[rocprofv3] SQLite3 database output (rocpd) support + rocprofiler-sdk-rocpd (#403)
* [rocprofv3] rocpd SQLite3 database output support
* Move counters xml and yaml to source/share/rocprofiler-sdk
- more representative of install hierarchy
* Add share/rocprofiler-sdk/rocpd SQL files
* Experimental rocprofiler-sdk SQL API
* rocprofv3 default output format is rocpd
* Fix rocpd event ids for counter collection w/o kernel dispatch
* Remove fktable entries from rocpd_tables.sql
* Fix rocpd schema path
* Fix install component for roctx python bindings
* rocprofiler-sdk-rocpd
- create include/rocprofiler-sdk-rocpd
- create rocprofiler-sdk-rocpd library, package, etc.
- default all "guid" fields to "{{guid}}" in tables
- remove "{{view_uuid}}" support (always unused)
* Migrate rocprofv3 to use rocprofiler-sdk-rocpd
* Fix missing foreign key reference
* Revert change
* Fix cmake comment
* Fix maybe-uninitialized compiler warning
* Fix maybe-uninitialized compiler warning
* Add logging to rocpd_sql_load_schema
* Improve string sanitization when inserting json strings
* Initialize rocpd logging on rocprofiler-sdk-rocpd library load
* Revert lib/output/generatePerfetto.cpp changes
* [temporary] Tweak rocprofv3-test-list-avail-trace-execute test log level
* Update get_install_path for lib/rocprofiler-sdk-rocpd/sql.cpp
- try to resolve issues on RHEL/SLES for dladdr
* Update lib/common/logging.cpp
- enable environ overrides
* dlsym for rocpd_sql_load_schema
* Make dl_info.dli_fname lexically normal
* Implement node_info alternatives if /etc/machine-id does not exist
* Misc include fixes
* SHA256 and UUIDv7 support
* Implement UUIDv7 in generateRocpd.cpp
* Support push/pop environment variables
* Minor tweak
* Fix glog segfaults when unsetting glog env
* Updated CHANGELOG
* Updates tests/pytest-packages
- rocpd_reader.py: RocpdReader
* Update tests / marker_views.sql
- add test_rocpd_data
* Update rocpd_tables.sql
- Use AUTOINCREMENT
- insert "uuid" and "guid" into rocpd_metadata
* Minor updates to generateRocpd.cpp
- don't quote GUID
- use sqlite3_open_v2
- use sqlite3_close_v2
* Update execute_raw_sql_statements_impl
- uses sqlite3_last_insert_rowid for autoincrement
* Update SQL deferred_transaction
- CI check for nullptr to connection
* Apply suggestions from code review
Co-authored-by: Welton, Benjamin <Benjamin.Welton@amd.com>
* Code review updates
- formatting
- replace if with switch
- remove loop for {{uuid}}
* Fix pmc_groups handling in rocprofv3
* Address code review feedback
- Include rocm_version in rocprofv3 version info
- Note `--version` option for `rocprofv3` in CHANGELOG.md
- remove commented out code
* Fix packaging dependencies
* Fix install package step of CI workflow
* Fix install package step of CI workflow
---------
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
Co-authored-by: Welton, Benjamin <Benjamin.Welton@amd.com>
This commit is contained in:
Madsen, Jonathan
GitHub
@@ -8,9 +8,14 @@ add_subdirectory(common)
|
||||
add_subdirectory(output)
|
||||
add_subdirectory(rocprofiler-sdk)
|
||||
|
||||
set(CMAKE_INSTALL_DEFAULT_COMPONENT_NAME "roctx")
|
||||
add_subdirectory(rocprofiler-sdk-roctx)
|
||||
|
||||
set(CMAKE_INSTALL_DEFAULT_COMPONENT_NAME "rocpd")
|
||||
add_subdirectory(rocprofiler-sdk-rocpd)
|
||||
|
||||
set(CMAKE_INSTALL_DEFAULT_COMPONENT_NAME "tools")
|
||||
add_subdirectory(att-tool)
|
||||
add_subdirectory(rocprofiler-sdk-roctx)
|
||||
add_subdirectory(rocprofiler-sdk-tool)
|
||||
|
||||
add_subdirectory(python)
|
||||
|
||||
@@ -9,11 +9,13 @@ set(common_sources
|
||||
environment.cpp
|
||||
logging.cpp
|
||||
md5sum.cpp
|
||||
sha256.cpp
|
||||
simple_timer.cpp
|
||||
static_object.cpp
|
||||
static_tl_object.cpp
|
||||
string_entry.cpp
|
||||
utility.cpp)
|
||||
utility.cpp
|
||||
uuid_v7.cpp)
|
||||
set(common_headers
|
||||
abi.hpp
|
||||
defines.hpp
|
||||
@@ -26,6 +28,7 @@ set(common_headers
|
||||
md5sum.hpp
|
||||
mpl.hpp
|
||||
scope_destructor.hpp
|
||||
sha256.hpp
|
||||
simple_timer.hpp
|
||||
static_object.hpp
|
||||
static_tl_object.hpp
|
||||
@@ -33,7 +36,8 @@ set(common_headers
|
||||
stringize_arg.hpp
|
||||
synchronized.hpp
|
||||
units.hpp
|
||||
utility.hpp)
|
||||
utility.hpp
|
||||
uuid_v7.hpp)
|
||||
|
||||
add_library(rocprofiler-sdk-common-library STATIC)
|
||||
add_library(rocprofiler-sdk::rocprofiler-sdk-common-library ALIAS
|
||||
|
||||
@@ -151,5 +151,56 @@ SPECIALIZE_SET_ENV(std::string_view)
|
||||
SPECIALIZE_SET_ENV(float)
|
||||
SPECIALIZE_SET_ENV(double)
|
||||
} // namespace impl
|
||||
|
||||
env_store::env_store(std::initializer_list<env_config>&& _container)
|
||||
{
|
||||
for(const auto& itr : _container)
|
||||
{
|
||||
m_original.emplace_back(env_config{itr.env_name, get_env(itr.env_name, ""), 1});
|
||||
m_modified.emplace_back(env_config{itr.env_name, itr.env_value, 1});
|
||||
}
|
||||
}
|
||||
|
||||
env_store::~env_store() { pop(); }
|
||||
|
||||
bool
|
||||
env_store::push()
|
||||
{
|
||||
// not that push ignored bc already pushed
|
||||
if(m_pushed) return false;
|
||||
|
||||
for(const auto& itr : m_modified)
|
||||
itr();
|
||||
|
||||
m_pushed = true;
|
||||
return true;
|
||||
}
|
||||
|
||||
bool
|
||||
env_store::pop(bool unset_if_empty)
|
||||
{
|
||||
if(!m_pushed) return false;
|
||||
|
||||
for(const auto& itr : m_original)
|
||||
{
|
||||
auto _current = get_env(itr.env_name, "");
|
||||
if(!unset_if_empty && itr.env_value.empty())
|
||||
continue;
|
||||
else if(_current == itr.env_value)
|
||||
continue;
|
||||
else if(_current != itr.env_value)
|
||||
{
|
||||
ROCP_INFO << fmt::format("[rocprofiler][env][pop] {}=\"{}\" => {}=\"{}\"",
|
||||
itr.env_name,
|
||||
_current,
|
||||
itr.env_name,
|
||||
itr.env_value);
|
||||
}
|
||||
itr();
|
||||
}
|
||||
|
||||
m_pushed = false;
|
||||
return true;
|
||||
}
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
|
||||
@@ -88,11 +88,49 @@ struct env_config
|
||||
|
||||
auto operator()(bool _verbose = false) const
|
||||
{
|
||||
if(env_name.empty()) return -1;
|
||||
ROCP_INFO_IF(_verbose) << "[rocprofiler][set_env] setenv(\"" << env_name << "\", \""
|
||||
<< env_value << "\", " << overwrite << ")\n";
|
||||
return setenv(env_name.c_str(), env_value.c_str(), overwrite);
|
||||
if(env_name.empty())
|
||||
return -1;
|
||||
else if(_verbose)
|
||||
{
|
||||
ROCP_INFO << "[rocprofiler][set_env] setenv(\"" << env_name << "\", \"" << env_value
|
||||
<< "\", " << overwrite << ")\n";
|
||||
}
|
||||
return (env_value.empty() && overwrite > 0)
|
||||
? unsetenv(env_name.c_str())
|
||||
: setenv(env_name.c_str(), env_value.c_str(), overwrite);
|
||||
}
|
||||
};
|
||||
|
||||
struct env_store
|
||||
{
|
||||
template <template <typename, typename...> class ContainerT, typename... TailT>
|
||||
explicit env_store(ContainerT<env_config, TailT...>&& _container);
|
||||
explicit env_store(std::initializer_list<env_config>&& _container);
|
||||
|
||||
~env_store();
|
||||
env_store(const env_store&) = default;
|
||||
env_store(env_store&&) noexcept = default;
|
||||
env_store& operator=(const env_store&) = default;
|
||||
env_store& operator=(env_store&&) noexcept = default;
|
||||
|
||||
bool push();
|
||||
bool pop(bool unset_if_empty = true);
|
||||
bool is_pushed() const { return m_pushed; }
|
||||
|
||||
private:
|
||||
bool m_pushed = false;
|
||||
std::vector<env_config> m_original = {};
|
||||
std::vector<env_config> m_modified = {};
|
||||
};
|
||||
|
||||
template <template <typename, typename...> class ContainerT, typename... TailT>
|
||||
env_store::env_store(ContainerT<env_config, TailT...>&& _container)
|
||||
{
|
||||
for(const auto& itr : _container)
|
||||
{
|
||||
m_original.emplace_back(env_config{itr.env_name, get_env(itr.env_name, ""), 1});
|
||||
m_modified.emplace_back(env_config{itr.env_name, itr.env_value, 1});
|
||||
}
|
||||
}
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
|
||||
@@ -53,6 +53,34 @@ struct log_level_info
|
||||
int32_t google_level = 0;
|
||||
int32_t verbose_level = 0;
|
||||
};
|
||||
|
||||
env_store
|
||||
get_glog_env_config(const logging_config& cfg)
|
||||
{
|
||||
auto as_env_config = [](std::string_view _var, auto _val) {
|
||||
return env_config{std::string{_var}, fmt::format("{}", _val), 1};
|
||||
};
|
||||
|
||||
auto _data = std::vector<env_config>{
|
||||
as_env_config("GLOG_minloglevel", cfg.loglevel),
|
||||
as_env_config("GLOG_logtostderr", cfg.logtostderr ? 1 : 0),
|
||||
as_env_config("GLOG_alsologtostderr", cfg.alsologtostderr ? 1 : 0),
|
||||
as_env_config("GLOG_stderrthreshold", cfg.loglevel),
|
||||
as_env_config("GLOG_v", cfg.vlog_level),
|
||||
};
|
||||
|
||||
if(!cfg.logdir.empty())
|
||||
{
|
||||
_data.emplace_back(as_env_config("GOOGLE_LOG_DIR", cfg.logdir));
|
||||
_data.emplace_back(as_env_config("GLOG_log_dir", cfg.logdir));
|
||||
}
|
||||
if(!cfg.vlog_modules.empty())
|
||||
{
|
||||
_data.emplace_back(as_env_config("GLOG_vmodule", cfg.vlog_modules));
|
||||
}
|
||||
|
||||
return env_store{std::move(_data)};
|
||||
}
|
||||
} // namespace
|
||||
|
||||
void
|
||||
@@ -139,7 +167,10 @@ init_logging(std::string_view env_prefix, logging_config cfg)
|
||||
}
|
||||
}
|
||||
|
||||
update_logging(cfg, !google::IsGoogleLoggingInitialized());
|
||||
auto _env_store = get_glog_env_config(cfg);
|
||||
|
||||
update_logging(cfg);
|
||||
_env_store.push();
|
||||
|
||||
if(!google::IsGoogleLoggingInitialized())
|
||||
{
|
||||
@@ -158,11 +189,13 @@ init_logging(std::string_view env_prefix, logging_config cfg)
|
||||
|
||||
ROCP_INFO << "logging initialized via " << fmt::format("{}_LOG_LEVEL", env_prefix)
|
||||
<< ". Log Level: " << loglvl << ". Verbose Log Level: " << vlog_level;
|
||||
|
||||
_env_store.pop(false);
|
||||
});
|
||||
}
|
||||
|
||||
void
|
||||
update_logging(const logging_config& cfg, bool setup_env, int env_override)
|
||||
update_logging(const logging_config& cfg)
|
||||
{
|
||||
static auto _mtx = std::mutex{};
|
||||
auto _lk = std::unique_lock<std::mutex>{_mtx};
|
||||
@@ -192,22 +225,6 @@ update_logging(const logging_config& cfg, bool setup_env, int env_override)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if(setup_env)
|
||||
{
|
||||
common::set_env("GLOG_minloglevel", cfg.loglevel, env_override);
|
||||
common::set_env("GLOG_logtostderr", cfg.logtostderr ? 1 : 0, env_override);
|
||||
common::set_env("GLOG_alsologtostderr", cfg.alsologtostderr ? 1 : 0, env_override);
|
||||
common::set_env("GLOG_stderrthreshold", cfg.loglevel, env_override);
|
||||
common::set_env("GLOG_v", cfg.vlog_level, env_override);
|
||||
if(!cfg.logdir.empty())
|
||||
{
|
||||
common::set_env("GOOGLE_LOG_DIR", cfg.logdir, env_override);
|
||||
common::set_env("GLOG_log_dir", cfg.logdir, env_override);
|
||||
}
|
||||
if(!cfg.vlog_modules.empty())
|
||||
common::set_env("GLOG_vmodule", cfg.vlog_modules, env_override);
|
||||
}
|
||||
}
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
|
||||
@@ -79,6 +79,6 @@ void
|
||||
init_logging(std::string_view env_prefix, logging_config cfg = logging_config{});
|
||||
|
||||
void
|
||||
update_logging(const logging_config& cfg, bool setup_env = false, int env_override = 0);
|
||||
update_logging(const logging_config& cfg);
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
|
||||
@@ -0,0 +1,229 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 2025 Advanced Micro Devices, Inc.
|
||||
//
|
||||
// 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.
|
||||
|
||||
#include "lib/common/sha256.hpp"
|
||||
#include "lib/common/defines.hpp"
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/mpl.hpp"
|
||||
|
||||
#include <unistd.h>
|
||||
#include <array>
|
||||
#include <cstdint>
|
||||
#include <cstring>
|
||||
#include <fstream>
|
||||
#include <iomanip>
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
#include <vector>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace common
|
||||
{
|
||||
sha256::sha256() { reset(); }
|
||||
|
||||
sha256::sha256(const std::string& data)
|
||||
{
|
||||
reset();
|
||||
update(data);
|
||||
finalize();
|
||||
}
|
||||
|
||||
void
|
||||
sha256::update(const uint8_t* data, size_t len)
|
||||
{
|
||||
ROCP_CI_LOG_IF(INFO, m_finalized) << "attempt to update sha256 after finalized";
|
||||
if(m_finalized) return;
|
||||
for(size_t i = 0; i < len; ++i)
|
||||
{
|
||||
m_data[m_datalen++] = data[i];
|
||||
if(m_datalen == 64)
|
||||
{
|
||||
transform();
|
||||
m_bitlen += 512;
|
||||
m_datalen = 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
sha256::update(const std::string& data)
|
||||
{
|
||||
ROCP_CI_LOG_IF(INFO, m_finalized) << "attempt to update sha256 after finalized";
|
||||
if(m_finalized) return;
|
||||
update(reinterpret_cast<const uint8_t*>(data.data()), data.size());
|
||||
}
|
||||
|
||||
void
|
||||
sha256::finalize()
|
||||
{
|
||||
if(m_finalized) return;
|
||||
|
||||
uint32_t idx = m_datalen;
|
||||
if(m_datalen < 56)
|
||||
{
|
||||
m_data[idx++] = 0x80;
|
||||
while(idx < 56)
|
||||
m_data[idx++] = 0x00;
|
||||
}
|
||||
else
|
||||
{
|
||||
m_data[idx++] = 0x80;
|
||||
while(idx < 64)
|
||||
m_data[idx++] = 0x00;
|
||||
transform();
|
||||
std::memset(m_data.data(), 0, 56);
|
||||
}
|
||||
|
||||
m_bitlen += m_datalen * 8;
|
||||
for(int j = 0; j < 8; ++j)
|
||||
m_data[63 - j] = static_cast<uint8_t>((m_bitlen >> (8 * j)) & 0xFF);
|
||||
|
||||
transform();
|
||||
|
||||
m_finalized = true;
|
||||
}
|
||||
|
||||
std::string
|
||||
sha256::hexdigest()
|
||||
{
|
||||
finalize();
|
||||
|
||||
auto oss = std::ostringstream{};
|
||||
for(int j = 0; j < 8; ++j)
|
||||
oss << std::hex << std::setfill('0') << std::setw(8) << m_state[j];
|
||||
return oss.str();
|
||||
}
|
||||
|
||||
std::array<uint32_t, 8>
|
||||
sha256::rawdigest()
|
||||
{
|
||||
finalize();
|
||||
|
||||
return m_state;
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::rotr(uint32_t x, uint32_t n)
|
||||
{
|
||||
return (x >> n) | (x << (32 - n));
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::ch(uint32_t x, uint32_t y, uint32_t z)
|
||||
{
|
||||
return (x & y) ^ (~x & z);
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::maj(uint32_t x, uint32_t y, uint32_t z)
|
||||
{
|
||||
return (x & y) ^ (x & z) ^ (y & z);
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::sig0(uint32_t x)
|
||||
{
|
||||
return rotr(x, 2) ^ rotr(x, 13) ^ rotr(x, 22);
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::sig1(uint32_t x)
|
||||
{
|
||||
return rotr(x, 6) ^ rotr(x, 11) ^ rotr(x, 25);
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::theta0(uint32_t x)
|
||||
{
|
||||
return rotr(x, 7) ^ rotr(x, 18) ^ (x >> 3);
|
||||
}
|
||||
|
||||
uint32_t
|
||||
sha256::theta1(uint32_t x)
|
||||
{
|
||||
return rotr(x, 17) ^ rotr(x, 19) ^ (x >> 10);
|
||||
}
|
||||
|
||||
void
|
||||
sha256::transform()
|
||||
{
|
||||
uint32_t m[64];
|
||||
for(int i = 0; i < 16; ++i)
|
||||
{
|
||||
m[i] = (m_data[i * 4] << 24) | (m_data[i * 4 + 1] << 16) | (m_data[i * 4 + 2] << 8) |
|
||||
(m_data[i * 4 + 3]);
|
||||
}
|
||||
for(int i = 16; i < 64; ++i)
|
||||
{
|
||||
m[i] = theta1(m[i - 2]) + m[i - 7] + theta0(m[i - 15]) + m[i - 16];
|
||||
}
|
||||
|
||||
uint32_t a = m_state[0];
|
||||
uint32_t b = m_state[1];
|
||||
uint32_t c = m_state[2];
|
||||
uint32_t d = m_state[3];
|
||||
uint32_t e = m_state[4];
|
||||
uint32_t f = m_state[5];
|
||||
uint32_t g = m_state[6];
|
||||
uint32_t h = m_state[7];
|
||||
|
||||
for(int i = 0; i < 64; ++i)
|
||||
{
|
||||
uint32_t t1 = h + sig1(e) + ch(e, f, g) + m_k[i] + m[i];
|
||||
uint32_t t2 = sig0(a) + maj(a, b, c);
|
||||
h = g;
|
||||
g = f;
|
||||
f = e;
|
||||
e = d + t1;
|
||||
d = c;
|
||||
c = b;
|
||||
b = a;
|
||||
a = t1 + t2;
|
||||
}
|
||||
|
||||
m_state[0] += a;
|
||||
m_state[1] += b;
|
||||
m_state[2] += c;
|
||||
m_state[3] += d;
|
||||
m_state[4] += e;
|
||||
m_state[5] += f;
|
||||
m_state[6] += g;
|
||||
m_state[7] += h;
|
||||
}
|
||||
|
||||
void
|
||||
sha256::reset()
|
||||
{
|
||||
m_state = {0x6a09e667,
|
||||
0xbb67ae85,
|
||||
0x3c6ef372,
|
||||
0xa54ff53a,
|
||||
0x510e527f,
|
||||
0x9b05688c,
|
||||
0x1f83d9ab,
|
||||
0x5be0cd19};
|
||||
m_datalen = 0;
|
||||
m_bitlen = 0;
|
||||
}
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
@@ -0,0 +1,81 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 2025 Advanced Micro Devices, Inc.
|
||||
//
|
||||
// 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/mpl.hpp"
|
||||
|
||||
#include <array>
|
||||
#include <cstdint>
|
||||
#include <string>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace common
|
||||
{
|
||||
// --- SHA-256 Implementation ---
|
||||
class sha256
|
||||
{
|
||||
public:
|
||||
sha256();
|
||||
explicit sha256(const std::string& data);
|
||||
|
||||
void update(const uint8_t* data, size_t len);
|
||||
void update(const std::string& data);
|
||||
|
||||
void finalize();
|
||||
std::string hexdigest();
|
||||
std::array<uint32_t, 8> rawdigest();
|
||||
|
||||
private:
|
||||
bool m_finalized = false;
|
||||
std::array<uint8_t, 64> m_data = {};
|
||||
std::array<uint32_t, 8> m_state = {};
|
||||
uint32_t m_datalen = 0;
|
||||
uint64_t m_bitlen = 0;
|
||||
|
||||
static constexpr std::array<uint32_t, 64> m_k = {
|
||||
0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4,
|
||||
0xab1c5ed5, 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe,
|
||||
0x9bdc06a7, 0xc19bf174, 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f,
|
||||
0x4a7484aa, 0x5cb0a9dc, 0x76f988da, 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7,
|
||||
0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967, 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc,
|
||||
0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85, 0xa2bfe8a1, 0xa81a664b,
|
||||
0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070, 0x19a4c116,
|
||||
0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
|
||||
0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7,
|
||||
0xc67178f2};
|
||||
|
||||
static uint32_t rotr(uint32_t x, uint32_t n);
|
||||
static uint32_t ch(uint32_t x, uint32_t y, uint32_t z);
|
||||
static uint32_t maj(uint32_t x, uint32_t y, uint32_t z);
|
||||
static uint32_t sig0(uint32_t x);
|
||||
static uint32_t sig1(uint32_t x);
|
||||
static uint32_t theta0(uint32_t x);
|
||||
static uint32_t theta1(uint32_t x);
|
||||
|
||||
void transform();
|
||||
void reset();
|
||||
};
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
@@ -0,0 +1,151 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 2025 Advanced Micro Devices, Inc.
|
||||
//
|
||||
// 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.
|
||||
|
||||
#include "lib/common/uuid_v7.hpp"
|
||||
#include "lib/common/defines.hpp"
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/mpl.hpp"
|
||||
#include "lib/common/sha256.hpp"
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <unistd.h>
|
||||
#include <array>
|
||||
#include <cstdint>
|
||||
#include <cstring>
|
||||
#include <fstream>
|
||||
#include <iomanip>
|
||||
#include <random>
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
#include <vector>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace common
|
||||
{
|
||||
uint64_t
|
||||
get_process_start_ticks_since_boot(pid_t pid)
|
||||
{
|
||||
auto line = std::string{};
|
||||
|
||||
// Read the stat file
|
||||
if(auto stat_file = std::ifstream{fmt::format("/proc/{}/stat", pid)}; !stat_file.is_open())
|
||||
{
|
||||
ROCP_CI_LOG(WARNING) << fmt::format("failed to open /proc/{}/stat for process start time",
|
||||
pid);
|
||||
return 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
// Read entire line
|
||||
std::getline(stat_file, line);
|
||||
}
|
||||
|
||||
// Locate the end of the comm field (")")
|
||||
size_t rparen = line.rfind(')');
|
||||
if(rparen == std::string::npos)
|
||||
{
|
||||
ROCP_CI_LOG(WARNING) << fmt::format("Malformed stat file for pid {}", pid);
|
||||
return 0;
|
||||
}
|
||||
|
||||
// Tokenize fields after ") "
|
||||
auto iss = std::istringstream{line.substr(rparen + 2)};
|
||||
auto token = std::string{};
|
||||
// Skip fields 3 through 21
|
||||
for(int i = 0; i < 20; ++i)
|
||||
{
|
||||
if(!(iss >> token))
|
||||
{
|
||||
ROCP_CI_LOG(WARNING) << fmt::format("Unexpected end of /proc/{}/stat", pid);
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
// Field 22: starttime in clock ticks since boot
|
||||
uint64_t start_ticks = 0;
|
||||
if(!(iss >> start_ticks))
|
||||
{
|
||||
ROCP_CI_LOG(WARNING) << fmt::format(
|
||||
"Unexpected end of /proc/{}/stat. Failed to read start ticks", pid);
|
||||
return 0;
|
||||
}
|
||||
|
||||
return start_ticks;
|
||||
}
|
||||
|
||||
uint64_t
|
||||
compute_system_seed(std::string_view machine_id, pid_t pid, pid_t ppid, uint64_t pstart_ticks)
|
||||
{
|
||||
// If no machine_id provided, read from /etc/machine-id
|
||||
ROCP_CI_LOG_IF(WARNING, machine_id.empty())
|
||||
<< fmt::format("compute_system_seed provided empty machine id");
|
||||
|
||||
// Hash for seed value
|
||||
return std::hash<std::string>{}(
|
||||
sha256{fmt::format("{}|{}|{}|{}", machine_id, pid, ppid, pstart_ticks)}.hexdigest());
|
||||
}
|
||||
|
||||
std::string
|
||||
generate_uuid_v7(uint64_t timestamp_ns, uint64_t seed, std::string_view delim)
|
||||
{
|
||||
constexpr auto nanosec_per_millisec = std::nano::den / std::milli::den;
|
||||
auto timestamp_ms = timestamp_ns / nanosec_per_millisec;
|
||||
|
||||
auto uuid = std::array<uint8_t, 16>{};
|
||||
|
||||
// First 6 bytes = timestamp
|
||||
for(int i = 0; i < 6; ++i)
|
||||
{
|
||||
uuid[i] = static_cast<uint8_t>((timestamp_ms >> (40 - 8 * i)) & 0xFF);
|
||||
}
|
||||
|
||||
// Set version to 7 with ordering based on timestamp.
|
||||
uuid[6] = static_cast<uint8_t>((timestamp_ms >> 8) & 0x0F);
|
||||
uuid[6] |= 0x70;
|
||||
|
||||
uuid[7] = static_cast<uint8_t>(timestamp_ms & 0xFF);
|
||||
|
||||
// Seeded RNG
|
||||
auto rand64 = std::mt19937_64{seed}();
|
||||
|
||||
for(int i = 0; i < 8; ++i)
|
||||
{
|
||||
uuid[8 + i] = static_cast<uint8_t>((rand64 >> (56 - 8 * i)) & 0xFF);
|
||||
}
|
||||
|
||||
// Set variant to RFC 4122
|
||||
uuid[8] = (uuid[8] & 0x3F) | 0x80;
|
||||
|
||||
// Format as UUID string
|
||||
auto oss = std::ostringstream{};
|
||||
oss << std::hex << std::setfill('0');
|
||||
for(int i = 0; i < 16; ++i)
|
||||
{
|
||||
oss << std::setw(2) << static_cast<int>(uuid[i]);
|
||||
if(i == 3 || i == 5 || i == 7 || i == 9) oss << delim;
|
||||
}
|
||||
|
||||
return oss.str();
|
||||
}
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
@@ -0,0 +1,49 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 2025 Advanced Micro Devices, Inc.
|
||||
//
|
||||
// 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/mpl.hpp"
|
||||
|
||||
#include <array>
|
||||
#include <cstdint>
|
||||
#include <random>
|
||||
#include <string>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace common
|
||||
{
|
||||
uint64_t
|
||||
get_process_start_ticks_since_boot(pid_t pid);
|
||||
|
||||
// use this function to create a deterministic random number seed for the system and process
|
||||
uint64_t
|
||||
compute_system_seed(std::string_view machine_id, pid_t pid, pid_t ppid, uint64_t pstart_ticks);
|
||||
|
||||
std::string
|
||||
generate_uuid_v7(uint64_t timestamp_ns,
|
||||
uint64_t seed = std::random_device{}(),
|
||||
std::string_view delim = "-");
|
||||
} // namespace common
|
||||
} // namespace rocprofiler
|
||||
@@ -17,6 +17,7 @@ set(TOOL_OUTPUT_HEADERS
|
||||
generateOTF2.hpp
|
||||
generatePerfetto.hpp
|
||||
generateStats.hpp
|
||||
generateRocpd.hpp
|
||||
generator.hpp
|
||||
kernel_symbol_info.hpp
|
||||
host_symbol_info.hpp
|
||||
@@ -41,6 +42,7 @@ set(TOOL_OUTPUT_SOURCES
|
||||
generateOTF2.cpp
|
||||
generatePerfetto.cpp
|
||||
generateStats.cpp
|
||||
generateRocpd.cpp
|
||||
metadata.cpp
|
||||
node_info.cpp
|
||||
output_config.cpp
|
||||
@@ -57,6 +59,7 @@ target_sources(rocprofiler-sdk-output-library PRIVATE ${TOOL_OUTPUT_SOURCES}
|
||||
${TOOL_OUTPUT_HEADERS})
|
||||
target_link_libraries(
|
||||
rocprofiler-sdk-output-library
|
||||
PUBLIC rocprofiler-sdk::rocprofiler-sdk-rocpd-library
|
||||
PRIVATE rocprofiler-sdk::rocprofiler-sdk-headers
|
||||
rocprofiler-sdk::rocprofiler-sdk-build-flags
|
||||
rocprofiler-sdk::rocprofiler-sdk-memcheck
|
||||
@@ -68,3 +71,8 @@ target_link_libraries(
|
||||
rocprofiler-sdk::rocprofiler-sdk-dw
|
||||
rocprofiler-sdk::rocprofiler-sdk-elf
|
||||
rocprofiler-sdk::rocprofiler-sdk-sqlite3)
|
||||
|
||||
target_compile_definitions(rocprofiler-sdk-output-library
|
||||
PRIVATE PROJECT_BINARY_DIR="${PROJECT_BINARY_DIR}")
|
||||
|
||||
add_subdirectory(sql)
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,76 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 2023 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 "agent_info.hpp"
|
||||
#include "generator.hpp"
|
||||
#include "metadata.hpp"
|
||||
#include "output_config.hpp"
|
||||
#include "stream_info.hpp"
|
||||
|
||||
#include <cstdint>
|
||||
#include <deque>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
void
|
||||
write_rocpd(
|
||||
const output_config& cfg,
|
||||
const metadata& tool_metadata,
|
||||
const std::vector<agent_info>& agent_data,
|
||||
const generator<rocprofiler_buffer_tracing_hip_api_ext_record_t>& hip_api_gen,
|
||||
const generator<rocprofiler_buffer_tracing_hsa_api_record_t>& hsa_api_gen,
|
||||
const generator<tool_buffer_tracing_kernel_dispatch_ext_record_t>& kernel_dispatch_gen,
|
||||
const generator<tool_buffer_tracing_memory_copy_ext_record_t>& memory_copy_gen,
|
||||
const generator<rocprofiler_buffer_tracing_marker_api_record_t>& marker_api_gen,
|
||||
const generator<tool_buffer_tracing_memory_allocation_ext_record_t>& memory_alloc_gen,
|
||||
const generator<rocprofiler_buffer_tracing_scratch_memory_record_t>& scratch_memory_gen,
|
||||
const generator<rocprofiler_buffer_tracing_rccl_api_record_t>& rccl_api_gen,
|
||||
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& rocdecode_api_gen,
|
||||
const generator<tool_counter_record_t>& counter_collection_gen);
|
||||
|
||||
// used in schema generation
|
||||
struct argument_info
|
||||
{
|
||||
uint32_t arg_number = 0;
|
||||
std::string arg_type = {};
|
||||
std::string arg_name = {};
|
||||
std::string arg_value = {};
|
||||
};
|
||||
|
||||
struct track_data
|
||||
{
|
||||
uint64_t node_id = 0;
|
||||
pid_t pid = 0;
|
||||
pid_t tid = 0;
|
||||
uint64_t name_id = 0;
|
||||
|
||||
size_t hash() const;
|
||||
};
|
||||
|
||||
bool
|
||||
operator==(const track_data& lhs, const track_data& rhs);
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
@@ -425,6 +425,25 @@ metadata::get_counter_dimension_info() const
|
||||
return _ret;
|
||||
}
|
||||
|
||||
metadata::string_index_map_t
|
||||
metadata::get_string_entries() const
|
||||
{
|
||||
return string_entries.rlock([](const auto& _inp) {
|
||||
auto _sorted = std::vector<std::string_view>{};
|
||||
_sorted.reserve(_inp.size());
|
||||
for(const auto& itr : _inp)
|
||||
_sorted.emplace_back(std::string_view{*itr.second});
|
||||
std::sort(_sorted.begin(), _sorted.end());
|
||||
|
||||
auto _ret = string_index_map_t{};
|
||||
size_t _idx = 1;
|
||||
for(const auto& itr : _sorted)
|
||||
_ret.emplace(itr, _idx++);
|
||||
|
||||
return _ret;
|
||||
});
|
||||
}
|
||||
|
||||
bool
|
||||
metadata::add_marker_message(uint64_t corr_id, std::string&& msg)
|
||||
{
|
||||
@@ -607,17 +626,12 @@ metadata::get_agent_index(rocprofiler_agent_id_t id, agent_indexing index) const
|
||||
return "UNK";
|
||||
};
|
||||
|
||||
switch(index)
|
||||
{
|
||||
case agent_indexing::node: return agent_index{"Agent", _agent->node_id, get_type()};
|
||||
case agent_indexing::logical_node_type:
|
||||
return agent_index{
|
||||
get_type(), static_cast<uint32_t>(_agent->logical_node_type_id), get_type()};
|
||||
|
||||
case agent_indexing::logical_node:
|
||||
default:
|
||||
return agent_index{"Agent", static_cast<uint32_t>(_agent->logical_node_id), get_type()};
|
||||
}
|
||||
return create_agent_index(
|
||||
index,
|
||||
_agent->node_id, // absolute index
|
||||
static_cast<uint32_t>(_agent->logical_node_id), // relative index
|
||||
static_cast<uint32_t>(_agent->logical_node_type_id), // type-relative index
|
||||
get_type());
|
||||
}
|
||||
|
||||
const std::string*
|
||||
@@ -691,5 +705,32 @@ metadata::decode_instruction(rocprofiler_pc_t pc)
|
||||
pc.code_object_offset);
|
||||
}
|
||||
|
||||
agent_index
|
||||
create_agent_index(const rocprofiler::tool::agent_indexing index,
|
||||
uint32_t agent_abs_index,
|
||||
uint32_t agent_log_index,
|
||||
uint32_t agent_type_index,
|
||||
const std::string_view agent_type)
|
||||
{
|
||||
switch(index)
|
||||
{
|
||||
case rocprofiler::tool::agent_indexing::node: // absolute
|
||||
{
|
||||
return agent_index{"Agent", agent_abs_index, agent_type};
|
||||
}
|
||||
case rocprofiler::tool::agent_indexing::logical_node: // relative (default)
|
||||
{
|
||||
return agent_index{"Agent", agent_log_index, agent_type};
|
||||
}
|
||||
case rocprofiler::tool::agent_indexing::logical_node_type: // type-relative
|
||||
{
|
||||
return agent_index{agent_type, agent_type_index, agent_type};
|
||||
}
|
||||
}
|
||||
|
||||
ROCP_CI_LOG(WARNING) << fmt::format(
|
||||
"Unsupported agent indexing {} for agent-{}", static_cast<int>(index), agent_abs_index);
|
||||
return agent_index{};
|
||||
}
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
|
||||
@@ -69,9 +69,26 @@
|
||||
} \
|
||||
}
|
||||
|
||||
#define ROCPD_CHECK_NESTED(VAR, RESULT, LEVEL) \
|
||||
{ \
|
||||
if(rocpd_status_t ROCPROFILER_VARIABLE(CHECKSTATUS, VAR) = RESULT; \
|
||||
ROCPROFILER_VARIABLE(CHECKSTATUS, VAR) != ROCPD_STATUS_SUCCESS) \
|
||||
{ \
|
||||
ROCP_##LEVEL << fmt::format( \
|
||||
"[{}] {} returned {} :: {}", \
|
||||
__FUNCTION__, \
|
||||
#RESULT, \
|
||||
rocpd_get_status_name(ROCPROFILER_VARIABLE(CHECKSTATUS, VAR)), \
|
||||
rocpd_get_status_string(ROCPROFILER_VARIABLE(CHECKSTATUS, VAR))); \
|
||||
} \
|
||||
}
|
||||
|
||||
#define ROCPROFILER_CHECK(RESULT) ROCPROFILER_CHECK_NESTED(__COUNTER__, RESULT, FATAL)
|
||||
#define ROCPROFILER_CHECK_WARNING(RESULT) ROCPROFILER_CHECK_NESTED(__COUNTER__, RESULT, WARNING)
|
||||
|
||||
#define ROCPD_CHECK(RESULT) ROCPD_CHECK_NESTED(__COUNTER__, RESULT, FATAL)
|
||||
#define ROCPD_CHECK_WARNING(RESULT) ROCPD_CHECK_NESTED(__COUNTER__, RESULT, WARNING)
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
@@ -118,6 +135,7 @@ struct agent_index
|
||||
struct metadata
|
||||
{
|
||||
using agent_info_ptr_vec_t = common::container::small_vector<const agent_info*, 16>;
|
||||
using string_index_map_t = std::unordered_map<std::string_view, size_t>;
|
||||
|
||||
struct inprocess
|
||||
{};
|
||||
@@ -147,6 +165,9 @@ struct metadata
|
||||
node_info node_data = {};
|
||||
std::vector<std::string> command_line = {};
|
||||
|
||||
// PMC event ids start at this number
|
||||
uint64_t pmc_event_offset = 1;
|
||||
|
||||
metadata() = default;
|
||||
metadata(inprocess);
|
||||
|
||||
@@ -182,6 +203,8 @@ struct metadata
|
||||
void add_decoder(rocprofiler_code_object_info_t* obj_data_v);
|
||||
code_object_load_info_vec_t get_code_object_load_info() const;
|
||||
|
||||
string_index_map_t get_string_entries() const;
|
||||
|
||||
template <typename Tp>
|
||||
Tp get_marker_messages(Tp&&);
|
||||
|
||||
@@ -232,5 +255,12 @@ metadata::get_marker_messages(Tp&& _inp)
|
||||
},
|
||||
std::move(_inp));
|
||||
}
|
||||
|
||||
agent_index
|
||||
create_agent_index(const agent_indexing index,
|
||||
uint32_t agent_abs_index,
|
||||
uint32_t agent_log_index,
|
||||
uint32_t agent_type_index,
|
||||
const std::string_view agent_type);
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
|
||||
@@ -21,10 +21,13 @@
|
||||
// SOFTWARE.
