7dcf95fb57
Cache packet creation in all cases to reduce the number of allocations/
destruction operations made down to KFD. There is a bug that we
encounter after a period of runtime in KFD where allocations fail to be
visable to the GPU (suspect this is a FW issue, similar to other FW
issues they have had along the same lines). This sidesteps that issue in
rocprof (and likely should be done regardless)
Co-authored-by: Benjamin Welton <bewelton@amd.com>
[ROCm/rocprofiler-sdk commit: 7e3ea0c58e]
302 خطوط
8.8 KiB
C++
302 خطوط
8.8 KiB
C++
// MIT License
|
|
//
|
|
// Copyright (c) 2023-2025 Advanced Micro Devices, Inc. All rights reserved.
|
|
//
|
|
// Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
// of this software and associated documentation files (the "Software"), to deal
|
|
// in the Software without restriction, including without limitation the rights
|
|
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
// copies of the Software, and to permit persons to whom the Software is
|
|
// furnished to do so, subject to the following conditions:
|
|
//
|
|
// The above copyright notice and this permission notice shall be included in all
|
|
// copies or substantial portions of the Software.
|
|
//
|
|
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
// SOFTWARE.
|
|
|
|
#pragma once
|
|
|
|
#include "lib/common/defines.hpp"
|
|
#include "lib/common/logging.hpp"
|
|
|
|
#include <sys/syscall.h>
|
|
#include <sys/utsname.h>
|
|
#include <unistd.h>
|
|
#include <atomic>
|
|
#include <chrono>
|
|
#include <condition_variable>
|
|
#include <cstddef>
|
|
#include <cstdint>
|
|
#include <cstring>
|
|
#include <ctime>
|
|
#include <functional>
|
|
#include <mutex>
|
|
#include <ratio>
|
|
#include <stdexcept>
|
|
#include <string>
|
|
#include <thread>
|
|
#include <type_traits>
|
|
#include <vector>
|
|
|
|
namespace rocprofiler
|
|
{
|
|
namespace common
|
|
{
|
|
template <typename... Tp>
|
|
void
|
|
consume_args(Tp&&...)
|
|
{}
|
|
|
|
uint64_t
|
|
get_clock_period_ns_impl(clockid_t _clk_id);
|
|
|
|
inline uint64_t
|
|
get_tid()
|
|
{
|
|
// system calls are expensive so store this in a thread-local
|
|
static thread_local uint64_t _v = ::syscall(__NR_gettid);
|
|
return _v;
|
|
}
|
|
|
|
inline uint64_t
|
|
get_ticks(clockid_t clk_id_v) noexcept
|
|
{
|
|
constexpr auto nanosec = std::nano::den;
|
|
auto&& ts = timespec{};
|
|
auto ret = clock_gettime(clk_id_v, &ts);
|
|
|
|
if(ROCPROFILER_UNLIKELY(ret != 0))
|
|
{
|
|
auto _err = errno;
|
|
ROCP_FATAL << "clock_gettime failed: " << strerror(_err);
|
|
}
|
|
|
|
return (static_cast<uint64_t>(ts.tv_sec) * nanosec) + static_cast<uint64_t>(ts.tv_nsec);
|
|
}
|
|
|
|
static constexpr int default_clock_id = CLOCK_BOOTTIME;
|
|
|
|
// CLOCK_MONOTONIC_RAW equates to HSA-runtime library implementation of os::ReadAccurateClock()
|
|
// CLOCK_BOOTTIME equates to HSA-runtime library implementation of os::ReadSystemClock()
|
|
template <int ClockT = default_clock_id>
|
|
inline uint64_t
|
|
timestamp_ns()
|
|
{
|
|
constexpr auto _clk = ClockT;
|
|
static auto _clk_period = get_clock_period_ns_impl(_clk);
|
|
|
|
if(ROCPROFILER_LIKELY(_clk_period == 1)) return get_ticks(_clk);
|
|
return get_ticks(_clk) / _clk_period;
|
|
}
|
|
|
|
// returns the process start time (in CLOCK_BOOTTIME nanoseconds) via /proc/<pid>/stat
|
|
uint64_t
|
|
get_process_start_time_ns(pid_t _pid);
|
|
|
|
std::vector<std::string>
|
|
read_command_line(pid_t _pid);
|
|
|
|
template <class Container, typename Key = typename Container::key_type>
|
|
const auto*
|
|
get_val(const Container& map, const Key& key)
|
|
{
|
|
auto pos = map.find(key);
|
|
