AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Roctracer flush activity fix + perfetto.cfg (#317)

Browse Source 
* Fix roctracer_flush_activity

- invoke roctracer_flush_activity() before disabling domains

* create comp::roctracer::flush()

- real issue was the global state when roctracer_flush_activity() was called

* formatting

* Update lib/omnitrace/library/components/roctracer.hpp

- provide definition of comp::roctracer::flush when OMNITRACE_USE_ROCTRACER is not defined

* omnitrace.cfg -> perfetto.cfg

- rename provided perfetto config file (omnitrace.cfg) to perfetto.cfg to avoid confusion

* Update lib/core

- gpu.hpp: defines for OMNITRACE_USE_{HIP,ROCTRACER,ROCPROFILER,ROCM_SMI}
- gpu.cpp
  - include core/hip_runtime.hpp
  - fix serialization of hipDeviceProp_t
- add hip_runtime.hpp
  -  ensure proper inclusion of hip_runtime.h
- add rccl.hpp
  - ensure proper inclusion of rccl.h

* Update lib/omnitrace/library

- rcclp.cpp
  - update includes for rccl
- roctracer.hpp
  - update includes for hip_runtime
- components/comm_data.hpp
  - update includes for rccl
- components/rcclp.hpp
  - update includes for rccl

* Update bin/omnitrace-avail/avail.cpp

- update includes for hip_runtime

* Update examples/rccl/CMakeLists.txt

- fix find_package for rccl when CI enabled

* Update CMakeLists.txt

- set cmake policy CMP0135 to NEW for cmake >= 3.24
  - Enable DOWNLOAD_EXTRACT_TIMESTAMP with ExternalProject_Add + URL download method

* Update timemory submodule

* Update pybind11 submodule

* Update pybind11 submodule

* Update lib/core/rccl.hpp

- include rccl.h only if OMNITRACE_USE_RCCL > 0

* Update lib/core/{gpu,hip_runtime}.hpp

* Update lib/core/gpu.cpp

- reintroduce some ppdefs

* Update lib/core/gpu.cpp

- fix ifdef on OMNITRACE_HIP_VERSION

* Update lib/core/gpu.cpp

- fix static assert for OMNITRACE_HIP_VERSION_MINOR when HIP version 4.x or older (unreliable minor versions)

* Update lib/core/gpu.cpp

- fix ifdef on OMNITRACE_HIP_VERSION

* Update lib/core/config.cpp

- disable OMNITRACE_PERFETTO_COMBINE_TRACES by default

* Update lib/core/perfetto.cpp

- if unable to open perfetto temp file, return the ReadTraceBlocking()

