too complex to implement for now. As we are improving shader loading, having the CPU timings is already helpful. Note that `Map<size_t, int>` does not compile on Clang. This is exposing the `--profile-gpu` option on all backends as the vulkan backend should follow shortly. Pull Request: https://projects.blender.org/blender/blender/pulls/139551
93 lines
2.3 KiB
C++
93 lines
2.3 KiB
C++
/* SPDX-FileCopyrightText: 2025 Blender Authors
|
|
*
|
|
* SPDX-License-Identifier: GPL-2.0-or-later */
|
|
|
|
#pragma once
|
|
|
|
#include "BLI_map.hh"
|
|
#include "BLI_mutex.hh"
|
|
#include "BLI_string_ref.hh"
|
|
|
|
#include <fmt/format.h>
|
|
#include <fstream>
|
|
#include <iostream>
|
|
#include <mutex>
|
|
#include <sstream>
|
|
#include <thread>
|
|
|
|
namespace blender::gpu {
|
|
|
|
class ProfileReport {
|
|
private:
|
|
std::fstream _report;
|
|
Mutex _mutex;
|
|
Map<uint64_t, int> _thread_ids;
|
|
|
|
ProfileReport()
|
|
{
|
|
_report.open("profile.json", std::ios::out);
|
|
_report << R"([{"name":"process_name","ph":"M","pid":1,"args":{"name":"GPU"}})"
|
|
",\n";
|
|
_report << R"({"name":"process_name","ph":"M","pid":2,"args":{"name":"CPU"}})";
|
|
}
|
|
|
|
~ProfileReport()
|
|
{
|
|
_report << "\n]\n";
|
|
_report.close();
|
|
}
|
|
|
|
public:
|
|
static ProfileReport &get()
|
|
{
|
|
static ProfileReport singleton;
|
|
return singleton;
|
|
}
|
|
|
|
void add_group(StringRefNull name,
|
|
uint64_t gpu_start,
|
|
uint64_t gpu_end,
|
|
uint64_t cpu_start,
|
|
uint64_t cpu_end)
|
|
{
|
|
std::scoped_lock lock(_mutex);
|
|
|
|
size_t thread_hash = std::hash<std::thread::id>()(std::this_thread::get_id());
|
|
int thread_id = _thread_ids.lookup_or_add(thread_hash, _thread_ids.size());
|
|
|
|
_report << fmt::format(
|
|
",\n"
|
|
R"({{"name":"{}","ph":"X","ts":{},"dur":{},"pid":1,"tid":{}}})",
|
|
name.c_str(),
|
|
gpu_start / uint64_t(1000),
|
|
(gpu_end - gpu_start) / uint64_t(1000),
|
|
thread_id);
|
|
|
|
_report << fmt::format(
|
|
",\n"
|
|
R"({{"name":"{}","ph":"X","ts":{},"dur":{},"pid":2,"tid":{}}})",
|
|
name.c_str(),
|
|
cpu_start / uint64_t(1000),
|
|
(cpu_end - cpu_start) / uint64_t(1000),
|
|
thread_id);
|
|
}
|
|
|
|
void add_group_cpu(StringRefNull name, uint64_t cpu_start, uint64_t cpu_end)
|
|
{
|
|
std::scoped_lock lock(_mutex);
|
|
|
|
size_t thread_hash = std::hash<std::thread::id>()(std::this_thread::get_id());
|
|
int thread_id = _thread_ids.lookup_or_add(thread_hash, _thread_ids.size());
|
|
|
|
_report << fmt::format(
|
|
",\n"
|
|
R"({{"name":"{}","ph":"X","ts":{},"dur":{},"pid":2,"tid":{}}})",
|
|
name.c_str(),
|
|
cpu_start / uint64_t(1000),
|
|
(cpu_end - cpu_start) / uint64_t(1000),
|
|
thread_id);
|
|
}
|
|
};
|
|
|
|
} // namespace blender::gpu
|