return (pos != map.end() ? &pos->second : nullptr);
|
|
}
|
|
|
|
template <class Container, typename Key = typename Container::key_type>
|
|
auto*
|
|
get_val(Container& map, const Key& key)
|
|
{
|
|
auto pos = map.find(key);
|
|
return (pos != map.end() ? &pos->second : nullptr);
|
|
}
|
|
|
|
template <typename Tp>
|
|
constexpr void
|
|
assert_public_data_type_properties()
|
|
{
|
|
static_assert(std::is_standard_layout<Tp>::value,
|
|
"public data type struct should have a standard layout");
|
|
static_assert(std::is_trivial<Tp>::value, "public data type should be trivial");
|
|
static_assert(std::is_default_constructible<Tp>::value,
|
|
"public data type struct should be default constructible");
|
|
static_assert(std::is_trivially_copy_constructible<Tp>::value,
|
|
"public data type struct should be trivially copy constructible");
|
|
static_assert(std::is_trivially_move_constructible<Tp>::value,
|
|
"public data type struct should be trivially move constructible");
|
|
static_assert(std::is_trivially_copy_assignable<Tp>::value,
|
|
"public data type struct should be trivially move assignable");
|
|
static_assert(std::is_trivially_move_assignable<Tp>::value,
|
|
"public data type struct should be trivially move assignable");
|
|
static_assert(std::is_trivially_copyable<Tp>::value,
|
|
"public data type struct should be trivially move assignable");
|
|
}
|
|
|
|
template <typename Tp>
|
|
constexpr void
|
|
assert_public_api_struct_properties()
|
|
{
|
|
assert_public_data_type_properties<Tp>();
|
|
static_assert(std::is_class<Tp>::value, "this is not a public API struct");
|
|
static_assert(offsetof(Tp, size) == 0, "public API struct should have a size field first");
|
|
static_assert(sizeof(std::declval<Tp>().size) == sizeof(uint64_t),
|
|
"public API struct size field should be 64 bits");
|
|
}
|
|
|
|
// used to set the "size" field to the offset of the "reserved_padding" field.
|
|
// The reserved_padding field is extra unused bytes added to the a struct to
|
|
// avoid an ABI break if/when new fields are added. This is only done
|
|
// for fields which are regularly passed by value
|
|
template <typename Tp, typename Up = Tp>
|
|
constexpr auto
|
|
compute_runtime_sizeof(int) -> decltype(std::declval<Up>().reserved_padding, size_t{})
|
|
{
|
|
return offsetof(Tp, reserved_padding);
|
|
}
|
|
|
|
template <typename Tp, typename Up = Tp>
|
|
constexpr auto
|
|
compute_runtime_sizeof(long)
|
|
{
|
|
return sizeof(Tp);
|
|
}
|
|
|
|
template <typename Tp>
|
|
constexpr auto
|
|
compute_runtime_sizeof()
|
|
{
|
|
return compute_runtime_sizeof<Tp>(0);
|
|
}
|
|
|
|
template <typename Tp, typename... Args>
|
|
decltype(auto)
|
|
init_public_api_struct(Tp&& val, Args&&... args)
|
|
{
|
|
assert_public_api_struct_properties<Tp>();
|
|
|
|
::memset(&val, 0, sizeof(Tp));
|
|
|
|
if constexpr(sizeof...(Args) == 0)
|
|
val.size = compute_runtime_sizeof<Tp>();
|
|
else
|
|
val = {compute_runtime_sizeof<Tp>(), std::forward<Args>(args)...};
|
|
|
|
return std::forward<Tp>(val);
|
|
}
|
|
|
|
template <typename Tp, typename... Args>
|
|
Tp&
|
|
init_public_api_struct(Tp& val, Args&&... args)
|
|
{
|
|
assert_public_api_struct_properties<Tp>();
|
|
|
|
::memset(&val, 0, sizeof(Tp));
|
|
|
|
if constexpr(sizeof...(Args) == 0)
|
|
val.size = compute_runtime_sizeof<Tp>();
|
|
else
|
|
val = {compute_runtime_sizeof<Tp>(), std::forward<Args>(args)...};
|
|
|
|
return val;
|
|
}
|
|
|
|
/**
|
|
* A simple wrapper that will call a function when the
|
|
* wrapper is being destroyed. This is primarily useful
|
|
* for static variables where we want to run some destruction
|
|
* operations when the program exits.