|
||||
|
||||
#include "lib/output/node_info.hpp"
|
||||
#include "lib/common/filesystem.hpp"
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/sha256.hpp"
|
||||
|
||||
#include <rocprofiler-sdk/cxx/details/tokenize.hpp>
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <sys/utsname.h>
|
||||
|
||||
#include <fstream>
|
||||
@@ -33,19 +36,121 @@ namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
namespace
|
||||
{
|
||||
using utsname_t = struct utsname;
|
||||
|
||||
std::string
|
||||
sha256_hex(const std::string& input)
|
||||
{
|
||||
auto sha = common::sha256{};
|
||||
sha.update(input);
|
||||
return sha.hexdigest();
|
||||
}
|
||||
|
||||
// --- Machine ID Utility ---
|
||||
std::string
|
||||
read_file_first_line(const std::string& path)
|
||||
{
|
||||
if(auto file = std::ifstream{path}; file.is_open())
|
||||
{
|
||||
auto line = std::string{};
|
||||
std::getline(file, line);
|
||||
return line;
|
||||
}
|
||||
return std::string{};
|
||||
}
|
||||
|
||||
std::string
|
||||
get_mac_address(std::string_view iface)
|
||||
{
|
||||
if(auto mac = read_file_first_line(fmt::format("/sys/class/net/{}/address", iface));
|
||||
!mac.empty())
|
||||
return mac;
|
||||
|
||||
return {};
|
||||
}
|
||||
|
||||
std::string
|
||||
read_file(const std::string& filePath)
|
||||
{
|
||||
auto file = std::ifstream{filePath, std::ios::in | std::ios::binary};
|
||||
if(file.is_open())
|
||||
{
|
||||
auto buffer = std::stringstream{};
|
||||
buffer << file.rdbuf();
|
||||
return buffer.str();
|
||||
}
|
||||
return std::string{};
|
||||
}
|
||||
|
||||
std::string
|
||||
get_mac_address(const std::vector<std::string>& interfaces = {"eth0", "enp0s3", "wlan0", "eno1"})
|
||||
{
|
||||
namespace fs = ::rocprofiler::common::filesystem;
|
||||
|
||||
auto remove_duplicates = [](auto _data) {
|
||||
std::sort(_data.begin(), _data.end());
|
||||
_data.erase(std::unique(_data.begin(), _data.end()), _data.end());
|
||||
return _data;
|
||||
};
|
||||
|
||||
for(std::string_view iface : interfaces)
|
||||
{
|
||||
if(auto mac = get_mac_address(iface); !mac.empty()) return mac;
|
||||
}
|
||||
|
||||
for(const auto& itr : fs::directory_iterator{fs::path{"/sys/class/net"}})
|
||||
{
|
||||
if(auto path = fs::path{itr}; fs::exists(path / "address"))
|
||||
{
|
||||
if(auto mac = get_mac_address(path.filename().string()); !mac.empty())
|
||||
{
|
||||
// some network interfaces have generic addresses like 00:00:00:00:00:00 or
|
||||
// ee:ee:ee:ee:ee:ee and we want to ignore these
|
||||
if(remove_duplicates(sdk::parse::tokenize(mac, ":")).size() > 1) return mac;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return std::string{};
|
||||
}
|
||||
|
||||
std::string
|
||||
get_machine_id()
|
||||
{
|
||||
// not all Linux distributions have /etc/machine-id so we need to fallback on various
|
||||
// alternatives to try to uniquely identify the system
|
||||
if(std::string id = read_file_first_line("/etc/machine-id"); !id.empty()) return id;
|
||||
|
||||
if(std::string id = read_file_first_line("/var/lib/dbus/machine-id"); !id.empty()) return id;
|
||||
|
||||
//
|
||||
// for all values beyond this point, encrypt the id with sha256 since this is potentially
|
||||
// sensitive information. prefix is used for salt separation
|
||||
//
|
||||
if(std::string id = read_file_first_line("/sys/class/dmi/id/product_uuid"); !id.empty())
|
||||
return sha256_hex(fmt::format("product_uuid:{}", id));
|
||||
|
||||
if(std::string id = read_file_first_line("/sys/class/dmi/id/board_serial"); !id.empty())
|
||||
return sha256_hex(fmt::format("board_serial:{}", id));
|
||||
|
||||
if(std::string id = read_file("/proc/cpuinfo") + read_file("/proc/version") +
|
||||
read_file("/proc/devices") + read_file("/proc/filesystems");
|
||||
!id.empty())
|
||||
return sha256_hex(fmt::format("procinfo:{}", id));
|
||||
|
||||
if(std::string id = get_mac_address(); !id.empty())
|
||||
return sha256_hex(fmt::format("mac_address:{}", id));
|
||||
|
||||
return std::string{};
|
||||
}
|
||||
} // namespace
|
||||
|
||||
node_info&
|
||||
read_node_info(node_info& _info)
|
||||
{
|
||||
{
|
||||
if(auto ifs = std::ifstream{"/etc/machine-id"})
|
||||
{
|
||||
auto _mach_id = std::string{};
|
||||
if((ifs >> _mach_id) && !_mach_id.empty())
|
||||
_info.machine_id = sdk::parse::strip(std::move(_mach_id), "\n\t\r ");
|
||||
}
|
||||
}
|
||||
_info.machine_id = get_machine_id();
|
||||
|
||||
auto _sys_info = utsname_t{};
|
||||
if(uname(&_sys_info) == 0)
|
||||
|
||||
@@ -22,7 +22,6 @@
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <rocprofiler-sdk/agent.h>
|
||||
#include <rocprofiler-sdk/cxx/serialization.hpp>
|
||||
|
||||
#include <unordered_map>
|
||||
|
||||
@@ -74,10 +74,11 @@ output_config::parse_env()
|
||||
for(const auto& itr : sdk::parse::tokenize(output_format, " \t,;:"))
|
||||
entries.emplace(to_upper(itr));
|
||||
|
||||
csv_output = entries.count("CSV") > 0 || entries.empty();
|
||||
csv_output = entries.count("CSV") > 0;
|
||||
json_output = entries.count("JSON") > 0;
|
||||
pftrace_output = entries.count("PFTRACE") > 0;
|
||||
otf2_output = entries.count("OTF2") > 0;
|
||||
rocpd_output = entries.count("ROCPD") > 0 || entries.empty();
|
||||
|
||||
const auto supported_formats =
|
||||
std::set<std::string_view>{"CSV", "JSON", "PFTRACE", "OTF2", "ROCPD"};
|
||||
|
||||
@@ -67,6 +67,7 @@ struct output_config
|
||||
bool json_output = false;
|
||||
bool pftrace_output = false;
|
||||
bool otf2_output = false;
|
||||
bool rocpd_output = false;
|
||||
bool summary_output = false;
|
||||
bool kernel_rename = false;
|
||||
bool group_by_queue = false;
|
||||
@@ -129,6 +130,7 @@ output_config::save(ArchiveT& ar) const
|
||||
CFG_SERIALIZE_MEMBER(pftrace_output);
|
||||
CFG_SERIALIZE_MEMBER(otf2_output);
|
||||
CFG_SERIALIZE_MEMBER(summary_output);
|
||||
CFG_SERIALIZE_MEMBER(rocpd_output);
|
||||
CFG_SERIALIZE_MEMBER(kernel_rename);
|
||||
CFG_SERIALIZE_MEMBER(group_by_queue);
|
||||
|
||||
|
||||
@@ -0,0 +1,8 @@
|
||||
#
|
||||
# add sql common sources to output library target
|
||||
#
|
||||
set(output_sql_headers common.hpp deferred_transaction.hpp extract_data_type.hpp)
|
||||
set(output_sql_sources common.cpp deferred_transaction.cpp)
|
||||
|
||||
target_sources(rocprofiler-sdk-output-library PRIVATE ${output_sql_sources}
|
||||
${output_sql_headers})
|
||||
@@ -0,0 +1,205 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 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.
|
||||
|
||||
#include "lib/output/sql/common.hpp"
|
||||
#include "lib/output/kernel_symbol_info.hpp"
|
||||
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/scope_destructor.hpp"
|
||||
|
||||
#include <rocprofiler-sdk/cxx/details/tokenize.hpp>
|
||||
#include <rocprofiler-sdk/cxx/hash.hpp>
|
||||
#include <rocprofiler-sdk/cxx/operators.hpp>
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <sqlite3.h>
|
||||
|
||||
#include <iomanip>
|
||||
#include <sstream>
|
||||
#include <thread>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
namespace sql
|
||||
{
|
||||
namespace sdk = ::rocprofiler::sdk;
|
||||
|
||||
void
|
||||
check(std::string_view function, int status, std::string_view stmt)
|
||||
{
|
||||
if(status != SQLITE_OK)
|
||||
{
|
||||
ROCP_FATAL << "[" << function << "] " << stmt << " failed with error code " << status;
|
||||
}
|
||||
}
|
||||
|
||||
int
|
||||
busy_handler(void* /*data*/, int count)
|
||||
{
|
||||
count = (count < 9) ? count : 8;
|
||||
std::this_thread::sleep_for(std::chrono::microseconds{1000 * (0x1 << count)});
|
||||
return 1;
|
||||
}
|
||||
|
||||
// invoked during SELECT operation; unused but kept for reference
|
||||
int
|
||||
exec_callback(void* user_data, int ncols, char** coltext, char** colnames)
|
||||
{
|
||||
ROCP_INFO << "SQL callback invoked with " << ncols << " columns";
|
||||
|
||||
if(!coltext || !colnames) return SQLITE_OK;
|
||||
|
||||
auto header = std::stringstream{};
|
||||
auto div = std::stringstream{};
|
||||
auto content = std::stringstream{};
|
||||
|
||||
header << fmt::format(
|
||||
"| {} |", fmt::join(std::vector<std::string_view>(colnames, colnames + ncols), " | "));
|
||||
div << fmt::format("|-{}-|", std::string(header.str().size() - 4, '-'));
|
||||
content << fmt::format(
|
||||
"| {} |", fmt::join(std::vector<std::string_view>(coltext, coltext + ncols), " | "));
|
||||
|
||||
ROCP_WARNING << "SQL callback for " << ncols << " columns contents: "
|
||||
<< "\n\t" << header.str() << "\n\t" << div.str() << "\n\t" << content.str();
|
||||
|
||||
return SQLITE_OK;
|
||||
|
||||
(void) user_data;
|
||||
}
|
||||
|
||||
int64_t
|
||||
execute_raw_sql_statements_impl(sqlite3* conn,
|
||||
std::string_view stmts,
|
||||
exec_callback_t callback,
|
||||
void* data,
|
||||
int line)
|
||||
{
|
||||
int64_t row_id = -1;
|
||||
// NOLINTNEXTLINE(performance-for-range-copy)
|
||||
for(auto stmt : sdk::parse::tokenize(stmts, ";"))
|
||||
{
|
||||
stmt = fmt::format("{};", std::move(stmt)); // make sure ends with semi-colon
|
||||
|
||||
ROCP_TRACE << "Executing SQLite3 statement: " << stmt;
|
||||
|
||||
char* msg = nullptr;
|
||||
auto ret = sqlite3_exec(conn, stmt.c_str(), callback, data, &msg);
|
||||
if(ret != SQLITE_OK)
|
||||
{
|
||||
// ensure no memory leak
|
||||
auto dtor = common::scope_destructor{[msg]() {
|
||||
if(msg) sqlite3_free(msg);
|
||||
}};
|
||||
|
||||
static constexpr auto full_file = std::string_view{__FILE__};
|
||||
static constexpr auto base_file = full_file.substr(full_file.find_last_of('/') + 1);
|
||||
ROCP_FATAL << "SQLite3 error " << ret << ": " << ((msg) ? msg : "unknown error")
|
||||
<< "\n\tSQLite3 error: " << base_file << ":" << line
|
||||
<< "\n\tSQL Statement: " << stmt;
|
||||
}
|
||||
else
|
||||
{
|
||||
if(stmt.find("INSERT") != std::string::npos) row_id = sqlite3_last_insert_rowid(conn);
|
||||
}
|
||||
}
|
||||
return row_id;
|
||||
}
|
||||
|
||||
int64_t
|
||||
execute_raw_sql_statements_impl(sqlite3* conn, std::string_view stmts, int line)
|
||||
{
|
||||
return execute_raw_sql_statements_impl(conn, stmts, exec_callback, nullptr, line);
|
||||
}
|
||||
|
||||
std::string
|
||||
extract_column_name(sqlite3_stmt* stmt, int32_t col)
|
||||
{
|
||||
return std::string{sqlite3_column_name(stmt, col)};
|
||||
}
|
||||
|
||||
int64_t
|
||||
extract_row_count(sqlite3* conn, std::string_view query)
|
||||
{
|
||||
auto _pos = query.find(';');
|
||||
auto _query = (_pos == std::string_view::npos) ? query : query.substr(0, _pos);
|
||||
auto _count_query = fmt::format("SELECT COUNT(*) AS count FROM ({}) x;", _query);
|
||||
|
||||
sqlite3_stmt* _stmt = nullptr;
|
||||
if(sqlite3_prepare_v2(conn, _count_query.c_str(), -1, &_stmt, nullptr) != SQLITE_OK)
|
||||
{
|
||||
ROCP_CI_LOG(ERROR) << fmt::format(
|
||||
"Error preparing select statement for '{}': {}", _count_query, sqlite3_errmsg(conn));
|
||||
return 0;
|
||||
}
|
||||
|
||||
ROCP_CI_LOG_IF(ERROR, _stmt == nullptr) << "Error preparing statment: " << query;
|
||||
|
||||
int64_t nrows = 0;
|
||||
if(_stmt && sqlite3_column_count(_stmt) == 1 && sqlite3_step(_stmt) == SQLITE_ROW)
|
||||
nrows = extract_column<int64_t>(_stmt, 0).value_or(0);
|
||||
|
||||
sqlite3_finalize(_stmt); // Finalize statement
|
||||
|
||||
ROCP_INFO << fmt::format("SQL query '{}' contains {} rows", query, nrows);
|
||||
return nrows;
|
||||
}
|
||||
|
||||
namespace
|
||||
{
|
||||
template <int Idx>
|
||||
struct sql_data_type_info;
|
||||
|
||||
#define SPECIALIZE_SQL_DATA_TYPE_INFO(VALUE) \
|
||||
template <> \
|
||||
struct sql_data_type_info<VALUE> \
|
||||
{ \
|
||||
static constexpr auto value = VALUE; \
|
||||
static constexpr auto name = #VALUE; \
|
||||
};
|
||||
|
||||
SPECIALIZE_SQL_DATA_TYPE_INFO(SQLITE_INTEGER)
|
||||
SPECIALIZE_SQL_DATA_TYPE_INFO(SQLITE_FLOAT)
|
||||
SPECIALIZE_SQL_DATA_TYPE_INFO(SQLITE_BLOB)
|
||||
SPECIALIZE_SQL_DATA_TYPE_INFO(SQLITE_NULL)
|
||||
SPECIALIZE_SQL_DATA_TYPE_INFO(SQLITE_TEXT)
|
||||
|
||||
template <size_t Idx, size_t... Tail>
|
||||
std::string_view
|
||||
get_sql_data_type(int val, std::index_sequence<Idx, Tail...>)
|
||||
{
|
||||
using info_type = sql_data_type_info<Idx + 1>;
|
||||
if(val == info_type::value) return std::string_view{info_type::name};
|
||||
if constexpr(sizeof...(Tail) > 0) return get_sql_data_type(val, std::index_sequence<Tail...>{});
|
||||
return std::string_view{"SQLITE_UNKNOWN"};
|
||||
}
|
||||
} // namespace
|
||||
|
||||
std::string_view
|
||||
get_sql_data_type(int val)
|
||||
{
|
||||
return get_sql_data_type(val, std::make_index_sequence<SQLITE_NULL>{});
|
||||
}
|
||||
} // namespace sql
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
@@ -0,0 +1,129 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 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/output/sql/extract_data_type.hpp"
|
||||
|
||||
#include "lib/common/container/ring_buffer.hpp"
|
||||
#include "lib/common/mpl.hpp"
|
||||
#include "lib/common/units.hpp"
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <sqlite3.h>
|
||||
|
||||
#include <chrono>
|
||||
#include <cstdint>
|
||||
#include <string_view>
|
||||
#include <type_traits>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
namespace sql
|
||||
{
|
||||
using exec_callback_t = int (*)(void* user_data, int ncols, char** coltext, char** colnames);
|
||||
|
||||
template <typename Tp>
|
||||
using ring_buffer_t = rocprofiler::common::container::ring_buffer<Tp>;
|
||||
|
||||
int
|
||||
exec_callback(void* user_data, int ncols, char** coltext, char** colnames);
|
||||
|
||||
void
|
||||
check(std::string_view function, int status, std::string_view stmt);
|
||||
|
||||
int
|
||||
busy_handler(void* data, int count);
|
||||
|
||||
int64_t
|
||||
execute_raw_sql_statements_impl(sqlite3* conn, std::string_view stmts, int line);
|
||||
|
||||
int64_t
|
||||
execute_raw_sql_statements_impl(sqlite3* conn,
|
||||
std::string_view stmts,
|
||||
exec_callback_t callback,
|
||||
void* data,
|
||||
int line);
|
||||
|
||||
std::string
|
||||
extract_column_name(sqlite3_stmt* stmt, int32_t col);
|
||||
|
||||
int64_t
|
||||
extract_row_count(sqlite3* conn, std::string_view query);
|
||||
|
||||
std::string_view
|
||||
get_sql_data_type(int val);
|
||||
|
||||
template <typename Tp>
|
||||
auto
|
||||
extract_column(sqlite3_stmt* stmt, int32_t col);
|
||||
|
||||
template <typename Tp, int ExpectedV>
|
||||
bool
|
||||
column_data_is_null(sqlite3_stmt* stmt, int32_t col);
|
||||
|
||||
//
|
||||
//
|
||||
// Template function definitions
|
||||
//
|
||||
//
|
||||
template <typename Tp>
|
||||
auto
|
||||
extract_column(sqlite3_stmt* stmt, int32_t col)
|
||||
{
|
||||
return extract_data_type<Tp>{}(stmt, col);
|
||||
}
|
||||
|
||||
template <typename Tp, int ExpectedV>
|
||||
bool
|
||||
column_data_is_null(sqlite3_stmt* stmt, int32_t col)
|
||||
{
|
||||
auto coltype = sqlite3_column_type(stmt, col);
|
||||
if(coltype == SQLITE_NULL) return true;
|
||||
|
||||
std::string column_name = extract_column_name(stmt, col);
|
||||
const char* sql_text = sqlite3_sql(stmt);
|
||||
|
||||
ROCP_CI_LOG_IF(WARNING, coltype != ExpectedV) << fmt::format(
|
||||
"Data in SQL column {} ('{}') is neither NULL nor the expected data type ({} == {}). "
|
||||
"Column data type is: ({} == {}), SQL: {}",
|
||||
col,
|
||||
column_name,
|
||||
ExpectedV,
|
||||
get_sql_data_type(ExpectedV),
|
||||
coltype,
|
||||
get_sql_data_type(coltype),
|
||||
sql_text);
|
||||
|
||||
return false;
|
||||
}
|
||||
} // namespace sql
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
|
||||
#define execute_raw_sql_statements(...) \
|
||||
::rocprofiler::tool::sql::execute_raw_sql_statements_impl(__VA_ARGS__, __LINE__)
|
||||
|
||||
#define SQLITE3_CHECK(RESULT) \
|
||||
::rocprofiler::tool::sql::check(__FUNCTION__, (RESULT), std::string_view{#RESULT})
|
||||
@@ -0,0 +1,62 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 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.
|
||||
|
||||
#include "lib/output/sql/deferred_transaction.hpp"
|
||||
#include "lib/output/sql/common.hpp"
|
||||
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/scope_destructor.hpp"
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <sqlite3.h>
|
||||
|
||||
#include <iomanip>
|
||||
#include <sstream>
|
||||
#include <thread>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
namespace sql
|
||||
{
|
||||
deferred_transaction::deferred_transaction(sqlite3* conn)
|
||||
: m_conn{conn}
|
||||
{
|
||||
ROCP_CI_LOG_IF(INFO, m_conn == nullptr) << "rocprofiler::tool::sql::deferred_transaction "
|
||||
"constructed will nullptr to sqlite3 connection";
|
||||
if(m_conn)
|
||||
{
|
||||
execute_raw_sql_statements(m_conn, "BEGIN DEFERRED TRANSACTION");
|
||||
}
|
||||
}
|
||||
|
||||
deferred_transaction::~deferred_transaction()
|
||||
{
|
||||
if(m_conn)
|
||||
{
|
||||
execute_raw_sql_statements(m_conn, "END TRANSACTION");
|
||||
}
|
||||
}
|
||||
} // namespace sql
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
@@ -0,0 +1,45 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 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/output/sql/common.hpp"
|
||||
|
||||
#include <sqlite3.h>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
namespace sql
|
||||
{
|
||||
struct deferred_transaction
|
||||
{
|
||||
explicit deferred_transaction(sqlite3* conn);
|
||||
~deferred_transaction();
|
||||
|
||||
private:
|
||||
sqlite3* m_conn = nullptr;
|
||||
};
|
||||
} // namespace sql
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
@@ -0,0 +1,141 @@
|
||||
// MIT License
|
||||
//
|
||||
// Copyright (c) 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/mpl.hpp"
|
||||
|
||||
#include <sqlite3.h>
|
||||
|
||||
#include <array>
|
||||
#include <cstdint>
|
||||
#include <optional>
|
||||
#include <string>
|
||||
#include <string_view>
|
||||
#include <type_traits>
|
||||
|
||||
namespace rocprofiler
|
||||
{
|
||||
namespace tool
|
||||
{
|
||||
namespace sql
|
||||
{
|
||||
template <typename Tp, typename CondT = void>
|
||||
struct extract_data_type;
|
||||
|
||||
template <typename Tp, int ExpectedV>
|
||||
bool
|
||||
column_data_is_null(sqlite3_stmt* stmt, int32_t col);
|
||||
|
||||
template <typename Tp>
|
||||
struct extract_data_type<Tp, std::enable_if_t<std::is_integral<Tp>::value>>
|
||||
{
|
||||
static constexpr auto value = SQLITE_INTEGER;
|
||||
|
||||
std::optional<Tp> operator()(sqlite3_stmt* stmt, int32_t col) const
|
||||
{
|
||||
if(column_data_is_null<Tp, value>(stmt, col)) return std::nullopt;
|
||||
|
||||
if constexpr(std::is_signed<Tp>::value)
|
||||
{
|
||||
if constexpr(sizeof(Tp) > sizeof(int32_t))
|
||||
return Tp{sqlite3_column_int64(stmt, col)};
|
||||
else
|
||||
return Tp{sqlite3_column_int(stmt, col)};
|
||||
}
|
||||
else
|
||||
{
|
||||
auto val = sqlite3_column_int64(stmt, col);
|
||||
return static_cast<Tp>(val);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Tp>
|
||||
struct extract_data_type<Tp, std::enable_if_t<std::is_floating_point<Tp>::value>>
|
||||
{
|
||||
static constexpr auto value = SQLITE_FLOAT;
|
||||
|
||||
std::optional<Tp> operator()(sqlite3_stmt* stmt, int32_t col) const
|
||||
{
|
||||
if(column_data_is_null<Tp, value>(stmt, col)) return std::nullopt;
|
||||
|
||||
return Tp{sqlite3_column_double(stmt, col)};
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Tp>
|
||||
struct extract_data_type<Tp, std::enable_if_t<common::mpl::is_string_type<Tp>::value>>
|
||||
{
|
||||
static constexpr auto value = SQLITE_TEXT;
|
||||
|
||||
std::optional<Tp> operator()(sqlite3_stmt* stmt, int32_t col) const
|
||||
{
|
||||
if(column_data_is_null<Tp, value>(stmt, col)) return std::nullopt;
|
||||
|
||||
const auto* ret = reinterpret_cast<const char*>(sqlite3_column_text(stmt, col));
|
||||
|
||||
if constexpr(std::is_constructible<Tp, const char*>::value) return Tp{ret};
|
||||
|
||||
return ret;
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Tp, size_t N>
|
||||
struct extract_data_type<
|
||||
std::array<Tp, N>,
|
||||
std::enable_if_t<std::is_integral<Tp>::value && sizeof(Tp) == sizeof(uint8_t)>>
|
||||
{
|
||||
static constexpr auto value = SQLITE_BLOB;
|
||||
|
||||
using value_type = std::array<Tp, N>;
|
||||
std::optional<value_type> operator()(sqlite3_stmt* stmt, int32_t col) const
|
||||
{
|
||||
if(column_data_is_null<value_type, value>(stmt, col)) return std::nullopt;
|
||||
|
||||
const auto* val = reinterpret_cast<const Tp*>(sqlite3_column_blob(stmt, col));
|
||||
|
||||
auto ret = value_type{};
|
||||
ret.fill(0);
|
||||
|
||||
uint64_t nbytes = std::min<uint64_t>(sqlite3_column_bytes(stmt, col), ret.size());
|
||||
for(uint64_t i = 0; i < nbytes; ++i)
|
||||
ret.at(i) = val[i];
|
||||
|
||||
return ret;
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Tp>
|
||||
struct extract_data_type<Tp, std::enable_if_t<std::is_same<Tp, std::nullptr_t>::value>>
|
||||
{
|
||||
static constexpr auto value = SQLITE_TEXT;
|
||||
|
||||
std::optional<Tp> operator()(sqlite3_stmt* stmt, int32_t col) const
|
||||
{
|
||||
if(column_data_is_null<Tp, value>(stmt, col)) return std::nullopt;
|
||||
return std::nullptr_t{};
|
||||
}
|
||||
};
|
||||
} // namespace sql
|
||||
} // namespace tool
|
||||
} // namespace rocprofiler
|
||||
@@ -101,7 +101,7 @@ function(rocprofiler_roctx_python_bindings _VERSION)
|
||||
install(
|
||||
FILES ${roctx_PYTHON_OUTPUT_DIRECTORY}/${_SOURCE}
|
||||
DESTINATION ${roctx_PYTHON_INSTALL_DIRECTORY}
|
||||
COMPONENT core)
|
||||
COMPONENT roctx)
|
||||
endforeach()
|
||||
|
||||
add_library(rocprofiler-sdk-roctx-python-bindings-${_VERSION} MODULE)
|
||||
@@ -129,7 +129,7 @@ function(rocprofiler_roctx_python_bindings _VERSION)
|
||||
install(
|
||||
TARGETS rocprofiler-sdk-roctx-python-bindings-${_VERSION}
|
||||
DESTINATION ${roctx_PYTHON_INSTALL_DIRECTORY}
|
||||
COMPONENT core)
|
||||
COMPONENT roctx)
|
||||
endfunction()
|
||||
|
||||
function(rocprofiler_rocpd_python_bindings_target_sources _VERSION)
|
||||
|
||||
@@ -0,0 +1,46 @@
|
||||
#
|
||||
# ROCm Profiling Data (rocpd) Library
|
||||
#
|
||||
set(CMAKE_INSTALL_DEFAULT_COMPONENT_NAME "rocpd")
|
||||
|
||||
add_library(rocprofiler-sdk-rocpd-shared-library SHARED)
|
||||
|
||||
foreach(_NAMESPACE rocprofiler-sdk rocprofiler-sdk-rocpd)
|
||||
foreach(_ALIAS library shared-library)
|
||||
add_library(${_NAMESPACE}::rocprofiler-sdk-rocpd-${_ALIAS} ALIAS
|
||||
rocprofiler-sdk-rocpd-shared-library)
|
||||
endforeach()
|
||||
endforeach()
|
||||
|
||||
target_sources(rocprofiler-sdk-rocpd-shared-library PRIVATE rocpd.cpp sql.cpp)
|
||||
|
||||
target_include_directories(
|
||||
rocprofiler-sdk-rocpd-shared-library
|
||||
INTERFACE $<BUILD_INTERFACE:${PROJECT_BINARY_DIR}/source/include>
|
||||
$<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}/source/include>
|
||||
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>)
|
||||
|
||||
target_link_libraries(
|
||||
rocprofiler-sdk-rocpd-shared-library
|
||||
PRIVATE rocprofiler-sdk::rocprofiler-sdk-headers
|
||||
rocprofiler-sdk::rocprofiler-sdk-build-flags
|
||||
rocprofiler-sdk::rocprofiler-sdk-memcheck
|
||||
rocprofiler-sdk::rocprofiler-sdk-common-library
|
||||
rocprofiler-sdk::rocprofiler-sdk-dl)
|
||||
|
||||
set_target_properties(
|
||||
rocprofiler-sdk-rocpd-shared-library
|
||||
PROPERTIES OUTPUT_NAME rocprofiler-sdk-rocpd
|
||||
LIBRARY_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/${CMAKE_INSTALL_LIBDIR}
|
||||
SOVERSION ${PROJECT_VERSION_MAJOR}
|
||||
VERSION ${PROJECT_VERSION}
|
||||
SKIP_BUILD_RPATH OFF
|
||||
BUILD_RPATH "\$ORIGIN"
|
||||
INSTALL_RPATH "\$ORIGIN"
|
||||
DEFINE_SYMBOL rocpd_EXPORTS)
|
||||
|
||||
install(
|
||||
TARGETS rocprofiler-sdk-rocpd-shared-library
|
||||
DESTINATION ${CMAKE_INSTALL_LIBDIR}
|
||||
COMPONENT rocpd
|
||||
EXPORT rocprofiler-sdk-rocpd-targets)
|
||||
@@ -0,0 +1,117 @@
|
||||
// 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.
