Integrated perfetto + roctracer (#5)

- hosttrace library automatically collects and merges timestamps for HIP API calls and kernels with the host-side instrumentation
  - mostly eliminates the need for using external rocprof
- added thread_instruction_count in perfetto output
- increased hosttrace min_loop_address_range to 512
- disabled instrumenting functions with dynamic callsites by default
- miscellaneous cmake updates

* roctracer support

- fully integrated perfetto + roctracer outputs
- thread_instruction_count in perfetto
- increased min_loop_address_range to 512
- disabled instrumenting functions with dynamic callsites by default
- updated timemory submodule

* hosttrace_launch_compiler

- support for using an alternative compiler as needed via launch compiler
- elfio added as submodule (not currently used)
- miscellaneous cmake updates

* README update + host/device categories + misc

- timemory fix for TIMEMORY_ROCTRACER_ENABLED
- transpose fix

* papi_tuple_t -> papi_tot_ins

- minor fix to Findroctracer.cmake

[ROCm/rocprofiler-systems commit: a061b7947f]
This commit is contained in:
Jonathan R. Madsen
2021-09-06 22:23:24 -05:00
zatwierdzone przez GitHub
rodzic 7435122726
commit 244b308cb5
23 zmienionych plików z 1119 dodań i 268 usunięć
@@ -1,5 +1,14 @@
cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
project(hosttrace-parallel-overhead
LANGUAGES CXX)
set(CMAKE_BUILD_TYPE "Release")
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}")
find_package(Threads REQUIRED)
add_executable(parallel-overhead parallel-overhead.cpp)
target_link_libraries(parallel-overhead Threads::Threads)
if(NOT CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME)
set_target_properties(transpose PROPERTIES
RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
endif()
@@ -1,106 +1,43 @@
cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
# uses make to avoid CMAKE_CXX_COMPILER issues
project(transpose
project(hosttrace-transpose
LANGUAGES CXX)
find_program(MAKE_EXECUTABLE NAMES make gmake)
find_program(HIPCC_EXECUTABLE NAMES hipcc)
if(NOT MAKE_EXECUTABLE)
message(AUTHOR_WARNING "make/gmake could not be found. Cannot build transpose target")
return()
endif()
if(NOT HIPCC_EXECUTABLE)
message(AUTHOR_WARNING "hipcc could not be found. Cannot build transpose target")
return()
endif()
string(TOUPPER "${CMAKE_BUILD_TYPE}" BUILD_TYPE)
set(transpose_CXX_FLAGS "-W -Wall ${CMAKE_CXX_FLAGS_${BUILD_TYPE}} ${CMAKE_CXX_FLAGS}")
set(transpose_LINK_FLAGS "")
if(CMAKE_CXX_COMPILER_IS_CLANG AND TARGET hosttrace::hosttrace-compile-options)
set(transpose_CXX_FLAGS "${transpose_CXX_FLAGS} ${hosttrace_CXX_FLAGS}")
get_target_property(COMPILE_OPTS hosttrace::hosttrace-compile-options INTERFACE_COMPILE_OPTIONS)
set(transpose_CXX_FLAGS "${transpose_CXX_FLAGS} ${COMPILE_OPTS}")
endif()
option(TRANSPOSE_USE_MPI "Enable MPI support in transpose exe" ${TIMEMORY_USE_MPI})
if(TRANSPOSE_USE_MPI)
find_package(MPI REQUIRED)
endif()
if(TARGET MPI::MPI_C)
set(transpose_CXX_FLAGS "${transpose_CXX_FLAGS} -DUSE_MPI")
get_target_property(COMPILE_OPTS MPI::MPI_C INTERFACE_COMPILE_OPTIONS)
foreach(_COPT ${COMPILE_OPTS})
if(NOT "${_COPT}" MATCHES "COMPILE_LANG")
set(transpose_CXX_FLAGS "${transpose_CXX_FLAGS} ${_COPT}")
endif()
endforeach()
get_target_property(INCLUDE_DIRS MPI::MPI_C INTERFACE_INCLUDE_DIRECTORIES)
foreach(_IDIR ${INCLUDE_DIRS})
set(transpose_CXX_FLAGS "${transpose_CXX_FLAGS} -I${_IDIR}")
endforeach()
if(MPI_C_LINK_FLAGS)
set(transpose_LINK_FLAGS "${transpose_LINK_FLAGS} ${MPI_C_LINK_FLAGS}")
endif()
set(_LINK_LIBS "")
foreach(_LIB ${MPI_C_LIB_NAMES})
string(APPEND _LINK_LIBS "-l${_LIB} ")
endforeach()
foreach(_IDIR ${INCLUDE_DIRS} ${MPI_mpich_LIBRARY} ${MPI_mpi_LIBRARY} ${MPI_LIBRARY_DIRS})
get_filename_component(_LIBDIR "${_IDIR}" DIRECTORY)
if(EXISTS "${_IDIR}/libmpi${CMAKE_SHARED_LIBRARY_SUFFIX}")
set(transpose_LINK_FLAGS "${transpose_LINK_FLAGS} -L${_IDIR} ${_LINK_LIBS}")
endif()
if(EXISTS "${_LIBDIR}/libmpi${CMAKE_SHARED_LIBRARY_SUFFIX}")
set(transpose_LINK_FLAGS "${transpose_LINK_FLAGS} -L${_LIBDIR} ${_LINK_LIBS}")
endif()
foreach(_LDIR lib lib64)
set(_LIBDIR_SAVE "${_LIBDIR}")
if(NOT EXISTS "${_LIBDIR}/${_LDIR}")
get_filename_component(_LIBDIR "${_LIBDIR}" DIRECTORY)
endif()
if(EXISTS "${_LIBDIR}/${_LDIR}")
set(transpose_LINK_FLAGS "${transpose_LINK_FLAGS} -L${_LIBDIR}/${_LDIR} ${_LINK_LIBS}")
endif()