* Update lib/core/config.*

- flush tmpfile before closing
This commit is contained in:
Jonathan R. Madsen
2024-01-10 05:02:22 -06:00
committed by GitHub
parent aeb346b6d6
commit 7bc50f5a0a
24 changed files with 441 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files
source/lib/core/gpu.cpp
+229 -16
View File
@@ -20,6 +20,8 @@
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.
#include "common/defines.h"
#if !defined(OMNITRACE_USE_ROCM_SMI)
# define OMNITRACE_USE_ROCM_SMI 0
#endif
@@ -28,15 +30,17 @@
# define OMNITRACE_USE_HIP 0
#endif
#include "core/hip_runtime.hpp"
#if OMNITRACE_USE_HIP > 0
# if !defined(TIMEMORY_USE_HIP)
# define TIMEMORY_USE_HIP 1
# endif
#endif
#include "gpu.hpp"
#include "debug.hpp"
#include "defines.hpp"
#include "gpu.hpp"
#include <timemory/manager.hpp>
@@ -45,10 +49,19 @@
#endif
#if OMNITRACE_USE_HIP > 0
# include <hip/hip_runtime.h>
# include <hip/hip_runtime_api.h>
# include <timemory/components/hip/backends.hpp>
static_assert(OMNITRACE_HIP_VERSION_MAJOR == HIP_VERSION_MAJOR,
"OMNITRACE_HIP_VERSION_MAJOR (detected by cmake) != HIP_VERSION_MAJOR "
"(from <hip/hip_version.h>)");
# if OMNITRACE_HIP_VERSION_MAJOR >= 5
// HIP versions 4.x and older have unreliable values for HIP_VERSION_MINOR
static_assert(OMNITRACE_HIP_VERSION_MINOR == HIP_VERSION_MINOR,
"OMNITRACE_HIP_VERSION_MINOR (detected by cmake) != HIP_VERSION_MINOR "
"(from <hip/hip_version.h>)");
# endif
# if !defined(OMNITRACE_HIP_RUNTIME_CALL)
# define OMNITRACE_HIP_RUNTIME_CALL(err) \
{ \
@@ -107,6 +120,91 @@ rsmi_init()
return _rsmi_init;
}
#endif
#if OMNITRACE_HIP_VERSION >= 60000
template <typename ArchiveT, typename ArgT,
std::enable_if_t<!std::is_pointer<ArgT>::value, int> = 0>
void
device_prop_serialize(ArchiveT& archive, const char* name, const ArgT& arg)
{
namespace cereal = tim::cereal;
using cereal::make_nvp;
archive(make_nvp(name, arg));
}
template <typename ArchiveT, typename ArgT, size_t N>
void
device_prop_serialize(ArchiveT& archive, const char* name, ArgT arg[N])
{
if constexpr(!std::is_same<ArgT, char>::value &&
!std::is_same<ArgT, const char>::value)
{
namespace cereal = tim::cereal;
using cereal::make_nvp;
auto data = std::array<int, N>{};
for(size_t i = 0; i < N; ++i)
data[i] = arg[i];
archive(make_nvp(name, data));
}
else
{
device_prop_serialize(archive, name, std::string{ arg });
}
}
template <typename ArchiveT>
void
device_prop_serialize(ArchiveT& archive, const char* name, hipUUID_t arg)
{
constexpr auto N = sizeof(arg.bytes);
namespace cereal = tim::cereal;
using cereal::make_nvp;
auto data = std::array<char, N + 1>{};
data.fill('\0');
for(size_t i = 0; i < N; ++i)
data[i] = arg.bytes[i];
auto str_v = std::string_view{ data.data() };
auto str = std::string{ str_v }.substr(0, str_v.find('\0'));
archive(make_nvp(name, str));
}
template <typename ArchiveT>
void
device_prop_serialize(ArchiveT& archive, const char* name, hipDeviceArch_t arg)
{
namespace cereal = tim::cereal;
using cereal::make_nvp;
# define OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(NAME) \
{ \
auto val = arg.NAME; \
archive(make_nvp(#NAME, val)); \
}
archive.setNextName(name);
archive.startNode();
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasGlobalInt32Atomics)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasGlobalFloatAtomicExch)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasSharedInt32Atomics)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasSharedFloatAtomicExch)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasFloatAtomicAdd)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasGlobalInt64Atomics)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasSharedInt64Atomics)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasDoubles)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasWarpVote)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasWarpBallot)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasWarpShuffle)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasFunnelShift)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasThreadFenceSystem)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasSyncThreadsExt)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasSurfaceFuncs)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(has3dGrid)
OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH(hasDynamicParallelism)
archive.finishNode();
# undef OMNITRACE_SERIALIZE_HIP_DEVICE_ARCH
}
#endif
} // namespace
int
@@ -161,11 +259,10 @@ template <typename ArchiveT>
void
add_hip_device_metadata(ArchiveT& ar)
{
#if OMNITRACE_USE_HIP > 0
namespace cereal = tim::cereal;
using cereal::make_nvp;
using intvec_t = std::vector<int>;
#if OMNITRACE_USE_HIP > 0
int _device_count = 0;
int _current_device = 0;
hipError_t _device_count_err = hipGetDeviceCount(&_device_count);
@@ -183,12 +280,6 @@ add_hip_device_metadata(ArchiveT& ar)
if(_current_device_err != hipSuccess || _device_count == 0) return;
# define OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(NAME) \
ar(make_nvp(#NAME, _device_prop.NAME));
# define OMNITRACE_SERIALIZE_HIP_DEVICE_PROP_ARRAY(NAME, ...) \
ar(make_nvp(NAME, __VA_ARGS__));
ar.setNextName("hip_device_properties");
ar.startNode();
ar.makeArray();
@@ -205,6 +296,16 @@ add_hip_device_metadata(ArchiveT& ar)
OMNITRACE_HIP_RUNTIME_CALL(hipRuntimeGetVersion(&_runtime_version));
ar.startNode();
# if OMNITRACE_HIP_VERSION < 60000
using intvec_t = std::vector<int>;
# define OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(NAME) \
ar(make_nvp(#NAME, _device_prop.NAME));
# define OMNITRACE_SERIALIZE_HIP_DEVICE_PROP_ARRAY(NAME, ...) \
ar(make_nvp(NAME, __VA_ARGS__));
ar(make_nvp("name", std::string{ _device_prop.name }));
ar(make_nvp("driver_version", _driver_version));
ar(make_nvp("runtime_version", _runtime_version));
@@ -215,11 +316,11 @@ add_hip_device_metadata(ArchiveT& ar)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(totalConstMem)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(clockRate)
# if OMNITRACE_HIP_VERSION >= 5000
# if OMNITRACE_HIP_VERSION >= 50000
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memoryClockRate)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memoryBusWidth)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(l2CacheSize)
# endif
# endif
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(sharedMemPerBlock)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(regsPerBlock)
@@ -247,7 +348,6 @@ add_hip_device_metadata(ArchiveT& ar)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pciBusID)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pciDeviceID)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(computeMode)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(computeMode)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(gcnArch)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(gcnArchName)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(isMultiGpuBoard)
@@ -259,8 +359,118 @@ add_hip_device_metadata(ArchiveT& ar)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(concurrentKernels)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSharedMemoryPerMultiProcessor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(asicRevision)
# else
# define OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(NAME) \
device_prop_serialize(ar, #NAME, _device_prop.NAME);
const char* _compute_mode_descr[] = {