|
||||
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/static_object.hpp"
|
||||
#include "lib/common/utility.hpp"
|
||||
|
||||
#include <rocprofiler-sdk-rocpd/defines.h>
|
||||
#include <rocprofiler-sdk-rocpd/rocpd.h>
|
||||
#include <rocprofiler-sdk-rocpd/types.h>
|
||||
|
||||
#include <array>
|
||||
#include <atomic>
|
||||
#include <cassert>
|
||||
|
||||
namespace rocpd
|
||||
{
|
||||
namespace
|
||||
{
|
||||
#define ROCPD_STATUS_STRING(CODE, MSG) \
|
||||
template <> \
|
||||
struct status_string<CODE> \
|
||||
{ \
|
||||
static constexpr auto name = #CODE; \
|
||||
static constexpr auto value = MSG; \
|
||||
};
|
||||
|
||||
template <size_t Idx>
|
||||
struct status_string;
|
||||
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_SUCCESS, "Success")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR, "General error")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR_INVALID_ARGUMENT, "Invalid function argument")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR_SQL_ERROR, "General SQL error")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR_SQL_INVALID_ENGINE, "Invalid SQL engine")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR_SQL_INVALID_SCHEMA_KIND, "Invalid SQL schema kind")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR_SQL_SCHEMA_NOT_FOUND, "SQL schema not found")
|
||||
ROCPD_STATUS_STRING(ROCPD_STATUS_ERROR_SQL_SCHEMA_PERMISSION_DENIED, "SQL schema could not be read")
|
||||
|
||||
template <size_t Idx, size_t... Tail>
|
||||
const char*
|
||||
get_status_name(rocpd_status_t status, std::index_sequence<Idx, Tail...>)
|
||||
{
|
||||
if(status == Idx) return status_string<Idx>::name;
|
||||
// recursion until tail empty
|
||||
if constexpr(sizeof...(Tail) > 0)
|
||||
return get_status_name(status, std::index_sequence<Tail...>{});
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
template <size_t Idx, size_t... Tail>
|
||||
const char*
|
||||
get_status_string(rocpd_status_t status, std::index_sequence<Idx, Tail...>)
|
||||
{
|
||||
if(status == Idx) return status_string<Idx>::value;
|
||||
// recursion until tail empty
|
||||
if constexpr(sizeof...(Tail) > 0)
|
||||
return get_status_string(status, std::index_sequence<Tail...>{});
|
||||
return nullptr;
|
||||
}
|
||||
} // namespace
|
||||
|
||||
// force initialization of logging on library load
|
||||
bool _rocpd_init_logging = (rocprofiler::common::init_logging("ROCPD"), true);
|
||||
} // namespace rocpd
|
||||
|
||||
ROCPD_EXTERN_C_INIT
|
||||
|
||||
rocpd_status_t
|
||||
rocpd_get_version(uint32_t* major, uint32_t* minor, uint32_t* patch)
|
||||
{
|
||||
if(major) *major = ROCPD_VERSION_MAJOR;
|
||||
if(minor) *minor = ROCPD_VERSION_MINOR;
|
||||
if(patch) *patch = ROCPD_VERSION_PATCH;
|
||||
return ROCPD_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
rocpd_status_t
|
||||
rocpd_get_version_triplet(rocpd_version_triplet_t* info)
|
||||
{
|
||||
*info = {
|
||||
.major = ROCPD_VERSION_MAJOR, .minor = ROCPD_VERSION_MINOR, .patch = ROCPD_VERSION_PATCH};
|
||||
return ROCPD_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
const char*
|
||||
rocpd_get_status_name(rocpd_status_t status)
|
||||
{
|
||||
return rocpd::get_status_name(status, std::make_index_sequence<ROCPD_STATUS_LAST>{});
|
||||
}
|
||||
|
||||
const char*
|
||||
rocpd_get_status_string(rocpd_status_t status)
|
||||
{
|
||||
return rocpd::get_status_string(status, std::make_index_sequence<ROCPD_STATUS_LAST>{});
|
||||
}
|
||||
|
||||
ROCPD_EXTERN_C_FINI
|
||||
@@ -0,0 +1,247 @@
|
||||
// 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.
|
||||
|
||||
#define _GNU_SOURCE 1
|
||||
|
||||
#include "lib/common/environment.hpp"
|
||||
#include "lib/common/filesystem.hpp"
|
||||
#include "lib/common/logging.hpp"
|
||||
#include "lib/common/mpl.hpp"
|
||||
#include "lib/common/string_entry.hpp"
|
||||
#include "lib/common/utility.hpp"
|
||||
|
||||
#include <rocprofiler-sdk-rocpd/rocpd.h>
|
||||
#include <rocprofiler-sdk-rocpd/sql.h>
|
||||
#include <rocprofiler-sdk-rocpd/types.h>
|
||||
|
||||
#include <rocprofiler-sdk/cxx/details/tokenize.hpp>
|
||||
|
||||
#include <fmt/core.h>
|
||||
#include <fmt/ranges.h>
|
||||
|
||||
#include <dlfcn.h>
|
||||
#include <initializer_list>
|
||||
#include <unordered_map>
|
||||
|
||||
namespace rocpd
|
||||
{
|
||||
namespace sql
|
||||
{
|
||||
namespace
|
||||
{
|
||||
namespace common = ::rocprofiler::common;
|
||||
namespace fs = ::rocprofiler::common::filesystem;
|
||||
|
||||
std::string
|
||||
get_install_path()
|
||||
{
|
||||
auto* _rocpd_sql_load_schema_sym = dlsym(RTLD_DEFAULT, "rocpd_sql_load_schema");
|
||||
|
||||
ROCP_CI_LOG_IF(WARNING, !_rocpd_sql_load_schema_sym)
|
||||
<< "[rocprofiler-sdk-rocpd] dlsym(RTLD_DEFAULT, 'rocpd_sql_load_schema') failed "
|
||||
"(unexpectedly) from within the rocprofiler-sdk-rocpd library";
|
||||
|
||||
if(!_rocpd_sql_load_schema_sym)
|
||||
_rocpd_sql_load_schema_sym = reinterpret_cast<void*>(&rocpd_sql_load_schema);
|
||||
|
||||
if(Dl_info dl_info = {};
|
||||
dladdr(_rocpd_sql_load_schema_sym, &dl_info) != 0 && dl_info.dli_fname != nullptr)
|
||||
{
|
||||
auto _share_path =
|
||||
fs::path{dl_info.dli_fname}.lexically_normal().parent_path().parent_path() /
|
||||
std::string{"share/rocprofiler-sdk-rocpd"};
|
||||
ROCP_INFO << fmt::format("[rocprofiler-sdk-rocpd] resolved rocprofiler-sdk-rocpd SQL "
|
||||
"schema path as '{}' (dli_fname: {})",
|
||||
_share_path.string(),
|
||||
dl_info.dli_fname);
|
||||
return _share_path;
|
||||
}
|
||||
|
||||
ROCP_CI_LOG(WARNING)
|
||||
<< "Failed to locate the installation path of rocprofiler-sdk-rocpd via dladdr of the "
|
||||
"'rocpd_sql_load_schema' symbol (which should be in librocprofiler-sdk-rocpd.so)";
|
||||
|
||||
return std::string{};
|
||||
}
|
||||
|
||||
template <typename Tp>
|
||||
auto
|
||||
replace_all(std::string val, Tp from, std::string_view to)
|
||||
{
|
||||
size_t pos = 0;
|
||||
while((pos = val.find(from, pos)) != std::string::npos)
|
||||
{
|
||||
if constexpr(std::is_same<common::mpl::unqualified_type_t<Tp>, char>::value)
|
||||
{
|
||||
val.replace(pos, 1, to);
|
||||
pos += to.length();
|
||||
}
|
||||
else
|
||||
{
|
||||
val.replace(pos, std::string_view{from}.length(), to);
|
||||
pos += to.length();
|
||||
}
|
||||
}
|
||||
return val;
|
||||
}
|
||||
} // namespace
|
||||
} // namespace sql
|
||||
} // namespace rocpd
|
||||
|
||||
extern "C" {
|
||||
rocpd_status_t
|
||||
rocpd_sql_load_schema(rocpd_sql_engine_t engine,
|
||||
rocpd_sql_schema_kind_t kind,
|
||||
rocpd_sql_options_t options,
|
||||
const rocpd_sql_schema_jinja_variables_t* variables,
|
||||
rocpd_sql_load_schema_cb_t callback,
|
||||
const char** schema_path_hints,
|
||||
uint64_t num_schema_path_hints,
|
||||
void* user_data)
|
||||
{
|
||||
namespace fs = ::rocpd::sql::fs;
|
||||
|
||||
switch(engine)
|
||||
{
|
||||
case ROCPD_SQL_ENGINE_SQLITE3:
|
||||
{
|
||||
break;
|
||||
}
|
||||
case ROCPD_SQL_ENGINE_NONE:
|
||||
case ROCPD_SQL_ENGINE_LAST:
|
||||
{
|
||||
return ROCPD_STATUS_ERROR_SQL_INVALID_ENGINE;
|
||||
}
|
||||
}
|
||||
|
||||
const auto kind_file_names = std::unordered_map<rocpd_sql_schema_kind_t, std::string_view>{
|
||||
{ROCPD_SQL_SCHEMA_ROCPD_TABLES, "rocpd_tables.sql"},
|
||||
{ROCPD_SQL_SCHEMA_ROCPD_INDEXES, "rocpd_indexes.sql"},
|
||||
{ROCPD_SQL_SCHEMA_ROCPD_VIEWS, "rocpd_views.sql"},
|
||||
{ROCPD_SQL_SCHEMA_ROCPD_DATA_VIEWS, "data_views.sql"},
|
||||
{ROCPD_SQL_SCHEMA_ROCPD_SUMMARY_VIEWS, "summary_views.sql"},
|
||||
{ROCPD_SQL_SCHEMA_ROCPD_MARKER_VIEWS, "marker_views.sql"},
|
||||
};
|
||||
|
||||
const auto _lib_schema_path = rocpd::sql::get_install_path();
|
||||
const auto _env_schema_path = rocprofiler::common::get_env("ROCPD_SCHEMA_PATH", "");
|
||||
const auto _usr_schema_path =
|
||||
(schema_path_hints)
|
||||
? fmt::format(
|
||||
"{}",
|
||||
fmt::join(schema_path_hints, schema_path_hints + num_schema_path_hints, ":"))
|
||||
: std::string{};
|
||||
const auto _schema_paths =
|
||||
fmt::format("{}:{}:{}", _usr_schema_path, _env_schema_path, _lib_schema_path);
|
||||
|
||||
if(kind_file_names.count(kind) == 0) return ROCPD_STATUS_ERROR_SQL_INVALID_SCHEMA_KIND;
|
||||
|
||||
auto _schema_file = std::optional<std::string>{};
|
||||
for(const auto& itr : rocprofiler::sdk::parse::tokenize(_schema_paths, ":"))
|
||||
{
|
||||
auto _fpath = fs::path{itr} / kind_file_names.at(kind);
|
||||
ROCP_TRACE << fmt::format("[rocprofiler-sdk-rocpd] Searching for schema file: '{}'",
|
||||
_fpath.string());
|
||||
if(fs::exists(_fpath))
|
||||
{
|
||||
ROCP_INFO << fmt::format("[rocprofiler-sdk-rocpd] Found schema file: '{}'",
|
||||
_fpath.string());
|
||||
_schema_file = _fpath;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if(!_schema_file) return ROCPD_STATUS_ERROR_SQL_SCHEMA_NOT_FOUND;
|
||||
|
||||
auto read_file = [](const std::string& _file_path) -> std::string {
|
||||
auto _ifs = std::ifstream{_file_path, std::ios::in | std::ios::binary};
|
||||
if(!_ifs.is_open()) return std::string{};
|
||||
|
||||
auto _buffer = std::stringstream{};
|
||||
_buffer << _ifs.rdbuf();
|
||||
return _buffer.str();
|
||||
};
|
||||
|
||||
auto _contents = read_file(*_schema_file);
|
||||
|
||||
if(_contents.empty()) return ROCPD_STATUS_ERROR_SQL_SCHEMA_PERMISSION_DENIED;
|
||||
|
||||
if(engine == ROCPD_SQL_ENGINE_SQLITE3)
|
||||
{
|
||||
if((options & ROCPD_SQL_OPTIONS_SQLITE3_PRAGMA_FOREIGN_KEYS) ==
|
||||
ROCPD_SQL_OPTIONS_SQLITE3_PRAGMA_FOREIGN_KEYS)
|
||||
_contents = fmt::format("PRAGMA foreign_keys = ON;\n\n{}", _contents);
|
||||
}
|
||||
|
||||
auto _substitutions = std::vector<std::pair<std::string_view, std::string>>{};
|
||||
|
||||
using jinja_init_list_t = std::initializer_list<std::pair<std::string_view, const char*>>;
|
||||
|
||||
if(variables != nullptr)
|
||||
{
|
||||
if(variables->size == 0)
|
||||
{
|
||||
return ROCPD_STATUS_ERROR_SQL_INVALID_SCHEMA_KIND;
|
||||
}
|
||||
|
||||
// {{uuid}} is used in table names and require special handling
|
||||
if(const auto* value = variables->uuid; value != nullptr)
|
||||
{
|
||||
auto _value = std::string{value};
|
||||
|
||||
// non-empty strings are prefixed with underscore for readability
|
||||
if(!_value.empty() && _value.find('_') != 0) _value = fmt::format("_{}", _value);
|
||||
|
||||
// replace hyphens with underscores since these are used in table/view names
|
||||
if(_value.find('-') != std::string::npos)
|
||||
_value = rocpd::sql::replace_all(_value, "-", "_");
|
||||
|
||||
// make substitutions
|
||||
_contents = rocpd::sql::replace_all(_contents, "{{uuid}}", _value);
|
||||
}
|
||||
|
||||
// make substitutions for remaining variables which do not require special handling like
|
||||
// {{uuid}}
|
||||
for(auto [key, value] : jinja_init_list_t{{"{{guid}}", variables->guid}})
|
||||
{
|
||||
if(value != nullptr)
|
||||
{
|
||||
_contents = rocpd::sql::replace_all(_contents, key, std::string_view{value});
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
const auto* cb_schema_path = _schema_file->c_str();
|
||||
const auto* cb_schema_contents = _contents.c_str();
|
||||
|
||||
if((options & ROCPD_SQL_OPTIONS_SQLITE3_PRAGMA_FOREIGN_KEYS) ==
|
||||
ROCPD_SQL_OPTIONS_SQLITE3_PRAGMA_FOREIGN_KEYS)
|
||||
{
|
||||
cb_schema_path = rocprofiler::common::get_string_entry(cb_schema_path)->c_str();
|
||||
cb_schema_contents = rocprofiler::common::get_string_entry(cb_schema_contents)->c_str();
|
||||
}
|
||||
|
||||
callback(engine, kind, options, variables, cb_schema_path, cb_schema_contents, user_data);
|
||||
|
||||
return ROCPD_STATUS_SUCCESS;
|
||||
}
|
||||
}
|
||||
@@ -1,335 +0,0 @@
|
||||
# rocprofv3 Multi-Node Profiling Data
|
||||
|
||||
## Overview
|
||||
|
||||
- rocprofv3 adds supports for a `--output-format rocpd` option which enables writing a SQLite database file (one per process) with the collected data
|
||||
- Use SQL schema from `rocpd` initially to support the rocpd post-processing analysis support
|
||||
- In order to visualize the data, users will convert the database(s) to their desired visualization formats
|
||||
- SQL has a relatively easy way to treat multiple separate databases as one database via views
|
||||
- rocprofv3 provides some command-line tools built on top of a python package designed for post-processing our databases
|
||||
|
||||
### Skills Required for Tasks
|
||||
|
||||
1. Rework rocprofv3 tool library output functions
|
||||
- __C++__: output functions written in C++ (`^/source/lib/rocprofiler-sdk-tool/generate*`)
|
||||
- __CMake__: move the output functions into stand-alone library
|
||||
2. Create Python package skeleton in `^/source/lib/python`
|
||||
- __Python__: organizing a Python package to be importable (`import rocpd`) and executable (i.e. `python -m rocpd --help`)
|
||||
3. Adding rocprofv3 SQLite support
|
||||
- __C++__: just a general skill requirement for working with rocprofiler-sdk
|
||||
- __CMake__: for integrating SQLite and python bindings into rocprofiler-sdk build
|
||||
- __SQL__: understanding of SQL statement meanings, knowledge of `rocpd` SQL schema
|
||||
4. Python bindings for output functions
|
||||
- __C++__: just a general skill requirement for working with rocprofiler-sdk
|
||||
- __PyBind11__: for writing Python bindings
|
||||
|
||||
#### Task #1: Rework `rocprofv3` Tool Library Output Functions
|
||||
|
||||
The problems with most of the output functions are:
|
||||
|
||||
- Problem: Access global memory via `tool_table` functions
|
||||
- Global memory access won't work well for invocation of these functions via Python bindings
|
||||
- Ideally, these functions should be written in the (pseudo-) functional programming style, i.e., function only accesses memory of arguments, communicates via return value, and avoids concepts like shared states but without restrictions such as immutable data arguments
|
||||
- Problem: Require all the profiling data to be loaded into memory
|
||||
- During runtime, rocprofv3 writes data to buffer and when buffer is full, writes the binary blob to a temporary intermediate binary file
|
||||
- During finalization, rocprofv3 reads _all_ of this data back into memory from the intermediate binary file and then writes to various output forms
|
||||
- This approach will not work when amount of collected data exceeds amount of available RAM, especially on systems with swap disabled; e.g., 1 TB of profiling data on system with 128 GB of RAM
|
||||
- We need to be able to stream data in chunks to these output functions
|
||||
- Proposed approach: function which creates a file handle, function which writes chunk of data to file (invoked multiple times), function which closes file handle
|
||||
|
||||
> Assigned: Markus, Olha, Jin, Araceli (i.e. onboarding group task) + Jonathan (CMake part)
|
||||
|
||||
##### Tasks
|
||||
|
||||
1. Move the `source/lib/rocprofiler-sdk-tool/generate*.{hpp,cpp}` functions into standalone (static) library: `source/lib/tool-data`
|
||||
- May require `source/lib/tool-common` (static) library if something is needed by both `tool-data` and `rocprofiler-sdk-tool` libraries
|
||||
- Please consult if you have any questions about where to put things and/or naming conventions
|
||||
- Pay attention to existing CMake and use similar style
|
||||
- We will link this library into `rocprofiler-sdk-tool` and link it into Python bindings library
|
||||
2. Solve global memory access problem
|
||||
- Probably need some additional data structures which represents the data currently stored/accessed from global memory which will be passed into function.
|
||||
|
||||
### Python Package for Converting Databases to Other Output Formats
|
||||
|
||||
> __Note__: We could potentially reuse `rocpd` for the the python package name since "ROCm Profiling Data" is a pretty appropriate name.
|
||||
|
||||
rocprofv3 will need to rework the output functions within the `librocprofiler-sdk-tool.so` library (underlying library used by `rocprofv3`) in order to support Python bindings.
|
||||
For example, `generateJSON(...)` currently fetches info from global memory stored during the run, we need these functions to be pure: the only memory operated on is from the function arguments.
|
||||
Furthermore, these output functions need to support partial writes: invocations with only a subset of the data so that all the data need not be loaded into memory at one time.
|
||||
|
||||
> __Example__ (workflow): get handle to output format, e.g. a Perfetto session, invoke `generatePerfetto(...)` with some of the data, repeat until all data has been passed, close handle to the output format.
|
||||
|
||||
These reworked functions should be moved to another library, e.g. `librocprofiler-sdk-tool-io.(a|so)`.
|
||||
Once the output functions are isolated and functional, we need to generate python bindings (via PyBind11) so that a python package can be built on top of them.
|
||||
Various command-line tools can be provided using `__main__.py` file(s) within our python package.
|
||||
Users can use the python package to write their own scripts.
|
||||
|
||||
> __Example__ (two databases, one Perfetto trace): `rocprofv3-merge --output-format pftrace --out mybenchmark.pftrace --in results-1000.db results-1001.db`
|
||||
|
||||
### Treating multiple SQL databases as one database
|
||||
|
||||
```python
|
||||
conn = sqlite3.connect('db1.db')
|
||||
conn.execute("ATTACH DATABASE 'db2.db' AS db2;")
|
||||
conn.execute("ATTACH DATABASE 'db3.db' AS db3;")
|
||||
|
||||
# Create a view that unifies the 'users' table from all three databases
|
||||
conn.execute("""
|
||||
CREATE VIEW all_users AS
|
||||
SELECT * FROM users
|
||||
UNION ALL
|
||||
SELECT * FROM db2.users
|
||||
UNION ALL
|
||||
SELECT * FROM db3.users;
|
||||
""")
|
||||
|
||||
# Now you can query the view as if it were a single table
|
||||
cursor = conn.execute("SELECT * FROM all_users;")
|
||||
for row in cursor:
|
||||
print(row)
|
||||
|
||||
# Close the connection
|
||||
conn.close()
|
||||
```
|
||||
|
||||
## Proposed SQL Schema
|
||||
|
||||
A more comprehensive SQL Schema is proposed below. This schema is intended to be more comprehensive with respect to the
|
||||
various types of data that profilers can collect (such as Omnitrace/RSP)
|
||||
|
||||
The schema consists of multiple interrelated tables to capture different categories of profiling data.
|
||||
Below is a high-level schema with the primary tables and relationships.
|
||||
|
||||
__*Please note, this is a very preliminary sketch of the schema*__.
|
||||
If you want to weigh in, please restrict comments to the high-level organization, comments that it doesn't contain
|
||||
fields for correlation IDs or something like that are not particularly helpful at the moment.
|
||||
|
||||
```sql
|
||||
CREATE TABLE strings (
|
||||
id SERIAL PRIMARY KEY,
|
||||
value VARCHAR(1024) UNIQUE
|
||||
);
|
||||
|
||||
CREATE TABLE process (
|
||||
id INT PRIMARY KEY,
|
||||
pid INT,
|
||||
process_name_id INT,
|
||||
executable_path_id INT,
|
||||
start_time BIGINT,
|
||||
end_time BIGINT,
|
||||
FOREIGN KEY (process_name_id) REFERENCES strings(id)
|
||||
FOREIGN KEY (executable_path_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE thread (
|
||||
id INT PRIMARY KEY,
|
||||
tid INT,
|
||||
process_id INT,
|
||||
thread_name_id INT,
|
||||
start_time BIGINT,
|
||||
end_time BIGINT,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (thread_name_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE cpu_info (
|
||||
id SERIAL PRIMARY KEY,
|
||||
core_id INT,
|
||||
socket_id INT,
|
||||
frequency_hz INT,
|
||||
model_id INT,
|
||||
cache_size_kb INT,
|
||||
FOREIGN KEY (model_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE gpu_info (
|
||||
id SERIAL PRIMARY KEY,
|
||||
device_name_id INT,
|
||||
compute_capability_id INT,
|
||||
memory_size_mb INT,
|
||||
multiprocessor_count INT,
|
||||
clock_rate_hz INT,
|
||||
FOREIGN KEY (device_name_id) REFERENCES strings(id)
|
||||
FOREIGN KEY (compute_capability_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE instrumentation_regions (
|
||||
id SERIAL PRIMARY KEY,
|
||||
process_id INT,
|
||||
thread_id INT,
|
||||
region_name_id INT,
|
||||
start_time BIGINT,
|
||||
end_time BIGINT,
|
||||
parent_region_id INT,
|
||||
duration_ns BIGINT GENERATED ALWAYS AS (end_time - start_time) STORED,
|
||||
file_id INT,
|
||||
line_number INT,
|
||||
additional_info JSONB,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (thread_id) REFERENCES thread(thread_id),
|
||||
FOREIGN KEY (region_name_id) REFERENCES strings(id),
|
||||
FOREIGN KEY (file_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE call_stacks (
|
||||
id SERIAL PRIMARY KEY,
|
||||
process_id INT,
|
||||
thread_id INT,
|
||||
timestamp BIGINT,
|
||||
stack_depth INT,
|
||||
function_id INT,
|
||||
file_id INT,
|
||||
line_number INT,
|
||||
parent_sample_id INT,
|
||||
call_site VARCHAR(1024),
|
||||
additional_info JSONB,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (thread_id) REFERENCES thread(thread_id),
|
||||
FOREIGN KEY (function_id) REFERENCES strings(id),
|
||||
FOREIGN KEY (file_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE hardware_counters (
|
||||
id SERIAL PRIMARY KEY,
|
||||
process_id INT,
|
||||
thread_id INT,
|
||||
timestamp BIGINT,
|
||||
event_id INT,
|
||||
value BIGINT,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (thread_id) REFERENCES thread(thread_id),
|
||||
FOREIGN KEY (event_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE memory_operations (
|
||||
id SERIAL PRIMARY KEY,
|
||||
process_id INT,
|
||||
thread_id INT,
|
||||
timestamp BIGINT,
|
||||
operation_type VARCHAR(50) CHECK (operation_type IN ('ALLOC', 'FREE', 'COPY')),
|
||||
source_address BIGINT,
|
||||
destination_address BIGINT,
|
||||
size_bytes BIGINT,
|
||||
duration_us BIGINT,
|
||||
additional_info JSONB,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (thread_id) REFERENCES thread(thread_id)
|
||||
);
|
||||
|
||||
CREATE TABLE gpu_kernel_launches (
|
||||
id SERIAL PRIMARY KEY,
|
||||
process_id INT,
|
||||
thread_id INT,
|
||||
gpu_id INT,
|
||||
kernel_id INT,
|
||||
dispatch_id INT,
|
||||
launch_time BIGINT,
|
||||
start_time BIGINT,
|
||||
end_time BIGINT,
|
||||
grid_size_x INT,
|
||||
grid_size_y INT,
|
||||
grid_size_z INT,
|
||||
block_size_x INT,
|
||||
block_size_y INT,
|
||||
block_size_z INT,
|
||||
shared_mem_bytes INT,
|
||||
duration_ns BIGINT GENERATED ALWAYS AS (end_time - start_time) STORED,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (thread_id) REFERENCES thread(thread_id),
|
||||
FOREIGN KEY (gpu_id) REFERENCES gpu_info(gpu_id),
|
||||
FOREIGN KEY (kernel_id) REFERENCES strings(id)
|
||||
);
|
||||
|
||||
CREATE TABLE binary_analysis_info (
|
||||
id SERIAL PRIMARY KEY,
|
||||
process_id INT,
|
||||
binary_name VARCHAR(1024),
|
||||
function_id INT,
|
||||
start_address BIGINT,
|
||||
end_address BIGINT,
|
||||
instruction_count INT,
|
||||
file_id INT,
|
||||
line_number INT,
|
||||
call_sites JSONB,
|
||||
additional_info JSONB,
|
||||
FOREIGN KEY (process_id) REFERENCES process(process_id),
|
||||
FOREIGN KEY (function_id) REFERENCES strings(id),
|
||||
FOREIGN KEY (file_id) REFERENCES strings(id)
|
||||
);
|
||||
```
|
||||
|
||||
Explanation of the design considerations:
|
||||
|
||||
1. __Separate String Tables__: Created unique string tables for function names, file names, kernel names, and event names to avoid storing redundant copies.
|
||||
- `function_names`, `file_names`, `kernel_names`, and `event_names` tables are created to hold unique strings. Each table has a surrogate primary key (`function_id`, `file_id`, `kernel_id`, `event_id`) that is referenced by the main tables.
|
||||
- This avoids storing redundant copies of long or frequently repeating strings in different tables, reducing the storage footprint and improving consistency.
|
||||
2. __Foreign Key References__: Main tables reference unique strings using foreign keys for consistency and space efficiency.
|
||||
- Main tables such as `instrumentation_regions`, `call_stacks`, `gpu_kernel_launches`, etc., reference these unique string tables using foreign keys.
|
||||
- This makes querying for specific function names or kernel names more efficient, as the strings are indexed separately.
|
||||
3. __Computed Columns__: Used computed columns for duration fields to facilitate quick analysis.
|
||||
- The `duration_us` columns are computed based on timestamps, providing useful metrics for quick analysis.
|
||||
4. __Extensibility__: Designed to be easily extensible with additional string categories if needed.
|
||||
- New string types or categories can be added by creating new tables, and the main tables can reference them with minor schema adjustments.
|
||||
5. __JSONB for Additional Metadata__:
|
||||
- JSONB columns (`additional_info`) are used to handle complex or variable metadata that doesn’t fit neatly into the structured schema (e.g., custom annotations, extra debug info).
|
||||
|
||||
### Example Data Insertion and Lookup
|
||||
|
||||
#### Adding a new function
|
||||
|
||||
```sql
|
||||
INSERT INTO function_names (function_name) VALUES ('my_function') ON CONFLICT (function_name) DO NOTHING;
|
||||
```
|
||||
|
||||
#### Linking a function in a call stack
|
||||
|
||||
```sql
|
||||
INSERT INTO call_stacks (process_id, thread_id, timestamp, stack_depth, function_id, file_id)
|
||||
VALUES (123, 456, '2024-09-27 10:00:00', 1, (SELECT function_id FROM function_names WHERE function_name = 'my_function'),
|
||||
(SELECT file_id FROM file_names WHERE file_name = 'my_file.c'));
|
||||
```
|
||||
|
||||
## Q & A
|
||||
|
||||
### All global variables are protected with locks in common synchronized library. How are we sending the data from these variables to the pure functions?
|
||||
|
||||
There is a new `rocprofiler::tool::metadata` struct in `lib/output/metadata.hpp` which will be populated with data from SQL.
|
||||
This struct is passed to the output functions.
|
||||
|
||||
### If we provide the functionality to flush the trace at regular intervals, do we delete the data in global memory after each flush? If not, how do we keep track of data already read at any given point time during runtime?
|
||||
|
||||
We will probably not delete the metadata (agent info, code objects, kernel symbols, etc.) after a flush.
|
||||
When we flush, we will swap out the temporary binary file with a new temporary binary file and write/append the database with
|
||||
the contents of the old temporary binary file.
|
||||
|
||||
### Can a user collect trace at regular flush interval and ask for counter collection at the end of application?
|
||||
|
||||
I am not sure what you mean here. We can write counter collection data when we flush. If the user is asking for periodic
|
||||
flushing, we will restrict the output format to the database. In other words, I suspect that only `--flush-rate X` will only
|
||||
be compatible with `--output-format rocpd` -- any additional or alternative data formats and we will throw an error in the
|
||||
rocprofv3 script. This is for simplicity sake, supporting periodically flushing to CSV, etc. is unnecessary in my opinion.
|
||||
|
||||
### I think hardware_counters table in database schema should have a dispatch_id field to represent the kernel it belongs to
|
||||
|
||||
Please note, the proposed schema states clearly:
|
||||
|
||||
> __*Please note, this is a very preliminary sketch of the schema*__.
|
||||
> If you want to weigh in, please restrict comments to the high-level organization, comments that it doesn't contain
|
||||
> fields for correlation IDs or something like that are not particularly helpful at the moment.
|
||||
|
||||
However, I will note that the hardware counters table is probably going to be generic, i.e. supporting CPU HW counters, which
|
||||
do not have dispatch IDs. Lastly, I will also note, device counter collection is not associated with a dispatch so even in
|
||||
the case of GPU HW counters, including this field is questionable.
|
||||
|
||||
### What is binary analysis info table?
|
||||
|
||||
More advanced tools such as Omnitrace/Rocprofiler-System do address to line translations. This could also potentally
|
||||
include the sort of data related to PC sampling
|
||||
|
||||
### What is the Key of gpu info table? Node_id/zero based numbering scheme?
|
||||
|
||||
That isn't defined. Very preliminary sketch.
|
||||
|
||||
### When is user allowed to access the database in case of flushing the trace at regular intervals? Is user allowed to read the database only after tool finalization? Or we create a database file for each interval?
|
||||
|
||||
TBD on the exact details but the user will certainly be able to read the database before tool finalization when it is flushed.