set(_LIBDIR "${_LIBDIR_SAVE}")
endforeach()
endforeach()
add_executable(transpose transpose.cpp)
if(CMAKE_CXX_COMPILER_IS_CLANG AND TARGET hosttrace::hosttrace-compile-options)
target_link_libraries(transpose PRIVATE hosttrace::hosttrace-compile-options)
else()
target_compile_options(transpose PRIVATE -W -Wall)
endif()
# remove generator expressions
string(REPLACE "$<" "" transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
string(REPLACE ">:" ":" transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
string(REPLACE ">" "" transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
string(REPLACE "COMPILE_LANGUAGE:CXX:" "" transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
# remove double spaces
string(REPLACE " " " " transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
# convert to list
string(REPLACE " " ";" transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
# remove duplicate args
list(REMOVE_DUPLICATES transpose_CXX_FLAGS)
# make single string
string(REPLACE ";" " " transpose_CXX_FLAGS "${transpose_CXX_FLAGS}")
if(TRANSPOSE_USE_MPI)
target_compile_definitions(transpose PRIVATE USE_MPI)
target_link_libraries(transpose PRIVATE MPI::MPI_C)
endif()
# configure files in build directory
configure_file(${PROJECT_SOURCE_DIR}/transpose.cpp
${CMAKE_BINARY_DIR}/transpose/transpose.cpp COPYONLY)
if(NOT CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME)
set_target_properties(transpose PROPERTIES
RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
configure_file(${PROJECT_SOURCE_DIR}/Makefile.in
${CMAKE_BINARY_DIR}/transpose/Makefile @ONLY)
# touch the file when cmake re-runs to ensure rebuilt
execute_process(COMMAND ${CMAKE_COMMAND} -E touch ${CMAKE_BINARY_DIR}/transpose/transpose.cpp)
# target for adding a dependency to libpytimemory target
add_custom_target(transpose ALL
COMMAND ${MAKE_EXECUTABLE}
WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/transpose
BYPRODUCTS ${CMAKE_BINARY_DIR}/transpose/transpose
SOURCES ${CMAKE_BINARY_DIR}/transpose/transpose.cpp)
if(NOT CMAKE_CXX_COMPILER_IS_HIPCC AND HIPCC_EXECUTABLE)
# defined in MacroUtilities.cmake
hosttrace_custom_compilation(
COMPILER ${HIPCC_EXECUTABLE}
TARGET transpose)
endif()
endif()
@@ -1,40 +0,0 @@
CXXFLAGS += -std=c++17 -O3 -Wno-unused-result -g
LINK_FLAGS :=
all: transpose
debug: CXXFLAGS += -ggdb -g
debug: transpose
transpose: transpose.cpp
hipcc ${CXXFLAGS} -o $@ $^ ${LINK_FLAGS}
.PHONY: clean
run:
./transpose
run.verbose:
AMD_LOG_LEVEL=3, AMD_LOG_MASK=128 ./transpose
rd:
rocgdb --args ./transpose
prof.trace:
rocprof --hip-trace --stats ./transpose
prof:
rocprof --obj-tracking on -i profile.txt ./transpose
prof.full:
rocprof --obj-tracking on -i profile.txt -m custom_metric.xml ./transpose
callgrind:
valgrind --tool=callgrind ./transpose
callgrind.ann:
callgrind_annotate --inclusive=yes --tree=both callgrind.out.* > callgrind.out
clean:
rm -f transpose *.o results.*
@@ -1,39 +0,0 @@
CXXFLAGS += -std=c++@CMAKE_CXX_STANDARD@ @transpose_CXX_FLAGS@
LINKFLAGS := @transpose_LINK_FLAGS@
all: transpose
debug: CXXFLAGS += -ggdb -g
debug: transpose
transpose: transpose.cpp
hipcc ${CXXFLAGS} -o $@ $^ ${LINKFLAGS}
.PHONY: clean
run:
./transpose
run.verbose:
AMD_LOG_LEVEL=3, AMD_LOG_MASK=128 ./transpose
rd:
rocgdb --args ./transpose
prof.trace:
rocprof --hip-trace --stats ./transpose
prof:
rocprof --obj-tracking on -i profile.txt ./transpose
prof.full:
rocprof --obj-tracking on -i profile.txt -m custom_metric.xml ./transpose
callgrind:
valgrind --tool=callgrind ./transpose
callgrind.ann:
callgrind_annotate --inclusive=yes --tree=both callgrind.out.* > callgrind.out
clean:
rm -f transpose *.o results.*
@@ -123,14 +123,15 @@ run(int argc, char** argv)
for(size_t i = 0; i < times; i++)
{
hipLaunchKernelGGL(transpose_a, grid, block, 0, 0, in, out, M, N);
check_hip_error();
HIP_API_CALL(hipDeviceSynchronize());
}
check_hip_error();
HIP_API_CALL(hipDeviceSynchronize());
t2 = std::chrono::high_resolution_clock::now();
double time =
std::chrono::duration_cast<std::chrono::duration<double>>(t2 - t1).count();
float GB = (float) size * times * 2 / (1 << 30);
std::cout << "The average performance of transpose is " << GB / time << " GBytes/sec"
std::cout << "Runtime of transpose is " << time << " sec\n"
<< "The average performance of transpose is " << GB / time << " GBytes/sec"
<< std::endl;
int* out_matrix = (int*) malloc(size);