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(name)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(uuid)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(luid)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(luidDeviceNodeMask)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(totalGlobalMem)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(sharedMemPerBlock)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(regsPerBlock)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(warpSize)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memPitch)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxThreadsPerBlock)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxThreadsDim)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxGridSize)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(clockRate)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(totalConstMem)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(major)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(minor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(textureAlignment)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(texturePitchAlignment)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(deviceOverlap)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(multiProcessorCount)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(kernelExecTimeoutEnabled)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(integrated)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(canMapHostMemory)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(computeMode)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture1D)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture1DMipmap)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture1DLinear)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture2D)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture2DMipmap)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture2DLinear)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture2DGather)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture3D)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture3DAlt)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTextureCubemap)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture1DLayered)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTexture2DLayered)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxTextureCubemapLayered)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurface1D)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurface2D)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurface3D)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurface1DLayered)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurface2DLayered)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurfaceCubemap)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSurfaceCubemapLayered)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(surfaceAlignment)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(concurrentKernels)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(ECCEnabled)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pciBusID)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pciDeviceID)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pciDomainID)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(tccDriver)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(asyncEngineCount)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(unifiedAddressing)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memoryClockRate)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memoryBusWidth)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(l2CacheSize)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(persistingL2CacheMaxSize)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxThreadsPerMultiProcessor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(streamPrioritiesSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(globalL1CacheSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(localL1CacheSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(sharedMemPerMultiprocessor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(regsPerMultiprocessor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(managedMemory)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(isMultiGpuBoard)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(multiGpuBoardGroupID)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(hostNativeAtomicSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(singleToDoublePrecisionPerfRatio)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pageableMemoryAccess)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(concurrentManagedAccess)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(computePreemptionSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(canUseHostPointerForRegisteredMem)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(cooperativeLaunch)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(cooperativeMultiDeviceLaunch)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(sharedMemPerBlockOptin)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(pageableMemoryAccessUsesHostPageTables)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(directManagedMemAccessFromHost)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxBlocksPerMultiProcessor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(accessPolicyMaxWindowSize)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(reservedSharedMemPerBlock)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(hostRegisterSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(sparseHipArraySupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(hostRegisterReadOnlySupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(timelineSemaphoreInteropSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memoryPoolsSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(gpuDirectRDMASupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(gpuDirectRDMAFlushWritesOptions)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(gpuDirectRDMAWritesOrdering)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(memoryPoolSupportedHandleTypes)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(deferredMappingHipArraySupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(ipcEventSupported)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(clusterLaunch)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(unifiedFunctionPointers)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(gcnArchName)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(maxSharedMemoryPerMultiProcessor)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(clockInstructionRate)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(arch)
// OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(hdpMemFlushCntl)
// OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(hdpRegFlushCntl)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(cooperativeMultiDeviceUnmatchedFunc)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(cooperativeMultiDeviceUnmatchedGridDim)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(cooperativeMultiDeviceUnmatchedBlockDim)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(cooperativeMultiDeviceUnmatchedSharedMem)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(isLargeBar)
OMNITRACE_SERIALIZE_HIP_DEVICE_PROP(asicRevision)
# endif
constexpr auto _compute_mode_descr = std::array<const char*, 6>{
"Default (multiple host threads can use ::hipSetDevice() with device "
"simultaneously)",
"Exclusive (only one host thread in one process is able to use "
@@ -271,8 +481,11 @@ add_hip_device_metadata(ArchiveT& ar)
"Unknown",
nullptr
};
auto _compute_mode = std::min<int>(_device_prop.computeMode, 5);
ar(make_nvp("computeModeDescription",
std::string{ _compute_mode_descr[_device_prop.computeMode] }));
std::string{ _compute_mode_descr.at(_compute_mode) }));
ar.finishNode();
}
#else