|
||||
@@ -49,6 +49,7 @@
|
||||
#include "lib/output/generateJSON.hpp"
|
||||
#include "lib/output/generateOTF2.hpp"
|
||||
#include "lib/output/generatePerfetto.hpp"
|
||||
#include "lib/output/generateRocpd.hpp"
|
||||
#include "lib/output/generateStats.hpp"
|
||||
#include "lib/output/metadata.hpp"
|
||||
#include "lib/output/output_stream.hpp"
|
||||
@@ -2444,6 +2445,24 @@ tool_fini(void* /*tool_data*/)
|
||||
rocjpeg_output.get_generator());
|
||||
}
|
||||
|
||||
if(tool::get_config().rocpd_output && outdata.num_output > 0 &&
|
||||
outdata.num_bytes >= tool::get_config().minimum_output_bytes)
|
||||
{
|
||||
tool::write_rocpd(tool::get_config(),
|
||||
*tool_metadata,
|
||||
agents_output,
|
||||
hip_output.get_generator(),
|
||||
hsa_output.get_generator(),
|
||||
kernel_dispatch_output.get_generator(),
|
||||
memory_copy_output.get_generator(),
|
||||
marker_output.get_generator(),
|
||||
memory_allocation_output.get_generator(),
|
||||
scratch_memory_output.get_generator(),
|
||||
rccl_output.get_generator(),
|
||||
rocdecode_output.get_generator(),
|
||||
counters_output.get_generator());
|
||||
}
|
||||
|
||||
if(tool::get_config().otf2_output && outdata.num_output > 0 &&
|
||||
outdata.num_bytes >= tool::get_config().minimum_output_bytes)
|
||||
{
|
||||
|
||||
@@ -23,9 +23,7 @@ set(ROCPROFILER_LIB_COUNTERS_HEADERS
|
||||
ioctl.hpp)
|
||||
target_sources(rocprofiler-sdk-object-library PRIVATE ${ROCPROFILER_LIB_COUNTERS_SOURCES}
|
||||
${ROCPROFILER_LIB_COUNTERS_HEADERS})
|
||||
add_subdirectory(xml)
|
||||
add_subdirectory(parser)
|
||||
add_subdirectory(yaml)
|
||||
|
||||
if(ROCPROFILER_BUILD_TESTS)
|
||||
add_subdirectory(tests)
|
||||
|
||||
@@ -1,10 +0,0 @@
|
||||
configure_file(basic_counters.xml
|
||||
${PROJECT_BINARY_DIR}/share/rocprofiler-sdk/basic_counters.xml COPYONLY)
|
||||
configure_file(derived_counters.xml
|
||||
${PROJECT_BINARY_DIR}/share/rocprofiler-sdk/derived_counters.xml COPYONLY)
|
||||
|
||||
install(
|
||||
FILES ${PROJECT_BINARY_DIR}/share/rocprofiler-sdk/basic_counters.xml
|
||||
${PROJECT_BINARY_DIR}/share/rocprofiler-sdk/derived_counters.xml
|
||||
DESTINATION share/rocprofiler-sdk
|
||||
COMPONENT core)
|
||||
@@ -1,780 +0,0 @@
|
||||
<gfx8 base="gfx8">
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
|
||||
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
|
||||
|
||||
<metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_INSTS_VALU" block=SQ event=26 descr="Number of VALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_WR" block=SQ event=27 descr="Number of VMEM write instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_RD" block=SQ event=28 descr="Number of VMEM read instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=30 descr="Number of SALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=31 descr="Number of SMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=32 descr="Number of FLAT instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT_LDS_ONLY" block=SQ event=33 descr="Number of FLAT instructions issued that read/wrote only from/to LDS (only works if EARLY_TA_DONE is enabled). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=34 descr="Number of LDS instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=35 descr="Number of GDS instructions issued. (per-simd, emulated)"></metric>
|
||||
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=61 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VALU" block=SQ event=69 descr="Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SALU" block=SQ event=86 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_THREAD_CYCLES_VALU" block=SQ event=89 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
|
||||
<metric name="SQ_LDS_BANK_CONFLICT" block=SQ event=97 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
|
||||
|
||||
<metric name="TA_TA_BUSY" block=TA event=15 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS" block=TA event=101 descr="Number of flat opcode reads processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS" block=TA event=102 descr="Number of flat opcode writes processed by the TA."></metric>
|
||||
|
||||
<metric name="TCC_HIT" block=TCC event=18 descr="Number of cache hits."></metric>
|
||||
<metric name="TCC_MISS" block=TCC event=19 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="TCC_MC_RDREQ" block=TCC event=35 descr="Number of 32-byte reads. The hardware actually does 64-byte reads but the number is adjusted to provide uniformity."></metric>
|
||||
<metric name="TCC_MC_WRREQ" block=TCC event=26 descr="Number of 32-byte transactions going over the TC_MC_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests."></metric>
|
||||
<metric name="TCC_MC_WRREQ_STALL" block=TCC event=28 descr="Number of cycles a write request was stalled."></metric>
|
||||
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES" block=TCP event=3 descr="TCP stalls TA data interface. Now Windowed."></metric>
|
||||
</gfx8>
|
||||
|
||||
<gfx9>
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
|
||||
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
|
||||
|
||||
<metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_INSTS_VALU" block=SQ event=26 descr="Number of VALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_WR" block=SQ event=27 descr="Number of VMEM write instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_RD" block=SQ event=28 descr="Number of VMEM read instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=30 descr="Number of SALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=31 descr="Number of SMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=32 descr="Number of FLAT instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT_LDS_ONLY" block=SQ event=33 descr="Number of FLAT instructions issued that read/wrote only from/to LDS (only works if EARLY_TA_DONE is enabled). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=34 descr="Number of LDS instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=35 descr="Number of GDS instructions issued. (per-simd, emulated)"></metric>
|
||||
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=63 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VALU" block=SQ event=71 descr="regspec 71? Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SALU" block=SQ event=84 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_THREAD_CYCLES_VALU" block=SQ event=85 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
|
||||
<metric name="SQ_LDS_BANK_CONFLICT" block=SQ event=93 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
|
||||
|
||||
<metric name="TA_TA_BUSY" block=TA event=15 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS" block=TA event=101 descr="Number of flat opcode reads processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS" block=TA event=102 descr="Number of flat opcode writes processed by the TA."></metric>
|
||||
|
||||
<metric name="TCC_HIT" block=TCC event=20 descr="Number of cache hits."></metric>
|
||||
<metric name="TCC_MISS" block=TCC event=22 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="TCC_EA_WRREQ" block=TCC event=29 descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands."></metric>
|
||||
<metric name="TCC_EA_WRREQ_64B" block=TCC event=30 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="TCC_EA_WRREQ_STALL" block=TCC event=33 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="TCC_EA_RDREQ" block=TCC event=41 descr="Number of TCC/EA read requests (either 32-byte or 64-byte)"></metric>
|
||||
<metric name="TCC_EA_RDREQ_32B" block=TCC event=42 descr="Number of 32-byte TCC/EA read requests"></metric>
|
||||
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES" block=TCP event=6 descr="TCP stalls TA data interface. Now Windowed."></metric>
|
||||
</gfx9>
|
||||
|
||||
<gfx900 base="gfx9">
|
||||
</gfx900>
|
||||
|
||||
<gfx906 base="gfx9">
|
||||
# EA1
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="TCC_EA1_WRREQ" block=TCC event=256 descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands."></metric>
|
||||
<metric name="TCC_EA1_WRREQ_64B" block=TCC event=257 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="TCC_EA1_WRREQ_STALL" block=TCC event=260 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="TCC_EA1_RDREQ" block=TCC event=267 descr="Number of TCC/EA read requests (either 32-byte or 64-byte)"></metric>
|
||||
<metric name="TCC_EA1_RDREQ_32B" block=TCC event=268 descr="Number of 32-byte TCC/EA read requests"></metric>
|
||||
</gfx906>
|
||||
|
||||
<gfx908 base="gfx9">
|
||||
<metric name="SQ_INSTS_VMEM_WR" block=SQ event=28 descr="Number of VMEM write instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_RD" block=SQ event=29 descr="Number of VMEM read instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=31 descr="Number of SALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=32 descr="Number of SMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=33 descr="Number of FLAT instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT_LDS_ONLY" block=SQ event=34 descr="Number of FLAT instructions issued that read/wrote only from/to LDS (only works if EARLY_TA_DONE is enabled). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=35 descr="Number of LDS instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=36 descr="Number of GDS instructions issued. (per-simd, emulated)"></metric>
|
||||
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=64 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VALU" block=SQ event=72 descr="regspec 71? Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SALU" block=SQ event=85 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_THREAD_CYCLES_VALU" block=SQ event=86 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
|
||||
<metric name="SQ_LDS_BANK_CONFLICT" block=SQ event=94 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
|
||||
|
||||
<metric name="TCC_HIT" block=TCC event=17 descr="Number of cache hits."></metric>
|
||||
<metric name="TCC_MISS" block=TCC event=19 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="TCC_EA_WRREQ" block=TCC event=26 descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands."></metric>
|
||||
<metric name="TCC_EA_WRREQ_64B" block=TCC event=27 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="TCC_EA_WRREQ_STALL" block=TCC event=30 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="TCC_EA_RDREQ" block=TCC event=38 descr="Number of TCC/EA read requests (either 32-byte or 64-byte)"></metric>
|
||||
<metric name="TCC_EA_RDREQ_32B" block=TCC event=39 descr="Number of 32-byte TCC/EA read requests"></metric>
|
||||
</gfx908>
|
||||
|
||||
<gfx90a>
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=91 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES" block=TCP event=6 descr="TCP stalls TA data interface. Now Windowed."></metric>
|
||||
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
|
||||
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
|
||||
<metric name="GRBM_CP_BUSY" block=GRBM event=3 descr="Any of the Command Processor (CPG/CPC/CPF) blocks are busy."></metric>
|
||||
<metric name="GRBM_SPI_BUSY" block=GRBM event=11 descr="Any of the Shader Pipe Interpolators (SPI) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GRBM_TA_BUSY" block=GRBM event=13 descr="Any of the Texture Pipes (TA) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GRBM_TC_BUSY" block=GRBM event=28 descr="Any of the Texture Cache Blocks (TCP/TCI/TCA/TCC) are busy."></metric>
|
||||
<metric name="GRBM_CPC_BUSY" block=GRBM event=30 descr="The Command Processor Compute (CPC) is busy."></metric>
|
||||
<metric name="GRBM_CPF_BUSY" block=GRBM event=31 descr="The Command Processor Fetchers (CPF) is busy."></metric>
|
||||
<metric name="GRBM_UTCL2_BUSY" block=GRBM event=34 descr="The Unified Translation Cache Level-2 (UTCL2) block is busy."></metric>
|
||||
<metric name="GRBM_EA_BUSY" block=GRBM event=35 descr="The Efficiency Arbiter (EA) block is busy."></metric>
|
||||
<metric name="CPC_ME1_BUSY_FOR_PACKET_DECODE" block=CPC event=13 descr="Me1 busy for packet decode."></metric>
|
||||
<metric name="CPC_UTCL1_STALL_ON_TRANSLATION" block=CPC event=24 descr="One of the UTCL1s is stalled waiting on translation, XNACK or PENDING response."></metric>
|
||||
<metric name="CPC_CPC_STAT_BUSY" block=CPC event=25 descr="CPC Busy."></metric>
|
||||
<metric name="CPC_CPC_STAT_IDLE" block=CPC event=26 descr="CPC Idle."></metric>
|
||||
<metric name="CPC_CPC_STAT_STALL" block=CPC event=27 descr="CPC Stalled."></metric>
|
||||
<metric name="CPC_CPC_TCIU_BUSY" block=CPC event=28 descr="CPC TCIU interface Busy."></metric>
|
||||
<metric name="CPC_CPC_TCIU_IDLE" block=CPC event=29 descr="CPC TCIU interface Idle."></metric>
|
||||
<metric name="CPC_CPC_UTCL2IU_BUSY" block=CPC event=30 descr="CPC UTCL2 interface Busy."></metric>
|
||||
<metric name="CPC_CPC_UTCL2IU_IDLE" block=CPC event=31 descr="CPC UTCL2 interface Idle."></metric>
|
||||
<metric name="CPC_CPC_UTCL2IU_STALL" block=CPC event=32 descr="CPC UTCL2 interface Stalled waiting on Free, Tags or Translation."></metric>
|
||||
<metric name="CPC_ME1_DC0_SPI_BUSY" block=CPC event=33 descr="CPC Me1 Processor Busy."></metric>
|
||||
<metric name="CPF_CMP_UTCL1_STALL_ON_TRANSLATION" block=CPF event=20 descr="One of the Compute UTCL1s is stalled waiting on translation, XNACK or PENDING response."></metric>
|
||||
<metric name="CPF_CPF_STAT_BUSY" block=CPF event=23 descr="CPF Busy."></metric>
|
||||
<metric name="CPF_CPF_STAT_IDLE" block=CPF event=24 descr="CPF Idle."></metric>
|
||||
<metric name="CPF_CPF_STAT_STALL" block=CPF event=25 descr="CPF Stalled."></metric>
|
||||
<metric name="CPF_CPF_TCIU_BUSY" block=CPF event=26 descr="CPF TCIU interface Busy."></metric>
|
||||
<metric name="CPF_CPF_TCIU_IDLE" block=CPF event=27 descr="CPF TCIU interface Idle."></metric>
|
||||
<metric name="CPF_CPF_TCIU_STALL" block=CPF event=28 descr="CPF TCIU interface Stalled waiting on Free, Tags."></metric>
|
||||
<metric name="SPI_CSN_WINDOW_VALID" block=SPI event=47 descr="Clock count enabled by perfcounter_start event. Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_CSN_BUSY" block=SPI event=48 descr="Number of clocks with outstanding waves (SPI or SH). Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_CSN_NUM_THREADGROUPS" block=SPI event=49 descr="Number of threadgroups launched. Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_CSN_WAVE" block=SPI event=52 descr="Number of waves. Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_RA_REQ_NO_ALLOC" block=SPI event=79 descr="Arb cycles with requests but no allocation. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_REQ_NO_ALLOC_CSN" block=SPI event=85 descr="Arb cycles with CSn req and no CSn alloc. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_RES_STALL_CSN" block=SPI event=91 descr="Arb cycles with CSn req and no CSn fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_TMP_STALL_CSN" block=SPI event=97 descr="Cycles where csn wants to req but does not fit in temp space."></metric>
|
||||
<metric name="SPI_RA_WAVE_SIMD_FULL_CSN" block=SPI event=103 descr="Sum of SIMD where WAVE can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_VGPR_SIMD_FULL_CSN" block=SPI event=109 descr="Sum of SIMD where VGPR can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_SGPR_SIMD_FULL_CSN" block=SPI event=115 descr="Sum of SIMD where SGPR can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_LDS_CU_FULL_CSN" block=SPI event=120 descr="Sum of CU where LDS can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_BAR_CU_FULL_CSN" block=SPI event=123 descr="Sum of CU where BARRIER can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_BULKY_CU_FULL_CSN" block=SPI event=125 descr="Sum of CU where BULKY can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_TGLIM_CU_FULL_CSN" block=SPI event=127 descr="Cycles where csn wants to req but all CU are at tg_limit"></metric>
|
||||
<metric name="SPI_RA_WVLIM_STALL_CSN" block=SPI event=133 descr="Number of clocks csn is stalled due to WAVE LIMIT."></metric>
|
||||
<metric name="SPI_SWC_CSC_WR" block=SPI event=189 descr="Number of clocks to write CSC waves to SGPRs (need to multiply this value by 4) Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_VWC_CSC_WR" block=SPI event=195 descr="Number of clocks to write CSC waves to VGPRs (need to multiply this value by 4) Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SQ_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1. Only accumulates once every 4 cycles."></metric>
|
||||
<metric name="SQ_CYCLES" block=SQ event=2 descr="Clock cycles. (nondeterministic, per-simd, global)"></metric>
|
||||
<metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. (nondeterministic, per-simd, global)"></metric>
|
||||
<metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_LEVEL_WAVES" block=SQ event=5 descr="Track the number of waves. Set ACCUM_PREV for the next counter to use this. (level, per-simd, global)"></metric>
|
||||
<metric name="SQ_WAVES_EQ_64" block=SQ event=6 descr="Count number of waves with exactly 64 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_64" block=SQ event=7 descr="Count number of waves with <64 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_48" block=SQ event=8 descr="Count number of waves with <48 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_32" block=SQ event=9 descr="Count number of waves sent <32 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_16" block=SQ event=10 descr="Count number of waves sent <16 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_BUSY_CU_CYCLES" block=SQ event=13 descr="Count quad-cycles each CU is busy. (nondeterministic, per-simd)"></metric>
|
||||
<metric name="SQ_ITEMS" block=SQ event=14 descr="Number of valid items per wave. (per-simd, global)"></metric>
|
||||
<metric name="SQ_INSTS" block=SQ event=25 descr="Number of instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU" block=SQ event=26 descr="Number of VALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_ADD_F16" block=SQ event=27 descr="Number of VALU ADD/SUB instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MUL_F16" block=SQ event=28 descr="Number of VALU MUL instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_FMA_F16" block=SQ event=29 descr="Number of VALU FMA/MAD instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_TRANS_F16" block=SQ event=30 descr="Number of VALU transcendental instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_ADD_F32" block=SQ event=31 descr="Number of VALU ADD/SUB instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MUL_F32" block=SQ event=32 descr="Number of VALU MUL instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_FMA_F32" block=SQ event=33 descr="Number of VALU FMA/MAD instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_TRANS_F32" block=SQ event=34 descr="Number of VALU transcendental instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_ADD_F64" block=SQ event=35 descr="Number of VALU ADD/SUB instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MUL_F64" block=SQ event=36 descr="Number of VALU MUL instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_FMA_F64" block=SQ event=37 descr="Number of VALU FMA/MAD instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_TRANS_F64" block=SQ event=38 descr="Number of VALU transcendental instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_INT32" block=SQ event=39 descr="Number of VALU 32-bit integer (signed or unsigned) instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_INT64" block=SQ event=40 descr="Number of VALU 64-bit integer (signed or unsigned) instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_CVT" block=SQ event=41 descr="Number of VALU data conversion instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_I8" block=SQ event=42 descr="Number of VALU V_MFMA_*_I8 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_F16" block=SQ event=43 descr="Number of VALU V_MFMA_*_F16 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_BF16" block=SQ event=44 descr="Number of VALU V_MFMA_*_BF16 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_F32" block=SQ event=45 descr="Number of VALU V_MFMA_*_F32 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_F64" block=SQ event=46 descr="Number of VALU V_MFMA_*_F64 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_I8" block=SQ event=47 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type I8. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_F16" block=SQ event=48 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type F16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_BF16" block=SQ event=49 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type BF16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_F32" block=SQ event=50 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type F32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_F64" block=SQ event=51 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type F64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_MFMA" block=SQ event=52 descr="Number of MFMA instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_WR" block=SQ event=53 descr="Number of VMEM write instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_RD" block=SQ event=54 descr="Number of VMEM read instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM" block=SQ event=55 descr="Number of VMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=56 descr="Number of SALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=57 descr="Number of SMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=58 descr="Number of FLAT instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT_LDS_ONLY" block=SQ event=59 descr="Number of FLAT instructions issued that read/wrote only from/to LDS (only works if EARLY_TA_DONE is enabled). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=60 descr="Number of LDS instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=61 descr="Number of GDS instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_EXP_GDS" block=SQ event=63 descr="Number of EXP and GDS instructions issued, excluding skipped export instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_BRANCH" block=SQ event=64 descr="Number of Branch instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SENDMSG" block=SQ event=65 descr="Number of Sendmsg instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VSKIPPED" block=SQ event=66 descr="Number of vector instructions skipped. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INST_LEVEL_VMEM" block=SQ event=67 descr="Number of in-flight VMEM instructions. Set next counter to ACCUM_PREV and divide by INSTS_VMEM for average latency. Includes FLAT instructions. (per-simd, level, nondeterministic)"></metric>
|
||||
<metric name="SQ_INST_LEVEL_SMEM" block=SQ event=68 descr="Number of in-flight SMEM instructions (*2 load/store; *2 atomic; *2 memtime; *4 wb/inv). Set next counter to ACCUM_PREV and divide by INSTS_SMEM for average latency per smem request. Falls slightly short of total request latency because some fetches are divided into two requests that may finish at different times and this counter collects the average latency of the two. (per-simd, level, nondeterministic)"></metric>
|
||||
<metric name="SQ_INST_LEVEL_LDS" block=SQ event=69 descr="Number of in-flight LDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_LDS for average latency. Includes FLAT instructions. (per-simd, level, nondeterministic)"></metric>
|
||||
<metric name="SQ_VALU_MFMA_BUSY_CYCLES" block=SQ event=72 descr="Number of cycles the MFMA ALU is busy (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_WAVE_CYCLES" block=SQ event=74 descr="Number of wave-cycles spent by waves in the CUs (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_WAIT_ANY" block=SQ event=85 descr="Number of wave-cycles spent waiting for anything (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_WAIT_INST_ANY" block=SQ event=88 descr="Number of wave-cycles spent waiting for any instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_ANY" block=SQ event=96 descr="Number of cycles each wave is working on an instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VMEM" block=SQ event=97 descr="Number of cycles the SQ instruction arbiter is working on a VMEM instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_LDS" block=SQ event=98 descr="Number of cycles the SQ instruction arbiter is working on a LDS instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VALU" block=SQ event=99 descr="Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_SCA" block=SQ event=100 descr="Number of cycles the SQ instruction arbiter is working on a SALU or SMEM instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_EXP_GDS" block=SQ event=101 descr="Number of cycles the SQ instruction arbiter is working on an EXPORT or GDS instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_MISC" block=SQ event=102 descr="Number of cycles the SQ instruction aribter is working on a BRANCH or SENDMSG instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_FLAT" block=SQ event=103 descr="Number of cycles the SQ instruction arbiter is working on a FLAT instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_VMEM_WR" block=SQ event=104 descr="Number of cycles needed to send addr and cmd data for VMEM write instructions. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_VMEM_RD" block=SQ event=105 descr="Number of cycles needed to send addr and cmd data for VMEM read instructions. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SMEM" block=SQ event=111 descr="Number of cycles needed to execute scalar memory reads. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SALU" block=SQ event=112 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_THREAD_CYCLES_VALU" block=SQ event=113 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
|
||||
<metric name="SQ_IFETCH" block=SQ event=115 descr="Number of instruction fetch requests from cache. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_IFETCH_LEVEL" block=SQ event=116 descr="Number of instruction fetch requests from cache. (per-simd, level)"></metric>
|
||||
<metric name="SQ_LDS_BANK_CONFLICT" block=SQ event=121 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
|
||||
<metric name="SQ_LDS_ADDR_CONFLICT" block=SQ event=122 descr="Number of cycles LDS is stalled by address conflicts. (emulated,nondeterministic)"></metric>
|
||||
<metric name="SQ_LDS_UNALIGNED_STALL" block=SQ event=123 descr="Number of cycles LDS is stalled processing flat unaligned load/store ops. (emulated)"></metric>
|
||||
<metric name="SQ_LDS_MEM_VIOLATIONS" block=SQ event=124 descr="Number of threads that have a memory violation in the LDS.(emulated)"></metric>
|
||||
<metric name="SQ_LDS_ATOMIC_RETURN" block=SQ event=125 descr="Number of atomic return cycles in LDS. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_LDS_IDX_ACTIVE" block=SQ event=126 descr="Number of cycles LDS is used for indexed (non-direct,non-interpolation) operations. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_ACCUM_PREV_HIRES" block=SQ event=185 descr="For counter N, increment by the value of counter N-1."></metric>
|
||||
<metric name="SQ_WAVES_RESTORED" block=SQ event=186 descr="Count number of context-restored waves sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_SAVED" block=SQ event=187 descr="Count number of context-saved waves. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM_NORM" block=SQ event=188 descr="Number of SMEM instructions issued normalized to match smem_level (*2 load/store; *2 atomic; *2 memtime; *4 wb/inv). (per-simd, emulated)"></metric>
|
||||
<metric name="SQC_DCACHE_INPUT_VALID_READYB" block=SQ event=260 descr="Input stalled by SQC (per-SQ, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_TC_REQ" block=SQ event=262 descr="Total number of TC requests that were issued by instruction and constant caches. (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_INST_REQ" block=SQ event=263 descr="Number of insruction requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_DATA_READ_REQ" block=SQ event=264 descr="Number of data read requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_DATA_WRITE_REQ" block=SQ event=265 descr="Number of data write requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_DATA_ATOMIC_REQ" block=SQ event=266 descr="Number of data atomic requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_STALL" block=SQ event=267 descr="Valid request stalled TC request interface (no-credits). (No-Masking, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_ICACHE_REQ" block=SQ event=270 descr="Number of requests. (per-SQ, per-Bank)"></metric>
|
||||
<metric name="SQC_ICACHE_HITS" block=SQ event=271 descr="Number of cache hits. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_ICACHE_MISSES" block=SQ event=272 descr="Number of cache misses, includes uncached requests. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_ICACHE_MISSES_DUPLICATE" block=SQ event=273 descr="Number of misses that were duplicates (access to a non-resident, miss pending CL). (per-SQ, per-Bank, nondeterministic)" ></metric>
|
||||
<metric name="SQC_DCACHE_REQ" block=SQ event=290 descr="Number of requests (post-bank-serialization). (per-SQ, per-Bank)"></metric>
|
||||
<metric name="SQC_DCACHE_HITS" block=SQ event=291 descr="Number of cache hits. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_DCACHE_MISSES" block=SQ event=292 descr="Number of cache misses, includes uncached requests. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_DCACHE_MISSES_DUPLICATE" block=SQ event=293 descr="Number of misses that were duplicates (access to a non-resident, miss pending CL). (per-SQ, per-Bank, nondeterministic)" ></metric>
|
||||
<metric name="SQC_DCACHE_ATOMIC" block=SQ event=298 descr="Number of atomic requests. (per-SQ, per-Bank)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_1" block=SQ event=323 descr="Number of constant cache 1 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_2" block=SQ event=324 descr="Number of constant cache 2 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_4" block=SQ event=325 descr="Number of constant cache 4 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_8" block=SQ event=326 descr="Number of constant cache 8 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_16" block=SQ event=327 descr="Number of constant cache 16 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="TA_TA_BUSY" block=TA event=15 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_TOTAL_WAVEFRONTS" block=TA event=32 descr="Total number of wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_WAVEFRONTS" block=TA event=44 descr="Number of buffer wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_READ_WAVEFRONTS" block=TA event=45 descr="Number of buffer read wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_WRITE_WAVEFRONTS" block=TA event=46 descr="Number of buffer write wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_ATOMIC_WAVEFRONTS" block=TA event=47 descr="Number of buffer atomic wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_TOTAL_CYCLES" block=TA event=49 descr="Number of buffer cycles issued to TC."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_READ_CYCLES" block=TA event=52 descr="Number of buffer coalesced read cycles issued to TC."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_WRITE_CYCLES" block=TA event=53 descr="Number of buffer coalesced write cycles issued to TC."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TC_CYCLES" block=TA event=54 descr="Number of cycles addr path stalled by TC. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TD_CYCLES" block=TA event=55 descr="Number of cycles addr path stalled by TD. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_DATA_STALLED_BY_TC_CYCLES" block=TA event=56 descr="Number of cycles data path stalled by TC. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_FLAT_WAVEFRONTS" block=TA event=100 descr="Number of flat opcode wavfronts processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS" block=TA event=101 descr="Number of flat opcode reads processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS" block=TA event=102 descr="Number of flat opcode writes processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_ATOMIC_WAVEFRONTS" block=TA event=103 descr="Number of flat opcode atomics processed by the TA."></metric>
|
||||
<metric name="TD_TD_BUSY" block=TD event=1 descr="TD is processing or waiting for data. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TD_TC_STALL" block=TD event=15 descr="TD is stalled waiting for TC data."></metric>
|
||||
<metric name="TD_SPI_STALL" block=TD event=18 descr="TD is stalled SPI vinit"></metric>
|
||||
<metric name="TD_LOAD_WAVEFRONT" block=TD event=25 descr="Count the wavefronts with opcode = load, include atomics and store."></metric>
|
||||
<metric name="TD_ATOMIC_WAVEFRONT" block=TD event=26 descr="Count the wavefronts with opcode = atomic."></metric>
|
||||
<metric name="TD_STORE_WAVEFRONT" block=TD event=27 descr="Count the wavefronts with opcode = store."></metric>
|
||||
<metric name="TD_COALESCABLE_WAVEFRONT" block=TD event=32 descr="Count wavefronts that TA finds coalescable."></metric>
|
||||
<metric name="TCP_GATE_EN1" block=TCP event=0 descr="TCP interface clocks are turned on. Not Windowed."></metric>
|
||||
<metric name="TCP_GATE_EN2" block=TCP event=1 descr="TCP core clocks are turned on. Not Windowed."></metric>
|
||||
<metric name="TCP_TD_TCP_STALL_CYCLES" block=TCP event=7 descr="TD stalls TCP"></metric>
|
||||
<metric name="TCP_TCR_TCP_STALL_CYCLES" block=TCP event=8 descr="TCR stalls TCP_TCR_req interface"></metric>
|
||||
<metric name="TCP_READ_TAGCONFLICT_STALL_CYCLES" block=TCP event=11 descr="Tagram conflict stall on a read"></metric>
|
||||
<metric name="TCP_WRITE_TAGCONFLICT_STALL_CYCLES" block=TCP event=12 descr="Tagram conflict stall on a write"></metric>
|
||||
<metric name="TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES" block=TCP event=13 descr="Tagram conflict stall on an atomic"></metric>
|
||||
<metric name="TCP_PENDING_STALL_CYCLES" block=TCP event=22 descr="Stall due to data pending from L2"></metric>
|
||||
<metric name="TCP_TA_TCP_STATE_READ" block=TCP event=27 descr="Number of state reads"></metric>
|
||||
<metric name="TCP_VOLATILE" block=TCP event=28 descr="Total number of L1 volatile pixels/buffers from TA"></metric>
|
||||
<metric name="TCP_TOTAL_ACCESSES" block=TCP event=29 descr="Total number of pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_READ+TCP_PERF_SEL_TOTAL_NONREAD"></metric>
|
||||
<metric name="TCP_TOTAL_READ" block=TCP event=30 descr="Total number of read pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_HIT_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_EVICT_READ"></metric>
|
||||
<metric name="TCP_TOTAL_WRITE" block=TCP event=32 descr="Total number of local write pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_MISS_LRU_WRITE+ TCP_PERF_SEL_TOTAL_MISS_EVICT_WRITE"></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITH_RET" block=TCP event=38 descr="Total number of atomic with return pixels/buffers from TA"></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITHOUT_RET" block=TCP event=39 descr="Total number of atomic without return pixels/buffers from TA"></metric>
|
||||
<metric name="TCP_TOTAL_WRITEBACK_INVALIDATES" block=TCP event=45 descr="Total number of cache invalidates. Equals TCP_PERF_SEL_TOTAL_WBINVL1+ TCP_PERF_SEL_TOTAL_WBINVL1_VOL+ TCP_PERF_SEL_CP_TCP_INVALIDATE+ TCP_PERF_SEL_SQ_TCP_INVALIDATE_VOL. Not Windowed."></metric>
|
||||
<metric name="TCP_UTCL1_REQUEST" block=TCP event=47 descr="Total CLIENT_UTCL1 NORMAL requests"></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_MISS" block=TCP event=48 descr="Total utcl1 translation misses"></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_HIT" block=TCP event=49 descr="Total utcl1 translation hits"></metric>
|
||||
<metric name="TCP_UTCL1_PERMISSION_MISS" block=TCP event=50 descr="Total utcl1 permission misses"></metric>