|
|
*/
|
|
template <typename Tp>
|
|
class static_cleanup_wrapper
|
|
{
|
|
public:
|
|
using data_type = Tp;
|
|
using functor_type = std::function<void(Tp&)>;
|
|
|
|
static_cleanup_wrapper(data_type&& data, functor_type&& destroy_func)
|
|
: m_data(std::move(data))
|
|
, m_destroy_func(std::move(destroy_func))
|
|
{}
|
|
|
|
static_cleanup_wrapper(functor_type&& destroy_func)
|
|
: m_destroy_func(std::move(destroy_func))
|
|
{}
|
|
|
|
~static_cleanup_wrapper() { m_destroy_func(m_data); }
|
|
|
|
void destroy() { m_destroy_func(m_data); }
|
|
|
|
data_type& get() { return m_data; }
|
|
const data_type& get() const { return m_data; }
|
|
|
|
private:
|
|
data_type m_data = {};
|
|
functor_type m_destroy_func = {};
|
|
};
|
|
|
|
template <typename Tp = long, typename RatioT = std::ratio<1, 1000>>
|
|
void
|
|
yield(std::chrono::duration<Tp, RatioT> duration = std::chrono::milliseconds{10})
|
|
{
|
|
std::this_thread::yield();
|
|
std::this_thread::sleep_for(duration);
|
|
}
|
|
|
|
template <typename PredicateT, typename Tp = long, typename RatioT = std::ratio<1, 1000>>
|
|
bool
|
|
yield(PredicateT&& predicate,
|
|
std::chrono::duration<Tp, RatioT> max_yield_time,
|
|
std::chrono::duration<Tp, RatioT> query_interval = std::chrono::milliseconds{10})
|
|
{
|
|
auto now = []() { return std::chrono::steady_clock::now(); };
|
|
auto start = now();
|
|
auto result = false;
|
|
while(!(result = predicate()))
|
|
{
|
|
yield(query_interval);
|
|
if((now() - start) > max_yield_time)
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
|
|
// return the result of the last predicate query
|
|
return result;
|
|
}
|
|
|
|
class assert_single_threaded
|
|
{
|
|
public:
|
|
assert_single_threaded(std::atomic<bool>& lock)
|
|
: m_is_initialized(lock)
|
|
{
|
|
bool expected = false;
|
|
if(!m_is_initialized.compare_exchange_strong(expected, true))
|
|
{
|
|
ROCP_FATAL << "This code must be run in a single thread!!!";
|
|
}
|
|
}
|
|
|
|
~assert_single_threaded() { m_is_initialized.store(false, std::memory_order_release); }
|
|
|
|
private:
|
|
std::atomic<bool>& m_is_initialized;
|
|
};
|
|
|
|
} // namespace common
|
|
} // namespace rocprofiler
|
|
|
|
extern "C" {
|
|
void
|
|
rocprofiler_debugger_block();
|
|
void
|
|
rocprofiler_debugger_continue();
|
|
}
|