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
2021-09-06 22:23:24 -05:00
parent b6e65d69f6
commit a061b7947f
23 changed files with 1119 additions and 268 deletions
@@ -0,0 +1,138 @@
+#!/bin/bash
+
+EXE=$(basename ${1})
+DIR=cpu.prof.${EXE}
+mkdir -p ${DIR}
+
+# gperf settings
+: ${N:=0}
+: ${GPERF_PROFILE:=""}
+: ${GPERF_PROFILE_BASE:=${DIR}/gperf}
+: ${MALLOCSTATS:=1}
+: ${CPUPROFILE_FREQUENCY:=250}
+: ${CPUPROFILE_REALTIME:=1}
+
+# rendering settings
+: ${INTERACTIVE:=0}
+: ${IMG_FORMAT:="png"}
+#: ${DOT_ARGS:='-Gsize=24,24\! -Gdpi=200'}
+: ${DOT_ARGS:=""}
+: ${PPROF_ARGS:="--no_strip_temp --functions"}
+
+if [ "$(uname)" = "Darwin" ]; then
+    if [ "${IMG_FORMAT}" = "jpeg" ]; then
+        IMG_FORMAT="jpg"
+    fi
+fi
+run-verbose()
+{
+    echo "${@}" 1>&2
+    eval ${@}
+}
+
+while [ -z "${GPERF_PROFILE}" ]
+do
+    TEST_FILE=${GPERF_PROFILE_BASE}.${N}
+    if [ ! -f "${TEST_FILE}" ]; then
+        GPERF_PROFILE=${TEST_FILE}
+    fi
+    N=$((${N}+1))
+done
+
+export MALLOCSTATS
+export CPUPROFILE_FREQUENCY
+export CPUPROFILE_REALTIME
+
+echo -e "\n\t--> Outputting profile to '${GPERF_PROFILE}'...\n"
+
+# remove profile file if unsucessful execution
+cleanup-failure() { set +v ; echo "failure"; rm -f ${GPERF_PROFILE}; exit 1; }
+trap cleanup-failure SIGHUP SIGINT SIGQUIT SIGILL SIGABRT SIGKILL
+
+ADD_LIBS()
+{
+    for i in $@
+    do
+        if [ -z "${ADD_LIB_LIST}" ]; then
+            ADD_LIB_LIST="--add_lib=${i}"
+        else
+            ADD_LIB_LIST="${ADD_LIB_LIST} --add_lib=${i}"
+        fi
+    done
+}
+
+ADD_PRELOAD()
+{
+    for i in $@
+    do
+        if [ -z "${LIBS}" ]; then
+            LIBS=${i}
+        else
+            LIBS="${LIBS}:${i}"
+        fi
+    done
+}
+
+# configure pre-loading of profiler library
+PROJECT_LIBRARIES="$(find ${PWD} -type f | egrep 'libtimemory|libctimemory' | egrep -v '\.a$|\.dSYM' | egrep '\.so$|\.dylib$')"
+run-verbose ADD_LIBS ${PROJECT_LIBRARIES}
+if [ "$(uname)" = "Darwin" ]; then
+    run-verbose ADD_PRELOAD $(otool -L ${1} | egrep 'profiler' | awk '{print $1}')
+    LIBS=$(echo ${LIBS} | sed 's/^://g')
+    if [ -n "${LIBS}" ]; then
+        export DYLD_FORCE_FLAT_NAMESPACE=1
+        echo "DYLD_INSERT_LIBRARIES=${LIBS}"
+    fi
+else
+    run-verbose ADD_PRELOAD $(ldd ${1} | egrep 'profiler' | awk '{print $(NF-1)}') /usr/lib/x86_64-linux-gnu/libprofiler.so
+    LIBS=$(echo ${LIBS} | sed 's/^://g')
+    if [ -n "${LIB}" ]; then
+        echo "LD_PRELOAD=${LIBS}"
+    fi
+fi
+
+set -e
+# run the application
+if [ "$(uname)" = "Darwin" ]; then
+    eval DYLD_INSERT_LIBRARIES=${LIBS} CPUPROFILE_FREQUENCY=${CPUPROFILE_FREQUENCY} CPUPROFILE=${GPERF_PROFILE} $@ | tee ${GPERF_PROFILE}.log
+else
+    eval LD_PRELOAD=${LIBS} CPUPROFILE_FREQUENCY=${CPUPROFILE_FREQUENCY} CPUPROFILE=${GPERF_PROFILE} $@ | tee ${GPERF_PROFILE}.log
+fi
+set +e
+
+echo-dart-measurement()
+{
+    local _NAME=${1}
+    local _TYPE=${2}
+    local _PATH=${3}
+    echo "<DartMeasurementFile name=\"${_NAME}\" type=\"image/${_TYPE}\">${_PATH}</DartMeasurementFile>"
+}
+
+# generate the results
+EXT=so
+if [ "$(uname)" = "Darwin" ]; then EXT=dylib; fi
+if [ -f "${GPERF_PROFILE}" ]; then
+    : ${PPROF:=$(which google-pprof)}
+    : ${PPROF:=$(which pprof)}
+    if [ -n "${PPROF}" ]; then
+        run-verbose ${PPROF} --text ${ADD_LIB_LIST} ${PPROF_ARGS} ${1} ${GPERF_PROFILE} 1> ${GPERF_PROFILE}.txt.tmp
+        run-verbose cat ${GPERF_PROFILE}.txt.tmp | c++filt -n -t &> ${GPERF_PROFILE}.txt
+        run-verbose ${PPROF} --text --cum ${ADD_LIB_LIST} ${PPROF_ARGS} ${1} ${GPERF_PROFILE} 1> ${GPERF_PROFILE}.cum.txt.tmp
+        run-verbose cat ${GPERF_PROFILE}.cum.txt.tmp | c++filt -n -t &> ${GPERF_PROFILE}.cum.txt
+        rm -f *.txt.tmp
+        # if dot is available
+        if [ -n "$(which dot)" ]; then
+            run-verbose ${PPROF} --dot ${ADD_LIB_LIST} ${PPROF_ARGS} ${1} ${GPERF_PROFILE} 1> ${GPERF_PROFILE}.dot
+            run-verbose dot ${DOT_ARGS} -T${IMG_FORMAT} ${GPERF_PROFILE}.dot -o ${GPERF_PROFILE}.${IMG_FORMAT}
+            echo-dart-measurement ${GPERF_PROFILE}.${IMG_FORMAT} ${IMG_FORMAT} ${PWD}/${GPERF_PROFILE}.${IMG_FORMAT}
+        fi
+        if [ "${INTERACTIVE}" -gt 0 ]; then
+            run-verbose ${PPROF} ${ADD_LIB_LIST} ${PPROF_ARGS} ${1} ${GPERF_PROFILE}
+        fi
+    else
+        echo -e "google-pprof/pprof not found!"
+    fi
+else
+    echo -e "profile file \"${GPERF_PROFILE}\" not found!"
+    ls -la
+fi