|
||||
<metric name="TCP_TOTAL_CACHE_ACCESSES" block=TCP event=60 descr="Count of total cache line (tag) accesses (includes hits and misses)."></metric>
|
||||
<metric name="TCP_TCP_LATENCY" block=TCP event=65 descr="Total TCP wave latency (from first clock of wave entering to first clock of wave leaving), divide by TA_TCP_STATE_READ to avg wave latency"></metric>
|
||||
<metric name="TCP_TCC_READ_REQ_LATENCY" block=TCP event=66 descr="Total TCP->TCC request latency for reads and atomics with return. Not Windowed."></metric>
|
||||
<metric name="TCP_TCC_WRITE_REQ_LATENCY" block=TCP event=67 descr="Total TCP->TCC request latency for writes and atomics without return. Not Windowed."></metric>
|
||||
<metric name="TCP_TCC_READ_REQ" block=TCP event=69 descr="Total read requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_WRITE_REQ" block=TCP event=70 descr="Total write requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITH_RET_REQ" block=TCP event=71 descr="Total atomic with return requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITHOUT_RET_REQ" block=TCP event=72 descr="Total atomic without return requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_NC_READ_REQ" block=TCP event=75 descr="Total read requests with NC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_NC_WRITE_REQ" block=TCP event=76 descr="Total write requests with NC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_NC_ATOMIC_REQ" block=TCP event=77 descr="Total atomic requests with NC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_UC_READ_REQ" block=TCP event=78 descr="Total read requests with UC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_UC_WRITE_REQ" block=TCP event=79 descr="Total write requests with UC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_UC_ATOMIC_REQ" block=TCP event=80 descr="Total atomic requests with UC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_CC_READ_REQ" block=TCP event=81 descr="Total write requests with CC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_CC_WRITE_REQ" block=TCP event=82 descr="Total write requests with CC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_CC_ATOMIC_REQ" block=TCP event=83 descr="Total atomic requests with CC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_RW_READ_REQ" block=TCP event=85 descr="Total write requests with RW mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_RW_WRITE_REQ" block=TCP event=86 descr="Total write requests with RW mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_RW_ATOMIC_REQ" block=TCP event=87 descr="Total atomic requests with RW mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCA_CYCLE" block=TCA event=1 descr="Number of cycles. Not windowable."></metric>
|
||||
<metric name="TCA_BUSY" block=TCA event=2 descr="Number of cycles we have a request pending. Not windowable."></metric>
|
||||
<metric name="TCC_CYCLE" block=TCC event=1 descr="Number of cycles. Not windowable."></metric>
|
||||
<metric name="TCC_BUSY" block=TCC event=2 descr="Number of cycles we have a request pending. Not windowable."></metric>
|
||||
<metric name="TCC_REQ" block=TCC event=3 descr="Number of requests of all types. This is measured at the tag block. This may be more than the number of requests arriving at the TCC, but it is a good indication of the total amount of work that needs to be performed."></metric>
|
||||
<metric name="TCC_STREAMING_REQ" block=TCC event=4 descr="Number of streaming requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_NC_REQ" block=TCC event=5 descr="The number of noncoherently cached requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_UC_REQ" block=TCC event=6 descr="The number of uncached requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_CC_REQ" block=TCC event=7 descr="The number of coherently cached requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_RW_REQ" block=TCC event=8 descr="The number of RW requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_PROBE" block=TCC event=9 descr="Number of probe requests. Not windowable."></metric>
|
||||
<metric name="TCC_PROBE_ALL" block=TCC event=10 descr="Number of external probe requests with with EA_TCC_preq_all== 1. Not windowable."></metric>
|
||||
<metric name="TCC_READ" block=TCC event=12 descr="Number of read requests. Compressed reads are included in this, but metadata reads are not included."></metric>
|
||||
<metric name="TCC_WRITE" block=TCC event=13 descr="Number of write requests."></metric>
|
||||
<metric name="TCC_ATOMIC" block=TCC event=14 descr="Number of atomic requests of all types."></metric>
|
||||
<metric name="TCC_HIT" block=TCC event=17 descr="Number of cache hits."></metric>
|
||||
<metric name="TCC_MISS" block=TCC event=19 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="TCC_WRITEBACK" block=TCC event=22 descr="Number of lines written back to main memory. This includes writebacks of dirty lines and uncached write/atomic requests."></metric>
|
||||
<metric name="TCC_EA_WRREQ" block=TCC event=26 descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands."></metric>
|
||||
<metric name="TCC_EA_WRREQ_64B" block=TCC event=27 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="TCC_EA_WR_UNCACHED_32B" block=TCC event=29 descr="Number of 32-byte write/atomic going over the TC_EA_wrreq interface due to uncached traffic. Note that CC mtypes can produce uncached requests, and those are included in this. A 64-byte request will be counted as 2"></metric>
|
||||
<metric name="TCC_EA_WRREQ_STALL" block=TCC event=30 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="TCC_EA_WRREQ_IO_CREDIT_STALL" block=TCC event=31 descr="Number of cycles a EA write request was stalled because the interface was out of IO credits."></metric>
|
||||
<metric name="TCC_EA_WRREQ_GMI_CREDIT_STALL" block=TCC event=32 descr="Number of cycles a EA write request was stalled because the interface was out of GMI credits."></metric>
|
||||
<metric name="TCC_EA_WRREQ_DRAM_CREDIT_STALL" block=TCC event=33 descr="Number of cycles a EA write request was stalled because the interface was out of DRAM credits."></metric>
|
||||
<metric name="TCC_TOO_MANY_EA_WRREQS_STALL" block=TCC event=34 descr="Number of cycles the TCC could not send a EA write request because it already reached its maximum number of pending EA write requests."></metric>
|
||||
<metric name="TCC_EA_WRREQ_LEVEL" block=TCC event=35 descr="The sum of the number of EA write requests in flight. This is primarily meant for measure average EA write latency. Average write latency = TCC_PERF_SEL_EA_WRREQ_LEVEL/TCC_PERF_SEL_EA_WRREQ."></metric>
|
||||
<metric name="TCC_EA_ATOMIC" block=TCC event=36 descr="Number of transactions going over the TC_EA_wrreq interface that are actually atomic requests."></metric>
|
||||
<metric name="TCC_EA_ATOMIC_LEVEL" block=TCC event=37 descr="The sum of the number of EA atomics in flight. This is primarily meant for measure average EA atomic latency. Average atomic latency = TCC_PERF_SEL_EA_WRREQ_ATOMIC_LEVEL/TCC_PERF_SEL_EA_WRREQ_ATOMIC."></metric>
|
||||
<metric name="TCC_EA_RDREQ" block=TCC event=38 descr="Number of TCC/EA read requests (either 32-byte or 64-byte)"></metric>
|
||||
<metric name="TCC_EA_RDREQ_32B" block=TCC event=39 descr="Number of 32-byte TCC/EA read requests"></metric>
|
||||
<metric name="TCC_EA_RD_UNCACHED_32B" block=TCC event=40 descr="Number of 32-byte TCC/EA read due to uncached traffic. A 64-byte request will be counted as 2"></metric>
|
||||
<metric name="TCC_EA_RDREQ_IO_CREDIT_STALL" block=TCC event=41 descr="Number of cycles there was a stall because the read request interface was out of IO credits. Stalls occur regardless of whether a read needed to be performed or not."></metric>
|
||||
<metric name="TCC_EA_RDREQ_GMI_CREDIT_STALL" block=TCC event=42 descr="Number of cycles there was a stall because the read request interface was out of GMI credits. Stalls occur regardless of whether a read needed to be performed or not."></metric>
|
||||
<metric name="TCC_EA_RDREQ_DRAM_CREDIT_STALL" block=TCC event=43 descr="Number of cycles there was a stall because the read request interface was out of DRAM credits. Stalls occur regardless of whether a read needed to be performed or not."></metric>
|
||||
<metric name="TCC_EA_RDREQ_LEVEL" block=TCC event=44 descr="The sum of the number of TCC/EA read requests in flight. This is primarily meant for measure average EA read latency. Average read latency = TCC_PERF_SEL_EA_RDREQ_LEVEL/TCC_PERF_SEL_EA_RDREQ."></metric>
|
||||
<metric name="TCC_TAG_STALL" block=TCC event=45 descr="Number of cycles the normal request pipeline in the tag was stalled for any reason. Normally, stalls of this nature are measured exactly from one point the pipeline, but that is not the case for this counter. Probes can stall the pipeline at a variety of places, and there is no single point that can reasonably measure the total stalls accurately."></metric>
|
||||
<metric name="TCC_NORMAL_WRITEBACK" block=TCC event=68 descr="Number of writebacks due to requests that are not writeback requests."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_WB_WRITEBACK" block=TCC event=73 descr="Number of writebacks due to all TC_OP writeback requests."></metric>
|
||||
<metric name="TCC_NORMAL_EVICT" block=TCC event=74 descr="Number of evictions due to requests that are not invalidate or probe requests."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_INV_EVICT" block=TCC event=80 descr="Number of evictions due to all TC_OP invalidate requests."></metric>
|
||||
<metric name="TCC_EA_RDREQ_DRAM" block=TCC event=102 descr="Number of TCC/EA read requests (either 32-byte or 64-byte) destined for DRAM (MC)."></metric>
|
||||
<metric name="TCC_EA_WRREQ_DRAM" block=TCC event=103 descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
|
||||
</gfx90a>
|
||||
|
||||
<gfx940>
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=96 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES" block=TCP event=6 descr="TCP stalls TA data interface. Now Windowed."></metric>
|
||||
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
|
||||
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
|
||||
<metric name="GRBM_CP_BUSY" block=GRBM event=3 descr="Any of the Command Processor (CPG/CPC/CPF) blocks are busy."></metric>
|
||||
<metric name="GRBM_SPI_BUSY" block=GRBM event=11 descr="Any of the Shader Pipe Interpolators (SPI) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GRBM_TA_BUSY" block=GRBM event=13 descr="Any of the Texture Pipes (TA) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GRBM_TC_BUSY" block=GRBM event=28 descr="Any of the Texture Cache Blocks (TCP/TCI/TCA/TCC) are busy."></metric>
|
||||
<metric name="GRBM_CPC_BUSY" block=GRBM event=30 descr="The Command Processor Compute (CPC) is busy."></metric>
|
||||
<metric name="GRBM_CPF_BUSY" block=GRBM event=31 descr="The Command Processor Fetchers (CPF) is busy."></metric>
|
||||
<metric name="GRBM_UTCL2_BUSY" block=GRBM event=34 descr="The Unified Translation Cache Level-2 (UTCL2) block is busy."></metric>
|
||||
<metric name="GRBM_EA_BUSY" block=GRBM event=35 descr="The Efficiency Arbiter (EA) block is busy."></metric>
|
||||
<metric name="CPC_ME1_BUSY_FOR_PACKET_DECODE" block=CPC event=13 descr="Me1 busy for packet decode."></metric>
|
||||
<metric name="CPC_UTCL1_STALL_ON_TRANSLATION" block=CPC event=24 descr="One of the UTCL1s is stalled waiting on translation, XNACK or PENDING response."></metric>
|
||||
<metric name="CPC_CPC_STAT_BUSY" block=CPC event=25 descr="CPC Busy."></metric>
|
||||
<metric name="CPC_CPC_STAT_IDLE" block=CPC event=26 descr="CPC Idle."></metric>
|
||||
<metric name="CPC_CPC_STAT_STALL" block=CPC event=27 descr="CPC Stalled."></metric>
|
||||
<metric name="CPC_CPC_TCIU_BUSY" block=CPC event=28 descr="CPC TCIU interface Busy."></metric>
|
||||
<metric name="CPC_CPC_TCIU_IDLE" block=CPC event=29 descr="CPC TCIU interface Idle."></metric>
|
||||
<metric name="CPC_CPC_UTCL2IU_BUSY" block=CPC event=30 descr="CPC UTCL2 interface Busy."></metric>
|
||||
<metric name="CPC_CPC_UTCL2IU_IDLE" block=CPC event=31 descr="CPC UTCL2 interface Idle."></metric>
|
||||
<metric name="CPC_CPC_UTCL2IU_STALL" block=CPC event=32 descr="CPC UTCL2 interface Stalled waiting on Free, Tags or Translation."></metric>
|
||||
<metric name="CPC_ME1_DC0_SPI_BUSY" block=CPC event=33 descr="CPC Me1 Processor Busy."></metric>
|
||||
<metric name="CPF_CMP_UTCL1_STALL_ON_TRANSLATION" block=CPF event=20 descr="One of the Compute UTCL1s is stalled waiting on translation, XNACK or PENDING response."></metric>
|
||||
<metric name="CPF_CPF_STAT_BUSY" block=CPF event=23 descr="CPF Busy."></metric>
|
||||
<metric name="CPF_CPF_STAT_IDLE" block=CPF event=24 descr="CPF Idle."></metric>
|
||||
<metric name="CPF_CPF_STAT_STALL" block=CPF event=25 descr="CPF Stalled."></metric>
|
||||
<metric name="CPF_CPF_TCIU_BUSY" block=CPF event=26 descr="CPF TCIU interface Busy."></metric>
|
||||
<metric name="CPF_CPF_TCIU_IDLE" block=CPF event=27 descr="CPF TCIU interface Idle."></metric>
|
||||
<metric name="CPF_CPF_TCIU_STALL" block=CPF event=28 descr="CPF TCIU interface Stalled waiting on Free, Tags."></metric>
|
||||
<metric name="SPI_CSN_WINDOW_VALID" block=SPI event=47 descr="Clock count enabled by perfcounter_start event. Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_CSN_BUSY" block=SPI event=48 descr="Number of clocks with outstanding waves (SPI or SH). Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_CSN_NUM_THREADGROUPS" block=SPI event=49 descr="Number of threadgroups launched. Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_CSN_WAVE" block=SPI event=52 descr="Number of waves. Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_RA_REQ_NO_ALLOC" block=SPI event=79 descr="Arb cycles with requests but no allocation. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_REQ_NO_ALLOC_CSN" block=SPI event=85 descr="Arb cycles with CSn req and no CSn alloc. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_RES_STALL_CSN" block=SPI event=91 descr="Arb cycles with CSn req and no CSn fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_TMP_STALL_CSN" block=SPI event=97 descr="Cycles where csn wants to req but does not fit in temp space."></metric>
|
||||
<metric name="SPI_RA_WAVE_SIMD_FULL_CSN" block=SPI event=103 descr="Sum of SIMD where WAVE can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_VGPR_SIMD_FULL_CSN" block=SPI event=109 descr="Sum of SIMD where VGPR can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_SGPR_SIMD_FULL_CSN" block=SPI event=115 descr="Sum of SIMD where SGPR can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_LDS_CU_FULL_CSN" block=SPI event=120 descr="Sum of CU where LDS can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_BAR_CU_FULL_CSN" block=SPI event=123 descr="Sum of CU where BARRIER can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_BULKY_CU_FULL_CSN" block=SPI event=125 descr="Sum of CU where BULKY can't take csn wave when !fits. Source is RA0"></metric>
|
||||
<metric name="SPI_RA_TGLIM_CU_FULL_CSN" block=SPI event=127 descr="Cycles where csn wants to req but all CU are at tg_limit"></metric>
|
||||
<metric name="SPI_RA_WVLIM_STALL_CSN" block=SPI event=133 descr="Number of clocks csn is stalled due to WAVE LIMIT."></metric>
|
||||
<metric name="SPI_SWC_CSC_WR" block=SPI event=189 descr="Number of clocks to write CSC waves to SGPRs (need to multiply this value by 4) Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SPI_VWC_CSC_WR" block=SPI event=195 descr="Number of clocks to write CSC waves to VGPRs (need to multiply this value by 4) Requires SPI_DEBUG_CNTL.DEBUG_PIPE_SEL to select source, DEBUG_PIPE_SEL = 1, source is CS1; DEBUG_PIPE_SEL = 2, source is CS2; DEBUG_PIPE_SEL = 3, source is CS3; default, source is CS0;"></metric>
|
||||
<metric name="SQ_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1. Only accumulates once every 4 cycles."></metric>
|
||||
<metric name="SQ_CYCLES" block=SQ event=2 descr="Clock cycles. (nondeterministic, per-simd, global)"></metric>
|
||||
<metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. (nondeterministic, per-simd, global)"></metric>
|
||||
<metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_LEVEL_WAVES" block=SQ event=5 descr="Track the number of waves. Set ACCUM_PREV for the next counter to use this. (level, per-simd, global)"></metric>
|
||||
<metric name="SQ_WAVES_EQ_64" block=SQ event=6 descr="Count number of waves with exactly 64 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_64" block=SQ event=7 descr="Count number of waves with <64 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_48" block=SQ event=8 descr="Count number of waves with <48 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_32" block=SQ event=9 descr="Count number of waves sent <32 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_LT_16" block=SQ event=10 descr="Count number of waves sent <16 active threads sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_BUSY_CU_CYCLES" block=SQ event=13 descr="Count quad-cycles each CU is busy. (nondeterministic, per-simd)"></metric>
|
||||
<metric name="SQ_ITEMS" block=SQ event=14 descr="Number of valid items per wave. (per-simd, global)"></metric>
|
||||
<metric name="SQ_INSTS" block=SQ event=25 descr="Number of instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU" block=SQ event=26 descr="Number of VALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_ADD_F16" block=SQ event=27 descr="Number of VALU ADD/SUB instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MUL_F16" block=SQ event=28 descr="Number of VALU MUL instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_FMA_F16" block=SQ event=29 descr="Number of VALU FMA/MAD instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_TRANS_F16" block=SQ event=30 descr="Number of VALU transcendental instructions on float16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_ADD_F32" block=SQ event=31 descr="Number of VALU ADD/SUB instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MUL_F32" block=SQ event=32 descr="Number of VALU MUL instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_FMA_F32" block=SQ event=33 descr="Number of VALU FMA/MAD instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_TRANS_F32" block=SQ event=34 descr="Number of VALU transcendental instructions on float32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_ADD_F64" block=SQ event=35 descr="Number of VALU ADD/SUB instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MUL_F64" block=SQ event=36 descr="Number of VALU MUL instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_FMA_F64" block=SQ event=37 descr="Number of VALU FMA/MAD instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_TRANS_F64" block=SQ event=38 descr="Number of VALU transcendental instructions on float64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_INT32" block=SQ event=39 descr="Number of VALU 32-bit integer (signed or unsigned) instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_INT64" block=SQ event=40 descr="Number of VALU 64-bit integer (signed or unsigned) instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_CVT" block=SQ event=41 descr="Number of VALU data conversion instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_I8" block=SQ event=42 descr="Number of VALU V_MFMA_*_I8 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_F16" block=SQ event=43 descr="Number of VALU V_MFMA_*_F16 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_BF16" block=SQ event=44 descr="Number of VALU V_MFMA_*_BF16 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_F32" block=SQ event=45 descr="Number of VALU V_MFMA_*_F32 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_F64" block=SQ event=46 descr="Number of VALU V_MFMA_*_F64 instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_I8" block=SQ event=49 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type I8. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_F16" block=SQ event=50 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type F16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_BF16" block=SQ event=51 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type BF16. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_F32" block=SQ event=52 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type F32. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VALU_MFMA_MOPS_F64" block=SQ event=53 descr="Number of VALU matrix math operations (add or mul) performed dividied by 512, assuming a full EXEC mask, of data type F64. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_MFMA" block=SQ event=56 descr="Number of MFMA instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_WR" block=SQ event=57 descr="Number of VMEM write instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM_RD" block=SQ event=58 descr="Number of VMEM read instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VMEM" block=SQ event=59 descr="Number of VMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=60 descr="Number of SALU instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=61 descr="Number of SMEM instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=62 descr="Number of FLAT instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=65 descr="Number of LDS instructions issued (including FLAT). (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=66 descr="Number of GDS instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_EXP_GDS" block=SQ event=68 descr="Number of EXP and GDS instructions issued, excluding skipped export instructions. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_BRANCH" block=SQ event=69 descr="Number of Branch instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_SENDMSG" block=SQ event=70 descr="Number of Sendmsg instructions issued. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INSTS_VSKIPPED" block=SQ event=71 descr="Number of vector instructions skipped. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INST_LEVEL_VMEM" block=SQ event=72 descr="Number of in-flight VMEM instructions. Set next counter to ACCUM_PREV and divide by INSTS_VMEM for average latency. Includes FLAT instructions. (per-simd, level, nondeterministic)"></metric>
|
||||
<metric name="SQ_INST_LEVEL_SMEM" block=SQ event=73 descr="Number of in-flight SMEM instructions (*2 load/store; *2 atomic; *2 memtime; *4 wb/inv). Set next counter to ACCUM_PREV and divide by INSTS_SMEM for average latency per smem request. Falls slightly short of total request latency because some fetches are divided into two requests that may finish at different times and this counter collects the average latency of the two. (per-simd, level, nondeterministic)"></metric>
|
||||
<metric name="SQ_INST_LEVEL_LDS" block=SQ event=74 descr="Number of in-flight LDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_LDS for average latency. Includes FLAT instructions. (per-simd, level, nondeterministic)"></metric>
|
||||
<metric name="SQ_VALU_MFMA_BUSY_CYCLES" block=SQ event=77 descr="Number of cycles the MFMA ALU is busy (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_WAVE_CYCLES" block=SQ event=79 descr="Number of wave-cycles spent by waves in the CUs (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_WAIT_ANY" block=SQ event=90 descr="Number of wave-cycles spent waiting for anything (per-simd, nondeterministic). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_WAIT_INST_ANY" block=SQ event=93 descr="Number of wave-cycles spent waiting for any instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_ANY" block=SQ event=101 descr="Number of cycles each wave is working on an instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VMEM" block=SQ event=102 descr="Number of cycles the SQ instruction arbiter is working on a VMEM instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_LDS" block=SQ event=103 descr="Number of cycles the SQ instruction arbiter is working on a LDS instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_VALU" block=SQ event=104 descr="Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_SCA" block=SQ event=105 descr="Number of cycles the SQ instruction arbiter is working on a SALU or SMEM instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_EXP_GDS" block=SQ event=106 descr="Number of cycles the SQ instruction arbiter is working on an EXPORT or GDS instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_MISC" block=SQ event=107 descr="Number of cycles the SQ instruction aribter is working on a BRANCH or SENDMSG instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_ACTIVE_INST_FLAT" block=SQ event=108 descr="Number of cycles the SQ instruction arbiter is working on a FLAT instruction. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_VMEM_WR" block=SQ event=109 descr="Number of cycles needed to send addr and cmd data for VMEM write instructions. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_VMEM_RD" block=SQ event=110 descr="Number of cycles needed to send addr and cmd data for VMEM read instructions. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SMEM" block=SQ event=116 descr="Number of cycles needed to execute scalar memory reads. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_INST_CYCLES_SALU" block=SQ event=117 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated). Units in quad-cycles(4 cycles)"></metric>
|
||||
<metric name="SQ_THREAD_CYCLES_VALU" block=SQ event=118 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
|
||||
<metric name="SQ_IFETCH" block=SQ event=120 descr="Number of instruction fetch requests from cache. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_IFETCH_LEVEL" block=SQ event=121 descr="Number of instruction fetch requests from cache. (per-simd, level)"></metric>
|
||||
<metric name="SQ_LDS_BANK_CONFLICT" block=SQ event=126 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
|
||||
<metric name="SQ_LDS_ADDR_CONFLICT" block=SQ event=127 descr="Number of cycles LDS is stalled by address conflicts. (emulated,nondeterministic)"></metric>
|
||||
<metric name="SQ_LDS_UNALIGNED_STALL" block=SQ event=128 descr="Number of cycles LDS is stalled processing flat unaligned load/store ops. (emulated)"></metric>
|
||||
<metric name="SQ_LDS_MEM_VIOLATIONS" block=SQ event=129 descr="Number of threads that have a memory violation in the LDS.(emulated)"></metric>
|
||||
<metric name="SQ_LDS_ATOMIC_RETURN" block=SQ event=130 descr="Number of atomic return cycles in LDS. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_LDS_IDX_ACTIVE" block=SQ event=131 descr="Number of cycles LDS is used for indexed (non-direct,non-interpolation) operations. (per-simd, emulated)"></metric>
|
||||
<metric name="SQ_ACCUM_PREV_HIRES" block=SQ event=184 descr="For counter N, increment by the value of counter N-1."></metric>
|
||||
<metric name="SQ_WAVES_RESTORED" block=SQ event=185 descr="Count number of context-restored waves sent to SQs. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_WAVES_SAVED" block=SQ event=186 descr="Count number of context-saved waves. (per-simd, emulated, global)"></metric>
|
||||
<metric name="SQ_INSTS_SMEM_NORM" block=SQ event=187 descr="Number of SMEM instructions issued normalized to match smem_level (*2 load/store; *2 atomic; *2 memtime; *4 wb/inv). (per-simd, emulated)"></metric>
|
||||
<metric name="SQC_ICACHE_INPUT_VALID_READYB" block=SQ event=257 descr=" Input stalled by SQC (per-SQ, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_DCACHE_INPUT_VALID_READYB" block=SQ event=260 descr="Input stalled by SQC (per-SQ, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_TC_REQ" block=SQ event=262 descr="Total number of TC requests that were issued by instruction and constant caches. (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_INST_REQ" block=SQ event=263 descr="Number of insruction requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_DATA_READ_REQ" block=SQ event=264 descr="Number of data read requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_DATA_WRITE_REQ" block=SQ event=265 descr="Number of data write requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_DATA_ATOMIC_REQ" block=SQ event=266 descr="Number of data atomic requests to the TC (No-Masking, nondeterministic)"></metric>
|
||||
<metric name="SQC_TC_STALL" block=SQ event=267 descr="Valid request stalled TC request interface (no-credits). (No-Masking, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_ICACHE_BUSY_CYCLES" block=SQ event=269 descr="Clock cycles while cache is reporting that it is busy. (No-Masking, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_ICACHE_REQ" block=SQ event=270 descr="Number of requests. (per-SQ, per-Bank)"></metric>
|
||||
<metric name="SQC_ICACHE_HITS" block=SQ event=271 descr="Number of cache hits. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_ICACHE_MISSES" block=SQ event=272 descr="Number of cache misses, includes uncached requests. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_ICACHE_MISSES_DUPLICATE" block=SQ event=273 descr="Number of misses that were duplicates (access to a non-resident, miss pending CL). (per-SQ, per-Bank, nondeterministic)" ></metric>
|
||||
<metric name="SQC_DCACHE_BUSY_CYCLES" block=SQ event=289 descr=" Clock cycles while cache is reporting that it is busy. (No-Masking, nondeterministic, unwindowed)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ" block=SQ event=290 descr="Number of requests (post-bank-serialization). (per-SQ, per-Bank)"></metric>
|
||||
<metric name="SQC_DCACHE_HITS" block=SQ event=291 descr="Number of cache hits. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_DCACHE_MISSES" block=SQ event=292 descr="Number of cache misses, includes uncached requests. (per-SQ, per-Bank, nondeterministic)"></metric>
|
||||
<metric name="SQC_DCACHE_MISSES_DUPLICATE" block=SQ event=293 descr="Number of misses that were duplicates (access to a non-resident, miss pending CL). (per-SQ, per-Bank, nondeterministic)" ></metric>
|
||||
<metric name="SQC_DCACHE_ATOMIC" block=SQ event=298 descr="Number of atomic requests. (per-SQ, per-Bank)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_1" block=SQ event=323 descr="Number of constant cache 1 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_2" block=SQ event=324 descr="Number of constant cache 2 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_4" block=SQ event=325 descr="Number of constant cache 4 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_8" block=SQ event=326 descr="Number of constant cache 8 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="SQC_DCACHE_REQ_READ_16" block=SQ event=327 descr="Number of constant cache 16 dw read requests. (per-SQ)"></metric>
|
||||
<metric name="TA_TA_BUSY" block=TA event=13 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_TOTAL_WAVEFRONTS" block=TA event=29 descr="Total number of wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_WAVEFRONTS" block=TA event=32 descr="Number of buffer wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_READ_WAVEFRONTS" block=TA event=33 descr="Number of buffer read wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_WRITE_WAVEFRONTS" block=TA event=34 descr="Number of buffer write wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_ATOMIC_WAVEFRONTS" block=TA event=35 descr="Number of buffer atomic wavefronts processed by TA."></metric>
|
||||
<metric name="TA_BUFFER_TOTAL_CYCLES" block=TA event=37 descr="Number of buffer cycles issued to TC."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_READ_CYCLES" block=TA event=40 descr="Number of buffer coalesced read cycles issued to TC."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_WRITE_CYCLES" block=TA event=41 descr="Number of buffer coalesced write cycles issued to TC."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TC_CYCLES" block=TA event=42 descr="Number of cycles addr path stalled by TC. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TD_CYCLES" block=TA event=43 descr="Number of cycles addr path stalled by TD. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_DATA_STALLED_BY_TC_CYCLES" block=TA event=44 descr="Number of cycles data path stalled by TC. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_FLAT_WAVEFRONTS" block=TA event=51 descr="Number of flat opcode wavfronts processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS" block=TA event=52 descr="Number of flat opcode reads processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS" block=TA event=53 descr="Number of flat opcode writes processed by the TA."></metric>
|
||||
<metric name="TA_FLAT_ATOMIC_WAVEFRONTS" block=TA event=54 descr="Number of flat opcode atomics processed by the TA."></metric>
|
||||
<metric name="TD_TD_BUSY" block=TD event=1 descr="TD is processing or waiting for data. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TD_TC_STALL" block=TD event=12 descr="TD is stalled waiting for TC data."></metric>
|
||||
<metric name="TD_SPI_STALL" block=TD event=15 descr="TD is stalled SPI vinit"></metric>
|
||||
<metric name="TD_LOAD_WAVEFRONT" block=TD event=16 descr="Count the wavefronts with opcode = load, include atomics and store."></metric>
|
||||
<metric name="TD_ATOMIC_WAVEFRONT" block=TD event=17 descr="Count the wavefronts with opcode = atomic."></metric>
|
||||
<metric name="TD_STORE_WAVEFRONT" block=TD event=18 descr="Count the wavefronts with opcode = store."></metric>
|
||||
<metric name="TD_COALESCABLE_WAVEFRONT" block=TD event=21 descr="Count wavefronts that TA finds coalescable."></metric>
|
||||
<metric name="TCP_GATE_EN1" block=TCP event=0 descr="TCP interface clocks are turned on. Not Windowed."></metric>
|
||||
<metric name="TCP_GATE_EN2" block=TCP event=1 descr="TCP core clocks are turned on. Not Windowed."></metric>
|
||||
<metric name="TCP_TD_TCP_STALL_CYCLES" block=TCP event=7 descr="TD stalls TCP"></metric>
|
||||
<metric name="TCP_TCR_TCP_STALL_CYCLES" block=TCP event=8 descr="TCR stalls TCP_TCR_req interface"></metric>
|
||||
<metric name="TCP_READ_TAGCONFLICT_STALL_CYCLES" block=TCP event=10 descr="Tagram conflict stall on a read"></metric>
|
||||
<metric name="TCP_WRITE_TAGCONFLICT_STALL_CYCLES" block=TCP event=11 descr="Tagram conflict stall on a write"></metric>
|
||||
<metric name="TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES" block=TCP event=12 descr="Tagram conflict stall on an atomic"></metric>
|
||||
<metric name="TCP_PENDING_STALL_CYCLES" block=TCP event=21 descr="Stall due to data pending from L2"></metric>
|
||||
<metric name="TCP_TA_TCP_STATE_READ" block=TCP event=25 descr="Number of state reads"></metric>
|
||||
<metric name="TCP_VOLATILE" block=TCP event=26 descr="Total number of L1 volatile pixels/buffers from TA"></metric>
|
||||
<metric name="TCP_TOTAL_ACCESSES" block=TCP event=27 descr="Total number of pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_READ+TCP_PERF_SEL_TOTAL_NONREAD"></metric>
|
||||
<metric name="TCP_TOTAL_READ" block=TCP event=28 descr="Total number of read pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_HIT_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_EVICT_READ"></metric>
|
||||
<metric name="TCP_TOTAL_WRITE" block=TCP event=30 descr="Total number of local write pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_MISS_LRU_WRITE+ TCP_PERF_SEL_TOTAL_MISS_EVICT_WRITE"></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITH_RET" block=TCP event=36 descr="Total number of atomic with return pixels/buffers from TA"></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITHOUT_RET" block=TCP event=37 descr="Total number of atomic without return pixels/buffers from TA"></metric>
|
||||
<metric name="TCP_TOTAL_WRITEBACK_INVALIDATES" block=TCP event=43 descr="Total number of cache invalidates. Equals TCP_PERF_SEL_TOTAL_WBINVL1+ TCP_PERF_SEL_TOTAL_WBINVL1_VOL+ TCP_PERF_SEL_CP_TCP_INVALIDATE+ TCP_PERF_SEL_SQ_TCP_INVALIDATE_VOL. Not Windowed."></metric>
|
||||
<metric name="TCP_UTCL1_REQUEST" block=TCP event=45 descr="Total CLIENT_UTCL1 NORMAL requests"></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_MISS" block=TCP event=47 descr="Total utcl1 translation misses"></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_HIT" block=TCP event=48 descr="Total utcl1 translation hits"></metric>
|
||||
<metric name="TCP_UTCL1_PERMISSION_MISS" block=TCP event=49 descr="Total utcl1 permission misses"></metric>
|
||||
<metric name="TCP_TOTAL_CACHE_ACCESSES" block=TCP event=60 descr="Count of total cache line (tag) accesses (includes hits and misses)."></metric>
|
||||
<metric name="TCP_TCC_READ_REQ" block=TCP event=65 descr="Total read requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_WRITE_REQ" block=TCP event=66 descr="Total write requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITH_RET_REQ" block=TCP event=67 descr="Total atomic with return requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITHOUT_RET_REQ" block=TCP event=68 descr="Total atomic without return requests from TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_NC_READ_REQ" block=TCP event=71 descr="Total read requests with NC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_NC_WRITE_REQ" block=TCP event=72 descr="Total write requests with NC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_NC_ATOMIC_REQ" block=TCP event=73 descr="Total atomic requests with NC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_UC_READ_REQ" block=TCP event=74 descr="Total read requests with UC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_UC_WRITE_REQ" block=TCP event=75 descr="Total write requests with UC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_UC_ATOMIC_REQ" block=TCP event=76 descr="Total atomic requests with UC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_CC_READ_REQ" block=TCP event=77 descr="Total write requests with CC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_CC_WRITE_REQ" block=TCP event=78 descr="Total write requests with CC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_CC_ATOMIC_REQ" block=TCP event=79 descr="Total atomic requests with CC mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_RW_READ_REQ" block=TCP event=80 descr="Total write requests with RW mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_RW_WRITE_REQ" block=TCP event=81 descr="Total write requests with RW mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCP_TCC_RW_ATOMIC_REQ" block=TCP event=82 descr="Total atomic requests with RW mtype from this TCP to all TCCs"></metric>
|
||||
<metric name="TCA_CYCLE" block=TCA event=1 descr="Number of cycles. Not windowable."></metric>
|
||||
<metric name="TCA_BUSY" block=TCA event=2 descr="Number of cycles we have a request pending. Not windowable."></metric>
|
||||
<metric name="TCC_CYCLE" block=TCC event=1 descr="Number of cycles. Not windowable."></metric>
|
||||
<metric name="TCC_BUSY" block=TCC event=2 descr="Number of cycles we have a request pending. Not windowable."></metric>
|
||||
<metric name="TCC_REQ" block=TCC event=3 descr="Number of requests of all types. This is measured at the tag block. This may be more than the number of requests arriving at the TCC, but it is a good indication of the total amount of work that needs to be performed."></metric>
|
||||
<metric name="TCC_STREAMING_REQ" block=TCC event=4 descr="Number of streaming requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_NC_REQ" block=TCC event=5 descr="The number of noncoherently cached requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_UC_REQ" block=TCC event=6 descr="The number of uncached requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_CC_REQ" block=TCC event=7 descr="The number of coherently cached requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_RW_REQ" block=TCC event=8 descr="The number of RW requests. This is measured at the tag block."></metric>
|
||||
<metric name="TCC_PROBE" block=TCC event=9 descr="Number of probe requests. Not windowable."></metric>
|
||||
<metric name="TCC_PROBE_ALL" block=TCC event=10 descr="Number of external probe requests with with EA_TCC_preq_all== 1. Not windowable."></metric>
|
||||
<metric name="TCC_INTERNAL_PROBE" block=TCC event=11 descr="Number of self-probes spawned by TCC for CC writes/atomic operations. Not windowable."></metric>
|
||||
<metric name="TCC_READ" block=TCC event=12 descr="Number of read requests. Compressed reads are included in this, but metadata reads are not included."></metric>
|
||||
<metric name="TCC_WRITE" block=TCC event=13 descr="Number of write requests."></metric>
|
||||
<metric name="TCC_ATOMIC" block=TCC event=14 descr="Number of atomic requests of all types."></metric>
|
||||
<metric name="TCC_HIT" block=TCC event=17 descr="Number of cache hits."></metric>
|
||||
<metric name="TCC_MISS" block=TCC event=19 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="TCC_WRITEBACK" block=TCC event=22 descr="Number of lines written back to main memory. This includes writebacks of dirty lines and uncached write/atomic requests."></metric>
|
||||
<metric name="TCC_EA0_WRREQ" block=TCC event=26 descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_64B" block=TCC event=27 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_PROBE_COMMAND" block=TCC event=28 descr="Number of probe commands going over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="TCC_EA0_WR_UNCACHED_32B" block=TCC event=29 descr="Number of 32-byte write/atomic going over the TC_EA_wrreq interface due to uncached traffic. Note that CC mtypes can produce uncached requests, and those are included in this. A 64-byte request will be counted as 2"></metric>
|
||||
<metric name="TCC_EA0_WRREQ_STALL" block=TCC event=30 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_IO_CREDIT_STALL" block=TCC event=31 descr="Number of cycles a EA write request was stalled because the interface was out of IO credits."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_GMI_CREDIT_STALL" block=TCC event=32 descr="Number of cycles a EA write request was stalled because the interface was out of GMI credits."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_DRAM_CREDIT_STALL" block=TCC event=33 descr="Number of cycles a EA write request was stalled because the interface was out of DRAM credits."></metric>
|
||||
<metric name="TCC_TOO_MANY_EA_WRREQS_STALL" block=TCC event=34 descr="Number of cycles the TCC could not send a EA write request because it already reached its maximum number of pending EA write requests."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_LEVEL" block=TCC event=35 descr="The sum of the number of EA write requests in flight. This is primarily meant for measure average EA write latency. Average write latency = TCC_PERF_SEL_EA_WRREQ_LEVEL/TCC_PERF_SEL_EA_WRREQ."></metric>
|
||||
<metric name="TCC_EA0_ATOMIC" block=TCC event=36 descr="Number of transactions going over the TC_EA_wrreq interface that are actually atomic requests."></metric>
|
||||
<metric name="TCC_EA0_ATOMIC_LEVEL" block=TCC event=37 descr="The sum of the number of EA atomics in flight. This is primarily meant for measure average EA atomic latency. Average atomic latency = TCC_PERF_SEL_EA_WRREQ_ATOMIC_LEVEL/TCC_PERF_SEL_EA_WRREQ_ATOMIC."></metric>
|
||||
<metric name="TCC_EA0_RDREQ" block=TCC event=38 descr="Number of TCC/EA read requests (either 32-byte or 64-byte)"></metric>
|
||||
<metric name="TCC_EA0_RDREQ_32B" block=TCC event=39 descr="Number of 32-byte TCC/EA read requests"></metric>
|
||||
<metric name="TCC_EA0_RD_UNCACHED_32B" block=TCC event=40 descr="Number of 32-byte TCC/EA read due to uncached traffic. A 64-byte request will be counted as 2"></metric>
|
||||
<metric name="TCC_EA0_RDREQ_IO_CREDIT_STALL" block=TCC event=41 descr="Number of cycles there was a stall because the read request interface was out of IO credits. Stalls occur regardless of whether a read needed to be performed or not."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_GMI_CREDIT_STALL" block=TCC event=42 descr="Number of cycles there was a stall because the read request interface was out of GMI credits. Stalls occur regardless of whether a read needed to be performed or not."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_DRAM_CREDIT_STALL" block=TCC event=43 descr="Number of cycles there was a stall because the read request interface was out of DRAM credits. Stalls occur regardless of whether a read needed to be performed or not."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_LEVEL" block=TCC event=44 descr="The sum of the number of TCC/EA read requests in flight. This is primarily meant for measure average EA read latency. Average read latency = TCC_PERF_SEL_EA_RDREQ_LEVEL/TCC_PERF_SEL_EA_RDREQ."></metric>
|
||||
<metric name="TCC_TAG_STALL" block=TCC event=45 descr="Number of cycles the normal request pipeline in the tag was stalled for any reason. Normally, stalls of this nature are measured exactly from one point the pipeline, but that is not the case for this counter. Probes can stall the pipeline at a variety of places, and there is no single point that can reasonably measure the total stalls accurately."></metric>
|
||||
<metric name="TCC_NORMAL_WRITEBACK" block=TCC event=68 descr="Number of writebacks due to requests that are not writeback requests."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_WB_WRITEBACK" block=TCC event=73 descr="Number of writebacks due to all TC_OP writeback requests."></metric>
|
||||
<metric name="TCC_NORMAL_EVICT" block=TCC event=74 descr="Number of evictions due to requests that are not invalidate or probe requests."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_INV_EVICT" block=TCC event=80 descr="Number of evictions due to all TC_OP invalidate requests."></metric>
|
||||
<metric name="TCC_PROBE_EVICT" block=TCC event=81 descr="Number of evictions/invalidations due to probes. Not windowable."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_DRAM" block=TCC event=102 descr="Number of TCC/EA read requests (either 32-byte or 64-byte) destined for DRAM (MC)."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_DRAM" block=TCC event=103 descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
|
||||
</gfx940>
|
||||
|
||||
<gfx941 base="gfx940"></gfx941>
|
||||
<gfx942 base="gfx940"></gfx942>
|
||||
|
||||
<gfx10>
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
|
||||
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
|
||||
<metric name="GRBM_CP_BUSY" block=GRBM event=3 descr="Any of the Command Processor (CPG/CPC/CPF) blocks are busy."></metric>
|
||||
<metric name="GRBM_SPI_BUSY" block=GRBM event=11 descr="Any of the Shader Pipe Interpolators (SPI) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GRBM_TA_BUSY" block=GRBM event=13 descr="Any of the Texture Pipes (TA) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GRBM_GDS_BUSY" block=GRBM event=25 descr="The Global Data Share (GDS) is busy."></metric>
|
||||
<metric name="GRBM_EA_BUSY" block=GRBM event=35 descr="The Efficiency Arbiter (EA) block is busy."></metric>
|
||||
<metric name="GRBM_GL2CC_BUSY" block=GRBM event=40 descr="The GL2CC block is busy."></metric>
|
||||
|
||||
<metric name="GL2C_HIT" block=GL2C event=42 descr="Number of cache hits"></metric>
|
||||
<metric name="GL2C_MISS" block=GL2C event=43 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="GL2C_MC_WRREQ" block=GL2C event=83 descr="Number of transactions (either 32-byte or 64-byte) going over the GL2C_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands"></metric>
|
||||
<metric name="GL2C_EA_WRREQ_64B" block=GL2C event=85 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="GL2C_MC_WRREQ_STALL" block=GL2C event=88 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="GL2C_MC_RDREQ" block=GL2C event=96 descr="Number of GL2C/EA read requests (either 32-byte or 64-byte or 128-byte)."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_32B" block=GL2C event=99 descr="Number of 32-byte GL2C/EA read requests"></metric>
|
||||
<metric name="GL2C_EA_RDREQ_64B" block=GL2C event=100 descr="Number of 64-byte GL2C/EA read requests"></metric>
|
||||
<metric name="GL2C_EA_RDREQ_96B" block=GL2C event=101 descr="Number of 96-byte GL2C/EA read requests"></metric>
|
||||
<metric name="GL2C_EA_RDREQ_128B" block=GL2C event=102 descr="Number of 128-byte GL2C/EA read requests"></metric>
|
||||
|
||||
<metric name="SQ_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1."></metric>
|
||||
<metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></metric>
|
||||
<metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. {emulated, global, C1}"></metric>
|
||||
<metric name="SQ_LEVEL_WAVES" block=SQ event=7 descr="Track the aggregated number of waves over certain period of time, Set next counter to ACCUM_PREV and divide by SQ_PERF_SEL_WAVES for average wave life."></metric>
|
||||
<metric name="SQ_WAVE_CYCLES" block=SQ event=26 descr="Number of clock cycles spent by waves in the SQs. Incremented by # of living (valid) waves each cycle. {nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_WAIT_INST_ANY" block=SQ event=28 descr="Number of clock cycles spent waiting for any instruction issue. In units of cycles. {nondeterministic}"></metric>
|
||||
<metric name="SQ_WAIT_ANY" block=SQ event=37 descr="Number of clock cycles spent waiting for anything. {nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_INSTS_WAVE32" block=SQ event=71 descr="Number of wave32 instructions issued, for flat, lds, valu, tex. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_WAVE32_LDS" block=SQ event=74 descr="Number of wave32 LDS indexed instructions issued. Wave64 may count 1 or 2, depending on what gets issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_WAVE32_VALU" block=SQ event=75 descr="Number of wave32 valu instructions issued. Wave64 may count 1 or 2, depending on what gets issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_WAVE32_INSTS" block=SQ event=84 descr="Number of instructions issued by wave32 waves. Skipped instructions are not counted. {emulated}"></metric>
|
||||
<metric name="SQ_WAVE64_INSTS" block=SQ event=85 descr="Number of instructions issued by wave64 waves. Skipped instructions are not counted. {emulated}"></metric>
|
||||
<metric name="SQ_INST_LEVEL_GDS" block=SQ event=98 descr="Number of in-flight GDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_GDS for average latency. {level, nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_INST_LEVEL_LDS" block=SQ event=99 descr="Number of in-flight LDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_LDS for average latency. Includes FLAT instructions. {level, nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_INST_CYCLES_VMEM" block=SQ event=120 descr="Number of cycles needed to send addr and data for VMEM (lds, buffer, image, flat, scratch, global) instructions, windowed by perf_en. {emulated, C1}"></metric>
|
||||
<metric name="SQC_LDS_BANK_CONFLICT" block=SQ event=285 descr="Number of cycles LDS is stalled by bank conflicts. (emulated, C1)"></metric>
|
||||
<metric name="SQC_LDS_IDX_ACTIVE" block=SQ event=290 descr="Number of cycles LDS is used for indexed (non-direct,non-interpolation) operations. {per-simd, emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_VALU" block=SQ event=64 descr="Number of VALU instructions issued excluding skipped instructions. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=60 descr="Number of SALU instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=61 descr="Number of SMEM instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=57 descr="Number of FLAT instructions issued. {emulated, C2}"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=59 descr="Number of LDS indexed instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=55 descr="Number of GDS instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=31 descr="Number of clock cycles spent waiting for LDS (indexed) instruction issue. In units of cycles. {nondeterministic, C1}"></metric>
|
||||
|
||||
<metric name="TA_TA_BUSY" block=TA event=15 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_FLAT_LOAD_WAVEFRONTS" block=TA event=101 descr=" Number of flat load vec32 packets processed by TA, same as flat_read_wavefronts in earlier IP"></metric>
|
||||
<metric name="TA_FLAT_STORE_WAVEFRONTS" block=TA event=102 descr="Number of flat store vec32 packets processed by TA, same as flat_write_wavefronts in earlier IP"></metric>
|
||||
</gfx10>
|
||||
|
||||
<gfx1010 base="gfx10">
|
||||
</gfx1010>
|
||||
|
||||
<gfx1030 base="gfx10">
|
||||
</gfx1030>
|
||||
|
||||
<gfx1031 base="gfx10">
|
||||
</gfx1031>
|
||||
|
||||
<gfx1032 base="gfx10">
|
||||
</gfx1032>
|
||||
|
||||
<gfx11>
|
||||
<metric name="MAX_WAVE_SIZE" expr=wave_front_size descr="Max wave size constant"></metric>
|
||||
<metric name="SE_NUM" expr=array_count/simd_arrays_per_engine descr="SE_NUM"></metric>
|
||||
<metric name="SIMD_NUM" expr=simd_per_cu/CU_NUM descr="SIMD Number"></metric>
|
||||
<metric name="CU_NUM" expr=cu_per_simd_array*array_count descr="CU_NUM"></metric>
|
||||
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
|
||||
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
|
||||
<metric name="GL2C_HIT" block=GL2C event=42 descr="Number of cache hits"></metric>
|
||||
<metric name="GL2C_MISS" block=GL2C event=43 descr="Number of cache misses. UC reads count as misses."></metric>
|
||||
<metric name="GL2C_MC_WRREQ" block=GL2C event=83 descr="Number of transactions (either 32-byte or 64-byte) going over the GL2C_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands"></metric>
|
||||
<metric name="GL2C_EA_WRREQ_64B" block=GL2C event=85 descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface."></metric>
|
||||
<metric name="GL2C_MC_WRREQ_STALL" block=GL2C event=88 descr="Number of cycles a write request was stalled."></metric>
|
||||
<metric name="GL2C_MC_RDREQ" block=GL2C event=96 descr="Number of GL2C/EA read requests (either 32-byte or 64-byte or 128-byte)."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_32B" block=GL2C event=99 descr="Number of 32-byte GL2C/EA read requests"></metric>
|
||||
<metric name="GL2C_EA_RDREQ_64B" block=GL2C event=100 descr="Number of 64-byte GL2C/EA read requests"></metric>
|
||||
<metric name="GL2C_EA_RDREQ_96B" block=GL2C event=101 descr="Number of 96-byte GL2C/EA read requests"></metric>
|
||||
<metric name="GL2C_EA_RDREQ_128B" block=GL2C event=102 descr="Number of 128-byte GL2C/EA read requests"></metric>
|
||||
<metric name="SQ_ACCUM_PREV" block=SQ event=1 descr="For counter N, increment by the value of counter N-1."></metric>
|
||||
<metric name="SQ_BUSY_CYCLES" block=SQ event=3 descr="Clock cycles while SQ is reporting that it is busy. {nondeterministic, global, C2}"></metric>
|
||||
<metric name="SQ_WAVES" block=SQ event=4 descr="Count number of waves sent to SQs. {emulated, global, C1}"></metric>
|
||||
<metric name="SQ_WAVE_CYCLES" block=SQ event=24 descr="Number of clock cycles spent by waves in the SQs. Incremented by number of living (valid) waves each cycle. {nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_WAIT_INST_ANY" block=SQ event=26 descr="Number of clock-cycles spent waiting for any instruction issue. In units of cycles. (nondeterministic)"></metric>
|
||||
<metric name="SQ_WAIT_ANY" block=SQ event=35 descr="Number of wave-cycles spent waiting for anything (nondeterministic, C1)"></metric>
|
||||
<metric name="SQ_INSTS_WAVE32" block=SQ event=70 descr="Number of wave32 instructions issued, for flat, lds, valu, tex. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_WAVE32_LDS" block=SQ event=72 descr="Number of wave32 LDS indexed instructions issued. Wave64 may count 1 or 2, depending on what gets issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_WAVE32_VALU" block=SQ event=73 descr="Number of wave32 valu instructions issued. Wave64 may count 1 or 2, depending on what gets issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_WAVE32_INSTS" block=SQ event=82 descr="Number of instructions issued by wave32 waves. Skipped instructions are not counted. {emulated}"></metric>
|
||||
<metric name="SQ_WAVE64_INSTS" block=SQ event=83 descr="Number of instructions issued by wave64 waves. Skipped instructions are not counted. {emulated}"></metric>
|
||||
<metric name="SQ_INST_LEVEL_GDS" block=SQ event=87 descr="Number of in-flight GDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_GDS for average latency. {level, nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_INST_LEVEL_LDS" block=SQ event=88 descr="Number of in-flight LDS instructions. Set next counter to ACCUM_PREV and divide by INSTS_LDS for average latency. Includes FLAT instructions. {level, nondeterministic, C1}"></metric>
|
||||
<metric name="SQ_INST_CYCLES_VMEM" block=SQ event=106 descr="Number of cycles needed to send addr and data for VMEM (lds, buffer, image, flat, scratch, global) instructions, windowed by perf_en. {emulated, C1}"></metric>
|
||||
<metric name="SQC_LDS_BANK_CONFLICT" block=SQ event=256 descr="Number of cycles LDS is stalled by bank conflicts. (emulated, C1)"></metric>
|
||||
<metric name="SQC_LDS_IDX_ACTIVE" block=SQ event=261 descr="Number of cycles LDS is used for indexed (non-direct,non-interpolation) operations. {per-simd, emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_VALU" block=SQ event=62 descr="Number of VALU instructions issued excluding skipped instructions. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_SALU" block=SQ event=58 descr="Number of SALU instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_SMEM" block=SQ event=59 descr="Number of SMEM instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_FLAT" block=SQ event=56 descr="Number of FLAT instructions issued. {emulated, C2}"></metric>
|
||||
<metric name="SQ_INSTS_LDS" block=SQ event=57 descr="Number of LDS indexed instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_GDS" block=SQ event=54 descr="Number of GDS instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_TEX_LOAD" block=SQ event=66 descr="Number of buffer load, image load, sample, or atomic (with return) instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_INSTS_TEX_STORE" block=SQ event=67 descr="Number of buffer store, image store, or atomic (without return) instructions issued. {emulated, C1}"></metric>
|
||||
<metric name="SQ_WAIT_INST_LDS" block=SQ event=29 descr="Number of clock cycles spent waiting for LDS (indexed) instruction issue. In units of cycles. {nondeterministic, C1}"></metric>
|
||||
<metric name="TA_TA_BUSY" block=TA event=15 descr="TA block is busy. Perf_Windowing not supported for this counter."></metric>
|
||||
<metric name="TA_BUFFER_LOAD_WAVEFRONTS" block=TA event=45 descr="Number of buffer load vec32 packets processed by TA"></metric>
|
||||
<metric name="TA_BUFFER_STORE_WAVEFRONTS" block=TA event=46 descr="Number of buffer store vec32 packets processed by TA"></metric>
|
||||
</gfx11>
|
||||
|
||||
<gfx1100 base="gfx11">
|
||||
</gfx1100>
|
||||
|
||||
<gfx1101 base="gfx11">
|
||||
</gfx1101>
|
||||
|
||||
<gfx1102 base="gfx11">
|
||||
</gfx1102>
|
||||
@@ -1,581 +0,0 @@
|
||||
<common_derived>
|
||||
# GPUBusy The percentage of time GPU was busy.
|
||||
<metric
|
||||
name="GPUBusy"
|
||||
descr="The percentage of time GPU was busy."
|
||||
expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT
|
||||
></metric>
|
||||
|
||||
# Wavefronts Total wavefronts.
|
||||
<metric
|
||||
name="Wavefronts"
|
||||
descr="Total wavefronts."
|
||||
expr=SQ_WAVES
|
||||
></metric>
|
||||
|
||||
# VALUInsts The average number of vector ALU instructions executed per work-item (affected by flow control).
|
||||
<metric
|
||||
name="VALUInsts"
|
||||
descr="The average number of vector ALU instructions executed per work-item (affected by flow control)."
|
||||
expr=SQ_INSTS_VALU/SQ_WAVES
|
||||
></metric>
|
||||
|
||||
# SALUInsts The average number of scalar ALU instructions executed per work-item (affected by flow control).
|
||||
<metric
|
||||
name="SALUInsts"
|
||||
descr="The average number of scalar ALU instructions executed per work-item (affected by flow control)."
|
||||
expr=SQ_INSTS_SALU/SQ_WAVES
|
||||
></metric>
|
||||
|
||||
# SFetchInsts The average number of scalar fetch instructions from the video memory executed per work-item (affected by flow control).
|
||||
<metric
|
||||
name="SFetchInsts"
|
||||
descr="The average number of scalar fetch instructions from the video memory executed per work-item (affected by flow control)."
|
||||
expr=SQ_INSTS_SMEM/SQ_WAVES
|
||||
></metric>
|
||||
|
||||
# GDSInsts The average number of GDS read or GDS write instructions executed per work item (affected by flow control).
|
||||
<metric
|
||||
name="GDSInsts"
|
||||
descr="The average number of GDS read or GDS write instructions executed per work item (affected by flow control)."
|
||||
expr=SQ_INSTS_GDS/SQ_WAVES
|
||||
></metric>
|
||||
|
||||
# MemUnitBusy The percentage of GPUTime the memory unit is active. The result includes the stall time (MemUnitStalled). This is measured with all extra fetches and writes and any cache or memory effects taken into account. Value range: 0% to 100% (fetch-bound).
|
||||
<metric
|
||||
name="MemUnitBusy"
|
||||
descr="The percentage of GPUTime the memory unit is active. The result includes the stall time (MemUnitStalled). This is measured with all extra fetches and writes and any cache or memory effects taken into account. Value range: 0% to 100% (fetch-bound)."
|
||||
expr=100*reduce(TA_TA_BUSY,max)/GRBM_GUI_ACTIVE/SE_NUM
|
||||
></metric>
|
||||
|
||||
# ALUStalledByLDS The percentage of GPUTime ALU units are stalled by the LDS input queue being full or the output queue being not ready. If there are LDS bank conflicts, reduce them. Otherwise, try reducing the number of LDS accesses if possible. Value range: 0% (optimal) to 100% (bad).
|
||||
<metric
|
||||
name="ALUStalledByLDS"
|
||||
descr="The percentage of GPUTime ALU units are stalled by the LDS input queue being full or the output queue being not ready. If there are LDS bank conflicts, reduce them. Otherwise, try reducing the number of LDS accesses if possible. Value range: 0% (optimal) to 100% (bad)."
|
||||
expr=400*SQ_WAIT_INST_LDS/SQ_WAVES/GRBM_GUI_ACTIVE
|
||||
></metric>
|
||||
|
||||
</common_derived>
|
||||
|
||||
<gfx8 base="common_derived">
|
||||
<metric name="SQ_WAVES_sum" expr=reduce(SQ_WAVES,sum) descr="Count number of waves sent to SQs. (per-simd, emulated, global). Sum over SQ instances."></metric>
|
||||
<metric name="TA_BUSY_avr" expr=reduce(TA_TA_BUSY,avr) descr="TA block is busy. Average over TA instances."></metric>
|
||||
<metric name="TA_BUSY_max" expr=reduce(TA_TA_BUSY,max) descr="TA block is busy. Max over TA instances."></metric>
|
||||
<metric name="TA_BUSY_min" expr=reduce(TA_TA_BUSY,min) descr="TA block is busy. Min over TA instances."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS_sum" expr=reduce(TA_FLAT_READ_WAVEFRONTS,sum) descr="Number of flat opcode reads processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS_sum" expr=reduce(TA_FLAT_WRITE_WAVEFRONTS,sum) descr="Number of flat opcode writes processed by the TA. Sum over TA instances."></metric>
|
||||
|
||||
<metric name="TCC_HIT_sum" expr=reduce(TCC_HIT,sum) descr="Number of cache hits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_MISS_sum" expr=reduce(TCC_MISS,sum) descr="Number of cache misses. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_MC_RDREQ_sum" expr=reduce(TCC_MC_RDREQ,sum) descr="Number of 32-byte reads. Sum over TCC instaces."></metric>
|
||||
<metric name="TCC_MC_WRREQ_sum" expr=reduce(TCC_MC_WRREQ,sum) descr="Number of 32-byte transactions going over the TC_MC_wrreq interface. Sum over TCC instaces."></metric>
|
||||
<metric name="TCC_WRREQ_STALL_max" expr=reduce(TCC_MC_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over TCC instances."></metric>
|
||||
|
||||
<metric name="FETCH_SIZE" expr=(TCC_MC_RDREQ_sum*32)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_SIZE" expr=(TCC_MC_WRREQ_sum*32)/1024 descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_REQ_32B" expr=TCC_MC_WRREQ_sum descr="The total number of 32-byte effective memory writes."></metric>
|
||||
<metric name="VFetchInsts" expr=(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector fetch instructions from the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that fetch from video memory."></metric>
|
||||
<metric name="VWriteInsts" expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."></metric>
|
||||
<metric name="FlatVMemInsts" expr=(SQ_INSTS_FLAT-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of FLAT instructions that read from or write to the video memory executed per work item (affected by flow control). Includes FLAT instructions that read from or write to scratch."></metric>
|
||||
<metric name="LDSInsts" expr=(SQ_INSTS_LDS-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of LDS read or LDS write instructions executed per work item (affected by flow control). Excludes FLAT instructions that read from or write to LDS."></metric>
|
||||
<metric name="FlatLDSInsts" expr=SQ_INSTS_FLAT_LDS_ONLY/SQ_WAVES descr="The average number of FLAT instructions that read or write to LDS executed per work item (affected by flow control)."></metric>
|
||||
<metric name="VALUUtilization" expr=100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE) descr="The percentage of active vector ALU threads in a wave. A lower number can mean either more thread divergence in a wave or that the work-group size is not a multiple of 64. Value range: 0% (bad), 100% (ideal - no thread divergence)."></metric>
|
||||
<metric name="VALUBusy" expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
|
||||
<metric name="SALUBusy" expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
|
||||
<metric name="FetchSize" expr=FETCH_SIZE descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WriteSize" expr=WRITE_SIZE descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="MemWrites32B" expr=WRITE_REQ_32B descr="The total number of effective 32B write transactions to the memory"></metric>
|
||||
<metric name="L2CacheHit" expr=100*reduce(TCC_HIT,sum)/(reduce(TCC_HIT,sum)+reduce(TCC_MISS,sum)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
|
||||
<metric name="MemUnitStalled" expr=100*reduce(TCP_TCP_TA_DATA_STALL_CYCLES,max)/GRBM_GUI_ACTIVE/SE_NUM descr="The percentage of GPUTime the memory unit is stalled. Try reducing the number or size of fetches and writes if possible. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
<metric name="WriteUnitStalled" expr=100*TCC_WRREQ_STALL_max/GRBM_GUI_ACTIVE descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."></metric>
|
||||
# LDSBankConflict The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad).
|
||||
<metric name="LDSBankConflict" expr=100*SQ_LDS_BANK_CONFLICT/GRBM_GUI_ACTIVE/CU_NUM descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
</gfx8>
|
||||
|
||||
<gfx9 base="common_derived">
|
||||
<metric name="SQ_WAVES_sum" expr=reduce(SQ_WAVES,sum) descr="Count number of waves sent to SQs. (per-simd, emulated, global). Sum over SQ instances."></metric>
|
||||
<metric name="TA_BUSY_avr" expr=reduce(TA_TA_BUSY,avr) descr="TA block is busy. Average over TA instances."></metric>
|
||||
<metric name="TA_BUSY_max" expr=reduce(TA_TA_BUSY,max) descr="TA block is busy. Max over TA instances."></metric>
|
||||
<metric name="TA_BUSY_min" expr=reduce(TA_TA_BUSY,min) descr="TA block is busy. Min over TA instances."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS_sum" expr=reduce(TA_FLAT_READ_WAVEFRONTS,sum) descr="Number of flat opcode reads processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS_sum" expr=reduce(TA_FLAT_WRITE_WAVEFRONTS,sum) descr="Number of flat opcode writes processed by the TA. Sum over TA instances."></metric>
|
||||
|
||||
<metric name="TCC_HIT_sum" expr=reduce(TCC_HIT,sum) descr="Number of cache hits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_MISS_sum" expr=reduce(TCC_MISS,sum) descr="Number of cache misses. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_32B_sum" expr=reduce(TCC_EA_RDREQ_32B,sum) descr="Number of 32-byte TCC/EA read requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_sum" expr=reduce(TCC_EA_RDREQ,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte). Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_sum" expr=reduce(TCC_EA_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_64B_sum" expr=reduce(TCC_EA_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_WRREQ_STALL_max" expr=reduce(TCC_EA_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over TCC instances."></metric>
|
||||
<metric name="GPU_UTIL" expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Percentage of the time that GUI is active"></metric>
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES_sum" expr=reduce(TCP_TCP_TA_DATA_STALL_CYCLES,sum) descr="Total number of TCP stalls TA data interface."></metric>
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES_max" expr=reduce(TCP_TCP_TA_DATA_STALL_CYCLES,max) descr="Maximum number of TCP stalls TA data interface."></metric>
|
||||
|
||||
<metric name="FETCH_SIZE" expr=(TCC_EA_RDREQ_32B_sum*32+(TCC_EA_RDREQ_sum-TCC_EA_RDREQ_32B_sum)*64)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_SIZE" expr=((TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum)*32+TCC_EA_WRREQ_64B_sum*64)/1024 descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_REQ_32B" expr=TCC_EA_WRREQ_64B_sum*2+(TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum) descr="The total number of 32-byte effective memory writes."></metric>
|
||||
<metric name="VFetchInsts" expr=(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector fetch instructions from the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that fetch from video memory."></metric>
|
||||
<metric name="VWriteInsts" expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."></metric>
|
||||
<metric name="FlatVMemInsts" expr=(SQ_INSTS_FLAT-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of FLAT instructions that read from or write to the video memory executed per work item (affected by flow control). Includes FLAT instructions that read from or write to scratch."></metric>
|
||||
<metric name="LDSInsts" expr=(SQ_INSTS_LDS-SQ_INSTS_FLAT_LDS_ONLY)/SQ_WAVES descr="The average number of LDS read or LDS write instructions executed per work item (affected by flow control). Excludes FLAT instructions that read from or write to LDS."></metric>
|
||||
<metric name="FlatLDSInsts" expr=SQ_INSTS_FLAT_LDS_ONLY/SQ_WAVES descr="The average number of FLAT instructions that read or write to LDS executed per work item (affected by flow control)."></metric>
|
||||
<metric name="VALUUtilization" expr=100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE) descr="The percentage of active vector ALU threads in a wave. A lower number can mean either more thread divergence in a wave or that the work-group size is not a multiple of 64. Value range: 0% (bad), 100% (ideal - no thread divergence)."></metric>
|
||||
<metric name="VALUBusy" expr=100*SQ_ACTIVE_INST_VALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
|
||||
<metric name="SALUBusy" expr=100*SQ_INST_CYCLES_SALU*4/SIMD_NUM/GRBM_GUI_ACTIVE descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
|
||||
<metric name="FetchSize" expr=FETCH_SIZE descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WriteSize" expr=WRITE_SIZE descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="MemWrites32B" expr=WRITE_REQ_32B descr="The total number of effective 32B write transactions to the memory"></metric>
|
||||
<metric name="L2CacheHit" expr=100*reduce(TCC_HIT,sum)/(reduce(TCC_HIT,sum)+reduce(TCC_MISS,sum)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
|
||||
<metric name="MemUnitStalled" expr=100*TCP_TCP_TA_DATA_STALL_CYCLES_max/GRBM_GUI_ACTIVE/SE_NUM descr="The percentage of GPUTime the memory unit is stalled. Try reducing the number or size of fetches and writes if possible. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
<metric name="WriteUnitStalled" expr=100*TCC_WRREQ_STALL_max/GRBM_GUI_ACTIVE descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."></metric>
|
||||
# LDSBankConflict The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad).
|
||||
<metric name="LDSBankConflict" expr=100*SQ_LDS_BANK_CONFLICT/GRBM_GUI_ACTIVE/CU_NUM descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
</gfx9>
|
||||
|
||||
<gfx900 base="gfx9">
|
||||
</gfx900>
|
||||
|
||||
<gfx906 base="gfx9">
|
||||
# EA1
|
||||
<metric name="TCC_EA1_RDREQ_32B_sum" expr=reduce(TCC_EA1_RDREQ_32B,sum) descr="Number of 32-byte TCC/EA read requests. Sum over TCC EA1s."></metric>
|
||||
<metric name="TCC_EA1_RDREQ_sum" expr=reduce(TCC_EA1_RDREQ,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte). Sum over TCC EA1s."></metric>
|
||||
<metric name="TCC_EA1_WRREQ_sum" expr=reduce(TCC_EA1_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Sum over TCC EA1s."></metric>
|
||||
<metric name="TCC_EA1_WRREQ_64B_sum" expr=reduce(TCC_EA1_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface. Sum over TCC EA1s."></metric>
|
||||
<metric name="TCC_WRREQ1_STALL_max" expr=reduce(TCC_EA1_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over TCC instances."></metric>
|
||||
|
||||
<metric name="RDATA1_SIZE" expr=(TCC_EA1_RDREQ_32B_sum*32+(TCC_EA1_RDREQ_sum-TCC_EA1_RDREQ_32B_sum)*64) descr="The total kilobytes fetched from the video memory. This is measured on EA1s."></metric>
|
||||
<metric name="WDATA1_SIZE" expr=((TCC_EA1_WRREQ_sum-TCC_EA1_WRREQ_64B_sum)*32+TCC_EA1_WRREQ_64B_sum*64) descr="The total kilobytes written to the video memory. This is measured on EA1s."></metric>
|
||||
|
||||
# both EA0 and EA1 should be included
|
||||
<metric name="FETCH_SIZE" expr=(TCC_EA_RDREQ_32B_sum*32+(TCC_EA_RDREQ_sum-TCC_EA_RDREQ_32B_sum)*64+RDATA1_SIZE)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_SIZE" expr=((TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum)*32+TCC_EA_WRREQ_64B_sum*64+WDATA1_SIZE)/1024 descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_REQ_32B" expr=(TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum)+(TCC_EA1_WRREQ_sum-TCC_EA1_WRREQ_64B_sum)+(TCC_EA_WRREQ_64B_sum+TCC_EA1_WRREQ_64B_sum)*2 descr="The total number of 32-byte effective memory writes."></metric>
|
||||
</gfx906>
|
||||
|
||||
<gfx908 base="gfx9">
|
||||
<metric name="TCC_HIT_sum" expr=reduce(TCC_HIT,sum) descr="Number of cache hits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_MISS_sum" expr=reduce(TCC_MISS,sum) descr="Number of cache misses. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_32B_sum" expr=reduce(TCC_EA_RDREQ_32B,sum) descr="Number of 32-byte TCC/EA read requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_sum" expr=reduce(TCC_EA_RDREQ,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte). Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_sum" expr=reduce(TCC_EA_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_64B_sum" expr=reduce(TCC_EA_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_WRREQ_STALL_max" expr=reduce(TCC_EA_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over TCC instances."></metric>
|
||||
|
||||
<metric name="CU_UTILIZATION" expr=GRBM_GUI_ACTIVE/GRBM_COUNT descr="The total number of active cycles divided by total number of elapsed cycles"></metric>
|
||||
</gfx908>
|
||||
|
||||
<gfx90a base="gfx9">
|
||||
<metric name="SQ_WAVES_sum" expr=reduce(SQ_WAVES,sum) descr="Count number of waves sent to SQs. (per-simd, emulated, global). Sum over SQ instances."></metric>
|
||||
<metric name="MeanOccupancyPerCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES/GRBM_GUI_ACTIVE/CU_NUM descr="Mean occupancy per compute unit."></metric>
|
||||
<metric name="MeanOccupancyPerActiveCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*4/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
|
||||
<metric name="TA_BUSY_avr" expr=reduce(TA_TA_BUSY,avr) descr="TA block is busy. Average over TA instances."></metric>
|
||||
<metric name="TA_BUSY_max" expr=reduce(TA_TA_BUSY,max) descr="TA block is busy. Max over TA instances."></metric>
|
||||
<metric name="TA_BUSY_min" expr=reduce(TA_TA_BUSY,min) descr="TA block is busy. Min over TA instances."></metric>
|
||||
<metric name="TA_TA_BUSY_sum" expr=reduce(TA_TA_BUSY,sum) descr="TA block is busy. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_TOTAL_WAVEFRONTS_sum" expr=reduce(TA_TOTAL_WAVEFRONTS,sum) descr="Total number of wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TC_CYCLES_sum" expr=reduce(TA_ADDR_STALLED_BY_TC_CYCLES,sum) descr="Number of cycles addr path stalled by TC. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TD_CYCLES_sum" expr=reduce(TA_ADDR_STALLED_BY_TD_CYCLES,sum) descr="Number of cycles addr path stalled by TD. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_DATA_STALLED_BY_TC_CYCLES_sum" expr=reduce(TA_DATA_STALLED_BY_TC_CYCLES,sum) descr="Number of cycles data path stalled by TC. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_WAVEFRONTS_sum" expr=reduce(TA_FLAT_WAVEFRONTS,sum) descr="Number of flat opcode wavfronts processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS_sum" expr=reduce(TA_FLAT_READ_WAVEFRONTS,sum) descr="Number of flat opcode reads processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS_sum" expr=reduce(TA_FLAT_WRITE_WAVEFRONTS,sum) descr="Number of flat opcode writes processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_ATOMIC_WAVEFRONTS_sum" expr=reduce(TA_FLAT_ATOMIC_WAVEFRONTS,sum) descr="Number of flat opcode atomics processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_WAVEFRONTS,sum) descr="Number of buffer wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_READ_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_READ_WAVEFRONTS,sum) descr="Number of buffer read wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_WRITE_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_WRITE_WAVEFRONTS,sum) descr="Number of buffer write wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_ATOMIC_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_ATOMIC_WAVEFRONTS,sum) descr="Number of buffer atomic wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_TOTAL_CYCLES_sum" expr=reduce(TA_BUFFER_TOTAL_CYCLES,sum) descr="Number of buffer cycles issued to TC. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_READ_CYCLES_sum" expr=reduce(TA_BUFFER_COALESCED_READ_CYCLES,sum) descr="Number of buffer coalesced read cycles issued to TC. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_WRITE_CYCLES_sum" expr=reduce(TA_BUFFER_COALESCED_WRITE_CYCLES,sum) descr="Number of buffer coalesced write cycles issued to TC. Sum over TA instances."></metric>
|
||||
<metric name="TD_TD_BUSY_sum" expr=reduce(TD_TD_BUSY,sum) descr="TD is processing or waiting for data. Perf_Windowing not supported for this counter. Sum over TD instances."></metric>
|
||||
<metric name="TD_TC_STALL_sum" expr=reduce(TD_TC_STALL,sum) descr="TD is stalled waiting for TC data. Sum over TD instances."></metric>
|
||||
<metric name="TD_LOAD_WAVEFRONT_sum" expr=reduce(TD_LOAD_WAVEFRONT,sum) descr="Count the wavefronts with opcode = load, include atomics and store. Sum over TD instances."></metric>
|
||||
<metric name="TD_ATOMIC_WAVEFRONT_sum" expr=reduce(TD_ATOMIC_WAVEFRONT,sum) descr="Count the wavefronts with opcode = atomic. Sum over TD instances."></metric>
|
||||
<metric name="TD_STORE_WAVEFRONT_sum" expr=reduce(TD_STORE_WAVEFRONT,sum) descr="Count the wavefronts with opcode = store. Sum over TD instances."></metric>
|
||||
<metric name="TD_COALESCABLE_WAVEFRONT_sum" expr=reduce(TD_COALESCABLE_WAVEFRONT,sum) descr="Count wavefronts that TA finds coalescable. Sum over TD instances."></metric>
|
||||
<metric name="TD_SPI_STALL_sum" expr=reduce(TD_SPI_STALL,sum) descr="TD is stalled SPI vinit, sum of TCP instances"></metric>
|
||||
<metric name="TCP_GATE_EN1_sum" expr=reduce(TCP_GATE_EN1,sum) descr="TCP interface clocks are turned on. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_GATE_EN2_sum" expr=reduce(TCP_GATE_EN2,sum) descr="TCP core clocks are turned on. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TD_TCP_STALL_CYCLES_sum" expr=reduce(TCP_TD_TCP_STALL_CYCLES,sum) descr="TD stalls TCP. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCR_TCP_STALL_CYCLES_sum" expr=reduce(TCP_TCR_TCP_STALL_CYCLES,sum) descr="TCR stalls TCP_TCR_req interface. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_READ_TAGCONFLICT_STALL_CYCLES_sum" expr=reduce(TCP_READ_TAGCONFLICT_STALL_CYCLES,sum) descr="Tagram conflict stall on a read. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum" expr=reduce(TCP_WRITE_TAGCONFLICT_STALL_CYCLES,sum) descr="Tagram conflict stall on a write. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum" expr=reduce(TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES,sum) descr="Tagram conflict stall on an atomic. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_VOLATILE_sum" expr=reduce(TCP_VOLATILE,sum) descr="Total number of L1 volatile pixels/buffers from TA. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_ACCESSES_sum" expr=reduce(TCP_TOTAL_ACCESSES,sum) descr="Total number of pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_READ+TCP_PERF_SEL_TOTAL_NONREAD. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_READ_sum" expr=reduce(TCP_TOTAL_READ,sum) descr="Total number of read pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_HIT_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_EVICT_READ. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_WRITE_sum" expr=reduce(TCP_TOTAL_WRITE,sum) descr="Total number of local write pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_MISS_LRU_WRITE+ TCP_PERF_SEL_TOTAL_MISS_EVICT_WRITE. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITH_RET_sum" expr=reduce(TCP_TOTAL_ATOMIC_WITH_RET,sum) descr="Total number of atomic with return pixels/buffers from TA. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITHOUT_RET_sum" expr=reduce(TCP_TOTAL_ATOMIC_WITHOUT_RET,sum) descr="Total number of atomic without return pixels/buffers from TA Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_WRITEBACK_INVALIDATES_sum" expr=reduce(TCP_TOTAL_WRITEBACK_INVALIDATES,sum) descr="Total number of cache invalidates. Equals TCP_PERF_SEL_TOTAL_WBINVL1+ TCP_PERF_SEL_TOTAL_WBINVL1_VOL+ TCP_PERF_SEL_CP_TCP_INVALIDATE+ TCP_PERF_SEL_SQ_TCP_INVALIDATE_VOL. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_REQUEST_sum" expr=reduce(TCP_UTCL1_REQUEST,sum) descr="Total CLIENT_UTCL1 NORMAL requests Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_MISS_sum" expr=reduce(TCP_UTCL1_TRANSLATION_MISS,sum) descr="Total utcl1 translation misses Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_HIT_sum" expr=reduce(TCP_UTCL1_TRANSLATION_HIT,sum) descr="Total utcl1 translation hits Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_PERMISSION_MISS_sum" expr=reduce(TCP_UTCL1_PERMISSION_MISS,sum) descr="Total utcl1 permission misses Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_CACHE_ACCESSES_sum" expr=reduce(TCP_TOTAL_CACHE_ACCESSES,sum) descr="Count of total cache line (tag) accesses (includes hits and misses). Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCP_LATENCY_sum" expr=reduce(TCP_TCP_LATENCY,sum) descr="Total TCP wave latency (from first clock of wave entering to first clock of wave leaving), divide by TA_TCP_STATE_READ to avg wave latency Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TA_TCP_STATE_READ_sum" expr=reduce(TCP_TA_TCP_STATE_READ,sum) descr="Number of state reads Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_READ_REQ_LATENCY_sum" expr=reduce(TCP_TCC_READ_REQ_LATENCY,sum) descr="Total TCP->TCC request latency for reads and atomics with return. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_WRITE_REQ_LATENCY_sum" expr=reduce(TCP_TCC_WRITE_REQ_LATENCY,sum) descr="Total TCP->TCC request latency for writes and atomics without return. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_READ_REQ_sum" expr=reduce(TCP_TCC_READ_REQ,sum) descr="Total read requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_WRITE_REQ_sum" expr=reduce(TCP_TCC_WRITE_REQ,sum) descr="Total write requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITH_RET_REQ_sum" expr=reduce(TCP_TCC_ATOMIC_WITH_RET_REQ,sum) descr="Total atomic with return requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum" expr=reduce(TCP_TCC_ATOMIC_WITHOUT_RET_REQ,sum) descr="Total atomic without return requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_NC_READ_REQ_sum" expr=reduce(TCP_TCC_NC_READ_REQ,sum) descr="Total read requests with NC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_NC_WRITE_REQ_sum" expr=reduce(TCP_TCC_NC_WRITE_REQ,sum) descr="Total write requests with NC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_NC_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_NC_ATOMIC_REQ,sum) descr="Total atomic requests with NC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_UC_READ_REQ_sum" expr=reduce(TCP_TCC_UC_READ_REQ,sum) descr="Total read requests with UC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_UC_WRITE_REQ_sum" expr=reduce(TCP_TCC_UC_WRITE_REQ,sum) descr="Total write requests with UC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_UC_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_UC_ATOMIC_REQ,sum) descr="Total atomic requests with UC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_CC_READ_REQ_sum" expr=reduce(TCP_TCC_CC_READ_REQ,sum) descr="Total write requests with CC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_CC_WRITE_REQ_sum" expr=reduce(TCP_TCC_CC_WRITE_REQ,sum) descr="Total write requests with CC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_CC_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_CC_ATOMIC_REQ,sum) descr="Total atomic requests with CC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_RW_READ_REQ_sum" expr=reduce(TCP_TCC_RW_READ_REQ,sum) descr="Total write requests with RW mtype from this TCP to all TCCs. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_RW_WRITE_REQ_sum" expr=reduce(TCP_TCC_RW_WRITE_REQ,sum) descr="Total write requests with RW mtype from this TCP to all TCCs. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_RW_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_RW_ATOMIC_REQ,sum) descr="Total atomic requests with RW mtype from this TCP to all TCCs. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_PENDING_STALL_CYCLES_sum" expr=reduce(TCP_PENDING_STALL_CYCLES,sum) descr="Stall due to data pending from L2. Sum over TCP instances."></metric>
|
||||
<metric name="TCA_CYCLE_sum" expr=reduce(TCA_CYCLE,sum) descr="Number of cycles. Sum over all TCA instances "></metric>
|
||||
<metric name="TCA_BUSY_sum" expr=reduce(TCA_BUSY,sum) descr="Number of cycles we have a request pending. Sum over all TCA instances."></metric>
|
||||
<metric name="TCC_BUSY_avr" expr=reduce(TCC_BUSY,avr) descr="TCC_BUSY avr over all memory channels."></metric>
|
||||
<metric name="TCC_WRREQ_STALL_max" expr=reduce(TCC_EA_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over TCC instances."></metric>
|
||||
<metric name="TCC_CYCLE_sum" expr=reduce(TCC_CYCLE,sum) descr="Number of cycles. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_BUSY_sum" expr=reduce(TCC_BUSY,sum) descr="Number of cycles we have a request pending. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_REQ_sum" expr=reduce(TCC_REQ,sum) descr="Number of requests of all types. This is measured at the tag block. This may be more than the number of requests arriving at the TCC, but it is a good indication of the total amount of work that needs to be performed. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_STREAMING_REQ_sum" expr=reduce(TCC_STREAMING_REQ,sum) descr="Number of streaming requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_NC_REQ_sum" expr=reduce(TCC_NC_REQ,sum) descr="The number of noncoherently cached requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_UC_REQ_sum" expr=reduce(TCC_UC_REQ,sum) descr="The number of uncached requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_CC_REQ_sum" expr=reduce(TCC_CC_REQ,sum) descr="The number of coherently cached requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_RW_REQ_sum" expr=reduce(TCC_RW_REQ,sum) descr="The number of RW requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_PROBE_sum" expr=reduce(TCC_PROBE,sum) descr="Number of probe requests. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_PROBE_ALL_sum" expr=reduce(TCC_PROBE_ALL,sum) descr="Number of external probe requests with with EA_TCC_preq_all== 1. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_READ_sum" expr=reduce(TCC_READ,sum) descr="Number of read requests. Compressed reads are included in this, but metadata reads are not included. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_WRITE_sum" expr=reduce(TCC_WRITE,sum) descr="Number of write requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_ATOMIC_sum" expr=reduce(TCC_ATOMIC,sum) descr="Number of atomic requests of all types. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_HIT_sum" expr=reduce(TCC_HIT,sum) descr="Number of cache hits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_MISS_sum" expr=reduce(TCC_MISS,sum) descr="Number of cache misses. UC reads count as misses. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_WRITEBACK_sum" expr=reduce(TCC_WRITEBACK,sum) descr="Number of lines written back to main memory. This includes writebacks of dirty lines and uncached write/atomic requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_sum" expr=reduce(TCC_EA_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_64B_sum" expr=reduce(TCC_EA_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WR_UNCACHED_32B_sum" expr=reduce(TCC_EA_WR_UNCACHED_32B,sum) descr="Number of 32-byte write/atomic going over the TC_EA_wrreq interface due to uncached traffic. Note that CC mtypes can produce uncached requests, and those are included in this. A 64-byte request will be counted as 2. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_STALL_sum" expr=reduce(TCC_EA_WRREQ_STALL,sum) descr="Number of cycles a write request was stalled. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_IO_CREDIT_STALL_sum" expr=reduce(TCC_EA_WRREQ_IO_CREDIT_STALL,sum) descr="Number of cycles a EA write request was stalled because the interface was out of IO credits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_GMI_CREDIT_STALL_sum" expr=reduce(TCC_EA_WRREQ_GMI_CREDIT_STALL,sum) descr="Number of cycles a EA write request was stalled because the interface was out of GMI credits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_DRAM_CREDIT_STALL_sum" expr=reduce(TCC_EA_WRREQ_DRAM_CREDIT_STALL,sum) descr="Number of cycles a EA write request was stalled because the interface was out of DRAM credits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_TOO_MANY_EA_WRREQS_STALL_sum" expr=reduce(TCC_TOO_MANY_EA_WRREQS_STALL,sum) descr="Number of cycles the TCC could not send a EA write request because it already reached its maximum number of pending EA write requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_LEVEL_sum" expr=reduce(TCC_EA_WRREQ_LEVEL,sum) descr="The sum of the number of EA write requests in flight. This is primarily meant for measure average EA write latency. Average write latency = TCC_PERF_SEL_EA_WRREQ_LEVEL/TCC_PERF_SEL_EA_WRREQ. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_LEVEL_sum" expr=reduce(TCC_EA_RDREQ_LEVEL,sum) descr="The sum of the number of TCC/EA read requests in flight. This is primarily meant for measure average EA read latency. Average read latency = TCC_PERF_SEL_EA_RDREQ_LEVEL/TCC_PERF_SEL_EA_RDREQ. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_ATOMIC_sum" expr=reduce(TCC_EA_ATOMIC,sum) descr="Number of transactions going over the TC_EA_wrreq interface that are actually atomic requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_ATOMIC_LEVEL_sum" expr=reduce(TCC_EA_ATOMIC_LEVEL,sum) descr="The sum of the number of EA atomics in flight. This is primarily meant for measure average EA atomic latency. Average atomic latency = TCC_PERF_SEL_EA_WRREQ_ATOMIC_LEVEL/TCC_PERF_SEL_EA_WRREQ_ATOMIC. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_sum" expr=reduce(TCC_EA_RDREQ,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte) Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_32B_sum" expr=reduce(TCC_EA_RDREQ_32B,sum) descr="Number of 32-byte TCC/EA read requests Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RD_UNCACHED_32B_sum" expr=reduce(TCC_EA_RD_UNCACHED_32B,sum) descr="Number of 32-byte TCC/EA read due to uncached traffic. A 64-byte request will be counted as 2 Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_IO_CREDIT_STALL_sum" expr=reduce(TCC_EA_RDREQ_IO_CREDIT_STALL,sum) descr="Number of cycles there was a stall because the read request interface was out of IO credits. Stalls occur regardless of whether a read needed to be performed or not. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_GMI_CREDIT_STALL_sum" expr=reduce(TCC_EA_RDREQ_GMI_CREDIT_STALL,sum) descr="Number of cycles there was a stall because the read request interface was out of GMI credits. Stalls occur regardless of whether a read needed to be performed or not. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_DRAM_CREDIT_STALL_sum" expr=reduce(TCC_EA_RDREQ_DRAM_CREDIT_STALL,sum) descr="Number of cycles there was a stall because the read request interface was out of DRAM credits. Stalls occur regardless of whether a read needed to be performed or not. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_TAG_STALL_sum" expr=reduce(TCC_TAG_STALL,sum) descr="Total number of cycles the normal request pipeline in the tag is stalled for any reason."></metric>
|
||||
<metric name="TCC_NORMAL_WRITEBACK_sum" expr=reduce(TCC_NORMAL_WRITEBACK,sum) descr="Number of writebacks due to requests that are not writeback requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_WB_WRITEBACK_sum" expr=reduce(TCC_ALL_TC_OP_WB_WRITEBACK,sum) descr="Number of writebacks due to all TC_OP writeback requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_NORMAL_EVICT_sum" expr=reduce(TCC_NORMAL_EVICT,sum) descr="Number of evictions due to requests that are not invalidate or probe requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_INV_EVICT_sum" expr=reduce(TCC_ALL_TC_OP_INV_EVICT,sum) descr="Number of evictions due to all TC_OP invalidate requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_RDREQ_DRAM_sum" expr=reduce(TCC_EA_RDREQ_DRAM,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte) destined for DRAM (MC). Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA_WRREQ_DRAM_sum" expr=reduce(TCC_EA_WRREQ_DRAM,sum) descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC). Sum over TCC instances."></metric>
|
||||
|
||||
<metric name="FETCH_SIZE" expr=(TCC_EA_RDREQ_32B_sum*32+(TCC_EA_RDREQ_sum-TCC_EA_RDREQ_32B_sum)*64)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_SIZE" expr=((TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum)*32+TCC_EA_WRREQ_64B_sum*64)/1024 descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_REQ_32B" expr=TCC_EA_WRREQ_64B_sum*2+(TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum) descr="The total number of 32-byte effective memory writes."></metric>
|
||||
<metric name="CU_OCCUPANCY" expr=(SQ_CYCLES/(SQ_WAVE_CYCLES*4))/MAX_WAVE_SIZE descr="The ratio of active waves on a CU to the maximum number of active waves supported by the CU"></metric>
|
||||
<metric name="CU_UTILIZATION" expr=GRBM_GUI_ACTIVE/GRBM_COUNT descr="The total number of active cycles divided by total number of elapsed cycles"></metric>
|
||||
<metric name="TOTAL_16_OPS" expr=(SQ_INSTS_VALU_FMA_F16*2+SQ_INSTS_VALU_ADD_F16+SQ_INSTS_VALU_MUL_F16+SQ_INSTS_VALU_TRANS_F16)*64+((SQ_INSTS_VALU_MFMA_MOPS_F16+SQ_INSTS_VALU_MFMA_MOPS_BF16)*512) descr="The number of 16 bits OPS executed"></metric>
|
||||
<metric name="TOTAL_32_OPS" expr=(SQ_INSTS_VALU_FMA_F32*2+SQ_INSTS_VALU_INT32+SQ_INSTS_VALU_ADD_F32+SQ_INSTS_VALU_MUL_F32+SQ_INSTS_VALU_TRANS_F32)*64+(SQ_INSTS_VALU_MFMA_MOPS_F32*512) descr="The number of 32 bits OPS executed"></metric>
|
||||
<metric name="TOTAL_64_OPS" expr=(SQ_INSTS_VALU_FMA_F64*2+SQ_INSTS_VALU_INT64+SQ_INSTS_VALU_ADD_F64+SQ_INSTS_VALU_MUL_F64)*64+(SQ_INSTS_VALU_MFMA_MOPS_F64*512) descr="The number of 64 bits OPS executed"></metric>
|
||||
|
||||
<metric name="AggSysCycles" expr=GRBM_GUI_ACTIVE*CU_NUM descr="Unit: cycles"></metric>
|
||||
## IP Block Utilization Metrics
|
||||
<metric name="GpuUtil" expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Unit: percent"></metric>
|
||||
<metric name="CpUtil" expr=100*GRBM_CP_BUSY/GRBM_GUI_ACTIVE descr="Unit: percent"></metric>
|
||||
<metric name="SpiUtil" expr=100*GRBM_SPI_BUSY/GRBM_GUI_ACTIVE descr="Unit: percent"></metric>
|
||||
<metric name="TaUtil" expr=100*GRBM_TA_BUSY/GRBM_GUI_ACTIVE descr="Unit: percent"></metric>
|
||||
<metric name="TcUtil" expr=100*GRBM_TC_BUSY/GRBM_GUI_ACTIVE descr="Unit: percent"></metric>
|
||||
<metric name="EaUtil" expr=100*GRBM_EA_BUSY/GRBM_GUI_ACTIVE descr="Unit: percent"></metric>
|
||||
## Instruction Fetch Metrics
|
||||
<metric name="InstrFetchLatency" expr=SQ_ACCUM_PREV_HIRES/SQ_IFETCH descr="Unit: cycles"></metric>
|
||||
## Wavefront Metrics
|
||||
<metric name="WaveOccupancy" expr=SQ_ACCUM_PREV_HIRES/GRBM_GUI_ACTIVE descr="Unit: wavefronts"></metric>
|
||||
<metric name="WaveDuration" expr=4*SQ_WAVE_CYCLES/SQ_WAVES descr="Unit: cycles"></metric>
|
||||
<metric name="WaveDepWait" expr=100*SQ_WAIT_ANY/SQ_WAVE_CYCLES descr="Unit: percent"></metric>
|
||||
<metric name="WaveIssueWait" expr=100*SQ_WAIT_INST_ANY/SQ_WAVE_CYCLES descr="Unit: percent"></metric>
|
||||
<metric name="WaveExec" expr=100*SQ_ACTIVE_INST_ANY/SQ_WAVE_CYCLES descr="Unit: percent"></metric>
|
||||
## Compute Unit Metrics
|
||||
<metric name="ValuIops" expr=(SQ_INSTS_VALU_INT32+SQ_INSTS_VALU_INT64)*64 descr="Unit: IOP"></metric>
|
||||
<metric name="MfmaFlops" expr=(SQ_INSTS_VALU_MFMA_MOPS_F16+SQ_INSTS_VALU_MFMA_MOPS_BF16+SQ_INSTS_VALU_MFMA_MOPS_F32+SQ_INSTS_VALU_MFMA_MOPS_F64)*512 descr="Unit: FLOP"></metric>
|
||||
<metric name="MfmaFlopsF16" expr=SQ_INSTS_VALU_MFMA_MOPS_F16*512 descr="Unit: FLOP"></metric>
|
||||
<metric name="MfmaFlopsBF16" expr=SQ_INSTS_VALU_MFMA_MOPS_BF16*512 descr="Unit: FLOP"></metric>
|
||||
<metric name="MfmaFlopsF32" expr=SQ_INSTS_VALU_MFMA_MOPS_F32*512 descr="Unit: FLOP"></metric>
|
||||
<metric name="MfmaFlopsF64" expr=SQ_INSTS_VALU_MFMA_MOPS_F64*512 descr="Unit: IOP"></metric>
|
||||
<metric name="ScaPipeIssueUtil" expr=100*SQ_ACTIVE_INST_SCA/(GRBM_GUI_ACTIVE*CU_NUM) descr="Unit: percent"></metric>
|
||||
<metric name="ValuPipeIssueUtil" expr=100*SQ_ACTIVE_INST_VALU/(GRBM_GUI_ACTIVE*CU_NUM) descr="Unit: percent"></metric>
|
||||
<metric name="VmemPipeIssueUtil" expr=400*(SQ_ACTIVE_INST_VMEM+SQ_ACTIVE_INST_FLAT)/(GRBM_GUI_ACTIVE*CU_NUM) descr="Unit: percent"></metric>
|
||||
<metric name="MfmaUtil" expr=100*SQ_VALU_MFMA_BUSY_CYCLES/(GRBM_GUI_ACTIVE*CU_NUM*4) descr="Unit: percent"></metric>
|
||||
<metric name="AvgNumActiveThreads" expr=SQ_THREAD_CYCLES_VALU/SQ_ACTIVE_INST_VALU descr="Unit: percent"></metric>
|
||||
<metric name="VmemLatency" expr=SQ_ACCUM_PREV_HIRES/SQ_INSTS_VMEM descr="Unit: cycles"></metric>
|
||||
<metric name="SmemLatency" expr=SQ_ACCUM_PREV_HIRES/SQ_INSTS_SMEM_NORM descr="Unit: cycles"></metric>
|
||||
## Local Data Share (LDS) Metrics
|
||||
<metric name="LdsUtil" expr=100*SQ_LDS_IDX_ACTIVE/(GRBM_GUI_ACTIVE*CU_NUM) descr="Unit: percent"></metric>
|
||||
<metric name="LdsPipeIssueUtil" expr=400*SQ_ACTIVE_INST_LDS/(GRBM_GUI_ACTIVE*CU_NUM*2) descr="Unit: percent"></metric>
|
||||
<metric name="LdsLatency" expr=SQ_ACCUM_PREV_HIRES/SQ_INSTS_LDS descr="Unit: cycles"></metric>
|
||||
<metric name="LdsBankConflict" expr=SQ_LDS_BANK_CONFLICT/(SQ_LDS_IDX_ACTIVE-SQ_LDS_BANK_CONFLICT) descr="Unit: conflicts/access"></metric>
|
||||
## L1I and sL1D Cache Metrics
|
||||
<metric name="L1iCacheHitRate" expr=100*SQC_ICACHE_HITS/SQC_ICACHE_REQ descr="Unit: percent"></metric>
|
||||
<metric name="sL1dCacheHitRate" expr=100*SQC_DCACHE_HITS/SQC_DCACHE_REQ descr="Unit: percent"></metric>
|
||||
## vL1D Cache Metrics
|
||||
<metric name="vL1dBufCoalesceRate" expr=6400*TA_TOTAL_WAVEFRONTS_sum/(TCP_TOTAL_ACCESSES_sum*4) descr="Unit: percent"></metric>
|
||||
<metric name="vL1dCacheUtil" expr=100*TCP_GATE_EN2_sum/TCP_GATE_EN1_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dCacheTcbHitRate" expr=100*TCP_UTCL1_TRANSLATION_HIT_sum/TCP_UTCL1_REQUEST_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dCacheWaveLatency" expr=TCP_TCP_LATENCY_sum/TCP_TA_TCP_STATE_READ_sum descr="Unit: cycles"></metric>
|
||||
<metric name="vL1dReadFromL2Latency" expr=TCP_TCC_READ_REQ_LATENCY_sum/(TCP_TCC_READ_REQ_sum+TCP_TCC_ATOMIC_WITH_RET_REQ_sum) descr="Unit: cycles"></metric>
|
||||
<metric name="vL1dWriteToL2Latency" expr=TCP_TCC_WRITE_REQ_LATENCY_sum/(TCP_TCC_WRITE_REQ_sum+TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum) descr="Unit: cycles"></metric>
|
||||
<metric name="vL1dRdTagConfStallRate" expr=100*TCP_READ_TAGCONFLICT_STALL_CYCLES_sum/TCP_GATE_EN2_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dWrTagConfStallRate" expr=100*TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum/TCP_GATE_EN2_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dAtomicTagConfStallRate" expr=100*TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum/TCP_GATE_EN2_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dMissReqStallRate" expr=100*TCP_TCR_TCP_STALL_CYCLES_sum/TCP_GATE_EN2_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dDataPendRate" expr=100*TCP_PENDING_STALL_CYCLES_sum/TCP_GATE_EN2_sum descr="Unit: percent"></metric>
|
||||
<metric name="vL1dDataRetStallRate" expr=100*TD_TC_STALL_sum/TD_TD_BUSY_sum descr="Unit: percent"></metric>
|
||||
## L2 Cache Metrics
|
||||
<metric name="L2CacheHitRate" expr=100*TCC_HIT_sum/(TCC_HIT_sum+TCC_MISS_sum) descr="Unit: percent"></metric>
|
||||
<metric name="L2CacheTagRamStallRate" expr=100*TCC_TAG_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaRdLatency" expr=TCC_EA_RDREQ_LEVEL_sum/TCC_EA_RDREQ_sum descr="Unit: cycles"></metric>
|
||||
<metric name="EaRdIoStallRate" expr=100*TCC_EA_RDREQ_IO_CREDIT_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaRdGmiStallRate" expr=100*TCC_EA_RDREQ_GMI_CREDIT_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaRdDramStallRate" expr=100*TCC_EA_RDREQ_DRAM_CREDIT_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaWrLatency" expr=TCC_EA_WRREQ_LEVEL_sum/TCC_EA_WRREQ_sum descr="Unit: cycles"></metric>
|
||||
<metric name="EaWrIoStallRate" expr=100*TCC_EA_WRREQ_IO_CREDIT_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaWrGmiStallRate" expr=100*TCC_EA_WRREQ_GMI_CREDIT_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaWrDramStallRate" expr=100*TCC_EA_WRREQ_DRAM_CREDIT_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaWrStarveRate" expr=100*TCC_TOO_MANY_EA_WRREQS_STALL_sum/TCC_BUSY_sum descr="Unit: percent"></metric>
|
||||
<metric name="EaAtomicLatency" expr=TCC_EA_ATOMIC_LEVEL_sum/TCC_EA_ATOMIC_sum descr="Unit: cycles"></metric>
|
||||
</gfx90a>
|
||||
|
||||
<gfx940>
|
||||
<metric name="SQ_WAVES_sum" expr=reduce(SQ_WAVES,sum) descr="Count number of waves sent to SQs. (per-simd, emulated, global). Sum over SQ instances."></metric>
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES_sum" expr=reduce(TCP_TCP_TA_DATA_STALL_CYCLES,sum) descr="Total number of TCP stalls TA data interface."></metric>
|
||||
<metric name="TCP_TCP_TA_DATA_STALL_CYCLES_max" expr=reduce(TCP_TCP_TA_DATA_STALL_CYCLES,max) descr="Maximum number of TCP stalls TA data interface."></metric>
|
||||
|
||||
<metric name="MeanOccupancyPerCU" expr=reduce(SQ_LEVEL_WAVES,sum)*0+reduce(SQ_ACCUM_PREV_HIRES,sum)/reduce(GRBM_GUI_ACTIVE,sum)/CU_NUM descr="Mean occupancy per compute unit."></metric>
|
||||
<metric name="MeanOccupancyPerActiveCU" expr=SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV_HIRES*4/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
|
||||
<metric name="VFetchInsts" expr=(SQ_INSTS_VMEM_RD-TA_FLAT_READ_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector fetch instructions from the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that fetch from video memory."></metric>
|
||||
<metric name="VWriteInsts" expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."></metric>
|
||||
<metric name="VALUUtilization" expr=100*SQ_THREAD_CYCLES_VALU/(SQ_ACTIVE_INST_VALU*MAX_WAVE_SIZE) descr="The percentage of active vector ALU threads in a wave. A lower number can mean either more thread divergence in a wave or that the work-group size is not a multiple of 64. Value range: 0% (bad), 100% (ideal - no thread divergence)."></metric>
|
||||
<metric name="VALUBusy" expr=100*reduce(SQ_ACTIVE_INST_VALU,sum)*4/SIMD_NUM/reduce(GRBM_GUI_ACTIVE,sum) descr="The percentage of GPUTime vector ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
|
||||
<metric name="SALUBusy" expr=100*reduce(SQ_INST_CYCLES_SALU,sum)*4/SIMD_NUM/reduce(GRBM_GUI_ACTIVE,sum) descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."></metric>
|
||||
<metric name="FetchSize" expr=FETCH_SIZE descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WriteSize" expr=WRITE_SIZE descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="MemWrites32B" expr=WRITE_REQ_32B descr="The total number of effective 32B write transactions to the memory"></metric>
|
||||
<metric name="MemUnitStalled" expr=100*TCP_TCP_TA_DATA_STALL_CYCLES_max/GRBM_GUI_ACTIVE/SE_NUM descr="The percentage of GPUTime the memory unit is stalled. Try reducing the number or size of fetches and writes if possible. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
<metric name="TA_BUSY_avr" expr=reduce(TA_TA_BUSY,avr) descr="TA block is busy. Average over TA instances."></metric>
|
||||
<metric name="TA_BUSY_max" expr=reduce(TA_TA_BUSY,max) descr="TA block is busy. Max over TA instances."></metric>
|
||||
<metric name="TA_BUSY_min" expr=reduce(TA_TA_BUSY,min) descr="TA block is busy. Min over TA instances."></metric>
|
||||
<metric name="TA_TA_BUSY_sum" expr=reduce(TA_TA_BUSY,sum) descr="TA block is busy. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_TOTAL_WAVEFRONTS_sum" expr=reduce(TA_TOTAL_WAVEFRONTS,sum) descr="Total number of wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TC_CYCLES_sum" expr=reduce(TA_ADDR_STALLED_BY_TC_CYCLES,sum) descr="Number of cycles addr path stalled by TC. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_ADDR_STALLED_BY_TD_CYCLES_sum" expr=reduce(TA_ADDR_STALLED_BY_TD_CYCLES,sum) descr="Number of cycles addr path stalled by TD. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_DATA_STALLED_BY_TC_CYCLES_sum" expr=reduce(TA_DATA_STALLED_BY_TC_CYCLES,sum) descr="Number of cycles data path stalled by TC. Perf_Windowing not supported for this counter. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_WAVEFRONTS_sum" expr=reduce(TA_FLAT_WAVEFRONTS,sum) descr="Number of flat opcode wavfronts processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_READ_WAVEFRONTS_sum" expr=reduce(TA_FLAT_READ_WAVEFRONTS,sum) descr="Number of flat opcode reads processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_WRITE_WAVEFRONTS_sum" expr=reduce(TA_FLAT_WRITE_WAVEFRONTS,sum) descr="Number of flat opcode writes processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_ATOMIC_WAVEFRONTS_sum" expr=reduce(TA_FLAT_ATOMIC_WAVEFRONTS,sum) descr="Number of flat opcode atomics processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_WAVEFRONTS,sum) descr="Number of buffer wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_READ_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_READ_WAVEFRONTS,sum) descr="Number of buffer read wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_WRITE_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_WRITE_WAVEFRONTS,sum) descr="Number of buffer write wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_ATOMIC_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_ATOMIC_WAVEFRONTS,sum) descr="Number of buffer atomic wavefronts processed by TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_TOTAL_CYCLES_sum" expr=reduce(TA_BUFFER_TOTAL_CYCLES,sum) descr="Number of buffer cycles issued to TC. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_READ_CYCLES_sum" expr=reduce(TA_BUFFER_COALESCED_READ_CYCLES,sum) descr="Number of buffer coalesced read cycles issued to TC. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_COALESCED_WRITE_CYCLES_sum" expr=reduce(TA_BUFFER_COALESCED_WRITE_CYCLES,sum) descr="Number of buffer coalesced write cycles issued to TC. Sum over TA instances."></metric>
|
||||
<metric name="TD_TD_BUSY_sum" expr=reduce(TD_TD_BUSY,sum) descr="TD is processing or waiting for data. Perf_Windowing not supported for this counter. Sum over TD instances."></metric>
|
||||
<metric name="TD_TC_STALL_sum" expr=reduce(TD_TC_STALL,sum) descr="TD is stalled waiting for TC data. Sum over TD instances."></metric>
|
||||
<metric name="TD_LOAD_WAVEFRONT_sum" expr=reduce(TD_LOAD_WAVEFRONT,sum) descr="Count the wavefronts with opcode = load, include atomics and store. Sum over TD instances."></metric>
|
||||
<metric name="TD_ATOMIC_WAVEFRONT_sum" expr=reduce(TD_ATOMIC_WAVEFRONT,sum) descr="Count the wavefronts with opcode = atomic. Sum over TD instances."></metric>
|
||||
<metric name="TD_STORE_WAVEFRONT_sum" expr=reduce(TD_STORE_WAVEFRONT,sum) descr="Count the wavefronts with opcode = store. Sum over TD instances."></metric>
|
||||
<metric name="TD_COALESCABLE_WAVEFRONT_sum" expr=reduce(TD_COALESCABLE_WAVEFRONT,sum) descr="Count wavefronts that TA finds coalescable. Sum over TD instances."></metric>
|
||||
<metric name="TD_SPI_STALL_sum" expr=reduce(TD_SPI_STALL,sum) descr="TD is stalled SPI vinit, sum of TCP instances"></metric>
|
||||
<metric name="TCP_GATE_EN1_sum" expr=reduce(TCP_GATE_EN1,sum) descr="TCP interface clocks are turned on. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_GATE_EN2_sum" expr=reduce(TCP_GATE_EN2,sum) descr="TCP core clocks are turned on. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TD_TCP_STALL_CYCLES_sum" expr=reduce(TCP_TD_TCP_STALL_CYCLES,sum) descr="TD stalls TCP. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCR_TCP_STALL_CYCLES_sum" expr=reduce(TCP_TCR_TCP_STALL_CYCLES,sum) descr="TCR stalls TCP_TCR_req interface. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_READ_TAGCONFLICT_STALL_CYCLES_sum" expr=reduce(TCP_READ_TAGCONFLICT_STALL_CYCLES,sum) descr="Tagram conflict stall on a read. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum" expr=reduce(TCP_WRITE_TAGCONFLICT_STALL_CYCLES,sum) descr="Tagram conflict stall on a write. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum" expr=reduce(TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES,sum) descr="Tagram conflict stall on an atomic. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_VOLATILE_sum" expr=reduce(TCP_VOLATILE,sum) descr="Total number of L1 volatile pixels/buffers from TA. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_ACCESSES_sum" expr=reduce(TCP_TOTAL_ACCESSES,sum) descr="Total number of pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_READ+TCP_PERF_SEL_TOTAL_NONREAD. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_READ_sum" expr=reduce(TCP_TOTAL_READ,sum) descr="Total number of read pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_HIT_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_LRU_READ + TCP_PERF_SEL_TOTAL_MISS_EVICT_READ. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_WRITE_sum" expr=reduce(TCP_TOTAL_WRITE,sum) descr="Total number of local write pixels/buffers from TA. Equals TCP_PERF_SEL_TOTAL_MISS_LRU_WRITE+ TCP_PERF_SEL_TOTAL_MISS_EVICT_WRITE. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITH_RET_sum" expr=reduce(TCP_TOTAL_ATOMIC_WITH_RET,sum) descr="Total number of atomic with return pixels/buffers from TA. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_ATOMIC_WITHOUT_RET_sum" expr=reduce(TCP_TOTAL_ATOMIC_WITHOUT_RET,sum) descr="Total number of atomic without return pixels/buffers from TA Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_WRITEBACK_INVALIDATES_sum" expr=reduce(TCP_TOTAL_WRITEBACK_INVALIDATES,sum) descr="Total number of cache invalidates. Equals TCP_PERF_SEL_TOTAL_WBINVL1+ TCP_PERF_SEL_TOTAL_WBINVL1_VOL+ TCP_PERF_SEL_CP_TCP_INVALIDATE+ TCP_PERF_SEL_SQ_TCP_INVALIDATE_VOL. Not Windowed. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_REQUEST_sum" expr=reduce(TCP_UTCL1_REQUEST,sum) descr="Total CLIENT_UTCL1 NORMAL requests Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_MISS_sum" expr=reduce(TCP_UTCL1_TRANSLATION_MISS,sum) descr="Total utcl1 translation misses Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_TRANSLATION_HIT_sum" expr=reduce(TCP_UTCL1_TRANSLATION_HIT,sum) descr="Total utcl1 translation hits Sum over TCP instances."></metric>
|
||||
<metric name="TCP_UTCL1_PERMISSION_MISS_sum" expr=reduce(TCP_UTCL1_PERMISSION_MISS,sum) descr="Total utcl1 permission misses Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TOTAL_CACHE_ACCESSES_sum" expr=reduce(TCP_TOTAL_CACHE_ACCESSES,sum) descr="Count of total cache line (tag) accesses (includes hits and misses). Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TA_TCP_STATE_READ_sum" expr=reduce(TCP_TA_TCP_STATE_READ,sum) descr="Number of state reads Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_READ_REQ_sum" expr=reduce(TCP_TCC_READ_REQ,sum) descr="Total read requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_WRITE_REQ_sum" expr=reduce(TCP_TCC_WRITE_REQ,sum) descr="Total write requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITH_RET_REQ_sum" expr=reduce(TCP_TCC_ATOMIC_WITH_RET_REQ,sum) descr="Total atomic with return requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum" expr=reduce(TCP_TCC_ATOMIC_WITHOUT_RET_REQ,sum) descr="Total atomic without return requests from TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_NC_READ_REQ_sum" expr=reduce(TCP_TCC_NC_READ_REQ,sum) descr="Total read requests with NC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_NC_WRITE_REQ_sum" expr=reduce(TCP_TCC_NC_WRITE_REQ,sum) descr="Total write requests with NC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_NC_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_NC_ATOMIC_REQ,sum) descr="Total atomic requests with NC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_UC_READ_REQ_sum" expr=reduce(TCP_TCC_UC_READ_REQ,sum) descr="Total read requests with UC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_UC_WRITE_REQ_sum" expr=reduce(TCP_TCC_UC_WRITE_REQ,sum) descr="Total write requests with UC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_UC_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_UC_ATOMIC_REQ,sum) descr="Total atomic requests with UC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_CC_READ_REQ_sum" expr=reduce(TCP_TCC_CC_READ_REQ,sum) descr="Total write requests with CC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_CC_WRITE_REQ_sum" expr=reduce(TCP_TCC_CC_WRITE_REQ,sum) descr="Total write requests with CC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_CC_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_CC_ATOMIC_REQ,sum) descr="Total atomic requests with CC mtype from this TCP to all TCCs Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_RW_READ_REQ_sum" expr=reduce(TCP_TCC_RW_READ_REQ,sum) descr="Total write requests with RW mtype from this TCP to all TCCs. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_RW_WRITE_REQ_sum" expr=reduce(TCP_TCC_RW_WRITE_REQ,sum) descr="Total write requests with RW mtype from this TCP to all TCCs. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_TCC_RW_ATOMIC_REQ_sum" expr=reduce(TCP_TCC_RW_ATOMIC_REQ,sum) descr="Total atomic requests with RW mtype from this TCP to all TCCs. Sum over TCP instances."></metric>
|
||||
<metric name="TCP_PENDING_STALL_CYCLES_sum" expr=reduce(TCP_PENDING_STALL_CYCLES,sum) descr="Stall due to data pending from L2. Sum over TCP instances."></metric>
|
||||
<metric name="TCA_CYCLE_sum" expr=reduce(TCA_CYCLE,sum) descr="Number of cycles. Sum over all TCA instances "></metric>
|
||||
<metric name="TCA_BUSY_sum" expr=reduce(TCA_BUSY,sum) descr="Number of cycles we have a request pending. Sum over all TCA instances."></metric>
|
||||
<metric name="TCC_BUSY_avr" expr=reduce(TCC_BUSY,avr) descr="TCC_BUSY avr over all memory channels."></metric>
|
||||
<metric name="TCC_WRREQ_STALL_max" expr=reduce(TCC_EA0_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over TCC instances."></metric>
|
||||
<metric name="TCC_CYCLE_sum" expr=reduce(TCC_CYCLE,sum) descr="Number of cycles. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_BUSY_sum" expr=reduce(TCC_BUSY,sum) descr="Number of cycles we have a request pending. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_REQ_sum" expr=reduce(TCC_REQ,sum) descr="Number of requests of all types. This is measured at the tag block. This may be more than the number of requests arriving at the TCC, but it is a good indication of the total amount of work that needs to be performed. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_STREAMING_REQ_sum" expr=reduce(TCC_STREAMING_REQ,sum) descr="Number of streaming requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_NC_REQ_sum" expr=reduce(TCC_NC_REQ,sum) descr="The number of noncoherently cached requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_UC_REQ_sum" expr=reduce(TCC_UC_REQ,sum) descr="The number of uncached requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_CC_REQ_sum" expr=reduce(TCC_CC_REQ,sum) descr="The number of coherently cached requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_RW_REQ_sum" expr=reduce(TCC_RW_REQ,sum) descr="The number of RW requests. This is measured at the tag block. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_PROBE_sum" expr=reduce(TCC_PROBE,sum) descr="Number of probe requests. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_PROBE_ALL_sum" expr=reduce(TCC_PROBE_ALL,sum) descr="Number of external probe requests with with EA_TCC_preq_all== 1. Not windowable. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_READ_sum" expr=reduce(TCC_READ,sum) descr="Number of read requests. Compressed reads are included in this, but metadata reads are not included. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_WRITE_sum" expr=reduce(TCC_WRITE,sum) descr="Number of write requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_ATOMIC_sum" expr=reduce(TCC_ATOMIC,sum) descr="Number of atomic requests of all types. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_HIT_sum" expr=reduce(TCC_HIT,sum) descr="Number of cache hits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_MISS_sum" expr=reduce(TCC_MISS,sum) descr="Number of cache misses. UC reads count as misses. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_WRITEBACK_sum" expr=reduce(TCC_WRITEBACK,sum) descr="Number of lines written back to main memory. This includes writebacks of dirty lines and uncached write/atomic requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_sum" expr=reduce(TCC_EA0_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the TC_EA_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests. This does not include probe commands. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_64B_sum" expr=reduce(TCC_EA0_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the TC_EA_wrreq interface. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WR_UNCACHED_32B_sum" expr=reduce(TCC_EA0_WR_UNCACHED_32B,sum) descr="Number of 32-byte write/atomic going over the TC_EA_wrreq interface due to uncached traffic. Note that CC mtypes can produce uncached requests, and those are included in this. A 64-byte request will be counted as 2. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_STALL_sum" expr=reduce(TCC_EA0_WRREQ_STALL,sum) descr="Number of cycles a write request was stalled. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_IO_CREDIT_STALL_sum" expr=reduce(TCC_EA0_WRREQ_IO_CREDIT_STALL,sum) descr="Number of cycles a EA write request was stalled because the interface was out of IO credits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_GMI_CREDIT_STALL_sum" expr=reduce(TCC_EA0_WRREQ_GMI_CREDIT_STALL,sum) descr="Number of cycles a EA write request was stalled because the interface was out of GMI credits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_DRAM_CREDIT_STALL_sum" expr=reduce(TCC_EA0_WRREQ_DRAM_CREDIT_STALL,sum) descr="Number of cycles a EA write request was stalled because the interface was out of DRAM credits. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_TOO_MANY_EA_WRREQS_STALL_sum" expr=reduce(TCC_TOO_MANY_EA_WRREQS_STALL,sum) descr="Number of cycles the TCC could not send a EA write request because it already reached its maximum number of pending EA write requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_LEVEL_sum" expr=reduce(TCC_EA0_WRREQ_LEVEL,sum) descr="The sum of the number of EA write requests in flight. This is primarily meant for measure average EA write latency. Average write latency = TCC_PERF_SEL_EA_WRREQ_LEVEL/TCC_PERF_SEL_EA_WRREQ. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_LEVEL_sum" expr=reduce(TCC_EA0_RDREQ_LEVEL,sum) descr="The sum of the number of TCC/EA read requests in flight. This is primarily meant for measure average EA read latency. Average read latency = TCC_PERF_SEL_EA_RDREQ_LEVEL/TCC_PERF_SEL_EA_RDREQ. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_ATOMIC_sum" expr=reduce(TCC_EA0_ATOMIC,sum) descr="Number of transactions going over the TC_EA_wrreq interface that are actually atomic requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_ATOMIC_LEVEL_sum" expr=reduce(TCC_EA0_ATOMIC_LEVEL,sum) descr="The sum of the number of EA atomics in flight. This is primarily meant for measure average EA atomic latency. Average atomic latency = TCC_PERF_SEL_EA_WRREQ_ATOMIC_LEVEL/TCC_PERF_SEL_EA_WRREQ_ATOMIC. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_sum" expr=reduce(TCC_EA0_RDREQ,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte) Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_32B_sum" expr=reduce(TCC_EA0_RDREQ_32B,sum) descr="Number of 32-byte TCC/EA read requests Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RD_UNCACHED_32B_sum" expr=reduce(TCC_EA0_RD_UNCACHED_32B,sum) descr="Number of 32-byte TCC/EA read due to uncached traffic. A 64-byte request will be counted as 2 Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_IO_CREDIT_STALL_sum" expr=reduce(TCC_EA0_RDREQ_IO_CREDIT_STALL,sum) descr="Number of cycles there was a stall because the read request interface was out of IO credits. Stalls occur regardless of whether a read needed to be performed or not. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_GMI_CREDIT_STALL_sum" expr=reduce(TCC_EA0_RDREQ_GMI_CREDIT_STALL,sum) descr="Number of cycles there was a stall because the read request interface was out of GMI credits. Stalls occur regardless of whether a read needed to be performed or not. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_DRAM_CREDIT_STALL_sum" expr=reduce(TCC_EA0_RDREQ_DRAM_CREDIT_STALL,sum) descr="Number of cycles there was a stall because the read request interface was out of DRAM credits. Stalls occur regardless of whether a read needed to be performed or not. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_TAG_STALL_sum" expr=reduce(TCC_TAG_STALL,sum) descr="Total number of cycles the normal request pipeline in the tag is stalled for any reason."></metric>
|
||||
<metric name="TCC_NORMAL_WRITEBACK_sum" expr=reduce(TCC_NORMAL_WRITEBACK,sum) descr="Number of writebacks due to requests that are not writeback requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_WB_WRITEBACK_sum" expr=reduce(TCC_ALL_TC_OP_WB_WRITEBACK,sum) descr="Number of writebacks due to all TC_OP writeback requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_NORMAL_EVICT_sum" expr=reduce(TCC_NORMAL_EVICT,sum) descr="Number of evictions due to requests that are not invalidate or probe requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_ALL_TC_OP_INV_EVICT_sum" expr=reduce(TCC_ALL_TC_OP_INV_EVICT,sum) descr="Number of evictions due to all TC_OP invalidate requests. Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_RDREQ_DRAM_sum" expr=reduce(TCC_EA0_RDREQ_DRAM,sum) descr="Number of TCC/EA read requests (either 32-byte or 64-byte) destined for DRAM (MC). Sum over TCC instances."></metric>
|
||||
<metric name="TCC_EA0_WRREQ_DRAM_sum" expr=reduce(TCC_EA0_WRREQ_DRAM,sum) descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC). Sum over TCC instances."></metric>
|
||||
<metric name="FETCH_SIZE" expr=(TCC_EA0_RDREQ_32B_sum*32+(TCC_EA0_RDREQ_sum-TCC_EA0_RDREQ_32B_sum)*64)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_SIZE" expr=((TCC_EA0_WRREQ_sum-TCC_EA0_WRREQ_64B_sum)*32+TCC_EA0_WRREQ_64B_sum*64)/1024 descr="The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WRITE_REQ_32B" expr=TCC_EA0_WRREQ_64B_sum*2+(TCC_EA0_WRREQ_sum-TCC_EA0_WRREQ_64B_sum) descr="The total number of 32-byte effective memory writes."></metric>
|
||||
<metric name="CU_OCCUPANCY" expr=(SQ_CYCLES/(SQ_WAVE_CYCLES*4))/MAX_WAVE_SIZE descr="The ratio of active waves on a CU to the maximum number of active waves supported by the CU"></metric>
|
||||
<metric name="CU_UTILIZATION" expr=GRBM_GUI_ACTIVE/GRBM_COUNT descr="The total number of active cycles divided by total number of elapsed cycles"></metric>
|
||||
<metric name="TOTAL_16_OPS" expr=(SQ_INSTS_VALU_FMA_F16*2+SQ_INSTS_VALU_ADD_F16+SQ_INSTS_VALU_MUL_F16+SQ_INSTS_VALU_TRANS_F16)*64+((SQ_INSTS_VALU_MFMA_MOPS_F16+SQ_INSTS_VALU_MFMA_MOPS_BF16)*512) descr="The number of 16 bits OPS executed"></metric>
|
||||
<metric name="TOTAL_32_OPS" expr=(SQ_INSTS_VALU_FMA_F32*2+SQ_INSTS_VALU_INT32+SQ_INSTS_VALU_ADD_F32+SQ_INSTS_VALU_MUL_F32+SQ_INSTS_VALU_TRANS_F32)*64+(SQ_INSTS_VALU_MFMA_MOPS_F32*512) descr="The number of 32 bits OPS executed"></metric>
|
||||
<metric name="TOTAL_64_OPS" expr=(SQ_INSTS_VALU_FMA_F64*2+SQ_INSTS_VALU_INT64+SQ_INSTS_VALU_ADD_F64+SQ_INSTS_VALU_MUL_F64)*64+(SQ_INSTS_VALU_MFMA_MOPS_F64*512) descr="The number of 64 bits OPS executed"></metric>
|
||||
<metric name="GPU_UTIL" expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Percentage of the time that GUI is active"></metric>
|
||||
</gfx940>
|
||||
|
||||
<gfx10 base="common_derived">
|
||||
<metric name="SQ_WAVES_sum" expr=reduce(SQ_WAVES,sum) descr="Count number of waves sent to SQs. (per-simd, emulated, global). Sum over SQ instances."></metric>
|
||||
<metric name="MeanOccupancyPerCU" expr=GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV/GRBM_GUI_ACTIVE/CU_NUM descr="Mean occupancy per compute unit."></metric>
|
||||
<metric name="MeanOccupancyPerActiveCU" expr=GRBM_COUNT*0+SQ_LEVEL_WAVES*0+SQ_ACCUM_PREV*4/SQ_BUSY_CYCLES/CU_NUM descr="Mean occupancy per active compute unit."></metric>
|
||||
|
||||
<metric name="GPU_UTIL" expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Percentage of the time that GUI is active"></metric>
|
||||
<metric name="CP_UTIL" expr=100*GRBM_CP_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that any of the Command Processor (CPG/CPC/CPF) blocks are busy"></metric>
|
||||
<metric name="SPI_UTIL" expr=100*GRBM_SPI_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that any of the Shader Pipe Interpolators (SPI) are busy in the shader engine(s)"></metric>
|
||||
<metric name="TA_UTIL" expr=100*GRBM_TA_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that any of the Texture Pipes (TA) are busy in the shader engine(s)."></metric>
|
||||
<metric name="GDS_UTIL" expr=100*GRBM_GDS_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that the Global Data Share (GDS) is busy."></metric>
|
||||
<metric name="EA_UTIL" expr=100*GRBM_EA_BUSY/GRBM_GUI_ACTIVE descr="Percentage of the GRBM_GUI_ACTIVE time that the Efficiency Arbiter (EA) block is busy."></metric>
|
||||
<metric name="WAVE_DEP_WAIT" expr=100*SQ_WAIT_ANY/SQ_WAVE_CYCLES descr="Percentage of the SQ_WAVE_CYCLE time spent waiting for anything."></metric>
|
||||
<metric name="WAVE_ISSUE_WAIT" expr=100*SQ_WAIT_INST_ANY/SQ_WAVE_CYCLES descr="Percentage of the SQ_WAVE_CYCLE time spent waiting for any instruction issue."></metric>
|
||||
|
||||
<metric name="TA_BUSY_avr" expr=reduce(TA_TA_BUSY,avr) descr="TA block is busy. Average over TA instances."></metric>
|
||||
<metric name="TA_BUSY_max" expr=reduce(TA_TA_BUSY,max) descr="TA block is busy. Max over TA instances."></metric>
|
||||
<metric name="TA_BUSY_min" expr=reduce(TA_TA_BUSY,min) descr="TA block is busy. Min over TA instances."></metric>
|
||||
<metric name="TA_FLAT_LOAD_WAVEFRONTS_sum" expr=reduce(TA_FLAT_LOAD_WAVEFRONTS,sum) descr="Number of flat load vec32 packets processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_FLAT_STORE_WAVEFRONTS_sum" expr=reduce(TA_FLAT_STORE_WAVEFRONTS,sum) descr="Number of flat store vec32 packets processed by the TA. Sum over TA instances."></metric>
|
||||
|
||||
<metric name="GL2C_HIT_sum" expr=reduce(GL2C_HIT,sum) descr="Number of cache hits. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_MISS_sum" expr=reduce(GL2C_MISS,sum) descr="Number of cache misses. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_32B_sum" expr=reduce(GL2C_EA_RDREQ_32B,sum) descr="Number of 32-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_64B_sum" expr=reduce(GL2C_EA_RDREQ_64B,sum) descr="Number of 64-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_96B_sum" expr=reduce(GL2C_EA_RDREQ_96B,sum) descr="Number of 96-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_128B_sum" expr=reduce(GL2C_EA_RDREQ_128B,sum) descr="Number of 128-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_MC_RDREQ_sum" expr=reduce(GL2C_MC_RDREQ,sum) descr="Number of GL2C/EA read requests (either 32-byte or 64-byte or 128-byte). Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_MC_WRREQ_sum" expr=reduce(GL2C_MC_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the GL2C_MC_wrreq interface. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_WRREQ_64B_sum" expr=reduce(GL2C_EA_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the GL2C_EA_wrreq interface. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_WRREQ_STALL_max" expr=reduce(GL2C_MC_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over GL2C instances."></metric>
|
||||
<metric name="L2CacheHit" expr=100*reduce(GL2C_HIT,sum)/(reduce(GL2C_HIT,sum)+reduce(GL2C_MISS,sum)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
|
||||
<metric name="FETCH_SIZE" expr=(GL2C_EA_RDREQ_32B_sum*32+GL2C_EA_RDREQ_64B_sum*64+GL2C_EA_RDREQ_96B_sum*96+GL2C_EA_RDREQ_128B_sum*128)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WriteUnitStalled" expr=100*GL2C_WRREQ_STALL_max/GRBM_GUI_ACTIVE descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."></metric>
|
||||
<metric name="LDSBankConflict" expr=100*SQC_LDS_BANK_CONFLICT/SQC_LDS_IDX_ACTIVE descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
</gfx10>
|
||||
|
||||
<gfx1030 base="gfx10">
|
||||
</gfx1030>
|
||||
|
||||
<gfx1031 base="gfx10">
|
||||
</gfx1031>
|
||||
|
||||
<gfx1010 base="gfx10">
|
||||
</gfx1010>
|
||||
|
||||
<gfx1032 base="gfx10">
|
||||
</gfx1032>
|
||||
|
||||
<gfx11 base="common_derived">
|
||||
<metric name="SQ_WAVES_sum" expr=reduce(SQ_WAVES,sum) descr="Count number of waves sent to SQs. (per-simd, emulated, global). Sum over SQ instances."></metric>
|
||||
<metric name="GPU_UTIL" expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT descr="Percentage of the time that GUI is active"></metric>
|
||||
<metric name="WAVE_DEP_WAIT" expr=100*SQ_WAIT_ANY/SQ_WAVE_CYCLES descr="Percentage of the SQ_WAVE_CYCLE time spent waiting for anything."></metric>
|
||||
<metric name="WAVE_ISSUE_WAIT" expr=100*SQ_WAIT_INST_ANY/SQ_WAVE_CYCLES descr="Percentage of the SQ_WAVE_CYCLE time spent waiting for any instruction issue."></metric>
|
||||
<metric name="TA_BUSY_avr" expr=reduce(TA_TA_BUSY,avr) descr="TA block is busy. Average over TA instances."></metric>
|
||||
<metric name="TA_BUSY_max" expr=reduce(TA_TA_BUSY,max) descr="TA block is busy. Max over TA instances."></metric>
|
||||
<metric name="TA_BUSY_min" expr=reduce(TA_TA_BUSY,min) descr="TA block is busy. Min over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_LOAD_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_LOAD_WAVEFRONTS,sum) descr="Number of buffer load vec32 packets processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="TA_BUFFER_STORE_WAVEFRONTS_sum" expr=reduce(TA_BUFFER_STORE_WAVEFRONTS,sum) descr="Number of buffer store vec32 packets processed by the TA. Sum over TA instances."></metric>
|
||||
<metric name="GL2C_HIT_sum" expr=reduce(GL2C_HIT,sum) descr="Number of cache hits. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_MISS_sum" expr=reduce(GL2C_MISS,sum) descr="Number of cache misses. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_32B_sum" expr=reduce(GL2C_EA_RDREQ_32B,sum) descr="Number of 32-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_64B_sum" expr=reduce(GL2C_EA_RDREQ_64B,sum) descr="Number of 64-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_96B_sum" expr=reduce(GL2C_EA_RDREQ_96B,sum) descr="Number of 96-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_RDREQ_128B_sum" expr=reduce(GL2C_EA_RDREQ_128B,sum) descr="Number of 128-byte GL2C/EA read requests. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_MC_RDREQ_sum" expr=reduce(GL2C_MC_RDREQ,sum) descr="Number of GL2C/EA read requests (either 32-byte or 64-byte or 128-byte). Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_MC_WRREQ_sum" expr=reduce(GL2C_MC_WRREQ,sum) descr="Number of transactions (either 32-byte or 64-byte) going over the GL2C_MC_wrreq interface. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_EA_WRREQ_64B_sum" expr=reduce(GL2C_EA_WRREQ_64B,sum) descr="Number of 64-byte transactions going (64-byte write or CMPSWAP) over the GL2C_EA_wrreq interface. Sum over GL2C instances."></metric>
|
||||
<metric name="GL2C_WRREQ_STALL_max" expr=reduce(GL2C_MC_WRREQ_STALL,max) descr="Number of cycles a write request was stalled. Max over GL2C instances."></metric>
|
||||
<metric name="L2CacheHit" expr=100*reduce(GL2C_HIT,sum)/(reduce(GL2C_HIT,sum)+reduce(GL2C_MISS,sum)) descr="The percentage of fetch, write, atomic, and other instructions that hit the data in L2 cache. Value range: 0% (no hit) to 100% (optimal)."></metric>
|
||||
<metric name="FETCH_SIZE" expr=(GL2C_EA_RDREQ_32B_sum*32+GL2C_EA_RDREQ_64B_sum*64+GL2C_EA_RDREQ_96B_sum*96+GL2C_EA_RDREQ_128B_sum*128)/1024 descr="The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account."></metric>
|
||||
<metric name="WriteUnitStalled" expr=100*GL2C_WRREQ_STALL_max/GRBM_GUI_ACTIVE descr="The percentage of GPUTime the Write unit is stalled. Value range: 0% to 100% (bad)."></metric>
|
||||
<metric name="LDSBankConflict" expr=100*SQC_LDS_BANK_CONFLICT/SQC_LDS_IDX_ACTIVE descr="The percentage of GPUTime LDS is stalled by bank conflicts. Value range: 0% (optimal) to 100% (bad)."></metric>
|
||||
</gfx11>
|
||||
|
||||
<gfx1100 base="gfx11">
|
||||
</gfx1100>
|
||||
|
||||
<gfx1101 base="gfx11">
|
||||
</gfx1101>
|
||||
|
||||
<gfx1102 base="gfx11">
|
||||
</gfx1102>
|
||||
|
||||
<gfx11 base="gfx11"></gfx11>
|
||||
#Mi300
|
||||
<gfx941 base="gfx940"></gfx941>
|
||||
<gfx942 base="gfx940"></gfx942>
|
||||
#Navi21
|
||||
<gfx1032 base="gfx1032"></gfx1032>
|
||||
@@ -1,6 +0,0 @@
|
||||
configure_file(counter_defs.yaml
|
||||
${PROJECT_BINARY_DIR}/share/rocprofiler-sdk/counter_defs.yaml COPYONLY)
|
||||
install(
|
||||
FILES ${PROJECT_BINARY_DIR}/share/rocprofiler-sdk/counter_defs.yaml
|
||||
DESTINATION share/rocprofiler-sdk
|
||||
COMPONENT core)
|
||||
File diff suppressed because it is too large
Load Diff
@@ -5,14 +5,18 @@ project(rocprofiler-sdk-tests-common LANGUAGES C CXX)
|
||||
|
||||
include(GoogleTest)
|
||||
|
||||
set(common_sources demangling.cpp environment.cpp md5sum.cpp mpl.cpp parse.cpp)
|
||||
set(common_sources demangling.cpp environment.cpp md5sum.cpp mpl.cpp parse.cpp sha256.cpp
|
||||
uuid_v7.cpp)
|
||||
|
||||
add_executable(common-tests)
|
||||
target_sources(common-tests PRIVATE ${common_sources})
|
||||
target_link_libraries(
|
||||
common-tests
|
||||
PRIVATE rocprofiler-sdk::rocprofiler-sdk-headers
|
||||
rocprofiler-sdk::rocprofiler-sdk-common-library GTest::gtest
|
||||
rocprofiler-sdk::rocprofiler-sdk-common-library
|
||||
rocprofiler-sdk::rocprofiler-sdk-output-library
|
||||
rocprofiler-sdk::rocprofiler-sdk-cereal
|
||||
GTest::gtest
|
||||
GTest::gtest_main)
|
||||
|
||||
gtest_add_tests(
|
||||
@@ -24,4 +28,4 @@ gtest_add_tests(
|
||||
set_tests_properties(
|
||||
${common-tests_TESTS}
|
||||
PROPERTIES TIMEOUT 45 LABELS "unittests" FAIL_REGULAR_EXPRESSION
|
||||
"${ROCPROFILER_DEFAULT_FAIL_REGEX}")
|
||||
"${ROCPROFILER_DEFAULT_FAIL_REGEX}" ENVIRONMENT "TEST_LOG_LEVEL=info")
|
||||
|
||||
@@ -22,8 +22,14 @@
|
||||
|
||||
#include "lib/common/environment.hpp"
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
namespace
|
||||
{
|
||||
bool _common_environment_test_init_logging = (rocprofiler::common::init_logging("TEST"), true);
|
||||
}
|
||||
|
||||
TEST(common, environment)
|
||||
{
|
||||
using rocprofiler::common::env_config;
|
||||
@@ -113,3 +119,42 @@ TEST(common, environment)
|
||||
EXPECT_TRUE(get_env("ROCPROFILER_ENV_TEST_BOOL", false));
|
||||
}
|
||||
}
|
||||
|
||||
TEST(common, environment_push_pop)
|
||||
{
|
||||
using rocprofiler::common::env_config;
|
||||
using rocprofiler::common::env_store;
|
||||
using rocprofiler::common::get_env;
|
||||
|
||||
namespace common = ::rocprofiler::common;
|
||||
|
||||
common::set_env("ROCPROFILER_ENV_TEST_A", "0", 1);
|
||||
common::set_env("ROCPROFILER_ENV_TEST_B", "2", 1);
|
||||
|
||||
auto _store = env_store{{env_config{"ROCPROFILER_ENV_TEST_A", "1"},
|
||||
env_config{"ROCPROFILER_ENV_TEST_B", "2", 1},
|
||||
env_config{"ROCPROFILER_ENV_TEST_C", "3", 0}}};
|
||||
|
||||
for(size_t i = 0; i < 5; ++i)
|
||||
{
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_A", 3), 0) << fmt::format("iteration={}", i);
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_B", 0), 2) << fmt::format("iteration={}", i);
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_C", 1), 1) << fmt::format("iteration={}", i);
|
||||
|
||||
EXPECT_FALSE(_store.is_pushed()) << fmt::format("iteration={}", i);
|
||||
EXPECT_TRUE(_store.push()) << fmt::format("iteration={}", i);
|
||||
EXPECT_FALSE(_store.push()) << fmt::format("iteration={}", i);
|
||||
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_A", 3), 1) << fmt::format("iteration={}", i);
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_B", 0), 2) << fmt::format("iteration={}", i);
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_C", 1), 3) << fmt::format("iteration={}", i);
|
||||
|
||||
EXPECT_TRUE(_store.is_pushed()) << fmt::format("iteration={}", i);
|
||||
EXPECT_TRUE(_store.pop()) << fmt::format("iteration={}", i);
|
||||
EXPECT_FALSE(_store.pop()) << fmt::format("iteration={}", i);
|
||||
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_A", 3), 0) << fmt::format("iteration={}", i);
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_B", 0), 2) << fmt::format("iteration={}", i);
|
||||
EXPECT_EQ(get_env("ROCPROFILER_ENV_TEST_C", 0), 0) << fmt::format("iteration={}", i);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -0,0 +1,57 @@
|
||||
// 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.
|
||||
|
||||
#include "lib/common/sha256.hpp"
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
#include <cstddef>
|
||||
#include <cstdint>
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
|
||||
TEST(common, sha256)
|
||||
{
|
||||
auto _val = rocprofiler::common::sha256{};
|
||||
|
||||
_val.update("rocprofiler-sdk|rocprofiler-sdk-roctx|rocprofiler-sdk-rocpd");
|
||||
|
||||
auto _hex_digest = _val.hexdigest();
|
||||
EXPECT_EQ(_hex_digest, "e58b701acc3c524f881e49fff2833879eb17975b6a072d9ecc27de5a5344aefc");
|
||||
|
||||
auto read_hex = [](std::string&& _inp) {
|
||||
uint32_t _ret = 0;
|
||||
std::stringstream{_inp} >> std::hex >> _ret;
|
||||
return _ret;
|
||||
};
|
||||
|
||||
auto _raw_digest = _val.rawdigest();
|
||||
for(size_t i = 0; i < _raw_digest.size(); ++i)
|
||||
{
|
||||
auto _sub = _hex_digest.substr(i * 8, 8);
|
||||
auto _extract = read_hex(fmt::format("0x{}", _sub));
|
||||
int _raw = _raw_digest.at(i);
|
||||
EXPECT_EQ(_raw, _extract) << fmt::format(
|
||||
"i={}, raw[i]={}, hex={}, hexdigest={}", i, _raw, _sub, _hex_digest);
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,80 @@
|
||||
// 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.
|
||||
|
||||
#include "lib/common/uuid_v7.hpp"
|
||||
#include "lib/common/utility.hpp"
|
||||
#include "lib/output/node_info.hpp"
|
||||
|
||||
#include <fmt/format.h>
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
#include <cstddef>
|
||||
#include <cstdint>
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
|
||||
TEST(common, uuid_v7)
|
||||
{
|
||||
namespace common = ::rocprofiler::common;
|
||||
namespace tool = ::rocprofiler::tool;
|
||||
|
||||
auto _node_data = tool::read_node_info();
|
||||
const auto& _mach_id = _node_data.machine_id;
|
||||
auto _this_pid = getpid();
|
||||
auto _this_ppid = getppid();
|
||||
auto _init_ns = common::timestamp_ns();
|
||||
auto _ticks = common::get_process_start_ticks_since_boot(_this_pid);
|
||||
|
||||
auto get_uuid_v7 = [&_mach_id, &_this_pid, &_this_ppid, &_ticks](auto _timestamp_ns) {
|
||||
auto _seed = common::compute_system_seed(_mach_id, _this_pid, _this_ppid, _ticks);
|
||||
return common::generate_uuid_v7(_timestamp_ns, _seed);
|
||||
};
|
||||
|
||||
auto reference_uuid_v7 = get_uuid_v7(_init_ns);
|
||||
|
||||
// uuid v7 has millisecond precision
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds{5});
|
||||
for(size_t i = 0; i < 1000; ++i)
|
||||
{
|
||||
auto varying_ns = common::timestamp_ns();
|
||||
auto varying_uuid_v7 = get_uuid_v7(varying_ns);
|
||||
|
||||
EXPECT_NE(reference_uuid_v7, varying_uuid_v7)
|
||||
<< fmt::format("machine-id={}, pid={}, ppid={}, ticks={}, init_ns={}, varying_ns={}",
|
||||
_mach_id,
|
||||
_this_pid,
|
||||
_this_ppid,
|
||||
_ticks,
|
||||
_init_ns,
|
||||
varying_ns);
|
||||
|
||||
// UUIDv7 is supposed to be lexicographically sortable
|
||||
EXPECT_LT(reference_uuid_v7, varying_uuid_v7)
|
||||
<< fmt::format("machine-id={}, pid={}, ppid={}, ticks={}, init_ns={}, varying_ns={}",
|
||||
_mach_id,
|
||||
_this_pid,
|
||||
_this_ppid,
|
||||
_ticks,
|
||||
_init_ns,
|
||||
varying_ns);
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user